Datasets:

timaeus

/

pile-pubmed_central

like 0

Follow

Timaeus 9

1. Introduction {#sec1} =============== The important recent achievement in Parkinson\'s disease (PD) research has been the identification of several causative and risk genes with their putative functions. Leucine-rich repeat kinase 2 (LRRK2) gene has been distinguishable from other known PD genes with a number of functional and epidemiologic features. LRRK2 is a highly conserved and widely expressed gene that encodes a unique multifunctional and multidomain protein, named dardarin \[[@B1]\]. Dardarin is a complex chain of 2 527 polypeptides containing two distinct enzymes, namely protein kinase and guanosine triphosphatase (GTPase), as well as multiple protein interaction domains \[[@B2]\]. These domains might interact with each other and other cell signaling proteins, thus playing a putatively key role in cellular function \[[@B3]--[@B5]\]. Intriguing is the fact that almost every LRRK2 domain is susceptible to PD-associated mutations resulting largely in idiopathic PD- (iPD-) like phenotype and pleomorphic neuronal pathology \[[@B2], [@B4]\]. Amongst the number of known LRRK2 mutations, p.Gly2019Ser mutation has emerged as an important determinant of familial autosomal dominant PD and iPD in North African Arabs, Ashkenazi Jews, and to a lesser degree in European and North American populations \[[@B6]\]. Interestingly, it appears that some LRRK2 mutations and disease-associated variants are specific to particular ethnic groups, most likely due to common founder effects \[[@B7]\]. This is evidently applicable to p.Gly2019Ser mutation, which is common in PD patients from the Western hemisphere and has not yet been reported in the big PD cohorts from East Asia. Similarly, several PD risk variants of LRRK2 including p.Gly2385Arg, p.Ala419Val, and p.Arg1628Pro have only been reported in East Asian populations and have been absent in the Western PD cohorts \[[@B8]\]. PD genetics has been largely unexplored in several world regions, including Central Asia. Here, we investigate 8 LRRK2 mutations and East Asian risk variants in an interesting population residing between Europe and Asia, in the first cohort of PD patients and healthy controls from Kazakhstan. 2. Study Methodology {#sec2} ==================== 2.1. Study Subjects {#sec2.1} ------------------- A total of 246 PD patients were consecutively recruited, with no regard to nationality, during 14 months from the National Center for Neurosurgery in Nur-Sultan city, movement disorders clinics in Almaty city, and a regional hospital in Shymkent city in Kazakhstan. The diagnoses of clinically established and clinically probable PD were made on the basis of the agreement between two movement disorder specialists according to the Movement Disorders Society (MDS) PD criteria \[[@B9]\]. Both iPD and familial PD cases with their available first-degree relatives were included in the study. Clinicodemographic characteristics of the cohort were uploaded to University College London (UCL) Research Data Capturing Database (Redcap) online secure database and its summary is given in [Table 1](#tab1){ref-type="table"}. There were 21 patients (8.5%, 21/246) with a family history of PD and/or tremor, and 31 (12.6%, 31/246) patients with the onset of PD before 40 years of age. The male to female ratio was 0.95 : 1. The mean age of PD onset was 55.06 ± 11.15 (range 14--77), and the mean age at the last examination was 61.7 ± 10.3 (range 28--83). Self-reported nationalities in 72.8% of the cohort were Kazakh, 20.7% were Russian, and the remaining nationalities were Uyghur (2.8%), Tatar (2%), Korean (1.3%), and Tajik (0.4%). Genomic DNAs of age- and gender-matched 200 unrelated control subjects were obtained from the research-ready database of neurologically healthy Kazakhs from the National Center for Biotechnology, Nur-Sultan (NCB). This study was approved by the Research Ethics Committee of NCB (4/29.08.2017) and the Institute of Neurology University College London (IoN UCL) (07/Q0512/26). Written informed consent for participation in the study was obtained from each subject. All personal information was hidden with unique study identifiers. 2.2. Genetic Analysis {#sec2.2} --------------------- Whole venous blood was collected from the subjects at the recruitment centers and sent to NCB for DNA extraction using the standardized laboratory protocols. DNAs were then shipped to IoN UCL for genetic analysis. At IoN, DNAs were checked for quality and concentrations using NanoDrop Spectrophotometer (Thermo Scientific, Waltham, MA, USA). Samples were uniformly diluted to 25 ng/*μ*l, and 25 *μ*l volume of DNA from each sample was transferred to 96-well plates. On the basis of the literature review, 8 LRRK2 single-nucleotide polymorphisms (SNPs) of interest were selected for the analysis. Five of them are commonly reported LRRK2 mutations: (c.6055G\>A) p.Gly2019Ser (rs34637584), (c.4322G\>A) p.Arg1441His (rs34995376), (c.5096A\>G) p.Tyr1699Cys (rs35801418), (c.6059T\>C) p.Ile2020Thr (rs35870237), and (c.4309A\>C) p.Asn1437His (rs74163686); and three of them are the East Asian-specific PD-associated variants: (c.7153G\>A) p.Gly2385Arg (rs34778348), (c.1256C\>T) p.Ala419Val (rs34594498), and (c.4883G\>C) p.Arg1628Pro (rs33949390) ([Figure 1](#fig1){ref-type="fig"}). The SNPs and surrounding 50 base-pairs were annotated in Ensembl genome browser (see S1 in the supplementary material for comprehensive analysis). The design of primers, adaptation of assays, and genetic analysis were performed in LGC Genomics, England. No positive and negative controls were available for the selected LRRK2 SNPs. LRRK2 genotyping was done using Kompetitive Allele-Specific Polymerase chain reaction (PCR) genotyping assay (KASP™, LGC Genomics. Herts, UK), a method that enables biallelic scoring of SNPs and insertion and deletions at specific loci through competitive allele-specific PCR. Genotyping followed the LGC Genomics protocol. SNP viewer Software (version 1.99, Hoddesdon, UK) was used to visualize the genotyping results (<https://www.biosearchtech.com/support/tools/genotyping-software/snpviewer>). Familial and young-onset PD cases with positive LRRK2 substitutions were sent for exome sequencing (WES) to Macrogen, South Korea. In addition, when LRRK2 mutations were identified in a proband, Sanger sequencing was performed for all available family members. Thus, we enrolled 13 additional living relatives for the p.Gly2385Arg and p.Ala419Val mutations from two families. 2.3. Statistical Analysis {#sec2.3} ------------------------- Statistical analysis was performed using IBM SPSPS version 21 (Chicago, USA). Genotype frequency distributions were tested for conformity to Hardy--Weinberg equilibrium (HWE). *T*-test and chi-squared tests were used at the level of significance 0.05. Odds ratios (OR) were calculated and presented with 95% confidence interval (CI) values. 3. Results {#sec3} ========== 3.1. KASP Coverage {#sec3.1} ------------------ 1.6--3.2% of samples were uncalled in KASP assay analysis and this was within the expected values \[[@B10]\]. The number of uncalled samples for each LRRK2 SNP is shown in Supplementary Materials ([](#supplementary-material-1){ref-type="supplementary-material"} and [](#supplementary-material-1){ref-type="supplementary-material"} for comprehensive analysis). 3.2. SPNs with Negative Findings {#sec3.2} -------------------------------- The following most pathogenic LRRK2 mutations p.Gly2019Ser, p.Arg1441His, p.Tyr1699Cys, p.Ile2020Thr, and p.Asn1437His were not found in our cohort of PD patients and controls. All of the 246 PD subjects and 200 controls were homozygous for wild-type alleles of these SNPs ([Table 2](#tab2){ref-type="table"}). The allelic frequencies for the aforementioned SNPs were in Hardy--Weinberg equilibrium (*p*=1). 3.3. SNPs with Positive Findings {#sec3.3} -------------------------------- ### 3.3.1. p.Gly2385Arg (c.7153G\>A) {#sec3.3.1} Monoallelic p.Gly2385Arg variant (rs34778348) was found in three PD cases (1.2%, *n* = 3/239) and two controls (1%, *n* = 2/199) ([Table 3](#tab3){ref-type="table"}). The first positive case was a young Kazakh male with an autosomal dominant family history of PD. He developed PD at the age of 38 years, and the onset symptoms were depression, anxiety, hyposmia, and left-hand tremor. He had an affected mother, who had developed PD at the age of 55 years and deceased at the age of 58 years. Interestingly, his mother had three female siblings with upper-limb tremor but no signs of bradykinesia ([Figure 2](#fig2){ref-type="fig"}). He had fast progression in the disease course and in two years from the disease onset developed difficulty in rising up from a chair, freezing episodes, urinary frequency, and gait abnormalities. His off-stage MDS UPDRS motor scale score was 79 with Hoehn--Yahr stage 3. There was a good response to levodopa. WES in the proband did not reveal any other known PD genes. p.Gly2385Arg variant was tested in six of his healthy relatives and in those with tremor (*n* = 4) by Sanger sequencing. The variant was present in one of the two unaffected siblings of the proband, unaffected 20-year-old daughter of the proband, and in only one out of the four relatives with tremor ([Figure 2](#fig2){ref-type="fig"}). The second PD case was a 71-year-old Kazakh female with sporadic PD onset at the age of 66 years. She had a mild and slowly progressive disease course. Her off-stage MDS UPDRS motor score was 16 and Hoehn--Yahr stage 1. Dopamine agonists effectively controlled her motor symptoms. The third PD case was a 65-year-old Kazakh female with sporadic PD with the onset at the age of 61 years. She had also a mild disease course and was not on levodopa. Her off-stage MDS UPDRS motor score was 14 and Hoehn--Yahr stage 2. p.Gly2385Arg positive healthy controls were 50-year-old and 54-year-old male and female subjects. The allelic and genotypic frequencies for p.Gly2385Arg were in Hardy--Weinberg equilibrium (*p*=0.9) and did not statistically differ between cases and controls (OR 1.25, 95% C.I.: 0.2071--7.5688, *p*=0.8) (Tables [2](#tab2){ref-type="table"} and [4](#tab4){ref-type="table"}). ### 3.3.2. p.Ala419Val (c.1256C\>T) {#sec3.3.2} LRRK2 p.Ala419Val variant (rs34594498) was positive in 9 PD cases (3.7%, *n* = 9/242) and 5 controls (2.5%, *n* = 5/199), giving the OR of 1.5 (95% C.I.: 0.4941--4.5463, *p*=0.4) ([Table 4](#tab4){ref-type="table"}). WES in the probands with young-onset and familial PD did not reveal any other known PD genes. One positive case had homozygous substitution in c.1256C\>T (T/T). This was a 52-year-old Kazakh patient with the onset of sporadic PD at the age of 48 years. The patient expressed akinetic-rigid PD with a good response to levodopa but early and severe motor complications. He reached HY stage 3 in four years from the onset of motor symptoms. Two unaffected children of the proband were heterozygous for LRRK2 c.1256C\>T substitution, whereas one unaffected sibling of the proband did not have LRRK2 c.1256C\>T substitution on Sanger sequencing segregation analysis ([Figure 3](#fig3){ref-type="fig"}) (see [](#supplementary-material-1){ref-type="supplementary-material"} in the supplementary material for comprehensive image analysis). Five out of the 9 LRRK2 p.Ala419Val carriers developed PD before the age of 50 years, the youngest manifestation being at the age of 26 years. The mean age at onset for the p.Ala419Val carriers was 48.3 ± 12.6 (range 26--69) and this did not significantly differ from the noncarriers (48.3 ± 12.6 v54.7 ± 12.3, *p*=0.19). The mean age at examination was 57.4 ± 12.9 (range 32--82), and mean disease duration was 9.1 ± 6.1 (range 2--20) ([Table 5](#tab5){ref-type="table"}). The mean HYS score for the positive cases was 2.5 ± 0.5. Interestingly, self-reported nationalities in three out of the nine positive cases were Russian, one case was half Kazakh and half Russian, and the remaining 5 cases were Kazakhs. Two of the Russian patients had an autosomal dominant history of PD with onset after the age of 50 years. All of the LRRK2 p.Ala419Val-positive cases had a good response to levodopa. The allele frequencies for LRRK2 p.Ala419Val in PD cases deviated from Hardy--Weinberg equilibrium (*p*=0.004) ([Table 2](#tab2){ref-type="table"}). ### 3.3.3. p.Arg1628Pro (c.4883G\>C) {#sec3.3.3} PD cases were negative for p.Arg1628Pro (rs33949390) variant, whereas two control subjects were found to be positive (1%, *n* = 2/199). The variant was in Hardy--Weinberg equilibrium (*p*=0.9) in controls. 4. Discussion {#sec4} ============= This study screened cohorts of PD patients and controls from Kazakhstan for five LRRK2 mutations and three Asian disease-associated variants. To date, over 80 LRRK2 disease-causing and disease-associated variants have been described in literature since the gene was discovered in 2004. However, only eight of them have been acknowledged as PD-causing mutations including p.Gly2019Ser, p.Arg1441His, p. Arg1441Cys, p.Arg1441Gly, p.Tyr1699Cys, p.Ile2020Thr, p. Asn1437His, and p. Ile2012Thr. All of these mutations affect the catalytic core of the LRRK2 enzyme \[[@B11]\]. Among these mutations, p.Gly2019Ser is the most common followed by the substitution of arginine (Arg) by glycine (Gly), cysteine (Cys), or histidine (His) in the position 1441 of LRRK2 gene \[[@B11]\]. p.Gly2019Ser mutation has a global prevalence of 1% in patients with iPD and around 4% in familial PD. It is noteworthy that p.Gly2019Ser has been predominantly reported in the North African population where it is responsible for 30--42% of familial and 30--34% of sporadic PD cases. Its prevalence shows high figures in Ashkenazi Jews (28% of familial PD and 10% in iPD) and among European as well as North American populations (6% and 3%, respectively) \[[@B12]\]. Conversely, p.Gly2019Ser mutation has not been reported in Asians (\<0.1%) \[[@B13], [@B14]\]. p.Gly2019Ser was reported in Russian PD cohorts (%), possibly in subjects of Ashkenazi Jewish origin \[[@B15]\]. The currently known most deleterious LRRK2 mutations associated with PD were not found in our cohort. The absence of these mutations in our cohort is similar to other Asian studies \[[@B6]\] This is probably due to the founder effects of these mutations, which seem to be specific to Western populations. 4.1. p.Gly2385Arg {#sec4.1} ----------------- LRRK2 p.Gly2385Arg substitution is located in the WD-40 domain, a toroidal beta-propeller structure responsible for protein-protein interactions \[[@B16]\]. There have been 5 independent case-control studies reporting this variant \[[@B6], [@B13], [@B16]--[@B18]\] and 14 studies screening their cohorts for p.Gly2385Arg \[[@B14], [@B19]\]. Originally, p.Gly2385Arg variant was identified in a small Taiwanese PD family, in a proband and his affected father \[[@B20]\]. Later, a number of studies in Chinese, Taiwanese, Korean, and Japanese populations have reported p.Gly2385Arg variant significantly more among PD patients than controls, with minor allele frequency (MAF) up to 0.4 in patients ([Table 6](#tab6){ref-type="table"}). Thus, p.Gly2385Arg variant was attributed to a risk factor for sporadic and familial PD. The population attributable risk for p.Gly2385Arg in Han-Chinese ethnicity was estimated to be around 4% \[[@B18]\], and the variant probably originated around 4800 years ago from one ancestor \[[@B16]\]. p.Gly2385Arg variant had a tendency for equal distribution across genders and age groups \[[@B6]\]. Regarding the clinical presentation, p.Gly2385Arg carriers expressed typical PD and homozygous cases were not clinically different from heterozygotes and non-carriers \[[@B13]\]. This variant seems not to influence the age of PD onset, as the mean ages of onset in p.Gly2385Arg carriers and non-carriers have not been consistently reported to significantly differ between these groups \[[@B6], [@B13], [@B17]\]. Reports on the association of p.Gly2385Arg carrier status with a family history of PD have also been inconsistent \[[@B16], [@B17]\]. In regards to non-Chinese populations, p.Gly2385Arg was found in 1.2% (2/166) of PD patients and 0.6% (2/306) of controls in Malay/Indian ethnicity from Singapore \[[@B8]\] The frequency of p.Gly2385Arg in cases and controls was not significantly different in this Singaporean study, and considerably lower than the frequency of 8--10% reported in Chinese PD subjects \[[@B6], [@B8]\]. A recent study on Malaysian PD subjects has found a significant association between p.Gly2385Arg and increased risk of PD \[[@B14]\] The variant was absent in 405 Iranian subjects \[[@B22]\] and positive in only one individual of Northern European origin among 14,002 screened Caucasian subjects \[[@B25]\]. Evidence from functional studies suggests that p.Gly2385Arg substitution leads to the replacement of hydrophobic glycine with the hydrophilic arginine and increases the net positive charge on the 40WD domain of LRRK2. Both LRRK2 proteins with wild-type and p.Gly2385Arg variant localized to the cytoplasm forming aggregates, but the intensity of apoptosis is higher in p.Gly2385Arg variant under oxidative stress conditions \[[@B18]\]. The frequency of p.Gly2385Arg variant in our study (1.2% patients and 1% controls) was almost similar to non-Chinese Singaporean subjects. If MAF for Gly2385Arg in Chinese and Japanese PD populations was between 0.05 and 0.4 ([Table 6](#tab6){ref-type="table"}), MAF in our study was 0.007. This might suggest that p.Gly2385Arg could be found in non-Chinese Asians but in considerably less proportion. Due to a small amount of non-Chinese subjects screened for p.Gly2385Arg and its equal distribution between patients and healthy controls, currently, it is difficult to ascertain the role of this LRRK2 variant in the risk of PD among Central Asian populations. Contemporary data suggest that p.Gly2385Arg could be a risk factor for PD only in selected Asian races. On the other hand, among three patients positive to p.Gly2385Arg in our study, we had an interesting familial PD case, where proband and his deceased mother had PD, whereas maternal siblings and one maternal cousin of the proband had unilateral asymmetric UL tremor. Although all other known PD genes have been excluded by WES in the proband, p.Gly2385Arg did not completely segregate in the family, being positive in some unaffected family members and negative in some affected ([Figure 2](#fig2){ref-type="fig"}). To date, the variant has been shown to segregate with PD in only one small Taiwanese family with affected proband, affected father of the proband, and unaffected sibling. We showed the segregation of p.Gly2385Arg, although incomplete, in a larger family presenting not only with PD but tremor. The incomplete segregation in our family could be due to reduced penetrance or other unknown genetic factors. 4.2. p.Ala419Val {#sec4.2} ---------------- There have been nine reports describing p.Ala419Val variant \[[@B26]\]. The variant resides in the LRRK2 Armadillo domain and is predicted deleterious by online prediction tools with high conservation in the vertebrates \[[@B26]\]. Initially, the association between this variant and PD was described by Ross et al. \[[@B23]\] where p.Ala419Val was tested in 2,338 Asian subjects from Japan, Korea, and Taiwan (1,376 PD cases and 962) in a large-scale multicenter study. The study results revealed the OR of 2.27 (95% CI: 1.35--3.83, *p*=0.0011). Several studies before and after the reported association of p.Ala419Val with PD have found either no carriers of this variant in large cohorts of PD patients and controls or insignificant OR ([Table 7](#tab7){ref-type="table"}), thus considering the variant as putatively nonpathogenic population-specific SNP. The MAF for p.Ala419Val in PD patients has been 0.002--0.018 in Chinese, and 0.026--0.029 in Japanese and Korean populations in studies reporting positive association \[[@B23], [@B31]\]. Interestingly, the replication studies in the Chinese and Taiwanese ethnicities have yielded inconsistent results. While p.Ala419Val was negative or positive with insignificant OR in some studies, several studies and their meta-analysis reported a significant association between this variant and predominantly early-onset PD ([Table 7](#tab7){ref-type="table"}, \[[@B26]\]). This has been explained by possible natural sampling variation and population heterogeneity \[[@B30]\] on the one hand. On the other hand, Li et al. \[[@B26]\] argued that the discrepancy is likely to result from different mean ages at onset (AAO) of PD patients in these studies. Thus, while the mean AAO in p.Ala419Val-positive reports on Chinese ethnicity was \<55 years \[[@B23], [@B26], [@B29]\], negative reports on the same population had AAO above 60 years \[[@B28], [@B30], [@B31], [@B33]\]. Provided the fact that p.Ala419Val might have a strong association with early-onset Chinese PD, the likelihood of yielding positive findings could be higher in young-onset PD cohorts. The MAF for p.Ala419Val in PD patients in our study was 0.02, which is higher than in Chinese and Taiwanese populations and closer to Japanese and Korean ([Table 7](#tab7){ref-type="table"}). Taking into account the young mean AAO in our Kazakhstani PD cohort (55.06 ± 11.15) and reference to Li et al. \[[@B26]\], one might explain the high MAF for p.Ala419Val. Moreover, the AAO in p.Ala419Val-positive PD patients was remarkably young (48.3 ± 12.6) in our study. However, the high frequency (MAF 0.012) of p.Ala419Val in healthy Kazakhstani controls, who also had a mean age of below 55 years, and insignificant OR do not allow us to ascertain the pathogenicity of p.Ala419Val in Kazakhstani PD. We have found a previously unreported homozygous carrier of p.Ala419Val variant with PD onset before 50 years and relatively aggressive disease course. If this LRRK2 variant is nonpathogenic and not rare in our population, the likelihood of p.Ala419Val homozygous carriers would increase and this might result in the HWE deviation. p.Ala419Val variant seems to be present not only in Kazakhs but also in self-reported Russian patients with late-onset or familial PD. Considering the reported specificity of p.Ala419Val to Asian populations, we could speculate that these self-reported Russian subjects in our study could have a mixed ethnic background, particularly with Tatars, an Asian population with some Russian phenotypic features. Alternatively, the variant could also be present in the Russian population. 4.3. p.Arg1628Pro {#sec4.3} ----------------- Regarding p.Arg1628Pro, data from a meta-analysis, including 19 studies with a total of 9,927 PD patients and 8,602 controls, suggest that the variant is significantly associated with the risk of PD in East Asian populations \[[@B34]\]. We failed to find this variant in our PD patients but it was present in controls. This might suggest that p.Arg1628Pro could be a common benign polymorphism in Kazakhstani population. We have to acknowledge the limitations in our study due to the relatively small sample size, the nonhomogeneous ethnic composition of the PD cohort, as only 72.8% of the PD cohort were Kazakhs. In addition, controls were not perfectly matched to cases by age, gender, and ethnicity. 5. Conclusions {#sec5} ============== The negative findings on common LRRK2 PD causing mutations, and the presence of LRRK2 Asian-specific variants in our PD patients, although at insignificant level compared with controls, suggest that further large-cohort genetic studies are required in Central Asia to ascertain the pathogenicity of LRRK2 Asian-specific variants in the Central Asian PD population. Cases were collected as part of the SYNaPS Study Group collaboration funded by The Wellcome Trust and Strategic Award (Synaptopathies) Funding (WT093205MA and WT104033AIA). This research was conducted as part of the Queen Square Genomics group at University College London, supported by the National Institute for Health Research University College London Hospitals Biomedical Research Centre. This research was funded by the Medical Research Council (MRC) (MR/S01165X/1, MR/S005021/1, and G0601943). Data Availability ================= VCF files from exome sequencing used to support the findings of this study are available from the corresponding author upon request. Conflicts of Interest ===================== The authors have no conflicts of interest. Supplementary Materials {#supplementary-material-1} ======================= ###### S1: the 8 LRRK2 SNPs and surrounding 50 base-pairs annotated in Ensembl genome browser S2: calculations of allelic and genotypic frequencies, odds ratios, and Hardy--Weinberg equilibrium S3: chromatogram for Ala419Val homozygous variant with family segregation S4: the results of KASP analysis. ###### Click here for additional data file. {#fig1} {#fig2} {#fig3} ###### Clinical and demographic characteristics of the cohort. ----------------------------------------------------------------------------------------------------------------------------------------------   Cases Controls ------------------------------------------------- --------------------------------------------------- ---------------------------------------- Number 246 200 Ethnic groups, abs number (%) Kazakhs 179 (72.8%)   Russians 51 (20.7%) Uygurs 7 (2.8%) Tatars 5 (2%) Koreans 3 (1.3%) Tajiks 1 (0.4%) Sex distribution Males---120, females---126\ Males---62, females---138\ M : F ratio−0.95 : 1 M : F ratio 0.4 : 1 Age at examination (mean) **61.7** ± 10.3 (range 28--83)\ Mean age---**54.93** ± 4.8 (47--66)\ For males---60.3 ± 10.6 (range 28--82), *p*=0.03\ For males---**55.27** ± 4.8 (47--66)\ For females---63.1 ± 9.9 (range---32--83) For females---**54.78** ± 4.7 (47--65) Age of onset (mean) **55.06** ± 11.15 (range 14--77)\   For males---53.3 ± 11.9 (range 14--76), *p*=0.01\ For females---56.8 ± 9.9 (range 26--77) Disease duration (mean) **13.2** ± 9.3 (range 1--24)   HY stage off (mean) **2.4** **±** 0.6 (range 1--5)\   For males---2.3 ± 0.7 *p*=0.6\ For females---2.4 ± 0.6 Family history of PD and tremor, abs number (%) **21 (8.5%)**   Young-onset cases, abs number (%) **31 (12.6%)** before the age of 40\   **65 (26.4%)** before the age of 50 ---------------------------------------------------------------------------------------------------------------------------------------------- M : F--male to female. ###### Allele frequency and distribution of the 8 tested LRRK2 variants. SNP Hardy--Weinberg equilibrium *p*-value Number of samples Allele *n* ^a^ Frequency Genotype *n* ^b^ Frequency --------------------------- --------------------------------------- ------------------- -------- --------- ----------- ---------- --------- ----------- p.Gly2019Ser (c.6055G\>A) 1 198 controls       GG 198 1.00 G 396 1.00 GA 0 0.00 A 0 0.00 AA 0 0.00 1 241 PD   482 1.00 GG 241 1.00 G 0 0.00 GA 0 0.00 A     AA 0 0.00 p.Arg1441His (c.4322G\>A) 1 199 controls       GG 199 1.00 G 398 1.00 GA 0 0.00 A 0 0.00 AA 0 0.00 1 240 PD   480 1.00 GG 240 1.00 G 0 0.00 GA 0 0.00 A     AA 0 0.00 p.Tyr1699Cys (c.5096A\>G) 1 196 controls       GG 196 1.00 G 392 1.00 GA 0 0.00 A 0 0.00 AA 0 0.00 1 239 PD   478 1.00 GG 239 1.00 G 0 0.00 GA 0 0.00 A     AA 0 0.00 p.Ile2020Thr (c.6059T\>C) 1 198 controls       TT 198 1.00 T 396 1.00 TC 0 0.00 C 0 0.00 CC 0 0.00 1 242 PD   484 1.00 TT 242 1.00 T 0 0.00 TC 0 0.00 C     CC 0 0.00 p.Asn1437His (c.4309A\>C) 1 199 controls       AA 198 1.00 A 398 1.00 AC 0 0.00 C 0 0.00 CC 0 0.00 1 240 PD   480 1.00 AA 240 1.00 A 0 0.00 AC 0 0.00 C     CC 0 0.00 p.Gly2385Arg (c.7153G\>A) 0.94 199 controls       GG 197 0.99 G 396 0.995 GA 2 0.01 A 2 0.005 AA 0 0.00 0.92 239 PD       GG 236 0.99 G 475 0.993 GA 3 0.01 A 3 0.007 AA 0 0.00 p.Ala419Val (c.1256C\>T) 0.85 199 controls       CC 194 0.98 C 393 0.988 CT 5 0.02 T 5 0.012 TT 0 0.00 0.004 242 PD     0.98 CC 233 0.97 C 474 0.02 CT 8 0.04 T 10   TT 1 0.00 p.Arg1628Pro (c.4883G\>C) 0.94 199 controls       GG 197 0.99 G 396 0.995 GT 2 0.01 C 2 0.005 TT 0 0.00 1 236 PD       GG 236 1.00   G 472 1.00 GT 0 0.00   C 0 0.00 TT 0 0.00 *n*1---number of alleles, *n*2---number of genotypes, PD--Parkinson\'s disease, HWE--Hardy--Weinberg equilibrium, and NA--not applicable. ###### p.Gly2385Arg-positive PD patients and their characteristics.   Number (246--7 uncalled = 239) Mean age at examination Mean age of onset Mean disease duration Family history Mean HY stage off ------------- -------------------------------- ------------------------- ------------------- ----------------------- ---------------- ------------------- Carriers 3 58.6 ± 13.4 55 ± 12.1 4 ± 0.8 1 2 ± 0.8 Noncarriers 236 61.7 ± 10.3 55 ± 11.1 6.8 ± 4.7 20 2.3 ± 1.5 *p* value   **0.77** **0.99** **0.01**   **0.56** ###### The allelic frequency and odds ratios for the positive LRRK2 Asian disease-associated variants. SNP Nucleotide change Amino acid change MAF Or (95% CI) *p* value ------------ ------------------- ------------------- ----------- ------------- ------------------------ ----- rs34778348 c.7153G\>A p.Gly2385Arg^a^ 0.007 (A) 0.005 (A) 1.25 (0.2071--7.5688) 0.8 rs34594498 c.1256C\>T p.Ala419Val^b^ 0.02 (T) 0.012 (T) 1.5 (0.4941 -- 4.5463) 0.4 rs33949390 c.4883G\>C p.Arg1628Pro^c^ 0.0 (C) 0.005 (C) NA NA MAF--minor allele frequency, NA--not applicable, OR--odds ratio, and PD--Parkinson\'s disease. ###### p.Ala419Val-positive PD patients and their characteristics.   Number 246-4uncalled = 242 Mean age at examination Mean age of onset Mean disease duration Family history Mean HY stage off ------------- ---------------------------- ------------------------- ------------------- ----------------------- ---------------- ------------------- Carriers 9 57.4 ± 12.9 48.3 ± 12.6 9.1 ± 6.1 2 2.5 ± 0.5 Noncarriers 233 61.6 ± 10.4 54.7 ± 12.3 6.8 ± 7 19 2.2 ± 0.7 *P* value   **0.38** **0.19** **0.3**   **0.02** HY--Hoehn--Yahr. ###### Studies investigating LRRK2 p.Gly2385Arg variant in Asian populations. --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------   Cases Controls Ethnicity OR MAF for PD patients ---------------------------------- ------------------------------------- ----------------- --------------------------------------- -------------------------------------------------------------------------------- ------------------------------------------------------------ Funayama et al., 2007 \[[@B13]\] 448/52 (11.6%) (2 homozygous cases) 457/22 (4.8%) Japanese OR for the frequency of A allele 2.63, 95% CI: 1.56--4.35, *p* = 1.24 × 10^−4^ 0.06 Di Fonzo at al., 2006 \[[@B6]\] 608/61 (10%) 373/18 (4.8%) Han Chinese from Taiwan OR = 2.24, 95% CI: 1.29--3.88, *p* = 0.004 0.05 Fung et al., 2006 \[[@B17]\] 305/27 (9%) 176/1 (0.5%) Han Chinese from Taiwan 16.99, 95% CI: 2.29 to 126.21, *p*=0.0002 0.4 No positive cases with FH An et al., 2008 \[[@B21]\] 600/71 (1 homozygous) (11.9%) 334/11 (3.3%) Han Chinese OR 3.9, 95% CI = 2.1--7.5, *p* \< 0.01 0.06 Tan et al., 2007 \[[@B18]\] 494/37 (7.27%) (1 homozygous) 495/18 (3.64%) Ethnic Chinese OR 2.1, 95% CI: 1.1--3.9, *p*=0.014 PAR of 4% for the Gly2385Arg heterozygous genotype. Tan et al., 2007 \[[@B8]\] 166/2 (Malays) 306/2 (Malays) Malay 98/173, Indian ethnicity 66/133 OR 2.83, 95% CI 0.40, 20.2, *p*\_0.3 0.003 for Malays Farrer et al., 2007 \[[@B16]\] 410/34 335/13 (3.9%) Ethnic Chinese OR 2.24 95% CI 1.16--4.32, *p* \< 0.014 MAF 0.08\ 23.1% (*n* = 6/26) of patients with familial parkinsonism. Ross et al., 2011 \[[@B23]\] 1,376 962 Japan, Korea, Taiwan OR: 1.73, 95% CI: 1.20--2.49, *p*=0.0026 MAF 3.3% Japan\ Control: 75 PD: 173 Korea\ *Control: 587* \|*PD: 844* *Taiwan*\ *Control: 300* *PD: 369* Mata et al., 2005 \[[@B20]\] 100 probands with PD FH/2 cases   Taiwan 1 family with 2 members   Zabetian et al., 2009 \[[@B24]\] 601/69 (11.5%) 1628/101 (6.2%) Japanese OR, 1.83; 95% CI: 1.31--2.54; *p* = 3.3 × 10^−4^   Gapalai et al., 2015 \[[@B14]\] 695 507 Malaysian OR 2.22 (*p*=0.019) MAF = 0.026 -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- OR--odds ratio, MAF--minor allele frequency, PAR--population attributable risk, and FH--family history. ###### Studies investigating LRRK2 p.Ala419Val variant in Asian populations. ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------   Cases Controls Ethnicity OR MAF for patients/controls ----------------------------------- ---------------- -------------- -------------------------------------------------------- ------------------------------------------ --------------------------- Di Fonzo et al., 2006 \[[@B6]\] 582/10 341/3 Han Chinese from Taiwan 1.95 (0.53--7.15), *p* \> 0.05 0.008/0.004 Nuytemans et al., 2009 \[[@B27]\] 620/1 540/0 Belgian *p* \> 0.05   Jasinska-Myga, 2010 \[[@B12]\] 165/0 364/0 Arab-Berber ethnicity n/a   Tan et al., 2010 \[[@B28]\] 250/0 250/0 Han Chinese n/a   Ross et al., 2011 \[[@B23]\] 1,376 962 Japan, Korea, Taiwan OR: 2.27, 95% CI: 1.35--3.83, *p*=0.0011     Japan PD: 173 Control: 75   Japan: OR 1.26 (0.38 to 4.22)     Korea PD: 844 Control: 587   Korea: OR 2.21 (1.2 to 4.06)     Taiwan PD: 369 Control: 300   Taiwan: OR 7.51 (0.95 to 59.6)   Li et al., 2012 \[[@B29]\] 729/22 585/4 Han Chinese OR, 4.14; 95% CI: 1.53--12.74 0.015 Wu et al., 2012 \[[@B30]\] 1517/13 1487/13 Han Chinese from China and Singapore Taiwanese 0.98 (0.45 to 2.18) 0.004 Gopalai et al., 2013 \[[@B31]\] 404/1 424/3 Chinese (223/236), Malay (122/110), and Indian (59/78) 0.35, 95% CI: 0.01 to 3.79; *p*=0.624 0.002 cases\ 0.004 Controls Heckman et al., 2013 \[[@B32]\] 369/10 300/1 Taiwan, South Korea, Japan OR 8.33 (1.06--65.43) Taiwanese 0.013\ Japanese -- 0.026\ South Korea 0.029 Wu-Chou et al., 2013 \[[@B33]\] 626/0 *473/0* *Han Chinese from Taiwan* n/a   Li et al., 2015 \[[@B26]\] 500/18 574/9 *Chinese* OR 2.57, 95% CI: 1.13--5.86, *p*=0.025 0.018 --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- OR--odds ratio and MAF--minor allele frequency. [^1]: Academic Editor: Carlo Ferrarese

Background {#Sec1} ========== Medical school is an emotional experience for students. The influence of emotion on cognition is well recognized. Emotions affect learning of complex skills and knowledge and transferring information into new scenarios \[[@CR1]\]. From the psychological and cognitive view it is believed that negative emotions narrow individuals' momentary thought-action repertoires by calling forth specific action tendencies (e.g., attack, flee), whereas many positive emotions broaden individuals' momentary thought-action repertoires, prompting them to pursue a wider range of thoughts and actions \[[@CR2], [@CR3]\]. In neurology, interviewing and examining the patient are crucial skills in both diagnosing and excluding a disease, as well as in the follow-up of disease progression. There is a need to develop effective educational experience in neurology as it has been reported that students experience neurology as a difficult topic, and some ascribe this to insufficient knowledge and poor teaching \[[@CR4]\]. These experiences may have effect on number of medical students who continue to pursue careers in clinical neurosciences. Other factors that lead students to pursue or avoid careers in neurology have been related to emotional experiences during training \[[@CR5]--[@CR8]\]. Jozefowicz first represented a term 'neurophobia' describing it as a phenomenon where medical students are unable to apply knowledge of basic neuroscience to a clinical situation \[[@CR5]\]. The 'symptoms' include intimidation, boredom and in some individuals, a cynical and nihilistic attitude towards neurological diseases in later career. A major 'sign' is inability to localise focal lesions in the nervous system. Although the term 'neurophobia' is informal, results of any such suggestion may in worst scenarios lead to real clinical consequences, and phenomena thus needs recognition, as many common neurological diseases among ageing populations worldwide are increasingly treated by primary health care physician \[[@CR6], [@CR9]\]. Previous studies suggested that integrating basic science and clinical neurology into medical school training in the form of group work and case-based exercises could reduce stress related to neurological studies \[[@CR10]\]. However, solid qualitative evidence showing the effect of such interventions is lacking \[[@CR11]\] nor have the causes for negative anticipation in neurological studies, which may differ from other issues in academic performance, such as procrastination, been extensively studied \[[@CR12]\]. The amount of negative anticipation towards neuroscience that exists in Finland and factors that could reduce the incidence are not known. Evaluation is valuable as although being a common phenomenon \[[@CR9]\], the degree of these attitudes and emotions may not be consistent between different countries or regions \[[@CR7]\]. It is commonly acknowledged that the development of clinical reasoning skills is the most important goal of medical studies \[[@CR9]\]. Clinical reasoning requires problem solving skills, which can be acquired by structured teaching \[[@CR13], [@CR14]\]. A good clinical teacher or an enthusiastic senior colleague may powerfully nurture learning by seeing the student's point of view and applying pedagogic theories to teaching \[[@CR14]\]. Teaching medical students how to perform a neurological examination is a challenge. It is not clear whether students are able to adopt hypothesis-based level of neurological examination, where clinical hypotheses that arises from the interview of a patient, steers the examination \[[@CR15], [@CR16]\]. Therefore, with regard to performing a neurological examination, some authors have concluded that it may be better to teach them more categorical screening type skills rather than how to formulate hypotheses \[[@CR15], [@CR17]\]. In Finland, all five universities use national learning objectives for medical students, although the curricula differ slightly between them. For example, in Oulu University, most neuroanatomy modules are incorporated into pre-clinical studies, and clinical neurology and clinical neuroanatomy are taught during the 4th year of study. In contrast, in Tampere University, neuroanatomy is integrated into clinical neurology during the 2nd to 3rd and 5th to 6th years of study. Approximately 135 students in Oulu University and 100 students in Tampere University attend these clinical courses each year. Clinical teaching is structured and takes place within small groups of 6--12 students, with the focus on clinical neuroanatomical knowledge. In practice, during a 4-week contact-teaching period, students examine patients with various neurological-related complaints, both in outpatient clinic and hospital ward settings. The training takes place within small groups, as well as individually, under the guidance of a clinical teacher. During this period, a thorough neurological examination is introduced but the focus is on the screening type of neurological examination, with some level of hypothesis-based examination in different clinical situations. The students receive constant, direct and individual feedback from teacher regarding their progress and performance. Students give both spontaneous oral feedback after each sessions as well as compulsory written feedback using a structured form and free word after the clinical course. Clinical teachers are mainly responsible for teaching students how to conduct a neurological examination, although senior staff in the hospitals as well as health care centres are involved in some educational tasks besides their clinical work. Some Finnish universities require that all clinical teachers have a pedagogic education background, whereas others recommend it. At present, practically all clinical teachers of neurology in Finland have studied university pedagogy, at least to some extent. This study was completed in two medical schools in Finland, Oulu and Tampere, during 2015. The aim of this study was to explore the types of negative emotions Finnish medical students have towards neurology and how to reduce these. Methods {#Sec2} ======= Phase 1: Online survey and content analysis {#Sec3} ------------------------------------------- One hundred and thirty-five students (Oulu: *n* = 95; Tampere: *n* = 40), were invited to take part in an anonymous survey, using an online survey tool (Webropol) after completing their clinical neurology course. No ethical approval was required by the Regional Ethics Committee of Northern Ostrobothnia Hospital District. A written consent was obtained from the study subjects before entering the study. The purpose of the first phase online survey was to study attitudes and experiences widely. The open-ended questions in the questionnaire for the assessment of experiences and attitudes in neurological studies were:'What were your experiences of your neurological studies compared to those of other specialties?' 'Please elaborate on the reason for these'.'Please describe your experience of performing a neurological examination on a patient. What was a) the most difficult aspect and b) the most interesting aspect?''Please state how capable you feel about a) conducting a neurological examination of a patient and b) interpreting the neurological findings of the examination'.'Has your attitude towards neurological patient examinations changed since you completed the clinical course and, if so, how?' Content analysis {#Sec4} ---------------- Open questions in the first phase of the study explored students' experiences. Both positive and negative experiences were screened. Descriptions of emotions were picked out and classified as negative in case of e.g. anger, anxiety, fear, and positive if they conveyed optimism, contentment, and happiness. In categorization we followed the widely used terms in Emotion Report Forms \[[@CR18]\]. We focused on negative experiences and in the narrative phase of the study explored this further among the students that expressed them. We used content analysis to explore the emotions and to examine trends and patterns in attitudes \[[@CR19], [@CR20]\]. The change of the nature of emotions before and after the clinical neurological course was evaluated. Emergent coding of categories expressed negative/positive emotion took place after a preliminary examination of the data. To control for reliability and validity, an intra- and inter-rater assessment was done. First, H.A. and M.S. independently reviewed the data and searched for descriptions related to emotions. The results were then compared, and differences were reconciled by consensus. After reconciliation of the data, a consolidated checklist was drawn up, and H.A. and M.S. applied the coding. The reliability of the coding was checked by comparing the results, which revealed a good level of reproducibility. To determine the intra-rate stability, a second round of coding took place 6 months later. The results showed good reliability. Phase 2: Narrative interview {#Sec5} ---------------------------- In the narrative phase of the study, an invitation to attend a personal interview was sent to the initial online survey responders (*n* = 58). Of these, 11 (21%) students were willing to relinquish their anonymity and participate in face-to-face interviews. Narrative method {#Sec6} ---------------- The narrative analysis is commonly used in social sciences \[[@CR21]--[@CR23]\]. The basic premise of narrative inquiry is that people make sense of themselves and their world by telling stories \[[@CR21]\]. The narrative interviews with the medical students lasted 20--30 min and were conducted as described previously \[[@CR22], [@CR23]\]. Confidentiality was assured, and a relationship between the interviewer and interviewees was established. The interview consisted of asking the students to tell stories related to their experiences of the clinical neurology course, using open-ended prompts. For this study, we systematically selected two cases, Tomi and Petri, for examination. We used a critical case strategy by selecting medical students who would contribute the most towards understanding students' experience of emotions and success in neurological studies. The cases selected made a point clearly, were particularly information rich \[[@CR24]\] and expressed themselves vividly \[[@CR25]\]. The literature presents various approaches to conducting narrative analysis \[[@CR23], [@CR26], [@CR27]\]. The narrative inquiry that we applied here involved emplotment \[[@CR26], [@CR27]\]. Like Ricoeur \[[@CR28]\], we take the view that the plot brings together goals, causes and chance within the temporal unity of a whole action. In particular, when emplotting Tomi's and Petri's narratives our goal was to explicate how their experiences in participating the clinical neurology course influenced their emotions towards neurology. The emplotment began by specifying the outcomes in Tomi's and Petri's narratives. The main outcome was considered a positive change in students´ emotions towards neurology after the course. Then, with reference to the data, the interviewee was asked about how the change happened, and we began seeking clues from the interviews that seemed to 'explain' the change and to reduce negative emotions towards neurology. When constructing the final version of the students' case descriptions, we arranged the data elements chronologically. To give a voice to the students we utilized many direct quotes from their talk. We also analysed the way the students talked, especially central expressions that they used when they talked about their experiences during the clinical neurology course, because it helped us to understand their purposes and actions. At the end of both case descriptions, we present a short summary of the cases where we explicate how the process of change happened, and what factors seemed to facilitate the change. We also connect the students' narratives to the broader theoretical framework that was used to interpret the narrative. Results {#Sec7} ======= Internet survey {#Sec8} --------------- Of 135 students, 58 (35 females and 23 males, 43%) responded to the initial online survey. Of these, 43 (74%) were from Oulu University, and 15 (26%) were from Tampere University. All the 58 questionnaires were completed correctly, and all the data were therefore usable. Content analysis {#Sec9} ---------------- At the beginning of the clinical neurology course, 20 (34%) of the responders (14 \[33%\] from Oulu University and 6 \[40%\] from Tampere University; 12 females and 8 males) conveyed negative emotions. The analysis of the data revealed that emotions could be categorised to: insecurity about personal performance (*n* = 19; 95%), anxiety (*n* = 9; 45%) and fear (*n* = 6; 30%). In half of the cases (*N* = 10), both sexes reported more than one emotion. All three emotions (insecurity, anxiety and fear) were more common among males (2/8; 25%) than females (2/12; 17%). The distribution of emotions experienced before the course is illustrated in Fig. [1](#Fig1){ref-type="fig"}.Fig. 1Distribution of reported emotions before the clinical neurology course, according to the student's gender During the course the combined negative emotions (insecurity, anxiety, fear) decreased in the majority of students (16/20, 80%), remained unchanged in some students (3/20, 15%) and could not be evaluated in one case (5%). Insecurity observed in 19 cases decreased significantly (18/19, 95%), but remained unchanged in one case (5%). Anxiety and fear decreased in most students (6/9, 67% anxiety; 4/6, 67% fear) and remained unchanged in one third of cases (3/9, 33% anxiety; 2/6, 33% fear). The histogram in Fig. [2](#Fig2){ref-type="fig"} illustrates the number of reports and the trends in change.Fig. 2Trends in the change of emotions among medical students before and after the course in clinical neurological studies Students' open ended answers unanimously pointed out that improvements in examination skills mainly explained the decrease in negative emotions. The students considered structured neurological examination instruction and practice beneficial. Most of the students viewed developing an understanding of how to interpret the neurological findings as the main reason for the decrease in their negative emotions. Narratives {#Sec10} ---------- Total of 11 (21%) of the 58 online survey responders, were willing to relinquish their anonymity and participate in 20--30 min face-to-face interviews. Below, we present the experiences of Tomi and Petri of their clinical neurology studies and the emotions they experienced. **Tomi's case:** Before the clinical neurological course, which took place during the 5th year, Tomi said: *'I was anxious about having to deal with difficult neurological diseases, such as Parkinson's disease and stroke. During the study break, I worked on a primary healthcare ward and met a patient with end-state Parkinson's disease who had swallowing difficulties. I found it difficult to make medical decisions. I was not ready to take responsibility for such a severe case'.* Above Tomi described well his anxious feelings towards neurology. It seems that his challenging experiences in working on a primary healthcare ward even increased his anxiety. The reason for his anxiety were '*difficult neurological diseases, such as Parkinson's disease and stroke'.* Tomi assumed that both acute and chronic neurological diseases were severe. This gave rise to anxiety about his ability to manage neurology patients. He concluded: *'My skills are not at an adequate level to treat difficult neurological cases'.* Tomi further noted:"*'Before starting the clinical course, we had all taken part in tutored practice on neurological examination, and I had practiced it also at work. I was able to do "the tricks", but I didn't understand their clinical significance'.*" Here Tomi's expression '*I was able to do the tricks'* shows well that his learning had been on a superficial level, and deeper learning with understanding was missing."*'My motivation to learn was good, due to primary experience. During the course, I recognised the learning objectives better. Even if I did not have the chance to study all the patients myself, it was helpful to observe the examination of patients with different sorts of neurological complaints'.*" He was aware that he had forgotten what he had learned earlier about anatomical nervous system structures and basic clinical neurology. Tomi also noted:"*'Pieces finally clicked into their place in the neurological examination'.*" During the clinical studies, Tomi felt comfortable examining a patient when the teacher was present, as the teacher guided him through the structured neurological examination, as illustrated below:"*'It was important to see how an examination should be done and to observe abnormal findings, such as a clonic reflex'.*" Although Tomi felt that it was important to be able to examine several patients, he also felt that the quality of the teaching was more important than the quantity of the patients observed. Furthermore, he remarked that a friendly and open-minded atmosphere made it easier to ask questions and discuss the patient cases freely. During the rounds at the university hospital, he met tertiary care patients with rare and severe neurological conditions. He met these patients without having a comprehensive knowledge on their diseases, although he had read their case records. He recounted the following:"*'Things proceeded too fast and the information was way too complex! For example, during the clinical rounds, one specialist immediately engaged in a complex discussion of the clinical problem, including a huge amount of detailed information. In that situation, his questions were difficult to comprehend, and I was unable to come up with answers. As a result, I felt stupid'.*" Tomi further elaborated on the problems experienced during the clinical rounds, as follows: *'The specialist did not seem to understand our level of knowledge or remember what it had felt like to be a student. It would have been helpful if the senior doctor had clarified his decision-making process. The decision-making seemed to be based on intuition, and he did not explain the process that led to the decision. As a result, I felt that I learned less than I should have in these situations. However, during other clinical rounds, an esteemed senior doctor talked casually before the rounds, and this created a relaxed atmosphere'.* "*'I think that it is important, particularly during rounds, to create a welcoming atmosphere, where even stupid questions are allowed and where a student can ask for clarification if he/she does not understand the question. To improve learning, I feel it is important to dare to be stupid!'*" Here Tomi expresses his frustration on the experienced poor teaching skills of a senior doctor and comments on the importance of pleasant atmosphere during teaching sessions."'*In the outpatient clinic, if I do not have a clear hypothesis at the outset, I just start to take a patient history and then examine the patient. Today, after I have taken a complete patient history and examined the patient, I feel I can arrive at a working diagnosis. I also feel more confident about consulting specialists'.*" The data above shows that Tomi's self-confidence is nowadays much better than it was in the in the outpatient clinic. Since completing his clinical neurology studies, Tomi has met several neurological patients at different clinics, and he feels at ease with the examinations. At the emergency clinic, he has also been able to incorporate hypothesis-based reasoning into the neurological examination."*'Concerning the patient examination, I now understand how to diagnose signs and symptoms at the neuroanatomical level. Since completing the course, I have seen a number of patients, most of whom have subacute cerebral symptoms. In all cases, I have been able to figure out the level of neuroanatomical symptoms and signs quite quickly'.*" "*'If I have studied the case carefully, I am better able to discuss the problem with the senior consultant. In many cases, I have identified the source of the problem. I feel much more confident when examining different patients, and I am not afraid anymore. Most of the time, I am able to consult the right specialist, and I reach the correct diagnosis'.*" Summary of Tomi's case {#Sec11} ---------------------- Tomi had earlier experience of dealing with patients with neurological diseases on a primary health care ward. Prior to starting his clinical neurology studies, he felt that his pre-clinical preparation was sufficient. He also had hands-on experience of conducting neurological examination. He felt that this experience provided a solid foundations for the clinical neurological course. He felt demotivated by the busy atmosphere during the rounds and rare tertiary care patients. He did not understand the level of knowledge that senior specialists expected him to have. He felt that the communication between the students and senior doctors was inadequate. His experiences of busy rounds and senior doctors made him feel inferior. He did not learn well in these situations. He also felt that he should have received more tuition in diagnostic reasoning skills. The main positive elements of his learning experience were structured teaching with varying teaching sessions. The increase in self-confidence decreased Tomi's anxious feelings. The negative elements were related to emotional experiences during the course. Petri's case {#Sec12} ------------ Petri had studied neuroanatomy (2nd year) and neurological diseases (3rd year). He remarked:"*'Neuroanatomy did not interest me... I knew that it was important, but I did not study it that well, and I was not interested either. However, this made the following courses difficult'.*" The data above shows that Petri's motivation of learning neuroanatomy was quite low although he seemed to appreciate the topic. Petri further noted:"*'The neurological diseases I knew about were gloomy and depressing. They are progressive, and there is no cure for them. They frightened me. Some members of my family had multiple sclerosis and amyotrophic lateral sclerosis, so I had personal experience of the diseases, and I felt anxious. At work, I had met stroke patients. As I had no neurology training, it found my dealings with them difficult, I felt I should have paid more attention to neurology studies'.*" "*'Neurological studies took a lot of time. I felt anxious and fearful because I did not have enough knowledge to do this course!'*" Above Petri used many string emotional expressions like "they frightened me" and "I felt anxious" that showed well that he had negative anticipation already before the studies. The main reason for this seemed to be the fact that he had "personal experience of the diseases" because some of his family members had had neurological diseases. He also felt that he did not have enough neurology training to deal with stroke patients. He had to study a lot to learn neuroanatomical basics, in addition to clinical practice, which was time consuming, as noted below:"*'I studied a lot. The integration of basic neuroscience and clinical examinations made me feel more confident. We met several patients and practiced neurological examinations so many times that I did not have to think about the mechanical performance and therefore had time to engage in clinical reasoning'.*" The data above shows that Petri's self-confidence was improved through the integration of basic neuroscience and clinical examinations. This was a turning point of his narrative. During the rounds, he found dealing with tertiary care patients (e.g. those with refractory epilepsy) confusing because of the complexity of symptomatology and treatment options. He also felt that managing acute stroke was more difficult and demanding than managing other neurological diseases and that his emotional stress level was greater when he met patients with progressive neurological diseases. However, he realized that there were many common neurological symptoms and disorders and that he should focus mainly on them."*'I felt uncomfortable when meeting patients who had been told they had a rare, fatal disease when their symptoms had at first seemed benign. This interfered with my diagnostic reasoning'.*" Further, he emphasised: *'Neurological diseases, they ARE just more complex than other diseases!'* Although Petri's self-confidence increased with the developing skills in neurology, anxious feelings did not disappear. After the clinical neurological course, while working during the summer break, he met several patients with neurological symptoms*.* Strokes made him feel anxious, as they are so common, and there is a lot to study, as shown below:"*'Some of the symptoms were difficult to define. It helped when I performed a thorough neurological examination. However, I had to keep an open mind. The differential diagnosis: that was difficult. Still, I felt I had sufficient knowledge on the most common types of neurological diseases to deal with the cases'.* He added: *'I felt good at work and liked neurology. I could even diagnose a cluster headache!'*" For students dealing with negative emotions towards neurology, Petri says they need to realise that it will take time to amass the knowledge needed to understand clinical neurology."'*Neurology is such a wide and difficult discipline, and I revere it. To learn, you have to study hard, more so than with other specialties. However, I know now that it can be done!'*" Here Petri's talk is decisive; he has found a resolution for overcoming his challenges in "study hard". Petri knows what he wants for the future and therefore uses utterances such as "*I know now that it can be done!'* Summary of Petri's case {#Sec13} ----------------------- Petri was not interested in neurology before the clinical course. He had neglected pre-clinical neurological studies because they caused him anxiety. Personal experience of neurological disease in his family had given rise to feelings of fear. During the course, structured teaching, practice and a good atmosphere motivated him to learn. Extensive studying further helped. Active participation in teaching sessions and self-directed learning increased his confidence. Poor preparedness on his part for clinical neurology and encounters with frightful diseases decreased his motivation. Today, Petri is confident about dealing with acute neurological patients but continues to feel emotional stress in relation to neurological diseases. Thus, he has not considered neurology as his future specialty. Discussion {#Sec14} ========== In this study, the complexity of neurology and the interpretation of clinical findings were the main causes of negative emotions among the students. Structured teaching effectively reduced these emotions towards neurology, whereas non-structured teaching seemed to increase such emotions. In structured medical teaching, the learning objectives are clear and appropriate, and teachers' didactic methods are suitable for small groups, with supervision and immediate feedback. The teaching focuses on common neurological symptoms and diseases and proceeds from the signs and symptoms to a diagnosis. In Finland, students are expected to acquire the skills needed to work in general practice during their clinical neurology studies. The findings of the present study provide further evidence that the integration of basic neuroscience, anatomy and clinical neurology into training improves problem solving in neurology \[[@CR29]--[@CR32]\]. The students in this study were in the final stages of their studies, and they were about to enter their working lives, with their current attitudes and experiences. We believe that this was an appropriate time to evaluate their learning experiences and self-assessment of their clinical neurological skills. In the voluntary internet survey, a 43% compliance rate was reached, and those who participated returned completed questionnaires, all of which were included in the study. Although the narrative examples are those of two male students, their attitudes were representative of those of the other students with negative anticipation in the cohort and logical generalizations are still possible in the sense of \"if it happens there, it can happen anywhere\" \[[@CR24]\]. The findings showed that students' preconceptions can change. The narrative part of the study demonstrated that Tomi and Petri did well in their neurological studies and that they are gifted students. Despite this, they had negative emotions towards neurology and their ability to learn it. The negative emotional experiences arose from past exposure to neurology. In Tomi's case, this was a patient with end-stage Parkinson's disease, and in Petri's case, it was severe neurological diseases in the family. Tomi also felt that poor communication with the instructors and poor teaching skills among senior doctors increased his anxiety and affected his self-esteem. In contrast, Petri had high demands towards his own level of knowledge, which caused feelings of inadequacy in neurology as well. In both students, these emotions disappeared during the clinical course. Not only Tomi and Petri but also the other participants reported that structured teaching and increased exposure to patients were the most helpful methods to enhance learning in neurology and to overcome the emotional obstacles to learning. It also seemed that with the developing clinical skills in neurology the self-confidence increased, which had a positive effect on the anxious emotions of the students. These two narratives represent cases who expressed negative emotions in both online survey and narrative interview. In both cases negative anticipation decreased during the course. We believe that they are representative examples in the cohort and also represent the substance relevant to our study question. The method used in this study combined qualitative and quantitative methods. The methodology used in this study is new in the field of neurological pedagogic research \[[@CR33]\]. A content analysis is considered a useful tool for examining trends and patterns and provides a basis for monitoring shifts in attitudes \[[@CR19], [@CR20]\] and it is also a powerful data-reduction technique. The narrative analysis is a commonly used qualitative method in social sciences \[[@CR21]--[@CR23]\]. The basic premise of narrative inquiry is that people make sense of themselves and their world by telling stories \[[@CR21]\]. By utilising these methodologies, we believed that we could achieve a broader understanding of students' perceptions of learning neurology. The use of open-ended rather than closed-ended questions in the online survey allowed the students to describe their experiences in their own words. The narrative method deepened the descriptions and helped us to better understand the processes underlying the expressed emotions. The small number of students in this study also directed the choice of methodology. As noted elsewhere, teaching methods may need to be revised to improve the integration of basic science knowledge and clinical neurology into medical training, for example, including virtual cases and group patient meetings \[[@CR29], [@CR34], [@CR35]\]. However, it also needs to be recognised that learning clinical neurology is not only a cognitive but also an emotional process, which should be consciously supported by teachers \[[@CR36]\]. We believe that such support is the best way to meet the individual learning trajectories of medical students. Furthermore, as also stated previously \[[@CR10], [@CR30], [@CR37]\], according to the opinion of the students in this study, pragmatic training with actual patients under supervision leads to the best possible results. As noted in an earlier study, the way in which negative emotions influence evolving professional self-esteem, in this case, that of medical students, is unclear \[[@CR38]\]. However, we can speculate that if there is a connection, it is adverse. In the present study, most types of negative emotions were already present in the content analysis. However, content analysis is best suited to small cohorts \[[@CR19], [@CR20]\]. To study a similar phenomenon in medical students in general, there is a need for validated methods for the assessment that would suit analysing larger study samples. Based on our results, neurology seems to make also Finnish medical students nervous. In this study, the concept of 'neurophobia', a phenomenon originally described by Josefowicz \[[@CR4]\] as not totally serious, became more precise, as more than fear, feelings of anxiety and insecurity were observed, in addition to the preconception that neurology is a difficult discipline in medicine. A neurological examination is certainly akin to a 3D jigsaw puzzle, where a diagnosis is reached by the clinical problem solving that is based on neuroanatomical knowledge, as also corroborated by others \[[@CR14]\]. Therefore, elaborating on the clinical thinking underlying individual cases may help to convert abstract concepts into concrete reasoning. Limitations of our study concern the small sample size, limiting inferences of the analyses. Another limitation concerns the self-reporting, as online survey and interviews took place only after the neurology course. Self-reports of current emotional experiences are likely to be more valid than are self-reports of emotions made somewhat distant in time from the relevant experience \[[@CR39]\]. Results observed in this study are aimed to be evaluated in a larger student cohort, where currently experienced emotions are assessed with suitable questionnaires for stress, anxiety and goal orientation \[[@CR40]\]. In the research literature, the concept of phobia has often been described as an irrational fear of specific objects that is not under voluntary control and that often leads to the avoidance of the phobic situation \[[@CR41]\]. For example, Tobias described 'mathphobia' as an irrational fear of mathematics \[[@CR42]\]. Cemen defined 'mathematics anxiety' as a state of discomfort that occurred in response to situations involving mathematical tasks, which were perceived as undermining the person's self-confidence \[[@CR43]\]. Based on the findings of the present study, we suggest that 'neurophobia' can also be manifested as 'neuroanxiety' because the negative emotions that the students had were not irrational. The findings in our study using content analysis complemented by narrative methodology however showed that students' preconceptions can change, enabling intervention with competent teaching and emotional support. Conclusion {#Sec15} ========== Although the perspective was to study negative repertoires, our interest was to explore which factors decreased them and brought up the positive action tendencies. We observed these positive tendencies and believe that teacher's awareness of them broaden the scope of attention and thought action repertoires during the neurological studies \[[@CR44]\]. Emotions may influence medical education in several ways that need further exploring. Our observations are in accordance with a concept that learning should not be treated simply as a rational, mechanistic process because emotional conditions are shown to affect the performance \[[@CR1]\]. Validated methodology to study the current emotions in larger student cohorts would help to adjust teaching to meet also the attitudes and emotional needs of medical students. We thank all students who participated in the study. We thank MSc Hanna Heikkinen, Faculty of Science, University of Oulu, for constructive comments during the preparation of the manuscript. Funding {#FPar1} ======= No funding. Availability of data and materials {#FPar2} ================================== All data is available through the first author of the manuscript. Authors' contributions {#FPar3} ====================== HA and MS conducted the literature review and the study, and they were the primary contributors to the paper. RK was the supervisor of the study. He was responsible for the conception and design of the research project. All the authors were responsible for the analysis and interpretation of the data. All the authors were involved in the draft of the manuscript, critical revision of the manuscript and approval of the final version. All the authors agree to be accountable for all aspects of the work and for ensuring that questions relating to the accuracy or integrity of any part of the work are appropriately investigated and resolved. Authors' information {#FPar4} ==================== Hanna Ansakorpi, MD, PhD, is a neurologist at the University Hospital of Oulu and a clinical teacher at the University of Oulu, Research Unit of Clinical Neuroscience in Finland. She has Special Competence in medical education and studies university pedagogics. Her clinical and research interest focuses mainly on epilepsy. Marja-Liisa Sumelahti, MD, PhD is a neurologist and a senior lecturer at the University Hospital of Tampere and at the Medical School of University of Tampere, Finland, where she graduated and has completed a subspeciality in medical pedagogics. Her clinical and research interests include MS and migraine. Raimo Kaasila, PhD, is a professor of educational sciences, especially teacher education at the University of Oulu in Finland. He is also adjunct professor in mathematics education. He organizes university pedagogy studies at the University of Oulu. His main research areas are the affective domain in education and the use of narrative and rhetorical methods. Competing interests {#FPar5} =================== The authors declare that they have no competing interest. Consent for publication {#FPar6} ======================= The participants of the interview gave informed consent to be recorded, and those participants selected as representative cases expressing negative emotions towards neurology, gave informed consent for their direct quotes to be published in a research article. Ethics approval and consent to participate {#FPar7} ========================================== The study was presented to The Regional Ethics Committee of the Northern Ostrobothnia Hospital District, and the approval was deemed unnecessary. A written consent was obtained from all students participating the study. Publisher's note {#FPar8} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

1. Introduction {#sec1-toxics-07-00005} =============== Endocrine disruptor compounds (EDCs) are mostly synthetic molecules from industrial origin \[[@B1-toxics-07-00005],[@B2-toxics-07-00005],[@B3-toxics-07-00005]\] but are also some natural molecules \[[@B4-toxics-07-00005],[@B5-toxics-07-00005]\] that are present in the environment and promote adverse modifications of endocrine homeostasis in humans and/or in wildlife animals. EDCs raise serious concerns about their potential health impact. Most of the receptors that are targeted by EDCs are nuclear receptors. These receptors are hormone-dependent transcription factors and, consequently, they exert long-term control of their target cells' phenotype. Membrane receptor signaling can also be affected by EDCs but this potentially leads to a short-term effect, since their signaling pathways exert more acute effects in target cells. An interesting case is that of estradiol, that binds to a transmembrane receptor named GPER (or GPR30) in addition to its nuclear receptors ERα and ERβ \[[@B6-toxics-07-00005]\]. Interestingly, the EDC bisphenol-A exhibits a higher affinity towards GPER than toward its nuclear ER receptors \[[@B7-toxics-07-00005]\]. The understanding of EDCs' mechanisms of action, as well as the extent to which their effects are responsible for health disorders, are the subject of scientific and public controversy. We present here information concerning EDCs compared to hormones in order to evaluate their particular properties and to estimate their potential risks for human and animal health. In a previous article \[[@B8-toxics-07-00005]\], we summarized the main mechanisms of action of EDCs. We present here a more precise view of the different mechanisms that EDCs can exhibit, including:(1)Binding to a hormone receptor leading to activation of its signaling pathway;(2)Binding to a hormone receptor leading to inhibition of its signaling pathway;(3)Interactions with components of hormone signaling pathway downstream of a receptor;(4)Stimulation or(5)Inhibition of an endogenous hormone biosynthesis;(6)Binding to circulating hormone-binding protein;(7)Stimulation or inhibition of hormone-binding protein synthesis or degradation;(8)Stimulation or(9)Inhibition of hormone receptor expression. Among these mechanisms, only the first one is common within the mechanisms of any action relating to hormones. The other mechanisms (numbers 2 to 9) lead to imbalances in endocrine homeostasis that are not consecutive to a direct hormonal-type mechanism of action by EDCs. We have thus taken these various mechanisms into consideration and summarized them in [Figure 1](#toxics-07-00005-f001){ref-type="fig"}, to differentiate the different types of EDCs according to their similarities or differences compared to hormones. 2. EDCs Directly Exhibiting Hormonal Activity Through Receptor Binding (Mechanism 1) {#sec2-toxics-07-00005} ==================================================================================== The most obvious EDCs are those exhibiting a hormone-type mechanism of action, i.e., those able to bind to and activate a hormone receptor (mechanism number 1 above). How is that possible, since hormones are considered to exhibit high affinity and high specificity towards their cognate receptor? Because EDCs exhibiting structures that are different from those of hormones, can sneak into their binding site and interfere with their mechanism of action. The present hormone receptor (HR) couples, in all today's living species, are the fruit of evolution. Natural selection during evolution does not work on each HR couple individually (and neither on interactions involving each enzyme, structural protein or others) but on organisms exhibiting all possible sets of protein forms and thus, various capabilities to thrive in various environmental situations. Concerning HR couples, it is not sufficient that they exhibit high affinity but also that they do not interfere with others. It must thus have a high specificity so that hormones with almost similar structures (androgens vs. estrogens for example) do not interfere. Thus, evolution has not only selected individual efficient HR couples but also couples without mutual interferences, i.e., receptors that recognize their cognate ligand but also inhibit the binding of other molecules with very near structures. Only functional isolation of HR couples allows harmonious endocrine controls. The massive introduction, in terms of number and quantity, of synthetic molecules having more or less the shape and size of hormones explains that some of them can lure the receptors and bind to them \[[@B9-toxics-07-00005]\]. Indeed, receptors have evolved to recognize their cognate ligand but also to impede the binding of other endogenous molecules with near structures. The receptors are thus protected against interaction with endogenous molecules resembling hormones. However, they cannot be protected against interaction with brand new molecules never encountered before during evolution. In addition, these new synthetic molecules can be present in rather high quantities, and therefore can compete with genuine hormones in spite of their lower affinity towards receptors, compared to hormones. Examples of this mechanism concern essentially xenobiotics interacting with hormone nuclear receptors. Indeed, these receptors have hormone-binding sites of rather small size that can potentially accommodate many synthetic organic molecules of industrial origin, but also natural molecules. Moreover, as these receptors exhibit transcriptional activity, their activation by xenobiotics can profoundly affect target cells phenotype. Some EDCs nevertheless interact with membrane receptors \[[@B10-toxics-07-00005]\]. There are many examples of this straightforward classical mechanism, in which EDCs act like hormones by direct interaction and activation hormone receptors. Nevertheless, EDCs might be less efficiently degraded than the natural hormones and thus be more active in vivo because of a longer half-life in blood or in cells. 3. EDCs Directly Inhibiting Endogenous Hormone Action Through Receptor Occupation (Mechanism 2) {#sec3-toxics-07-00005} =============================================================================================== Among the new molecules resembling hormones and able to bind to receptors, some can freeze the receptors conformation in their inactive state, and thus antagonize endogenous hormone action (mechanism number 2 above). Through such a mechanism, exogenous molecules can clearly exert endocrine disruption. For example, polychlorinated biphenyls (PCBs) can suppress transcription through inhibiting the binding of T3 to the thyroid hormone receptor (TR) and consequently, by dissociating the transcriptionally active TR/retinoid X receptor heterodimer complex from the thyroid response element (TRE) \[[@B11-toxics-07-00005]\]. Additionally, anti-estrogenic, anti-androgenic, anti-progesteronic, and anti-ER activities were detected in samples from wastewater treatment plants \[[@B12-toxics-07-00005]\]. In this mechanism, EDCs are able to bind to receptors like hormones by exerting an antagonistic effect, in contrast to hormones. This mechanism of action can be tested in vitro but the cytotoxic effects of unknown substances can confound in vitro assays. This can make the interpretation of results difficult and uncertain, particularly when assessing antagonistic activity \[[@B13-toxics-07-00005]\]. 4. EDCs Interacting with Hormone Signaling Pathways (Mechanism 3) {#sec4-toxics-07-00005} ================================================================= A number of EDCs interplay with endocrine regulations through direct interaction. This means interaction is not with hormone receptors, but with hormone signaling pathway components downstream of receptor activation. Such molecules can exhibit structures that are largely different from those of hormones. For example, fluoxetine (FLX) that is the SSRI (selective serotonin reuptake inhibitor) active substance in the Prozac™ antidepressant has been shown to modify a number of intracellular signaling pathways in various cell types \[[@B14-toxics-07-00005],[@B15-toxics-07-00005],[@B16-toxics-07-00005]\]. Several bisphenols have been shown to interact with Ras small G proteins (particularly K-Ras4B) leading to the activation of the Ras signaling cascade, as shown by raised pERK and pAKT levels \[[@B17-toxics-07-00005]\]. Atrazine, one of the most commonly used herbicides worldwide, acts as an endocrine disruptor by inhibiting cAMP-specific phosphodiesterase PDE4 \[[@B18-toxics-07-00005]\] and thus, favors cAMP intracellular accumulation. Tolylfluanid impairs insulin signaling in human adipocytes through a reduction in insulin receptor substrate-1 (IRS-1) levels downstream from the insulin receptor \[[@B19-toxics-07-00005]\]. The plasticizer di-(2-ethylhexyl)-phthalate (DEHP) is classified as an endocrine disruptor but also as an obesogen and has been shown to act through the peroxisome proliferator activated receptors (PPARs) \[[@B20-toxics-07-00005]\], provoking downstream effects on AMPK, ERK1, ERK2 and ACC activation through phosphorylation. DEHP can also exhibit non-endocrine reprotoxic effects by directly affecting these pathways in gametes \[[@B21-toxics-07-00005]\]. Recently, neonicotinoid pesticides have been shown to induce a change in *CYP19 (aromatase)* promoter usage in Hs578t breast cancer cells, leading to increased aromatase catalytic activity and the activation of MAPK 1/3 and/or PLC pathways \[[@B22-toxics-07-00005]\]. This promoter usage change is similar to that observed in patients with hormone-dependent breast cancer. Another example concerns the effect of triclosan, a broad-spectrum antibacterial and antifungal compound, on ovarian \[[@B23-toxics-07-00005]\] and testicular steroidogenesis through miRNAs which are involved in endocrine regulation and disease development in humans \[[@B24-toxics-07-00005]\]. For example, the miR-6321/Map3k1-regulated JNK/c-Jun/Nur77 cascade contributes to a triclosan endocrine disrupting effect \[[@B25-toxics-07-00005]\]. Histone methylation events are a general component of nuclear receptor mediated transcriptional regulation, for example in the testis \[[@B26-toxics-07-00005]\]. DNA methylation of a Wnt2 promoter, under bisphenol-A (BPA) exposure, is implicated in preeclampsia-like effects in mice \[[@B27-toxics-07-00005]\]. BPA also affects cell proliferation of human placental first trimester trophoblasts \[[@B28-toxics-07-00005]\] and is thus of concern for the sensitive window that is fetal development. In this mechanism, EDCs do not interfere with hormone receptors but downstream of them, at numerous possible sites which can be difficult to identify. Potentially, this type of mechanism should be detectable and quantitated in vitro in cell culture systems. It must be kept in mind that this mechanism can lead to direct, non-endocrine, and toxic effects ([Figure 1](#toxics-07-00005-f001){ref-type="fig"}). 5. EDCs Affecting Endogenous Hormone Concentration (Mechanisms 4 and 5) {#sec5-toxics-07-00005} ======================================================================= Many molecules can exert endocrine disruption, not by interfering directly with hormone receptors, but by affecting, positively or negatively, endogenous hormone(s) biosynthesis (mechanism 4) or degradation (mechanism 5). Such molecules generally exhibit structures that are different from those of hormones, since they do not compete with hormones at the receptor level. 5.1. Mechanism 4 {#sec5dot1-toxics-07-00005} ---------------- One example of this mechanism is that of BPA which, at a low dose, inhibits adiponectin secretion in vitro in human adipocytes \[[@B29-toxics-07-00005],[@B30-toxics-07-00005],[@B31-toxics-07-00005],[@B32-toxics-07-00005]\]. It has been shown that EDC 4-nonyphenol (4-NP) inhibits the secretion of testosterone by Leydig cells stimulated by human chorionic gonadotropin \[[@B33-toxics-07-00005]\] and triclosan induces Vascular Endothelial Growth Factor (VEGF) secretion by human prostate cancer cells \[[@B34-toxics-07-00005]\]. 5.2. Mechanism 5 {#sec5dot2-toxics-07-00005} ---------------- Flame retardants such as polybrominated diphenyl ethers (PBDEs) have been described to act through the induction of hepatic enzymes involved in glucuronidation \[[@B11-toxics-07-00005]\], thus potentially leading to an increase in T4 elimination and the lowering of its concentration in blood. Parabens, which are effective preservatives widely used in cosmetic products, inhibit 17β-hydroxysteroid dehydrogenase (17β-HSD) and consequently inhibit estrogen degradation \[[@B35-toxics-07-00005]\], potentially leading to an increased hormone concentration in blood. In this mechanism again, EDCs do not interfere with hormone receptors but, by affecting endogenous hormone concentration, impact either their biosynthesis or degradation. Such a mechanism has to be studied in vivo but can be tested in vitro when a specific step has been identified. 6. EDCs Affecting Endogenous Free Active Hormone Concentration (Mechanisms 6 and 7) {#sec6-toxics-07-00005} =================================================================================== Many hormones, particularly the hydrophobic ones (steroids and thyroid hormones), are transported by binding proteins in blood. Since EDCs are generally hydrophobic, they are susceptible to compete with small hydrophobic hormones in relation to these transport proteins. 6.1. Mechanism 6 {#sec6dot1-toxics-07-00005} ---------------- A number of EDCs directly interfere with hormone-binding transport proteins, thus competing with the endogenous hormones' concentration in blood. For example, numerous chemicals have been shown to interact with SHBG (steroid hormone-binding protein) or AFP (α-fetoprotein) \[[@B36-toxics-07-00005],[@B37-toxics-07-00005]\] and thus, able to interfere with steroid hormones transport and concentration in blood. The EDCs exerting their effect through this mechanism exhibit some structural resemblance with the hormones, so that they can compete with them for binding with hormone-binding transport proteins. In this mechanism, EDCs do not compete with hormones at the receptor level, but at the level of their circulating binding proteins. They can thus exhibit structural resemblance with the hormones they compete with, and this competition can be studied in vitro. 6.2. Mechanism 7 {#sec6dot2-toxics-07-00005} ---------------- Other EDCs affect the biosynthesis or degradation of hormone-binding transport proteins, so that both the total hormone concentration and/or its free active fraction can be affected. The EDCs acting this way can exhibit chemical structures very different from those of hormones. For example, PBDEs act through the down-regulation of the transport protein transthyretin (TTR) \[[@B11-toxics-07-00005]\] and therefore can lower T4 concentration in blood. Through this mechanism, many toxicants can be catalogued as EDCs because hormone-binding transport proteins are often synthesized and/or degraded by the liver, which, as a (degrading) organ, is the main target of toxicants. 7. EDCs Affecting Endogenous Hormone Receptor Turn-Over (Mechanisms 8 and 9) {#sec7-toxics-07-00005} ============================================================================ 7.1. Mechanism 8 {#sec7dot1-toxics-07-00005} ---------------- Stimulation of endogenous hormone receptors is a way by which a number of EDCs interfere with endocrine homeostasis. BPA has been shown to stimulate leptin receptor expression in ovarian cancer cells in vitro \[[@B31-toxics-07-00005]\]. Cadmium exposure of endothelial HUVEC cells in vitro induced a significant increase of estradiol receptor β (ERβ) and Cyp19a1 enzymes at both mRNA and protein levels, while a drastic dose-dependent decrease of androgen receptor (AR) expression levels was observed after 24 h of exposure \[[@B38-toxics-07-00005]\]. 7.2. Mechanism 9 {#sec7dot2-toxics-07-00005} ---------------- Inhibition of receptor expression is also a mechanism responsible for EDC alteration of the endocrine system. It has also been described that a low oral dose of BPA given to rats can inhibit estrogen receptor expression in their hypothalamic cells \[[@B39-toxics-07-00005]\]. Likewise, inhibition of androgen receptor expressions by BPA has been described in vivo \[[@B29-toxics-07-00005]\] and in vitro in cells from breast or prostate cancer patients \[[@B40-toxics-07-00005]\]. Such an inhibition has also been observed in newborn rats exposed to BPA and was attributed to hypermethylation of the androgen receptor promoter \[[@B41-toxics-07-00005]\]. Moreover, BPA can selectively affect the expression of the ecdysone receptor gene expression in insects \[[@B42-toxics-07-00005]\], whereas it promotes a decrease in ERα, ERβ and GPR30 in fetal mammary gland \[[@B43-toxics-07-00005]\]. In these mechanisms, EDCs generally do not need to resemble hormones to exert their adverse effect by modifying receptor availability. Nevertheless, receptor synthesis and/or degradation are often controlled by its cognate hormone \[[@B43-toxics-07-00005]\]. In this case, EDC structural similarity with hormones can be responsible for this effect. These mechanisms can potentially be identified and studied in vitro, using cell culture assays. 8. In Vitro Tests vs. In Vivo Tests {#sec8-toxics-07-00005} =================================== Mechanisms 1 and 2 are relatively easily amendable to in vitro tests to replace in vivo tests, making use of living animals \[[@B44-toxics-07-00005],[@B45-toxics-07-00005],[@B46-toxics-07-00005]\]. Mechanisms 8 and 9 can also be potentially studied in vitro. In vivo tests in aquatic animals reproducing EDC concentrations recorded in polluted places in the environment are particularly useful \[[@B47-toxics-07-00005],[@B48-toxics-07-00005]\], but not always easy to interpolate to terrestrial species, including humans. The limited capabilities of in vitro models to metabolically activate or inactivate xenobiotics may lead to misinterpretation of the in vitro data if such information is missing \[[@B49-toxics-07-00005]\]. These authors have shown that HC11 cells did not show any biotransformation capability, while the major biotransformation pathways in HepG2 and MCF7 cells were conjugated to sulfate and, to a lesser extent, glucuronic acid. These results suggest that HC11 cells should be a valuable cellular system to study the intrinsic estrogenic activity of the tested compound. In these cells, it is thus the concentrations of EDCs in active form that must be taken into consideration. Using HepG2 and MCF7 cells that are able to metabolize activity can help to take into account part of the metabolic fate of the tested compound that occur in vivo. Since a number of metabolizing enzymes are poorly or not at all expressed in standard in vitro systems, their use in endocrine disruptor testing may result in false negatives for compounds in which bioactivation is a prerequisite. In vitro and in vivo tests are complementary but in vivo tests have to be kept at a minimum for ethical reasons, providing, nevertheless, that in vitro tests give sufficient reliable information. 9. Endocrine Disruption vs. Other Toxicological Mechanisms {#sec9-toxics-07-00005} ========================================================== Molecules with recognized endocrine disruption activity can also have additional adverse effects through other toxicological mechanisms. They can directly be cytotoxic or reprotoxic (i.e., direct alteration of gametogenesis or other reproductive steps) \[[@B50-toxics-07-00005]\], or teratogenic or genotoxic (alteration of DNA: either by epigenetic alterations or through mutations), possibly leading to cancers independent of endocrine-related cancers. For example, BPA exhibits cytotoxic and genotoxic effects not related to its EDC properties \[[@B50-toxics-07-00005],[@B51-toxics-07-00005],[@B52-toxics-07-00005]\]. Likewise, dioxin that acts as an EDC through the aryl hydrocarbon receptor (AhR) \[[@B53-toxics-07-00005]\] has also been shown to be a potent genotoxic \[[@B54-toxics-07-00005]\]. Although these cytotoxic and genotoxic effects are generally observed at higher concentrations than endocrine-disturbing effects, this possibility must be taken into consideration. As with the other toxicants, EDCs exhibit longer effects during early developmental steps, such as embryonic, fetal, neonatal, childhood, and puberty periods \[[@B52-toxics-07-00005],[@B55-toxics-07-00005]\]. Obviously, defects during developmental steps will have consequences during the whole life of the exposed individual \[[@B56-toxics-07-00005]\] and sometimes in its descendants, as these effects often occur through epigenetic mechanisms \[[@B57-toxics-07-00005]\]. Whatever the mechanism in action, it would be wise to study EDC effects in these sensitive windows and observe the consequences in adults. Nevertheless, numerous and long-term experiments for testing individual EDCs or mixtures in animals would be nearly impossible for all suspected molecules. It is therefore advantageous to classify individual molecules according to their disturbing mechanisms, in order to get a better analysis of their synergies in mixtures. 10. EDCs Mechanisms of Action and Risk Assessment {#sec10-toxics-07-00005} ================================================= As stated in the WHO-UNEP 2012 document \[[@B58-toxics-07-00005]\], EDCs represent a challenge as their effects depend on both the level and timing of exposure, being especially critical when exposure occurs during development. "Risk assessment" is the term generally used to refer to the characterization of the potential adverse effects of exposure to hazards. The evaluation of EDC risk assessment is an issue leading to controversies \[[@B58-toxics-07-00005],[@B59-toxics-07-00005],[@B60-toxics-07-00005],[@B61-toxics-07-00005]\]. Therefore, the timing of exposure and of its acceptable quantitative limits are of major interest to assess risk \[[@B62-toxics-07-00005]\]. Nevertheless, the existence of dose-thresholds for endocrine disruptors continues to be debated \[[@B63-toxics-07-00005],[@B64-toxics-07-00005],[@B65-toxics-07-00005],[@B66-toxics-07-00005],[@B67-toxics-07-00005]\] because non-monotonous dose response (NMDR) curves are often considered as an intrinsic property of EDCs in the non-scientific press and general public. It is rather a property derived from the complexity of endocrine regulations \[[@B64-toxics-07-00005]\]. It remains nevertheless, that it can be difficult to distinguish a valid true threshold from an apparent threshold, which merely arises from the limits of detection of the experimental system used. If a molecule exhibits a U-shape dose-response curve in a given experimental system, the U descending branch of the curve should be used as the basis for determining the threshold if the registered response is related to the adverse effect of the molecule. This can lead to exceedingly low limits but, at least, this is more satisfying than, by principle, refusing any limit. Of course, the determination of the control value in the total absence of the molecule under test is primordial to demonstrate a significative, positive or negative, effect at these very low doses. Both authors, Y.C. and T.M.D.N., carried out the writing and editing of this Review. This research received no external funding. The authors declare no conflict of interest. {#toxics-07-00005-f001}

Background {#Sec1} ========== Hemorrhagic fever with renal syndrome (HFRS) is a rodent-borne disease caused by hantaviruses. Each hantavirus tends to be specific to a different rodent or insectivore host \[[@CR1], [@CR2]\]. Two dominant hantaviruses, Seoul virus (SEOV) and Hantaan virus (HTNV), carried by *Rattus norvegicus* and *Apodemus agrarius*, respectively, cause HFRS in China \[[@CR3]\]. China is one of the countries most affected by hantaviruses (mainly HTNV and SEOV). The reported cases in China account for 90% of the total global burden of the HFRS \[[@CR4]--[@CR6]\]. HFRS has become an important public health problem in Asia. The mortality rates have reached 12% in some outbreaks \[[@CR7]\]. In recent years, the incidence of HFRS has significantly decreased. However, 30,000--60,000 cases are reported annually in China \[[@CR8]\]. Hunan Province is one of the most seriously affected areas in mainland China \[[@CR2], [@CR6], [@CR9]\]. Since HFRS was first detected in Hunan Province in 1963, more than 90% of the cities in the province have reported cases \[[@CR4], [@CR10], [@CR11]\]. Hunan Province has reported several hantavirus strains, predominantly SEOV, and various species of rodent host, including *A. agrarius*, *R. norvegicus*, *Mus musculus*, and *Rattus flavipectus* \[[@CR12], [@CR13]\]. All of these species can carry hantaviruses \[[@CR14]\] and were found to carry and transmit hantavirus frequently in recent years \[[@CR15]\]. Rodent population densities, virus prevalence in rodents, diversity of rodents, rodent community composition, and species distributions have important influences on HFRS transmission \[[@CR16]--[@CR22]\]. Different rodent species thrive in different habitats. For example, *A. agrarius* prefers humid and food-rich environments, and is found predominantly in forested regions and fields. *R. norvegicus* is abundant in residential areas, and is the main vector for zoonotic diseases in rural and urban populations \[[@CR4]\]. The main routes of transmission to humans are inhalation of aerosolized urine or feces, contact with the saliva of infected rodents, or via contaminated food, all of which require humans and rodent hosts to share the same space \[[@CR4], [@CR23]\]. Previous studies revealed that land use influences HFRS transmission through the effect on the reservoir, host, and environmental conditions \[[@CR6], [@CR24]\]. To date, few studies have examined the relationships among landscape, rodent community composition and HFRS occurrence. In 2006--2008, the rodent density in different habitats and the prevalence of major rodent-borne diseases (including HFRS) in Nanchang City in Jiangxi Province were investigated and the risks of the rodent-borne diseases were assessed \[[@CR25]\]; The spatial as well as temporal variation in the occurrence of HFRS is linked to geographic differences in the population dynamics of the reservoir rodents in different biomes of Europe \[[@CR26]\]. These studies showed that studying the relationships among landscape, rodent community composition and HFRS occurrence are beneficial works to promote the progress of the understanding of HFRS epidemiology. The first case in Shaoyang, one of the prefecture-level cities most seriously affected by HFRS in Hunan Province, was reported in 1965 \[[@CR27]\]. In 2006, 135 cases were reported in Shaoyang, accounting for 24.1% of the total cases in Hunan Province. There were more than 1000 cases, in total, from 1980 to 1999, but the incidence decreased for unknown reasons by 54.3% during this time period. In prefecture-level city of Loudi, after the first case emerged in the 1970s, the incidence of HFRS increased substantially in the 1990s. Despite a decline in incidence in Loudi that began in the early 2000s, there was still one area with high incidence. The annual incidence in Loudi increased to 3.7 cases per 100,000 people in 2007, and was the highest in Hunan Province. The aims of this study were to: 1) investigate how the community composition of the hosts influences the risk of HFRS among different landscapes; 2) to identify dominant rodent species in different environments; and 3) to investigate the spatiotemporal distribution of hantavirus infection risk at small spatial scales. Methods {#Sec2} ======= Study area {#Sec3} ---------- The study was conducted in the prefecture-level cities of Shaoyang and Loudi, in the southwest of Hunan Province (Fig. [1](#Fig1){ref-type="fig"}). Shaoyang has mainly mountainous terrain, an annual average temperature of 16.1--17.1 °C, and annual rainfall of 1000--1300 mm. Shaoyang has a total land area of 20,829 km^2^ and a population of about 7.1 million people. Loudi covers 8117 km^2^ and has a population of 4.67 million people. In Loudi, the mean annual temperature is about 16.5--17.5 °C, and the annual rainfall is about 1300--1400 mm.Fig. 1Land use and location of trapping sites in the study area, the prefecture-level cities of Loudi and Shaoyang Data collection {#Sec4} --------------- Data on HFRS cases in Shaoyang and Loudi from 2006 to 2013 were obtained from the Hunan Notifiable Disease Surveillance System (HNDSS). The HNDSS is a passive surveillance system. All HFRS cases were first diagnosed based on clinical symptoms, as defined by a national standard \[[@CR28]\]. The diagnosis was confirmed by detection of specific IgM and IgG antibodies to hantaviruses in acute phase serum specimens by enzyme-linked immunosorbent assay (ELISA). Information recorded for each case included sex, age, residential address, and the date of onset of symptoms. The HFRS data in this analysis did not differentiate HTNV from SEOV infections. Cases were geo-coded according to the residential address using Google Earth (Google, Mountain View, California, USA). As most patients' occupations are farmer, their working places were mainly farmland, closing to their family address. We hypothesize that people usually have the most frequent activities near their address and working places. Thus the residential address of HFRS cases could reflect the environmental condition where the infected persons exposure to rodents. The rodent monitoring data in Loudi and Shaoyang were collected by 45 permanent trapping sites covering different environments; 36 in cultivated areas and three, each, in forests, grasslands, and urban areas. As the trapping sites were geocoded at township-level, some sites in the same township are represented by one point in the map (Fig. [1](#Fig1){ref-type="fig"}). A total of 48,328 trap-nights occurred between 2006 and 2013. According to the HFRS monitoring program of Hunan, rodents were trapped in March, April, September, and October \[[@CR29]\]. The traps were baited with peanuts, placed at each trapping site each night, and checked in the morning. More than 100 traps were placed per site in peridomestic environments, at approximately 12--15 m intervals, for 3 consecutive nights. More than 200 traps were placed per site outdoors, for 3 consecutive nights (every 5 m in each row, with 50 m between rows). The trapped rodents were numbered and the species and sex were identified. Land use data were extracted from the GlobCover 2009 land cover map, provided by Université Catholique de Louvain (UCL) and ESA (<http://due.esrin.esa.int/page_globcover.php> \[[@CR30]\]), with a resolution of 300 m. The original GlobCover 2009 global land cover data were collected by the Medium Resolution Imaging Spectrometer (MERIS) sensor data from the Envisat satellite. The study areas were categorized into five land use types, cultivated land, forest, grassland, urban land, and water bodies (such as rivers and lakes). Maps were created using ArcGIS 10.0 (ESRI Inc., Redlands, CA, USA). Statistical analysis {#Sec5} -------------------- The same data analysis was conducted for Loudi and Shaoyang. First, the annual proportions of HFRS cases for the five land use types were calculated. A matrix (*R*) was constructed, with rows representing the proportion of HFRS cases for one land use type in different years, and columns representing the proportion of HFRS cases in the same year for different land use types. Second, the annual proportions of different rodent species were calculated based on rodent surveillance data. The rodents were classified mainly as *R. norvegicus*, *M. musculus*, *A. agrarius*, *R. flavipectus*, and other rodent species (including *Rattus losea* and *Microtus fortis calamorum*). The rodent community composition was quantified as matrix *C*, with rows representing the proportion of the same rodent species in different years, and columns representing the proportion of different rodent species in the same year. Elements of each column in matrix R and matrix C should add up to one. After that, the coefficient matrix *β* was calculated from *R* and *C* according to Eq. ([1](#Equ1){ref-type=""}) using the method of matrix right division to determine how rodent community composition influences the HFRS occurrence probability: $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \left(\begin{array}{cccc}{R}_{11}& {R}_{12}& \cdots & {R}_{1j}\\ {}{R}_{21}& {R}_{22}& & \\ {}\vdots & & \ddots & \vdots \\ {}{R}_{i1}& & \cdots & {R}_{ij}\end{array}\right)=\left(\begin{array}{cccc}{\beta}_{11}& {\beta}_{12}& \cdots & {\beta}_{1k}\\ {}{\beta}_{21}& {\beta}_{22}& & \\ {}\vdots & & \ddots & \vdots \\ {}{\beta}_{i1}& & \cdots & {\beta}_{ik}\end{array}\right)\cdot \left(\begin{array}{cccc}{C}_{11}& {C}_{12}& \cdots & {C}_{1j}\\ {}{C}_{21}& {C}_{22}& & \\ {}\vdots & & \ddots & \vdots \\ {}{C}_{k1}& & \cdots & {C}_{\mathrm{kj}}\end{array}\right) $$\end{document}$$ where *R*~*ij*~ is the proportion of HFRS cases in area *i* in year *j*, *β*~*ik*~ shows the potential contact rate of HFRS from rodent species *k* to humans in area *i*, and *C*~*kj*~ is the proportion of rodent species *k* in year *j*. The *β* matrices for Loudi and Shaoyang are shown in Fig. [2](#Fig2){ref-type="fig"}, with low values in dark blue and high values in red. Each value in the figure is a coefficient for one rodent species and one land use type. All the values are dimensionless. Positive values represent positive association among the HFRS occurrence, rodent species, and land use types, and negative values represent negative associations.Fig. 2Visualized coefficient matrix showing the relationships among rodent community composition, land use types and HFRS occurrence in (**a**) Loudi, (**b**) Shaoyang. The coefficient values are color coded from blue (low values) to red (high values) All data were divided into two categories. Training data (75%), collected from 2006 to 2011, were used to develop the model and estimate the coefficient matrix. Validation data (25%), collected from 2012 to 2013, were used for model evaluation. The matrices *R* and *C*, constructed with data from 2006 to 2011, were used to calculate the coefficient matrix *β*. Based on the testing matrix *C*, constructed with data from 2012 to 2013, and the coefficient matrix *β*, we estimated the proportion of HFRS occurrence among different land use types in 2012--2013. The calculated results and the observed data from 2012 to 2013 in both Loudi and Shaoyang were used to perform a linear fitting to assess the accuracy of our predicted results. The accuracy of prediction was reflected by the R^2^ and was thought as better when the R^2^ was closer to 1. All statistical analyses were performed using SPSS 19 software (SPSS Inc., Chicago, IL, USA) and Matlab (vR2012b) (Math Works Inc., Natick, MA, USA). Results {#Sec6} ======= Species distribution and HFRS occurrence {#Sec7} ---------------------------------------- A total of 906 rodents were trapped in Loudi, where *A. agrarius*, the main reservoir of HTNV, was the predominant species, accounting for 91.4% of all trapped rodents in 2009. The number of *M. musculus* from 2006 to 2013 varied, with none trapped in 2009 and 109 trapped in 2012. The number of *R. norvegicus* decreased yearly from 2006 and none were trapped in 2010 and 2011. Other species (mainly *Rattus losea and Microtus fortis calamorum*) were trapped starting in 2009 (Figs. [3a](#Fig3){ref-type="fig"} and [4a](#Fig4){ref-type="fig"}).Fig. 3Distribution of rodent species and HFRS cases in Loudi, 2006--2013. **a** Proportion of each rodent species, (**b**) Proportion of HFRS cases among different land use typesFig. 4Number of rodents trapped and HFRS cases reported in (**a**) Loudi, (**b**) Shaoyang A total of 742 HFRS cases were confirmed in Loudi between 2006 and 2013. Figure [4a](#Fig4){ref-type="fig"} shows the annual distribution of cases. More cases occurred in 2007 than in any other year. The number of HFRS cases declined from 2007 to 2010. Case reports increased in 2011 and 2013, with 94 and 115 cases reported, respectively (Fig. [4a](#Fig4){ref-type="fig"}). There was little annual variation in the proportion of HFRS cases for each land use type in Loudi (Fig. [3b](#Fig3){ref-type="fig"}). Cultivated land consistently had the largest proportion of cases. The distributions of HFRS cases in urban areas, forests, and grasslands were similar. Except in 2007 and 2013, no cases occurred in bodies of water. In 2010, consistent with the increase in other rodent species, there was an increase in HFRS cases in grassland areas. In Shaoyang, a total of 858 rodents were trapped during the study period. *R. norvegicus* (67.7%) and *A. agrarius* (17.8%) were the predominant rodent species. *R. flavipectus* and *M. musculus* accounted for 71.9% of all trapped species in 2006, but this proportion declined over the next 7 years. In 2010 and 2011, no *M. musculus* were trapped, but they appeared again in 2012 and 2013. No *R. flavipectus* were trapped from 2011 to 2013. There were no other rodent species trapped in Shaoyang (Figs. [5a](#Fig5){ref-type="fig"} and [4b](#Fig4){ref-type="fig"}).Fig. 5Distribution of rodent species and HFRS cases in Shaoyang, 2006--2013. **a** Proportion of each rodent species, (**b**) Proportion of HFRS cases among different land use types Shaoyang had 797 HFRS cases during the study period and the incidence was highest in 2007. The number of cases declined in 2008 and increased between 2009 and 2013 (Fig. [4b](#Fig4){ref-type="fig"}). The proportion of cases for each land use type varied over the study period. Cultivated land consistently had the highest proportion of cases. In urban areas, the proportion of cases declined annually. However, annual occurrence increased in forests. There was little variation in HFRS cases in grassland areas over the study period. There were a few HFRS cases reported in bodies of water from 2006 to 2010, but no cases were reported in these areas after 2010 (Fig. [5b](#Fig5){ref-type="fig"}). Relationships between rodent hosts, land use types, and HFRS occurrences {#Sec8} ------------------------------------------------------------------------ The coefficient matrix, *β*, identified the potential influence of rodent species distributions in different land use types on the occurrence of HFRS. In Loudi, HFRS cases in cultivated land were positively associated with all rodent species. In forests, HFRS cases were positively associated with *R. flavipectus* and *M. musculus* and negatively associated with *R. norvegicus* and other rodent species. In grasslands, HFRS cases were positively associated with *R. norvegicus* and other rodent species and negatively associated with *R. flavipectus* and *M. musculus*, while the opposite occurred in forests. In urban land and water bodies, HFRS cases were negatively associated with *R. norvegicus* and other rodent species. There was a weak positive association of *M. musculus* and *A. agrarius* with HFRS cases in urban land, and *R. flavipectus* was negatively associated with HFRS cases. In water bodies, there was a weak negative association of HFRS cases with *A. agrarius* and a positive association with *R. flavipectus* (Fig. [2a](#Fig2){ref-type="fig"}). In Shaoyang, HFRS cases were positively associated with *R. norvegicus*, *A. agrarius*, and *R. flavipectus*, and negatively associated with *M. musculus* in cultivated land, forest, and grassland. In urban land, HFRS cases were positively associated with *R. norvegicus* and *M. musculus*, and negatively associated with *A. agrarius* and *R. flavipectus*. In water bodies, there was a weak negative association of all rodent species, except *A. agrarius*, with HFRS cases. Additionally, there was a weak association of other species with HFRS cases in all land use types (Fig. [2b](#Fig2){ref-type="fig"}). Risk of potential contacts between humans and rodents in different land use types {#Sec9} --------------------------------------------------------------------------------- The proportions of HFRS cases among different land use types in Loudi and Shaoyang in 2012--2013 were predicted. In Loudi, the land use type with the highest predicted risk of HFRS was cultivated land, following by forest and urban land in both 2012 and 2013. The model predicted that grassland and water bodies would have a low risk of HFRS in these 2 years. In Shaoyang, cultivated land had the highest predicted risk of HFRS in both years, followed by urban land, forest, grassland, and water bodies in 2012, and followed by forest, grassland, urban land, and water bodies in 2013. Figure [6](#Fig6){ref-type="fig"} shows the predicted probability of occurrence of HFRS cases as well as the corresponding observations in both Loudi and Shaoyang in 2012--2013. The predicted and observed proportions in the same area in the same year were paired to assess the accuracy of the predictive model. The scatterplot in Fig. [7](#Fig7){ref-type="fig"} shows the concordant relationship of the predictions and observations. The R^2^ reflected that our prediction was accurate (R^2^ = 0.934).Fig. 6Predicted and observed HFRS occurrence probability among different land use types in Loudi and Shaoyang, 2012--2013. The HFRS occurrence probability is predicted by Eq. [1](#Equ1){ref-type=""}Fig. 7Scatterplot showing the predicted and observed HFRS occurrence probabilities. The HFRS occurrence probability is predicted by Eq. [1](#Equ1){ref-type=""} Discussion {#Sec10} ========== This study investigated the relationships among HFRS occurrence, land use type, and rodent community composition. The results indicated that different rodent species influenced the HFRS occurrence for different land use types. Overall, the highest probability of HFRS was on cultivated land, followed by urban areas, forests, and grasslands. Relatively few cases of HFRS were identified in water covered areas in both Shaoyang and Loudi. For the same land use type, the probability of HFRS occurrence varied between cities. The high probability of HFRS on cultivated land may be due to the humid environment with adequate food for rodents to survive. Farmers working on cultivated land increase the potential for contact between rodents and humans, thereby increasing the risk of HFRS transmission. In 2012, relatively high HFRS risk was predicted in urban areas in Shaoyang, but in 2013, the predicted HFRS risk was lower. This might have resulted from increases in *R. norvegicus* and *A. agrarius* in 2012 and 2013. *R. norvegicus* was positively associated with HFRS in urban land while *A. agrarius* was negatively associated with the HFRS in urban land. The negative correlation was stronger than the positive correlation (Fig. [2b](#Fig2){ref-type="fig"}). Therefore, the increase in *A. agrarius* had a greater influence on HFRS in urban land in Shaoyang. An increase in intensive human activities, such as farming, leading to agricultural encroachment on forests, grassland areas, and water covered areas, and large human populations in urban areas changing the geographical landscape \[[@CR20]\], has an important impact on the spread of viruses. A previous study, focused on HFRS cases caused by HTNV in rural areas, found that agricultural land use and cultivated soil were related to high risk for HFRS \[[@CR6]\]. We found the risks among different land use types varied in relation to rodent community composition. Hantaviruses are transmitted to humans by infected rodents. Different land uses lead to different rodent community composition and distribution \[[@CR31], [@CR32]\]. Moreover, each land use type has a predominant rodent species. In the current study, the risks of hantavirus infection in cultivated land were associated with different rodent species in Loudi and Shaoyang. The risks of HFRS occurrence in other land use types varied for different rodent species. In Loudi, *A. agrarius* was the most predominant species (Figs. [3](#Fig3){ref-type="fig"} and [4a](#Fig4){ref-type="fig"}). However, the highest risk of hantavirus infection was on cultivated land, and mainly correlated with *R. norvegicus* (Fig. [2a](#Fig2){ref-type="fig"})*.* This suggests Loudi city may be a mixed-type epidemic area. In Shaoyang, *R. norvegicus* was the predominant species (Figs. [5](#Fig5){ref-type="fig"} and [4b](#Fig4){ref-type="fig"}). Cultivated and urban areas had higher risk of HFRS and HFRS in these areas was predominantly associated with *A. agrarius* and *M. musculus*, respectively (Fig. [2b](#Fig2){ref-type="fig"}), indicating that Shaoyang may be a mixed-type epidemic area. It can be concluded that both of the cities are mixed-type HFRS epidemic areas with various reservoir rodents. The corresponding risks of potential contact between humans and rodents in different landscapes may also change over time with varied rodent community composition. Different rodent population dynamics have disparate influences on HFRS occurrence. *A. agrarius* and *R. norvegicus* were the predominant species in Loudi and Shaoyang, respectively. According to monitoring data from the last 20 years in China, the highest virus-carrying indexes among host animals in wild and residential areas are for *A. agrarius* and *R. norvegicus*, respectively \[[@CR33]--[@CR35]\]. Different rodents have their own preferred habitats and different abilities to carry and transmit pathogenic viruses. *A. agrarius* are more active outdoors and *M. musculus*, *R. norvegicus*, and *R. flavipectus* are active both outdoors and indoors \[[@CR13]\]. We found different rodent species in both Loudi and Shaoyang, so the occurrences of HFRS cases in both outdoor (cultivated land, forest, grass, and water) and indoor (urban land) environments are consistent with prior knowledge. The coefficient matrix of Loudi indicated that *R. norvegicus* was the dominant species affecting HFRS risk on cultivated land. HFRS occurrence and the related dominant rodent host varied for each environment. This suggests that rodent community composition has a significant influence on the epidemic pattern and transmissions of hantaviruses. Based on these findings, preventative measures can be developed for different land use types, in different cities and epidemic areas. HFRS is related to the number of rodents in different environments. Therefore, we can effectively identify the dominant rodent species in different areas and enact preventative measures to reduce the risk of hantavirus transmission. Cultivated land was a high-risk area for HFRS in our study. The dominant rodent species in this environment has an important impact on the HFRS risk. Therefore, more attention should be spent reducing rodent numbers in these environments. This is consistent with a previous study that found that HFRS cases commonly occur in rural areas \[[@CR36]\]. *R. norvegicus* was the main vector for hantavirus in Loudi and the main vectors in Shaoyang were primarily *A. agrarius*, *R. flavipectus*, and *M. musculus*. Differences in rodent community composition may result in different epidemic characteristics, infection risks, and even control measures. For example, when *A. agrarius* is the predominant species in the rodent population, as in Loudi, the main risk of HFRS is from cultivated land, so the prevention of HFRS should focus more on the farmlands. In contrast, when *R. norvegius* is the predominant species in the rodent population, such as in Shaoyang, the main risk of HFRS is from cultivated land, forest, and urban land, which indicates that more attention should be paid to all these types of land. Our study indicates that rodent community composition and land use types are associated with the epidemiology of HFRS. This information can be used to develop species-specific control measures to reduce the risk of potential contact between hantavirus and humans in different environments. Several limitations for this study should be noted. First, it only considered the influence of rodents on HFRS. Hantavirus transmission results from a combination of environment, climate change, change in biotope, hantavirus species, and social factors \[[@CR13], [@CR31], [@CR37]\]. Second, more detailed information about both rodents and humans needs to be considered, including rodent density, virus-carrying index, and population density. In the absence of the virus-carrying and population density data, we cannot investigate the actual role of rodent species in viral transmission from rodents to humans. Third, change in land use was not considered in our model because these data were not available. Finally, we used the postal addresses of patients to represent the sites of contact, which might have induced measurement error. The accuracy of address resolution was also limited. Further studies are needed to determine the effect of rodent community composition, density, distribution and virus-carrying index on the risks of HFRS transmission. Additionally, potential seasonal variation in prevalence is critical and should be considered when studying contact rate. It is therefore prevalence, seasonal variations of prevalence, and other environmental factors should also be considered in future studies. Conclusions {#Sec11} =========== This study identified the dominant rodent species for different land use types in areas with HFRS, and provides support for the development of regional rodent monitoring programs to prevent HFRS in different environments. We also found that change in rodent community composition was associated with risk of hantavirus infection in different land use types. In addition, this study provides baseline data for HFRS incidence in Loudi and Shaoyang, China. HFRS : Hemorrhagic fever with renal syndrome HNDSS : Hunan notifiable disease surveillance system HTNV : Hantaan virus SEOV : Seoul virus We thank James N. Mills and Gregory E. Glass for their valuable comments. Funding {#FPar1} ======= This work was supported by Construct Program of the Key Discipline in Hunan Province of China (2011001), National Natural Science Foundation of China (81673234, 31500383, 71473264, 41476161), Science and Technology Planning Project of Hunan Province, China (2015JC3063), Fundamental Research Funds for the Central Universities, and Key Subject Construction Project of Hunan Normal University (geographic information systems). Availability of data and materials {#FPar2} ================================== The data that support the findings of this study are available from Hunan Provincial Center for Disease Control and Prevention but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission of Hunan Provincial Center for Disease Control and Prevention. XT, RH, HX, LG, SH, HTian were involved in the conceptualization, research design, execution and write-up of the first draft of the manuscript. HTan, YL, HG and PZ contributed to database design and data analysis. HY and ZY.X.H advised on the study design and the analysis and interpretation of results. All authors were involved in preparation of the manuscript. All authors read and approved the final manuscript. Ethics approval and consent to participate {#FPar3} ========================================== The present study was reviewed and approved by the research institutional review board of the Hunan Provincial Centre for Disease Control and Prevention (CDC). In this study, all the patient medical data analyzed were anonymized for the consideration of confidentiality, only aggregated data were used in the data analysis and no personal information has been used. The whole rodent trapping campaign obtained new samples specifically for this study and was validated by the Animal Ethics Committee of the Hunan CDC. Because the methods did not include animal experimentation, it was not necessary to obtain an animal ethics license. Furthermore, none of the rodent species investigated in the present study are protected in China and none of the species captured are included in the China Species Red List. Consent for publication {#FPar4} ======================= Not applicable. Competing interests {#FPar5} =================== We have read and understood BMC Infectious Diseases policy on declaration of interests and declare that we have no competing interests. Publisher's Note {#FPar6} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Introduction {#sec1-1} ============ Tuberculosis (TB) is a common infectious disease in the developing countries. Antitubercular therapy with the first line drugs is very effective and well-tolerated with only few major adverse reactions. Cutaneous adverse drug reactions (CADR) with antitubercular treatment (ATT) can make further management of TB challenging.\[[@ref1]\] Isoniazid is the first-line antitubercular drug par excellence and an essential component of all antitubercular regimens unless the patient is intolerant to it, or bacilli show resistance. Most common adverse effects of isoniazid are peripheral neuritis and hepatitis, which are more common in alcoholics and older patients, but CADR are rare with incidence of \<0.001% to maximum of 3/1000 patients treated.\[[@ref1][@ref2]\] Case Report {#sec1-2} =========== A 63-year-old Indian woman was diagnosed as a case of pulmonary TB when she was evaluated for fever and cough with expectoration. She was started on the first-line antitubercular drugs as per Directly Observed Treatment Short Course regimen for TB. After 8 weeks of ATT, she reported to dermatology department of our hospital, for complaints of acute onset erythema along with severe itching all over the body for the last 7--10 days. Dermatological examination revealed generalized involvement of the body with extensive nonuniform dusky erythematous scaly plaques involving the scalp, face, nape of the neck, trunk, arms, legs, palms, and soles \[[Figure 1](#F1){ref-type="fig"}\]. Erythema and scaling were more pronounced over trunk and legs. Furthermore, the pruritic plaques were first noted on trunk and legs which increased in size and coalesced to involve the entire body in a few days. Scalp lesions formed red-yellow scales with hair loss \[[Figure 1](#F1){ref-type="fig"}\]. On palms and soles, the exfoliative eruption led to severe sloughing of the epidermis. No significant lymphadenopathy or hepatosplenomegaly was observed. {#F1} There was no history of any other drug intake, or history of jaundice, chest pain, palpitation, and dyspnea on exertion. There was no preexisting dermatosis or prior exposure to chemical precipitants of dermatitis or any other medical problem. Family history was negative for similar conditions or skin disorders. General physical examination was unremarkable while systemic examination revealed edema feet. Laboratory investigations were within normal limits. HIV-ELISA was nonreactive. ATT was stopped immediately. Oral antihistaminics were started along with supportive therapy and topical emollients. The patient improved over a period of 1--2 weeks with decreased erythema and a significant reduction in scaling \[[Figure 2a](#F2){ref-type="fig"}\]. {#F2} On rechallenge after 2 weeks of stopping ATT, individual drugs were reintroduced in a sequential manner starting with ethambutol, followed by pyrazinamide, then rifampicin, and isoniazid at last at an interval of 1 week between the drugs. Prior to isoniazid rechallenge, she did not develop any signs of CADR but on introducing isoniazid, she rapidly developed similar erythematous lesions with intense itching within 48 h \[[Figure 2b](#F2){ref-type="fig"}\]. Isoniazid was withdrawn and diagnosis of "isoniazid induced erythroderma" was made. At present, the patient was given symptomatic treatment with topical emollients and oral antihistaminics. The lesions subsided in 1 week, and the patient was prescribed alternative regimen of ATT excluding isoniazid. The causality assessment was "certain" on WHO-UMC causality assessment scale; whereas "probable" on Naranjo\'s scale (Score 7) for isoniazid. The severity of ADR was found to be "moderate (level 3)" as per the modified Hartwig and Siegel Scale. Discussion {#sec1-3} ========== Erythroderma is an intense generalized redness of the skin, first described by Von Hebra in 1868. It is an inflammatory disorder and an extreme state of dysmetabolism characterized by extensive erythema and scaling all over the body classically involving more than 90% of the body surface. It is of great concern because of significant risk of morbidity and mortality owing to dysmetabolism and its complications, in addition to the risks inherent to the underlying disease and its therapy.\[[@ref3]\] Various causes of erythroderma in adults include pre-existing eczema, psoriasis, lymphoma, leukemia, and drugs such as phenylbutazone, hydantoin derivatives, carbamazepine, sulfonamides, penicillins, cimetidine, diltiazem, dapsone, allopurinol, gold salts, and lithium. Exposure to the causative drug may last for 2 weeks to several months before the reaction emerges. Drug-induced erythroderma has the best prognosis among all the causes of erythroderma often resolving in 2--6 weeks.\[[@ref4]\] In the present case, the patient presented with erythema and scaling involving more than 90% of the body surface area along with itching within 8 weeks of ATT. ATT was stopped, and a significant improvement was noted within 1 week. However, the patient developed exfoliative dermatitis again after rechallenge test with isoniazid and improved after stopping it leading to the diagnosis of isoniazid-induced erythroderma. Prompt resolution of the lesions after withdrawal of the ATT and start of oral antihistaminics further supported the diagnosis. The patient was now prescribed rifampicin 450 mg, pyrazinamide 1500 mg, ethambutol 1000 mg, and levofloxacin 750 mg for a period of 2 months in intensive phase followed by levofloxacin 750 mg, rifampicin 450 mg for a period of 4 months in continuation phase. The underlying pathogenesis of this hypersensitivity, whether immune-mediated and/or toxic in nature, is unclear. Predisposing factors for hypersensitivity reactions to ATT include HIV infection, polypharmacy, advanced age, autoimmune disease, and renal or liver impairment.\[[@ref5]\] In a large tertiary care center study on CADR with antitubercular drugs, pyrazinamide was the most common offending drug (2.38%), followed by streptomycin (1.45%), ethambutol (1.44%), rifampicin (1.23%), and isoniazid (0.98%).\[[@ref1]\] In contrast to the findings of the above-mentioned study, pyrazinamide, ethambutol, and rifampicin were well-tolerated by our patient; however, she developed reaction to isoniazid. There are many case reports of exfoliative dermatitis with other antitubercular drugs; but, to the best of our knowledge, only 3 cases of erythroderma induced by isoniazid alone is reported so far.\[[@ref6][@ref7][@ref8][@ref9][@ref10][@ref11][@ref12][@ref13]\] Higher incidence of TB and CADR in HIV-infected persons poses a challenge for clinicians, particularly in high HIV-prevalence settings. It is important to recognize the CADRs to ATT so that severe and potentially life-threatening adverse reaction can be identified and managed early. It is equally important to continue ATT in minor CADRs so that patients suffering from TB can be cured and rendered noninfectious as early as possible from uninterrupted ATT along with prevention of development of drug resistance. To conclude, erythroderma as a rare but potentially fatal drug reaction with isoniazid. Immediate withdrawal of offending drug alongwith supportive measures carries good prognosis. Hence, cautious use of isoniazid can help in early identification and management of this ADR. {#sec2-1} ### Financial Support and Sponsorship {#sec3-1} Nil. ### Conflicts of Interest {#sec3-2} There are no conflicts of interest.

1. Introduction {#sec1-micromachines-10-00324} =============== Tactile sensors, most commonly referred to as strain and pressure sensors, can collect mechanical property data from the human body and the local environment, and then provide valuable insights into the human health status or artificial intelligence systems \[[@B1-micromachines-10-00324],[@B2-micromachines-10-00324],[@B3-micromachines-10-00324]\]. Meanwhile, it can also be equipped on robots in order to be aware of their surroundings, keep away from potentially destructive effects and provide information for subsequent tasks such as in-hand manipulation \[[@B4-micromachines-10-00324],[@B5-micromachines-10-00324]\]. Change of resistance, capacitance, electrical charge and optical distribution can be used in various sensing systems \[[@B6-micromachines-10-00324]\], and the typical sensing mechanisms for tactile sensors includes piezo-resistive, capacitive, piezo-electric and optical. Compared with other types of tactile sensors, capacitive sensors have high sensitivity and fast frequency response \[[@B7-micromachines-10-00324],[@B8-micromachines-10-00324],[@B9-micromachines-10-00324]\]. Thus, many resonance capacitive sensors, which are very sensitive to small changes, are designed and optimized \[[@B10-micromachines-10-00324],[@B11-micromachines-10-00324],[@B12-micromachines-10-00324],[@B13-micromachines-10-00324]\]. The electrostatic actuation method is an important method of Micro electromechanical systems (MEMS). The system actuated by the electrostatic force is parametric excitation system \[[@B14-micromachines-10-00324],[@B15-micromachines-10-00324],[@B16-micromachines-10-00324]\]. Among the electrostatic actuation structures, parallel-plates configuration is widely used in MEMS \[[@B17-micromachines-10-00324],[@B18-micromachines-10-00324],[@B19-micromachines-10-00324],[@B20-micromachines-10-00324],[@B21-micromachines-10-00324],[@B22-micromachines-10-00324]\]. However, this actuation method can lead to instabilities like pull-in \[[@B23-micromachines-10-00324],[@B24-micromachines-10-00324],[@B25-micromachines-10-00324]\]. In order to obtain large amplitude and avoid the instabilities, the fringing-field actuation technique has got its attention. This actuation approach has several advantages, including the ability to obtain large amplitude displacements without the limitation of the proximity of the electrodes and the possibility of significantly tuning the resonant frequency response range \[[@B26-micromachines-10-00324]\]. This structure has attracted attention and has been studied by some communities. Experimental and theoretical analysis of micro-cantilevers actuated by fringing-field electrostatics was performed \[[@B26-micromachines-10-00324]\]. The different behaviors of curved micro beam with low initial elevation and relatively high initial elevation were studied when actuated by fringing-field electrostatic force \[[@B27-micromachines-10-00324]\]. A parametrically excited electrostatic resonator, which had a flexible support, used weaker electrostatic fringe fields to get higher vibrational amplitude \[[@B28-micromachines-10-00324]\]. These researches show that the device actuated by the fringing electrostatic force can reveal many new static and dynamic behaviors. In the present work, researchers always care about the situations when the beam thickness is near the electrode thickness \[[@B26-micromachines-10-00324]\]. In these cases, the relationship between the electrostatic force and the initial displacement is linear when the beam vibrates near the middle of the electrode in the thickness direction. However, we find that the relationship between the electrostatic force and the initial displacement is nonlinear when the thickness of the electrode is much more than the thickness of the beam. This situation can reveal some new dynamic behaviors. Based on the dynamic investigations into the micro cantilevered beam actuated by fringing electrostatic force, the working principle and the usage patterns of a new micro tactile sensor are presented in this paper. The paper is organized as follows. In [Section 2](#sec2-micromachines-10-00324){ref-type="sec"}, a new resonance tactile sensor is designed and the dynamic modeling of a micro cantilevered beam actuated by fringing electrostatic fields is given. This section consists of three parts: in [Section 2.1](#sec2dot1-micromachines-10-00324){ref-type="sec"}, the working principle of the micro tactile sensor is presented; in [Section 2.2](#sec2dot2-micromachines-10-00324){ref-type="sec"}, the fringing electrostatic force and the influences of geometric parameters on this force are found; in [Section 2.3](#sec2dot3-micromachines-10-00324){ref-type="sec"}, the governing equation of this new kind of fringing electrostatic actuation mode is outlined. In [Section 3](#sec3-micromachines-10-00324){ref-type="sec"}, the dynamic characteristics of the micro cantilevered beam which actuated by fringing electrostatic force are analyzed. In [Section 4](#sec4-micromachines-10-00324){ref-type="sec"}, an experiment is designed to observe the dynamic analysis of this structure. In [Section 5](#sec5-micromachines-10-00324){ref-type="sec"}, the dynamic behaviors of this structure are summarized and the usage patterns of this micro tactile sensor are presented. 2. Problem Formulation and Dynamic Modeling {#sec2-micromachines-10-00324} =========================================== 2.1. Working Principle {#sec2dot1-micromachines-10-00324} ---------------------- The proposed micro tactile sensor consists of a micro cantilevered beam and a pair of micro electrodes. The concept structure of the micro tactile sensor is shown in [Figure 1](#micromachines-10-00324-f001){ref-type="fig"}. One end of the micro cantilevered beam is fixed on the base. The edge of the micro electrode is also connected to the base. The base is nonconductive. The micro cantilevered beam and the micro electrode are conductive, but they are disconnected. When the electrostatic voltage is applied to the micro beam and the micro electrode is connected to the ground wire, the fringing electrostatic force appears between the micro beam and the micro electrode. Under the actuation of the electrostatic force, the micro beam vibrates. When the pressure *P* is applied to the micro electrode, the parts which link electrode and base deform linearly, the micro electrode moves from position 1 to position 2. Then the relative position between the micro cantilevered beam and the micro electrode changes $d_{p}$. This relative position change leads to the change of fringing electrostatic force, which effects the vibration behaviors of the micro beam. In this way the vibration behaviors of the micro beam can reflect the pressure applied to the micro electrode. Some obvious advantages are as following: a large deflection can be obtained without the limitation of the proximity of the electrodes; this structure which is compatible with the circuit can be designed to smaller scale; a smaller scale leads to a better sensitivity. It is the key issue to grasp the relative position change effects the vibration behaviors of the micro cantilevered beam. The paper focuses on the influence of the relative position change in the vibration behaviors of the micro cantilevered beam. This can provide support for sensor design. [Figure 2](#micromachines-10-00324-f002){ref-type="fig"} is the schematic illustration of micro cantilevered beam and micro electrode. In this picture, $l_{b}$ $w_{b}$ and $t_{b}$ denote the length, width and thickness of the micro beam respectively; $l_{s}$ $w_{s}$ and $t_{s}$ denote the length, width and thickness of the micro electrode respectively; $d_{g}$ denotes the slit gap in the width direction; $d_{}$ denotes the initial displacement in the thickness direction, it is the placement position of the beam. In order to study the vibration behaviors of the micro beam, the geometric parameters of the structure are taken as $l_{b} = l_{s} = 5{\ {mm}}$, $w_{b} = 0.4{\ {mm}}$, $t_{b} = 0.01{\ {mm}}$, $w_{s} = 1.5{\ {mm}}$, $t_{s} = 0.3{\ {mm}}$ and $d_{g} = 0.04{\ {mm}}$. The range of the initial displacement is less than half of the electrode thickness. 2.2. Fringing Electrostatic Force {#sec2dot2-micromachines-10-00324} --------------------------------- In this paper, the thickness of the electrode is much more than the thickness of the beam, as shown in [Figure 2](#micromachines-10-00324-f002){ref-type="fig"}. The distributed electrostatic force is $q_{e} = f_{e} \times v_{e}^{2}$, in which $f_{e}$ is electrostatic force on the beam per unit length per square voltage; $v_{e}\left( t \right)$ is a combined DC/AC voltage applied on the beam, i.e., $v_{e}\left( t \right) = v_{DC} + v_{AC}\cos\omega_{e}t$. When the cantilevered beam is in a position of the fringing electrostatic actuation structure, the voltage applied to the cantilevered beam is defined as $+ v_{e}$, the voltage applied to the electrode is defined as $- v_{e}$, and the capacitance between the cantilevered beam and electrode is $4C_{e}$. Then, the charge on the cantilevered beam is $+ 8C_{e}v_{e}$, The charge on the electrode is $- 8C_{e}v_{e}$. Assume that the charges on the cantilevered beam are evenly distributed at the two points, and the charges on the electrode are evenly distributed at the eight points, as shown in [Figure 2](#micromachines-10-00324-f002){ref-type="fig"}. Then, the charge of each point on the cantilevered beam is $q_{b} = + 4C_{e}v_{e}$, and the charge of each point on the electrode is $q_{s} = - C_{e}v_{e}$. According to Coulomb's law and the geometric position of the plate, the electrostatic force acting on the cantilevered beam in the direction of thickness is $$F_{e} = 2k_{e}q_{b}q_{s}\left( \begin{array}{l} {r_{11}{}^{- 2}\sin\theta_{11} + r_{21}{}^{- 2}\sin\theta_{21} + r_{12}{}^{- 2}\sin\theta_{12} + r_{22}{}^{- 2}\sin\theta_{22} +} \\ {r_{13}{}^{- 2}\sin\theta_{13} + r_{23}{}^{- 2}\sin\theta_{23} + r_{14}{}^{- 2}\sin\theta_{14} + r_{24}{}^{- 2}\sin\theta_{24}} \\ \end{array} \right)$$ See [Appendix A](#app1-micromachines-10-00324){ref-type="app"} for the symbolic meaning. In the expression of electrostatic force---the proportion of $r_{12}{}^{- 2}\sin\theta_{12}$ and $r_{22}{}^{- 2}\sin\theta_{22}$ are the largest. When the fringing electrostatic force near *d* = 0 is concerned, the Taylor expansion of the above approximate expression near *d* = 0 is $$F_{e} = 2k_{e}q_{b}q_{s}\left\{ {\left\lbrack {\frac{C_{1}}{C_{2}} + \frac{1}{3}\frac{t_{s}\left( {6t_{s}/C_{2}{}^{5/2} - 1.75t_{s}C_{1}/C_{2}} \right)}{C_{2}}} \right\rbrack d^{3} + \left\lbrack {\frac{2}{C_{2}{}^{3/2}} - \frac{1.5t_{s}{}^{2}}{C_{2}{}^{5/2}}} \right\rbrack d} \right\}$$ in which $C_{1} = 3.75t_{s}{}^{2}/C_{2}{}^{5/2} - 3/C_{2}{}^{3/2}$, $C_{2} = d_{g}{}^{2} + 0.25t_{s}{}^{2}$. From this Taylor expansion, we can see that the expression of fringing electrostatic force near *d* = 0 can be fitted by polynomials containing the first and third terms of the distance *d*. By using the finite element software, [Figure 3](#micromachines-10-00324-f003){ref-type="fig"} shows the equipotential lines when actuated voltage of 1 Volt is applied across the two electrodes and the beam for $d = 0{\ {mm}}$, $d = 0.02{\ {mm}}$, $d = 0.04{\ {mm}}$, $d = 0.06{\ {mm}}$, respectively. When $d = 0{\ {mm}}$, the equipotential lines are symmetrical around the beam, When the initial displacement increases, the equipotential lines become unsymmetrical, which leads to the appearance of the fringing electrostatic force. Under different electrode thickness, the values of fringing electrostatic force responding to initial displacement are calculated. In [Figure 4](#micromachines-10-00324-f004){ref-type="fig"}, the data points and fitted curves of fringing electrostatic force per unit length are recorded. When the thickness of the electrode is much greater than the thickness of the beam, the electrostatic force increases monotonically and nonlinearly. Since the thickness of the electrode dominates the deflection range of the beam, the thickness of the electrode should not be too small. Thus, the fit function of the fringing electrostatic force responding to initial displacement should consist of linear term and nonlinear term in this case. The polynomial fit function which includes linear and cubic parameters is proposed, and the fringing electrostatic force is approximated as $f_{e} = r_{1}d + r_{3}d^{3}$, in which $r_{1}$ and $r_{3}$ are fitting parameters. The assumption of the fringing electrostatic force can reflect the trend of electrostatic force change, and this assumption is convenient for analysis. The beam vibrates around the initial displacement *d*, the fringing electrostatic force responding to the vibration amplitude $u$ can be written as $f_{e} = r_{1}\left( {d + u} \right) + r_{3}\left( {d + u} \right)^{3} = e_{p}\left( {1 + e_{p1}u + e_{p2}u^{2} + e_{p3}u^{3}} \right)$, in which $e_{p} = r_{1}d + r_{3}d^{3}$, $e_{p1} = \left( {r_{1} + 3r_{3}d^{2}} \right)/\ e_{p}$, $e_{p2} = 3r_{3}d/\ e_{p}$, $e_{p3} = r_{3}/\ e_{p}$. The fit function of electrostatic force in this paper is different from that in reference \[[@B26-micromachines-10-00324]\]. [Figure 5](#micromachines-10-00324-f005){ref-type="fig"} shows the fitted curves of electrostatic force based on different fit function when $t_{s} = 0.3{\ {mm}}$. In this figure, FC. 0 is the fitted curves which based on this paper, FC. 1 and FC. 2 are the fitted curves which based on reference \[[@B26-micromachines-10-00324]\]. As shown in [Figure 5](#micromachines-10-00324-f005){ref-type="fig"}, FC. 1 and FC. 2 cannot fit the data when ${d\ = \ 0\ {mm}\ –\ 0.30}{\ {mm}}$. The beam vibrates in the range of the thickness of the electrode, i.e., ${d\ = \ 0\ {mm}\ –\ 0.15}{\ {mm}}$, which is what we care about in this paper. FC. 0 can fit the data when ${d\ = \ 0\ {mm}\ –\ 0.15}{\ {mm}}$. It means that the fit functions, which based on reference \[[@B26-micromachines-10-00324]\], cannot fit the electrostatic force in this case. The fit function in this paper can reflect the change of the electrostatic force, when the vibration amplitude of the cantilevered beam is less than the thickness of the electrode. 2.3. Dynamic Modeling {#sec2dot3-micromachines-10-00324} --------------------- The influences of the initial displacement change in the beam's dynamic behaviors are focused in this paper. By using the elastic beam theory, the equation of motion and the boundary conditions of the beam are written as \[[@B26-micromachines-10-00324]\] $$EI\frac{\partial^{4}u(x,t)}{\partial x^{4}} + \rho w_{b}t_{b}\frac{\partial^{2}u(x,t)}{\partial t^{2}} = q_{e} + q_{a}$$ $${u\left( {0,t} \right) = \frac{\partial u(0,t)}{\partial x} = 0}{,\ \frac{\partial^{2}u(l_{b},t)}{\partial x^{2}} = \frac{\partial^{3}u(l_{b},t)}{\partial x^{3}} = 0}$$ where $u\left( {x,t} \right)$ is the deflection of the beam in the thickness direction; *E* is the Young's modulus; $I = w_{b}t_{b}^{3}/12$ is the inertial moment of the beam cross section; $\rho$ is the mass density of the beam. The material parameters of the beam, which is made of brass, are taken as $E = 108{\ {GPa}}$ and $\rho = 8500{\ {kg}}/m^{3}$. Actuation force $q_{e}$ and damping force $q_{a}$ represent the fringing electrostatic force and the aerodynamic force per unit length respectively. The distributed aerodynamic force $q_{a}$ is \[[@B29-micromachines-10-00324],[@B30-micromachines-10-00324]\] $\left| q_{a} \right| = 0.5\rho_{a}w_{b}c_{a}\left( {\partial u/\partial t} \right)^{2}$, in which the direction of the aerodynamic force is opposite to the direction of the velocity; $c_{a}$ is the drag coefficient; $\rho_{a}$ is the density of the air. For analytical convenience, we obtain the following non-dimensional equation of motion and the boundary conditions. $$\frac{\partial^{4}U}{\partial X^{4}} + \frac{\partial^{2}U}{\partial T^{2}} = \mathsf{Ε}_{0}\left( {1 + \mathsf{Ε}_{1}U + \mathsf{Ε}_{2}U^{2} + \mathsf{Ε}_{3}U^{3}} \right)\left( {1 + V_{AC}\cos W_{e}T} \right)^{2} - \mathsf{Α}\frac{\partial U}{\partial T}\left| \frac{\partial U}{\partial T} \right|$$ $${U\left( {0,T} \right) = \frac{\partial U(0,T)}{\partial X} = 0}{,\frac{\partial^{2}U(1,T)}{\partial X^{2}} = \frac{\partial^{3}U(1,T)}{\partial X^{3}} = 0}$$ See [Appendix B](#app2-micromachines-10-00324){ref-type="app"} for the symbolic meaning. We use the Galerkin discretization method to transform the partial differential equation to the ordinary differential equations \[[@B31-micromachines-10-00324]\]. The model in this paper is based on the fundamental frequency vibration. The steady-state solution of the non-dimensional governing equation is written by $U\left( {X,T} \right) = \mathsf{\Phi}\left( X \right)\mathsf{\Theta}\left( T \right)$, and the mode function $\mathsf{\Phi}\left( X \right)$ for cantilevered beam is $\mathsf{\Phi}\left( X \right) = {ch}\lambda_{r}X - cos\lambda_{r}X + \xi_{r}\left( {{sh}\lambda_{r}X - \sin\lambda_{r}X} \right)$, in which $\lambda_{r} = 1.875$, $\xi_{r} = - \left( {{ch}\lambda_{r} + \cos\lambda_{r}} \right)/\left( {{sh}\lambda_{r} + \sin\lambda_{r}} \right)$. Substituting $U\left( {X,T} \right) = \mathsf{\Phi}\left( X \right)\mathsf{\Theta}\left( T \right)$ into the governing equation. Then multiplying the outcome by the mode function, and integrating the resultant equation from X = 0 to 1. An ordinary differential equation with respect to time is obtained as $$\begin{array}{l} {\frac{\partial^{2}\mathsf{\Theta}}{\partial T^{2}} + \alpha_{k}\mathsf{\Theta} =} \\ {\alpha_{Ep}\left( {1 + \alpha_{ep1}\mathsf{\Theta} + \alpha_{ep2}\mathsf{\Theta}^{2} + \alpha_{ep3}\mathsf{\Theta}^{3}} \right)\left\lbrack {\left( {1 + \frac{V_{AC}{}^{2}}{2}} \right) + 2V_{AC}\cos W_{e}T + \frac{V_{AC}{}^{2}}{2}\cos 2W_{e}T} \right\rbrack - \alpha_{a}\frac{\partial\mathsf{\Theta}}{\partial T}\left| \frac{\partial\mathsf{\Theta}}{\partial T} \right|} \\ \end{array}$$ See [Appendix C](#app3-micromachines-10-00324){ref-type="app"} for the symbolic meaning. The deflection splits into a static deflection and a dynamic deflection, i.e., $\mathsf{\Theta}\left( T \right) = \mathsf{\Theta}_{0p} + \vartheta_{p}\left( T \right)$, then we obtain static equation and dynamic equation. $$S_{0} + S_{1}\mathsf{\Theta}_{0p} + S_{2}\mathsf{\Theta}_{0p}^{2} + S_{3}\mathsf{\Theta}_{0p}^{3} = 0$$ $$\vartheta_{p}^{''} + K_{1}\vartheta_{p} + K_{2}\vartheta_{p}{}^{2} + K_{3}\vartheta_{p}{}^{3} + \alpha_{a}\vartheta_{p}^{\prime}\left| \vartheta_{p}^{\prime} \right| = \left\lbrack \begin{array}{l} {\left( {F_{E1}\cos W_{e}T + F_{E2}\cos 2W_{e}T} \right) +} \\ {\left( {K_{E1P1}\vartheta_{p} + K_{E1P2}\vartheta_{p}{}^{2} + K_{E1P3}\vartheta_{p}{}^{3}} \right)\cos W_{e}T +} \\ {\left( {K_{E2P1}\vartheta_{p} + K_{E2P2}\vartheta_{p}{}^{2} + K_{E2P3}\vartheta_{p}{}^{3}} \right)\cos 2W_{e}T} \\ \end{array} \right\rbrack$$ where (′) denotes the derivate with respect to $T$. See [Appendix D](#app4-micromachines-10-00324){ref-type="app"} for the symbolic meaning. 3. Dynamic Analysis {#sec3-micromachines-10-00324} =================== The static equation is a one variable cubic equation, which can be solved by using the method of reference \[[@B32-micromachines-10-00324]\]. The nonlinear dynamic equation is solved by using the Method of Multiple Scales (MMS). Then the dynamic equation can be written by $$\vartheta_{p}^{''} + K_{1}\vartheta_{p} + \varepsilon K_{2}\vartheta_{p}{}^{2} + \varepsilon K_{3}\vartheta_{p}{}^{3} + \varepsilon\alpha_{a}\vartheta_{p}^{\prime}\left| \vartheta_{p}^{\prime} \right| = \varepsilon\left\lbrack \begin{array}{l} {\left( {F_{E1}\cos W_{e}T + F_{E2}\cos 2W_{e}T} \right) +} \\ {\left( {K_{E1P1}\vartheta_{p} + K_{E1P2}\vartheta_{p}{}^{2} + K_{E1P3}\vartheta_{p}{}^{3}} \right)\cos W_{e}T +} \\ {\left( {K_{E2P1}\vartheta_{p} + K_{E2P2}\vartheta_{p}{}^{2} + K_{E2P3}\vartheta_{p}{}^{3}} \right)\cos 2W_{e}T} \\ \end{array} \right\rbrack$$ where $\varepsilon$ is regarded as a small non-dimensional bookkeeping parameter, $\sigma$ is a detuning parameter, $W_{e}{}^{2} = K_{1}{+ \mathsf{\varepsilon}}\sigma$. Introduce $W_{0}^{2} = \alpha_{k}$, when the static deflection is small and $\mathsf{\sigma} = 0$, the resonance frequency ratio is $$W_{e}{}^{2}/W_{0}{}^{2} = {K_{1}/\alpha_{k}} = \mathsf{Κ}_{w0} - \mathsf{Κ}_{w2}d^{2}$$ in which the coefficients are $${\mathsf{Κ}_{w0} = 1 - r_{1}\frac{\left( {\int_{0}^{1}\mathsf{\Phi}dX} \right)^{2}}{\int_{0}^{1}\frac{\partial^{4}\mathsf{\Phi}}{\partial X^{4}}\mathsf{\Phi}dX}\frac{l_{b}{}^{4}v_{DC}^{2}}{EI}\left\lbrack {1 + \frac{1}{2}\left( \frac{v_{AC}}{v_{DC}} \right)^{2}} \right\rbrack}{,\ \mathsf{Κ}_{w2} = 3r_{3}\frac{\left( {\int_{0}^{1}\mathsf{\Phi}dX} \right)^{2}}{\int_{0}^{1}\frac{\partial^{4}\mathsf{\Phi}}{\partial X^{4}}\mathsf{\Phi}dX}\frac{l_{b}{}^{4}v_{DC}^{2}}{EI}\left\lbrack {1 + \frac{1}{2}\left( \frac{v_{AC}}{v_{DC}} \right)^{2}} \right\rbrack}$$ The primary resonance of nonlinear dynamic equation is analyzed, yields $$\begin{array}{l} {- W_{e}\alpha_{0}\beta_{0}{}^{\prime} = \frac{1}{2}\sigma\alpha_{0} - \frac{1}{4}K_{3}\alpha_{0}{}^{3} +} \\ {\left( {\frac{1}{2}F_{E1} + \frac{1}{4}K_{E1P2}\alpha_{0}{}^{2}} \right)\cos\beta_{0} + \left( {\frac{1}{4}K_{E2P1}\alpha_{0} + \frac{1}{4}K_{E2P3}\alpha_{0}{}^{3}} \right)\cos 2\beta_{0}} \\ \end{array}$$ $$- W_{e}\alpha_{0}{}^{\prime} = - \frac{W_{e}^{2}}{\pi}\alpha_{a}\alpha_{0}{}^{2} + \frac{1}{2}F_{E1}\sin\beta_{0} + \left( {\frac{1}{4}K_{E2P1}\alpha_{0} + \frac{1}{8}K_{E2P3}\alpha_{0}{}^{3}} \right)\sin 2\beta_{0}$$ where $\alpha_{0}$ is the amplitude of $\vartheta_{p}$, $\beta_{0}$ is the phase difference with $W_{e}$, (′) denotes the derivate with respect to $T_{1} = \varepsilon T$. The complete proof is given in [Appendix E](#app5-micromachines-10-00324){ref-type="app"}. The steady-state periodic motion corresponds to the solution of the system of equations, by conditions $\alpha_{0}^{\prime} = 0$ and $\beta_{0}^{\prime} = 0$. Finally, the frequency response equation of the primary resonance can be derived as $$\left\lbrack \frac{- P_{C1} - \sqrt{P_{C1}{}^{2} - 8P_{C2}\left( {P_{C0} - P_{C2}} \right)}}{4P_{C2}} \right\rbrack^{2} + \left\lbrack \frac{- 4P_{C2}P_{S0}}{4P_{C2}P_{S1} - 2P_{C1}P_{S2} - 2P_{S2}\sqrt{P_{C1}{}^{2} - 8P_{C2}\left( {P_{C0} - P_{C2}} \right)}} \right\rbrack^{2}{= 1}$$ in which the coefficients are$P_{C0} = \frac{1}{2}\sigma\alpha_{0} - \frac{1}{4}K_{3}\alpha_{0}{}^{3}$, $P_{C1} = \frac{1}{2}F_{E1} + \frac{1}{4}K_{E1P2}\alpha_{0}{}^{2}$, $P_{C2} = \frac{1}{4}K_{E2P1}\alpha_{0} + \frac{1}{4}K_{E2P3}\alpha_{0}{}^{3}$,$P_{S0} = - \frac{W_{e}^{2}}{\pi}\alpha_{a}\alpha_{0}{}^{2}$, $P_{S1} = \frac{1}{2}F_{E1}$, $P_{S2} = \frac{1}{4}K_{E2P1}\alpha_{0} + \frac{1}{8}K_{E2P3}\alpha_{0}{}^{3}$. The relationship between the electrostatic force and the initial displacement is nonlinear. It can reveal nonlinear dynamic behaviors of the vibration beam. When $v_{DC} = 30{\ V}$ and $v_{AC} = 0.5{\ V}$, based on the frequency response equation, the frequency response curve of the primary resonance is shown in [Figure 6](#micromachines-10-00324-f006){ref-type="fig"}. It exhibits linear behavior. Based on Equation (11) and setting σ = 0, i.e., when the vibration beam resonates, the relationship between the square of resonance frequency and that of initial displacement is linear, as shown in [Figure 7](#micromachines-10-00324-f007){ref-type="fig"}. From [Figure 7](#micromachines-10-00324-f007){ref-type="fig"}, the resonance frequency rises, with the increase of the initial displacement and the decrease of the slit gap. In the same range of the initial displacement, a smaller slit gap makes marked change of the resonance frequency. So, the initial displacement can be obtained from resonance frequency measurement. What's more, decreasing the slit gap can enhance the sensitivity of the sensor to the initial displacement. The vibration peak value can be found by the frequency response equation, [Figure 8](#micromachines-10-00324-f008){ref-type="fig"} shows the effects of initial displacement and slit gap on the vibration amplitude of the primary resonance. Increase of the vibration amplitude is linear, with the increase of the initial displacement. Linearity is an important parameter for sensors because linearity indicates a directly proportional relationship between output and input signals of a sensing system \[[@B1-micromachines-10-00324]\]. So, it is available that the initial displacement is obtained from vibration amplitude measurement of the primary resonance. When $v_{DC} = 300{\ V}$ and $v_{AC} = 5{\ V}$, the nonlinear dynamic equation Equation (9) is solved by using the MMS and the Fourth-Order Runge--Kutta Method (RK4) \[[@B33-micromachines-10-00324]\], and the effect of the initial displacement on the vibration amplitude is shown in [Figure 9](#micromachines-10-00324-f009){ref-type="fig"}. The results of the MMS and the RK4 are in a good agreement when the amplitude is small. The error between the results of the MMS and the RK4 increases, when the vibration amplitude increases. [Figure 9](#micromachines-10-00324-f009){ref-type="fig"} proves the results of the MMS verified. As shown in [Figure 9](#micromachines-10-00324-f009){ref-type="fig"}, when *d* = 0.03 mm, there are two values of the vibration amplitude when *W~e~ / W~0~* = 1.0750 and 1.0775. The jump phenomenon has been found, when the initial displacement increases, the jump frequency increases. Primary resonance's vibration amplitude versus initial displacement under different excitation frequency ratio is shown in [Figure 10](#micromachines-10-00324-f010){ref-type="fig"}. In [Figure 10](#micromachines-10-00324-f010){ref-type="fig"}, the frequency ratio is 1.0750, 1.0775 and 1.0800 respectively. Within the range of initial displacement \[0.01 mm, 0.05 mm\], the amplitude is calculated once every 0.002 mm interval, and the bifurcation diagram of amplitude with respect to initial displacement is drawn. It can be clearly observed that the number of equilibrium points of parameter *u* changes with the change of parameter *d*. For the case of frequency ratio 1.0750, parameter *u* has three equilibrium points when parameter *d* = 0.03; for the case of frequency ratio 1.0775, parameter *u* has three equilibrium points when parameter *d* = 0.032 and 0.034; for the case of frequency ratio 1.0800, parameter *u* has three equilibrium points when parameter *d* = 0.034 and 0.036. With the change of parameter *d*, there is a jump phenomenon in parameter *u*. When the jump phenomenon occurs, the corresponding excitation frequency is the resonance frequency of the vibration system. By comparing the frequencies of the three equilibrium points, it is found that with the increase of initial displacement, the smaller excitation frequencies first appear three equilibrium points. Because there is a corresponding relationship between the excitation frequency ratio and the initial displacement of the cantilever beam when there are three equilibrium points in the vibration system. Therefore, the initial displacement of the cantilever beam can be described by measuring the excitation frequency of the three equilibrium points. So, jump phenomenon can be used to locate the demand initial displacement. The excitation frequency can be changed to adjust the change of the demand initial displacement. When *d* = 0.03 mm and $W_{e}/W_{0}$ = 1.075, phase trajectory is drawn in [Figure 11](#micromachines-10-00324-f011){ref-type="fig"}. It can be seen from the figure that the amplitude of the cantilever beam is related to its initial state, that is, when the initial state energy is high, the amplitude of the cantilever beam corresponds to the higher equilibrium point S1; when the initial state energy is low, the amplitude of the cantilever beam corresponds to the lower equilibrium point S3. The equilibrium point S2 is unstable saddle. The jumping amplitude change can be detected more easily and more quickly. In the following, the jump phenomenon is studied, based on the frequency response equation. And the impact of the different parameters, which include the material parameters, the beam length and the actuated voltage, on the nonlinear dynamic characteristic is presented. The effects of the initial displacement and the material parameters on the frequency response curve of the primary resonance are shown in [Figure 12](#micromachines-10-00324-f012){ref-type="fig"}. The frequency ratio $W_{e}/W_{0}$ of the vibration peak value of the aluminum beam, whose Young's modulus is the least, is the biggest. That because that when the Young's modulus decreases, the value of the parameter $E_{0}$ increases, which leads the increase of the frequency ratio $W_{e}/W_{0}$. It means that the fringing electrostatic force has a largest impact on the resonance frequency of the aluminum beam. As show in this figure, the amplitude $u$ of the brass beam, whose density is the largest, is the biggest. The nonlinear response of the brass beam is obvious. That because that when the density increases, the value of the parameter $\alpha_{a}$ decreases, which leads the decrease of the damping and the increase of the vibration amplitude. To increase the vibration amplitude, we use the brass beam. When $l_{s} = 6{\ {mm}}$, the effects of the initial displacement on the frequency response curve of the primary resonance are shown in [Figure 13](#micromachines-10-00324-f013){ref-type="fig"}. Compared with [Figure 9](#micromachines-10-00324-f009){ref-type="fig"}, the vibration amplitude $u$ and frequency ratio $W_{e}/W_{0}$ of the vibration peak value are larger, the nonlinear behavior is more obvious. Meanwhile, with the increase of the initial displacement, the increase of the jump frequency changes is more obviously. It means that the fringing electrostatic force has a larger impact on the frequency response, when the length of beam is larger. When $v_{DC} = 350{\ V}$ $v_{AC} = 5{\ V}$ and $v_{DC} = 300{\ V}$ $v_{AC} = 10{\ V}$, [Figure 14](#micromachines-10-00324-f014){ref-type="fig"} shows the frequency response curve of the primary resonance under different initial displacement. Compared with [Figure 9](#micromachines-10-00324-f009){ref-type="fig"}, when DC/AC voltage are big enough, the electrostatic force can lead to obvious nonlinear vibration. Furthermore, as the increase of the actuated voltage, the nonlinear vibration strengthens. When a larger amplitude is expected, the actuated voltage is always set big enough, and the nonlinear vibration must be considered. 4. Experimental Setup and Results {#sec4-micromachines-10-00324} ================================= An experiment is designed to observe the dynamic analysis of this structure. The experimental setup for the dynamic tests consists of excitation powers (high voltage power and waveform generation), mechanical parts (cantilevered beam and electrode) and detection parts (laser displacement sensor and oscilloscope). The schematic of experimental setup is depicted in [Figure 15](#micromachines-10-00324-f015){ref-type="fig"}. Because the experimental conditions are limited, the geometric parameters of the structure are magnified, which are taken as $l_{b} = 50{\ {mm}}$, $w_{b} = 4{\ {mm}}$, $t_{b} = 0.1{\ {mm}}$, $l_{s} = 50{\ {mm}}$, $w_{s} = 15{\ {mm}}$, $t_{s} = 3{\ {mm}}$, $d_{g} = 0.5{\ {mm}}$ and $v_{DC} = 300{\ V}$, $v_{AC} = 5{\ V}$. The cantilevered beam is directly actuated by a periodic wave produced from a waveform generator, while DC voltage produced by high voltage power is applied to the cantilevered beam at the same time. The electrode is connected with ground. In this way, both DC voltage and AC voltage are applied between the cantilevered beam and electrode. The moving platform can change the initial displacement between the cantilevered beam and electrode. The vibration of the cantilevered beam is more obvious, when the excitation frequency is close to the resonance frequency. And the vibration amplitude is the largest, when the excitation frequency equals to the resonance frequency. The laser displacement sensor is used to detect the vibration amplitude, by transforming the amplitude change to voltage change. Then this voltage change signal is given to the oscilloscope. The oscilloscope is used to display and record the data of both frequency and amplitude. It means that the resonance frequency change can be detected by using both the laser displacement sensor and the oscilloscope. When $v_{DC} = 300{\ V}$ and $v_{AC} = 5{\ V}$, [Figure 16](#micromachines-10-00324-f016){ref-type="fig"} shows that the resonance frequency rises with the increase in initial displacement, in which *d* = 0 mm -- 0.8 mm. As the same as the theory analysis, it is expected that relationship between the input (the square of initial displacement) and the output (the square of resonance frequency) is linear, when the beam is far from the middle and the end of the electrode in the thickness direction. In this range, the initial displacement can be obtained from resonance frequency measurement. The vibration amplitude of the cantilevered beam is relatively small, when the cantilevered beam is near the middle of the electrode in the thickness direction. It is not easy to find the resonance frequency. So, the error is large near this point in [Figure 16](#micromachines-10-00324-f016){ref-type="fig"}. When the beam is near the end of the electrode in the thickness direction, the frequency response curves are shown in [Figure 17](#micromachines-10-00324-f017){ref-type="fig"}. When $v_{DC} = 100{\ V\ {and}\ }200{\ V}$, the frequency response curve is linear., while the amplitude of the beam is relatively small. But when $v_{DC} = 300{\ V}$, the frequency response curve is nonlinear, while the nonlinearity of the system is of the softening type, as shown in \[[@B26-micromachines-10-00324]\]. If the amplitude of the beam is large enough, the electrostatic force does not agree with linear and cubic fit function at all, when the beam is near the end of the electrode in the thickness direction. So, it causes an error near the end of the electrode in [Figure 16](#micromachines-10-00324-f016){ref-type="fig"}. When the excitation frequency equals to the resonance frequency, the vibration amplitude is the largest. The largest vibration amplitude values responding to different initial displacement are recorded by the test. When the initial displacement is 0.4 mm to 0.8 mm, the vibration amplitudes which are given by the MMS and the TEST are shown in [Figure 18](#micromachines-10-00324-f018){ref-type="fig"}. When the initial displacement is 0.4 mm to 0.8 mm, the increase in the vibration amplitude is linear, and the initial displacement can be obtained from amplitude measurement. As show in [Figure 18](#micromachines-10-00324-f018){ref-type="fig"}, the result of the MMS agrees well with the experimental result. Base on [Figure 8](#micromachines-10-00324-f008){ref-type="fig"}, the measurement error of slit gap leads the error between the results of the MMS and the TEST in [Figure 18](#micromachines-10-00324-f018){ref-type="fig"}. In the above test, the hardening effect is not obvious, since the cubic fitting parameter is relatively small. [Figure 13](#micromachines-10-00324-f013){ref-type="fig"} shows that the fringing electrostatic force has a larger impact on the frequency response, when the length of beam is larger. To increase cubic fitting parameter, the slit gap is reduced to 0.3 mm, and another beam with $l_{b} = 100{\ {mm}}$ is used. The frequency response curve exhibits hardening behavior, and there is a jump in 5.140 Hz, as shown in [Figure 19](#micromachines-10-00324-f019){ref-type="fig"}. 5. Conclusions {#sec5-micromachines-10-00324} ============== In this paper, a new fringing electrostatic actuation mode is developed. Some obvious advantages of this new actuation mode are as following: a large deflection can be obtained without the limitation of the proximity of the electrodes; this structure which is compatible with the circuit can be designed to smaller scale; a smaller scale leads to better sensitivity. Through the combination of theoretical modeling, analytical calculation, numerical verification and experimental research, the mechanism of fringing electrostatic force and the complex response law of fringing electrostatic actuation vibration system are revealed, the relationship between system parameters and vibration response is analyzed, the application scheme of this actuation mode is put forward. The fringing electrostatic force and the dynamic investigations into the micro cantilevered beam actuated by fringing electrostatic force are of great concern. Through analysis, the expression of fringing electrostatic force is found; the effects of the some parameters on the dynamic behaviors are investigated. Results shows that the fringing electrostatic force is nonlinear, which leads to nonlinear vibration of the micro-cantilevered beam; the resonance frequency rises with the increase of the initial displacement and the decrease of the slit gap; in the same range of the initial displacement, a smaller slit gap makes marked change of the resonance frequency; with the increase of the initial displacement, the increase of the vibration amplitude is linear; when the initial displacement increases, the jump frequency increases; the fringing electrostatic force has a larger impact on the frequency response, when the length of beam is larger; as the increase of the actuated voltage, the nonlinear vibration strengthens. Moreover, that are the influences of initial displacement change in the dynamic behaviors of the micro cantilevered beam that helps us to design a new micro tactile sensor. This sensor can measure the pressure based on the initial displacement change. The initial displacement can be derived by measuring resonance frequency and vibration amplitude of the micro cantilevered beam. And the jump phenomenon can be used to locate the initial displacement. Z.W. and Q.Z. conceived and designed the model; Z.W. and W.W. contributed theoretical analysis; Z.W. and J.H. analyzed the data; Z.W. wrote the paper. This work is supported by the National Natural Science Foundation of China (Grant Nos. 11772218, 11872044 and 11702192). The authors declare no conflict of interest. $r_{11}{}^{2} = \left( {w_{s} + d_{g}} \right)^{2} + \left( {d - 0.5t_{s}} \right)^{2}$, $r_{21}{}^{2} = \left( {w_{s} + d_{g}} \right)^{2} + \left( {d + 0.5t_{s}} \right)^{2}$, $r_{12}{}^{2} = d_{g}{}^{2} + \left( {d - 0.5t_{s}} \right)^{2}$, $r_{22}{}^{2} = d_{g}{}^{2} + \left( {d + 0.5t_{s}} \right)^{2}$, $r_{13}{}^{2} = \left( {w_{b} + d_{g}} \right)^{2} + \left( {d - 0.5t_{s}} \right)^{2}$, $r_{23}{}^{2} = \left( {w_{b} + d_{g}} \right)^{2} + \left( {d + 0.5t_{s}} \right)^{2}$, $r_{14}{}^{2} = \left( {w_{s} + w_{b} + d_{g}} \right)^{2} + \left( {d - 0.5t_{s}} \right)^{2}$, $r_{24}{}^{2} = \left( {w_{s} + w_{b} + d_{g}} \right)^{2} + \left( {d + 0.5t_{s}} \right)^{2}$, $\sin\theta_{11} = \left( {d - 0.5t_{s}} \right)/r_{11}$, $\sin\theta_{21} = \left( {d + 0.5t_{s}} \right)/r_{21}$, $\sin\theta_{12} = \left( {d - 0.5t_{s}} \right)/r_{12}$, $\sin\theta_{22} = \left( {d + 0.5t_{s}} \right)/r_{22}$, $\sin\theta_{13} = \left( {d - 0.5t_{s}} \right)/r_{13}$, $\sin\theta_{23} = \left( {d + 0.5t_{s}} \right)/r_{23}$, $\sin\theta_{14} = \left( {d - 0.5t_{s}} \right)/r_{14}$, $\sin\theta_{24} = \left( {d + 0.5t_{s}} \right)/r_{24}$. $U = u/\left( {0.5t_{s} - d} \right)\ $, $X = x/l_{b}$, $T = \omega t$, $\omega = \sqrt{EI/\left( {\rho w_{b}t_{b}l_{b}^{4}} \right)}$, $W_{e} = \omega_{e}/\omega$, $V_{AC} = v_{AC}/v_{DC}$, $\mathsf{Ε}_{0} = l_{b}^{4}e_{p}v_{DC}^{2}/\left\lbrack {EI\left( {0.5t_{s} - d} \right)} \right\rbrack$, $\mathsf{Ε}_{1} = e_{p1}\left( {0.5t_{s} - d} \right)$, $\mathsf{Ε}_{2} = e_{p2}\left( {0.5t_{s} - d} \right)^{2}$, $\mathsf{Ε}_{3} = e_{p3}\left( {0.5t_{s} - d} \right)^{3}$, $\mathsf{Α} = \rho_{a}c_{a}\left( {0.5t_{s} - d} \right)/\left( {2\rho t_{b}} \right)$. $\alpha_{k} = \int_{0}^{1}\frac{\partial^{4}\mathsf{\Phi}}{\partial X^{4}}{\mathsf{\Phi}d}X$, $\alpha_{Ep} = \mathsf{Ε}_{0}\int_{0}^{1}{\mathsf{\Phi}d}X$, $\alpha_{ep1} = \mathsf{Ε}_{1}\int_{0}^{1}{\mathsf{\Phi}d}X$, $\alpha_{ep2} = \mathsf{Ε}_{2}\int_{0}^{1}\mathsf{\Phi}^{2}dX$, $\alpha_{ep3} = \mathsf{Ε}_{3}\int_{0}^{1}\mathsf{\Phi}^{3}dX$, $\alpha_{a} = \mathsf{Α}\int_{0}^{1}\mathsf{\Phi}^{3}dX$. $S_{0} = \left( {1 + \frac{V_{AC}^{2}}{2}} \right)\alpha_{Ep}$, $S_{1} = \left( {1 + \frac{V_{AC}^{2}}{2}} \right)\alpha_{Ep}\alpha_{ep1} - \alpha_{k}$, $S_{2} = \left( {1 + \frac{V_{AC}^{2}}{2}} \right)\alpha_{Ep}\alpha_{ep2}$, $S_{3} = \left( {1 + \frac{V_{AC}^{2}}{2}} \right)\alpha_{Ep}\alpha_{ep3}$, $C_{0} = 1 + \alpha_{ep1}\mathsf{\Theta}_{0p} + \alpha_{ep2}\mathsf{\Theta}_{0p}^{2} + \alpha_{ep3}\mathsf{\Theta}_{0p}^{3}$, $C_{1} = \alpha_{ep1} + 2\alpha_{ep2}\mathsf{\Theta}_{0p} + 3\alpha_{ep3}\mathsf{\Theta}_{0p}^{2}$, $C_{2} = \alpha_{ep2} + 3\alpha_{ep3}\mathsf{\Theta}_{0p}$, $C_{3} = \alpha_{ep3}$, $K_{1} = \alpha_{k} - \alpha_{Ep}\left( {1 + \frac{V_{AC}^{2}}{2}} \right)C_{1}$, $K_{2} = - \alpha_{Ep}\left( {1 + \frac{V_{AC}^{2}}{2}} \right)C_{2}$, $K_{3} = - \alpha_{Ep}\left( {1 + \frac{V_{AC}^{2}}{2}} \right)C_{3}$, $F_{E1} = 2V_{AC}\alpha_{Ep}C_{0}$, $F_{E2} = 0.5V_{AC}^{2}\alpha_{Ep}C_{0}$, $K_{E1P1} = 2V_{AC}\alpha_{Ep}C_{1}$, $K_{E1P2} = 2V_{AC}\alpha_{Ep}C_{2}$, $K_{E1P3} = 2V_{AC}\alpha_{Ep}C_{3}$, $K_{E2P1} = 0.5V_{AC}^{2}\alpha_{Ep}C_{1}$, $K_{E2P2} = 0.5V_{AC}^{2}\alpha_{Ep}C_{2}$, $K_{E2P3} = 0.5V_{AC}^{2}\alpha_{Ep}C_{3}$. The nonlinear dynamic equation is solved using the MMS. The solution can be represented by an expansion having the form as $$\vartheta_{p} = \vartheta_{0}(T_{0},T_{1}) + \varepsilon\vartheta_{1}(T_{0},T_{1})$$ where, $T_{n} = \varepsilon^{n}T$. A detuning parameter σ is introduced and defined by $K_{1} = W_{e}^{2} - \varepsilon\sigma$, then equating coefficients of like powers of $\varepsilon$, yield $${\mathsf{Ο}(\varepsilon^{0}):}{D_{0}^{2}\vartheta_{0} + W_{e}{}^{2}\vartheta_{0} = 0}$$ $${\mathsf{Ο}(\varepsilon^{1}):}\begin{array}{l} {D_{0}^{2}\vartheta_{1} + 2D_{0}D_{1}\vartheta_{0} - \sigma\vartheta_{0} + W_{e}{}^{2}\vartheta_{1} + K_{2}\vartheta_{0}^{2} + K_{3}\vartheta_{0}^{3} + \alpha_{a}D_{0}\vartheta_{0}\left| {D_{0}\vartheta_{0}} \right| =} \\ {F_{E1}\cos W_{e}T_{0} + \left( {K_{E1P1}\vartheta_{0} + K_{E1P2}\vartheta_{0}^{2} + K_{E1P3}\vartheta_{0}^{3}} \right)\cos W_{e}T_{0} +} \\ {F_{E2}\cos 2W_{e}T_{0} + \left( {K_{E2P1}\vartheta_{0} + K_{E2P2}\vartheta_{0}^{2} + K_{E2P3}\vartheta_{0}^{3}} \right)\cos 2W_{e}T_{0}} \\ \end{array}$$ when $D_{n} = \frac{\partial}{\partial T_{n}}$. The general solution of Equation $O\left( \varepsilon^{0} \right)$ can be written as $$\vartheta_{0} = \alpha\left( T_{1} \right)\exp\left( {iW_{e}T_{0}} \right) + {cc}$$ where cc represents the complex conjugate terms. Substituting the solution of Equation $O\left( \varepsilon^{0} \right)$ into Equation $O\left( \varepsilon^{1} \right)$, yields $${\mathsf{Ο}(\varepsilon^{1}):}\begin{array}{l} {D_{0}^{2}\vartheta_{1} + W_{e}{}^{2}\vartheta_{1} = \sigma\alpha e^{iW_{e}T_{0}} - 2iW_{e}D_{1}\alpha e^{iW_{e}T_{0}}} \\ {- K_{2}\left( {\alpha e^{iW_{e}T_{0}} + \overline{\alpha}e^{- iW_{e}T_{0}}} \right)\alpha e^{iW_{e}T_{0}} - K_{3}\left( {\alpha e^{iW_{e}T_{0}} + \overline{\alpha}e^{- iW_{e}T_{0}}} \right)^{2}\alpha e^{iW_{e}T_{0}}} \\ {- \alpha_{a}\left( {iW_{e}\alpha e^{iW_{e}T_{0}}} \right)\left| {iW_{e}\alpha e^{iW_{e}T_{0}} - iW_{e}\overline{\alpha}e^{- iW_{e}T_{0}}} \right|} \\ {+ \frac{1}{2}F_{E1}e^{iW_{e}T_{0}} + \left\lbrack \begin{array}{l} {K_{E1P1}\alpha e^{iW_{e}T_{0}} + K_{E1P2}\left( {\alpha e^{iW_{e}T_{0}} + \overline{\alpha}e^{- iW_{e}T_{0}}} \right)\alpha e^{iW_{e}T_{0}}} \\ {+ K_{E1P3}\left( {\alpha e^{iW_{e}T_{0}} + \overline{\alpha}e^{- iW_{e}T_{0}}} \right)^{2}\alpha e^{iW_{e}T_{0}}} \\ \end{array} \right\rbrack\frac{e^{iW_{e}T_{0}} + e^{- iW_{e}T_{0}}}{2}} \\ {+ \frac{1}{2}F_{E2}e^{2iW_{e}T_{0}} + \left\lbrack \begin{array}{l} {K_{E2P1}\alpha e^{iW_{e}T_{0}} + K_{E2P2}\left( {\alpha e^{iW_{e}T_{0}} + \overline{\alpha}e^{- iW_{e}T_{0}}} \right)\alpha e^{iW_{e}T_{0}}} \\ {+ K_{E2P3}\left( {\alpha e^{iW_{e}T_{0}} + \overline{\alpha}e^{- iW_{e}T_{0}}} \right)^{2}\alpha e^{iW_{e}T_{0}}} \\ \end{array} \right\rbrack\frac{e^{2iW_{e}T_{0}} + e^{- 2iW_{e}T_{0}}}{2}} \\ {+ {cc}} \\ \end{array}$$ At this point, it is convenient to express $\mathsf{\alpha}$ in the polar form $$\alpha = \frac{1}{2}\alpha_{0}e^{i\beta_{0}}$$ where $\alpha_{0}$ and $\beta_{0}$ are real functions of $T_{1}$. In order to eliminate the secular term, one needs $$\begin{array}{l} {\sigma\alpha - 2iW_{e}D_{1}\alpha - 2K_{3}\alpha^{2}\overline{\alpha} + \frac{1}{2}F_{E1} + \frac{1}{2}K_{E1P2}\left( {\alpha^{2} + \alpha\overline{\alpha}} \right) +} \\ {\left\lbrack {\frac{1}{2}K_{E2P1}\overline{\alpha} + \frac{1}{2}K_{E2P3}\left( {\alpha^{3} + 3\alpha{\overline{\alpha}}^{2}} \right)} \right\rbrack +} \\ {- \frac{W_{e}}{2\pi}{\int_{0}^{2\pi/W_{e}}{\alpha_{a}\left( {iW_{e}\alpha e^{iW_{e}T_{0}}} \right)\left| {iW_{e}\alpha e^{iW_{e}T_{0}} - iW_{e}\overline{\alpha}e^{- iW_{e}T_{0}}} \right|e^{- iW_{e}T_{0}}dT_{0}}} = 0} \\ \end{array}$$ Substituting $\mathsf{\alpha}$ into the secular term and separating the result into real and imaginary parts, we obtain Equation (12) and Equation (13). {#micromachines-10-00324-f001} {#micromachines-10-00324-f002} {#micromachines-10-00324-f003} {#micromachines-10-00324-f004} {#micromachines-10-00324-f005} {#micromachines-10-00324-f006} {#micromachines-10-00324-f007} {#micromachines-10-00324-f008} {#micromachines-10-00324-f009} {#micromachines-10-00324-f010} {#micromachines-10-00324-f011} {#micromachines-10-00324-f012} {#micromachines-10-00324-f013} {#micromachines-10-00324-f014} {#micromachines-10-00324-f015} {#micromachines-10-00324-f016} {#micromachines-10-00324-f017} {#micromachines-10-00324-f018} {#micromachines-10-00324-f019}

C-type natriuretic peptide (CNP), the third discovered member of the natriuretic peptide (NP) family, received its name because it shares a common amino acid core structure with the previously isolated cardiac hormones atrial and B-type NPs (ANP and BNP) (reviewed in ([@CIT0001])). However, human and murine genetic studies later revealed that CNP's major physiological function is not the regulation of renal natriuresis. Instead, CNP is a critical regulatory hormone in the bone, where it stimulates physiological endochondral bone growth by auto/paracrine activation of its specific cyclic GMP-forming natriuretic peptide receptor-B (NPR-B, also named guanylyl cyclase-B) in chondrocytes ([@CIT0001]). In addition, local CNP/NPR-B signalling participates in oocyte maturation, sensory axon bifurcation, and moderation of arterial blood pressure ([@CIT0001], [@CIT0004], [@CIT0005]). NPR-B presents as a homodimer, and each subunit contains an extracellular CNP-binding domain, a short transmembrane region and an intracellular region with a short submembrane segment followed by a kinase homology regulatory domain, a coiled-coil dimerization region and a C-terminal guanylyl cyclase domain ([@CIT0001], [@CIT0006]). After extracellular binding of CNP, the intracellular guanylyl cyclase domain catalyses the conversion of guanosine triphosphate to cyclic guanosine monophosphate (cGMP) as a second messenger. In chondrocytes, this fosters hypertrophy, differentiation, and extracellular matrix deposition necessary for longitudinal bone growth ([@CIT0002], [@CIT0003]). The kinase-like domain does not exert kinase activity; however, phosphorylation of specific residues within this region is necessary for CNP-dependent stimulation of cGMP synthesis ([@CIT0006]). The essential role of the chondrocyte CNP/NPR-B/cGMP signalling pathway during long bone growth has been demonstrated in many functional and genetic studies. In mice, global or chondrocyte-restricted CNP (gene name *Nppc*) or NPR-B (*Npr2*) deletions provoke severe dwarfism ([@CIT0003], [@CIT0007]). In human, homozygous or compound heterozygous loss-of-function variants of NPR-B (MIM \#108961) lead to extreme short stature, with shortening of the limbs, severe brachydactyly of the hands and feet, and skeletal dysplasia, a condition named acromesomelic dysplasia, type Maroteaux (AMDM, MIM \#602875) ([@CIT0008]). Consistently, heterozygous loss-of-function NPR-B variants are associated with mildly reduced stature (MIM \#616255) ([@CIT0009], [@CIT0010]). Conversely, gain-of-function NPR-B variants lead to tall stature, sometimes associated with a Marfanoid habitus, arachnodactyly, and scoliosis, a condition named epiphyseal chondrodysplasia, Miura type (MIM \#615923) ([@CIT0011]). Three such intracellular activating variants of NPR-B have been identified in humans: the Val883Met variant is located in the cyclase domain ([@CIT0011]); the Ala488Pro and Arg655Cys variants affect the submembrane and kinase-homology domains, respectively ([@CIT0012], [@CIT0013]). Because such variants enhanced both the basal and CNP-stimulated cGMP-synthesis of NPR-B, it was suggested that they provoke a conformational change which is similar to that adopted by NPR-B in response to CNP binding ([@CIT0014]). In this report, we describe a novel heterozygous gain-of-function variant c.1444_1449delATGCTG in exon 8 of *NPR2*, predicted to result in the deletion of 2 amino acids Met482-Leu483 in the cytosolic submembrane region of NPR-B. Similarly to the previously identified Ala488Pro variant in this region ([@CIT0012]), this novel variant is associated with tall stature and macrodactyly of the great toes (epiphyseal chondrodysplasia, Miura type). Research Design and Methods {#s1} =========================== Patients {#s2} -------- We studied a family of Dutch origin, including a mother and her 2 daughters. Informed consent and assent were obtained from the 3 probands. Blood samples were obtained from the mother and saliva was obtained from her daughters for genotyping. Fibroblasts from a skin biopsy were obtained from the mother for ex vivo studies of NPR-B activity. We compared fibroblast NPR-B expression and activity with fibroblasts from a wild-type (WT) NPR-B healthy volunteer and with the previously described proband with the activating Arg655Cys NPR-B variant ([@CIT0013]). Sanger sequencing and analysis {#s3} ------------------------------ DNA was isolated from blood samples using the chemagic Prime instrument (PerkinElmer, Waltham, MA, USA). DNA from saliva was isolated using the prepIT-L2P kit (DNA Genotek, Kanata, Ontario, Canada). Primer design for *NPR2* was based on a template sequence available on Ensembl transcript ENST00000342694.6 and ENST00000265074.13, respectively. Primers were designed for all coding exons and the corresponding intron/exon boundaries using Primer3, version 4.1.0 (sequences available upon request). Amplification of the selected gene regions was performed using GoTaq G2 DNA polymerase-mediated PCR (Promega Corporation, Madison, WI, USA) and was verified by agarose gel electrophoresis. Primer sequences and PCR conditions are available on request. After amplification by PCR, primers, and unincorporated dNTPs (Promega Corporation) were removed using exonuclease I (New England Biolabs, Inc, Ipswich, MA, USA) and calf intestine alkaline phosphatase (CIAP, Roche Applied Science, Hoffmann--La Roche AG, Basel, Switzerland). Sequencing was carried out directly on purified fragments with the ABI 310 Genetic Analyzer (Applied Biosystems, Foster City, CA, USA), using an ABI Prism BigDye terminator cycle sequencing ready reaction kit, version 1.1 (Applied Biosystems). The BigDye XTerminator purification kit was used as purification method for DNA sequencing with the purpose of removing unincorporated BigDye terminators. The presence of the *NPR2* variant in both affected daughters was also confirmed using Sanger sequencing. Isolation and culture of skin fibroblasts and cGMP determinations {#s4} ----------------------------------------------------------------- Skin biopsies were taken from the proband and a healthy 37-year-old female (height 171 cm, +0.07 standard deviation score \[SDS\]), and fibroblast cultures were established as previously described ([@CIT0013]). Fibroblasts of the previously described proband (not related to the currently investigated family) with the activating Arg655Cys NPR-B variant were included for comparison ([@CIT0013]). Experiments were performed in mitogen-free, serum-reduced DMEM (0.5% fetal calf serum 4 hours before experimentation). Cells were pretreated with 0.1 mM of the phosphodiesterase inhibitor 3-isobutyl-1-methylxanthine (IBMX; Sigma) for 15 minutes, and then stimulated with CNP (0.1-100 nM; Bachem) for another 10 minutes. Intracellular cGMP contents were extracted with 70% (v/v) ice-cold ethanol and determined by radioimmunoassay ([@CIT0005], [@CIT0013]). Site-directed mutagenesis and cGMP responses of transfected HEK-293 cells to CNP {#s5} -------------------------------------------------------------------------------- To generate the construct encoding the mutant protein NPR-B Met482-Leu483del, a PCR fragment was amplified from a *pFLAG-CMV1* vector containing the WT *NPR2* cDNA (Swiss-Prot entry Q6VVW5 ([@CIT0013])). The reverse primer: 5- CCATAGCATGCTAGCCAGCTCCTTCTCCAGCTTCCGGAAAATTAGGAAACTG-3, spanning the novel c.1444_1449delATGCTG deletion, and the forward primer: 5-GGGCACTTCAATTGGACAGCTCG-3 were used for the PCR amplification. The PCR fragment containing the variant was digested with the MfeI and NheI restriction enzymes, and subcloned into the *pFLAG-CMV1-NPR2* vector restricted with the same enzymes. The desired variant and the absence of unwanted variants were verified by sequencing. To compare the expression and activity levels of the WT and mutant NPR-B proteins, HEK-293 cells were transiently transfected with 10 μg of the respective plasmids in 10-cm dishes using the *X-tremeGene HP* DNA transfection reagent (Roche). For immunoblotting, membrane proteins were isolated 48 hours posttransfection. To assess intracellular cGMP responses, the HEK-293 cells were transferred into 24-well plates (100 000 cells per well) 1 day after transfection ([@CIT0013]). After 20 hours, the cells were serum starved for 4 hours, pretreated with 0.1 mM IBMX for 15 minutes, and then incubated with various concentrations of CNP for another 10 minutes. Intracellular cGMP contents were determined as stated previously ([@CIT0005], [@CIT0013]). Guanylyl cyclase activity assays {#s6} -------------------------------- Guanylyl cyclase activity of crude membranes prepared from transfected HEK-293 cells was assessed as described before ([@CIT0013]). Assays were performed in a 50 mM HEPES buffer, pH 7.4, containing 50 mM NaCl, 5% glycerol, 0.05% BSA, 1 mM IBMX, 2 mM guanosine triphosphate, 30 mM creatine phosphate, 1.5 U/mL creatine phosphokinase, and 2 mM ATP. Membranes (20 μg protein) were incubated with CNP (to assess ligand-dependent activity), or with 1% (v/v) Triton X-100 (for detergent stimulated, maximal activity) during 10 minutes ([@CIT0013]). cGMP formation was measured by radioimmunoassay ([@CIT0005], [@CIT0013]). Basal and CNP-stimulated cGMP responses were calculated as percentage of maximal Triton-stimulated activity. Cell membrane fractionations and western blotting {#s7} ------------------------------------------------- Membrane protein fractions were obtained from cultured skin fibroblasts and HEK-293 cells using a Protein Fractionation Kit (Thermo Fisher Scientific). For Western blot analyses, 25 μg protein was resolved by sodium dodecyl sulfate-PAGE ([@CIT0013]). The primary antibodies were against NPR-B (1:5000 ([@CIT0005]);) and Na^+^/K^+^-ATPase (1:5000; Abcam). In silico modelling {#s8} ------------------- To elucidate the possible impact of the reported Met482-Leu483 deletion within the intracellular membrane near region of NPR-B (UniProt code P20594) a theoretical 3-dimensional (3D) model was built for the region spanning Thr451 to Arg495. First, the location of the transmembrane helix of NPR-B was predicted using the software tool TMHMM 2. 0 (<http://www.cbs.dtu.dk/>). Subsequently, secondary structure prediction was performed for the region comprising residues Leu456 to Trp492 using the software JPRED4 (<http://www.compbio.dundee.ac.uk>). A 3D model was then built using the software package Quanta2008 (BIOVIA, San Diego, CA, USA) and the tool ProteinDesign. A 3D model for the deletion variant was similarly obtained by the building routine using Quanta2008/ProteinDesign removing residues Met482 and Leu483. Statistical analyses {#s9} -------------------- Data are presented as mean ± SEM. The 2-tailed unpaired Student *t-*test was used to analyze significant differences between 2 groups. *P* \< 0.01 was considered significant. Results {#s10} ======= Clinical characteristics and investigations {#s11} ------------------------------------------- The proband is a currently 37-year-old female. She was born at term to nonconsanguineous parents after a normal pregnancy with a birth weight of 3400 g (-0.39 SDS) and a birth length of 49 cm (-0.92 SDS). She reached developmental milestones at appropriate ages. At the age of 8 years, she was treated for the first time for markedly long great toes by bilateral percutaneous epiphysiodesis. Throughout adolescence, the proband exhibited tall stature and low weight for height. At 14 years of age, her height was 181 cm, bone age was 13 years according to Greulich and Pyle, and predicted adult height was 186.8 cm based on the Bayley-Pinneau tables. Physical examination at that time showed macrodactyly of the great toes, long thumbs, an arm span to height ratio of 0.9, and a positive "thumb sign" (a sign of arachnodactyly; positive if the distal phalanx of the thumb reaches beyond the ulnar border of the palm, after wrapping the fingers around the thumb in adduction) without other signs of Marfan syndrome. Family history was unremarkable except for osteoporosis in the proband's mother and grandmother. The proband's parents and brother are of normal height. From age 14 years, she was treated with supraphysiological doses of estrogens (200 mcg of ethinylestradiol per day) in combination with progesterone (5 mg of hydroxyprogesterone during the first 10 days of each month) for adult height reduction. Menarche occurred 3 months after start of treatment. Treatment had little effect on her growth and was discontinued at the age of 16 years. The proband has been experiencing irregular menses since menarche (menses every 25-60 days). She has had 2 spontaneous pregnancies and possibly 1 miscarriage. The proband is presently diagnosed with fibromyalgia and monitored by a rheumatologist. Her current anthropometric data are shown in [Table 1](#T1){ref-type="table"}. She has a minor ankle valgus, no scoliosis, and no joint hypermobility. She reports nearsightedness, with a correction of -5 diopters. Arterial blood pressure is normal. ###### Anthropometric Data H (HSDS) SHHR (SDS) AHR ------------ ----------------- ------------- ------ Proband 188.1 cm (+2.8) 0.50 (-1.7) 0.90 Daughter 1 136.8 cm (+2.0) 0.53 (-0.5) 0.94 Daughter 2 110.2 cm (+1.3) 0.54 (-0.7) 0.94 Daughter 1: eldest daughter (age 7.1 years), daughter 2: youngest daughter (age 4.0 years). Abbreviations: AHR, arm span to height ratio; H, height; SDS, standard deviation scores; SHHR, sitting height to height ratio. The proband and her nonconsanguineous spouse gave birth to 2 daughters. The girls were born at term following uneventful pregnancies. Birth weight of the eldest daughter at 38 + 1 weeks gestation was 2740 g (-1.13 SDS); length at 1 month was 53.5 cm (-0.13 SDS). She reached developmental milestones at an appropriate age, except for idiopathic toe walking until the age of 3 years. She has pes planovalgus. She underwent epiphysiodesis of both great toes when she was 5 years old. She wears glasses (+3.75/+3.5). The youngest daughter was born at 37 + 2 weeks gestation with a birth weight of 2740 g (-0.66 SDS); length at the age of 5 weeks was 54 cm (-0.43 SDS). She has a history of minor gross motor developmental delay and receives speech therapy. They were referred for clinical genetic evaluation at the age of 6 and 4 years, respectively, because of progressive macrodactyly of the great toes. At physical examination, the girls had markedly long great toes ([Fig. 1](#F1){ref-type="fig"}), ankle valgus, and long thumbs with a positive thumb sign. Skeletal surveys of the girls demonstrate pseudo-epiphyses of mid- and proximal phalanges in hands and feet, but not in other long bones ([Fig. 1](#F1){ref-type="fig"}). Beighton score (joint hypermobility score in which a maximum of 9 points can be allocated for specific maneuvers; ≥ 4 points indicates generalized hypermobility) was 0 in both girls. No specific facial dysmorphic features were noted. At anthropometric evaluation, the girls had a diminished span to height ratio resulting from shortened forearms (mesomelia). For current anthropometric data, see [Table 1](#T1){ref-type="table"} and [Fig. 1](#F1){ref-type="fig"}. The father of the girls is 178 cm (-0.58 SDS), which results in a target height of 175 cm (+0.68 SDS) with a range of 165 to 185 cm. Skeletal surveys of the girls showed pseudo-epiphyses of the mid- and proximal phalanges of all fingers and both great toes ([Fig. 1](#F1){ref-type="fig"}). The growth curve of the eldest daughter shows progressive growth acceleration until the age of 4.5 years up to a height of +2 SDS ([Fig. 2A](#F2){ref-type="fig"}); the curve of the youngest daughter seems to follow this pattern ([Fig. 2B](#F2){ref-type="fig"}). Bone age was according to calendar age in both girls. No other skeletal abnormalities were observed. {#F1} {#F2} Identification of a heterozygous variant in the *NPR2* gene {#s12} ----------------------------------------------------------- Because of the similarities with the phenotypes of the patients published by Miura et al. ([@CIT0011], [@CIT0012]), we selected *NPR2* as a candidate gene for further investigations. Sanger sequencing identified a heterozygous c.1444_1449delATGCTG deletion in exon 8 of *NPR2* (NM_003995.3), which is predicted to lead to an in-frame deletion of 2 highly conserved amino acids, methionine 482 and leucine 483, within the intracellular membrane near region of NPR-B ([Fig. 3](#F3){ref-type="fig"}). The same heterozygous *NPR2* variant was sequenced in DNA of the daughters. Other family members were not studied. [Fig. 3](#F3){ref-type="fig"} illustrates that the amino acid sequence of this region of NPR-B is very conserved across species. ![Alignments of the amino acids 459--519 of NPR-B comparing the sequences of wild-type NPR-B across species, human NPR1 and mutant NPR-B. TMD represents the transmembrane domain and PKD the start of the protein kinase-homology domain. The black arrows label the 2 amino acids (Met482-Leu483) that are deleted in the cytosolic sub-membrane region of NPR-B in the proband (last line). For comparison, the sequence of the human natriuretic peptide receptor type A (NPR-A, also named guanylyl cyclase-A \[GC-A\]) is also included. Please note the high sequence conservation of the trans- and submembrane regions of NPR-B among species and between NPR-B and NPR-A, which suggests that the here studied deletion variant affects a functionally critical region. NPR, natriuretic peptide receptor.](dgaa190f0003){#F3} Fibroblasts from the proband with the Met482-Leu483del NPR-B variant have enhanced basal cGMP levels and augmented cGMP responses to CNP {#s13} ---------------------------------------------------------------------------------------------------------------------------------------- To study the impact of the variant on the expression and activity of native NPR-B, we cultured fibroblasts from skin biopsies of the forearms of the proband and an age-/gender-matched control donor. Western blot analyses of fractionated membrane proteins revealed that membrane NPR-B expression levels were lower in the proband´s fibroblasts as compared to control ([Fig. 4A](#F4){ref-type="fig"}). Despite this attenuated expression, the basal cGMP contents of such fibroblasts were significantly increased ([Fig. 4B](#F4){ref-type="fig"}). Moreover, the cGMP responses to low, physiological CNP concentrations (0.1 and 1 nM) were significantly augmented ([Fig. 4C](#F4){ref-type="fig"}). The effects of higher, supraphysiological CNP concentrations (10 and 100 nM) were also greater in fibroblasts from the proband; however, the difference to controls did not reach statistical significance ([Fig. 4C](#F4){ref-type="fig"}). Although we could only study fibroblasts from 1 control donor, the results suggest increased baseline and ligand-evoked activity of the mutant NPR-B naturally expressed in fibroblasts. {#F4} Site-directed mutagenesis demonstrates enhanced baseline and CNP-stimulated guanylyl cyclase activity of the mutant NPR-B {#s14} ------------------------------------------------------------------------------------------------------------------------- Receptor-independent alterations may influence NPR-B activity in native cells. Therefore, we replicated the variant by site-directed mutagenesis for studies in a heterologous expression system. Transfection of HEK-293 cells with cDNA encoding Met482-Leu483del NPR-B produced slightly less amounts of membrane-bound immunoreactive protein compared with WT NPR-B in 3 separate experiments ([Fig. 5A](#F5){ref-type="fig"}). Together with the experiments with cultured fibroblasts this suggests diminished synthesis and/or membrane localization of the mutant. Despite, basal intracellular cGMP levels of HEK-293 cells expressing the mutant were \~9.9-fold higher in comparison with cells expressing WT NPR-B ([Fig. 5B](#F5){ref-type="fig"}). CNP evoked concentration-dependent increases of intracellular cGMP levels. The responses to low CNP concentrations (0.01-1 nM CNP) were significantly greater in cells expressing the mutant, whereas the responses to a high CNP concentration (10 nM) were similar between cells expressing the WT or the mutant receptor ([Fig. 5B](#F5){ref-type="fig"}). {#F5} Guanylyl cyclase activity assays were performed with crude membranes from transfected HEK-293 cells in the presence of 2 mM ATP to mimic cytoplasmic ATP levels. Such assays harbor the advantage that the response to CNP can be normalized to the maximal, detergent (Triton)-induced NPR-B activity. This facilitates the comparison of WT and mutant receptor activities when their membrane expression levels are unequal. [Fig. 5C](#F5){ref-type="fig"} corroborates the previous experiments showing that baseline, ligand-independent activity of Met482-Leu483del NPR-B as well as the responses to low CNP concentrations (0.01-1 nM CNP) were significantly augmented. The responses to the highest CNP concentration (10 nM) were similar to the WT receptor ([Fig. 5C](#F5){ref-type="fig"}). In concordance with the reduced protein expression levels, the maximal, Triton-stimulated activity of the mutant NPR-B was slightly reduced ([Fig. 5C](#F5){ref-type="fig"}). To account for the differences in membrane receptor expression levels, within each experiment the CNP stimulated activity data were normalized to the respective maximal, Triton-stimulated NPR-B activity (considered as 100%) ([Fig. 5D](#F5){ref-type="fig"}). This normalization demonstrated the mutant NPR-B exhibits augmented catalytic activity at baseline and in response to all tested CNP concentrations. In other words, the previous observations that the effects of high CNP concentrations (10-100 nM) on intracellular cGMP levels of fibroblasts ([Fig. 4C](#F4){ref-type="fig"}) or HEK-293 cells ([Fig. 5B](#F5){ref-type="fig"}) were similar between cells expressing WT or mutant NPR-B, is probably linked to the attenuated membrane expression of the mutant. The Met482-Leu483del NPR-B variant has lower baseline and CNP-stimulated guanylyl cyclase activities than the Arg655Cys NPR-B variant {#s15} ------------------------------------------------------------------------------------------------------------------------------------- Last, we compared in fibroblasts and transfected HEK-293 cells the CNP responsiveness of the novel deletion variant with the previously published variant Arg655Cys ([@CIT0013]). This point mutation variant, which we found in an extremely tall proband ([@CIT0013]), affects the KHD of NPR-B. [Fig. 6A](#F6){ref-type="fig"} illustrates that membrane NPR-B expression levels in fibroblasts cultured from both probands were lower, as in fibroblasts from the control donor. The cGMP-responses to a physiological CNP concentration (0.1 nM) were greatest in fibroblasts from the previously published proband with the Asp655Cys NPR-B variant ([Fig. 6B](#F6){ref-type="fig"}). {#F6} HEK-293 cells were transfected with the cDNAs encoding each mutant. Western blot analysis showed that the expression of the Met482-Leu483del NPR-B mutant was slightly lower than the expression of Arg655Cys NPR-B ([Fig. 6C](#F6){ref-type="fig"}). Guanylyl cyclase activity assays using crude membranes from transfected HEK-293 cells revealed that Met482-Leu483del NPR-B has significantly lower baseline (ligand-independent) activity and lower responses to CNP compared with Arg655Cys NPR-B ([Fig. 6D](#F6){ref-type="fig"}). As also shown, in concordance with the reduced protein expression levels, the maximal, Triton-stimulated activity of the Met482-Leu483del NPR-B mutant was also slightly lower. To account for such differences in membrane expression levels, the CNP-stimulated activity data were again normalized to the respective maximal, Triton-stimulated NPR-B activity. These evaluations confirmed that the Met482-Leu483del NPR-B mutant exhibits lower responsiveness to CNP as compared to the Arg655Cys NPR-B mutant ([Fig. 6E](#F6){ref-type="fig"}). Molecular modelling suggests that the Met482-Leu483 deletion changes the 3D structure of the submembrane region of NPR-B {#s16} ------------------------------------------------------------------------------------------------------------------------ Crystal structure analyses of the extracellular domain (ECD) of natriuretic peptide receptor-A (NPR-A), shared by ANP and BNP, provided insights into the molecular mechanisms of ligand recognition and a model of ligand-induced allosteric NPR-A activation ([@CIT0015], [@CIT0016]). These experimental studies revealed that ANP binding induces a rotation of the ECDs of the 2 monomer subunits of the receptor dimer with respect to each other. It was postulated that this movement transduces across the membrane being an allosteric trigger for activation of the intracellular guanylyl cyclase domain to cGMP synthesis ([@CIT0015]). Unfortunately, no experimentally validated structure is currently available for the transmembrane and intracellular regions of NPR-B or a homologous protein. To generate a working hypothesis about how the deletion of residues Met482-Leu483 could impact NPR-B activity, a theoretical 3D model of the transmembrane domain and the submembrane segment containing residues Met482-Leu483 was built using bioinformatic tools. Sequence-based tools to predict location and secondary structure indicated that the residues Leu456 to Phe478 form a single transmembrane helix, as expected for a type I transmembrane receptor ([Fig. 3](#F3){ref-type="fig"}). This is consistent with the notion of basic positively charged amino acids at the C-terminal end of transmembrane helices (here, as residues Arg479 and Lys480), which are important to vertically fix such helices in the lipid bilayer ([@CIT0017]). Subsequently secondary structure modelling was performed for this transmembrane sequence and flanking extra- and intracellular protein segments. This predicted that the transmembrane and intracellular membrane near regions very likely form a continuous α-helical structure running from Leu456 to Trp492 ([Fig. 7A](#F7){ref-type="fig"}). The 3D models were built to estimate how the deletion of residues Met482 and Leu483 would change these single helices. Because an α-helix has 3.6 residues per full turn, a 2-residue deletion would result in a reorientation of the helix surface ahead and past the deletion site by 120° ([Fig. 7A](#F7){ref-type="fig"}, right panel; and [Fig. 7B](#F7){ref-type="fig"}, left panel). The amphipathic signature of the helix markedly changes as a result of that. The 3D model of WT NPR-B indicated that this segment does not form a continuous hydrophobic patch, whereas in the Met482-Leu483 deletion variant, a continuous hydrophobic patch is predicted to run along the transmembrane region into the submembrane region ([Fig. 7B](#F7){ref-type="fig"}, middle panel). Such change in the surface chemistry of this intracellular membrane-near segment of the here described mutant NPR-B might can affect dimer stability or conformation thereby leading to a preactivated state already in the absence of CNP or a hyperactivated state by stabilizing the CNP-induced active receptor conformation. ![A theoretical 3D model of the trans- and cytosolic submembrane segments of WT and mutant NPR-B. Bioinformatic modelling predicted that Leu456 is the N-terminal start and Phe478 the end of the transmembrane region, which is consistent with the notion that basic residues at the C-term of transmembrane helices (here, residues Arg479 and Lys480) are important to vertically fix such helices in the lipid bilayer ([@CIT0017]). Secondary structure prediction indicated an uninterrupted α-helical element comprising residues Leu456 to Trp492. (A) A theoretical 3D model of this helix was built for WT NPR-B (A, left helix) and mutant NPR-B (A, right helix; the deleted residues are marked with C-atoms colored in magenta). The deletion Met482-Leu483 results in a rotational rearrangement of the helix past the deletion site (in panel A, the right panel compares the WT and mutant helix, the rotation is marked). This in silico model suggests that in the mutant NPR-B the residues Glu484 to Trp492 are rotated by 120° clockwise and that the amphipathic structure of the helix is markedly changed. (B) Left panel: 3D model of the helix covering the transmembrane and cytoplasmic submembrane regions is shown for WT and Met482-Leu483del NPR-B (residues Thr451 to Arg495). Color coding was done according to amino acid polarity, with hydrophobic amino acid residues colored in gray, positively charged residues colored in blue, negatively charged residues marked in red, and polar but uncharged amino acids indicated in green. Middle panel: The model is depicted with a semitransparent surface representation showing a continuous hydrophobic patch for Met482-Leu483del past the transmembrane segment. WT NPR-B does not show such a hydrophobic surface patch. Right panel: The model is rotated clockwise by 90° around the y-axis. 3D, 3-dimensional; NPR, natriuretic peptide receptor; WT, wild-type.](dgaa190f0007){#F7} Discussion {#s17} ========== Loss-of-function sequence variants in NPR-B (encoded by the *NPR2* gene) cause short stature ([@CIT0008]), whereas gain-of-function variants lead to tall stature ([@CIT0011]). Here, we describe a mother and 2 daughters with tall stature and macrodactyly of the great toes (epiphyseal chondrodysplasia, Miura type) associated with an in-frame deletion in *NPR2* resulting in the deletion of the amino acids Met482-Leu483 in the intracellular membrane-near region of NPR-B. Assays with cultured native fibroblasts and transfected HEK-293 cells demonstrated that Met482-Leu483del NPR-B responds to CNP with markedly enhanced cGMP production. Also, the ligand-independent (baseline) cGMP-synthesis activity of this mutant receptor is markedly increased. Because the CNP/NPR-B/cGMP signalling pathway is critically involved in bone development by stimulating growth plate chondrocyte differentiation and proliferation ([@CIT0001]), we conclude that the Met482-Leu483del variant is responsible for the observed skeletal overgrowth of the proband. The current growth of the 2 daughters is not as pronounced compared with the proband. This can be explained by their father's height, which is slightly below average. Because height is a highly polygenic trait, this observation is congruent with the notion that monogenic growth disorders may present with a height within normal ranges. This report adds to the three earlier reports of gain-of-function NPR-B variants: 2 families with variants in the GCD or submembrane domains exhibited tall stature including macrodactyly of the great toes, scoliosis, coxa valga, and slipped capital femoral epiphysis ([@CIT0011], [@CIT0012]); and 1 man with a variant in the KHD had isolated extreme tall stature ([@CIT0013]). Elegant biochemical studies in a heterologous expression system showed that these variants have distinct effects on NPR-B-mediated cGMP production, NPR-B dephosphorylation/desensitization or the allosteric modulation of the GCD by ATP ([@CIT0014], [@CIT0018]). Our comparative studies of the Met482-Leu483del and Arg655Cys NPR-B mutants indicate that the extent of cGMP-overproduction correlates with height (present study and ([@CIT0013])). Ultimately it remains unclear why different activating NPR-B variants are associated with distinct overgrowth phenotypes. Pronounced macrodactyly of the great toes is a striking feature shared by almost all known gain-of-function NPR-B variants (([@CIT0011], [@CIT0012]); see [Fig. 1](#F1){ref-type="fig"} of the present study). In general, 3 cGMP-regulated proteins participate in chondrogenesis and chondrocyte proliferation during long bone growth: cGMP-dependent protein kinases type I and II; and phosphodiesterase (PDE) 3A, a cGMP-inhibited cAMP-hydrolyzing enzyme ([@CIT0001]). Notably, during development PDE3A is highly expressed in domains involved in digit formation. PDE3A gain-of-function variants, with increased cAMP-hydrolytic activity, are associated with severe brachydactyly ([@CIT0019]). We suggest that augmented CNP/NPR-B/cGMP-dependent inhibition of PDE3A (diminishing cAMP-hydrolytic activity and thereby enhancing chondrocyte cAMP levels) may contribute to macrodactyly in gain-of-function NPR-B variants. However, why overactivation of CNP/NPR-B/cGMP signalling preferentially affects the growth of the big toes is presently unknown. The proband has a history of delayed puberty, which also occurred in our previous proband ([@CIT0013]). This increases final height by allowing for prolonged growth before epiphyseal closure. We assume that in the proband, pubertal delay could have been a result of her low body weight, leading to hypothalamic dysfunction. Another novel clinical feature in this NPR-B gain-of-function phenotype is mesomelia. Until now, this was only reported in loss-of-function NPR-B variants. Unfortunately, segregation analysis of the *NPR2* in-frame deletion with mesomelia or delayed puberty could not be established in other family members. Therefore, it remains unclear if these features are part of the phenotypic spectrum of activating NPR-B variants, and this needs to be explored in future diagnosed families with NPR-B gain-of-function variants. The presence of pseudo-epiphyses ([Fig. 1](#F1){ref-type="fig"}) has been reported before by Boudin et al. ([@CIT0020]), who reviewed X-rays of previously published cases with activating NPR-B variants after they found that biallelic loss-of-function variants in natriuretic peptide receptor-C (NPR-C; also named NPR3) are associated with tall stature, macrodactyly of the great toes, and pseudo-epiphyses ([@CIT0019]). NPR-C binds and internalizes all 3 natriuretic peptides (ANP, BNP, CNP) with similar affinities, thereby acting as a "clearance" receptor ([@CIT0001]). It was suggested that loss-of-function variants in NPR-C result in reduced clearance of CNP, thereby increasing local CNP levels in the bone and CNP/NPR-B signalling ([@CIT0001], [@CIT0020]). The reported NPR-B deletion variant Met482-Leu483 as well as the previously reported substitution Ala488Pro ([@CIT0012]) affect a topological domain between the transmembrane region and the kinase-homology domain of NPR-B. As shown in [Fig. 3](#F3){ref-type="fig"}, the amino acid sequence of this region is highly conserved in NPR-B across species and among natriuretic peptide receptors. Notably, both variants resulted in augmented baseline as well as ligand-dependent NPR-B/cGMP activities. Moreover, they were associated with very similar clinical phenotypes, with pronounced macrodactyly of the great toes. Interestingly, a recently published novel NPR-B variant Arg495Cys, affecting the same protein region, was associated with diminished CNP-simulated NPR-B catalytic activity and idiopathic short stature ([@CIT0021]). The marked impact of these 3 variants on baseline and CNP-stimulated NPR-B function (([@CIT0012], [@CIT0021]) and present study) indicates that this submembrane region has a critical role, although the precise function is unknown. Different NPRs can bind the same ligand (e.g., both NPR-A and NPR-C bind ANP and BNP, both NPR-B and NPR-C bind CNP), indicating that the ECD of all NPRs have similar ligand recognition and activation mechanisms ([@CIT0001]). By sedimentation equilibrium analyses, it was found that the crystalized ECD of NPR-A undergoes spontaneous dimerization with a dissociation constant *K*~*d*~ of \~500 nM, suggesting that most NPR-As preexist in the cell's membrane as homodimers ([@CIT0016], [@CIT0022]). Furthermore, crystal structure analyses of the ECDs of NPR-A and NPR-C revealed that a single ANP (in NPR-A) or either ANP or CNP molecule (in NPR-C) binds within the interface of a receptor homodimer and induces a rotation of the ECDs of the 2 dimer subunits with respect to each other ([@CIT0015], [@CIT0023]). It is postulated that this movement transduces across the membrane being an allosteric trigger for intracellular signalling by natriuretic peptides ([@CIT0015], [@CIT0023]). Such experimental studies have not been performed for NPR-B. However, because of the high amino acid homologies to NPR-A (in the intracellular domains) and NPR-C (in the extracellular domain), it is likely that this allosteric activation mechanism is valid also for NPR-B. To provide a potential molecular explanation for the effect of the described Met482-Leu483 deletion here, we performed in silico modelling of this receptor region. Our 3D model predicted a single uninterrupted α-helical element comprising the transmembrane (Leu456-Phe478) and submembrane regions of NPR-B (until residue Trp492). Because of the architecture of α-helices, the deletion reported here Met482-Leu483 would result in a rotational rearrangement of the helix past the deletion site (clockwise by 120°) and thereby markedly change its amphipathic surface. Based on the results of our functional studies, we assume that this change in amphilicity could stabilize the active conformation of the NPR-B dimer and further facilitate its CNP-induced allosteric activation. This could lead to a hyperactivated state of NPR-B at baseline and upon ligand-binding. However, the exact molecular mechanism is unknown. Because the probands are heterozygous carriers of the variant, they will possibly express WT/WT (\~25%), WT/mut (\~50%), and mut/mut (\~25%) NPR-B dimers. Our present study and also the previously published studies ([@CIT0011]) cannot distinguish whether the resulting gain of function of the CNP/NPR-B system is caused by the small population of mut/mut dimers or linked to the prevalent WT/mut dimers. In summary, with this report of a family with tall stature and macrodactyly of the great toes because of a gain-of-function in-frame deletion within the submembrane region of NPR-B, the genotypical and phenotypical spectrum of NPR-B-related tall stature is expanded. Our observations emphasize the important role of this domain in the regulation of cyclic GMP production and bone growth in response to CNP/NPR-B signaling. The authors are grateful for the participation of the proband and her family. ***Financial Support:*** The studies of CNP/NPR-B signalling by M.K.'s group are funded by the Deutsche Forschungsgemeinschaft (DFG KU 1037/7-1 and DFG KU 1037/8-1) and the Bundesministerium für Forschung und Technik (BMBF 01EO1004). The work of E.B. was supported by grants of the 'Fonds voor Wetenschappelijk Onderzoek Vlaanderen' (12A3814N and 1507317N) and the 'Bijzonder Onderzoeksfonds' (BOF) of the University of Antwerp (BOF KP: 34417). ***Author Contributions:*** P.L., D.v.d.K., M.K., and H.v.D. planned the studies and wrote the manuscript. P.L., A.v.H., and D.v.d.K. gathered clinical data on the proband and her family. Functional assays were performed by E.M.L., E.B., F.W., H.S., and M..K. T.M. performed in silico modellings. All authors critically reviewed manuscript drafts and approved the final manuscripts as submitted. 3D : 3-dimensional cGMP : cyclic guanosine monophosphate CNP : C-type natriuretic peptide ECD : extracellular domain IBMX : 3-isobutyl-1-methylxanthine NP : natriuretic peptide NPR : natriuretic peptide receptor PDE : phosphodiesterase SDS : standard deviation score WT : wild-type. Additional Information {#s0105} ====================== ***Disclosure Summary:*** The authors have no conflicts of interests to state. The authors have no ethical conflicts to disclose. This study was approved by the Medical Ethical Committee of Leiden University Medical Center. The proband and her family agreed to participate and signed informed consent. ***Data Availability:*** The datasets generated during and/or analyzed during the current study are not publicly available but are available from the corresponding author on reasonable request. [^1]: P.L. and E.M.-L. contributed equally; M.K. and D.C.M.vd.K. contributed equally.

{#sp1 .338} {#sp2 .339} {#sp3 .340} {#sp4 .341} {#sp5 .342} {#sp6 .343}

1. Introduction {#sec0005} =============== In space, microgravity affects the central circulation in humans and induces a number of adaptive changes within the cardiovascular system. Previous investigations showed that the baroreflex sensitivity fluctuates along with altered blood volume distribution \[[@bib0005], [@bib0010], [@bib0015]\], which affects neural mechanisms involved in dynamic cardiovascular coordination. Several reports indicate that heart rate is maintained at preflight values \[[@bib0020], [@bib0025], [@bib0030]\] and that parasympathetic activity is reduced [@bib0020] in space. Cardiac output and stroke volume are reportedly increased in space as a result of an increase in preload to the heart induced by upper body fluid shift from the lower body segments with no major difference in sympathetic nerve activity [@bib0030]. However, high sympathetic nervous activity, measured invasively by microneurography in peroneal nerves, has been simultaneously detected in space in three astronauts [@bib0035] compared to the ground-based supine posture. Physiologic acclimation to space flight is a complex process involving multiple systems [@bib0040]. How the neural cardiovascular coordination adapts to the space environment is still poorly understood in humans. When faced with a new environment, humans must first acclimate to it in order to survive. This includes the cardiovascular system. Adjustment to the new environment to improve quality of life follows, involving the autonomic, endocrine and immune systems, among others. But, as we reported previously [@bib0045], the "intrinsic" cardiovascular regulatory system, reflected by the fractal scaling of HRV \[[@bib0045], [@bib0050], [@bib0055]\], did not adapt to the new microgravity environment in space during long-duration (about 6-month) spaceflights. By contrast, after 6 months in space, the circadian rhythm of heart rate had adapted to the new microgravity environment in space [@bib0060], an important observation since disruption of circadian rhythms adversely affects human health \[[@bib0065], [@bib0070]\]. As humans plan for long-term space exploration, it is critical to ascertain that the regulatory system can function well in a microgravity environment. The power-law fractal scaling of heart rate variability (HRV) relates to the autonomic [@bib0075], endocrine [@bib0075], immune, inflammatiory \[[@bib0080], [@bib0085]\], mental, cognitive [@bib0090], and behavioral systems, which operate at multiple frequency ranges, from the 1 Hz cardiac cycle to circadian and even secular variations, as part of a broad time structure, the chronome [@bib0095]. Herein, we examine how the space environment affects HRV in specific frequency regions, broken down into 8 different frequency ranges. We focus on the basic rest-activity cycle (BRAC), well known since Kleitman [@bib0100], who showed regularly occurring alternations between non-REM and REM (Rapid Eye Movement) sleep. The BRAC is involved in the functioning of the central nervous system and manifests time-dependent changes in human performance, including oral activity cycles (e.g., eating, drinking, smoking). 2. Methods {#sec0010} ========== 2.1. Subjects {#sec0015} ------------- Ten healthy astronauts (8 men, 2 women) participated in this study. Their mean (± SD) age was 49.1 ± 4.2 years. Their mean stay in space was 171.8 ± 14.4 days. On the average, astronauts had already experienced spaceflight 0.9 ± 0.7 times and had passed class III physical examinations from the National Aeronautics and Space Administration (NASA). This study obtained consent from all subjects and gained approval from the ethics committee jointly established by the Johnson Space Center and Japan Aerospace Exploration Agency (JAXA). A detailed explanation of the study protocol was given to the subjects before they gave written, informed consent, according to the Declaration of Helsinki Principles. 2.2. Experimental protocols {#sec0020} --------------------------- Ambulatory around-the-clock 24-hour electrocardiographic (ECG) records were obtained by using a two-channel Holter recorder (FM-180; Fukuda Denshi). Measurements were made five times: once before flight (Control), three times during flight (International Space Station (ISS) 01, ISS02, and ISS03), and once after return to Earth (After flight). The before-flight measurement session (Control) was conducted on days 234.4 ± 138.4 (63 to 469) before launch in all but one astronaut who had technical problems with his before-flight record. In his case, a replacement control record was obtained 3.5 years after return to Earth. The three measurement sessions during flight were taken on days 20.8 ± 2.9 (18 to 28, ISS01), 72.5 ± 3.9 (67 to 78, ISS02) and 152.8 ± 16.1 (139 to 188, ISS03) after launch, the latter corresponding to 19.1 ± 4.1 days (11 to 27) before return (ISS03). The last measurement session was performed on days 77.2 ± 14.4 (37 to 127 days) after return to Earth (After flight). 2.3. Analysis of heart rate variability and measurement of 1/f fluctuations in HR dynamics {#sec0025} ------------------------------------------------------------------------------------------ The measurement procedures and data collection were conducted as previously reported \[[@bib0045], [@bib0060]\]. Briefly, for HRV measurements, QRS waveforms were read from continuous electrocardiographic (ECG) records. The RR intervals between normal QRS waveforms were extracted as the normal-to-normal (NN) intervals. The measured NN intervals were A/D converted (125-Hz) with 8-ms time resolution. After the authors confirmed that all artifacts were actually removed and that the data excluded supraventricular or ventricular arrhythmia, frequency-domain measures [@bib0075] were obtained with the MemCalc/CHIRAM (Suwa Trust GMS, Tokyo, Japan) software [@bib0105]. Time series of NN intervals covering 5-min intervals were processed consecutively, and the spectral power in different frequency regions was computed, namely in the "high frequency (HF)" (0.15--0.40 Hz; spectral power centered around 3.6 sec), "low frequency (LF)" (0.04--0.15 Hz; spectral power centered around 10.5 sec), and "very low frequency (VLF)" (0.003--0.04 Hz; 25 sec to 5 min) regions of the Maximum Entropy Method (MEM) spectrum. VLF power was further broken down into "VLF band-1" (0.005--0.02 Hz; 50 sec to 3.3 min), "VLF band-2" (0.02--0.03 Hz; 33 to 50 sec) and "VLF band-3" (0.03--0.15 Hz; 6.7 to 33 sec). Time series of NN intervals were also processed consecutively in 180-min intervals, progressively displaced by 5 min, to estimate the "ultra-low frequency" (ULF) component (0.0001--0.003 Hz; periods of 2.8 hours to 5 min), further broken down into: "ULF band-1" (0.0001--0.0003 Hz; 166.7 to 55.5 min), "ULF band-2" (0.0003--0.001 Hz; 55.5 to 16.6 min), and "ULF band-3" (0.001--0.005 Hz; 16.6 to 3.3 min). Thus, 8 different frequency regions were examined: "HF", "LF", "VLF01", "VLF02", "VLF03", "ULF01", "ULF02", and "ULF03". Results representing each HRV component were averaged over the entire 24-hour. To evaluate the 1/f^β^-type scaling in HRV, the log~10~\[power\] (ordinate) was plotted against log~10~\[frequency\] (abscissa) and a regression line fitted to estimate the slope β, as reported earlier [@bib0045]. Focus was placed on the frequency range of 0.0001--0.01 Hz (periods of 2.8 hours to 1.6 minutes), as previously reported [@bib0045]. 2.4. Fit of 4-component cosine model {#sec0030} ------------------------------------ A multiple-component model consisting of cosine curves with anticipated periods of 24, 12, 8 and 1.5 hours was fitted to various HRV endpoints by cosinor [@bib0110] to assess their time structure and to determine how the latter may have been modified in space. The model includes the usually prominent circadian rhythm (24-hour period) and its first two harmonic terms with periods of 12 (circasemidian) and 8 (circaoctohoran) hours, as well as the BRAC (with a period of about 90-min). Using a (least squares) regression approach, the cosinor does not require the data to be equidistant, and can thus handle missing values in cases when artifacts prevented the computation of HRV endpoints in some of the 5-min or 180-min intervals. Analyses considered primarily the Midline Estimating Statistic Of Rhythm (MESOR, a rhythm-adjusted mean) and the amplitude of each of the 4 components, as a measure of the extent of predictable change within each cycle. The 4-component model was fitted to 24-hour records of NN intervals, total power (TF), and power in the ULF (separately also in the ULF01, ULF02, and ULF03), VLF, LF, and HF regions of the MEM spectrum. 2.5. Inter-individual differences in HRV response to microgravity {#sec0035} ----------------------------------------------------------------- Consistent differences in various HRV endpoints were noted in the way astronauts responded to microgravity. Examination of the inter-individual differences prompted the classification of the 10 astronauts into 2 clearly distinct groups. Hence, the influence of the space environment was also assessed separately in each group. 2.6. Statistical analyses {#sec0040} ------------------------- Since we previously showed that the fractal scaling of HRV did remain altered in space as compared to Earth during long-term (∼ 6-month) spaceflights, this study specifically examines the behavior of HRV in 8 different frequency regions of the spectrum (ULF01, ULF02, ULF03, VLF01, VLF02, VLF03, LF, and HF), which can be considered to provide independent information. Adjustment for multiple testing thus uses a P-value of 0.05/8 to indicate statistical significance, using Bonferroni\'s inequality to adjust for multiple testing. The same correction is applied to other HRV endpoints shown for the sake of completeness, noting the high degree of correlation existing among different indices. We test whether HRV endpoints differ between space and Earth while showing no change among the 3 records obtained in space. In order to do so, estimates of HRV endpoints averaged over 24 hours were expressed as mean ± SD (standard deviation). To minimize inter-individual differences in HR and HRV among the 10 astronauts that may obscure an effect of the space environment, 24-hour mean values of each variable were expressed as a percentage of mean, computed across the 5 sessions (before flight, ISS01, ISS02, ISS03, and after return to Earth) contributed by each astronaut. In this way, astronauts serve as their own longitudinal control. The two-sided paired-t and one-way analysis of variance (ANOVA) for repeated measures were applied on these relative values for the space vs. Earth difference and for comparing the 3 records in space, respectively. Estimates of the MESOR and of the relative amplitude of each of the 4 anticipated components (with periods of 24, 12, 8, and 1.5 hours, expressed as a percentage of MESOR) of the selected HRV endpoints were considered as imputations for a comparison of HRV endpoints obtained during ISS03 versus before-flight. The statistical significance of change between the two sessions was determined using the 2-tailed paired t test. Inter-group differences were determined using the two-tailed Student t-test. P-values less than 0.05, adjusted for multiple testing according to Bonferroni\'s inequality, were considered to indicate statistical significance. The Stat Flex (Ver. 6) software (Artec Co., Ltd., Osaka, Japan) was used. 3. Results {#sec0045} ========== 3.1. Change in time structure of heart rate variability during long-duration spaceflight {#sec0050} ---------------------------------------------------------------------------------------- Average HRV endpoints during each of the 5 sessions are shown in [Table 1A](#tbl0005){ref-type="table"}. Results from a comparison of their relative values between space and Earth and across the 3 sessions on the ISS are summarized in [Table 1B](#tbl0010){ref-type="table"}. On average, among the 10 astronauts, no differences were found in HR (or NN) or in SDNN, the standard deviation of NN intervals. As reported earlier, the fractal scaling of HRV (slope β) was statistically significantly less steep in space than on Earth, while no changes were observed across the 3 records obtained in space, [Table 1A](#tbl0005){ref-type="table"}, [Table 1B](#tbl0010){ref-type="table"}. This result may be accounted for by the large space-Earth difference observed in the ULF frequency region of the spectrum, which is statistically significant for ULF02 and ULF03, as well as for ULF01 once it is normalized by the total spectral power (TF). These HRV endpoints did not differ among the 3 sessions recorded on the ISS, [Table 1A](#tbl0005){ref-type="table"}, [Table 1B](#tbl0010){ref-type="table"}. Of all the HRV endpoints considered herein, apart from β and the spectral power in the 3 ULF bands, only SDmean5 and SDmean30 show a lasting difference in space as compared to Earth, [Table 1A](#tbl0005){ref-type="table"}, [Table 1B](#tbl0010){ref-type="table"}. Differences in β and the spectral power in the 3 ULF bands may stem from changes occurring around a frequency of one cycle in about 90 min. Indeed, β is computed over a frequency range centered around one cycle in about 90 min (1.7--166 min). Its absolute value decreased from 1.087 ± 0.130 (control, before flight) to 0.924 ± 0.095 (ISS03) (p \< 0.01). Correspondingly, ULF01/TF, also centered around 90 min, increased from 0.207 ± 0.053 to 0.310 ± 0.090, whereas ULF02/TF and ULF03/TF decreased from 0.189 ± 0.037 to 0.136 ± 0.030 and from 0.219 ± 0.035 to 0.151 ± 0.034, respectively. 3.2. Individual HRV response to microgravity associated with change in parasympathetic nerve activity {#sec0055} ----------------------------------------------------------------------------------------------------- Individual 24-hour records of NN intervals (and hence instantaneous HR values) showed striking differences among the 10 astronauts. In 7 of them (Group 1), the 24-hour standard deviation (SD) of NN intervals was much lower (74.7--105.4 msec) than in the other 3 (Group 2) (171.7--196.0 msec) (Student t = 10.462, p \< 0.001). The two groups also differed in their average NN intervals (820. 8 ± 44.6 vs. 1023.2 ± 54.2, Student t = 2.610, p = 0.031). The inter-group difference in SD (NN) persisted during ISS01 (t = 3.451, p = 0.009), ISS02 (t = 4.615, p = 0.002), and ISS03 (t = 3.430, p = 0.009), as well as after return to Earth (t = 3.287, p = 0.011), when a difference in average NN intervals was also observed (t = 2.610, p = 0.031). Moreover, astronauts in Group 1 tended to respond to the space environment by increasing their average NN interval (decreasing their HR). The inter-group difference in response was statistically significant during ISS02 (t = 2.814, p = 0.023) and ISS03 (t = 3.515, p = 0.008), when the average NN intervals of all 7 astronauts of Group 1 was increased (on average by 85.4 ± 59.0 msec, t = 3.825, p = 0.009) and that of all 3 astronauts of Group 2 was decreased (on average by 41.9 ± 23.6 msec, t = 3.072, p = 0.092). [Table 2](#tbl0015){ref-type="table"} lists individual results during each of the 5 recordings, illustrating strong inter-individual differences in the HRV response to the space environment. 3.3. Power-law scaling β and ULF component of astronauts whose heart rate decreased in space {#sec0060} -------------------------------------------------------------------------------------------- As seen for all 10 astronauts, the absolute value of β was also statistically significantly decreased in space (ISS03: 0.944 ± 0.097) as compared to preflight (1.144 ± 0.102) for the 7 astronauts of Group 1. Their ULF02 and ULF03 power was statistically significantly decreased from 915.0 ± 320.4 msec^2^ to 673.6 ± 275.3 msec^2^ and from 1017.4 ± 268.1 msec^2^ to 647.6 ± 192.5 msec^2^, respectively. In Group 2, there were no statistically significant differences in any of the HRV endpoints. 3.4. Change in chronome components (notably the basic rest-activity cycle) of heart rate variability during long-duration exposure to microgravity in space {#sec0065} ----------------------------------------------------------------------------------------------------------------------------------------------------------- Changes during the 6-month spaceflight in the relative amplitudes of the 24-, 12-, 8-, and 1.5-hour components, expressed as a percentage of the MESOR, are shown in [Table 3](#tbl0020){ref-type="table"} for NN intervals, β, TF, and the different frequency ranges of the spectrum. On the average, the 90-min amplitude of TF, ULF and ULF01 increased 2- to 3-fold in space in astronauts of Group 1, whereas it decreased in those of Group 2, [Table 3](#tbl0020){ref-type="table"}. During ISS03 as compared to preflight, the BRAC amplitude of TF increased from 154.9 ± 105.0 to 532.7 ± 301.3 msec^2^, or from 3.2 to 11.3% of MESOR (n = 7), that of ULF increased from 117.9 ± 57.5 to 442.4 ± 202.9 msec^2^, or from 4.1 to 15.8% of MESOR (n = 7) and that of ULF01 increased from 124.3 ± 82.8 to 427.6 ± 214.8 msec^2^, or from 8.9 to 31.2% of MESOR (n = 7). In astronauts of Group 2, the 90-min amplitude of ULF01 decreased from 801.6 ± 155.6 before flight to 452.0 ± 239.9 during ISS02, or from 30.8 to less than 20% of the MESOR in space (n = 3), [Table 3](#tbl0020){ref-type="table"}. Two examples of the fitted model to the TF data are shown in [Fig. 1](#fig0005){ref-type="fig"}, comparing the record during ISS03 (right) with the preflight record (left). In one case ([Fig. 1](#fig0005){ref-type="fig"}A), the 90-min amplitude increased from 59.5 to 684.5 msec^2^, with practically no change in the circadian amplitude. In another case ([Fig. 1](#fig0005){ref-type="fig"}B), the 90-min amplitude also increased from 71.4 to 754.5 msec^2^, but it was accompanied by an increase in the 24-hour amplitude from 529.8 to 3196.4 msec^2^. 3.5. Implications of heart rate response to space environment for adaptation to microgravity {#sec0070} -------------------------------------------------------------------------------------------- To better understand the meaning of a difference in HRV response to the space environment, we compared the characteristics of the 4-component model fitted to some HRV endpoints before flight and during ISS03 between Groups 1 and 2. Before flight, the MESOR of TF, ULF and VLF spectral power was statistically significantly lower, on average, in astronauts of Group 1 as compared to those of Group 2, [Table 4](#tbl0025){ref-type="table"}. These differences became smaller during ISS03, to the point of no longer reaching statistical significance, except for TF and VLF spectral power, [Table 4](#tbl0025){ref-type="table"}. In other words, the two groups differed less in space (ISS03) than before flight. Before flight, the BRAC amplitude was found to be much smaller in Group 1 as compared to Group 2, the difference being statistically significant for all considered HRV endpoints, except for LF, [Table 4](#tbl0025){ref-type="table"} (left). During ISS03, the 90-min amplitude increased in Group 1 and mostly decreased in Group 2 (except for LF), so that differences between the two groups were no longer statistically significant after spending several months in space, [Table 4](#tbl0025){ref-type="table"} (right). Similar results were observed for the 24-hour amplitude, and to a lesser extent for the 12-hour and 8-hour amplitudes of these HRV endpoints. These results suggest that the HRV of astronauts in Group 2, but not in Group 1, may have been sufficiently large to be exposed to the space environment. 4. Discussion {#sec0075} ============= Spaceflight dramatically alters cardiovascular dynamics, as illustrated by changes in HRV [@bib0060] and a less negative slope β of the fractal scaling [@bib0045] confirmed herein. Kleitman\'s about 90-min BRAC [@bib0100] was found to be amplified about 3-fold in space, notably among astronauts of Group 1, in keeping with a corresponding increase in ULF01/TF (0.0001--0.0003 Hz, i.e., 55--166 min) and corresponding decreases in ULF02/TF and ULF03/TF. Major changes observed in space all relate to the same frequency range centered around one cycle in about 90 min, including β. Beyond the partly built-in circadian rhythms [@bib0115], there are many other oscillations of different frequencies, including the BRAC, observed in the sleep-wake (REM/NREM) cycle and also in heart rate variability. Some neuropeptides can have more prominent ultradian (with a frequency higher than one cycle per day; e.g., 8-hour periodicity) than circadian changes [@bib0120]. We previously showed that the circadian rhythm persisted in space in HR and β \[[@bib0045], [@bib0060]\]. Herein, we confirm the presence in space of 24-, 12-, and 8-hour components in several HRV endpoints by the fit of a model including 4 anticipated components. The question may be raised, however, whether different daily routines before and during flight (including higher or lower frequency of physical activities) as well as different sleep patterns in space may have contributed to the findings [@bib0125]. Whereas further work is needed to address this question, it should be noted that the space environment had a different effect on astronauts from the 2 groups. Amplitudes of all 4 anticipated components were markedly increased in astronauts of Group 1, whereas they were mostly decreased in astronauts of Group 2. It thus seems unlikely that the daily routine on the ISS fully accounts for the results observed in this study. Unlike short-term (\<24 h) analysis of HRV \[[@bib0125], [@bib0130], [@bib0135]\], transient changes of body movement related to the daily routine were not associated with measurements of long HRV signals, including the ULF and VLF components and the slope β. Aoyagi et al. \[[@bib0050], [@bib0055]\] reported that during both usual daily-routine and constant-routine protocols in healthy men, HRV at frequencies between 0.0033 Hz and 10^−3.5^ Hz (25 sec to 57 min periods) was behavior-independent, possibly reflecting intrinsic mechanisms of the regulatory system. Amaral LAN et al. [@bib0140] also reported that the complexity of heartbeat dynamics showed behavioral-independent features during a constant-routine protocol. As reported previously \[[@bib0125], [@bib0130], [@bib0135]\], however, body movement was lower and the HF component of HRV was higher during sleep than during wakefulness. The less negative slope β in space versus Earth was also seen more prominently during the awake span [@bib0045]. Future studies are thus needed to examine how different daily routines before and during flight, including different sleep patterns in space, may contribute to our findings herein. The presence of the BRAC in HRV endpoints observed herein is supported by different studies in a number of physiological systems. Based on 24-hour polygraphic tracings, Othmer E et al. [@bib0145] inferred that the so-called sleep-dream cycle of human sleep is a general activity pattern of the brain. Bailey D et al. [@bib0150] found regular oscillations with periods of 1--2 hours in their subjects\' oxygen consumption. Orr WC et al. [@bib0155] noted that their subjects\' heart rate showed the same about 90-min periodicity in performance of a complex vigilance task. Hiatt JF and Kripke DF [@bib0160] reported on 90- to 120-min ultradian rhythms in gastric motility. Lavie P and Kripke DF [@bib0165] discerned a cycle of 80--133 min in urine flow of awake subjects. The rhythm in urine flow was, however, clearly out of phase with those of electrolyte concentrations and osmolarity. Lavie P and Scherson A [@bib0170] observed rhythmic variations in subjects\' ability to fall asleep throughout the day. Conversely, an expected variation in vigilance was reported by Okawa M et al., [@bib0175]; the ultradian rhythms in vigilance had periods of 90--120 min. The BRAC may play an important and unique role in keeping the quality of life in space independently of or in conjunction with the circadian rhythm. It is involved in the functioning of the central nervous system which integrates many somatic, visceral, and neurobehavioral functions and manifests itself in the alternation of non-REM and REM sleep. Ultradians may be the basic signature of life [@bib0180]. Effects of space weather are enormous, which have acted as selective forces in humans on Earth and shaped human life as we know it today. Using 61 worldwide populations, Hancock AM et al. [@bib0185] elucidated the genetic basis for adaptation to the climate-mediated selection in a scan of the human genome. They identified genes that are key to the differentiation of brown adipocytes, and genes whose regulation makes a difference in response to ultraviolet radiation [@bib0185]. Among the circadian clock components, cryptochrome may have played a pivotal role in evolution because it coordinates light-induced effects and protects from hazards of ultraviolet radiation [@bib0190]. Brown adipocytes and their cryptochromes may not only be relevant to survival and adaptation, but they may also be targeted by natural selection [@bib0195]. Circadian clocks in brown adipocytes are relevant to mammalian adaptation and the cryptochromes in particular are of key importance because of their evolutionary roots of circadian clocks. Brown adipose tissue expressing BRAC may be an active pacemaker tissue, participating in the arrangement of ultradian [@bib0200] to infradian [@bib0205] oscillations. Circadian clocks may thus be built on properties generating metabolic oscillations in the ultradian range [@bib0190]. Brown adipose tissue may be a site of interaction between metabolic and circadian systems. A non-transcriptional pathway for the metabolic cycle engages the circadian clock, thereby enhancing clock performance [@bib0210]. As cryptochromes are key components of the core of the transcription-translation feedback loops on which circadian clocks are built, the question may thus be raised whether the amplification of the BRAC in space observed herein is a sign of early adaptation to microgravity. 5. Conclusion {#sec0080} ============= Whether the increase in space of the BRAC amplitude is a sign that the intrinsic autonomic regulatory system may start to adapt requires further investigation, as β remains disturbed throughout the 6-month spaceflight. Whether some features of the HRV may indicate suitability for space travel also deserves further work as the BRAC amplification in space was only observed in some but not all astronauts. Most HRV changes observed in space relate to a frequency window centered around one cycle in about 90 min, although astronauts follow regular 24-hour rest-activity and feeding schedules on the ISS. Since the BRAC component is amplified in space for only specific HRV endpoints, it is likely to represent a physiologic response rather than an artifact from the ISS orbit. If so, it may offer a way to help adaptation to microgravity during long-duration spaceflight. Declarations {#sec0085} ============ Author contribution statement {#sec0090} ----------------------------- Kuniaki Otsuka: Conceived and designed the experiments; Analyzed and interpreted the data; Wrote the paper. Germaine Cornelissen, Yutaka Kubo, Mitsutoshi Hayashi, Koichi Shibata, Koh Mizuno: Analyzed and interpreted the data; Wrote the paper. Satoshi Furukawa, Tatsuya Aiba, Hiroshi Ohshima, Chiaki Mukai: Conceived and designed the experiments; Performed the experiments; Wrote the paper. Competing interest statement {#sec0095} ---------------------------- The authors declare no conflict of interest. Additional information {#sec0100} ---------------------- No additional information is available for this paper. Funding statement {#sec0105} ----------------- The JAXA Chronobiology Project was supported by the Japan Aerospace Exploration Agency (KO, YK, MH, NY, KS, TA, SF, HO, CM), Halberg Chronobiology Fund (GC). The authors thank Dr. I. Tayama and S. Ishida from the Space Biomedical Research Group, Japan Aerospace Exploration Agency (JAXA), for cooperation in our study. The authors also acknowledge the cooperation of the astronauts, the engineers, staff and managers of JAXA and NASA. {ref-type="fig"}A), it is also amplified in another case ([Fig. 1](#fig0005){ref-type="fig"}B). A fixed model is used, considering only anticipated periodicities. As such, the model is not optimal for any given record, even if on a group basis it conveys the behavior of components that are the most commonly detected in such records. Because it is a fixed model, the residual variance may exhibit lack of fit. Nevertheless, the amplitude of the about 90-min component is increased during ISS03 in astronauts of Group 1, when it resembles that of astronauts of Group 2.](gr1){#fig0005} ###### Change in characteristics of heart rate variability associated with 6-month mission in space: Numerical results.^\*^ Table 1A Variable Units Target period (range) n Control (Before flight) ISS01 ISS02 ISS03 After flight ---------------------------- ---------- ------------------------- ----------------------- -------- ------------------------- -------- -------- -------- -------------- -------- -------- -------- -------- ------- Time- domain measures HR (beats/min) 24 hours 10 69.9 10.9 66.7 8.5 66.9 7.0 66.6 7.4 69.2 8.9 NN-interval (msec) 24 hours 10 878.2 146.7 914.1 126.4 906.4 97.6 911.9 104.3 880.5 120.9 SDNN (msec) 24 hours 10 132.5 45.2 148.4 29.5 140.1 52.6 151.0 43.2 144.7 43.5 SDANN (5 min) (msec) 24 hours 10 115.8 43.6 129.0 27.0 121.4 46.0 130.0 39.3 125.1 43.2 SDANN (30 min) (msec) 24 hours 10 109.3 44.2 125.2 27.1 117.9 44.7 129.1 38.5 116.8 44.6 TINN (msec) 24 hours 10 571.5 178.9 638.0 144.9 523.5 186.7 552.7 128.7 612.3 146.9 HRVI (--) 24 hours 10 35.7 11.2 39.9 9.1 32.7 11.7 34.5 8.0 38.3 9.2 Triangular Index (TI) (--) 24 hours 10 34.2 10.4 38.3 9.2 30.8 10.8 31.8 7.2 36.8 9.0 Lorenz Plot Length (msec) 24 hours 10 627.9 228.3 690.7 160.7 659.0 284.5 745.0 252.2 707.1 234.9 Lorenz Plot Width (msec) 24 hours 10 54.9 16.5 50.9 15.6 51.5 13.7 61.7 15.9 58.9 15.8 Length/Width ratio (--) 24 hours 10 11.5 2.5 14.5 4.2 13.0 4.9 12.5 4.2 12.5 4.8 SDNN index (30 min) (msec) 30 min 10 72.3 19.1 66.6 16.7 63.9 15.1 68.0 17.6 76.6 17.4 SDNN index (5 min) (msec) 5 min 10 56.5 14.8 53.1 13.2 50.9 11.3 55.1 13.4 58.8 13.4 CVNN (%) 5 min 10 16.3 5.1 17.5 4.9 16.0 4.6 17.3 3.6 17.7 5.9 r-MSSD (msec) 5 min 10 23.9 5.9 23.1 5.9 22.6 5.2 26.4 5.8 24.7 6.6 NN50 (number) 5 min 10 4048.2 2841.3 3603.4 2514.5 3142.4 2605.1 4442.1 2610.0 4226.3 2616.1 pNN50 (%) 5 min 10 4.360 3.536 4.050 3.143 3.430 2.819 5.820 3.594 5.090 4.260 Frequency- domain measures \|β\| (log(msec^2^)/log(Hz)) 90 min (1.7--166 min) 10 1.087 0.130 0.977 0.098 0.910 0.130 0.924 0.095 1.135 0.147 TF-component msec^2^ 90 min (2 sec--166 min) 10 6417.1 3238.0 5932.1 2453.4 5297.5 2806.2 6530.9 3562.3 6897.6 2823.3 ULF-component msec^2^ 90 min (5--166 min) 10 3479.8 1636.4 3255.5 1295.1 2857.4 1982.5 3624.3 2362.4 3815.4 1605.2 ULF01 msec^2^ 90 min (55--166 min) 10 1361.2 775.7 1788.0 747.4 1450.9 1146.7 2080.8 1399.7 1389.3 640.7 ULF02 msec^2^ 36 min (17--55 min) 10 1190.3 561.6 885.2 433.2 849.6 561.5 878.2 520.4 1378.1 548.7 ULF03 msec^2^ 10 min (3--17 min) 10 1360.3 596.7 920.2 522.1 868.0 488.2 1034.7 764.3 1533.5 860.7 VLF-component msec^2^ 5 min (25 sec--5 min) 10 2113.7 1361.6 1928.7 1034.7 1741.5 827.4 2105.8 1211.2 2210.5 1127.5 VLF01 msec^2^ 2 min (50 sec-3.3 min) 10 1177.2 834.2 1114.4 605.5 1002.5 464.8 1245.3 758.5 1209.6 666.5 VLF02 msec^2^ 42 sec (33--50 sec) 10 291.9 185.9 275.2 151.7 250.3 132.3 287.6 147.5 297.8 134.0 VLF03 msec^2^ 20 sec (6.7--33 sec) 10 911.3 425.2 836.9 416.2 773.5 367.4 864.6 389.8 960.4 391.9 LF-component msec^2^ 15 sec (6--25 sec) 10 698.8 316.1 635.8 329.3 595.8 296.1 661.1 306.0 742.9 310.6 HF-component msec^2^ 4.3 sec (2.5--6 sec) 10 116.5 55.1 104.6 60.2 94.9 45.9 127.6 63.7 120.1 52.2 LF/HF ratio (--) 10 6.428 2.711 6.398 1.760 6.305 0.744 5.606 1.761 6.506 2.129 ULF/TF (--) 90 min (5--166 min) 10 0.549 0.079 0.556 0.071 0.511 0.123 0.542 0.091 0.557 0.092 ULF01/TF (--) 90 min (55--166 min) 10 0.207 0.053 0.314 0.078 0.251 0.095 0.310 0.090 0.202 0.070 ULF02/TF (--) 36 min (17--55 min) 10 0.189 0.037 0.145 0.025 0.154 0.051 0.136 0.030 0.204 0.045 ULF03/TF (--) 10 min (3--17 min) 10 0.219 0.035 0.151 0.034 0.164 0.024 0.151 0.034 0.219 0.047 VLF-/TF (--) 5 min (25 sec--5 min) 10 0.316 0.057 0.319 0.064 0.347 0.088 0.323 0.055 0.312 0.065 VLF01/TF (--) 2 min (50 sec-3.3 min) 10 0.173 0.041 0.186 0.043 0.200 0.050 0.189 0.039 0.169 0.042 VLF02/TF (--) 42 sec (33--50 sec) 10 0.044 0.010 0.045 0.013 0.051 0.021 0.045 0.011 0.043 0.013 VLF03/TF (--) 20 sec (6.7--33 sec) 10 0.147 0.045 0.139 0.038 0.159 0.055 0.143 0.055 0.143 0.047 LF-/TF (--) 15 sec (6--25 sec) 10 0.114 0.039 0.106 0.034 0.122 0.044 0.110 0.047 0.111 0.040 HF-/TF (--) 4.3 sec (2.5--6 sec) 10 0.019 0.009 0.018 0.007 0.020 0.008 0.023 0.014 0.018 0.008 ^\*^For definition of HRV endpoints, see [@bib0075]. ###### Comparison of relative HRV endpoints in Space and on Earth.[\*](#tblfn0005){ref-type="table-fn"} Table 1B *Means (10 astronauts)* Space vs. Earth ISS01-03 ----------- ------------------------- -------- -------- ----------------- ---------- -------- -------- ------- ---------- ------- ------- Primary endpoints ULF01 83.36 121.19 81.97 124.52 88.95 86.16 109.23 1.933 NS 3.106 NS ULF02 115.85 85.24 76.54 85.12 137.25 126.55 82.30 6.265 0.001 0.431 NS ULF03 123.56 80.66 75.23 84.86 135.70 129.63 80.25 7.344 \< 0.001 0.924 NS VLF01 97.16 99.97 91.02 107.01 104.83 101.00 99.34 0.250 NS 2.135 NS VLF02 100.69 97.34 90.57 103.60 107.80 104.24 97.17 1.354 NS 2.141 NS VLF03 104.67 94.54 90.20 99.38 111.21 107.94 94.71 2.345 NS 1.327 NS LF 105.48 93.15 90.01 98.99 112.37 108.92 94.05 2.160 NS 1.153 NS HF 103.99 90.11 85.09 112.54 108.27 106.13 95.91 1.121 NS 4.582 NS 

 Secondary endpoints TF 102.32 97.44 83.44 103.19 113.61 107.96 94.69 2.482 NS 3.778 NS ULF 102.95 99.96 78.42 103.52 115.15 109.05 93.97 1.910 NS 2.621 NS VLF 100.96 96.99 88.87 103.20 109.98 105.47 96.36 1.630 NS 2.321 NS ULF/TF 101.16 103.07 93.37 99.96 102.44 101.80 98.80 0.906 NS 1.214 NS  ULF01/TF 81.13 124.29 96.43 119.89 78.25 79.69 113.54 4.376 0.014 2.416 NS  ULF02/TF 114.49 88.38 91.39 82.87 122.87 118.68 87.55 6.199 0.001 0.562 NS  ULF03/TF 121.54 83.79 91.27 83.26 120.15 120.84 86.11 6.945 0.001 1.100 NS VLF/TF 97.60 99.31 107.00 100.06 96.03 96.81 102.12 1.145 NS 0.555 NS  VLF01/TF 93.69 101.89 109.48 103.35 91.59 92.64 104.91 2.137 NS 0.397 NS  VLF02/TF 96.91 100.07 108.81 100.47 93.74 95.32 103.12 1.530 NS 0.621 NS  VLF03/TF 101.92 96.84 107.65 97.09 96.50 99.21 100.53 0.175 NS 1.169 NS LF/TF 102.99 95.35 107.26 96.97 97.42 100.21 99.86 0.038 NS 1.189 NS HF/TF 101.09 92.76 102.21 109.37 94.58 97.83 101.45 0.436 NS 1.312 NS LF/HF 100.55 102.16 103.28 89.54 104.47 102.51 98.33 0.484 NS 1.974 NS HR 102.68 98.28 98.82 98.22 102.00 102.34 98.44 1.793 NS 0.043 NS NN 97.40 101.77 101.19 101.64 98.00 97.70 101.53 1.788 NS 0.035 NS CVRR 94.98 103.46 94.89 102.43 104.23 99.61 100.26 0.119 NS 0.613 NS SDNN 91.44 106.14 96.17 105.38 100.86 96.15 102.57 1.139 NS 1.053 NS r-MSSD 98.74 95.33 93.78 110.11 102.04 100.39 99.74 0.161 NS 4.545 NS NN 105.87 100.09 97.18 96.38 100.48 103.17 97.88 1.544 NS 0.698 NS NN50 104.53 92.78 79.83 113.64 110.59 107.56 94.28 1.093 NS 1.980 NS NN50+ 96.13 92.98 78.64 126.63 105.63 100.88 99.42 0.113 NS 2.541 NS NN50- 95.24 82.15 75.63 137.51 109.47 102.36 98.43 0.257 NS 4.117 NS pNN50 90.37 87.90 77.29 135.50 108.94 99.65 100.23 0.035 NS 3.388 NS pNN50+ 87.99 93.91 79.83 130.65 107.61 97.80 101.46 0.233 NS 2.578 NS pNN50− 90.21 86.15 73.10 140.81 109.74 99.97 100.02 0.002 NS 4.296 NS SDANN5 92.10 107.08 95.96 104.85 100.01 96.05 102.63 1.046 NS 1.187 NS SDANN30 90.01 107.93 96.77 108.65 96.63 93.32 104.45 1.537 NS 1.339 NS SDmean5 102.51 96.81 93.19 100.27 107.22 104.87 96.76 4.004 0.025 3.693 NS SDmean30 103.63 95.80 92.12 97.68 110.77 107.20 95.20 6.551 0.001 1.954 NS N 94.98 99.95 106.92 107.43 90.72 92.85 104.77 2.359 NS 1.181 NS X 98.29 100.10 100.54 98.89 102.18 100.24 99.84 0.184 NS 0.237 NS M 96.69 105.11 98.51 101.23 98.46 97.58 101.61 2.598 NS 3.447 NS TINN 97.65 111.12 88.84 95.94 106.45 102.05 98.63 0.898 NS 6.227 0.048 HRVI 97.64 111.14 88.83 95.95 106.44 102.04 98.64 0.891 NS 6.241 0.047 TI 98.60 112.06 88.28 93.14 107.92 103.26 97.83 1.374 NS 7.098 0.027 Length 90.65 103.84 94.15 108.23 103.14 96.89 102.07 0.823 NS 1.231 NS Width 97.90 91.20 92.98 111.42 106.50 102.20 98.53 0.766 NS 4.756 NS Len/Wid 91.94 113.80 101.01 96.52 96.73 94.34 103.78 1.606 NS 1.371 NS 

 Trend (β) 108.03 97.23 90.33 91.80 112.61 110.32 93.12 4.298 0.016 1.958 NS P-values adjusted for multiple testing, using Bonferroni\'s inequality, considering that 8 different tests were conducted (in 8 independent frequency regions). Secondary endpoints also used the same correcting factor, considering the large correlation among different endpoints, shown here for sake of completeness only (rather than for testing per se). For definition of HRV endpoints, see [@bib0075]. 24-hour mean HRV endpoints expressed as a percentage of 5-session average for each astronaut, then averaged during each session across the 10 astronauts. ###### Individual HRV responses of astronauts.[\*](#tblfn0010){ref-type="table-fn"} Table 2 Subjects Variables units Control (Before flight) ISS01 ISS02 ISS03 After flight -------------- ------------ ------------ ------------------------- ------- ------- ------- -------------- ------- ------- ------- ------- ------ ------ Group 1 Case 1 Heart Rate (b/min) 74.8 10.6 81.5 16.1 75.5 7.3 71.6 12.2 73.5 16.9 r-MSSD (msec) 23.1 8.2 22.6 5.7 23.3 5.3 27.0 7.1 28.9 9.3 pNN50 (%) 3.9 5.9 3.2 2.8 3.4 2.5 5.4 3.8 8.0 6.2 HF-component msec^2^ 127.5 112.9 92.1 54.3 100.4 49.7 127.5 61.1 193.1 103.4 LF/HF ratio (--) 6.0 3.4 6.6 3.0 7.1 2.7 7.3 3.7 4.9 2.5 Case 2 Heart Rate (b/min) 78.8 13.5 67.0 9.2 77.7 9.8 74.8 7.1 78.1 10.0 r-MSSD (msec) 23.3 7.6 29.5 6.7 23.6 6.3 26.1 6.6 23.8 7.4 pNN50 (%) 4.5 4.8 8.4 5.2 4.4 3.8 6.0 5.9 4.8 4.7 HF-component msec^2^ 169.1 114.7 235.0 106.9 162.5 101.9 239.4 170.0 165.1 123.1 LF/HF ratio (--) 6.0 4.1 6.0 3.0 6.5 3.5 5.4 3.1 6.3 4.4 Case 3 Heart Rate (b/min) 89.9 11.9 72.2 14.4 70.9 8.0 78.5 12.5 82.3 7.5 r-MSSD (msec) 22.9 3.9 20.3 6.6 18.2 4.5 32.9 13.7 19.2 4.7 pNN50 (%) 3.0 2.5 2.4 3.7 1.4 2.2 13.1 13.8 1.5 2.0 HF-component msec^2^ 105.3 48.0 86.0 49.3 71.3 34.3 180.1 153.0 98.0 64.0 LF/HF ratio (--) 5.8 2.8 5.8 3.6 6.2 3.8 3.3 2.6 6.2 3.2 Case 4 Heart Rate (b/min) 77.3 10.0 70.3 15.0 64.9 6.2 66.4 14.6 66.9 6.1 r-MSSD (msec) 16.8 4.9 17.8 3.9 19.7 3.7 20.2 4.4 23.0 6.2 pNN50 (%) 1.3 1.9 1.2 1.4 1.5 1.6 1.9 1.8 3.8 3.8 HF-component msec^2^ 55.7 33.2 56.3 27.7 74.0 30.1 69.2 29.3 102.7 66.9 LF/HF ratio (--) 13.4 7.2 10.7 6.1 7.4 3.2 9.2 5.5 8.1 4.6 Case 5 Heart Rate (b/min) 61.6 5.1 59.9 6.7 56.5 9.0 57.2 6.7 62.1 10.6 r-MSSD (msec) 15.9 2.9 11.9 2.2 15.1 3.7 15.2 3.6 13.1 3.8 pNN50 (%) 0.5 0.6 0.2 0.4 0.7 1.2 0.6 1.5 0.3 0.7 HF-component msec^2^ 37.9 15.7 22.0 10.3 31.5 15.3 31.8 17.1 26.2 19.2 LF/HF ratio (--) 4.6 2.7 7.5 4.6 7.5 4.6 6.6 4.7 7.4 4.8 Case 6 Heart Rate (b/min) 69.0 7.4 62.4 6.7 67.7 12.9 63.4 12.4 71.7 12.0 r-MSSD (msec) 19.0 4.4 22.3 7.1 20.4 5.7 31.1 8.5 23.4 5.6 pNN50 (%) 1.8 1.8 3.5 4.1 2.3 2.6 10.2 6.8 4.0 3.6 HF msec^2^ 69.1 38.6 90.6 54.7 84.7 62.6 144.5 88.9 110.2 55.0 LF/HF ratio (--) 10.8 6.4 8.7 5.4 8.9 5.2 5.7 3.3 12.2 7.3 Case 7 Heart Rate (b/min) 64.6 5.4 66.2 10.4 64.5 5.9 62.9 5.6 66.6 6.4 r-MSSD (msec) 21.3 3.8 25.2 7.1 18.1 5.5 22.6 4.6 18.8 3.5 pNN50 (%) 2.2 1.8 5.7 5.4 1.5 2.1 3.2 3.1 1.4 1.3 HF-component msec^2^ 79.1 32.9 85.6 42.3 50.4 25.2 78.2 30.6 62.6 25.5 LF/HF ratio (--) 5.9 3.8 4.7 3.0 6.1 3.7 4.8 2.7 7.5 4.7 Group 2 Case 8 Heart Rate (b/min) 65.8 13.5 65.5 11.2 69.4 15.8 67.7 14.4 64.5 7.2 r-MSSD (msec) 28.8 9.7 21.0 4.8 28.1 4.7 25.2 6.3 36.8 6.0 pNN50 (%) 8.4 9.1 2.4 2.4 6.5 3.7 4.8 4.8 15.3 6.2 HF-component msec^2^ 160.9 129.4 73.4 39.9 91.7 39.5 90.7 48.2 163.1 65.9 LF/HF ratio (--) 8.6 6.9 10.3 7.3 6.7 3.3 8.4 4.9 5.9 2.7 Case 9 Heart Rate (b/min) 67.8 16.5 71.8 19.0 65.2 10.5 66.4 13.0 76.6 23.0 r-MSSD (msec) 33.6 12.5 27.6 8.4 32.9 8.0 32.7 10.6 27.7 10.3 pNN50 (%) 12.2 11.3 7.0 6.4 11.1 8.0 10.9 9.7 7.3 8.8 HF-component msec^2^ 200.7 137.5 147.7 80.0 179.8 92.5 201.3 135.5 123.1 90.0 LF/HF ratio (--) 7.8 4.2 7.5 3.7 7.2 3.1 6.3 2.7 8.9 4.5 Case 10 Heart Rate (b/min) 51.6 8.5 49.8 7.0 57.1 17.9 54.4 13.5 53.1 7.4 r-MSSD (msec) 35.3 8.8 32.9 6.5 26.5 6.3 31.6 7.7 31.7 6.6 pNN50 (%) 13.2 8.8 10.9 6.4 5.0 4.3 9.7 6.7 9.7 6.2 HF-component msec^2^ 161.7 78.7 157.7 55.6 102.4 42.6 119.9 45.2 139.8 60.2 LF/HF ratio (--) 6.7 4.7 5.7 3.6 6.0 3.6 6.3 3.6 6.9 4.7 r-MSSD: square root of mean squared differences of successive NN intervals; pNN50: fraction of consecutive NN intervals that differ by more than 50 ms; HF-component: spectral power centered around 3.6 sec; LF/HF ratio: ratio of low-frequency (LF, centered around 10.5 sec) and high-frequency (HF) spectral power; all indices obtained from 5-min segments, averaged over the entire 24-hour span. Astronauts were grouped in terms of their NN records (see text). Each record contains 254 to 286 values, except for case 8 after return to Earth (N = 70 or 71). ###### Change in relative amplitude of 24-, 12-, 8-, and 1.5-hour components of some HRV endpoints during 6-month mission in space.[\*](#tblfn0015){ref-type="table-fn"} Table 3 Group 1 (N = 7) ---------- ----------------- -------- ------- -------- ------- ------- -------- ------- ------------- ------- ----------- NN  24h-A 8.07 12.52 9.80 12.51 9.85 8.96 11.61 1.940 NS 2.332 NS  12h-A 5.14 5.95 6.26 6.70 6.70 5.92 6.30 0.399 NS 1.749 NS  8h-A 3.74 4.42 3.03 4.31 3.66 3.70 3.92 0.219 NS 0.831 NS  1.5h-A 0.94 1.16 0.99 1.61 1.42 1.18 1.25 0.294 NS 1.977 NS β  24h-A 18.35 20.67 20.86 20.48 19.60 18.98 20.67 0.275 NS 0.250 NS  12h-A 13.31 19.97 20.67 19.06 12.45 12.88 19.90 1.195 NS 0.843 NS  8h-A 13.67 13.35 12.31 12.07 12.25 12.96 12.58 0.380 NS 0.581 NS  1.5h-A 2.07 1.68 1.76 1.99 1.33 1.70 1.81 0.326 NS 0.222 NS TF  24h-A 24.59 56.39 53.79 62.60 33.99 29.29 57.59 3.978 **0.044** 2.326 NS  12h-A 23.94 43.43 44.11 49.68 28.03 25.98 45.74 2.531 NS 1.916 NS  8h-A 18.99 38.89 39.86 41.00 19.76 19.38 39.92 5.149 **0.013** 2.877 0.169  1.5h-A 3.17 8.14 7.76 11.29 4.93 4.05 9.06 3.367 ***0.091*** 3.240 0.106 ULF  24h-A 37.33 97.94 72.79 104.14 49.16 43.25 91.62 3.686 ***0.062*** 2.382 NS  12h-A 33.64 84.49 66.30 88.82 37.76 35.70 79.87 3.417 ***0.085*** 2.644 NS  8h-A 30.91 63.26 54.71 60.19 32.81 31.86 59.39 3.484 ***0.078*** 2.278 NS  1.5h-A 4.06 11.14 8.47 15.80 5.71 4.88 11.80 7.485 **0.002** 4.923 **0.016** ULF01  24h-A 42.91 158.90 95.59 166.67 43.62 43.26 140.39 4.465 **0.026** 2.601 NS  12h-A 41.33 144.99 92.88 145.30 43.58 42.46 127.72 3.302 ***0.098*** 2.188 NS  8h-A 39.90 110.80 74.42 105.45 40.78 40.34 96.89 3.014 0.141 2.145 NS  1.5h-A 8.87 22.71 16.30 31.32 11.55 10.21 23.44 4.706 **0.020** 4.052 **0.040** ULF02  24h-A 44.92 61.12 70.78 51.16 71.49 58.20 61.02 0.549 NS 0.742 NS  12h-A 44.73 55.49 58.22 48.84 49.83 47.28 54.18 0.778 NS 0.426 NS  8h-A 34.01 32.05 47.92 35.97 51.07 42.54 38.65 0.655 NS 0.183 NS  1.5h-A 4.44 4.71 4.79 6.13 4.76 4.60 5.21 0.917 NS 1.647 NS ULF03  24h-A 62.70 36.69 40.41 49.54 67.57 65.13 42.22 2.574 NS 0.772 NS  12h-A 41.72 32.60 26.97 28.96 34.97 38.35 29.51 1.480 NS 1.653 NS  8h-A 29.02 19.52 26.48 33.48 26.66 27.84 26.49 0.178 NS 0.529 NS  1.5h-A 3.01 2.01 2.38 3.01 2.50 2.75 2.47 0.621 NS 0.008 NS ULF/TF  24h-A 17.86 28.05 20.27 18.86 14.31 16.09 22.39 1.938 NS 0.228 NS  12h-A 16.58 25.73 20.24 22.24 15.18 15.88 22.74 1.659 NS 1.168 NS  8h-A 16.19 15.74 15.62 13.76 11.14 13.67 15.04 0.514 NS 0.557 NS  1.5h-A 3.97 4.83 4.87 5.28 3.75 3.86 4.99 1.753 NS 0.869 NS ULF01/TF  24h-A 45.30 68.04 31.24 71.25 88.99 67.14 56.84 0.507 NS 1.408 NS  12h-A 32.19 67.20 41.99 59.94 27.71 29.95 56.38 2.437 NS 1.721 NS  8h-A 34.96 41.44 30.34 47.92 29.27 32.11 39.90 0.950 NS 1.299 NS  1.5h-A 10.25 10.52 11.04 14.97 7.09 8.67 12.18 2.771 NS 1.772 NS ULF02/TF  24h-A 35.43 29.49 33.98 18.14 30.42 32.93 27.20 1.266 NS 3.291 0.100  12h-A 33.34 30.88 26.52 28.13 27.31 30.32 28.51 0.319 NS 0.646 NS  8h-A 15.88 15.95 18.20 27.22 33.30 24.59 20.45 0.963 NS 1.769 NS  1.5h-A 6.22 6.01 5.94 6.98 7.83 7.02 6.31 0.658 NS 0.424 NS ULF03/TF  24h-A 53.53 24.10 26.26 41.43 39.95 46.74 30.60 3.376 **0.090** 1.266 NS  12h-A 26.80 22.56 13.09 16.51 18.06 22.43 17.39 1.215 NS 1.599 NS  8h-A 16.01 13.82 14.19 17.23 23.01 19.51 15.08 1.052 NS 0.363 NS  1.5h-A 5.33 4.07 5.23 6.62 4.51 4.92 5.31 0.491 NS 1.837 NS VLF  24h-A 29.89 41.13 38.29 42.63 34.64 32.27 40.68 0.859 NS 1.013 NS  12h-A 20.71 24.11 20.84 29.59 21.58 21.15 24.85 0.964 NS 1.102 NS  8h-A 17.50 26.51 25.08 33.56 10.67 14.08 28.38 2.857 NS 1.621 NS  1.5h-A 9.68 17.06 14.18 17.14 11.36 10.52 16.13 1.717 NS 1.436 NS LF  24h-A 18.12 15.07 18.21 24.30 27.81 22.97 19.19 1.267 NS 0.760 NS  12h-A 14.30 18.91 13.14 17.69 24.28 19.29 16.58 0.729 NS 0.639 NS  8h-A 19.79 14.11 11.82 17.08 22.48 21.14 14.34 1.485 NS 0.524 NS  1.5h-A 7.24 7.96 5.75 8.92 5.88 6.56 7.54 0.802 NS 0.952 NS HF  24h-A 26.30 27.31 26.59 55.76 36.39 31.35 36.56 0.482 NS 1.266 NS  12h-A 18.17 17.89 20.99 38.32 30.69 24.43 25.73 0.152 NS 1.794 NS  8h-A 16.34 17.95 16.92 25.72 18.22 17.28 20.20 0.635 NS 1.502 NS  1.5h-A 6.79 6.77 6.85 10.87 12.54 9.67 8.16 0.647 NS 1.358 NS LF/HF  24h-A 24.67 24.76 14.48 26.18 23.65 24.16 21.81 0.564 NS 0.289 NS  12h-A 11.93 15.23 10.84 15.63 12.33 12.13 13.90 0.596 NS 1.087 NS  8h-A 10.44 13.20 13.98 10.21 8.82 9.63 12.46 1.959 NS 0.054 NS  1.5h-A 9.06 11.16 5.31 11.39 11.76 10.41 9.29 1.775 NS 0.875 NS Group 2 (N = 3) ---------- ----------------- ------- ------- -------- ------- ------- ------- -------- ------------- ------- ------------- NN  24h-A 19.82 15.86 17.48 18.32 17.99 18.91 17.22 0.621 NS 0.499 NS  12h-A 5.80 8.78 9.45 8.40 8.80 7.30 8.88 1.113 NS 0.777 NS  8h-A 3.61 4.28 4.35 3.83 4.14 3.88 4.15 0.271 NS 0.129 NS  1.5h-A 1.06 1.62 1.90 2.40 1.90 1.48 1.97 1.907 NS 2.546 NS β  24h-A 37.55 24.94 30.95 35.59 31.28 34.42 30.49 0.406 NS 0.210 NS  12h-A 23.78 29.58 30.72 27.21 17.56 20.67 29.17 1.394 NS 0.486 NS  8h-A 19.68 17.39 15.66 16.70 18.84 19.26 16.58 0.455 NS 1.043 NS  1.5h-A 2.24 2.11 1.86 1.67 1.75 1.99 1.88 0.317 NS 0.557 NS TF  24h-A 60.22 31.50 51.61 39.07 35.60 47.91 40.73 0.760 NS 1.368 NS  12h-A 48.94 17.76 45.86 50.37 33.90 41.42 38.00 0.338 NS 0.043 NS  8h-A 47.92 26.41 35.01 33.84 18.12 33.02 31.75 0.916 NS 7.198 **0.002**  1.5h-A 9.80 3.28 5.68 7.49 4.73 7.27 5.48 2.802 0.186 0.732 NS ULF  24h-A 87.09 53.23 73.05 58.51 45.77 66.43 61.60 0.283 NS 0.861 NS  12h-A 75.45 44.29 65.30 71.00 38.67 57.06 60.20 0.238 NS 0.128 NS  8h-A 76.57 46.30 55.71 63.55 35.12 55.85 55.19 0.085 NS 2.082 NS  1.5h-A 13.05 8.83 8.23 5.60 6.02 9.54 7.55 1.275 NS 3.629 ***0.066*** ULF01  24h-A 108.64 71.78 88.03 96.08 32.58 70.61 85.30 0.967 NS 0.375 NS  12h-A 120.47 64.16 92.57 121.20 57.76 89.12 92.64 0.269 NS 0.050 NS  8h-A 114.42 72.49 91.25 107.47 59.20 86.81 90.40 0.182 NS 0.339 NS  1.5h-A 30.75 17.55 16.54 12.27 12.29 21.52 15.45 2.639 NS 6.623 **0.003** ULF02  24h-A 115.88 41.77 77.19 41.78 79.06 97.47 53.58 3.810 ***0.053*** 4.063 **0.040**  12h-A 103.51 39.62 54.65 41.31 47.27 75.39 45.19 4.227 **0.033** 3.225 0.108  8h-A 64.38 33.04 41.18 37.45 36.76 50.57 37.22 2.777 NS 2.236 NS  1.5h-A 10.37 5.16 7.12 3.82 5.24 7.81 5.37 1.047 NS 2.217 NS ULF03  24h-A 65.84 32.72 48.60 36.56 44.94 55.39 39.29 0.675 NS 0.810 NS  12h-A 79.34 15.71 44.70 43.96 26.36 52.85 34.79 1.003 NS 0.886 NS  8h-A 55.65 21.07 21.60 35.66 26.79 41.22 26.11 1.023 NS 0.677 NS  1.5h-A 2.73 3.12 1.96 2.71 2.02 2.37 2.60 0.483 NS 0.008 NS ULF/TF  24h-A 24.97 22.74 29.37 31.76 29.00 26.99 27.96 0.107 NS 0.958 NS  12h-A 23.48 26.92 26.49 26.60 15.56 19.52 26.67 3.320 ***0.096*** 0.328 NS  8h-A 23.18 20.18 17.73 17.54 19.29 21.23 18.48 0.726 NS 1.441 NS  1.5h-A 3.93 4.16 8.16 5.99 6.37 5.15 6.11 0.615 NS 0.676 NS ULF01/TF  24h-A 42.32 44.30 48.37 50.48 31.22 36.77 47.72 0.776 NS 1.437 NS  12h-A 47.50 46.01 42.67 58.27 39.76 43.63 48.98 0.904 NS 1.354 NS  8h-A 50.49 55.09 50.52 49.11 49.18 49.83 51.57 0.267 NS 0.354 NS  1.5h-A 9.85 11.57 18.14 11.70 9.94 9.90 13.81 1.159 NS 0.450 NS ULF02/TF  24h-A 30.10 29.26 41.42 27.29 54.72 42.41 32.66 1.034 NS 0.825 NS  12h-A 56.17 26.10 26.97 25.47 35.02 45.60 26.18 2.146 NS 2.269 NS  8h-A 22.72 13.42 10.26 17.50 42.38 32.55 13.73 13.648 **\<0.001** 1.159 NS  1.5h-A 8.32 4.50 5.87 7.41 5.46 6.89 5.93 0.276 NS 0.167 NS ULF03/TF  24h-A 28.57 13.69 19.92 8.49 36.54 32.55 14.03 2.902 NS 2.763 NS  12h-A 24.10 13.50 16.68 19.87 24.19 24.14 16.69 1.471 NS 0.489 NS  8h-A 16.60 21.78 19.32 18.71 16.52 16.56 19.94 1.459 NS 0.378 NS  1.5h-A 8.47 3.22 5.00 8.68 5.01 6.74 5.63 1.081 NS 0.192 NS VLF  24h-A 67.35 36.26 43.68 58.98 44.16 55.76 46.31 0.654 NS 0.982 NS  12h-A 43.61 25.58 22.78 46.55 33.76 38.68 31.64 0.948 NS 0.219 NS  8h-A 22.40 23.38 11.35 30.15 20.26 21.33 21.63 0.052 NS 1.717 NS  1.5h-A 15.94 8.09 13.54 13.61 14.55 15.25 11.75 3.172 0.116 2.117 NS LF  24h-A 31.02 20.76 29.00 38.00 29.65 30.34 29.25 1.026 NS 1.241 NS  12h-A 23.15 14.79 15.99 23.03 23.85 23.50 17.94 2.142 NS 0.018 NS  8h-A 24.38 11.74 11.28 15.81 18.18 21.28 12.95 2.199 NS 2.031 NS  1.5h-A 9.27 6.77 12.33 10.12 6.35 7.81 9.74 1.401 NS 0.185 NS HF  24h-A 62.47 24.21 30.45 42.66 35.59 49.03 32.44 1.170 NS 1.314 NS  12h-A 38.30 20.22 16.52 21.89 34.08 36.19 19.55 3.440 ***0.083*** 1.211 NS  8h-A 22.78 16.29 13.50 11.87 14.90 18.84 13.89 1.065 NS 1.861 NS  1.5h-A 13.45 5.53 3.74 8.55 4.95 9.20 5.94 2.947 0.154 3.345 ***0.093*** LF/HF  24h-A 32.88 14.76 11.37 7.33 15.48 24.18 11.15 4.222 **0.033** 9.487 **\<0.001**  12h-A 15.90 24.65 7.34 8.31 10.68 13.29 13.43 0.079 NS 1.407 NS  8h-A 11.69 9.97 12.10 6.25 19.31 15.50 9.44 1.242 NS 1.233 NS  1.5h-A 10.80 10.41 12.23 6.91 9.03 9.91 9.85 0.018 NS 0.535 NS Amplitudes expressed as a percentage of MESOR, P-values adjusted for multiple testing, considering 6 different frequency regions (ULF01, ULF02, ULF03, VLF, LF, and HF). Based on results from [Tables 1A, 1B and 2](#tbl0005){ref-type="table"}, significant results were anticipated to be found in the ULF rather than in other spectral regions. NN: normal-to-normal intervals; β: slope of fractal scaling; TF: total spectral power; ULF, VLF, LF and HF: spectral power in ultra-low, very low, low, and high frequency regions of the spectrum. Non-sinusoidal waveform may occasionally be associated with overfit (A \> 100%). ###### Characteristics of model of 4 anticipated components fitted to some HRV endpoints compared between astronauts whose HR did or did not decrease in space.[\*](#tblfn0020){ref-type="table-fn"} Table 4 Control (Before flight) ISS03 ------------------- ------------- ------------------------- --------- -------- -------- ------------ ------------ -------- --------- -------- -------- ------------ ---- MESOR NN-interval 837.1 102.1 1021.9 143.6 −2.35 NS 904.1 97.6 1002.0 134.5 −1.31 NS TF 4674.1 1216.8 12924.0 2750.1 −6.90 **0.0005** 5040.0 1631.2 10287.5 3043.6 −3.66 **0.0320** ULF 2686.4 798.7 7229.4 1661.5 −6.09 **0.0015** 2893.6 1138.2 5478.2 2099.0 −2.60 NS ULF01 971.8 443.0 2161.8 632.8 −3.47 **0.0425** 1744.5 1054.8 2889.3 1064.6 −1.57 NS VLF 1337.1 337.0 4536.4 1425.1 −6.02 **0.0015** 1431.4 348.7 3687.2 894.7 −6.06 **0.0015** LF 572.7 220.4 1067.8 268.3 −3.08 0.0760 575.4 291.2 853.1 302.4 −1.37 NS HF 92.4 45.4 174.6 23.4 −2.91 0.0985 122.7 70.5 137.3 56.9 −0.31 NS 

 24-hour Amplitude TF 1182.7 493.5 6430.0 2363.0 −6.05 **0.0015** 3120.4 2248.0 4525.0 3652.8 −0.76 NS ULF 1066.3 570.5 5055.7 1652.0 −6.01 **0.0015** 2947.5 1944.6 3891.7 3382.3 −0.57 NS ULF01 566.9 444.9 2885.1 974.0 −5.41 **0.0030** 2322.2 1631.2 2698.9 2172.9 −0.31 NS VLF 407.2 220.1 2514.0 1794.6 −3.33 0.0520 580.3 385.9 2149.4 1842.9 −2.32 NS LF 116.7 84.9 297.9 209.0 −2.05 NS 126.6 93.9 377.4 329.6 −1.98 NS HF 30.8 34.3 88.8 41.3 −2.32 NS 61.5 63.7 65.0 58.0 −0.08 NS 

 12-hour Amplitude TF 1255.5 1006.9 6072.5 7806.4 −1.75 NS 2558.4 1965.8 5357.6 1545.4 −2.17 NS ULF 1006.0 784.8 5433.8 6761.2 −1.86 NS 2601.2 1900.6 4454.7 2538.6 −1.29 NS ULF01 575.4 512.8 3333.7 2524.9 −2.99 0.0870 2054.6 1686.9 3336.2 2088.9 −1.03 NS VLF 287.0 219.6 1662.6 964.5 −3.85 **0.0246** 385.1 184.9 1731.1 1259.6 −3.00 0.0851 LF 73.6 59.9 214.6 110.8 −2.69 NS 86.3 56.3 231.2 209.1 −1.82 NS HF 20.2 25.2 53.7 21.8 −1.99 NS 42.3 33.2 34.0 32.3 0.37 NS 

 8-hour Amplitude TF 924.6 503.4 5448.3 4121.8 −3.11 0.0720 2105.5 1417.6 3907.8 3526.6 −1.22 NS ULF 898.5 423.2 4904.8 3394.5 −3.34 0.0510 1834.5 1428.1 4036.8 2549.7 −1.80 NS ULF01 545.8 258.6 3025.5 1192.0 −5.64 **0.0025** 1507.4 1222.9 2962.5 1889.2 −1.49 NS VLF 251.7 164.4 847.5 445.7 −3.27 0.0570 455.0 247.3 1133.0 633.0 −2.57 NS LF 111.0 72.6 229.6 150.4 −1.75 NS 87.4 29.8 150.0 93.6 −1.70 NS HF 16.3 16.9 32.1 7.5 −1.53 NS 28.9 23.8 18.4 17.8 0.67 NS 

 90-min Amplitude TF 154.9 105.0 1063.1 549.0 −4.55 **0.0095** 532.7 301.3 872.3 738.1 −1.09 NS ULF 117.9 57.5 790.4 292.0 −6.32 **0.0010** 442.4 202.9 303.2 84.7 1.12 NS ULF01 124.3 82.8 801.6 155.6 −9.28 **0.0001** 427.6 214.8 303.0 74.2 0.95 NS VLF 126.0 77.2 629.5 554.7 −2.56 NS 236.3 121.3 544.1 565.3 −1.48 NS LF 38.6 31.7 73.2 33.1 −1.57 NS 48.3 28.6 86.4 63.8 −1.36 NS HF 6.0 3.7 19.4 9.1 −3.39 **0.0470** 12.9 14.8 12.8 9.6 0.04 NS NN: Normal-to-normal inter-beat interval; TF: Total spectral power; ULF: Ultra low frequency spectral power (0.0001--0.003 Hz). ULF01: ULF band-1 (0.0001--0.0003 Hz); VLF: very low frequency spectral power (0.005--0.02 Hz). LF: low frequency spectral power (0.04--0.15 Hz); HF: high frequency spectral power (0.15--0.40 Hz). P-values adjusted for multiple testing, using Bonferroni inequality (considering 5 tests per endpoint: MESOR, amplitude of each of 4 anticipated components). MESOR: Midline Estimating Statistic Of Rhythm, a rhythm-adjusted mean.

Introduction {#s1} ============ Plant cell walls are a terrifically important source of raw material for food, fuel, and industrially chemicals ([@B8]). In addition, they are responsible for not only conferring a definitive shape to the cell and enabling overall growth, but it also provides material transportation and protection of the cells\' inner contents. Plant cell walls are rich in complex biopolymers carbohydrates that comprise cellulose, hemicellulose, and pectin ([@B12]) as well as lignin ([@B45]). Cellulose, the linear (1→4)-β-D-glucan, is an important structural and functional component of both primary and secondary cell walls ([@B14]). Plant primary wall surrounds growing and dividing plant cells while the secondary wall provides structural support to the xylem and the plant body ([@B42]; [@B55]). In most terrestrial plants, cellulose is thought to be synthesized at the plasma membrane by rosette cellulose synthase complexes which consist of cellulose synthase proteins (CESA) ([@B26]). The *CesA* genes belong to membrane-bound glycosyltransferase family II (GT-2) enzymes ([@B43]). These enzymes are characterized by two domains (A and B domains) which possess the D, D, and D, QxxRW motif, respectively ([@B40]). The first plant *CesA* gene was successfully identified from cotton, by screening its expressed sequence tags ([@B35]). In *Arabidopsis*, at least three CesA subunits encoded by the *AtCesA1*, *AtCesA3*, and one of the *AtCesA6*-related genes (*AtCesA2*, *AtCesA5*, *AtCesA6*, or *AtCesA9*) are required for cellulose biosynthesis in primary cell walls, while *AtCesA4*, *AtCesA7*, and *AtCesA8* are required for secondary cell walls formation ([@B48]; [@B28]). In aspen (*Populus tremuloides*), the three *PtrCesA1*, *PtrCesA2*, and *PtrCesA3* are highly expressed during secondary cell wall enriched xylem tissues ([@B18]). *GhCesA* gene identification revealed that *CesA2* was a predominant gene for secondary cell wall formation ([@B23]). [@B33] have shown that the phylogenetic and expression analysis of three loblolly pine *CesA* genes representing they are orthologous to the *CesA* genes in angiosperms which is responsible for cellulose synthesis in the secondary cell walls. These data suggested that these three genes *PtCesA1*, *PtCesA2*, and *PtCesA3* have been linked to secondary xylem development in gymnosperms *Pinus taeda*. [@B11] observed that the knockout of *AtCesA2* caused severe defects in cell wall formation and microtubule orientation that led to abnormal plant growth and development. Furthermore, [@B11] concluded that cellulose biosynthesis was needed for the microtubule orientation. It is noteworthy that microtubule orientation plays a critical role in controlling cell expansion and elongation. Correspondingly, mutations in *CesA* genes are known to loss of impair cellulose synthesis. For example, *cesa5/cesa6* double mutants were seedling lethal ([@B13]), a mutation in *CesA6* -related genes (*CesA2*, *CesA5*, and *CesA9)* revealed only a mild phenotype ([@B41]; [@B7]) and *cesa2/cesa6/cesa9* triple mutants showed pollen lethality ([@B36]). *Populus* plants are versatile and semi-evergreen forest trees with a wide distribution in northern China ([@B24]). Their ease and simplicity of clonal propagation, rapid growth, and small genome size have made *Populus* tree species a well-established model organism for woody plant research ([@B3]; [@B6]; [@B52]). Although the functions of CesA proteins are well studied in plants ([@B2]), their genetic manipulation to enhance cellulose production, especially in timber trees, has remained demanding. For example, earlier overexpression of *CesA* genes has not resulted in improved plant growth in *Arabidopsis*, barley, and poplar ([@B54]; [@B19]; [@B47]). In this work, the *pBI121:CesA2* binary vector was constructed and introduced to the poplar hybrid clone "Nanlin895" (*Populus deltoides* × *Populus euramericana*) *via* the *Agrobacterium-*mediated transformation system. We found that *PmCesA2* overexpression in *Populus* influenced its secondary cell wall thickening as well as morphological and physiological traits. Analysis of molecular and morphological data and chemical composition of cell walls in transgenic poplars indicated a significantly increased *PmCesA2* transcript abundance along with cell wall thickening. Materials and Methods {#s2} ===================== Source Plant {#s2_1} ------------ The biennial *Pinus massoniana* tree was propagated in the greenhouse of Nanjing Forestry University (NFU), in Jiangsu Province, China. Different tissues samples were directly frozen in liquid nitrogen and kept at -80°C before the RNA extraction. Total RNA was isolated from each sample using RNAprep Pure Plant Kit (Polysaccharides & Polyphenolics-rich) (Tiangen Biotech, Beijing, China) by following the manufacturer\'s instructions. Cloning of the *PmCesA2* Open Reading Frame {#s2_2} ------------------------------------------- *PmCesA2* was cloned from a cDNA library constructed of RNA isolated from the needle tissue of *P. massoniana* using the Prime Script 1st Strand cDNA Synthesis Kit (Takara, Dalian, China). A pair of primers was designed according to the full-length coding region of the *CesA2* gene sequence of *P. taeda* (GenBank: AY789651.1). The PCR products were cloned into the pEASY-T1 (Transgen, Beijing, China) vector and transformed into *E. coli* DH5α, and then sequenced. All of the primers used in these assays are listed in [**Supplementary Table S2**](#SM4){ref-type="supplementary-material"}. Sequence and Phylogenetic Analyses {#s2_3} ---------------------------------- A total of 60 putative CesA protein sequences from various plant species were aligned using the ClustalW2 (<http://www.ebi.ac.uk/Tools/clustalw2/index.html>). A phylogenetic tree of CesAs protein family members of (*Arabidopsis thaliana*), (*Zea mays*), (*Oryza sativa*), (*Eucalyptus grandis*), (*Betula luminifera*), (*Populus trichocarpa*), (*P. taeda*), (*Pinus radiata*), (*Cunninghamia lanceolata*), and (*Picea glauca*) was constructed by the neighbor-joining method using MEGA7 with 1,000 replicates for the bootstrap analysis, and a 50% cutoff value ([@B21]). *Agrobacterium*-Mediated Transformation of (P. Deltoides × P. Euramericana 'Nanlin895') {#s2_4} --------------------------------------------------------------------------------------- The binary vector pBI121 plasmid harboring the desired *CesA2* gene, where *PmCesA2* is under the control of the CaMV 35S constitutive promoter, was introduced into *Agrobacterium tumefaciens* strain EHA105 using the freeze--thaw method ([@B16]). The *P. deltoides × P. euramericana* 'Nanlin895' leaf disks were inoculated with an infective suspension (OD~600~ = 0.7) of regenerated *A. tumefaciens*, under gentle shaking of 200 rpm for 1 h. Then, the leaf disks were dried using sterile paper towels and co-cultivated on MS medium ([@B32]) with 0.5 mg/L N-6-benzyladenine (6-BA), 0.004 mg/L thidiazuron (TDZ), 6 g/L agar, 25 g/L sucrose, 200 μM acetosyringone (AS), pH 5, and incubated in the dark at 28°C for 2 days ([@B30]). This was followed by transferring the leaf disks to MS medium supplemented with 0.5 mg/L 6-BA, 0.004 mg/L TDZ, 6 g/L agar, 25 g/L sucrose, 400 mg/L cefotaxime, and 50 mg/L kanamycin, pH 5.8, under 16/8 h light/dark conditions at 23 ± 1°C in a phytotron to screen for the putative transformant explants. Afterward, the selected shoots were transferred to half-strength MS rooting medium; then, transferred to soil and propagated for complementary experiments. All the transgenic and wild type plants were acclimated and grown in the greenhouse at 18--23°C, 60% humidity, and with 18 h of light and 6 h of dark daily at the NFU. Plant Height and Biomass Measurements {#s2_5} ------------------------------------- Height from the basal stem to the peaks and stem diameter from 5 cm above the soil of 3-month‐old poplars were measured from each transgenic and WT lines. Fresh weight was immediately measured after sample collection. Then, the dry weight of the same plant material was determined after drying at 80°C for 48 h. Extraction of Chlorophyll Pigments {#s2_6} ---------------------------------- Total chlorophyll (TChl) content from the leaves of WT and transgenic plants was measured spectrophotometrically, by using 0.1 g of tissue ground in a pre-chilled mortar and pestle and extracted with 80% acetone (10 ml). After centrifugation at 10,000×*g*, the absorbance of a given extract was recorded at 663.8, 646.8, and 470.0 nm. The concentrations of chlorophyll a and b, and of total chlorophyll, were calculated following [@B25]. Determination of Cell Wall Composition {#s2_7} -------------------------------------- The 10th to 13th internodes of 3-month-old WT and transgenic plants used for cell wall composition analysis. To determine the cellulose content, 100-mg dried samples were degraded with a mixture of nitric and acetic acid (1:8, v/v, 30 min, 100°C) and centrifuged followed by dilution with 60% H~2~SO~4~. Cellulose levels were measured using a cold anthrone reagent at 620 nm ([@B17]) and the determination of lignin content carried out as previously described ([@B53]). The percentages of cellulose and lignin content were then averaged for three biological and technical replicates experiments. To determine the monosaccharide composition, 5 mg of the extract-free samples were extracted with 50 μl of sulfuric acid (72% w/w) at 37°C for 60 min, diluted with 4% H~2~SO~4~, autoclaved for 60 min, then allowed to cool to room temperature and an aliquot was neutralized with CaCO~3~. Analysis of monosaccharide composition was done using high performance liquid chromatography ([@B44]). PCR and Real-Time Quantitative PCR {#s2_8} ---------------------------------- Leaf tissues of all transgenic lines were collected for genomic DNA extraction from 1-month-old transgenic plants, by using the DNeasy Plant Mini Kit (Qiagen, Germany) following the manufacturer\'s instructions. The ensuing genomic DNA from each line was then used for PCR to confirm the integration of *PmCesA2*. RNAprep Pure Plant Kit (Polysaccharides & Polyphenolics-rich) (Tiangen Biotech, Beijing, China) was used to extract RNA from the stem segment of WT and transgenic plant lines. The RNA was then used as a template in a reverse transcription reaction to produce cDNA, following the instructions of the PrimeScript RT Reagent Kit (Perfect Real Time) (TaKaRa Biotechnology, Dalian, China). QRT-PCR was used to assess the copy number and relative expression level of the *PmCesA2* gene (2^−ΔΔCT^) in the transgenic and WT lines using an ABI quantitative real-time RT-PCR system (Applied Biosystems, USA) and the SYBR Green PCR Master Mix according to the manufacturer\'s instruction. The relative expression levels of related *PmCesA2* genes were determined by same method. For the standard curve method cDNA was diluted (1,000-, 500-, 250-, 125-, and 62.5-fold) and two *PmCesA2* primers were used to amplify a product of 130 bp, and expression of the housekeeping gene *β-actin* primers was used for normalization expression to verify the real-time quantitative PCR reaction. All of the primers are listed in [**Supplementary Table S2**](#SM4){ref-type="supplementary-material"}. Southern Blot Analysis {#s2_9} ---------------------- Genomic DNA was extracted using cetyltrimethylammonium bromide method from WT plants and three transgenic poplar lines ([@B37]). Approximately 10 μg of total genomic DNA was digested with EcoRI restriction enzyme, separated on a 0.8% agarose gel at 25 V overnight, and transferred onto Hybond N+ membrane. A 404-bp digoxin-labeled CaMV 35S was used as a probe for hybridization according to the instruction manual (DIG High Prime DNA Labeling and Detection Starter Kit I, Roche). Primers used for DIG labeling of CaMV 35S are listed in [**Supplementary Table S2**](#SM4){ref-type="supplementary-material"}. Scanning Electron Microscopy {#s2_10} ---------------------------- The *Populus* stem (10th internode) of 3‐month‐old transgenic plants and the wild type were used for scanning electron microscopy (SEM), according to the previously described protocol by [@B100] and the image analysis software IMAGEJ (<https://imagej.nih.gov/ij/>) was employed for quantifying morphological parameters of xylem cells (μm) and wall thickness. Statistical Analysis {#s2_11} -------------------- All data for measured height, stem diameter, number of leaves, fresh weight, dry weight, chlorophyll content, carbohydrate content, cellulose, and lignin content were analyzed using the Student\'s *t*-test calculated in Microsoft Excel. Three biological replications with three technical replicates were performed each experiment. A one-way analysis of variance was applied to determine the significance of differences at *p* \< 0.05. Results {#s3} ======= Cloning and Phylogenetic Tree Analysis {#s3_1} -------------------------------------- A 3,147 bp cDNA fragment was obtained by RT-PCR using the primers derived from the *CesA2* gene of *P. taeda*, encoding a protein of 1,057 amino acid residues, and having a molecular weight of 119.767 kDa and an isoelectric point of 8.4. Sequence analysis showed 98% similarity between the reported *CesA2* gene (AY789651.1) of *P. taeda* and *P. massoniana*. We used the 60 CESA protein sequences from plant species, *A. thaliana*, *P. trichocarp*a, *P. taeda*, and *Z. mays* to generate a phylogenetic tree ([**Figure 1**](#f1){ref-type="fig"}). Protein sequence information was collected from NCBI database. To identify the species of origin for each CESA, the corresponding species name was included before each sequence name: *At*, *A. thaliana*; *Zm*, *Z. mays*, *Pt*, *P. trichocarpa*; *Pta*, *P. taeda*; *Pr*, *P. radiata*; *Cl*, *C. lanceolata*; *Pg*, *P. glauca*; *Eug*, *E. grandis*; *Os*, *O. sativa*; *Pm*, *P. massoniana*; and *Bl*, *B. luminifera.* Phylogenetic tree shows that *PmCesA2* in pine is ortholog of *PtCesA2* in poplar related secondary wall and the results confirmed the same function of these genes. All information about plant species and gene accession number are placed in [**Supplementary Table S1**](#SM3){ref-type="supplementary-material"}. {#f1} Tissue-Specific Expression Analysis of *PmCesA2* {#s3_2} ------------------------------------------------ The quantitative real-time expression analyses of RNAs were performed to investigate the *PmCesA2* gene expression patterns in various tissues of *P. massoniana*. These RT-PCR results demonstrated that *PmCesA2* gene was expressed in all examined plant tissues ([**Figure 2A**](#f2){ref-type="fig"}). In addition, the quantitative PCR ([**Figure 2B**](#f2){ref-type="fig"}) analysis showed a high level of expression of *PmCesA2* in the stem and root than in the needle. {#f2} Analysis of Expression and Integration of *PmCesA2 in* Nanlin895 Poplars {#s3_3} ------------------------------------------------------------------------ Ten transgenic poplar lines were verified by PCR analysis using the *PmCesA2*-specific primers. This indicated that the *PmCesA2* gene had been integrated into the genomes of 10 independent transgenic plant lines ([**Figure 3A**](#f3){ref-type="fig"}). Relative expression levels of the *PmCesA2* gene were analyzed in 2-month-old stems of these verified transgenic lines, for which the quantitative real-time PCR analysis showed notable variation ([**Figure 3B**](#f3){ref-type="fig"}). Specifically, *PmCesA2* expression levels in the L15 and L7 were relatively higher than in the other transgenic line and the wild type. The transgene copy numbers of *PmCesA2* were determined *via* Real time PCR based on formula *X* = *Y* -- intercept/slope degrees (*X* = copy number, *Y* = Ct) ([@B31]). The results revealed that the average gene copy number of *PmCesA2* in transgenic plants is 7.45 with a slope of −3.36 and efficiency 0.994. Southern blot hybridization confirmed the stable integration of the *PmCesA2* into the genome of transgenic lines. Southern blotting analysis ([**Supplementary Figure S1**](#SM1){ref-type="supplementary-material"}) revealed that the transgene had integrated into the genome stably at two to three copies per genome. No bands were detected in the WT lanes. {#f3} Growth and Morphological Characteristics in *PmCesA2* Transgenic Poplars {#s3_4} ------------------------------------------------------------------------ To investigate whether overexpression of *PmCesA2* could improve plant growth, we monitored the individual growth of three transgenic plants from each line and untransformed controls. Three months after planting into the soil, significant growth phenotype differences were observed between plants overexpressing *PmCesA2* and the wild type (WT). The former showed a fast‐growing phenotype with increased plant height, stem diameter, and leaf number ([**Figure 4A**](#f4){ref-type="fig"}). Height measurements of the 10 lines clearly indicated that both L15 and L3 were significantly taller than WT ([**Figure 4B**](#f4){ref-type="fig"}). With the exception of L2, which had a smaller stem diameter than WT, transgenic lines had a significantly greater stem diameter (31.83%) than did WT ([**Figure 4C**](#f4){ref-type="fig"}). Similarly, concerning the number of leaves of the poplars, overexpression lines produced more leaves (48%) than did WT ([**Figure 4D**](#f4){ref-type="fig"}). Generally, lines overexpressing the *CesA2* gene showed altered growth characteristics, demonstrating significantly increased height, stem diameter, and number of leaves. {#f4} As [**Figure 5A**](#f5){ref-type="fig"} clearly shows, compared with WT, the transgenic lines 15 and 12 both displayed high levels of fresh weight while lines 12 and 19 had the largest dry weight increase ([**Figure 5B**](#f5){ref-type="fig"}). Both the fresh and dry weights of the transgenic plants exceeded those of WT plants which were 45% and 36% higher than WT, respectively. Apart from L21, which had lower chlorophyll content than the WT, the transgenic plants maintained higher chlorophyll contents (34.3%) than did WT ([**Figure 5C**](#f5){ref-type="fig"}). The changes in the chlorophyll contents could increase photosynthesis and hence increase plant growth. {#f5} Changes in Secondary Cell Wall Composition {#s3_5} ------------------------------------------ SEM observations clearly showed a thickened secondary cell wall in transgenic line when compared with the WT. To investigate how the cell wall changed due to *PmCesA2* overexpression, the content of the cell wall\'s monosaccharide composition was analyzed in transgenic and wild-type plants. As [**Table 1**](#T1){ref-type="table"} shows, all transgenic lines showed significantly higher (up to 48%) xylose content with variations of other monosaccharides, compared to WT. ###### Cell wall monosaccharide composition (mg g ^-1^) from stem of the control and *PmCesA2* transgenic plants. Plant Glucose Xylose Mannose Galactose Rhamnose Arabinose ------- ------------------ ------------------ ------------- ------------- ------------- ------------- WT 35.16 ± 1.10 15.45 ± 0.36 2.83 ± 1.40 1.48 ± 0.19 0.67 ± 0.07 0.76 ± 0.18 L-2 36.05 ± 0.9 21.49 ± 0.78 2.19 ± 0.46 1.18 ± 0.09 0.66 ± 0.05 0.6 ± 0.03 L-3 35.59 ± 1.39 23.43 ± 0.60^\*^ 2.49 ± 0.53 1.33 ± 0.28 0.53 ± 0.11 0.61 ± 0.21 L-7 38.24 ± 1.22 22.56 ± 0.55 2.44 ± 0.29 1.16 ± 0.07 0.53 ± 0.13 0.71 ± 0.09 L-8 39.79 ± 2.16^\*^ 24.72 ± 2.63^\*^ 2.23 ± 0.78 1.19 ± 0.11 0.5 ± 0.05 0.58 ± 0.04 L-9 38.06 ± 1.20 24.4 ± 1.51^\*^ 2.53 ± 0.55 1.25 ± 0.17 0.6 ± 0.14 0.63 ± 0.06 L-12 39.71 ± 1.35^\*^ 25.61 ± 0.39^\*^ 2.83 ± 1.18 1.34 ± 0.10 0.65 ± 0.09 0.68 ± 0.23 L-15 37.19 ± 1.0 25.03 ± 1.01^\*^ 2.63 ± 0.74 1.35 ± 0.13 0.64 ± 0.11 0.81 ± 0.03 L-19 36.81 ± 1.25 22.64 ± 1.62 2.32 ± 0.51 1.17 ± 0.14 0.54 ± 0.08 0.6 ± 0.09 L-21 38.04 ± 1.20 22.63 ± 0.60 2.27 ± 0.43 1.2 ± 0.11 0.48 ± 0.12 0.57 ± 0.19 L-23 36.23 ± 1.40 22.79 ± 0.69 2.03 ± 0.15 1.09 ± 0.08 0.57 ± 0.07 0.6 ± 0.11 Data are the mean value ± SD of three biological replicates. \*Denotes significance at p \< 0.05. Cellulose production is correlated with the level of cellulose synthesis activity of *CesA* ([@B4]; [@B39]). To assess which cell wall components drove the increased thickness of the secondary cell wall, cellulose and lignin content of transgenic stems were measured. The overexpressing lines contained more cellulose ([**Figure 6A**](#f6){ref-type="fig"}) and lignin levels ([**Figure 6B**](#f6){ref-type="fig"}) compared with the control plants. These results showed *PmCesA2* could enhance biomass yields in the transgenic plants due to high cellulose and lignin content. {#f6} Changes in the Thickness of the Secondary Cell Walls in Transgenic *Populus* {#s3_6} ---------------------------------------------------------------------------- Cell wall thickness arises from increased deposition levels of xylose and cellulose ([@B46]). To better understand the contribution of *PmCesA2* overexpression to secondary cell wall biosynthesis, microscopic analyses were conducted to measure the thicknesses with the stems of WT and transgenic plants. Notably, SEM images showed the entire cell wall had increased, including the secondary cell wall (at least twofold more), in the overexpressing lines when compared with that of WT ([**Figure 7**](#f7){ref-type="fig"}). The xylem width of poplar transgenic lines had significantly increased compared to the wild-type plants ([**Supplementary Table S3**](#SM5){ref-type="supplementary-material"}). These results showed that overexpression of *PmCesA2* positively regulated the secondary cell wall formation in transgenic poplars. {#f7} Alternation of Gene Expression in *PmCesA2* Transgenic Poplars {#s3_7} -------------------------------------------------------------- To determine whether *PmCesA2* impacts the expression of other genes involved in cellulose or lignin biosynthesis, we performed quantification analysis of the expression of cellulose and lignin biosynthetic genes in the stems of transgenic lines. Transcript abundance of *CesA5*, *CesA6* (two primary *CesA* genes), *Susy2* (key enzyme for secondary growth), and *PAL1*, *4CL1* (lignin biosynthesis genes) were up‐regulated in overexpressing lines compared to those in WT. Overexpression of *PmCesA2* gene could increase the expression levels of other primary wall *CesAs*, *Susy2* and lignin biosynthetic genes ([**Supplementary Figure S2**](#SM2){ref-type="supplementary-material"}). Discussion {#s4} ========== The cell wall in plant cells provides structural support, underpinning plant growth and development ([@B22]). Cellulose, a major load-bearing structure of growing cell walls, has drawn much research attention for its various industrial applications. Therefore, researchers have tried to alter the process of cellulose biosynthesis to improve the growth, biomass production, and wood quality of plants. Here, we reported on the molecular and physiological behavior of transgenic poplar overexpressing *PmCesA2* under natural conditions. Over-expression of *PmCesA2* resulted in improved cellulose synthesis, plant growth, and biomass production in transgenic poplar lines compared to WT control plants, together with increased secondary wall thickening and width of the xylem. Although in plants, cellulose synthesis (*CesA*) genes have been shown to be fundamental for growth and development ([@B36]), in the last three decades, much effort has been met with limited success for improving cellulose synthesis through the overexpression of various the secondary *CesA~S~* ([@B5]). Attempts to overexpress a secondary wall-associated *CesA* gene (*CesA7*) in *Arabidopsis*, (*CesA8*) in *Populus*, and (*CesA4*) in barley did not show any improvement in plant growth and plant biomass production ([@B54]; [@B19]; [@B47]). A recent study of *Panicum virgatum* highlighted that both *CesA4* and *CesA6* overexpression and knock-down to extreme levels in the transgenic lines resulted in decreased biomass production ([@B27]). In the present study, hybrid poplars overexpressing the structural *PmCesA2* gene from the source pine tree displayed considerable improvements in biomass production. In terms of cell wall composition, the overexpressing transgenic plants also showed higher cellulose and lignin levels. The present work therefore suggests that increases in cellulose and lignin led to enhanced biomass yields in these plants. In order to verify whether overexpression of *CesA2* could influence the transcription of carbohydrate metabolism and cellulose production, monosaccharide composition in the stem was studied. This observation showed that glucose and xylose were the most abundant sugars in all samples as expected and suggested that increased glucose content in the cell walls of transgenic lines compared with the corresponding wild-type could be ascribed to increases in the cellulose content. Compared with WT, we observed that the chlorophyll contents of all transgenic mature lines except one were significantly increased. The greatly increased chlorophyll contents would be expected to augment the photosynthetic capacity of plants which could potentially increase the biomass of the transgenic plants. *EgCesA1,2,3* which are orthologous to *PtrCesAl*, *PtrCesA3*, and *PtrCesA2*, respectively, and *ZmCesA10*, *11*, *12* which are orthologous to *AtCesA4*, *AtCesA8*, and *AtCesA7*, respectively, all presented high expression in stem or stalk, tissues that undergo secondary cell wall biosynthesis in xylems ([@B1]; [@B38]). Expression patterns of *PtaCesA1*, *PtaCesA2*, and *PtaCesA3* in loblolly pine are consistent with functional roles to their orthologous of secondary cell wall *CesA* genes in angiosperms which are highly expressed in developing xylem. *PtrCesA2* was isolated from a xylem cDNA library that exhibited a high degree of identity 82% with *AtCesA7* cDNA that has been associated with xylem development in *Arabidopsis* ([@B50]), concluded that *PtrCesA2* from aspen is orthologous to *Arabidopsis AtCesA7*. Multiple alignments of full-length CesA protein sequences showed that secondary *CesAs* in pine are orthologs of *Arabidopsis* and poplar secondary wall CesAs ([@B33]). *PmCesA2* cDNA shows a high degree of similarity 98% with *PtaCesA2* cDNA that has been associated with secondary cell wall development in *P. taeda*. This indicates *PmCesA2* could be involved in secondary cell wall synthesis. In this respect, it seems that growth improvement and biomass production were achieved *via* genetic manipulation, at least for poplar trees. Cellulose biosynthesis and secondary wall thickness of *Arabidopsis* are affected by mutations in each of the secondary *CesAs* (*CESA4/IRX5*,*CESA7/IRX3*,and *CESA8/IRX1*), leading to collapsed xylem phenotype ([@B51]; [@B50]; [@B49]; [@B48]). Mutations in each of the primary *CesAs* can lead to reduced organ growth, which has been interpreted as the consequence of growth anisotropy being lost ([@B34]; [@B15]; [@B10]). *CesA5* and *CesA2* are responsible for secondary wall cellulose biosynthesis in *Arabidopsis* seed coat epidermis ([@B29]). Using various methods, both *in vitro* and *in planta*, it was shown that the primary wall *CesAs* interacts with other secondary wall *CesAs*, thus raising the possibility that mixed complexes of primary and secondary wall structure *CesAs* could occur at particular times ([@B9]). Overexpression *PmCesA2* gene enhances the expression of other primary wall *CesAs* as well as changes in expression of gene related to cell growth, cellulose (*Susy2*) or lignin (*PAL1* and *4CL1*) production. In conclusion, our results demonstrate the overexpressing of the *PmCesA2* gene is directly relevant to plant growth and development in poplar due to enhanced cellulose synthesis which led to a thickened secondary cell wall. Based on our results, we propose that the *CesA2* genes\' overexpression may cause to enhance the expression of other genes linked to cell growth and cellulose production in transgenic plants. Our approach could serve as an efficient biotechnological modification tool for producing enhanced plant biomass. Data Availability Statement {#s5} =========================== This article contains previously unpublished data. The name of the repository and accession number(s) are not available. Author Contributions {#s6} ==================== SM and KM designed and directed the project. SM, KM, and AM performed the experiments. SM and FW processed the experimental data. SM and KM wrote the manuscript with input from all authors. KJ supervised the project. All authors discussed the results and commented on the manuscript. Conflict of Interest {#s7} ==================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. We acknowledge the financial support from the National Key R&D Program of China (2017YFD0600304) and a project funded by the Priority Academic Program Development of Jiangsu Higher Education Institutions (PAPD). Supplementary Material {#s8} ====================== The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fpls.2020.00110/full#supplementary-material> ###### Click here for additional data file. ###### Click here for additional data file. ###### Click here for additional data file. ###### Click here for additional data file. ###### Click here for additional data file. [^1]: Edited by: Agnieszka Ludwików, Adam Mickiewicz University, Poland [^2]: Reviewed by: Guohua Chai, Qingdao Institute of Bioenergy and Bioprocess Technology (CAS), China; Bo Yang, Leibniz-Institut für Pflanzenbiochemie (IPB), Germany; Changjun Ding, Chinese Academy of Forestry, China; Chaofeng Li, The University of Tokyo, Japan; Bingyu Zhang, Chinese Academy of Forestry, China [^3]: †These authors have contributed equally to this work [^4]: This article was submitted to Plant Biotechnology, a section of the journal Frontiers in Plant Science

Introduction {#Sec1} ============ Competition tends to select for large size of individual offspring, balanced by a relatively smaller total number of offspring^[@CR1]--[@CR8]^. The competitive advantage of large offspring size is generally thought to occur during dominance interactions, such as during interference competition, in which organisms defend a resource and thus directly interfere with conspecifics^[@CR2],\ [@CR9]^. Interference competition can create a non-linear disparity between size and resource acquisition^[@CR2]^, which has been shown to favor the evolution of larger offspring, with more developed weapons^[@CR10]^. In contrast, exploitative competition^[@CR11],\ [@CR12]^, in which organisms indirectly compete over limited available resources, is not predicted to result in the evolution of size disparity among offspring; an organism's effectiveness at obtaining limited resources should be proportional to its size alone^[@CR2],\ [@CR13]^. Despite this, the Trinidadian guppy (*Poecilia reticulata*) has shown repeated convergence of large offspring size in environments where individuals indirectly compete over limited trophic resources (Fig. [1](#Fig1){ref-type="fig"})^[@CR3],\ [@CR14],\ [@CR15]^.Figure 1Geographic locations and average neonatal size among five Trinidadian guppy populations in the Northern Range Mountains. Repeated evolution of large offspring size has occurred in low-predation (LP) localities on both the northern (Yarra drainage) and the southern slopes (Caroni and Aripo drainages). The Caroni HHP exhibit an extremely high-predation (HHP) phenotype relative to Aripo HP. Standard length (SL) is defined as a straight line drawn from the tip of the lower jaw (mouth closed) to the base of the tail (caudal peduncle). Adobe Illustrator CS6 (Version 16.0.4; http://www.adobe.com/products/illustrator.html) was used to generate the map. The guppy is a species of live-bearing fish (Poeciliidae) that inhabits freshwater streams flowing from the Northern Range Mountains of Trinidad (Fig. [1](#Fig1){ref-type="fig"}). Guppies have repeatedly invaded and subsequently evolved within environments above waterfall barriers, which preclude large piscivorous fish from establishing^[@CR14],\ [@CR16]^. These low-predation (LP) environments are characterized by high guppy biomass and low invertebrate abundance, which force LP guppies to feed on encrusting algae and diatoms from the benthos^[@CR17]^. Primary production within LP environments has been shown to constrain guppy population size, indicating that individuals indirectly compete over a limited supply of available food^[@CR18]^. In such environments, large guppy offspring have higher fitness than their smaller high-predation (HP) counterparts^[@CR19]^. It is likely that increased ability to exploit these limited resources, by more effectively or efficiently removing the encrusting food source, leads to an increased fitness advantage among larger offspring at birth. Poeciliids as a group are specialized scrapers, possessing jaws adapted for removal of encrusting food items from the benthos^[@CR20]--[@CR22]^. The lower jaw of most ray-finned fishes (Actinopterygii), is comprised of two bones (angular and dentary) that have been shown to allow some torsion along the long axis of the mandible during suction feeding^[@CR23]^. In poeciliids, however, this joint, termed the intramandibular joint (IMJ), allows for rotation about the medio-lateral axis of the lower jaw and adult poeciliids produce up to 90° of IMJ rotation (Fig. [2](#Fig2){ref-type="fig"})^[@CR21]^. The fitness advantages of mobility at the IMJ have been demonstrated through increased benthic scraping performance across several unrelated groups of fishes that all have highly mobile IMJs and feed by scraping^[@CR24],\ [@CR25]^. The IMJ allows for greater gape, enhanced bite force and the ability to close the mouth even while the oral jaws are protruded^[@CR20],\ [@CR21],\ [@CR24]--[@CR27]^. Among poeciliids, mobility at the IMJ correlates with reliance on encrusting food sources^[@CR21],\ [@CR22]^. It is not fully understood, however, the degree to which the joint functions in early life stages, particularly within the ecological context of newborn guppies competing over limited benthic resources.Figure 2Schematic of the jaw while closed (**a**) and while applied to the substrate to scrape food (**b**). Close-up of the jaws (**c**) illustrates that upon full mouth opening much of the gape is due to rotation about the intramandibular joint (IMJ). The IMJ sits at the intersection of the angular and dentary bones, which are connected via Meckel's cartilage (gray band running deep to the lower jaw). The IMJ is known to be mobile, with up to 90° of rotation, among adult poeciliids^[@CR21]^. Abbreviations: pmx, premaxilla; max, maxilla; pal, palantine; dent, dentary; ang, angular. The effect of offspring size on fitness (growth rate, time to first reproduction, etc.) in the guppy depends on the availability of resources. When resources are unlimited, smaller HP offspring experience catch up growth, where higher growth rates increase HP size to meet LP juvenile size by one month post natal^[@CR28],\ [@CR29]^. When guppy densities are high and resources are limiting, however, larger juveniles experience higher growth rates than their smaller counterparts, maintaining their size advantage throughout ontogeny^[@CR19]^. The finding that larger offspring grow faster than smaller offspring when resources are limiting suggests a non-linear disparity between size and resource acquisition. It is possible that this disparity is the product of morphological features that confer a competitive advantage to the LP offspring. Since increased mobility at the IMJ among poeciliids correlates with improved performance when scraping encrusting food, we predict that LP offspring are born with greater mobility at the IMJ. From high-speed video of offspring feeding from the benthos, we will measure jaw joint kinematics among five populations of Trinidadian guppies. We predict that LP offspring will exhibit increased function of their feeding apparatus, as evidenced by increases in intramandibular joint rotation, quadratomandibular joint rotation and total gape. By applying immunohistochemical and traditional histological staining techniques to visualize the musculoskeletal system, we will quantify morphological development among neonatal guppy head skeletons. To distinguish between size and population effects, we will collect data at two time points for each population: first at birth, where size is predicted to vary among populations; and second at later stages, where juvenile size ranges overlap. This functional and morphological investigation of neonates will offer insights into the advantage of large offspring size and will provide a better understanding of how size and maturity influence organismal growth and development, and how evolution shapes these parameters at the population level. Results {#Sec2} ======= A total of 45 neonatal offspring (n = 3 per brood, 3 broods per population, 5 populations) were filmed with high-speed video (Fig. [3](#Fig3){ref-type="fig"}) in their brood groups feeding on an encrusting food source (5--6 feeding events per brood). Following the feeding trials, neonates were euthanized and fixed for morphological analysis. Juveniles (n = 45) were reared for 10--20 days postnatal to obtain overlapping size ranges for all populations and the same filming and fixing procedures were followed. All kinematic data are reported as brood averages.Figure 3Feeding kinematics in neonates from HP and LP populations. Standard length of the two individuals is shown at top. (**a**,**b**) Intramandibular (IMJ) rotation angle between high-predation (**a**) and low-predation (**b**) neonates. (**c**,**d**) Quadratomandibular (QMJ) rotation between high (**c**) and low (**d**) predation neonates. Rotation at the IMJ, but not at the QMJ, increases with neonatal size. As expected from previous studies on guppy life histories^[@CR30]--[@CR33]^, standard length (Fig. [1](#Fig1){ref-type="fig"}) of guppy neonates was significantly different among populations (ANOVA: F (4, 40) = 394.73; p \< 0.0001). Low-predation (LP) neonates were significantly larger than their high-predation (HP) counterparts across drainages (Tukey HSD: p \< 0.0001). Feeding kinematics: neonates {#Sec3} ---------------------------- Guppy neonates showed substantial differences in IMJ rotation among populations (Fig. [4](#Fig4){ref-type="fig"}). There was a significant effect of population on intramandibular joint (IMJ) rotation (ANOVA: F (4, 10) = 12.23; p = 0.0007; Figs [3](#Fig3){ref-type="fig"} and [4](#Fig4){ref-type="fig"}). Post-hoc Tukey-HSD indicates that the two LP populations have greater IMJ mobility than their HP counterparts. The neonates from the LP populations are also significantly longer at birth (Fig. [4](#Fig4){ref-type="fig"}). Quadratomandibular joint (QMJ) rotation showed no significant difference among population (ANOVA: F (4, 10) = 2.81; p = 0.085).Figure 4Mean intramandibular joint (IMJ) rotation among neonates for each population. Low-predation offspring exhibit nearly twice the IMJ mobility during feeding on encrusting food sources compared to their high-predation counterparts (asterisk indicates significance: p = 0.0007; error bars indicate standard error). Mean neonatal standard length (±standard error) indicated for each population. HHP, high-high-predation; HP, high-predation; LP, low-predation. Feeding kinematics: neonates and juveniles {#Sec4} ------------------------------------------ Among neonates and juveniles examined herein, size explained a significant amount of variation in jaw kinematics (Fig. [5](#Fig5){ref-type="fig"}). Mobility at the IMJ is significantly and positively correlated with standard length (Linear regression: R^2^ = 0.82; F (1, 28) = 126.36; p \< 0.0001). This trend continues to increase with size where 12 mm juveniles produce upwards of 50 degrees of rotation and adults (\>20 mm) produce upwards of 90 degrees of rotation (Fig. [5a](#Fig5){ref-type="fig"}).Figure 5Intramandibular and quadratomandibular joint rotation in relation to body length in neonates and juveniles. Intramandibular joint rotation (**a**) increases with size among neonates and throughout ontogeny, but quadratomandibular joint rotation (**b**) does not change with size. Five populations and two age classes are presented (neonates indicated by closed symbols; juveniles indicated by open triangles; one adult indicated by open square for reference). Each data point shows mean brood joint rotation and mean standard length [+]{.ul} standard error for 5--6 scraping events per brood. Regression lines in A are distinct from linear mixed model reported in Results. Regression analyses for individual populations indicate that within each population, IMJ rotation increases with increasing standard length (Fig. [5a](#Fig5){ref-type="fig"}): Aripo HP: R^2^ = 0.88; F (1, 5) = 30.52; p \< 0.0052; Aripo LP: R^2^ = 0.92; F (1, 5) = 43.20; p \< 0.0028; Caroni HHP: R^2^ = 0.85; F (1, 5) = 24.01; p \< 0.0080; Yarra HP: R^2^ = 0.90; F (1, 5) = 35.04; p \< 0.0041; Yarra LP: R^2^ = 0.67; F (1, 5) = 8.10; p \< 0.0466. To determine the effect of population on the data, an ANCOVA with standard length as the covariate was performed. Prior to conducting the ANCOVA, we found no interaction between our independent variable (population) and covariate (standard length; p = 0.5573). ANCOVA results indicate that population was not a significant predictor of IMJ mobility when taking size into account (p = 0.7189), therefore we modeled population as a random effect with size as a fixed effect and ran a linear mixed model to determine the significance of size in predicting IMJ rotation. Linear mixed effects model with population as the random effect indicated that standard length accounts for a significant amount of variation in IMJ rotation (p \< 0.0001). Quadratomandibular joint rotation does not change with size throughout the ontogenetic series examined herein (Linear regression: R^2^ = 0.01; F (1, 28) = 0.25; p = 0.62; Fig. [5b](#Fig5){ref-type="fig"}). Population does not explain variation in QJR across the size series examined herein (ANCOVA (F (5, 24) = 1.8828; p = 0.1349; interaction pop\*SL p = 0.1868). Linear mixed effects model with population as the random effect and size as the fixed effect indicated that standard length does not account for a significant amount of variation in QMJ rotation (p = 0.707). Maximum gape {#Sec5} ------------ Maximum gape while feeding increases with positive allometry among guppy neonates, scaling on a log-transformed plot with a slope of 1.20 ± 0.12 (R^2^ = 0.80; F (1, 52) = 101.5; p \< 0.0001; Fig. [6](#Fig6){ref-type="fig"}). A scaling exponent significantly greater than 1 indicates that the linear distance between the oral jaws during maximum gape is relatively greater in larger offspring. The scaling of gape size among guppy offspring (SL^1.2^) indicates larger offspring not only take larger absolute bites, but they also take relatively larger bites.Figure 6Positive allometric scaling of maximum gape among guppy offspring. Neonates are shown in open circles, postnatal juveniles shown in closed circles. If gape increased isometrically with body length, the points would fall along the dotted line with a slope of 1. Positive allometry indicates maximum gape is relatively and absolutely greater in larger fish. Each data point indicates mean brood gape (n = 5--6 scraping events per brood) and standard length [+]{.ul} standard error. Head ossification {#Sec6} ----------------- The degree of cranial ossification observed in neonatal guppies varied substantially among neonates (Fig. [7a](#Fig7){ref-type="fig"}). The smallest neonates (Caroni HHP and Yarra HP) are born with fewer than 20% of the bones in their head skeleton showing evidence of ossification (red staining), whereas the larger, low-predation neonates (Aripo LP and Yarra LP) are born with over 90% of their head skeleton ossified (Fig. [7b](#Fig7){ref-type="fig"}). The increase in neonatal head ossification is steep among populations, such that a 20% increase in size yields over four-fold increase in skeletal ossification (Fig. [7b](#Fig7){ref-type="fig"}). Ossification of the following skeletal elements distinguishes LP neonates from their HP counterparts: angular, hyomandibula, quadrate, ceratohyal and autopalatines. Caroni HHP and Yarra HP neonates lack ossification of these key elements, whereas Aripo LP and Yarra LP neonates possess full ossification of these elements. The largest juvenile guppies cleared and stained (10 mm) show that head ossification plateaus around 95% (the hypohyal and basihyal elements remain cartilaginous into adulthood).Figure 7(**a**) Guppy neonates exhibit substantial variation in ossification of the head skeleton at birth. Cleared and stained neonatal specimens indicate cartilage (blue stain) and bone (red), from each of the five populations studied, aligned by standard length. North and South slope of the Northern Range Mountains in Trinidad indicated. Scale bar = 1 mm. All images to scale. (**b**) Development of the skull (as assessed by cranial ossification) increases rapidly with size among neonates. Of 18 skeletal elements examined within the head, the smallest neonates are born with fewer than 20% of these elements ossified. Neonates from LP environments are born \~20% larger, but with 400% greater ossification within the head skeleton. Populations are coded by color. Adductor mandibulae scaling {#Sec7} --------------------------- The surface area of the guppy adductor mandibulae scales with significant positive allometry (Fig. [8](#Fig8){ref-type="fig"}), such that larger guppy offspring possess relatively larger jaw-closing musculature. Adductor mandibulae area among neonatal individuals and juveniles increases with positive allometry throughout the first month of guppy postnatal development (up to 13 mm SL). The regression line of log adductor mandibulae area against log standard length yields a slope of 2.72 ± 0.10 (R^2^ = 0.94; F (1, 52) = 750.97; p \< 0.0001), which is significantly higher than the scaling value expected for geometric similarity (isometry = 2).Figure 8Positive allometric scaling of surface area of the adductor mandibulae relative to body length. Inset shows adductor mandibulae muscle area outlined in black (scale bar = 1 mm). If area of the adductor mandibulae increased isometrically, the points would fall along the dotted line with a slope of 2, which indicates isometry of surface area relative to standard length. Positive allometry indicates that growth of the adductor mandibulae outpaces growth of the body. Discussion {#Sec8} ========== We predicted that rotation at both the intramandibular and quadratomandibular joints would increase with neonatal size, but the data show that guppy offspring only exhibit a positive relationship between joint mobility and standard length at the intramandibular joint (IMJ; Fig. [5](#Fig5){ref-type="fig"}). Intramandibular joint rotation is positively allometric among neonates and postnatal juveniles, such that larger offspring employ greater movement at the IMJ while feeding (Figs [3](#Fig3){ref-type="fig"}, [4](#Fig4){ref-type="fig"} and [5](#Fig5){ref-type="fig"}). This increase in joint rotation with size is correlated with a relative increase in maximal gape (Fig. [6](#Fig6){ref-type="fig"}). In other fishes, IMJ mobility has been shown to increase maximal gape as well as contact area of oral jaws with the substrate and force production^[@CR21],\ [@CR24],\ [@CR27]^. In the guppy, the advantages associated with increased IMJ mobility might lead to the increase in scraping performance observed among larger offspring^[@CR19]^. An increase in mobility at the IMJ is unlikely due to changes of size per se. Joint rotation should not change with organismal size, because the degree of rotation about any given joint depends on the potential excursion afforded by the joint itself, and not the length of the elements involved^[@CR34],\ [@CR35]^. The observed increase in IMJ mobility with size among guppy offspring (Fig. [5](#Fig5){ref-type="fig"}) suggests that the morphology of the lower jaw is different among neonates. The observed increase in IMJ mobility occurs concurrent with increases in ossification pattern and muscular development, two separate measures of maturity that indicate guppies are born at varying stages of morphological maturity. Throughout ontogeny the guppy skeleton is ossified in a regular pattern^[@CR36]^, and we find that populations of Trinidadian guppies are born at varying points along this developmental trajectory, with the largest neonates possessing the greatest degree of cranial ossification (Fig. [7](#Fig7){ref-type="fig"}). Over the size range of guppy neonates investigated in this study (5.5 mm -- 7.2 mm), we observe a four-fold increase in the number of cranial elements that are ossified at birth (Fig. [7b](#Fig7){ref-type="fig"}). The cranial skeleton is nearly fully ossified among the largest offspring, and only a few additional elements ossify during postnatal development. Growth of the muscles responsible for scraping behaviors provides secondary evidence that guppy neonates are born at different stages of morphological maturity. Adductor mandibulae area scales with significant positive allometry among guppy neonates and among postnatal juveniles (Fig. [8](#Fig8){ref-type="fig"}). The positive allometric scaling of adductor mandibulae size indicates that the growth of this muscle outpaces the growth of the entire organism, a pattern consistent with the development of early stage morphologies among larval fishes^[@CR37]^. These profound differences in degree of ossification and muscular growth provide evidence that guppy neonates are born at varying levels of morphological maturity, and we hypothesize that maturation of the mouth opening mechanism explains the observed positive allometry of IMJ rotation among neonates. The mouth opening mechanism (i.e., jaw depression) in fishes continues to develop even after first feeding has initiated^[@CR38]^. Several jaw depression mechanisms are thought to actuate the lower jaw, and in all cases, mouth opening is the product of morphological maturation of jaw depression musculature^[@CR39],\ [@CR40]^. One of the last jaw opening muscles to develop and become functional is the geniohyoideus (=protractor hyoideus), which extends from the hyoid apparatus (ceratohyals) to the tip of the lower jaw (dentary symphysis)^[@CR38]^. This muscle is the most likely candidate for causing IMJ rotation in the guppy^[@CR38]^ and we propose that its function in causing IMJ rotation develops over the range of neonates investigated in this study. We hypothesize that the geniohyoideus muscle is not fully developed in the smallest HP neonates; indeed, the certatohyals are not ossified until later stages. We suggest that guppy offspring are born across a range of developmental stages where maturation of the jaw opening system varies widely and influences jaw function and competitive ability of neonates. The positive effects of size on musculoskeletal development and feeding kinematics suggests that larger guppy neonates in low-predation (LP) environments are better equipped to compete within the resource limited environments into which they are born. Competition experiments have shown that when resources are limiting, large guppy juveniles exhibit higher growth rates than their smaller counterparts^[@CR19]^. The current finding that larger offspring are both more mature and possess increased IMJ mobility provides a potential mechanism underlying this competitive advantage. The advantage of producing larger offspring in low resource, high competition environments, where these larger and more developed offspring can more efficiently obtain^[@CR19]^ and utilize^[@CR41]^ limited resources, might help to maintain selection for the LP phenotype. In the environmental context of limited resources, the LP, high competition guppy phenotype has evolved repeatedly and rapidly^[@CR16]^. Female guppies are selected to provision more yolk per ova, gestate for a greater period of time and ultimately produce larger offspring, among other life history traits^[@CR42]^. Here, we provide evidence that associated with these life history features are morphological and functional traits important for exploitative competition. Low-predation guppy offspring are born not only larger, but also more mature and with more jaw mobility than their smaller counterparts. The observed disparity between size and resource acquisition among guppy offspring competing over limited resources is perhaps the product of the additive effects of maturation and size. We suggest that size per se may not be the only adaptive benefit of producing larger offspring in highly competitive environments; large size may instead be of secondary importance to the maturation of morphological features that allow for enhanced foraging ability at birth. Methods {#Sec9} ======= Collection and housing {#Sec10} ---------------------- Pregnant female guppies were collected from five populations throughout Trinidad's Northern Range Mountains (Fig. [1](#Fig1){ref-type="fig"}). Three populations were collected from the Caroni drainage, the major southern confluence draining westward into the Atlantic: Aripo low-predation (LP); Aripo high-predation (HP); and Caroni extreme high-predation (HHP). Two additional populations were collected from the Yarra River on the north slope of the mountain range: Yarra LP and Yarra HP. The predator communities differ between northern and southern slopes; the former contain mainly gobiid predators, while the latter contain a dominant cichlid (*Crenicichla alta*)^[@CR16],\ [@CR43]^. The selective pressures have been shown to be similar between slopes: high predation selects for accelerated life histories, while high competition in LP sites selects for prolonged life histories (i.e., larger offspring, longer gestation times, longer inter-brood intervals, later age of first reproduction, less overall investment in reproduction)^[@CR3],\ [@CR14],\ [@CR16],\ [@CR31]^. Pregnant female guppies were housed within a field lab in isolated 2-liter tanks, where they later gave birth. Upon parturition, each mother was promptly removed to mitigate potential cannibalism and loss of the litter. Litter sizes were separated into five or fewer individuals per tank to standardize feeding amounts and other density effects. Fish were fed twice daily on a diet of *Artemia* nauplii in the morning and algae flakes in the evening. Tanks were housed within an open-air laboratory facility and exposed to ambient light, which maintained light:dark schedule at approximately 12:12 hours. All procedures were approved by the Brown University Institutional Animal Care and Use Committee (Protocol \#: 1211035 to E. L. Brainerd). All experiments were performed in accordance with relevant guidelines and regulations. High-speed video {#Sec11} ---------------- Filming of offspring while feeding was staged within a separate arena, equipped with lights, camera, and small chamber. The tank itself was constructed of acrylic (32 × 121 mm base) and filled with stock water to 50 mm height. Individuals were allowed a minimum of 30 minutes acclimation time prior to filming. All individuals were fasted the night prior to the feeding trial to facilitate feeding. Fish were fed a gelatin substance composed of dried brine shrimp, fruit, calcium and gelatin powder (hereafter referred to as 'gel'). A small bolus of gel (approx. 0.1 g) was pressed onto a small, flat rock, and allowed to adhere for 5 minutes before being placed at the edge of the acrylic panel closest to the camera. Video sequences were captured using a Photron FASTCAM 1024PCI (Photron USA, Inc., San Diego, CA, USA) fitted with a Nikon 105 mm, f/2.8 macro lens (Nikon Inc., Melville, NY, USA). Video was captured at 500 frames per second, at 1/1000 s shutter speed. The filming setup was illuminated with two LED lights (2.0 amp, 28 volt; Visual Instrumentation Corp., Lancaster, CA, USA). Each filming event consisted of three individuals from each brood feeding concurrently upon the encrusting substrate (individual fish did not readily feed when isolated within the filming chamber). Filming sessions lasted on average between 30--60 minutes, which allowed 5--6 scraping events to be captured per brood. Data was collected from three broods per population, five populations and at two time points. This resulted in a total of n = 45 neonates, n = 45 juveniles and n = 5 adult females. Individuals were sacrificed at the end of each trial. Following the feeding trials, guppies were sacrificed by overdose of tricaine methanesulfonate (Tricaine-S, Western Chemical Inc., Ferndale, WA, USA). Specimens were fixed in 4% buffered paraformaldehyde (Sigma, St. Louis, MO, USA) overnight and transferred to 70% EtOH for long-term storage. Staining {#Sec12} -------- Skeletons of the guppy offspring were differentially stained and the bodies cleared to enable us to characterize and quantify degree of head skeletal ossification. The use of this technique stains cartilage blue and bone red. Bone and cartilage stains were performed by dehydrating specimens in 100% EtOH for 30 minutes prior to transfer into alcian blue solution (80% ethanol, 20% acetic acid) overnight. Specimens were rehydrated through a graded series of ethanol solutions (30 minutes in each) and were then placed overnight in a neutralizing solution of saturated sodium borate. They were bleached in a 30--35% H~2~O~2~ per 50 ml of 0.5% potassium hydroxide (KOH) for 1 hour, and then digested in a trypsin solution of 3 parts saturated sodium borate solution to 7 parts distilled H~2~O (10 g trypsin per liter) for 1 hour. Specimens were transferred to a solution of alizarin red and 0.5% KOH overnight, and finally cleared through a graded series (1:3, 1:1, 3:1) of glycerol:KOH before being imaged in 3:1 glycerol:KOH. Specimens were visualized using a Nikon dissecting microscope (Nikon SMZ800 dissecting scope and Nikon DXM1200C digital camera). To determine the degree of muscle development during ontogeny, neonates were immunostained with the muscle-specific antibody MF-20, visualized after performing HRP color reaction and the perimeter of the adductor mandibulae traced from a lateral view. Specimens were skinned to promote penetration of the antibodies, were washed in phosphate buffered saline (PBS) + tween (PBST) and then bleached overnight in 2:1 Dent's fixative:H~2~O~2~ (Dent's fixative = 4 parts methanol to 1 part dimethyl sulfoxide). Specimens were washed in PBST and then treated with Proteinase K (10 μl/ml PBS) for 20 min, washed again in PBST and blocked in PBST/bovine albumin/goat serum (PBN) for two hours before being incubated in MF-20 primary antibody (DSHB, Iowa, USA) overnight at 20 °C. Specimens were washed a third time in PBST and blocked in PBN (2 hours) prior to application of horseradish peroxidase (HRP) secondary antibody (1:200 dilution; Promega Corp., Madison, WI, USA) overnight. Color reaction was performed using 3,3′-Diaminobenzidine (DAB chromogen kit, Biocare Medical, Concord, CA, USA) and specimens were imaged in 1:1 glycerol:PBS solution under a dissecting microscope (Nikon SMZ800 dissecting scope and Nikon DXM1200C digital camera). Analysis {#Sec13} -------- Quantification of ossification within the cranial skeleton was performed by identifying (and assigning a value) to the presence (1), partial presence (0.5) or absence (0) of alizarin red uptake within each of the following feeding-associated skeletal elements: pharyngeal jaws, premaxillae, acrodont teeth, dentaries, hyomandibula, opercles, angulars, maxillae, prevomer, quadrate, urohyal, epihyal, ceratohyal, frontal, prefrontal-lateral ethmoids, autopalatines, hypohyal, basihyal. Ossification of cranial elements is presented as % of total head skeleton ossified, which was quantified by determining the mean of the values for presence (1), partial presence (0.5) or absence (0) of red coloration in all 18 elements measured in the head skeleton. This type of averaging yields, for example, a skeleton with 9 elements fully ossified (9 × 1 = 9) and 9 elements partially ossified (9 × 0.5 = 4.5) a % ossification of 75% (13.5/18 \* 100). The surface area of the superficial adductor mandibulae muscles was measured using IMAGEJ v.1.42 (National Institutes of Health, Bethesda, MD, USA). The stained muscle was outlined for each specimen to determine area. These values were log-transformed and regressed against log standard length to yield scaling relationships. Kinematic variables during scraping of the gel substrate were quantified using IMAGEJ v.1.42 (National Institutes of Health, Bethesda, MD, USA). Linear gape distance was defined as a straight line connecting the rostral-most tip of the premaxillary and dentary bones. Quadratomandibular joint (QMJ) rotation was defined as a line extending rostrally from the ventral margin of the orbit to the QMJ, a line drawn down the length of the angular-articular bone complex and a vertex where these two lines intersect. Intramandibular joint (IMJ) rotation was defined as a line extending rostrally from the QMJ along the angular-articular bone complex, a line extending rostrally from the IMJ along the dentary to the point of lower jaw contact with the substrate and a vertex where these two lines intersect (Fig. [3](#Fig3){ref-type="fig"}). Quadratomandibular joint rotation was referenced to the closed position of the jaw (defined as a line extending rostrally from the ventral-most position of the orbit to the QMJ, a line drawn along the ventral margin of the lower jaw from the QMJ rostrally to the tip of the dentary during closed mouth and a vertex where these two lines intersect; Fig. [3](#Fig3){ref-type="fig"}). IMJ rotation was referenced to the closed jaw position of the lower jaw (i.e., a straight line, equaling 180°). Kinematic analysis was performed on several feeding trials per filming event and joint angles were each measured five separate times and averaged to yield the reported joint angle (standard error of five measures did not exceed ± 1 degree). Statistical analysis {#Sec14} -------------------- To determine scaling relationships morphological variables were log-transformed and regression lines (reduced major axis) fit against log standard length. The resultant scaling coefficients were compared against expected isometric scaling values for each variable. Kinematic data were presented as brood averages. Regression analyses were performed on kinematic variables, fitted against standard length. One-way analyses of variance (ANOVA) were performed on kinematic variables with population as the factor. Where relevant, Tukey HSD post hoc tests were performed to identify means that are significantly different from each other. Analysis of covariance (ANCOVA) analyses were used to control for the effect of size among populations (an interaction analysis was first performed to determine homogeneity of slopes). Linear mixed effects models were performed to account for random effects of population. All statistical analyses were performed using JMP v.11 (SAS Institute, Cary, NC, USA). Data accessibility {#Sec15} ------------------ Cranial morphology and kinematic data among populations of neonatal guppies will be deposited in the Dryad Digital Repository. Videos will be available on the Zoological Motion Analysis (ZMA) Portal. **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. The authors are grateful for conceptual and editorial contributions provided by D.N. Reznick. Natalie Dial assisted in the capture, care and filming of guppies in Trinidad. We thank M. Rosario for statistical advice. We also thank C. Anderson, E. Tavares, T. Martin and two anonymous reviewers for comments that greatly improved the manuscript. This work was funded in part by the US National Science Foundation (grant nos 1655756, 1601377 and 1025845) and by the Bushnell Research and Education fund. T.D. and E.B. designed the experiment. T.D. gathered and analyzed all the data. P.H. assisted with immunohistochemistry methods and prepared Figure. 2. T.D. wrote the manuscript. All authors reviewed and revised the manuscript. Competing Interests {#FPar1} =================== The authors declare that they have no competing interests.

1. Introduction {#s0005} =============== Tarlov cysts are perineural fluid-filled sacs that are usually found in the lumbosacral spine around the junction of peripheral nerve roots and the respective dorsal root ganglia. The cysts are extradural meningeal cysts with collections of cerebrospinal fluid (CSF) within the nerve root sheath \[[@bb0005]\]. Tarlov cysts are found in roughly 1%--5% of the general population. \[[@bb0010],[@bb0015]\]. Interestingly, some studies have demonstrated that women account for 59%--70% of cases \[[@bb0010],[@bb0020]\]. Symptomatic Tarlov cysts are rare but can sometimes grow in size and compress adjacent nerve root structures, leading to localized or radicular pain, bowel and bladder symptoms, or other neurological sequelae. The most common presenting symptoms are back pain, pelvic pain, perineal pain, lumbar radiculopathy, or bowel/bladder symptoms. These are generally exacerbated by maneuvers that raise the CSF pressure such as standing, coughing, sneezing, or straining to void or defecate. Diagnosis is generally done with imaging and magnetic resonance imaging (MRI) is the modality of choice. Tarlov cysts were first described by Tarlov in 1938 as an incidental finding during an autopsy \[[@bb0025]\]. Since then a variety of surgical and non-surgical interventions have been described for the treatment of Tarlov cysts. Unfortunately, there is a high rate of complications associated with surgical treatment of these cysts including hemorrhage, neurological deficit, CSF leak, infection, chronic pain, and cyst recurrence. A systematic review of surgical and non-surgical management of Tarlov cysts demonstrated a 21% and 12.47% complication rate respectively. There was a higher rate of transient sciatica, CSF-related complications, and bladder/bowel complications. Interestingly, both groups reported symptomatic improvement in 83.5% of patients. Transient exacerbation of symptoms and cyst recurrence was noted to be higher in non-surgically managed groups, suggesting surgical management is associated with higher postprocedural complication rates but better long-term efficacy for symptoms and cyst resolution \[[@bb0030]\]. Given the rarity of the condition and the high complication rates associated with surgical management, there is currently no consensus regarding the best treatment for patients who are symptomatic. We describe a patient who had a longstanding history of chronic pelvic pain secondary to multiple sacral Tarlov cysts. She underwent surgery for her condition, but it worsened her symptoms and the patient ultimately obtained relief with the use of a high-frequency spinal cord stimulator (SCS). 2. Case Description {#s0010} =================== The patient was a 66-year-old woman with a longstanding history of chronic pelvic pain secondary to multiple large Tarlov cysts affecting the S1-S4 nerve roots. She had attempted medication management with acetaminophen, NSAIDs, and neuropathic pain medications. She had also undergone a series of epidural steroid injections and a trial of superior hypogastric plexus blocks but obtain minimal relief with these. She ultimately had surgery for the removal of the Tarlov cysts as well as sacral lamina reconstruction ([Fig. 1](#f0005){ref-type="fig"}). However, the patient\'s surgery resulted in worsening of her chronic pelvic pain, and also produced new-onset back pain and lumbar radiculopathy down both legs. The patient reported constant debilitating back and pelvic pain with intermittent stabbing and burning pain in her legs. Her worsening symptoms were uncontrolled with continued medication management as well as a repeat series of epidural steroid injections.Fig. 1Lumbosacral MRI demonstrating post-surgical changes after the patient\'s initial surgery.Fig. 1 Given that the patient\'s worsening symptoms were unrelieved with both medication management and interventional pain procedures, we tried high-frequency spinal cord stimulation for her worsening pain and new-onset radiculopathy. The patient was counseled regarding the risks and benefits of the procedure and elected to proceed. A spinal cord stimulator lead was introduced into the epidural space and advanced to the superior endplate of T8. A second lead was placed at the superior endplate of T9 ([Fig. 2](#f0010){ref-type="fig"}).Fig. 2Fluoroscopic imaging demonstrating SCS lead placement in the posterior epidural space at the superior levels of T8 and T9.Fig. 2 The patient presented for follow-up after the procedure and reported significant improvement in her symptoms. She noted that the use of SCS had resulted in a 90% improvement of her back pain, a 95% improvement in her pelvic pain, and \>50% improvement in her radiculopathy. Additionally, she reported she was much more active and was able to decrease her medication use with the pain relief she obtained from spinal cord stimulation. 3. Discussion {#s0015} ============= Tarlov cysts can be a challenging condition to recognize and diagnose given the rarity of the condition. Although these cysts most commonly affect the sacral nerve roots, they have been found in lumbar, thoracic, and cervical regions as well. The condition is generally asymptomatic. However, a small percentage of patients may demonstrate symptoms related to nerve root compression. Tarlov cysts tend to expand over time and can cause nerve root irritation, leading to pain or other neurological disturbances. There is no consensus on the optimal management of symptomatic Tarlov cysts. Percutaneous cyst drainage is a nonsurgical intervention has been used to treat this condition \[[@bb0015]\]. This treatment is only temporary though, as cysts tend to gradually reform and symptoms recur. In addition to percutaneous drainage, one study has demonstrated that cyst aspiration with the placement of fibrin glue can prevent recurrence of the cysts. However, these patients are also at significant risk for postprocedural aseptic meningitis \[[@bb0035]\]. Surgical treatment of symptomatic cysts varies and can involve complete cyst removal with excision of the affected posterior root and ganglion, decompressive laminectomy, cyst wall resection, and cyst fenestration \[[@bb0040], [@bb0045], [@bb0050], [@bb0055]\]. The success and complication rates vary greatly by procedure. Again, there is no consensus regarding when surgical management for Tarlov cysts is warranted, though one study suggested that cysts larger than 1.5 cm with associated radicular pain or bowel/bladder dysfunction may benefit the most from surgical intervention \[[@bb0050]\]. We recommend that patients presenting with symptomatic Tarlov cysts diagnosed by imaging should initially undergo conservative management with medication management, including acetaminophen, nonsteroidal anti-inflammatory drugs, and neuropathic pain medications such as gabapentin. Additionally, patients may respond to interventional pain procedures such as epidural steroid injections. Surgery should be reserved for patients who fail conservative management, given the high complication rates. We would also like to comment on the success of spinal cord stimulation in this patient. SCS has been proven to be effective for treating intractable neuropathic pain such as lumbar radiculopathy and post-laminectomy syndrome \[[@bb0060], [@bb0065], [@bb0070]\]. There is also growing evidence that SCS can even be helpful for treating debilitating chronic visceral pelvic pain \[[@bb0075]\]. We believe this case is of importance as it describes the complicated management of patients with symptomatic Tarlov cysts who ultimately fail to respond to conservative therapy. We also describe the use of SCS in this patient and the benefit it can provide for patients who are suffering from severe radiculopathy after surgery as well as those with chronic pelvic pain ([Fig. 3](#f0015){ref-type="fig"}).Fig. 3An example of an implantable high-frequency spinal cord stimulator system showing the percutaneous leads and implantable pulse generator.Fig. 3 4. Conclusion {#s0020} ============= Tarlov cysts are an uncommon condition that present more often in women. This condition rarely becomes symptomatic, producing a variety of neurologic symptoms and painful conditions. Given the rarity of the condition and the high complication rates associated with surgery, there is no clear treatment algorithm for symptomatic patients. These cysts can be managed conservatively or surgically; surgical management has higher complication rates but better long-term results for symptom and cyst resolution. Patients who develop new or worsening pain after Tarlov cyst surgery may benefit from spinal cord stimulation. This case may provide guidance for physicians managing patients suffering from symptomatic Tarlov cysts, or worsening pain symptoms after surgical management of these cysts. Contributors {#s0025} ============ Jamal Hasoon contributed to the drafting, literature review, and critical revision of the article. Amnon A. Berger contributed to the drafting, literature review, and critical revision of the article. Ivan Urits contributed to review and critical revision of the article. Vwaire Orhurhu contributed to review and critical revision of the article. Omar Viswanath contributed to review and critical revision of the article. Musa Aner provided supervision, and contributed to review and critical revision of the article. Conflict of Interest {#s0030} ==================== The authors declare that they have no conflict of interest regarding the publication of this case report. Funding {#s0035} ======= No funding from an external source supported the publication of this case report. Patient Consent {#s0040} =============== Obtained. Provenance and Peer Review {#s0045} ========================== This case report was peer reviewed.

Human airway epithelia function in barrier formation, defense against pathogens, and mucociliary clearance ([@b11-ehp-119-784]). They represent the first barrier against airborne environmental pollutants, and they coordinate recruitment of pivotal inflammatory cells in several pathologies, including chronic obstructive pulmonary disease (COPD) ([@b14-ehp-119-784]; [@b17-ehp-119-784]; [@b30-ehp-119-784]). The inhalation of diesel exhaust particles (DEP), produced by vehicular traffic contributing to urban smog, leads to serious respiratory diseases (e.g., COPD, emphysema, bronchial cancer, chronic asthma) ([@b28-ehp-119-784]). The particles' carbonaceous cores are coated with thousands of organics and heavy metals. Because large numbers of hazardous chemicals are present on DEP, its pathological effects on human airways are pleiotropic. We and others have found that DEP evokes the secretion of matrix metalloproteinase-1 (MMP-1) from human bronchial epithelia ([@b3-ehp-119-784]; [@b14-ehp-119-784]). Matrix metalloproteinase-1 (MMP-1) plays a role in tissue remodeling during development, inflammation, migration of inflammatory and malignant cells, and COPD and emphysema pathogenesis ([@b22-ehp-119-784]). It also has neurotropic effects, possibly enhancing sensitization of airway-innervating sensory neurons, contributing to airway hypersensitization and chronic cough ([@b6-ehp-119-784]). We recently identified a novel pathway that results in DEP-induced *MMP-1* activation and entails activation of RAS-RAF-MEK-extracellular signal--regulated kinase (ERK) signaling, dependent on β-arrestins ([@b14-ehp-119-784]). From a global health perspective, one important finding was that the human *MMP-1* polymorphism at position −1607(1G/2G) of the *MMP-1* promoter yielded, after DEP exposure, either a diminutive (1G) or large (2G) response. The 2G polymorphism is found in 75% of humans. Against this background, we sought to identify critical elements upstream of RAS in human airways in response to DEP. The pathogenic component of DEP that activates *MMP-1* is primarily retained in its organic extract (OE), such that DEP carbonaceous core particles shuttle water-insoluble OE to the ciliary plasma membrane. The DEP/OE initially activates proteinase-activated receptor 2 (PAR-2), which, via G~i/o~ G-protein, phospholipase-Cβ3 (PLCβ3), and phosphatidylinositol 3 kinase (PI3-K), activates Ca^2+^-permeable TRPV4 (transient receptor potential vanilloid, family member 4) ion channels ([@b15-ehp-119-784]; [@b16-ehp-119-784]; [@b25-ehp-119-784]; [@b26-ehp-119-784]). A uniquely protracted Ca^2+^ influx through TRPV4 follows, which is critical for mitogen-activated protein kinase (MAPK)--mediated *MMP-1* activation. Localization studies show that PAR-2, PLCβ3, and TRPV4 colocalize to cilia of human differentiated airway epithelia. DEP exposure greatly enhances protein--protein complex formation between these signaling molecules and calmodulin. Importantly, we observed that TRPV4~P19S~, a human genetic polymorphism previously identified as a COPD susceptibility locus ([@b32-ehp-119-784]), increases *MMP-1* activation via increased Ca^2+^ influx, providing a mechanistic link between human airway epithelia signaling, airway disease, and air pollution. Materials and Methods ===================== DEP --- Particles were generated at the U.S. Environmental Protection Agency (EPA; Research Triangle Park, NC) from a Deutz four-cylinder diesel engine, running at three defined engine loads before collection, as described previously ([@b14-ehp-119-784]). For experiments, we used DEP at 100 μg/mL. DEP organic extract (OE) was prepared by washing organic chemicals off of DEP using methylene chloride, followed by solvent exchange with dimethyl sulfoxide (DMSO). In experiments, we used 20 μg/mL OE, which is equivalent to the organic compounds contained in 100 μg DEP. We used Degussa Printex 90 carbon nanospheres (P90; provided by W. Moeller, GSF, Munich, Germany) as controls. Chemicals --------- We used the following compounds: pertussis toxin (G~i/o~ inhibitor; Sigma Chemical Co., St. Louis, MO), U73122 (PLC inhibitor; Tocris, Ellisville, MO), LY294002 and PI828 (PI3-K inhibitors; Tocris), 4α-phorbol 12,13 didecanoate (4α-PDD; TRPV4 activator; Tocris), GSK205 \[TRPV4 blocker ([@b19-ehp-119-784])\], ruthenium red \[TRP(V) blocker; Tocris\], gadolinium(III) chloride \[GdCl~3~ (Sigma); inhibitor of store-operated calcium entry (SOCE) at 5 μM ([@b4-ehp-119-784])\], thapsigargin (Ca^2+^-store depletion; Tocris), GM1489 and Z-PDLDA-NHOH (pan-MMP inhibitors; Endogen, Rockford, IL), and W-7 (calmodulin blocker at 10 μM; Sigma). Cell culture ------------ BEAS-2B cells were obtained from ATCC (Rockville, MD), maintained as previously described ([@b14-ehp-119-784]), and used for stimulation with DEP or OE and for all DNA and small interfering RNA (siRNA) transfection experiments. Primary human bronchial epithelial (HBE) cells were tracheobronchial cells derived from healthy, nonsmoking adult volunteers. We obtained institutional review board approval for this study from the participating institutions, and volunteer donors provided informed consent for use of the cells in research. Additional details on cell culture are available in Supplemental Material (doi:10.1289/ehp.1002807). *MMP-1* reporter gene assays were conducted as described previously ([@b14-ehp-119-784]). A set of *MMP-1* promoters of different lengths was available for both human polymorphisms, −1607G and −1607GG. siRNA was transfected into BEAS-2B cells following previously published methods ([@b14-ehp-119-784]). siRNA was directed against PAR-2, PAR-1, β-arrestins 1 and 2, and TRPV4. Scrambled controls were used as provided by the manufacturer (Dharmacon, Lafayette, CO). siRNA efficiency was confirmed by quantitative reverse-transcriptase polymerase chain reaction (PCR) and Western blotting. TRPV3 and TRPV4 dominant-negative (DN) isoforms were generated by isolating a truncated form of each channel, from 10 amino- acids N-terminal to the fifth transmembrane domain to 10 amino-acids C-terminal to the sixth transmembrane domain. In addition, two point mutations were generated as M680K and D682K for TRPV4 and as L619K and D621K for TRPV3 in order to render the channel fragments Ca^2+^ impermeable. Both of these constructs were C-terminally fused to monomeric red fluorescent protein (RFP). Dominant-negative isoforms of STIM1 and ORAI1, -2 and -3 were provided by L. Birnbaumer and S. Muallem. These cDNAs were driven by CMV promoters in eukaryotic expression plasmids, the coding region fused to eGFP. DN-STIM1 and DN-ORAI1-3 have been shown to specifically interfere with function of their cognate wildtype isoforms and inhibit them. Enzyme-linked immunosorbent assays (ELISAs) for MMP-1, RANTES (regulated on activation, normal T-expressed and secreted), and IP-10 (interferon-γ--induced protein 10 kDa, CXCL10) were conducted using commercially available kits. For MMP-1 secretion, we previously demonstrated its correlation to the specific proteolytic activity of MMP-1 ([@b14-ehp-119-784]). Ca^2+^ imaging of BEAS-2B cells was conducted using 2 μM fura-2 acetoxymethyl ester for loading and following a protocol for ratiometric Ca^2+^ imaging using 340/380 nm blue light for dual excitation, recording emissions with specific filter sets. Ratios of the emissions were acquired every 5 sec. ΔR/R~0~ is the fraction of the increase of a given ratio over the baseline ratio, divided by baseline ratio. For stimulation of cells with DEP, we used particles at 100 μg/mL and analyzed only cells with microscopically verified contact with particles. For stimulation with OE, all cells were analyzed. To stimulate TRPV4, hypotonicity was used at 260 mOsmol/L, and 4α-PDD at 10 μM; Ca^2+^ stimulation was accomplished by switching from 0 to 2 mM. Ca^2+^ imaging of primary HBE cells was conducted by excising the air--liquid interface matrix with a scalpel and affixing it to the opening of a glass-bottom dish, with other procedures as for BEAS-2B cells. Electrophysiological recordings ------------------------------- Extracellular Ca^2+^ was precipitated by addition of EGTA. We conducted PI3-K Förster resonance energy transfer (FRET) imaging based on FRET of membrane-targeted enhanced green fluorescent protein (eGFP; donor) and the pleckstrin homology domain of Bruton's tyrosine kinase, fused to mCherry fluorescent protein (acceptor). With low phosphatidylinositol (3,4,5)-trisphosphate (PIP3) levels, the mCherry is cytoplasmic; with increased PIP3 levels, it translocates to the membrane leading to FRET, which we quantified using two-photon fluorescence lifetime imaging. Additional details for electrophysiological recordings are given in Supplemental Material (doi:10.1289/ehp.1002807). The phosphorylated ERK (phospho-ERK) trafficking assay was performed as described previously ([@b14-ehp-119-784]). Briefly, BEAS-2B cells were stimulated either with DEP or with OE and fixed (4% paraformaldehyde) at 5-, 10-, 20-, and 30-min time points. Phospho-ERK~1/2~ was verified by fluorescent immunodetection and quantified densitometrically (≥ 75 cells/time point), corrected for background, in the nuclear area using ImageJ software (version 1.42q; [@b20-ehp-119-784]). Confocal imaging was conducted after immunocytochemical staining for acetylated α-tubulin, PLCβ3, TRPV4, and PAR-2. Fluorescently labeled sections were visualized using a Zeiss LSM710 confocal imaging suite with lasers tuned to the emission spectra of the secondary fluorescent antibodies. Coimmunoprecipitation studies were conducted using 10^6^ BEAS-2B cells per experiment; cells were harvested in lysis buffer (1% NP40 detergent). Exactly 100 μg protein was incubated with rabbit anti--PLC-β3, rabbit anti-TRPV4, or mouse monoclonal anti--PAR-2 overnight at 4°C, and then solutions were exposed to 15 μL protein-A/G-Sepharose for 4 hr (4°C). After stringency washing, complexes were investigated by Western blotting using antibodies specific for TRPV4, PAR-2, or calmodulin. Normal rabbit or mouse isotype antibodies were used as controls. Western blotting was performed following standard methodology with chemoluminescence detection. Statistical analysis -------------------- We compared mean and SE of quantified outcome parameters after stimulation with their respective controls. Group comparisons were performed using Student's *t*-test or analysis of variance with post hoc Scheffe test for multigroup comparison, applying the statistics program StatPlus:mac (AnalystSoft, Vancouver, British Columbia, Canada). Minimum significance was set at *p* \< 0.05. Results ======= The OE of DEP contains the active component to activate *MMP-1* --------------------------------------------------------------- To understand which component(s) of DEP activate *MMP-1*, we investigated effects of DEP and its OE in human BEAS-2B and primary HBE cells, the latter exposed at air--liquid interface \[see Supplemental Material, Figure 1 (doi:10.1289/ehp.1002807)\]. Our findings suggest that the carbonaceous core of DEP, by size a carbon nanoparticle, acts as a vehicle carrier for delivery of the highly active, water-insoluble organic fraction to the plasma membrane of human airway epithelia to elicit *MMP-1* activation. Extracellular Ca^2+^ influx is necessary for activation of *MMP-1* ------------------------------------------------------------------ Previous studies in lung cells and neurons have shown that particulate matter evokes Ca^2+^ transients ([@b1-ehp-119-784]); other studies have shown that Ca^2+^ increases activated RAS ([@b13-ehp-119-784]) and that DEP activates RAS ([@b14-ehp-119-784]). Therefore, we examined whether DEP and/or OE causes Ca^2+^ influx and whether this can activate *MMP-1*. We found that DEP and OE evoke extracellular Ca^2+^ influx ([Figure 1A--D](#f1-ehp-119-784){ref-type="fig"}), as indicated by curtailing of the response by addition of EGTA ([Figure 1C,D](#f1-ehp-119-784){ref-type="fig"}). P90 control carbon nanoparticles had no effect on Ca^2+^ ([Figure 1B](#f1-ehp-119-784){ref-type="fig"}), whereas DEP activated a uniquely protracted and monotonically increasing response with a peak at approximately 60 min that gradually declined (data not shown). In comparison, the response to OE increased more rapidly, reaching a maximum at approximately 20 min and decreasing to baseline in the next 10 min ([Figure 1B](#f1-ehp-119-784){ref-type="fig"}), indicating that the particle core retarded Ca^2+^ influx by slowing delivery of the organic fraction to the plasma membrane. To determine whether DEP-induced Ca^2+^ influx was necessary for transcriptional activation of *MMP-1*, we exposed cells to DEP in the presence and absence of extracellular Ca^2+^ and measured *MMP-1* transcriptional activation at 2 and 24 hr and the appearance of nuclear phospho-ERK at 30 min ([Figure 1E--I](#f1-ehp-119-784){ref-type="fig"}). These experiments indicated that extracellular Ca^2+^ was necessary for both nuclear translocation of phospho-ERK and *MMP-1* activation in response to DEP or OE. In primary HBE cells, EGTA eliminated MMP-1 secretion in response to DEP or OE ([Figure 1H](#f1-ehp-119-784){ref-type="fig"}), thus confirming the validity of this mechanism. We were unable to demonstrate functionality of SOCE in DEP/OE--evoked Ca^2+^ influx \[see Supplemental Material, Figure 2A--C (doi:10.1289/ehp.1002807)\]. Preincubation with 5 μM thapsigargin or 5 μM GdCl~3~ ([@b4-ehp-119-784]) did not markedly change DEP-evoked Ca^2+^ increase. Furthermore, cotransfection of STIM1-DN and ORAI1-3--DN, both known to function in SOCE, did not significantly alter Ca^2+^ responses evoked by DEP or OE. Finally, we found that DEP and OE also caused the Ca^2+^-dependent secretion of proinflammatory mediators RANTES ([@b12-ehp-119-784]) and IP-10 ([@b29-ehp-119-784]) \[see Supplemental Material, Figure 2D,E (doi:10.1289/ehp.1002807)\]. Thus, in human lung cells, Ca^2+^ influx is necessary for DEP/OE--evoked activation of *MMP-1*, *IP-10*, and *RANTES*. PAR-2 is a DEP-sensitive G-protein--coupled receptor (GPCR) that activates G~i/o~, PLCβ3, and PI3-K --------------------------------------------------------------------------------------------------- We focused on PAR-2 because an earlier low-throughput proteomics screen revealed that, compared with DEP alone, DEP plus PAR-2--activating peptide (PAR-2-AP) increased MMP-1 secretion (data not shown). We first conducted experiments in BEAS-2B cells by siRNA-mediated knockdown of PAR-2. *PAR-2* mRNA was significantly reduced by PAR-2 siRNA but not by the scrambled control \[see Supplemental Material, Figure 3A (doi:10.1289/ehp.1002807)\]. Cells treated with PAR-2 siRNA exhibited significantly reduced Ca^2+^ influx, *MMP-1* reporter gene activation, and MMP-1 secretion ([Figure 2A](#f2-ehp-119-784){ref-type="fig"}; see also Supplemental Material, Figure 3B,C). This demonstrates that PAR-2 functions upstream of Ca^2+^-mediated *MMP-1* activation. In addition to scrambled siRNA controls, PAR-1--specific siRNA had no effect on *MMP-1* activation (data not shown). Costimulation of BEAS-2B and primary HBE cells with DEP or OE and PAR-2-AP potentiated *MMP-1* activation \[[Figure 2B](#f2-ehp-119-784){ref-type="fig"}; see also Supplemental Material, Figure 3D,E (doi:10.1289/ehp.1002807)\]. To boost its moderate expression level in BEAS-2B cells, we overexpressed PAR-2. This led to increased baseline and DEP-evoked *MMP-1* activation, indicating that PAR-2 overexpression is sufficient to increase *MMP-1* expression and to render the cell more responsive to DEP (see Supplemental Material, Figure 3F). Thus, specific activation and inhibition of PAR-2 imply that this receptor is critical in DEP-mediated Ca^2+^ influx that leads to *MMP-1* activation. We next investigated whether secreted MMP-1 activates PAR-2 proteolytically, as it does for PAR-1 ([@b5-ehp-119-784]), which might explain the protracted Ca^2+^ influx in response to DEP. This was not the case, because MMP inhibitors accelerated *MMP-1* reporter gene activity in response to DEP \[see Supplemental Material, Figure 3G (doi:10.1289/ehp.1002807)\]. We addressed whether β-arrestins are necessary for PAR-2--mediated Ca^2+^ influx in response to DEP or OE ([@b7-ehp-119-784]). This was not the case in view of Ca^2+^ increase in the absence of β-arrestins 1 and 2 \[siRNA-mediated knockdown; see Supplemental Material, Figure 3H (doi:10.1289/ehp.1002807)\]. We previously verified elimination of *MMP-1* activation by siRNA-mediated β-arrestin knockdown ([@b14-ehp-119-784]). Together, these results implicate β-arrestins as MAPK scaffolds necessary for the DEP--MMP-1 response yet dispensable for PAR-2--mediated Ca^2+^ influx in response to DEP or OE. We examined G~i/o~ signaling because of PAR-2's known signal transduction mechanisms via this G-protein ([@b18-ehp-119-784]). We found that the DEP--MMP-1 response, namely, Ca^2+^ influx, *MMP-1* transcription, and MMP-1 secretion, depends on G~i/o~, which we targeted specifically with pertussis toxin in both BEAS-2B and primary HBE cells \[[Figure 2A,B](#f2-ehp-119-784){ref-type="fig"}; see also Supplemental Material, Figure 4A (doi:10.1289/ehp.1002807)\]. Because G~i/o~ is known to activate PLC ([@b8-ehp-119-784]), we next treated cells with PLC-selective inhibitor, U73122, which led to a marked DEP--MMP-1 response ([Figure 2A,B](#f2-ehp-119-784){ref-type="fig"}; see also Supplemental Material, Figure 4B). PLC has several isoforms; we investigated the β-isoforms because of PLCβ's link to GPCRs, specifically PLCβ3, in view of its previously established link to G~i/o~ ([@b23-ehp-119-784]). When we immunolabeled for PLCβ1--4, we found the most robust expression for PLCβ3 in primary HBE cells (data not shown). Interestingly, using a phospho-specific antibody against PLCβ3, we documented phospho-PLCβ3 up-regulation within 30 min after DEP application ([Figure 2C](#f2-ehp-119-784){ref-type="fig"}). This finding can help explain the protracted Ca^2+^ influx because PLCβ3, being upstream of extracellular Ca^2+^ influx, was previously demonstrated to be attenuated by phosphorylation ([@b31-ehp-119-784]). Another phospholipid-metabolizing enzyme that signals downstream of G~i/o~ and upstream of TRP channel Ca^2+^ conductances is PI3-K ([@b33-ehp-119-784]). We identified its critical role in response to DEP or OE using the PI3-K inhibitor LY294002 by documenting significant reduction of Ca^2+^ influx and subsequent *MMP-1* activation in both BEAS-2B and primary HBE cells \[[Figure 2D](#f2-ehp-119-784){ref-type="fig"}; see also Supplemental Material, Figure 4C (doi:10.1289/ehp.1002807)\], suggesting the signaling position of PI3-K upstream of Ca^2+^ influx. Moreover, using a novel FRET-based assay, we could visualize the enzymatic activity of PI3-K (change in PIP3) in BEAS-2B cells in response to DEP or OE, which indicated PI3-K activity as an early signaling event ([Figure 2D](#f2-ehp-119-784){ref-type="fig"}). Furthermore, in addition to time-scale resolution after DEP or OE exposure, this method illustrates the confinement of PI3-K signaling to the plasma membrane. TRPV4 forms a DEP-sensitive Ca^2+^ pathway downstream of PI3-K/PLC-β3 --------------------------------------------------------------------- PAR-2 has been shown to sensitize TRP channels, including TRPV1, TRPV4, and TRPA1 ([@b2-ehp-119-784]; [@b9-ehp-119-784]; [@b10-ehp-119-784]). Because TRPV4 is expressed in tracheobronchial epithelia ([@b16-ehp-119-784]), we addressed whether it functions downstream of the above signaling cascade, initially by inhibiting its function in BEAS-2B cells expressing TRPV4-DN \[see Supplemental Material, Figure 5A--C (doi:10.1289/ehp.1002807)\], which produced strong reduction of Ca^2+^ influx in response to DEP yet no reduction for TRPV3-DN ([Figure 3A](#f3-ehp-119-784){ref-type="fig"}). We also knocked down *TRPV4* using specific siRNA, which effectively down-regulated *TRPV4* mRNA and protein (see Supplemental Material, Figure 5B,C). Compared with the scrambled control, the siRNA-TRPV4 knockdown reduced Ca^2+^ influx in response to DEP or OE ([Figure 3B](#f3-ehp-119-784){ref-type="fig"}). Thus, TRPV4 is necessary for DEP-evoked Ca^2+^ influx. Next we addressed whether DEP-evoked, TRPV4-mediated Ca^2+^ influx activates *MMP-1*. Using *MMP-1* reporter assays, we found that TRPV4-specific siRNA decreased *MMP-1* transcriptional activation, thus implying that TRPV4 is critical for DEP/OE--evoked Ca^2+^ influx, which then activates *MMP-1* \[see Supplemental Material, Figure 5D (doi:10.1289/ehp.1002807)\]. Based on these results, we used the *MMP-1* reporter platform to determine that TRPV4 functions downstream of PAR-2 because siRNA-mediated TRPV4 knockdown virtually eliminated potentiated *MMP-1* activation by DEP or OE plus PAR-2-AP (see Supplemental Material, Figure 5E). This effect of TRPV4-specific siRNA was recapitulated for DEP-evoked MMP-1 secretion in BEAS-2B cells ([Figure 3C](#f3-ehp-119-784){ref-type="fig"}). Finally, we found that ruthenium red, an unspecific TRP(V) blocker, decreased *MMP-1* activation (see Supplemental Material, Figure 5F). TRPV4 activation by 4α-PDD or hypotonicity strongly increased MMP-1 secretion, indicating that in airway epithelia, TRPV4 activation is sufficient to up-regulate *MMP-1* ([Figure 3D](#f3-ehp-119-784){ref-type="fig"}). Furthermore, TRPV4 transfection in BEAS-2B cells increased *MMP-1* reporter activation. Because these findings were obtained in the BEAS-2B cell line, we also tested TRPV4 function in primary HBE cells. First, we were able to significantly attenuate the DEP-evoked Ca^2+^ response by GSK205, a specific small-molecule TRPV4 inhibitor ([@b19-ehp-119-784]), in a dose-dependent manner ([Figure 3E](#f3-ehp-119-784){ref-type="fig"}). In addition, secreted MMP-1 in response to DEP was significantly reduced by two concentrations of GSK205 ([Figure 3F](#f3-ehp-119-784){ref-type="fig"}). Thus, the cornerstones of TRPV4's involvement in the DEP--*MMP-1* response, namely, dependence of the Ca^2+^ response and MMP-1 secretion on TRPV4, were recapitulated in primary HBE cells. Whenever possible, we performed the same experiment in human primary HBE cells as in permanent human BEAS-2B cells. Taken together, these findings point toward critical functioning of TRPV4 in Ca^2+^ influx into human airway epithelia evoked by DEP, a globally relevant air pollutant. TRPV4 signaling complex is located on motile cilia of primary human airway epithelia ------------------------------------------------------------------------------------ Because TRPV4 channels have been found in primary motile cilia of mouse tracheal epithelia ([@b16-ehp-119-784]), we determined TRPV4's subcellular location in human ciliated airway epithelia. Primary HBE cells were differentiated in culture until they became ciliated. They showed ciliary location of TRPV4, PAR-2, and PLCβ3 ([Figure 4](#f4-ehp-119-784){ref-type="fig"}). Thus, critical DEP-responsive membrane-bound components all localize to motile cilia of primary human HBE cells. DEP-facilitated recruitment to a membrane-associated receptor-signaling multicomplex ------------------------------------------------------------------------------------ Given the ciliary localization of the DEP-evoked transduction cascade, we asked whether these membrane-associated signaling molecules coaggregate in response to DEP or OE ([Figure 5](#f5-ehp-119-784){ref-type="fig"}). In aggregate, DEP-responsive membrane-bound signaling is characterized by a nonincremental interaction between PLCβ3 and TRPV4, with subsequent recruitment of PAR-2 and calmodulin caused by DEP exposure. Calmodulin's response to DEP is inhibitory, because specific inhibition of calmodulin with W-7 increased Ca^2+^ signaling and *MMP-1* activation ([Figure 5B,C](#f5-ehp-119-784){ref-type="fig"}). This suggests as explanatory mechanism(s) an increase in intracellular Ca^2+^ concentration (\[Ca^2+^\]*~i~*) via disinhibited TRPV4 and/or activation of PLCβ3, both of which have previously been shown to bind calmodulin. COPD-associated TRPV4~P19S~ increases *MMP-1* activation in response to DEP or OE --------------------------------------------------------------------------------- For human *TRPV4*, a number of genetic polymorphisms enhance susceptibility for COPD; one of them, P19S, is located in the coding region ([@b32-ehp-119-784]). In another study, TRPV4~P19S~ was reported as a DN channel in transfected HEK cells in response to weak but not strong hypotonicity ([@b27-ehp-119-784]). Because our finding that DEP-evoked Ca^2+^ influx via TRPV4 causes *MMP-1* activation rationalizes airway injury by MMP-1 as caused by TRPV4 channel activity, not by DN channels, we attempted to resolve these seemingly contradictory concepts. Compared with wild-type TRPV4 (TRPV4~wt~), TRPV4~P19S~ exhibited gain-of-function effects in Ca^2+^ influx, patch clamp, *MMP-1* reporter gene activation, and MMP-1 secretion \[[Figure 6](#f6-ehp-119-784){ref-type="fig"}; see also Supplemental Material, Figure 6 (doi:10.1289/ehp.1002807)\]. For *MMP-1* transcriptional activation, TRPV4~P19S~ gain-of-function effects were strictly dependent on Ca^2+^ influx. This was evidenced by inhibitory effects of TRPV4~P19S/M680K~, where the selectivity-filter-- blocking M680K mutation causes Ca^2+^ impermeability, leading to elimination of gain of function ([Figure 6D](#f6-ehp-119-784){ref-type="fig"}). Furthermore, Ca^2+^ influx in response to changes in Ca^2+^ concentration and to DEP or OE were significantly increased in TRPV4~P19S~ versus TRPV4~wt~, as was nonstimulated \[Ca^2+^\]*~i~* \[extracellular Ca^2+^ concentration, 2 mM) ([Figure 6A--C](#f6-ehp-119-784){ref-type="fig"}; see also Supplemental Material, Figure 6). Thus, in a human airway epithelium--derived cell line with robust similarity to primary HBE cells, TRPV4~P19S~ functions as Ca^2+^-permeable gain-of-function channel to hyperactivate the pathogenic mediator gene *MMP-1* in response to the common air pollutant DEP. Discussion ========== We identified a novel DEP-activated signaling pathway in human airway epithelia that consists of a GPCR (PAR-2) signaling to a TRP channel (TRPV4). Specific activation of PAR-2 by OE leads to Ca^2+^ influx mediated by TRPV4 channels, via membrane phospholipid signaling involving PLCβ3 and PI3-K. This pathway can be used to develop effective medical therapy for human airway injury caused by airborne particulate pollution, a well-recognized global health problem that contributes to development of COPD and other respiratory illnesses. Regarding COPD, we identified a molecular, cellular, and subcellular mechanism for how the COPD-predisposing nonsynonymous genetic polymorphism TRPV4~P19S~ may potentiate DEP-evoked Ca^2+^ influx and activation of airway-pathogenic *MMP-1.* A schematic of our findings is presented [Figure 7](#f7-ehp-119-784){ref-type="fig"}. In the context of our results, three qualifiers should be mentioned. First, TRPV4 may not be the only critical Ca^2+^ conductance that is activated. Second, regarding generation of phospholipid molecules with modulatory activity on TRPV4 (e.g., by PI3-K and PLCβ3), we speculate that ratios of active small molecules in the immediate vicinity of the channel are critical for channel function. These two aspects deserve further study. Third, mice do not have an *MMP-1*--orthologous gene, thus restricting the direct translatability of our findings into a rodent model that is amenable to genetic engineering. The critical participants of the DEP response, PAR-2, PLCβ3, and TRPV4, were localized to motile cilia of differentiated primary human respiratory epithelia. Motile cilia represent a cellular extension that enhances transfer of DEP organics to ciliary membranes via aerogenic exposure, given the enormous increase in cellular surface ([Figure 7A](#f7-ehp-119-784){ref-type="fig"}). This delivery mechanism can be viewed as "slow-release" delivery of organic chemicals from coated particles to lipid membranes. We also intend to address the question of how this novel signaling pathway has been shaped by evolution. PAR-2 signaling, as yet another variant of sensory signaling ([@b21-ehp-119-784]), has evolved in respiratory cilia likely as a sentinel for activation by microbial proteases, such as granzyme and chitinase. PAR-2 signaling thus would have conferred survival benefits in host defenses of airway integrity yet has subsequently been hijacked by man-made air pollution, a detrimental turn that occurred very recently, during the last milliseconds of the evolutionary clock. A recently published landmark report showed that bitter taste receptors and PLCβ2 localized to motile cilia of human primary airway epithelia ([@b24-ehp-119-784]); however, we found a different isoform, PLCβ3, that localized to respiratory cilia. Whether bitter taste receptor--mediated Ca^2+^ increase will trigger a different set of responses compared with PAR-2/TRPV4--mediated Ca^2+^ influx remains to be established. Our results in the present study suggest that TRPV4-mediated Ca^2+^ influx in response to DEP or OE leads to maladaptive, propathogenic reprogramming of gene-regulatory mechanisms in human airway epithelia. In the novel signaling mechanism in human airway epithelia described here, Ca^2+^ influx is characterized by uniquely slow kinetics. Two possible causes for the slow kinetics are *a*) DEP-dependent PLCβ3 phosphorylation, and *b*) DEP-enhanced calmodulin binding to a receptor-signaling multiplex containing TRPV4 and PLCβ3 ([Figures 5](#f5-ehp-119-784){ref-type="fig"}, [7](#f7-ehp-119-784){ref-type="fig"}), both of which attenuate signaling via known properties of the modified phospholipase or channel. The attractive hypothesis of proteolytic activation of PAR-2 by MMP-1 has not been not corroborated. This study is relevant for global human health because of the global presence of DEP. However, we also discovered a possibly novel mechanism of airway injury that is caused by DEP yet enhanced by the human COPD-susceptibility polymorphism TRPV4~P19S~. Our identification of TRPV4~P19S~ as a gain-of-function Ca^2+^-permeable channel in a human respiratory epithelial cell line, in response to DEP, links COPD pathogenesis to pathologically increased Ca^2+^ influx into human airway epithelia elicited by a globally relevant air pollutant. Furthermore, our results imply that two human genetic polymorphisms are linked to respiratory health, TRPV4~P19S~ and *MMP-1* (−1607G/GG), thus highlighting the concept of disease susceptibility as a function of genetic "makeup" combined with environmental insults. Finally, we note yet another translational medical implication: The novel pathway described here can be targeted by inhalation of compounds that can specifically inhibit critical signaling molecules. In other words, although DEP injures respiratory epithelia via a luminal--apical unloading mechanism of DEP organics delivered by carbonaceous nanoparticles, this very same route could become the avenue for safe and effective therapy now that key participants are known. Supplemental Material is available online (doi:10.1289/ehp.1002807 via <http://dx.doi.org/>). R. Lefkowitz (Duke University) provided invaluable discussions and suggestions regarding signaling and β-arrestin--related reagents, and J. Putney \[National Institute of Environmental Health Sciences (NIEHS)\] provided enlightened discourse on Ca^2+^ entry mechanisms and their pharmacology. The following reagents were kindly supplied by the following colleagues: P90 carbon particles, W. Moeller (GSF, Munich, Germany); *MMP-1* reporter gene constructs, C.E. Brinckerhoff (Dartmouth University); dominant-negative ORAI constructs, L. Birnbaumer (NIEHS); and dominant-negative STIM1 construct, S. Muallem (University of Texas Southwestern Medical Center. This research was supported by startup funds from Duke University and financial support from R.J. Reynolds, Inc., to W.L.; by a grant from Philip Morris USA Inc. and Philip Morris International to S.A.S.; and by U.S. Environmental Protection Agency (EPA) internal funds to A.G. Fellowship support for J.L. (Duke University Integrated Toxicology and Environmental Health Program, Leon-Goldberg Fellowship) and S.H.C. (Research Participation Program in U.S. EPA administered by the Oak Ridge Institute for Science and Education) is acknowledged. S.A.S. and W.L. received funding from Philip Morris USA and Philip Morris International, and W.L. received funding from R.J. Reynolds. The other authors declare they have no actual or potential competing financial interests. {#f1-ehp-119-784} ![Signal transduction in response to DEP or OE involves PAR-2, G~i/o~, and PLCβ3. (*A*) In BEAS-2B cells, Ca^2+^ response (ΔR/R~0~ vs. time) to OE is significantly attenuated with PAR-2 siRNA \[for proof of efficiency, see Supplemental Material, Figure 3A (doi:10.1289/ehp.1002807)\], pertussis toxin (PTX; G~i/o~ inhibitor), the PLC inhibitor U73122, and the PI3-K inhibitor LY294002 (*n* ≥ 50 cells per condition). (*B*) In primary HBE cells, modulation of these pathways affects MMP-1 secretion; PAR-2 gain of function potentiates the response, whereas inhibition of G~i/o~, PLC, and PI3-K significantly attenuates/eliminates it. For results in BEAS-2B cells, including PAR-2 gain and loss of function (siRNA), plus all other pathways, see Supplemental Material, Figures 3 and 4 (doi:10.1289/ehp.1002807). (*C*) Increase in phospho-PLCβ3 in response to DEP shown by representative confocal micrographs of immunolabeled primary HBE cells (top; bar = 5 μm) and densitometry (bottom; *n* ≥ 50 cells per time point). Results show the phospho-PLCβ3 increase peaking at 30 min and decreasing at 60 min. (*D*) PI3-K activity in live transfected BEAS-2B cells shown by real-time imaging in response to OE using a novel FRET-based methodology. FRET measurements are shown in micrographs (top; bar = 4 μm), reiterating that PI3-K activity resides in the plasma membrane. The time course of PIP3 generation (bottom) shows that FRET generated by membrane-targeted eGFP and mCherry-tagged pleckstrin homology domain of Bruton's tyrosine kinase increases robustly after application of OE and became significantly attenuated in response to subsequent application of the specific PI3-K inhibitor LY294002 (10 μM). The graph (bottom) shows the time course of the change in PIP3 evoked by OE (quantified at the membrane) and its attenuation by preincubation with another specific PI3-K inhibitor, PI828 (50 μM).\ \**p* \< 0.05, and *^\#^p* \< 0.001 compared with the 0-min time point.](ehp-119-784f2){#f2-ehp-119-784} ![TRPV4 channels are critical for DEP-evoked *MMP-1* activation and function downstream of PAR-2. Abbreviations: +, with; −, without. (*A*) The DEP-evoked Ca^2+^ response in BEAS-2B cells was significantly and specifically attenuated by inhibition of TRPV4 upon transfection of cells with RFP-tagged TRPV4-DN (*n* = 33) and unaffected by transfection with TRPV3-DN \[*n* = 44; see also Supplemental Material, Figure 4A (doi:10.1289/ehp.1002807)\]. (*B*) TRPV4 loss of function is illustrated by robust inhibition of OE-evoked Ca^2+^ responses in BEAS-2B cells transfected with TRPV4 siRNA compared with transfection with scrambled control siRNA; for proof of efficiency of TRPV4 siRNA, see Supplemental Material, Figure 5B,C. (*C*) siRNA-mediated TRPV4 knockdown in BEAS-2B cells virtually eliminated *MMP-1* activation, thus confirming that TRPV4 functions downstream of PAR-2; note the elimination of MMP-1 secretion in response to DEP or DEP plus PAR-2-AP when TRPV4 is knocked down with TRPV4 siRNA. (*D*) TRPV4 activation is sufficient to activate *MMP-1* hypotonicity (HYPO; 260 mosmol/L) and the specific TRPV4 activator 4α-PDD (10 μM) induced MMP-1 secretion from BEAS-2B cells. (*E*) Specific inhibition of TRPV4 with GSK205 led to dose-dependent reduction of DEP-evoked Ca^2+^ responses in primary HBE cells. (*F*) Specific inhibition of TRPV4 with GSK205 led to a significant reduction of DEP-evoked MMP-1 secretion in primary HBE cells; note the increasing effect with increasing dose.\ \**p* \< 0.05, \*\**p* \< 0.01, and *^\#^p* \< 0.001.](ehp-119-784f3){#f3-ehp-119-784} {#f4-ehp-119-784} {#f5-ehp-119-784} ![In BEAS-2B cells, TRPV4~P19S~ functions as a gain-of-function channel in response to DEP, causing increased *MMP-1* activation via influx of Ca^2+^. (*A*) Time course showing increased Ca^2+^-facilitated activation of TRPV4~P19S~ compared with TRPV4~wt~ (*n* ≥ 24 cells per group). (*B*) Time course showing OE-evoked Ca^2+^ influx in Ca^2+^-free buffer (note the presence of Ca^2+^ in OE) followed by the addition of external Ca^2+^ (2 mM), which leads to accelerated Ca^2+^ influx for TRPV4~P19S~ versus TRPV4~wt~. The signal of TRPV4~wt~ declined after the peak, whereas TRPV4~P19S~-transfected cells did not desensitize; the signal remained high at least during this observation period. Arrows in *A* and *B* indicate the time of Ca^2+^ or OE exposure. (*C*) Internal \[Ca^2+^\]*~i~* was significantly elevated in TRPV4~P19S~-expressing BEAS-2B cells cultured in external media containing Ca^2+^. (*D*) In keeping with Ca^2+^ responses shown in *B*, TRPV4~P19S~ increased *MMP-1* transcriptional activation. First, transfection of TRPV4~wt~ increased *MMP-1* reporter gene activation in response to OE by approximately 2-fold. Second, transfection of TRPV4~P19S~ strikingly increased baseline *MMP-1* reporter gene activation in response to OE by a factor of approximately 4--5 versus control-transfected cells and by a factor of \> 2 versus TRPV4~wt~. TRPV4~wt~- and TRPV4~P19S~-mediated increases were virtually eliminated with a second point mutation, M680K (selectivity-filter block). This finding indicates that the effects of TRPV4~wt~ and TRPV4~P19S~ are mediated by an influx of external Ca^2+^ through the channel's pore. (*E*) Validity of results and conclusions for MMP-1 secretion by transfected BEAS-2B cells.\ \**p* \< 0.05, and \*\**p* \< 0.01.](ehp-119-784f6){#f6-ehp-119-784} {ref-type="fig"}). (*B*) The signaling cascade begins with activation of PAR-2 (green); this ultimately leads to the influx of Ca^2+^ (dark green circles) via TRPV4 (blue) by GPCR signaling encompassing G~i/o~, which in turn leads to activation of PLCβ3 and PI3-K. PLCβ3 is phosphorylated (Phos) in response to DEP, partially accounting for the protracted Ca^2+^ response. PLCβ3 and PI3-K then regulate Ca^2+^ influx through TRPV4, which binds calmodulin (CaM), which is enhanced by DEP exposure; the increased CaM also protracts Ca^2+^ influx. (*C*) TRPV4-mediated Ca^2+^ entry activates RAS-RAF-MEK MAPK signaling ([@b14-ehp-119-784]), resulting in reprogramming of transcriptional mechanisms that orchestrate remodeling of the extracellular matrix via activation of *MMP-1*. The COPD-susceptibility polymorphism TRPV4~P19S~ functions as a gain-of-function channel for additional Ca^2+^ influx and *MMP-1* activation, thus being relevant to human health.](ehp-119-784f7){#f7-ehp-119-784} [^1]: These authors contributed equally to this work. [^2]: Deceased.

1. Introduction {#sec1-ijerph-17-01164} =============== Diabetes represents a clustering of heterogeneous risk factors that affects all aspects of quality of life due to the lifetime demands of diabetes care \[[@B1-ijerph-17-01164]\]. Health-related quality of life (HRQoL) refers to a multi-dimensional concept of quality of life, which can be affected by illnesses or treatments. Thus, assessment of HRQoL is an important outcome measure to evaluate overall disease management for patients with diabetes, including disease course, early detection of complications, and effectiveness and efficacy of interventions \[[@B2-ijerph-17-01164]\]. Comorbidity is defined as the presence of chronic conditions existing concurrently with a primary disease \[[@B3-ijerph-17-01164]\]. Patients with diabetes often have multiple comorbidities, including hypertension, overweight or obesity, hyperlipidemia, chronic kidney disease, and cardiovascular disease, all of which can impact the patient's HRQoL \[[@B4-ijerph-17-01164],[@B5-ijerph-17-01164]\]. Consequently, the management of chronic conditions in patients with diabetes will benefit from a focus on HRQoL. Cardiorespiratory fitness (CRF), which is defined as the minimum volume of oxygen consumption during a maximal exercise test, has been found to protect against chronic diseases and premature death from those diseases \[[@B6-ijerph-17-01164],[@B7-ijerph-17-01164]\]. For example, higher CRF is associated with lower risks of morbidity and mortality in younger and older adults \[[@B8-ijerph-17-01164]\]. Some studies have examined the association between physical activity and HRQoL in healthy persons, as well as those with chronic diseases, reporting positive relationships for both groups \[[@B9-ijerph-17-01164],[@B10-ijerph-17-01164]\]. Unlike physical activity (PA), less attention has been paid to the relationship between CRF and HRQoL. Although research in this area is limited, the current evidence suggests a positive relationship between CRF and HRQoL in patients with chronic and metabolic diseases \[[@B11-ijerph-17-01164]\]. In older persons, however, objectively measuring CRF can be difficult due to the requirements related to its measurement, such as specialized equipment; trained personnel; amount of time, volitional exertion; and mobility. Importantly, studies have found that CRF can be estimated with acceptable accuracy from routinely obtained health indicators \[[@B12-ijerph-17-01164],[@B13-ijerph-17-01164]\]. We have previously shown that non-exercise-based estimation of cardiorespiratory fitness (eCRF) can serve as a prognostic tool for estimating the risk of mortality from all and specific causes in older Korean adults \[[@B14-ijerph-17-01164]\]. Nationwide data in South Korea revealed that a majority of patients with diabetes have one or more comorbidities which can affect HRQoL, including obesity, hypertension, hypercholesterolemia, and others \[[@B15-ijerph-17-01164]\]. Health behaviors, including PA and CRF, likely mediate the impact of comorbidities on HRQoL among these patients. To the best of our knowledge, however, no previous studies have examined the mediating effect of CRF on the relationships between comorbidities and HRQoL in patients with diabetes in South Korea. Therefore, we investigated whether eCRF mediates the impact of comorbidities on HRQoL in older Korean adults with type-2 diabetes. 2. Materials and Methods {#sec2-ijerph-17-01164} ======================== 2.1. Study Design and Participants {#sec2dot1-ijerph-17-01164} ---------------------------------- The cross-sectional data used for this study were drawn from the Korea National Health and Nutrition Examination Surveys (KNHANES) IV and V, which are nationwide surveys that were conducted from 2008 until 2011 in South Korea. A detailed description of the survey sampling method is available elsewhere \[[@B16-ijerph-17-01164],[@B17-ijerph-17-01164]\]. For the current study, we initially selected a total of 2757 adults aged 60 years and older from those who participated in the 2008--2011 KNHANES IV and V. We then excluded 1386 individuals because no baseline data were available regarding body composition (*n* = 352), resting heart rate (*n* = 799), PA (*n* = 17), and HRQoL (*n* = 218). Consequently, a total of 1371 older adults with diabetes (604 men; 767 women) were included in the final data analyses. The presence of diabetes was determined with a self-reported questionnaire that asked whether the participants had ever received a diagnosis of diabetes from a physician. The institutional review board of human study reviewed and approved the study protocol participants (SKKU 2017-06-009). Informed consent was obtained from all participants in the study. 2.2. Study Variables {#sec2dot2-ijerph-17-01164} -------------------- ### 2.2.1. Assessment of HRQoL (Dependent Variable, Y) {#sec2dot2dot1-ijerph-17-01164} HRQoL was assessed with the EuroQoL group, which consists of a health-status descriptive system (EQ-5D) and a visual analogue scale (EQ-VAS). The EQ-5D records the level of self-reported problems in five dimensions: mobility, self-care, usual activities, pain/discomfort, and anxiety/depression \[[@B18-ijerph-17-01164],[@B19-ijerph-17-01164]\]. Each of the dimensions is assessed based on a single question with three response levels (no problems, some problems, and extreme problems). Scores on the EQ-5D index range from −0.171 to 1, where 1 indicates no problems in any of the five dimensions, zero indicates death, and negative values indicate a health status worse than death. Next, patients report their health status with the EQ-VAS, which involves a VAS ranging from 0 (worst imaginable health) to 100 (best imaginable health) \[[@B18-ijerph-17-01164]\]. ### 2.2.2. Assessment of Comorbidities (Independent Variable, X) {#sec2dot2dot2-ijerph-17-01164} Participants were asked if they had ever been diagnosed by a physician with any of the following medical condition(s): malignancy, hypertension, heart disease (acute myocardial infarction or angina), stroke, arthritis, and/or chronic renal disease. ### 2.2.3. Estimation of Cardiorespiratory Fitness (Mediator, M) {#sec2dot2dot3-ijerph-17-01164} Non-exercise-based eCRF was calculated as one-minute peak volume of oxygen consumption (VO~2peak~) in units of metabolic equivalents (METs), in accordance with previously reported procedures \[[@B13-ijerph-17-01164]\]: Once the algorithms were implemented, participants were classified into low (lowest 25%), middle (middle 50%), and high (highest 25%) categories on the basis of sex-specific tertiles of the estimated peak VO~2~ distributions. ### 2.2.4. Covariates {#sec2dot2dot4-ijerph-17-01164} Measured covariates included age, sex, household income, education level (lower than elementary school, middle/high school, or college or higher), marital status (yes or no), current smoker (never or past/current), frequency of alcohol consumption (more or less than twice per week), and regular exercise (yes or no). 2.3. Statistical Analyses {#sec2dot3-ijerph-17-01164} ------------------------- All variables were checked for normality, both visually and through the Kolmogorov--Smirnov test, and subjected to an appropriate transformation, if necessary, prior to statistical analyses. Descriptive statistics are presented as means and standard deviations for continuous variables and as frequencies and percentages for categorical variables. Analysis of variance (ANOVA) was used to test linear trends in outcome variables according to number of comorbidities and eCRF categories. We examined the relationships between number of comorbidities, eCRF, and HRQoL using parametric and non-parametric statistics. Then, the impact of comorbidities on HRQoL through eCRF was tested based on four criteria for the mediation paths proposed by Baron and Kenny \[[@B20-ijerph-17-01164]\], as illustrated in [Figure 1](#ijerph-17-01164-f001){ref-type="fig"}: (1) the coefficient of path "a" is significant in identifying the effect of the independent variable (IV) on the mediating variable (MV); (2) the MV is significantly related to the dependent variable (DV) of the IV (path b); (3) a significant direct association (path c) between the IV and DV is confirmed; and (4) the association between the IV and DV is weakened when the MV is controlled (path c′). The PROCESS macro in SPSS-PC (version 23.0, IBM Corporation, Armonk, NY, USA) was used to carry out the mediation analyses. No covariates were included in Model 1, demographics and socio-economic status (SES) variables were included in Model 2, and variables of health behaviors were added to Model 3. The possible mediating effect of eCRF on the impact of comorbidities on HRQoL was identified using the Sobel mediation test with a bootstrapping process, which overcomes the assumption of a normal distribution of the classical Sobel mediation test. The robustness of *p*-values was confirmed via the bootstrapping process. An α level of 0.05 was used for all statistical analyses. 3. Results {#sec3-ijerph-17-01164} ========== [Table 1](#ijerph-17-01164-t001){ref-type="table"} summarizes descriptive statistics of participants. Overall, male participants were older (*p* \< 0.001), heavier (*p* \< 0.001), more educated (*p* \< 0.001), and less likely to be alone (*p* \< 0.001) than female participants. Men had higher values for smoking (*p* \< 0.001) and alcohol intake (*p* \< 0.001), but lower values for resting heart rate (*p* = 0.023) and number of comorbidities (*p* \< 0.001) than women. Men were less physically active (*p* \< 0.001) but had higher values on the EQ-5D index (*p* \< 0.001) and EQ-VAS (*p* \< 0.001) than women. We found no significant differences in marital status or income between older men and women. [Table 2](#ijerph-17-01164-t002){ref-type="table"} compares the measured parameters in relation to number of comorbidities. Significantly, positive, linear trends in mean age (*p* \< 0.001), solitary status (*p* = 0.022), and smoking status (*p* \< 0.001) and significant, negative linear trends in education (*p* = 0.001), alcohol consumption (*p* = 0.004), EQ-5D index (*p* \< 0.001), and EQ-VAS score (*p* = 0.002) were found in relation to the number of comorbidities. In general, patients with one or more comorbidities were older, more likely to live alone, less likely to smoke, consumed alcohol less frequently, and had worse HRQoL than patients with no comorbidities. [Table 3](#ijerph-17-01164-t003){ref-type="table"} compares the measured parameters according to eCRF category. Significantly, negative linear trends in mean age (*p* \< 0.001), solitary status (*p* = 0.003), physical activity (*p* \< 0.001), and number of comorbidities (*p* = 0.008) and significant, positive linear trends in education (*p* = 0.002), alcohol consumption (*p* = 0.034), EQ-5D index (*p* \< 0.001), and EQ-VAS score (*p* \< 0.001) were found in relation to increased eCRF (from low to high). Patients who were physically fit were younger, less likely to live alone, less physically active, had more education, consumed alcohol more frequently, had fewer comorbidities, and had higher values on the EQ-5D index and EQ-VAS than patients who were less physically fit. No significant linear trends in sex, marital status, income, or smoking were found in relation to eCRF category. [Table 4](#ijerph-17-01164-t004){ref-type="table"} represents the impact of eCRF on the associations between comorbidities and HRQoL. Mediation analysis showed that number of comorbidities had a direct effect on HRQoL independent of eCRF (*β~c'~* = −1.913, *p* \< 0.005; *c'* path). However, the number of comorbidities also indirectly affected HRQoL through its effect on eCRF. The number of comorbidities was negatively associated with eCRF (*βa* = −0.528, *p* \< 0.001; *a* path), and eCRF was positively associated with HRQoL (*β~b~* = 1.434, *p* \< 0.001; *b* path). Both comorbidity and eCRF remained significant predictors of HRQoL even after adjusting for demographics and SES (i.e., age, marital status, living condition, and education in Model 2 (*β~c'~* = −2.327, *p* \< 0.001; *c'* path; *β~b~* = 0.996, *p* \< 0.001; b path, respectively), as well as health behaviors (i.e., smoking, alcohol consumption, and regular exercise) in Model 3 (*β~c'~* = −2.039, *p* \< 0.005; *c'* path and *β~b~* = 1.153, *p* \< 0.005; b path, respectively). The mediating effect of eCRF on the impact of comorbidities on HRQoL was further tested using the Sobel mediation test with a bootstrapping procedure ([Table 4](#ijerph-17-01164-t004){ref-type="table"} and [Figure 1](#ijerph-17-01164-f001){ref-type="fig"}). The Sobel mediation test showed a significant indirect effect of eCRF on the impact of comorbidities on HRQoL (Z = −4.632, *p* \< 0.001). The results of the bootstrap procedure corroborated those of the Sobel test: the 95% bias-corrected confidence interval (95% CI −1.104 to −0.453) was non-zero, indicating that eCRF mediated the relationship between comorbidities and HRQoL (Model 1) and accounted for 28.4% of the total effect on HRQoL. The Sobel mediation effect of eCRF on the relationship between comorbidities and HRQoL remained significant even after adjusting for demographics and SES in Model 2 (Z = −3.004, *p* = 0.001), with 16.1% of the total effect being explained. Likewise, this relationship remained significant after adjustments for Model 2 plus inclusion of parameters for health behaviors in Model 3 (Z = −2.753, *p* = 0.005), with 12.1% of the total effect explained. 4. Discussion {#sec4-ijerph-17-01164} ============= In this study, we examined the mediating effect of eCRF on the relationships between comorbidities and HRQoL in older Korean adults with diabetes. Our findings show that both comorbidities and eCRF are important predictors of HRQoL among older Korean patients. This is the first study to report that eCRF mediates the impact of comorbidities on HRQoL partially and independently among patients with diabetes. The current findings agree with previous studies reporting an inverse relationship between comorbidities and HRQoL in patients with diabetes. For example, Wexler et al. \[[@B21-ijerph-17-01164]\] examined the impacts of medical comorbidities and depression on HRQoL in a large primary care cohort of patients with type 2 diabetes and showed that treatment of depression and prevention of complications were important determinants of HRQoL. By analyzing the treatment options for type 2 diabetes in adolescents and youth (TODAY) study data, Larkin et al. \[[@B22-ijerph-17-01164]\] showed that HRQoL was negatively associated with depressive symptoms and number of comorbidities in youth with type 2 diabetes. This inverse relationship between HRQoL and comorbidities has also been observed in patients with chronic conditions \[[@B23-ijerph-17-01164]\], dementia \[[@B24-ijerph-17-01164]\], psoriatic arthritis \[[@B25-ijerph-17-01164]\], and survivors of breast \[[@B26-ijerph-17-01164]\] and colorectal cancer \[[@B27-ijerph-17-01164]\]. Likewise, the current findings are consistent with previous studies reporting a positive relationship between CRF and HRQoL in patients with type 2 diabetes \[[@B28-ijerph-17-01164]\], pediatric patients with type 1 diabetes \[[@B29-ijerph-17-01164]\], and women at risk of gestational diabetes \[[@B30-ijerph-17-01164]\]. Using patient records from the 2002 to 2012 Calidad de Vida Center, Clennin et al. \[[@B11-ijerph-17-01164]\] examined the relationship between CRF and HRQoL in a Uruguayan cohort at risk for developing cardiovascular disease (CVD) and found positive associations between CRF and several dimensions of HRQoL (women: vitality, physical health, physical role, bodily pain, and general health; men: physical health). In children and adolescents with type 1 diabetes, Lukács et al. \[[@B31-ijerph-17-01164]\] examined associations among HRQoL, clinical profiles, anthropometric measures, and physical activity and CRF and found that good CRF was significantly associated with better HRQoL and favorable metabolic control among the youth. Similarly, this positive association between CRF and HRQoL was observed in community-dwelling older adults \[[@B32-ijerph-17-01164]\] and workers \[[@B33-ijerph-17-01164]\], as well as in patients with CVD \[[@B11-ijerph-17-01164]\], McArdle disease \[[@B34-ijerph-17-01164]\], severe mental illness \[[@B35-ijerph-17-01164]\], and survivors of cancer \[[@B36-ijerph-17-01164]\]. Our study is the first to extend these findings to show that CRF may mediate the impact of comorbidities on HRQoL in patients with diabetes. The mediating effect of eCRF on the relationship between comorbidities and HRQoL was independent of potential covariates, including demographics (age, sex, marital status), SES (income and education), health behaviors (smoking, alcohol consumption, and regular exercise) and depressive symptoms. In support of the current findings, the findings of the FIT study involving 46,979 patients showed that achieving ≥12 METs reduced the incidence of diabetes compared to achieving \<6 METs in both non-obese (hazard ratio = 0.73; 95% CI = 0.59−0.91; *p* \< 0.001) and obese individuals (hazard ratio = 0.56; 95% CI 0.42−0.73; *p* \< 0.001) \[[@B37-ijerph-17-01164],[@B38-ijerph-17-01164]\]. In that study \[[@B37-ijerph-17-01164]\], the protective effect of achieving high CRF was found to be independent of a number of confounders, including age, sex, race, BMI, history of hypertension, hypertension medication use, ACE inhibitor use, angiotensin II receptor blocker use, b-blocker use, diuretic use, history of hyperlipidemia, lipid lowering medication use, statin use, history of obesity, family history of coronary heart disease, current smoking status, sedentary lifestyle, treated pulmonary disease, depression medication use, and indication for stress testing. Taken together, those findings suggest that high CRF may alleviate the impact of comorbidities on HRQoL in patients with diabetes. Thus, promotion of CRF in conjunction with management of comorbidities targeted at increasing HRQoL should be a key component of the overall management of such patients. HRQoL worsens as the number of diabetes-related comorbidities increases \[[@B21-ijerph-17-01164]\]. Thus, information regarding determinants and/or mediators of HRQoL is essential for development and implementation of an effective care program to promote physical and mental well-being for older patients with diabetes. Higher levels of physical activity and fitness are associated with lower risks of morbidity and mortality in healthy adults and patients with chronic diseases. Higher CRF is also associated with better HRQoL in healthy adults and in patients with chronic disease. The current study has some limitations. First, the cross-sectional study design does not allow causal inference regarding the relationships among comorbidities, eCRF, and HRQoL. A randomized controlled trial on exercise training would be necessary to confirm the mediating effect of CRF on the relationship between comorbidities and HRQoL in a cause-and-effect manner. Second, several potential mediators such as sex, SES, depression, and social networking were identified in previous studies \[[@B29-ijerph-17-01164]\]. A more complex mediation model or examination of a larger number of mediation variables should help elucidate the roles of mediator(s) in determining the relationships between exposure and outcomes. Third, the questionnaire of the survey used in this study was not designed to identify the type of diabetes, which should be addressed in a future study. Lastly, a further study will be necessary to investigate the mechanism(s) through which CRF mediates the relationship between comorbidities and HRQoL in diabetes. 5. Conclusions {#sec5-ijerph-17-01164} ============== In summary, our findings indicate the mediating effect of eCRF on the relationship between comorbidities and HRQoL in older Korean adults with diabetes. Assessment of eCRF should be considered an important measure for evaluating the overall management of diabetes. From a public health perspective, these findings could help guide public agencies when developing future healthcare policies to promote health equality for geriatric patients with diabetes in South Korea. The authors thank to Kelly O'Keefe (English editor) and anonymous Reviewers who allowed us to improve the manuscript. Conceptualization, I.L. and H.K.; Methodology, I.L., S.K. and H.K.; Formal Analysis, I.L., S.K. and H.K.; Investigation, I.L., S.K. and H.K.; Writing-Original Draft Preparation, I.L. and H.K.; Writing-Review & Editing, I.L. and H.K.; Funding Acquisition, National Research Foundation (NRF), S.K. All authors have read and agreed to the published version of the manuscript. This study was supported by the National Research Foundation Grant funded by the Korean Government (NRF-2017R1A2B4010684). The authors declare no conflict of interest. {#ijerph-17-01164-f001} ijerph-17-01164-t001_Table 1 ###### Descriptive statistics of study participants (Mean ± SD). ---------------------------------------------------------------------------------------------- Variables All\ Men\ Women\ *p*-Value (*n* = 1371) (*n* = 604) (*n* = 767) ---------------------------------- --------------- --------------- --------------- ----------- *Demographics* Age (years) 69.1 ± 6.1 68.3 ± 5.9 69.8 ± 6.2 \<0.001 Marital status, *n* (%) 0.606   Yes 1360 (99.2) 600 (99.3) 760 (99.1)   No 11 (0.8) 4 (0.7) 7 (0.9) Living condition, *n* (%) \<0.001 Live someone 1176 (85.8) 564 (93.4) 612 (79.8) Alone 195 (14.2) 40 (6.6) 155 (20.2) *Socioeconomic Status* Income (10,000 won/month) 228.1 ± 113.0 217.1 ± 319.9 236.8 ± 148.6 0.752 Education, *n* (%) \<0.001   Lower than elementary school 558 (40.7) 122 (20.2) 436 (56.8)   Middle/high school 597 (43.5) 304 (50.3) 293 (38.2)   Over than college 216 (15.8) 178 (29.5) 38 (5.0) *Health Behavior Factors* \<0.001 Smoking, *n* (%)   Never 769 (56.1) 87 (14.4) 682 (88.9)   Past/current 602 (43.9) 517 (85.6) 85 (11.1) Weekly alcohol intake, *n* (%) \<0.001   \<2 1137 (82.9) 394 (65.2) 743 (96.9)   ≥2 234 (17.1) 210 (34.8) 24 (3.1) Regular exercise, *n* (%) \<0.001   Yes 430 (31.4) 243 (40.2) 187 (24.4)   No 941 (68.6) 361 (59.8) 580 (75.6) *Number of Comorbidity, n (%)* \<0.001 0 220 (16.0) 133 (22.0) 87 (11.3) 1 567 (41.4) 286 (47.4) 281 (36.6) ≥2 584 (42.6) 185 (30.6) 399 (52.1) *eCRF Variables* eCRF (METs) 7.1 ± 2.3 9.1 ± 1.6 5.6 ± 1.5 \<0.001 BMI (kg/m^2^) 24.7 ± 3.3 24.0 ± 3.0 25.3 ± 3.5 \<0.001 RHR (beats/min) 73.0 ± 10.2 72.3 ± 10.5 73.5 ± 9.8 0.023 Physical activity score, *n* (%) \<0.001   Level 1 224 (16.3) 66 (10.9) 158 (20.6)   Level 2 642 (46.9) 283 (46.9) 359 (46.8)   Level 3 23 (1.7) 8 (1.3) 15 (2.0)   Level 4 95 (6.9) 40 (6.6) 55 (7.2)   Level 5 387 (28.2) 207 (34.3) 180 (23.4) *Health-Related Quality of Life* EQ-5D problems   Mobility, *n* (%) 621 (45.3) 189 (31.3) 432 (56.3) \<0.001   Self-care, *n* (%) 205 (15.0) 55 (9.1) 150 (19.6) \<0.001   Usual activities, *n* (%) 409 (29.8) 121 (20.0) 288 (37.5) \<0.001   Pain/discomfort, *n* (%) 585 (42.7) 189 (31.3) 396 (51.6) \<0.001   Anxiety/depression, *n* (%) 260 (19.0) 79 (13.1) 181 (23.6) \<0.001 EQ-5D index 0.85 ± 0.18 0.90 ± 0.15 0.81 ± 0.19 \<0.001 EQ-VAS score 67.0 ± 22.1 70.1 ± 19.0 64.6 ± 24.0 \<0.001 ---------------------------------------------------------------------------------------------- SD---standard deviation; eCRF---estimated cardiorespiratory fitness; MET---metabolic equivalents; BMI---body mass index; RHR---resting heart rate; EQ-5D---euroqol-5 dimension; EQ-VAS---euroqol-visual analogue scale. ijerph-17-01164-t002_Table 2 ###### Comparison of measurement variables according to number of comorbidity (Mean ± SD). Variables Number of Comorbidity *p* for Linear Trend -------------------------------- ----------------------- ---------------------- --------------- --------- *Socio-Demographic Status* Age (years) 67.6 ± 6.0 69.3 ± 6.3 69.6 ± 5.9 \<0.001 Marital status, *n* (%) 0.194   Yes 219 (99.5) 564 (99.5) 577 (98.8)   No 1 (0.5) 3 (0.5) 7 (1.2) Living condition, *n* (%) 0.022   Live someone 197 (89.5) 491 (86.6) 488 (83.6)   Alone 23 (10.5) 76 (13.4) 96 (16.4) *Socio-Economic Status* Income (10,000 won/month) 236.5 ± 386.3 268.0 ± 172.2 186.8 ± 289.2 0.583 Education, *n* (%) 0.001   Lower than elementary school 71 (32.3) 234 (41.3) 253 (43.3)   Middle/high school 106 (48.2) 234 (41.3) 257 (44.0)   Over than college 43 (19.5) 99 (17.4) 74 (12.7) *Health Behavior Factor* Smoking, *n* (%) \<0.001   Never 102 (46.4) 285 (50.3) 382 (65.4)   Past/current 118 (53.6) 282 (49.7) 202 (34.6) Weekly alcohol intake, *n* (%) 0.004   \<2 181 (82.3) 444 (78.3) 512 (87.7)   ≥2 39 (17.7) 123 (21.7) 72 (12.3) Regular exercise, *n* (%) 0.251   Yes 79 (35.9) 172 (30.3) 179 (30.7)   No 141 (64.1) 395 (69.7) 405 (69.3) *HRQoL* EQ-5D problems   Mobility, *n* (%) 69 (31.4) 221 (39.0) 331 (56.7) \<0.001   Self-care, *n* (%) 24 (10.9) 58 (10.2) 123 (21.1) \<0.001   Usual activities, *n* (%) 41 (18.6) 136 (24.0) 232 (39.7) \<0.001   Pain/discomfort, *n* (%) 73 (33.2) 209 (36.9) 303 (51.9) \<0.001   Anxiety/depression, *n* (%) 33 (15.0) 93 (16.4) 134 (22.9) \<0.001 EQ-5D index 0.90 ± 0.14 0.87 ± 0.16 0.81 ± 0.20 \<0.001 EQ-VAS score 69.2 ± 22.7 69.4 ± 20.9 63.9 ± 22.5 0.002 ijerph-17-01164-t003_Table 3 ###### Descriptive statistics of measured parameters according to eCRF categories (Mean ± SD). Variables eCRF Categories *p* for Linear Trend ---------------------------------- ----------------- ---------------------- --------------- --------- eCRF (METs) 5.3 ± 1.8 7.0 ± 1.9 9.3 ± 1.8 \<0.001 *Demographics* Age (years) 74.3 ± 5.5 67.9 ± 5.3 66.5 ± 5.2 \<0.001 Marital status, *n* (%) 0.198   Yes 339 (99.1) 678 (98.8) 343 (100.0)   No 3 (0.9) 8 (1.2) 0 (0.0) Living condition, *n* (%) 0.003   Live someone 274 (80.1) 600 (87.5) 302 (88.0)   Alone 68 (19.9) 86 (12.5) 41 (12.0) *Socioeconomic Status* Income (10,000 won/month) 291.4 ± 220.3 191.7 ± 253.6 238.0 ± 418.4 0.547 Education, *n* (%) 0.002   Lower than elementary school 155 (45.3) 288 (42.0) 115 (33.5)   Middle/high school 144 (42.1) 284 (41.4) 169 (49.3)   Over than college 43 (12.6) 114 (16.6) 59 (17.2) *Health Behavior Factors* Smoking, *n* (%) 0.676   Never 186 (54.4) 391 (57.0) 192 (56.0)   Past/current 156 (45.6) 295 (43.0) 151 (44.0) Weekly alcohol intake, *n* (%) 0.034   \<2 301 (88.0) 555 (80.9) 1.9)   ≥2 41 (12.0) 131 (19.1) 62 (18.1) Regular exercise, *n* (%) \<0.001   Yes 83 (24.3) 216 (31.5) 131 (38.2)   No 259 (75.7) 470 (68.5) 212 (61.8) *Number of Comorbidity, n (%)* 0.008 0 36 (10.5) 115 (16.8) 69 (20.1) 1 156 (45.6) 270 (39.4) 1.1) ≥2 150 (43.9) 301 (43.8) 133 (38.8) *Health-Related Quality of Life* EQ-5D problems   Mobility, *n* (%) 206 (60.2) 288 (42.0) (37.0) 127 \<0.001   Self-care, *n* (%) 85 (24.9) 85 (12.4) 35 (10.2) \<0.001   Usual activities, *n* (%) 141 (41.2) 178 (25.9) 90 (26.2) \<0.001   Pain/discomfort, *n* (%) 170 (49.7) 271 (39.5) 144 (42.0) 0.041   Anxiety/depression, *n* (%) 64 (18.7) 140 (20.4) 56 (16.3) 0.425 EQ-5D index 0.80 ± 0.22 0.87 ± 0.16 0.87 ± 0.16 \<0.001 EQ-VAS score 62.9 ± 24.9 67.3 ± 20.9 70.6 ± 20.7 \<0.001 ijerph-17-01164-t004_Table 4 ###### The association between comorbidity and health related quality of life, mediated by eCRF, in diabetes. Path Model 1 Model 2 Model 3 ------------------------------------------------ --------------------- ------------------ --------------------- ------------------ --------------------- ------------------ Comorbidity → eCRF, a −0.528 (0.063) \*\* −0.652 to −0.404 −0.388 (0.055) \*\* −0.495 to −0.281 −0.244 (0.046) \*\* −0.335 to −0.154 eCRF → Quality of life, b 1.434 (0.258) \*\* 0.928 to 1.940 0.996 (0.300) \*\* 0.407 to 1.586 1.153 (0.358) \* 0.452 to 1.855 Total effect, c −2.670 (0.611) \*\* −3.868 to −1.472 −2.327 (0.609) \*\* −3.522 to −1.133 −2.321 (0.613) \*\* −3.522 to −1.119 Direct effect, c′ −1.913 (0.619) \* −3.128 to −0.698 −1.941 (0.618) \* −3.153 to −0.729 −2.039 (0.617) \* −3.249 to −0.829 Indirect effect, ab −0.757 (0.168) −1.104 to −0.453 −0.387 (0.135) −0.672 to −0.145 −0.282 (0.107) −0.517 to −0.095 Ratio of indirect to total effect mediated (%) 28.4 16.1 12.1 Sobel test −4.632 \*\* −3.004 \*\* −2.753 \*\* Model 1 was non-adjusted. Model 2 was adjusted for age, marital status, living condition, and education. Model 3 was adjusted for Model 2 + smoking, weekly alcohol consumption, and regular exercise. Number of bootstrap samples for bias-corrected bootstrap confidence intervals: 5000. eCRF---non-exercise-based estimation of cardiorespiratory fitness; CI---confidence interval; SE---standard error. \* *p* \< 0.005, \*\* *p* \< 0.001.

Introduction {#Sec1} ============ Major depressive disorder (MDD) is a disease affecting around 16% of people over the age of 13 in the US^[@CR1]^. This prevalent mood disorder greatly impacts a patient's quality of life and is the leading cause of disability around the world^[@CR2]^. Experiencing stressful events throughout one's life, particularly childhood trauma, further increases the likelihood of being diagnosed with MDD^[@CR3]^. These stresses often exacerbate chronic medical issues, such as heart disease, diabetes and other inflammatory disorders^[@CR4]--[@CR6]^. To quantify biological stress, studies have used measured levels of cortisol, a stress hormone, as well as Interleukin (IL-6) and C-reactive protein (CRP), two markers of inflammation. Increased levels of all three of these measures have been linked to both early life stress and MDD^[@CR7]--[@CR9]^. Psychological stress and MDD have also been reported to affect brain anatomy, with numerous brain regions shown to have decreased volumes in patients^[@CR10]^. Indeed, the hippocampus, a subcortical structure critically involved in memory and cognition, has been shown to be sensitive to stress and smaller hippocampal volumes are characteristic of various mood disorders, including MDD^[@CR11]--[@CR13]^. Although global hippocampal volume may differ between populations, the hippocampus is in fact made up of anatomically and functionally distinct subfields that may be differentially sensitive to stressors and play varying roles in mood disorder pathophysiology^[@CR14]^. Therefore, investigating whether specific subfields are linked to MDD may serve to provide greater clarity as to the mechanisms behind the disease^[@CR15]^. There have been several studies published on hippocampal subfields in MDD, some of which have reported volume decreases in the cornu ammonis (CA) and dentate gyrus, while others have found no significant differences between MDD subjects and healthy controls (HCs)^[@CR16]--[@CR18]^. More recent studies have conducted their analyses using the automated segmentation algorithm released in Freesufer 6.0. This algorithm was developed using an atlas based on ex vivo brain tissue, which was scanned under ultra-high field strength. It has been shown to result in more accurate segmentation of the hippocampus, and it is able to label the molecular layer (ML) and granule cell layer (GCL), both which were not available in previous segmentation algorithms^[@CR19]^. Yet, there are relatively few studies published that use the novel algorithm, and conflicting results as to whether differences between the subfields of MDDs and HCs are significant^[@CR20]--[@CR23]^. Only one of these studies used high-field 7T scanners to acquire their hippocampal neuroimaging data. Due to its complex structure and distinct subfields, the hippocampus can particularly benefit from the enhanced resolution and contrast acquired by a 7T scan. Indeed, a study comparing 7T images with ex vivo anatomy found 7T images to provide excellent anatomic detail of the hippocampus^[@CR24]^. Recent developments in ultra-high-field (UHF ≥ 7T) imaging allow for enhanced exploration of both structure and function of the brain. Due to increased signal-to-noise (SNR), spatial resolution, and contrast-to-noise (CNR) ratios, finer details and difficult to scan structures can be better visualised^[@CR25]^. UHF imaging simultaneously poses a number of challenges, such as non-uniform radiofrequency fields, enhanced susceptibility artifacts and higher radiofrequency energy deposition in the tissue creating a higher likelihood of focal RF heating and tissue damage. These issues, however, can be solved at the level of hardware (such as coils), software (such as using particular scanning parameters) and post-scanning data processing (such as bias correction)^[@CR26]^. UHF imaging may also induce transient physiological effects such as nausea, dizziness, metallic taste, magnetophosphenes and increased blood pressure in some subjects. However, these effects are temporary, pose minimal risk to the subject, and do not create major obstacles to scanning^[@CR26]^. Therefore, given the great potential for more accurate and reliable hippocampal segmentations, using 7T imaging data to study the hippocampal subfields of MDD patients may provide more definitive conclusions about the neurobiology of the hippocampus in MDD. This study aimed to explore the brain anatomical and biological stress signatures of MDD and how early exposure to psychological trauma may affect the presentation of the illness using 7T MRI scans. We hypothesised that our MDD subjects would have higher levels of biological stress measures, specifically cortisol, IL-6 and CRP. Despite the hippocampal subfield literature being mixed, we hypothesised that we would find smaller global hippocampal volumes and reductions in the dentate gyrus and cornu ammonis in the MDD group. Methods {#Sec2} ======= Subjects {#Sec3} -------- Adult volunteers between the ages of 18 and 65 were recruited for this study. Patients meeting DSM-IV diagnosis of MDD without other Axis 1 conditions, such as bipolar disorder, psychosis, or substance dependence (determined using the Standard Clinical Interview for Diagnostic and Statistical Manual for Mental Health Disorders---Fourth Edition)^[@CR27]^ were included. Patients with a clinically significant risk of suicidal behaviour, or a need for urgent drug treatment were excluded. Healthy controls without a current or past history of Axis I disorders on DSM-IV were included. Any potential subject with contraindications to magnetic resonance (MR) imaging, who was pregnant or breastfeeding, or was involved in a research project during the month preceding the study was excluded from the study sample as well. Mood ratings were measured using the Hamilton Rating Scale for Depression (HAM-D)^[@CR28]^ and the Beck Depression Inventory (BDI)^[@CR29]^, while anxiety ratings were scored using the Spielberger State Anxiety Inventory (STAI)^[@CR30]^. We also measured anhedonia with the Snaith--Hamilton Pleasure Scale (SHAPS)^[@CR31]^ and fatigue with the Chalder Fatigue Scale (CFS)^[@CR31],[@CR32]^. Collectively, these scores were used to capture illness severity of a given subject. Childhood trauma was measured using the Childhood Trauma Questionnaire (CTQ)^[@CR33]^. All subjects included in this study gave informed written consent, and this study was approved by the National Research Ethics Service Committee (NRES), South-Central Oxford B. Biological stress measures {#Sec4} -------------------------- Venous blood samples were taken at the time of scanning. Five millilitres were assayed on the same day for high sensitivity C-Reactive Protein (hsCRP), using a standard immunoturbidimetric method on an Abbott c 16 000 automatic chemistry analyzer (Abbott Diagnostics, Maidenhead, UK). Another 5 mL were collected into an EDTA tube, centrifuged for 10 min at 1250*×g* within 30 min to separate plasma, which was then stored at −30 °C until assay. On study completion, the plasma samples were transferred in a single batch, in unlinked anonymized form, for Human IL-6 Immunoassay using a Quantikine® HS (high sensitivity) ELISA kit (R&D Systems, Abingdon, UK). The assay was performed in duplicates. Salivary cortisol was assayed in duplicate using a commercial, ELISA-based method according to manufacturer's instructions (Salimetrics). Imaging {#Sec5} ------- All participants were scanned at the Functional Magnetic Resonance Imaging of the Brain (FMRIB) Centre in Oxford using a 7T Siemens MAGNETOM scanner (Siemens, Erlangen, Germany) with a Nova Medical 32 channel receive array head coil. The whole-brain 1 mm isotropic T1-MPRAGE image was obtained (MPRAGE: repetition time = 8.60 ms, echo time = 4.00 ms, flip angle = 7°, field of view = 192 mm, GRAPPA factor = 4). All MRI scans were visually inspected for anatomical abnormalities before being processed. Due to greater intensity inhomogeneity in 7T scans compared to 3T, images were bias corrected in SPM12 before processing as per the Freesurfer recommendations, with FWHM set at 18 mm, sampling distance at 2 mm, and bias regularisation at 0.001 (<https://surfer.nmr.mgh.harvard.edu/fswiki/>). All other parameters remained at default. Freesurfer 6.0 was then used for motion correction, intensity normalisation, automated topology corrections and fully automatic segmentation of cortical and subcortical regions^[@CR34]--[@CR36]^. Segmentation of the hippocampal subfields was conducted using the automated segmentation method developed by Iglesias et al, which, as shown in Fig. [1](#Fig1){ref-type="fig"}, results in the parcellation of 12 hippocampal subfields^[@CR19]^. All segmentations were visually assessed by taking the segmented image and superimposing it on the corresponding structural T1-weighted image. Only one subject was excluded from analysis due to poor subfield segmentation.Fig. 1A sample segmentation of the left hippocampal subfields of a healthy subject.CA cornu ammonis. In order to optimise the power of our study, we selected eight hippocampal subfields to include in our analysis based on prior findings in the literature^[@CR20],[@CR37],[@CR38]^. The following subfields were included in our analysis: CA1, CA2 and CA3 (noted as CA3 due to the indistinguishable MR contrast between CA2 and CA3), CA4, Granule cell layer, molecular layer, presubiculum, subiculum and the hippocampal tail. Statistical analysis {#Sec6} -------------------- All statistics were performed in RStudio 3.5.0. Chi-square and independent, two-sample *t*-tests, and Mann--Whitney tests were used to test for demographic differences between groups. Due to the visibly skewed nature of the mood scores and biological stress measure data, shapiro tests of normality were used to evaluate whether parametric testing would be appropriate to evaluate group differences. Mann--Whitney tests were used in lieu of independent two-sample *t*-tests if the measures were not normally distributed. As an exploratory investigation into biological stress, for each of the biological stress measures, we used the average level across all subjects to divide MDD subjects into 'low' and 'high' level groups and tested whether the groups differed in subfield volume, CTQ scores, illness duration, and illness severity using Mann--Whitney tests. General linear models were conducted to assess the effect of diagnostic group on hippocampal subfield volumes. We also tested whether MDD subjects in the 'high' biological stress measure groups had different subfield volumes than HCs. In order to fully explore gender effects, we also divided the data into male and female datasets and repeated the general linear model analysis to investigate hippocampal subfield differences between MDDs and HCs. In addition, partial correlations were used to examine the relationship between subfield volumes, biological stress markers, mood scores, and illness duration for MDD subjects. FDR corrections were used to adjust for multiple comparisons for all the general linear models and partial correlations. General linear models and partial correlations also controlled for age, gender and total intracranial volume. Significance threshold was set at *p* \< 0.05. Results {#Sec7} ======= Biological stress measures {#Sec8} -------------------------- Demographic, mood, and biological stress data are summarised in Table [1](#Tab1){ref-type="table"}. There were significant group differences in IL-6 levels, while CRP differences trended towards significance (*W* = 770.5, *p* = 0.02; W = 829.5, *p* = 0.08). There was a positive correlation between IL-6 and CRP levels. Cortisol, however, did not differ significantly between groups and did not have a significant relationship with the other stress measures collected. In addition, there were no significant relationships between any biological stress measures and CTQ or illness duration. We used the average level of each of the biological stress markers to subdivide the MDD subjects into 'low' and 'high' groups; these averages were: cortisol, 0.209 μg/dL; IL-6, 1.27 pg/mL; CRP, 0.88 mg/L. Mann--Whitney tests analysing hippocampal subfield differences between "low" and 'high' groups were not significant for any of the biological stress measures. We also tested for differences in illness duration and mood rating scale scores. There was a trend that indicated subjects with higher CRP scores having higher SHAS scores (low: *m* = 32.69 ± 5.98; high: *m* = 35.94 ± 7.84; *W* = 243.5, *p* = 0.066). No other differences were significant or approached significance.Table 1Demographic information of study participants.HC (46)MDD (71)*tWX*^2^*p*Age (years)31.5 ± 10.531.6 ± 10.20.070.94Gender\<0.0011.00 Male45.7% (21)45.1% (32) Female54.3% (25)54.9% (39)Cortisol (μg/dL)0.2 ± 0.20.2 ± 0.11118.50.89CRP (mg/L)0.5 ± 0.51.1 ± 1.6829.50.08IL-6 (pg/mL)1.2 ± 1.31.4 ± 1.1770.50.02CTQ Emotional abuse6.6 ± 2.412.6 ± 5.7455.5\<0.001 Physical abuse5.3 ± 1.07.7 ± 3.8781.5\<0.001 Sexual abuse5.2 ± 0.76.6 ± 3.91127\<0.001 Emotional neglect8.1 ± 3.514.1 ± 5.0501.5\<0.001 Physical neglect6.0 ± 2.18.5 ± 3.8781.5\<0.001 Denial0.5 ± 1.00.2 ± 0.61662.50.07Currently on medication--22.5% (16)Illness duration (years)--10.9 ± 11.0BDI2.1 ± 4.730.8 ± 8.626.5\<0.001HAM-D0.6 ± 2.422.1 ± 5.510\<0.001STAI26.4 ± 7.152.9 ± 12.0100\<0.001CFS10.1 ± 3.223.2 ± 4.823\<0.001SHAS19.2 ± 5.533.4 ± 6.6165.5\<0.001*HC* healthy controls, *MDD* major depressive disorder, *BDI* beck depression inventory, *HAM-D* Hamilton Rating Scale for depression, *STAI* Spielberger state anxiety inventory, *CFS* Chalder Fatigue Scale, *SHAPS* Snaith--Hamilton Pleasure Scale. Hippocampal subfields {#Sec9} --------------------- There were no significant group differences in any of the hippocampal subfields examined (Fig. [2](#Fig2){ref-type="fig"}). Global hippocampal volume differences were also not significantly different. Although we found IL-6 levels to increase along with right CA3 volumes (*t* = 2.751, *p* = 0.007), this did not survive multiple comparison correction. There were also no hippocampal subfield differences between MDD subjects with high levels of our biological stress measures and MDDs with normal levels. There were no significant relationships between any subfield and any other biological stress measure, CTQ score, illness duration, or mood rating scale. We separated males and females and repeated the hippocampal subfield analysis on both datasets and found no differences between MDDs and HCs. We also compared the hippocampal subfield volumes of our MDD subjects who had high cortisol, IL-6, and CRP levels relative to the rest of the study sample to our HC group. We found the right CA 1 to be smaller in high cortisol MDD patients compared to controls \[MDD high cortisol- M = 567.6 ± 47.2 mm^3^, HC- *M* = 597.4 ± 61.9 mm^3^, *F*(1,66) = 5.0, *p* = 0.03\]. This, however, did not survive FDR correction for multiple comparisons across subfields. The hippocampal subfield volumes of MDD subjects who had high levels of IL-6 and CRP did not differ significantly than those of HCs.Fig. 2Hippocampal subfield volumes of all subjects (mm^3^).Subfield volumes MDDs and HCs were compared using general linear models that were then FDR corrected for multiple comparisons. No subfield differences were found between subject groups. HC Health Controls, MDD Major Depressive disorder, F Female, M Male, CA Cornu Ammonis, GCL Granule Cell Layer, HT Hippocampal Tail, ML Molecular Layer. Discussion {#Sec10} ========== Our study sought to investigate how stress and trauma affect the pathophysiology and symptomatology of MDD. Our MDD cohort had higher levels of trauma than our HC's in all CTQ domains except for denial. Furthermore, MDD subjects were shown to have higher levels of IL-6 and CRP, two common markers of biological stress that we also found to be correlated with one another. Our biological stress measures did not have significant relationships with childhood trauma and did not appear to be related to hippocampal subfield volumes. Although we expected to find hippocampal subfield volume differences between MDD and HC subjects and those with low versus high CTQ scores, our data did not show these trends. As aforementioned, due to the lack of consistent findings regarding hippocampal subfields in MDD, more studies are needed for a clearer characterisation of the role of the hippocampus in the pathophysiology of MDD^[@CR20]--[@CR23]^. In addition, these discrepancies in the literature may be due to the heterogeneity of inherent to mood disorders, and future work emphasising computational approaches to classifying subjects has the potential to better explore MDD and other mood disorders^[@CR39]^. Additionally, a power analysis based on the effect size of 0.35 reported in a meta-analysis on the hippocampus in MDD indicated that our study had a power statistic of 0.38^[@CR11]^. Regarding the association between CTQ scores and hippocampal volumes, studies have reported subjects with higher scores having smaller global hippocampal volume as well as deficits in certain subfields, specifically the CA3, subiculum, and the dentate gyrus^[@CR40],[@CR41]^. However, a meta-analysis examining the effects of childhood maltreatment on global hippocampal volume found an effect size of just 0.08^[@CR42]^. Such a small effect size, in turn, leaves our study with a power statistic equal to 0.07. Therefore, despite having a relatively large sample when compared to published hippocampal subfield studies, the power of our analysis remains limited due to the small effect sizes characteristic of hippocampal abnormalities. Future studies with even larger cohorts and multi-site datasets will have greater power and may provide a clearer picture on the role of hippocampal subfields in MDD. Our findings did, however, imply greater illness severity in MDD subjects with CRP levels above the calculated threshold compared to MDDs in the lower range. As this pattern trended, a more balanced sample of MDD subjects with and without high CRP levels will serve to further elucidate the role of CRP in MDD symptomology. Moreover, when examining the distribution of CRP and IL-6 levels in our sample, we noted that our subjects had lower levels than those reported in studies showing associations between these two biological stress measures and grey matter volumes, which may also factor into our null findings^[@CR43],[@CR44]^. Ultimately our study reports on the intersection between trauma, stress, and hippocampal structure. Due to the rarity of our imaging data being acquired using high-field 7T scanner and the relatively large size sample, we believe that our lack of significant findings regarding the impact that childhood trauma and MDD have on hippocampal subfields remains a useful contribution to a mixed body of literature in need of further investigation. **Publisher's note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Jonika Tannous, Beata R. Godlewska, Phil J. Cowen, Sudhakar Selvaraj This work was supported by an MRC program grant to P.J.C. (MR/K022202/1) and the Oxford Health NIHR Biomedical Research Centre. The views expressed are those of the authors and not necessarily those of the National Health Service, NIHR or the Department of Health. We thank all the participants, as well as Jon Campbell, David Parker, Michael Sanders and Caroline Young for expert radiographic assistance, and care of the participants during scanning. Anonymized individual subject data are available at the Open Source Framework: osf.io/2cwya/ S.S. has received speaking honoraria from Global Medical Education and honoraria from British Medical Journal Publishing Group; own shares at Flow Med Tech; Research support from Compass pathways. P.J.C has no conflicting interests during the last 3 years. B.R.G. has received travel expenses from Janssen UK. P.J.C. has no conflict of interest. J.C.S. has received grants/research support from BMS, Forrest, J&J, Merck, Compass pathways, Stanley Medical Research Institute, NIH and has been a speaker for Pfizer and Abbott. J.T. has no conflict of interest.

1. Introduction {#sec1-ijerph-17-01067} =============== Nasolacrimal duct obstruction (NLDO) is the most common cause of childhood epiphora \[[@B1-ijerph-17-01067]\]. Failed canalisation of the distal nasolacrimal duct, associated with a membranous obstruction at the level of Hasner's valve, is the main cause of congenital NLDO; acquired causes of NLDO include infections and traumatic obstructions \[[@B2-ijerph-17-01067]\]. Congenital NLDO is managed conservatively in the first year of life, usually resolving spontaneously, coincident with elongation and volume expansion of the nasolacrimal duct \[[@B3-ijerph-17-01067],[@B4-ijerph-17-01067],[@B5-ijerph-17-01067]\]. Epiphora which persists after the age of one year old may require surgical intervention, as the success rates of conservative treatment decline with increasing age \[[@B6-ijerph-17-01067]\]. The surgical treatment of NLDO in children is classically based on a stepwise paradigm, with probing as the primary procedure, followed by balloon catheter dilation \[[@B7-ijerph-17-01067]\]. Intubation has traditionally been reserved for congenital NLDO refractory to other measures \[[@B7-ijerph-17-01067],[@B8-ijerph-17-01067],[@B9-ijerph-17-01067]\]. Intubation involves the placement of a stent within the nasolacrimal duct to prevent re-closure of the membranous obstruction. With the advent of technologically superior instrumentation and surgical skills, lacrimal intubation is not only an increasingly popular alternative to dacryocystorhinostomy (DCR) for cases which fail conservative management and probing, but also serves as an adjunct during balloon dacryoplasty \[[@B9-ijerph-17-01067]\] and DCR \[[@B10-ijerph-17-01067],[@B11-ijerph-17-01067]\]. This systematic review aims to present the current role of intubation in the management of children with NLDO requiring surgical intervention. 2. Materials and Methods {#sec2-ijerph-17-01067} ======================== 2.1. Search Strategies {#sec2dot1-ijerph-17-01067} ---------------------- The search was conducted over a period of 5 months (November 2018 to March 2019) in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines \[[@B12-ijerph-17-01067]\] and the Cochrane Handbook \[[@B13-ijerph-17-01067]\] when appropriate. The literature search was conducted by searching for English-language articles from the electronic databases PubMed, SCOPUS, and the COCHRANE library. The following keywords were used either individually or in combination to aid in retrieving the articles: stents, children, lacrimal intubation, endoscopic DCR (EDCR), external DCR (extDCR), NLDO, dacryocystitis, congenital, acquired. Literature for inclusion in the review was restricted to the period of 1997 to 2019 in order to keep the information as relevant and up to date as possible. 2.2. Eligibility Criteria {#sec2dot2-ijerph-17-01067} ------------------------- Articles were included in the systematic review if they fulfilled the following eligibility criteria: (1) prospective comparative design (e.g., randomised and non-randomised controlled trials (RCT), cohort study), retrospective with comparative group design (e.g., case-control, cross-sectional), retrospective or prospective non-comparative design (e.g., case series, before and after study); (2) included participants were less than 18 years of age; (3) intubation was part of the management of NLDO. Articles were excluded if they were (1) case reports; (2) abstract only studies; (3) published in a language other than English. 2.3. Study Outcomes {#sec2dot3-ijerph-17-01067} ------------------- The primary outcome was defined as the success of the intervention, determined by the resolution of symptoms and the patency of the lacrimal anatomy confirmed by the fluorescein dye disappearance test, syringing or irrigation of the nasolacrimal duct. The secondary outcome was the presence of complications. 2.4. Screening and Data Extraction {#sec2dot4-ijerph-17-01067} ---------------------------------- The authors selected the studies according to the predetermined inclusion and exclusion criteria for this systematic review. This was done by reading the abstracts and/or the full articles. A standardised data extraction form was developed by the authors and used in the present study. The variables extracted from the studies included study location (country), number of patients, age, gender, intervention procedure, duration of tube removal, duration of follow up, overall successful outcome, and post-operative complications. Data extraction from the included studies was done by two authors independently. Any discrepancy between the two authors with respect to the data extracted were discussed. When disagreements remained, a third author was consulted for his/her opinion and decision. 2.5. Quality Assessment {#sec2dot5-ijerph-17-01067} ----------------------- The quality assessment of the included studies was conducted by using the Effective Public Health Practice Project (EPHPP) checklist \[[@B14-ijerph-17-01067]\]. The EPHPP has been widely used in assessing the quality of public health intervention studies of varying study designs \[[@B15-ijerph-17-01067],[@B16-ijerph-17-01067],[@B17-ijerph-17-01067]\]. The EPHPP checklist has six components of assessment of study methodology; selection bias, study design, confounders, blinding, data collection methods, withdrawal and dropouts. These components were scored as either weak, moderate, or strong. The overall quality rating for each included study was also scored as either weak, moderate, or strong. An overall quality rating of 'strong' was assigned when there were no weak ratings, 'moderate' was assigned when there was one weak rating, and 'weak' was assigned when there were two or more weak ratings on the components of EPHPP. The quality assessment was conducted by two authors. Any discrepancy of scoring was discussed to reach consensus. Some components of EPHPP were labelled as not applicable for some studies. For example, the component of withdrawals and dropouts were not applicable for studies with retrospective study design, while that of blinding was not applicable for non-comparative studies, case series, or studies with a single group. 3. Results {#sec3-ijerph-17-01067} ========== 3.1. Literature Search {#sec3dot1-ijerph-17-01067} ---------------------- A total of 144 articles were identified from the electronic databases. One hundred and twenty- eight articles remained after duplicates were removed. Seventy-nine articles were excluded after screening the titles and abstract as they did not meet the review criteria. Of the remaining 49 articles, data extraction was done by two authors independently; 14 articles that included adult samples and did not report subgroup analysis result for children samples below 16 years old were excluded. A total of 35 studies which fulfilled the selection criteria were included in the review (see [Figure 1](#ijerph-17-01067-f001){ref-type="fig"}). From the included studies, several types of study designs were used by the authors. These were randomised controlled trials (5 studies), non-randomised controlled trials (5 studies), retrospective with comparative groups (4 studies), non-comparative or single group (20 studies), and retrospective record review with descriptive study (1 study). 3.2. Description of the Studies {#sec3dot2-ijerph-17-01067} ------------------------------- A total of 2953 patients were pooled. The total number of patients for each study ranged from 4 to 635 patients. The mean age of patients ranged from 15 months to 11 years old. All studies involved the use of silicone stents. The majority of the included studies involved lacrimal intubation (85.7%, 30 studies), with or without a comparative group; followed by intubation as an adjunctive procedure to extDCR (2 studies), EDCR (1 study), both extDCR and EDCR (1 study), and balloon dacryoplasty (1 study). The duration of the stent placements varied, while the average follow-up after removal of stent ranged from 9 weeks to 40 months. [Table 1](#ijerph-17-01067-t001){ref-type="table"} summarises the studies included in this systematic review. 3.3. Outcomes {#sec3dot3-ijerph-17-01067} ------------- A meta-analysis was not performed due to the heterogeneity of all the included studies. Hence, meaningful interpretation of the study outcomes in the included studies required expert discussion and clinical judgement. The two main outcomes, percentage of success and presence of complications, are narratively described in [Table 2](#ijerph-17-01067-t002){ref-type="table"}. The overall success outcome of the studies' interventions ranged from 41.1% to 100%. Post-operative complications were reported in 23 studies, while nine studies reported no complications. Three studies did not report the complication rate. 3.4. Quality Assessment {#sec3dot4-ijerph-17-01067} ----------------------- Using the EPHPP global rating decision tool, four studies were assessed as being of moderate quality and 31 of weak quality (see [Table 3](#ijerph-17-01067-t003){ref-type="table"}). Most of the studies were considered weak due to the study design and non-control of confounding factors. However, based on individual methodology component assessment, all studies were assessed as being of strong quality in terms of selection bias. All eligible patients in the included studies were from hospital-based samples. As clinical cases of children with nasolacrimal duct obstruction requiring surgical intervention were limited in nature, probability sampling methods such as the random sampling method were considered infeasible. Therefore, the authors considered that the studies assessed had included samples that were representative of their target population and were thus of strong quality in terms of selection bias. Five studies were rated as strong quality in terms of study design because the authors used randomised controlled trials in their intervention study. Four studies were rated as strong quality in terms of controlling for confounding variables (e.g., age, gender), as confounders were either balanced at baseline, or controlled for during the analysis. Data collection methods were considered strong for all studies because the authors used a standard assessment criteria of success, such as complete resolution of epiphora and the dye disappearance test. 4. Discussion {#sec4-ijerph-17-01067} ============= The management of children with epiphora is challenging, not only because of the miniaturized and variable anatomy of the lacrimal drainage pathways, but also because of the lack of high quality evidence regarding the optimal treatment of NLDO in children. Probing, which involves pushing a metal wire through the punctum, canaliculi, lacrimal sac, and nasolacrimal duct into the nose, is the standard of care for congenital NLDO. Although probing is successful in uncomplicated obstructions of the distal nasolacrimal duct \[[@B50-ijerph-17-01067]\], which comprise the majority of cases in children \[[@B51-ijerph-17-01067]\], NLDO due to anatomical variations or scarred tissue, such as in Down syndrome, trauma or craniofacial malformations, is more difficult to manage \[[@B28-ijerph-17-01067],[@B52-ijerph-17-01067],[@B53-ijerph-17-01067],[@B54-ijerph-17-01067],[@B55-ijerph-17-01067]\]. In cases which fail primary probing, treatment options include repeat probing, lacrimal intubation or balloon dacryoplasty, before resorting to DCR as a last measure \[[@B6-ijerph-17-01067],[@B56-ijerph-17-01067]\]. The role of stenting in the management of children with NLDO is poorly defined. Unlike in adults with acquired NLDO, where trials have shown no benefit of intubation on the 12-month success rate of endonasal DCR \[[@B57-ijerph-17-01067],[@B58-ijerph-17-01067]\], the value of stenting in cases of paediatric NLDO requires further elucidation. Most of the studies reviewed involved only lacrimal intubation, which generally had high success rates. Two RCTs compared stenting with other interventions for NLDO in children. Elsawaby et al. found no statistically significant difference in success rates between stenting and probing as a primary treatment for patients with congenital NLDO aged 6 months to 36 months \[[@B21-ijerph-17-01067]\]. Unfortunately, the lack of blinding or controlling for confounders resulted in a weak overall global rating for this study. Other non-randomized studies comparing stenting to probing as a primary procedure were likewise rated weak; Eshraghi et al. found a significantly higher success rate (73.3% vs. 48.9%) in the Masterka stent group compared with the probing group in children older than 18 months \[[@B27-ijerph-17-01067]\], while Al-Faky et al. noted a success rate of 88% for stenting and 80.3% for probing \[[@B28-ijerph-17-01067]\]. In general, intubation has been found to have an advantage over probing alone in certain groups, such as those with bilateral congenital NLDO, Down syndrome, history of acute dacryocystitis, and other causes of complex NLDO \[[@B28-ijerph-17-01067],[@B59-ijerph-17-01067],[@B60-ijerph-17-01067]\]. Unfortunately, beyond a certain age, the success rate of intubation as a primary or secondary procedure after failed probing appears to decline \[[@B32-ijerph-17-01067],[@B55-ijerph-17-01067]\]. A RCT by Ceylan et al. observed that intubation was inferior to balloon dilation for the primary surgical treatment of congenital NLDO in children older than three years of age \[[@B20-ijerph-17-01067]\]. Two types of intubation were evaluated in the studies assessed; intubation using monocanalicular versus bicanalicular stents. The monocanalicular stent passes through a single canaliculus to the lacrimal sac and NLD, whereas the bicanalicular stent courses through both canaliculi into the sac and nasal cavity, after which the ends are tied in a loop inside the nasal cavity. Using a monocanalicular or bicanalicular stent had no statistically significant effect on outcome \[[@B18-ijerph-17-01067],[@B23-ijerph-17-01067]\], although some authors prefer monocanalicular intubation for its ease of insertion and tube removal as well as a lower incidence of canalicular slit \[[@B25-ijerph-17-01067],[@B26-ijerph-17-01067]\]. The duration of stenting ranged from three weeks to six months. In most studies, stents were removed after three months. It may be important to note that retention of stents for longer than 12 months has been associated with a significantly lower success rate \[[@B61-ijerph-17-01067]\]. On the contrary, early stent removal (at approximately two months) has not been shown to affect the success rate among younger children, particularly those less than two years of age \[[@B22-ijerph-17-01067],[@B37-ijerph-17-01067],[@B62-ijerph-17-01067]\]. In older children, higher reoperation rates are associated with stent removal prior to 4--6 weeks \[[@B37-ijerph-17-01067],[@B62-ijerph-17-01067]\]. The rationale for intubation is that the stent may maintain patency of the newly-created lacrimal drainage passage by preventing the formation of granulation-related obstruction \[[@B63-ijerph-17-01067]\]. The latter may be a particular problem in children due to their anatomically narrower nasolacrimal ducts \[[@B64-ijerph-17-01067]\], elevated inflammatory tendencies, and unpredictable remodeling in response to probing-induced trauma \[[@B65-ijerph-17-01067]\], all of which may contribute to a greater risk of restenosis and failure after probing. When used as adjunctive treatment in DCR, intubation has been associated with a significantly lower incidence of operative revision \[[@B41-ijerph-17-01067]\]. Although there is a paucity of histopathological evidence of the effect of stenting in children, comparison of lacrimal sac biopsies in adults with and without silicone stents has not demonstrated any significant differences in mucosal histopathology \[[@B66-ijerph-17-01067]\]. The duration of stenting in the aforementioned study was approximately three months. A study of tear inflammatory cytokines after endoscopic endonasal DCR observed that levels of interleukin (IL)-1β, IL-2, IL-6, vascular endothelial growth factor and fibroblast growth factor-2 were higher in patients post DCR than in the control group, but rapidly returned to control group levels after stent removal \[[@B67-ijerph-17-01067]\]. This may explain the lower success rates associated with prolonged stent retention, as persistently elevated cytokines may cause sustained inflammation and fibrosis \[[@B61-ijerph-17-01067]\]. Another downside of intubation is that stent-related complications are not uncommon \[[@B68-ijerph-17-01067]\]. The most common of these is early stent dislodgement or loss, occurring in up to 50% of cases \[[@B22-ijerph-17-01067],[@B23-ijerph-17-01067],[@B25-ijerph-17-01067],[@B26-ijerph-17-01067],[@B30-ijerph-17-01067],[@B38-ijerph-17-01067],[@B68-ijerph-17-01067]\]. The prognostic value of this complication depends on age and the timing of tube displacement; the risk of reoperation is higher in children older than two years with stent retention of less than two months \[[@B36-ijerph-17-01067],[@B62-ijerph-17-01067]\]. Stent displacement may also cause corneal abrasions \[[@B10-ijerph-17-01067],[@B26-ijerph-17-01067],[@B55-ijerph-17-01067]\] and ulceration \[[@B21-ijerph-17-01067],[@B38-ijerph-17-01067]\]. Minor complications include those related to the lacrimal passages, such as punctal or canalicular slitting due to cheese-wiring \[[@B25-ijerph-17-01067],[@B26-ijerph-17-01067],[@B30-ijerph-17-01067],[@B37-ijerph-17-01067],[@B43-ijerph-17-01067]\] and granuloma formation \[[@B26-ijerph-17-01067],[@B37-ijerph-17-01067],[@B45-ijerph-17-01067]\]. This systematic review observed that lacrimal intubation is most commonly performed as a primary procedure in children with congenital NLDO, with good outcome. It is preferred over probing alone in complex NLDO \[[@B28-ijerph-17-01067],[@B59-ijerph-17-01067],[@B60-ijerph-17-01067]\]. Up to approximately ten years old, age is not predictive of intubation failure \[[@B32-ijerph-17-01067],[@B61-ijerph-17-01067]\]. The optimal timing of stent removal is two months post insertion, although children more than two years old may require a longer duration of stenting. The main complication is stent dislodgement. Considering the technical ease of stent manipulation and high success rates, it seems reasonable to perform primary intubation in children undergoing initial probing. However, well-designed, adequately powered RCTs are required to define the role of intubation as a primary or adjunctive procedure in the surgical management of children with NLDO. 5. Conclusions {#sec5-ijerph-17-01067} ============== This systematic review identified only two RCTs evaluating the benefit of stenting over other surgical modalities in the management of children with NLDO. In the absence of high-quality evidence, the decision of whether to perform lacrimal intubation in children with NLDO requiring surgical intervention depends on clinical judgement and other low-level evidence, such as observational non-randomised trials. The authors would like to express their gratitude and heartfelt thanks to the librarian in Universiti Sains Malaysia for providing articles needed in the present review. Conceptualisation, B.A., E.L.M.T., and Y.C.K.; Methodology, B.A., E.L.M.T., and Y.C.K.; Validation, B.A., E.L.M.T., and Y.C.K; Formal analysis, B.A., E.L.M.T., and Y.C.K.; Resources, B.A., E.L.M.T., and Y.C.K.; Writing---Original Draft Preparation, B.A., E.L.M.T., and Y.C.K.; Writing---Review and Editing, B.A., E.L.M.T., and Y.C.K. All authors have read and agreed to the published version of the manuscript. This research received no external funding. The authors declare no conflicts of interest. {#ijerph-17-01067-f001} ijerph-17-01067-t001_Table 1 ###### Characteristics of the studies reviewed. -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Study Design First Author, Year Country Procedure Number of Patients\|Number of Eyes Mean Age in mo/yr\ Gender in % (male) Timing of Postoperative Stent Removal in d/wk/mo Mean Follow--up in wk/mo (min--max) (min--max) -------------------------------------------------------------------------------- ------------------------------------------------ ------------------- ------------------------------------------------------------- ------------------------------------ -------------------- --------------------------- -------------------------------------------------- ------------------------------------- Randomised controlled trials Andalib, 2010 \[[@B18-ijerph-17-01067]\] Iran LI 57\|70 MCI: 34.9mo NR 3mo NR (13--71mo) BCI: 38.7mo (14--84mo) Andalib, 2014 \[[@B19-ijerph-17-01067]\] Iran LI 49\|53 MCI: 26.25mo NR 3mo NR (13--49mo) PMCI: 26.85mo (16--68mo) Ceylan, 2007 \[[@B20-ijerph-17-01067]\] Turkey LI 20\|24 (BCI) 50.8mo (36--120mo) NR Average 6.2 mo NR (12mo--NR) Elsawaby, 2016 \* \[[@B21-ijerph-17-01067]\] Egypt LI 27\|30 14.85mo 50 At least 3wk 16wk (NR) (7--30mo) Kominek, 2010 \[[@B22-ijerph-17-01067]\] Czech Republic LI 83\| 95 (Group 1: 42\|48; Group 2:41\|47) NR (15--30mo) NR Group 1: 2mo NR (NR--6mo) Group 2: 5mo Non--randomised controlled trials Eshraghi, 2017a \[[@B23-ijerph-17-01067]\] Iran LI 99\|99 (MCI:52\|52; BCI:47\|47) 3.56yr 57.6 3mo NR (NR--12mo) (1.3--10yr) Fayet, 2011 ^\#^ \[[@B24-ijerph-17-01067]\] France LI 68\|68 (Group 2:6\|6; Group 3:62\|62) Group 2: NR NR Group 2: 39d Group 3: 29d Group 2: 14wk (3--30wk) (1--9yr) Group 3: NR Group 3: 16wk (3--74wk) (1--6yr) Lee, 2012 \[[@B25-ijerph-17-01067]\] South Korea LI 46\|60 (BCI:22\|30; MCI:24\|30) BCI:23.3mo 52.2 BCI: 5--22wk BCI: 16.4 wk (NR) (9--52mo) MCI: 23.1mo MCI:5--15wk MCI: 11.6 wk (NR) (8--62mo) Kominek, 2011 \[[@B26-ijerph-17-01067]\] Czech Republic LI 53\|70 (BCI:24\|35; MCI:29\|35) NR (10--36mo) 44.3 3--4mo NR (NR--6mo) Eshraghi, 2017b \[[@B27-ijerph-17-01067]\] Iran LI 45\|45 (LI only, study compared LI and probing) 28mo (NR) NR Average 9.2 wk NR (1wk--6mo) Retrospective with comparative groups Al--Faky, 2012 ^\$^ \[[@B28-ijerph-17-01067]\] Iran LI 350\|454 32.6mo 46 3mo 15.3mo (12--132mo) (3--108mo) Kaufman, 1998 ^&^ \[[@B29-ijerph-17-01067]\] United States LI 64\|73 (Prospective:39\|48 31.8mo NR 4--6mo NR (3--12wk) Retrospective:25\|25) (12--87mo) Rajabi, 2016 \[[@B30-ijerph-17-01067]\] Iran LI 338\|338 (Crawford:248\|248; Monoka:52\|52; Masteka:38\|38) NR 56.1 3mo Schedule follow up 3mo (1--4yr) Khatib, 2017 \[[@B31-ijerph-17-01067]\] United States LI 53\|72 (complex; simple) 22mo NR 2--3mo 14mo (5--65mo) (6--29mo) Retrospective/prospective with single group/non--comparative/consecutive cases Okumus, 2016 \[[@B32-ijerph-17-01067]\] Turkey LI 30\|30 10.7yr 60 Average 4.6mo 8.8mo (7--15yr) (6--16mo) Orhan, 1997 \[[@B33-ijerph-17-01067]\] Turkey LI 16\|18 25mo 43.8 4--7mo 12mo (18--48mo) (4--24mo) Eshraghi, 2014 \[[@B34-ijerph-17-01067]\] Iran LI 44\|44 3.2yr (NR) 45.5 2mo 9mo (6.5--13mo) Ali, 2013 \[[@B35-ijerph-17-01067]\] India ExtDCR 10\|11 9.4yr 30 12--16wk NR (3--6mo) (6--15yr) Engel, 2007 \[[@B10-ijerph-17-01067]\] United States LI 635\|803 18mo 45 Median of 8wk Median of 12wk (6.5--103.8mo) Dotan, 2015 \[[@B36-ijerph-17-01067]\] Israel LI 46\|54 37.6mo (NR) 52.2 4--6mo NR El--Essawy, 2013 \[[@B37-ijerph-17-01067]\] Egypt LI 192\|236 21.2mo 51 3--6mo 5mo (3--16mo) (8--48mo) Fayet, 2012 \[[@B38-ijerph-17-01067]\] France LI 88\|110 2.4yr NR 3wk 33.7wk (4--139wk) (1--8yr) Casady, 2006 \[[@B7-ijerph-17-01067]\] United States LI NR\|7 NR NR 3--3.5mo NR (4--6wk) (12--18mo) Eloy, 2009 \[[@B39-ijerph-17-01067]\] Belgium EDCR 8\|10 4.3yr 87.5 1--3mo 10.5mo (8mo--9yr) (6--15wk) Han, 2015 \[[@B40-ijerph-17-01067]\] South Korea LI 56\|77 29.8mo 53.6 2--3mo NR (6mo--12yr) Nemet, 2008 \[[@B41-ijerph-17-01067]\] Australia ExtDCR/ EDCR 82\|104 6.6yr (NR) 51.2 6mo 1.44yr (6mo--8yr) Napier, 2016 \[[@B42-ijerph-17-01067]\] United Kingdom LI 177\|246 2.1yr (0--9.8yr) 50.4 At least 12wk NR (6--12wk) Yazici, 2006 \[[@B43-ijerph-17-01067]\] Turkey LI 42\|50 37.3mo 47.6 3mo 18.1mo (9mo--7yr) (3--48mo) Pelit, 2009 \[[@B44-ijerph-17-01067]\] Turkey LI 30\|34 5yr (2--10yr) 53.3 6mo 40.32mo (12--96mo) Yalaz, 2004 \[[@B45-ijerph-17-01067]\] Turkey LI 26\|29 4.85yr (2--12yr) 46.2 6mo 8.3mo (6--25mo) Fayet, 2010a \[[@B46-ijerph-17-01067]\] France LI 14\|18 26.2mo (14--46mo) NR Average of 34d 8.7wk (3--26wk) Pe, 1998 \[[@B47-ijerph-17-01067]\] United States LI 28\|34 19.5mo (5mo--5yr 3mo) 39.3 2--6mo NR (NR) Fayet, 2010b \[[@B48-ijerph-17-01067]\] France LI 4\|6 33mo (30--37mo) NR 3wk NR (2--3mo) Huang 2009 \[[@B9-ijerph-17-01067]\] Taiwan Balloon dacryocystoplasty and LI (MCI) 25\|33 3.5yr 60 4--6mo 6mo Five year record review (descriptive study) Abdu, 2014 \[[@B49-ijerph-17-01067]\] Nigeria ExtDCR 17\|NA NR (NR--15yr) 52.9 6wk Up to 1yr -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Notes. \* refer to group B, the intubation group; ^\#^ Group 1 was excluded (aged over 16 years); ^\$^ comparison based on age groups; ^&^ comparison based on two different cohorts (prospective and retrospective groups had monocanalicular and bicanalicular silastic tube intubation respectively); MCI, monocanalicular; BCI, bicanalicular; PMCI, pushed monocanalicular; NR, not reported; d, day; wk, week; mo, month; yr, year; min, minimum; max, maximum; %, percentage; LI, lacrimal intubation; EDCR, endoscopic dacryocystorhinostomy; ExtDCR, external dacryocystorhinostomy. ijerph-17-01067-t002_Table 2 ###### Summary of reported outcomes. First Author, Year Criteria for Successful Outcome Overall Successful Outcome % Post--Operative Complication --------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------ Andalib, 2010 \[[@B18-ijerph-17-01067]\] Munk score of 0 or 1 at 3 months after tube removal MCI: 86.2 None BCI: 89 Andalib, 2014 \[[@B19-ijerph-17-01067]\] Complete resolution of epihora at 3 months after tube removal MCI: 90 Slit punctum in PMCI PMCI: 50 Ceylan, 2007 \[[@B20-ijerph-17-01067]\] Complete remission of epiphora at 12 months, maintained for 4 months 62.5 Ocular irritation, false lumen in the inferior meatus, iatrogenic punctal laceration Elsawaby, 2016 \[[@B21-ijerph-17-01067]\] Munk's score 0 or 1 after 3 months from surgery 83.3 \* Corneal ulcer, epistaxis Kominek, 2010 \[[@B22-ijerph-17-01067]\] Fluorescein dye disappearance grade 0--1, corresponding to complete resolution of previous symptoms Group 1(removal at 2 mo): 89.6 None Group 2 (removal at 6 mo): 91.5 Eshraghi, 2017 \[[@B23-ijerph-17-01067]\] Dye disappearance test grade 0--1 & complete resolution of symptom at 12 months' follow up MCI: 59.6 Loss of tubes BCI: 74.4 Fayet, 2011 ^&^ \[[@B24-ijerph-17-01067]\] Absence of epiphora, absence of mucous discharge Group 2 (age 1--9 years): 100 Group 2: none Group 3 (age 1--6 years): 88.3 Group 3: Loss of tube, keratitis Lee, 2012 \[[@B25-ijerph-17-01067]\] Complete disappearance of symptoms BCI: 93.3 Tube prolapse, punctal slitting MCI: 90 Kominek, 2011 \[[@B26-ijerph-17-01067]\] Fluorescein dye disappearance test grade 0--1 = complete resolution from symptoms BCI: 82.86 Dislodging of tube, premature removal, loss of tube, slitting of punctum and canaliculi, granuloma pyogenicum, corneal erosion MCI: 88.57 Eshraghi, 2017b \[[@B27-ijerph-17-01067]\] Complete absence of clinical signs and symptoms of congenital nasolacrimal duct obstruction at 6 months after the procedure 73.3 Epiphora with tubes in place Al--Faky, 2012 \[[@B28-ijerph-17-01067]\] Normal dye disappearance test, positive Jones primary dye test 88 NR Kaufman, 1998 \[[@B29-ijerph-17-01067]\] Negative dye disappearance test 79 Bilateral preseptal cellulitis, migration of punctal anchor into canaliculus, corneal abrasion, corneal ulcer, premature removal of tube Rajabi, 2016 \[[@B30-ijerph-17-01067]\] No sign and symptom of tearing or discharge BCI:80.2 Tube dislodging, spontaneous extrusion, corneal abrasion, punctual slitting due to cheese wiring, punctal plug migration to canaliculus MCI:41.1 Khatib, 2017 \[[@B31-ijerph-17-01067]\] Complete resolution of symptoms, negative dye disappearance test 75 Early tube loss Okumus, 2016 \[[@B32-ijerph-17-01067]\] Complete resolution of previous signs and symptoms and DDT grade 0--1 73.3 None Orhan, 1997 \[[@B33-ijerph-17-01067]\] Resolution of symptoms and previous signs 100 Epiphora with tubes in place Eshraghi, 2014 \[[@B34-ijerph-17-01067]\] Complete resolution or partial improvement 82.6 None Ali, 2013 \[[@B35-ijerph-17-01067]\] Resolution of symptoms 91 NR Engel, 2007 \[[@B10-ijerph-17-01067]\] Good clearance of fluorescein dye and/or absence of symptomatic testing 96 Conjunctival--corneal abrasion Dotan, 2015 \[[@B36-ijerph-17-01067]\] Complete resolution of all preoperative CNLDO symptoms and signs 85 Spontaneous tube loss El--Essawy, 2013 \[[@B37-ijerph-17-01067]\] Complete resolution of symptoms, no epiphora, no discharge, no increase tear lake 82.2 Cheesewiring of canaliculi, late postoperative granuloma formation Fayet, 2012 \[[@B38-ijerph-17-01067]\] Absence of symptoms after stent removal or loss 85 Keratitis, tube loss, epiphora with tubes in place Casady, 2006 \[[@B7-ijerph-17-01067]\] Complete resolution of symptoms 100 None Eloy, 2009 \[[@B39-ijerph-17-01067]\] Complete resolution of symptoms 90 Transient slight epiphora Han, 2015 \[[@B40-ijerph-17-01067]\] Disappearance of epiphora symptoms by a minimum of 2 months 89.6 Tube prolapse, tube loss Nemet, 2008 \[[@B41-ijerph-17-01067]\] Objective confirmation of free fluorescein flow to the nose 95.2 Jones tube placement Napier, 2016 \[[@B42-ijerph-17-01067]\] Complete resolution of symptoms and signs 91.7 NR Yazici, 2006 \[[@B43-ijerph-17-01067]\] Resolution of lacrimal symptoms and signs, normal tear meniscus, and in cooperative patients, normal dye disappearance test and/or patent nasolacrimal duct on irrigation at the last examination. 86 Slit punctum Pelit, 2009 \[[@B44-ijerph-17-01067]\] Complete resolution of previous lacrimal symptoms and signs 100 None Yalaz, 2004 \[[@B45-ijerph-17-01067]\] Relief from symptom and/or positive results in fluorescein dye disappearance test 93.2 (initial intubation); Granuloma 100 (reintubation) Fayet, 2010a \[[@B46-ijerph-17-01067]\] Absence of epiphora, absence of mucous discharge 88 Mildly watery eye Pe, 1998 \[[@B47-ijerph-17-01067]\] Easy, uncomplicated retrieval of the Prolene guide thread during intubation; complete resolution of previous signs and symptoms and a normal result of the dye disappearance test 97 None Fayet, 2010b \[[@B48-ijerph-17-01067]\] Residual epiphora after ablation 100 None Huang 2009 \[[@B9-ijerph-17-01067]\] Complete resolution of symptoms and normal dye disappearance test 97 None Abdu, 2014 \[[@B49-ijerph-17-01067]\] Patent nasolacrimal duct 1 year after surgery 88 Extrusion of the tube, infection Notes. NNR, not reported; MCI, monocanalicular; BCI, bicanalicular; PMCI, pushed monocanalicular; \* study consisted of probing and stent groups, the value refers to stent group; ^&^ study consisted of three groups; group 1 aged 44--77 years was excluded. ijerph-17-01067-t003_Table 3 ###### EPHPP quality assessment tool rating for individual studies. First Author, Year Selection Bias Study Design Confounders Blinding Data Collection Methods Withdrawals and Dropouts Global Rating -------------------------------------------- ---------------- -------------- ------------- ---------- ------------------------- -------------------------- --------------- Andalib, 2010 \[[@B18-ijerph-17-01067]\] S S S W S S M Andalib, 2014 \[[@B19-ijerph-17-01067]\] S S S W S M M Ceylan, 2007 \[[@B20-ijerph-17-01067]\] S S W W S S W Elsawaby, 2016 \[[@B21-ijerph-17-01067]\] S S W W S S W Kominek, 2010 \[[@B22-ijerph-17-01067]\] S S W W S S W Eshraghi, 2017 \[[@B23-ijerph-17-01067]\] S M W W S S W Fayet, 2011 \[[@B24-ijerph-17-01067]\] S M W W S S W Lee, 2012 \[[@B25-ijerph-17-01067]\] S M S W S S M Kominek, 2011 \[[@B26-ijerph-17-01067]\] S M W W S S W Eshraghi, 2017b \[[@B27-ijerph-17-01067]\] S M W W S S W Al-Faky, 2012 \[[@B28-ijerph-17-01067]\] S M W W S NA W Kaufman, 1998 \[[@B29-ijerph-17-01067]\] S M W W S S W Rajabi, 2016 \[[@B30-ijerph-17-01067]\] S M W W S NA W Khatib, 2017 \[[@B31-ijerph-17-01067]\] S M W W S NA W Okumus, 2016 \[[@B32-ijerph-17-01067]\] S W W NA S S W Orhan, 1997 \[[@B33-ijerph-17-01067]\] S W W NA S S W Eshraghi, 2014 \[[@B34-ijerph-17-01067]\] S W S NA S S M Ali, 2013 \[[@B35-ijerph-17-01067]\] S W W NA S S W Engel, 2007 \[[@B10-ijerph-17-01067]\] S W W NA S NA W Dotan, 2015 \[[@B36-ijerph-17-01067]\] S W W NA S NA W El-Essawy, 2013 \[[@B37-ijerph-17-01067]\] S W W NA S NA W Fayet, 2012 \[[@B38-ijerph-17-01067]\] S W W NA S NA W Casady, 2006 \[[@B7-ijerph-17-01067]\] S W W NA S NA W Eloy, 2009 \[[@B39-ijerph-17-01067]\] S W W NA S NA W Han, 2015 \[[@B40-ijerph-17-01067]\] S W W NA S NA W Nemet, 2008 \[[@B41-ijerph-17-01067]\] S W W NA S NA W Napier, 2016 \[[@B42-ijerph-17-01067]\] S W W NA S NA W Yazici, 2006 \[[@B43-ijerph-17-01067]\] S W W NA S NA W Pelit, 2009 \[[@B44-ijerph-17-01067]\] S W W NA S NA W Yalaz, 2004 \[[@B45-ijerph-17-01067]\] S W W NA S NA W Fayet, 2010a \[[@B46-ijerph-17-01067]\] S W W NA S NA W Pe, 1998 \[[@B47-ijerph-17-01067]\] S W W NA S NA W Fayet, 2010b \[[@B48-ijerph-17-01067]\] S W W NA S NA W Huang 2009 \[[@B9-ijerph-17-01067]\] S W W NA S NA W Abdu, 2014 \[[@B49-ijerph-17-01067]\] S W W NA S NA W Notes. EPHPP: Effective Public Health Practice Project; S: strong; M: medium; W: weak; NA: not applicable.

Introduction {#sec1-1} ============ Chicken is the cheapest source of protein available for human consumption, but it cannot tolerate a wide range of climatic variations which affects the production and reproduction. Climate change and animal production always are complementary to each other and its effect on livestock and poultry production is witnessed all over the world \[[@ref1]\]. India is more vulnerable due to demographic pressure on natural resources and poor coping up mechanisms. Models are there, which can predict effects of rising temperature, increased climatic variability, and extreme weather events on livestock and poultry production in coming decades. Each individual phenotype is the result of an interaction between the specific genotype and a particular environment. Therefore, genotype × environment interaction is used to describe the situation where, different genotypes (breeds, lines, strains) respond differently to different environments \[[@ref2]\]. Temperature, rainfall, solar radiation, atmospheric pressure, etc., were related with fertility. The most potent environmental measures that affect fertility might vary depending on geographical locations. The levels of performance of poultry, does not depend only on inherited capacity but, also to a great extent upon the environment \[[@ref3]\]. Poultry production and reproduction are affected by various factors such as, feeding, management, disease control, stock density, housing, climate, sire effect, hatch effect, etc. Research has generated information and techniques to deal with most of these factors except the climate change in order to maximize production. All poultry farmers agree that poultry keeping is an excellent tool in poverty alleviation due to the quick turnover and low investment. Improvement in poultry production implies to create an opportunity for development of the poor section of the society \[[@ref4]\]. However, climate change is emerging as a great challenge for poultry industries to sustain the level of production. The worst effects of such climate change are experienced in tropical countries where, a common practice is to house the birds in the open side sheds. Climate variation is one of the major threats to poultry production. Birds of different breeds/strains and of different age, sex, stage of production, and reproduction respond differently to climatic variations \[[@ref5]\]. It is highly desirable that data on such effects in different flocks should be generated and analyzed to develop strategies to deal with adverse effects of climate change. Available literature on this aspect is scanty to develop a definite strategy and hence, the present study was an attempt to examine and find out important climatic variables affecting production and reproduction in a broiler breeder flock. Materials and Methods {#sec1-2} ===================== Ethical approval {#sec2-1} ---------------- Ethical approval was not necessary. However, the present research was carried out as per the standard procedures and guidelines of the institution. Study period {#sec2-2} ------------ The study was carried out for a period of 1 year starting from January 2012 to December 2012. Study area, population size and collection of data {#sec2-3} -------------------------------------------------- Flock of colored synthetic female line broiler breeder birds of All India Co-ordinated Research Project on poultry improvement (AICRP on poultry) were maintained at College of Veterinary Science and Animal Husbandry, Orissa University of Agriculture and Technology, Bhubaneswar, Odisha, India. The birds were product of selection for 16 generations for their 5^th^ week, 11^th^ week, and 20^th^ week age body weights. The selection criteria for the progeny selected were individual selection based on 5^th^ week, 11^th^ week, and 20^th^ week age body weights. 630 female progenies and 194 male progenies from 69 sires and 552 dams produced in four consecutive hatches at an interval of 10 days were utilized for this study. The selection process was conducted for 69 sires and 552 dams, and all of them contributed to each hatch in order to produce the next generation progeny, although unequally, hence respective hatch correction was applied for all the individuals. Data on environmental variables were also collected from the observatory of Orissa University of Agriculture and Technology located at Bhubaneswar, Odisha on a daily basis. Feeding, management, vaccination and disease control of the flock under trial were followed adopting a standard procedure. Traits considered for this study were: body weight of male at 5^th^-week of age (g), body weight of male at 20^th^-week of age (g), body weight of female at 5^th^-week of age (g), body weight of female at 20^th^-week of age, female age at sexual maturity (ASM) (days), egg number up to 40 weeks of age, and egg shape index. Egg shape index was equal to 100\* (width of egg/length of the egg). The width and length of eggs were calculated through an electronic slide caliper. Environmental variables such as climatic variables considered for this study were: Rainfall (mm), maximum temperature (°C), evaporation (mm), minimum temperature (°C), morning RH (%), afternoon RH (%), bright sunshine (hours), wind velocity (km/h), and rainy days (number). The daily environmental data recorded were totaled from the date of hatch to the date of measurement for production and reproduction traits of individual progeny. Statistical analysis {#sec2-4} -------------------- Since the chicks were hatched out in several hatches, all the production and reproduction data were corrected for hatch effect through BLUE estimates \[[@ref6]\]. The mathematical model for each trait was Y~ijk~=h~i~+s~j~+e~ijk~ or in matrix notations, Y=Xh+Zs+e where, Y = observation data of production reproduction traits (n × 1 vector), n=number of observations, h = p × 1 vector of hatch effect, p = number of levels of hatch effects, s = q × 1 vector of sire effects, q = number of levels of sire effects, e = n × 1 vector of random residual effects, X = design matrix of order n × p, Z = design matrix of order n × q. Under these assumptions, best linear unbiased estimator (BLUE) equations were reduced to the form,  Where, G^−1^ = (1-h^2^)/h^2^; h^2^ is the prior known value of heritability for the trait from literature. Now,  *ĥ*= vector of all hatch effects; *ŝ*= vector of all sire effects. Estimates for the effects of four hatches were obtained through MS-excel in computer and were subtracted from the production/reproduction data of the progeny belonging to the respective hatch. Then, analysis of variance was conducted using the hatch corrected data in MS-excel following a mathematical model for each trait as described below \[[@ref7]\]. Y~ijk~ = µ + s~i~ + d~ij~ + e~ijk;~ dams were nested in sire and i = 1,2.,s sires; j= 1,2.,d dams per sire; k = 1,2.,n progeny per dam per sire. Y~ijk~ = observation of k^th^ progeny of the j^th^ dam mated to i^th^ sire. µ = overall mean; s~i~ = effect of i^th^ sire; d~ij~ = effect of j^th^ dam mated to i^th^ sire; e~ijk~ = the random error associated with Y~ijk~ which is assumed to be normally and independently distributed with mean zero and variance σ^2^. F-test was carried out to know the significance of sire component. Those traits that were significant (p\<0.05) for sire component in the analysis of variance, were subjected to sire effect correction. The individual sire effects calculated through BLUE estimates were taken and were subtracted from the hatch corrected progeny data belonging to respective sires. The effect of climatic variables on each trait was then calculated through multiple linear regressions according using MS-excel data analysis-regression options in the computer \[[@ref8]\]. Results and Discussion {#sec1-3} ====================== Weekly mean ranges of several climatic variables during different experimental periods are considered because short-term climatic variations affect meat production and quality in livestock and poultry ([Table-1](#T1){ref-type="table"}) \[[@ref9]\]. Mean square values through analysis of variance with nested classification with sire, dam, and error components as sources of variation having the sire component for 20 week age body weight of females and egg shape index were significant (p\<0.01), however, all other traits were non-significant for the same ([Table-2](#T2){ref-type="table"}). ###### Weekly values (ranges) of climatic variables at different ages of the poultry flock. Age groups ------------------------------- ------------ ----------- ----------- ----------- ----------- Weekly total rainfall (mm) 0-24.6 0-91.7 0-113.6 0-215.3 0-215.3 Mean maximum temperature (°C) 30.2-37.3 30.2-38.4 30.2-38.4 30.2-38.4 30.1-34.1 Mean evaporation (mm) 2.6-6.4 2.6-8.4 2-8.4 1.9-8.4 1.9-5 Mean minimum temperature (°C) 15.7-24.7 15.7-27.2 15.7-27.2 15.7-27.2 16-25.8 Mean morning RH (%) 87-96 84-96 84-96 84-97 84-97 Mean afternoon RH (%) 30-59 30-81 30-89 30-94 35-94 Total bright sunshine (hours) 263.2 929.3 1044.4 1607.93 563.5 Mean wind velocity (km/h) 2.2-13.2 2.2-13.2 2.2-13.2 2.1-13.2 2.1-7 Total rainy days 2 35 60 105 40 ###### Mean square values of production and reproduction traits of broiler breeder poultry. Traits ------- ----------------- --------------- -------------- ---------------- -------------- --------------------------------------------- ------------------------------------------------ Sire 37315.10 (64) 12561.28 (69) 177.43 (68) 191387.20 (66) 479.31 (69) 29.56[\*](#t2f1){ref-type="table-fn"} (67) 124886[\*](#t2f1){ref-type="table-fn"} (69) Dam 3252076.76 (83) 9272.66 (264) 255.36 (203) 216402.20 (87) 382.54 (230) 24.56[\*](#t2f1){ref-type="table-fn"} (148) 110903.90[\*](#t2f1){ref-type="table-fn"}(233) Error 16788.33 (10) 23427.41 (74) 203.40 (155) 184547.40 (40) 352.95 (212) 17.72 (92) 79453.38 (209) Values in parenthesis indicate respective degrees of freedom. Significance at P\<0.05, Traits: 1=5th week body weight of male (g); 2=5^th^ week body weight of female (g); 3=Egg number up to 40 week of age, 4=20^th^ week body weight (males); 5=Age at sexual maturity of females (days); 6=Egg shape index; 7=20^th^ week body weight (females) Regression of different traits on environmental factors with coefficients (b values) along with R^2^ values and F ratio for the overall regression are presented in [Table-3](#T3){ref-type="table"}. The R^2^ value represents percent variation in the trait which could be explained by the linear regression of nine independent environmental variables. It is the multiple coefficient of determination for the collective effect of all of the independent variables (total rainfall, maximum temperature, evaporation, minimum temperature, morning RH, afternoon RH, bright sunshine hours, wind velocity, and rainy days). F ratio is the F value found in the Analysis of Variance for testing overall significance of the regression. ###### Multiple linear regressions on production and reproduction traits of environmental variables in broiler breeder poultry. Coefficients 5^th^ week body weight (males) 5^th^ week body weight (females) 20^th^ week body weight (males) 20^th^ week body weight (females) Egg number up to 40 week age Egg shape index Age at sexual maturity (females) ----------------- ----------------------------------------- ---------------------------------------- ---------------------------------------- --------------------------------------- ---------------------------------------- ----------------------------------------- ----------------------------------------- a 545.06 −122.45 −69.63 173.99 −24.30 −356.05 −471 b~1~ 17.19 6.31 4.36[\*](#t3f1){ref-type="table-fn"} −0.36[\*](#t3f1){ref-type="table-fn"} −0.01[\*](#t3f1){ref-type="table-fn"} −0.11 −0.05 b~2~ 0.15[\*](#t3f1){ref-type="table-fn"} 0.11 −0.22 0.10 0.06[\*](#t3f1){ref-type="table-fn"} −0.01[\*](#t3f1){ref-type="table-fn"} −0.17 b~3~ 3.84 0.59[\*](#t3f1){ref-type="table-fn"} −0.75 −0.29[\*](#t3f1){ref-type="table-fn"} −0.67[\*](#t3f1){ref-type="table-fn"} 0.02\* 2.07\* b~4~ −0.69[\*](#t3f1){ref-type="table-fn"} −0.94 0.05 0.17 -0.02 0.02 −0.26 b~5~ −0.22 0.15[\*](#t3f1){ref-type="table-fn"} −0.15[\*](#t3f1){ref-type="table-fn"} −0.15[\*](#t3f1){ref-type="table-fn"} −0.01[\*](#t3f1){ref-type="table-fn"} −0.003 0.004 b~6~ −1.14 −0.51 0.15[\*](#t3f1){ref-type="table-fn"} −0.07[\*](#t3f1){ref-type="table-fn"} 0.04[\*](#t3f1){ref-type="table-fn"} 0.025[\*](#t3f1){ref-type="table-fn"} 0.10[\*](#t3f1){ref-type="table-fn"} b~7~ 2.33[\*](#t3f1){ref-type="table-fn"} 0.29 −0.31[\*](#t3f1){ref-type="table-fn"} 0.19[\*](#t3f1){ref-type="table-fn"} −0.17[\*](#t3f1){ref-type="table-fn"} 0.132 −0.32 b~8~ 3.63 2.40 0.95 1.28[\*](#t3f1){ref-type="table-fn"} 0.11[\*](#t3f1){ref-type="table-fn"} −0.141 −0.18 b~9~ −88.66 −65.21 −22.69[\*](#t3f1){ref-type="table-fn"} 1.09[\*](#t3f1){ref-type="table-fn"} −0.33[\*](#t3f1){ref-type="table-fn"} 1.69 0.80 R^2^ 0.50 0.12 0.29 0.003 0.08 0.30 0.90 SE (regression) 129.04 102.89 386.42 303.06 14.46 4.04 7.58 F (regression) 19.69[\*\*](#t3f2){ref-type="table-fn"} 5.34[\*\*](#t3f2){ref-type="table-fn"} 8.45[\*\*](#t3f2){ref-type="table-fn"} 0.154 4.32[\*\*](#t3f2){ref-type="table-fn"} 12.55[\*\*](#t3f2){ref-type="table-fn"} 44.43[\*\*](#t3f2){ref-type="table-fn"} Significant at *P*\<0.05; Significant at *P*\<0.01, Subscripts of coefficients: 1. Total rainfall (mm); 2. Maximum temperature (°C); 3. Evaporation (mm); 4. Minimum temperature (°C); 5. Morning RH (%); 6. Afternoon RH (%); 7. Bright sunshine hours; 8. Wind velocity (km/h); 9. Rainy days. F (regression) = F-value calculated through ANOVA for the regression, SE=Standard error Unpredictable day-length pattern, increase in temperature, unexpected rainfall, high RH, excessive wind velocity, increase in sunshine hours is reported to have a detrimental effect on poultry production. It leads to reduced feed intake, reduced egg production, less body weight gain, small egg size, decreased egg weight, fragile egg shell quality, yolkless egg, reduced feed conversion efficiency \[[@ref10]\]. Climate and 5^th^-week body weight {#sec2-5} ---------------------------------- The result of the regression analysis for the effect of the climatic elements on the 5^th^-week body weight shows that R^2^ is 0.56 for males and 0.12 for females. The implication is that about 56% of the variance of the body weight in males and 12% of the variance in females has been explained by the climatic elements. The F-ratio values are 19.69 and 5.34 for males and females, respectively, and were highly significant at 1% level. This shows that there is a strong relationship between climatic variable selected and the 5^th^-week body weight in both the sexes. However, the difference in variation level between males and females may be due to sex linkage, resulting in higher rate of growth in males than those of the females. Testosterone promotes protein anabolism resulting in increased body size in male as compared to the female. The skeleton also responds to testosterone, with bones becoming larger and thicker \[[@ref11]\]. Of the nine climatic elements studied, only maximum temperature, minimum temperature, and bright sunshine hours had significant (p\<0.05) influence on 5^th^-week body weight of males. The maximum temperature and bright sunshine hours had a positive effect and minimum temperature had a negative effect on this trait. The chicks were hatched out in cold months of January-February and the atmospheric temperature was low during this time. Any temperature below the ideal brooding temperature (32°C) is likely to affect growth and performance of the birds \[[@ref12]\]. It is established that the 5^th^-week body weight is considered for initial selection of progenies. As during the period, the minimum environmental temperature ranged from 15.7°C to 24.7°C ([Table-1](#T1){ref-type="table"}), this could be the reason for the negative relationship between minimum temperature and 5^th^-week body weight. The maximum temperature is positively correlated with bright sunshine hours as with an increase in bright sunshine hours, the maximum temperature is bound to increase. The positive relationship of 5^th^-week body weight with a maximum temperature and bright sunshine hours can be explained on the basis that the chicks needed more heat during this period \[[@ref13]\]. The lethal temperature for birds was about 47°C at which birds could not dissipate body temperature to atmosphere. Consequently, panting occurred and birds died. In hot weather, birds minimized heat production by less eating, mating, and more sitting. When air temperature was low, birds increased body heat production by increasing feed consumption and activity. Birds hatched in hot months (April-May) achieved lesser 5^th^ week body weight in the present study which is in close agreement with the earlier finding reporting that the day length influenced the body weight positively at all ages in broiler breeders \[[@ref14]\] In contrast to the above, the morning RH and evaporation had significant (p\<0.05) and positive effect on 5^th^-week body weight of females. The morning RH was 87-96% and evaporation was 2.6-6.4 mm during this period ([Table-3](#T3){ref-type="table"}) which could be due to the fact that, cool climate in February- March was negated with high RH resulting in increased body weight of females. Climate and 20^th^-week body weight {#sec2-6} ----------------------------------- The result of the regression analysis on the effect of the climatic elements on the 20^th^-week body weight shows that R^2^ is 0.29 for males and 0.003 for females. The implication is that about 29% of the variance of the body weight in males and 0.3% of the variance in females has been explained by the climatic elements. The F-ratio values are 8.45 and 0.154, respectively, for males and females. However, the F-ratio for males were highly significant (p\<0.01) and in females it was non-significant. This shows that there is a strong relationship between climatic variable selected and the 20^th^-week body weight in males only and not in females. Furthermore, the difference in variation level between males and females may be due to sex linkage, resulting in higher rate of growth in males than the females \[[@ref13]\] Of the nine climatic elements studied, only total rainfall and afternoon RH had significant (p\<0.05) and positive effect whereas morning RH, bright sunshine hours, and number of rainy days had significant (p\<0.05) but negative effect on 20^th^ week body weight of males ([Table-1](#T1){ref-type="table"}). Up to the period of 20 weeks, which included the summer months (March-June), there was more sunshine hours leading to high environmental temperature (30.2-38.4°C). As this temperature is higher than the comfort zone of the birds, the feed consumption might have been reduced leading to lower rate of growth \[[@ref10]\]. Feed intake decreased by 1.5 g a day for every degree centigrade rise in temperature above 30°C and the negative effect of the sun shine hours on the 20^th^-week body weight can be explained \[[@ref13]\]. The morning RH is always higher than the afternoon RH ([Table-1](#T1){ref-type="table"}). The more the number of rainy days, the more is the RH. Humidity more than 75% affect breathing, feed intake, and its utilization. The RH recorded in the present study ranged from 84-96 in the morning which is above 75% \[[@ref15]\]. The high morning RH might be attributed to affect the feed consumption and utilization resulting in a negative effect on growth and 20^th^-week body weight. The positive relationship of 20^th^-week body weight in males with the total rainfall and afternoon RH can be explained on the basis that with more rainfall particularly in summer months, the environmental temperature goes down. Further, as the afternoon RH is always lower than or around the threshold value of 75%, it does not adversely affect the growth. Chickens regardless of age could not withstand high temperature and high humidity. When the surrounding air was moist, it could not absorb much moisture from respiratory tract and birds pant rapidly \[[@ref13]\]. Similarly, when high ambient temperature and humidity prevailed, birds might not be able to exchange enough air by panting to remove heat from the body. Ordinarily, sweating is not the method of heat removal in birds as seen in humans. That might be the probable reason that morning RH, bright sunshine hours, and rainy days had a negative impact on 20-week body weight of males. Opposite to this, afternoon RH and rainfall had a positive impact. Rainfall and wind in summer brings relief to heat stressed birds. Birds reared in hot and humid environment with detrimental heat stress were lighter at 9 weeks of age and birds grew faster in cooler environment of tropical climate and also resulted in reduced growth and egg production which supports the present findings \[[@ref16],[@ref17]\]. Climate and ASM {#sec2-7} --------------- The result of the regression analysis on the effect of the climatic elements on the ASM shows that R^2^ is 0.90 ([Table-1](#T1){ref-type="table"}). It implied that about 90% of the variance of the ASM in females has been explained by the climatic elements. The F-Ratio value was 44.43 and was highly significant at 1% level. This shows that there strong relationship exists between climatic variable selected and ASM. Among all the environmental elements, only evaporation and afternoon RH were regressed positively (p\<0.05) with ASM ([Figure-1](#F1){ref-type="fig"}). During 15-20^th^ week of age, the afternoon RH was recorded to be the highest ([Table-1](#T1){ref-type="table"}). The RH coupled with high environmental temperature might have created an uncomfortable atmosphere delaying sexual maturity. Similar findings have been reported by several workers \[[@ref14],[@ref18]\]. A statistically significant sire family × temperature interaction for ASM confirmed the present finding \[[@ref19]\]. {#F1} Climate and 40 weeks egg production {#sec2-8} ----------------------------------- The result of the regression analysis on the effect of the climatic elements on the 40 weeks egg production shows that R^2^ is 0.08 ([Table-3](#T3){ref-type="table"}). The implication is that about 8% of the variance of the egg production up to 40 weeks has been explained by the climatic elements. The F-ratio value was 4.32 ([Table-1](#T1){ref-type="table"}) and was highly significant (p\<0.01). This showed that a strong relationship exists between climatic variable selected and the 40 weeks egg production. All environmental coefficients were significant (p\<0.05) for egg number up to 40 weeks of age except the evaporation. Egg number was negatively regressed with total rainfall, maximum temperature, minimum temperature, morning RH, and rainy days whereas it was positively regressed with afternoon RH, bright sunshine hours, and wind velocity ([Table-3](#T3){ref-type="table"}). The egg production started approximately at 25 week of age i.e., in July (rainy season) and continued till October-November (40 week of age), when the climate was relatively cool from scorching heat ([Table-1](#T1){ref-type="table"}). Hence, wind velocity and afternoon RH were conducive for comfort living of birds leading to more egg production. Moreover, longer bright sunshine hours having photo-stimulation effect on laying hens has resulted in higher egg production \[[@ref16]\]. On the contrary, rainfall, morning RH, temperature, and rainy days were more of uncomforting to the breeder hens during that period. Higher rainfall and RH have been reported to increase the disease incidences in the flock leading to low production \[[@ref20]\]. Temperature above 80°F was reported to depress egg production, egg size, and shell quality, which is also in agreement with the present findings \[[@ref21]\]. Climate and egg shape index {#sec2-9} --------------------------- The result of the regression analysis on the effect of the climatic elements on the egg shape index shows that R^2^ is 0.30 ([Table-1](#T1){ref-type="table"}). The implication is that about 30% of the variance of the egg shape index has been explained by the climatic elements. The F-Ratio value was 12.55 ([Table-1](#T1){ref-type="table"}) and was highly significant (p\<0.01). This shows that there is a strong relationship exists between climatic variable selected and egg shape index. Egg shape index was regressed negatively (p\<0.05) with a maximum temperature indicating that higher the environmental temperature, poorer was the quality of eggshell \[[@ref20]\]. Temperature above 21°C decreased feed intake, weight gain, egg production, poor shell quality, and egg size, which supports the present finding \[[@ref13],[@ref22]\]. Conclusions {#sec1-4} =========== Based on the present finding, it may be concluded that when the chicks are hatched during the months of January-February, environmental temperature, RH, sunshine hours, and number of rainy days are the prominent climatic factors affecting growth, production, and reproduction. The environmental temperature is positively related with growth and negatively related with egg production. Similarly, afternoon RH positively affects growth, ASM, egg number, and egg shape index whereas it negatively influences growth and egg production. Growth and egg production are positively related with Sunshine hours, but negatively influenced by number of rainy days. The findings of the present study can be used to develop models to reduce the influence of potent climatic factors adversely affecting growth, production, and reproduction in breeder poultry hatched during the winter months (January-February). Although the minimum temperature and wind velocity has very less effect on response parameters, but these have been recorded in the present study for further research. Authors' Contributions {#sec1-5} ====================== GDN and NCB designed the experiment. GDN and PKM collected and analyzed the data. GDN, NCB, and KKS provided technical guidance and participated in the scientific investigation and discussion. GDN and NCB drafted the final manuscript. All authors read and approved the final manuscript. The facilities provided by AICRP on poultry for meat (ICAR, New Delhi) at College of Veterinary Science and Animal Husbandry, Orissa University of agriculture and Technology, Bhubaneswar, Odisha for this study are duly acknowledged by the authors. Competing Interests {#sec1-6} =================== The authors declare that they have no competing interests.

1. Introduction {#sec1-sensors-15-26039} =============== A variety of devices have been developed to extract energy from the environment through piezoelectric, electromagnetic and thermoelectric energy conversion \[[@B1-sensors-15-26039],[@B2-sensors-15-26039],[@B3-sensors-15-26039],[@B4-sensors-15-26039],[@B5-sensors-15-26039],[@B6-sensors-15-26039]\]. We are interested in energy harvesting to power systems deep in an oil well where ambient pressures of 200 MPa and temperatures higher than 160 °C can occur. Among various transduction mechanisms, vibration-based piezoelectric energy harvesting is attractive for such an application due to the availability of piezoelectric materials with Curie temperatures in the appropriate range as well as their high electromechanical coupling constants. A piezoelectric energy harvester allows many distinct vibration modes to be used to generate electrical power \[[@B7-sensors-15-26039]\], and can be operated with limited strain and wear to promote longevity. A variety of studies have assessed methods to convert flow energy into vibration. These include vortex shedding \[[@B8-sensors-15-26039],[@B9-sensors-15-26039],[@B10-sensors-15-26039],[@B11-sensors-15-26039],[@B12-sensors-15-26039]\], flapping motions \[[@B13-sensors-15-26039]\], hydraulic pressure \[[@B14-sensors-15-26039]\], and impeller structures \[[@B15-sensors-15-26039],[@B16-sensors-15-26039]\]. The approach we are investigating consists of a cantilever mounted in an internal flow channel where either the cantilever itself is a piezoelectric bimorph, or a non-piezoelectric cantilever drives a transducer \[[@B17-sensors-15-26039],[@B18-sensors-15-26039]\]. The internal flow passage is a spline-shaped nozzle-diffuser. [Figure 1](#sensors-15-26039-f001){ref-type="fig"} shows the concept design for an array of piezoelectric energy harvester segments that may be combined in series and/or parallel flow paths (and electrically) to generate power levels required by devices down in an oil well. In a standard well with a flow-control valve, the flow enters an annulus between the formation (reservoir rock) and an internal metal pipe (completion tubing), flows through an annular valve, enters the center tubing, and is carried to surface. The illustration makes use of the fluid flow to generate electricity in this annular section. {#sensors-15-26039-f001} Clearly, a piezoelectric energy harvester segment with high energy-conversion efficiency from fluid flow to electricity is beneficial, as it reduces both the complexity of system design electronics and the pressure loss in an internal flow environment. Bimorph cantilever-type piezoelectric harvesters are typically used in existing vibration-induced energy harvesting devices \[[@B19-sensors-15-26039],[@B20-sensors-15-26039],[@B21-sensors-15-26039]\]. Since power output is linked to internal stresses generated within the piezoelectric material, the advantage of this type of harvester is its low transverse bending stiffness, which can create large stresses with small amplitude forces relative to other piezoelectric actuators. A maximum output power can be achieved when the harvesting device vibrates at its resonant frequency; thus, a low resonant frequency harvester (\<1 kHz) is generally preferred for both robustness to fatigue (less cycles per total lifetime) and ease of coupling to a mechanical system (resonances usually \<1 kHz). However, a major drawback is its brittle nature, precipitating short lifetimes due to low fatigue limits. These problems can be mitigated with ruggedized designs, which include the addition of both passive and active mechanisms for limiting the maximum vibration stress/strain levels. A different approach that improves on the robustness of a bimorph-type piezoelectric harvester is the implementation of a flextensional as the transduction actuator. A flextensional actuator is generally comprised of a multilayer piezoelectric stack inside a metal frame. Since the stack is kept under a compressive load during operation, its design offers a high fatigue limit and a high energy density transducing structure: a lifetime test of piezoelectric stacks revealed no catastrophic failures or degradation after 100 billion cycles \[[@B22-sensors-15-26039]\]; another reported that piezoelectric stacks are capable of producing high electrical power, specifically in the order of 200--300 mW \[[@B23-sensors-15-26039]\]. Our flextensional-based flow energy harvester uses a non-piezoelectric cantilever in a converging-diverging flow channel. The cantilever is mounted between two flextensional actuators at its clamped end, with its free end extending downstream of the channel throat. The channel geometry, fluid and flow properties, along with cantilever geometry and material properties dictate the fluid-structure force amplitude and driving frequency at the cantilever's clamped end. The flextensional resonant frequencies must match those driven by the mounted cantilever, which couples to the fluid flow and provides the forcing function to the actuators. The dimension of the flextensional metal frame governs its resonant frequencies; thus, a frequency-matched flextensional system can be designed and fabricated by controlling the frame's length and thickness. The integration of flextensional actuators facilitates a design where the piezoelectrics are completely isolated from the working/producing fluids, reducing the effects of corrosion/degradation on the piezoelectric material. The emphasis of this work is to design a robust system, such as a highly efficient, piezoelectric harvester that is capable of generating suitable electrical power for sensors and actuators through a range of well production flow rates. This paper presents the preliminary results of experimental and simulation data for various designed, modeled, and prototyped flow energy harvesters. 2. Theoretical Background {#sec2-sensors-15-26039} ========================= A cantilever beam with two piezoelectric layers in parallel connection (bimorph), where the two piezoelectric layers have the same polarization directions, has been used to design the initial prototype set of energy harvesting devices. The voltage is generated between the intermediate electrode and the top/bottom electrodes. The constitutive equations describing the behavior of the cantilever type piezoelectric bimorph were first derived by Smits *et al.* \[[@B24-sensors-15-26039],[@B25-sensors-15-26039]\], where the deflection at the free end, $\delta$, and the charge on the electrode, $Q$, are related to an applied force at the free end, $F$ and an applied voltage over the electrodes, $V$ through a 2 × 2 matrix. The matrix equation under static conditions can be written: $$\begin{pmatrix} \delta \\ Q \\ \end{pmatrix} = \begin{pmatrix} \frac{s_{11}^{E}L^{3}}{2wh^{3}} & {- \frac{3d_{31}L^{2}}{4h^{2}}} \\ {- \frac{3d_{31}L^{2}}{4h^{2}}} & {\frac{2\varepsilon_{33}^{T}wL}{h}(1 - \frac{k_{31}^{2}}{4})} \\ \end{pmatrix}\begin{pmatrix} F \\ V \\ \end{pmatrix}$$ where $d_{31}$, $k_{31}$, and $s_{11}^{E}$ are the piezoelectric strain coefficient, electromechanical coupling, and elastic compliance in matrix notation, respectively. $\varepsilon_{33}^{T}$ is the dielectric permittivity. $L$, $w$, and $h$ are length, width, and thickness of a piezoelectric plate, respectively. If an external force $F$ is acting at the bimorph tip (free end), and no voltage is present at the deflected end, (*i.e.*, closed circuit), the generated electrical charge can be expressed as follows: $$Q_{b} = - \frac{3d_{31}L^{2}}{4h^{2}}F$$ The generated open circuit voltage can be obtained by using the following relation: $$V_{\text{OC,b}} = \frac{Q}{C_{p}} = \frac{3}{8}(\frac{L}{wh})\frac{d_{31}}{\varepsilon_{33}^{T}}F$$ where $C_{p}$ is the static capacitance of a piezoelectric plate, $$C_{p} = \frac{\varepsilon_{33}^{T}A}{t}$$ where A is the surface area of a piezoelectric plate, $t$ is the thickness of the composite beam (*t* = 2 h). From the constitutive equation, Equation (1), the stiffness of bimorph, defined as $k = \frac{F}{\delta}$, under closed circuit condition ($V = 0$) is given by: $$k = (\frac{s_{11}^{E}L^{3}}{2wh^{3}})^{- 1}$$ Since the bimorph is generally composed of several protective layers, the beam stiffness can be expressed as a function of effective Young's modulus ($Y_{b}$) of composite beam: $$k_{eq} = \frac{Y_{b}wt^{3}}{4L^{3}}$$ Considering the bimorph in a short-circuit condition as a single degree of freedom spring-mass system, a natural resonant frequency arises as $\omega = \sqrt{\frac{k}{m}}$; thus, the natural frequency of the undamped composite beam ($\omega_{b}$) is given by: $$\omega_{b} = \sqrt{\frac{Y_{b}wt^{3}}{4m_{eq}L^{3}}} \approx \frac{t}{L^{2}}\sqrt{\frac{Y_{b}}{\rho}}$$ where m~eq~ is an equivalent mass placed at the free end of the cantilever beam and $\rho$ is the density of the composite bimorph. For a piezoelectric bar operating in a 33 mode, the stress is parallel to the polarization direction, and the generated voltage can be obtained using the piezoelectric constitutive equation: $$E_{3} = - \mathbf{g}_{\mathbf{33}}T_{3} + \beta_{33}^{T}D_{3}$$ where $T_{3}$ and $D_{3}$ are the stress on the element in the direction of polarization and dielectric displacement, $E_{3}$ is electric field $( = V/t$), $\beta_{33}^{T}$ is free dielectric impermeability constant. For the piezoelectric voltage coefficient $g_{33}$, the following equation can be used: $$g_{33} = \frac{d_{33}}{\varepsilon_{33}^{T}}$$ Under open-circuit condition ($D_{3} = 0$), Equation (8) reduces to: $$|V_{OC,l}| = \frac{d_{33}}{\varepsilon_{33}^{T}}tT_{3}\operatorname{~~}\text{=}~\frac{d_{33}}{C_{p}}F_{3}$$ The generated electrical charge of a longitudinal bar is linearly proportional to the piezoelectric charge coefficient under an applied force, and is expressed as $Q_{l} = d_{33}F_{3}$. The natural frequency of a longitudinal piezoelectric bar ($\omega_{l}$) is given by: $$\omega_{l} = \sqrt{\frac{Y_{l}}{\rho}}$$ where $Y_{l}$ is the Young's modulus of the poled piezoelectric bar, which is the reciprical of $s_{33}^{E}$. For the case of multilayer piezoelectric stack, where a number of thin alternately poled piezoelectric layers are connected mechanically in series and electrically in parallel, the effective piezoelectric strain coefficient and capacitance is proportional to the number of the piezoelectric layers given by: $$d_{33}^{*} = nd_{33}$$ $$\varepsilon_{33}^{T*} = \frac{nL\varepsilon_{33,piezo}^{T}}{t_{p}} \approx n^{2}\varepsilon_{33,piezo}^{T}$$ where $n$ is a number of a piezoelectric layers, $L$ and $t_{p}$ are the total length of the piezoelectric stack and the thickness of a single piezoelectric layer, respectively. {#sensors-15-26039-f002} When an external resistive load ($R_{L}$) is connected to the piezoelectric harvester, as shown in [Figure 2](#sensors-15-26039-f002){ref-type="fig"}, the voltage across the resistive load is maximum when the load resistance ($R_{L}$) is matched to the source impedance ($Z_{p}$). If the piezoelectric material is an ideal capacitor and the dielectric loss of the piezoelectric structure is infinitely small, the voltage across the resistive load can be obtained as follows: $$V_{\text{L}} = V_{\text{OC}}\left| \frac{R_{\text{L}}}{Z_{p} + R_{\text{L}}} \right|$$ where $Z_{p} \approx \left( {\omega C_{p}} \right)^{- 1}$, and the maximum electrical voltage to a resistive load is half of the open circuit voltage, (*i.e.*, $V_{L,max}~ = ~V_{OC}/2$). Under dynamic conditions (simple sinusoid), the RMS power delivered to the resistive load is given by: $$P_{\text{rms}} = \frac{V_{\text{L}}^{2}}{2R_{\text{L}}}$$ The RMS voltage for a time dependent signal can be calculated directly from the following equation: $$V_{rms} = \sqrt{\frac{1}{T}\int_{0}^{T}V{(t)}^{2}\mspace{2mu} dt}$$ 3. Flow Loop Experimental Setup {#sec3-sensors-15-26039} =============================== A schematic of the flow loop used in experiments is shown in [Figure 3](#sensors-15-26039-f003){ref-type="fig"}. The loop contains a motor/pump and a reservoir, and has digital pressure gauges (G1 Pressure transducers manufactured by ASHCROFT, Stratford, CT, USA) on the inlet and outlet of the flow energy harvester test section. For the measurement of flow rate, a digital flow meter (Stainless Steel Flowmeter with 4--20 mA module, GPI, Wichita, KS, USA) is used and is located at the reservoir inlet. A safety relief valve is positioned between the pump and the accumulator. Accumulator is used to reduce vibration and noise from the pump. Pressure regulators are placed before and after the flow energy harvester. The pump speed and flow rate are adjusted dictating the harvester test section inlet and outlet pressures. A filter between the reservoir and the pump is used to catch particles in the flow. The tubing is made of copper and the joints are connected with AN fittings. The back-pressure on the flow energy harvester test section can be maintained above a critical level to suppress cavitation via a needle valve. ![Schematic of the flow loop that is currently being used to measure the performance of the harvesters (reprinted with permission from \[[@B18-sensors-15-26039]\]; copyright 2014 SPIE Publications).](sensors-15-26039-g003){#sensors-15-26039-f003} Once the test section is fitted accordingly, the pump motor rotating speed was set and the flow rate, inlet, and outlet pressures were recorded. For power output measurements, the load resistance across the electrical output of the harvester and corresponding voltage $V$ were measured using an oscilloscope (TDS 2024B, Beaverton, OR, USA). The waveform of $V$ in time was downloaded to the computer and the instantaneous power was time averaged to find the average maximum recoverable power generated. Nozzle-diffusers, the housing, and mechanical cantilever designs for one set of tests are shown in [Figure 4](#sensors-15-26039-f004){ref-type="fig"}. Different spline-shaped nozzle-diffuser configurations were designed with varying throat sizes. In order to accommodate the maximum flow loop pressure (1.7 MPa), the harvester housings were made of aluminum and the flow profiles (nozzle-diffuser inserts) were mounted with dowel pins and screws. An o-ring was inserted into a groove at the front face of the housing, and a Plexiglass cover was then screwed-in against the o-ring to seal the surface. {#sensors-15-26039-f004} 4. Bimorph Harvester {#sec4-sensors-15-26039} ==================== A piezoelectric bimorph connected in parallel (QP21B produced by Mide Technology Corporation, Medford, MA, USA \[[@B26-sensors-15-26039]\]) was used to demonstrate power generation in our first flow energy harvester design. This bimorph consists of two thin 0.008 inch (0.2032 mm) of PZT5A type piezoelectric elements (3195HD) that are 1.33 inch (33.782 mm) long and 0.56 inch (14.224 mm) wide. They are covered with thin 0.001 inch (0.0254 mm) polyimide layers to protect and electrically isolate the electrodes. A 0.002 inch (0.0508 mm) stainless steel shim was added to either side of the commercially available QP21B bimorph actuator in order to increase its fatigue life and erosion resistance (armored QP21B). An exploded view of the armored QP21B actuator is shown in [Figure 5](#sensors-15-26039-f005){ref-type="fig"}, where the total length and width are 1.63 inch (42.418 mm) and 0.67 inch (17.018 mm), respectively. The total thickness is 0.031 inch (0.7874 mm). A photograph and schematic illustration of our typical bimorph flow energy harvester are shown in [Figure 6](#sensors-15-26039-f006){ref-type="fig"}. {#sensors-15-26039-f005} {#sensors-15-26039-f006} [Figure 7](#sensors-15-26039-f007){ref-type="fig"} shows the electrical impedance and phase spectra of armored QP21B in air and in water as a function of frequency, measured using an HP4294A impedance/gain-phase analyzer (Hewlett-Packard, Palo Alto, CA, USA). This impedance analysis determines the target resonant frequency to be excited by the flow as well as the optimum resistance of the load resistor. Note that the first resonant frequency of the armored QP21B is located at around 400 Hz in air, but decreases to around 140 Hz in water. Its optimal electrical resistance at 140 Hz in water is between 7 and 10 kOhm. For power output measurements of the bimorph, the harvester system is placed in the flow loop described. An example snapshot showing the deflection of piezoelectric bimorph at its maximum can be found in [Figure 8](#sensors-15-26039-f008){ref-type="fig"}. This frame was taken at a flow rate of 15 L/min with a nozzle throat size of 1.25 mm, using a high speed (1200 frame per second) camera (Nikon 1 J4, Chiyoda, Tokyo, Tokyo, Japan). As shown, the max deflection of this bimorph is found to be on the order of 2 mm. {#sensors-15-26039-f007} {#sensors-15-26039-f008} The measured power from the armored QP21B for different nozzle throat sizes is shown in [Figure 9](#sensors-15-26039-f009){ref-type="fig"}. Note that the output power with all tested nozzle-diffuser inserts showed the same maximum of 25 mW albeit at different flow rates (ranging from 10 to 20 L/min). Notice that the data clearly shows a critical flow rate exists at which power generation increases rapidly ("turns on"), then appears to further increase linearly with increasing flow rate. The larger the throat size the larger this critical flow rate. Another trend in the data is an increase in throat size leads to a decrease in the average pressure drop across the test section; for example, at large nozzle throat size of 3.5 mm one could only generate non-negligible amounts of power near the maximum flow rate of 20 L/min. Yet the pressure drop is much smaller than the subsequent runs with smaller nozzle throat sizes (35--70 kPa for the larger throat). {#sensors-15-26039-f009} [Figure 10](#sensors-15-26039-f010){ref-type="fig"} shows the armored QP21B voltage waveforms for multiple flow rates and a nozzle throat size of 1.25 mm. {#sensors-15-26039-f010} The frequency-domain signal of the voltage waveforms were also obtained by taking their respective fast Fourier transform (FFT). Notice that the dominant frequency at 15 L/min corresponds to the first mode, located between 140 and 150 Hz; however, as the flow rate is increased, the voltage signal shows multifrequency response peaks, exhibiting first (140 Hz) and second (1 kHz) resonance mode vibrations. Although not shown in [Figure 9](#sensors-15-26039-f009){ref-type="fig"}, by further increasing the flow rate to 20 L/min power levels up to 35 mW could be generated, yet the life of the device, as expected, was found to be short. Testing at reduced flow rates, under 10 L/min, produced 5--6 mW and was shown to run for 9 h without failure (3,888,000 cycles); however after 9 hours a crack in the metal shim developed where the actuator was clamped. The photograph of the armored QP21B after the 9 h life test, shown in [Figure 11](#sensors-15-26039-f011){ref-type="fig"}, suggests that the cause of the failure was that the actuator was being driven past its metal shim fatigue limits. {#sensors-15-26039-f011} The crack location suggests that the rigid fixture and the right angle joint clamping the bimorph may have been responsible for stress concentrations that caused the fatigue failures observed. In order to investigate the effect of mounting on the fatigue of the piezoelectric cantilever, a stress analysis was performed using the finite element method. Note that the fatigue strengths of the strainless steel and piezoelectric material in the bimorph actuators are around 210 MPa and 50 MPa, respectively, below which fatigue failure is not expected \[[@B27-sensors-15-26039],[@B28-sensors-15-26039]\]. Hence, the fatigue failure of the bimorph harvester can be eliminated with a design that ensures cyclic stresses are sufficiently lower than these limits. To develop the finite element model of the multilayered, armored QP21B, the commercial finite element software package ABAQUS was used. A geometric model of the armored QP21B was created using Siemens PLM Software NX software, and then imported into ABAQUS for the stress analysis. [Figure 12](#sensors-15-26039-f012){ref-type="fig"} shows the geometry of the model with applied the load and fixed boundary conditions. The FEA model includes epoxy layers, Espanex polyimide layers, stainless steel shims and the piezoelectric layers, whose properties are summarized in [Table 1](#sensors-15-26039-t001){ref-type="table"} and [Table 2](#sensors-15-26039-t002){ref-type="table"}. Hexahedral (brick) elements with reduced integration are used for both piezoelectric elements (C3D8E) and non-piezoelectric layers (C3D8R). All the layers are meshed with an element size of 0.4 mm in length and width, and four elements along the thickness direction. A finer mesh, 0.2 mm in element size, was used in the region where the actuator was mechanically clamped in order to improve the accuracy of localized stress concentrations. {#sensors-15-26039-f012} sensors-15-26039-t001_Table 1 ###### Material parameters of non-piezoelectric layer. Property Epoxy Polyimide Stainless Steel ----------------------- ------- ----------- ----------------- Density (kg/m^3^) 0.73 1.42 7800 Young's Modulus (GPa) 3.5 2.5 200 Poisson's ratio 0.35 0.34 0.3 sensors-15-26039-t002_Table 2 ###### Piezoelectric properties (CTS-3195HD) from Mide Technology Corporation. Property Symbol Value --------------------------------------- --------------------------------------------------------------------------------------------------------- -------------------------------- Density (kg/m^3^) ρ 7800 Relative permittivity $\varepsilon_{11}^{s}/\varepsilon_{0}\,,\,\varepsilon_{33}^{s}/\varepsilon_{0}$ 916, 830 Compliance (pm^2^/N) $\text{s}_{11}^{D}$, $\text{s}_{12}^{D}$, $\text{s}_{13}^{D}$, $\text{s}_{33}^{D}$, $\text{s}_{44}^{D}$ 14.4, −4.24, −2.98, 9.43, 25.2 Piezoelectric charge constants (pC/N) d~33~, d~31~, d~15~ 390, −190, 585 [Figure 13](#sensors-15-26039-f013){ref-type="fig"}a,b show the static analysis results exposing the stress concentrations at the clamped joint of the armored QP21B. A load of 1 N is applied at the tip of the beam and causes a tip deflection of around 0.18 mm. The stainless steel layer shows the highest stress level (50 MPa), in part due to the higher Young's modulus of stainless steel compared to other layers. The piezoelectric layer exhibits a maximum von Mises stress level of around 15 MPa. The electric potential (open circuit voltage), maximum bending stress, and tip displacement of the piezoelectric layer as a function of position along the actuator length are shown in [Figure 13](#sensors-15-26039-f013){ref-type="fig"}c, with an open circuit voltage of 20 V at a tip deflection of 0.18 mm. The electrical RMS power output can be estimated at 5 mW from Equation (15), assuming that the bimorph is vibrating at 150 Hz, that the matched electrical impedance is 10 kOhm, $R \approx \left( {\omega C_{p}} \right)^{- 1}$, and $V_{L} = 10~V$. The stiffness can be estimated from these results using its definition, $k = \frac{F}{\delta}$, and the effective Young's modulus of the armored QP21B struture can be calculated according to Equation (6). The former and latter values are 5.55 N/mm and 205 GPa, respectively. {#sensors-15-26039-f013} Under an unsteady fluid flow driving force, the armored QP21B tip deflection reaches an approximate maximum of 2 mm (see [Figure 8](#sensors-15-26039-f008){ref-type="fig"}), meaning that the bending stresses on the structure are about 10 times higher than those predicted in the previous analysis. For example, the stress on the stainless steel layer would be on the order of 500 MPa, which is above the fatique limit of the material. A variety of methods to reduce the stress concentration has been investigated, including adding a fillet or a glue bond line that extends along the mounting edge line of the bimorph. The stepped joint mounting design is found to be simple and effective in reducing these stress concentrations without affecting the bending stiffness of the structure. [Figure 14](#sensors-15-26039-f014){ref-type="fig"} shows the stress analysis results of armored QP21B with a stepped joint (0.0508 mm in thick) under a 2 mm tip displacement. For comparison, the stress analysis result of armored QP21B without a stepped joint under a 2 mm tip displacement is also shown in the figure. Note that as expected, extending the mount and stepping it can have a 26% reduction in the stress field at the mounting line. {#sensors-15-26039-f014} An alternate means of calculating power output and a first order approximation of the stress profile along the length of different layers of the bimorph actuator used video data that mapped the beam shape. Although the method needs refinement, this section aims at discussing its details and preliminary results as compared to the finite element simulations and experimentally measured power output. The analysis considers the cantilever as a forced Euler-Bernoulli beam with clamped- and free-end boundary conditions. This implies that bending stresses are dominant over shear stresses, and that small strain and small beam displacement relative to the beam length$~L$ exist. The shape of the beam was approximated by an eigenfunction expansion that satisfies the above boundary conditions for the eigenvalue problem of the biharmonic operator \[[@B29-sensors-15-26039]\]: $$\phi_{n}(x) = \cosh(\beta_{n}x) - \cos(\beta_{n}x) + a_{1}(\sin(\beta_{n}x) - \sinh(\beta_{n}x))$$ where the wave number $\beta_{n}$ satisfies the constraint: $$\cos(\beta_{n}L)\cosh(\beta_{n}L) = - 1$$ and the constants are: $$a_{1} = \frac{(\cos(\beta_{n}L) + \cosh(\beta_{n}L))}{(\sin(\beta_{n}L) + \sinh(\beta_{n}L))}$$ with $x$ as the coordinate along the beam length. The displacement $\delta$ is: $$\delta(x,t) = {\sum\limits_{n = 1}^{\infty}\xi_{n}}(t)\phi_{n}(t)$$ where $\xi_{n}\left( t \right)$ is a periodic function of time. [Figure 15](#sensors-15-26039-f015){ref-type="fig"} shows the beam deflection $\delta$ as a function of $x$ and time $t$. ![Illustration of deformed beam represented by $\delta\left( {x,t} \right)$ with coordinate system (reprinted with permission from \[[@B18-sensors-15-26039]\]; copyright 2014 SPIE Publications.)](sensors-15-26039-g015){#sensors-15-26039-f015} Camera data taken as slow motion video (Nikon 1 J4, 1200 frame per second) was processed frame by frame and the edges of the vibrating cantilever were mapped in the x-y plane, shown in [Figure 15](#sensors-15-26039-f015){ref-type="fig"}, using the Canny edge filter implementation in MATLAB \[[@B30-sensors-15-26039]\]. The filter parameters were chosen as necessary to consistently capture the edges near the same location (±1 pixel) for each video file. The actuator experiments chosen for the initial processing consisted of vibration almost entirely in actuator's fundamental mode. Hence, only the $n = 1$ solution to Equation (17) is considered for the cases shown. At each frame, the x-y actuator edge data is least squares fitted to Equation (16), yielding a constant coefficient for $\xi_{1}\left( t_{i} \right)$, where index $i$ is the frame number. $\xi_{1}\left( t \right)$ is a sinusoid fit to the time series $\xi_{1}\left( t_{i} \right)$ with the appropriate frequency and phase parameters. As shown by Sodano *et al.* \[[@B31-sensors-15-26039]\], the power output can be calculated from $\delta\left( {x,t} \right)$. The piezoelectric constituent equations applied to the bimorph actuator geometry yields a first order ODE for the charge $q\left( t \right)$ as: $$R\overset{˙}{q}(t) - C_{p}^{- 1}\Theta_{i}\xi_{i}(t) + C_{p}^{- 1}q(t) = 0$$ where $R$ is the circuit resistance and the dot represents the derivative in time. The capacitance is: $$C_{p} = {\iiint_{V_{p}}\psi^{2}}(y)\varepsilon dV$$ with $V_{p}$ as the bounds to the piezoelectric volume. The electromechanical coupling constant is: $$\Theta_{1} = - {\iiint_{V_{p}}y}e\phi_{1}^{\prime\prime}(x)\psi(y)dV$$ where $e$ is the piezoelectric coupling coefficient, $y$ is the coordinate variable defined from the neutral plane of the beam, and the function: $$\psi(y) = \left\{ \begin{matrix} {- \frac{1}{t_{p}},} & {\frac{t}{2} < y < \frac{t}{2} + t_{p}} \\ {0,} & {- \frac{t}{2} < y < \frac{t}{2}} \\ {\frac{1}{t_{p}},} & {- \frac{t}{2} - t_{p} < y < - \frac{t}{2}} \\ \end{matrix} \right.$$ for $y$ values shown in [Figure 16](#sensors-15-26039-f016){ref-type="fig"}b. ![Illustration of (**a**) bimorph and (**b**) cross-section. Blue represents the piezoelectric material layer, grey the polyimide coating, and red the 301 stainless steel layer for the geometry of the armored QP21B actuator (reprinted with permission from \[[@B18-sensors-15-26039]\]; copyright 2014 SPIE Publications).](sensors-15-26039-g016){#sensors-15-26039-f016} Variation of parameters yields the steady state solution to Equation (21): $$q_{ss} = {\int_{0}^{t}\exp}\lbrack{(RC_{p})}^{- 1}(t - \tau)\rbrack{(RC_{p})}^{- 1}\Theta_{1}\xi_{1}(\tau)d\tau$$ and the instantaneous power output $P$: $$P(t) = {\overset{˙}{q}}_{ss}^{2}R$$ Video for the armored QP21B was analyzed for the experiments with a nozzle throat size of 1.25 mm and flow rates of 9.5 L/min, 12.4 L/min, 15.3 L/min, and 18 L/min. The power generated is shown in [Figure 17](#sensors-15-26039-f017){ref-type="fig"}. The piezoelectric coupling coefficient is $e = 1.5052\text{E} - 08\text{~m}/\text{V}$ and resistance is $R = 10$ kOhms. The figure shows the results of the analysis, plotting the predicted power output with an error bound of ±1 pixel and the time average of RMS oscilloscope measured voltage and derived power (error ranges from ±1.4 mW at the lowest flow rate to ±7.1 mW at the highest). ![Video processed power data (x) and oscilloscope measurement (dots) (reprinted with permission from \[[@B18-sensors-15-26039]\]; copyright 2014 SPIE Publications).](sensors-15-26039-g017){#sensors-15-26039-f017} Although the model is limited to assumptions of the Euler-Bernoulli beam and the boundary conditions mentioned, it seems to do a reasonable job at predicting the measured RMS power. Consequently, stresses in the stainless steel can be estimated from the fitted eigenfunction as: $$\sigma_{ss} = \frac{My}{I}$$ where is the $M$ is the bending moment: $$M = EI\frac{\partial^{2}\delta}{\partial x^{2}}$$ and $I$ is the area moment of inertia for each section shown in [Figure 16](#sensors-15-26039-f016){ref-type="fig"}. For the flow rates tested, [Figure 18](#sensors-15-26039-f018){ref-type="fig"} shows the maximum bending stresses in the stainless steel section. The corresponding tip displacements from the processed video data are shown in [Figure 19](#sensors-15-26039-f019){ref-type="fig"}. From the FEA results shown in [Figure 14](#sensors-15-26039-f014){ref-type="fig"}, the stress level magnitudes of contours outside of the stress concentration zone are within the ballpark of the surface stresses calculated using the video data. For example, with a tip displacement of \~1.68 mm, the maximum surface bending stress of the stainless-steel element is \~238 MPa. {#sensors-15-26039-f018} {#sensors-15-26039-f019} 5. Double Flextensional-Cantilever Harvester {#sec5-sensors-15-26039} ============================================ A second promising prototype is the Double Flextensional Cantilever Harvester (Double- FCH). A schematic representation of the Double- FCH is shown in [Figure 20](#sensors-15-26039-f020){ref-type="fig"}, where the metal cantilever is mounted and coupled between two flextensional actuators. The principle of operation is the flow-induced vibration onto the non-piezoelectric cantilever transfers forces into the flextensional frame along the y axis (minor axis). The frame, in turn, amplifies the forces along its x axis (major axis). The force amplification is related to the ratio of the long axis length to the short axis height, which can be approximated as shown in [Figure 20](#sensors-15-26039-f020){ref-type="fig"}, (*i.e.*, $F_{x} = F_{y}\cot\theta$). {#sensors-15-26039-f020} Commercially available APA 400M flextensional actuators (Cedrat Technologies S.A., Meylan, France) were selected due to their resonant frequency of \~350 Hz under a blocked-free boundary condition. This resonant frequency can be tuned further with an added mass at either support point of the flextensional actuators. In order to use these flextensional actuators in water, the piezoelectric stacks were replaced with water resistant stacks purchased from American Piezoelectric Ceramic (Mackeyville, PA, USA) APC − 30 × 45 − 1130 Pst150/5 × 5/20. The stack has the cross section 5 mm × 5 mm and a length of 18 mm. Finite element analysis of the Double-FCH was performed using ABAQUS in order to predict its performance. An assumption for the model consisted of the stack as an isotropic material, with effective Young's modulus and dielectric permittivity derived based on the measured and reported dimensions, stiffnesses, piezoelectric strain coefficients and capacitances of stacks \[[@B32-sensors-15-26039]\] (shown in [Table 3](#sensors-15-26039-t003){ref-type="table"}). Piezoelectric materials were meshed with a global element size of 0.5 mm, generating 14400 hexahedral linear elements (C3D8E), while non-piezoelectric materials were meshed with total 53064 hexahedral linear elements, which have a global element size of 0.5 mm with finer elements in the thickness of flextensional frame and cantilever beam. sensors-15-26039-t003_Table 3 ###### Material parameters of piezoelectric multlayer stack (APC − 30 × 45 − 1130 Pst150/5 × 5/20). Materials l/w/h k C ε~33~^T\*^ d~33~\* Y~l~ ----------------- -------- -------- ------ ------------ ---------- ------- \(mm\) (N/um) (µF) (F/m) (m/V) (Gpa) Pst150/5 × 5/20 5/5/18 60 1.63 0.00117 8.41E−08 43 A modal analysis was performed in order to identify the mode shapes, the resonant frequencies and electromechanical coupling of the Double-FCH structure. [Figure 21](#sensors-15-26039-f021){ref-type="fig"} illustrates the deformed shapes of the double-FCH at resonance (374.9 Hz) in a short circuit condition. The undeformed shape is superimposed on the deformed shape. An open circuit modal analysis was also performed by removing the voltage boundary condition, and its first natural frequency in resonance found to be 385.5 Hz. Note that the resonance frequency $f_{r}$ represents the mechanical resonance vibrating under short-circuit conditions, while the anti-resonance frequency $f_{a}$ represents the mechanical resonance vibrating under an open-circuit condition, indicating that *f~r~* and *f~a~* are 374.9 Hz and 385.5 Hz, respectively. The effective electromechanical coupling factor can be calculated using Equation (29), and found to be 0.23. Note that higher electromechanical coupling factor implies better mechanical coupling between flextensional actuators and cantilever beam, allowing for effective vibration transfer from the cantilever beam to the flextensional actuators, and to piezoelectric stacks: $$k_{eff} = \sqrt{1 - {(\frac{f_{r}}{f_{a}})}^{2}}$$ {#sensors-15-26039-f021} [Figure 22](#sensors-15-26039-f022){ref-type="fig"} shows the static analysis results, exhibiting the open circuit voltage from the piezoelectric stacks when a load of 1 N is applied at the free end of the beam. The reaction forces at the piezoelectric stack are found to be 0.8 N along the y direction, but are amplified to 10 N along the x direction due to the frame lever arm magnification of the applied force. Note that the open circuit voltage can be analytically calculated using the Equation (10), where $d_{33}$, $C_{p}$ and $F_{x}$ are 0.0841 µm/V, 1.63 µF and 10 N, respectively, yielding around 0.7 V. An estimate of power output can be made assuming that the structure is vibrating at its natural resonant frequency of 374 Hz and has a matched resistance $R_{L}$) of 235 Ohm based on $R_{L} = \left( {\omega C_{p}} \right)^{- 1}$. The generated instantaneous power output per a stack would be around 0.2 mW, and since the FCH is composed of 4 stacks, the FCH yields a total of around 0.8 mW. This is a factor of 5--6 smaller than when compared to that of the armored QP21B harvester. The lower power with respect to applied force is due to the stack's higher stiffness (60 N/um) as compared to the armored QP21B (5.5 N/um). Note, however, that this design can survive higher stress levels than the bimorph actuator and thus produce higher power when an appropriate flow passage and cantilever design provides a frequency-matched, high amplitude (\~10 N) fluid-structure forcing function. Another method to further increase output power is to use a lower stack stiffness. Since capacitance is inversely proportional to the cross-sectional area of the stack, the use of a stack with smaller cross-sectional area allows for higher output power under the same boundary conditions assuming that the applied force and piezoelectric charge coefficient are kept constant (see Equation (30)): $$P = \frac{\omega d_{33}^{2}}{2C_{p}}F^{2}$$ {#sensors-15-26039-f022} The Double-FCH performance was determined experimentally in both air and water. The generated output voltages were investigated as a function of the load resistance at the various pressure levels in air first. [Figure 23](#sensors-15-26039-f023){ref-type="fig"} shows the corresponding power output from a single flextensional actuator as a function of the resistance. The power shows a flat peak of about 35 mW at about 200--350 Ohms at the maximum inlet pressure of 410 kPa, and was decreased above 350 Ohm. A flat peak in the range of 200--350 Ohm, rather than a sharp peak at a matched resistance (\~235 Ohm), is believed to be due to combined resonance effects from both cantilever and flextensional actuators. This induces multi-frequency harmonic excitations on piezoelectric materials, resulting in a flat peak in the range of 200--350 Ohm. Note that since this power is from a single flextensional actuator equipped with two stacks, the flow energy harvesting device is technically capable of generating 70 mW (four stacks per flow energy harvester). The Double-FCH was then tested in the flow loop system described. The measured power and pressure drop as a function of the flow rate are shown in [Figure 24](#sensors-15-26039-f024){ref-type="fig"}. The maximum power corresponds to about 25 mW across a 100 Ohm resistor at a flow rate of 20 L/min and a pressure drop of 165 kPa. It should be noted that to achieve this power level, the cantilever was shortened to 90 mm from the original 100 mm tested in air. The voltage waveform and corresponding frequency content across a 100 Ohm resistor at a flow rate of 20 L/min are shown in [Figure 25](#sensors-15-26039-f025){ref-type="fig"}, where the dominant frequency excited for this harvester is found to be about 305 Hz with \~1.5 of RMS voltage. {#sensors-15-26039-f023} {#sensors-15-26039-f024} {#sensors-15-26039-f025} 6. Conclusions and Future Work {#sec6-sensors-15-26039} ============================== Two designs of piezoelectric transducers were investigated in this study. Both the bimorph- and flextensional-based flow energy harvesters rely on fluid motion coupled to structural vibration via a cantilever placed in a converging-diverging flow channel. The bimorph is the cantilever itself, while the flextensional clamps a non-piezoelectric cantilever that provides the forces it converts into electricity. The two designs experimentally generated power at a level of 20 mW and above, with the bimorph type harvester specifically prone to fatigue failure caused by stress concentrations at its mounting point. A stepped joint mounting design was shown via FEA to ameliorate this issue, with a reduction of 26% in stress concentration without a reduction in power output. The flextensional actuator based harvester was found to be a viable alternative to the bimorph, with a power generation of \~20 mW from a single flextensional actuator. Greater robustness (unexposed piezoelectric to flow) and more design flexibility (cantilever can be designed independently of piezoelectric fatigue limits) are advantages the flextensional type harvester (FCH) provides. These results have prompted further investigation into different designs using this type of actuator. For example, the ability to tune the resonance frequency of the actuator by adding mass to the flextensional bodies may allow another design parameter that can be optimized to maximize power output. Further research is currently underway in order to increase fluid-structure coupling efficiency and further ruggedize harvesters. The research at the Jet Propulsion Laboratory (JPL), a division of the California Institute of Technology, was carried out under a contract with the National Aeronautics Space Agency (NASA). Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise, does not constitute or imply its endorsement by the United States Government or the Jet Propulsion Laboratory, California Institute of Technology. All authors contributed to the design of the experiment and discussed the results. H.J.L. and L.P.T. co-wrote the manuscript, and all authors contributed to the refinement of the paper. The authors declare no conflict of interest.

Background ========== There is an increasing interest in studying genotype by environment interactions in terrestrial and aquatic organisms to clarify how different populations might have diverged to become adapted to local habitats, and to understand how populations may respond to human-induced selection. In marine fishes the population structure is expected to be weak because of extensive gene flow, and it might be questioned whether adaptive responses to changes in physico-chemical conditions or selective harvesting have occurred in this large vertebrate group. These objections have been challenged by an increasing number of genetic studies of population structure \[[@B1]-[@B3]\]. The focus has recently changed from the rather descriptive population genetic analyses to an intensive search for functional genes subjected to natural selection \[[@B4],[@B5]\]. The limited number of fitness-related genes reported to be associated with adaptive traits in marine or euryhaline fishes includes the three-spined stickleback ectodysplasin \[[@B6]\], European flounder Hsc70 \[[@B7]\], killifish lactate dehydrogenase \[[@B8]\], and Atlantic cod pantophysin \[[@B9],[@B10]\] and hemoglobin \[[@B11]\]. The latter study provided evidence for local adaptation in cod populations by identifying molecular mechanisms underlying the different oxygen binding properties of the cod hemoglobins in temperate and Arctic waters. An alternative strategy for identifying adaptive population divergence is to scan transcriptome databases, or the whole genome when available, for polymorphic loci to be analysed in different populations \[[@B5]\]. Evidence of directional selection in Atlantic cod was provided by genotyping multiple gene-associated single nucleotide polymorphisms (SNPs) in Northeast and Northwest Atlantic populations \[[@B12]-[@B14]\]. Outlier loci were shown to be correlated with temperatures and/or salinity conditions that might be associated with local adaptation. However, documentation of molecular changes underlying adaptive traits requires additional information about the specific genes involved and the regulatory or structural implications of the identified mutations. Iron is an essential element required for the growth and survival of most organisms, and the iron balance is therefore tightly regulated by several interacting iron-binding factors \[[@B15]\]. Serum transferrin plays a crucial role in iron metabolism as it provides most of the iron required for organismal functions \[[@B16]\]. The two homologous iron-binding lobes of modern transferrins are believed to have evolved by gene duplication of a primitive monolobal form \[[@B17],[@B18]\], which has been reported in a surprisingly few species. Most insect transferrins bind only one ferric iron, but this is because of extensive deletions in the C-terminal lobe \[[@B19]\]. Evidence of a 44-kDa hagfish transferrin with only one iron-binding site was contradicted in a later study \[[@B20]\], and a report on monolobal rat hemiferrin turned out to be erroneous \[[@B21]\]. True monolobal transferrin has therefore been claimed to be present only in ascidian species of the Urochordates \[[@B22]-[@B24]\]. A transferrin-like otolith matrix protein (OMP) with a single potential metal binding site was identified in rainbow trout and zebrafish \[[@B25]-[@B27]\], but the protein was proposed to represent a partial sequence of the membrane-bound melanotransferrin (MTf) \[[@B28]\]. The possibility for an alternatively spliced monolobal variant of the *MTf*gene might be excluded by the identification of both genes in the fish genome. By limiting the availability of iron for replicating pathogens transferrin provides resistance to bacterial infections \[[@B29],[@B30]\], but an iron-independent role of transferrin in the immune system is evidenced by the involvement of transferrin and its receptor in early T cell differentiation in the thymus \[[@B31]\]. The conserved dual functions of transferrin in the iron metabolism and immune response were recently demonstrated in sea bass, which responded to both bacterial infection and altered iron status by modulating the liver and brain expression of transferrin \[[@B32]\]. Bacterial species have a variety of mechanisms for obtaining transferrin-bound iron \[[@B33]\], and competition for iron from bacterial pathogens could potentially be a strong source of natural selection on vertebrate transferrins. For example, Newton et al. \[[@B34]\] provided evidence for a transferrin allele as a host-level risk factor in naturally occurring equine respiratory disease, and Jurecka et al. \[[@B35]\] showed that particular transferrin alleles were associated with the resistance of common carp to the blood parasite *Trypanoplasma borrelli*probably as a direct effect of binding to the transferrin receptor of the parasite. Atlantic cod has been one of the major fish resources on the continental shelf and banks on both sides of the North Atlantic Ocean, and has undergone a complex pattern of phylogenetic evolution, including population fluctuations attributable to long-term geological events, short-term ecological history and contemporary antropogenic fishing and environmental shifts \[[@B36],[@B37]\]. Serum transferrin polymorphism in Atlantic cod was reported almost 50 years ago \[[@B38]\], and the presence of two distinct cod populations at the Faroe Islands was proposed by analysing the frequencies of five transferrin types \[[@B39]\]. The important role played by transferrin in the immune response makes this polymorphic gene highly valuable for identifying adaptive divergence between local populations. Here we report on selective signatures of a tandem duplicated transferrin gene in trans-Atlantic populations, and provide novel evidence for the evolution of modern transferrins from a monolobal transferrin locus. Results ======= Four cod transferrin genes -------------------------- Two full-length Atlantic cod *Tf*cDNAs with different 3\'UTR length were identified in the GAFFA database of the Norwegian coastal cod population. The open reading frame (ORF) of 2073 nucleotides (nt) was identical in the two cDNAs, and the 689 amino acids (aa) of the designated cod Tf1 were predicted (Figure [1](#F1){ref-type="fig"}). The corresponding *Tf1*gene spanned about 8 kb of scaffold GmG100427sc4451 in the reference genome of Atlantic cod representing the Northeast Arctic population (Figure [2](#F2){ref-type="fig"}). Aligning the *Tf1*genomic and cDNA sequences revealed 17 exons sharing 100% sequence identity with the *Tf1*cDNA. The allelic variant of cod *Tf1*identified in the two Northeast Atlantic populations was named *Tf1*-NE. {#F1} {#F2} A second transferrin gene designated cod *Tf2*was found about 16.5 kb downstream of *Tf1*(Figure [2](#F2){ref-type="fig"}). The 17 exons of *Tf2*were identified by aligning the tandem duplicated *Tf*genes, and the predicted 683 aa of Tf2 shared 84% identity with cod Tf1 (Figure [1](#F1){ref-type="fig"}). The two serum transferrins showed 82 and 78% identity, respectively, with haddock Tf, but only 53-63% identity with the non-gadoid fish Tfs presented in the constructed phylogenetic tree (Figure [3](#F3){ref-type="fig"}). Although the gadoids clustered with the Antarctic notothenioids examined, the phylogenetic tree seems to be consistent with the taxonomic classification of teleosts, while man and *Ciona*formed separate branches. Four potential iron-binding residues forming a DYYH motif \[[@B40]\] are conserved in both the N- and C-lobe of cod Tf1 and Tf2, and the two lobes contained 12 and 14 conserved Cys residues, respectively, involved in the formation of disulphide bridges (Figure [1](#F1){ref-type="fig"}). However, the basic Arg residue serving as the principal anchor for the synergistic carbonate anion has been changed to Lys131 in the N-lobe of Tf1 and to Thr455 in the C-lobe of Tf2. The corresponding Arg124Lys mutation in the N-lobal anion binding site of human Tf generated a protein much more facile in releasing iron \[[@B41]\]. The 3D structure of Tf1 and Tf2 illustrates the bilobal nature of the proteins in which each lobe is divided into two subdomains connected by a hinge that give rise to a deep cleft containing the iron-binding residues (Figure [4A, B](#F4){ref-type="fig"}). The expression of the two genes was semi-quantified by RT-PCR using gene specific primers. The transcripts of both *Tf1*and *Tf2*were identified in the liver and brain, but *Tf2*was expressed at very low levels compared to *Tf1*(Figure [5A](#F5){ref-type="fig"}). Reducing the number of PCR cycles from 35 to 25 diminished the *Tf1*amplicon considerably only in the brain, indicating abundant expression of *Tf1*in the liver. {#F3} {#F4} {#F5} A cod melanotransferrin (*MTf*) gene and its cDNA were identified in scaffold GmG100427s526 of the reference genome (Figure [2](#F2){ref-type="fig"}) and in the GAFFA database, respectively. Sequence alignment revealed 17 exons, and the ORF of 2166 nt encodes the predicted 722 aa of cod MTf, which showed only 36% identity with cod Tf1 and Tf2 (Figure [1](#F1){ref-type="fig"}). In comparison, cod MTf shared 62-69% identity with the MTf of stickleback, medaka, pufferfish and zebrafish. The cluster of fish and human MTf in the phylogenetic tree was, however, not supported by the low boot strap value (\<50%), and *Ciona*Tf formed a separate branch (Figure [3](#F3){ref-type="fig"}). All four iron-binding residues (Asp393, Tyr423, Tyr527, His596) are present in the C-lobe of cod MTf, whereas Tyr56 and Arg257 in the N-lobe have replaced the crucial Asp and His residues, respectively (Figure [4C](#F4){ref-type="fig"}). Thus, the iron-binding activity is apparently intact only in the C-lobe of cod MTf, in contrast to the binding of iron to the N-lobe of mammalian MTf, in which the C-lobe is involved in membrane anchoring \[[@B42]\]. Although the residues for iron binding are not conserved in MTf, the overall sequence identity between the two lobes in MTf is higher than those of Tf in both man (48% and 42%, respectively) and cod (42% and 36%; Tf1, 38%; Tf2). Consistently, the N- and C-lobes of cod MTf aligned closely in the superimposition of the two lobes (Figure [4C](#F4){ref-type="fig"}), and a root mean square deviation (RMSD) of 1.5 Å over 264 aligned C-α atoms was calculated. A monolobal transferrin of only 370 aa was predicted from the 9 exons identified in a gene spanning the entire scaffold GmG100427s1898 of 37.8 kb in the reference cod genome (Figure [1](#F1){ref-type="fig"} and [2](#F2){ref-type="fig"}). The completeness of the gene was confirmed by the identification of the orthologous gene and the conserved flanking genes *Senp2*, *Tra2b*, *D2hgdh*and *Slitrk3*in the other fish genomes available, including medaka and stickleback (Figure [2](#F2){ref-type="fig"}). Whole mount *in situ*hybridization revealed the exclusive staining of the otoliths in the cod larvae (Figure [5B](#F5){ref-type="fig"}), and the predicted protein showed 79 and 76% identity, respectively, with the otolith matrix protein (OMP) of rainbow trout and zebrafish \[[@B25],[@B27]\]. Cod OMP shared low identity with the N- and C-lobes, respectively, of Tf1 (32% and 31%), Tf2 (34% and 32%) and MTf (38% and 35%), which are slightly higher than the identities of 31 and 30% between the monolobal nicaTf and the N- and C- lobes, respectively, of Tf in *Ciona intestinalis*\[[@B43]\]. The monolobal nature of cod OMP was illustrated by the 3D model of the protein superimposed on the C-lobe of cod MTf (Figure [4D](#F4){ref-type="fig"}). The C-α RMSD of 1.52 and 0.60 Å, respectively, between OMP and the N- and C-lobes of MTf (over 271 and 315 aligned C-α atoms, respectively) indicates a major structural similarity of OMP to the C-lobe, although cod OMP is lacking all the key residues for iron and carbonate anion binding, except for the conserved Tyr111 (Figure [1](#F1){ref-type="fig"}). Cod Tf1 polymorphisms --------------------- Polymorphisms in the cod *Tf1*were found by comparing the *Tf1*-NE variant with a *Tf*cDNA isolated from a Northwest (NW) Atlantic cod population \[[@B44]\]. The missing N-terminal coding sequence of the latter variant named *Tf1*-NW was PCR amplified from a heterozygous Faroe cod, and a total of 22 SNPs were identified by aligning the protein coding sequences of the two cod *Tf1*variants (Additional file [1](#S1){ref-type="supplementary-material"}; Figure S1). The 18 non-synonymous substitution sites were found to cause 16 amino acid replacements, including eight surface residues (Figure [4E](#F4){ref-type="fig"}). Additionally, E160 is deleted in the NW variant probably as the result of alternative splicing of exons 5 and 6. The 3D model of the NE and NW variants revealed no structural differences in the iron-binding sites, and the high structural similarity between the two variants was expressed by the calculated RMSD of only 0.64Å between Cα atoms. The *d*~*N*~/*d*~*S*~ratio of 5.8 between the NW and NE variants is significantly greater than 1 (*p*= 0.0013) indicating strong positive selection on the cod *Tf1*gene. Population genetic analyses --------------------------- The cod *Tf1*polymorphisms were investigated in 14 cod populations covering the North Atlantic (Table [1](#T1){ref-type="table"}, Additional file [2](#S2){ref-type="supplementary-material"}; Figure S2) by genotyping six of the 22 identified SNPs. The NE haplotype was almost fixed in the Baltic cod samples and predominated in the other eastern Atlantic samples, although at slightly different frequencies in the Northeast Arctic cod and the Greenland Nuuk sample (Figure [6](#F6){ref-type="fig"}, Additional file [3](#S3){ref-type="supplementary-material"}: Table S1). In contrast, the NW alleles of the three SNPs tf-8, tf-10 and tf-11 dominated in the Canadian samples examined, whereas alleles of the N-terminal (tf-6) and the C-terminal SNPs (tf-13, tf-22) were in similar proportions among all west Atlantic samples, including the Greenland Sisimiut cod. Population pair-wise *F*~ST~values based on all six SNP loci demonstrated a clear genetic separation between eastern and western cod samples, with intermediate Northeast Arctic (Båtsfjord) and Greenland cod (Figure [7](#F7){ref-type="fig"}; Additional file [4](#S4){ref-type="supplementary-material"}: Table S2). Heterozygosity showed an increasing trend from almost zero in the Baltic to close to 50% in western Atlantic samples (Table [1](#T1){ref-type="table"}, Additional file [3](#S3){ref-type="supplementary-material"}: Table S1). ###### Atlantic cod sampling locations and heterozygosities. Locality Lat Long Sampling year Sample size ***H***~**o**~ ***H***~**e**~ --------------------------- ------- -------- --------------- ------------- ---------------- ---------------- Baltic Öland 56.04 16.41 2004 29 0.006 0.006 Baltic Bornholm 55.50 16.00 2004 30 0.039 0.039 Kattegat 56.90 12.15 2004 29 0.058 0.056 North Sea 55.57 05.85 2002 29 0.126 0.119 Norwegian coast, Molde 62.80 06.44 2003 12 0.083 0.083 Norwegian coast, Malangen 69.71 17.33 2003 18 0.222 0.203 Faeroe Bank 61.10 -08.30 2008 50 0.154 0.143 Faeroe Plateau 61.96 -06.02 2008 49 0.109 0.104 Barents Sea, Båtsfjord 70.65 29.81 2003 10 0.220 0.304 Greenland Nuuk 64.73 -50.45 2003 25 0.419 0.344 Greenland Sisimiut 66.84 -52.89 2003 25 0.573 0.469 Labrador 52.06 -53.39 2004 19 0.534 0.466 Nova Scotia 45.74 -58.41 2003 25 0.456 0.459 Georges bank 42.15 -67.01 2003 25 0.546 0.478 *H*~o~observed heterozygosity across six *Tf1*SNP loci, *H*~e~expected heterozygosity. {ref-type="supplementary-material"}; Figure S2 for sampling locations, and Additional file [3](#S3){ref-type="supplementary-material"}: Table S1 for SNP alleles.](1471-2156-12-51-6){#F6} {#F7} Discussion ========== The identification of four transferrin genes in the Atlantic cod genome adds novel information to our knowledge about the evolution of the transferrin gene family by documenting 1) the expression of tandem duplicated transferrin genes, 2) presence of both bilobal and monolobal transferrin forms in the fish genome, 3) highly conserved synteny between fish monolobal and tetrapod bilobal transferrin loci, and 4) evidence of positive selection acting on a tandem duplicated transferrin gene in trans-Atlantic cod populations. The close linkage of the *Tf1*and *Tf2*genes in the cod genome confirms the suggested tandem duplication of transferrin genes in salmonids and channel catfish based on Southern blotting, pedigree and RFLP analyses \[[@B45]-[@B48]\]. The higher sequence identity between the Tf genes in salmonids (96%) than in cod (84%) indicates separate events of tandem gene duplication of this iron-binding protein, and are supported by the separate TF1/Tf2 bifurcations in salmonids, cod and flounder in the phylogenetic tree. The evolution of bilobal transferrins from an ancestral monolobal form is thought to have occurred in the last common ancestor of vertebrates and arthropods at least 600 MYA \[[@B28],[@B49]-[@B51]\]. However, Williams et al. \[[@B52]\] proposed that the gene duplication event occurred in the prochordates about 500 MYA and correlated with the evolution of a filtration kidney in higher vertebrates that would cause the excretion of monolobal transferrins of about 40 kDa. Tinoco et al. \[[@B24]\] argued that the major advantage for the evolution of bilobal transferrins is the stronger iron affinity because of cooperativity, while monolobal transferrins may have evolved as more general metal ion transporters such as the proposed role of ascidian nicaTf in vanadium trafficking using carbonate as synergistic anion \[[@B53]\]. Contrasting with nicaTf, the iron-binding residues are not conserved in the fish OMP, which seems to be involved in the otolith formation by providing calcium carbonate \[[@B25]\]. The otoliths grow by the continuous deposition of calcium carbonate, and knockdown of zebrafish *Omp*caused a reduction in otolith size \[[@B27]\]. Thus, this monolobal fish transferrin form has apparently acquired a novel function within the otoliths without being excreted by the kidney. The evolution of modern transferrins might be elucidated by searching for conserved synteny between monolobal and bilobal transferrin genes. Similar to the linkage of the different transferrin genes in mammals and chicken \[[@B54]\], the *Ciona nicaTf*and *Tf*-like genes are both located on chromosome 7q \[[@B43]\], but we were not able to find any syntenic segments between these linkage groups. In contrast, the fish *Tf*, *MTf*and *Omp*genes are positioned on three linkage groups, and the *Omp*flanking genes *Senp2*, *Tra2b*and *D2hgdh*are closely linked to chicken *Tf*on chromosome 9, while *Senp2*, *Tra2b*and *Slitrk3*are flanked by human *MTF*and *Tf*on chromosome 3 (Figure [2](#F2){ref-type="fig"}). The conserved synteny between fish monolobal and tetrapod bilobal transferrin loci infers close evolutionary relationship, and we propose that an *Omp*-like gene became duplicated and gave rise to a bilobal form in the common ancestor of the fish and tetrapod lineages. The presence of a single *Omp*gene and the chromosomal separation of the different transferrin genes in the extant fish genome could be explained by two whole genome duplication events, which are believed to have occurred in the ancestral vertebrate \[[@B55]\]. Our speculative model implies that the melanotransferrin and the other transferrins did not evolve from a common bilobal protein, but originated from at least two distinct single gene duplication and fusion events. This is supported by the much higher similarity between the N- and C-lobes of melanotransferrin compared to serum transferrin, although the difference might also be explained by their different functions \[[@B56]\]. The success of competition for the transferrin-bound iron by pathogens depends on the shape and surface charge of the transferrin molecule \[[@B57]\]. Selection for new replacement alleles has been proposed to play a large role in the evolution of transferrin within salmonids, carp and notothenioids \[[@B35],[@B58]-[@B60]\]. We show here that positive selection on the *Tf1*gene in Atlantic cod is significant as evidenced from the high *d*~*N*~/*d*~*S*~ratio between western and eastern Atlantic cod sequences, and suggest that the replacement of surface residues might affect the binding of pathogen transferrin receptors. Another way to test for positive selection is to compare patterns of genetic divergence among populations for the transferrin SNPs with the population structure in neutral genetic markers. Two microsatellite datasets \[[@B61],[@B62]\] and one dataset based on 88 putatively neutral SNPs \[[@B13]\] cover most of the transect from the Baltic along the Norwegian coast across to Greenland, Canada and USA. These neutral markers show several times less divergence compared to the transferrin SNPs (Figure [7](#F7){ref-type="fig"}). Although one should be cautious when comparing genetic markers with different mutational properties such as microsatellites and SNPs, this large difference would suggest positive selection on the transferrin gene, rather than divergence due to geographic isolation. Atlantic cod has been shown in previous studies to be subject to positive selection on several genetic loci in relation to different environmental factors on both local and global scales. Most notable are the observed clines in PanI and hemoglobin in relation to water temperature \[[@B9],[@B11],[@B63],[@B64]\]. Baltic cod is highly divergent in the hemoglobin \[[@B65]\] and heat shock protein *HSP90*genes \[[@B13]\] compared to other populations, and show specific adaptations such as egg buoyancy and sperm motility to the low salinity environment \[[@B66]\]. Intriguingly, additional biological functions of transferrin were suggested by the relation of the carp transferrin polymorphism to sperm motility characteristics \[[@B67]\], and the transferrin polymorphism in tilapia was proposed to be associated with saltwater tolerance \[[@B68]\]. However, transferrin does not seem to be under a particular selection pressure in the Baltic cod population, since SNP allele frequencies in Baltic samples were similar to those collected in the Kattegat, Skagerrak and North Sea. In a recent large scale population genomic study of 1641 SNP loci, Bradbury et al. \[[@B14]\] demonstrated temperature-associated clines in SNP allele frequencies on both sides of the Atlantic, suggesting selection on multiple independent genes in response to ocean temperature. This large-scale study also showed a clear pattern of reduced heterozygosity in eastern cod populations \[[@B14]\]. Whereas a similar cline in heterozygosity across the Atlantic was shown in the present tranferrin study, Nielsen et al. \[[@B13]\] did not find such a pattern for 88 neutral SNP loci. The contrasting patterns of heterozygosity may be due to ascertainment bias, since the SNP loci showing reduced heterozygosity in the eastern Atlantic were developed from Canadian cod \[[@B14]\], whereas the SNPs in Nielsen et al. \[[@B13]\] were developed from Norwegian cod in the east Atlantic. In the present study, the transferrin SNPs were identified by comparing cDNA sequences from Canadian and Norwegian cod specimen, thus excluding ascertainment bias due to different population histories. Instead the difference in heterozygosity between eastern and western Atlantic cod is most likely due to selection. The issue whether selection has acted on the NE or NW cod populations or both might be addressed by comparing the different gadoid Tf sequences. The phylogenetic analysis showed that the Tf1-NE variant had higher similarity than the NW variant to the closest outgroups haddock Tf and cod Tf2. Comparison of the 16 aa changing sites in the cod Tf1 shows that the NE variant is identical to haddock Tf in 8 sites, while only 3 sites in the NW variant are identical to haddock Tf (Table [2](#T2){ref-type="table"}). In addition, the NE variant of Tf1 is identical to Tf2 at 12 sites, whereas the NW variant and Tf2 are identical at only one site of the 16 substitutions. Thus, the most parsimonious explanation is that the NE variant is ancestral to the NW variant of cod Tf1, and that cod in the NW populations has undergone adaptive evolution. If environmental conditions vary temporally, multiple alleles may be selected for, increasing heterozygosity, as has been suggested for immune resistance genes \[[@B69],[@B70]\]. We have not investigated polymorphism in cod Tf2 in the different populations, which might further elucidate the origin and divergence of this iron-binding protein. ###### Comparison of the 16 substituted amino acids in the two cod Tf1 variants with cod Tf2 and haddock Tf SNP tf-6 tf-8 tf-10 tf-11 tf-13 tf-22 ------------- -------- ---------- ---------- --------- --------- ---------- --------- ---------- --------- --------- ---------- --------- ---------- --------- ---------- ---------- ---------- **AA site** **52** **160** **161** **264** **270** **280** **285** **286** **337** **369** **407** **475** **653** **668** **680** **686** **688** Cod TF1-NE **R** [E]{.ul} [A]{.ul} **D** G [R]{.ul} **N** [L]{.ul} T Y [G]{.ul} **T** [Q]{.ul} V [D]{.ul} [T]{.ul} [F]{.ul} Cod TF1-NW K S \- V *E* T *S* F P S Q H L *I* E I S Cod TF2 **R** [E]{.ul} [A]{.ul} **D** R [R]{.ul} **N** [L]{.ul} \- N [G]{.ul} **T** [Q]{.ul} *I* [D]{.ul} [T]{.ul} [F]{.ul} Haddock \- [E]{.ul} [A]{.ul} K *E* [R]{.ul} *S* [L]{.ul} S Q [G]{.ul} I [Q]{.ul} *I* [D]{.ul} [T]{.ul} [F]{.ul} Ala161 is deleted in cod Tf1-NW. Could historical factors have shaped the cod *Tf1*SNP allele frequencies? Paleoecological modelling, as well as nuclear and mitochondrial genetic markers, suggests that cod populations have survived as least for 100 000 years on both sides of the Atlantic \[[@B36],[@B37]\]. Cod populations were more fragmented in the western Atlantic during the last glacial maximum 20 KYA \[[@B36]\], and due to drift and bottlenecks different alleles may have increased in frequency in more or less isolated NW populations. However, the higher substitution rate in nonsynonymous sites, and the fact that multiple alleles are present also in the NE populations, though in low frequencies, argues against the idea that historical expansion in the NW populations alone could explain the observed genetic pattern. Other genetic data suggest that the Greenland populations of Atlantic cod post-dates the last glacial period \[[@B36]\], which possibly could help explain the observed intermediate allele frequencies in Greenland samples. Conclusion ========== The fish transferrin genes provide novel evidence for the evolution of modern transferrins from monolobal transferrins by documenting highly conserved synteny between fish and tetrapods. We propose that the evolution of the iron-binding and iron-independent functions of the different transferrin family members probably involved more than a single event of gene duplication and fusion of monolobal transferrins. Although the specific forces driving evolution of the cod Tf1 are uncertain, the multiple surface residue changes suggest an evolutionary competition for transferrin-bound iron between the host and invading pathogens. We did not find any difference in transferrin allele frequency between the two Faroe populations as previously suggested \[[@B39]\], and also recently based on microsatellites \[[@B61]\]. Neither was the distinctness of Baltic cod previously demonstrated for both neutral markers and coding genes, eg. hemoglobin, supported by the transferrin polymorphisms. Our results underline that functional genetic differences should not be overlooked in fisheries management, but also that markers under selection may give a completely different view of stock structure compared to neutral markers. Methods ======= Identification of cod transferrin genes and alleles --------------------------------------------------- The genomic sequences of Atlantic cod *Tf1*, *Tf2*, *MTf*and *Omp*were found by BLAST search of the reference genome representing the Northeast Arctic population (<http://www.codgenome.no>), and the cDNAs were identified by screening the GAFFA database of the Norwegian coastal cod population. The designated NE variant of cod *Tf1*differed at multiple positions from a published cod *Tf*sequence \[[@B44]\]. This was verified by PCR using templates of liver cDNA and genomic DNA from three Faroe cod specimen, including one heterozygous fish possessing both variants. Primer sequences are available upon request. The liver samples were stored in RNA later (Ambion) before extraction of genomic DNA (Qiagen) and total RNA using Trizol Reagent (Invitrogen Life Technologies). Liver cDNA was synthesized from 1 μg total RNA using the SMART RACE cDNA amplification kit (Clontech Laboratories Inc.) after DNAse treatment (TURBO DNA-free kit, Ambion). The PCR reactions were cycled in a standard thermocycler using the Advantage 2 PCR Enzyme system (Clontech) at standard conditions recommended by the manufacturer. The PCR products were sequenced in both directions with Big Dye sequencing kit (v.3.1) on 3730 ABI DNA Analyser (Applied Biosystems). Gene expression analyses ------------------------ ### RT-PCR Tissue expression of *Tf1*and *Tf2*was semi-quantified by reverse transcription (RT)-PCR using liver and brain from two adult Atlantic cod from the Northeast Arctic population. PCR was run for 35 or 25 cycles with gene specific primers (Additional file [5](#S5){ref-type="supplementary-material"}: Table S3) on liver and brain cDNA templates synthesized as described above. ### Whole mount in situ hybridization (WISH) WISH analysis of *Omp*expression in cod larvae was carried out as described \[[@B71]\]. A 906-bp region of cod *Omp*was PCR amplified using gene specific primers (Additional file [5](#S5){ref-type="supplementary-material"}: Table S3), and sense- and antisense probes, respectively, were synthesized from Sp6- and T7-tailed PCR product and labelled with digoxigenin (Roche, Basel, Switzerland). Twenty newly hatched larvae were fixed for WISH analysis. Generation of 3D protein structures ----------------------------------- All computational experiments were conducted on a Hewlett-Packard xw8600 workstation running Red Hat Enterprise Linux 5. Sequence analysis was performed using BLAST \[[@B72]\] through the Protein Data Bank (BLOSUM62 matrix). Human serum transferrin (Protein Data Bank ID [2HAU](2HAU)) \[[@B73]\] was selected as the most appropriate template to generate comparative models and showed sequence identities with cod Tf1-NE, Tf1-NW, Tf2, MTf and OMP of 45%, 45%, 44%, 40% and 35%, respectively. The template structure was then aligned against the cod sequences using ClustalW program \[[@B74]\]. Each sequence alignment was then checked to ensure that (i) all the secondary structural elements had a minimum number of insertions or deletions within them, and (ii) Cys residues forming consensus disulfide bridges were conserved \[[@B75]\]. Once an acceptable alignment had been produced, an ensemble of 50 models of the five different cod proteins were built using MODELLER v. 9.7 \[[@B76]\] as implemented in Discovery Studio (Accelrys Inc., San Diego, CA, USA) and ranked using the MODELLER objective function, which is highly efficient in ranking different models calculated from the same alignment. It should be noted that the disulfide bridges between conserved Cys were not included as a restraint in the modeling process, but that the relative position of two Cys side chains in the resulting (unrestrained) model led us to suggest the existence of a disulfide. The stereochemical quality of the structures was assessed by PROCHECK \[[@B77]\] supplemented by the profile programs VERIFY3D \[[@B78]\] and ProSA-web \[[@B79]\]. In order to assess the reliability of each model, the corresponding energy graphs were compared with the template [2HAU](2HAU) structure. Phylogenetic analyses --------------------- Database searches were used to identify in total 40 amino acid sequences from teleost species, *Ciona intestinalis*and man (Additional file [6](#S6){ref-type="supplementary-material"}: Table S4). The C-terminal domain of Tf and MTf sequences was aligned with OMP sequences using ClustalX \[[@B80]\] and by manual editing. The final alignment consisted of 40 taxa and 330 characters. The best evolution model based on the sequence alignment was determined using ProtTest \[[@B81]\]. The sequences were used to infer the phylogeny in a Bayesian framework applying the program MrBayes v3.1.2 \[[@B82]\]. The Bayesian inferences were done as follows: two independent runs, each with three cold and one heated MCMC (Markov Chain Monte Carlo) chains were started from a random starting tree. The two runs lasted for 5,000,000 generations. The covarion (COV) model was used together with the WAG+G+I to accommodate for different substitution rates across sites (G + proportion of invariable sites (I)) and across sequences (COV). The maximum likelihood (ML) tree was estimated using the program RAxML v.6 \[[@B83]\]. The topology with the highest likelihood score out of 100 heuristic searches, each from a random starting tree, was selected, and bootstrapping was done with 100 pseudoreplicates and one heuristic search per replicate. In the ML analyses, the WAG model with a gamma-distributed rate of variation across sites (G) was employed. All phylogenetic analyses were done on the freely available Bioportal at University of Oslo <http://www.bioportal.uio.no>. The Nei-Gojobori method \[[@B84]\] implemented in DNAsp 5.0 was used to calculate rates of synonymous and non-synonymous substitutions in the identified Norwegian *Tf1*cDNA sequence compared to the Canadian sequence \[[@B44]\]. We then used the *z*-test in MEGA (4.0) to test for positive selection. Population sampling and SNP genotyping -------------------------------------- In total 375 adult individuals were collected during 2002-2008 from 14 localities covering the geographical region of the North Atlantic inhabited by Atlantic cod (Table [1](#T1){ref-type="table"}, Additional file [2](#S2){ref-type="supplementary-material"}; Figure S2). Genomic DNA was extracted from fin clips, muscle tissue or gill arches, and six SNPs loci (Additional File [1](#S1){ref-type="supplementary-material"}: Figure S1) spanning the *Tf1*gene were genotyped for all individuals using the MassARRAY system from Sequenom (San Diego, USA). PCR primers and extension primers were designed using the software SpectroDESIGNER v3.0 (Sequenom) (Additional file [5](#S5){ref-type="supplementary-material"}: Table S3). Genomic DNA was PCR amplified as described \[[@B85]\], and the SNP genotyping was performed according to the iPLEX protocol from Sequenom available at <http://www.sequenom.com/iplex>. For allele separations the Sequenom MassARRAY Analyzer (Autoflex mass spectrometer) was used. Genotypes were assigned in real time \[[@B86]\] using the MassARRAY SpectroTYPER RT v3.4 software (Sequenom) based on the mass peaks present. All results were manually inspected using the MassARRAY Typer Analyzer v3.3 software (Sequenom). Statistical analyses of population genetic data ----------------------------------------------- SNP heterozygosities and population differentiation (*F*~ST~) were calculated using Genepop 4.0.10 \[[@B87]\]. Statistical significance of population differentiation was assessed using Fishers exact test implemented in Genepop. List of abbreviations ===================== Tf: Transferrin; OMP: otolith matrix protein; Mtf: melanotransferrin; RMSD: root mean square deviation; WISH: whole mount in situ hybridization; RT-PCR: reverse transcription polymerase chain reaction; UTR: untranslated region; ORF: open reading frame. Authors\' contributions ======================= ØA and CA conceived and designed the study, and wrote the manuscript. MCDR and DP performed the computer modeling study and the generation of 3D structures. ATK carried out the phylogenetic analyses. PEP was responsible for the Faroe cod samples. CA performed the statistical analyses. All authors critically read the manuscript drafts and approved the final version of the manuscript. Supplementary Material ====================== ###### Additional file 1 **Figure S1 Alignment of the NE and NW variants of Atlantic cod *Tf1*cDNAs**. The Atlantic cod Tf1 cDNA derived from two Northeast Atlantic populations was compared with that of a Northwest Atlantic population \[[@B44]\]. The 22 SNPs identified are numbered, and the six analysed SNPs are shown in bold. ###### Click here for file ###### Additional file 2 **Figure S2 Map of the Atlantic cod populations examined**. Six out of 22 SNPs identified in cod Tf1 were analysed in 14 populations across the North-Atlantic. ###### Click here for file ###### Additional file 3 **Table S1 Genotype frequencies for six Atlantic cod Tf1 SNP loci in 14 samples across the North Atlantic**. Sample sizes of cod collected at each location in the first row. ###### Click here for file ###### Additional file 4 **Table S2 Genetic differentiation among pairs of Atlantic cod samples**. Below diagonal are pairwise *F*~ST~-values, and above diagonal statistical significance values \**P*\< 0.05; \*\**P*\< 0.01; \*\*\**P*\< 0.001. ###### Click here for file ###### Additional file 5 **Table S3 PCR primers for SNP analysis, RT-PCR and WISH**. The sequences are shown in 5\'-3\' direction. ###### Click here for file ###### Additional file 6 **Table S4 Accesion numbers of the proteins included in the phylogenetic analysis**. Several distinct transferrin genes have been described in some teleost species. ###### Click here for file Acknowledgements ================ We thank Lou van Eeckhaute, Marie Storr-Paulsen, John Brattey, Henrik Svedäng, Yvonne Walther, Halvor Knutsen, Per-Erik Jorde and Svein-Erik Fevolden for providing cod samples. Hege Munk, Arne Roseth, Paul Berg, Hanne Johnsen and Jacob Torgersen are acknowledged for excellent technical assistance. Financial support was provided by the Norwegian Research Council and Swedish Research Councils Formas and VR. Parts of the analyses were conducted within the framework of the BaltGene project funded by BONUS Baltic Organisations\' Network for Funding Science EEIG, and the Linnaeus Centre for Marine Evolutionary Biology at University of Gothenburg (<http://www.cemeb.science.gu.se>).

Introduction ============ Mussels of the genus *Mytilus* are widespread in both the Northern and Southern Hemispheres and are important components of coastal ecosystems ([@ref-79]; [@ref-68]; [@ref-51]). Mussels are a bivalve species of significant ecological and commercial value worldwide with annual production of 1.2M tonnes ([@ref-24]; [@ref-37]; [@ref-80]). Due to their filter-feeding abilities and sessility, they also play an important role as biosensors for pollution and other environmental changes (including ocean acidification) in coastal areas ([@ref-11]; [@ref-17]). Rising levels of atmospheric CO~2~ affects the ability of molluscs to incorporate calcium carbonate, which is essential for building and maintaining shells ([@ref-75]). In molluscs biocalcitic structures fulfil a wide range of functions such as protecting the inner soft body from predators and environmental damage, detoxification, and attachment of muscles ([@ref-52]). The shell is an exoskeleton made up of a mineral phase (calcium carbonate, CaCO~3~ ∼95%) and organic macromolecules: proteins, glycoproteins, polysaccharides and lipids ([@ref-52]). Despite being a minor component (\<5%), proteins in the shell, referred to as shell matrix proteins (SMPs), may play a key role in biomineralization including size regulation, nucleation, morphology and organization of the growing of CaCO~3~ crystals ([@ref-56]). Shells of bivalves differ in color, shape and marking, according to environmental conditions, for example the shell of the *Mytilus edulis* Linnaeus, 1758 in the intertidal zone is blue--black and heavy, while in the sublittoral region, where mussels are continuously immersed, it is thin and brown ([@ref-24]). Shell morphology and thickness of various mussels from the genus *Mytilus* have also been associated with variability of environmental conditions, such as sediment types, trophic conditions, wave impact ([@ref-69]; [@ref-60]; [@ref-71]). The mussel shell is formed by the heavy fold mantle tissue, which encloses the internal organs of the animal ([@ref-24]). The mantle is a complex structure underneath the shell, which consists of connective tissue with hemolymph, neural tissue, muscles and gonads ([@ref-24]). Moreover, it has a sensory function and plays a role in directing particles onto the gills and in deflecting heavier material, storage of nutrient reserves, in bioaccumulation of metals and other compounds and in shell repair and growth ([@ref-24]). To identify candidate biomineralization genes, over a dozen mantle transcriptomes have been assembled for mussels ([@ref-18]; [@ref-5]; [@ref-78]), clams ([@ref-9]; [@ref-3]; [@ref-66]), scallops ([@ref-65]; [@ref-70]) and oysters ([@ref-14]; [@ref-46]). However, there remains a lack of mantle tissue transcriptome from *M. trossulus* Gould, 1850 and *M. chilensis* Hupé, 1854 as well as candidate biomineralization transcripts, and their comparative analysis within the genus *Mytilus*. In this study, the mantle transcriptomes of the four mussel taxa: *M. edulis*, *M. galloprovincialis* Lamarck, 1819---the Mediterranean Sea and Tasmania, *M. trossulus*, *M. chilensis*, were assembled and characterized using a Roche GS-FLX 454 pyrosequencing system. The aim of the present study was twofold: (1) to identify transcripts putatively involved in biomineralization and melanogenesis, and (2) to perform a comparative transcriptome analysis among different taxa of the genus *Mytilus* based on mantle tissue. A large number of proteins putatively involved in biomineralization and pigmentation processes were identified in different *Mytilus* taxa. This study is the first to present the mantle tissue transcriptome in *M. trossulus* and *M. chilensis*. The generated data (accession number [PRJNA419475](PRJNA419475)) add to the growing sequence database for mussels and provide the first comparison of the mantle tissue using mussel transcriptomes from different taxa. Materials and Methods ===================== RNA extraction and pyrosequencing --------------------------------- Five populations of mussels (*Mytilus* spp.) representing four taxa were collected from several localities: *M. edulis* (North Sea, EduN), *M. trossulus* (Vancouver, TroV), *M. galloprovincialis* (Mediterranean Sea, GalM), *M. galloprovincialis* (Tasmania, GalT), *M. chilensis* (Chile, Chil) ([Table 1](#table-1){ref-type="table"}). All taxa had been identified by single nucleotide polymorphism (SNP) genotyping ([@ref-79]; [@ref-73]; [@ref-40]). Specimens chosen for RNA extraction were adults of indeterminate sex, spent. Tissue sampling was standardized across individuals: extraction of total RNA from the central and outer mantle tissue from right and left mussel side was performed using a GenElute Mammalian Total RNA Miniprep Kit (Sigma-Aldrich, St. Louis, MO, USA) according to the manufacturer's protocol, with minor modifications. Homogenization was carried out in a custom buffer composed of 100 mM Tris, 1.4M NaCl, 20 mM EDTA, 2% CTAB, proteinase K (20 mg/ml) and 0.03 mM 2-mercaptoethanol ([@ref-54]). Samples were stored at −20 °C. RNA samples from two to four individuals were pooled for each group in equal proportions to make a total of five libraries. The quantity and quality of total RNA was determined at 260 nm on microplate using an Epoch Microplate Spectrophotometer (BioTek Instruments, Incorporated, Winooski, VT, USA) and gel electrophoresis. RNA integrity was checked using an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA). cDNA libraries were prepared using a cDNA Rapid Library Preparation Kit (Roche Applied Science, Basel, Switzerland) and pyrosequencing was performed by Macrogen Incorporation (Seoul, Korea), using Roche GS-FLX sequencing system (Roche, Basel, Switzerland) according to manufacturer's instructions. 10.7717/peerj.6245/table-1 ###### Information about mussel (*Mytilus* spp.) samples included in the present study.  Group/library name Species Location Year of sampling *N* -------------------- ------------------------ --------------------------------------------------------------- ------------------ ----- EduN *M. edulis* The North Sea: The Oosterschelde estuary 2006 2 TroV *M. trossulus* Canada: Vancouver 2012 2 GalM *M. galloprovincialis* The Mediterranean Sea: Trieste and Chioggia and the Black Sea 2012 4 GalT *M. galloprovincialis* Australia: Tasmania, Spring Bay 2012 2 Chil *M. chilensis* Chile: Punta Arenas and Concepción 2012 2 **Note:** *N*, number of specimens. De novo assembly, annotation, SSR and SNP discovery --------------------------------------------------- All reads from the mantle tissue of five *Mytilus* groups were processed separately, through the same pipeline. Raw reads were pre-processed by removing adapters and low quality sequences (quality score = 0.05, discard reads \<50 bp) with CLC Genomics Workbench software (v.7.5.5, CLC Bio, Qiagen, Aarhus, Denmark). Quality reads were assembled into contigs using de-Bruijn graphs in CLC Genomics Workbench with default parameters ([@ref-53]; [@ref-54]). Contigs were first searched against proteins from the NCBI non-redundant (NR) database using a BLASTX tool implemented in BLAST+ (v.2.2.29) ([@ref-2]), with an *E*-value threshold of 10^−5^ and other default parameters. A list of the top BLAST hits was achieved by using bit score (12th column) and *E*-value values (11th column) with the code: sort -k1,1 -k12,12 nr -k11,11 g -k3,3 gr blastout.txt sort -u -k1,1--merge \> besthit. For functional annotation, gene ontology (GO) terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways were assigned to the transcripts using Blast2GO software ([@ref-10]) with *E*-value Hit-Filter = 1.0 *E*^−10^ (other default parameters) and the single directional best-hit method in the KEGG Automatic Annotation Server ([@ref-59]), respectively. The software, MSDB ([@ref-12]) was used to discover potential microsatellite markers. Di- to hexa- nucleotide motifs, with minimum thresholds of 6-5-4-4-4 repeats, respectively and 50 bp of flanking sequence length were identified. SNP analyses were carried out using CLC Genomics Workbench. For each transcriptome, clean reads were mapped back to corresponding contigs with an overlap criterion of 70% and a similarity of 90%. Candidate variants were called with the following minimum value parameters: neighborhood quality = 15; quality of central bases = 20; coverage = 20; count = 8; variant frequency = 35.0. Identification of potential biomineralization and melanogenesis-related contigs ------------------------------------------------------------------------------- Candidate contigs related to biomineralization and melanogenesis were identified by searching top BLASTX hits with key words from a reference list of previously reported candidate biomineralization genes ([Table S1](#supp-10){ref-type="supplementary-material"}). Melanogenesis-related transcripts were identified based on KEGG pathways analysis. Several biomineralization proteins were selected for additional analysis: carbonic anhydrase (CA), chitinase (CHIT), tyrosinase (TYR), dermatopontin (DPT), arginine kinase (AK) and glyceraldehyde-3-phosphate dehydrogenase (GAPDH). Proteins of selected sequences were predicted using OrfPredictor ([@ref-57]). Functional domains and motifs were identified using Simple Modular Architecture Research Tool (SMART) software, allowing identification and annotation of domain architectures (<http://smart.embl-heidelberg.de>) ([@ref-43]). To further scrutinize and compare selected contigs, representative reference bivalve databases were created. Protein sequences described as CA, CHIT, DPT, AK and GAPDH, available at the GenBank database ([www.ncbi.nlm.nih.gov](www.ncbi.nlm.nih.gov)) were downloaded (217, 155, 8, 43, 24 sequences, respectively). Only sequences with annotated domains, were selected (199, 123, 8, 43, 24 sequences, respectively). Following this, mussel contigs from the present study were searched against the created databases with an *E*-value threshold of 10^−15^. For phylogenetic reconstruction of *Mytilus* spp. TYR proteins, contigs from this study, amino acid sequences from GenBank (Acc. No: [AKS48166](AKS48166), [ALA16023](ALA16023), [OPL3388](OPL3388)), and sequences from previous studies ([@ref-1]; [@ref-58]; [@ref-30]; [@ref-66]) were used. Homologous sequences were aligned using multiple alignment program MUSCLE ([@ref-13]) with default parameters, and trimmed using TrimAl ([@ref-8]) with an 80% threshold. Best substitution models were inferred using ModelFinder ([@ref-35]), as follow: CA---the Le Gascuel (LG) model ([@ref-41]), gamma distribution, four discrete rate categories (+G4); GH18---the Whelan and Goldman (WAG) model ([@ref-74]) + G4 with invariant sites (+I); AK and GAPDH---WAG+ G4; DPT---the Block substitution matrices (Blosum62) model ([@ref-26]) + G4, with frequencies (+F); TYR---Blosum62+F+I+G4 model. Three phylogenetic models were run for each alignment. Neighbor-joining (NJ) trees were performed using MEGA X ([@ref-38]) and the Jones--Taylor--Thornton (JTT) substitution model ([@ref-34]). Maximum likelihood reconstructions were performed using IQ-TREE ([@ref-61]) with 1,000 bootstrap replicates. Bayesian inferences were constructed using MrBayes v.3.2 ([@ref-64]) with parameters: ngen = 1,500,000; samplefreq = 100; printfreq = 100; nchain = 4; starttree = random. Trees were visualized and edited using MEGA X and manually combined to produce a consensus tree for each protein alignment. Comparative transcriptomic analysis ----------------------------------- Homologous transcripts present in all five groups were identified by using bi-directional tBLASTx analyses conducted between each transcriptome with an *E*-value cut-off of 10^−15^. A list of top BLAST hits was achieved by using bit score (12th column) and *E*-value values (11th column) with the code: sort -k1, 1 -k12, 12 nr -k11, 11 g -k3,3 gr blastout.txt \| sort -u -k1,1--merge\>besthit. Further, results from this study were compared with available *Mytilus* spp. mantle sequences. The accession numbers are: for *M. edulis*: [PRJEB4516](PRJEB4516) ([@ref-18]), [PRJNA30561](PRJNA30561) ([@ref-78]); for *M. galloprovincialis*: [PRJNA230138](PRJNA230138) ([@ref-58]), [PRJNA295512](PRJNA295512) ([@ref-5]). For this purpose, a de novo transcriptome assembly was performed using the raw reads deposited in the NCBI Sequence Read Archive (SRA). The Illumina HiSeq raw reads were filtered using Trimmomatic v.0.32 ([@ref-7]) and retained reads were assembled using Trinity program with default parameters ([@ref-25]). To find enriched GO terms between groups, Fisher's exact test was performed with Blast2GO software, applying a one-tailed test, removing double IDs, with a false discovery rate (FDR) cut-off of 0.05. As a test-set, list of contigs from each group was used. As a background (reference-set), a fusion of the five transcriptomes was used. To identify putative orthologs among the genus of *Mytilus*, transcripts from this study and from *M. coruscus* ([@ref-77]), *M. californianus* ([@ref-63]) were analyzed using OrthoMCL software with default settings ([@ref-45]). The putative orthologous clusters were aligned with a MUSCLE tool and then concatenated in CLC Genomics Workbench. A splits network was created with SplitsTree 4.13.1 ([@ref-31]), using the neighbor-net algorithm under pdistance model. A phylogenetic tree was inferred using a NJ model in MEGA X, with the JTT + G4 substitution model and 1,000 bootstrap replicates. Results ======= Assembly, annotation, SSR and SNP detection ------------------------------------------- A total of 753,135 raw reads with an average read length of 342 bp were generated from sequencing of the *Mytilus* spp. mantle transcriptomes. After filtering, 743,773 (98.75%) high-quality reads were used for de novo assembly. In total, 20,982 contigs with an average length of 529 bp were obtained ([Table S2](#supp-11){ref-type="supplementary-material"}). Mussel raw sequences from this study were deposited in NCBI SRA under accession number [PRJNA419475](PRJNA419475). BLAST analysis against NCBI NR database showed that between 43.19% (TroV) and 52.18% (GalM) of transcripts had a significant hit ([Tables S2](#supp-11){ref-type="supplementary-material"} and [S3](#supp-12){ref-type="supplementary-material"}). As expected, most of the annotated transcripts were homologous to proteins from the Mollusca phylum: Pacific oyster (*C. gigas*, av. 50%), *Mytilus* spp. (av. 10%) and owl limpet (*L. gigantea*, av. 5%) ([Fig. S1](#supp-1){ref-type="supplementary-material"}). KEGG analysis showed that a total of 3,679 (17.53%) transcripts were assigned to 271 biochemical pathways. The number of sequences assigned to each pathway was different across groups ([Table S2](#supp-11){ref-type="supplementary-material"}). However, the most represented pathways in all groups included focal adhesion, phagosome, ribosome and apoptosis ([Table S4](#supp-13){ref-type="supplementary-material"}). The signal transduction, endocrine system, immune system and cellular community were the most represented KO (KEGG orthology) subcategories in all groups ([Fig. S2](#supp-2){ref-type="supplementary-material"}). In total, 35,383 GO terms were assigned to 7,824 (37.28%) contigs ([Table S2](#supp-11){ref-type="supplementary-material"}). The main GO category distribution was similar in all *Mytilus* spp. groups, the dominant was the biological process category (BP), followed by cellular component and molecular function (MF) ([Fig. S3](#supp-3){ref-type="supplementary-material"}). In this study, 483 potential simple sequence repeats (SSRs) were identified in 645 contigs (3.07% of the total number of contigs) ([Table S5A](#supp-14){ref-type="supplementary-material"}). Among the identified putative markers, the most highly represented in the EduN, TroV, GalM and GalT groups were di-nucleotide repeats, and tri-nucleotide in Chil. AT was the dominating motif in all groups ([Table S5B](#supp-14){ref-type="supplementary-material"}). In addition, a total of 1,497 putative SNPs were detected in all groups. The A/G variation type was dominant in EduN and TroV groups, whereas in other groups the C/T was dominant ([Fig. S4](#supp-4){ref-type="supplementary-material"}). Contigs potentially involved in biomineralization and melanogenesis ------------------------------------------------------------------- BLASTx searching revealed 1,292 contigs with significant sequence similarity to candidate biomineralization proteins ([Table S6](#supp-15){ref-type="supplementary-material"}). SMPs such as, nacrein, calmodulin, collagens, perlucin, perlustrin, PIF were found among identified contigs (see [Table S6](#supp-15){ref-type="supplementary-material"}). Transcripts involved in ion and acid-base regulation, protein homeostasis and anti-oxidation were also found including V-type proton ATPases, sodium/potassium-transporting ATPases, heat shock 70 kDa protein (HSP70) and glutathione peroxidase. Other annotations included C1q-domain-containing protein, cAMP-response element-binding protein and cold-shock protein (CSP) ([Table S6](#supp-15){ref-type="supplementary-material"}). Of annotated biomineralization contigs, 68.65% had domains, including chitin binding, EF hand, sushi, Von Willebrand factor A ([Table S6](#supp-15){ref-type="supplementary-material"}). Several biomineralization proteins were selected for more comprehensive analyses. In *M. galloprovincialis*, *M. edulis* and *M. trossulus* groups, five contigs (GalT_1066, GalT_1436, EduN_2761, GalM_2384, TroV_1436) were identified as DPT with DERM domain. A phylogenetic analysis of DPT proteins in Bivalvia generated a separate, well-supported clade of Mytiloida members, suggesting that these contigs are *Mytilus* spp. orthologues of DPT ([Fig. 1A](#fig-1){ref-type="fig"}). Contigs with 81--95% sequence identity to AK were identified in all groups. The phylogenetic analysis showed two separate clades of Pterimorphia and Heteroconchia subclasses ([Fig. 1B](#fig-1){ref-type="fig"}). Contigs Chil_958, EduN_492, GalM_470 and TroV_123 encoded proteins with similarity to GAPDH and possessed the Gp_dh_C domain. Alignment with other known GAPDHs from Ostreoida, Pectinoida and Pterioida members revealed many conserved regions among these species ([Fig. S5](#supp-5){ref-type="supplementary-material"}). However, a phylogenetic analysis showed two well-supported separate clades of Mytiloida and Ostreoida with Pectinoida orders ([Fig. 1C](#fig-1){ref-type="fig"}). TYR amino acid sequences derived were aligned with available molluscan TYR sequences. A phylogenetic analysis revealed that mussel TYR, in this study, created two separate clades: clade A and clade B ([Fig. 2](#fig-2){ref-type="fig"}), described earlier as ancestral and bivalve-specific ([@ref-1]). It is possible that there were two small expansions within *Mytilus*, which is a common feature in bivalves. In *M. galloprovincialis* there were TyrA2, TyrA3 orthologues shared with the genera *Crassostrea* and *Pinctada*, indicating gene duplication before the divergence of lineages A and B. Searches in mussel transcriptomes revealed five contigs (in EduN, TroV and GalT groups) that pertain to the CA gene family and 10 contigs (in all transcriptomes) that showed similarity to the glycoside hydrolase family 18 (GH18) CHIT members ([Table 2](#table-2){ref-type="table"}). A SMART analysis revealed CA domains in annotated CA sequences, and GH18_CHIT-like superfamily domain in GH18 sequences. The chitin binding peritrophin-A domain (CBM_14 superfamily) was found in transcripts recognized as acidic mammalian CHIT (GalT_650, GalT_649 and TroV_158). The phylogenetic analysis of CA in the Bivalvia class revealed three major clusters: α-carbonic anhydrases related proteins (α-CARPs), cytosolic/extracellular α-carbonic anhydrases (α-CAs) and membrane-bound α-CAs. There were two visible sub-clusters: CA2 and CA14 ([Fig. 3A](#fig-3){ref-type="fig"}). Two major clades, A and B were resolved in the GH18 CHIT family phylogenetic tree. The A cluster consisted of CHIT domain-containing proteins (CHIDs, with GH18_SI-CLP domain), and Di-N-acetylchitobiases. In contrast to A, cluster B contained genes involved in chitin degradation (CHIT and CHIT-like proteins), creating three sub-clades (B1, B2, B3) ([Fig. 3B](#fig-3){ref-type="fig"}). Presented phylogenetic reconstructions were supported by the high values of the nodes. {#fig-1} ![Molecular phylogenetic analysis of tyrosinase (TYR) protein sequences in Bivalvia class.\ A consensus tree based on maximum likelihood (ML) topology. The numbers of the nodes show the bootstrap values (BV) \>50% and Bayesian posterior probabilities (BPP) \>0.50, as follows: ML BV/NJ BV/BIs BPP. A black dot at the node represents BV \>90% and BPP \>0.90. Bivalve TyrA ortholog groups are annotated as A1--A3, *Pinctada* spp. TyrB ortholog groups are annotated as B1--B5. Color code: green and yellow---two possible *Mytilus* expansions; orange---*Crassostrea*, *Mya*, *Pinctada* expansions, described in previous studies ([@ref-1]; [@ref-66]).](peerj-07-6245-g002){#fig-2} 10.7717/peerj.6245/table-2 ###### α-CA and GH18 proteins identified in mantle transcriptomes, according to BLASTX searches.  Contig Size (nt) Acc. number (NCBI) Homology in NCBI (NR) database ----------- ----------- ---------------------- ------------------------------------------------------ ---------------------------------- ---------- EduN_1982 697 [MG827120](MG827120) *M. coruscus* carbonic anhydrase-like protein [AKS48148.1](AKS48148.1) 1*E*-46 EduN_2638 1,264 [MG827121](MG827121) *C. virginica* carbonic anhydrase 14-like isoform X3 [XP_022339697.1](XP_022339697.1) 8*E*-81 EduN_4714 404 [MG827122](MG827122) *M. galloprovincialis* carbonic anhydrase II [ALF62133.1](ALF62133.1) 3*E*-65 TroV_766 1,475 [MG827123](MG827123) *C. gigas* carbonic anhydrase 14-like isoform X2 [XP_011435377.1](XP_011435377.1) 2*E*-59 GalT_2236 347 [MG827124](MG827124) *C. gigas* carbonic anhydrase 2-like [XP_011429671.1](XP_011429671.1) 5*E*-40 EduN_7774 362 [MG827131](MG827131) *M. yessoensis* chitinase 3 [OWF51745.1](OWF51745.1) 7*E*-37 EduN_2068 1,194 [MG827125](MG827125) *M. yessoensis* Acidic mammalian chitinase [OWF47140.1](OWF47140.1) 1*E*-52 TroV_279 1,678 [MG827132](MG827132) *M. yessoensis* Acidic mammalian chitinase [OWF47140.1](OWF47140.1) 4*E*-62 TroV_158 1,647 [MG827133](MG827133) *M. coruscus* mytchitin 5 [AIF74559.1](AIF74559.1) 0.00 GalM_153 195 [MG827126](MG827126) *M. coruscus* mytchitin 4 [AIF74558.1](AIF74558.1) 2*E*-25 Chil_296 1,026 [MG827134](MG827134) *M. coruscus* chitinase-like protein-1 [AKS48165.1](AKS48165.1) 8*E*-157 Chil_188 500 [MG827127](MG827127) *M. galloprovincialis* chitinase-like protein-3 [AKS48199.1](AKS48199.1) 9*E*-88 GalT_650 2,725 [MG827128](MG827128) *M. yessoensis* Acidic mammalian chitinase [OWF47140.1](OWF47140.1) 3*E*-122 GalT_649 705 [MG827129](MG827129) *M. yessoensis* Acidic mammalian chitinase [OWF47140.1](OWF47140.1) 1*E*-36 GalT_1838 847 [MG827130](MG827130) *M. yessoensis* probable chitinase 10 [XP_021358847.1](XP_021358847.1) 2*E*-33 {#fig-3} A total of 42, 99, 19 and 62 contigs from the mussel mantle transcriptomes were assigned to 13, 40, 8 and 15 KO terms in melanogenesis, endocytosis, TYR and calcium signaling KEGG pathways, respectively ([Table S7](#supp-16){ref-type="supplementary-material"}). Comparative transcriptomics --------------------------- Bi-directional Best Hit (BBH) method was used to identify 6,130 homologous pairs between transcriptomes of which 552 were shared between all five groups ([Fig. S6](#supp-6){ref-type="supplementary-material"}). The highest number of homologs were found between pairs: EduN_TroV, EduN_GalT and EduN_GalM (1,029, 946, 818, respectively). Putative homologous pairs were ordered according to sequence similarity of BLAST alignments ([Fig. S6](#supp-6){ref-type="supplementary-material"}). In general most homologous pairs exhibited a sequence similarity of 90--100%. The highest similarity per no. of homologs was found between EduN_GalM (85.45%), followed by EduN_Chil (84.89%), GalM_Chil (84.55%), GalM_GalT (84.12%). The lowest similarity to other groups was shown by TroV. Transcriptomes obtained from this study were compared to available mantle sequences for *M. edulis* and *M. galloprovincialis* using tBLASTx. In total, 18,456 (87.96%) contigs from this study were homologous with one of the four created mantle transcriptomes, but 12.04% might represent new sequence information. Additionally, between 47.8% and 81.5% of homologous contigs had high similarity (\>90%; [Fig. S7](#supp-7){ref-type="supplementary-material"}). For more accurate analysis, Fisher's exact test was applied to discover significantly over-represented GO terms in each group. The most interesting GO enrichments associated with BP and MF were identified only in EduN and GalM groups. The results were reduced to the most specific terms (FDR = 0.01) ([Figs. S8](#supp-8){ref-type="supplementary-material"} and [S9](#supp-9){ref-type="supplementary-material"}). A total of 25 GO terms were enriched in the *M. edulis* from the North Sea including cellular response to chemical stimulus (<GO:0070887>), positive regulation of cellular process (<GO:0048522>) in the BP category, ATP binding (<GO:0005524>), calcium ion binding (<GO:0005509>) in the MF category ([Fig. S8](#supp-8){ref-type="supplementary-material"}). A total of 46 terms were enriched in the *M. galloprovincialis* from the Mediterranean Sea such as organonitrogen compound biosynthetic process (<GO:1901566>), regulation of biological quality (<GO:0065008>) in the BP category, metal ion binding (<GO:0046872>), purine ribonucleotide binding (<GO:0032555>) in the MF category ([Fig. S9](#supp-9){ref-type="supplementary-material"}). Both the neighbor-net splits network ([Fig. 4A](#fig-4){ref-type="fig"}) and NJ phylogenetic tree ([Fig. 4B](#fig-4){ref-type="fig"}) showed that *M. chilensis* was the most similar to the *M. edulis* taxon, followed by *M. galloprovincialis* and *M. trossulus*. A separate clade was identified containing *M. californianus* Conrad, 1837 and *M. coruscus* Gould, 1861 (presently accepted as *M. unguiculatus* Valenciennes, 1858) mussels ([Fig. 4](#fig-4){ref-type="fig"}). {#fig-4} Discussion ========== The mollusc shell, formed by calcium carbonate deposition (secreted by the mantle tissue), forms the first barrier against marine predators, mechanical or toxic damage and protects the soft body. This calcified structure is sensitive to environmental conditions: CO~2~ level, temperature, humidity. Moreover, mussels have adapted to different environments that affect the shell formation process and the stability of the secreted composite biomineral ([@ref-55]). The sequencing of mantle transcriptome from multiple taxa of *Mytilus* spp. has added to the growing transcriptome resources for mussels and has provided contigs putatively involved in biomineralization. High-quality reads were assembled totaling 20,982 contigs, 9,773 (46.57%) of them were annotated by BLAST searches against NCBI NR protein database ([Tables S2](#supp-11){ref-type="supplementary-material"} and [S3](#supp-12){ref-type="supplementary-material"}). The top hits showed homology with other bivalves and most annotations were extracted from *C. gigas* Thunberg, 1793 ([Fig. S1](#supp-1){ref-type="supplementary-material"}), as expected due to the availability of an oyster fully sequenced genome. Transcriptome-based markers are a valuable resource for determining functional genetic variation ([@ref-54]). In this study, 483 potential SSRs and 1,497 putative mantle-related SNPs were identified. SSRs ([@ref-76]) and SNPs ([@ref-75]) have previously been developed in mussels using the transcriptome approach, however, this study aimed to identify molecular markers in mantle tissue from several *Mytilus* taxa. These results might be a resource for molecular marker studies associating candidate biomineralization genes with different environmental conditions and/or taxa. Both GO and KEGG analyses showed a wide array of categories and pathways potentially involved in biomineralization and pigmentation mechanisms ([Tables S4](#supp-13){ref-type="supplementary-material"} and [S7](#supp-16){ref-type="supplementary-material"}). In all groups, cellular processes: focal adhesion, tight junction, adherens junction and signaling pathways: insulin, calcium, chemokine, ErbR, MAPK and VEGF were identified ([Table S4](#supp-13){ref-type="supplementary-material"}). The importance of these pathways in shell formation has been proposed by [@ref-72] and [@ref-50]. Tyrosine metabolism and melanogenesis pathways detected in the mantle transcriptomes presented here might play a fundamental role in the shell pigmentation process. Numerous genes in those pathways were identified such as calmodulin, protein kinase, TYR, β-catenin ([Table S7](#supp-16){ref-type="supplementary-material"}). Moreover, it seems that the Wnt signaling pathway regulation of the transcription factor essential for melanocyte development (MITF) might play a crucial role in melanogenesis (melanocyte development) in bivalves. Another important pathway may be endocytosis, a mechanism for cells to remove nutrients, ligands and to influence pigment granule formation ([@ref-16]). [@ref-16] further suggested, that, endocytosis and genes VPS or Rab might have an effect on shell coloration in *Crassostrea gigas* (Thunberg, 1793). These results are in line with previous studies in scallop ([@ref-70]) and oysters ([@ref-46]). A total of 1,292 contigs putatively involved in the biomineralization mechanism were also identified. Although most of these transcripts have been reported in *M. edulis* ([@ref-30]) and *M. galloprovincialis* ([@ref-20]; [@ref-58]) mantle transcriptomes, new transcripts were identified for *M. trossulus* and *M. chilensis*. This data may contribute to future comparison between different localities of the same taxon, to assess the influence of environment on the shell formation. All presented contigs should be considered as candidate biomineralization proteins since their functions were hypothesized mainly based on expression data, but were not validated ([@ref-27]). In mussels, the shell is composed of two polymorph forms of calcium carbonate (calcite and aragonite) ([@ref-48]). Some matrix proteins, potentially related to the formation of calcite (e.g., TYR) as well as aragonite crystals (e.g., Pif, nacrein) and a number of enzymes metabolizing chitin protein, including CHIT (may be responsible for chitin degradation) and chitin synthase (may be active in construction of the chitin framework) ([@ref-15]) were identified in the presented transcriptomes. Perlucin (nucleates growth of CaCO~3~ crystals) ([@ref-6]) and calponin (calcium binding protein, may ensure shell elasticity) ([@ref-66]) were also confirmed in the transcriptomes of *Mytilus* spp. by this study ([Table S6](#supp-15){ref-type="supplementary-material"}). In EduN, TroV, GalM and GalT groups, 15 contigs were identified rich in immunoglobulin domains with homology from 34% to 69% sequence identity to hemicentin from *M. yessonis* Jay, 1857, *Crassostrea virginica* Gmelin, 1791 and *C. gigas*. Hemicentin is an extracellular ion-binding protein with an unclear function in shell formation, however, it was previously found in the coral skeletal organic matrix ([@ref-62]). Another significant protein is chorion peroxidase (TroV_3244) with an unknown role in biomineralization, though it has been suggested that it might be involved in melanogenesis in *S. officinalis* Linnaeus, 1758 ([@ref-22]). In this study, several candidate biomineralization proteins were selected for more comprehensive analysis: DPT, AK and GAPDH, since they were represented in at least four of the five *Mytilus* groups. DPT is an extracellular matrix protein and may participate in nacre formation ([@ref-32]). GAPDH is an enzyme involved in carbohydrate metabolism, which catalyze the formation of 1,3-bisphospho glycerate from glyceraldehyde 3-phosphate and might be involved in the shell growth process ([@ref-3]). AK is a protein with ATP-guanido phosphotransferases domain, which may play a role in buffering intracellular pH, important in the extrapallial space with the supersaturation of shell matrix components, including calcium ions ([@ref-9]). A phylogenetic reconstruction of these proteins showed a well-supported, separate clade of Mytiloida order in a sister relationship to other Bivalves ([Fig. 1](#fig-1){ref-type="fig"}). However, the trees presented did not support relationships between the four orders: Mytiloida, Ostreoida, Pterioida and Pectinoida. Only AK phylogeny clearly pictures the relationships among major bivalve lineages Pterimorphia and Heteroconchia, although it should be noted that there is a lack of sequences for DPT and GAPDH proteins. In the case of AK and GAPDH high conservation was observed between *M. edulis, M. galloprovincialis* and *M. trossulus* taxa, suggesting the importance of these candidate biomineralization genes in mussels. Only DPT phylogeny confirmed the expected evolutionary relationships of *M. trossulus*, *M. edulis* and *M. galloprovincialis*. Rising atmospheric CO~2~ emissions and the direct consequences of it, ocean acidification and ocean warming, have the potential to limit the ability of calcifying marine organisms to produce their protective exoskeleton and shell, and lead to decalcification ([@ref-17]; [@ref-23]). Changes in the structural order potentially impact on mussel shell strength, limiting protection from predators and the environment. In order to project the future implications of ocean acidification on marine calcifying organisms the biomineralization mechanism need to be fully understood. Previous studies have explored the impact of acidification on shell formation and on the activity of candidate biomineralization molecules: α-CA, CHIT and TYR in mussels ([@ref-29]; [@ref-17]). The evolutionary relationships of these genes in metazoan have been previously studied ([@ref-1]; [@ref-42]; [@ref-66]; [@ref-15]), however, there has still been a lack of sequence resources especially from mussels. Given the importance of these genes in biomineralization process, phylogenetic reconstructions using contigs obtained from this study were performed. α-CAs are metalloenzymes which catalyze the reversible hydration of CO~2~ to form bicarbonate ion and proton ([@ref-42]). In mammals the α-CA family is characterized by 13 different enzymatically active isoforms and three CARPs which appear to lack CA activity ([@ref-4]). CHIT is hydrolyzing β-1,4-linkages in chitin, which is a major component of the organic matrix in the shell. TYR is a type-3 copper protein superfamily, a multifunctional, shell associated protein, melanocyte-specific marker, potentially involved in pigmentation and essential for the varied coloration of shell in bivalves ([@ref-16]; [@ref-70]). It has been suggested that TYR represents a protective mechanism against external shell corrosion ([@ref-17]). This study identified several copies (isoforms) of α-CA, CHIT and TYR proteins in the genus *Mytilus*, which might have been a result of several duplication events and diversification resulting in multiple functionality: immune response, potentially biomineralization and pigmentation ([Figs. 2](#fig-2){ref-type="fig"} and [3](#fig-3){ref-type="fig"}). This diversity might be also correlated with wide geographical distribution, adaptation to different environments and different selection pressures driving evolution of these specific genes. The contigs identified in this study have provided a new resource for future studies on α-CAs, CHIT and TYR, however, the phylogeny reveals a complex evolutionary history of these genes that requires more data. To identify homologs and similarity of mantle transcriptomes between different taxa and localizations, BBH method was used. The highest similarity (\>80%), was observed between *M. edulis* (EduN), *M. galloprovincialis* (GalM, GalT) and *M. chilensis* (Chil) taxa ([Fig. S6](#supp-6){ref-type="supplementary-material"}). Respectively, 848, 488, 549, 343 and 298 contigs in EduN, TroV, GalM, Chil, GalT groups, did not show homology (*E*-value cut-off 10^−15^) to mantle transcriptomes from previous studies in mussels ([Fig. S7](#supp-7){ref-type="supplementary-material"}). These contigs may represent novel or specific genes for each group or might be related to environment conditions. To further investigate diversity between different *Mytilus* spp. groups, a GO Fisher exact test was carried out. In *M. edulis* from the North Sea (EduN), GO terms related to heat stress response were over-represented, including: apoptotic process, regulation of programmed cell death, cellular protein complex assembly, microtubule-based process and calcium ion binding ([Fig. S8](#supp-8){ref-type="supplementary-material"}). These terms were associated with high pH exposure in the Antarctic pteropod ([@ref-33]) and thermal response in the genus *Crassostrea* ([@ref-81], [@ref-47]). In *M. galloprovincialis* from the Mediterranean Sea (GalM) metal ion binding, purine ribonucleoside triphosphate binding, lipid transport and oxidative phosphorylation were enriched ([Fig. S9](#supp-9){ref-type="supplementary-material"}). These terms might be associated with pH and temperature response, as found in *P. fucata* Gould, 1850 ([@ref-44]) and *C. gigas* ([@ref-81]). Thermal and pH response of mantle tissue in both EduN and GalM groups, might be a direct consequence of increasing carbon dioxide (connected with ocean acidification and global warming), which significantly affects marine calcifiers ([@ref-44]). However, these results should be interpreted with care, since different number of sequences were used to retrieve GO terms. Over the last decade, understanding of *Mytilus* spp. evolutionary characteristics significantly increased. Several approaches using molecular methods were used, such as molecular markers ([@ref-36]), SNP genotyping ([@ref-21]) and mitochondrial DNA ([@ref-67]). In this study phylogenetic relationships were investigated between the six taxa using transcriptome data ([Fig. 4](#fig-4){ref-type="fig"}). Our analyses showed that *M. edulis*, *M. chilensis*, *M. galloprovincialis* and *M. trossulus* created a separate clade from *M. californianus* and *M. coruscus*, which is consistent with earlier studies of their mitochondrial DNA ([@ref-19]). Conclusions =========== This study investigated mantle transcriptomes of five groups from four *Mytilus* spp. using 454 pyrosequencing technology. Of generated transcripts 46.57% were annotated by BLAST search. Of these 6.16% were contigs potentially involved in biomineralization and pigmentation processes, such as CA, CHIT and TYR. GO enrichment analysis revealed stress response in *M. edulis* (from the North Sea) and *M. galloprovincialis* (from the Mediterranean Sea), which might be related to ocean acidification and global warming. To our knowledge, this study is the first to describe mantle transcriptomes from multiple *Mytilus* spp and provide a basis for further work (using RNA-seq and gene expression patterns) on biomineralization proteins, which may help to further account for mussels diversity and responses to environment conditions. Supplemental Information ======================== 10.7717/peerj.6245/supp-1 ###### Characteristics of the homology search of *Mytilus* transcripts. Distribution of an E-value (A) and species annotation (B) based on the NR best hit results. ###### Click here for additional data file. 10.7717/peerj.6245/supp-2 ###### KEGG analysis of the five *Mytilus* transcriptomes. CP--Cellular Processes, EIP--Environmental Information Processing, GIP--Genetic Information Processing, M--Metabolism, OS--Organismal Systems. ###### Click here for additional data file. 10.7717/peerj.6245/supp-3 ###### Gene Ontology classification of *Mytilus* transcriptomes: biological process (A), molecular function (B), cellular component (C). ###### Click here for additional data file. 10.7717/peerj.6245/supp-4 ###### Putative single nucleotide polymorphisms (SNPs) distribution in *Mytilus* spp. mantle transcriptomes. ###### Click here for additional data file. 10.7717/peerj.6245/supp-5 ###### Alignment of glyceraldehydes-3-phosphate (GAPDH) protein sequences in Bivalvia class. ###### Click here for additional data file. 10.7717/peerj.6245/supp-6 ###### Sequence similarity distribution of pairwise comparisons between *Mytilus* spp. groups from present study. Color code indicate the percentage of similarity between two groups. ###### Click here for additional data file. 10.7717/peerj.6245/supp-7 ###### Sequence similarity distribution of pairwise comparisons between *Mytilus* spp. groups: EduN (A), GalM (B), Chil (C), TroV (D), GalT(E) and mantle transcriptomes from *M. edulis* and *M. galloprovincialis*. *M. edulis*: EduG ([@ref-18]), EduS ([@ref-78]), *M. galloprovincialis*: GalP ([@ref-5]), GalS ([@ref-58]). Color code indicate the percentage of similarity between two groups. ###### Click here for additional data file. 10.7717/peerj.6245/supp-8 ###### Over-represented biological process (BP) and molecular function (MF) GO terms in *M. edulis* (EduN). ###### Click here for additional data file. 10.7717/peerj.6245/supp-9 ###### Over-represented biological process (BP) and molecular function (MF) GO terms in *M. galloprovincialis* (GalM). ###### Click here for additional data file. 10.7717/peerj.6245/supp-10 ###### A reference list of previously reported candidate biomineralization genes. ###### Click here for additional data file. 10.7717/peerj.6245/supp-11 ###### Statistical summary of five *Mytilus* transcriptomes. ###### Click here for additional data file. 10.7717/peerj.6245/supp-12 ###### Table listing *Mytilus* spp. contigs, including description (NCBI NR database), GO and KEGG IDs ascribed to each sequence. ###### Click here for additional data file. 10.7717/peerj.6245/supp-13 ###### KEGG pathway annotations in mussel transcriptomes. ###### Click here for additional data file. 10.7717/peerj.6245/supp-14 ###### Summary of simple sequences repeats (SSRs) analysis (A) and distribution based on motif types (B) in the *Mytilus* spp. mantle transcriptomes. ###### Click here for additional data file. 10.7717/peerj.6245/supp-15 ###### Contigs potentially involved in the biomineralization mechanism identified in mussel transcriptomes, including description, domain ascribed and sequences in FASTA format. ###### Click here for additional data file. 10.7717/peerj.6245/supp-16 ###### Contigs identified in tyrosinase, endocytosis, calcium signaling and melanogenesis pathways. ###### Click here for additional data file. 10.7717/peerj.6245/supp-17 ###### α-CA and GH18 sequences with accesion numbers (NCBI) identified in mantle transcriptomes. ###### Click here for additional data file. The authors thank Dr. Tomasz Sanko for preparation of RNA samples for 454 sequencing. This research was supported in part by PL-Grid Infrastructure. Additional Information and Declarations ======================================= The authors declare that they have no competing interests. [Magdalena Malachowicz](#author-1){ref-type="contrib"} performed the experiments, analyzed the data, prepared figures and/or tables, authored or reviewed drafts of the paper, approved the final draft. [Roman Wenne](#author-2){ref-type="contrib"} conceived and designed the experiments, performed the experiments, contributed reagents/materials/analysis tools, authored or reviewed drafts of the paper, approved the final draft. The following information was supplied regarding data availability: The raw data is available on Sequence Read Archive under accession number [PRJNA419475](PRJNA419475). The CA and CHIT sequences described here are accessible via GenBank accession numbers [MG827120](MG827120)--[MG827134](MG827134) and are included in [Supplemental_Data_File_1](#supp-17){ref-type="supplementary-material"}. Other sequences are included in [Table S6](#supp-15){ref-type="supplementary-material"}.

Introduction {#S1} ============ Background to Time Pressure {#S1.SS1} --------------------------- Within modern society, perceived shortage of time is a frequently experienced aspect of daily life ([@B37]; [@B57]; [@B96]). Recognizing this, a range of academic disciplines have investigated time pressure. This breadth of interest has resulted in a variety of conceptualizations ([@B88]). These include (time): pressure ([@B91]), stress ([@B23]), crunch ([@B97]), famine ([@B78]), poverty ([@B25]), deficit ([@B3]), squeeze ([@B47]), sickness ([@B28]), and scarcity ([@B43]). Examination of these terms reveals they possess a similar underlying meaning. Accordingly, researchers often use these labels interchangeably ([@B78]). In this context, at a general level, subjective 'time pressure' refers to the notion that there is insufficient time available to complete necessary tasks (perceived acute time shortage) ([@B79]; [@B48]). Concomitant with the notion of 'pressure' is the concept of temporal overload. This denotes the busy, hastening pace of life, the need to multitask, and the necessity to perform tasks faster. Indicative terms include tyranny of the moment ([@B32]), fast time ([@B32]), 24-h society ([@B51]), pace of life ([@B34]), hyperculture ([@B2]), and work intensification ([@B63]) (see [@B88]). At a phenomenological level, subjective time pressure comprises experience of both tempo (accelerated pacing of time) and limits/choices (having to select one action over another) ([@B24]). At a higher conceptual level, consideration of the perceived time pressure literature reveals a focus on global temporal concerns or consequences ([@B85]; [@B86]). Particularly, feeling harried and pressed for time. Harried signifies the need to adhere to myriad social practices within precise timeframes, and 'pressed for time' designates a general absence of free time ([@B86]). Thus, harried embodies the negative sense of feeling rushed to a point where time concerns produce worry and anxiety ([@B6]). Relatedly, [@B67] introduced the term 'extended present' to describe hastened moment-by-moment daily living. This encompasses perceptions of speeded up time, fragmentation resulting from multitasking, and the awareness that constant interruptions frustrate task completion. These acuities result in constant feelings of busyness, unnoticed passing of time, and a lack of opportunity to plan prospectively. Furthermore, [@B78] employ the concept of time deepening to explain how people under temporal constraints work more intensely. Time deepening involves adopting strategies, such as tight scheduling, speeding up activities, multitasking and selecting faster (leisure) activities. Time pressure is also conceptualized in other related ways. For instance, [@B85] observed that people create 'hot spots' for work, which free up time for 'cold spots' of slow time allotted to family and leisure. The limitations of 'hot spots' are that they can create mental strain, a sense of being rushed, and require disciplined adherence. Negative and Positive Consequences of Time Pressure {#S1.SS2} --------------------------------------------------- There is strong evidence to suggest that experience of persistent time pressure is associated with poorer health and reduced quality of life (e.g., lower health and life satisfaction) ([@B40]). This is especially true when individuals are unable to mentally detach or "switch off" from work ([@B84]). Indeed, [@B96] reported a negative correlation between extreme levels of time pressure and mental health. This relationship may arise from the fact that acute time pressure can act as a physical and psychological stressor, attendant with a range of undesirable outcomes (i.e., inability to cope, sleeping difficulties, tension and fatigue). Although investigations of chronic time pressure typically report that sustained perceived temporal constraints negatively affect health and psychological wellbeing, there exists a substantial body of work that has demonstrated that time pressure can facilitate positive social behavior (see review by [@B12]). Illustratively, in a series of experiments using economic games, [@B76] found that participants who reached decisions faster were more cooperative. Other researchers have produced similar outcomes (see [@B74]). However, studies also report null results ([@B93]; [@B5]). Researchers have also produced commensurate positive results for altruism ([@B75]), honesty ([@B83]; [@B11]; [@B59]; [@B14]), and the equity-efficiency trade-off ([@B13]). In the case of altruism, it is important to note that [@B75] only observed a positive effect for women (there was no significant effect for men). Collectively, investigations demonstrate that perceived time pressure or shortage can have both positive and negative consequences depending on duration, context and individual perceptions ([@B80]). This reflects the subjective perception of time pressure. Particularly the fact that individuals can view it either as motivating, or as something to endure ([@B54]). Conceptualization of Chronic Time Pressure ([@B88]) {#S1.SS3} --------------------------------------------------- Noting that lack of consensus and heterogeneity of terminology hindered the conceptual development of time pressure, [@B88] identified core elements that define the construct. These embrace the notion that subjective experience of time shortage is negative (i.e., aversive, unwanted, undesirable, and apprehension inducing) and coalesce around two factors, time shortage (perceived lack of time), and being rushed (sense of time passing quickly). The latter two experiences overlap, but represent discrete factors. Time shortage implies objective problems with time allocation, involves cognitive based judgment, and produces minimal affect. Hence, the term is often synonymous with time-management. Contrastingly, feeling rushed focuses on subjective emotional experiences, and is experiential in nature. Accordingly, feeling rushed typically gives rise to feelings of apprehension, worry, anxiety and frustration. Noting this distinction, in an attempt to establish theoretical clarity, [@B88], p. 339) defined time pressure as, "a temporary, overarching designation that would subsume all the terms related to time shortage as well as to being rushed." This delineation acknowledged cognitive awareness of insufficient time and the emotional experience of hastened pace. Furthermore, [@B88] definition linked time pressure with stress-related research. Explicitly, it recognized that perceived time pressure was transactional, arising from the interaction between the individual and their environment. An additional conceptual advantage of linking time pressure with stress is that the association emphasizes the chronic consequences that time shortage can have on the individual. The word 'chronic' in this circumstance is important because it delimits time pressure as a habitual, reoccurring, repetitive, and potentially aversive process. Noting these features, [@B88] advocated adoption of the nomenclature chronic time pressure (CTP). Theoretically, CTP provides a solid platform for empirical inquiry. Explicitly, it facilitates the investigation of time pressure at a variety of levels (i.e., from individual physiology to societal life--work balance). Furthermore, the term recognizes that sustained time pressure at high levels is potentially harmful ([@B48]). Indeed, intense time pressure can act as a stressor ([@B37]; [@B79]; [@B88]), which is detrimental to mental well-being ([@B96]; [@B79]). This notion has received attention in stress-related research (e.g., Type A behaviors) ([@B33]). Szollos' ([@B88]) conceptualization of chronic time pressure is important because researchers have recently applied the construct to important real-world contexts, outlined implications and produced interventions. Work has embraced behavioral, cognitive and theoretical perspectives. For example, [@B17] reported that time pressure has an effect on risky street-crossing decision-making. Relatedly, [@B46] considered the impact of time pressure on driver behavior. More generally, authors have evaluated the effect of temporal constraints on work life balance ([@B82]), well-being ([@B81]), and lifestyle choices (e.g., eating habits, [@B45]). Conceptually, studies have investigated whether time pressure is stable across situations (home and work; [@B48]), cultures (Nordic countries; [@B39]), and genders ([@B92]). It is important to consider whether cultural and individual differences effect the perception, and consequences of time pressure because variations limit the usefulness of generalizations. Additionally, researchers have investigated relationships between chronic time pressure and psychological well-being. A notable example is [@B38], who found an association between parents' subjective time pressure and increased mental health problems among children. Acknowledging that chronic time pressure is an unavoidable experience of daily life in industrialized societies, several studies have examined the effectiveness of alleviating strategies and therapeutic interventions. Examples include using teamwork (clarifying demands and setting priorities) to moderate the relationship between time pressure and exhaustion ([@B50]), mindfulness training (to counterbalance acceleration in social and working life and social relationships) ([@B53]), and downtime (a state of physical relaxation and psychological detachment) ([@B30]). The Present Study {#S1.SS4} ----------------- Researchers frequently use self-report measures (i.e., questionnaires and diaries) to assess feelings and perception of time pressure. Hence, scales represent an established, reliable and valid methodological tool for assessing the construct. Noting this, the present study designed a scale that incorporated the multidimensional aspects of CTP outlined by [@B88]. This was necessary for two reasons. Firstly, it facilitated psychometric evaluation of [@B88] conceptualization of CTP. Particularly, analysis enabled the authors to determine whether measurement models provided adequate evidence for the existence of related, but discrete factors (time shortage proper and being rushed). At a general level, the study also offered insights into the dimensionality of CTP. Secondly, previous self-report measures have been limited because there is no universal measure of CTP. Studies typically use either single questions (i.e., the General Social Survey), or small item groups, which assess only global (overall) perceptions of time pressure. Illustratively, the General Social Survey measures time pressure via a single item questioning how often the respondent feels rushed ([@B48]). Regarding small item groups, a typical example is [@B1] (i.e., "I feel a lot of time pressure in my life," "I really feel the pressure of time passing in my life," and "I am always in a hurry."). A further often cited measure is [@B73]. This comprises five-items related to shopping, which index: feeling pressed for time, being in a hurry, having limited/enough amount of time, and fastness. Accordingly, researchers frequently adapt these items for the purpose of their studies (e.g., [@B90]; [@B56]). The current study represented the first step in the development of an established measure of CTP. This is an important development in time pressure research because it builds on [@B88] review of literature and is conceptually congruent with his conclusions. Moreover, developing a robust measure of CTP facilitates context comparisons. This is important because the use of myriad measures conflates assessment of environment variations. An additional objective was to examine convergent validity of the resultant measure. Specifically, this comprised a comparison with an established and relevant scale (the 10-item Perceived Stress Scale by [@B21]; PSS-10). Convergent validity is useful when designing a scale because it indicates the extent to which a measure aligns with a construct it should relate to according to theoretical predictions. Previous research consistently demonstrates that perceptions of stress correlate largely with time pressure. For example, [@B49] evidenced a correlation of 0.50 between time management pressures/issues among teachers (i.e., if a teacher can find the time to meet all significant professional or personal needs) and perceived stress. This study anticipated a similar pattern. Materials and Methods {#S2} ===================== Participants {#S2.SS1} ------------ ### Study 1 {#S2.SS1.SSS1} A sample of 401 respondents (326 women, 81% and 74 men, 19%) participated in this study. The overall mean age was 26.20 (*SD* = 11.799, range 18--68 years). The mean age for men was 28.64 (*SD* = 12.488; range = 18--68 years), and the mean age for women was 25.66 (*SD* = 11.604; range = 18--67 years). Respondents included university students (24%) and employees from various occupational sectors (76%). Specifically, 21% from the educational sector; 13% public services and administration; 8% accountancy, banking and finance; 16% healthcare; 3% recruitment and HR; 6% retail and sales; 7% business and management. Recruitment was via social media, university staff email, and through local stakeholders (businesses and vocational classes). Involvement was voluntary and responses anonymized. Participants could withdraw up to 4 weeks after data collection. Exclusion criteria required that respondents were at least 18 years of age. Assessment of univariate skewness and kurtosis ([Table 1](#T1){ref-type="table"}) indicated no concerns, as values fell within the recommended range of −2.0 to +2.0 ([@B8]). Conversely, multivariate non-normality existed, as [@B61] kurtosis (*b2p* = 12.178, *p* \< 0.001) and skewness (*b1p* = 13.980, *p* \< 0.001) tests suggested significant deviation from normal distribution. ###### Summary statistics for Study 1 and Study 2 items. **Study 1** **Study 2** ----------------- ------------- ------------- --------- --------- ------ ------- --------- --------- Q1 3.70 1.003 --0.741 0.047 3.66 0.971 --0.655 0.068 Q2 3.27 1.038 --0.315 --0.947 3.15 1.010 --0.240 --0.974 Q3 3.51 1.096 --0.469 --0.699 3.38 1.078 --0.179 --1.050 Q4 3.58 0.949 --0.388 --0.654 3.39 0.884 --0.246 --0.876 Q5 3.01 0.955 0.054 --1.034 2.84 0.895 0.268 --0.646 Q6 2.96 1.095 0.079 --1.153 2.93 1.037 0.069 --1.033 Q7 2.91 1.059 0.190 --0.927 2.74 1.069 0.255 --0.855 Q8 3.67 0.968 --0.698 --0.157 3.55 0.693 --0.590 --0.464 Q9 3.34 1.051 --0.297 --0.968 3.20 1.071 --0.125 --1.126 Q10 2.55 1.001 0.445 --0.622 2.65 1.003 0.341 --0.743 Q11 3.78 1.067 --0.802 --0.140 3.85 1.052 --0.849 --0.099 Q12 3.13 1.085 --0.083 --0.901 2.99 1.130 0.064 --0.925 Q13 3.86 1.113 --0.947 0.047 3.93 1.136 --1.081 0.390 Q14 3.30 0.982 --0.359 --0.731 3.07 0.937 --0.136 --0.665 Q15 2.97 1.014 0.056 --1.027 2.96 0.990 0.164 --1.016 **PSS-10 item** Q1 2.93 1.078 0.048 --0.627 2.96 1.102 0.086 --0.736 Q2 3.08 1.088 --0.005 --0.671 3.05 1.104 --0.014 --0.612 Q3 3.74 1.053 --0.468 --0.523 3.82 1.036 --0.716 0.055 Q4 2.66 0.966 0.300 --0.425 2.63 0.943 0.368 --0.560 Q5 2.87 0.867 0.033 --0.366 2.83 0.836 0.077 --0.449 Q6 3.02 1.024 0.216 --0.559 2.96 0.987 0.152 --0.516 Q7 2.78 0.876 0.022 --0.285 2.68 0.859 --0.044 --0.418 Q8 2.99 0.897 --0.001 --0.305 2.89 0.903 0.220 --0.357 Q9 3.04 1.073 --0.048 --0.692 2.95 1.082 --0.109 --0.798 Q10 2.76 1.181 0.245 --0.788 2.72 1.179 0.382 --0.657 CTPI, Chronic Time Pressure Inventory; PSS-10, 10-item Perceived Stress Scale. ### Study 2 {#S2.SS1.SSS2} The Study 2 sample comprised 163 respondents (133 women, 82% and 30 men, 18%). Mean sample age was 19.15 (*SD* = 2.886, range = 18--36 years). For men, mean age was 20.21 (*SD* = 5.747, range = 18--36 years), and 18.92 for women (*SD* = 1.676, range = 18--29 years). Of the sample, 71% were university students and 29% were employees from different occupations. Specifically, 10% from the educational sector; 8% public services and administration; 7% healthcare; 4% business and management. The same recruitment procedure occurred as for Study 1. As with the Study 1 data, acceptable univariate skewness and kurtosis existed (i.e., all between −2.0 to +2.0) ([Table 1](#T1){ref-type="table"}). In addition, non-normality existed, as [@B61] kurtosis (*b2p* = 5.351, *p* \< 0.001) and skewness (*b1p* = 35.401, *p* \< 0.001) inferred significant deviation. Measures {#S2.SS2} -------- Study 1 and Study 2 both used the Chronic Time Pressure Inventory (newly devised by the study authors) and the Perceived Stress Scale ([@B20]). The Chronic Time Pressure Inventory {#S2.SS3} ----------------------------------- An extensive review of literature on time pressure by the study authors occurred resulting in different but related operationalizations, for example 'work demand' (e.g., [@B4]), 'time constraints' (e.g., [@B9]), 'time urgency' (e.g., [@B55]), 'time pressure' (e.g., [@B42]). Notably (as alluded to in the Introduction), an absence of psychometrically validated measures existed, and in some instances single-item measures assessed time pressure (e.g., [@B42]). [@B88] review offered a thorough assessment of the status of time pressure literature and a plausible conceptualization of the construct. Thus, using this theoretical framework, the development of initial statements occurred. The study authors independently devised the statements to reflect core features of [@B88] conceptualization. This resulted in 15 items following the removal of items indicative of duplication. A group of four experienced academics reviewed the measure, concluding that the scale appeared to assess features of time shortage, pressure, and feeling rushed. This procedure offered satisfactory face validity (e.g., see [@B60]). The scale comprised 15 statements (e.g., "I always run out of time") where participants are asked to rate each as it applies to them using a Likert response format from 1 (Strongly disagree) to 5 (Strongly agree). The Perceived Stress Scale ([@B20]) {#S2.SS4} ----------------------------------- The Perceived Stress Scale, a global stress measure, assessed the extent that respondents perceive life to be unpredictable, uncontrollable, and overloading ([@B35]). The PSS detects existing stressful circumstances and background extraneous stressors. The present study used the 10-item version (PSS-10), which asks about thoughts and feelings over the last month. Items appear in the form of statements (e.g., "In the last month, how often have you felt nervous and stressed?") and participants indicate their level of agreement on a 5-point Likert scale ranging from 0 (Never) to 4 (Very often). Summation of item scores produces an overall stress score; higher scores indicate greater levels of perceived stress. In addition, the scale possesses two underlying factors of 'Distress' and 'Coping' ([@B27]). The PSS-10 possesses established psychometric properties (internal consistency, test-retest reliability and factor structure) ([@B21]; [@B27]). Procedure {#S2.SS5} --------- Prior to participation, potential respondents read the study brief. This contained background information about the nature of the study, and outlined the conditions and requirements of involvement. Only consenting participants progressed to the online measures hosted by Qualtrics. Further instructions asked participants to take their time, complete all questions, and answer questions openly/honestly. The online self-report measure comprised three subsections: demographic information (completed first), time pressure and perceived stress. To eliminate order effects, measure sequence rotated across respondents. This procedure was identical for Study 1 and Study 2. Ethics {#S2.SS6} ------ The research team gained ethical authorization for the project (Evaluating the Chronic Time Pressure Inventory). The study investigated the development of a new measure of chronic time pressure in two independent samples. Following formal submission, the Director of the Research Institute for Health and Social Change and the Manchester Metropolitan University Faculty of Health, Psychology and Social Care Ethics Committee granted ethical approval. Analytic Strategy {#S2.SS7} ----------------- Psychometric evaluation of the Chronic Time Pressure Inventory (CTPI) advanced through a number of sophisticated analytical procedures. These involved Horn's parallel analysis, exploratory factor analysis (EFA using maximum likelihood), and confirmatory factor analysis (CFA). Parallel analysis (PA) determined the number of factors representing the CTPI. This is an empirically supported approach for establishing the quantity of factors underlying a measure ([@B70]). PA involved random resampling of raw data ([@B68]). Eigenvalues higher than random data eigenvalues represented underlying factors. EFA (SPSS 25) with the recommended number of factors subsequently provided details of item loadings ([@B22]). Next, CFA (AMOS25) examined data-model fit. Given the presence of non-normality, CFA used ML estimation with bootstrapping (2000 resamples) to produce standard error estimates and confidence intervals (bias-corrected at the 95% confidence level) and *p*-values ([@B8]). Naïve bootstrapping is a robust alternative to other ML robust approaches (e.g., Satorra--Bentler chi-square), and operates successfully even when data evidences extreme non-normality ([@B65]). Assessment of model fit included the chi-square statistic (χ*^2^*), Comparative Fit Index (CFI), Root-Mean-Square Error of Approximation, RMSEA, and Standardized Root-Mean-Square Residual, SRMR. RMSEA scrutiny involved reference to its 90% confidence interval (CI). Values \> 0.90 imply good fit for CFI ([@B44]). Values of 0.05, 0.06--0.08, and 0.08--1.0 indicate good, satisfactory and marginal RMSEA and SRMR ([@B7]). Model comparison comprised Akaike's Information Criterion (AIC). Lower values signify superior fit. Internal consistency assessment followed guidelines recommended by [@B94]. This comprised the testing of equality constraints via SEM-based congeneric, tau equivalent and parallel models prior to estimating reliability coefficients (in this instance alpha and omega). The congeneric imposes the least restrictive assumptions, supposing that all scale items measure the same latent construct, with potentially different degrees of precision and error. The tau equivalent assumes that scale items measure the same latent construct with the same degrees of precision, but with potentially different error. The parallel imposes the most restrictive assumptions, supposing that items measure the same construct, with identical degrees of precision and error ([@B64]). Analysis considered chi-square differences among these models. However, because Study 1 involved a moderately large sample, consultation of fit indices less sensitive to sample size (CFI and SRMR) occurred, with 0.05 as an acceptable cut-off for changes in these ([@B58]). Cronbach's alpha examined internal consistency of the CTPI following data-model inspection and testing of equality constraints within a single group. Coefficient omega (ω) also measured reliability (using the Omega program; [@B95]) given this more successfully estimates reliability than alpha ([@B26]). Subsequently, convergent validity testing occurred. This included correlating CTPI with the PSS-10. Lastly, analysis involved inspecting associations between gender, age, CTPI and PSS-10. This project adopted [@B19] criteria for judging the magnitude of associations. Specifically, 0.1--0.29 indicates a weak correlation; 0.3--0.49 suggests a moderate relationship; and 0.50 or greater infers a strong correlation. The analysis procedure for Study 2 included (in an independent sample) replication and testing of the superior Study 1 model via CFA and assessment of alpha and omega reliability. Additionally, multi-group CFA evaluated invariance of the superior model by comparing Study 1 and Study 2 data at the configural (factor structure), metric (factor loadings), scalar (item intercepts), and residual (item residuals) levels. Assessment of fit at each stage involved consultation of [@B15] criteria (CFI difference ≤ 0.01 and RMSEA ≤ 0.015). As with Study 1, the final stages of analysis comprised a test of convergent validity (i.e., comparing CTPI with PSS-10) alongside exploring associations between gender, age, CTPI and PSS-10. Results {#S3} ======= Study 1 {#S3.SS1} ------- Parallel analysis using 1000 resamples indicated that two factors existed, given two factors evinced higher values compared with random data (factor 1 eigenvalue = 5.297 vs. 1.34; factor 2 eigenvalue = 1.533 vs. 1.265). EFA indicated a satisfactory item correlation matrix, Bartlett's Test of Sphericity *p* \< 0.001; and good sampling adequacy, Kaiser-Meyer-Olkin = 0.900. The factors accounted for 45.534% of variance. Consideration of factor loadings revealed that item 14 ('I am not concerned about being late and missing things') loaded poorly (0.212). In addition, item 7 ('I am so busy I mess things up') evidenced cross loading (factor 1 loading = 0.389; factor 2 loading = 0.329). EFA following removal of these items explained 48.792% of the variance. All items loaded above the minimum threshold of 0.32 ([@B89]) and, apart from item 6 ('I feel in control of how I spend my time', loading of 0.387), exceeded 0.4 ([@B66]). Five items loaded on Factor 1, and eight items loaded on Factor 2 ([Table 2](#T2){ref-type="table"}). Items belonging to Factor 1 (labeled as 'Feeling Harried') referred to a negative sense of feeling rushed to an extent where time concerns generate anxiety and worry. Items informing Factor 2 (named 'Cognitive Awareness of Time Shortage') referred to a cognizance of not having enough time to complete tasks, to do the things they enjoy (see [Appendix](#A1){ref-type="app"} for final scale). ###### EFA and CFA factor loadings for Study 1 and Study 2. **Study 1** **Study 2** ----- ------------- ------------- ----------- ----------- ----------- -------- ----------- ----------- -------- ----------- ----------- Q4 0.511 0.673^∗∗^ 0.681^∗∗^ 0.001 0.723^∗∗^ 0.180 0.481^∗∗^ Q8 0.659 0.479^∗∗^ 0.549^∗∗^ −0.348 0.459^∗∗^ 0.685 0.545^∗∗^ Q10 0.685 0.570^∗∗^ 0.632^∗∗^ −0.170 0.600^∗∗^ −0.069 0.703^∗∗^ Q11 0.853 0.584^∗∗^ 0.676^∗∗^ −0.638 0.564^∗∗^ 0.152 0.563^∗∗^ Q12 0.645 0.703^∗∗^ 0.757^∗∗^ −0.193 0.718^∗∗^ −0.256 0.768^∗∗^ Q1 0.592 0.450^∗∗^ 0.508^∗∗^ 0.444^∗^ 0.339^∗∗^ 0.411^∗^ 0.373^∗∗^ Q2 0.553 0.445^∗∗^ 0.508^∗∗^ 0.475^∗^ 0.317^∗∗^ 0.372^∗^ 0.251^∗^ Q3 0.555 0.610^∗∗^ 0.648^∗∗^ 0.345^∗∗^ 0.539^∗∗^ 0.309^∗^ 0.512^∗∗^ Q5 0.661 0.484^∗∗^ 0.539^∗∗^ 0.369^∗∗^ 0.408^∗∗^ 0.403^∗^ 0.483^∗∗^ Q6 0.387 0.615^∗∗^ 0.631^∗∗^ 0.312^∗^ 0.537^∗∗^ 0.060 0.557^∗∗^ Q7 0.629 0.378^∗∗^ 0.455^∗∗^ 0.498^∗∗^ 0.243^∗∗^ 0.470^∗∗^ 0.282^∗^ Q9 0.490 0.578^∗∗^ 0.616^∗∗^ 0.365^∗^ 0.491^∗∗^ 0.287^∗^ 0.397^∗∗^ Q13 0.449 0.680^∗∗^ 0.672^∗∗^ 0.207^∗∗^ 0.660^∗∗^ 0.389^∗^ 0.558^∗∗^ CTPI, Chronic Time Pressure Inventory. a \* p \< 0.05; ∗∗ p \< 0.001 (using bootstrap significance estimates). Analysis of a correlated two-factor model resembling the EFA results reported satisfactory fit overall ([Table 3](#T3){ref-type="table"}). All factor loadings ([Table 2](#T2){ref-type="table"}) exceeded 0.4 ([@B66]), with 8 of the 13 items (62%) loading above the strict requirement of 0.6 by [@B41]. Further examination revealed that the subfactors correlated highly (0.747), suggesting conceptual overlap. Therefore, a bifactor model tested the notion of multidimensionality. Findings ([Figure 1](#F1){ref-type="fig"}) indicated good fit overall. ###### Fit indices for measurement and invariance models of the CTPI. **Model** **χ^2^** ***df*** **CFI** **SRMR** **RMSEA (90% CI)** **AIC** ---------------------------- ------------- ---------- --------- ---------- ---------------------- --------- **Study 1 (*N* = 401)** One-factor 313.609^∗∗^ 65 0.834 0.072 0.100 (0.087--0.109) 391.609 Two-factor 208.053^∗∗^ 64 0.904 0.058 0.075 (0.064--0.087) 288.053 Bifactor 125.672^∗∗^ 52 0.951 0.038 0.060 (0.046--0.073) 229.672 Reliability  Congeneric 125.672^∗∗^ 52 0.951 0.038 0.060 (0.046--0.073)  Tau equivalent 212.846^∗∗^ 75 0.907 0.067 0.068 (0.058--0.079)  Parallel 244.249^∗∗^ 86 0.894 0.074 0.068 (0.058--0.078) **Study 2 (*N* = 163)** Bifactor 84.972^∗∗^ 52 0.935 0.056 0.063 (0.037--0.086) **Invariance (*N* = 564)** Configural 216.500^∗∗^ 106 0.945 0.045 0.043 (0.035--0.051) Metric 257.485^∗∗^ 128 0.935 0.045 0.042 (0.035--0.050) Scalar 282.369^∗∗^ 141 0.929 0.046 0.042 (0.035--0.049) Residual 302.872^∗∗^ 157 0.927 0.047 0.041 (0.034--0.048) ∗∗ Indicates p \< 0.001. {#F1} All factor loadings (apart from item 6, loading of 0.243) exceeded 0.32 on the general factor. In comparison, some of the items pertaining to Factor 1 did not load significantly, suggesting that these more directly inform a general factor. Furthermore, negative item loadings existed on Factor 1, which can happen unexpectedly in bifactor solutions (e.g., [@B16]) due to a crossover suppression effect ([@B71]). All items (apart from 6 and 13) loaded above 0.32 on Factor 2. However, direct comparison of factor loadings and observation of the average weights (specifically, Factor 1 = −0.269; Factor 2 = 0.376; Factor 3 = 0.507) suggested that, overall, items loaded more highly on a general factor. Discrete variance does appear to exist, particularly in the case of Factor 2. Lastly, CFA analysis considered a unidimensional model (as a null test of whether a single factor explains sufficient variance; [@B29]). Unsatisfactory fit existed on all criteria but SRMR ([Table 3](#T3){ref-type="table"}). This indicated that a single factor solution did not represent a good fit to the data. In addition to better data-model fit across indices, AIC supported superior fit of the bifactor model, given this was lower (229.672) than the two-factor (288.053) and unidimensional (391.609) solutions. [Table 3](#T3){ref-type="table"} demonstrates fit of the reliability models for the bifactor solution. The congeneric model fitted the best. This is unsurprising, however, given this is the least restrictive and avoids assumptions about constant means and variances ([@B31]). The tau equivalent and congeneric model exhibited a significant chi-square difference, χ^2^ (*df* = 23) = 87.174, *p* \< 0.001, yet CFI and SRMR differences less than 0.05 existed. Similarly, a significant chi-square difference occurred for the parallel and tau equivalent models, χ^2^ (*df* = 11) = 31.403, *p* \< 0.001, but analysis evidenced CFI and SRMR differences below recommended cut-offs. To exercise caution, however, analysis considered alpha as a lower bound estimate of the CTPI's reliability, with omega representing a more accurate model-based interpretation of scale reliability ([@B16]). Alpha was good for the total scale (α = 0.854), Feeling Harried (FH) (α = 0.795), and Cognitive Awareness of Time Shortage (CA) (α = 0.800). Coefficient omega conveyed similar (yet slightly higher) outcomes: good reliability for a general factor (ω = 0.878), for FA (ω = 0.819), and for CA (ω = 0.809). Omega hierarchical was high for a general CTPI factor (ω*h* = 0.702); however, lower estimates existed for FH (ω*h* = 0.133) and CA (ω*h* = 0.341). Common variance (ECV) exhibited comparable results, as total CTPI accounted for 66.9% whereas FH and CA explained 11 and 22.1% respectively. The percentage of uncontaminated correlations (PUC) was 51.3%. [@B77] advise that if PUC \< 0.80 and ECV \> 0.60 and ωh \> 0.70, then a scale can be interpreted as largely unidimensional. A test of convergent validity with the PSS-10 reported large correlations between total CTPI with total PSS-10, PSS-10 Distress, and PSS-10 Coping factors ([Table 4](#T4){ref-type="table"}). Similarly, FH evidenced large correlations with total PSS-10, PSS-10 Distress, and PSS-10 Coping. CA demonstrated moderate correlations with the PSS-10 outcomes. Examining associations between age and gender with CTPI, CTPI subfactors, PSS-10, and PSS-10 subfactors indicated significant (albeit small) positive correlations between gender, total PSS-10 and PSS-10 Distress. Further inspection revealed that females reported higher average levels of total PSS and PSS-10 Distress (Males: total PSS = 28.337, PSS-10 Distress = 17.297; Females: total PSS = 30.217, PSS-10 Distress = 18.862). This finding is consistent with previous studies investigating the PSS-10 (see [@B27]). In addition, females indexed greater means of CTPI, FH and CA than males (though this was non-significant). FH and all PSS-10 variables demonstrated small negative correlations with age ([Table 4](#T4){ref-type="table"}). ###### Intercorrelations among total CTPI, CTPI subscales, total PSS-10, PSS-10 subscales, gender and age for Study 1 and Study 2. **Variable** ***M*** ***SD*** **1** **2** **3** **4** **5** **6** **7** **8** -------------------------------------------- --------- ---------- ------- ----------- ----------- ------------ ----------- ----------- ---------- ------------- **Study 1** \(1\) Total CTPI 43.107 8.071 0.847^∗∗^ 0.910^∗∗^ 0.600^∗∗^ 0.599^∗∗^ 0.500^∗∗^ 0.079 0.030 \(2\) Feeling Harried 16.281 4.00 0.552^∗∗^ 0.596 ^∗∗^ 0.588^∗∗^ 0.508^∗∗^ 0.088 0.122^∗^ \(3\) Cognitive Awareness of Time Shortage 26.825 5.138 0.479^∗∗^ 0.482^∗∗^ 0.390^∗∗^ 0.046 --0.097 \(4\) Total PSS-10 29.870 7.418 0.967^∗∗^ 0.888^∗∗^ 0.098^∗^ −0.169^∗^ \(5\) Distress 18.573 5.074 0.741^∗∗^ 0.120^∗^ --0.191^∗∗^ \(6\) Coping 11.296 2.829 0.043 −0.101^∗^ \(7\) Gender 1.820 0.388 −0.098^∗^ \(8\) Age 26.200 11.799 **Study 2** \(1\) Total CTPI 42.690 7.673 0.856^∗∗^ 0.926^∗∗^ 0.585^∗∗^ 0.576^∗∗^ 0.475^∗∗^ 0.131 0.024 \(2\) Feeling Harried 16.831 3.607 0.599^∗∗^ 0.533^∗∗^ 0.507^∗∗^ 0.469^∗∗^ 0.131 0.070 \(3\) Cognitive Awareness of Time Shortage 25.858 4.948 0.518^∗∗^ 0.524^∗∗^ 0.396^∗∗^ 0.099 --0.045 \(4\) Total PSS-10 31.089 7.079 0.962^∗∗^ 0.858^∗∗^ 0.174^∗^ --0.021 \(5\) Distress 19.438 4.992 0.685^∗∗^ 0.207^∗^ --0.013 \(6\) Coping 11.651 2.654 0.076 --0.033 \(7\) Gender 1.820 0.389 −0.165^∗^ \(8\) Age 19.150 2.886 ∗ Indicates p \< 0.05; ∗∗ indicates p \< 0.001. Study 2 {#S3.SS2} ------- Replication of the superior bifactor solution from Study 1 ([Figure 2](#F2){ref-type="fig"}) resulted in satisfactory to good fit. Factor loadings ([Table 2](#T2){ref-type="table"}) followed a similar pattern to Study 1. Specifically, direct comparison across factors and observation of average weights (Factor 1 = 0.138; Factor 2 = 0.337; Factor 3 = 0.497) suggested that a general factor possessed higher factor loadings overall. In addition, Factor 2 accounted for a reasonable quantity of discrete variance. {#F2} Multi-group analysis comparing Study 1 and Study 2 reported good model fit at the configural stage ([Table 3](#T3){ref-type="table"}). At the metric level, a satisfactory CFI difference of 0.01 existed alongside an RMSEA difference of 0.001. Testing scalar invariance reported acceptable CFI (0.006) and no change in RMSEA. At the residual invariance level, satisfactory CFI and RSMEA differences existed (0.002 and 0.001 respectively). Findings support invariance of factor structure, loadings, intercepts and residuals. Consistent with Study 1, alpha reliability was good for the total scale (α = 0.858), FH (α = 0.820), and CA (α = 0.759). Coefficient omega additionally indicated good reliability for these components (general factor ω = 0.863; FH ω = 0.799; CA ω = 0.778). For a general Chronic Time Pressure factor, omega hierarchical was high (ω*h* = 0.728); yet lower for FH (ω*h* = 0.039) and CA (ω*h* = 0.299). In terms of ECV, total CTPI explained 68.3%, and FH and CA accounted for 11.7% and 20.1%. PUC was identical to Study 1 (51.3%). These results support reliability outcomes from Study 1, and suggest that the CTPI is principally unidimensional according to [@B77]. Convergent validity analysis for Study 2 (i.e., compared with the PSS-10) demonstrated large correlations between total CTPI with total PSS-10 and PSS-10 Distress, and a moderate association with PSS-10 Coping ([Table 4](#T4){ref-type="table"}). FH and CA also evinced large associations with total PSS-10 and PSS-10 Distress, and a moderate correlation with PSS-10 Coping. Consistent with Study 1, gender evidenced small (yet significant) positive correlations with total PSS-10 and PSS-10 Distress only. Females reported higher mean levels of total PSS and PSS-10 Distress (Males: total PSS = 28.498, PSS-10 Distress = 17.270; Females: total PSS = 31.674, PSS-10 Distress = 19.927). Females indexed greater averages of CTPI, FH and CA than males (though non-significant). Age, however, did not demonstrate any significant associations with CTPI or PSS-10 variables ([Table 4](#T4){ref-type="table"}). Discussion {#S4} ========== Psychometric evaluation of the Chronic Time Pressure Inventory (CTPI) revealed that the scale reflected a general dimension comprising two discrete, but overlapping temporal factors: Cognitive Awareness of Time Shortage (CA) and Feeling Harried (FH). Consistent with [@B88], CA and FA indexed negative features associated with the subjective experience of time shortage. These specifically referenced worry, feeling rushed and a sense of pressure. Congruent with this notion, analysis found positive correlations in the medium to large range between CTPI measures (overall and factors) and perceived stress. Concomitantly, items centered thematically on lack of control and the inability to schedule and complete tasks. This conceptualization is consistent with preceding research, which has reported an association between lack of apparent control and the negative effects of time pressures ([@B91]). Indeed, feelings of control can generally decrease the negative effects of time pressure, although this relationship varies as a function of situation (i.e., workload) and individual differences (i.e., level of neuroticism) ([@B91]). The emergent factors CA and FH shared features with the notions of 'time shortage' (perceived lack of time), and 'being rushed' (sense of time passing quickly) outlined by [@B88]. Although, FH and CA correlated positively (these factors shared between 30 and 36% variance) there was considerable theoretical divergence. Thus, despite possessing two items referencing 'hurry' and 'pressure,' CA aligned with time shortage. Explicitly, CA comprised items that were generally synonymous with time management issues. Specifically, reflected judgments about perceived absence of time, especially insufficiency (e.g., "There aren't enough hours in the day"). Thus, the emphasis with CA was objective awareness of time shortage rather than negative, affective consequences of time pressure. Contrastingly, FH more closely aligned to Szollo's delimitation of feeling rushed. Particularly, FH items indexed subjective feelings of apprehension, worry, anxiety and frustration arising from the experience of the perceived rapid passage of time (e.g., "The days fly by without me ever getting anything done" and "I feel rushed to do the things that I have do"). Although these observations are intuitively consistent, it is important to interpret them cautiously because the CTPI possessed a complex structure, which best fitted a bifactor solution. This indicated the presence of a latent structure where items loaded on a general factor indexing negative features associated with the subjective experience of time shortage. In practice, findings recommend the use of total scores rather than independent subscales when assessing general population samples as the CTPI is, for the most part, unidimensional. A degree of non-redundant variance existed, however, particularly for CA. Thus, the subscales can be used when administering the scale, but in the company of general scores. This inference accords with existing research concerning bifactor solutions (e.g., [@B62]). In addition, omega reliability indicated that the CTPI was reliable and that a general factor accounted for a sufficient proportion of variance in scale scores. Therefore, in terms of administering the CTPI in practice, the authors recommend summing the values of the Likert scale to form composite scores to represent levels of chronic time pressure, with the subscales usable for information purposes. Furthermore, although CA in particular possessed unique variance the subfactors were orthogonal and accordingly demonstrated conceptual overlap. The overall analysis was compatible with the notion that the concept of time pressure is an overarching designation that subsumes terms related to time shortage and being rushed. Additionally, the CTPI demonstrated invariance at the strictest level across the two studies. This indicated that the scale measured the same construct across the two studies without any notable measurement bias ([@B36]). While this study supported the presence of factors that corresponded with [@B88] conceptualization and evidenced factorial validity, further research is required to establish fully the legitimacy of the proposed theoretical distinction. Ensuing work on CTPI needs also to assess the temporal stability of the measure to ensure that it possesses test--retest reliability. Supplementary evidence supporting the psychometric robustness of the CTPI was apparent. Explicitly, the measure demonstrated content-related validity. Concerning face validity, scrutiny of the items by an academic panel ensured that the CTPI accurately assessed core elements of perceived time shortage (see [@B60]). Additionally, CTPI analyses indicated convergent validity; the CPTI strongly positively correlated with the PSS-10. The observed relationship was similar to those noted in other related studies (e.g., [@B1], time pressure and stress; [@B49], time management pressures/issues and perceived stress). Following studies should seek to establish concurrent validity by comparing the performance of the CPTI alongside similar extant measures. Although, myriad studies investigated time pressure these have used a range of instruments. Hence, consideration of the CTPI alongside a subset of these would usefully help to establish the scale's psychometric credibility. In this context, studies could examine CTPI performance alongside the Time Pressure Scale ([@B79]) and the single-item measure used by [@B42] (i.e., "How often do you feel rushed or pressed for time?"). Another related extension to the present paper could examine the extent to which CTPI predicted scores on time pressure-related variables, e.g., "Lack of efficiency" and "Forced to cut down on lunch time" (cf. [@B92]). Limitations and Future Research {#S4.SS1} ------------------------------- Several limitations existed. Notably, although the studies utilized samples comprising a range of occupations, both samples were predominantly female. In addition, the age range was rather restrictive with relatively youthful mean ages and the majority of participants being under the age of 35 (80% for Study 1; 98% for Study 2). Study 2 sample also contained a preponderance of university students (71%). These features limit the generalizability of the results to samples of various ages and replication is required with populations that are more heterogeneous. A second limitation relates to use of self-report data, which is associated with recognized limitations including response bias ([@B27]). Including supplementary assessment methods (e.g., physiological assessment), when measuring chronic time pressure would be useful in future. Lastly, as aforementioned analysis did not include test-retest reliability. The present study did adopt, nonetheless, certain strategies to circumvent typical issues inherent with the use of a single time point (a cross-sectional design). Specifically, this approach is frequently criticized because it can result in common method variance (CMV) and the inability to draw causal conclusions ([@B87]). The present study minimized the possibility of CMV by using procedural remedies ([@B52]). Importantly, the researchers created methodological separation between the CTPI and the PSS-10. Particularly, the authors made it clear within the participant instructions that the measures assessed different constructs. Furthermore, the two measures used dissimilar response scales. These factors created psychological distance between the CTPI and the PSS-10 ([@B72]). Finally, the researchers reduced the likelihood of evaluation apprehension and social desirability effects by providing instructions that emphasized that there were no were no right or wrong answers and that respondents should answer questions as honestly as possible. The notion of causality was not important in the context of the present study, as assessment of the CTPI against a criterion (PSS-10) was correlational in nature and intended to assess convergence. Relatedly, the strength of correlations suggested that the scale was not simply indexing stress, but captured additional variance. Accordingly, subsequent research may wish to examine the extent to which time pressure causes stress. This is important because significant previous research acknowledges that, whilst stress and time pressure result from an interaction between the individual and the environment, there are circumstances where the assumption is that time pressure causes ensuing stress. This is certainly true within medical literature, which attributes high stress and contemporary stress-related conditions (e.g., hypertension) to time shortage/pressure ([@B28]). A way to determine the direction of the chronic time pressure-stress relationship is to conduct a longitudinal study with multiple intervals. This approach, combined with sophisticated analytical techniques such as latent growth curve modeling, would determine how chronic time pressure and stress change (or remain stable) as a function of time. The outcome would inform the development of interventions designed to alleviate the negative effects of these factors. Relatedly, future research should consider also the degree to which perception of chronic time pressure varies as a function of context (i.e., work vs. home life). It may be that the situational factors within occupational settings, such as deadlines, targets and organizational level may exacerbate feeling of time pressure and accordingly have a more negative effect on the individual. Alongside this, studies could evaluate the mediating/moderating influence of individual differences. Illustratively, work historically has demonstrated that Type A personality (hard driving, persistent, involved in work) is associated with stress-related illness ([@B10]). From this perspective, an evaluation of the effects of time management training may also prove informative. In terms of applications, the CTPI provides a solid platform for further empirical investigation of time pressure at an individual and societal level. Explicitly, it provides an expedient measure, which researchers can use in myriad contexts (occupational, educational, health, etc.) for complementing understanding of stress, wellbeing, life satisfaction and work-life balance. In terms of implications, the CTPI provides a solid platform for further empirical investigation of time pressure at an individual and societal level. Explicitly, it provides an expedient measure, which researchers can use in myriad contexts (occupational, educational, health, etc.) for complementing understanding of stress, wellbeing, life satisfaction and work-life balance. Specifically, the measure will assist researchers to identify groups that are vulnerable to chronic time pressure. This information will usefully inform policymaking and facilitate the design and implementation of appropriate health policies and interventions. This is important because preceding research has found that time pressure varies as a function of role. For example, [@B69] reported that managing multiple roles (i.e., child rearing, income-earner, and a caregiver) was a cause of time pressure in a sample of Australian women born between 1973 and 1978. This finding accorded with [@B48], who noted that time pressure spanned contextual boundaries (e.g., home and at work). In this instance, strategies to ameliorate the negative effects of time pressure should focus on balancing diverse demands. This could involve identifying key stress points/times and making best use of available resources (i.e., health policies, family friendly leave and child-care policies) ([@B69]). Using the CTPI to identify groups experiencing chronic time pressure recognizes the subjective nature of time pressure and the fact that health risks vary across contexts. Illustratively, time pressure resulting from professional activities (i.e., time constraints, challenges and uncertainties) produced negative emotions in hospital-in-the-home nurses that led them to take more risks on the road ([@B18]). Furthermore, development of occupation specific coping strategies is important because rigid and inflexible coping patterns can exacerbate the risk of stress and health disorders ([@B50]). Data Availability Statement {#S5} =========================== The datasets generated for this study are available on request to the corresponding author. Ethics Statement {#S6} ================ The studies involving human participants were reviewed and approved by The Manchester Metropolitan University Faculty of Health, Psychology and Social Care Ethics Committee. The patients/participants provided their written informed consent to participate in this study. Author Contributions {#S7} ==================== AD: questionnaire development, theoretical focus, analysis and write-up. ND: questionnaire development, theoretical focus, and write-up. Conflict of Interest {#conf1} ==================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. ###### Chronic Time Pressure Inventory. Instructions: Here are a number of statements that focus on whether you feel you have enough time in your life to do the things you want to do. For each statement, select the response (from 1 "*Strongly disagree*" to 5 "*Strongly agree*") that most accurately captures your thoughts or feelings. **Strongly disagree (1)** **Disagree (2)** **Neither disagree nor agree (3)** **Agree (4)** **Strongly agree (5)** ----------------------------------------------------------------- ------------------------------ ------------------------------ ------------------------------------ ------------------------------ ------------------------------ \(1\) There aren't enough hours in the day      \(2\) I have enough time to do the things that I want to do (R)      \(3\) I feel pressured to fit everything in      \(4\) The days fly by without me ever getting everything done      \(5\) I am often in a hurry      \(6\) I feel in control of how I spend my time (R)      \(7\) I should have more free time to do the things I enjoy      \(8\) I worry about how well I use my time      \(9\) I have enough time to properly prepare for things (R)      \(10\) I think I won't finish work that I set out to do      \(11\) I feel disappointed with how I spend my time      \(12\) I always run out of time      \(13\) I feel rushed to do the things that I have to do      Summing all the items produces a total score. Items 4, 8, 10, 11, and 12 belong to 'Feeling Harried' (Factor 1); items 1, 2, 3, 5, 6, 7, 9, and 13 comprise 'Cognitive Awareness of Time Shortage' (Factor 2). Items with (R) need to be reverse-scored prior to analysis. [^1]: Edited by: Maicon Rodrigues Albuquerque, Federal University of Minas Gerais, Brazil [^2]: Reviewed by: Laiss Bertola, Federal University of Minas Gerais, Brazil; Valerio Capraro, Middlesex University, United Kingdom [^3]: This article was submitted to Quantitative Psychology and Measurement, a section of the journal Frontiers in Psychology

[^1]: Academic Editor: Fray Marshall

The worldwide incidence of thyroid cancer is increasing ([@bib6]) with considerable international variation in incidence ([@bib28]). Thyroid cancer is an indolent disease often detected by ultrasound of incidentally discovered nodules and this may be due to diagnostic bias rooted in differential access to health care. ([@bib3]). However, other studies suggest changes in risk factors such as iodine supplementation, obesity and the frequency of use of medical diagnostic radiation may be partly responsible ([@bib2]; [@bib32]). Studying incidence in different ethnic groups in a single country can help to understand this variation (and offer insights into aetiology) as similar diagnostic, reporting and registration procedures are used regardless of ethnic group ([@bib18]). In England, there were 2208 new cases of thyroid cancer in 2010 with incidence rates having increased by more than 150% since 1975 ([@bib16]). As a multiethnic nation (14% of England\'s population were 'non-White\' in 2011 ([@bib17]) with a unified health-care system, England provides an ideal setting in which to do this. Since 1995, self-assigned ethnicity has also been recorded in the National Health System\'s Hospital Episodes Statistics (HES) database (using the same classification system as used in the census) and HES records can now be linked to cancer registrations, providing more reliable information on ethnicity ([@bib26]) and allowing individual ethnic groups to be analysed separately for the first time ([@bib8]). In this paper, we compare the incidence of thyroid cancer among the six largest 'non-White\' ethnic groups in England with each other and with Whites using self-assigned ethnicity. Materials and methods ===================== We obtained data from the National Cancer Intelligence Network for all cancer registrations from 2001 to 2007 in England with the following information: cancer site coded to the International Classifications of Diseases, 10th Revision (ICD-10; [@bib30]); morphology coded to the ICD of Oncology, 2nd and 3rd Revisions (ICD-O-2 and ICD-O-3; [@bib29]; [@bib31]); deprivation assessed from the income domain of the Index of Multiple Deprivation 2007 (IMD 2007; [@bib14]); age at diagnosis of cancer; sex; ethnicity and regional cancer registry. To determine population incidence data, we used mid-year population estimates produced by the Office of National Statistics from 2001 to 2007 stratified by age, sex and ethnicity. We used ICD-10 code C73 to identify all thyroid cancers and morphology codes to identify follicular and papillary subtypes. NCRS obtained the self-assigned ethnicity for each cancer registration by record linkage to the HES database. We classified ethnicity as White, Indian, Pakistani and Bangladeshi, with the three groups combined to form the category 'South Asian\', Black African, Black Caribbean (again both combined to form the category 'Black\') and Chinese. We estimated age-standardised rates of cancer per 100 000 person-years for all ethnic groups using direct standardisation to the 1960 Segi world population ([@bib22]) with age divided into six categories: (\<40, 40--49, 50--59, 60--69, 70--79, 80+). We used Poisson regression to estimate incidence rate ratios (RRs) comparing each ethnic group (and 'South Asians\' and 'Blacks\') with Whites adjusting for sex, age and deprivation. We performed pre-specified subgroup analyses by sex (male *vs* female), age (\<50 *vs* ⩾50 years), deprivation (quintile 1 *vs* quintiles 2--4 of the income domain of the index of IMD 2007) and by tumour type (follicular *vs* papillary). We chose the age division so that cancer rates in first *vs* later generations of South Asians could be examined -- the percentage of South Asians born outside the United Kingdom is 97% for those aged ⩾50 years, whereas for those aged \<50 years the majority (58%) were born in the United Kingdom ([@bib15]). Subgroup analysis by age for Blacks and Chinese was also done for completeness but it does not allow the same discrimination by generation. When comparing 'South Asians\' and \'Blacks\' with Whites, we present results as RRs with 99% confidence intervals (owing to multiple tests performed across subgroups.) When comparing the individual ethnic groups, we use 99% floating confidence intervals (FCIs), calculated using the method of floating absolute risks, which enable valid comparisons between any two groups, even if neither one is the baseline ([@bib5]; [@bib20]). As ethnicity information was not complete for all registered cancers, we did a sensitivity analysis using multiple imputations of the missing ethnicity values age, sex, income domain of IMD 2007, site of cancer and region. Analyses were performed using Stata (version 12; StataCorp, College Station, TX, USA) and R statistical software packages (R Foundation for Statistical Computing, Vienna, Austria). Results ======= [Table 1](#tbl1){ref-type="table"} shows socio-demographic information from the 2001 census by ethnic group. All non-White groups are younger than Whites and all except Chinese are also more deprived, with Pakistanis, Bangladeshis and Black Africans being the most deprived. About half of the South Asian and Black Caribbean populations were born in the United Kingdom compared with only about 30% of Black Africans and Chinese. [Table 2](#tbl2){ref-type="table"} shows the total number of thyroid cancer registrations with missing ethnicity values for each subtype. For all thyroid cancers ([Figure 1](#fig1){ref-type="fig"}), there was a statistically significantly higher incidence in all ethnic groups (except Indians) compared with Whites, with significant heterogeneity between the groups (*P*\<0.001). Among South Asians, the rates were statistically significantly higher in both British Pakistanis (RR 1.79, 99% FCI 1.47--2.19) and British Bangladeshis (RR 1.99, 99% FCI 1.46--2.71), but not in British Indians (RR 1.09, 99% FCI 0.90 to 1.32), demonstrating heterogeneity between these groups (*P*\<0.001). In Blacks, the incidence of thyroid cancer was also statistically significantly higher in both Africans (RR 1.69, 99% FCI 1.34--2.13) and Caribbeans (RR 1.56, 99% FCI 1.25--1.93) but with no heterogeneity between these groups (*P*=0.5). The risk for thyroid cancer was highest in Chinese (RR 2.14, 99% FCI 1.63--2.80). The increased risk in the non-White ethnic groups was evident in men and women, in those aged \<50 and ⩾50 years and in those who were most deprived (quintile 1), as well as those in quintiles 2--5. However, as also shown in [Figure 1](#fig1){ref-type="fig"}, in South Asians the rate of follicular thyroid cancer was not statistically significantly higher than in British Whites, whereas the RR for papillary thyroid cancer was statistically significantly higher (RR 1.47, 99% CI 1.25--1.73). This difference is mainly because of the statistically significantly lower incidence of follicular thyroid cancer in Indians (RR 0.55, 99% FCI 0.31--0.98), whereas the incidence of both follicular and papillary thyroid cancers were statistically significantly higher in both the Pakistanis (follicular: RR 1.95, 99% FCI 1.29--2.96, papillary: RR 1.85, 99% FCI 1.46--2.36) and Bangladeshis (follicular: RR 3.15, 99% FCI 1.84--5.41, papillary: RR 1.63, 99% FCI 1.07--2.07). In Blacks, the incidence of both follicular and papillary thyroid cancers was statistically significantly higher than in Whites. However, the incidence rate ratios were statistically significantly higher in follicular (RR 2.09, 99% CI 1.53--2.86) than in papillary (RR 1.34, 99% CI 1.07--1.68), with significant heterogeneity between the two (*P*=0.003). The opposite pattern was seen in Chinese, with incidence rate ratios being statistically significantly higher for papillary cancer (RR 2.64, 99% FCI 1.94--3.58) than follicular cancer (RR 1.38, 99% FCI 0.68--2.83), again with significant heterogeneity between the two (*P*=0.03). In the sensitivity analysis, which assigned missing values using multiple imputation, results similar to those shown in [Figure 1](#fig1){ref-type="fig"} were obtained as shown in [Supplementary Figure 2](#sup1){ref-type="supplementary-material"} (online). Discussion ========== In this study, we compared, for the first time, incidence rates for thyroid cancers in the main 'non-White\' ethnic groups in England-South Asian (Indian, Pakistani and Bangladeshi), Black (African and Caribbean) and Chinese with Whites and with each other. There was considerable variation by ethnic group, even when gender, age and socio-economic factors are taken into account. Overall, the risk of thyroid cancer was significantly higher in all 'non-White\' ethnic groups except Indians, with the increased risk also seen in the subgroupings by gender, age, deprivation and histology. There were significant differences in the incidence of thyroid cancer among South Asians with the risk of both follicular and papillary cancer being higher in Pakistanis and Bangladeshis but not in Indians. The higher rate of thyroid cancer in Blacks was driven principally by an increased risk of follicular cancer, whereas in Chinese, the higher rate was due to an increased risk of papillary cancer. There is only one previous report of thyroid cancer incidence by ethnicity in England (using name analysis), which showed a higher thyroid cancer incidence in South Asians compared with non-South Asians, but only in females ([@bib27]). Studies from the United States have shown a lower incidence in African Americans compared with Whites ([@bib21]), in contrast to our findings, but also found the highest incidence in South East Asians and Chinese, consistent with our results ([@bib23]). The different patterns of cancer risk seen across each of the different ethnic groups as well as differences by sex, age, deprivation and tumour subtype suggest that our findings are unlikely to be due to systematic over-reporting of thyroid cancer in the ethnic minority groups. Our previous work using the same data set also showed reduced risks of gastrointestinal cancers in the same ethnic groups that further supports the absence of an over-reporting bias ([@bib1]). The differences we found are also between populations with equal access to health care ([@bib13]) and there is evidence that non-White ethnic groups are less likely to access services such as cancer screening ([@bib25]). It is, therefore, very unlikely that increased access (diagnostic bias) could explain the increased incidence in the non-White groups, although of course there may be other confounding factors, and studies with individual-level exposure are needed to address this. The environmental and genetic factors that lead to thyroid cancer are not fully known, but there are some established risk factors -- pre-existing thyroid disease, iodine status and exposure to radiation ([@bib11]). Insufficient iodine in the diet is associated with an increase in the risk of follicular thyroid cancer and it is therefore less prevalent in areas where fortification of salt with iodine is the norm. By contrast, a diet high in iodine, such as one rich in sea food, has been associated with an increased risk in papillary thyroid cancer ([@bib4]). In the United Kingdom, salt iodisation is long standing and there is no evidence of difference in iodine status by ethnic group and this is therefore unlikely to explain the ethnic variation. The reduced incidence of follicular thyroid cancer in Indians and increased risk of papillary thyroid cancer is striking (a similar pattern is also seen for Chinese) and would be consistent with increased iodine levels but there is no evidence of this. However, some groups -- Pakistanis, Bangladeshis and Blacks -- have an increased risk of both follicular and papillary cancers, and this cannot be explained by their iodine status. Other risk factors that are more contentious include an association between increased BMI and thyroid cancer, diabetes, female reproductive factors and exposure to endocrine-disrupting agents ([@bib9]; [@bib19]; [@bib32]). Although there are some differences in these risk factors (for example, obesity) by ethnic group ([@bib24]), it is unlikely that this could explain the significant differences in risk we have observed. Our finding of an increased risk in Blacks is in contrast to studies in the United States but this is likely to be mainly due to a reduction in the recording of thyroid cancer cases in African Americans owing to their inferior access to health care ([@bib10]) compared with US Whites. It could also reflect differences in the ancestry of the US and UK Black populations, with UK Blacks having migrated relatively recently and coming from the Caribbean and East Africa, whereas US Black immigration is more historic and largely from West Africa ([@bib12]). The increased risk we found in Chinese is consistent with studies in the United States where a higher frequency of thyroid nodules and goitre and reduced consumption of carotenoids explained more than half the increased risk ([@bib7]). Furthermore, the incidence of thyroid cancer in Hong Kong, where the majority of British Chinese originate, is even higher than that in British Chinese ([@bib6]). In contrast, rates in the countries of origin for all other ethnic groups in our study is much lower ([@bib6]). The main strength of our study is the use of a reliable and self-assigned measure of ethnicity. We also adjusted for socio-economic status, which is of particular importance for comparisons involving Pakistanis, Bangladeshis and Blacks owing to their higher levels of deprivation. The main limitation of this type of descriptive study is the lack of individual-level information available on exposures. Ethnicity information was also missing for 16.7% of cancer registrations but the similar results found in the sensitivity analyses suggest that this did not affect our results. In conclusion, the higher incidence of thyroid cancer in most ethnic minority groups compared with Whites, and the differences by subtype cannot be explained by known risk factors and requires further investigation. Establishing the determinants of this variation with individual-level data of exposures and prevalence of known thyroid cancer genetic risk factors could offer new insights into its aetiology. Our findings also have important public health implications; clinicians serving those areas with large non-White populations need to be aware of the increased risk and commissioners need to consider the implications of the increased thyroid cancer incidence for these areas. We thank the National Cancer Registration Service (NCRS) and the Office for National Statistics (ONS) for providing the data. **Disclaimer** The sponsor of the study had no role in design and conduct of the study; collection, management, analysis and interpretation of the data; and preparation, review or approval of the manuscript. **Author contributions** RA and IB conceived and designed the study. RA, IB, AF and SS contributed to the analysis and interpretation of the data. AF drafted the report, which was critically revised for important intellectual content by RA, IB and SS. All authors approved the report. RA is the guarantor. [Supplementary Information](#sup1){ref-type="supplementary-material"} accompanies this paper on British Journal of Cancer website (http://www.nature.com/bjc) RA, IB, AF, SS and VB are employed by the Cancer Epidemiology Unit at the University of Oxford, which is supported by Cancer Research UK. BM is employed by the Mayo Clinic in Rochester, MN, USA. Supplementary Material {#sup1} ====================== ###### Click here for additional data file. {#fig1} ###### Comparison of demographic characteristics by ethnic group in England in 2001     **White** **Indian** **Pakistani** **Bangladeshi** **Black African** **Black Caribbean** **Chinese** ------------------------------------- ------------ ----------- ------------ --------------- ----------------- ------------------- --------------------- ------------- --------- --------- --------- --------- --------- --------- **Census data for 2001** Total population 42 747 136 (100.0) 1 028 546 (100.0) 706 539 (100.0) 275 394 (100.0) 475 938 (100.0) 561 246 (100.0) 220 681 (100.0) **Sex** Male 20 828 644 (48.7) 511 204 (49.7) 358 043 (50.7) 138 972 (50.5) 229 103 (48.1) 259 881 (46.3) 105 913 (48.0) **Age** \<50 27 665 393 64.7 828 200 80.5 625 118 88.5 248 841 90.4 432 985 91.0 426 424 76.0 184 675 83.7 50+ 15 081 743 35.3 200 346 19.5 81 421 11.5 26 553 9.6 42 953 9.0 134 822 24.0 36 006 16.3 **Deprivation** Low income (quintile 1) 7 305 527 (17.1) 347 098 (33.7) 455 710 (64.5) 198 884 (72.2) 277 858 (58.4) 292 537 (52.1) 49 427 (22.4) Middle income (quintiles 2,3 and 4) 26 315 786 (61.6) 563 939 (54.8) 222 038 (31.4) 69 325 (25.2) 177 234 (37.2) 245 103 (43.7) 123 994 (56.2) High income 9 125 823 (21.3) 117 509 (11.4) 28 791 (4.1) 7185 (2.6) 20 846 (4.4) 23 606 (4.2) 47 260 (21.4) (quintile 5)                             **Country of birth** United Kingdom 41 911 150 (98.0) 472 545 (45.9) 387 198 (54.8) 127 902 (46.4) 161 050 (33.8) 324 764 (57.9) 62 209 (28.2) Other 835 986 (2.0) 556 001 (54.1) 319 341 (45.2) 147 492 (53.6) 314 888 (66.2) 236 482 (42.1) 158 472 (71.8) ###### Distribution of registered cancers from 2001--2007 in England by ethnic group (percentages in brackets)   **White** **Indian** **Pakistani** **Bangladeshi** **Black African** **Black Caribbean** **Chinese** **All other ethnicities** **No ethnicity recorded** **Total** ------------------- ------------- ------------ --------------- ----------------- ------------------- --------------------- ------------- --------------------------- --------------------------- ----------- All cancers 7396 (65.7) 178 (1.6) 170 (1.5) 70 (0.6) 124 (1.1) 142 (1.3) 90 (0.8) 1216 (10.8) 1877 (16.7) 11 263 Follicular cancer 1762 (70.7) 20 (0.8) 39 (1.6) 23 (0.9) 32 (1.3) 43 (1.7) 13 (0.5) 203 (8.2) 357 (14.3) 2492 Papillary cancer 4195 (63.8) 128 (2.0) 115 (1.8) 38 (0.6) 67 (1.0) 73 (1.1) 72 (1.1) 808 (12.3) 1076 (16.4) 6572 Other cancer 1439 (65.4) 30 (1.4) 16 (0.7) 9 (0.4) 223 (1.1) 26 (1.2) 5 (0.2) 205 (9.3) 444 (20.2) 2199

There is mounting evidence to support increased thrombo-embolic risk associated with coronavirus disease 2019 (COVID-19) infection. Here, we report a case of COVID-19 complicated by a right ventricular apical thrombus and aortic thrombi. A 61-year-old woman with type 2 diabetes mellitus, hypertension, and hyperlipidaemia presented to the emergency department (ED) with severe sharp epigastric pain and shortness of breath, with an oxygen saturation of 85%. In the ED, her baseline ECG and physical exam were unremarkable but she was placed on 6 L of oxygen due to persistent hypoxia. Subsequent arterial blood gas analysis revealed a P~a~O~2~ of 69 mmHg, P~a~CO~2~ of 31 mmHg, pH of 7.49, and sO~2~ of 93%. Initial chest radiograph demonstrated multifocal bilateral patchy opacities, and reverse transcription--PCR (RT--PCR) tested positive for SARS-CoV-2. Laboratory tests revealed mild leukocytosis (13.8 × 10^3^/μL), elevated D-dimer (8264 ng/mL), elevated fibrinogen (682 mg/dL), and elevated anticardiolipin IgM (31.1 MPL) with normal anticardiolipin IgG. CT pulmonary angiogram and CT of the abdomen/pelvis demonstrated bilateral peripherally located ground-glass and consolidative opacities without evidence of pulmonary embolism (*Panel A*). There was a prominent hypoattenuating lesion in the right ventricular apex along the interventricular wall representing thrombus (*Panel B*). Additional rounded filling defects were noted in the thoracic and abdominal aorta, suggestive of vascular thrombi (*Panels C* and *D*). Transthoracic agitated saline intravenous echocardiography confirmed the presence of mobile thrombus in the right ventricular apex without evidence of an intracardiac shunt (*Panel E*). The patient was treated with intravenous tissue plasminogen activator for thrombolysis and was subsequently transitioned to anticoagulation (low molecular weight heparin followed by oral apixaban). After 8 days of hospitalization, the patient was discharged home in stable condition. Coagulopathy and vascular endothelial damage have been suggested as potential complications of COVID-19 infection. Several reports have shown increased risk for venous thrombo-embolism; in four retrospective studies, the risk of thrombo-embolism ranged from 25% to 69%. The present case raises concern that thrombo-embolic risk associated with COVID-19 infection may not be limited to microthrombosis of small to medium size vessels. Left unrecognized, intracardiac and large vessel thrombi can lead to significant morbidity and mortality. Learning pointsSmall vessel thrombosis is a well described risk in the context of COVID-19 infection, yet there is little discussion of large vessel thrombo-embolism in these patients.Our case is a patient with no history of hypercoagulability or embolism who presented multiple large vessel thrombo-emboli in the context of COVID-19 infection. **Consent:** Patient informed consent was obtained for publication of this case report. {#ytaa199-F1} **Conflict of interest:** B.P.L. is a textbook author and editor for Elsevier and receives royalties for his previous work. [^1]: Nicholas Reid, Denston Carey and Min Lang contributed equally to this work.

Introduction ============ Mosquitoes including *Culex pipiens* are responsible for several serious diseases and able to transmit diseases agents infecting over 600 million people per year[@R1],[@R2]. *Culex pipiens pipiens* are known by their ecologic plasticity which explains their wide geographical distribution (temperate and tropical regions). This species may be the cause of strong nuisance and can transmit several parasitosis such as the Japanese Encephalitis, the West Nile fever[@R3]--[@R6], the Rift Valley virus and certain filariasis[@R7]--[@R11]. *Culex pipiens* mosquito has been strongly suspected as the most likely vector in the transmission of West Nile virus outbreaks that have affected Tunisia[@R12]--[@R15] in 1997, 2003, 2007, 2010, 2011 and 2012. This situation forces us to investigate one of the major obstacles to vectors disease control, the mosquito resistance to insecticides since the use of chemical insecticides for mosquito control remains the most widespread. In general, resistance to organophosphates including temephos insecticides can be metabolic or due to target site modifications. For target-site resistance, several authors have been identified and reported a relatively small number of highly conserved point mutations in the gene encoding the acetylcholinesterase (AChE) enzyme of many species[@R16],[@R17]. For metabolic resistance, other studies on a range of insect species identified and reported genomic changes which lead to gene amplification, over expression and/or modification of genes encoding members of the glutathione S-transferases (GSTs), cytochrome P450 (CYTP450) and carboxylesterases (CEs)[@R18]. Unfortunately, the massive use of insecticides during the malaria eradication program between 1967 and 1978 led to the development of strong resistance worldwide in *Culex pipiens* from Tunisia[@R19]--[@R22]. The aim of the present study was to evaluate the status of temephos resistance in larval populations of *Culex pipiens pipiens* (Diptera: Culicidae) in Tunisia and estimate the involved mechanisms using different synergists. The cross-resistance between temephos and propoxur, and the polymorphism of over-produced esterases and AChE 1 were also investigated. Methods ======= Standard methods of Raymond et al.[@R23] were performed on larvae populations of *Culex pipiens pipiens* mosquitoes collected in three breeding sites in northern and Southern Tunisia ([Figure 1](#F1){ref-type="fig"}). Identification of collected populations was done morphologically using the key of Brunhes et al[@R24]. Collected larvae were transported to the laboratory and directly transferred into plastic trays containing distilled water with rabbit croquette which served as food under standard insectary conditions (25 ± 1°C and 70 ± 5% RH). Bioassays were performed on both early third and late fourth instars. However, young larvae were reared until advanced instars. Field collected strains were tested against temephos insecticide and compared to bioassays of a susceptible reference strain. Five concentrations of temephos (100, 10, 1, 0,1, 0,01 ppm) providing between 0 and 100% mortality were used in a total volume of 100 ml of water containing 1 ml of ethanol solution of the tested insecticide. The tests were replicated five times per concentration. In each replicate, 20 early third and late fourth instar larvae were released. After a period of 24 hours, larval mortality was recorded. The Mazzarri and Georghiou[@R25] criteria were followed to classify the resistance level of each population tested as follows: low (RR\<5), moderate (5≤RR≤10) or high (RR\>10). We assessed also the effect of the synergists piperonyl-butoxide (Pb) on metabolic insecticide resistance. Esterase\'s activities were characterized on homogenates of adult thorax and abdomen according to the method of Pasteur et al.[@R26]. Propoxur bioassays which included one dose (1mg/l) was done to estimate the common mechanism of resistance to temephos insecticide: acetylcholinesterase resistance. According the method described by Bourguet et al.[@R27], AChE1 polymorphism was analyzed to comparing AChE1 activity of homogenates of adult heads in the absence or presence of propoxur. Three phenotypes including ace-1R, ace-1S alleles, and duplicated haplotype were separated using this enzyme bioassay. The obtained results were analyzed by using the log probit program of Raymond[@R28], based on Finney[@R29] to obtain LC~50~, LC~95~ and regression line. {#F1} Results ======= The resistance to temephos insecticide in field populations of *Culex pipiens pipiens* are presented in [Table 1](#T1){ref-type="table"}. The resistance ratio (LC~50~) values of temephos ranged from 1.34 to 114 and showed their low and high levels according Mazzarri et al[@R25] criteria. ###### Temephos resistance characteristics of Tunisian *Culexpipienspipiens* Population Temephos Temephos +DEF Temephos +Pb -------------- ------------- ----------------------------------------- -------------- -------------- -------- ---- -------------- ---- ------------ ---------------------------------------------- -------------- -------------- ------- Slab 1,2 2,34 \- 0,32 4,99 \- 3,84 \- 2,2 1,94 \- 0,56 \- (1,1--1,4) ± 0,22 (0,28--0,36) ± 0,69 (2,89--5,09) (1,7--2,8) ±0,28 (0,44--0,72) 1-Sidi Hcine 142 2,41 114 \- \- \- \- \- 7,3 3,44 ± 2,48[\*\*](#TF1){ref-type="table-fn"} 3,32 19,4 34,33 (65,5--308) ± 0,75[\*\*](#TF1){ref-type="table-fn"} (60,2--216) (3--17) (1,12--9,84) (6,00--63,0) 2-El Fahs 3,1 1,69 2,55 \- \- \- \- \- \- \- \- \- \- (2,1--4,4) ± 0,31 (1,93--3,37) 3-Dgach 1,6 2,05 1,34 \- \- \- \- \- \- \- \- \- \- 0,98--2,8) ± 0,35 (0,87--2,05) (a), 95% CI; Parallelism test positif but without probability. RR~50~, resistance ratio at LC~50~ (RR~50~=LC~50~ of the population considered/LC~50~ of Slab); SR~50~, synergism ratio (LC~50~ observed in absence of synergist/LC~50~ observed in presence of synergist).RR and SR considered significant (P\<0.05) if their 95%CI did not include the value 1. RSR, relative synergism ratio (RR for insecticide alone / RR for insecticide plus synergist). The addition of Pb synergist in sample \# 1 increased the toxicity to used insecticide (RSR\>30), indicating that the resistance mechanisms inhibited by synergist (CYTP450) were involved in the resistance of the studied sample. Starch electrophoresis was realized to verify the involvement of detoxification (esterase) in the recorded resistance because it was not possible to do all synergists bioassays mainly DEF synergist. Two esterases were detected in the most resistant sample with high frequency of A4-B4 (45%). Three esterases including A4-B4, B12 and C1 were revealed in sample \# 2 with low frequencies. Low frequencies of two esterases including A4-B4 and C1 were reported in sample \# 3 which showed the lowest level of resistance to temephos and propoxur. Cross-resistance temephos/propoxur was detected using bioassays and the mortality due to propoxur was significantly correlated with the LC~50~ of temephos insecticide indicating the resistance of the common target site: AChE 1. The propoxur mortality was low in the most resistant population (32%). The high moralities, 38% and 99%, were associated to the susceptibility of samples \# 2 and 3, respectively. The polymorphism of AChE 1 showed that the highest frequencies of acetylcholinesterase 1 (AChE 1) resistant phenotypes was recorded in the most resistant population (sample \# 1) with a percentage of 74% including an important percentage of duplicate phenotype \[RS\]. However, the highest frequencies of acetylcholinesterase 1 (AChE 1) susceptible phenotypes was reported in the two susceptible populations (samples \# 2 and 3) with a percentages of 80 and 75%, respectively. Discussion ========== Unfortunately, the massive use of insecticides during the malaria eradication program between 1967 and 1978 has led to the development of strong resistance worldwide in *Culex pipiens* from Tunisia[@R19]--[@R22]. Furthermore, organophosphates actually remain one of the major tools for culicinae control including *Culex pipiens*. It is clear that closer collaboration between resistance experts in agriculture and public health is needed. Indeed and despite the absence of agricultural control in the studied areas, we can consider that the migration including passive transportation of mosquitoes and gene flow play major roles in the dispersion of resistance genes between distant populations[@R30]. Except the previous and ancient studies of Ben Cheikh and his collaborators[@R19]--[@R22],[@R31], there are no previous published studies about the resistance status of Tunisian *Culex pipiens pipiens* populations to chemical insecticides. Therefore, these populations of *Culex pipiens pipiens* need to be monitored for insecticide resistance in this area. Previous findings[@R19]--[@R22],[@R31] showed that organophosphate and carbamate resistance was widespread in Tunisian populations of *Culex pipiens*. Ben Cheikh et al[@R19] showed low resistance ratio levels of *Culex pipiens pipiens* to temephos insecticide, not exceeding 10-fold. However, high resistance level of this species to temephos was recorded in Tunisia and reached 400-fold[@R31]. The use of studied insecticides in public health explained the recorded resistance of *Culex pipiens pipiens*. Carbamate and organophosphate resistance was strongly correlated with the presence of an insensitive acetylcholinesterase which results in reduced sensitivity to inhibition of the enzyme. This target site resistance has been found and documented in many vectors including *Culex pipiens* and other mosquitoes[@R19],[@R32]--[@R34]. In Benin, a cross-resistance between organophosphatess (fenitrothion) and carbamates (bendiocarb and propoxur) in *Anopheles gambiae* was reported with the presence of insensitive acetylcholinesterase[@R35]. The polymorphism of AChE 1 showed that the highest frequencies of acetylcholinesterase 1 (AChE 1) resistant phenotypes was recorded in the most resistant population with an important percentage of duplicate phenotype \[RS\]. This could be probably due to the fitness costs associated to different alleles. Indeed, the overall fitness advantage of the duplicated haplotype may result from a much lower fitness cost[@R36]. Esterases, CYTP450 and glutathione-S-transferases are the three detoxification enzymes known to confer resistance to insecticides in mosquitoes vectors for all major classes of insecticides currently used for vector control, including organochlorine, organophosphates, carbamates, and pyrethroids. In the present study, synergist bioassays and biochemical characterization showed that esterases and CYTP450 were involved in the recorded resistance. Our results are in agreement with previous studies on the role of the CYTP450, esterases and the glutathione-S-transferases in the organophosphate including temephos resistance[@R19],[@R37],[@R38]. However, authors showed that AChE 1 identified above is known to be a dominant feature in resistance insects[@R39]. Conclusion ========== In the present study, both resistant target site and detoxification enzymes were identified and therefore confer the recorded resistance to temephos and propoxur compounds in *Culex pipiens pipiens*. The selection pressure for resistance could have risen mainly from the use of these insecticides in agriculture, as well as in public health. Given the increase of resistance to temephos insecticide in *Culex pipiens pipiens* compared to previous studies, there is an urgent need for field and laboratory monitoring of insecticide resistance. This work was kindly supported by the Ministry of Higher Education and Scientific Research of Tunisia by funds allocated to the Research Unit (Génétique 02/UR/08-03) and by DHMPE of the Minister of Public Health of Tunisia. We are very grateful to S. Ouanes, for technical assistance, A. Ben Haj Ayed and I. Mkada for help in bioassays, S. Saïdi, Tunisian hygienist technicians for help in mosquito collecting, and M. Nedhif and M. Rebhi for their kind interest and help. Conflict of interest ==================== The authors declare that they have no conflict of interest. [^1]: Ahmed Tabbabi and Jabeur Daaboub contributed equally to this work.

Objective ========= Reduction of β-cell mass is increasingly recognized as one of the main contributing factors to the pathogenesis of type 2 diabetes. Chronic free fatty acid (FFA) exposure has been shown to induce endoplasmic reticulum (ER) stress that may contribute to promoting pancreatic β-cell apoptosis. In the present study, we first investigated whether anticonvulsant sodium valproate (VPA), at clinically relevant doses, protects pancreatic β-cell from palmitate-induced apoptosis and the mechanism underlying anti-apoptosis. Methods and results =================== INS1 cells exposed to 0.25\~1.0mM palmitate for 24\~48h under serum-free conditions showed marked apoptosis in time- and concentration-dependent as assessed by CCK-8 assay, Hoechst 33342 / PI, flow cytometric cell apoptosis assay and electron microscopy. Palmitate triggered ER stress and apoptosis in INS1 cells as evidenced by increased mRNA levels of C/EBP homologous transcription factor (CHOP), activating transcription factor 4 (ATF4) and X box-binding protein 1 (XBP-1) in a time-dependent fashion. Western blot analysis also showed significant increase of CHOP and caspase-3 in protein level. We also found that palmitate activated GSK3β by inhibiting phosphorylation at serine 9. While chronic, not acute, 1\~2mM VPA and 2mM LiCl remarkable reduced palmitate-induced cytotoxicity. Furthermore, INS1 cells treated with 10\~20μM TDZD-8, a specific GSK3β inhibitor, also elicited cytoprotective responses against 0.25\~0.5mM palmitate for 6\~48h and decreased mRNA level of CHOP, but not ATF4 or XBP-1. The protein levels of CHOP, caspase-3 and GSK3β activity were remarkable reduced by co-treatment of INS1 cells with 0.25mM palmitate and 1mM VPA, compared with 0.25mM palmitate only. Finally, down-regulation of CHOP expression in INS1 cells by small interfering RNA (SiRNA) did not show apparent cytoprotective responses against 0.25mM palmitate. Conclusion ========== ER stress and GSK3β involved in palmitate-induced β-cell apoptosis, however, GSK3β other than ER stress is likely playing a more prominent role. Valproate protected pancreatic β-cell from palmitate-induced apoptosis and ER stress by inhibiting GSK3β.

{.155} {.156} {.157} {.158} {.159} {.160} {.161} {.162} {.163} {.164} {.165}

Introduction {#s1} ============ Depression has become one of the leading causes of disability worldwide. In 2015, 7.5% of depression patients lived with disability ([@B1]). However, many people living with this condition are not aware of their illness. In some cultures, it is also considered shameful to disclose one's mental health problems to family members or even doctors, partially due to the tradition of presenting a positive self-image to others ([@B2]). Similar to many other countries, China is facing a severe shortage of mental health professionals; China has 23,000 psychiatrists---about 1.7 for every 100,000 participants ([@B3]). The social stigma related to cultural and moral beliefs also deters people in China from seeking treatment. Therefore, it has become essential to help individuals be aware of their symptoms so that they can decide when to seek support from professionals. In the last decade, researchers have begun to explore the possibility of using digital footprints to monitor social media users' depression symptoms because social media data have provided a record of users' emotional and behavioral patterns. In this work, we introduce a new approach to analyze the affect patterns disclosed on social media posts and explore how the social media affective language is associated with depression symptoms. In psychology, affect refers to nonreflective feelings towards stimuli, e.g., the feelings of pleasure or tension ([@B4], [@B5]). Reduced positive affect (PA) and increased negative affect (NA) have been found to be classic markers of major depressive disorder (MDD) ([@B6], [@B7]). Increased NA signals negative interpretation bias and negative repetitive thinking, and decreased PA indicates a reduction of interest or pleasure in response to stimuli. Existing empirical studies often examine people's affect using measurement scales or professional interviews ([@B8]). The emergence of sentiment analysis provides an alternative way to study affect. Sentiment analysis is a process of extracting affective words or phrases from text. It has been found that affect, especially NA expressed in social media text, reflects social media users' psychological characteristics and mental health status ([@B9]--[@B11]). NA is a summary of a variety of negative emotions, including anger, sadness, fear, etc. Findings from both empirical studies and social media data have shown that users with high depression symptoms tend to use more words/phrases containing NA ([@B12], [@B13]). Why does the use of NA words relate to depression symptoms? On the one hand, depressed individuals tend to have cognitive vulnerabilities, which are cognitive processing biases in attention, memory, interpretation, and repetitive negative thoughts ([@B14]). For example, "No one cares about my problem," is a negative cognitive bias. On the other hand, frequent occurrences of NA reflect dysfunctional stress response. Individuals with a dysfunctional stress response system often fixate on the causes, consequences, and meanings of stressors, which results in "stress-reactive rumination," a passive comparison of one's current situation with some acknowledged standard ([@B15]). Contrary to NA, PA is in general beneficial to health and cognitive function ([@B16]). Studies have found that nondepressed individuals are often motivated to downregulate negative emotions and upregulate positive ones ([@B17], [@B18]), but depressed individuals usually experience reduction of pleasure ([@B19]). Similar to the empirical findings, social media users also tend to post positive content to seek social approval and/or form positive impression ([@B20], [@B21]). However, Locatelli and colleagues have recently found that people with depression symptoms also use more positive affective expressions in social media ([@B19]). Accordingly, they hypothesize that the relationship between PA and depression symptoms is possibly mediated by rumination. Although there exists a large amount of evidence to support that affect expressed in social media texts can reflect mental health status, few of the studies examine the life stressors that may trigger NA and the fixation behavior. In addition, although NA has been extensively studied, there is a very limited amount of literature that examines the relationship between PA in social media content and depression symptoms. In order to fill these gaps in this line of research, this paper investigates the patterns of positive and negative affect, as well as the rumination language following a stressful life event, targeting a popular microblogging social media website in China: Weibo. Examining life stressors presents a challenging question: What kind of events are considered to be more stressful, as opposed to those that are less so? Some stressors are uniformly perceived as more damaging to mental health than others. By asking participants from diverse cultures to rate how much readjustment was required for 42 life events, Masuda and colleagues identified a set of life events that were perceived as detrimental to mental health ([@B22]). Among others, death of a spouse, divorce, and marital separation were ranked as the top three events requiring the most life readjustment. Later studies found similar rankings in the life events requiring much life adjustment, but different rankings in those requiring moderate to low levels of adjustment, e.g., being "fired" from work dropped from the 8th in ([@B22]) to the 47th in ([@B23]). Here, we focus on three life stressors that respectively bring severe (e.g., marital separation), medium (e.g., severe illness of a family member), and low (e.g., being fired from work) levels of impact to mental health. In light of the above discussions, this paper aims to address the following two research questions: RQ1: Does social media affect reflect cognitive vulnerability to depression symptoms?RQ2: What are the characteristics of stress-reactive affective content on Weibo? In particular, we focus on rumination (NA), and investigate what stressful life events people tend to ruminate on social media. For RQ 1, we first visualized the positive and negative affect patterns of Weibo users in multiple time windows. Then, we examined the relationship between social media affect and depression symptom scores when specific life stressors were presented to the users. For RQ 2, we examined the rumination in postevent reaction. Stress reactive rumination reinforces the interpretation bias of an individual, thus putting an individual at higher risk of developing depression symptoms. We looked into how social media users ruminated on specific life events and summarized the characteristics of the rumination. Contribution {#s1_1} ------------ This paper provides an opportunity to advance the understanding of how positive and negative affect reflects cognitive vulnerabilities to depression. By examining the stress reactive affective language on social media data, we seek to identify affective content that links to cognitive vulnerability. Addressing these issues would help better understand the pattern of affect in a social media text and its association with people's vulnerability to affective disorders in general. Moreover, by observing what types of stressors social media users tend to ruminate on, this paper offers essential insights into cognitive biases in social media data, thus promoting future research on life events and affect in the social media context to take into account these biases. Data and Data Collection {#s2} ======================== Weibo {#s2_1} ----- Weibo is a social media platform where users can publish a short piece of text, video, or photo to customized lists of friends or followers. Before 2016, users could write up to 140 words on each post. Since then, the maximum number of words per post has been increased to 2,000. In Weibo, users can follow or unfollow others, like or dislike others' posts, make comments to those posts, or share some of those posts to their social networks. By 2017, there were nearly 300 million users on Weibo, accounting for one fifth of the population in China ([@B24]). Multiple survey studies show that the majority of Weibo users are in their 20s and 30s ([@B25], [@B26]). Female users are more likely to mention they were diagnosed with depression than male users ([@B27]). In this study, we collected a sample of participants' Weibo posts, and assessed their depression symptoms using a depression symptom screening test and subjective stresses of daily lives. Data Collection {#s2_2} --------------- We posted a recruitment notice for this study on a personal Weibo account on June 10, 2016. The survey was open from June to September, 2016. A few science bloggers and entertainment bloggers reposted our recruitment notice voluntarily. We also promoted our survey with paid advertisements to increase participation rates while the survey was open. The study targeted users residing in China, aged over 18. Participants of the study were asked to complete a survey containing the Center for Epidemiologic Studies Depression (CES-D) scale and a stressful life event survey. Participants could optionally sign a consent form (see Appendix A) to allow us to collect and analyze their Weibo posts by computer programs. A total of 1,918 participants responded to the survey between June and September 2016. Among them, 1,629 allowed us to access their Weibo data. We used a custom Python script to automatically collect 198,485 Weibo posts from these 1,629 users. All the posts were posted from January 2009 to September 2016. Depression Symptom Screening Test {#s2_3} --------------------------------- We used a depression screening test, namely, the Center for Epidemiologic Studies Depression Scale (CES-D), to infer participants' depression symptom levels. The original CES-D was a 20-item self-reported scale designed to measure depression symptoms in the general public ([@B28]). We adopted the short form developed by Kohout ([@B29]) and translated into Chinese by Chin et al. ([@B30]). A back translation version of the short form is provided in Appendix A. This short form sacrifices little precision and taps the same symptom dimensions as the original CES-D. Both the original CES-D and the short form were tested in the Chinese population. The internal consistency of CES-D 10 in the Chinese population was satisfactory (Cronbach *α* = 0.78 ± 0.79). Reliability over a period of 3 years was also found to be significant (*r* = 0.44, p 50.01) ([@B31]). In the short form, participants were asked to rate the extent to which they experienced depression symptoms. An example item could be: "My sleep was restless." Responses are on a Likert-type scale, including 1 ("Rarely or none of the time"), 2 ("1--2 days last week"), 3 ("3--4 days last week"), 4 ("Every day"), and 5 ("Every day for at least two weeks"). Appendix A shows the standard Chinese questionnaire used in this study (with English translation). Stressful Life Event Survey {#s2_4} --------------------------- We examined the stressful life events recently encountered by participants before completing the survey. A major problem in assessing life stress is that whether an event is considered stressful or not could be very subjective. Some stressors can be motivating to one person but stressful to others. Therefore, we focused on objective stressful life events. After asking the participants whether they had experienced, up to 3 months before completing the survey, any life events that they perceived as stressful, we asked them whether they had recently experienced one of three specific stressful events: relationship breakdown, a family member/close friend being diagnosed with severe illness, or being fired from work. The first two were ranked within the top three to top ten in the Social Readjustment Rating Scale (SRRS) ([@B32]); and the last one was found to drop from the top 10 to the top 50 in a recent revisit to SRRS ([@B23]). In addition to these specific events, we provided the option of "others" on the survey, which allowed participants to include any events that they themselves perceived as significant life stressors. Methodology {#s3} =========== Summary Statistics {#s3_1} ------------------ To compare the patterns of affect score from participants with high level of symptoms with those with low level of symptoms, we split participants into two groups according to their CES-D survey scores using a cutoff point of 22, which has been applied in multiple studies ([@B9], [@B28], [@B33]). Here, we present the summary statistics of the two groups. [**Figure 1A**](#f1){ref-type="fig"} shows the basic statistics of users' CES-D scores, and [**Figure 1B**](#f1){ref-type="fig"} shows the statistics in the H-group and L-group. Internal consistency of the CES-D scale was high (Cronbach's *α* = 0.62). Participants posted more positive Weibo posts (*N*=50,731) than negative ones (*N*= 35,651). Their mean affect was negatively correlated with their CES-D scores (*r* = -0.13, *p \<* 0.001). {#f1} Computing Affect Score {#s3_2} ---------------------- We used a sentiment analysis service provided by Lexalytics (<https://www.lexalytics.com/>) to assign a continuous sentiment score ranging from 1 to −1 to each document (i.e., Weibo post). Lexalytics has performed satisfactorily compared with other popular sentiment annotation tools ([@B34]), such as OpinionFinder and Sentistrength. Lexalytics uses part-of-speech tagging to identify adjective-noun combinations, and then counts the number of affective words in a sentence. The algorithm adds weights to the word count according to a sentiment library developed by Lexalytics. The sentiment library contains an extensive collection of adjectives, each manually scored by human annotators according to their judgment of how negative or positive the word is. The sentiments of the words are inverted in the presence of negators (e.g., "not") or some conjunction (e.g. "but" and "however"). Lexalytics also accounts for multilayered sentiment; if a sentence contains both positive and negative affective words, the two types of words may cancel each other out, thus making the document neutral. Before computing the affect score, we preprocessed the Weibo posts following some simple procedures, including removing Email addresses and hyperlinks and encoding emoticons using descriptive words within square brackets, e.g., \[sad\]. Visualizing Affect Pattern {#s3_3} -------------------------- We visualized the affect pattern of each user over a time series in the unit of day, and applied a generalized linear model to smooth the time series. The timeline was aligned in a backward manner, with the day when users completed the CES-D scale as "Day 0" and the day before "Day 0" was "Day 1." Note that "Day" here is not a calendar day. A calendar day might include events that influence public affect in general. For example, extreme weather might lead to more NA, and holidays to PA. To reduce the noise from holidays, weather, and other confounding factors on calendar dates, we residualized the daily affect of each participant *vi* by subtracting it with the mean affect score of the whole sample on that day *µ*. Therefore, the adjusted post affect score would be *vi* − *µ*. We were interested in participants' affect patterns at different stages before they developed depression symptoms. Examining the stages presented a challenge: How to define the time window of each stage? Note that the self-reported score is not a gold standard for diagnosis, participants might develop the symptoms long before they completed the measurement scale. Therefore, we first defined the time window (T3) as Day 0 to Day 30 to observe the affect score while participants were experiencing high symptoms. Literature suggests that early symptoms happened in a time ranging between 6 weeks and 23 months ([@B35], [@B36]). Hence, we defined T2 as Day 0 to Day 365 to examine the development of early symptoms within a year. We were also interested in the affect pattern beyond the flare-up of symptom, so we set up T1 as Day 0 to Day 1095 to observe the longitudinal affect over the three years (see [**Figure 2**](#f2){ref-type="fig"}). Note that depression is a persistent condition that can last for years if left untreated, thus, some participants might have been living with symptoms for years. {#f2} Cognitive Vulnerability Analysis {#s3_4} -------------------------------- Individuals with cognitive vulnerability are more likely to develop depression symptoms if experiencing a stressful life event. Therefore, we conducted a within-subject correlation analysis between affect scores and CES-D scores before the occurrence of specific stressful life events. We divided the participants into two groups based on whether they had experienced certain life events in the recent 3 months. We focused on examining the events that could bring more severe impact on participants' depression symptoms. Among the 250 participants who answered the life event questions, 77 reported that they had encountered at least one stressful life event recently (CES-D median = 26) (see [**Table 1**](#T1){ref-type="table"}). Among them, those who reported having a breakup (CES-D median = 29) or being fired from work (CES-D median = 29) tended to develop more depression symptoms. Accordingly, we grouped the participants according to whether they had been through these two life events in the recent 3 months. ###### Center for Epidemiologic Studies Depression (CES-D) scores of the participants who experienced stressful life events. events N CES-D ----------------------- ---- ------- -------------- total 77 26.00 21.00--32.00 break up 33 29.00 24.00--34.00 illness 23 23.00 20.00--24.52 being fired from work 12 29.00 23.50--31.25 other 31 25.00 21.00--32.00 more than two 17 26.00 21.00--35.00 Since stressful life events occurred in the recent 3 months (90 days) prior to the completion of the survey, we selected three time windows: 90 days to 1 year, 90 days to 2.5 years and 90 days to 3 years before the completion of the survey. In each window, we computed the residualized daily mean affect score and conducted a correlation analysis between the mean affect score and CES-D score. Due to multiple correlation tests in the analysis, we used a permutation test to reduce the uncertainty of *p*. In the permutation test, the labels of the data were rearranged in each computation and the *p*-value achieved in the statistical test was estimated based on 20,000 simulations. Rumination Language {#s3_5} ------------------- Before analyzing the characteristics of rumination language, we first annotated the posts that contained rumination language following one of the three stressful life events (c.f Section 2.4). We selected the Weibo posts from 77 participants who reported having encountered at least one stressful life event in the recent 3 months, and focused on their Weibo posts between Day 0 and Day 90, because this is the time window closest to the time point when these participants self-reported their depression symptoms. Instead of using a keyword approach to capture the stressful life events, we manually annotated life events. In some cases, we identify life stressors from contextual information. For example, a post such as "She left me, my heart is broken" indicates a relationship breakup. Then, we analyzed whether these posts reflected any of the three types of cognitive tendencies ([@B37]), including the tendency to focus on negative attributions and inferences, the tendency to focus on hopeless thoughts, and the tendency to focus on coping strategies. To protect users' privacy, we removed the name entities and other sensitive information that might reveal the identities of these persons from the Weibo examples. Our annotation was carried out with an in-house, online annotation tool. Annotation guidelines for both rumination and stressful life events can be found in Appendix B. Two authors annotated life stressors. The interrater reliability was 0.80 for rumination language and 0.92 for life events. Appendix B shows the annotation guideline for both annotation tasks. Result: Cognitive Vulnerabilities {#s4} ================================= Visualizing Affect Patterns in Multiple Time Windows {#s4_1} ---------------------------------------------------- We computed the average affect score of each individual and examined the affect patterns in the three time windows. [**Table 2**](#T2){ref-type="table"} shows the statistics of the affect scores in the H-group and L-group, respectively. It is evident that the H-group consistently shows lower affect in T1, T2, and T3. ###### Basic statistics of affect and Center for Epidemiologic Studies Depression (CES-D). Time H-group L-group all participants ------ --------- --------- ------------------ ---- -------- ----- ------- ------ ---- -------- ------- ------ T1 781 −0.009 0.11 27 24--34 861 0.01 0.09 27 24--34 0.00 1.32 T2 732 −0.002 0.07 19 18--20 678 0.006 0.06 19 18--20 0.002 1.55 T3 794 −0.02 0.05 27 24--34 500 0.03 0.06 19 18--20 0.005 2.22 M, mean; sd, standard deviation; N, number of participants in each time window; T1, three-year time window; T2, one-year time window; T3, 30-day time window; Affect, mean affect score on a day. We plotted the participants' affect against the time (see [**Figure 3**](#f3){ref-type="fig"}). Note that Day 0 was the day when the participants completed the CES-D survey. We conducted a Welch t-test to detect the significance of the affect differences between the two groups, and the *p*-values were adjusted following the Bonferroni correction. {#f3} We found distinguished affect patterns between the two groups. As for NA, the H-group showed significantly higher NA than the L-group in T1 (3 years) (*t*(2044) = 4.65, *p \<* 0.001, Cohen's *d* = 0.20) and T3 (30 days) (*t*(58.6) = 3.45, *p \<* 0.05, Cohen's *d* = 0.88), but not in T2 (1 year) (*t*(724.7) = 1.05, *p \>* 0.05, Cohen's *d* = 0.13). As for PA, the H-group showed significantly higher PA than the L-group in T2 (*t*(721.3) = 1.98, *p \<* 0.05, Cohen's *d* = 0.14), but not in T1 (*t*(2046) = 1.76, *p* = 0.07, Cohen's *d* = 0.08) and T3 (*t*(59.8) = 0.44, *p* = 0.6, Cohen's *d* = 0.11). The persistently high level of NA signals a negative cognitive bias in the H-group. The elevated level of PA might indicate that the participants in the H-group incorporated various coping skills to tackle life stress. However, this coping ability might be impaired when an individual has already developed severe depression symptoms. Life Stressors and Vulnerability {#s4_2} -------------------------------- We first looked at how many participants actually mentioned on their social media posts the life stressors they reported to us. Our annotation result shows that only seven participants indicated a breakup experience, but no one mentioned being fired from work. Among the 23 participants who reported to us that their significant others were diagnosed with severe illness, only one mentioned it on Weibo. Our result suggests that participants are very selective regarding what life stressors they want to share to the public, of which, relationship breakup is the one most commonly mentioned on social media. In the previous sections, we observed that people with high CES-D scores used more negative affective words on their social media posts over years, which indicates certain degrees of cognitive bias. In this part of analysis, we further raise the question, do people displaying cognitive bias in their social media posts over a long period of time have higher risks of developing depression symptoms when they are under stress? To answer this question, we conducted a correlation analysis between affect and depression symptoms before the life stressor occurred. Since participants reported that life stressors happened between day 0 and day 90, we used day 90 as a pivot. We broke down the timeline into three time windows and observed the correlations between depression symptoms and social media affect in each time window (see [**Table 3**](#T3){ref-type="table"}). Here, the time line starts from the pivot to 1 year, 2 year and 3 years before the pivot. Participants in table 3 all completed the life event survey. The 'Yes' group refers to participants who reported a relationship breakdown or being fired from work (N = 45), the 'No' group (N = 205) refers to participants who experienced other life stressors or no life stressors. We divided participants according to life stressors they experienced because those who experienced a relationship breakdown or being fired from work have highest self-reported symptom score (see [**Table 1**](#T1){ref-type="table"}). We found that the amount of negative affective words in the 'Yes' group is moderately correlated with their self-reported symptom score long before the life stressor occurred. Participants in this group seem to be less satisfied with their daily life activities in general. Our result suggests cognitive vulnerability can be observed in longitudinal social media data. ###### Correlation between Center for Epidemiologic Studies Depression (CES-D) score and social media affect before the occurrences of life stressors. Relationship breakdown/fired from work Yes No ---------------------------------------- -------- --------- --------- -------- --------- --------- time window (from 90 days to...) 1 year 2 years 3 years 1 year 2 years 3 years positive affect −0.17 −0.17 0.06 −0.017 −0.025 −0.025 negative affect 0.38 0.40\* 0.40\* −0.038 0.021 0.021 'Yes' refers to the fact that the participants experienced relationship breakdown or being fired from work in the recent 3 months before they completed the CES-D measurement. The number shows the correlation between participants' CES-D score and their average affect over a period of time (1 year, 2 years, 3 years before they completed CES-D measurement). \*p \< 0.05; p \< 0.1. Results: Stress-Reactive Rumination {#s5} =================================== Among the three life events we examined in this study, relationship breakdown was the most commonly mentioned stressor on Weibo. Therefore, we focused on examining the rumination language from the participants who had experienced relationship breakdown (N = 33). Among them, only seven mentioned the word "breakup" on their Weibo posts. We annotated 151 Weibo posts from the seven participants, these posts were posted between day 0 and day 90. We found that 23% (N = 33) of their posts contained rumination language, and all of the rumination contents were related to relationship breakup. We observed that the rumination language indicated various types of cognitive tendencies, and the most common one was focused on negative attribute. People tend to ruminate on the loss of a relationship. For example, "I can't face reality." "I can't move on." "Your favorite Mr. Z has left you." (see Examples 1 and 2 in [**Table 4**](#T4){ref-type="table"}). Occasionally, people have hopeless thoughts, such as "life is meaningless" (see Example 3 in [**Table 4**](#T4){ref-type="table"}). They tend to linger on the negative emotions, for example, "My tears keep pouring down when I'm not busy with anything." Meanwhile, we also observed that people adopted various coping strategies, such as reappraisal, e.g., "I don't like you anymore." and problem-solving, e.g., "I have to treat myself well." "stay strong and still" (see Example 4 in [**Table 4**](#T4){ref-type="table"}). These coping strategies often contain positive emotions. ###### Examples of rumination contents. Participant Rumination content Translation ---------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Example 1: Focus on negative attribute 1 你最爱的Z先生，已经离开你了 Your favorite Mr. Z has left you. 1 隐藏了关于你的一切，不后悔曾经爱过你，也没力气再向前。如果我的心痛全世界没有一个人懂，我也不后悔曾经爱过 I hide everything related to you, I never regret loving you, but I can't dare to move on. Even if no one else in the world knows my sorrow, I will never regret I fell in love with you. 1 曾经最爱的那个人怎么就不爱了呢，想起曾经的快乐，和再也回不去的困惑，总是做不到头也不回的回到现实中去。总想着有一天春风和煦，我们还是可以一起离开这里，忘掉所有不愉快。可是改变了就是改变了啊 How come I don't love the person I used to love anymore? I remember all the joy and confusion, I can't face the reality. I am always thinking about that one day, we will leave this place together and forget all the sorrows. However, something has changed. 1 一闲下来眼泪就往上涌，都会过去的，会过去的 My tears keep pouring down when I am not busy with anything, and everything will be fine, will be fine. Example 2: Focus on negative attribute 2 可能你很久以后才学会爱人，我很遗憾只做了你途中の风景 Maybe you will learn how to love again, and it's a shame that I'm only scenery in your life. 2 一万次的道别难道还不够，也许再见只是一个承诺，你在夕阳里回首的轮廓，我到现在依然记得 No matter how many times we say farewell, it's never enough. Maybe this is just a promise to you. I still remember the way you look at me at sunset. Example 3: Focus on hopeless thoughts 2 一句话也不想说，不是淡薄而是呵呵，不是不多想而是乏了。静的只听得见自己的呼吸，淋一场大雨不管不顾，放声大哭。然而眼泪也出不来，没有意义没有寻找的生活，人早已经麻木。在这么下去得抑郁不可 I don't want to say anything, because I'm too tired to think of anything else. The world is so quiet that I can hear my breath. I showered in the rain and cried out loud. There are no tears in my cry, and life is meaningless, I feel numb about my life. I will be depressed soon. Example 4: Focus on coping strategies 3 不做白日梦了，认真做事，好好生活 Stop daydreaming, work hard and enjoy life. (problem solving) 3 终于可以不再爱你了，真好 Finally, I don't love you anymore, that's great. (reappraisal) 3 25岁，不矫情，不任性，不抱怨！摒弃外界眼光，只为自己而活。 I'm 25 years old now, and I'll be strong, mature and stop complaining. (problem solving) 3 自己也要对自己好啊 I have to treat myself well. (problem solving) 2 真正能依靠的唯有内心的强大，坚强难得，却定心 I should rely on the power inside me, stay strong and still. (problem solving) 2 其实真的不用那么敏感，我不在喜欢你了，我的名字只是来自一句歌词，没翅膀也做梦想家 Don't be over-sensitive, I don't like you anymore. My name comes from a song lyric: "I want to be a dreamer even without wings". (reappraisal) Discussions {#s6} =========== Implications {#s6_1} ------------ In this paper, we applied a data-driven approach to analyze individuals' affect patterns on a Chinese social media platform. Overall, we found that people's affective expressions on social media could reflect their risks of developing depression symptoms long before early symptoms were expressed. Therefore, researchers should examine social media posts over a longer time frame when studying depression symptoms. By looking at NA and PA separately, we found that individuals with high depression symptoms tended to use more negative and positive affective words on their social media status updates in general. This finding is in contrast with the findings in traditional empirical studies but aligns well with the recent findings from Locatelli and colleagues ([@B19]). We speculate this is related to the fact that social media users tend to present themselves positively ([@B20], [@B21]). This finding might suggest that users with high depression symptoms are more likely subject to a greater level of self-presentation bias. Accordingly, researchers should take into account the characteristic of specific social media behaviors while using social media data to study psychological symptoms. We also found that users rarely mentioned significant life stress on social media. Among 77 people who told us on our survey that they had experienced a stressful life event, only about 10% of them had mentioned it on a Weibo post. This encourages the researcher to be aware of a highly biased sample when conducting research on life stressors with social media data. Since female users are more likely to disclose their mental state ([@B27]), our sample for rumination might also be biased toward female users. So far, most of the existing studies that make use of social media data to infer depression symptoms have only used a quantitative approach to analyze the language in the posts. Few studies have attempted to examine the content that is directly linked to negative cognitive biases. In our study, we examined the rumination language from the participants who had recently experienced a breakup. This group of people also had exceptionally high symptom scores (M = 29). We found that 23% of their Weibo posts contained rumination contents. Our finding aligns well with the literature on depression symptoms and post-event rumination ([@B38], [@B39]). Although their rumination often focused on negative attributes, we also found evidence of problem-solving coping strategies ([@B40]). These findings provide insights into identifying social media content that is directly associated with depression symptoms, and call for a more calibrated approach to measure depression symptoms by looking at cognitive biases in social media data. Limitations {#s6_2} ----------- Chinese Weibo has a sophisticated filtering system to censor Internet data; contents considered "harmful" to the community will be immediately tagged and discarded ([@B41]). Hence, swear words and some negative opinions are often censored in such a social media platform. In order to evade the censorship, social media users start to use metaphorical language or change the written forms of swear words. Simple natural language processing techniques are less reliable in detecting such variations of negative or sarcastic expressions. In addition, there are confounding factors that might affect our conclusions, such as the offline behaviors not observable in social media data. Therefore, affect expressed on social media data only reflects a small portion of daily life affect. Furthermore, the results of this study are also biased toward the data generated by active Weibo users. All these limitations prevent us from making stronger or more general claims, but our study still provides useful insights about cultural dependent symptoms and vulnerability as indicated by social media data. Conclusions and Future Work {#s7} =========================== We presented a comprehensive study of negative and positive affects shown on the Weibo posts of Chinese social media users. First, we collected Weibo status updates from users who completed a survey to measure their depression symptom levels and detect their life stressors. We visualized users' social media affect in a temporal manner and proceed to examine their language after they experienced specific life stressors. Our results show that increased negative and positive affects in social media status updates are closely related to elevated depression symptoms. Such a unique pattern reflects cognitive vulnerability to developing depression symptoms. Users with cognitive vulnerability have higher depression symptom scores after they experience specific stressors in life. Finally, we proposed to study the rumination language in social media content with negative affect, because rumination language is associated with dysfunctional stress response. This study reveals how social media based measures serve as a longitudinal resource to monitor participants' vulnerability to mental problems. It is potentially useful for clinicians to identify individuals at risk. Some of the findings could be limited to Chinese culture. More cross-cultural studies are necessary to identify the cultural differences and their influences on mental disorders. Data Availability Statement {#s8} =========================== The data of this paper can be made available based on appropriate requests to the corresponding authors. Ethics Statement {#s9} ================ This study was approved by the College Research Ethics Committee of City University of Hong Kong. The methods were carried out in accordance with the approved guidelines from the College Research Ethics Committee. Written informed consents were obtained from all the participants. Author Contributions {#s10} ==================== LC contributed to the design and data collection. LC and TG contributed to data analysis and interpretation, drafting, and revising the paper. CC contributed to supervision and paper revision. Funding {#s11} ======= TG is supported in part by the Natural Science Foundation Committee of Guangdong Province (Grant No. 2018A0303130235) and the MOE Project of the Centre for Linguistics and Applied Linguistics, Guangdong University of Foreign Studies. Conflict of Interest {#s12} ==================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The authors thank Dr. Zheng Pan from the South China Agricultural University for the help on web scraping and Dr. Rob Davidson for suggestions on this work. **Introduction:** This document contains the survey questions and the translated/back translated version of the CES-D survey. Data Collection Consent Form Options --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------- 本人是香港城市大学心理系硕士生，此研究目的是用微博数据建计算机模型自动预测心理特质。我们需要搜集你的微博数据，如同意我们收集您的数据，请在本问卷填写您的微博用户名称。如不同意，请忽略。本问卷收集的数据只用于研究用途。参与研究同意书: 1\. 本人已经明白上述的资料，并同意参与这次研究。2. 不同意参与这次研究 I am a master student from the City University of Hong Kong, HKSAR. We are looking for Weibo users to complete this survey. The purpose of this study is to identify the characteristics of social media users with depression symptoms. You are opting to fill out the review below and allow us to collect your Weibo data. Your data will only be used for this study. 1\. I understand the information, and I agree to participate in this study. 2. I don't want to participate in this study. Life Events Survey Question Translation Options 1\. 你的微博名称（不是你邮箱的名称，例如：luciasalar) 最关键的一步，大家务必要填写。实在没有微博请填写"无" Your Weibo account (Not your email address, e.g.: luciasalar) 1\. 你过去三个月有经历什么重大压力吗？（例如：失恋，亲人患病) Have you experienced any major stress in the past 3 months? (e.g., breakup, family member being diagnosed with illness) \(a\) Yes (b) No 2\. 如选择有，请选择（可多选）： If you selected Yes in the previous question, please specify the stressful life event (multiple choice): \(a\) breakup (b) family member being diagnosed with severe illness (c) being fired from work (d) Others 3\. 你曾经得过抑郁症吗？ Have you ever been diagnosed with depression? \(a\) Yes (b) No (c) I don't know Below is the translated CES-D scale and its back translation, the answers for each questions are: a. I barely feel that way; b. 1-2 days last week; c. 3-4 days last week; d. every day; e. every day and it has been for 2 weeks. Question Back Translation ---------------------------------- ----------------------------------------------------------------------- 1\. 你最近不想吃东西、胃口不好 You have a poor appetite recently and you don't want to eat anything. 2\. 你觉得心情很不好 You feel like having a bad mood. 3\. 你觉得做事情很不顺利 You feel that things don't go through smoothly 4\. 你睡不安稳 You have an uneasy sleep 5\. 你觉得很快乐 You feel happy 6\. 你觉得人人都不友善 You feel others are not friendly 7\. 你觉得日子过得很好很享受人生 You are enjoying your life 8\. 你觉得人不喜欢您 You think that people don't like you 9\. 你提不起劲做任何事 You don't have a mood to work on anything 10\. 你觉得很悲哀 You feel sad 11\. 你觉得寂寞，孤单 You feel lonely **Introduction:** This document describes the annotation guidelines for marking up Weibo that indicate specific stressful life events and ruminating behaviors. The document contains two tasks. Task 1: Annotate stressful life events, including relationship problem, unemployment and severe illnesses. Task 2: Annotate rumination in the post. **Task 1: Annotate stressful life events.** In this task, you are provided three sets of posts from Weibo users who have reported that they experienced at least one of the following LIFE EVENTs: Relationship breakdown, unemployment, and severe illness of a friend or family member. The title of the page indicates which LIFE EVENT that the author had been through. How to Annotate LIFE EventsRelationship breakdown. In this task, please annotate the earliest post that indicates that the poster experiences a breakup, divorce, or loss of a relationship. You only need one label for each user in this task. Once finding a post as such, you can skip the rest of the posts from the same user.Unemployment. In this task, please annotate the earliest post that indicates the state of unemployment. You only need one label for each user in this task. Once finding a post as such, you can skip the rest of the posts from the same user.Severe illness. In this task, please annotate the earliest post that indicates a family member or close friend being diagnosed with terminal illness. You only need one label for each user in this task. Once finding a post as such, you can skip the rest of the posts from the same user. **Task 2: Annotate rumination in the post.** Rumination is repetitively going over a thought or a problem without completion. The ruminating behavior on Weibo should be the act of posting multiple Weibo that show deep thoughts or reflections about the same topic. Content of ruminative thought can be positive or negative. Please select RUMINATION if the post fulfill all of the following conditions:Author describes a specific thought, emotion or a problem in the post.The thought or problem reoccur within one week after the previous post (Read the timestamp). Here shows an example of rumination on relationship breakup:可能你很久以后才学会爱人，我很遗憾只做了你途中の风景。Timestamp: 2015-2-14 (Translation: It might take longer for you to learn how to love a person. It's a shame that I am just scenery in your life.)有时候会想，为什么偏偏是天籁。以后每次坐他的车，都难以忘记第一次你用挂挡の那只手牵着我，跟我说手动挡都无所谓，何况这是自动挡。真的希望回忆不能欺骗时光太久，愿无岁月可回首，愿无深情共白头。Timestamp: 2015-2-15 (Translation: I can't forget how he held my hand the first time I was in his car. I hope my memory does not lie to the time for too long. Time flies by, and I hope we can grow old together.)隐藏了关于你的一切，不后悔曾经爱过你，也没力气再向前。如果我的心痛全世界没有一个人懂，我也不后悔曾经爱过。Timestamp: 2015-2-15 (Translation: I hide everything about you. I never regret to fall in love with you. But I can't move on anymore. Even if no one can understand how painful I am, I will never regret being in love with you.) All the above examples describes feelings toward the breakup and or memory toward the ex-lover and these posts occur within a few days. [^1]: Edited by: Meichun Mohler-Kuo, University of Applied Sciences and Arts of Western Switzerland, Switzerland [^2]: Reviewed by: Jessica L. Hamilton, University of Pittsburgh, United States; Oliver Gruebner, University of Zurich, Switzerland [^3]: This article was submitted to Public Mental Health, a section of the journal Frontiers in Psychiatry

Introduction {#s1} ============ In the research field of drug addiction the mesolimbic dopaminergic system has always received great attention since the evidence that most of the substances of abuse are able to increase dopamine extracellular concentration in the Nucleus Accumbens (Di Chiara and Imperato, [@B13]). Only recently it has been suggested that also dopaminergic neurons projecting to the medial prefrontal cortex (mPFC) might play a crucial role in the development of dependence (for reviews, see Kalivas and Volkow, [@B25]; Koob and Volkow, [@B30]; Jentsch et al., [@B22]). mPFC is part of the mesocorticolimbic dopaminergic system whose neurons project from the VTA also to the Nucleus Accumbens and Amydgala (Björklund and Dunnet, [@B3]). These areas together form the reward and motivation circuitry which is crucial in the regulation of functions that are altered in drug addiction, such as attribution of incentive salience to a stimulus (Robinson and Berridge, [@B38]), activation of goal-directed behaviors (Salamone et al., [@B39]), evaluation of reward (Koob and Volkow, [@B30]). In this circuitry, the mesocortical dopaminergic pathway has been proved to be fundamental in regulating impulsivity and action inhibition (Jentsch et al., [@B22]), which are key feature in all the stages of drug addiction. Accordingly, disruption of the processes that regulate inhibitory control and reward sensitivity has been suggested to be important mechanism in the development of addiction which has been proposed to involve neuroadaptive changes in the mesocorticolimbic circuitry that in turn alter the mechanisms that regulate reward, motivation, memory consolidation, sensitivity to stress, executive and inhibitory control (Koob and Volkow, [@B30]). The ventromedial PFC is highly involved in the evaluation of reward and in the process of decision making (Peters and Büchel, [@B36]). Thus, PFC dysfunction may exacerbate the loss of control associated with compulsive drug use and facilitate the progression to drug addiction (Jentsch et al., [@B22]; Koob et al., [@B31]). During abstinence from alcohol, mPFC functionally disconnects from the Amygdala while retaining connection to the Nucleus Accumbens; this functional disconnection has been suggested to be crucial for impaired executive control over motivated behavior suggesting that disregulation of mPFC interneurons may be an early index of neuroadaptation in alcohol dependence (Koob et al., [@B31]). In the development of addiction, stress is known to be a key factor, which can increase the vulnerability to drug abuse (Koob et al., [@B31]). Accordingly, in a model of chronic exposure to stress, like social isolation at weaning, socially isolated (SI) rats show several evidences of a high propensity to addiction. Thus, SI rats have been shown to be more prone to self-administer amphetamine (Bardo et al., [@B2]; Whitaker et al., [@B43]), cocaine (Howes et al., [@B20]) and ethanol (Hall et al., [@B18]; Lodge and Lawrence, [@B33]; Deehan et al., [@B12]; McCool and Chappell, [@B34]; Whitaker et al., [@B43]; Lesscher et al., [@B32]). In light of these evidences, in our study we evaluated the possibility that social isolation, as a model of chronic stress exposure, might induce a change in the sensitivity of mesocortical dopaminergic neurons to ethanol exposure which, in turn, would increase the vulnerability of SI rats to develop ethanol addiction. We hypothesize that a decreased response of mesocortical dopaminergic neurons to ethanol would induce a loss of control and a compulsive behavior toward drug use and facilitate the progression to drug addiction. In our study, we evaluated the effect of both acute and chronic administration of ethanol to test whether the ethanol-induced response of mesocortical dopaminergic neurons might be different after acute or chronic administration of the drug. Moreover, to evaluate whether a decrease in prefrontal cortical function was associated with a vulnerability to ethanol addiction, we also measured the amount of ethanol consumed during self-administration protocol in SI and group housed (GH) rats, as a possible index of a higher propensity of SI animals to develop dependence from the drug. Materials and Methods {#s2} ===================== Animals {#s2-1} ------- Male Sprague Dawley CD young-adult rats (Charles River, Como, Italy) were bred in our animal facility and maintained under an artificial 12 h-light, 12 h-dark cycle (lights on from 08:00 to 20:00 h), at a constant temperature of 22 ± 2°C, and a relative humidity of 65%. They had free access to water and standard laboratory food at all times. All efforts were made to minimize animal suffering. Animal care and handling throughout the whole experimental procedures were made in accordance with the European Communities Council Directive of 24 November 1986 (86/609/EEC). The experimental protocols were also approved by the Animal Ethics Committee of the University of Cagliari and by the Italian Ministry of Health (authorization \#353/2015-PR). Social Isolation and Voluntary Consumption of Ethanol Paradigm {#s2-2} -------------------------------------------------------------- At weaning, at postnatal day (PND) 21, the animals were housed individually SI or in groups of five per cage GH. From 28 PND to rats from both groups an ethanol solution was made available for self administration for 3 weeks, 2 h a day (from 11:00 to 13:00 h). To instigate ethanol reinforcement without food or fluid deprivation we used a modified initiation procedure (Samson et al., [@B40]) that involved the use of sucrose in the ethanol solution; sucrose concentration was progressively decreased, contextually keeping constant that of ethanol, according to the following paradigm: (days 1--2) 5% (v/v) ethanol + 5% (v/v) sucrose; (days 3--4) 5% ethanol + 4% sucrose; (days 5--6) 5% ethanol + 3% sucrose; (days 7--8) 5% ethanol + 2% sucrose; (day 9-end of treatment) 5% ethanol + 1% sucrose. By day 11 of treatment to the beginning of the experiment, the solution was kept constant to 5% ethanol (v/v) and 1% sucrose. Both SI and GH animals were placed in individual cages for the 2 hof daily exposure to ethanol to allow a precise measure of ethanol consumption. The weight of each rat, the amount of fluid (both water and ethanol), and food intake were monitored daily at the end of the session. Surgery and Experimental Procedures {#s2-3} ----------------------------------- Rats were anesthetized by intraperitoneal (i.p.) injection of chloral hydrate (0.4 g/kg), and a concentric dialysis probe was inserted at the level of the mPFC (A +3.2, ML +0.8, V −5.3 relative to the bregma), according to the Paxinos and Watson ([@B35]) Atlas. The active length of the dialysis membrane (Hospal Dasco, Bologna, Italy) was restricted to 4 mm. As previously described (Dazzi et al., [@B11]), the length of the dialitycal membrane allowed to sample from both infralimbic and prelimbic cortices. Experiments were performed in freely moving rats, 24 hafter probe implantation to allow recovery from surgery procedures. Ringer's solution \[3 mM KCl, 125 mM NaCl, 1.3 mM CaCl~2~, 1 mM MgCl~2~, 23 mM NaHCO~3~, 1.5 mM potassium phosphate (pH 7.3)\] was pumped through the dialysis probe at a constant rate of 2 μl/min. Samples of dialysate were collected every 20 min from 8:30 to 15:00 h and immediately analyzed for dopamine by high performance liquid chromatography (HPLC) with electrochemical detection as previously described (Dazzi et al., [@B10]); the detection limit for dopamine was 2 fmol per injection. The average neurotransmitter concentration in the first two samples was taken as 100%, and all subsequent values were expressed as mean ± SEM relative to the basal value. The mean *in vitro* recovery of the probes was 15 ± 3%. All probes were tested before implantation, and those with a recovery value outside of this range were not used. The absolute concentration of dopamine was not corrected for this value. At the end of each experiment, the placement of the probe was verified histologically. All rats in which the probe was located outside of the target region were excluded from the analysis. A group of rats were subjected to an acute treatment with ethanol (0.5, 2 g/kg, i.p., 20% solution v/v); the drug was injected after three stable samples (variation in dopamine concentrations less than 20%). For the acute ethanol administration experiments the average neurotransmitter concentration in the first three samples was taken as 100%, and all subsequent values were expressed as mean ± SEM relative to the basal value. Statistical Analysis {#s2-4} -------------------- Data are presented as Mean ± SEM of at least five animals per group. Microdialysis data were compared among groups with one- or two-way analysis of variance (ANOVA) for repeated measures, factors being treatment and time points. The raw values of dopamine concentration were used for the analysis, with absolute basal concentrations given in "Results" Section. Normal distribution of data was verified by Skewness and Kurtosis evaluation with Graph Pad Prism 5.0. *Post hoc* comparisons were performed with Newman-Keuls test. A *p* value of \<0.05 was considered statistically significant for all experiments. Results {#s3} ======= Basal Extracellular Concentration of Dopamine in SI and GH Rats {#s3-1} --------------------------------------------------------------- Basal extracellular concentration of dopamine in the mPFC of SI rats was not significantly different from that of GH rats (14.86 ± 2.926 fmol per 40 μl sample for GH rats vs. 20.22 ± 3.048 fmol per 40 μl sample for SI rats; Figure [1](#F1){ref-type="fig"}). One way ANOVA revealed a not significant effect between the two experimental groups \[*F*~(1,28)~ = 0.60; *P* = 0.4498\]. {#F1} Effect of Acute Administration of Ethanol on Extracellular Dopamine Concentration in the mPFC in SI and GH Rats {#s3-2} --------------------------------------------------------------------------------------------------------------- We have previously shown (Dazzi et al., [@B10]), that acute administration of ethanol is able to induce a biphasic effect on dopamine extracellular concentration in the mPFC, with lower doses inducing an increase and higher doses a decrease, respectively, in dopamine output. In this article, to evaluate whether in SI rats mesocortical dopaminergic neurons show a different sensitivity to the acute administration of ethanol, we used the same doses we used in our previous article (0.5--2 g/kg, i.p.). The present observations confirm our previous data and show that, in GH animals, the acute administration of a low dose of ethanol (0.5 g/kg, i.p., 20% v/v) induced an increase in dopamine extracellular concentration in the mPFC that was maximal (+60%) 40 min after administration and returned to basal values after 120 min (Figure [2A](#F2){ref-type="fig"}). An higher dose of ethanol (2 g/kg, i.p., 20% v/v), on the contrary, induced a significant decrease in the same parameter (Figure [2B](#F2){ref-type="fig"}) with the maximal effect (−50%) observed 80 min after ethanol administration and values returning to basal after 120 min. {#F2} In contrast, in SI animals the acute administration of the lower dose of ethanol (0.5 g/kg, i.p.) failed to significantly modify basal dopamine extracellular concentration (Figure [2A](#F2){ref-type="fig"}) while the higher dose (2 g/kg, i.p.) induced a significant increase (+90% vs. basal values) in dopamine output 20 min after administration, reaching its maximum after 40 min. The increase persisted for 100 min after the injection and returned to basal values in 120 min (Figure [2B](#F2){ref-type="fig"}). The effect induced in SI animals by administration of the higher dose was similar to that observed in GH rats after injection of the lower dose of ethanol (0.5 g/kg, i.p.; Dazzi et al., [@B10]; present data), which, on the contrary, failed to induce any significant change in dopamine output in SI animals. Thus, social isolation induced a shift in the dose-response relation on the effect of ethanol on dopamine output in the mPFC (Figure [2C](#F2){ref-type="fig"}). For the dose of 0.5 g/kg, two-way ANOVA revealed a significant effect over time \[*F*~(8,64)~ = 1.89; *P* \< 0.01\]; a significant effect of treatment \[*F*~(1,64)~ = 7.28; *P* \< 0.01\]; and a significant effect of the interaction between factors \[*F*~(8,64)~ = 5.75; *P* \< 0.0001\]. For the dose of 2 g/kg, two-way ANOVA revealed a significant effect over time \[*F*~(8,64)~ = 2.92; *P* \< 0.01\]; a significant effect of treatment \[*F*~(1,64)~ = 8.83; *P* \< 0.05\]; and a significant effect of the interaction between factors \[*F*~(8,64)~ = 6.12; *P* \< 0.0001\]. Voluntary Ethanol Consumption in GH and SI Rats {#s3-3} ----------------------------------------------- To evaluate whether a chronic stress exposure like social isolation might alter the preference for ethanol and/or the amount of the drug consumed, we measured the amount of ethanol consumed by SI and GH animals. As shown in Figure [3](#F3){ref-type="fig"}, there was a significant difference in voluntary ethanol consumption relative to housing history, starting from the 5th day of the training protocol with SI animals consuming a significantly greater amount of ethanol than GH rats. The amount of ethanol consumed reached a plateau at the 5th day of the treatment for SI rats (\~6 g of ethanol/kg of body weight) and on the 9th day for GH (\~3.8 g of ethanol/kg of body weight). Two-way ANOVA revealed a significant effect over time \[*F*~(29,406)~ = 3.603; *P* \< 0.0001\]; a significant effect between the experimental groups \[*F*~(1,20)~ = 12.55; *P* \< 0.01\]; and a not significant interaction between the factors \[*F*~(29,406)~ = 1.130; *P* = 0.2962\]. {#F3} Figure [4](#F4){ref-type="fig"} shows that body weight did not differ significantly between the two experimental groups during the voluntary ethanol intake protocol. SI and GH rats also showed a similar intake of total fluid and food (data not shown). {#F4} Effect of Prolonged Voluntary Ethanol Intake on the Extracellular Concentration of Dopamine in the mPFC of SI and GH Rats {#s3-4} ------------------------------------------------------------------------------------------------------------------------- In GH rats the chronic voluntary consumption of ethanol induced a significant increase of extracellular dopamine concentration as early as 120 min before ethanol presentation and increased further to reach a maximal value of +70% of basal values by 60 min before ethanol consumption. It then slightly declined to a value of +50% during ethanol intake to return to basal values 40 min before removal of the alcoholic solution (Figure [5A](#F5){ref-type="fig"}). In contrast, social isolation markedly reduced the sensitivity of mesocortical dopaminergic neurons to anticipation of ethanol. Indeed, in SI rats, in contrast to GH animals, the extracellular concentration of dopamine didn't show any significant variation during the anticipatory phase; during the consummatory phase there was a slight but not significant decrease (−25% below basal values) in dopamine output. ANOVA revealed a significant effect over time \[*F*~(18,198)~ = 1.628; *P* \< 0.0001\]; a significant effect of housing \[*F*~(1198)~ = 9.130; *P* \< 0.0001\]; and a significant interaction between factors \[*F*~(18,198)~ = 1.628; *P* \< 0.05\]. {#F5} The amount of ethanol consumed by the two groups of rats during the experiment is shown in Figure [5B](#F5){ref-type="fig"}, and was significantly higher for SI than for GH rats (*P* \< 0.01; Figure [5B](#F5){ref-type="fig"}). Discussion {#s4} ========== Our results have shown that social isolation induces, in rats, a shift in the dose-response relation for the effect of the acute administration of ethanol in mesocortical dopaminergic neurons. In SI rats, in fact, the administration of 2 g/kg of ethanol, which in GH rats elicited a significant decrease in dopamine extracellular concentration in the PFC, induced an increase in the same parameter that was similar to the effect elicited by a lower dose (0.5 g/kg) in GH rats (present data; Dazzi et al., [@B10]). The lower dose of ethanol failed to induce any significant effect in SI animals, while increased dopamine output in GH animals. One possible explanation for the observed effect is that SI rats show a marked decrease in plasma and cerebrocortical concentrations of neuroactive steroids (Serra et al., [@B41]) and the effect of ethanol on mesocortical dopaminergic neurons is particularly sensitive to brain concentrations of progesterone and 3α, 5α-THDOC. In fact, an increase in the concentrations of these neurosteroids by prolonged administration of progesterone, is able to increase the response of mesocortical dopaminergic neurons to an acute administration of ethanol (Dazzi et al., [@B10]). This effect was completely abolished by finasteride, which, by inhibiting 5α-reductase induces a marked decrease in plasma and brain concentration of 3α, 5α-THDOC (Concas et al., [@B7]; Dazzi et al., [@B9]), suggesting that this progesterone metabolite is crucially involved in mediating this effect. Alternatively, in SI rats the extracellular dopamine might be governed through a different mechanism or even circuitry with respect to GH animals. The effect of ethanol on mesocortical dopaminergic neurons shown here is opposite to that recently observed by Karkhanis et al. ([@B29], [@B28]) who, instead, described an increase in dopamine response to acute ethanol in the Nucleus Accumbens and Amygdala of SI rats. Although discordant, these results are in agreement with the suggested alteration in the balance between the activity of mesocortical *vs* mesolimbic pathway in addiction (Kalivas and Volkow, [@B25]; Koob and Volkow, [@B30]), where an increased sensitivity to reward (Bonci et al., [@B5]), an increase in impulsivity and a decrease in inhibitory control (Jentsch et al., [@B22]) have been identified as key features to the development of addiction. Accordingly, our data show that SI rats consumed significantly higher amounts of ethanol compared to their GH counterpart, suggesting that chronic exposure to stress, in a critical period like adolescence, might increase the vulnerability to develop addiction in this experimental group. The observation that SI and GH rats did not significantly differ in their body weight nor in the amount of food or total fluid consumed during the entire protocol, suggests that this effect is specific for ethanol. Our results are in line with previous studies showing that social isolation increase voluntary ethanol intake (Hall et al., [@B18]; Lodge and Lawrence, [@B33]; Deehan et al., [@B12]; McCool and Chappell, [@B34]; Lesscher et al., [@B32]). However, this effect is strictly depending on the strain of rats used, on the drinking protocol and on the concentration of the ethanol solution used for the experiments, and some authors found a decrease in the amount of ethanol consumed by SI rats (Fahlke et al., [@B16]), and others found no effect (Ehlers et al., [@B14]). We show that basal extracellular concentration of dopamine in the mPFC wasn't significantly different in SI vs. GH rats; in fact, although there was a tendency toward an increase in this parameter, it did not reach statistical significance. These results are in agreement with previous microdialysis data (Dalley et al., [@B8]) but not with measurements of dopamine function in postmortem tissues of SI rats (Blanc et al., [@B4]; Jones et al., [@B24]; Heidbreder et al., [@B19]) that showed a significant increase in DOPAC/dopamine ratio in homogenates from the mPFC of SI vs. GH rats. However, as suggested by Dalley et al. ([@B8]), microdialysis allows a more accurate measurement of the extracellular content of dopamine. Our observation that in SI animals the increase in dopamine output induced by both anticipation and consumption of ethanol was dramatically blunted, together with the increased consumption of ethanol by rats in this experimental group, suggest that early and prolonged exposure to stress might increase the vulnerability to drug addiction. Previous evidence that social isolation didn't change ethanol metabolism (Karkhanis et al., [@B29]), suggest that this effect may be due to a perturbation of the function of mesocortical dopaminergic neurons by chronic exposure to stress rather than to pharmacokinetic differences between the two experimental groups. Our results showing that a blunted response of mesocortical dopaminergic neurons to ethanol is accompanied by an increase in the consumption of ethanol in SI rats, confirm the link between addictive behavior and dopaminergic hypofrontality. Together with the previous observation that in SI rats there is, on the contrary, an increase in the sensitivity of dopaminergic neurons in the nucleus accumbens and amygdala to ethanol (Karkhanis et al., [@B29], [@B28]), they suggest that SI might be able to alter the balance between mesocortical and mesolimbic dopaminergic pathways, to increase vulnerability to addiction. A number of studies have in fact suggested that progression of addiction from a social use to compulsive use of a drug might be consequence of a decrease in the executive control and/or of a strengthening of the cortico-striatal circuitry that regulates habitual behavior. In fact, once a given behavior is acquired, the cortico-striatal-thalamic pathway allows the behavior to be efficiently performed without the activation of the prefrontal cortical circuitry (Jog et al., [@B23]; Canales, [@B6]) which is then able to integrate new information that can modify and drive the acquired behavior. Thus, by processing the environmental stimuli, the PFC has the ability to modify a given behavior whenever it results dangerous or inappropriate to the subject (Kalivas et al., [@B26]). Addiction is characterized by the inability to stop or modify a behavior even when it clearly has negative consequences on the individual (Everitt and Robbins, [@B15]; Kalivas et al., [@B26]). Accordingly, a number of studies described a functional "hypofrontality" as a key feature in addiction (for review, see Jentsch and Taylor, [@B21]; Goldstein and Volkov, [@B17]; Jentsch et al., [@B22]). Together with the decline in frontal executive control, it has been shown a progressive strengthening of the compulsive behavior of seeking and taking the drugs (Everitt and Robbins, [@B15]). The latest researches on drug addiction have pointed out that drugs of abuse, as well as certain palatable foods, are able to induce neuroadaptive changes in the activity of the mesocorticolimbic circuitry (Volkow et al., [@B42]; Koob et al., [@B31]); in particular, in animals, withdrawal from drug addiction has been characterized by an increased responsiveness to reward and a decreased activity in the mesocortical dopamine system (Volkow et al., [@B42]). Our observation of a blunted sensitivity of mesocortical dopaminergic neurons to ethanol anticipation in SI rats suggests that chronic stress is able to reduce the response of this pathway to ethanol. This data, together with the previous observation that in SI rats mesolimbic dopaminergic neurons show, on the contrary, an increased response to acute ethanol administration (Karkhanis et al., [@B29]), further support the hypothesis that an important mechanism for the development of addiction is a disruption in the balance of the function between mesolimbic and mesocortical dopaminergic pathway. They also suggest that chronic exposure to social isolation stress, by blunting the sensitivity of mPFC projecting neurons to ethanol exposure, might increase the vulnerability to drug addiction. Our observation of a decreased efficacy of ethanol in SI rats already after a single administration, seems to suggest that the increased vulnerability is not dependent on previous ethanol exposure. Accordingly, in previous studies SI rats have shown greater increases in dopamine extracellular concentration in the Nucleus Accumbens after an acute administration of amphetamine (Hall et al., [@B18]), an enhanced Conditioned Place Preference for amphetamine and ethanol after a single conditioning session (Whitaker et al., [@B43]), an increased locomotor response to cocaine (Phillips et al., [@B37]), as well as an increased self-administration of drugs of abuse (Hall et al., [@B18]; Lodge and Lawrence, [@B33]; Deehan et al., [@B12]; McCool and Chappell, [@B34]). In conclusion, our data show that chronic exposure to social isolation stress in a critical period such as early adolescence is able to modify the neurochemical response of mesocortical dopaminergic neurons to both acute and prolonged administration of ethanol. They suggest that a blunted sensitivity of mesocortical dopaminergic neurons might be a neuroadaptive adjustment to chronic stressful stimuli, which, in turn, would disrupt the balance between mesocortical and mesolimbic system's function that has been suggested to be crucial for the development of addictive behavior (Kalivas and Volkow, [@B25]). Thus, ours and others result (Karkhanis et al., [@B29]), seem to suggest that both a decrease in mPFC responsiveness and an increased sensitivity in the Nucleus Accumbens and Amygdala to the effect of ethanol, are already present at the first administration of the drug, and that prolonged administration abolishes the motivational salience of ethanol anticipation, thus building in SI rats a vulnerability to ethanol addiction. Author Contributions {#s5} ==================== VL: performed the experiments, analyzed the data, wrote the article. LM: performed the experiments. LD: conceived and designed the experiments, analyzed the data, wrote the article. Conflict of Interest Statement {#s6} ============================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer CSC and handling Editor declared their shared affiliation, and the handling Editor states that the process nevertheless met the standards of a fair and objective review. [^1]: Edited by: Marco Martina, Northwestern University, USA [^2]: Reviewed by: C. Savio Chan, Northwestern University, USA; Hau-Jie Yau, National Institutes of Health (NIH)/National Institute on Drug Abuse (NIDA), USA

Background {#Sec1} ========== Uterine artery embolization (UAE) for the treatment of uterine fibroids was first reported in 1995 \[[@CR1]\], and since then, many reports \[[@CR2]--[@CR15]\] have described the risks and benefits of UAE. Some reports \[[@CR12]--[@CR15]\] stated that preoperative UAE can help reduce bleeding, and most reports indicated that UAE is safe for uterine fibroid. David et al. \[[@CR15]\] reported that patients undergoing a hysterectomy with a uterine weight of more than 1000 g have a significantly higher risk of perioperative complications and are at greater risk of requiring a blood transfusion. David and Kröncke \[[@CR15]\] also reported that only two of the three patients with myomata weighing more than 1100 g were able to avoid blood transfusion, because of preoperative UAE. The occurrence of hyperkalemia and acute kidney failure as complications of preoperative UAE has not been reported previously. Here we report the occurrence of hyperkalemia and acute kidney failure after preoperative UAE for a large uterine fibroid. To our knowledge, this is the first report on the occurrence of hyperkalemia and acute kidney failure after preoperative UAE. Case presentation {#Sec2} ================= A 48-year-old Japanese woman with a medical history of multiple sclerosis presented to our hospital complaining of compression in her abdomen and an abdominal mass. Magnetic resonance imaging revealed a uterine fibroid measuring 37.5×27×13.5 cm along with some small fibroids (Figs. [1](#Fig1){ref-type="fig"} and [2](#Fig2){ref-type="fig"}). We planned total abdominal hysterectomy and bilateral salpingo-oophorectomy 3 days after UAE.Fig. 1A sagittal T2-weighted magnetic resonance image of the uterus Fig. 2A coronal T2-weighted magnetic resonance image of the uterus Embolization of her bilateral uterine arteries and selective embolization of her left bladder artery were performed using a gelatin sponge (Figs. [3](#Fig3){ref-type="fig"} and [4](#Fig4){ref-type="fig"}), because her left bladder artery (Fig. [3](#Fig3){ref-type="fig"}, arrow) supplied the uterine fibroid. However, 12 hours after embolization, she experienced cold sweats and vomiting, and 15 hours after embolization, hyperkalemia was noted on venous blood analysis and acute kidney failure was identified (Table [1](#Tab1){ref-type="table"}). Arterial blood gas analysis showed compensated metabolic acidosis: pH, 7.368; partial pressure of carbon dioxide (pCO~2~), 27.3 mmHg; base excess, −8.2; and bicarbonate (HCO~3~), 15.4 mmol/L. Glucose-insulin therapy was administered; however, it was not successful in resolving her condition. She then received continuous hemodiafiltration in our intensive care unit; however, her hyperkalemia and kidney failure did not improve. Therefore, she underwent emergency surgery. Total abdominal hysterectomy and bilateral salpingo-oophorectomy were performed, and her intraoperative blood loss was 105 g (Figs. [5](#Fig5){ref-type="fig"}, [6](#Fig6){ref-type="fig"}, and [7](#Fig7){ref-type="fig"}). The weight of her uterus was 10.8 kg and the volume was 9964 cm^3^. The volume was calculated using the formula: volume = length (cm) × width (cm) × diameter (cm) × 0.5233 (Fig. [8](#Fig8){ref-type="fig"}). She underwent autotransfusion (800 mL) and received 1200 mL of packed red blood cells. Her uterus had necrotic tissue, and the pathological finding was uterine fibrosis with necrosis (Figs. [9](#Fig9){ref-type="fig"} and [10](#Fig10){ref-type="fig"}). Following surgery, her hyperkalemia and kidney failure resolved.Fig. 3An angiography image obtained before uterine artery embolization. The left bladder artery (*arrow*) supplied a uterine fibroid Fig. 4An angiography image obtained after uterine artery embolization Table 1Results of venous blood analysis 15 hours after uterine artery embolizationTestValueWhite blood cell count (μL)24800Red blood cell count (×10^4^/μL)454Hemoglobin (g/dL)13.3Hematocrit (%)39.3Platelet count (×10^4^/μL)32.5Total protein (g/dL)9.4Albumin (g/dL)4.7Potassium (mEq/L)8.0Sodium (mEq/L)134Chloride (mEq/L)96Blood urea nitrogen (mg/dL)28Creatinine (mg/dL)2.02Creatine kinase (IU/L)301Lactate dehydrogenase (IU/L)363Aspartate transaminase (IU/L)33Alanine transaminase (IU/L)17Alkaline phosphatase (IU/L)262C-reactive protein (mg/dL)1.49Activated partial thromboplastin time (s)25.6Prothrombin time international normalized ratio1.04Fibrinogen (mg/dL)452 Fig. 5An image of the patient's abdomen before surgery Fig. 6An image of the abdomen obtained during the surgery. The uterus occupies a large part of the intraperitoneal space Fig. 7An image of the uterus during the surgery Fig. 8An image of the excised specimen Fig. 9An image showing necrosis within the excised specimen Fig. 10A histological image showing necrotic cells (*yellow circle*). Hematoxylin and eosin stain, ×400 Discussion {#Sec3} ========== Previous reports \[[@CR2]--[@CR15]\] mentioned that most women who underwent UAE for uterine fibroids were satisfied with the clinical outcome and had few complications. Although complications have been reported after UAE, hyperkalemia and kidney failure have not been reported previously. The complication rate associated with UAE has been reported to be very low, and most complications have been found to be transient \[[@CR2]\]. The most serious complication associated with UAE is endometritis/uterine infection, with a reported incidence of approximately 2 % \[[@CR2]--[@CR4]\]; however, the associated morbidity rate was found to be very low \[[@CR2]\]. Complications following UAE can be classified into immediate (peri-procedure), early (within 30 days), and late (beyond 30 days) complications \[[@CR5]\]. Most immediate complications are local complications, such as hematoma, arterial thrombosis, dissection, and pseudoaneurysm, and other complications include spasm and non-target embolization \[[@CR5]\]. Non-target embolization is relatively rare, and it does not occur if a good technique is used \[[@CR5]\]. Most early complications are associated with post-embolization syndrome and include pain, nausea, fever, and malaise, and other complications are rare \[[@CR5]\]. Most complications of UAE have been shown to occur more than 30 days after the procedure \[[@CR5]\]. Late complications include vaginal discharge, fibroid expulsion and impaction, infection, amenorrhea, and sexual dysfunction \[[@CR5]\]. The rate of hysterectomy subsequent to UAE ranges from 0.25 to 1.6 % \[[@CR2]--[@CR4]\]. Uterine necrosis is a rare complication after UAE, and it necessitates hysterectomy and treatment with antibiotics to prevent bacteremia, sepsis, and death \[[@CR2], [@CR6]\]. In our case, we believe that the uterine fibroids became necrotic following UAE and the necrotic tissue caused hyperkalemia and acute kidney failure. Some reports have mentioned that patients with a septic uterus required urgent surgery 7 or more days after the initial procedure \[[@CR2], [@CR6]\]. However, in our case, necrotic tissue caused serious complications and necessitated surgical intervention within 48 hours. In addition, our patient needed a blood transfusion in spite of the small intraoperative blood loss; we think a blood transfusion was needed because after UAE the necrotic uterine fibroids lost blood. Previous reports found that the size of the uterine fibroid was not associated with complications after UAE \[[@CR7]--[@CR10]\]. In one report, complications were found to be associated with a large uterus size (500 mL), large dominant tumor volume (100 cm^3^), and high post-intervention creatine kinase level (170 U/L) \[[@CR11]\]. Among previous reports, the largest fibroid was approximately 4000 cm^3^ \[[@CR7]--[@CR10]\], while the uterus in the present case was 9964 cm^3^; therefore, it was not possible to generalize the previous findings to the present case. In our case, fibroid volume was the most important risk factor for serious complications. It has been reported that surgical intervention should be performed within 48 hours after preoperative UAE \[[@CR12]--[@CR15]\]. Therefore, for the prevention of serious complications, such as those in the present case, we suggest that surgical intervention should be performed immediately after preoperative UAE. Conclusions {#Sec4} =========== We reported the occurrence of serious complications, including hyperkalemia and acute kidney failure, after preoperative UAE for a large uterine fibroid. Preoperative UAE is effective for preventing blood loss. The findings of the present case indicate that UAE performed for a large uterine fibroid can cause hyperkalemia and acute kidney failure. We would like to thank the consultants of the Division of Gynaecology, the Department of Emergency Medicine, and the Department of Pathological Medicine, Hachinohe City Hospital. Funding {#FPar1} ======= There is no source of funding for the preparation of this manuscript. Availability of data and materials {#FPar2} ================================== Not applicable. Authors' contributions {#FPar3} ====================== KT, TK and TH carried out surgical intervention. KT, TK, TH, and NI contributed to the discussion of this case. All authors read and approved the final manuscript. Competing interests {#FPar4} =================== The authors declare that they have no competing interests. Consent for publication {#FPar5} ======================= Written informed consent was obtained from the patient for publication of this case report and any accompanying images. A copy of the written consent is available for review by the Editor-in-Chief of this journal. Ethics approval and consent to participate {#FPar6} ========================================== Not applicable.

Background {#Sec1} ========== "Homeless" is a term generally conceptualised to refer to a group of individuals who have no regular access to a decent and conventional housing \[[@CR1]\]. However, the concept of decent housing is influenced by local cultural values, hence the difficulty to find a worldwide accepted definition \[[@CR2]\]. The homelessness continues to be a global and local social problem, but its true prevalence is underestimated. Urban centres have many people, seemingly homeless, poor and ragged, living all their lives in the street \[[@CR3]\]. This is an worldwide and complex phenomenon of big cities that has been increasing in the last decades, particularly due to socio-economic factors \[[@CR4], [@CR5]\]. Despite this evolving epidemiology there is scarcity of studies on the prevalence of mental illness in the homeless \[[@CR4]\]. Homelessness is known to increase the risk of mental illness, thus raising concerns in mental health providers \[[@CR6], [@CR7]\]. In the United States between 33 and 50% of homeless had schizophrenia, and a considerable proportion had substance use disorder, alcoholism and personality disorders \[[@CR3], [@CR8]\]. Particularly in children, the high level of exposure of the homeless to abuse and to urban violence can increase risk of developing mental illness \[[@CR9], [@CR10]\] in adulthood. On the other hand, schizophrenia can affect cognition and promote impairment of social and professional functioning \[[@CR11]\]. Effective intervention strategies may be adapted to promote familial reintegration of homeless people. Intervention models such as the assertive community treatment, were conceived aiming at hospital as well as those community-based approaches \[[@CR12]\]. They focus not only on the homeless but also involve the relatives, so as to reduce risk factors for violence, physical abuse and victimization \[[@CR1], [@CR9]\]. Between 25 and 50% of homeless populations have some sort of mental disorder in high-income countries. Alcohol dependence, drug addiction and psychotic disorders are among the most common mental health problems identified in previous studies \[[@CR8]\]. Little is known about the situation in low middle-income countries, but the impact of the mental disorders among the homeless people is higher in these settings due to shortage of mental health services. Particularly in Africa studies on homeless people are scarce, and thus the size of the problem in the region remains unknown \[[@CR13]\]. In Mozambique, although the National Mental Health Program has been providing and implementing services to reduce the gap in treatment of mental disorders, using strategies that include training of psychiatric technicians, there are no specific programs for homeless people. Moreover, there are no specific interventions for the homeless mentally ill in the country, which would combine hospital and community interventions, offered by a multidisciplinary team, composed by psychiatrists, psychologists, psychiatric technicians, nurses and social workers. A pilot study was designed to evaluate the profile of homeless with apparent mental illness in one urban and one suburban area in a low-income country. Using a referral strategy from community to hospital settings, we aimed to understand the mental health status of the homeless people, as well as to assess potential predictors of family integration, implementing a multidisciplinary approach. Thus, the present work had two main objectives: (a) to characterise the mental health status of the homeless people in Maputo and Matola utilising standardised clinical and sociodemographic assessment; and (b) to look at potential predictors of family integration among homeless submitted to psychiatric and psychosocial treatment. Methods {#Sec2} ======= Target population {#Sec3} ----------------- The study took place from August 2008 to December 2010, in Maputo city and Matola area, both in the southern region of Mozambique. These cities were chosen because of their accessibility and due to the dimension of the homeless population, representing an urban and suburban area respectively. The homeless were assisted at Infulene Psychiatric Hospital, the main psychiatric health facility in the country admitting patients with mental disorders, as well as providing psychiatric, psychological, occupational and social assistance. Study design {#Sec4} ------------ Homeless subjects from Maputo city and Matola area were recruited and referred to the Infulene Psychiatric Hospital. The sample comprised homeless people selected by convenience and included in the study using a snowball strategy. Patients with severe physical illness, under the age of 18 who were unable to give informed consent or those who were not in contact with their relatives were excluded. Patients discharged home from Infulene Psychiatric Hospital after being identified and referred by the researchers from homeless communities in Maputo and Matola to hospital, assessed, treated and provided group psychotherapy were eligible for this study. Study sample {#Sec5} ------------ After psychiatric assessment, patients started the recommended psychotropic treatment. Laboratorial analyses were performed to exclude organic conditions; in positive cases a full clinical examination was requested. Once the symptoms were controlled, the patients were submitted to a psychological evaluation and initiated group psychotherapy sessions and psychopharmacological treatment according to the individual needs. Group psychotherapy sessions included the following: training of social skills (communication, social interaction and assertive behaviors); cognitive stimulation and training of activities of daily living (personal hygiene, hygiene of spaces and standardized mealtimes). All patients were submitted to the same group psychotherapy and occupational therapy. Patients were also exposed to occupational therapy sessions to develop and improve autonomy and self-confidence. In a subsequent phase, group psychotherapy sessions and psychoeducation were also held involving patients' relatives, as part of family reintegration preparation process. Hospital stay varied from 3 to 6 months. Meanwhile, social workers performed 1--4 home visits to the relatives of the participants so as to create the appropriated environment to the family reintegration process. These visits were done in coordination with the community leaders and the municipal authorities after permission was obtained from patients and their relatives. Throughout these visits, the family social situation, the level of family involvement regarding the therapeutic process, the patient's health status, and the search information of their family member's disease were assessed. We defined family reintegration as the process of return of the homeless from an institution or shelter to its original, extensive or adopted family \[[@CR14]\]. In our study, we consider family reintegration as the return of the participant to family of origin after completing the treatment at the hospital level. We followed every patient for three months after family reintegration. Social workers conducted domiciliar monitoring of the families to provide additional information about disease and its management. Additionally, patients were registered and followed in a health unit near their residence areas as per usual standard of care in Mozambique. The support of clinical staff and social workers was maintained after the study period. Data collection method {#Sec6} ---------------------- The diagnosis was established by experienced psychiatrists through a structured clinical interview, the Mini International Neuropsychiatric Interview (MINI), which determines the likelihood of a psychiatric diagnosis by ICD10 \[[@CR15]\]. Socio-demographic data was collected using a questionnaire that also included a dichotomized question (yes/no), addressing whether the patient was present or not in the first, second and third monthly visit made by the health professionals. The questionnaire included questions for assessment of social skills adapted from social skills evaluation scale (EEHS) \[[@CR16]\], namely: verbal skills, non-verbal skills and conflict resolution that were explored in the three visits. Data analysis {#Sec7} ------------- Data analysis was performed using the Statistical Package for Social Sciences version 22, for the quantitative data analysis. All the collected information was coded including the identification of patients, relatives, health professionals and community leaders involved in the study. After three months of discharge a health professional was sent to their homes to check if the patient was leaving with their relatives and these was treated as a dichotomy variable. The presence of the patient after three months was defined as "reintegration". The present study was waived of approval by the National Bioethics Committee for Health (Ref. 343/CNBS), as it was part of an activity held by the Ministry of Health. Results {#Sec8} ======= Eighty-three homeless individuals were identified with apparent mental illness. Of these 12 refused to participate in the study. The 71 homeless recruited (85.5%) had a mean age of 37.83 ± 6.61 years, and 66 (93.0%) were male. Following identification and informed consent all 71 were referred for in-hospital treatment, and after clinical evaluation and social status assessment 38 (53.5%) (n = 38) were reintegrated in their nuclear or foster families. The remaining 33 (46.5%) stayed in the hospital as residents, or returned to the streets. From the participants in the study 59 (83.1%) were single and 12 (17.0%) married. Fifty-seven (80.3%) homeless were unemployed and 14 (19.6%) had an informal job such car washer, street or market seller, as well as other occasional jobs. The educational level was primary or secondary in 59 (83.1%) and 6 (8.4%) of the homeless had pre-university or university level; the latter have been entirely reintegrated. Regarding their origin 34 (47.9%) street residents were from provinces outside the southern region of Mozambique; Maputo City followed with 30 (42.3%). Before living on the street 32 (45.1%) lived in the suburban neighbourhoods of Maputo City, whilst 24 (33.4%) could not provide reliable data regarding their previous address (Table [1](#Tab1){ref-type="table"}).Table 1Comparison between socio-demographic data and familial reintegrationVariablesFamilial reintegrationNo\ n (%)Yes\ n (%)Total\ N (%)Gender Male31 (43.7)35 (49.3)66 (93.0) Female2 (2.8)3 (4.2)5 (7.0)Marital status Single29 (40.8)30 (42.3)59 (83.1) Married/partners2 (2.8)4 (5.6)6 (8.5) Divorced/separated2 (2.8)4 (5.6)6 (8.5)Occupational status Unemployed27 (38.0)30 (42.3)57 (80.3) Formally employed1 (1.4)4 (5.6)5 (7.0) Informally employed4 (5.6)5 (7.0)9 (12.6)Education Illiterate4 (5.6)2 (2.8)6 (8.5) Primary21 (29.6)15(21,1)36 (50.7) Secondary8 (11.3)15 (21.1)23 (32.4) Pre-university0 (0.0)5 (7.0)5 (7.0) Superior0 (0.0)1 (1.4)1 (1.4)Place of birth Maputo10 (14.1)20 (28.2)30 (42.3) Maputo province4 (5.6)3 (4.2)7 (9.9) Outside Maputo19 (26.8)15 (21.1)34 (47.9)Province residence before living on the street Maputo city9 (12.7)23 (32.4)32 (45.1) Maputo province4 (5.6)4 (5.6)8 (11.3) Outside Maputo3 (4.2)4 (5.6)7 (9.9) No information17 (23.9)7 (9.9)24 (33.4) Schizophrenia and other psychoses were present in 46 (64.8%) participants, mental and behavioural disturbance due to psychoactive substances in 21 (29.6%) and intellectual disability in 4 (5.6%). In general patients did not provide data related to previous treatment; for many the study provided the first contact with mental health professionals. Patients who gave information about their previous illness and hospital admissions reported diagnosis ranging from one to 30 years. Discharge by drop-out occurred in 28 (39.4%) patients. A statistically significant association ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi_{(2)}^{2}$$\end{document}$ = *46.1; p* = *0.000)* was verified between type of discharge and family reintegration (Table [2](#Tab2){ref-type="table"}).Table 2Association between diagnostic, type of discharge and familial reintegrationVariablesFamilial reintegration$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi_{(df)}^{2}$$\end{document}$ *; p value*NoYesTotalDiagnosis (ICD-10)Schizophrenia and other psychosis22 (31.0)24 (33.8)46 (64.8)$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi_{(2)}^{2}$$\end{document}$ = *6.1; p* = *0.047*Intellectual disability4 (5.6)0 (0.0)4 (5.6)Alcohol and substance use disorder7 (9.9)14 (19.7)21 (29.6)Type of discharge Clinical5 (7.0)33 (46.5)38 (53.5)$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi_{(2)}^{2}$$\end{document}$ = *46.1; p* = *0.000* Drop-out26 (36.6)2 (2.8)28 (39.4) By request0 (0.0)3 (4.2)3 (4.2) Transference2 (2.8)0 (0.0)2 (2.8) Reintegration in family was associated with the clinical diagnosis ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\chi_{(2)}^{2}$$\end{document}$ = *6.1; p* = *0.047*); 24 patients reintegrated (33.8%) had schizophrenia/other psychoses, 14 (19.7%) had mental and behavioural disturbance due to psychoactive substances, and none had intellectual disability. Reintegrated patients' relatives possessed reasonable to good information about the mental illness in 30 cases and 24 (63.2%) were from low socio-economical level (Table [3](#Tab3){ref-type="table"}). Of these 38 families with reintegrated patients, 23 (60.6%) had reasonable to very good involvement with the patient's situation.Table 3Results of the familial reintegration, level of family involvement and family information about the illnessVariablen (%)Reintegration No33 (46.5) Yes38 (53.5)Level of family involvement (n = 38) Very low7 (18.4) Low8 (21.1) Reasonable6 (15.8) Good12 (31.6) Very good5 (13.2)Information that the family possesses about the illness (n = 38) Weak8 (21.1) Reasonable16 (42.1) Good14 (36.8)Family socio-economical level (n = 38) Medium high2 (5.3) Medium10 (26.3) Low24 (63.2) Very low2 (5.3) Discussion {#Sec9} ========== In this study in two areas of southern Mozambique we were able to map, screen, treat and reintegrate homeless with mental illness. The majority of participants were male, single and unemployed. More than half of the sample presented the diagnosis of schizophrenia/other psychoses, followed by mental disorders resulting from addiction, corroborating findings from studies that showed high prevalence of these conditions in people living in the streets \[[@CR8]\], and consider them as a driving force for the individual to live in the streets \[[@CR6]\]. Like others \[[@CR17]\] our results suggests an advantage for patients with schizophrenia/psychosis and mental/behavioural disorders resulting from substance abuse being to being reintegrated, compared to those with intellectual disability. The reduced reintegration of participants with intellectual disability might be explained by the stigma associated to intellectual disability. Intellectual disability being not curable may probably lead to relatives thinking about the patient as a permanent burden. Additionally, feelings of shame due to non-biological explanations for their condition \[[@CR18]\] may hamper the possibility of relatives taking their family member back home and assume again the responsibility for their care. Indeed, the conceptual framework for mental disorders may influence relative's attitudes and determine seeking care pattern. In our setting people attribute intellectual disability to unknown rather than supernatural causes, in contrast with psychosis and epilepsy considered to be have supernatural origins \[[@CR17]\]. Socio-demographic factors such as being married, having a nuclear family, being employed, having higher level of education before living in street appear to be protective for homeless people \[[@CR5]\] and to improve the chances of reintegration \[[@CR19], [@CR20]\]. In our study the reinsertion rate was higher in the female gender, suggesting that women are culturally more protected than men; additionally, having children and living with a partner seem to be protective factors \[[@CR19]\] and to influence reintegration in this particular group \[[@CR20]\]. The majority of reintegrated patients' relatives had reasonable to good information about the mental illness; this seems to have contributed to patients' reintegration and reinforces the importance of family involvement throughout process, including in the pharmacological treatment, rehabilitation and psychotherapy. Although this is not a causal study the fact that most participants presented mental illness, suggests that this condition may be a driving force (*driver*) for a person to live in the street \[[@CR6]\]. Severity of symptoms can lead to neglect of basic personal care (hygiene, seeking for services or resources for social support and health care in the community) \[[@CR3]\], besides decreasing the *coping* abilities \[[@CR5]\]. Schizophrenia and substance abuse are the two most prevalent disturbances in our sample, like in several epidemiological studies \[[@CR7], [@CR21]\], and are risk factor for becoming homeless \[[@CR21]\] with strong influence in the socio-family reintegration. Institutionalization can lead to lack of privacy for residents, but it is known that living in the street may be related to difficulty in socialisation \[[@CR22]\]. Our drop-outs were related to the difficulty to adhere to treatment plan, especially in learning social skills. Usually patients were unable to meet schedules and conform to living rules for the group. Additionally, some participants had had problems with the neighborhood before leaving their residences (burning houses or attacking people) leading to their relatives to avoid living with them again. Relatives would rather visit them in the hospital regularly. Difficult socialisation in hospital environment led to patients abandoning the institution. Dropout from hospital treatment may also be related to the way health providers take care of their clients, some of whom have already created forms of resilience to adapt to life in the street \[[@CR22]\] and interact with people around. Moreover, patients discharged following clinical improvement presented higher chances of being reintegrated (46.5%) than those who were discharged at their own request (4.2%) or drop out (2.8%), suggesting that clinical improvement may be an important factor to facilitate family reintegration. There are several limitations to this study namely lack of preliminary information on the number of homeless and the fact that its findings cannot be generalised to other Mozambican cities. The sample selection procedure chosen (convenience and snowball effect) can induce selection biases related to choosing patients who belong to a specially exposed and stigmatised population. The sample size also constitutes a challenge, considering that the total population of homeless is known. Because we did not have a control group we cannot assess if the success of reintegration was due to the study itself or to unknown confounding factors. Moreover, since only those who were willing to participate were included (or those whose relatives were contacted), homeless with no identified relatives were not represented. Finally, the fact that additional assessment of the global functioning and comorbidities (such as personality and evaluation of traumatic events) had not been included in the study constitutes also a limitation. Despite these limitations our study positively contributed to enhance comprehension regarding potential factors associated to family reintegration that may be considered in strategies to reduce the gap in the treatment of mental illness in Mozambique, specially through utilisation of available resources, and ensuring sustainable implementation of culturally accepted strategies directed to homeless people with mental illness. Conclusions {#Sec10} =========== Homelessness coexists with mental illness and is aggravated by low socio-economical level, low school level, unemployment and low healthcare access. Unfavourable socio-demographic factors such as being single or divorced, unemployment, poverty and lack of family involvement influenced family reintegration of the homeless. Schizophrenia, other psychotic disturbances and substance abuse were the conditions with facilitated family reintegration. Reintegration of the homeless that can be achieved with relatively inexpensive integrated approaches, using locally available resources, may help reduce the number of homeless people with mental disturbance in the streets. There is need for research to better understand homelessness and elaborate tailored and culturally adapted psychosocial interventions with multisectoral involvement to reduce the gap in mental care in low-income settings. LG and HM wrote and designed the protocol, performed and supervised data collection and wrote the manuscript; FM revised the protocol and did statistical analysis; WF and DM wrote and revised the manuscript, AOM and JM did overall revision of the protocol and manuscript. All authors read and approved the final manuscript. Acknowledgements {#FPar1} ================ We would like to express our gratitude to all the participants of the study. The success of this study was possible due to the commitment and professionalism of a large team composed by professionals from the mental health teams of the Department of Mental Health in the Ministry of Health, in the City and Province of Maputo, researchers, community leaders, and relatives. Our thankfulness is extended to the different institutions and political decision-makers, who made the successful implementation of this work possible. We could not end without addressing our special appreciation to Prof. Dr. Paulo Ivo Garrido, who was by that time the Minister of Health of Mozambique. The authors are also grateful to the comments and suggestions made by the referees of this paper. This publication was possible because of financial support from Mental Health Implementation Research Project for Official Portuguese Speakers Countries-FOGARTY/National Institute of Mental Health---D43TW009675. Competing interests {#FPar2} =================== The authors declare that they have no competing interests. Availability of data and supporting materials section {#FPar3} ===================================================== Data will not be shared because there is a need of changing the content in it in order not to break participant confidentiality. This process is still in course. Ethical considerations and consent to participate {#FPar4} ================================================= The present study was waived of approval by the National Bioethics Committee for Health, since it was realized on behalf of an activity held by the Ministry of Health: Ref. 343/CNBS. It was explained to the participants and their relatives the objectives of the study and they were asked to sign the informed consent after the explanation. Confidentiality of the information and the freedom to not accept participate or to discontinue participation without any damage resulting of it were guaranteed as well. Funding {#FPar5} ======= The funding for the design of study and collection of data was made by the Ministry of Health of Mozambique. National Institute of Mental Health---Fogarty D43TW009675 supported all the aspects of the research related to analysis, interpretation of data and in writing the manuscript. Publisher's Note {#FPar6} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

1. Introduction {#sec1-sensors-20-02641} =============== In recent years, the field of indoor localization has increased in popularity due to both the increasing number of applications \[[@B1-sensors-20-02641]\] in domains such as surveillance \[[@B2-sensors-20-02641]\], navigation (both assistive and general purpose) \[[@B3-sensors-20-02641]\], robotics \[[@B4-sensors-20-02641],[@B5-sensors-20-02641],[@B6-sensors-20-02641]\], and Augmented Reality (AR) \[[@B7-sensors-20-02641]\] and the many proposed solutions that differ in terms of the devices used for tracking, the type of sensor data, and the localization algorithms. This paper focuses on computer vision based localization methods; therefore, the solutions presented are based on input from cameras. Most navigation systems use cameras carried by the subject, which represents the mobile entity (e.g., person, robot) that requires positioning or tracking, as illustrated in the left-hand side of [Figure 1](#sensors-20-02641-f001){ref-type="fig"}. The other type of solutions uses an infrastructure of static cameras positioned at known locations throughout the building to track the subject, as shown in the right-hand side of [Figure 1](#sensors-20-02641-f001){ref-type="fig"}. The vision based localization systems use 2D or 3D cameras (e.g., stereo, depth, RGB-D cameras) and perform the localization by identifying artificial markers (such as Quick Response (QR) codes and fiducial markers like AprilTags, ARTags, and CALTags \[[@B8-sensors-20-02641]\]) or objects that are part of the environment. In many cases, the cameras are used in combination with other sensors such as WiFi, beacon, or inertial sensors \[[@B1-sensors-20-02641]\]. Depending on the application, the required level of location accuracy varies \[[@B9-sensors-20-02641]\]. Navigation solutions for guiding people to find specific rooms in a building or when changing underground lines accept an accuracy of several meters. In the same accuracy range are assistive solutions for the elderly that monitor their approximate location to confirm their compliance with certain routines and to detect situations of emergency such as a person ceasing to move. Other tracking or surveillance applications require 1--2 m accuracy to assess risky situations such as a person getting too close to an exhibit item in a museum. However, some of these surveillance applications require a higher accuracy of 10--20 cm when aiming to detect whether restricted areas/perimeters are only entered by authorized people. Applications for indoor autonomous robots or assistive systems for the visually impaired that perform obstacle detection cannot rely on an approximate localization and need centimeter-level accuracy. Modern AR solutions take the accuracy requirements even further. To offer seamless integration of the multimedia content, superimposed over video flows on smartphones or over smart glasses' lenses, these applications require centimeter to millimeter accuracy of the position and orientation of the user's mobile device. Even though a considerable number of surveys on indoor localization have been published \[[@B1-sensors-20-02641],[@B4-sensors-20-02641],[@B5-sensors-20-02641],[@B6-sensors-20-02641],[@B9-sensors-20-02641],[@B10-sensors-20-02641],[@B11-sensors-20-02641],[@B12-sensors-20-02641],[@B13-sensors-20-02641],[@B14-sensors-20-02641],[@B15-sensors-20-02641]\], as this research space continuously developed and the types of localization solutions diversified, we find that for the area of vision based localization, the majority of the previous surveys could have better focus as they are too general (encompassing all kinds of sensing devices) or too specific (addressing only a segment of the vision based localization problem, such as Simultaneous Localization and Mapping (SLAM) \[[@B16-sensors-20-02641],[@B17-sensors-20-02641],[@B18-sensors-20-02641],[@B19-sensors-20-02641],[@B20-sensors-20-02641],[@B21-sensors-20-02641]\], Structure from Motion (SfM) \[[@B22-sensors-20-02641]\], or image matching \[[@B23-sensors-20-02641]\]). Other surveys discuss indoor positioning solutions particular to certain application domains. For instance, Huang et al. \[[@B24-sensors-20-02641]\] analyzed only localization solutions that combined visual and inertial information. Marchand et al. \[[@B7-sensors-20-02641]\] provided a survey of pose estimation methods used only for AR. Silva and Wimalaratne \[[@B25-sensors-20-02641]\] presented a survey of navigation and positioning aids for the visually impaired. In comparison, we offer a comprehensive survey of image based localization solutions regardless of the application domain and propose a new classification. The paper is organized as follows: the next section describes the most impacted domains by indoor localization; [Section 3](#sec3-sensors-20-02641){ref-type="sec"} classifies 70 selected computer vision based indoor localization methods and details their main characteristics; [Section 4](#sec4-sensors-20-02641){ref-type="sec"} focuses on benchmarks used for evaluating image based indoor localization solutions; and [Section 5](#sec5-sensors-20-02641){ref-type="sec"} presents our conclusions. 2. Application Domains {#sec2-sensors-20-02641} ====================== 2.1. Assistive Devices {#sec2dot1-sensors-20-02641} ---------------------- With the advance in technology, researchers have focused on improving the lifestyle of people with various disabilities, including visual impairment \[[@B26-sensors-20-02641]\]. Two of their main problems are navigating and perceiving unknown environments. There are several solutions to this problem that map the environment with images containing different visual cues such as QR codes, bar codes, or other simple synthesized geometric shapes like circles and triangles. Idrees et al. \[[@B27-sensors-20-02641]\] proposed an indoor navigation system that used QR codes that were placed on the floor at certain locations. The system guided the user to a selected destination, checking the user's location every time a QR code was scanned. Fusco and Coughlan \[[@B28-sensors-20-02641]\] used sign detection and visual-inertial odometry to estimate the user's location inside a building, requiring only a digital map of that building that contained the locations of the signs. Other solutions create a 3D reconstruction of the environment in a configuration stage or create a database with images of the indoor space, annotated with location information. Endo et al. \[[@B29-sensors-20-02641]\] proposed a navigation system for visually impaired people, which applied Large-Scale Direct SLAM (LSD-SLAM) to estimate the user's position while constructing a 3D map of the environment. The 3D model of the environment allowed for the construction of an occupancy grid map, divided into quadrate cells, which stored information about the presence or absence of an obstacle in that location. The system created a cost map and conducted path finding through the navigation stack provided by the Robot Operating System (ROS) framework \[[@B30-sensors-20-02641]\]. Li et al. \[[@B19-sensors-20-02641]\] detected dynamic obstacles and applied path planning to improve navigation safety for people with visual disabilities. They built a 3D reconstruction of the environment with the visual positioning service provided by the Google Tango device. With a time-stamped map Kalman filter, they implemented an obstacle detection and avoidance algorithm that guided the user to a specified destination. As previously mentioned, some localization solutions use an infrastructure of static cameras located at known positions in the environment. Heya et al. \[[@B31-sensors-20-02641]\] employed color detection in an indoor localization system used for visually impaired people. A static camera located on the ceiling tracked the screen of a smartphone that was placed on the user's shoulder. Chaccour and Badr \[[@B32-sensors-20-02641]\] proposed an ambient navigation system that was composed of static cameras attached to the ceiling. The system detected the users' location and orientation based on markers located on their heads. Static and dynamic obstacles were identified based on their shape or predefined images that were stored in a database. The proposed solution provided navigation assistance and obstacle avoidance and allowed the visually impaired users to locate missing objects. Many navigation systems for visually impaired people assume the existence of a map for the current environment or create such a map in a configuration stage \[[@B25-sensors-20-02641]\]. However, there are some solutions that allow the user to navigate in unknown environments by simply translating the visual information through audio or haptic signals, allowing the user to create a mental map of the surroundings. Sound of Vision \[[@B33-sensors-20-02641],[@B34-sensors-20-02641]\] is such an example, which identifies the most important objects in the proximity of the user and sends information about their characteristics (weight, height, elevation, distance to the user, etc.) through headphones and a haptic belt. Sound of Vision is not only a navigation system, but also a solution for perceiving the environment. However, the understanding of the audio and haptic information that characterize the environment can be accomplished only through intensive training \[[@B35-sensors-20-02641]\]. More information about indoor positioning systems for visually impaired people can be found in Siva's and Wimalaratne's survey \[[@B25-sensors-20-02641]\]. 2.2. Autonomous Robots {#sec2dot2-sensors-20-02641} ---------------------- Designing autonomous robot applications can represent a challenge, since the localization methods cannot rely on external information. When navigating an environment, a human being can use the five senses, especially vision, touch, and hearing, to create a mental representation of the surroundings. This is not the case for robots, which navigate the environment only based on the information provided by the localization system. Therefore, localization solutions for autonomous robots require continuous computation of the robot's position and orientation relative to a digital representation of the environment, as well as obstacle detection and path planning. On the other hand, designing applications for specific robots can simplify the localization problem. Various characteristics of a robot, such as degrees of freedom, width, height, position of the sensors mounted on the robot, or wheel diameter can be used to make assumptions about the movement of the robot (dead reckoning), thus reducing the complexity of the localization algorithms. Since most robots operate in controlled environments, a popular approach is to configure the space with artificial landmarks such as QR codes. Li and Huang \[[@B18-sensors-20-02641]\] presented a system that assisted robots, as well as human beings, in navigating indoor environments. A Kinect device acquired color information that allowed the detection of QR codes attached to the walls at known locations in a room. The depth sensor measured the distance from the Kinect camera to the identified QR codes. Babu and Markose \[[@B36-sensors-20-02641]\] proposed a navigation system for Internet of Things (IoT) enabled robots. A QR code based detection solution estimated the position of the robot, while a path optimization step based on Dijkstra's algorithm assisted the robot in reaching a destination node. Nazemzadeh and Macii \[[@B37-sensors-20-02641]\] described a localization solution for unicycle-like wheeled robots. It computed the position of the robots by fusing information from QR codes, odometry based on dead reckoning, and a gyroscope platform. Cavanini et al. \[[@B38-sensors-20-02641]\] proposed a low-cost QR code based localization system for robots operating indoors, experimentally validated on smart wheelchairs. Other approaches used actual images of the environment, acquired with 2D or 3D cameras. Correa et al. \[[@B39-sensors-20-02641]\] described a Kinect based reactive navigation system that guided robots while performing obstacle avoidance. It recognized different configurations of the indoor space with an artificial neural network. Xin et al. \[[@B40-sensors-20-02641]\] introduced an RGB-D SLAM method that combined the Oriented FAST and Rotated BRIEF (ORB) and Random Sample Consensus (RANSAC) algorithms for feature extraction and matching. They created a 3D volumetric map of the environment that could be used for the navigation of a mobile robot. Kao and Huy \[[@B41-sensors-20-02641]\] proposed an indoor navigation system that performed smartphone based visual SLAM with ORB features using a wheel-robot. They combined WiFi signals, information from inertial sensors, and monocular images for the computation of the robot's position. Surveys dedicated to positioning solutions for autonomous robots \[[@B4-sensors-20-02641],[@B5-sensors-20-02641]\] present in-depth information on this topic. 2.3. Augmented Reality {#sec2dot3-sensors-20-02641} ---------------------- Currently, Augmented Reality (AR) has become very popular, due to the new technologies that are bringing it closer to the greater public. Smartphones are the most commonly used devices for displaying augmented content. However, smart glasses, such as Moverio \[[@B42-sensors-20-02641]\] or Google Glass \[[@B43-sensors-20-02641]\], are gaining ground \[[@B44-sensors-20-02641]\]. For a seamless integration of the multimedia content within the real environment, the position and orientation of the display device must be estimated with high accuracy. Gerstweiller \[[@B45-sensors-20-02641]\] presented HyMoTrack , a tracking solution that generated a 3D model of the environment out of a vectorized 2D floor plan, and an AR path concept, called FOVPath, for guiding people. Using the FOVPath approach, the display of a trajectory depended on the user's position and orientation and also on the Field Of View (FOV) capabilities of the device. Wang et al. \[[@B46-sensors-20-02641]\] described the development of a 3D augmented reality mobile navigation system that provided indoor localization based on Radio-Frequency Identification (RFID) readings and computer vision. They created 3D representations of internal and external structures of Oxford College, from the present and the past. Based on the position and orientation of the user's device, they displayed the 3D architectural appearance of the college during important time periods, as well as multimedia content such as texts, pictures, or 3D models related to various exhibits. Balint et al. \[[@B47-sensors-20-02641]\] presented an AR multiplayer treasure hunt game, which combined GPS position information with localization based on image recognition. The treasures were virtual 3D objects that were displayed when the user reached a checkpoint, and the device was oriented towards the respective direction. Baek et al. \[[@B48-sensors-20-02641]\] proposed an AR system for facility management, which computed the user's pose relative to the building, with a deep learning approach. The visualization module displayed location-specific information, holographic pipes in this case, which in reality were not visible because they were built within the walls. Marchand et al. \[[@B7-sensors-20-02641]\] offered more information on the topic of pose estimation for augmented reality, presenting the most important approaches on vision based positioning. ### AR Commercial Solutions This section presents some of the most popular commercial tools for developing augmented reality content, which have indoor localization capabilities. Wikitude \[[@B49-sensors-20-02641]\] is an augmented reality Software Development Kit (SDK) that uses the SLAM technology to reconstruct the environment. It also performs 2D and 3D image recognition and tracking, which can trigger the display of digital content, overlaid on the real world. ARKit \[[@B50-sensors-20-02641]\] is an iOS AR platform that provides scene understanding capabilities by combining inertial data with visual information to detect horizontal and vertical planes. It also recognizes images and 3D objects, determining the position and orientation of the camera relative to the target. ARCore \[[@B51-sensors-20-02641]\] is another SDK, launched by Google, which allows developers to build augmented reality experiences that seamlessly integrate the digital content into the real world. ARCore provides motion tracking capabilities, as well as environmental understanding based on plane detection. It performs SLAM, making use of inertial sensors and the data acquired with a smartphone camera, estimating the position and orientation of the user's device relative to a 3D coordinate system. Vuforia \[[@B52-sensors-20-02641]\] is a popular engine that provides detection and tracking of image targets and pose estimation of any tracked target or marker, allowing the rendering module to display the 3D virtual content naturally, depending on the position and orientation of the user. The pose computation is performed only relative to an image target or a marker, therefore not offering actual indoor positioning capabilities (relative to an entire room or another type of indoor space). ARToolKit \[[@B53-sensors-20-02641]\] is an open-source library intended for the development of augmented reality applications, which overlays 2D and 3D multimedia content on the real world. It is a tracking library that computes the camera position and orientation relative to square markers or to natural feature markers in real time. It works with both monocular and stereo cameras, providing calibration capabilities. MAXST \[[@B54-sensors-20-02641]\] is a cross-platform engine that provides a variety of tracking features for the development of augmented reality applications. It recognizes and tracks planar target images or planar surfaces, as well as markers with regular patterns or QR codes. Their implementation of visual SLAM can be used to create map files of the environment that are later loaded up by the Object Tracker module, which superimposes AR experiences on them. Another module, AR Fusion tracker, generates world representations and performs environment tracking by combining information from the other tracking modules. Other popular commercial solutions for developing augmented reality applications are EasyAR \[[@B55-sensors-20-02641]\], Kudan \[[@B56-sensors-20-02641]\], Onirix \[[@B57-sensors-20-02641]\], Pikkart \[[@B58-sensors-20-02641]\], and DeepAR \[[@B59-sensors-20-02641]\]. 2.4. Surveillance and Monitoring {#sec2dot4-sensors-20-02641} -------------------------------- Indoor positioning can also be used for surveillance or monitoring purposes, detecting whether an unauthorized person has breached a perimeter, or tracking a certain person throughout an entire building or house. Generally, surveillance systems use an infrastructure of static cameras for the purpose of detecting and tracking the users. However, there are cases when information from other sensors, such as WiFi access points or beacons, is fused with the video frames. Sun et al. \[[@B60-sensors-20-02641]\] proposed a localization solution that used panoramic cameras and a map of the indoor environment. They applied a background subtraction method to detect human beings, matching their location to a corresponding position on the indoor map. Desai and Rattan \[[@B61-sensors-20-02641]\] used a pan/tilt camera and wireless sensor networks to track objects within an indoor space. The estimation of an object's position was performed with the time difference of arrival method. The camera, equipped with a laser pointer, followed the object continuously, by computing the pan and tilt angles based on a listener Cricket mote carried by the object. Grzechca et al. \[[@B62-sensors-20-02641]\] fused Received Signal Strength Indication (RSSI) information with data from a static video camera to track human beings in indoor environments. Zhang et al. \[[@B63-sensors-20-02641]\] acquired video sequences with a surveillance camera and recognized a target person by matching the information provided by the inertial sensor of the person's smartphone with gait and heading azimuth features extracted from the videos. They applied a Convolutional Neural Networks (CNN) based object tracking technique in order to handle occlusion. As far as we know, there is no recent survey on computer vision based indoor localization dedicated to surveillance and monitoring, but further information on this topic can be found in the work of Shit et al. \[[@B64-sensors-20-02641]\], which presented localization solutions with static cameras, and in the survey of Jiao et al. \[[@B65-sensors-20-02641]\], which discussed deep learning methods for object positioning. 3. Indoor Localization Solutions {#sec3-sensors-20-02641} ================================ 3.1. Selection of Papers Included in the Survey {#sec3dot1-sensors-20-02641} ----------------------------------------------- Image based indoor localization has been intensely researched. Among existing scientific publications, we chose 70 papers based on publication date and relevance to the domain, using only prestigious research databases (IEEE Explore, ACMDigital Library, SpringerLink, MDPI, and Elsevier). [Figure 2](#sensors-20-02641-f002){ref-type="fig"} shows the distribution of selected papers over time, illustrating an increased interest in the image based localization domain in the last five years. Since it takes time for research papers to acquire visibility, the number of citations was not one of the selection criteria, as it disadvantaged the more recent research. The purpose of this survey was not to be exhaustive in terms of listing the work performed in the field of vision based indoor positioning, but to illustrate the main characteristics of the existing technologies and techniques. This allows the reader to attain an overview of the domain while understanding the advantages and drawbacks of the various methods. Choosing an appropriate solution does not boil down to just the application domain, but also to the particular requirements of the applications such as accuracy, computing time, equipment, and dynamic and static aspects (the properties of the objects contained in the environment). 3.2. Classification {#sec3dot2-sensors-20-02641} ------------------- Recent surveys in the indoor positioning domain have proposed various classifications. For instance, the survey of Yassin et al. \[[@B1-sensors-20-02641]\], which addressed the entire domain of indoor localization (not limited to vision based solutions), proposed a two-level classification. The first level grouped the solutions based on the positioning algorithms, which were divided into three classes: triangulation, scene analysis, and proximity detection. The second level classified the solutions within the first level classes based on the measurement techniques as follows: the triangulation class had two sub-classes: lateration and angulation; the scene analysis class had only one sub-class: fingerprinting based; and the proximity detection class had two sub-classes: cell-ID and RFID. Another general survey in indoor localization \[[@B10-sensors-20-02641]\] classified existing research solutions into local infrastructure dependent techniques (ultra-wideband, wireless beacons), local infrastructure independent techniques (ultrasound, assisted global navigation satellite systems, magnetic localization, inertial navigation systems, visual localization, infrared localization), and visual/depth sensors (structured light technology, pulsed light technology, stereo cameras). Mendoza-Silva et al. \[[@B9-sensors-20-02641]\] presented a meta-review of indoor positioning systems, resulting from the analysis of 62 indoor localization-related surveys. They reviewed the most commonly used technologies for localization applications and proposed the following classes: light, computer vision, sound, magnetic fields, dead reckoning, ultra-wideband, WiFi, Bluetooth Low Energy, RFID, and Near-Field Communication (NFC). In the computer vision class, they discussed several positioning techniques, such as visual odometry and vision based SLAM, and mentioned different acquisition devices (monocular, stereo, omnidirectional). However, they did not propose any classification for this domain. They also observed the complete lack of recent surveys on computer vision based indoor localization solutions and claimed the necessity of such a work. Analyzing the research papers mentioned in the previous section, several discriminating characteristics emerged. Therefore, we propose a new classification of computer vision based indoor localization solutions, as illustrated in [Figure 3](#sensors-20-02641-f003){ref-type="fig"}. All indoor positioning methods have a configuration stage, in which the environment is filled with landmarks and sensors, images from the environment are saved into a database, or a 3D representation of the indoor space is created. Therefore, environment data could consist of information about the position of the markers (e.g., QR codes, geometric synthetic identifiers) or the location of the static cameras placed within the scene. Another type of environment data is represented by databases with images or features from images, annotated with position and orientation information. Lastly, environment data could consist of a 3D model of the environment, a point cloud, a 3D mesh, or a 3D map, obtained with various methods such as manual modeling, SLAM, or SfM. Another element that helps discriminate between methods is the type of employed sensing devices. As previously mentioned in [Section 1](#sec1-sensors-20-02641){ref-type="sec"} and illustrated in [Figure 1](#sensors-20-02641-f001){ref-type="fig"}, the main acquisition devices are static and mobile cameras. Furthermore, the input information can be enriched with data from other sensors, such as WiFi access points or IMU devices. Another differentiating aspect of image based localization methods is the type of visual input, which can be either 2D or 3D. The localization methods can search for artificial markers (e.g., QR codes and other fiducial markers such as AprilTags \[[@B66-sensors-20-02641]\], ARTags, and CALTags \[[@B8-sensors-20-02641]\]) or for features from the real environment. The latter category includes any type of element that can be extracted from the real environment (without the need to insert synthesized items into the scene), either features of interest such as Speeded Up Robust Features (SURF) and Scale-Invariant Feature Transform (SIFT) or semantic objects. Therefore, we propose a new level of classification, namely the *detected elements*, which refers to the type of features (artificial markers or natural, real elements from the environment) that are tracked or matched within the images. Indoor positioning solutions employ various localization methods, which range from low-level feature matching to complex scene understanding. We divided the techniques into traditional image analysis and artificial intelligence. The ones belonging to the second category include any type of artificial intelligence, such as Bayesian approaches, Support-Vector Machine (SVM), and neural networks. We applied the proposed classification to the selected indoor localization solutions. [Table 1](#sensors-20-02641-t001){ref-type="table"} assigns each of the chosen research papers to a class, based on environment data, sensing devices, detected elements, and localization method. Out of all the classes that could result from combining the differentiating elements from [Figure 3](#sensors-20-02641-f003){ref-type="fig"}, we chose only 17 of the more popular ones, which were represented by a large number of research papers. The following sub-sections analyze each category, presenting representative indoor localization solutions and discussing their advantages and drawbacks. For each examined scientific paper, we include in [Table 2](#sensors-20-02641-t002){ref-type="table"}, [Table 3](#sensors-20-02641-t003){ref-type="table"}, [Table 4](#sensors-20-02641-t004){ref-type="table"}, [Table 5](#sensors-20-02641-t005){ref-type="table"}, [Table 6](#sensors-20-02641-t006){ref-type="table"}, [Table 7](#sensors-20-02641-t007){ref-type="table"}, [Table 8](#sensors-20-02641-t008){ref-type="table"}, [Table 9](#sensors-20-02641-t009){ref-type="table"}, [Table 10](#sensors-20-02641-t010){ref-type="table"}, [Table 11](#sensors-20-02641-t011){ref-type="table"}, [Table 12](#sensors-20-02641-t012){ref-type="table"}, [Table 13](#sensors-20-02641-t013){ref-type="table"}, [Table 14](#sensors-20-02641-t014){ref-type="table"}, [Table 15](#sensors-20-02641-t015){ref-type="table"}, [Table 16](#sensors-20-02641-t016){ref-type="table"}, [Table 17](#sensors-20-02641-t017){ref-type="table"} and [Table 18](#sensors-20-02641-t018){ref-type="table"} information about the characteristics of the datasets used for evaluation, the computing time or refresh rate (related to a certain running platform), and the achieved accuracy. If there were papers that did not report information about a certain characteristic, the field corresponding to that characteristic is marked with "-". Some papers evaluated their solutions only visually, while others applied various metrics, such as average and/or absolute errors for position and orientation, percentage of tested cases when accuracy was within certain intervals, Detection Success Rate (DSR), Root Mean Squared Error (RMSE), Navigation Success Rate (NSR), Relative Pose Error (RPE), and Absolute Trajectory Error (ATE). ### 3.2.1. Indoor Localization Solutions with 2D Static Cameras, Markers, and Traditional Image Analysis {#sec3dot2dot1-sensors-20-02641} This class of indoor localization methods uses an infrastructure of 2D static cameras with known locations. The images from these cameras are processed with traditional computer vision algorithms in order to detect synthetic identifiers carried by people or robots. Belonging to this class is the work of Heya et al. \[[@B31-sensors-20-02641]\], where the screen of the user's smartphone was detected with a simple color tracking algorithm. Each user was assigned a color, which was displayed on the smartphone, and the system tracked the screen of the device, which was placed on the user's shoulder. Another example of indoor localization solution using static 2D cameras and traditional image processing was an ambient navigation system proposed by Chaccour and Badr \[[@B32-sensors-20-02641]\], which detected the users' location and orientation based on markers located on their heads. The system was evaluated within a home composed of three rooms, kitchen, living room, and bedroom, each containing an IP camera placed on the ceiling. The tests performed with eight people, including dynamically added obstacles, proved the reliability of the system. The methods in this class require the map of the building and a configuration step that consists of annotating the positions of the static cameras on the map. They can achieve good, centimeter-level, accuracy, as can be seen in [Table 2](#sensors-20-02641-t002){ref-type="table"}, which makes them viable solutions for scenarios requiring high accuracy positioning in small spaces. However, maintaining this accuracy level in large indoor spaces comes with high costs in terms of both effort and infrastructure, due the cumbersome configurations and the high number of cameras required. ### 3.2.2. Indoor Localization Solutions with 2D Static Cameras, Real Features, and Traditional Image Analysis {#sec3dot2dot2-sensors-20-02641} In this class of indoor localization solutions, the images from the static cameras are processed with traditional computer vision algorithms in order to track objects or people and compute their positions within a certain room. Localization solutions within this class identify people or robots without the need for the tracked entities to carry devices or artificial markers. Bo et al. \[[@B67-sensors-20-02641]\] recursively updated the position of multiple people based on the detected foreground and the previous known locations of each person. The foreground was identified by analyzing changes in image structure (edges) based on the computation of the normalized cross-correlation for each pixel. They applied a greedy algorithm to maximize the likelihood of observing the foreground for all people. The efficiency of their algorithm was evaluated on public datasets, using the Multiple Object Tracking Accuracy (MOTA), a metric computed based on object misses, false positives, and mismatches. Shim and Cho \[[@B69-sensors-20-02641]\] employed a homography technique to create a 2D map with accurate object position, using several surveillance cameras. Dias and Jorge \[[@B68-sensors-20-02641]\] tracked people using multiple cameras and a two level processing strategy. Firstly they applied region extraction and matching to track people, and secondly, they fused the trajectories detected from multiple cameras in order to obtain the positions relative to a global coordinate system, using homography transformations between image planes. Sun et al. \[[@B60-sensors-20-02641]\] proposed a device-free human localization method using a panoramic camera. They employed pre-processing, human detection with background subtraction (with mean filtering and a Gaussian low pass filtering), and an association between the location of users in the image space and their location on a given map of the indoor environment. Compared to the previous class of solutions that use artificial markers, the methods in this class have slightly higher localization errors, as can be observed in [Table 3](#sensors-20-02641-t003){ref-type="table"}. However, this accuracy level (tens of centimeters) is still good for many types of applications, and these methods have a wider applicability, especially in the monitoring and surveillance domains, due to them not requiring the tracked entities to carry devices or markers. ### 3.2.3. Indoor Localization Solutions with 2D Static Cameras, Real Features, and Artificial Intelligence {#sec3dot2dot3-sensors-20-02641} This class of indoor localization methods differs from the class described in [Section 3.2.2](#sec3dot2dot2-sensors-20-02641){ref-type="sec"} by the type of employed algorithms for determining the entities' positions. An alternative to traditional image processing algorithms is artificial intelligence, in the form of Bayesian approaches, SVM, or neural networks. For instance, Utasi and Benedek \[[@B70-sensors-20-02641]\] proposed a Bayesian method for people localization in multi-camera systems. First, pixel-level features were extracted, providing information about the head and leg positions of pedestrians. Next, features from multiple camera views were fused to compute the location and the height of people with a 3D Marked Point Process (MPP) model, which followed a Bayesian approach. They evaluated their method on two public datasets and used the Ground Position Error (GPE) and Projected Position Error (PPE) metrics for accuracy computation. Cosma et al. \[[@B73-sensors-20-02641]\] described a location estimation solution based on 2D images from static surveillance cameras, which used pose estimation from key body points' detection to extend the pedestrian skeleton in case of occlusion. It achieved a location estimation accuracy of approximately 45 cm, as can be observed in [Table 4](#sensors-20-02641-t004){ref-type="table"}, in complex scenarios with a high level of occlusion, using a power efficient embedded computing device. See-your-room \[[@B74-sensors-20-02641]\] represents another localization solution that uses cameras placed on the ceiling. It employs Mask R-CNN and OpenPose \[[@B128-sensors-20-02641]\] to detect people and their pose (standing, sitting) and the perspective transformation to obtain the position of the users on a map. Hoyer et al. \[[@B71-sensors-20-02641]\] presented a localization framework for robots based on Convolutional Neural Networks (CNN) using static cameras. In a first stage, they used a CNN object detection to estimate the type and the bounding box of a robot. In the second stage, they ran two more neural networks, one for computing the orientation of the robot and another one to provide identification (based on a code placed on the robot). An algorithm was also proposed for generating synthetic training data by placing contour-cropped images of robots on background images. The solution described by Jain et al. \[[@B72-sensors-20-02641]\] was based on the assumption that, in an office, employees tend to keep their phones lying on the table and that the ceiling layout is unique throughout the building, containing different tiles. They used a combination of artificial intelligence and traditional image processing to detect landmarks such as ceiling tiles, heating or air conditioning vents, lights, sprinklers, audio speakers, or smoke detector sensors. First, they applied the Hough transform to extract tiles, then SURF for feature extraction, and SVM to classify the type of landmark with the ECOCframework \[[@B129-sensors-20-02641]\]. [Table 4](#sensors-20-02641-t004){ref-type="table"} presents the characteristics of the localization methods that use 2D static cameras, real features, and artificial intelligence based algorithms. The computational challenge of using neural networks or other AI based implementations can be met with the use of GPUs, as can be observed for several methods \[[@B71-sensors-20-02641],[@B73-sensors-20-02641]\], which achieve interactive or real-time performance. Although a higher complexity of the algorithms would lead to expecting a higher accuracy level compared to the previous class of solutions, relevant accuracy comparisons cannot be made due to the evaluations being performed on different datasets/scenarios. ### 3.2.4. Indoor Localization Solutions with 2D Mobile Cameras, Markers with Known Positions, and Traditional Image Analysis {#sec3dot2dot4-sensors-20-02641} This class of indoor localization solutions employs a configuration step, in which artificial landmarks, predominantly QR codes, are placed at known locations inside a building (generally on the ceiling, walls, or floor). These solutions make use of cameras attached to people or robots and apply traditional image processing during the localization stage. Each QR image codifies its position within the coordinate system of the building. Based on the appearance of the QR code in the acquired images during the localization stage, compared to the raw images of the QR codes, the orientation of the camera can also be estimated by computing the projective transform matrices. QR codes allow for fast detection and decoding of stored information. However, in cases where the video camera is moving fast, the detection of these codes can be difficult. This led Lee et al. \[[@B75-sensors-20-02641]\] and Goronzy et al. \[[@B76-sensors-20-02641]\] to surround their codes with simple borders such as circles or rectangles, which can be detected faster than QR codes with Hough transform. Ooi et al. \[[@B79-sensors-20-02641]\] used QR codes to reposition mobile sensor networks, in the form of four wheeled robots. When QR codes were not in range, the system estimated the position of the robot using dead reckoning. Lightbody et al. \[[@B78-sensors-20-02641]\] proposed WhyCode, a new family of circular markers that enable faster detection and pose estimation, of up to two orders of magnitude compared to other popular fiducial marker based solutions. They extended the WhyCon algorithm \[[@B130-sensors-20-02641]\], which localizes a large number of concentric black and white circles with adaptive thresholding, flood fill, and a circularity test. The position of a marker, along with the pitch and roll, was estimated based on eigenvalues with a method proposed by Yang et al. \[[@B131-sensors-20-02641]\]. The yaw was computed by detecting the Necklace code contained in the WhyCode marker. Benligiray et al. \[[@B80-sensors-20-02641]\] presented STag, a fiducial marker system that used geometric features to provide stable position estimation. The markers contained an inner circular border and an outer square border used for detection and homography estimation. They compared their detection capabilities against the ARToolkit, ArUco \[[@B132-sensors-20-02641]\], and RUNE-Tag \[[@B133-sensors-20-02641]\] fiducial markers. Khan et al. \[[@B81-sensors-20-02641]\] proposed a generic approach for indoor navigation and pathfinding using simple markers (ARToolkit) printed on paper and placed on ceilings. The orientation of the smartphone relative to a marker enabled the computation of the user's direction along a certain path. As can be observed in [Table 5](#sensors-20-02641-t005){ref-type="table"}, the performance of these methods is quite impressive. The centimeter or even sub-centimeter level position accuracy is achieved due to the precise matching mechanism when dealing with synthesized images. The fast detection and decoding of QR codes and fiducial markers enables real-time applications. Compared to the previously presented static camera based solutions, even though deploying such a system in a large built environment also comes with a considerable effort in the configuration stage, it is significantly less expensive (artificial markers are practically free in comparison to static cameras). However, the tracked entity is required to carry a mobile camera, which in certain scenarios can represent an inconvenience, and mapping a building with artificial images can have a negative impact on the building's appearance. Another important aspect when choosing marker based localization solutions is their detection success when facing occlusion. This problem was addressed in the solution proposed by Garrido-Jurado et al. \[[@B132-sensors-20-02641]\], which combined multiple markers with an occlusion mask computed by color segmentation. Sagitov et al. \[[@B8-sensors-20-02641]\] compared three fiducial marker systems, ARTag, AprilTag, and CALTag, in the presence of occlusion, claiming that CALTags showed a significantly higher resistance for both systematic and arbitrary occlusions. ### 3.2.5. Indoor Localization Solutions with 3D Mobile Cameras, Markers with Known Positions, and Traditional Image Analysis {#sec3dot2dot5-sensors-20-02641} Localization based on fiducial markers can also be performed by analyzing RGB-D images with traditional image processing methods. Li et al. \[[@B82-sensors-20-02641]\] used RGB-D images in order to detect and recognize QR landmarks with the Zbar \[[@B135-sensors-20-02641]\] code reader. The distance to the QR code was computed based on the depth image. Dutta \[[@B83-sensors-20-02641]\] proposed a real-time application for localization using QR codes from RGB-D images, based on the keystone effect in images from range cameras (the apparent distortion of an image caused by projecting it onto an angled surface). Some solutions achieve centimeter accuracy when computing the distance from the camera to the artificial marker (see [Table 6](#sensors-20-02641-t006){ref-type="table"}). These solutions are very practical, since 3D cameras already offer a depth map of the environment, allowing for a faster and less complex computation of the position in a 3D coordinate system. However, as can be observed in [Section 3.2.4](#sec3dot2dot4-sensors-20-02641){ref-type="sec"}, detection and pose computation for markers is very fast for 2D cameras as well, due to the geometric properties of the synthetic images. Therefore, using 3D cameras could represent an unnecessary excess of resources. Furthermore, RGB-D cameras usually have a lower resolution than RGB cameras, both for the color and depth maps. Thus, their use is rarely justified for marker based solutions. ### 3.2.6. Indoor Localization Solutions with 2D Cameras + Other Sensors, Markers with Known Positions, and Traditional Image Analysis {#sec3dot2dot6-sensors-20-02641} Synthetic identifiers represent a very powerful tool when estimating the subject's position and orientation in indoor scenarios. However, the use of other sensors, such as inertial sensors, WiFi, or beacons, could enrich the information, thus increasing the accuracy, or could help reduce the number of necessary synthetic landmarks. Nazemzadeh et al. \[[@B37-sensors-20-02641]\] proposed a localization solution for unicycle-like wheeled robots, using Zbar and OpenCV to detect QR codes that were placed on the floor. They applied an Extended H-Infinity Filter (EHF) to compute the odometry based on dead reckoning and on a gyroscope platform. Babu and Markose \[[@B36-sensors-20-02641]\] also invoked dead reckoning with accelerometer and gyroscope information, increasing the accuracy of their QR based localization solution. Gang and Pyun \[[@B84-sensors-20-02641]\] configured the indoor space, in an offline phase, by creating a fingerprint map with the RSSI of the beacon signals and the intensity of the geomagnetic field at each reference point. In the localization stage, they combined the information from the beacons and the inertial sensors with the coordinates extracted from QR codes, obtaining an accuracy of approximately 2 m, as can be observed in [Table 7](#sensors-20-02641-t007){ref-type="table"}. The use of other sensors besides cameras can add many benefits to a localization solution, especially if there is no need to acquire supplementary equipment. This is the case for WiFi access points, already installed in a building for other purposes. However, most of the WiFi localization solutions are based on the WiFi fingerprinting procedure, a manual and cumbersome configuration stage in which the signal strengths of the access points are recorded for known locations on the map of the building. Since smartphones have become very popular and their cameras have reached impressive capabilities, they can be successfully used as acquisition devices in computer vision based localization solutions. Another advantage of using a smartphone is represented by the built-in inertial sensors. Thus, an application that combines input from the camera and the inertial sensors of a smartphone does not require equipment that is not already owned by the users. ### 3.2.7. Indoor Localization Solutions with Real Image/Feature Databases, 2D Mobile Cameras, and Traditional Image Analysis {#sec3dot2dot7-sensors-20-02641} Using a database of real images or features from real images of the environment in localization solutions represents an alternative to decorating the indoor space with QR codes or other synthesized images. In a configuration stage, images or features, labeled with location and orientation information, are stored in a database. For instance, Hu et al. \[[@B85-sensors-20-02641]\] obtained a panoramic video of the scene, which was processed with traditional computer vision algorithms for computing omni-projection curves. Bai et al. \[[@B86-sensors-20-02641]\] constructed a landmark database by using a laser distance meter to measure the distance between the location of the camera and selected landmarks. In the localization stage, the images acquired with the mobile camera were compared with the ones from the database using feature matching algorithms such as SIFT, SURF, or ORB. The processing time in this stage is highly affected by the number of images/features in the database, which must be compared against the images from the mobile camera's video flow. The first line of [Table 8](#sensors-20-02641-t008){ref-type="table"} is a good example, as it shows that running the localization algorithm with a database of 1000 frames was eight times faster than with a database of 8000 frames. To reduce the processing time, Elloumi et al. \[[@B87-sensors-20-02641]\] limited the similarity search of two images to only a selection of areas within the images, thus reducing the number of features by 40%. These areas were considered to contain the most important characteristics and were selected based on a metric that combined orientation, color, intensity, flickering effects, and motion. Compared to solutions that use artificial markers, the solutions in this class do not require decorating the indoor space with visual markers, thus not affecting the aesthetics of the indoor space. Although they have a higher localization error (few meters), this error level can still be acceptable for certain applications. ### 3.2.8. Indoor Localization Solutions with Real Image/Feature Databases, 2D Mobile Cameras, and Artificial Intelligence {#sec3dot2dot8-sensors-20-02641} Artificial intelligence includes a plethora of localization algorithms for systems that use mobile cameras. For instance, Lu et al. \[[@B89-sensors-20-02641]\] proposed a multi-view regression model to determine the location and orientation of the user accurately. Xiao et al. \[[@B90-sensors-20-02641]\] determined the location of a smartphone, based on the detection of static objects within images acquired with the smartphone's cameras. Faster-RCNN was used for static object detection and identification. Another deep CNN, Convnet, was used in the localization system proposed by Akal et al. \[[@B91-sensors-20-02641]\]. This network uses compound images from four non-overlapping monocular images placed on a ground robot, achieving centimeter accuracy, but requiring a sizeable dataset of compound images for training. As can be observed in [Table 9](#sensors-20-02641-t009){ref-type="table"}, the machine learning based solutions achieved interactive computing times or even real-time performance and a localization accuracy of under one meter to tens of centimeters. These solutions seemed to have better accuracy performance compared to the solutions in the previous class, while benefiting from the same advantages of not requiring deploying visual markers in the indoor space. ### 3.2.9. Indoor Localization Solutions with Real Image/Feature Databases, 3D Mobile Cameras, and Artificial Intelligence {#sec3dot2dot9-sensors-20-02641} Another class of indoor localization methods uses RGB-D images acquired with mobile cameras that are processed with the help of CNN. Guo et al. \[[@B92-sensors-20-02641]\] used a CNN (PoseNet network) for exploiting the vision information and the long short-term memory network for incorporating the temporal information. Zhang et al. \[[@B63-sensors-20-02641]\] applied visual semantic information for performing indoor localization. A database with object information was constructed using Mask-RCNN, extracting the category and position for each object. Then, using the SURF descriptor, keypoints of the recognized objects were detected. Furthermore, CNN features were obtained using a pre-trained ResNet50 network. The visual localization was performed in two steps: the most similar key frames were obtained using the selected CNN features; the bundle adjustment method \[[@B137-sensors-20-02641]\] was used to estimate the matrix between the current image and candidate frames. Both methods were tested on public datasets. Localization results were within 0.3 m and 0.51 m (as shown in [Table 10](#sensors-20-02641-t010){ref-type="table"}). 3D cameras give access to a depth map of the environment, either through built-in algorithms, as in the case of structured light or time-of-flight devices, or through stereo matching algorithms that have multiple implementations, available to the public. However, these cameras come with various limitations. For instance, the estimation of the depth map with stereo cameras in the case of untextured surfaces (such as white walls) is very inaccurate. Furthermore, structured light and time-of-flight depth cameras cannot estimate the distance to reflective surfaces or in case of sunlit environments. Moreover, although 3D cameras have gained popularity, they are not as common as 2D cameras, and therefore, their applicability is reduced. While localization solutions with 2D mobile cameras can be easily deployed, using generally available smartphones, 3D cameras are more appropriate for specialized applications, in areas like assistive devices or autonomous robots. ### 3.2.10. Indoor Localization Solutions with Real Image/Feature Databases, 2D Cameras + Other Sensors, and Traditional Image Analysis {#sec3dot2dot10-sensors-20-02641} If WiFi signals, inertial sensors, beacons, or other sensors can increase the accuracy of marker based localization solutions or can help reduce the number of synthesized images that should be placed on the ceiling/floor/walls of the building (as discussed in [Section 3.2.6](#sec3dot2dot6-sensors-20-02641){ref-type="sec"}), a hybrid approach can be even more useful when dealing with natural features from the environment. Acquiring additional information from various sensors can help reduce the search space in the image matching stages. Yan et al. \[[@B94-sensors-20-02641]\] also used WiFi information to increase the accuracy and improve the processing time of a natural feature extraction algorithm, which combined Features from Accelerated Segment Test (FAST) with SURF. Marouane et al. \[[@B93-sensors-20-02641]\] used accelerometer data for step counting and gyroscope information for orientation and transformation of images into histograms for more efficient image matching. Rotation invariance was achieved by adding the perspective transformation of two planes. Another solution that used inertial sensors was the one proposed by Huang et al. \[[@B95-sensors-20-02641]\]. They applied the vanishing points method and indoor geometric reasoning, taking advantage of rules for 3D features, such as the ratio between width and height, the orientation, and the distribution on the 2D floor map. Arvai and Dobos \[[@B96-sensors-20-02641]\] applied the perspective-n-point algorithm to estimate the user's position inside the 2D floor-plan of a building, relative to a series of landmarks that were placed in the configuration stage. They used an extended Kalman filter to estimate the position by combining visual and inertial information. [Table 11](#sensors-20-02641-t011){ref-type="table"} presents the characteristics of indoor localization solutions that combine data from 2D cameras and other sensors, estimating the position and orientation of the subject with traditional image processing. Several such solutions achieved centimeter location accuracy, due to this fusion between images and information from inertial sensors, WiFi signals, RFID devices, or beacons. However, this fusion of data from several sensors brings a computational load. ### 3.2.11. Indoor Localization Solutions with Real Image/Feature Databases, 2D Cameras + Other Sensors, and Artificial Intelligence {#sec3dot2dot11-sensors-20-02641} The solutions based on the detection of objects or markers from RGB images offer a relative position and orientation estimation, but are unreliable when markers or objects are not visible. Furthermore, detection is influenced by camera exposure time. Thus, images combined with data from other sensors can increase the precision of the localization. Rituerto et al. \[[@B97-sensors-20-02641]\] estimated the user's location using values acquired from inertial sensors combined with computer vision methods applied on RGB images. The particle filtering method was used for combining all these data. A map with walls, corridors, and rooms and some important signs (such as exit signs and fiducial markers) was also considered. Neges et al. \[[@B98-sensors-20-02641]\] combined an IMU step based counter with video images for performing indoor localization. IMU data were used to estimate the position and orientation of the mobile device, and different semantic objects were extracted from the video (e.g., exit signs, fire extinguishers, etc.) for validation of the obtained position. The recognition of different markers was achieved using Metaio SDK \[[@B140-sensors-20-02641]\], a machine learning based development tool. In Sun et al. \[[@B99-sensors-20-02641]\], RSS samples, surveillance images, and room map information were used for performing indoor localization. People were detected using background subtraction from images acquired with a camera placed on the ceiling of the room. The foreground pixel that was the nearest to the location of the camera would approximate the person position in the image. Then, this position was mapped to a localization coordinate using a multi-layer neural network (with three layers). The iStart system \[[@B100-sensors-20-02641]\] combines WiFi fingerprints and RGB images for indoor localization. The system proposed by Zhao et al. \[[@B101-sensors-20-02641]\] was based on a combination of CNN with a dual-factor enhanced variational Bayes adaptive Kalman filter. Channel State Information (CSI) was extracted from an MIMO-OFDM PHY layer as a fingerprint image to express the spatial and temporal features of the WiFi signal. CSI features were learned with a CNN inspired by the AlexNet network obtaining the mapping relationship between the CSI and the 2D coordinates. Results were processed with the Bayes adaptive Kalman filter in order to achieve noise attenuation. These methods were evaluated on their own datasets with good results (position accuracy of approximately 1 m), as shown in [Table 12](#sensors-20-02641-t012){ref-type="table"}. Even though artificial intelligence and especially deep convolutional networks have become very popular, they still come with certain limitations. First, they require a large amount of training data, usually manually annotated. Second, the training stage is both time consuming and hardware demanding. Even though in the online stage, the already trained network requires less resources, adding the complexity of fusing the visual data with information from other sensors can have a negative impact on the runtime, as can be observed for several selected papers \[[@B97-sensors-20-02641],[@B100-sensors-20-02641]\]. ### 3.2.12. Indoor Localization Solutions with Real Image/Feature Databases, 3D Mobile Cameras + Other Sensors, and Traditional Image Analysis {#sec3dot2dot12-sensors-20-02641} Localization precision can be increased by matching of RGB-D images using traditional feature descriptors combined with information obtained from an IMU sensor. In Gao et al. \[[@B102-sensors-20-02641]\], key points were extracted from the RGB-D images using an improved SIFT descriptor. Then, the RANSAC algorithm \[[@B141-sensors-20-02641]\] eliminated mismatched points from the matching pairs. Their corresponding depth coordinates were obtained from the depth images. Using this information, the rotation matrix and translation vector were computed from two consecutive frames. Furthermore, IMU data were used to eliminate the noise, improving the stability and positioning accuracy. Adaptive fading extended Kalman filter fused the position information of Kinect and IMU outputs. Furthermore, this fusion eliminated the noise and improved the stability and accuracy of the system. A similar idea was proposed by Kim et al. \[[@B103-sensors-20-02641]\]. Their solution generated 3D feature points using the SURF descriptor, which were next rotated using IMU data to have the same rigid body rotation component between two consecutive images. The RANSAC algorithm \[[@B141-sensors-20-02641]\] was used for computing the rigid body transformation matrix. [Table 13](#sensors-20-02641-t013){ref-type="table"} shows the dataset characteristics and obtained accuracy for the localization methods based on RGB-D images processed with traditional image analysis algorithms and sensor fusion. Since robots can be equipped with many sensors, including 3D cameras and inertial units, the solutions in this class have been successfully applied to the autonomous robots domain. ### 3.2.13. Indoor Localization Solutions with a 3D Model of the Environment, 2D Mobile Cameras, Real Features, and Traditional Image Analysis {#sec3dot2dot13-sensors-20-02641} Simultaneous Localization and Mapping is a very popular algorithm in several domains, such as autonomous robots or Augmented Reality. During recent years, various solutions to the problem of localization and mapping have been proposed. For instance, Endo et al. \[[@B29-sensors-20-02641]\] used LSD-SLAM for map construction, localization, and detection of obstacles in real time. Teixeira et al. \[[@B104-sensors-20-02641]\] used the pattern recognition SURF method to locate natural markers and reinitialize Davison's Visual SLAM \[[@B142-sensors-20-02641]\]. Several SLAM based solution use 3D cameras in the configuration stage, to create a 3D reconstruction of the environment, and then change the acquisition device to a monocular camera in the localization stage. Sinha et al. \[[@B105-sensors-20-02641]\] applied RGBD-SLAM on images acquired with Microsoft Kinect to reconstruct 3D maps of indoor scenes. In the localization stage, they used monocular images acquired with a smartphone camera and estimated the transformation matrix between frames using RANSAC on the feature correspondences. They applied SIFT or SURF for feature extraction, in order to detect landmarks, which were cataloged as sets of distinguished features regularly observed in the mapping environment, being stationary, distinctive, repeatable, and robust against noise and lighting conditions. Deretey et al. \[[@B106-sensors-20-02641]\] also applied RGBD-SLAM in an offline, configuration stage, to create 3D point clouds that contained intensity information. 2D features were extracted with a matching algorithm (SIFT, SURF or ORB), and then, a projection matrix of matched features between 2D images and 3D points was computed. A comparison with RGBD-SLAM was offered by Zhao et al. \[[@B109-sensors-20-02641]\], which used Kinect to collect the 3D environment information in a configuration stage. They also built a 2D map of the indoor scene with Gmapping, an ROS package that used Rao--Blackwellized Particle Filters (RBPF) \[[@B143-sensors-20-02641]\] to learn grid maps. In the online phase, they applied Monte Carlo localization based on the previously created 2D map. Ruotsalainen et al. \[[@B107-sensors-20-02641]\] performed Visual SLAM for tactical situational awareness by applying a Kalman filter to combine a visual gyroscope and a visual odometer. The visual gyroscope estimated the position and orientation of the camera by detecting straight lines in three orthogonal directions. The visual odometer computed the transformation of the camera from the motion of image points matched using SIFT in adjacent images. A similar approach, which took into account the structural regularity of man-made building environments and detected structure lines along dominant directions, was the solution proposed by Zhou et al. \[[@B108-sensors-20-02641]\]. They also applied an extended Kalman filter to solve the SLAM problem. Ramesh et al. \[[@B110-sensors-20-02641]\] combined imaging geometry, visual odometry, object detection with aggregate channel features, and distance-depth estimation algorithms into a Visual SLAM based navigation system for the visually impaired. A different approach was the one proposed by Dong et al. \[[@B111-sensors-20-02641]\], which reused a previous traveler's (leader) trace experience to navigate future users or followers. They used ORB features for the mobile Visual SLAM. To combat environmental changes, they culled non-rigid contexts and kept only the static contents in use. SLAM based approaches can attain centimeter or even millimeter location accuracy, but at a high computational cost. They also require significant memory resources to store the 3D representation of the scene. [Table 14](#sensors-20-02641-t014){ref-type="table"} presents the characteristics of some solutions that create a 3D reconstruction of the environment in an offline stage, acquire images with a monocular camera in the localization stage, and perform low-level image processing to estimate the position and orientation of the user/robot. ### 3.2.14. Indoor Localization Solutions with a 3D Model of the Environment, 2D Mobile Cameras, Real Features, and Artificial Intelligence {#sec3dot2dot14-sensors-20-02641} Artificial intelligence based 2D localization methods can also be applied on 3D representations of the space. Han et al. \[[@B112-sensors-20-02641]\] removed obstacles detected with the Mask-RCNN network to enhance the performance of the localization. It detected persons as potential obstacles and split these obstacles from the background. Then, ORB-SLAM2 \[[@B148-sensors-20-02641]\] was used for localization. Xiao et al. \[[@B113-sensors-20-02641]\] proposed Dynamic-SLAM for solving SLAM in dynamic environments. It was based on ORB-SLAM. First, a CNN was used for static or dynamic object detection. Then, applying a missed detection compensation algorithm based on the speed invariance from adjacent frames, the detection recall rate was improved. Finally, tracking was performed using ORB features extracted from each keyframe image for performing feature based visual SLAM by processing feature points of dynamic objects. The pose estimation was obtained by solving the perspective-n-point problem with the bundle adjustment method. [Table 15](#sensors-20-02641-t015){ref-type="table"} presents the characteristics of some solutions belonging to the current class. The neural networks introduce a high computational load, but can help not only with the localization, but also with the scene understanding problem. ### 3.2.15. Indoor Localization Solutions with a 3D Model of the Environment, 3D Mobile Cameras, Real Features, and Traditional Image Analysis {#sec3dot2dot15-sensors-20-02641} Several localization solutions use 3D cameras in the configuration step, as well as in the actual localization stage. For instance, Du et al. \[[@B114-sensors-20-02641]\] created an interactive mapping system that partitioned the registration of RGB-D frames into local alignment, based on visual odometry, and global alignments, using loop closure information to produce globally consistent camera poses and maps. They combined RANSAC inlier count with visibility conflict in the three point matching algorithm to compute 6D transformations between pairs of frames. Paton and Kosecka \[[@B115-sensors-20-02641]\] applied feature extraction and mapping on RGB-D data with SIFT, motion estimation and outlier rejection with RANSAC, and estimation refinement to compute the position and orientation of a camera. Correspondences established between SIFT features could initialize a generalized Iterative Closest Point (ICP) algorithm. Salas-Moreno et al. \[[@B116-sensors-20-02641]\] proposed a GPGPUparallel 3D object detection algorithm and a pose refinement based on ICP. Their real-time incremental SLAM was designed to work even in large cluttered environments. Prior to SLAM, they created a database of 3D objects with KinectFusion. The scene was represented by a graph, where each node stored the pose of an object with a correspondent entry in the database. Their object level scene description offered a huge representation compression in comparison with the usual reconstruction of the environment into point clouds. A robust key-frame selection from RGB-D image streams, combined with pose tracking and global optimization based on the depth camera model, vertex-weighted pose estimation, and edge-weighted global optimization, was described by Tang et al. \[[@B118-sensors-20-02641]\]. Most solutions acquire images with structured light or time-of-flight cameras, but stereo cameras can also provide 3D information. For instance, Albrecht and Heide \[[@B117-sensors-20-02641]\] acquired images with a stereo camera and applied ORB-SLAM2 for poses of the keyframes, creating a 3D reconstruction of the environment with OpenCV's Semi-Global Block Matching (SGBM) algorithm. Then, they condensed the point cloud into a blueprint-like map of the reconstructed building, based on ground and wall segmentation. Martin et al. \[[@B119-sensors-20-02641]\] applied Monte Carlo based probabilistic self-localization on a map of colored 3D points, organized in an octree. They demonstrated that their algorithm recovered quickly from cases of unknown initial position or kidnappings (the robot was manually displaced from one place of the environment to another). [Table 16](#sensors-20-02641-t016){ref-type="table"} presents the computing capabilities and obtained accuracy for several SLAM based localization solutions that apply low-level image processing on data that contain both color and depth information. It can be observed that some of the researchers evaluated their algorithms only through visual inspection. Even so, inspection of the obtained 3D reconstruction and especially loop closure can demonstrate the performance in the case of SLAM based solutions. This class is reduced to a 3D to 3D matching problem, much less complex than the 3D to 2D matching problem described in [Section 3.2.13](#sec3dot2dot13-sensors-20-02641){ref-type="sec"} and [Section 3.2.14](#sec3dot2dot14-sensors-20-02641){ref-type="sec"}. However, the requirement to have a 3D camera both in the configuration stage and in the online phase greatly reduces the applicability of this kind of solution. ### 3.2.16. Indoor Localization Solutions with a 3D Model of the Environment, 3D Mobile Cameras, Real Features, and Artificial Intelligence {#sec3dot2dot16-sensors-20-02641} Another class of indoor localization solutions is the one that uses 3D cameras in the configuration stage, to create a reconstruction of the scene with SLAM or other algorithms, but also in the localization stage, applying high level computer vision techniques for computing the position and orientation of the user. Guclu et al. \[[@B121-sensors-20-02641]\] proposed an SLAM method applied on RGB-D images using a graph based approach. The keyframe autocorrelogram database estimated motion between frames. Keyframes were indexed based on their image autocorrelograms \[[@B150-sensors-20-02641]\], using a priority search k-means tree. Adaptive thresholding was used to increase the robustness of loop closure detection. Kuang et al. \[[@B120-sensors-20-02641]\] improved ORB-SLAM. A combination between quasi-physical sampling algorithm (based on BING features \[[@B151-sensors-20-02641]\], obtained by SVM training) with depth information was used to pre-process an image for decreasing the computing time of the ORB algorithm. Then, improved KD-trees were used to increase the matching speed of the ORB algorithm. Furthermore, using RGB-D images, a 3D dense point cloud map system was constructed, instead of a sparse map from ORB-SLAM. As can be observed in [Table 17](#sensors-20-02641-t017){ref-type="table"}, the use of 3D cameras can improve the accuracy of known localization methods such as ORB-SLAM or ORB-SLAM2. Still, the dimensionality of the data introduces a high computational cost. Furthermore, the lack of 3D training data could represent a limitation of this class. ### 3.2.17. Indoor Localization Solutions with a 3D Model of the Environment, 2D Mobile Cameras + Other Sensors, Real Features, and Traditional Image Analysis {#sec3dot2dot17-sensors-20-02641} A hybrid approach that fuses information from 2D cameras and other sensors can be applied on 3D models of the environment as well. For instance, Wang et al. \[[@B46-sensors-20-02641]\] used RFID readers for an approximate estimation of the location and calculation of 3D image coordinates with low-level image matching. Kao and Huy \[[@B41-sensors-20-02641]\] combined information from WiFi access points with the K-nearest neighbor method, inertial sensors (accelerometer and gyroscope), and a CMOS camera. They chose ORB features in their SLAM implementation to navigate Bluetooth connected wheeled robots in indoor environments. Yun et al. \[[@B122-sensors-20-02641]\] saved the WiFi access point information in a configuration stage and assembled the images acquired with an Xtion PRO LIVE depth camera, building a 3D indoor map of the indoor location. In the localization stage, they reduced the per-frame computation by splitting a video frame region into multiple sub-blocks and processing only a sub-block in a rotating sequence at each frame. They applied SIFT based keypoint detection and optical flow for tracking. Huang et al. \[[@B95-sensors-20-02641]\] applied an extended Kalman filter to fuse data from LSD-SLAM computed on RGB images, ZigBee localization, and IMU sensors (accelerometer, gyroscope, and magnetometer). Ullah et al. \[[@B125-sensors-20-02641]\] combined data from a monocular visual SLAM and an IMU with an unscented Kalman filter. Gerstweiler \[[@B45-sensors-20-02641]\] also fused IMU information with SLAM, using the HyMoTrack framework \[[@B153-sensors-20-02641]\], a hybrid tracking solution that uses multiple clusters of SLAM maps and image markers, anchored in the 3D model. Chan et al. \[[@B124-sensors-20-02641]\] computed a laser based SLAM and a RBPF based visual SLAM. Perspective trajectories obtained from the laser SLAM were mapped into images, and the essential matrix between two sets of trajectories was combined with the monocular camera based SLAM. Even if the fusion between sensor data and visual information introduces a high computational load, several solutions achieve real-time frame rates on commodity computers, as can be observed in [Table 18](#sensors-20-02641-t018){ref-type="table"}. 3.3. Discussion {#sec3dot3-sensors-20-02641} --------------- This section draws conclusions from our analysis of the proposed classes of vision based localization solutions, enabling readers to make better informed choices in terms of indoor positioning technologies to accommodate their specific requirements or particularities. While positioning technologies are numerous and do not limit themselves to image processing, vision based solutions have become popular due to the increasing affordability of cameras and their integration in pervasive devices such as smartphones. Localization methods that use static cameras can benefit from the camera surveillance infrastructure already available in most modern large office and public buildings. Furthermore, since most robotic platforms have RGB or RGB-D cameras, it makes it easier to port visual positioning solutions on the different platforms, enabling their use in assisted living scenarios. Other applications in the autonomous robots domain can take advantage of 2D/3D cameras already integrated in the robots. Smart glasses with cameras can enable a more seamless user experience for indoor localization applications; however, until they reach a wide market adoption, most applications that localize people (especially in the domains of assistive devices and augmented reality) use smartphones. Even though 2D cameras have larger applicability due to their ubiquity and the dimensionality of the acquired data, 3D cameras have several advantages. 3D cameras offer a depth map of the environment, either obtained from a disparity map computed with stereo matching algorithms or estimated with time-of-flight and structured light technologies. Stereo cameras require optimal lighting conditions and are affected by lens distortion, similar to 2D cameras. Furthermore, depth cannot be estimated in untextured environments through stereo matching. On the other hand, structured light and time-of-flight cameras work even in unlit environments and can estimate the depth regardless of texture properties. Although these cameras are affected by bright light and reflective surfaces, typical indoor environments contain untextured surfaces (especially uniformly painted walls) and are rarely characterized by bright sunlight. Therefore, we considered that among 3D cameras, structured light, and time-of-flight devices are the most suited for indoor applications. While cameras have many advantages, they are affected by lighting conditions, occlusion, and position changes of objects from the environment. In order to increase localization accuracy or to decrease the computational load of the computer-vision algorithms, visual data can be combined with data from other sensors. Other popular indoor localization solutions are those based on sensors such as WiFi, beacons, and RFID. WiFi based solutions use the received signal strength and the media access control address of access points to determine the position. WiFi based methods also enjoy the advantage of using existing infrastructure in buildings, as WiFi access points are even more widely available in buildings than camera surveillance systems. While beacon based positioning technologies can reach higher accuracy than WiFi based solutions, they require deploying additional hardware. The RFID technology poses even more limitations in terms of range. Although the positioning algorithms that use sensors such as WiFi, beacons, or RFID have a lower accuracy compared to vision based methods, they also have a lower complexity. Thus, possible localization solutions can benefit from a two-step positioning algorithm: firstly obtaining a quick, approximate location using beacons or WiFi, which tightens the search space of computer vision algorithms; secondly achieving an accurate and also quick location and orientation estimation of the tracked entity. Vision based indoor localization solutions can detect fiducial markers or features from real images of the environment. The use of artificial markers enables extremely fast detection and position estimation. Due to the geometric properties of the fiducial markers and their accurate localization with 2D cameras, the use of 3D cameras is unjustified. The biggest disadvantage of using markers is the requirement of covering the space with synthetic images, which can have a negative visual impact on the environment. Therefore, the applicability of such solutions is reduced. Features or semantic objects detected from real images of the environment do not visually influence the environment. However, setting up a database of features/images, annotated with position and orientation information, or creating a 3D model of the environment represent cumbersome processes. Furthermore, changes in the environment, such as rearranging furniture or paintings and posters, would require another configuration stage for rebuilding the feature/image database or the 3D model of the scene. Objects or features from images can be detected using traditional image processing or artificial intelligence methods. Traditional image processing methods perform detection by comparing different features that are extracted from the images, and the recognition success depends on the selected features. On the other hand, the artificial intelligence methods used for object recognition are mainly based on convolutional neural networks, thus not needing to select the features for recognizing objects, as convolutional neural networks learn specific objects directly from images. One disadvantage of these networks is the high number of images required for training the network. Training data can be obtained either by manual acquisition and annotation or from publicly available datasets. Public datasets are very helpful; however, they are only a few available (especially containing 3D data), and they are limited to several semantic classes. 4. Benchmarks {#sec4-sensors-20-02641} ============= The evaluation methods used for the indoor localization solutions presented in this paper differ. Some are based on visual inspection, some on testing the solutions in certain scenarios or testbeds, and some on using public datasets. This section presents benchmarks created for evaluating localization methods. They can differ based on the input information, which consists of monocular or RGB-D images and WiFi, along with other sensor readings. Some testbeds are designed with the purpose of evaluating only the location and orientation accuracy, while others can also evaluate the correctness of 3D reconstructions in SLAM based methods. Several research papers released to the public a series of datasets and evaluation tools \[[@B139-sensors-20-02641],[@B155-sensors-20-02641],[@B156-sensors-20-02641]\], while others proposed reference systems that can be used for testing the accuracy of localization solutions \[[@B157-sensors-20-02641]\], the latter enabling fair comparisons of existing localization systems in similar conditions \[[@B158-sensors-20-02641],[@B159-sensors-20-02641],[@B160-sensors-20-02641]\]. Sturm et al. \[[@B139-sensors-20-02641]\] proposed a benchmark for the evaluation of RGB-D SLAM based localization solutions. The TUM dataset and this benchmark represent a popular testing tool. This is noticeable in the tables in [Section 3](#sec3-sensors-20-02641){ref-type="sec"}. The database consists of images acquired with a Kinect sensor, containing both color and depth information, at a resolution of $640 \times 480$. They provide ground truth trajectories that are computed with a motion-capture system composed of eight cameras that acquire images at 100 Hz. These sequences cover a variety of cases, from short to long trajectories, with or without loop closure. Their benchmark offers automatic evaluation tools to assess the drift of visual odometry solutions, as well as the global pose error of SLAM based methods. Another popular benchmark, which contains the ICL-NUIM dataset, was provided by Handa et al. \[[@B138-sensors-20-02641]\]. The database consists of RGB-D frames within synthetically generated scenes with the point of view of handheld cameras. It contains ground truth camera poses and surface models, which enable not only the evaluation of localization solutions, but also of the surface reconstruction accuracy of SLAM based methods. Sun et al. \[[@B155-sensors-20-02641]\] proposed a dataset for evaluating computer vision based localization solutions that compute the pose of a 2D camera with respect to a 3D representation of the scene. Their database contained training data acquired with cameras and a LiDAR scanner, which measured the distance to a target by illuminating it with a laser light and computing the difference in return times for the reflected light. The LiDAR point clouds were used as a reference in a semi-automatic localization workflow that estimated the camera pose with six degrees of freedom. They compared this dataset with several image based localization datasets produced with the SfM algorithm \[[@B136-sensors-20-02641],[@B161-sensors-20-02641],[@B162-sensors-20-02641]\], claiming the creation of a point cloud with much higher density and precision. EgoCart \[[@B156-sensors-20-02641]\] is another benchmark dataset, comprising of almost 20,000 RGB-D images, annotated with information of the camera position and orientation. The authors made the dataset public, along with the evaluation (in terms of accuracy, computing time, and memory requirements) of various machine learning based localization solutions \[[@B163-sensors-20-02641],[@B164-sensors-20-02641],[@B165-sensors-20-02641],[@B166-sensors-20-02641]\]. Schmitt et al. \[[@B157-sensors-20-02641]\] presented an indoor localization system that relied on visual information provided by two Microsoft Kinect devices and on wheel-odometry data acquired with a Roomba robot. Within the ROS framework, they enhanced a pre-drawn floor plan with SLAM, achieving an average error of 6.7 cm for the position estimation. The authors claimed that the accuracy was sufficient to use the system as a reference, when testing the performance of other systems. Their robot could carry the components of the system under test and collect data, without interfering with the localization process. Ibragimov and Afanasyev \[[@B158-sensors-20-02641]\] analyzed the feasibility of using different visual SLAM based localization methods for robot systems in homogeneous indoor spaces. Their evaluation testbed was built with a monocular camera, a LiDAR sensor, a ZEDstereo camera, and a Kinect device. LIDAR based HECTORSLAM and a tape measure are considered ground truth for comparing trajectories obtained with ORB-SLAM \[[@B152-sensors-20-02641]\], Dense Piecewise Parallel Tracking and Mapping (DPPTAM) \[[@B167-sensors-20-02641]\], Stereolabs' ZedFu \[[@B168-sensors-20-02641]\] 3D mapping tool, and Real-Time Appearance Based Mapping (RTAB-MAP) \[[@B169-sensors-20-02641]\]. Filipenko and Afanasyev \[[@B159-sensors-20-02641]\] compared various SLAM based methods integrated in the ROS framework: GMapping \[[@B170-sensors-20-02641]\], Parallel Tracking and Mapping (PTAM) \[[@B171-sensors-20-02641]\], HectorSLAM \[[@B172-sensors-20-02641]\], Semi-direct Visual Odometry (SVO) \[[@B173-sensors-20-02641]\], LSD-SLAM \[[@B174-sensors-20-02641]\], RTAB-MAP \[[@B169-sensors-20-02641]\], ORB-SLAM \[[@B152-sensors-20-02641]\], DPPTAM \[[@B167-sensors-20-02641]\], Direct Sparse Odometry (DSO) \[[@B175-sensors-20-02641]\], Cartographer \[[@B176-sensors-20-02641]\], and Stereo Parallel Tracking and Mapping (S-PTAM) \[[@B177-sensors-20-02641]\]. They built a robot equipped with a 2D LiDAR, a monocular camera, and a ZED stereo camera. ATE was the chosen means of evaluation, represented through statistical metrics such as RMSE or the standard deviation. Ragot et al. \[[@B160-sensors-20-02641]\] performed an evaluation of two Visual SLAM algorithms, ORB-SLAM2 \[[@B148-sensors-20-02641]\] and RTAB-MAP \[[@B169-sensors-20-02641],[@B178-sensors-20-02641],[@B179-sensors-20-02641]\]. During the comparison of the two solutions, they used a VICON motion capture system as the ground truth. They performed various experiments, with straight-line, straight-line and back, circular paths, or trajectories containing loop closure. 5. Conclusions {#sec5-sensors-20-02641} ============== Indoor localization has an increasingly vast applicability in domains such as AR, navigation systems, assistive devices (especially for the visually impaired), autonomous robots, surveillance, and monitoring. Since surveillance cameras and smartphone cameras represent a commodity currently, researchers have developed a plethora of indoor localization solutions based on visual input. This paper offers an overview of the computer vision based indoor localization domain, discussing applications areas, commercial solutions, and benchmarks and presenting some of the most relevant contributions in the area. It also provides a survey of selected positioning solutions, proposing a new classification that organizes the solutions according to the use of known environment data, the sensing devices, the type of detected elements (artificial markers or real features), and the employed localization methods. The research papers selected in the 17 classes of the proposed taxonomy were chosen from prestigious research databases based on their relevance to the domain and publication date, and their purpose was to be illustrative for the reader in terms of the indoor positioning technologies. Since many relevant papers were too recent to have a considerable number of citations, we decided to not use this criterion for selection. The focus was on providing short descriptions of the solutions, highlighting the advantages and disadvantages and presenting the achieved performances (in terms of running time and location estimation accuracy), along with the properties of the datasets used for testing. [Table 2](#sensors-20-02641-t002){ref-type="table"}, [Table 3](#sensors-20-02641-t003){ref-type="table"}, [Table 4](#sensors-20-02641-t004){ref-type="table"}, [Table 5](#sensors-20-02641-t005){ref-type="table"}, [Table 6](#sensors-20-02641-t006){ref-type="table"}, [Table 7](#sensors-20-02641-t007){ref-type="table"}, [Table 8](#sensors-20-02641-t008){ref-type="table"}, [Table 9](#sensors-20-02641-t009){ref-type="table"}, [Table 10](#sensors-20-02641-t010){ref-type="table"}, [Table 11](#sensors-20-02641-t011){ref-type="table"}, [Table 12](#sensors-20-02641-t012){ref-type="table"}, [Table 13](#sensors-20-02641-t013){ref-type="table"}, [Table 14](#sensors-20-02641-t014){ref-type="table"}, [Table 15](#sensors-20-02641-t015){ref-type="table"}, [Table 16](#sensors-20-02641-t016){ref-type="table"}, [Table 17](#sensors-20-02641-t017){ref-type="table"} and [Table 18](#sensors-20-02641-t018){ref-type="table"} show that many papers did not report computing times and the specifics of their testbeds, and few papers discussed the financial implications of implementing such solutions in real life. The evaluation methodologies for the presented solutions differed. While some used visual observations to test their solutions, others chose to use private datasets. Although public datasets and benchmarks exist, they come with limitations in terms of required environment data and input from cameras, limiting their use to only certain localization solutions. We considered this paper a good guide to the field of computer vision based indoor localization. Our proposed classification and the selection of localization solutions for each category aimed to allow the reader to easily grasp the advantages and applicability of each class of solutions. Conceptualization, A.M. (Alin Moldoveanu), A.M. (Anca Morar) and I.M.; writing---original draft preparation, A.M. (Anca Morar), I.M. and I.E.R.; writing---review and editing, all authors; funding acquisition, A.M. (Alin Moldoveanu), A.M. (Anca Morar) and A.B. All authors have read and agreed to the published version of the manuscript. This work was supported by the grant "Ecosistem de cercetare, inovare si dezvoltare de produse si servicii TIC pentru o societate conectata la Internet of Things--NETIO" (MySMIS Code = 105976), subsidiary contract "Platformă imersivă de realitate virtuală și augmentată pentru spații culturale--CultReal", Number 4192/18.03.2019. The authors declare no conflict of interest. {#sensors-20-02641-f001} {#sensors-20-02641-f002} {#sensors-20-02641-f003} sensors-20-02641-t001_Table 1 ###### Classification of computer vision based localization research papers considering the environment data, the sensing devices, the detected elements, and the localization algorithm. Classification Research Papers ------------------------ --------------------- ------------ ---------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Marker/camera position 2D static cameras Artificial Image analysis \[[@B31-sensors-20-02641],[@B32-sensors-20-02641]\] Marker/camera position 2D static cameras Real Image analysis \[[@B60-sensors-20-02641],[@B67-sensors-20-02641],[@B68-sensors-20-02641],[@B69-sensors-20-02641]\] Marker/camera position 2D static cameras Real AI \[[@B70-sensors-20-02641],[@B71-sensors-20-02641],[@B72-sensors-20-02641],[@B73-sensors-20-02641],[@B74-sensors-20-02641]\] Marker/camera position 2D mobile cameras Artificial Image analysis \[[@B38-sensors-20-02641],[@B75-sensors-20-02641],[@B76-sensors-20-02641],[@B77-sensors-20-02641],[@B78-sensors-20-02641],[@B79-sensors-20-02641],[@B80-sensors-20-02641],[@B81-sensors-20-02641]\] Marker/camera position 3D mobile cameras Artificial Image analysis \[[@B82-sensors-20-02641],[@B83-sensors-20-02641]\] Marker/camera position 2D cameras, sensors Artificial Image analysis \[[@B36-sensors-20-02641],[@B37-sensors-20-02641],[@B84-sensors-20-02641]\] Image/feature database 2D mobile cameras Real Image analysis \[[@B85-sensors-20-02641],[@B86-sensors-20-02641],[@B87-sensors-20-02641],[@B88-sensors-20-02641]\] Image/feature database 2D mobile cameras Real AI \[[@B89-sensors-20-02641],[@B90-sensors-20-02641],[@B91-sensors-20-02641]\] Image/feature database 3D mobile cameras Real AI \[[@B63-sensors-20-02641],[@B92-sensors-20-02641]\] Image/feature database 2D cameras, sensors Real Image analysis \[[@B93-sensors-20-02641],[@B94-sensors-20-02641],[@B95-sensors-20-02641],[@B96-sensors-20-02641]\] Image/feature database 2D cameras, sensors Real AI \[[@B97-sensors-20-02641],[@B98-sensors-20-02641],[@B99-sensors-20-02641],[@B100-sensors-20-02641],[@B101-sensors-20-02641]\] Image/feature database 3D cameras, sensors Real Image analysis \[[@B102-sensors-20-02641],[@B103-sensors-20-02641]\] 3D model 2D mobile cameras Real Image analysis \[[@B29-sensors-20-02641],[@B104-sensors-20-02641],[@B105-sensors-20-02641],[@B106-sensors-20-02641],[@B107-sensors-20-02641],[@B108-sensors-20-02641],[@B109-sensors-20-02641],[@B110-sensors-20-02641],[@B111-sensors-20-02641]\] 3D model 2D mobile cameras Real AI \[[@B112-sensors-20-02641],[@B113-sensors-20-02641]\] 3D model 3D mobile cameras Real Image analysis \[[@B114-sensors-20-02641],[@B115-sensors-20-02641],[@B116-sensors-20-02641],[@B117-sensors-20-02641],[@B118-sensors-20-02641],[@B119-sensors-20-02641]\] 3D model 3D mobile cameras Real AI \[[@B120-sensors-20-02641],[@B121-sensors-20-02641]\] 3D model 2D cameras, sensors Real Image analysis \[[@B41-sensors-20-02641],[@B45-sensors-20-02641],[@B46-sensors-20-02641],[@B122-sensors-20-02641],[@B123-sensors-20-02641],[@B124-sensors-20-02641],[@B125-sensors-20-02641]\] sensors-20-02641-t002_Table 2 ###### Characteristics of indoor localization solutions with 2D static cameras (with known positions), markers, and traditional image analysis. Research Paper Dataset Characteristics Computing Time and Platform Accuracy ----------------------------- -------------------------------------------------------- ---------------------------------- -------------------------------------------------- \[[@B31-sensors-20-02641]\] own dataset: 1 static camera, covering 1.26 m × 1.67 m avg. 0.2 s per frame on a server err. between 0.0002 and 0.01 m (max. err.: 1 cm) \[[@B32-sensors-20-02641]\] own dataset: 3 rooms, each with 1 IP camera real-time observational sensors-20-02641-t003_Table 3 ###### Characteristics of indoor localization solutions with 2D static cameras (with known positions), real features, and traditional image analysis. Research Paper Dataset Characteristics Computing Time and Platform Accuracy ----------------------------- ----------------------------------------------------------------------------------------------------- ---------------------------------------------------------------- ------------------------------------------------------------------------------------------------- \[[@B67-sensors-20-02641]\] public datasets: PETS2009 \[[@B126-sensors-20-02641]\], TUD-Stadtmitte \[[@B127-sensors-20-02641]\] approximately 140 ms on Intel Core2Quad 2.66 GHz with 8 GB RAM Multiple Object Tracking Accuracy (MOTA): 87.8% on the PETS2009 and 64.2% on the TUD-Stadtmitte \[[@B69-sensors-20-02641]\] own dataset: indoor space with 2.2 m × 6 m, images with $320 \times 240$ pixels from 2 cameras \- less than 7.1 cm \[[@B68-sensors-20-02641]\] own dataset: 12,690 frames acquired with 3 cameras; public dataset: PETS2001 \- 95.7% hit rate and 96.5% precision \[[@B60-sensors-20-02641]\] own dataset: office with 5.1 m × 8.5 m × 2.7 m \- mean err. of 0.37 m sensors-20-02641-t004_Table 4 ###### Characteristics of indoor localization solutions with 2D static cameras (with known positions), real features, and artificial intelligence. Research Paper Dataset Characteristics Computing Time and Platform Accuracy ----------------------------- ------------------------------------------------------------------------------------------------------------ --------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------- \[[@B70-sensors-20-02641]\] public datasets: PETS 2009 (City Center), EPFLterrace dataset \- Ground Position Error (GPE) metric with total err. rate 0.122/0.131, Projected Position Error (PPE) metric with total err. rate 0.107/0.140 \[[@B71-sensors-20-02641]\] training: 1542 images (own) + 25,608 images from MS COCO; evaluation: 1400 images/robot type + 110/pattern 50 Hz on a GPU and 10 Hz on a CPU detection rate between 70% and 97.9%; orientation err. between 1.6 and 11.9 degrees \[[@B72-sensors-20-02641]\] own dataset: 47 employees, 18 rooms and 6 cubicles, 960 ceiling images 2.8 s per image (offline computation) 88.2% accuracy for identifying locations \[[@B73-sensors-20-02641]\] own dataset: over 2100 frames in 42 scenarios 6.25 fps on Jetson TX2 approximately 45 cm mean err. \[[@B74-sensors-20-02641]\] own dataset: office room with 1 camera and supermarket with 6 cameras 5 fps on a server detection success rate of 90% and avg. localization err. of 14.32 cm sensors-20-02641-t005_Table 5 ###### Characteristics of indoor localization solutions with 2D mobile cameras, markers with known positions, and traditional image analysis. Research Paper Dataset Characteristics Computing Time and Platform Accuracy ----------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ --------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------- \[[@B75-sensors-20-02641]\] own dataset: classroom with area 2.4 m × 1.8 m and 4 QR codes Nexus 4 Google (fps not mentioned) localization err. 6--8 cm, heading direction err. 1.2 angles \[[@B76-sensors-20-02641]\] own dataset, simplified and complex scenarios 47 ms for QR code extraction on a Raspberry Pi 2 complex scenario: planar position err. 17.5 cm, 3D pose estimation self-localization err. 10.4 cm \[[@B77-sensors-20-02641]\] public dataset proposed by Mikolajczyk and Schmid \[[@B134-sensors-20-02641]\], 4 image pairs 0.11 s, 0.16 s, 0.27 s, 0.14 s; 1--2 iterations to reach the threshold similarity threshold similarity 0.8 \[[@B38-sensors-20-02641]\] own dataset: hall with 6 QR codes and 2 possible trajectories (circular and 8-shape) 10 Hz on Linux Ubuntu 12.04 OS running ROS framework err. for circular path 0.2 m, err. for 8-shape path 0.14 m, orientation err. 0.267 radians \[[@B78-sensors-20-02641]\] own dataset acquired with an RGB camera fixed on an FLIR Pan Tilt Unit mounted on a mobile platform with an SICK s300 laser scanner (ground truth), markers placed on a wall 0.07 s avg processing time of a scene with 550 markers (up to 200 times faster than AprilTags) avg. error of angle estimates: 0.02 rad.for pitch/roll \[[@B79-sensors-20-02641]\] own dataset: space covered with 4 × 4 QR codes, placed 50 cm apart \- the robot can travel more than 7 times on the same route \[[@B80-sensors-20-02641]\] own dataset, images of resolution 1280 × 720 18.1 ms on an image with a cluttered scene and a single marker, using a single core 3.70 GHz Intel Xeon less than 0.6 degrees std. dev. for rotation, less than 0.4 cm std. dev. for translation \[[@B81-sensors-20-02641]\] own dataset in an academic building, four different paths, markers on the ceiling, guidance test with 10 blindfolded users \- 0 miss detections, 2 false detections out of 40 tests sensors-20-02641-t006_Table 6 ###### Characteristics of indoor localization solutions with 3D mobile cameras, markers with known positions, and traditional image analysis. Research Paper Dataset Characteristics Computing Time and Platform Accuracy ----------------------------- ------------------------- ----------------------------- ------------------------------------------------------------------------------------------- \[[@B82-sensors-20-02641]\] own dataset \- distance from the camera to the QR code: 1 cm err. \[[@B83-sensors-20-02641]\] own dataset real time maximum distance and angles from which the robot can see the QR code are: 270 cm and 51∘. sensors-20-02641-t007_Table 7 ###### Characteristics of indoor localization solutions with 2D cameras + other sensors, markers with known positions, and traditional image analysis. Research Paper Dataset Characteristics Computing Time and Platform Accuracy ----------------------------- ---------------------------------------------------------------------- ---------------------------------------------------------------------------- -------------------------------------------------------------------- \[[@B37-sensors-20-02641]\] own dataset less than 4 ms per frame; convergence time 18 s less than 0.2 m for position and less than 0.1 orientation for EHF \[[@B36-sensors-20-02641]\] own dataset computational load increases if dead reckoning is invoked with IMU sensors visual (performance affected if dead reckoning is not used) \[[@B84-sensors-20-02641]\] own dataset: corridor with 100 m × 2.25 m and hall with 14 m × 6.5 m \- accuracy is within 2 m 80% of the time sensors-20-02641-t008_Table 8 ###### Characteristics of indoor localization solutions with real image/feature databases, 2D mobile cameras, and traditional image analysis. Research Paper Dataset Characteristics Computing Time and Platform Accuracy ----------------------------- -------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------- \[[@B85-sensors-20-02641]\] own datasets: 862 frames, 3674 frames single CPU/Kepler K 20 chip: 20 ms/1.13 ms for a database of 1000 frames, 160 ms/8.82 ms for 8000 frames within 2 m in most cases \[[@B86-sensors-20-02641]\] own datasets: hallways 15 m long \- estimated moving speed compared to ground truth: max absolute err. 0.0643 m/s, RMSE for speed 0.24--0.37 m/s, RMSE for distance 0.16--0.23 m \[[@B87-sensors-20-02641]\] own dataset: 1866 images, 40 key frames \- \- \[[@B88-sensors-20-02641]\] own dataset: over 90,000 annotated frames out of 60 videos from six corridors (approximately 3.5 km of data) \- 4 m avg. absolute err. for HOG3D and 1.3 m for SF GABOR, over a 50 m traveling distance sensors-20-02641-t009_Table 9 ###### Characteristics of indoor localization solutions with real image/feature databases, 2D mobile cameras, and artificial intelligence. Research Paper Dataset Characteristics Computing Time and Platform Accuracy ----------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------- \[[@B89-sensors-20-02641]\] own dataset: 1800 images from 30 different locations, 480 indoor videos of buildings (each lasting around 2--3 s), public dataset: Dubrovnik \[[@B136-sensors-20-02641]\] 0.00092 s for image based localization and 0.0012 s for the video based method 95.56%/94.44% accuracy for location/orientation with image based localization and 98% with the video based method \[[@B90-sensors-20-02641]\] own dataset: 302 training images, resolution 3024 × 4032 object detection phase takes 0.3 s location accuracy is within 1 m \[[@B91-sensors-20-02641]\] own dataset: 112,919 compound images (composed of 4 images taken by 4 Google Nexus phones) of resolution 224 × 224 close to real-time avg. median err. after a 20 step moving for compound images is 12.1 cm sensors-20-02641-t010_Table 10 ###### Characteristics of indoor localization solutions with real image/feature databases, 3D mobile cameras, and artificial intelligence. Research Paper Dataset Characteristics Computing Time and Platform Accuracy ----------------------------- ------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------ ----------------------------------------------------------------------------------------------------------- \[[@B92-sensors-20-02641]\] ICL-NUIMdataset \[[@B138-sensors-20-02641]\] and TUM dataset \[[@B139-sensors-20-02641]\] 296 ms to find the most similar frame and 277 ms to estimate the final pose on Intel Xeon E5-1650 v3 CPU 3.5 GHz, NVidia TITAN GPU more than 80% of the images are localized within 2.5 degrees and more than 90% are localized within 0.3 m \[[@B63-sensors-20-02641]\] ICL-NUIM dataset \[[@B138-sensors-20-02641]\] \- 0.51 m living room, 0.41 m office sensors-20-02641-t011_Table 11 ###### Characteristics of indoor localization solutions with real image/feature databases, 2D cameras + other sensors, and traditional image analysis. Research Paper Dataset Characteristics Computing Time and Platform Accuracy ----------------------------- -------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------ -------------------------------------------------------------------------------------------------------------- \[[@B93-sensors-20-02641]\] own dataset: 75 location images from 1.5--2 m distance query time 40--230 ms mean distance err. rate is 2.5/2.21 m for extended distance estimation method/hybrid approach \[[@B94-sensors-20-02641]\] own dataset 90 ms FAST-SURF, 100--130 ms indoor positioning, 3--7 ms character detection, 30--45 ms tracking and registration on Honor 3C smartphone \- \[[@B95-sensors-20-02641]\] own dataset (offices and hallways) 0.5 s per frame 90% of location and orientation errors are within 25 cm and 2 degrees \[[@B96-sensors-20-02641]\] \- iPhone5s, iPhone X, LG Nexus 5X, Samsung Galaxy S7, S9, Huawei Mate tablet best results, on Samsung S9: 4.5 deg. avg. rotation and 250 mm position err. from 1 m in front of the marker sensors-20-02641-t012_Table 12 ###### Characteristics of indoor localization solutions with real image/feature databases, 2D cameras + other sensors, and artificial intelligence. Research Paper Dataset Characteristics Computing Time and Platform Accuracy ------------------------------ ---------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------ ----------------------------------------------------------------------------------------------------------------------------------------- \[[@B97-sensors-20-02641]\] own dataset, 3 blind volunteers 2 fps on a laptop visual inspection \[[@B98-sensors-20-02641]\] own dataset, 5 people with different weight and height, walking at 3 different speeds, on two tracks (straight or zig-zag) real time 93% accuracy in case of normal speed \[[@B99-sensors-20-02641]\] own dataset: office floor 51 m × 20 m × 2.7 m and 7 WiFi routers \- panoramic camera based method: mean err. for localization 0.84 m, cumulative probability within localization err. of 1 m/2 m is 70%/86% \[[@B100-sensors-20-02641]\] own dataset: room-level environment and open large environment 4 s per image (0.8 s fingerprint location on server, 2.9 s image location on smartphone, 1 s data transmission) less than 0.6 m avg. location err. and less than 6 degrees avg. direction err., 90% location deviations are less than 1 m \[[@B101-sensors-20-02641]\] own dataset: 50 m${}^{2}$ office and 2 cases (with and without line-of-sight); ground divided into 42 reference points real time on Intel5300 NIC laptop with 3 antennas as signal receiver and Ubuntu server with Intel Xeon e5-2609 CPU, GeForce GTX TITAN X GPU and 256 GB RAM avg. position err. is 0.98/1.46 m for line-of-sight/none line-of-sight sensors-20-02641-t013_Table 13 ###### Characteristics of indoor localization solutions with real image/feature databases, 3D mobile cameras + other sensors, and traditional image analysis. Research Paper Dataset Characteristics Computing Time and Platform Accuracy ------------------------------ ------------------------------------------------------------------------ ----------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- \[[@B102-sensors-20-02641]\] own dataset: small number of experiments and short tested trajectories real time avg. err. in the X-axis direction is 0.06 m with IMU \[[@B103-sensors-20-02641]\] own dataset real time translation error: 0.1043 m, rotation err.: 6.6571 degrees for static environments; translation err.: 0.0431 m ± 0.0080 m, rotation err.: 2.3239 degrees ± 0.4241 degrees for dynamic environments sensors-20-02641-t014_Table 14 ###### Characteristics of indoor localization solutions with an existing/generated 3D model of the environment, 2D mobile cameras, real features, and traditional image analysis. Research Paper Dataset Characteristics Computing Time and Platform Accuracy ------------------------------ ----------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------- \[[@B29-sensors-20-02641]\] own dataset: simple experiment with obstacles along a route real time on a single CPU visual inspection \[[@B104-sensors-20-02641]\] own dataset: images of resolution $320 \times 240$ 3 fps for SURF on 2.20 GHz dual-core computer 90% detection success rate and 14.32 cm avg. localization err. \[[@B105-sensors-20-02641]\] 3 own datasets: images of resolution $640 \times 480$ (from Galaxy S4 camera); public dataset: feature set of Liang et al. \[[@B144-sensors-20-02641]\] avg. search time for Dataset 1 (176 frames) is 10 ms and for Dataset 4 (285 frames) is 28 ms 80--100% accuracy, depending on dataset; 0.173--0.232 m localization error \[[@B106-sensors-20-02641]\] own dataset: reconstruction with RGBD-SLAM, Dataset 1 (50 frames, 139 mp), dataset 2 (33 frames, 37 mp) avg. localization time is 0.72 s per frame on an Intel Core i7 with 8 GB RAM avg. localization error is less than 10 mm: translation err. for Dataset 1 is 0.9--35 mm and for Dataset 2 is 0.3--17 mm \[[@B107-sensors-20-02641]\] own dataset: office environment, with 154 m long route images captured at 0.8 Hz using a smartphone camera, computing time not mentioned 1.5 degrees mean accuracy for visual gyroscope, 0.3 m/s mean accuracy for visual odometer, 1.8 m localization err. \[[@B108-sensors-20-02641]\] own dataset: synthetic scenes (20 m × 20 m scene with 88 lines and 160 points, 794 generated images); public dataset: Biccoca_2009 \[[@B145-sensors-20-02641]\] 0.5--1 s for MATLAB version, 25.8 ms avg. running time for C++ version 0.79 accuracy err. in position on a 967 m path; 0.2 m accuracy err. for synthetic scenes \[[@B109-sensors-20-02641]\] own dataset: Guangdong Key Laboratory, Shantou University \- only visual inspection, in comparison with the RGB-D SLAM method \[[@B146-sensors-20-02641]\] \[[@B110-sensors-20-02641]\] own dataset: indoor environment; public dataset: Karlsruhe outdoor datasets \[[@B147-sensors-20-02641]\] real time on an i7 processor 94--98% distance and depth measurements accuracy; absolute err. of 5.72/9.63 m for 2 outdoor datasets and 4.07/1.35 cm for 2 indoor datasets \[[@B111-sensors-20-02641]\] own datasets: office building (400 m${}^{2}$), gymnasium (1000 m${}^{2}$), and a shopping mall (6000 m${}^{2}$), 21 navigation paths, 274 checkpoints approximately 0.1 s for relocalization and navigation on Huawei P10, Nexus 6, Nexus 7, Lenovo Phab2 pro 98.6% immediate NSR, 93.1% NSR after 1 week, 83.4% NSR after 2 weeks sensors-20-02641-t015_Table 15 ###### Characteristics of indoor localization solutions with an existing/generated 3D model of the environment, 2D mobile cameras, real features, and artificial intelligence. Research Paper Dataset Characteristics Computing Time and Platform Accuracy ------------------------------ -------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------- \[[@B112-sensors-20-02641]\] public dataset: TUM Dynamic Object dataset (RGB images, depth information, ground truth trajectory) 5fps for Mask-RCNN on NVidia Tesla M40 GPU \[[@B149-sensors-20-02641]\] RMSE between 0.006134 and 0.036156 \[[@B113-sensors-20-02641]\] own dataset: 370 m from route; public datasets: TUM dynamic dataset, KITTI dataset (outdoor large scenarios) real time the trajectory RMSE err. is 2.29 m, the accuracy is 7.48--62.33% higher than ORB-SLAM2 \[[@B148-sensors-20-02641]\] sensors-20-02641-t016_Table 16 ###### Characteristics of indoor localization solutions with an existing/generated 3D model of the environment, 3D mobile cameras, real features, and traditional image analysis. Research Paper Dataset Characteristics Computing Time and Platform Accuracy ------------------------------ ------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------- \[[@B114-sensors-20-02641]\] own dataset: a room, check if the virtual objects have the same size as the real ones 3--4 fps for map building on a laptop with i7-720qm CPU difference in dimensions between 3D reconstructed and real objects: dm/cm level accuracy \[[@B115-sensors-20-02641]\] own dataset: barren office hallway; public dataset: TUM dataset \[[@B139-sensors-20-02641]\] \- 0.1--1.5 m translation RPE, 2--18 degrees rotational RPE, 0.02--1.1 m ATE \[[@B116-sensors-20-02641]\] own dataset: room of size 15 × 10 × 3 m${}^{3}$ 20 fps on a gaming laptop visual inspection, checking loop closure \[[@B117-sensors-20-02641]\] own dataset: a path of 70 m through a building 25 Hz visual inspection \[[@B118-sensors-20-02641]\] own datasets: taken with a handheld structure sensor; public datasets: Freiburg Benchmark, TUM dataset \[[@B139-sensors-20-02641]\] 5 fps 0.011--0.062 RMSE of ATE for public datasets; 1.4--4.1 cm closing distance and 1.08--3.32 degrees closure angle for own datasets \[[@B119-sensors-20-02641]\] own dataset collected with 2 wheeled robots (RB-1 and Kobuki) with Asus Xtion RGB-D sensors less than 30 s tracking mode (robot starts from a known position): x-mean: 0.082 m, y-mean 0.078 m; global mode (robot starts from unknown position): x-mean 0.27 m, y-mean 0.43 m sensors-20-02641-t017_Table 17 ###### Characteristics of indoor localization solutions with an existing/generated 3D model of the environment, 3D mobile cameras, real features, and artificial intelligence. Research Paper Dataset Characteristics Computing Time and Platform Accuracy ------------------------------ ---------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------------------------------- \[[@B120-sensors-20-02641]\] public dataset: TUM dataset \[[@B139-sensors-20-02641]\] 37.753 ms per frame on Intel i5 2.0 GHz CPU with 3 GB RAM 0.015 and 0.103 RMSE for the error size of the posture, better than ORB-SLAM \[[@B152-sensors-20-02641]\] \[[@B121-sensors-20-02641]\] public dataset: ICL-NUIM \[[@B138-sensors-20-02641]\] and TUM dataset \[[@B139-sensors-20-02641]\] 119.0 ms (average) on a desktop PC running Ubuntu 12.04 with an Intel Core i7-2600 CPU at 3.40 GHz and 8 GBRAM Absolute Trajectory Error ATE: 1 cm--5 cm, mostly competing with ORB-SLAM2 \[[@B152-sensors-20-02641]\] sensors-20-02641-t018_Table 18 ###### Characteristics of indoor localization solutions with an existing/generated 3D model of the environment, 2D mobile cameras + other sensors, real features, and traditional image analysis. Research Paper Dataset Characteristics Computing Time and Platform Accuracy ------------------------------ ----------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------ \[[@B46-sensors-20-02641]\] own dataset \- visual evaluation \[[@B41-sensors-20-02641]\] own dataset: robot moves along a specific path in a lab room, QVGA resolution 200 ms for basic image processing on LG P970, whole pipeline processed offline (manual extraction of features) position err. converges from 35--50 cm to less than 3 m \[[@B122-sensors-20-02641]\] own dataset: 120 m indoor hallway with 5200 video frames of size $640 \times 480$ from 3.2 fps to 23.3 fps on commodity laptop (2.6 GHz quad-core CPU and 4 GB RAM) 0.17 m position err. \[[@B95-sensors-20-02641]\] own dataset \- visual inspection \[[@B124-sensors-20-02641]\] own dataset map building and fusion process: real time on Intel Core i7-8550U CPU relative error of ORB-SLAM2 \[[@B148-sensors-20-02641]\] calibrated with proposed mapping matrix is less than 5% \[[@B125-sensors-20-02641]\] own dataset and public dataset: EuRoCdataset \[[@B154-sensors-20-02641]\] real time on an embedded board (1.92 GHz processor and 2 GB DDR3L RAM) 0.01--0.15 m position err. for own dataset; 0.234 m max. err. for EuRoC dataset \[[@B45-sensors-20-02641]\] own dataset: Vienna airport, path of 200 m 23 s for the proposed method to complete a guiding task visual inspection

Background ========== The intervertebral disc (IVD) lies between the vertebral bodies and links them together. The components of the disc are Nucleus pulposus (NP), Annulus fibrosus (AF) and the end-plates. Although the phenotype of IVD cells and the composition of the extracellular matrix (ECM) is still the subject of considerable debate, they appear quite comparable to those of articular cartilage particularly for NP cells \[[@B1],[@B2]\]. This composite nature of IVD endows the disc with both the tension-resisting properties of a ligament and the compression-resisting properties of articular cartilage. Unfortunately, disc structure and function does not remain optimal throughout life, but undergoes a progressive age-dependent degeneration. The IVD aging initiates early in the NP, as seen by a loss of cellularity and alteration of the ECM, thus compromising the mechanical properties of IVD \[[@B3]\]. It is well acknowledged that IVD degeneration encompasses several age-related processes influenced primarily by mechanical, nutritional and genetic factors. However, the underlying cellular and molecular mechanisms involved in the premature initiation and progression of IVD aging and degeneration still remains poorly deciphered (for review see \[[@B4]\]). In this context, the development of appropriate animal models capable of providing new insights into the IVD physiopathology should be further investigated. IVD blood supply terminates at the end-plate, making NP and AF non-vascularized tissue \[[@B3]\]. In addition, IVD is a poorly cellularized tissue \[[@B3]\]. Both these characteristics are responsible for the limited intrinsic repair capacity of IVD. Accordingly, IVD damages are irreversible and often result in clinical symptoms, such as low back pain, that require medical intervention \[[@B5]\]. Such treatments currently involve removal or replacement of the injured disc by surgery rather than its repair, which would be the preferred course of action. Successful transplantation of IVD autografts, allografts (fresh and fresh-frozen) have also been considered in primate models and in humans \[[@B6]\]. However the safety and efficiency of such techniques remain to be clarified. In this context, the use of cell-seeded biomaterials for tissue engineering of the IVD has been recently investigated \[[@B7],[@B8]\]. Although results thus far are promising, the development of an in vivo model that can closely mimic human IVD aging and degeneration is crucial to test the efficacy of such future regenerative cell-based biotherapies. Among the different animal models described in the literature \[[@B9],[@B10]\], the use of the rabbit disc model because in addition to being cost effective and the most widely investigated, it also appears to be a relevant model of age-linked altered proteoglycan metabolism and disc injury. The present study was conducted to address whether the rabbit could be considered (i) as a valuable model to provide new insights into the tissue and cellular changes occurring during IVD aging and degeneration and (ii) as an appropriate tool to investigate the efficacy of IVD biotherapies. To this end, lumbar IVD from rabbits with increasing ages (1, 6 and 30 months old) were compared by MRI and histological staining. In an effort to determine whether a close correlation may exist between the tissue and cellular changes observed during the early course of aging, we also analyze the variation of transcript expression in NP cells with increasing age. Methods ======= Animals and surgical procedures ------------------------------- All animal handling and surgical procedures were conducted according to European Community guidelines for the care and use of laboratory animals (DE 86/609/CEE). New Zealand White rabbits (Charles River, L\'Arbresle, France), 1, 6 and 30-month old, were used. The study protocol has been approved by the ethics committee at the National Veterinary School of Nantes (ENVN-ONIRIS). Three rabbits per age group were used to perform MRI and histological analyses as well as five additional rabbits per age group for the RT-PCR analysis. MRI scanning and image assessment --------------------------------- MRI scans were performed using a one Tesla clinical magnet (Siemens Magnetom Harmony/Syngo). One, six, 30 month old rabbits (n = 3 for every ages) were anesthetized by intramuscular injection of xylazine (Bayer, Puteaux, France) and ketamine (Merial, Lyon, France). A 2.5 mm midsagittal section image was obtained, using a T2-weighted imaging sequence (T2ws) (TR, 5000 milliseconds; TE, 111 milliseconds). Images were analysed by Pfirrmann\'s grading \[[@B11]\]. This grading system is based on MRI signal intensity, disc structure, distinction between NP and AF and disc height on T2-weighted magnetic resonance images. The degree of disc degeneration was ranked from grade 1 (normal disc) to grade 5 (severe degeneration) as previously described \[[@B11]\]. An evaluation of MRI scans was performed by three independent investigators expert in MRI image reading. Histological analysis --------------------- After rabbit euthanasia (n = 3 for every age group), the lumbar IVD were collected from five consecutive levels (L2-3, L3-4, L4-5, L5-6 and L6-7). IVD were dissected \[[@B2]\] and fixed in 10% paraformaldehyde for four days, decalcified for 24 hours in Decalcifier II^®^(Surgipath, Richmond, USA). After dehydration and incubation with Histosol^®^(Shandom, Brussels, Belgium), specimens were embedded in paraffin and sectioned into 3-μm slices. For histological analysis, 3 μm thick paraffin sections were deparaffined using toluene, rehydrated through a graded series of ethanol, and rinsed in distilled water. Sections were stained with hematoxylin phloxin safran (HPS) and with 0.1% alcian blue (Sigma-Aldrich, St Louis, USA) \[[@B12]\]. Histological sections were analyzed using Boos\' modified scoring \[[@B13]\]. Four parameters were specifically assessed to classify age-related changes in IVD: cell density, mucous degeneration, tears and cleft formations and granular changes \[[@B13]\]. Each parameter was ranked from 0 to 4 according to the intensity of the tested parameters (0: lowest; 4: highest). A blind evaluation of histological samples was performed by three independent investigators expert in reading histological slides. Transcript expression analysis ------------------------------ Five rabbits of each age were sacrificed. NP tissues from five consecutive lumbar levels (L2-3, L3-4, L4-5, L5-6 and L6-7) were isolated and enzymatically digested as previously described \[[@B2]\]. NP cells were frozen for subsequent real time PCR analysis. Total RNA was extracted using Trizol^®^reagent according to the manufacturer\'s instructions. Real-time PCR experiments were conducted as described extensively in \[[@B2]\]. Briefly, after deoxyribonuclease I digestion, RNA samples (2 μg) were reverse-transcribed using AMV-RT and random primers in a total volume of 30 μl. Complementary DNA (cDNA) was amplified in a total volume of 25 μl PCR reaction containing 12.5 μl of brilliant SYBR Green Master Mix (1X) and 30 nM of SYBR Green reference dye. The sequence of each primer (MWG Biotech, Ebersberg, Germany) set used are given in Table [1](#T1){ref-type="table"}. Real-time PCR was performed using the Mx3000P QPCR system (Stratagene, La Jolla, CA, USA). Cycle thresholds were normalized to β-actin in order to correct for potential cDNA quantification differences. The expression levels of each gene were normalized to the average of all the 1 month-old rabbits and the results are reported as fold change in gene expression. ###### real-time PCR investigation. --------------------------------------------------------------------------------------------------------------------------------------- Gene (Abbreviation) Accession number (Gene Bank) Forward primer (5\'-3\') Reverse primer (5\'-3\') Product length --------------------------------- ------------------------------ -------------------------- -------------------------- ---------------- *GAPDH* [NM_002046](NM_002046) agccacatcgctcagaca gcccaatacgaccaaatcc 66 pb Type II collagen\ [D83228](D83228) acagcaggttcacctataccg cccacttaccggtgtgtttc 60 pb *COL2A1* Aggrecan\ [L38480](L38480) gaggatggcttccaccagt tggggtacctgacagtctga 61 pb *AGC1* Type I colagen\ [D49399](D49399) agcgatggtcctccaggt gccagggtaaccacgttct 63 pb *COL1A1* Matrix metalloproteinase 13\ [NM_001082037](NM_001082037) ttttgaagacacgggcaag tcatcatagctccagacttggtt 60 pb *MMP13* Bone Morphogenic Protein 2\ [NM_001082650](NM_001082650) tgcagaacttcaggtttttcg tggaaaccgctgtcgtct 63 pb *BMP2* Matrix Gla protein\ [D21265](D21265) tggatataatgctgcttacaatcg tttccaatcttattcagctctgc 64 pb *MGP* P21-activated protein kinase I\ [NM_001082756](NM_001082756) agaaagaaaaagaacgaccagaga cgtggatggtgtgctcaa 60 pb *P21* --------------------------------------------------------------------------------------------------------------------------------------- GenBank reference of the genes evaluated and oligonucleotide primers used for real-time PCR. Statistical analysis -------------------- Each analysis was performed on three rabbits per age group for MRI and histology and on five rabbits per age group for RT-PCR. Results are expressed as mean +/- SD of triplicate samples. Comparative studies of means were performed using one-way ANOVA followed by a post hoc test (Fisher\'s projected least significant difference) with a statistical significance at p \< 0.05. Results ======= MRI and histological characterization of IVD aging -------------------------------------------------- In order to characterize the aging process of IVD, the T2ws intensity of lumbar spine from one, six and 30-month-old rabbits was assessed. Representative MRI picture of one-month-old IVD on Figure [1A](#F1){ref-type="fig"} indicates a grade I according to the Pfirrmann\'grading to IVD from L2 to L6 levels. The lowest lumbar IVD (L6-L7) appears non-homogeneous with a dark transverse band. A grade II was assigned to this disc. The six-month-old rabbits\' (Figure [1B](#F1){ref-type="fig"}) NP from L2 to L7 level showed a nonhomogeneous T2ws intensity with the presence of a dark transverse band and were assigned as grade II. Six-month-old disc height was not altered as compared to that of one-month old disc but the distinction between NP and AF became less discernable. For the 30-month-old rabbits (Figure [1C](#F1){ref-type="fig"}), all the IVD at lumbar level exhibited a non-homogeneous T2 signal intensity (gray). The IVD height was slightly decreased, and no distinction between NP and AF could be observed. These were scored as grade III. In summary, MRI data indicate that rabbits exhibit scalable IVD changes evolving as early as one-month old. {#F1} To further address whether the age-dependent MRI changes of rabbit IVD may correlate with some tissue alteration, we then performed histological stainings. Analysis of histological staining revealed a dramatic decrease in cell density as a function of age (Figure [2A, B, C](#F2){ref-type="fig"}). In addition, mucous degeneration and granular changes, shown by the presence of large decoloured and heterogeneous areas in granulated regions, were markedly increased (Figure [2D, E, F](#F2){ref-type="fig"}). Finally, the formation of tears and clefts shown by the presence of small thin defects was also affected by aging. To quantitatively assess the histological damage observed as a function of age, we next used a modified Boos\' scoring \[[@B13]\] (Figure [2G](#F2){ref-type="fig"}). Our results show a significant increase in the modified Boos\' scoring from one to 30 months old. This scoring was 5.5-times higher in the 30-month-old rabbits compared to one-month old rabbits (p \< 0.05). {#F2} Alteration of transcript expression patterns during IVD aging ------------------------------------------------------------- To address whether the structural age-related changes identified by MRI and histology may be related to alteration of cell phenotype, we analysed the expression levels of transcript coding for several IVD associated genes in freshly isolated NP cells from increasing ages. We first focused our attention on transcript coding for ECM genes (type II collagen COL2A1, type I collagen COL1A1 and aggrecan AGC1). Our data indicated that COL2A1 mRNA levels decreased significantly with aging in NP cells (Figure [3A](#F3){ref-type="fig"}). The COL2A1 mRNA level was 100-fold lower in the 30-month-old rabbit compared to that of a one-month-old rabbit (p \< 0.05). In parallel, a significant 2.5-fold decrease in AGC1 mRNA levels was also noted between one and 30-month-old rabbits (p \< 0.05) (Figure [3B](#F3){ref-type="fig"}). In contrast to the decrease in COL2A1 and AGC1 mRNA levels, the COL1A1 mRNA level increased significantly with aging in NP cells (Figure [3C](#F3){ref-type="fig"}) with a 2.5-fold increase in COL1A1 mRNA at the age of 30-months compared to one-month. Taken together, these data indicate that the expression of chondrogenic markers (COL2A1 and AGC1) decreases with age whereas that of the dedifferentiation marker (COL1A1) increases as early as thirty months. {#F3} Since it has been reported that the expression of matrix metalloprotease 13 (MMP13), bone morphogenetic protein-2 (BMP2) and the matrix Gla protein (MGP) were affected during the course of osteoarthritis \[[@B14]\] and IVD aging \[[@B15]\], we were interested in deciphering whether these genes may also be affected by aging in rabbit NP. Whilst a significant decrease in MMP-13 mRNA levels was found between one and six-month-old NP, our data indicate that MMP13 mRNA levels increased significantly between 6 and 30-month old rabbit (2.8-fold higher) (p \< 0.05) (Figure [3D](#F3){ref-type="fig"}). In addition, BMP2 mRNA levels was found to be significantly up-regulated between one and 30-month-old rabbits with a 2.5 fold increase (p \< 0.05) (Figure [3E](#F3){ref-type="fig"}). MGP mRNA levels were significantly increased as a function of age with a significant seven-fold increase between one-and 30-month-old rabbits (Figure [3F](#F3){ref-type="fig"}). Finally, since IVD aging has recently been associated with cellular senescence \[[@B16]\], we investigated whether the level of p21 mRNA may vary as a function of age (Figure [3G](#F3){ref-type="fig"}). Interestingly, our transcript analysis revealed that p21 mRNA level was significantly up-regulated (2 fold increase) between six month and 30-month-old rabbits. Viewed together, our data strongly suggest that the age-dependent tissue changes evidenced by MRI and histology were associated with alteration of the phenotype of NP cells. Discussion ========== The current therapeutic strategies for patients with lower back pain remain symptomatic and are mainly dedicated to relieving painful symptoms. Our current understanding of the physiopathology of IVD degeneration allows us to consider regenerative medicine as a promising strategy. Indeed, recent reviews have considered the imbalance between anabolism and catabolism as a pivotal factor of the IVD degeneration process. This imbalance between anabolism and catabolism contributes to the disorganization of the ECM. Accordingly, increasing attention has been paid to the regeneration of functional tissue based on the restoration of the ECM integrity by cell therapy and/or tissue engineering \[[@B3],[@B17]\]. Nevertheless, before these promising biotherapies may enter the therapeutic arsenal, some preclinical tests in adapted animal models that closely mimic the physiological aging and degeneration process of human IVD should be performed. In this context, the present study aims at characterizing IVD aging and degeneration in the rabbit to propose this model as a suitable tool to i) gain new insights in the complex mechanisms of IVD degeneration and ii) to perform preclinical experiments dedicated to the evaluation of the safety and efficacy of IVD tissue engineering strategies. In humans, MRI is the gold standard for the clinical investigation of IVD integrity. This technique allows the definition of IVD based on the tissue hydration shown by the intensity of the T2ws in the NP. The pictures of rabbit lumbar spines showed an age-dependent decrease in the T2ws and the appearance of a dark transverse band. This data is quite similar to that observed in humans during the course of IVD degeneration \[[@B18]\]. The decrease in T2ws intensity in human IVD after 20 years of age is well-acknowledged and reflects the decrease in water and proteoglycan contents during aging \[[@B19]\]. The dark transverse band is considered to be a normal structure in persons of 30 years or older \[[@B18]\]. Histological studies have thus suggested that this dark transverse band may likely to be due to the accumulation of fibrous tissue in the central zone of NP \[[@B20]\]. To precisely assess disc degeneration on routine T2ws in human, Pfirrmann et al. have developed a reliable grading system \[[@B11]\]. Interestingly, the use of Pfirrmann\'s system to grade IVD degeneration in rabbit reveals the same profile of T2ws variation as a function of age, thereby confirming the reliability of the different periods of life in rabbits previously described by Murakami et al. \[[@B21]\]: grade I for childhood (one-month old in rabbits is similar to a child in humans), grade II for adolescent rabbit (six months old in rabbits is similar to an adolescent in humans) and grade III-IV in adult age (30 months old in rabbits is similar to an adult in humans). All these MRI data suggest that the age-related changes that occurs in rabbit IVD share consistent similarities with those traditionally observed during the course of human IVD aging. To further address whether the age-related changes observed by MRI may correlate with some tissue alteration, we then performed histological stainings. We were interested in addressing whether the modified Boos\' scoring, originally developed to histologically assess IVD changes in humans, could be transposed to rabbit IVD aging. Accordingly, the main features of human IVD degeneration were found in rabbit IVD with mucous degeneration, granular changes as well as cracks and tears being formed as a function of age. Interestingly, these histological changes perfectly corroborate the changes observed by MRI (loss of T2ws intensity and formation of a dark band). Taken together, our MRI and histological data show some similarities between the process of IVD aging in rabbits and humans thereby strongly suggesting that rabbits could be considered as a valuable spontaneous model of age-dependent IVD changes. In addition to being a reliable model of IVD aging, this spontaneous model may provide a closer representation of the pathophysiology of IVD aging in humans compared to the degeneration induced by needle aspiration or chemonucleolysis \[[@B9]\]. Despite our data strongly suggesting that the rabbit is a relevant model of IVD aging, some limitations should, however, be considered. Contrarily to humans, rabbits have notochordal cells in their NP at least up to 6 months of age \[[@B22]\]. In addition, in small quadrupeds and despite a relative discrepancy in disc diameter, the spine is likely to be loaded with smaller forces as compared to humans \[[@B23],[@B24]\]. Whether both of these limits affect the relevance of the rabbit model is not yet known. To our knowledge, only two studies have partially described IVD aging in rabbits \[[@B25],[@B26]\]. In these previous studies, rabbit IVD aging was evaluated longitudinally between the age of 6 and 42 months. In this previous study, rabbit IVD aging is longitudinally evaluated between the age of six and 42 months. On the contrary to our study, no result is available for the period ranging from one to six months, which could be an essential period for the onset of IVD aging. In fact, it is well known that the NP of less than twelve-month-old rabbits contain some remnant cells originating from the embryonic notochord \[[@B22]\]. Whether the age-dependent loss of notochordal cells, most likely by apoptosis, is intimately related to IVD aging is not yet fully deciphered but has been proposed as a possible initiating mechanism for IVD degeneration \[[@B27]\]. NP is the structure in which IVD degeneration is initiated by dehydration followed by an alteration of the structural organization of the tissue \[[@B13]\]. Therefore, to strengthen our MRI and histological data, we finally sought to decipher whether the above described age-dependent tissue changes may correlate with alteration in the NP cell phenotype. To address this issue, we focused our attention on the expression levels of molecules that were previously shown to be modulated during the onset of IVD degeneration or osteoarthritis (COL1A1, COL2A1, AGC1, BMP2, MMP13, MGP and P21) \[[@B14],[@B16],[@B28]\]. Since it remains difficult to quantitatively evaluate the level of the corresponding proteins, particularly in the rabbit, we embarked on the analysis of the corresponding transcript levels by real-time RT-PCR. Our data describing an age-dependent increase in COL1A1 expression with a concomitant decrease in COL2A1 and AGC1 expression suggest that NP cells experience a process of dedifferentiation. Interestingly, this dedifferenciation process has been well described in cultured articular chondrocytes \[[@B29]\] and in osteoarthritic joints \[[@B30]\]. Our data also suggests that during the process of IVD aging, the molecules with a longer half-life (collagen type II) exhibit an early decrease in the corresponding transcript levels. Conversely, the molecules with a shorter half-life (aggrecan) have a transcript expression levels that starts to decline later on. Whether this differential regulation of ECM molecules may be of importance for the maintenance of IVD integrity deserves further attention. Among the various genes modulated in osteoarthritic chondrocytes \[[@B14]\], MMP-13 is probably the most trustworthy. We were therefore interested to find a significant modulation of these genes during IVD degeneration. MMP-13 is known to degrade collagens and glycosaminoglycans \[[@B31]\]. The increase in MMP13 could therefore be a major contributor of IVD degeneration \[[@B16]\] as it has been extensively reported in cartilage degradation during OA \[[@B14]\]. BMP-2, a growth factor that stimulates the production of ECM components in IVD \[[@B21]\] and articular cartilage \[[@B32]\], was also shown to be stimulated as a function of age in rabbit IVD. One can assume that the increase in BMP2 expression during IVD aging is related to the existence of reparative mechanisms that could contribute to slowing down ECM degradation. In line with this suggestion, MGP, one of the most potent inhibitors of calcification in mammals \[[@B33]\], was also found to be up-regulated as a function of age \[[@B15]\] thereby supporting our hypothesis of the existence of such compensatory mechanisms. Of note, similar compensatory mechanisms have been described in osteoarthritic cartilage, which draws attention to the physiopathological vicinity of IVD degeneration and osteoarthritis. Finally, the age-dependent increase in p21, a cycline dependant kinase inhibitor that in addition to being involved in cellular senescence \[[@B16]\] has also been reported to inhibit type II collagen expression in articular chondrocytes \[[@B34]\], further highlights the similarities between OA and IVD degeneration. Conclusions =========== This study describes the consistency of the rabbit as a model of age-dependent IVD changes. Our data also highlight some similarities between the physiopathological processes involved in the onset of IVD aging and those described for osteoarthritis. Finally, our study makes the rabbit a valuable tool (i) to gain new insights into the complex molecular mechanisms that govern IVD aging and (ii) to test the preclinical efficacy of tissue engineering strategy that may offer the possibility of regenerating damaged IVD. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= JC and MPV have conceived and coordinated the study. They participated in MRI analysis, Boos\'scoring and RT-PCR analysis. They also performed the statistical analysis and drafted the manuscript. MM performed the PCR analysis. JL performed the histological stainings. BHF, OG participated in the design of the study, particularly in animal management. MF performed MRI imaging and participated in analysis. YC participated in histological analysis. GG and YM have helped in the writing of the manuscript. JG and CV helped to the conception of the study, and participated in its design and have helped in the writing of the manuscript. All authors read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2474/12/147/prepub> Acknowledgements ================ This study was supported by grants from Société Française de Rhumatologie, Fondation de l\'Avenir pour la Recherche Médicale Appliquée (ET8-491 and ET9-541), Agence Nationale de la Recherche AAP TeCSAN (chondrograft project), Agence de la Biomédecine, Institut National de la Santé et de la Recherche Médicale, région des Pays de la Loire (BIOREGOSproject), and the University Hospital of Nantes. The authors gratefully acknowledge the technical assistance of Pierre Weiss, Maïthe Gatius-Perré, Pierre Monmousseau, Patrice Roy, Stéphane Madec, Dominique Rouleau and Christian Raphael, as well as Joanna Ashton-Chess for the critical reading of the manuscript.

Current constraints in blood glucose monitoring and insulin delivery technologies limit the ability of most individuals with type 1 diabetes to safely achieve and maintain recommended glucose and hemoglobin A~1c~ (HbA~1c~) targets. Severe hypoglycemia (SH) remains a common side effect of intensive treatment and a major barrier to achieving normoglycemia in type 1 diabetes. Several prior studies have evaluated factors associated with an increased risk of SH. In a study of 1,190 children and adolescents with type 1 diabetes, Craig et al. ([@B1]) reported that younger age, male sex, longer duration of diabetes, and intensive insulin therapy (≥3 injections/day) were associated with an increased risk of SH. In a study of 60 individuals, mainly adults, with insulin-dependent diabetes, Gold et al. ([@B2]) reported that the occurrence of SH was associated with prior SH, hypoglycemia unawareness, older age, and autonomic dysfunction. In the Diabetes Control and Complications Trial (DCCT) ([@B3]), an analysis of the first 424 intensively treated subjects found that predictors of SH in the intensive group included prior SH, longer duration of diabetes, higher baseline HbA~1c~, lower recent HbA~1c~, and higher baseline insulin doses. A later analysis of all 1,441 subjects found that a higher SH rate in both treatment groups occurred in subjects with prior SH, longer duration of diabetes, absent residual C-peptide secretion, younger age (adolescents compared with adults), and higher baseline insulin doses; the rate was higher in females than males in the conventional group but not in the intensive group and higher in those with lower baseline HbA~1c~ in the conventional group but not the intensive group ([@B4]). Recurrent episodes of mild hypoglycemia appear to cause defects in counterregulatory hormone responses to subsequent hypoglycemia placing patients with type 1 diabetes at increased risk of severe hypoglycemia. This sequence of events has been termed hypoglycemia-associated autonomic failure. The evidence supporting the development of hypoglycemia-associated autonomic failure was initially demonstrated in clinical research center--based hypoglycemic clamp studies, and a relationship between the risk of SH and antecedent biochemical hypoglycemia in the free living condition also has been reported ([@B5]). The Juvenile Diabetes Research Foundation (JDRF) Continuous Glucose Monitoring Study Group recently reported the results of a 6-month randomized clinical trial and a 6-month extension study that evaluated the effectiveness of real-time continuous glucose monitoring (CGM) in 451 intensively treated type 1 diabetes subjects who had baseline HbA~1c~ levels both within and above the target range ([@B6]--[@B11]). These studies provided a large dataset to evaluate the association of clinical and demographic factors with the development of SH. In addition, longitudinal CGM glucose data were available to evaluate the relationship between biochemical hypoglycemia detected by CGM and subsequent SH. RESEARCH DESIGN AND METHODS {#s5} =========================== The study protocol and clinical characteristics of enrolled subjects have been described in detail ([@B7]--[@B9]). Major eligibility criteria included age ≥8 years, type 1 diabetes for at least 1 year, use of either an insulin pump or multiple (at least three) daily insulin injections, and HbA~1c~ level \<10.0%. Prior SH was not an exclusion and 8% of subjects in both treatment groups self-reported at least one SH event in the 6 months prior to study entry. The study consisted of a 6-month randomized trial in which subjects were randomized to either a control group that used standard home blood glucose monitoring or a CGM group that used one of the following three CGM devices: the FreeStyle Navigator (Abbott Diabetes Care, Inc., Alameda, CA), the MiniMed Paradigm REAL-Time Insulin Pump and Continuous Glucose Monitoring System (Medtronic MiniMed, Inc., Northridge, CA), or the DexCom SEVEN (DexCom, Inc., San Diego, CA). The randomized trial was followed by a 6-month extension study in which CGM was initiated in the control group and continued in the CGM group. Analysis was limited to 436 (97%) of 451 randomized subjects who completed 12 months of follow-up. The 15 subjects with incomplete follow-up included one subject who was believed to be factitiously producing SH by intentional insulin overdose and 14 others who did not experience SH before dropping out of the study. SH was defined as an event that required assistance from another person to administer carbohydrate, glucagon, or other resuscitative actions ([@B3]). The proportional hazards model was used to evaluate the association of baseline demographic and clinical factors with the occurrence of SH events in univariate models. Factors in the univariate models with a *P* value \< 0.20 were included in an initial multivariate model and then a backward elimination procedure was used to remove variables with a *P* value \> 0.05. However, because of multiple statistical comparisons, only *P* values \< 0.01 were considered significant. A forward selection process resulted in a similar model. To avoid colinearity in the model building, only one baseline CGM measure of hypoglycemia (percentage of values ≤70 mg/dL) was included in the models. Results were similar for the highly correlated hypoglycemic area under the curve (AUC) and the low blood glucose index (LBGI) ([@B12]) calculated from CGM data (data not shown). CGM measures of glycemic variation such as SD, coefficient of variation (defined as SD divided by the mean glucose), and the absolute rate of change ([@B13]) were also confounded with percentage of CGM values ≤70 mg/dL and were excluded from the models. Subjects with missing values for covariates were excluded from the corresponding univariate models. For the multivariate models, missing values were treated as a separate category for discrete covariates, and an indicator for missing values was added to the model for continuous covariates. The SH rates in the control group and CGM group during their first 6 months of usage were compared using a Wilcoxon rank sum test. A paired signed rank test was used to compare the SH rate of the CGM group between the first and second 6 months. A repeated-measures logistic regression with generalized estimating equations was used to compare the SH rate between days with and without CGM use. A second analysis evaluated the association of four CGM hypoglycemia indices (% ≤70 mg/dL, hypoglycemic AUC, LBGI, and at least 30 consecutive min ≤54 mg/dL) during 1 day with the occurrence of SH on the following day using repeated-measures logistic regression with generalized estimating equations to account for correlated data. Inclusion in this analysis was limited to those subjects who had at least one SH event for which there was at least 12 h of CGM glucose data available from the preceding day. When an additional hypoglycemic event occurred within 3 days after a prior hypoglycemic event, the event was not considered as a new event and was not counted (*N* = 1). Operating characteristics (sensitivity, specificity, false alarm rate, and positive predictive value \[PPV\]) are given for various cut points for the four CGM hypoglycemic indices. RESULTS {#s6} ======= One or more SH events occurred in 54 (12%) of the 436 subjects; 36 (8%) subjects experienced one event, 13 (3%) subjects had two events, 4 (0.9%) subjects had three events, and 1 (0.2%) subject had four events. The overall incidence rate of SH was 17.9 events per 100 person-years, being 21.3 in the 160 subjects ≥25 years of age, 16.0 in the 138 subjects 15--24 years of age, and 15.9 in the 138 subjects 8--14 years of age. The rate was 21.5 in the first 6 months of use by the CGM group and 15.0 during the 6 months of CGM use in the control group (which followed the 6-month randomized trial) (*P* = 0.56). Within the CGM group, there was a trend toward less SH during the second 6 months compared with the first 6 months (8.0 vs. 21.5 events per 100 person-years, respectively; *P* = 0.02). The clinical characteristics of the 436 subjects are shown in [Supplementary Table 1](http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc10-1111/-/DC1) according to whether or not an SH event occurred during the study. In a univariate analysis, SH was more likely to occur in subjects who had experienced SH in the 6 months prior to study entry (*P* \< 0.001), and there were suggestive trends for more frequent SH in adults (*P* = 0.06), females (*P* = 0.05), subjects with higher scores on the Hypoglycemia Fear Questionnaire (*P* = 0.02), those with a higher percentage of baseline CGM values ≤70 mg/dL (*P* = 0.02), and those who had higher glucose variability as assessed with the coefficient of variation (*P* = 0.08). In general, these factors also were associated with previous SH; consequently, in multivariate analysis only SH during the prior 6 months (hazard ratio \[HR\] 6.2 \[95% CI 3.4--11.6\]; *P* \< 0.001) and female sex (2.3 \[1.3--4.1\]; *P* = 0.006) ([Table 1](#T1){ref-type="table"}) were independent predictors of SH during the study. Although the associations of SH during the prior 6 months and female sex with the occurrence of SH were highly statistically significant, the PPV for each was low (42 and 15%, respectively). The occurrence of SH was not associated with baseline HbA~1c~ level. SH occurred in similar proportions of subjects who used an insulin pump and those who used multiple daily injections of insulin. ###### Proportional hazards models of baseline factors predictive of SH (*N* = 436 subjects who completed the 52-week visit)[\*\*](#t1n7){ref-type="table-fn"} *N* \% SH[‡](#t1n3){ref-type="table-fn"} Univariate Initial multivariate[\*](#t1n1){ref-type="table-fn"} Final multivariate[†](#t1n2){ref-type="table-fn"} ---------------------------------------------------------------------- ----- -------------------------------------- ------------ ------------------------------------------------------ --------------------------------------------------- ----- ------------- ------------------------------------- ----- ------------- --------- Overall 436 12 Age (years) 0.06[§](#t1n4){ref-type="table-fn"} 0.29[§](#t1n4){ref-type="table-fn"}  8--14 138 10 1.0 1.0  15--24 138 12 1.2 (0.6--2.5) 0.8 (0.4--1.7)  ≥25 160 14 1.4 (0.7--2.8) 1.0 (0.5--2.0) Sex 0.05 0.02 0.006  Male 199 9 1.0 1.0 1.0  Female 237 15 1.8 (1.0--3.1) 2.2 (1.2--4.1) 2.3 (1.3--4.1) *n* SH events in 6 months prior to study \<0.001 \<0.001 \<0.001  None 400 10 1.0 1.0 1.0  ≥1 36 42 5.0 (2.8--9.2) 5.5 (2.8--10.6) 6.2 (3.4--11.6) Fingersticks per day\|\| 0.65[§](#t1n4){ref-type="table-fn"}  ≤5 135 17 1.0  6--8 179 8 0.5 (0.2-- 0.9)  ≥9 69 13 0.8 (0.4--1.7) Insulin delivery 0.70  Injections 80 14 1.0  Pump 356 12 0.9 (0.5--1.7) HbA~1c~ (%) 0.32[§](#t1n4){ref-type="table-fn"}  \<7.0 127 13 1.0  7.0 to \<8.0 197 14 1.1 (0.6--2.1)  ≥8.0 112 9 0.7 (0.3--1.5) Hypo Fear Score[¶](#t1n5){ref-type="table-fn"} 0.02[§](#t1n4){ref-type="table-fn"} 0.46[§](#t1n4){ref-type="table-fn"}  \<20 151 8 1.0 1.0  20 to \<30 96 15 1.9 (0.9--4.2) 1.5 (0.7--3.3)  ≥30 184 15 2.0 (1.0--3.9) 1.4 (0.7--2.9) \% CGM values ≤70 mg/dL (%)[\#](#t1n6){ref-type="table-fn"} 0.02[§](#t1n4){ref-type="table-fn"} 0.75[§](#t1n4){ref-type="table-fn"}  None 25 8 1.0 1.0  \<5 207 9 1.2 (0.3--5.0) 0.9 (0.2--3.8)  5 to \<15 160 16 2.1 (0.5--8.9) 1.2 (0.2--5.6)  ≥15 44 16 2.1 (0.4-- 10.2) 1.3 (0.2--8.3) Glucose coefficient of variation (%)[††](#t1n8){ref-type="table-fn"} 0.08[§](#t1n4){ref-type="table-fn"} 0.54[§](#t1n4){ref-type="table-fn"}  \<35 116 7 1.0 1.0  35 to \<40 115 13 2.0 (0.8--4.6) 1.9 (0.8--4.5)  40 to \<45 88 16 2.5 (1.0--5.9) 2.4 (0.9--6.2)  ≥45 117 15 2.2 (0.9--5.0) 1.5 (0.5--4.3) \*Factors with *P* value ≤ 0.20 in univariate model are included in the initial multivariate model. †Factors with *P* value ≤ 0.05 in the initial multivariate model are kept in the final multivariate model. ‡Percentage of subjects with at least one SH event during the study. §*P* value calculated as a continuous variable. Categories are for display purposes in this table. \|\|Self-reported number of home glucose meter tests per day. Data collected after study initialization and are therefore missing for 53 subjects. ¶Hypoglycemia Fear Questionnaire ([@B20]) consists of 15 5-point Likert scale items, with scores scaled to a 0--100 range. Higher score denotes more fear of hypoglycemia. Missing for five subjects. \#CGM data based on blinded use at baseline for approximately 1 week prior to randomization. Results were similar for hypoglycemic AUC and LBGI ([@B12]) (data not shown). \*\*Diabetes duration was not associated with SH. Data not shown because this factor was highly confounded with age. ††Coefficient of variation is the SD divided by the mean glucose from the CGM expressed as a percentage. The second analysis evaluated the predictive value of CGM-measured hypoglycemia during 1 day with the occurrence of SH on the following day. During the full 12 months of follow-up of the CGM group and the last 6 months of follow-up of the control group (the time period during which CGM was used), 48 SH events occurred in 40 subjects. For 31 of the 48 events (65%), CGM was used on the day of the event, which was comparable with a usage rate of 71% on the 11,994 days without an SH event (*P* = 0.40). For 27 of the 48 events (*N* = 24 subjects), a sensor was used on the day prior to the event (for at least 12 h). Median percentage of time with glucose levels ≤70 mg/dL was 3% during the 24 h of the calendar day prior to SH compared with 2% of the time on other days (*P* \< 0.001). Although this association was strong statistically, the PPV was extremely low (∼5%), and the false alarm rate was extremely high (∼95%) even when 30% or more of the glucose values were ≤70 mg/dL on the day prior to a SH event ([Table 2](#T2){ref-type="table"}). Findings were similar for hypoglycemic AUC (0.2 on the day prior to SH vs. 0.1 on other days, *P* = 0.002) and LBGI (1.1 vs. 0.8, *P* = 0.003), with PPVs being low and false-positive rates being high for each ([Supplementary Table 2](http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc10-1111/-/DC1)). Results also were similar when assessing the predictive value of 30 consecutive min below 54 mg/dL ([Supplementary Table 2](http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc10-1111/-/DC1)). Median glucose was 131 mg/dL on the day prior to an SH event and 141 mg/dL on other days (*P* = 0.86). ###### Sensitivity, specificity, false alarm rates, and PPV of CGM-measured hypoglycemia on 1 day for the occurrence of SH on the following day CGM glucose readings ≤70 mg/dL on prior day (%) *n* of days Sensitivity[\*](#t2n1){ref-type="table-fn"} Specificity[†](#t2n2){ref-type="table-fn"} False alarm[‡](#t2n3){ref-type="table-fn"} PPV[§](#t2n4){ref-type="table-fn"} ------------------------------------------------- ------------- --------------------------------------------- -------------------------------------------- -------------------------------------------- ------------------------------------ ------- ------ 0 2,009 1,999 10 \>0 3,286 3,269 17 63% 38% 99.5% 0.5% ≤5 3,292 3,278 14 \>5 2,003 1,990 13 48% 62% 99.4% 0.7% ≤15 4,613 4,596 17 \>15 682 672 10 37% 87% 98.5% 1.5% ≤30 5,184 5,162 22 \>30 111 106 5 19% 98% 95.5% 4.5% All 5,295 5,268 27 \*Sensitivity, Proportion of true SH events where the CGM indices correctly predicted the prior days as positive. †Specificity, Proportion of days without SH where the CGM indices correctly predicted the prior days as negative. ‡False alarm, Proportion of days with CGM indices predicted as positive where there were no SH in the following days. §PPV, Proportion of days with CGM indices predicted as positive where there were SH events in the following days (this is 100% minus the false alarm rate). CONCLUSIONS {#s7} =========== We found the rate of occurrence of SH during the study to be most strongly associated with a history of SH in the 6 months prior to entry into the study. In addition, the rate was higher in females than males. Both of these findings are consistent with prior findings in the DCCT ([@B3],[@B4]). As in our study, multivariate analyses conducted on the DCCT data did not identify a predictive model with high sensitivity ([@B3]). The incidence rate of SH in this study (17.9 events per 100 person-years) was similar to that of the conventional therapy group in the DCCT (18.7 events per 100 person-years), but significantly lower than the rate in the intensive treatment group (61.2 events per 100 person-years) in the DCCT ([Supplementary Fig. 1](http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc10-1111/-/DC1)) ([@B4]). A similar SH rate was found in the Sensor-Augmented Pump Therapy for A1c Reduction (STAR) 3 trial (∼13 events per 100 person-years in both the CGM group and control group) ([@B14]). Our results need to be viewed in the context of the study participants who were well-versed in self-management, were receiving intensive insulin management with either an insulin pump or multiple daily injections of insulin, and were performing frequent home blood glucose monitoring. We also found that CGM-measured hypoglycemia occurred more often on days prior to SH than on other days. However, although the statistical association was strong, the predictive value of biochemical hypoglycemia for subsequent SH was very low. This is because on any given day, SH is a rare event (\<1% probability). This probability increases eightfold when more than 30% of CGM values the day prior are in the hypoglycemic range, but there is still less than a 5% chance of SH on the following day. Thus, if a CGM were programmed to sound a warning whenever 30% of values over a 24-h period were ≤70 mg/dL, more than 95% of alarms would be false. The four CGM measures of hypoglycemia studied here (% ≤70 mg/dL, AUC, LBGI, and ≤54 mg/dL for at least 30 consecutive min) are all highly correlated, and results were similar regardless which was used. One possibility to in part explain the low predictive value could be that subjects modified their diabetes management based on the presence of CGM-measured hypoglycemia, and this reduced their risk of an SH event on the next day. Evidence against this explanation, however, is that during the randomized trial phase of the study, the SH rate in the CGM group was similar to that in the control group ([@B8],[@B9]). Another possible factor contributing to the low PPV is measurement error from CGM. Studies of CGM accuracy have shown that the median error during hypoglycemia ranges from 13 to 24 mg/dL ([@B15],[@B16]) so that some episodes of true biochemical hypoglycemia are missed by CGM, and some CGM readings in the hypoglycemic range occur when the true glucose concentration is \>70 mg/dL. Kovatchev et al. ([@B17]) studied 96 adults with insulin-dependent diabetes and found that history of SH and LBGI calculated from 1 month of home glucose meter data accounted for 40% of the variance of SH episodes over the following 6 months. In another study of 85 adults with type 1 diabetes, Kovatchev et al. ([@B18]) reported that LBGI values from home glucose meter data were significantly higher in the 24 h prior to and immediately following an SH episode compared with other days in the same subjects. Cox et al. ([@B19]) reported that LBGI was predictive of SH with a sensitivity rate of 58--60% among 100 adults with type 1 diabetes, but did not report the false-positive rate. Our results with CGM data were similar to these studies in that hypoglycemic indices were significantly higher on the day prior to an SH event and that over 50% of SH events could be predicted from these measures depending on the threshold used. However, our data also show a very large false alarm rate (≥95%) when these indices are used to predict SH events. The SH rates in these previous studies, ranging from 192 to 803 events per 100 person-years ([@B17]--[@B19]), were much larger than that observed in the current study (17.9 events per 100 person-years) and in the DCCT. In conclusion, the ability to predict the likelihood that SH will occur in the near future remains elusive. The strongest predictor is the occurrence of prior SH. Although biochemical hypoglycemia substantially increases the risk of the occurrence of SH on the next day, SH only occurs in about 1 in 20 days after preceding biochemical hypoglycemia, and thus this is a poor predictor. Supplementary Material ====================== ###### Supplementary Data Clinical trial reg. no. NCT00406133, clinicaltrials.gov. This article contains Supplementary Data online at <http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc10-1111/-/DC1>. \*Members of the writing committee are listed in [APPENDIX]{.smallcaps}, and a full group listing is available in the [Supplementary Data](http://care.diabetesjournals.org/lookup/suppl/doi:10.2337/dc10-1111/-/DC1) online. The study funding was provided by JDRF (grant numbers 22-2006-1107, 22-2006-1117, 22-2006-1112, 22-2006-1123, and 01-2006-8031). Continuous glucose monitors and sensors were purchased at a bulk discount price from DexCom, Inc. (San Diego, CA), Medtronic MiniMed, Inc. (Northridge, CA), and Abbott Diabetes Care, Inc. (Alameda, CA). Home glucose meters and test strips were provided to the study by LifeScan, Inc. and Abbott Diabetes Care, Inc. The companies had no involvement in the design, conduct, or analysis of the trial or the manuscript preparation. Below is a listing of relationships of the investigators with companies that make products relevant to the manuscript. Research funds where listed below were provided to the legal entity that employs the individual and not directly to the individual. B.B. reports having received consulting fees, honoraria, travel reimbursement, and research funds from Medtronic MiniMed, Inc. and grant support from DexCom, Inc. B.A.B. reports having received grant support and serving on the Medical Advisory Board for Medtronic MiniMed, Inc., grant support and a speaker honorarium from Abbott Diabetes Care, Inc., and grant support from DexCom, Inc. C.K. reports having received consulting fees from Medtronic MiniMed, Inc. L.L. reports having received consulting fees and a speaker honorarium from Abbott Diabetes Care, Inc., and consulting fees and research funding from Medtronic MiniMed, Inc. W.V.T. reports having received consulting fees from Medtronic MiniMed, Inc. S.W. reports having received research support, a speaker honorarium and travel reimbursement from Medtronic MiniMed, Inc., and a speaker honorarium from Animas Corp/LifeScan, Inc. No other potential conflicts of interest relevant to this article were reported. The study was designed and conducted by the investigators. The writing group collectively wrote the manuscript and vouches for the data. The investigators had complete autonomy to analyze and report the trial results. There were no agreements concerning confidentiality of the data between JDRF, the authors, or their institutions. The Jaeb Center for Health Research had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. R.F.-S. researched data, contributed to discussion, wrote the manuscript, and reviewed and edited the manuscript. J.C. contributed to discussion, wrote the manuscript, and reviewed and edited the manuscript. R.W.B. contributed to discussion and reviewed and edited the manuscript. B.B. researched data, contributed to discussion, and reviewed and edited the manuscript. B.A.B. researched data, contributed to discussion, and reviewed and edited the manuscript. H.P.C. researched data, contributed to discussion, and reviewed and edited the manuscript. L.L. researched data, contributed to discussion, and reviewed and edited the manuscript. J.M.L. researched data, contributed to discussion, and reviewed and edited the manuscript. C.K. contributed to discussion and reviewed and edited the manuscript. N.M. researched data, contributed to discussion, and reviewed and edited the manuscript. K.J.R. researched data, contributed to discussion, and reviewed and edited the manuscript. W.V.T. researched data, contributed to discussion, and reviewed and edited the manuscript. S.W. researched data, contributed to discussion, and reviewed and edited the manuscript. D.M.W. researched data, contributed to discussion, and reviewed and edited the manuscript. H.W. researched data, contributed to discussion, and reviewed and edited the manuscript. D.X. contributed to discussion and reviewed and edited the manuscript. The JDRF CGM Study Group would like to recognize the efforts of the subjects and their families and thanks them for their participation. Co-authors: Rosanna Fiallo-Scharer, MD^1^; Jing Cheng, MS^2^. Additional authors (in alphabetical order): Roy W. Beck, MD, PHD^2^; Bruce A. Buckingham, MD^3^; H. Peter Chase, MD^1^; Craig Kollman, PHD^2^; Lori Laffel, MD, MPH^4^; Jean M. Lawrence, SCD, MPH, MSSA^5^; Nelly Mauras, MD^6^; William V. Tamborlane, MD^7^; Darrell M. Wilson, MD^3^; Howard Wolpert, MD^4^, of the JDRF Continuous Glucose Monitoring Study Group. ^1^Barbara Davis Center for Childhood Diabetes, Aurora, Colorado; the ^2^Jaeb Center for Health Research, Tampa, Florida; ^3^Stanford University, Stanford, California; the ^4^Joslin Diabetes Center, Boston, Massachusetts; ^5^Kaiser Permanente, San Diego, California; the ^6^Nemours Children's Clinic, Jacksonville, Florida; and ^7^Yale University, New Haven, Connecticut. Additional writing committee members include Bruce Bode, MD^1^; Katrina J. Ruedy, MSPH^2^; Stuart Weinzimer, MD^3^; Dongyuan Xing, MPH^2^. ^1^Atlanta Diabetes Associates, Atlanta, Georgia; the ^2^Jaeb Center for Health Research, Tampa, Florida; and ^3^Yale University, New Haven, Connecticut.

Introduction ============ The use of bicycles by children is a common and desirable activity in Canadian society. Several types of bicycle helmet legislation for young people have recently been adopted in various Canadian jurisdictions in an attempt to make bicycling a safer activity. In this paper, we examine bicycle helmet use and bicycling-related injury across population subgroups defined socially, economically and geographically in order to identify sub-population patterns and describe possible unfair and remediable differences, namely inequities that might exist and require attention. There is a significant amount of existing literature related to the use and effectiveness of bicycle helmets and bicycle helmet legislation, including three relevant Cochrane Collaboration systematic reviews \[[@B1]-[@B3]\]. The literature is roughly divided into two types. First, there are studies that examine the relationship between helmets and injuries (see for example \[[@B1],[@B2],[@B4]-[@B9]\], and second, studies that examine helmet use itself including the many determinants of helmet use or non-use and the interventions associated with encouragement of the use of helmets (see for example \[[@B3],[@B10]-[@B14]\]. There has been limited previous work that has examined the distribution of bicycling-related injury or helmet use. Bicycle and transportation safety education appears to be more successful with children under the age of 12 than with adolescents or young adults \[[@B15]-[@B17]\], there is some evidence suggesting the occurrence of sex and racial differences in helmet use \[[@B16],[@B18]\] as well as evidence linking helmet nonuse and bicycling-related injury with lower socioeconomic status \[[@B17]-[@B23]\]. There remains a need for further research into group differences in bicycle helmet use and bicycle-related injury among young people to identify vulnerable population sub-groups beyond, and in addition to, those divided by sex, age and socioeconomic status and particularly youth in different social and geographic contexts. The aim of our study was to identify sub-population patterns and possible inequities associated with bicycling, bicycle helmet use and bicycling-relating injury in young Canadians in order to better understand the health protection or health risk profile for these behaviours. Using data from Cycle 6 (2009/10) of the Canadian Health Behaviour in School-Aged Children Survey (HBSC), we addressed two study objectives: (1) To examine national patterns of bicycle and bicycle helmet use among Canadian youth by age, sex, school grade, socioeconomic status, number of years in Canada and urban--rural geographic location, and (2) To examine the point prevalence of bicycling-related injury and the association between these injury estimates and the same socio-demographic characteristics of Canadian youth. Methods ======= Data source ----------- The Health Behaviour in School-aged Children (HBSC) study is a cross-national survey conducted in collaboration with the World Health Organization. The 6^th^ cycle of the HBSC was undertaken in Canada in the 2009/10 school-year with 26,078 young Canadians, mostly aged 11 to 15, from 436 schools in 8 provinces and the 3 territories (all but Prince Edward Island and New Brunswick participated). The national sample was stratified by province/territory, type of school board (public vs. separate), urban--rural geographic status, school population size, and language of instruction (French vs. English) with standardized population weights generated to account for the over- or under-sampling in specific regions. Young people were not included in the survey if they were home schooled, attended private school, attended school on First Nations reserves, were incarcerated or did not provide informed consent (explicit or implicit, as per local school board protocol). The national study protocol was approved by the Queen's University General Research Ethics Board. Response rates were 11/13 (84.6%) at the province/territorial level, 436/765 (57.0%) at the level of schools, and 26,078/33,868 (77.0%) at the individual student level. Variables for analysis ---------------------- ### Bicycle and bicycle helmet Use Question \#86 of the survey asked: During the past 12 months, how often did you wear a helmet when you rode a bicycle? The response options were: I did not ride a bicycle; Never; Sometimes; Most of the time; and Always. Students who responded that they did not ride a bicycle were identified as non-riders. All other students were considered riders. Among riders, consistency in wearing a bike helmet (Never, Sometimes, Most of the time, and Always) was also reported. We collapsed the "sometimes" and "most of the time" categories into one category that we labeled *inconsistent* helmet use. ### Youth Sub-groups Demographic items were recorded for each student including sex (male/female), age (in years), school grade (\<8, 8--9, 10+), years in Canada (lifetime/born in Canada, immigrant \>5 years, immigrant ≤ 5 years); socio-economic status (self-perceived below average, average or above average) \[[@B24]\] and urban--rural status by population size of the students' school's census subdivision (rural = \< 1000; small town = 1000--29,999; medium urban centre ≥ 30,000-99,999 and large urban centre ≥100,000). These urban--rural groupings are based on Statistics Canada suggested approach and cut-offs \[[@B25]\]. The grade categories roughly divide the students by younger (elementary), middle (equivalent to junior high school ages) and older (high school levels). ### Injury Experiences with bicycling-related injuries were derived from a question that asked whether participants had one or more injuries during the past 12 months that required treatment by a doctor or nurse. If injured, the students were then asked to think of their one most serious injury and then, what they were doing when the injury occurred. One of the response options to this follow-up question was bicycling/cycling. In this way we were able to identify a specific subset of bicycling-related injuries. Due to the lack of specificity of the available items, we were not able to specify further the exact type of bicycling-related injury that occurred. Approach to analysis -------------------- Bicycle ridership and bicycle helmet use were profiled using conventional descriptive statistics for population subgroups defined by age, sex, school grade, socioeconomic status, number of years in Canada and urban--rural geographic location. Poisson regression analyses, using the SAS Procedure PROC GLIMMIX which considered the weighted and multi-level (clustered) nature of the HBSC sampling design, were conducted to relate age, sex, urban--rural geographic location, socioeconomic status and number of years in Canada to reports of bicycle-related injury. Relative risks and associated 95% confidence intervals (with inflation for clustering) were estimated. All data analyses were conducted using SAS (version 9.3). We did not analyze injury by helmet use because the specific focus of this project was to profile bicycling and bicycling-related injury outcomes by different population subgroups, not to determine associations between injuries and use of helmets. Results ======= Of the overall sample of 26,078 students, 74% (95% CI ±0.53) of them reported riding a bike in the last 12 months. A description of the overall study sample and bicycle ridership by sex, age, school grade, socioeconomic status, urban--rural geographic location and number of years in Canada is provided in Table [1](#T1){ref-type="table"}. Independent factors associated with bicycle ridership include being male, a younger student, more affluent, and from a small town. ###### **A description of the overall study sample and the bicycle ridership**\[[@B1]\] **Sub-group** **Overall sample** **Bicycle ridership** ------------------------- -------------------- ---------------------------- Sex:     Boys 12 815 (49.2) 10 085 (52.0) (row % 78.7) Girls 13 254 (50.8) 9322 (48.0) (row % 70.3) Age:     Mean (SD) 13.3 (1.6) 13.3 (1.6) Range in years 9 to 19 10 to 19 Grade level:     \< 8 10 370 (39.8) 8180 (42.2) (row % 78.9) 8-9 10 661 (40.9) 7708 (39.8) (row % 72.3) ≥ 10 5047 (19.3) 3521 (18.1) (row % 69.8) Socio-Economic status:     Above average 13 998 (56.9) 10 817 (58.0) (row % 77.3) Average 8276 (33.6) 6176 (33.1) (row % 74.6) Below average 2339 (9.5) 1652 (8.9) (row % 70.6) Urban--rural Location:     Large Urban Centre 8589 (32.9) 6161 (31.7) (row % 71.7) Medium Urban Centre 5739 (22.0) 4174 (21.5) (row % 72.7) Small Town 10 767 (41.3) 8329 (42.9) (row % 77.4) Rural 983 (3.8) 746 (3.8) (row % 75.9) Years in Canada:     Lifetime (born in Can.) 18 466 (71.5) 13 855 (72.0) (row % 75.0) Immigrant \> 5 yrs 6143 (23.8) 4642 (24.1) (row % 75.6) Immigrant ≤ 5 yrs 1212 (4.7) 750 (3.9) (row 61.9) ^1^. Totals in each cell may not add exactly to corresponding "n" as the number of responses varies slightly by survey question. Among the 19,410 students who reported that they rode bicycles, we examined patterns of bicycle helmet use. Overall, 43% (95%CI ±0.6%) of riders reported never wearing a helmet, 32% (95%CI ±0.6%) inconsistently wore a helmet and 26% (95%CI ±0.5%) reported always wearing a helmet while riding a bicycle in the past 12 months. Table [2](#T2){ref-type="table"} provides a breakdown of helmet use patterns by sex, age, school grade, socioeconomic status, urban--rural geographic location and years in Canada. Independent factors associated with always wearing a helmet include being a younger student, above average socioeconomically, and from a medium sized urban area. ###### **Helmet use patterns among bicyclists by sub-group**\[[@B2]\] **Sub-group** **Never wears a helmet, (count and row %)** **Inconsistently wears a helmet (count and row %)** **Always wears a helmet, (count and row %)** --------------------------- --------------------------------------------- ----------------------------------------------------- ---------------------------------------------- Sex:       Boys (n= 10085) 4394 (43.6) 3102 (30.8) 2589 (25.7) Girls (n= 9322) 3871 (41.5) 3019 (32.4) 2432 (26.1) Age:       Mean (SD) 13.9 (1.5) 13.1 (1.5) 12.5 (1.4) Range in years 10 to 18 10 to 19 10 to 17 Grade level:       \< 8 (n= 8178) 2139 (26.2) 2852 (34.9) 3187 (39.0) 8-9 (n= 7708) 3794 (49.2) 2462 (32.0) 1452 (18.8) ≥ 10 (n= 3522) 2332 (66.2) 809 (23.0) 381 (10.8) Socio-Economic status:       Above average (n=10818) 4378 (40.5) 3501 (32.4) 2939 (27.2) Average (n= 6176) 2785 (45.1) 1917 (31.0) 1474 (23.9) Below average (n= 1653) 811 (49.1) 467 (28.3) 375 (22.7) Urban--rural Location:       Large Urban (n= 6161) 2597 (42.2) 1980 (32.1) 1584 (25.7) Medium Urban (n= 4174) 1605 (38.5) 1345 (32.2) 1224 (29.3) Small Town (n= 8329) 3714 (44.6) 2539 (30.5) 2076 (24.9) Rural (n= 745) 349 (46.8) 257 (34.5) 139 (18.7) Years in Canada:       Born in Can. (n= 13,855) 5987 (43.2) 4255 (30.7) 3613 (26.1) Immigrant \> 5y (n= 4642) 1883 (40.6) 1544 (33.3) 1215 (26.2) Immigrant ≤ 5y (n= 750) 331 (44.1) 275 (36.7) 144 (19.2) ^2^. Totals in each cell may not add exactly to corresponding "n" as the number of responses varies slightly for each survey question. Among only the reported cyclists, we also examined the prevalence of bicycling-related injury and the association between the experience of an injury and a student's demographic characteristics (age, sex, school grade, socioeconomic status, number of years in Canada and urban--rural geographic location). Percentages of students in each subgroup reporting a medically treated bicycle-related injury ranged from a high of 6.9% (among boys or new immigrant students) to a low of 3.4% among girls. Table [3](#T3){ref-type="table"} provides the crude and adjusted relative risk for bicycling-related injury for levels of the different demographic characteristics. ###### Results of multiple logistic regression analysis examining direct effects of specific socio- demographic characteristics on risks for bicycling-related injury **Socio-demographic characteristic** **Crude relative risk** **Adjusted relative risk** -------------------------------------- ------------------------- ---------------------------- ------ ---------------- ---------- ------------------- **Sex**             Female 310 3.4 1.00   1.00   Male 671 6.9 2.13 \[1.85-2.46\] **2.02** **\[1.78-2.30\]** **Age**             ≥ 15 yrs 429 6.4 1.00   1.00   13-14 yrs 362 4.9 0.77 \[0.66-0.90\] **0.77** **\[0.66-0.90\]** \< 13 yrs 188 4.0 0.62 \[0.53-0.78\] **0.62** **\[0.51-0.76\]** **Socioeconomic Status**             Above Average 542 5.1 1.00       Average 314 5.2 1.01 \[0.88--1.16\]     Below Average 84 5.2 0.99 \[0.79-1.25\]     **Urban--rural Geographic Location**             Large urban 307 5.1 1.00       Medium urban 189 4.7 1.01 \[0.79-1.30\]     Small Town 442 5.5 1.10 \[0.89-1.37\]     Rural 42 5.9 1.21 \[0.79-1.85\]     **Years in Canada**             Born in Canada 654 4.9 1.00   1.00   Immigrant \> 5 yrs 267 5.9 1.23 \[1.06-1.42\] 1.14 \[0.98-1.32\] Immigrant ≤ 5 yrs 50 6.9 1.43 \[1.07-1.93\] **1.35** **\[1.00-1.82\]** ^3^. Estimated using multi-level procedures; students nested within schools, and SAS PROC GLIMMIX Procedure. ^4^. Model was adjusted for sex, age group, socio-economic status, urban--rural geographic status, and years in Canada. We have calculated adjusted relative risk estimates only for those variables that were included in the final regression model. Adjusting for all other demographic characteristics, boys had a 2.02-fold increase (95% CI: 1.78-2.30) in the relative risk of bicycle-related injury relative to girls, while new immigrants a 1.35-fold increase (95%CI: 1.00-1.82) in the relative risk compared to those students born in Canada. Being younger was protective against bicycling-related injury. Although not statistically significant, students from small towns and rural settings showed slightly higher risk for bicycle-related injury relative to their peers from more urban settings. We found no statistically significant relationship between risk of bicycling-related injury and socioeconomic status. Discussion ========== We have estimated that about three-quarters of all young Canadians between the ages of 11 and 15 years ride bicycles. Given what is known about the potential risks for injury while cycling \[[@B7]-[@B21]\] and the protective effects of bicycle helmets \[[@B7],[@B8],[@B10]-[@B14]\] one would ideally see high rates of bicycle helmet use in all Canadian youth. Unfortunately, our estimates indicate that currently only about a quarter of young Canadian cyclists wear a bicycle helmet all of the time. Indeed, 43% of the cyclists in our study reported that they *never* wore a helmet. These aggregate trends are concerning, and when bicycle helmet uptake patterns are examined within specific socio-demographic groups of riders there also appear to be distinct, and potentially troubling, patterns. It seems quite clear that helmet use is less frequent among older students. Two-thirds of bicycle riders (66.2%) in grade ten or higher reported never wearing a helmet, compared to 26.2% in students grade 8 or lower. In many Canadian jurisdictions, there are helmet laws requiring use in younger riders and there has been a gradual shift towards increased use in the younger population over time \[[@B9],[@B11],[@B13]\]. There is also a social trend towards non-compliance as children age and take responsibility for their own health behaviours \[[@B15]\]. Helmet use patterns also appear to vary by sex, SES and geographic location indicating a possible association between the non-use of helmets and male sex, lower SES and more rural locations. We do encourage caution in the interpretation of these unadjusted point prevalence estimates however. In addition to bicycle ridership and associated helmet use, the estimated relative risk of bicycling-related injury was also modeled for the 26,078 students in the study sample. Adjusting for all other factors, age, sex and immigration status were all independent predictors of the risk of injury. Indeed, new immigrants had a 1.35-fold increase in risk and male students a two-fold increase in risk for bicycling-related injuries than students in the comparison groups. While not strongly statistically significant, bicycling-related injury does appear to be more likely among youth in rural and small town areas as well. When taken in the context of existing literature, these findings are not entirely unexpected, but they do point to the importance of disaggregated analyses especially if research is to be used to informed public health intervention. Health inequities are differences in health that are judged to be unfair and remediable \[[@B26]-[@B28]\]. Over the past twenty years, there has been a notable increase in attention and scholarship around health equity and inequity in Canada and globally. In the past five years, there have been many key publications such as "Closing the Gap in a Generation: Health equity through action on the social determinants of health" \[[@B29]\], Integrating Social Determinants of Health and Health Equity Into Canadian Public Health Practice \[[@B30]\]; Reducing Health Inequalities- A challenge for our times \[[@B31]\], the Chief Public Health Officer's Report for Canada in 2008 \[[@B32]\] and Promoting Action on Equity Issues: A Knowledge-to-Action Handbook \[[@B33]\]. Striving for health equity means supporting all people to reach their full health potential and not be disadvantaged because of specific socio-demographic factors that influence their health and health behaviours. The fact that certain subgroups of populations are at greater apparent risks for bicycling-related injury is concerning, and particularly if these differences are unfair and remediable, they should be highlighted for attention and intervention. Canadian youth who are male, students who are older adolescents and students who are new immigrants to Canada are more at risk for bicycling-related injury. An aggregated analysis examining mean levels of risk, or overall health behaviours in young people as a homogenous group, would mask these differences. Of specific concern in this equity analysis are systemic factors that might exist as barriers to safe cycling for specific groups. While previous research indicates that older male students may, as a matter of choice, engage in higher risk activities that put them at greater risk for injury \[[@B16]\], this is not likely a similar pattern for new immigrant students. In this case, immigrant young people may be at greater risk as a result of transitions across different cycling environments and "safety cultures" for themselves and their families \[[@B34]\], or inadequate preparation or access to appropriate resources to support their safe cycling in Canada \[[@B35]\]. Previously, interventions to encourage bicycle helmets have tended towards whole population-level approaches including legislation, enforceable by law, requiring that all young people under the age of 18 years wear helmets. Two recent systematic reviews of bicycle helmet legislation for the reduction of bicycling-related head injury indicate that these approaches are useful \[[@B1],[@B2]\]. There have also been non-legislative interventions aimed at the general public or broad groups of young people \[[@B3]\] and interventions to increase helmet uptake among lower income Canadians, for example \[[@B17],[@B18],[@B23]\]. In our review of the literature, we did not find examples of interventions catering specifically to male students or the older adolescent and young adult population specifically. These kinds of targeted interventions may be warranted. Also, from a policy perspective, there is a need to understand more about the mechanisms underpinning differences in bicycling-related outcomes so that interventions can be appropriately designed and evaluated. For example, are boys more likely to sustain bicycling related injuries because of the risk-taking behaviours they undertake that are different from girls? Is it just that they spend more time cycling than female youth? Or is it that they are somehow differentially impacted by environmental or policy factors? Have certain immigrant students been socialized within different bicycling cultures than Canadian-born students? If so, how do these possible differences influence their injury risk or their uptake of protective measures such as helmets? Are bicycling-related injuries more common in young people with older, less expensive bicycles or newer, perhaps more expensive ones? Answering these kinds of questions can inform targeted interventions to address root causes of potential inequities. The most striking of the injury results from this study relates to the sex differences in estimated risk of bicycling-related injury (boys were 2.02 times more likely than girls to report an injury). Studies of gender equity, and the development of gender equity theory, tools and methodologies have a long history \[[@B36]-[@B38]\]. The Canadian government mandates sex and gender-based analysis (SGBA) and supports efforts across Ministries with policy frameworks and tools to address this form of inequity. This material could be more effectively engaged with in current public health and epidemiological research. For example, scholars of gender-based analyses have articulated the process of standardizing the assessment of health outcomes by gender \[[@B37]\] and this may be a helpful model for how disaggregation by various sub-groups could be realized more comprehensively across research and practice settings. In the spring of 2012, the Ministry of Health and Long-term Care of Ontario updated their Health Equity Impact Assessment Tool (HEIA) \[[@B39]\]. This is a good example of a government tool and concerted initiative to support health across populations. The HEIA is underpinned by equity theory and health impact assessment methodologies \[[@B40]-[@B42]\]. The aim of the tool is to support health equity and reduce avoidable health disparities between population groups. What we find helpful about this tool is the explicit list of population sub-groups that might be considered in a health equity analysis. Table [4](#T4){ref-type="table"} is an adaptation of this list. While the sub-groups that are necessary to consider may differ on a case-by-case basis, further disseminating this kind of sub-population breakdown may better inform disaggregated analyses and support health equity research. ###### Examples of population subgroups that can be considered in disaggregated analyses (Adapted from Ministry of Health and long-term care Ontario, Health Equity Impact Assessment tool, 2012) **Sub-population group** **Examples** ------------------------------------------------------ -------------------------------------------------------------------------------------------------------------------------------------------- Aboriginal peoples First Nations, Inuit, Métis Age-related groups Children, youth, seniors Disability communities Physical, deaf, deafened or hard of hearing, visual, intellectual/developmental, learning, mental illness, addictions/substance use, etc.) Ethno-racial communities Racial/racialized or cultural minorities, immigrants and refugees Francophone or other linguistic groups New immigrant francophones, deaf communities using sign language Homeless populations living on the street, marginally or under-housed Low income Unemployed, underemployed, single parents Religious/faith communities Muslim, Christian, Jewish Populations as defined by geographic characteristics Rural/remote, inner-urban, geographic or social isolation, under-serviced areas Sex/gender Male, female, women, men, trans, transsexual, transgendered, two-spirited Sexual orientation Heterosexual, lesbian, gay, bisexual Other Depending of the health issue or outcome of interest, other population subgroups may apply. This study had a number of strengths. These included the survey and sample itself. The survey has been administered in four year cycles over 24 years in Canada (the 2009--2010 response rate was 75%) and it has been developed and honed by a strong and experienced international team (for a complete list of HBSC publications please see: <http://www.hbsc.org/publications/>). With an overall sample of more than 26,000 young people, sub-population group sizes remained sufficiently large to support analyses. Injury prevention and the health and health equity of children and youth clear priorities in Canada today and so the study has an important focus. The study also has several limitations. As in other population health surveys, HBSC relies on accurate self-report data from children. Considerable efforts have been made over decades internationally to maximize validity and reliability of the survey items \[[@B24]\], however, some misclassification and non-response is inevitable especially for factors that are inherently difficult to quantify such as socioeconomic status. There is a possibility that non-responders varied significantly from responders and this could have biased our estimations of prevalence and, in reducing the size and heterogeneity of the sample, minimized our power to detect effects. In addition, we do not have any information about the frequency or context of bicycle usage or the specific types or frequency of bicycle-related injuries over the past 12 months. We also were not able to assess potential differences related to some of the HEIA population subgroups such as students with disabilities, students from different religious or faith communities, students from different racial or ethnic groups and students with different sexual orientations. Given our growing understanding of the social determinants of health equity \[[@B29],[@B31],[@B42]\], these kinds of analyses may have been helpful. In summary, bicycling is a popular activity among Canadian youth and the health behaviours and possible risks associated with bicycling are not equally distributed across the Canadian youth population. More than 5% of young cyclists will experience a medically-treated bicycle-related injury each year in Canada. Injuries are more than twice as common among male students than female, and a third more common among new immigrants than non-immigrants. These differences are concerning. This study highlights the importance of considering disaggregated analyses when studying health behaviours, health outcomes and health equity in Canadian youth. These kinds of analyses can inform prevention interventions that, for bicycling-related injury for example, could be especially sensitive to gender, age, and socioeconomic differences. Beyond bicycles, there is evident value across health and health behavior topics, in analyzing and reporting data that are disaggregated by specific population subgroups. These kinds of analyses may help point to troubling disparities that may represent underlying, systemic inequities requiring targeted intervention and our continued public health efforts. Competing interests =================== The authors declare that they have no competing interests. Authors' contributions ====================== CD led the overall conduct of the study and preparation of the manuscript. MT, PW, WT, and SM represented the funders and contributed to the design of the study, the interpretation of findings and the review and editing of the final manuscript. WP led the analysis and made important contributions to the interpretation of findings and the writing of the final manuscript. All authors read and approved the final manuscript. Acknowledgements ================ HBSC is coordinated internationally by Dr. Candace Currie, University of St. Andrews, Scotland. HBSC in Canada is directed by Dr. John Freeman and Dr. William Pickett, and coordinated by Mr. Matthew King, Queen's University. In 2010, data collection efforts were supported by national and territorial partners from the Joint Consortium for School Health. We thank Dr. Ian Janssen and Mr. Andrei Rosu for collection of geographic information integral to this analysis. Financial support for this study was provided by the Public Health Agency of Canada and Health Canada as well as operating grants from the Canadian Institutes of Health Research and Heart and Stroke Foundation (CIHR Grant MOP 97962; CIHR/HSF Grant PCR 101415), the CIHR Team in Child and Youth Injury Prevention and support from the Injury Section, Health Surveillance and Epidemiology Division of the Public Health Agency of Canada (Contract 4500267124). CD was supported by an Emerging Researcher Award from the Population Health Improvement Research Network of Ontario. Alyssa Parpia helped format this manuscript for publication.

Introduction {#sec1_1} ============ Stroke is the fifth leading cause of mortality in the United States. Nearly 90% of all stroke cases are ischemic in nature \[[@B1]\]. Rapid assessment of patient eligibility for intravenous recombinant tissue plasminogen activator (rt-PA) is critical in these cases to re-establish blood flow to the ischemic area \[[@B2], [@B3], [@B4]\]. The use of warfarin by itself is not an absolute contraindication to rt-PA administration; though due to increased bleeding risk, an international normalized ratio (INR) of \>1.7 is a contraindication \[[@B5]\]. Because of this, tests of coagulation are included in the acute workup of patients on anticoagulation who present with new neurologic deficits \[[@B6]\]. Another reason to obtain an INR on admission in these patients may be its ability to help determine the likelihood of ischemia. There is little published data regarding the incidence of ischemic stroke in patients already on warfarin with varying INR cutoffs, or the predictive value of INR in informing the likelihood of ischemia versus a stroke mimic. For the majority of patients on warfarin, available guidelines recommend an INR goal of 2--3 \[[@B7]\]. Unfortunately, patients are often sub-therapeutic given the difficulty in controlling INR due to medication and food interactions \[[@B8], [@B9]\]. For many indications, such as atrial fibrillation or a hypercoagulable state, a sub-therapeutic INR could increase risk for an ischemic stroke \[[@B10]\]. The aim of the present study is to determine the likelihood of ischemia (stroke or transient ischemic attack \[TIA\]) in patients on warfarin who present with stroke-like symptoms, based on their INR result. Materials and Methods {#sec1_2} ===================== Patient Selection {#sec2_1} ----------------- A retrospective analysis was performed of all patients presenting to our urban, academic stroke center with symptoms suggestive of acute stroke within the last 6 h, who were also on warfarin, between January 1, 2013, and December 31, 2014. The study was approved by the Institutional Review Board of the Johns Hopkins Health System. Eligible patients included adults (age ≥18 years), with warfarin documented as home medication, who had an INR obtained at presentation, and were included in our Brain Attack Registry. This registry is maintained by our Stroke Program to track patients presenting to the Emergency Department with symptoms concerning for acute stroke. Patients were excluded if they failed to meet the above criteria, had lack of further follow-up information regarding diagnosis, or were on warfarin because of mechanical valve (different INR goal: 2.5--3.5). Patients were evaluated by a neurologist as part of the "brain attack (BAT)" or acute stroke code. They were categorized by final diagnosis using history, examination, laboratory data, and neuroimaging (MRI and non-contrast head CT) into ischemic (ischemic stroke or TIA) versus non-ischemic (hemorrhagic stroke or stroke mimic \[other determined diagnosis\]). In addition to demographic information, admission INR, and indication for warfarin therapy; baseline laboratory results, stroke characteristics when applicable (Trial of Org 10172 in Acute Stroke Treatment \[TOAST\] criteria), thrombolytic administration, and outcome (discharge modified Rankin scale score \[mRS\]) data were collected. The primary objective was to determine the association between initial INR and ischemia in patients on warfarin presenting with acute neurologic deficits. The secondary objective was to compare outcomes in patients with ischemic stroke who presented with a therapeutic versus sub-therapeutic INR. We hypothesized that those with a therapeutic INR would be less likely to have ischemia than individuals who were sub-therapeutic, and that when they did present with strokes, they would be less severe. Analysis {#sec2_2} -------- Statistical analyses were performed using Stata® (version 13·2, College Park, TX, USA). Continuous and normally distributed variables were compared using Student *t* tests. Categorical and nominal variables were calculated using χ^2^ tests. A *p* value of \<0.05 was considered to be statistically significant. Multivariable regression was used to determine the odds ratio for any variable significant in univariate analysis and those predetermined to be potentially clinically significant (age, race, sex). A subsequent sensitivity and specificity analysis was performed to evaluate the relationship between INR and ischemic risk. Positive and negative predictive values were determined for multiple INR cut-points. In a secondary analysis, patients presenting with hemorrhage were removed from the "non-ischemic" group and analyses were repeated. Results {#sec1_3} ======= Patients and Baseline Characteristics {#sec2_3} ------------------------------------- One-hundred eighteen patients met inclusion/exclusion criteria for this study (Fig. [1](#F1){ref-type="fig"}). Two additional patients were excluded due to indication for anticoagulation (mechanical valve). The baseline characteristics of the 116 patients are shown in Table [1](#T1){ref-type="table"}. The median age was 77 years; 57% were women; 74% were white. Indications for warfarin therapy could be divided into two groups: arterial (atrial fibrillation \[*n* = 86\] and antiphospholipid antibody syndrome \[*n* = 2\]) and venous (deep vein thrombosis and pulmonary embolism \[*n* = 28\]). Thirty-four patients had ischemic stroke; 12 TIA; 24 hemorrhagic strokes; and 46 stroke mimics. Four patients received thrombolytic therapy (rt-PA) and 1 patient received intra-arterial intervention. Common stroke mimics included: seizure (*n* = 6), infection (*n* = 6), dementia/delirium (*n* = 3), hypoglycemia (*n* = 3), recrudescence (*n* = 4), drug toxicity (*n* = 4), systemic impairment (*n* = 4), and functional deficit (*n* = 2). Initial INR and Ischemic Stroke {#sec2_4} ------------------------------- Univariate analysis showed a statistically significant difference in the initial INR for ischemic versus non-ischemic patients (1.7 vs. 2.8, *p* \< 0.001). The initial INR values ranged from 1 to 3.4 for ischemic patients versus 1 to 6.4 for those diagnosed with hemorrhagic strokes. Twenty-five percent of hemorrhages were supra-therapeutic and the mean INR was 2.7. Individuals diagnosed with ischemic stroke or TIA were more likely to be on warfarin therapy because of arterial indications, namely atrial fibrillation, compared to non-ischemic patients (87 vs. 67%, *p* = 0.024). The cause of stroke was most often cardioembolic (76%). Aside from a significant difference in platelet number between the ischemic and non-ische mic groups (252 vs. 215, *p* = 0.024), there were no other differences seen in patient char­acteristics, including initial National Institute of Health Stroke Scale (NIHSS), CHADS~2~, CHA~2~DS~2~VASc, and serum creatinine (Table [1](#T1){ref-type="table"}). Results were unchanged when patients with hemorrhage (*n* = 24) were removed from analysis. A multivariable regression was performed with 5 independent variables: age, sex, race, initial INR, and indication for warfarin. Initial INR (OR per 1.0-point increase 0.30; 95% confidence interval \[CI\] 0.17--0.55) and arterial indication for warfarin (OR 4.11; 95% CI 1.33--12.67) remained statistically significant (Table [2](#T2){ref-type="table"}). Table [3](#T3){ref-type="table"} displays the sensitivity and specificity analysis for INR values. When INR was used to predict ischemia, the area under the curve was high at 0.77 indicating its ability to correctly classify patients. As the INR value increased, the number of patients incorrectly identified as "non-ischemic" significantly decreased (increased specificity). No patient had ischemia with an INR value of 3.6 or greater (specificity 100%). Predictive values were calculated to determine the likelihood of a patient with a therapeutic INR at ≥2 and ≥2.5 to have a mimic or hemorrhage rather than ischemia. The predictive value for both an INR of ≥2 (sensitivity of 69% and specificity of 67%) and an INR of ≥2.5 (sensitivity of 47% and specificity of 80%) was 79% for a non-ischemic etiology. Initial INR and Clinical Outcome in Stroke Patients {#sec2_5} --------------------------------------------------- The effect of INR result on clinical outcome was examined using discharge mRS. A mRS of 0--2 was defined as good outcome and a mRS 0--1 was considered a great outcome. There was a lack of an effect of INR on mRS \<3 (OR 0.9; 95% CI 0.3--2.5) or mRS \<2 (OR 4.7; 95% CI 0.6--35.3) within the ischemic group. In the entire patient population, initial INR also did not affect mRS \<3 (OR 0.5, 95% CI 0.2--1.1) or mRS \<2 (OR 1.1, 95% CI 0.5--2.5). Discussion {#sec1_4} ========== Our results indicate that for patients on warfarin presenting with a potential stroke, indication for anticoagulation and INR values are associated with cerebral ischemia. Within our cohort, patients diagnosed with ischemic stroke or TIA were more likely to present with a sub-therapeutic INR than those with non-ischemic etiologies in both univariate testing and multivariable regression analyses. Arterial indications for warfarin, mainly atrial fibrillation, further increased the likelihood for ischemic stroke or TIA. This is consistent with prior research. Several studies have shown that a sub-therapeutic INR and atrial fibrillation increase the incidence of ischemic stroke in patients on warfarin with concurrent acute neurologic deficits \[[@B11], [@B12], [@B13]\]. In addition to showing an association, our data suggests that admission INR can be used to reliably predict the likelihood of ischemia. The predictive value of INR for ischemic stroke in patients on warfarin who present with acute neurologic symptoms has not been previously well established in literature. The sensitivity and specificity results of this study elucidate the value of INR in this group. When diagnostic imaging does not reveal evidence of hemorrhage or other clear etiology, the INR and indication for anticoagulation may be used in conjunction with pre-test probability (e.g., vascular risk factors) to direct immediate further management. There should be high suspicion for ischemia in individuals with atrial fibrillation who present with a sub-therapeutic INR. Alternatively, the negative predictive value of INR emphasizes the low likelihood of ischemia in patients on warfarin with therapeutic INR values. In our cohort, no ischemia occurred in patients with an INR ≥3.6. Interestingly, patients who presented with ischemia had a small but statistically significant elevation in platelets compared to those who were not ischemic. Elevated platelets may represent an acute phase reactant \[[@B14], [@B15]\]. Advanced age and higher NIHSS scores were not associated with ischemia. This may illustrate that older individuals are at higher risk not only for ischemic stroke but also for other causes of acute neurologic symptoms such as urinary tract infection \[[@B16]\]. While assessing for suspected stroke, physicians\' attention should also be given to identifying any clinical or laboratory signs of stroke mimics \[[@B15]\]. We also evaluated the association between INR and patient outcome in our stroke patients using mRS at discharge. We hypothesized that patients with a therapeutic INR would have better outcomes than those with a non-therapeutic INR because emboli may break into smaller pieces, leading to smaller infarcts and less severe strokes in those on anticoagulation. Wakita et al. \[[@B17]\] showed that individuals who are anticoagulated do tend to have smaller infarcts. We found no such association despite that, as expected, the majority of our population had cardioembolic strokes due to atrial fibrillation \[[@B18]\]. There are several possible explanations including our relatively small sample and the numerous medical comorbidities of our population (Table [1](#T1){ref-type="table"}) that may have negatively impacted the outcome. It is also possible that a larger difference in mRS would have been observed 90 days after the stroke, allowing time for divergence of the recovery curves for varying stroke severity. Many previous trials found that therapeutic INR was associated with more favorable stroke outcomes in patients with cardioembolic stroke types \[[@B19], [@B20]\], though some failed to detect such a correlation \[[@B21]\]. A larger prospective trial including infarct size and longer-term follow-up is needed. Our study is not without limitations. The patient population was from a single stroke center - with a relatively small sample size and the majority of patients above the age of 75 years, white, and at high risk for ischemic stroke give their numerous vascular risk factors. The relatively small sample size may have limited our ability to find statistically significant results, and our sensitivities and specificities were based on the prevalence of stroke within our population and therefore our results may not be generalizable to all other populations. In addition, given the retrospective nature, we relied on physician notes and test results within the electronic medical record to determine final diagnosis. Fortunately, the underlying etiology was clear in the majority of cases and any questions were reviewed by a stroke neurologist to ensure that the cases most likely represented mimics rather than ischemic strokes. Finally, the INR can be influenced by numerous factors (foods, medications) and the value on presentation represents only a single data point rather than coagulability over time. A patient may have been sub-therapeutic for several days at home but present with a single therapeutic INR. In spite of this, we feel that our data support that admission INR is a fairly good indicator of likelihood of ischemia and may have important clinical applications. Conclusions {#sec1_5} =========== This study illustrates the usefulness of both INR values and the indication for anticoagulation, specifically atrial fibrillation, in determining the likelihood of ischemic stroke in patients on anticoagulation presenting to the Emergency Department with an acute onset of neurologic symptoms. A sub-therapeutic INR in a patient with atrial fibrillation should significantly increase the suspicion for ischemia; while a supra-therapeutic INR, or lack of atrial fibrillation, is more reassuring that symptoms are more likely due to a mimic. While it is not our intention for a laboratory test to replace physician acumen, it may help to more accurately risk stratify patients and determine the need and urgency of further neurologic imaging and workup. Disclosure Statement {#sec1_6} ==================== The authors declare no conflicts of interest. {#F1} ###### Patient characteristics: predictors of stroke in patients on warfarin Total (*n* = 116) Ischemia (*n* = 46) Non-ischemia (*n* = 70) *p* value ---------------------------------------------------------------------------- ------------------- --------------------- ------------------------- ------------- *Demographic information* Median age (IQR), years 77 (67--84) 77 (67--82) 77 (67--84) 0.830 Race, *n* (%) 0.716  White 86 (74) 35 (76) 51 (73) \-  Black 23 (20) 9 (20) 14 (20) \-  Asian 2 (2) 0 (0) 2 (3) \- Women, *n* (%) 66 (57) 28 (61) 38 (54) 0.484 Smoking, *n* (%) 26 (23) 9 (20) 17 (25) 0.497 Alcohol, *n* (%) 26 (23) 6 (13) 20 (29) **0.041** *Present medical history, n (%)* Coronary artery disease 74 (64) 30 (65) 44 (64) 0.874 History of stroke 54 (47) 21 (46) 33 (47) 0.875 History of TIA 36 (31) 14 (30) 22 (31) 0.910 Chronic heart failure 38 (33) 15 (33) 23 (33) 0.978 Hypertension 110 (95) 44 (96) 66 (94) 0.745 Diabetes mellitus 46 (40) 17 (37) 29 (41) 0.630 *Warfarin information* Mean initial INR (SD) 2.4 (1.5) 1.7 (0.7) 2.8 (1.7) **\<0.001** Arterial indication for warfarin, *n* (%)[^1^](#T1F1){ref-type="table-fn"} 88 (76) 40 (87) 48 (67) **0.024** *Laboratory values* Mean platelets (SD), 10^3^/mm^3^ 230 (85) 252 (99) 215 (71) **0.021** Mean LDL (SD), mg/dL 75 (30) 78 (33) 70 (22) 0.309 Mean SBP (SD), mm Hg 154 (28) 155 (27) 152 (29) 0.533 Mean DBP (SD), mm Hg 84 (16) 85 (15) 82 (17) 0.337 Mean glucose (SD), mg/dL 138 (52) 132 (41) 143 (58) 0.254 Mean albumin (SD), g/dL 3.5 (0.7) 3.4 (0.6) 3.5 (0.7) 0.377 Mean HbA~1c~ (SD) 6.5 (1.3) 6.4 (1.1) 6.8 (1.5) 0.277 Mean WBC (SD), 10^3^/mm^3^ 9.1 (3.9) 8.8 (2.9) 9.3 (4.4) 0.488 Median CHADS~2~ (IQR) 3 (2--4) 3 (2--4) 3 (2--4) 0.940 Median CHA~2~DS~2~VASc (IQR) 4 (3--5) 3 (3--5) 4 (3--5) 0.670 Median NIHSS (IQR) 5 (2--11) 4.5 (2--11) 5 (2--14) 0.180 Mean serum creatinine (SD), mg/dL 1.3 (0.6) 1.3 (0.6) 1.2 (0.6) 0.752 *Medication use, n (%)* rt-PA 4 (4) 4 (9) 0 **0.010** Antiplatelet 43 (37) 14 (30) 29 (41) 0.230 Statin 65 (56) 22 (48) 43 (61) 0.149 *Final diagnosis* Ischemic stroke, *n* (%) 34 (29) 34 (74) 0 \- TIA, *n* (%) 12 (10) 12 (26) 0 \- Hemorrhagic stroke, *n* (%) 24 (21) 0 24 (34) \- Mimics, *n* (%) 46 (40) 0 46 (66) \- TOAST criteria, *n* 34 34 \- \-  Cardioembolism 26 26 \- \-  Lacune 3 3 \- \-  Large vessel 1 1 \- \-  Undetermined 4 4 \- \- DBP, diastolic blood pressure; INR, international normalized ratio; LDL, low-density lipoprotein; rt-PA, recombinant tissue plasminogen activator; SBP, systolic blood pressure; TIA, transient ischemic attack; WBC, white blood cells. Arterial indications include atrial fibrillation and antiphospholipid syndrome; venous indications include deep vein thrombosis and pulmonary embolism. ###### Multivariable analysis of likelihood of ischemia Odds ratio 95% confidence interval --------------------------------- ------------ ------------------------- Entire cohort (*n* = 116)  Age 1.03 0.99--1.08  Race 1.18 0.80--1.73  Sex 1.03 0.41--2.57  Initial INR 0.30 0.17--0.55  Indication for warfarin 4.11 1.33--12.67 Hemorrhages excluded (*n* = 92)  Age 1.04 0.99--1.09  Race 0.97 0.58--1.61  Sex 0.84 0.29--2.42  Initial INR 0.32 0.17--0.59  Indication for warfarin 4.91 1.46--16.57 INR; international normalized ratio. ###### Sensitivity and specificity analysis of INR result INR result Sensitivity, % Specificity, % ------------ ---------------- ---------------- ≥1 100 0 ≥1.1 4.35 4.35 ≥1.2 91.43 13.04 ≥1.3 85.71 34.78 ≥1.4 78.57 45.65 ≥1.5 78.57 52.17 ≥1.6 75.71 58.70 ≥1.7 75.71 63.04 ≥1.8 71.43 65.22 ≥1.9 68.57 67.39 ≥2.1 64.29 73.91 ≥2.2 60.00 73.91 ≥2.3 51.43 78.26 ≥2.4 48.57 80.43 ≥2.5 47.14 80.43 ≥2.6 42.86 86.96 ≥2.7 38.57 86.96 ≥2.8 37.14 86.96 ≥2.9 32.86 86.96 ≥3 31.43 89.13 ≥3.1 30.00 97.83 ≥3.2 28.57 97.83 ≥3.3 27.14 97.83 ≥3.4 24.29 97.83 ≥3.6 21.43 100 ≥3.7 20.00 100 ≥4.3 15.71 100 ≥4.6 12.86 100 ≥5.3 11.43 100 ≥5.4 10.00 100 ≥5.6 8.57 100 ≥6.2 5.71 100 ≥6.4 4.29 100 ≥9.1 2.86 100 ≥9.1 0.00 100

In the original article, there was a mistake in Figure 1 and Figure 2 as published. The coefficients for the effects of the experimental intervention condition on thought related test anxiety were mistakenly reported as negative even though the coefficients for these effects are positive. The corrected Figure [1](#F1){ref-type="fig"} and Figure [2](#F2){ref-type="fig"} appear below. {#F1} {#F2} Because of the error mentioned above, a correction has been made to the **Results, Model 1: Combined Analyses**: "The fit statistics for model 1 were as follows, χ^2^(3) = 1.26, *p* = 0.739; CFI = 1.00; RMSEA = 0; SRMR = 0.03. Allover, a significant degree of the variance of thought related test anxiety measured at time 2 (*R*^2^ = 0.59, *SE* = 0.05, *p* \< 0.001) and time 3 (*R*^2^ = 0.57, *SE* = 0.06, *p* \< 0.001) was explained. All path coefficients are depicted in Figure [1](#F1){ref-type="fig"}. Trait test anxiety proved to be a significant positive predictor of thought related test anxiety at time 2 (β = 0.34, *SE* = 0.08, *p* \< 0.001) and time 3 (β = 0.34, *SE* = 0.08, *p* \< 0.001). The same was true for exams\' personal value regarding thought related test anxiety measured at time 2 (β = 0.09, *SE* = 0.05, *p* = 0.037), but not regarding thought related test anxiety measured at time 3 (β = 0.04, *SE* = 0.05, *p* = 0.215). As expected, we also found a significant effect of the dummy variable d1 (IBSR vs. control groups) on thought related test anxiety measured at time 2 (β = 0.19, *SE* = 0.05, *p* \< 0.001) and at time 3 (β = 0.15, *SE* = 0.06, *p* \< 0.001). The direction of the coefficients indicates that exploration of an individual worry thought with the IBSR technique is effective in reducing thought related test anxiety compared to reflecting on or distracting oneself from a worry thought." A correction has also been made to the **Results, Model 2: Differential Analyses**: "The fit statistics for model 2 were as follows, χ^2^(6) = 1.58, *p* = 0.954; CFI = 1.00; RMSEA = 0; SRMR = 0.02. Model 2 explained a significant degree of the variance of thought related test anxiety measured at time 2 (*R*^2^ = 0.63, *SE* = 0.04, *p* \< 0.001) and time 3 (*R*^2^ = 0.58, *SE* = 0.06, *p* \< 0.001). All path coefficients are depicted in Figure [2](#F2){ref-type="fig"}. Again, trait test anxiety proved to be a significant positive predictor of thought related test anxiety at time 2 (β = 0.34, *SE* = 0.08, *p* \< 0.001) and time 3 (β = 0.34, *SE* = 0.08, *p* \< 0.001). The same was true for exams\' personal value regarding thought related test anxiety measured at time 2 (β = 0.08, *SE* = 0.05, *p* = 0.046), but not regarding thought related test anxiety measured at time 3 (β = 0.04, *SE* = 0.06, *p* = 0.245). Results also revealed a significant effect of the dummy variable d2a (IBSR vs. reflection) on thought related test anxiety measured at time 2 (β = 0.29, *SE* = 0.05, *p* \< 0.001) and time 3 (β = 0.21, *SE* = 0.06, *p* \< 0.001). The effect of the dummy variable d2b (IBSR vs. distraction) on thought related test anxiety measured at time 2 (β = 0.08, *SE* = 0.06, *p* = 0.084) and time 3 (β = 0.09, *SE* = 0.07, *p* = 0.088) was statistically non-significant. These results indicate that exploration of an individual worry thought with the IBSR technique is effective in reducing thought related test anxiety in comparison to reflecting on a worry thought. The β-values also indicated that IBSR was associated with lower thought related test anxiety than distraction, however, this effects was statistically non-significant." The authors apologize for these errors and state that this does not change the scientific conclusions of the article in any way. The original article has been updated. Conflict of Interest Statement ============================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. [^1]: Edited and reviewed by: Jesus de la Fuente, University of Navarra, Spain [^2]: This article was submitted to Educational Psychology, a section of the journal Frontiers in Psychology

Introduction ============ Nowadays, online crowdsourcing is one of the widely used methods for conducting behavioral experiments with large numbers of people. In comparison to the traditional way of performing offline experiments, crowdsourcing on the Internet enables researchers to collect massive amounts of data and requires less effort to recruit participants, set up the experimental environments, and run the experiments ([@B24]). In addition, it has been shown that the quality of experimental results from crowdsourcing is almost the same as that from offline-recruited participants ([@B4]). One of the important issues in conducting crowdsourcing-based experiments is ensuring the quality of experimental data. It is crucial to determine a proper way to motivate people to participate in the experiment, which includes financial reward, altruism, enjoyment, reputation, and implicit work ([@B18]). For example, Amazon Mechanical Turk (MTurk) is a financial reward-based crowdsourcing marketplace in which one can post small tasks, called Human Intelligence Tasks (HITs), and people who fulfill those tasks get monetary reward from the task provider. reCAPTCHA is an example of implicit crowdsourcing work in which users are asked to type both scanned and distorted words in order to prevent automated programs that perform abnormal activities such as registering numerous user accounts ([@B23]). Enjoyment-based applications inspired by gaming also exist in many fields, such as music annotation ([@B2]), image labeling ([@B22]), and protein structure prediction ([@B6]). Many crowdsourcing-based research studies have relied on financial reward to motivate participants due to its simplicity and similarity to traditional offline experiments. MTurk -- which offers an efficient interface to induce mass participation and give financial reward directly to participants, called workers, after completion of the experiment's tasks -- has been widely used. It has been shown that MTurk-based experiments have not only almost the same reliability of results as traditional offline experiments but also higher immediacy for finishing given tasks ([@B3]; [@B10]). In addition, the characteristics and efficiency of financial reward-based experiments have been analyzed extensively in the literature. For example, [@B3] evaluated the efficiency of MTurk in terms of quality of experimental results with respect to the amount of payment. [@B17] showed that enlightening workers is more important to obtaining high-quality results than raising payment rates. As alternative crowdsourcing media, online social networks have often been considered in recent days. Facebook, one of the most popular social network services with more than 1 billion daily active users around the world ([@B7]), has been used in several experiments, such as those testing quality of experience (QoE) for videos ([@B8]) and photo album summarization ([@B21]). Facebook allows researchers to retrieve users' personal information and interests, so it is easy to recruit participants who have specific interests or to analyze experimental results with respect to demographic characteristics. Moreover, it is not required to pay participants directly, so motivations other than monetary reward can be chosen to prevent issues arising from the discrepancy between the incentives and experimental purpose. Despite these advantages, research studies analyzing the characteristics and efficiency of social networks, which are pivotal for the successful design of crowdsourcing experiments, are rarely found. As with crowdsourcing experiments based on financial reward, it is necessary to analyze the properties of social network-based crowdsourcing to help researchers plan their own experiments tactically to maximize both the efficiency and quality of data. Recently, we conducted a crowdsourcing-based experiment called "EvoTunes" ([@B5]). It aimed at collecting data regarding user satisfaction with given music playlists that could be used for automatic playlist recommendations. Each participant listened to two specific songs in a row whose order was determined by a recommendation algorithm and provided satisfaction ratings. This procedure was repeated with different song pairs for many participants. When a sufficient amount of feedback data was gathered, the recommendation algorithm calculated the probability of satisfaction for each song-to-song transition to generate new playlists expected to have higher levels of satisfaction. Optimization of the recommendation process required a large amount of experimental data on users' satisfaction with song transitions. Therefore, it was important to design an efficient and reliable way of conducting a crowdsourcing-based experiment. For this, MTurk was the first candidate as the most popular crowdsourcing medium. However, two problems arose: first, the gathered data could have been largely compromised due to unfaithful responses or cheating if workers pursued only monetary reward, and it was extremely difficult to detect such responses due to the lack of ground-truth data. Second, the environment of music listening through MTurk was significantly different from real music listening situations; thus, even if the gathered data were reliable, they might not have been suitable for playlist generation. To address these issues, we decided to create our own web-based experiment interface. Participants listened to music clips suggested by the recommendation algorithm through the simple music player shown in **Figure [1](#F1){ref-type="fig"}**, which is similar to that of common music streaming services. If two particular songs were played until the end, the listener was considered satisfied with the playlist; however, if the listener pressed the "skip" button to listen to another song, the listener was considered dissatisfied. This interface was able to provide a natural music listening experience that minimized the feeling of being involved in an experiment. {#F1} Since the experiment required voluntary participation, we needed an appropriate way to recruit participants interested in music streaming services. In the end, we decided to exploit an online social network, Facebook, which provides a fully customizable means of public relations, including advertisement and promotional pages^[1](#fn01){ref-type="fn"}^. We recruited a sufficient number of participants over 30 days through Facebook and ran our experiment successfully. Moreover, we collected a massive amount of data regarding users' behaviors on both Facebook and the experiment interface. Based on the collected data, in this paper, we investigate the efficiency and usability of online social networks to attract participants for crowdsourcing experiments, especially focusing on the following research questions. 1. How can participants be effectively attracted for crowdsourcing experiments in online social networks? Since monetary reward are not given to participants, ways of attracting participants is a critical issue in social network-based crowdsourcing experiments. For this, advertisement, which is a unique tool in social networks such as Facebook, can be exploited. Previous studies including [@B1] concluded that it is hard to attract participants via advertisement of online social networks due to low response rates. However, they may be still exploitable if a sufficiently large number of participants can be hired at a reasonable cost. Moreover, different types of advertisement have not been compared in existing studies. In this paper, different tools for advertisement on Facebook are tested and compared in terms of effects and efficiency. 1. How can particular groups of users be appropriately targeted in online social networks? Users' behaviors within a social network-based crowdsourcing experiment may vary with the users' demographic characteristics, such as gender and age group. There have been studies examining gender-dependence of participation in crowdsourcing (e.g., [@B20]), but other characteristics such as age have been rarely studied. Understanding such variability is helpful for designing a recruiting strategy to effectively target particular groups of users suited for the goal of the experiment. We try to answer this question by examining the demographic characteristics of the participants in our Facebook-based experiment. 1. Do online social networks offer reliability and efficiency comparable to other recruitment methods? To take advantage of online social networks to attract participants for reliable crowdsourcing experiments, their efficiency must first be proved. As aforementioned, this has not been studied well in literature. We analyze how efficient it is to use online social networks for multimedia-involved crowdsourcing experiments in comparison to other media such as MTurk. The methods used to conduct our experiment are described in Section "Materials and Methods." The analysis of the experimental results in various aspects is shown in "Results." In "Discussion" discusses the outcomes, compares our approach with existing crowdsourcing methods, discusses the advantages and disadvantages of online social networks, makes suggestions for successful recruitment using social networks, and finally presents our conclusions. Materials and Methods {#s1} ===================== Procedure --------- **Figure [2](#F2){ref-type="fig"}** shows an overview of how participants were attracted in our experiment. The two components that we exploited were advertisements of our experiment interface using the Facebook advertisement system and articles posted on the Facebook page for the interface. The advertisements guided users to visit our experiment interface directly or to click the "Like" button to subscribe to our Facebook page. For those who subscribed to the page, new articles on the page were shown on their main homepage that encouraged them to participate in the experiment repeatedly. The study was carried out in compliance with the ethical recommendations in the Declaration of Helsinki, and informed consent was provided on the experiment interface. In addition, the advertisement data were aggregated on an anonymous basis to abide by the Facebook advertising policies. {#F2} ### Facebook Advertisement The Facebook advertisement system allows advertisers to choose various options, such as targeting users who have specific interests and locating an appropriate position to display the advertisement. Once the options are configured, Facebook randomly selects users within the targeted user group to show advertisements. To maximize the efficiency of our advertisement, we targeted users who were over the age of 18, had expressed interest in music-related topics, and lived in English-speaking countries including the United States, United Kingdom, Canada, Australia, and the Philippines. Advertisers need to set a daily budget and choose one of the two charging options: charge per 1,000 exposures or charge per click. Because inducing users to click the advertisement was more important than exposing the advertisement in our case, we chose the click-based charging option. In this option, the maximum cost per click (CPC) of the advertisement can be set. Then, Facebook automatically decides an optimized CPC based on the maximum costs of other advertisements having similar targets. During our experiment, we often altered the daily budget of our advertisement so that the number of exposures changed accordingly. Here, we set only the total budget for both advertisement types; the budget for each advertisement type was not specified. Although Facebook provides detailed targeting options, such as gender, education level, and so on, we did not set any other particular targeting configurations except for those mentioned above in order to prevent any bias in the experimental results and enable general conclusions to be drawn from the results. For example, we set the age range as over 18, which is the default targeting option. In addition, we used both types of Facebook advertisements appearing in the desktop version of Facebook, as shown in **Figure [3](#F3){ref-type="fig"}**. {#F3} #### Newsfeed advertisement A newsfeed advertisement appears directly on a user's main homepage of Facebook, called "News Feed." It is marked as "Sponsored" to be distinguished from posts published by friends. Users could click the inner box of the advertisement containing a thumbnail image, title, description, and link to go to our experiment interface or click the "Like" button at the bottom to subscribe to our Facebook page. The number of "Likes" for our Facebook page was shown next to the "Like" button. We activated this type of advertisement for 30 days with adjustment of the daily cost, *N = 30, M = \$7.45, SD = \$8.45*. #### Right-column advertisement A right-column advertisement appears in the "Sponsored" section on the right side of the Facebook site. Unlike the newsfeed advertisement, this type of advertisement did not contain the "Like" button for our Facebook page. Users could click any part of the advertisement to launch our experiment interface. We activated this type of advertisement only for 7 days due to its inefficiency (see Differences by Advertisement Type) with adjustment of the daily cost, *N = 7, M = \$5.18, SD = \$5.73*. ### Facebook Page Facebook allows users to create pages for promoting various goods and services (e.g., a website, entertainment program, shoe brand). Any Facebook user can create a page, and there are no fees. Posts can be written on such a page, which are shown on the "News Feed" of the users who have pressed the "Like" button of the page. We created a page on the day we launched our experiment interface and maintained it for more than 30 days. Each article that we regularly posted on the page contained the link to the interface to encourage repeated participation in our experiment. The typical content of the articles was as follows: "The algorithm for music recommendation has been updated. \[the link to the interface\]." Measures -------- ### Statistics from Facebook We obtained the following statistics from the daily reports produced by Facebook Ads Manager, grouped by advertisement type, age, and gender: 1. Reaches: number of users our advertisements reached 2. Exposures: total number of times our advertisements were exposed 3. Clicks: number of clicks of link to the experiment interface on advertisements 4. Cost per mille (CPM): average cost per 1,000 exposures of advertisements 5. Click-through rate (CTR): number of clicks of link per exposure 6. Cost per click: average CPC of link on advertisements In addition, we determined the number of people who pressed the "Like" button on our Facebook page from the daily reports of users' activities offered by Facebook. ### User Statistics of Experiment Interface Our experiment interface was designed in such a way that participants could login to the interface with their Facebook accounts in order to eliminate the inconvenience of creating new accounts. This enabled us to retrieve personal information about each user. We collected only basic personal information because retrieving additional information requires the users' permission, which might have been unacceptable for some users and consequently lowered the number of participants ([@B21]). Every activity in our experiment interface was tracked in terms of type and time stamp, including logging in and pressing the music player buttons (e.g., play, pause, skip). Results ======= During the experimental period, our Facebook advertisement cost \$266.27 in total. It had 1,122,751 reaches, 4,343,736 exposures, and 2,431 clicks, which resulted in a CPM of \$0.06, CTR of 0.056%, and CPC of \$0.11. At the same time, our Facebook page received 768 "Likes." A total of 395 participants joined our experiment interface, the total number of their activities was 3,145, and the average time spent on the experiment interface was 18.59 min. Daily Flows of Participants --------------------------- **Figure [4](#F4){ref-type="fig"}** shows the daily statistics of promotional elements and participants' actions. The number of articles posted on our Facebook page and the cost of the Facebook advertisement are shown at the top in blue. The number of "Likes" that the Facebook page received, the number of users who registered on our experiment interface, and the number of actions of the participants are shown at the bottom in green. The days when articles were posted are marked with circles (●), and the days when the advertisement cost was set high are marked with squares (■). The days when both articles were posted and the cost was set high are marked with diamonds (♦). {#F4} It is noticeable that the number of participants' actions was significantly higher when we promoted our experiment. For example, when we posted an article on our Facebook page on Days 7, 18, 22, 25, and 26 and when we set the daily budget of the advertisement high on Days 7, 14, 22, 25, and 26, the number of actions prominently increased. When Day 18 (an article was posted on the Facebook page without budget adjustment) and Day 14 (the daily budget was set high without posting an article) were compared, the latter was found to be more effective in facilitating users' actions, which directly influenced the number of exposures. **Tables [1](#T1){ref-type="table"}** and **[2](#T2){ref-type="table"}** present statistical analysis results regarding the relationship between attractions and responses. As shown in the tables, both articles posting and increasing the advertisement cost have a strong influence on every type of response, especially the number of daily users in our experiment interface. ###### *z*-statistics and *p*-values of one-sided Wilcoxon rank-sum tests comparing responses (number of daily "Likes" of Facebook page, number of daily users of experiment interface, and number of daily actions in interface) when articles were posted on Facebook page (*N = 5*) and those when no articles were posted (*N = 25*). Page: Daily "Likes" Interface: Daily users Interface: Daily actions ---------------------- ------------------------ -------------------------- -2.090 (*p* = 0.018) -3.296 (*p* \< 0.001) -3.062 (*p* = 0.001) ###### Correlation coefficients (Kendall's τ) and *p*-values between advertisement cost and responses (number of daily "Likes" of Facebook page, number of daily users of experiment interface, and number of daily actions in interface). Page: Daily "Likes" Interface: Daily users Interface: Daily actions ---------------------- ------------------------ -------------------------- 0.558 (*p* \< 0.001) 0.637 (*p* \< 0.001) 0.511 (*p* \< 0.001) Differences by Advertisement Type --------------------------------- For the two types of Facebook advertisements, CPM, CTR, and CPC are compared in **Figure [5](#F5){ref-type="fig"}**. As shown in the figure, the CPM of the right-column advertisement is much smaller than that of the newsfeed advertisement; the right-column advertisement was exposed to users about 19 times more often than the newsfeed advertisement for the same cost. However, the CTR of the right-column advertisement is much smaller than that of the newsfeed advertisement. {#F5} As a result, the CPC of the newsfeed advertisement (*N = 30, M = 0.11, SD = 0.03*) is lower than that of the right-column advertisement (*N = 7, M = 0.21, SD = 0.04*). A one-sided Wilcoxon rank-sum test also confirms the statistical significance of this difference (*z = -3.897, p \< 0.001*). It is because the CTR of the newsfeed advertisement is significantly higher than that of the right-column advertisement (*0.198% \> 0.006%*) even though the latter has a lower CPM (*\$0.19 \> \$0.01*). This implies that the newsfeed advertisement is more effective for attracting participants, while the right-column advertisement is better if frequent exposures of certain information are the aim. Because we needed to maximize the number of clicks rather than the number of exposures, we stopped delivering the right-column advertisement after 7 days and focused our budget on the newsfeed advertisement thereafter. Differences by Gender --------------------- In total, the newsfeed advertisement was exposed 603,660 times to 302,112 males and 474,819 times to 254,393 females. From these statistics, we examine gender differences in terms of daily CPM, CTR, and CPC for the newsfeed advertisement via one-sided Wilcoxon signed-rank tests. The test result for CTR (*z = -2.016, p = 0.022*) shows that women clicked more than men for the same number of exposures to the advertisement. In addition, the results for both CPM (*z = 3.949, p \< 0.001*) and CPC (*z = 2.715, p = 0.003*) strongly suggest that a higher budget was required to attract men than women. **Figure [6](#F6){ref-type="fig"}** shows how differently males and females acted in response to the Facebook advertisements. It is observed that females clicked the "Like" button of the Facebook page 23.2% more than males, while males actually joined the experiment interface 54.8% more than females. One-sided Wilcoxon signed-rank tests also confirm the statistical significance of the gender differences for both the number of page "Likes" (*z = -2.400, p = 0.008*) and the number of participants who joined the interface (*z = 3.491, p \< 0.001*). These imply that females were more active on Facebook than males. However, males showed a higher tendency to use the service immediately, which required going outside of Facebook. {#F6} Differences by Age ------------------ **Figure [7](#F7){ref-type="fig"}** compares different age groups (as defined by Facebook) for the newsfeed advertisement in terms of CTR, CPC, and CPM. It can be seen that CTR increases and CPC decreases with increasing age. Kruskal--Wallis tests also confirm significant differences between the age groups for CTR (*χ = 22.052, df = 4, p \< 0.001*) and CPC (*χ = 20.433, df = 4, p \< 0.001*) but not for CPM (*χ = 1.878, df = 4, p = 0.758*). {#F7} Discussion ========== Previous studies have raised concerns about the inefficiency of using online social networks to attract participants. [@B1] noted that it is difficult to recruit participants from social networks, because they have very low response rates. For example, [@B13] got a response rate of 0.024% from MySpace, and [@B14] and [@B11] reported that the CTR of their Facebook advertisement ranged from 0.02 to 0.04% and 0.024 to 0.033%, respectively. In our case, we obtained a CTR of 0.056% on average, which is slightly higher than those in the previous studies but has the same order of magnitude. However, we argue that it is more meaningful to consider the cost than the response rate in evaluating the efficiency of online social networks for crowdsourcing. When a crowdsourcing problem is dealt with, a certain minimum number of participants or responses is usually required in order to be able to conduct reliable statistical analyses. For a fixed value of CTR, this can be achieved via increasing exposures of the advertisement in social networks by, for example, increasing the budget for the advertisement. Of course, the user pool the advertisement will reach must be sufficiently large, but this will be mostly satisfied due to the extremely large number of Facebook users. In our experiment, the overall CPC was \$0.11. In Section "Comparison with other Crowdsourcing Projects," we will show that this result corresponds to reasonable cost efficiency that is comparable to other crowdsourcing experiments using MTurk. We showed that both advertisements and page articles were able to attract participants, and the former was more effective. When the newsfeed type and right-column type of advertisements were compared, our results showed that the former was more effective than the latter, recording CPMs of 0.198 and 0.006%, respectively. [@B20] recruited participants for a self-report questionnaire about drugs and alcohol via the right-column advertisement of Facebook. They achieved a CTR of 0.054%, which is significantly higher than that of our experiment, because they advertised the chance to win a prize as a reward for participation. Note that, however, such a monetary reward-based approach was not applicable in our case due to the reasons mentioned in the introduction. Instead, we raised the CTR to 0.056% on average by employing the newsfeed advertisement. In the study of [@B14], another method of improving the efficiency of the Facebook advertisement was shown. They did not offer any incentive to participants but achieved a CTR of up to 0.04% by targeting a specific group of interest for the Facebook advertisement (i.e., the advertisement was shown only on the Facebook pages of 24 selected colleges and universities). We also note that the users' behavioral patterns with respect to their demographic characteristics in our experiment are in line with existing findings, particularly for gender differences. It is known that females usually engage in more Facebook activities than males ([@B9]), but males participate in experiments more than females ([@B20]). Our experimental results also follow the same patterns: women engaged in more Facebook activities but less external engagements. Regarding age differences, our results show a higher CTR as age increases. This seems to be because older people either have higher interest in the advertisement or have more difficulty in distinguishing advertisements from posts published by their friends. It is hard to compare this result with existing findings due to the lack of comparable studies. [@B16] reported less Facebook activities of older people in terms of, for example, number of hours on Facebook, number of friends, and photo activities, but such activities are very different from clicking advertisements. Comparison with other Crowdsourcing Projects -------------------------------------------- To examine the suitability of attracting participants via online social networks, we compare our analysis results with existing crowdsourcing projects, especially those that recruit participants via MTurk and whose topics are related to music information retrieval. **Table [3](#T3){ref-type="table"}** shows the comparison of our study and three other studies in terms of experiment type, crowdsourcing medium, data quality-control method, number of participants, number of accepted HITs, number of rejected HITs, average cost per accepted HIT, and completion time per HIT. ###### Comparison of our experiment with existing MTurk-based crowdsourcing experiments. EvoTunes [@B12] [@B15] [@B19] ----------------------------- ----------------------- --------------------------- ----------------------- ------------------------- Topic Music recommendation Music similarity judgment Music tagging Musical mood annotation Medium Facebook MTurk MTurk MTurk Quality control Statistical filtering Duplicated questions Statistical filtering Statistical filtering Number of participants 395 N/A 209 272 Accepted HITs 935^∗^ 583 2566 634 Rejected HITs 2210^∗^ 464 305 756 Cost per accepted HIT (\$) 0.28^∗^ 0.22 ≥0.03 0.54 Completion time per HIT (s) 248.2^∗^ N/A N/A ≥330 ∗ Human Intelligence Tasks are defined only in MTurk, and in our case, HITs are not clearly defined. However, for comparison, we consider each action in our experiment (e.g., listening to as long until the end, pressing the skip button) as a HIT, based on which we estimate HIT-related statistics. [@B12] examined the efficiency of MTurk for music similarity judgment. Each HIT contained 13--15 pairs of music samples. Workers listened to given pairs of samples and rated them on a 3-point scale (not similar, somewhat similar, and very similar). Each task contained one duplicated pair so that task results containing different answers for the same pairs were rejected. Moreover, tasks that were completed too quickly were also rejected. [@B15] used MTurk for associating tags to songs. They used 925 10-s song clips. Workers had to provide 5--15 tags for each clip in five categories. Quality control was performed by rejecting tasks based on the length of words, size of vocabulary, or the frequency of stop words. The overall cost spent was approximately \$100, so the average cost per accepted HIT can be estimated as at least \$0.03. [@B19] used MTurk to annotate the emotional mood of songs. Each HIT contained 11 30-s song clips, including one dummy clip. Workers had to annotate the emotion that they felt on the two-dimensional arousal--valence (A--V) space while listening to each clip. Two clips were identical, so the tasks showing largely different results for that pair were rejected. Some other tasks in which workers did not seem to understand the instruction were also rejected. These experiments were suitable to be conducted on MTurk, because each of them could be divided into small, distinct tasks and concealing their objectives was not important. In contrast, our experiment could not be constructed with separate, well-defined tasks, and natural music listening situations, which are difficult to be implemented within MTurk, were important to verify the music recommendation algorithm. Because our experiment did not use MTurk, we need to define which data should be regarded as "accepted HITs" or "rejected HITs" for comparison with the aforementioned experiments. Our experiment interface gathered participants' actions between transitions of song pairs (i.e., listening until the end and pressing the skip button). During the experiment, we implemented the following data validation mechanisms to reject untruthful responses ([@B5]). 1. The system occasionally made the "wrong" recommendation. If the user kept listening without skipping, the actions of the user were filtered out. 2. The system occasionally showed a dialog box containing a "continue" button. If the user did not press the button for a long time, the recent actions of the user were filtered out. Only the actions that passed these validation mechanisms were used to improve the recommendation algorithm. Therefore, we consider these actions accepted HITs and the rest rejected HITs. The cost per accepted HIT in our experiment is also estimated based on the overall cost paid for the Facebook advertisement. From the table, it can be concluded that our experiment has reasonable efficiency compared to the MTurk-based experiments. First, our experiment was able to recruit a larger number of participants than the other experiments. Moreover, the estimated cost per accepted HIT in our experiment is within the range of those of the other experiments even though our experiment rejected a larger number of tasks. Here, it should be noted that the HITs of each experiment have different complexities. The time complexity of the HIT in our experiment is similar to (or slightly less than) that in the experiment of [@B19] where a HIT contained 11 30-s song clips. However, the cost per accepted HIT is much less in our case, which shows the efficiency of our crowdsourcing strategy. Although [@B15] did not mention the completion time per HIT, each HIT of the experiment contained only one 10-s song clip, which is very simple, and thus, the cost per accepted HIT is relatively low in comparison to that of our experiment. Advantages and Disadvantages of Social Networks for Crowdsourced Experiments ---------------------------------------------------------------------------- Based on our experiment and the analysis presented above, we find both advantages and disadvantages of using a social network to recruit participants for crowdsourcing-based experiments. First, the experimenters can easily analyze the demographic characteristics of workers or trace them over time. Most social networks support Application Program Interfaces (APIs) to retrieve the basic information and interests of each user under authorized permission. This allows researchers to categorize collected data with respect to various aspects of personal information, which may be beyond country, age, or gender. In addition, those who participate in the experiment repeatedly can be easily traced. In addition, social networks enable researchers to choose motivations other than financial reward, including enjoyment and implicit work. These types of motivation do not require direct payment to users. Participants are encouraged to join the experiment voluntarily, and those who wish to participate only for the money are excluded. Moreover, the experiment can proceed as though it is a real service, which allows observation of participants' natural responses without bias. Moreover, social networks allow experimenters to communicate with workers bilaterally, while most MTurk-based experiments implement only one-way feedback functions ([@B15]; [@B19]). For example, our experiment used a Facebook page to communicate with and receive feedback from participants. Some users may have difficulty participating in experiments due to their particular environmental configurations, such as operating systems, network states, and web browsers. However, one-way feedback may not be sufficiently effective to identify and resolve such problems. Communication in both directions enables researchers to deal with unexpected errors or problems in the experiment in a timely manner, which is critical for ensuring the quality of data. Finally, attracting participants via social networks is suitable for long-term crowdsourcing-based experiments. If the Facebook page or Twitter account for an experiment is maintained, the number of subscribers or followers can increase, and thus, more and more participants can be involved in the experiment as time goes on. Moreover, posting an article does not incur any charge, so the overall required cost to promote the experiment can be reduced in comparison to the case with one-time financial motivation when the experiment is operated for a long time. However, there are also drawbacks of using social networks to attract participants. While the amount of money paid to participants can be easily defined and estimated in financial reward-based experiments, it may be rather difficult to estimate the cost required for advertisements to gather a sufficient amount of data in social networks. Moreover, privacy issues must be considered if detailed information of each user needs to be collected. Permission is required for retrieving some types of information, such as Facebook pages that people like and e-mail addresses, and people may not be willing to give such information readily ([@B21]). Suggestions for Recruitment Based on Social Networks ---------------------------------------------------- In order to exploit the advantages of attracting participants via social networks, proper strategies are required, which are discussed below. First, apart from paid advertisements, maintaining promotional pages such as the Facebook page is highly beneficial to increase the constant influx of participants. While advertisements are shown only for a limited period at a cost, articles on promotional pages can be permanently accessed at no cost. Moreover, most social network services offer so-called word-of-mouth features, including the "Like" button on Facebook and "Retweet" on Twitter. These features can facilitate the propagation of articles on the page for many people. Our experiment showed that the articles on the Facebook page were effective at attracting participants. Next, advertisements placed in the same place as normal articles are more effective than those in other places. In our analysis, the newsfeed advertisement was shown to be better than the right-column advertisement on Facebook due to its higher CTR and lower CPC. In addition, the newsfeed advertisement offers socially connected features, such as the "Like" button of the Facebook page and a list of friends who like the page, which can increase interest in the advertisement. Furthermore, adopting different strategies for different target user groups is important to enhance efficiency due to group-dependent behavioral patterns. According to our results, experiments performed directly on Facebook would engage women more easily than men, while Facebook-independent experiments that require users to go outside of Facebook would attract men better than women. In addition, a higher advertisement budget would be required to draw men than women and younger age groups than older age groups. Limitations of the Study ------------------------ In these days, people use online social networks via mobile devices. However, the advertisement in our experiment targeted only the desktop environment because the validity of the experimental data gathered from the mobile environment was questionable; for instance, users may lose attention to the experiment easily in the mobile environment. Therefore, it is inconclusive whether our observations are valid in mobile-based experiments. In addition, our advertisement target was limited to English-speaking countries. This was necessary because all the text in the experiment interface, advertisement, and Facebook page articles was English and the lyrics of the songs were also English. However, this might limit general applicability of our observations to non-English-speaking countries. Conclusion ========== In this paper, we presented and analyzed an approach of using a social network to attract participants for crowdsourced multimedia-involved behavioral testing. The analyzed results demonstrated that promoting the experiment on Facebook via a dedicated page and advertisements is effective to attract users. A comparison with other crowdsourcing projects having similar topics in multimedia showed that our approach has competitive efficiency and usability in recruiting a sufficient number of users and retrieving a high quality of data while keeping the cost for running the experiment comparably reasonable. In addition, we suggested methods of using online social networks as crowdsourcing media for achieving high efficiency. We anticipate that our analysis results will help researchers find efficient ways of gathering fine data from crowds. Conflict of Interest Statement ============================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. This research was supported by the MSIP (Ministry of Science, ICT and Future Planning), Korea, under the "IT Consilience Creative Program" (IITP-2015-R0346-15-1008) supervised by the IITP (Institute for Information & Communications Technology Promotion), and by the Basic Science Research Program through the National Research Foundation of Korea funded by the MSIP (2013R1A1A1007822). Other social networks (e.g., Twitter) may be used, but we chose Facebook because it has the largest number of active users. [^1]: Edited by: *Javier Jaen, Universitat Politecnica de Valencia, Spain* [^2]: Reviewed by: *Andrej Košir, University of Ljubljana, Slovenia; Amael Arguel, Macquarie University, Australia* [^3]: This article was submitted to Human-Media Interaction, a section of the journal Frontiers in Psychology

All relevant data are within the manuscript and its Supporting Information files. Introduction {#sec001} ============ The improvement of the performance of commercial banks is of great significance for the rational use of financial resources and the enhancement of the comprehensive competitiveness of China's banking industry. This improvement leads to considerable progress and the development of commercial banks. Intuitively, the efficiency evaluation of banks in China plays a vital role in providing policy implications to managers and decision-makers. Issues related to banking efficiency have long been emphasized in the literature, wherein financial ratio analysis \[[@pone.0204559.ref001]\], data envelopment analysis (DEA) \[[@pone.0204559.ref002]\], stochastic semi-nonparametric envelopment of data (StoNED) \[[@pone.0204559.ref003]\], and stochastic frontier analysis (SFA) \[[@pone.0204559.ref004],[@pone.0204559.ref005]\] have been applied to find the production or cost frontier and analyze productivity efficiency. Among these evaluation methods, the DEA (radial or non-radial) is one of the most widely used tools to estimate banking efficiency. Examples include Halkos et al. \[[@pone.0204559.ref001]\] for Greece; Ohsato et al. \[[@pone.0204559.ref006]\] and Fukuyama and Weber \[[@pone.0204559.ref007],[@pone.0204559.ref008]\] for Japan; Ebrahimnejad et al. \[[@pone.0204559.ref009]\] for East Virginia; Eken et al. \[[@pone.0204559.ref010]\] for Turkey; Wu et al. \[[@pone.0204559.ref011]\] for Canada; Park et al. \[[@pone.0204559.ref012]\] for Korea; Khodabakhshi et al. \[[@pone.0204559.ref013]\] and Kwon et al. \[[@pone.0204559.ref014]\] for the US; Webb \[[@pone.0204559.ref015]\] for the UK retail Banks; Puri et al. \[[@pone.0204559.ref016]--[@pone.0204559.ref018]\] for India; Nguyen et al. \[[@pone.0204559.ref019]\] for Vietnam; Matthews \[[@pone.0204559.ref020]\], Wang et al. \[[@pone.0204559.ref021]\], and Zha et al. \[[@pone.0204559.ref022]\] for China; and Huang et al. \[[@pone.0204559.ref023]\] for a panel of 17 Central and Eastern European countries. Decision making units (DMUs) are treated as a 'black box' in traditional DEA models, where one of the drawbacks of these approaches is the neglect of intermediate products or linked activities \[[@pone.0204559.ref024], [@pone.0204559.ref025]\]. In fact, some production systems have a network (multiple stages) structure; these include new energy enterprises \[[@pone.0204559.ref026]\], the iron and steel industry \[[@pone.0204559.ref027], [@pone.0204559.ref028]\], the service industry \[[@pone.0204559.ref029]\], and commercial banks and supply chains \[[@pone.0204559.ref030], [@pone.0204559.ref031]\]. To open the 'black box' and obtain greater insight into the production process, network DEA models have been constructed to analyze the network structure of production. Some examples include Huang et al. \[[@pone.0204559.ref002]\], Fukuyama et al. \[[@pone.0204559.ref008]\], Wang et al. \[[@pone.0204559.ref021]\], Tone et al. \[[@pone.0204559.ref025]\], Färe et al. \[[@pone.0204559.ref032]\], Zhu \[[@pone.0204559.ref033]\], Sexton et al. \[[@pone.0204559.ref034]\], Lozano \[[@pone.0204559.ref035]\] and Huang et al. \[[@pone.0204559.ref036]\]. On the basis of these studies, DEA methods, especially network structures or multistage DEA approaches, attracted interest for measuring banking efficiency from the perspective of macro-, meso-, and micro-levels. However, several key issues should be highlighted when measuring bank efficiency, which are explained in detail as follows. Although many previous studies have assessed banking performance by considering undesirable outputs (such as non-performing loans; see, for example, Huang et al. \[[@pone.0204559.ref002]\], Park et al. \[[@pone.0204559.ref012]\], Zha et al. \[[@pone.0204559.ref022]\], Fukuyama and Weber \[[@pone.0204559.ref037]\] and Fukuyama et al. \[[@pone.0204559.ref038]\]) and super efficiency (Khodabakhshi et al. \[[@pone.0204559.ref013]\], Andersen et al. \[[@pone.0204559.ref039]\], Chiu et al. \[[@pone.0204559.ref040]\], Chen \[[@pone.0204559.ref041]\], Minh et al. \[[@pone.0204559.ref042]\], Avkiran et al. \[[@pone.0204559.ref043]\] and Zhou et al. \[[@pone.0204559.ref044]\]) either separately or simultaneously \[[@pone.0204559.ref002]\], these models do not consider both super efficiency and heterogeneous factors, while they assume that banks share a common production frontier. Actually, the production sets of different banks may differ due to differences in physical and capital stocks, as well as the social and economic infrastructure in which the production occurs \[[@pone.0204559.ref045]\]. If these differences are left out, the measures of banking efficiency may be biased. Metafrontier analysis is a mainstream approach to consider the heterogeneity of factors \[[@pone.0204559.ref046]--[@pone.0204559.ref048]\], and it proceeds in two steps. First, banks are classified into different groups according to their internal characteristics (such as state-owned commercial banks, joint-stock commercial banks, and foreign banks), including some stylized environments \[[@pone.0204559.ref049]--[@pone.0204559.ref052]\] to estimate a group-specific production frontier for each group. Admittedly, external environments such as the spatial effect of tourism building investments on tourist revenues \[[@pone.0204559.ref053]\], the asymmetric impact of oil price shock on the stock market \[[@pone.0204559.ref054]\], and portfolio optimization problems \[[@pone.0204559.ref055]\] also exert influences on the group division. Second, the metafrontier is estimated by enveloping the group-specific frontiers \[[@pone.0204559.ref029], [@pone.0204559.ref056], [@pone.0204559.ref057]\]. In their latest work, Chiu et al. \[[@pone.0204559.ref057]\] developed a new model to decompose the source of metafrontier inefficiency for various banks based on a two-stage network system with undesirable outputs. Their empirical results suggest that foreign banks are less efficient in developed countries, indicating that there are technology gaps among different types of commercial banks. However, the metafrontier constructed in the abovementioned works is concave, and the meta-technology gap may be greater than unity when evaluated by a slack-based measure (SBM) in a two-stage framework. This outcome may occur because the concave metafrontier exhibited some piecewise differences located on the area labeled 'Infeasible Input-Output combinations' \[[@pone.0204559.ref058]\], which is consistent with the indivisibility of technology proposed by Tone and Sahoo \[[@pone.0204559.ref059]\]. In such cases, it is necessary to extend the concave metafrontier analysis to a non-concave metafrontier one. Based on the need to include the abovementioned issues in the DEA analytical framework, we propose an innovative measurement approach that incorporates a non-concave metafrontier and undesirable outputs into a slack-based network DEA model (NCMeta-US-NSBM), providing more accurate results and evidence when banking efficiency is estimated. To the best of our knowledge, this is the first paper to consider a non-concave metafrontier, super efficiency, and undesirable outputs in a network SBM framework. The strength of the NCMeta-US-NSBM (compared to traditional DEA models) is its comparability treatment of efficiency of the same DMUs in different years due to possible unobserved exogenous technical changes and the ranks of the efficient DMUs on the efficient frontier. Methodology {#sec002} =========== The proposed model {#sec003} ------------------ Assuming that the number of observed DMUs (banks) is *N* and they can be divided into *G*(*G*\>1) groups in terms of heterogeneous factors, each group contains *N*~*g*~ DMUs and ${\sum_{g = 1}^{G}N_{g}} = N$. Furthermore, suppose that there are two stages in the production process of a bank. They are the deposit stage (stage 1) and the loan stage (stage 2), which are also known as the productivity stage and the profitability stage. The efficiency scores of these two stages are called productivity efficiency and profitability efficiency in this study, respectively. Then, each DMU obtains *Q* intermediate products on the consumption of *M* original inputs in the productivity stage and utilizes *Q* intermediate products to produce *R* desirable and *J* undesirable outputs in the second stage. These are denoted by vectors **x,z,y** and **b**, respectively, where $\mathbf{x} = \left\lbrack {x_{1},x_{2},\cdots,x_{M}} \right\rbrack \in {\mathbb{R}}_{+}^{M}$, $\mathbf{z} = \left\lbrack {z_{1},z_{2},\cdots,z_{Q}} \right\rbrack \in {\mathbb{R}}_{+}^{Q}$, $\mathbf{y} = \left\lbrack {y_{1},y_{2},\cdots,y_{R}} \right\rbrack \in {\mathbb{R}}_{+}^{R}$, and $\mathbf{b} = \left\lbrack {b_{1},b_{2},\cdots,b_{J}} \right\rbrack \in {\mathbb{R}}_{+}^{J}$. We define the intensity variable column for the first stage as $\mathbf{\lambda} = (\lambda_{1},\lambda_{2},\cdots,\lambda_{N}) \in {\mathbb{R}}_{+}^{N}$, and, for the second stage, the intensity column vector is $\mathbf{\gamma} = (\gamma_{1},\gamma_{2},\cdots,\gamma_{N}) \in {\mathbb{R}}_{+}^{N}$. All the variables are strictly greater than 0. [Fig 1](#pone.0204559.g001){ref-type="fig"} illustrates the production procedure of a bank. {#pone.0204559.g001} ### Production possibility sets {#sec004} The production possibility sets of the productivity and profitability stages can be given by Eqs ([1](#pone.0204559.e008){ref-type="disp-formula"}) and ([2](#pone.0204559.e009){ref-type="disp-formula"}). {#pone.0204559.e008g} P ( 1 ) = { ( x , z ) \| x can produce z } . {#pone.0204559.e009g} P ( 2 ) = { ( z , y , b ) \| z can produce ( y , b ) } . By combing ([1](#pone.0204559.e008){ref-type="disp-formula"}) and ([2](#pone.0204559.e009){ref-type="disp-formula"}), the network production possibility set is $$\begin{array}{l} {N_{PPS} = \left\{ \left. {(\mathbf{x},\mathbf{z},\mathbf{y},\mathbf{b})} \right| \right.(\mathbf{x},\mathbf{z}) \in P^{(1)}} \\ {\mspace{148mu}\left. {(\mathbf{z},\mathbf{y},\mathbf{b}) \in P^{(2)}} \right\}} \\ \end{array}$$ A function that is homogeneous of degree 1 is said to either have constant returns to scale or neither economies or diseconomies of scale. A function that is homogeneous of a degree greater (less) than 1 is said to have either increasing (decreasing) returns to scale or economies (diseconomies) of scale, known collectively as variable returns to scale. If the production function is not homogeneous, the returns to scale can still be determined by summing the respective ratios of the marginal to average products. Holod et al. \[[@pone.0204559.ref060]\] suggested that the assumption of variable returns to scale (VRS) was a better alternative to constant returns to scale (CRS). Moreover, the performance of the efficiency scores obtained by the assumption of VRS is better than that of the assumption of CRS \[[@pone.0204559.ref061]\]. Thus, we first estimate the banking efficiency under the assumption of VRS. Specifically, the network production technology of the group frontier *g*′(*g*′ = 1,2,⋯,*G*) under the assumption of VRS can be expressed as follows: $$\begin{array}{l} {N_{PPS}^{g^{\prime}} = \left\{ {\left. {(x_{m},z_{q},y_{r},b_{j})} \right|x_{mg^{\prime}o} \geq {\sum\limits_{n \in g^{\prime},n \neq o}{\lambda_{g^{\prime}n}x_{mg^{\prime}n}}},m = 1,2,\cdots,M;} \right.} \\ {\mspace{184mu} y_{rg^{\prime}o} \leq {\sum\limits_{n \in g^{\prime},n \neq o}{\gamma_{g^{\prime}n}y_{rg^{\prime}n}}},r = 1,2,\cdots,R;} \\ {\mspace{184mu} b_{jg^{\prime}o} \geq {\sum\limits_{n \in g^{\prime},n \neq o}{\gamma_{g^{\prime}n}b_{jg^{\prime}n}}},j = 1,2,\cdots,J;} \\ {\mspace{184mu}{\sum\limits_{n \in g^{\prime},n \neq o}\lambda_{g^{\prime}n}} = 1;{\sum\limits_{n \in g^{\prime},n \neq o}\gamma_{g^{\prime}n}} = 1;} \\ {\mspace{184mu}\lambda_{g^{\prime}n},\gamma_{g^{\prime}n} \geq 0,n \in g^{\prime},n \neq o;} \\ {\mspace{184mu}\left. {z_{q}\ free,\ q = 1,2,\cdots,Q.} \right\}} \\ \end{array}$$ Similarly, the network production technology of the overall DMUs, with respect to the concave metafrontier, under the assumption of VRS assumption can be expressed as follows: $$\begin{array}{l} {N_{PPS}^{c - meta} = \left\{ {\left. {(x_{m},z_{q},y_{r},b_{j})} \right|x_{mg^{\prime}o} \geq {\sum\limits_{g = 1}^{G}{\sum\limits_{n \in g^{\prime},n \neq o\ if\ g = g^{\prime}}{\lambda_{gn}x_{mgn}}}},m = 1,2,\cdots,M;} \right.} \\ {\mspace{216mu} y_{rg^{\prime}o} \leq {\sum\limits_{g = 1}^{G}{\sum\limits_{n \in g^{\prime},n \neq o\ if\ g = g^{\prime}}{\gamma_{gn}y_{rgn}}}},r = 1,2,\cdots,R;} \\ {\mspace{216mu} b_{jg^{\prime}o} \geq {\sum\limits_{g = 1}^{G}{\sum\limits_{n \in g^{\prime},n \neq o\ if\ g = g^{\prime}}{\gamma_{gn}b_{jgn}}}},j = 1,2,\cdots,J;} \\ {\mspace{216mu}{\sum\limits_{g = 1}^{G}{\sum\limits_{n \in g^{\prime},n \neq o\ if\ g = g^{\prime}}\lambda_{gn}}} = 1;{\sum\limits_{g = 1}^{G}{\sum\limits_{n \in g^{\prime},n \neq o\ if\ g = g^{\prime}}\gamma_{gn}}} = 1;} \\ {\mspace{216mu}\lambda_{gn},\gamma_{gn} \geq 0,g = 1,2,\cdots,G;n \in g^{\prime},n \neq o\ if\ g = g^{\prime};} \\ {\mspace{216mu}\left. {z_{q}\ free,\ q = 1,2,\cdots,Q.} \right\}} \\ \end{array}$$ [Fig 2](#pone.0204559.g002){ref-type="fig"} portrays the standard (concave) metafrontier with respect to the two technologies (I and II) in the two-stage setting. It is clear from [Fig 2](#pone.0204559.g002){ref-type="fig"} that the metafrontier can encompass input/output combinations that are not feasible in either of the two technologies. These observations are in the triangle labeled 'Infeasible Input--Output combinations' (such as *A*\* and *B*\* in stage 1 and stage 2, respectively; see [Fig 2](#pone.0204559.g002){ref-type="fig"}). Tiedemann et al. \[[@pone.0204559.ref058]\] introduced a non-concave metafrontier (piecewise part of bold form in [Fig 2](#pone.0204559.g002){ref-type="fig"}) that enveloped those input--output combinations that are part of the technology set of at least one of the technologies, eliminating the area labeled 'Infeasible Input--Output combinations.' They also provided a two-step procedure to estimate a non-concave metafrontier. More details can be found in Huang et al. \[[@pone.0204559.ref062]\]. This may ensure that the TGR in different stages under the network framework lies in (0,1\]. Meanwhile, the TGRs of some DMUs in different stages are greater than unity when considering the concave frontier since it contains the area labeled 'Infeasible Input--Output combinations', thus resulting in incorrect and irrational estimations. {#pone.0204559.g002} Following Tiedemann et al. \[[@pone.0204559.ref058]\] and Huang et al. \[[@pone.0204559.ref062]\], we extended the non-concave metafrontier to a network framework, which is expressed as follows: $$\begin{array}{l} {N_{PPS}^{nc - meta} = \left\{ {\left. {(x_{m},z_{q},y_{r},b_{j})} \right|x_{mg^{\prime}o} \geq {\sum\limits_{g = 1}^{G}{\sum\limits_{n \in g^{\prime},n \neq o\ if\ g = g^{\prime}}{\lambda_{gn}x_{mgn}}}},m = 1,2,\cdots,M;} \right.} \\ {\mspace{216mu} y_{rg^{\prime}o} \leq {\sum\limits_{g = 1}^{G}{\sum\limits_{n \in g^{\prime},n \neq o\ if\ g = g^{\prime}}{\gamma_{gn}y_{rgn}}}},r = 1,2,\cdots,R;} \\ {\mspace{216mu} b_{jg^{\prime}o} \geq {\sum\limits_{g = 1}^{G}{\sum\limits_{n \in g^{\prime},n \neq o\ if\ g = g^{\prime}}{\gamma_{gn}b_{jgn}}}},j = 1,2,\cdots,J;} \\ {\mspace{216mu}{\sum\limits_{g = 1}^{G}{\sum\limits_{n \in (g^{\prime} = 1),n \neq o\ if\ g = g^{\prime}}\lambda_{gn}}} = \phi_{1},{\sum\limits_{g = 1}^{G}{\sum\limits_{n \in (g^{\prime} = 2),n \neq o\ if\ g = g^{\prime}}\lambda_{gn}}} = \phi_{2},\cdots,{\sum\limits_{g = 1}^{G}{\sum\limits_{n \in (g^{\prime} = G),n \neq o\ if\ g = g^{\prime}}\lambda_{gn}}} = \phi_{G};} \\ {\mspace{216mu}{\sum\limits_{g = 1}^{G}{\sum\limits_{n \in (g^{\prime} = 1),n \neq o\ if\ g = g^{\prime}}\gamma_{gn}}} = \varphi_{1},{\sum\limits_{g = 1}^{G}{\sum\limits_{n \in (g^{\prime} = 2),n \neq o\ if\ g = g^{\prime}}\gamma_{gn}}} = \varphi_{2},\cdots,{\sum\limits_{g = 1}^{G}{\sum\limits_{n \in (g^{\prime} = G),n \neq o\ if\ g = g^{\prime}}\gamma_{gn}}} = \varphi_{G};} \\ {\mspace{216mu}{\sum\limits_{g = 1}^{G}\phi_{g}} = 1;{\sum\limits_{g = 1}^{G}\varphi_{g}} = 1;\phi_{g} = 1\ or\ 0;\varphi_{g} = 1\ or\ 0;\lambda_{gn},\gamma_{gn} \geq 0;n \in g^{\prime},n \neq o\ if\ g = g^{\prime};} \\ {\mspace{216mu}\left. {z_{q}\ free,\ q = 1,2,\cdots,Q.} \right\}} \\ \end{array}$$ Consequently, we have the following relationship between $N_{PPS}^{g^{\prime}}$, $N_{PPS}^{conc - meta}$ and $N_{PPS}^{nc - meta}$: $$N_{PPS}^{g^{\prime}} \subseteq N_{PPS}^{nc - meta} \subseteq N_{PPS}^{conc - meta}.$$ ### NCMeta-US-NSBM model {#sec005} There are two types of models in DEA: radial and non-radial efficiency measures. The shortcoming of the former approach is that it neglects the non-radial input/output slacks (i.e., it cannot provide detailed information regarding the inefficiency of a specific input/output). As indicated by Tone \[[@pone.0204559.ref063]\], slack-based measures used as non-radial measures can directly address the 'input excess' and 'output shortfall' problems, where the objective function value can be interpreted as the ratio of mean input and output mix inefficiencies. Slack-based measure approaches can effectively explore the sources of inefficiency behind the operational process of each bank from this point of view. In this subsection, we mainly focus on the proposed model NCMeta-US-NSBM, which simultaneously incorporates the non-concave metafrontier technique, super efficiency, and undesirable outputs into the network slack-based measure model. Assume that *N* DMUs (*n* = 1,2,⋯,*N*) consist of *K* divisions (stages). Let *M*~*k*~ be the number of original inputs, *R*~*k*~ and *J*~*k*~ respectively stand for the number of final undesirable and desirable outputs of division, and let *Q*~*k*~ be the number of intermediate products (*k* = 1,2,⋯,*K*). Denote the link leading from division (*k*−1) to division *k* by (*k*−1,*k*) and the set of links by *L*. The observations are $\left\{ {x_{g^{\prime}n}^{k} \in {\mathbb{R}}_{+}^{M_{k}}} \right\}$ (original inputs to DMU *n* at division *k* in group *g*′), $\left\{ {y_{g^{\prime}n}^{k} \in {\mathbb{R}}_{+}^{R_{k}}} \right\}$ (final desirable outputs to DMU *n* at division *k* in group *g*′), $\left\{ {b_{g^{\prime}n}^{k} \in {\mathbb{R}}_{+}^{J_{k}}} \right\}$ (final undesirable outputs to DMU *n* at division *k* in group *g*′) and $\left\{ {z_{g^{\prime}n}^{(k - 1,k)} \in {\mathbb{R}}_{+}^{\pi(k - 1,k)}} \right\}$ (linked intermediate product from division (*k*−1) to division *k* in group *g*′), where *π*(*k*−1,*k*) is the number of items in link (*k*−1,*k*). Note that $z_{g^{\prime}n}^{(k - 1,k)}$ demonstrates the outputs from (*k*−1) and the inputs to *k*. With the group frontier and metafrontier defined, we can measure bank efficiency by a Group-US-SBM model by considering the undesirable outputs and super efficiency in the SBM model with respect to the group frontier. The optimal objective value for the *o*th DMU in group *g*′(*o* = 1,2,⋯,*N*~*g*′~; *g*′ = 1,2,⋯,*G*) under the group frontier is estimated as: $$\begin{array}{l} {\lbrack{Group}‑{US}‑{NSBM}\rbrack\ \rho_{g^{\prime}o}^{group*} = \min\frac{{\sum\limits_{k = 1}^{K}\omega^{k}}\left\lbrack {1 + \frac{1}{M_{k}}\left( {\sum\limits_{m = 1}^{M_{k}}\frac{t_{mg^{\prime}o}^{xk}}{x_{mg^{\prime}o}^{k}}} \right)} \right\rbrack}{{\sum\limits_{k = 1}^{K}\omega^{k}}\left\lbrack {1 - \frac{1}{R_{k} + J_{k}}\left( {{\sum\limits_{r = 1}^{R_{k}}\frac{t_{rg^{\prime}o}^{yk}}{y_{rg^{\prime}o}^{k}}} + {\sum\limits_{j = 1}^{J_{k}}\frac{t_{jg^{\prime}o}^{bk}}{b_{jg^{\prime}o}^{k}}}} \right)} \right\rbrack}} \\ {s.t.\ x_{mg^{\prime}o}^{k} - {\sum\limits_{n \in g^{\prime},n \neq o}{\lambda_{g^{\prime}n}^{k}x_{mg^{\prime}n}^{k}}} + t_{mg^{\prime}o}^{xk} \geq 0,m = 1,2,\cdots,M_{k};k = 1,2,\cdots,K;} \\ {{\sum\limits_{n \in g^{\prime},n \neq o}{\gamma_{g^{\prime}n}^{k}y_{rg^{\prime}n}^{k}}} - y_{rg^{\prime}o}^{k} + t_{rg^{\prime}o}^{yk} \geq 0,r = 1,2,\cdots,R_{k};k = 1,2,\cdots,K;} \\ {b_{jg^{\prime}o}^{k} - {\sum\limits_{n \in g^{\prime},n \neq o}{\gamma_{g^{\prime}n}^{k}b_{jg^{\prime}n}^{k}}} + t_{jg^{\prime}o}^{bk} \geq 0,j = 1,2,\cdots,J_{k};k = 1,2,\cdots,K;} \\ {1 - \frac{1}{R_{k} + J_{k}}\left( {{\sum\limits_{r = 1}^{R_{k}}\frac{t_{rg^{\prime}}^{yk}}{y_{rg^{\prime}o}^{k}}} + {\sum\limits_{j = 1}^{J_{k}}\frac{t_{jg^{\prime}}^{bk}}{b_{jg^{\prime}o}^{k}}}} \right) \geq \varepsilon,k = 1,2,\cdots,K;} \\ {{\sum\limits_{n \in g^{\prime},n \neq o}{\lambda_{g^{\prime}n}^{k}z_{g^{\prime}n}^{({k - 1,k})}}} = {\sum\limits_{n \in g^{\prime},n \neq o}{\gamma_{g^{\prime}n}^{k}z_{g^{\prime}n}^{({k,k - 1})}}},k = 1,2,\cdots,K;} \\ {{\sum\limits_{n \in g^{\prime},n \neq o}\lambda_{g^{\prime}n}^{k}} = 1,k = 1,2,\cdots,K;} \\ {{\sum\limits_{n \in g^{\prime},n \neq o}\gamma_{g^{\prime}n}^{k}} = 1,k = 1,2,\cdots,K;} \\ {{\sum\limits_{k = 1}^{K}\omega^{k}} = 1,k = 1,2,\cdots,K;} \\ {\omega^{k},t^{xk},t^{yk},t^{bk},\lambda^{k},\gamma^{k} \geq 0;k = 1,2,\cdots,K.} \\ \end{array}$$ The same applies for the *o*th DMU in group *g*′(*o* = 1,2,⋯,*N*~*g*′~; *g*′ = 1,2,⋯,*G*) under the concave metafrontier, which is estimated as: $$\begin{array}{l} {\lbrack{CMeta}‑{US}‑{NSBM}\rbrack\ \rho_{g^{\prime}o}^{c - meta*} = \min\frac{{\sum\limits_{k = 1}^{K}\omega^{k}}\left\lbrack {1 + \frac{1}{M_{k}}\left( {\sum\limits_{m = 1}^{M_{k}}\frac{s_{mg^{\prime}o}^{xk}}{x_{mg^{\prime}o}^{k}}} \right)} \right\rbrack}{{\sum\limits_{k = 1}^{K}\omega^{k}}\left\lbrack {1 - \frac{1}{R_{k} + J_{k}}\left( {{\sum\limits_{r = 1}^{R_{k}}\frac{s_{rg^{\prime}o}^{yk}}{y_{rg^{\prime}o}^{k}}} + {\sum\limits_{j = 1}^{J_{k}}\frac{s_{jg^{\prime}o}^{bk}}{b_{jg^{\prime}o}^{k}}}} \right)} \right\rbrack}} \\ {s.t.\ x_{mg^{\prime}o}^{k} - {\sum\limits_{g = 1}^{G}{\sum\limits_{n \in g^{\prime},n \neq o\ if\ g = g^{\prime}}{\lambda_{gn}^{k}x_{mgn}^{k}}}} + s_{mg^{\prime}o}^{xk} \geq 0,m = 1,2,\cdots,M_{k};k = 1,2,\cdots,K;} \\ {{\sum\limits_{g = 1}^{G}{\sum\limits_{n \in g^{\prime},n \neq o\ if\ g = g^{\prime}}{\gamma_{gn}^{k}y_{rgn}^{k}}}} - y_{rg^{\prime}o}^{k} + s_{rg^{\prime}o}^{yk} \geq 0,r = 1,2,\cdots,R_{k};k = 1,2,\cdots,K;} \\ {b_{jg^{\prime}o}^{k} - {\sum\limits_{g = 1}^{G}{\sum\limits_{n \in g^{\prime},n \neq o\ if\ g = g^{\prime}}{\gamma_{gn}^{k}b_{jgn}^{k}}}} + s_{jg^{\prime}o}^{bk} \geq 0,j = 1,2,\cdots,J_{k};k = 1,2,\cdots,K;} \\ {1 - \frac{1}{R_{k} + J_{k}}\left( {{\sum\limits_{r = 1}^{R_{k}}\frac{s_{rg^{\prime}}^{yk}}{y_{rg^{\prime}o}^{k}}} + {\sum\limits_{j = 1}^{J_{k}}\frac{s_{jg^{\prime}}^{bk}}{b_{jg^{\prime}o}^{k}}}} \right) \geq \varepsilon,k = 1,2,\cdots,K;} \\ {{\sum\limits_{g = 1}^{G}{\sum\limits_{n \in g^{\prime},n \neq o\ if\ g = g^{\prime}}{\lambda_{gn}^{k}z_{gn}^{({k - 1,k})}}}} = {\sum\limits_{g = 1}^{G}{\sum\limits_{n \in g^{\prime},n \neq o\ if\ g = g^{\prime}}{\gamma_{gn}^{k}z_{gn}^{({k,k - 1})}}}},k = 1,2,\cdots,K;} \\ {{\sum\limits_{g = 1}^{G}{\sum\limits_{n \in (g^{\prime} = 1),n \neq o\ if\ g = g^{\prime}}\lambda_{gn}^{k}}} = 1,k = 1,2,\cdots,K;} \\ {{\sum\limits_{g = 1}^{G}{\sum\limits_{n \in (g^{\prime} = 1),n \neq o\ if\ g = g^{\prime}}\gamma_{gn}^{k}}} = 1,k = 1,2,\cdots,K;} \\ {{\sum\limits_{k = 1}^{K}\omega^{k}} = 1,k = 1,2,\cdots,K;} \\ {\omega^{k},s^{xk},s^{yk},s^{bk},\lambda^{k},\gamma^{k} \geq 0;k = 1,2,\cdots,K.} \\ \end{array}$$ For a two-stage (two-division) bank production procession (that is, *K* =2), the bad outputs (such as NPLs) are part of stage 2 instead of stage 1. Accordingly, we can also define the divisional efficiency score of each stage with respect to the group frontier as follows: $$\rho_{g^{\prime}o}^{group*1} = \frac{1 + \frac{1}{M_{1}}\left( {\sum\limits_{m = 1}^{M_{1}}\frac{t_{mg^{\prime}o}^{x1*}}{x_{mg^{\prime}o}^{1}}} \right)}{1 - \frac{1}{Q}\left( {\sum\limits_{q = 1}^{Q}\frac{t_{qg^{\prime}o}^{z*}}{z_{qg^{\prime}o}}} \right)};$$ $$\rho_{g^{\prime}o}^{group*2} = \frac{1 + \frac{1}{Q}\left( {\sum\limits_{q = 1}^{Q}\frac{t_{qg^{\prime}o}^{z*}}{z_{qg^{\prime}o}}} \right)}{\left\lbrack {1 - \frac{1}{R_{2} + J_{2}}\left( {{\sum\limits_{r = 1}^{R_{2}}\frac{t_{rg^{\prime}o}^{y2*}}{y_{rg^{\prime}o}^{2}}} + {\sum\limits_{j = 1}^{J_{2}}\frac{t_{jg^{\prime}o}^{b2*}}{b_{jg^{\prime}o}^{2}}}} \right)} \right\rbrack}.$$ where the slack variables with the superscript "\*" denote the optimal slacks of model ([8](#pone.0204559.e023){ref-type="disp-formula"}) in the corresponding stages. The same applies for a concave metafrontier. The divisional efficiency score of each stage is computed as follows: $$\rho_{g^{\prime}o}^{c - meta*1} = \frac{1 + \frac{1}{M_{1}}\left( {\sum\limits_{m = 1}^{M_{1}}\frac{s_{mg^{\prime}o}^{x1*}}{x_{mg^{\prime}o}^{1}}} \right)}{1 - \frac{1}{Q}\left( {\sum\limits_{q = 1}^{Q}\frac{s_{qg^{\prime}o}^{z*}}{z_{qg^{\prime}o}}} \right)};$$ $$\rho_{g^{\prime}o}^{c - meta*2} = \frac{1 + \frac{1}{Q}\left( {\sum\limits_{q = 1}^{Q}\frac{s_{qg^{\prime}o}^{z*}}{z_{qg^{\prime}o}}} \right)}{\left\lbrack {1 - \frac{1}{R_{2} + J_{2}}\left( {{\sum\limits_{r = 1}^{R_{2}}\frac{s_{rg^{\prime}o}^{y2*}}{y_{rg^{\prime}o}^{2}}} + {\sum\limits_{j = 1}^{J_{2}}\frac{s_{jg^{\prime}o}^{b2*}}{b_{jg^{\prime}o}^{2}}}} \right)} \right\rbrack}.$$ where the slack variables with the superscript "\*" denote the optimal slacks of model ([9](#pone.0204559.e024){ref-type="disp-formula"}) in the corresponding stages. The technology gap ratio (TGR) based on the optimal objective values $\rho_{g^{\prime}o}^{group*}$ and $\rho_{g^{\prime}o}^{c - meta*}$ can be computed as follows: $$TGR_{g^{\prime}o}^{c} = \frac{\rho_{g^{\prime}o}^{c - meta*}}{\rho_{g^{\prime}o}^{group*}}.$$ The TGR with respect to each stage under a concave metafrontier is given by: $$TGR_{g^{\prime}o}^{c1} = \frac{\rho_{g^{\prime}o}^{c - meta*1}}{\rho_{g^{\prime}o}^{group*1}};$$ $$TGR_{g^{\prime}o}^{c2} = \frac{\rho_{g^{\prime}o}^{c - meta*2}}{\rho_{g^{\prime}o}^{group*2}}.$$ $$\begin{array}{l} {\lbrack{NCMeta}‑{US}‑{NSBM}\rbrack\ \rho_{g^{\prime}o}^{nc - meta*} = \min\frac{{\sum\limits_{k = 1}^{K}\omega^{k}}\left\lbrack {1 + \frac{1}{M_{k}}\left( {\sum\limits_{m = 1}^{M_{k}}\frac{\tau_{mg^{\prime}o}^{xk}}{x_{mg^{\prime}o}^{k}}} \right)} \right\rbrack}{{\sum\limits_{k = 1}^{K}\omega^{k}}\left\lbrack {1 - \frac{1}{R_{k} + J_{k}}\left( {{\sum\limits_{r = 1}^{R_{k}}\frac{\tau_{rg^{\prime}o}^{yk}}{y_{rg^{\prime}o}^{k}}} + {\sum\limits_{j = 1}^{J_{k}}\frac{\tau_{jg^{\prime}o}^{bk}}{b_{jg^{\prime}o}^{k}}}} \right)} \right\rbrack}} \\ {s.t.\ x_{mg^{\prime}o}^{k} - {\sum\limits_{g = 1}^{G}{\sum\limits_{n \in g^{\prime},n \neq o\ if\ g = g^{\prime}}{\lambda_{gn}^{k}x_{mgn}^{k}}}} + \tau_{mg^{\prime}o}^{xk} \geq 0,m = 1,2,\cdots,M_{k};k = 1,2,\cdots,K;} \\ {{\sum\limits_{g = 1}^{G}{\sum\limits_{n \in g^{\prime},n \neq o\ if\ g = g^{\prime}}{\gamma_{gn}^{k}y_{rgn}^{k}}}} - y_{rg^{\prime}o}^{k} + \tau_{rg^{\prime}o}^{yk} \geq 0,r = 1,2,\cdots,R_{k};k = 1,2,\cdots,K;} \\ {b_{jg^{\prime}o}^{k} - {\sum\limits_{g = 1}^{G}{\sum\limits_{n \in g^{\prime},n \neq o\ if\ g = g^{\prime}}{\gamma_{gn}^{k}b_{jgn}^{k}}}} + \tau_{jg^{\prime}o}^{bk} \geq 0,j = 1,2,\cdots,J_{k};k = 1,2,\cdots,K;} \\ {1 - \frac{1}{R_{k} + J_{k}}\left( {{\sum\limits_{r = 1}^{R_{k}}\frac{\tau_{rg^{\prime}}^{yk}}{y_{rg^{\prime}o}^{k}}} + {\sum\limits_{j = 1}^{J_{k}}\frac{\tau_{jg^{\prime}}^{bk}}{b_{jg^{\prime}o}^{k}}}} \right) \geq \varepsilon,k = 1,2,\cdots,K;} \\ {{\sum\limits_{g = 1}^{G}{\sum\limits_{n \in g^{\prime},n \neq o\ if\ g = g^{\prime}}{\lambda_{gn}^{k}z_{gn}^{({k - 1,k})}}}} = {\sum\limits_{g = 1}^{G}{\sum\limits_{n \in g^{\prime},n \neq o\ if\ g = g^{\prime}}{\gamma_{gn}^{k}z_{gn}^{({k,k - 1})}}}},k = 1,2,\cdots,K;} \\ {{\sum\limits_{g = 1}^{G}{\sum\limits_{n \in (g^{\prime} = 1),n \neq o\ if\ g = g^{\prime}}\lambda_{gn}^{k}}} = \phi_{1}^{k},{\sum\limits_{g = 1}^{G}{\sum\limits_{n \in (g^{\prime} = 2),n \neq o\ if\ g = g^{\prime}}\lambda_{gn}^{k}}} = \phi_{2}^{k},\cdots,{\sum\limits_{g = 1}^{G}{\sum\limits_{n \in (g^{\prime} = G),n \neq o\ if\ g = g^{\prime}}\lambda_{gn}^{k}}} = \phi_{G}^{k};} \\ {{\sum\limits_{g = 1}^{G}{\sum\limits_{n \in (g^{\prime} = 1),n \neq o\ if\ g = g^{\prime}}\gamma_{gn}^{k}}} = \varphi_{1}^{k},{\sum\limits_{g = 1}^{G}{\sum\limits_{n \in (g^{\prime} = 2),n \neq o\ if\ g = g^{\prime}}\gamma_{gn}^{k}}} = \varphi_{2}^{k},\cdots,{\sum\limits_{g = 1}^{G}{\sum\limits_{n \in (g^{\prime} = G),n \neq o\ if\ g = g^{\prime}}\gamma_{gn}^{k}}} = \varphi_{G}^{k};} \\ {{\sum\limits_{k = 1}^{K}{\sum\limits_{g = 1}^{G}\phi_{g}^{k}}} = 1;{\sum\limits_{k = 1}^{K}{\sum\limits_{g = 1}^{G}\varphi_{g}^{k}}} = 1;\phi_{g}^{k} = 1or0;\varphi_{g}^{k} = 1or0;} \\ {{\sum\limits_{k = 1}^{K}\omega^{k}} = 1,k = 1,2,\cdots,K;} \\ {\omega^{k},\tau^{xk},\tau^{yk},\tau^{bk},\lambda^{k},\gamma^{k} \geq 0;k = 1,2,\cdots,K.} \\ \end{array}$$ where *t*^*x*^(*s*^*x*^,*τ*^*x*^), *t*^*y*^(*s*^*y*^,*τ*^*y*^) and *t*^*b*^(*s*^*b*^,*τ*^*b*^) are the slacks of the inputs, desirable outputs, and undesirable outputs with respect to group frontiers (concave metafrontier and non-concave metafrontier), respectively. The term *ε* is non-Archimedean infinitely small, and the corresponding constraint ensures that the denominator in the objective function is greater than zero. For the sake of considering the continuity of the production activities of the two stages, the linkage between the productivity stage and the profitability stage is added in the proposed models. Specifically, we have: $$\sum\limits_{g = 1}^{G}\sum\limits_{n \in g^{\prime},n \neq o\ if\ g = g^{\prime}}\lambda_{gn}^{k}z_{gn}^{({k - 1,k})} = \sum\limits_{g = 1}^{G}\sum\limits_{n \in g^{\prime},n \neq o\ if\ g = g^{\prime}}\gamma_{gn}^{k}z_{gn}^{({k,k - 1})},k = 1,2,\cdots,K.$$ Considering a non-concave metafrontier, the divisional efficiency score of each stage is computed as follows: $$\rho_{g^{\prime}o}^{nc - meta*1} = \frac{1 + \frac{1}{M_{1}}\left( {\sum\limits_{m = 1}^{M_{1}}\frac{\tau_{mg^{\prime}o}^{x1*}}{x_{mg^{\prime}o}^{1}}} \right)}{1 - \frac{1}{Q}\left( {\sum\limits_{q = 1}^{Q}\frac{\tau_{qg^{\prime}o}^{z*}}{z_{qg^{\prime}o}}} \right)};$$ $$\rho_{g^{\prime}o}^{nc - meta*2} = \frac{1 + \frac{1}{Q}\left( {\sum\limits_{q = 1}^{Q}\frac{\tau_{qg^{\prime}o}^{z*}}{z_{qg^{\prime}o}}} \right)}{\left\lbrack {1 - \frac{1}{R_{2} + J_{2}}\left( {{\sum\limits_{r = 1}^{R_{2}}\frac{\tau_{rg^{\prime}o}^{y2*}}{y_{rg^{\prime}o}^{2}}} + {\sum\limits_{j = 1}^{J_{2}}\frac{\tau_{jg^{\prime}o}^{b2*}}{b_{jg^{\prime}o}^{2}}}} \right)} \right\rbrack}.$$ where the slack variables with the superscript "\*" denote the optimal slacks of model ([8](#pone.0204559.e023){ref-type="disp-formula"}) in the corresponding stages. The technology gap ratio (TGR) based on the optimal objective values $\rho_{g^{\prime}o}^{group*}$ and $\rho_{g^{\prime}o}^{nc - meta*}$ can be computed as follows: $$TGR_{g^{\prime}o}^{nc} = \frac{\rho_{g^{\prime}o}^{nc - meta*}}{\rho_{g^{\prime}o}^{group*}}.$$ The TGR with respect to each stage under a concave metafrontier is given by: $$TGR_{g^{\prime}o}^{nc1} = \frac{\rho_{g^{\prime}o}^{nc - meta*1}}{\rho_{g^{\prime}o}^{group*1}};$$ $$TGR_{g^{\prime}o}^{nc2} = \frac{\rho_{g^{\prime}o}^{nc - meta*2}}{\rho_{g^{\prime}o}^{group*2}}.$$ Dagum's decomposition and Gini coefficient for subpopulations {#sec006} ------------------------------------------------------------- The Gini decomposition method proposed by Dagum \[[@pone.0204559.ref064]\] can be used to describe the sources of between group disparity and the distribution of subsamples free from the influence of sample overlap. Eq ([24](#pone.0204559.e043){ref-type="disp-formula"}) below can express it: $$G = \frac{\sum\limits_{j = 1}^{k}{\sum\limits_{h = 1}^{k}{\sum\limits_{i = 1}^{n_{j}}{\sum\limits_{r = 1}^{n_{k}}\left| {y_{ji} - y_{hr}} \right|}}}}{2\mu n^{2}}$$ where *y*~*ji*~(*y*~*hr*~) is the efficiency score of a bank in the group *j*(*h*), *μ* is the mean value of banking efficiency, *n* is the total number of banks, *k* is the number of groups, and *n*~*j*~(*n*~*h*~) is the number of banks in the group *j*(*h*). The groups must be sorted by their average banking efficiency using Eq ([25](#pone.0204559.e044){ref-type="disp-formula"}) before performing Dagum's decomposition: $$\overline{Y_{1}} \leq \overline{Y_{2}} \leq \cdots \leq \overline{Y_{j}} \leq \cdots \leq \overline{Y_{k}}$$ The Gini coefficient can be decomposed into three components according to Dagum's approaches. 1) The contribution of differences within groups *G*~*w*~ (i.e., the spatial differences between the banking efficiency within groups) refers to such differences between groups within SOB, JSB, FB and CCB in China in this study. 2) The contribution of differences between groups *G*~*nb*~ (i.e., the bank efficiency differences between groups) refers to such differences between SOB, JSB, FB and CCB in this study. 3) The contribution of the intensity of transvariation *G*~*t*~ is the overlapping contribution of banking efficiency between groups. The components satisfy *G* = *G*~*w*~ + *G*~*nb*~ + *G*~*t*~. *G*~*jj*~ is the Gini coefficient within group *j*. *G*~*jh*~ is the Gini coefficient between groups *j* and *h*. Then, the contribution rate of *G*~*w*~, *G*~*nb*~ and *G*~*t*~ can be calculated as $\frac{G_{w}}{G} \times 100\%$, $\frac{G_{nb}}{G} \times 100\%$ and $\frac{G_{t}}{G} \times 100\%$, respectively. $$G_{jj} = \frac{\frac{1}{2\overline{Y_{j}}}{\sum\limits_{i = 1}^{n_{j}}{\sum\limits_{r = 1}^{n_{k}}\left| {y_{ji} - y_{hr}} \right|}}}{n_{j}^{2}}$$ $$G_{w} = {\sum\limits_{j = 1}^{k}{G_{jj}p_{j}s_{j}}}$$ $$G_{jh} = \frac{\sum\limits_{i = 1}^{n_{j}}{\sum\limits_{r = 1}^{n_{k}}\left| {y_{ji} - y_{hr}} \right|}}{n_{j}n_{h}(\overline{Y_{j}} + \overline{Y_{h}})}$$ $$G_{nb} = {\sum\limits_{j = 2}^{k}{\sum\limits_{h = 1}^{j - 1}{G_{jh}(p_{j}s_{h} + p_{h}s_{j})D_{jh}}}}$$ $$G_{t} = {\sum\limits_{j = 2}^{k}{\sum\limits_{h = 1}^{j - 1}{G_{jh}(p_{j}s_{h} + p_{h}s_{j})(1 - D_{jh})}}}$$ $$D_{jh} = \frac{d_{jh} - p_{jh}}{d_{jh} + p_{jh}}$$ where *p*~*j*~ = *n*~*j*~/*n*, $s_{j} = {{n_{j}\overline{Y_{j}}}/n}\overline{Y}$, and *j* = 1,2,⋯,*k*. *D*~*jh*~ is the relative contribution rate of banking efficiency between groups *j* and *h*. *d*~*jh*~ is the difference of the contribution rates of banking efficiency between groups (i.e., the weighted average of all samples with *y*~*ji*~ − *y*~*hr*~ \> 0 in groups *j* and *h*). *p*~*jh*~ is the first-order moment of transvariation (i.e., the weighted average of all samples with *y*~*hr*~ − *y*~*ji*~ \> 0 in groups *j* and *h*). $$d_{jh} = {\int\limits_{0}^{\infty}{dF_{j}(y)}}{\int\limits_{0}^{y}{(y - x)dF_{h}(x)}};$$ $$p_{jh} = {\int\limits_{0}^{\infty}{dF_{h}(y)}}{\int\limits_{0}^{y}{(y - x)dF_{j}(x)}}$$ where *F*~*j*~ and *F*~*h*~ are the cumulative density distribution functions of groups *j* and *h*, respectively. Kernel density estimation {#sec007} ------------------------- Kernel density estimation (KDE) is an important approach to nonparametric estimation. It describes the general distribution of a random variable using a continuous density curve obtained by estimating its probabilistic density \[[@pone.0204559.ref065]\]. Specifically, suppose *x*~1~,*x*~2~,⋯,*x*~*n*~ are samples from a continuous population *X*. Then, the KDE for the population density function *f*(*x*) at any point *x* can be defined as: $$\hat{f_{h}}(x) = \frac{1}{nh}{\sum\limits_{i = 1}^{n}{K\left( \frac{x - x_{i}}{h} \right)}}$$ where *K*(∙) is the kernel function, and *h* is the bandwidth. Since the shape of the kernel function has little effect on the accuracy of the estimation result, this study bases its estimation on the Gaussian kernel function \[[@pone.0204559.ref066]\]: $$K(x) = \frac{1}{\sqrt{2\pi}}\exp\left( {- \frac{x^{2}}{2}} \right)$$ Empirical analysis {#sec008} ================== Data sources and variable descriptions {#sec009} -------------------------------------- The datasets used in this paper include 93 commercial banks from China's mainland taken from the Bureau van Dijk (BvD) (Bankscope) over the period of 2005--2016. The appendix includes a sample list. Following previous studies \[[@pone.0204559.ref002], [@pone.0204559.ref057]\], we use three inputs: fixed assets (*fixed_asset*), equity (*equity*), and personnel expenses (*personnel_expenses*). The deposits (*deposits*) are treated as intermediates in the two-stage network DEA framework. The desirable outputs are gross loans (*gross_loans*) and other earning assets (*other_earning_assets*). Non-performing loans (*NPLs*) are an undesirable output. All financial data are deflated using the GDP deflator with a base = 100 in 2005. The descriptive statistics of all inputs, intermediates, and outputs are summarized in [Table 1](#pone.0204559.t001){ref-type="table"}, which reveals that significant differences exist among banks. 10.1371/journal.pone.0204559.t001 ###### Descriptive statistics of samples (Unit: Million RMB). {#pone.0204559.t001g} Total Total SOB JSB FB CCB ------------------------ -------- ------------- ----------- ------------ ------------ ---------- ----------- **Original inputs** *fixed_asset* 0.0740 1339.7830 63.5013 1002.9660 120.2605 1.4609 15.9305 *equity* 1.5590 5761.0680 328.6731 3592.2850 947.2279 51.1371 125.7505 *personnel_expenses* 0.0200 576.1070 30.7610 313.2472 105.8187 3.9377 11.0294 **Intermediates** *deposits* 7.0360 107182.7000 5371.0440 62074.6800 17626.1500 325.4474 1711.6410 **Final outputs** *gross_loans* 3.5610 64309.5900 3209.6400 39371.4400 10104.3800 212.9052 901.2157 *other_earning_assets* 1.5230 44212.8500 2176.4720 24714.9200 7187.1240 166.7921 698.5564 *npls* 0.0160 13397.8300 322.3043 6638.5430 256.8248 1.8208 21.8842 SOB, JSB, FB and CCB represent state-owned banks, joint-stock banks, foreign banks and city commercial banks, respectively. ### Evaluation and evolution of bank efficiency {#sec010} Different groups may reflect different productivity performances, and the efficiency scores may vary among different banks. As reported in [Table 2](#pone.0204559.t002){ref-type="table"}, on average, the pooled descriptive statistics of overall efficiency indicate that SOBs perform the best, followed by JSBs, and the FBs are the worst. However, the productivity efficiency of the JSBs is the highest, followed by SOBs, and the FB's is the smallest. Meanwhile, the controversy rank can be found in the profitability stage. Thus, for different bank types, the efficiency development is unbalanced, especially for FBs and CCBs, with respect to a non-concave metafrontier. [Table 3](#pone.0204559.t003){ref-type="table"} presents both a paired *t*-test and a sign rank test that show that there are significance differences among the four groups in different stages. Consequently, it is necessary to consider heterogeneity when measuring banking efficiency. 10.1371/journal.pone.0204559.t002 ###### Pooled descriptive statistics of overall efficiency and stage efficiencies. {#pone.0204559.t002g} Overall Stage 1 Stage 2 ------------- -------------------------- ------ --------- --------- --------- -------- -------- -------- Total banks Non-Concave Metafrontier 1116 0.4977 0.2520 0.0373 1.0281 0.6339 0.5935 Group frontier 1116 0.5264 0.2495 0.0373 1.0514 0.6659 0.7942 Technology Gap Ratio 1116 0.9452 0.1186 0.4665 1.0000 0.9446 0.7753 SOB Non-Concave Metafrontier 48 0.8369 0.1871 0.4916 1.0281 0.8823 0.1130 Group frontier 48 0.8957 0.1100 0.6882 1.0514 0.9392 0.9521 Technology Gap Ratio 48 0.9280 0.1425 0.4676 1.0000 0.9377 0.1201 JSB Non-Concave Metafrontier 108 0.8312 0.1157 0.6710 1.0052 0.9168 0.4052 Group frontier 108 0.8317 0.1166 0.6710 1.0505 0.9170 0.9118 Technology Gap Ratio 108 0.9995 0.0047 0.9520 1.0000 0.9998 0.4555 FB Non-Concave Metafrontier 384 0.4084 0.2076 0.0373 1.0000 0.5082 0.7165 Group frontier 384 0.4092 0.2083 0.0373 1.0514 0.5573 0.7871 Technology Gap Ratio 384 0.9982 0.0124 0.8858 1.0000 0.8937 0.9433 CCB Non-Concave Metafrontier 576 0.4665 0.2288 0.1374 1.0000 0.6441 0.5868 Group frontier 576 0.5165 0.2239 0.1402 1.0514 0.6684 0.7638 Technology Gap Ratio 576 0.9012 0.1453 0.4665 1.0000 0.9687 0.7778 10.1371/journal.pone.0204559.t003 ###### The comparison of stage efficiencies of different groups. {#pone.0204559.t003g} Groups *t*-test (*t*-value) Mean difference Sign rank test(Z-value) -------------------------- ------------------ --------------------------------------------------- ----------------- --------------------------------------------------- Overall efficiency SOB *versus* JSB 0.2323[\*\*\*](#t003fn001){ref-type="table-fn"} 0.0057 1.5100 SOB *versus* FB 13.6209[\*\*\*](#t003fn001){ref-type="table-fn"} 0.4284 9.4120[\*\*\*](#t003fn001){ref-type="table-fn"} SOB *versus* CCB 10.9138[\*\*\*](#t003fn001){ref-type="table-fn"} 0.3704 8.6790[\*\*\*](#t003fn001){ref-type="table-fn"} JSB *versus* FB 20.2850[\*\*\*](#t003fn001){ref-type="table-fn"} 0.4227 13.7860[\*\*\*](#t003fn001){ref-type="table-fn"} JSB *versus* CCB 16.1756[\*\*\*](#t003fn001){ref-type="table-fn"} 0.3647 13.0390[\*\*\*](#t003fn001){ref-type="table-fn"} FB *versus* CCB -3.9950[\*\*\*](#t003fn001){ref-type="table-fn"} -0.0580 -3.9110[\*\*\*](#t003fn001){ref-type="table-fn"} Productivity efficiency SOB *versus* JSB -1.5761 -0.0345 -0.1790 SOB *versus* FB 9.0732[\*\*\*](#t003fn001){ref-type="table-fn"} 0.3741 8.2520[\*\*\*](#t003fn001){ref-type="table-fn"} SOB *versus* CCB 7.4795[\*\*\*](#t003fn001){ref-type="table-fn"} 0.2382 7.6690[\*\*\*](#t003fn001){ref-type="table-fn"} JSB *versus* FB 14.7253[\*\*\*](#t003fn001){ref-type="table-fn"} 0.4086 11.9340[\*\*\*](#t003fn001){ref-type="table-fn"} JSB *versus* CCB 12.7071[\*\*\*](#t003fn001){ref-type="table-fn"} 0.2727 11.5020[\*\*\*](#t003fn001){ref-type="table-fn"} FB *versus* CCB -8.4326[\*\*\*](#t003fn001){ref-type="table-fn"} -0.1359 -7.5910[\*\*\*](#t003fn001){ref-type="table-fn"} Profitability efficiency SOB *versus* JSB -9.3232[\*\*\*](#t003fn001){ref-type="table-fn"} -0.2922 -8.8580[\*\*\*](#t003fn001){ref-type="table-fn"} SOB *versus* FB -20.6942[\*\*\*](#t003fn001){ref-type="table-fn"} -0.6035 -11.3270[\*\*\*](#t003fn001){ref-type="table-fn"} SOB *versus* CCB -23.9605[\*\*\*](#t003fn001){ref-type="table-fn"} -0.4737 -11.5200[\*\*\*](#t003fn001){ref-type="table-fn"} JSB *versus* FB -13.9715[\*\*\*](#t003fn001){ref-type="table-fn"} -0.3113 -10.8610[\*\*\*](#t003fn001){ref-type="table-fn"} JSB *versus* CCB -11.4197[\*\*\*](#t003fn001){ref-type="table-fn"} -0.1816 -7.7570[\*\*\*](#t003fn001){ref-type="table-fn"} FB *versus* CCB 11.8982[\*\*\*](#t003fn001){ref-type="table-fn"} 0.1298 11.0900[\*\*\*](#t003fn001){ref-type="table-fn"} \*\*\*, \*\*, and \* denote significance at the levels of 1%, 5%, and 10%, respectively. In terms of the overall efficiency, the TGR of the JSB group is highest at 0.9995, FB comes in second at 0.9982, SOB comes in third at 0.9280, and CCB is the lowest at 0.9012. The metafrontier implies the potential payoffs from the meta-technology for each heterogeneous group \[[@pone.0204559.ref045]\]. The results of TGR evaluation indicate that JSB achieves 99.95% of the potential payoffs, which is greater than the case with the other technology groups. This indicates that the technology of JSB is the best technology type relative to the other types since it can produce the most outputs under the given input level. The percentages of achievable potential payoffs in FB, SOB and CCB are 99.82%, 92.80% and 90.12%, respectively. Moreover, regarding productivity efficiency (stage 1), the percentages of achievable potential payoffs in SOB, JSB, FB and CCB are 93.77%, 99.98%, 89.73% and 96.87%, respectively This declines to 12.01%, 45.55% and 77.78% for SOB, JSB, and CCB, respectively, with respect to profitability efficiency (stage 2). The above findings indicate that most of the banks have a large space for improvement, especially for SOB and JSB in the profitability stage. [Fig 3](#pone.0204559.g003){ref-type="fig"} displays the sources and contribution rates of the overall difference in the four groups. Since we mainly focus on the evaluation and evolution of bank efficiency with respect to the non-concave metafrontier, we do not provide the evolution of the banking efficiency with respect to the group frontier and TGR. From 2005 to 2016, the contribution rate of the between group differences and the intensity of transvariation showed a fluctuating tendency, while the within group differences had a contribution rate that was stable. The curve of the intensity of transvariation was always the highest, thus making it the major source of the overall differences in banking efficiency in mainland China. For overall efficiency, productivity efficiency and profitability efficiency, the gap between the group differences and the intensity of transvariation decreased until 2007 and then increased after 2007. However, the contribution rate of the intensity of transvariation is stable and at a lower rate (smaller than 20%) for each type of efficiency. {#pone.0204559.g003} [Fig 4](#pone.0204559.g004){ref-type="fig"} shows the evolutionary path of the differences in the 93 banks' efficiencies during 2005--2016. Generally, the basic characteristics of the kernel density curve have similar shapes for the overall efficiency, productivity efficiency and profitability, including the shape of the curve, the location of the curve and the peak values. More importantly, the central point of the density function is located at approximately 0.5, indicating that the level of banking efficiency is not very high overall. In addition, there is one major peak value and the interval of the variation changed little. This indicated that the banking efficiency presents a trend of polarization and the group differences changed little from 2005 to 2016. With regard to the overall efficiency (productivity efficiency and profitability efficiency are similar), we specifically get the following. {#pone.0204559.g004} Peak values and numbers. From 2005 to 2016, the peaks of the distribution curve showed an increasing trend. The increase of a peak's corresponding area indicates an increase in each bank's efficiency in the period. There are a prominent peak and a side peak in 2005, thus indicating that there are significant differences in the banking efficiency in China. In addition, there are a prominent peak and more than one side peak since 2008, thus showing that the multilevel differentiation phenomenon of bank efficiency appears compared to the initial period. Shape of the curve. The tail of the banking efficiency distribution in China becomes longer over time, thus indicating that the gaps in banking efficiency among different groups are gradually increasing during the study period. Conclusions and directions for further research {#sec011} =============================================== The primary concerns of this paper are the evaluation and evolution of banking efficiency in China over the period of 2005--2016. On the one hand, considering a non-concave metafrontier framework and super efficiency simultaneously in a network SBM model to open the black box of traditional DEA methods provides more accurate and comprehensive measurements of banking efficiency. This paper extends the US-NSBM model introduced by Huang et al. \[[@pone.0204559.ref002]\] to a new two-stage network model called NCMeta-US-NSBM by combining the model with a non-concave metafrontier, undesirable outputs, and super efficiency. An empirical analysis of the proposed model is provided that is based on the data of Chinese commercial banks from 2005 to 2016 (complete panel data; 93 banks, and 1116 observations). On the other hand, we employ the Dagum Gini index decomposition method and kernel density estimation technique to investigate the evolution of banking efficiency. The main empirical findings are summarized as follows. First, the statistical analysis shows that the disparity of efficiency occurs in banks in terms of the average level. Therefore, the efficiencies of different bank types are unbalanced. For the overall efficiency, SOB and JSB perform better than FB and CCB. The same conclusion can be found in the productivity stage, but a controversial result is obtained in the profitability stage. Second, both a paired *t*-test and a sign rank test show that there are significance differences among the four groups for overall efficiency, productivity efficiency and profitability efficiency. Third, the results of the TGR evaluation of SOB, JSB, and CCB in the productivity stage are higher than are those in the profitability stage, indicating that most of the banks have a large space for improvement, especially for SOB and JSB in the profitability stage. Finally, although the kernel density estimations for different efficiency scores have similar distributions in corresponding years, the multilevel differentiation phenomenon of bank efficiency may appear after 2008. For further research studies, our NCMeta-US-NSBM can be extended to measure and compare productivity changes for banks in different groups under the framework of the Malmquist--Luenberger productivity indicator. With the same metafrontier, these indicators are comparable and can provide insightful information. More specifically, they can show whether productivity change is driven by technological change or efficiency change, which has different implications for managers and policymakers. Essentially, the production process of banking industry may be treated as a complex network or dynamic rather than a two-stage mode or static, and it is a far-reaching attempt to introduce the complex network analysis \[[@pone.0204559.ref067]\] into the measurement of bank efficiency. Supporting information {#sec012} ====================== ###### Dataset of 93 banks in the sample. (DOCX) ###### Click here for additional data file. ###### Excel spreadsheet listing inputs and outputs for the proposed DEA model, along with the measures of banking efficiency. (XLSX) ###### Click here for additional data file. This research is supported by the National Natural Science Foundation of China (No. 41571524), the National Social Science Foundation of China (No. 15BGL041), the Social Science Foundation of Hunan Province (No. 15YBA014), and the Collaborative Innovation Center for the Development of Modern Services and New Urbanization in Hunan Province (No. CIC1502). [^1]: **Competing Interests:**The authors have declared that no competing interests exist.

CASE REPORT {#S0001} =========== A 68-year old woman presented with urge-incontinence, resistant to anticholinergic therapy. Urinalysis revealed a urinary tract infection caused by E. coli and cytology showed several neutrophils, but no malignant cells. Despite treatment with ciprofloxacin, the urge-incontinence persisted. Ultrasonography revealed right hydronephrosis and thickening of the bladder wall. In addition a cystoscopy was performed, displaying abnormal tissue at the level of the trigon and on both sides of the bladder wall. Transurethral biopsies were obtained revealing tumor cells of unknown primary origin. The immunohistochemical studies disclosed positivity for CK7, estrogen and progesterone receptors compatible with metastatic lobular carcinoma of the breast. In 1989 the patient had undergone a right radical mastectomy and adjuvant radiotherapy for a lobular breast carcinoma pT1N0M0. Analysis of the family history revealed that her mother had pancreatic cancer; her father esophageal cancer and her daughter a teratoma of the ovary. Genetic analysis for gene mutations, including BRCA1, was negative. Further staging using positron emission tomography (PET)-CT displayed local recurrence in the right breast and confirmed the unique metastatic location at the level of the trigon. Treatment with tamoxifen, 20 mg daily, was started with progressive improvement of the urge-incontinence within 5 months. A follow-up PET-scan showed decreased intensity of the breast lesion and trigon lesion. Nearly 10 months after diagnosis the patient is in clinical and partial radiological remission and the urge-incontinence has dissipated, while she has no overt toxicities from treatment with tamoxifen. DISCUSSION {#S0002} ========== The typical symptoms of urge-incontinence are frequent and urgent micturitions with incontinence. In most of the cases, no underlying cause is observed. However, when adequate behavioral attitude, pharmacological treatment and intense physiotherapy do not lead to improvement of urge-incontinence, an underlying disease should be ruled out. Primary bladder cancer should be excluded with a cystoscopy. Metastasis in the bladder occurs in 2% of all malignant disease and only 2.4% of these bladder metastases arise from a primary breast carcinoma \[[@CIT0001]\]. Only a few cases of an isolated bladder metastasis of breast cancer have been reported \[[@CIT0002]--[@CIT0009]\]. Various therapeutic options have been employed in these cases. Isolated metastases of primary breast carcinoma to the bladder are mostly treated by systemic chemotherapy \[[@CIT0002], [@CIT0003], [@CIT0007], [@CIT0009]\]. In only two estrogen-receptor positive cases, successful hormonal treatment with tamoxifen has been reported. Soon et al. reported an 87-year old patient presenting with genuine stress incontinence as a first sign of breast cancer \[[@CIT0005]\]. Unlike in our case, family history for breast cancer was positive. Choudhary et al. describe a patient with mixed incontinence, 18 years after an infiltrating ductal breast carcinoma \[[@CIT0006]\]. In both cases, a cystoscopy with biopsy revealed a metastasis of the breast, positive for estrogen and progesterone receptors. Tamoxifen treatment resulted in improvement of the urinary symptoms. No estrogen-receptor negative cases treated with tamoxifen have been described. In our case the patient suffered from a lobular subtype of breast cancer. Of the 19 cases with bladder metastases from breast cancer reviewed by Feldman et al. about one third had a lobular carcinoma subtype \[[@CIT0010]\]. The two cases previously described as successfully treated by tamoxifen presented one patient with a history of lobular subtype and another with a ductal subtype. This current third case report describes tamoxifen as a successful treatment of urge-incontinence, due to an isolated bladder metastasis. In the advent of an unknown cancer history or inconclusive malignant history, we propose to take a biopsy and to look for the presence of estrogen receptors as this offers an opportunity for a well-tolerated treatment that can produce long-term disease control and allows a conservative approach with regard to the local treatment. Zagha et al. promoted a more invasive approach \[[@CIT0004]\]. After discovering bladder metastasis of a primary breast cancer, they performed a partial cystectomy, followed by long-term treatment with fulvestrant. CONCLUSION {#S0003} ========== There is no standard therapy to the rare occurrence of isolated bladder metastasis from breast cancer. This current case report describes tamoxifen as a successful treatment of urge-incontinence due to an isolated bladder metastasis.

I. INTRODUCTION =============== Thyroid carcinoma is rare in children, especially in patients under 10 years of age (1). In Europe, its incidence is estimated at around 1--2 cases per million in children aged 10--14 years compared to 70 cases per million in adults (2). The incidence rate is higher in females than in males, with a sex ratio ranging from 2:1 in children aged 0--14 years to 3,5:1 between 15--19 years (3). From the histological point of view, four major types of thyroid cancer are recognized: papillary and follicular carcinoma (differentiated carcinoma), medullary carcinoma and anaplastic carcinoma. The aetiology of thyroid cancer is still not completely understood. Exposure to ionizing radiation may promote the development of a differentiated thyroid cancer, especially in children, probably due to increased sensitivity and proliferative activity of thyrocytes compared to adults (4). However, medullary carcinoma correlates with some hereditary factors such as the mutation of the RET proto-oncogene and recurs in families with MEN syndrome (multiple endocrine cancers). Follicular cancer is related to both dietary iodine deficiency and predisposing genetic factors that are not yet clear. The authors describe the diagnostic and therapeutic process of three clinical cases of patients suffering from papillary thyroid carcinoma, observed at University Hospital of Salerno. Case 1 ------ Male, 14 years old. At the age of 12, an indolent tumefaction was detected in the thyroid region. In another hospital he underwent a biopsy with a histological result of papilliferous carcinoma. Then, the child underwent subtotal thyroidectomy surgery at the same site. After 2 years, because of the appearance of some right latero-cervical tumefactions, he came to our department. At the physical examination, the right laterocervical region presented ovoid masses, arranged along the sternocleidomastoid muscle, with a well-defined contour, with an irregular anterior surface and a hard-elastic consistency, not adhering to the skin, but was united to the deep cutaneous tissues and it was painless. An otorhinolaryngological specialist consultation revealed a hemiparesis from the right vocal cord compensated by the left chord, probably related to a injury of the right recurrent nerve. We performed a neck ultrasound (US) and the chest and skeletal x-ray which excluded pleuro-parenchymal, mediastinal and bone alterations. Also thyroid scintigraphy was performed; it showed persistence of captive tissue in the thyroid region. The "Total Body" scintigraphic examination, performed with I131, revealed two modest adsorptive thyroid residues, one at the isthmus and the other less absorbing at the level of the upper portion of the right lobe of the thyroid, with no lesions in the rest of the organism. It was decided to schedule surgery to remove the two known formations and a lymph node package at the right jugular vein below the sternocleido-mastoid muscle. The histological examination of the removed lymph nodes confirmed neoplastic infiltration by papilliferous carcinoma ([Fig. 1](#f1-tm-22-028){ref-type="fig"}), while the two formations were made up of healthy thyroid tissue. The patient was discharged, on the 15th post-operative day, with iodine thyroglobulin replacement therapy. Periodic clinical, radiological and laboratory tests performed for up to three years were within normal limits. Case 2 ------ Female, 16 years old. She came to our observation for the presence of a right laterocervical tumefaction which had been persistent for a year. The three US controls performed after 1, 2 and 3 months revealed laterocervical lymph-adenomegalies with a maximum diameter of 17x10 mm, an inhomogeneous echo-structure and peripheral vascularization ([Fig. 2](#f2-tm-22-028){ref-type="fig"}). The thyroid, which was evaluated during the same tests, did not present any pathological element. At the local physical examination, we observed the presence of a right laterocervical oval tumefaction, of hard-elastic consistency, mobile on the underlying planes, slightly painful on palpation, not painful spontaneously. Because of the persistence of the clinical and diagnostic situation, it was decided to perform the surgical exeresis and biopsy of the lymph node. Histology reported tissue infiltration of papillary differentiated thyroid carcinoma. For this reason, after 20 days, a further ultrasound examination was performed and it revealed a nodule at the right thyroid lobe of 13 mm, with microcalcifications and intralesional vascularization, associated with lymph adenomegalies in the latero-cervical site and in the thyroid lodge and small nodules size in the isthmic site and left lobe. At this point, total thyroidectomy surgery was necessary with right laterocervical compartment resection and positioning of two drains at the level of the thyroid lodges. On the third postoperative day, the following parameters were measured: FT4, FT3, TSH, TG, anti-TG, anti-TPO, anti-TSH receptors, calcitonin; (within the normal limits). On the fourth and fifth day the two drainages were removed and on the seventh day the patient was discharged in good general condition and with replacement therapy with levothyroxine. She subsequently performed a cycle of radioiodiotherapy. At the follow-up at 1 and 3 months the patient appeared in good conditions. Case 3 ------ Female, 6 years old. At the age of 5 years, during a routine pediatric check, a nodule was noted in the right anterior region of the neck. For this reason, a thyroid ultrasound was performed, which highlighted the presence of "a single nodule at the level of the middle third level of the right thyroid lobe with a solid echo-structure, isoechogen with halo calcifications, vascularized type III, of the size of 7.8x10x14 mm ". An ago-biopsy was also performed but, since the examination was not executed in narcosis, the removal of sufficient material for diagnosis was not possible. The blood values of anti-TPO, anti-TG, TSH, FT4 and FT3 were normal. After 2 months, the ultrasound examination showed an increased isoechogenic nodule (17x12 mm), with hypoechogenic halo and peripheral vascularization, in a right thyroid lobe increased in volume. This picture was compatible with hyperplastic nodule. At the age of 6, the patient comes to our observation in good general condition. The physical examination of the neck showed a right front rounded formation, with a smooth surface, mobile on the levels below and with swallowing, not mobile with the tongue protrusion, not adhering to the skin and not sore nor painful. A new ultrasound examination ([Fig. 3](#f3-tm-22-028){ref-type="fig"}) confirmed a further right thyroid nodule increased in volume (21x13 mm), with internal calcifications and peripheral vascularization, and with reactive bilateral laterocervical lymphadenopathies (maximum diameter of 15 mm on the right and 13 mm to the left). It was therefore considered necessary to perform a new aspiration in narcosis, in order to have a clear histological diagnosis of this neoformation. The result was compatible with papilliferous carcinoma. Total thyroidectomy surgery was performed, with positioning of two drainages in the two thyroid lodges. FT3, FT4, TSH, anti-TG, FSH were all in the normal ranges. The chest x-ray was negative. On the second day, the patient underwent thyroid replacement therapy with levothyroxine. On the third and fourth day the two drainages were removed. The girl was discharged after a 7-day hospital stay. The child began radioiodiotherapy cycle. II. DISCUSSION ============== Differentiated thyroid carcinoma in children has a good prognosis in the majority of cases. In paediatric age, it is a rare condition and its annual incidence is about 0.5 cases/100,000 worldwide. The affected population is very heterogeneous, and it is necessary to distinguish epidemiologically and prognostically between prepubertal children and adolescents. The 5-year survival rate is estimated 98% in children aged 0--14 and 99% in adolescents aged 15--19 (3). We considered two cases in adolescence and one in prepubescent age. Papillary thyroid carcinoma in children differs significantly from adulthood particularly in terms of size and multifocality (5--6). Generally, children have large sizes and multifocal tumors; in our experience all cases presented an indolent mass at diagnosis (13mm in case 2 and 17x12mm in case 3). The main features of the three cases are summarized in [Table 1](#t1-tm-22-028){ref-type="table"}. In pediatric age there is an increased incidence of lymph node metastases at diagnosis, which we observed either in the 1st case or in the 2nd case. Moreover, in the 2nd case lymph node involvement was the first sign of illness. Considering the incidence of distant metastases this is higher in pediatric age compared to adults (7% of patients in the age \<20 years compared to 2% in subjects \> 20 years) with a higher risk of recurrence. In fact, at 16.6 years of follow-up, recurrence occurs in 40% of patients in whom the tumor occurred at \<20 years of age and in 20% of patients with 20 to 50 years at diagnosis(7). Despite the increased incidence of relapse ( 8) and the increased number of metastases at diagnosis, the survival of children with differentiated thyroid cancer is greater than in adults. This better prognosis is certainly to be connected with a greater responsiveness of children's functionally and metabolically active cancer cells to radioiodiotherapy and a more effective immune response of the body. In conclusion, compared to adults, in children there is a clear prevalence of the papillary type against the follicular, an element that can be linked to the child's increased thyroid susceptibility to ionizing radiation, which is at the basis of the onset of papillary carcinoma. Our choice to perform a radical surgery instead of the subtotal thyroidectomy in presence of lymph node involvement, arises precisely from the need to reduce the risk of recurrence. In fact, analyzing our 3 clinical cases we observed that in case 1 the subtotal thyroidectomy did not guarantee a complete removal of the tumor, which relapsed after two years. Because of the high incidence of lymph node metastasis and the increased frequency of relapse, early diagnosis is extremely important in differentiated thyroid carcinoma in the pediatric age. In all three cases an accurate diagnostic process, characterized by careful clinical examination, ultrasound examination and histological confirmation was performed. US may highlight nodules with a diameter of less than 15 mm, with irregular margins and increased intranodular vascularization (9, 10). However, in case 2 the initially US examination showed exclusively laterocervical lymphadenopathy without thyroid alterations, while a further US revealed a nodule at right thyroid lobe of 13 mm. Finally, through the fine-needle aspiration and the subsequent histological examination of the neoformation, it is possible to have a diagnosis of certainty. This examination presents a high sensitivity, (from 94% up to 100%) (11, 12), a specificity around 74.9% and a positive predictive value of 89% (13). The fine-needle aspiration allows us to evaluate the histopathological characteristics of a thyroid nodule and directs the therapeutic choice in order to avoid unnecessary surgical treatments in healthy children. In children with papillary thyroid carcinoma the treatment to perform is surgery. Although there is the possibility of complications in total thyroidectomy such as hypoparathyroidism (in case of ablation of the parathyroid glands) and dysphonia up to the aphonia (in case of injury of the right and left recurrent nerves) we did not find post-operative alterations. Total thyroidectomy is associated with the removal of the lymph nodes of the thyroid lodge, especially if they are macroscopically involved (14--15). A total thyroidectomy involves subsequent life-long replacement therapy. Controversial is the utility of laterocervical lymph nodes dissection that is mainly performed in case of certain diagnosis of tumor infiltration (16); two of our patients with lateral-cervical lymph node involvement performed the lymphadenectomy while case 3 underwent total thyroidectomy without lymph node dissection. It is also important consider the usefulness of post-surgical radioiodine therapy, as it has a low rate of complications or side effects and has a high specificity for residual or metastatic thyroid tissues. The association of surgery and radioiodiotherapy allows a higher selectivity on the residual cancer (5-18-19). For this reason, we subjected two patients (case 2 and 3) to post-surgical treatment with radioiodium. III. CONCLUSION =============== Differentiated thyroid carcinoma in children is a rare tumor, but not to be underestimated. The importance of early diagnosis is linked to the high frequency with which it given repetitions and relapses. However, proper treatments still allow a good survival rate and an excellent quality of life. The most appropriate therapeutic strategy should be implemented early and includes a first surgical phase of complete ablation of the gland associated or not with exeresis of the latero-cervical lymph nodes and a second phase of radioiodiotherapy that further completes the healing of young patients. {#f1-tm-22-028} {#f2-tm-22-028} {#f3-tm-22-028} ###### main features of the three cases Case 1 Case 2 Case 3 --------------------------------------- --------------------------------------------------------------------- ------------------------------------------------------------------ ------------------------------------------------- **Sex** Male Female Female **Age at diagnosis** 12 years old 16 years old 6 years old **Clinical presentation** Latero-cervical tumefactions Latero-cervical tumefaction which had been persistent for a year Nodule in the right anterior region of the neck **Lymph node involvement** Yes Yes No **Medical tests** Thyroid ultrasound, Ches t and skeletal x-ray, Thyroid scintigraphy Thyroid ultrasound, biopsy of the lymph node. Thyroid ultrasoud, ago-biopsy **Surgical treatment** -Total thyroidectomy - latero-cervical compartment resection -Total thyroidecto my - latero-cervical compartme nt resection -Total thyroidect omy **Post-surgical radioiodine therapy** No Yes Yes

Introduction {#Sec1} ============ It has been well established that the cyclical development of the mammary gland in mammals represents a valuable model of the processes underlying cell differentiation for the dynamic remodeling of the organ and of its function, in this case, the production and secretion of milk. This interest was further strengthened when in 2006 two studies showed the presence of a single pluripotent cell possessed the ability to completely regenerate the functional unit of the gland, the mammary alveolus^[@CR1],[@CR2]^. Further, a single cell clone has been reported to be able to differentiate in a bilayered system organized into alveolar structures connected by a common ductal system. The two cell layers are separate but contiguous throughout and consist of an inner layer of cytokeratin CK18^+^ luminal cells and an outer layer of cytokeratin CK14^+^ myoepithelial cells^[@CR2]^. In subsequent years, further studies were carried out on the characterization of this niche of mammary pluripotent stem cells in the human or in laboratory animals and on the maintenance and activation processes of adult stem cells both in physiological and pathological contexts^[@CR3]--[@CR6]^. The purification and isolation of these cells in other species is made more difficult due to the lack of species-specific cellular information and the availability of specific antibodies for the individual species although cell isolation and characterization has been reported in the bovine, caprine, canine and feline species ^[@CR7]--[@CR10]^. Interestingly, among mammalians species, with respect to the development and interaction of the alveolar functional unit and the extracellular matrix of the mammary gland, the morphological development of the mammary gland in humans and ruminants is similar^[@CR11]^, revealing the potential of ruminant species to serve as useful models for a better understanding of human breast development and cancer. In the phenotypic characterization of this small stem cell population, some surface markers appear to be established and in part common among species, such as alpha-6 integrin (CD49f) and CD24^[@CR12]--[@CR15]^ as well as intracellular enzymes which have been evaluated for some species (i.e. absence of ALDH1 activity^[@CR16]^. Furthermore, it is well accepted that this stem cell niche is localized in close association with the basal layer of the alveolus^[@CR17]^. P-Cadherin is a calcium-dependent cell--cell adhesion glycoprotein with a homeostatic function in several tissues and its name is derived from the site where this adhesion molecule was firstly characterized, the placenta^[@CR18]^. It exerts a crucial role in the conservation of the structural integrity of epithelial tissues and regulates processes involved in embryonic development and maintenance of adult tissue architecture, important for cell differentiation, cell shape, cell polarity, growth and migration^[@CR19]--[@CR21]^. P-Cadherin is of interest in that it plays a role both during embryonic development and during the processes of differentiation of adult tissue, maintaining its architecture through cell differentiation, polarity organization, the induction of migration and proliferation^[@CR22]^. Despite being expressed in human fetal structure^[@CR23],[@CR24]^, P-Cadherin is present in several adult tissues, usually co-expressed with E-Cadherin, such as the basal layer of the epidermis, the mammary gland and the prostate, as well as the mesothelium, the ovary, the cervix, the hair follicle, and the corneal endothelium^[@CR25]^. It has been reported that P-Cadherin may be involved in the biology of mammary stem cells as its expression was observed in a cellular subpopulation with a phenotype similar to the basal epithelial cells and which co-expressed with SLUG, a transcription factor that regulates cadherin genes and the genes involved in mammary epithelial cell lineages^[@CR26]^, but within the basal or myoepithelial layer where the niche of multipotent cells are believed to reside^[@CR27]--[@CR29]^. Moreover during the evolution of the gland, P-Cadherin is found in stem cells that give rise to the myoepithelial cell layer, termed cap cells^[@CR30],[@CR31]^. In contrast, in the mature lactating gland, P-Cadherin is expressed in human alveolar cells, the units responsible for milk protein production^[@CR32]^. To date, however, there is no functional evidence of the role of P-Cadherin for the generation of a mammary alveolus and normal function. While it is interesting to confirm the possible identification of a new marker linked to mammary stem cell populations, in particular for a species where few molecular markers are available, it is necessary to examine the cellular processes that occur in this cell niche in the presence of P-Cadherin. The aim of this research is to demonstrate the utility of P-Cadherin in the characterization of the cellular phenotype of adult stem cells in the mammary gland and to report which genes are activated in this role during the process of mammary differentiation through transcriptome analysis. Results {#Sec2} ======= Frequencies of the mammary epithelial subpopulations {#Sec3} ---------------------------------------------------- The expression of CD49f. and P-Cadherin were analyzed by flow cytometry in mammary epithelial cells recovered from eight Piedmontese heifers. After exclusion of dead cells, four subpopulations were identified according to the expression levels of both markers (Fig. [1](#Fig1){ref-type="fig"}). A double negative population (CD49f^neg^/P-Cadherin^neg^), a highly CD49f positive/P-Cadherin negative population (CD49f^high^/P-Cadherin^neg^), a medium CD49f/medium P-Cadherin positive population (CD49f^mid^/P-Cadherin^mid^) and a medium CD49f/highly P-Cadherin positive population (CD49f^mid^/P-Cadherin^high^) were identified in all samples analyzed. The distribution of these populations shows variability among samples: CD49f^neg^/P-Cadherin^neg^ ranged from 5 to 30.7%, CD49f^high^/P-Cadherin^neg^ ranged from 10.3 to 52.6%, CD49f^mid^/P-Cadherin^mid^ ranged from 8.6 to 40.6% and CD49f^mid^/P-Cadherin^high^ ranged from 3.9 to 40%.Figure 1Sorting strategy for separating mammary subpopulations. The figure shows representative scatterplots of a sample of bovine mammary epithelial cells and the general sorting strategy adopted. A broad gate was set for the physical parameters, then events that were positive for 7-AAD (dead cells) and CD45 + (leucocytes) were excluded from the population of interest. Cells were then separated according to the expression levels of CD49f and P-Cadherin. The expression of CD45 was also evaluated and CD45^+^ cells accounted for less than 5% of the starting population in all analyzed samples. Distribution of mammary progenitor cells among sorted fractions {#Sec4} --------------------------------------------------------------- As we have previously shown, in adult bovine mammary tissue, both stem cells and committed unipotent progenitor cells can be observed^[@CR38]^. A clonogenic assay such as the CFC allows for the detection of highly proliferating cells that are committed to either a myoepithelial or a luminal phenotype. We carried out the CFC assay with the four sorted fractions in order to investigate whether the progenitors that were present in the mammary tissue were associated with one or more of the phenotypes that we identified and therefore in which of the subpopulations they segregated. In all the dishes where cells from the CD49f^neg^/P-Cadherin^neg^ fractions were seeded, the overall frequency of colonies was lower than 0.05%, while within cell populations of the other fractions were seeded, the colony frequency ranged from 2.7 to 4.1% (Fig. [2](#Fig2){ref-type="fig"}i). Interestingly, when the colonies were scored according to their morphology as previously described \[both in the CD49f^mid^/P-Cadherin^mid^ and in the CD49f^mid^/P-Cadherin^high^ fractions\], no myoepithelial-like colonies could be observed and only luminal-like colonies were apparent. In the CD49f^igh^/P-Cadherin^neg^ fraction however, myoepithelial-like colonies could be observed and their relative frequency was 71.4 ± 3.8% (number of myoepithelial colonies / total number colonies × 100 ± s.e.) (Fig. [2](#Fig2){ref-type="fig"}ii).Figure 2Frequency distribution of mammary progenitors among sorted fractions. (**i**) frequency of mammary progenitors in the four isolated cell fraction is expressed as number of colonies per 1,000 cells seeded (*n* = 3) ± s.e. (**ii**) relative frequency of luminal and myoepithelial colonies within the CD49f^high^/P-Cadherin^neg^ subpopulation when cells were seeded immediately after sorting and after cells of the same fraction were engrafted in NOD/SCID mice. Mammary epithelial stem cells segregate in the CD49f^high^*/P-Cadherin*^neg^ subpopulation {#Sec5} ------------------------------------------------------------------------------------------ While the CFC assay is efficient in detecting highly proliferating cells of different lineages (i.e. committed progenitors), it is not suitable to detect cells that retain stem-like properties. Therefore cells from different subpopulations were xenografted into immunocompromised mice as an in vivo model that has been previously described^[@CR3],[@CR33]^. In a xenotrasplantation model, only mammary stem cells are able to engraft and regenerate outgrowths that mirror ducts and acini found in the bovine mammary tissue. In previous experiments, it was established that 7.5 × 10^5^ cells were an optimal number to obtain a sufficient number of outgrowths to enable analysis. Therefore for each sorted subpopulation, the number of cells that were engrafted was calculated as equal to the frequency of the subpopulation measured when sorting, applied to a 7.5 × 10^5^ total cell number. Three bovine mammary samples were selected and for each subpopulation, multiple gels were prepared. Two to three gels were engrafted into each kidney for a total of five to six gels per bovine sample/experiment. When recovered, two to three gels per subpopulation were dissociated and seeded in CFC assays to assess the progenitor cell content of the xenografts, while the remaining gels were paraffin-embedded and sectioned for immunohistological analysis. After inclusion and sectioning of the gels, outgrowths were detected only in those seeded with the CD49f^high^/P-Cadherin^neg^ fraction (*n* = 9) (Fig. [3](#Fig3){ref-type="fig"}). The regenerated structures exhibited an internal lumen. Both round-shaped and slightly elongated outgrowths were visible even in the same section and consisted of an inner layer of cuboidal cells and an outer layer of flat, more elongated CK14^+^ cells (Fig. [4](#Fig4){ref-type="fig"}). Occasionally, multilayered structures could be detected. In gels seeded with cells from the other fractions (*n* = 9 for each subpopulation), no outgrowths could be detected.Figure 3Histological sections of xenografted gels. Representative images of xenografted gels from each of the mammary epithelial cell subpopulations after 4 weeks of in vivo engraftment. Hematoxylin staining, scale bar is 250 µm.Figure 4Regenerated structures in CD49f^high^/P-Cadherin^neg^ seeded gels. (**i**) and (**ii**) show representative images of regenerated outgrowths in which an inner lumen and a double layer is visible. Hematoxylin staining, scale bar is 50 µm. (**iii**) and (**iv**) shows immunofluorescence staining for cytokeratin (CK) 14. CK14 is in green, nuclei are stained blue with DAPI, scale bar is 50 µm. By dissociating the recovered gels and seeding the cells in a CFC assay, evaluation could be undertaken of whether committed progenitor cells were produced by stem cells during the xenograft. Only CD49f^high^/P-Cadherin^neg^-engrafted cells were able to produce distinct colonies. Such colonies included myoepithelial-like (with a frequency of 36.7 ± 1.8%) and luminal-like ones (63.3 ± 6.9%) (Fig. [2](#Fig2){ref-type="fig"}ii). Whole transcriptome analysis of mammary subpopulations {#Sec6} ------------------------------------------------------ The first step in the analysis of the whole transcriptome of the four different populations consisted of fitting the mixed model. Since normalization of arrays occurred prior to analysis, no significant effect of cell population is expected, as was observed here (*P* = 0.36). Also no between-sample variation is expected, as supported by the small variance component estimate ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{\sigma }_{S}^{2} = 0.036 \pm 0.019$$\end{document}$). However as expected there was substantial overall between-gene variation ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{\sigma }_{G}^{2} = 5.44 \pm 0.070$$\end{document}$), and also substantial Gene.CellPop effects ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{\sigma }_{G.CP}^{2} = 0.672 \pm 0.006$$\end{document}$) suggesting the existence of DE genes. Differentially expressed genes {#Sec7} ------------------------------ Table [1](#Tab1){ref-type="table"} shows the results of fitting of the mixture model for DE and non-DE genes. Based on the model, a large proportion of genes showed evidence of differential expression from 23% of genes (CD49f^mid^/P-Cadherin^mid^) to 39.7% (CD49f^high^/P-Cadherin^neg^). Using the threshold of a probability of 0.8, individual genes were classified as DE or non-DE. The false discovery rates (FDR) of DEGs was also calculated using this threshold, for each cell population. Overall, it is estimated that 3.2% of genes identified as being DE are 'false discoveries'. A complete list of DE genes is available in Supplementary Table [1](#MOESM1){ref-type="media"}.Table 1Results of fitting of the mixture model for DE and non-DE genes.Cell populationCD49f^neg^\ P-cad^neg^CD49f^mid^\ P-cad^mid^CD49f^mid^\ P-cad^high^CD49f^high^\ P-cad^neg^π~1~0.2270.2250.3070.375σ~0~0.3020.2040.3500.261σ~1~1.2900.6121.1191.109*n*~DE~1,4339801,6442,669FDR0.0300.0410.0350.029π~1~ = estimated proportion of genes that are differentially expressed (DE); σ~0~ = SD of the Gene × Cell population gene effects in each cell population for the non-DE genes; σ~1~ = corresponding SD for DE genes; *n*~DE~ = number of DE genes detected; FDR = false discovery rate of a gene being DE based on a threshold probability of \> 0.8. A general comparison of the four different subpopulations was initially undertaken. Genes were clustered according to their pattern of expression and a heatmap of DE genes was generated (Fig. [5](#Fig5){ref-type="fig"}). A list of DE genes split into 8 clusters (3rd level of splitting, 2^3^) was generated and a functional analysis for enrichment for biological processes (BP) was carried out. The analysis was performed both on the annotated bovine transcriptome and by translating bovine transcripts in their human orthologs.Figure 5Clustering of DE genes. The graph shows the eight clusters of DE genes distributed among the four different mammary epithelial subpopulations that were used to perform the analysis for enrichment of biological processes, molecular function and cell components associated genes. When the bovine annotated transcripts were used, Cluster 3 showed an enrichment for transcripts associated with cell-to-cell communication (signal transduction, 68 transcripts; signaling, 71; cell communication, 71; regulation of cellular process, 92; cellular response to stimulus, 75). In Cluster 8, an enrichment in metabolic process (183 transcripts) was found. Cluster 2 showed an enrichment in genes associated with immune response and leukocytes activation although the number of transcripts per each specific category was between 3 and 9 and many transcripts were shared among the categories. When the analysis was repeated with the human orthologs, an increased number of categories of enrichment were found. Again, in Cluster 2 biological processes associated with an immune response and activation of different leukocyte populations (in particular lymphocyte) were observed. In contrast to what was identified when using bovine annotated transcripts, Cluster 3 showed an enrichment in biological processes again associated with immune response, similar to Cluster 2. In Cluster 4, biological processes associated with angiogenesis and cell junction assembly were observed to be enriched in these transcripts. More interestingly, in the same cluster, the following biological processes could also be found: anatomical structure formation involved in morphogenesis (104 transcripts), cell migration (114 transcripts), tube morphogenesis (81 transcripts), epithelial cell migration (34 transcripts), epithelial cell differentiation (47 transcripts), epithelial cell proliferation (36 transcripts). In Cluster 8, an enrichment in transcripts was found for biological processes either associated with chromosome/chromatid segregation (20 and 14 transcripts respectively), mitotic nuclear division (21 transcripts) and DNA repair (19 transcripts), or with cofactor metabolic process (46 transcripts), oxidation--reduction process (76 transcripts), carboxylic acid metabolic process (82 transcripts), cellular amide metabolic process (50 transcripts). When analyzing differentially expressed genes, 445 genes were overexpressed exclusively in the CD49f^high^/P-Cadherin^neg^ population while 333 were downregulated in the same subpopulation when compared to the other three (that is they were not detected as DE in the other subpopulations) (Supplementary Fig. [1](#MOESM2){ref-type="media"}). Among the overexpressed genes, FGFR1, members of MAP kinase family (MAP3K11, MAPK11), ADAMTS1 and ADAMTS7 metalloproteases, MEF2A transcription factor, CADM4 cell adhesion molecule and several RHO GTPases (RHOB, RHOD, RHOF, RHOJ) were detected. Notably, Fos was among the downregulated gene along with HOXC9. The same functional analysis for enrichment was subsequently performed on these two gene subsets. Only the analysis on the human orthologs was performed for better annotation. At first an analysis for enrichment of biological process-related genes was performed. No functional enrichment was found when the analysis was restricted to the subset of the upregulated genes. However, in the subset of the downregulated genes, an enrichment for genes associated with chromosome organization, mitotic cell cycle, mitotic cell cycle process and sister chromatid segregation was found (Supplementary Fig. [2](#MOESM2){ref-type="media"}i). When single genes in these enrichment categories were examined and their specific function investigated in the literature, many were related to chromatin condensation, kinetochore formation, histone acetylation and cell-cycle progression checkpoints. We then performed an analysis for enrichment of molecular function (MF)-related genes. In the subset of the upregulated genes, we could detect specific enrichments for genes associated with the cadherin binding and cell adhesion molecule binding functions (Supplementary Fig. [2](#MOESM2){ref-type="media"}ii). When single genes in these categories were examined and their function investigated in literature, most of them promoted cytoskeleton organization, cell adhesion, cell migration and focal adhesion. No statistically significant enrichment was found for MF in the downregulated subset of genes. Discussion {#Sec8} ========== This study reports for the first time the utility that P-Cadherin may have in the purification of adult stem cells of the mammary gland in the bovine species. P-Cadherin has been observed to be crucial for preserving the structural integrity of epithelia in general^[@CR22]^ and for maintaining important functions related to cellular differentiation and polarity, cell growth and cell migration also in adult tissues, although its role has proved prominent mostly during embryonic development^[@CR25],[@CR39]^. At the level of the mammary gland it has been reported that P-Cadherin is co-expressed with markers of the basal layer of the mammary alveolus in the mouse in association with CK5 and CD49f^[@CR40]^ but it has not been investigated in more detail if it has a relevance for the regulation of the stem and / or progenitor cells. Our results propose that P-Cadherin may have an influence in cellular differentiation and its segregation outside a progenitor-rich compartment is important for the characterization of stem cells in the basal mammary phenotype when analyzed along with CD49f, a recognized marker of mammary stemness. Our study reports that mammary stem cells exhibit a cellular phenotype (CD49f^high^ /P-Cadherin^neg^) which identifies a subpopulation of cells capable of regenerating polarized mammary acini unlike the other three subpopulations sorted with different levels of co-expression of the two membrane markers. The overall frequency of colonies was lower than 0.05% in the double negative cells, while when cells of the other fractions were seeded, it ranged from 2.7 to 4.1%. Such frequencies would exclude the presence of progenitors in the double negative compartment, while indicating their presence in the other three. While the overall frequency of progenitor cells did not differ among the three subpopulations, it is important to emphasis that colony types differed among the remaining cell fractions. Interestingly, the analysis of cellular colonies resulting from the clonal proliferation of these populations shows the capacity of the CD49f^high^/P-Cadherin^neg^ population to differentiate into luminal-only colonies and basal-only colonies as demonstrated by expression of markers that we previously reported^[@CR16],[@CR33]^. In fact, both in the CD49f^mid^/P-Cadherin^mid^ and in the CD49f^mid^/P-Cadherin^high^ fractions, no myoepithelial-like colonies could be found and only luminal-like colonies were apparent. In the CD49f^high^/P-Cadherin^neg^ fraction however, myoepithelial-like colonies could be detected with relative high frequency (71%). As for the out-growths, only CD49f^high^/P-Cadherin^neg^-engrafted cells were able to produce colonies at all. Moreover myoepithelial-like and luminal-like colonies were both evident. Such behavior adds proof to the stem identity of the engrafted cells from the CD49f^high^/P-Cadherin^neg^ subpopulation. The organization of the mammary stem niche in the bovine appears, as previously reported, more similar to the mammary stem niche of the humans rather than the murine^[@CR8],[@CR33]^. The observed prevalence of mixed colonies in assays of cells isolated from regenerated structures is similar to what we have seen in similar assays of human cells and likely reflects the early regenerative phase of the structures from which they were obtained. In addition, the niche requirements of these primitive bovine mammary stem cells appear to more closely resemble those of their human rather than their murine counterparts since they regenerated structures in the kidney capsule assay but not when injected directly into cleared mouse mammary fat pads^[@CR3],[@CR41]^. Higher expression of P-Cadherin seems to be restricted to luminal-only progenitors and therefore might play a role in luminal differentiation in adult mammary tissue. This observation has been reported in human cancer cell lines^[@CR28]^ that underlined the correlation with ALDH1, a well-known cancer stem cells marker^[@CR42]^ and an association with a phenotype of luminal progenitor more than stem cells. In our study a specific functional test for stem cells relate strictly the absence of P-Cadherin to a normal primitive niche. The CD49f^mid^/P-Cadherin^mid^ and the CD49f^mid^/P-Cadherin^high^ fractions do not give rise to myoepithelial colonies when seeded at clonal density. Such observations lead us to hypothesize that P-Cadherin expression becomes relevant when stem cells have already started differentiating towards the phenotype of the innermost layer of the alveolus, which is formed by secretory epithelial cells. Previously, similar observations have always been reported in the bovine for E-Cadherin for its co-expression with the luminal phenotype^[@CR14]^. However, we report here that P-Cadherin absence is associated with the maintenance of a stem state. Some authors have reported that mammary stem cells are associated with a high expression of CD49f^[@CR43]^. Therefore, separating different mammary subpopulations just by using CD49f can result in an enrichment in the stem cell fraction by selecting for the cells with the highest expression of this marker. However, such sorting strategy would suffer from a major drawback, specifically the variability in setting a gate for such a population. The use of a secondary marker, such as P-Cadherin, allows for a better separation of mammary subpopulations since P-Cadherin expression is different in the CD49f^mid^ and CD49f^high^ populations. This approach results in a higher purity of the mammary stem cell population. The similarity with the behavior reported above with the stem niche characterized for the human species allows us to hypothesize that this surface marker in association with CD49f increases the purity of cellular subpopulation for stemness. The transcriptome analysis of the different cell subpopulations highlights that P-Cadherin^neg^ cellular subpopulation has a downregulation of genes related to the progression through cell cycle, which is consistent with a quiescent nature of adult stem cells. We have reported that 445 genes were overexpressed in the stem-cell enriched mammary subpopulation identified in CD49f^high^/P-Cadherin^neg^ phenotype while 333 were downregulated in the same subpopulation. In particular, in the subset of the downregulated genes, an enrichment for genes associated with chromosome organization (such as SMC2), mitotic cell cycle (such as CDC23), mitotic cell cycle progress, sister chromatid segregation (such as SYCP3, KIF22 and AURKB), chromatin condensation, kinetochore formation (such as KIF2C and CENPQ), histone acetylation (such as EYA3 and HDAC11), and cell-cycle progression checkpoints (such as CCNE1 and MYB) was observed. It was reported that some types of stem cells, such as long-term hematopoietic stem cells and skeletal muscle stem cells, are characterized by prolonged periods of quiescence^[@CR44],[@CR45]^. The identity of the niche signals required for long-term stem cell quiescence, and the mechanisms that underlie the transition from quiescence to activation, are not completely understood however interesting results have been proposed^[@CR46]^ N-Cadherin and M-Cadherin have been proposed as niche-based regulators of SC quiescence assuming that cell--cell adhesion exerts a key role in niche-regulated stem cell behavior, with cadherin-based adhesive junctions anchoring stem cells to niche cells^[@CR47]^. In our study, a correlation for the activation of the mammary stem niche could be hypothesized due to the presence of P-Cadherin which would initiate a partial differentiation towards cellular precursors. In these mammary epithelial subpopulations that lack the expression of P-Cadherin we show that genes related to cell cycle are downregulated while genes related to cytoskeleton organization are upregulated, thus when this cadherin is expressed the cells loose the focal adhesion, start to migrate from their stem niche and proliferate. In conclusion, this study reports for the first time the role of P-Cadherin to isolate the adult mammary stem cells subpopulation at higher purity in association with CD49f, indicating how the expression of this cadherin is important for the recruitment towards a more differentiated phenotype and, as already reported for other cadherins, its absence is linked to the activation or shutdown of genes associated with cell quiescence. This observation could be useful for identifying new strategies for controlling stem cell mobilization in mammary tissue. The data discussed in this publication have been deposited in NCBI\'s Gene Expression Omnibus^[@CR48]^ and are accessible through GEO Series accession number GSE148447 (<https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE148447>. Materials and methods {#Sec9} ===================== Tissue collection and cell isolation {#Sec10} ------------------------------------ The study on cows/animals were carried out in accordance with relevant guidelines and regulations: whole udders were collected from Piedmontese breed cows (age ranging from 17 to 28 months) at a local abattoir within 2 h of the time of slaughter. Sample collection, after the slaughter, was performed with the authorization and under the supervision of representatives of the Veterinary Services of the Italian National Health Service branch of the Ministry of Health. Animal work described in this study (NOD-SCID mice experiments) has been reviewed and approved by the Italian Ministry of Health (Authorization note no. 149401894128) and by the Bioethics Committee of the University of Turin. Mammary tissue was then processed as previously described^[@CR33]^ in order to obtain a single cell suspension to be used for subsequent assays. Briefly, tissue samples were dissociated overnight in Dulbecco's Modified Eagle's Medium/Nutrient Mixture F12 Ham (DMEM/F12) 1:1 v/v mixture supplemented with 2% bovine serum albumin (BSA), 300 U/ml collagenase type IV, 100 U/ml hyaluronidase type IV-S, 100 U/ml penicillin and 100 μg/ml streptomycin (all from Sigma-Aldrich Corp., St. Louis, MO, USA). Epithelial aggregates were then separated by differential centrifugation and enzymatically digested with a 0.5 mg/ml trypsin solution supplemented with 0.2 mg/ml EDTA followed by a digestion with 5 mg/ml dispase and 100 µg/ml DNAseI (all from Sigma-Aldrich Corp.). Flow cytometry and sorting {#Sec11} -------------------------- After the dissociation of the epithelial organoids, a single cell suspension was obtained. The cell suspension was then further filtered through a 40-µm mesh. Cells were then counted, centrifuged and resuspended in a suitable volume of Hank's Balanced Salt Solution (HBSS) supplemented with 2% Fetal Bovine Serum (FBS, Gibco, Thermo Fisher Scientific, Waltham, MA USA). Cell concentration was kept lower than 0.5 × 10^7^ cells/ml. Antibodies were then added to the cell suspension at the following concentration: 5 µl/10^6^ cells CD49f.-PE (clone GOH3, Santa Cruz Biotechnology, Dallas, TX USA), 10 µl/10^6^ cells P-Cadherin-FITC (N-19, Santa Cruz Biotechnology), 20 µl/10^6^ cells CD45-Alexa647 (CACTB51A, Monoclonal Antibody Center---Washington State University, Pullman, WA USA that was conjugated with Alexa647 using an APEX Antibody Labeling kit from Thermo Fisher Scientific). Cells were incubated for 20 min at 4 °C in the dark and then washed with HBSS supplemented with 2% FBS and resuspended in a suitable volume of HBSS. Immediately before sorting 7-Aminoactinomycin D (7-AAD) was added to each tube for live/dead cells discrimination at a final concentration of 1 µg/ml. 7-AAD negative events were selected to exclude dead cells and CD45 negative events to exclude immune cells. 7-AAD^-^/CD45^-^ cells were then sorted according to the expression of CD49f. and P-Cadherin by use of a FACSAria III (BD-Becton Dickinson, Franklin Lakes, NJ USA) equipped with an 85-µm nozzle tip and three lasers (405, 488 and 633 nm). Colony forming cell (CFC) assay {#Sec12} ------------------------------- CFC assays were carried out as previously described^[@CR33]^. Briefly, single cell suspensions of bovine mammary epithelial cells were prepared from the sorted fractions and seeded at very low density (2 × 10^3^ cells per 60-mm dish) along with 2 × 10^5^ NIH 3T3 fibroblasts that were previously treated with mitomycin C (10 μg/ml for 2 h). Cells were cultured for 24 h in human EpiCult B medium (STEMCELL Technologies, Vancouver, BC, Canada) supplemented with 5% fetal bovine serum (FBS) and 10^-6^ M hydrocortisone, 100 U/ml penicillin and 100 μg/ml streptomycin (Sigma-Aldrich Corp). Subsequently the medium was replaced with fresh media but without FBS and cells were cultured for seven additional days. Colonies were then examined under a Leica DM IL inverted contrast phase microscope in order to assess their morphology and scored accordingly. Xenograft {#Sec13} --------- Immediately after sorting, cells from each fraction were resuspended and mixed with 10T1/2 mouse fetal fibroblasts that were previously treated with mitomycin C at a concentration of 2 μg/ml for 16 h. Cells were then embedded in rat tail collagen as previously described^[@CR33]^ and transferred under the kidney capsule of female NOD/SCID mice. Mice received a silicone pellet (MED-4011, NuSil Technology, Carpinteria, CA, USA) containing 2 mg of 17β-estradiol and 4 mg of progesterone (both from Sigma-Aldrich Corp) at the time of surgery. The gels were recovered 4 weeks later. Staining of paraffin embedded sections {#Sec14} -------------------------------------- When gels were recovered from mice, they were processed as previously described for paraffin embedding^[@CR33]^. Briefly, gels were fixed in 4% neutral buffered formalin overnight and subsequently dehydrated and paraffin embedded. Sections of 4 µm were cut and processed for either hematoxylin eosin staining or for immunofluorescence staining. After dewaxing, sections were placed in Tris--HCl buffered saline supplemented with Tween-20 (TBS-Tween, 0.1 M Tris HCl, 0.14 M NaCl, 0.05% Tween-20, pH7.6, all reagents from Sigma-Aldrich) for 10 min, then incubated at room temperature for 1 h in TBS-Tween supplemented with 10% goat serum (Sigma-Aldrich) and then for 1 h at room temperature with the primary antibody. Sections were then washed 3 times with TBS-Tween, the secondary antibody applied and incubated for another hour. Sections were then counterstained with 4′,6-diamidino-2-phenylindole (DAPI, Sigma-Aldrich) and then mounted. As primary antibody, human cytokeratin 14 (CK14, 1:500 dilution, polyclonal AF-64, Covance, Princeton, NJ, USA) was used, while as secondary antibody, AlexaFluor 488-labelled goat anti-rabbit IgG was used (1:500, Invitrogen, Thermo Fisher Scientific). RNA extraction and sequencing {#Sec15} ----------------------------- In order to perform a whole transcriptome analysis, total RNA was extracted from cells of the four different fractions. Cells were directly sorted in 1.5 ml tubes containing 1 ml of TRI Reagent (Sigma-Aldrich Corp). For each fraction, at least 5 × 10^5^ cells were collected and lysed for RNA extraction. RNA was purified by adding 200 µl of chloroform (Sigma-Aldrich Corp.) and centrifuging at 12,000×*g* for 15 min at 4 °C. The upper clear phase was recovered and RNA was precipitated with 500 µl of isopropanol (Sigma-Aldrich Corp.) followed by a wash with 70% ethanol (Sigma-Aldrich Corp.). The RNA pellet was then resuspended in DEPC water (approximately 20 µl) and quantified with a Nanodrop 2000 (Themo Fisher Scientific). RNA samples were then shipped to IRCCS Ospedale San Raffaele, Italy, where they were processed for an Illumina TruSeq sequencing protocol with a reads depth of 30 M and expression data were normalized as RPKM. Gene expression analysis {#Sec16} ------------------------ The data set allowed to compare patterns of gene expression across the four cell types, namely CD49f^−^/P-cad^-^ (*n* = 3); CD49f^+^/P-cad^+^ (*n* = 2); CD49f^++^/P-cad^-^ (*n* = 3); and CD49f^+^/P-cad^++^ (*n* = 3). Data analysis was conducted by using a two-stage approach, as outlined by Singh et al.^[@CR24]^ and Trabzuni et al.^[@CR34]^. Firstly, a large-scale linear mixed model was fitted to all the gene expression data, of the form$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {\text{logExpr}} = {\text{constant}} + {\text{CellPop}} + {\text{Sample}} + {\text{Gene}} + {\text{Gene}}.{\text{CellPop}} + \varepsilon $$\end{document}$$where logExpr is the logarithm of the expression value, CellPop is the fixed effect of the cell population type, Sample is the random effect of the array, Gene is the random effect of a particular gene, and Gene.CellPop is the specific random effect of a gene in a particular cell population, and ε is the random error. Of main interest it will be the estimates of the Gene.CellPop terms. Fitting of the linear mixed model was conducted using ASReml-R^[@CR35]^ within the R computing environment. The second stage of the analysis involved fitting a two-component mixture model for these Gene.CellPop effect estimates, separately for each cell population. The two components are a set of differentially expressed (DE) and non-differentially expressed (non-DE) genes. Genes are assigned as DE when the (posterior) probability of being DE exceeds 0.8. Following this, some descriptive approaches were used, particularly to investigate patterns of differential expression across the four cell population types. All analyses were conducted using R. Gene annotation and functional analysis {#Sec17} --------------------------------------- Genes named after their ENSEMBL ID have been translated to their common gene name in order to have the same identifier for all genes considered, the translation has been run with data from BioMart tool as in Ensembl Release 96 (April 2019) based on the bovine genes ARS-UCD1.2 assembly. Gene ontology enrichments and gene functional analysis have been conducted in R environment, release 3.6.1, through Bioconductor (<https://www.bioconductor.org/>) package ClusterProfiler, version 3.12.0, a 0.05 cutoff value has been chosen for false discovery rate values. Bovine and human functional annotation were based on org.Bt.eg.db^B^ and org.Hs.eg.db^C[@CR36]^, respectively, and homologene^D[@CR37]^ package has been used for cow-human gene orthology conversion. Supplementary information ========================= {#Sec18} Supplementary file1Supplementary file2Supplementary file3 **Publisher\'s note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information ========================= is available for this paper at 10.1038/s41598-020-71179-4. This work was supported by FIL 2015 and 2016 of the University of Turin and IRCA University of Sydney 2015. E.M.: Conception and design, writing manuscript. U.A.: Data analysis and interpretation. P.A.S.: financial support, manuscript revision. PCT: data analysis and interpretation. M.B.: Conception and design, writing manuscript, financial support. The authors declare no competing interests.

Introduction ============ Compared with endoscopic mucosal resection (EMR) for early gastric cancer (EGC), endoscopic submucosal dissection (ESD) has considerable advantages regarding rates of en-bloc resection, curative resection, and local recurrence [@JR221-1] [@JR221-2] [@JR221-3] [@JR221-4] [@JR221-5]. EMR has subsequently been replaced by ESD. The main problem with ESD using conventional knives is its technical difficulty. Consequently, it is associated with a high rate of complications, foremost of which are a long procedure time and the need for advanced endoscopic techniques [@JR221-6] [@JR221-7]. Conventional devices such as the insulation-tipped electrosurgical knife and needle knife gently push the knife to the tissue and cut using electrosurgical current. Because these tools lack the ability to grasp (accurate targeting and hemostatic effect) and to pull the target tissue (away from the proper muscle layer), they are associated with the potential for major complications such as perforation and bleeding [@JR221-8] [@JR221-9]. To reduce the risk of complications related to ESD using a conventional knife, Akahoshi and FUJIFILM developed the Clutch Cutter (CC), which can accurately grasp, pull, coagulate, and/or incise the targeted tissue using electrosurgical current [@JR221-10] [@JR221-11]. In our previous pilot study for early gastric neoplasms, we resected tumors safely and easily without unintentional incision by ESD using the CC (ESD-CC) [@JR221-12]. However, outcomes in a large number of patients with EGC treated by this new method of ESD-CC have not been previously reported. In this study, we assessed the clinical outcomes of ESD-CC for EGC in a large population. Patients and methods ==================== Inclusion criteria/curability criteria and ethical considerations ----------------------------------------------------------------- A total of 325 consecutive patients (228 men, 97 women; mean age 73.7 years, range 35 -- 95) with EGCs were treated by ESD-CC at Aso Iizuka Hospital from June 2007 to March 2014. They were enrolled based on the clinical indication criteria for ESD proposed by Gotoda et al. [@JR221-13] and the Japanese Gastric Cancer Association (JGCA) [@JR221-14] ([Table 1](#TB221-1){ref-type="table"}). Resection is considered as curative when all of the following post-ESD histopathological conditions are fulfilled: En-bloc resection, negative horizontal margin, negative vertical margin, no lymphovascular infiltration, and: A) "absolute indication group"-differentiated-type adenocarcinoma without ulcerative findings, of which the depth of invasion is diagnosed as mucosa (pT1a) and the diameter is ≤ 2 cm; B) "Expanded indication group" -- B1) differentiated type mucosal cancer (pT1a) without ulcer findings irrespective of tumor size, B2) differentiated type mucosal cancer (pT1a) with ulcer findings ≤ 3 cm in diameter, B3) differentiated type minute (\< 500 micron from the muscularis mucosae) submucosal invasive cancer (pT1b) ≤ 3 cm in size without ulceration, B4) undifferentiated type mucosal cancer (pT1a) ≤ 2 cm in size without ulceration. Gastrectomy with removal of lymph nodes was recommended for patients with EGCs that did not meet these criteria by post ESD-histological analysis (exclusion criteria group). To evaluate the learning curve of ESD-CC, 325 cases were grouped chronologically into five periods: (1^st^): cases 1 -- 50; (2^nd^): cases 51 -- 100; (3^rd^): cases 101 -- 150; (4^th^): cases 151 -- 200; (5^th^): cases 201 -- 325. This study was carried out at Aso Iizuka Hospital and was approved by its ethics committee. Written informed consent was obtained from all the patients in accordance with the Declaration of Helsinki. ###### JGCA indication criteria for ESD for early gastric cancer. ------------------------------------------------------------------- **Absolute indication**  Tumor size \< 2 cm, histologically of differentiated type,  T1a, ulcer (--) **Expanded indication**  (a) Tumor size \> 2 cm, histologically of differentiated type,  T1a, ulcer (--)  (b) Tumor size \< 3 cm, histologically of differentiated type,  T1a, ulcer ( + )  (c) Tumor size \< 2 cm, histologically of undifferentiated type,  T1a, ulcer (--) **Exclusion indication**  Not satisfying any of the above criteria ------------------------------------------------------------------- Clutch Cutter (CC) ------------------ The CC (DP2618DT, FUJIFILM Corporation, Tokyo, Japan) ([Fig. 1](#FI221-1){ref-type="fig"}) can grasp and cut or coagulate a piece of tissue with electrosurgical current. It has a 0.4 mm-wide and 3.5-mm or 5-mm long serrated cutting edge to facilitate grasping the tissue. The outer side of the forceps is insulated so that electrosurgical current energy is concentrated at the closed blade to avoid unintentional incision. Furthermore, the forceps is rotatable to the desired orientation. The diameter of this forceps is 2.7 mm. This device is available for all steps of ESD. The forced coagulation mode (VIO 300D; Erbe, Tübingen, Germany) 30 W (effect 3) was used for marking, endo cut Q mode (effect 2, duration 3, interval 1) was used for cutting (mucosal incision and submucosal dissection), and the soft coagulation mode 100 W (effect 5) was used for pre-cut coagulation and hemostatic treatment. {#FI221-1} ESD --- ESD was performed by two endoscopists (one endoscopist maneuvered the scope and the other endoscopist maneuvered the CC.). ESD-CC was carried out using a single-channel therapeutic endoscope (EG-450RD5, Fujifilm) or a two-channel multi-bending endoscope (GIF-2T240M; Olympus, Tokyo, Japan). A long transparent hood (F-01, Top Co. Ltd., Tokyo, Japan) was attached to the endoscopic tip to facilitate submucosal dissection by elevating the lesion. The ESD-CC technique was as follows ([Fig. 2](#FI221-2){ref-type="fig"}) (VTR. 1). Marking dots were placed a few millimeters outside the margin of the lesion by CC in closed mode. Next, hyaluronic acid solution (MucoUp: Johnson and Johnson Co., Tokyo, Japan) mixed with a small volume of epinephrine and indigo carmine dye was injected into the submucosal layer. A mucosal incision and submucosal dissection were performed to completely remove the lesion using the CC. The bleeding vessel (artery or vein) was grasped, pulled/lifted and coagulated with the CC using electrosurgical current to stop the bleeding. Finally, the lesion was completely resected. All cutting steps consisted of 1) grasping step; 2) pull or lift up step; 3) pre-cut coagulation step with soft coagulation (if existence of blood vessel was suspected); and 4) cutting step with endo cut Q ([Fig. 3](#FI221-3){ref-type="fig"}). {#FI221-2} {#FI221-3} Histopathological evaluation ---------------------------- The excised specimens were sectioned perpendicularly at 2-mm intervals. According to the Japanese classification for Gastric Cancer, the histology was divided into differentiated adenocarcinoma (well or moderately differentiated adenocarcinoma or papillary adenocarcinoma) or undifferentiated adenocarcinoma (poorly differentiated adenocarcinoma or signet-ring-cell carcinoma). Tumor size, depth of invasion, presence of ulcerative changes, lymphatic and vascular involvement, and tumor involvement to the horizontal and vertical margins were assessed. Assessment of therapeutic efficacy and complications ---------------------------------------------------- The operating time was calculated as the time from the beginning of submucosal injection to the end of submucosal dissection. Resections were defined as "en bloc" when a lesion was resected in one piece and the resection margins were macroscopically tumor-free. Extension of cancer cells to the resected margin was evaluated as R0 (en-bloc resection with the lateral and basal resection margins free of cancer), R1 (incomplete resection when the cancer extended into the lateral or basal margins), or Rx (not evaluable when the margins were not evaluable as a result of the artificial effects of coagulation necrosis or multi-piece resection). All patients stayed in the hospital for 7 days following the procedure, after which follow-up endoscopic examinations were conducted at 2 days, and 2 (or 3), 9, and 15 months, and annually thereafter. All patients were given a proton pump inhibitor for a minimum of 8 weeks. Statistical analysis -------------------- All data analysis was conducted with a statistical software package (SAS version 9.2 and JMP version 8.0.1, SAS Institute Inc, NC, USA) The significance of differences in the lesion's features and the R0 resection rate of ESD-CC was determined using the χ^2^test. The significance of any differences in the lesion's features and the mean operating time of ESD-CC were determined using the Kruskal-Wallis test. A *P* value of less than 0.05 was considered to be significant. Results ======= Clinicopathological characteristics are shown in [Table 2](#TB221-2){ref-type="table"}. The male : female ratio was 2.4 : 1 (228/97) and the patients' mean age was 73.7 years (range 35 -- 95 years). After histological analysis of the resected specimens obtained by this procedure, the 325 patients were divided into three groups ([Table 1](#TB221-1){ref-type="table"}): Group 1 fulfilled the traditionally accepted indications (absolute indication group); Group 2 did not meet the absolute indication criteria but did meet the expanded indication criteria (the expanded indication group), and Group 3 did not meet the proposed criteria (exclusion indication group). The 38 patients in the exclusion-indication group were urged to undergo surgery. Nineteen patients have been referred for gastrectomy with lymph node dissection. Postoperative histological findings were no gastric residual cancer without lymph node metastasis in 17 patients, no gastric residual cancer with lymph node metastasis in one patient, and residual cancer (subserosal cancer) without lymph node metastasis in the other patient. The remaining 19 patients refused surgery and were followed up.  ###### Clinicopathological characteristics (N = 325). --------------------------------------- ------------------------ Sex, male/female 228/97 Mean ± SD (range) age (years)  73.7 ± 9.0 (35 -- 95) Histologic type (%)  Differentiated type adenocarcinoma 315 (97 %)  Undifferentiated type adenocarcinoma  10 (3 %) Location (%)  Lower 131 (40 %)  Middle  99 (31 %)  Upper  95 (29 %) Indication criteria for ESD  Absolute indication group 204 (63 %)  Expanded indication group  83 (25 %)  Exclusion indication group  38 (12 %) --------------------------------------- ------------------------ The technical outcome is summarized in [Table 3](#TB221-3){ref-type="table"}. The grasping and pull or lift-up steps before cutting the targeted tissue provided good visualization of the target area and allowed the use of sufficient pre-cut coagulation. In three patients (0.9 %), difficulties were encountered with the endoscopic approach to the target tissue during submucosal dissection. These lesions were successfully resected (en-bloc resection) using a polypectomy snare. The mean size of the EGCs and resected specimens was 17.3 ± 12.1 mm and 46.7 ± 15.5 mm, respectively. The en-bloc resection rate and R0 resection rate were 99.7 % and 95.3 % respectively. The mean operating time was 97.2 minutes. ###### Technical results of ESD with the Clutch Cutter (N = 325). ------------------------------------------------- ------------------------- Mean ± SD size of lesion, mm (range)  17.3 ± 12.1 (2 -- 74) Mean ± SD size of resected specimen, mm (Range)  46.7 ± 15.5 (18 -- 95) En-bloc resection rate (%) 324/325 (99.7 %) R0 resection rate (%) 310/325 (95.3 %) Complication rate  12/325 (3.6 %)  Intraoperative perforation rate   1/325 (0.3 %)  Intraoperative uncontrollable bleeding rate   0/325 (0 %)  Postoperative perforation rate   0/325 (0 %)  Postoperative bleeding rate  11/325 (3.4 %) ------------------------------------------------- ------------------------- [Table 4](#TB221-4){ref-type="table"} lists the R0 resection rate and operating time according to clinicopathological factors. R0 resection rates differed significantly according to tumor size and indication criteria. Mean operating time also differed significantly according to tumor size, histologic type, location, and indication criteria. ###### R0 resection rate and operating time for ESD using the Clutch Cutter according to clinicopathological factors (N = 325). ------------------------------------------------------------------------------- -------------------------- **R0 resection rate according to clinicopathological factors (%)** Tumor size  0 -- 20 mm 230/235 (97.9 %)  21-mm  80/90 (88.9 %)  *P* value *P* \< 0.0001 Histologic type  Differentiated type adenocarcinoma 301/315 (95.6 %)  Undifferentiated type adenocarcinoma   9/10 (90 %)  *P* value NS Location  Lower 127/131 (96.9 %)  Middle  91/99 (91.9 %)  Upper  92/95 (96.8 %)  *P* value NS Indication criteria  Absolute indication group 201/204 (98.5 %)  Expanded indication group  81/83 (97.6 %)  Exclusion indication group  28/38 (73.7 %)  *P* value *P* \< 0.0001 **Operating time (minutes) according to clinicopathological factors (Range)** Tumor size  0 -- 20 mm  82.4 ± 58.8 (14 -- 423)  21-mm 135.4 ± 63.2 (20 -- 340)  *P* value *P* \< 0.0001 Histologic type  Differentiated type adenocarcinoma  95.5 ± 62.3 (14 -- 379)  Undifferentiated type adenocarcinoma 149 ± 106.4 (51 -- 423)  *P* value *P* \< 0.05 Location  Lower  73.9 ± 44.9 (15 -- 230)  Middle 108.8 ± 58.8 (19 -- 300)  Upper 117.2 ± 81.6 (14 -- 423)  *P* value *P* \< 0.0001 Indication criteria  Absolute indication group  78.8 ± 54.7 (14 -- 379)  Expanded indication group 128.7 ± 68.0 (33 -- 423)  Exclusion indication group 126.3 ± 68.9 (20 -- 340)  *P* value *P* \< 0.0001 ------------------------------------------------------------------------------- -------------------------- During ESD-CC, minor bleeding occurred in most cases, but prompt hemostasis was achieved with the CC (VTR. 2). Unsuccessful hemostasis using the CC for intraoperative bleeding occurred in eight cases (2.4 %), but hemostasis was obtained with the injection of hypertonic saline epinephrine solution and/or coagulation with a Coagrasper (FD-410LR; Olympus). No uncontrollable bleeding was encountered. Postoperative bleeding was seen in 3.4 % (11 of 325). However, all of the hemorrhagic episodes were successfully managed with endoscopic clipping or coagulation. In one case (0.3 %), endoscopic perforation occurred during cutting when proper muscle was wrongly identified as fibrotic submucosal tissue. That patient was successfully treated with endoscopic clipping. There was no delayed perforation after ESD-CC. The learning curve demonstrated changes in operator proficiency over time ([Fig. 4](#FI221-4){ref-type="fig"}). Proficiency was expressed as operating time only and did not reflect the rates of R0 resection or perforation. The operating time in the first period was significantly longer than in the remaining periods (*P* \< 0.01). {#FI221-4} Discussion ========== ESD for EGC is a widely accepted and well-established procedure because of it has a high tumor eradication rate and is less invasive than surgery. The major advantage of ESD over EMR is the high rate of en-bloc resection, regardless of the size, shape, coexisting ulcer, or location of the lesion [@JR221-1] [@JR221-2] [@JR221-3] [@JR221-4] [@JR221-5]. However, ESD is a more difficult and meticulous technique than EMR, and sometimes is associated with serious adverse events. Despite its technical difficulties, ESD using various knife devices has been shown by experts in limited high-volume centers to be effective and relatively safe for treatment of EGC [@JR221-1] [@JR221-2] [@JR221-3] [@JR221-4] [@JR221-5] [@JR221-6] [@JR221-7]. ESD using knife devices is technically difficult and carries a substantial risk of perforation and bleeding [@JR221-2] [@JR221-4] [@JR221-6]. The basic counter-measure for such complications is sufficient submucosal lifting by injected solution, which is very effective for all ESD methods including ESD-CC. When circumferential incision is completed prior to submucosal dissection, submucosal lifting often is difficult because of escape of the injected solution. We usually perform partial mucosal incision followed by partial submucosal dissection and repeat it up to completion of the en-bloc resection to prevent insufficient submucosal lifting, especially for broad lesions. The inevitable risk factors for ESD-related complications that are associated with knife devices are defects of fixation (inaccurate targeting) and compression (hemostatic effect) and pushing the knife to the target tissue (push force basically in the direction of the proper muscle) with electric discharge [@JR221-8]. Our new device, CC, can be used to grasp, pull (or lift up), and incise or coagulate the targeted tissue with electrosurgical current. The CC has four characteristic safe mechanical procedures: 1) fixation (accurate targeting); 2) pull or lift up (away from the proper muscle layer); 3) compression (high hemostatic capability); and 4) outside insulation (minimization of outside electric damage) [@JR221-8] [@JR221-9] [@JR221-10] [@JR221-11] [@JR221-12]. In this study, there was a 0.3 % perforation rate (only one case), no unintentional incision, and we were able to stop the intraoperative bleeding quickly and easily by CC without changing the device (VTR. 2). ESD-CC is accomplished merely with grasp, pull (or lift up), and cut or coagulate steps, and it is as easy and simple as a standard-bite biopsy technique ([Fig. 3](#FI221-3){ref-type="fig"}). Furthermore, it was possible to perform all steps of ESD with only one instrument: the CC. We hypothesize that the advantages of ESD-CC will reduce the difficulty, risk of complication, and costs involved in performing ESD [@JR221-8] [@JR221-9]. The ESD-CC is available for different organs. As of February 2015 (including our previously reported studies[@JR221-9]), in our institute, we had performed ESD-CC (long-type for the stomach, short-type for the remaining organs) on 852 consecutive patients with a diagnosis of early-stage digestive tract tumor including adenoma (organ, number of patients, en-bloc resection rate, perforation rate, bleeding rate; lower pharynx, five patients, 80 %, 0 %, 0 %; esophagus, 66 patients, 100 %, 1.5 %, 1.5 %; stomach, 524 patients, 99 %, 0.2 %, 3 %; duodenum, eight patients, 100 %, 0 %, 0 %; ileum, one patient, 100 %, 0 %, 0 %;colon and rectum, 248 patients, 95 %, 1.6 %, 1.6 %) using the same method (unpublished data). In this ESD-CC study, en-bloc resection was achieved in 99.7 % of the patients. That is comparable to results in other studies (94.9 % -- 97.7 %) [@JR221-3] [@JR221-5] [@JR221-15] [@JR221-16] [@JR221-17]. The rate of R0 resection by ESD-CC was 95.3 %. The reported rate of R0 resection using a knife device ranges from 82 % to 95.5 % [@JR221-4] [@JR221-7] [@JR221-18]. The mean operating time for ESD-CC was 97.2 minutes, versus reported mean operating times for ESD using knife devices ranging from 47.8 to 108.1 minutes [@JR221-5] [@JR221-7] [@JR221-19]. The R0 resection rate and procedure times for ESD-CC appear to be similar to those for conventional knife ESD. Furthermore, the R0 resection rate with ESD-CC was high (over 88 %) irrespective of differences in tumor size, histologic type, location, and indication type (not counting the exclusion indication group). Neuhaus [@JR221-20] reported that European results for ESD indicate lower rates of complete resection and higher morbidity compared to most trials from Asia, and they also demonstrate an unfavorable outcome at the beginning of the learning curve even after training in an experimental setting. Probst et al. [@JR221-21] showed a learning curve for ESD using knife devices resulting in a decreased procedural duration and an increased rate of complete en-bloc resection over time. However, ESD-CC is a simple technique (grasp \> \> pull \> \> cut or coagulate) similar to a standard bite biopsy. Because to the ease or learning ESD-CC, we achieved a high complete resection rate and an extremely low rate of morbidity, although we had little experience with conventional knife ESD. Based on our analysis of the learning curve, approximately 50 procedures must be carried out to acquire skill with performing ESD-CC for early gastric cancers. Risk of perforation associated with the ESD procedure is divided into two groups, depending on the time of onset. In­traoperative perforation is due mainly to penetration of an electrosurgical knife through the proper muscle layer dur­ing ESD. The other type of perforation is postoperative, which mainly occurs 1 to 2 days after the ESD procedure. The rate of intraoperative perforation during gastric ESD ranges from 1.2 % to 8.2 % [@JR221-5] [@JR221-22] [@JR221-23] [@JR221-24]. To prevent intraoperative perforation, it is necessary to avoid electric sparkling to the proper muscle layer, which occurs mainly because of unintentional incision. The safe mechanism of CC, such as the pull effect and outside insulation, is very effective in preventing intraoperative perforation. In our ESD-CC study, we encountered only one endoscopic perforation (0.3 %) due to misidentification of the proper muscle layer as fibrotic tissue, not due to unintentional incision. This patient was successfully treated with immediate closure of the perforation by endoscopic clipping. Postoperative perforation is reported to be a rare complication; however, once it occurs, it can lead to serious conditions that often require emergency surgery [@JR221-25] [@JR221-26]. The frequency is report­ed to be about 0.45 % [@JR221-26]. In theory, excessive thermal damage to the muscle layer may be one of the causes of postopera­tive perforation. Usually, hemostasis using the conventional knife requires a gentle push of the knife to the visible vessel with electrosurgical coagulation. Furthermore, most hemostatic devices have no outside insulation. Such shortcomings of conventional ESD devices may lead to postoperative perforation. The pull effect and outside insulation of the CC is very effective in preventing postoperative perforation. In this ESD-CC study, we encountered no postoperative perforations. Bleeding is the most common major complication of ESD and is divided into two types, based on time of onset. Intraoperative bleeding is defined as any bleeding occurring during the ESD procedure, whereas postoperative bleeding occurs after the ESD procedure. Although intraoperative bleeding occurs frequently, it is difficult to measure its severity. Oda et al. reported that the rate of significant intraoperative bleeding was 7 % [@JR221-27]. Prevention and early control of any intraoperative bleeding is vital because bleeding can impair the endoscopic view, resulting in an increase in procedure time and intraoperative perforation. Preventive hemostatic coagulation of visible blood vessels and the quick cessation of unintentional bleeding are very important for obtaining a safe cure. The CC has accurate and sufficient hemostatic effects such as fixation (accurate targeting), pull or lift up (from the proper muscle layer), and compression (high hemostatic capability)[@JR221-8] [@JR221-9]. Therefore, we were able to perform sufficient pre-cut coagulation and stop the unintentional hemorrhage quickly using the CC without changing the device. There was no uncontrollable intraoperative bleeding in our study. Postoperative bleeding is reported to occur in 5.3 % to 15.6 % of gastric ESD cases [@JR221-5] [@JR221-22] [@JR221-27] [@JR221-28] [@JR221-29] [@JR221-30], usually within a week after ESD. Therefore, in our institution, patients are observed in the hospital for 7 days following the procedure. ESD-CC is available for pre-cut coagulation and preventive coagulation for visible blood vessels of the ESD ulcer base. In this study, the postoperative bleeding rate was relatively low (3.4 %) as compared with that for conventional knife ESD. In our series, we performed second-look endoscopy in all cases. If a risky exposed vessel with clot or oozing bleeding was found during second-look endoscopy, we performed additional coagulation. The second-look endoscopy may have had a positive effect on our low rate of delayed postoperative hemorrhage. Several specific knives or devices are required for each step in conventional ESD [@JR221-8] [@JR221-9]. With ESD-CC, on the other hand, a single device is used (CC). Before ESD-CC was introduced in our institute, we used a needle knife for marking and making a starting hole, an insulation -- tipped diathermic knife for circumferential incision and submucosal excision steps, and a hemostatic forceps for intraoperative bleeding. At least three devices were required for each session of ESD. In this study, we used only one device (CC) for all ESD steps. The ESD-CC reduces the number of devices and the cost of ESD [@JR221-9]. In conclusion, the results of our large, single-center series suggest that ESD-CC is an excellent endoscopic treatment for EGC because it is a single-device method, effective, safe, and technically simple to perform. Extensive controlled, randomized studies, e. g., vs. IT knife, vs. needle knife or vs. other devices, are necessary to fully evaluate the usefulness of our method, not only in Japan but also in Western countries, given the significant difference in ESD experience. **Video 1** VTR of the procedure of ESD-CC of type 0-I + IIa early gastric cancer. **Video 2** VTR of the hemostasis of intraoperative bleeding by the CC. **Competing Interests:** Kazuya Akahoshi and Hidefumi Akahane (FUJIFILM) have applied for a patent in Europe for the Clutch Cutter describes in this article. Japan, China, and the USA have already granted the patent.

Introduction ============ Aggressive mature B-cell lymphomas harboring concurrent translocations of 8q24/*MYC* mainly with 18q21/*BCL2* are called "double-hit lymphomas (DHL)" now referred to as "high grade B-cell lymphoma with *MYC* and *BCL2* and/or *BCL6* rearrangements (DH-HGBL)" according to the current World Health Organization (WHO) classification of lymphoid neoplasms.^[@b1-1041417]^ The concurrent translocations of 8q24/*MYC* and 18q21/*BCL2* usually lead to overexpression of both proteins, and DH-HGBL clinically forms a specific group among "double-protein-expression lymphomas (DPL)".^[@b1-1041417]--[@b3-1041417]^ The most common histological type of DH-HGBL is diffuse large B-cell lymphoma (DLBCL), which has heterogeneous clinicopathological, immunophenotypic, and genetic features.^[@b1-1041417],[@b4-1041417]^ Gene expression signatures have stratified DLBCL into germinal center B-cell (GCB)-like, activated B-cell (ABC)-like, and other subtypes, each of which results from different pathogenic mechanisms.^[@b1-1041417],[@b5-1041417],[@b6-1041417]^ DH-HGBL cases with DLBCL morphology frequently result in disastrous consequences in spite of showing the GCB phenotype, which is regarded as a relatively favorable marker for survival.^[@b1-1041417],[@b2-1041417],[@b4-1041417]^ Thus, to be DHL and DPL (DH-DPL) seems to have a negative impact on survival, especially in GCB-like DLBCL cases.^[@b1-1041417]--[@b3-1041417]^ MYC is a powerful transcriptional activator, target genes of which are associated with cell proliferation, DNA replication, protein synthesis, and cell metabolism, and its overexpression is a hallmark of tumor aggressivity.^[@b7-1041417],[@b8-1041417]^ In contrast, BCL2 is the first identified anti-apoptotic regulator that contributes to the survival of lymphoma cells.^[@b9-1041417],[@b10-1041417]^ Dysregulation of both genes likely generates aggressive lymphoma cells showing a fast growth rate and resistance to apoptotic stimuli. Clinically, DH-DPL has a poor prognosis when treated with the standard rituximab-combined cyclophosphamide, doxorubicin, vincristine, and prednisone (R-CHOP) regimen, with a median survival of around 20 months.^[@b2-1041417],[@b11-1041417]^ Until now, optimal therapeutic strategies against DH-DPL remain to be determined. Recent reports suggest that targeting MYC and BCL2 may be a promising strategy to control DH-DPL.^[@b12-1041417]--[@b15-1041417]^ BRD4, a member of the bromodomain and extra-terminal domain (BET) family, is considered to be a convenient target for MYC-driven lymphomas.^[@b16-1041417],[@b17-1041417]^ BET family proteins recognize acetylated chromatin and act as transcription co-factors.^[@b18-1041417]^ BRD4 is upregulated in DLBCL and Burkitt lymphoma cells, and its inhibition leads to a strong downregulation of MYC and its regulating genes, resulting in suppression of their cell growth.^[@b16-1041417],[@b17-1041417]^ Meanwhile, the selective BCL2 inhibitor venetoclax demonstrated excellent antitumor effects in chronic lymphocytic leukemia.^[@b19-1041417],[@b20-1041417]^ BCL2 and its family proteins function as inhibitors and activators of the intrinsic apoptotic pathway at the mitochondrial membrane level.^[@b10-1041417],[@b21-1041417]^ They contain at least one of four BCL2 homology (BH) domains (BH1-4) and are classified into three groups based on their structure and function: i.e., the pro-survival proteins (BCL2, BCL-xL, MCL1, BFL1, and BCLw) sequester the pro-apoptotic BH3-only proteins (BID, BIM, BAD, NOXA, PUMA, BMF, HRK, and BIK), which in turn activate the pore-forming proteins (BAX and BAK).^[@b10-1041417],[@b21-1041417]^ Oligomerization of BAX/BAK permeabilizes the mitochondrial membrane, resulting in cytochrome c release and apoptosis.^[@b10-1041417],[@b21-1041417]^ The BH3 mimetic venetoclax binds to the BH3 domain of BCL2, releases BH3-only proteins, and induces apoptosis.^[@b10-1041417],[@b21-1041417]^ Although short exposure to venetoclax can trigger significant antitumor effects in DLBCL cells,^[@b12-1041417]--[@b15-1041417],[@b19-1041417],[@b22-1041417]--[@b24-1041417]^ this drug's clinical efficacy in DLBCL is less promising,^[@b25-1041417]^ probably because the apoptotic sensitivity to venetoclax is influenced not only by total amounts of BCL2, but also by its phosphorylation status, especially at serine 70 (Ser70), and the further presence of other pro-survival proteins.^[@b14-1041417],[@b15-1041417],[@b22-1041417]--[@b24-1041417],[@b26-1041417]--[@b28-1041417]^ Among the pro-survival proteins, MCL1 is considered the major determinant of resistance to venetoclax.^[@b22-1041417]--[@b24-1041417],[@b28-1041417]^ Therefore, the therapeutic application of venetoclax to DH-DPL needs further investigation. In this study, we examined the apoptotic sensitivity of GCB-like DLBCL cells to the BRD4 inhibitor JQ-1 and BH3 mimetics, focusing on the association of BCL2 with MCL1. Methods ======= Reagents -------- The BCL2 inhibitor venetoclax (Selleck Chemicals, Houston, TX, USA), MCL1 inhibitor S63845 (ApexBio, Houston, TX, USA), BCL-xL inhibitor A-1155463 (Selleck Chemicals), and BRD4 inhibitor JQ-1 (Sigma-Aldrich, St. Louis, MO, USA) were dissolved with dimethyl sulfoxide (DMSO; Nacalai Tesque, Kyoto, Japan) and added to the culture medium. Cell lines ---------- We used four GCB-like DLBCL cell lines: BJAB,^[@b29-1041417]^ SU-DHL10 (ATCC, Manassas, VA, USA), Karpas231 (ECACC, Salisbury, UK), and OCI-Ly8 (kindly provided by Dr. Masao Seto, Kurume University, Kurume, Japan). They were maintained as described previously.^[@b29-1041417]^ Fluorescence *in situ* hybridization analyses --------------------------------------------- Fluorescence *in situ* hybridization analyses were performed using a Vysis LSI MYC dual-color, break-apart rearrangement probe, a Vysis LSI IGH/MYC/CEP8 tri-color dual-fusion probe, and a Vysis LSI *IGH/BCL2* dual-color, dual-fusion translocation probe (Abbott Molecular, Des Plaines, IL, USA). Clinical samples and immunohistochemistry ----------------------------------------- Clinical samples were obtained at biopsy performed for initial diagnosis at our institution between September 2012 and October 2018. Diagnoses were made according to the current WHO classification,^[@b1-1041417]^ and GCB-like DLBCL was defined using the Hans criteria.^[@b30-1041417]^ A bone marrow specimen from one patient (UPN3) was subjected to *in vitro* susceptibility testing using venetoclax after having obtained written informed consent. Immunohistochemistry was performed on formalin-fixed, paraffin-embedded specimens using standard techniques. The primary antibodies used are listed in the *Online Supplementary Methods*. The percentages of stained lymphoma cells were evaluated by visual estimation and recorded in 10% increments by at least two observers. The study protocol was approved by the Ethics Review Board of St. Marianna University. Cell proliferation and annexin V-binding assays ----------------------------------------------- We performed direct cell counting using a trypan blue-exclusion test (Thermo Fisher Scientific, Carlsbad, CA, USA) and annexin-V binding/7-amino-actinomycin D (7-AAD) rejection assays (Beckman Coulter, Brea, CA, USA) with a FACSCalibur flow cytometer (BD Bioscience, Franklin Lakes, NJ, USA). Acquired data were analyzed using FlowJo software (BD Bioscience). Western blot analysis --------------------- Cell lysates were prepared as described previously.^[@b29-1041417]^ Equal amounts of protein (30 μg/well) were separated on a discontinuous sodium dodecyl sulfate-10% polyacrylamide gel and blotted onto a nitrocellulose membrane (Bio-Rad, Hercules, CA, USA). The primary antibodies used are listed in the *Online Supplementary Methods*. Antibody signals were visualized using Western blue (Promega, Madison, WI, USA). Immunoprecipitation ------------------- Whole cell lysate (100 μg of total protein) was added to 1/200 volume of each rabbit antibody and adjusted to 680 μL of volume with the lysis buffer. The antibodies used are listed in the *Online Supplementary Methods*. Rabbit serum-saturated protein G-Sepharose beads (20 μL) (GE Health Care, Uppsala, Sweden) were added to each lysate, and immunoprecipitation was performed for 1 h while rotating at 4°C. After incubation, the beads were washed three times with ice-cold lysis buffer and boiled for 3 min with 20 μL/tube of loading buffer. The supernatant was subjected to western blot analysis. Statistical analysis -------------------- Differences in positive rates of MYC and BCL2 family proteins between DH-HGBL and other GCB-like DLBCL were calculated using a two-tailed Mann-Whitney U test and Student *t* test. *P* values \<0.05 were considered statistically significant. All analyses were performed using EZR (Saitama Medical Center, Jichi Medical University, v.1.33).^[@b31-1041417]^ Results ======= MCL1 overexpression is less frequently observed in DH-HGBL ---------------------------------------------------------- We initially evaluated the protein expression of MYC and BCL2 family members in clinical samples. A total of 27 cases were analyzed because they had minimal data and specimens for immunohistochemical and cytogenetic evaluations (*Online Supplementary Table S1*). Immunohistochemistry and fluorescence *in situ* hybridization analyses using *MYC* split and *IGH-BCL2* fusion probes confirmed that these clinical samples comprised eight DH-HGBL cases harboring both *MYC* and *BCL2* gene rearrangements and 19 GCB-like DLBCL cases. Of course, all DH-HGBL cases showed the GCB phenotype, and seven of the eight cases corresponded to DH-DPL (*Online Supplementary Table S1*). Among the 19 GCB-like DLBCL cases, five showed overexpression of both MYC and BCL2 proteins and were regarded as non-DH DPL (*Online Supplementary Table S1*). Although immunohistochemistry clearly detected overexpression of both MYC and BCL2 proteins more frequently in DH-HGBL than in other GCB-like DLBCL (Mann-Whitney U test, *P*=0.028 in MYC and *P*=0.013 in BCL2), differences between DH-HGBL and non-DH DPL were not seen ([Figure 1A](#f1-1041417){ref-type="fig"}). Among the BCL2 family proteins, immunhistochemistry clearly confirmed the expression of BCL2, MCL1 and BIM, but rarely detected BAX regardless of the groups ([Figure 1A,B](#f1-1041417){ref-type="fig"}, and *Online Supplementary Figure S1*). There were no differences in the rates of positivity for BIM and BAX among the three groups, whereas MCL1 overexpression was detected significantly more frequently in other GCB-like DLBCL than in DH-HGBL (Mann-Whitney U test, *P*=0.005) ([Figure 1A](#f1-1041417){ref-type="fig"}). This significance was also observed between DH-HGBL and non-DH DPL (Student *t* test, *P*=0.019) ([Figure 1A](#f1-1041417){ref-type="fig"}). Concerning the staining pattern for MCL1, we found differences between DH-HGBL and other GCB-like DLBCL including non-DH DPL. All DH-HGBL cases showed the cytoplasmic staining pattern, whereas two of five non-DH DPL (40%) and six of 14 other GCB-like DLBCL cases (43%) exhibited the nuclear staining pattern ([Figure 1B](#f1-1041417){ref-type="fig"}, *Online Supplementary Table S1*). These results suggest that the intrinsic anti-apoptotic activities in DHL-HGBL or DH-DPL may be dependent on BCL2 rather than MCL1. {#f1-1041417} Karpas231 and OCI-Ly8 are DH-DPL cells -------------------------------------- We next performed *in vitro* investigations to characterize DH-DPL cells. Fluorescence *in situ* hybridization analyses confirmed that all four GCB-like DLBCL-derived cell lines have the 8q24/*MYC* translocation ([Figure 2A](#f2-1041417){ref-type="fig"}). *MYC* was fused to *IGH* in BJAB, SU-DHL10, and OCI-Ly8 ([Figure 2A](#f2-1041417){ref-type="fig"}). In addition to the rearrangement, *IGH-BCL2* fusion resulting from t(14;18)(q32;q21) was detected in SU-DHL10, Karpas231, and OCI-LY8 ([Figure 2A](#f2-1041417){ref-type="fig"}). Western blot analysis showed that the four lines express BCL6, MYC, BRD4, MCL1, BCL-xL, BIM, BAD, BAK, and BAX at a variety of levels ([Figure 2B](#f2-1041417){ref-type="fig"}). Among the pro-apoptotic BH3-only proteins, BIM was abundantly expressed in each cell line ([Figure 2B](#f2-1041417){ref-type="fig"}). The results were consistent with those from clinical samples. In contrast, protein expression levels of BCL-xL and MCL1 were relatively low in SU-DHL10 and OCI-Ly8, respectively ([Figure 2B](#f2-1041417){ref-type="fig"}). Although SU-DHL10 has the *IGH-BCL2* fusion, we failed to detect BCL2 protein expression in this line ([Figure 2B](#f2-1041417){ref-type="fig"}). Also, NOXA was not detected in BJAB cells ([Figure 2B](#f2-1041417){ref-type="fig"}). These results indicate that BJAB represents non-DH DPL cells and that SU-DHL10, Karpas231 and OCI-Ly8 are regarded as DH-HGBL cells. Furthermore, Karpas231 and OCI-Ly8 correspond to DH-DPL cells having relatively abundant BCL2 proteins, a considerable part of which is phosphorylated at Ser70. {#f2-1041417} Exposure to 200 nM of venetoclax effectively induces apoptosis in DH-DPL cells ------------------------------------------------------------------------------ We subsequently evaluated the growth inhibitory effect and apoptotic sensitivity to JQ-1 and three BH3 mimetics in each cell line. Except for the BJAB cell line, JQ-1 similarly suppressed cell proliferation in a dose-dependent manner, while the inhibitory effects of BH3 mimetics were different among the four lines ([Figure 3A](#f3-1041417){ref-type="fig"}). Venetoclax effectively suppressed proliferation of Karpas231 and OCI-Ly8 cells even at low concentrations (20 nM and 200 nM), whereas this agent had no effect on either BJAB or SU-DHL10 cells ([Figure 3A](#f3-1041417){ref-type="fig"}). In contrast, S63845 clearly showed an inhibitory effect at low concentrations (10 nM and 100 nM) only in SU-DHL10 cells ([Figure 3A](#f3-1041417){ref-type="fig"}). Although A-1155463 inhibits cell growth at levels less than 1 μM in sensitive cells,^[@b28-1041417],[@b32-1041417]^ at nanomolar levels this agent failed to suppress the proliferation of any cell line ([Figure 3A](#f3-1041417){ref-type="fig"}). {#f3-1041417} Flow cytometry detected apoptotic changes 24 h after exposure to these drugs. Although 50 μM of JQ-1 efficiently suppressed proliferation of SU-DHL10, Karpas231, and OCI-Ly8 cells, the exposure was insufficient to induce apoptosis in the two DH-DPL-cell lines. Approximately 70% of SU-DHL10 cells were positive for annexin V, whereas the positive rates were limited to around 18\~40% in Karpas231 and OCI-Ly8 cells ([Figure 3B](#f3-1041417){ref-type="fig"}). Exposure to 200 nM of venetoclax effectively led to apoptotic changes in Karpas231 and OCI-Ly8 cells. More than 80% of the cells were positive for annexin V in both DH-DPL-cell lines ([Figure 3B](#f3-1041417){ref-type="fig"}). Conversely, exposure to 100 nM of S63845 induced cell death only in SU-DHL10. Although around 60% of the cells were annexin V-positive, more than 90% of SU-DHL10 cells were stained with 7-AAD ([Figure 3B](#f3-1041417){ref-type="fig"}). In this study, exposure to 1 μM of A-1155637 did not induce even modest cell death in each cell line ([Figure 3B](#f3-1041417){ref-type="fig"}). These results indicate that anti-apoptotic activities in the four lines were less related to BCL-xL. We further evaluated the alteration of apoptotic sensitivity to venetoclax in combination with S63845. In BJAB cells, the addition of S63845 to venetoclax led to a clear increase in annexin V-positive cells, indicating that S63845 and venetoclax had a synergic effect on inducing apoptosis ([Figure 4A](#f4-1041417){ref-type="fig"}). In contrast, S63845 did not show even an additive effect in Karpas231, OCI-Ly8, and SU-DHL1 cells ([Figure 4B](#f4-1041417){ref-type="fig"}). Although the expression levels of MCL1 were nearly equal between BJAB and Karpas231 cells, S63845 failed to exert additive pro-apoptotic effects in Karpas231 cells. These results indicate that the intrinsic anti-apoptotic activities in BJAB are equally dependent on BCL2 and MCL1, but those in the two DH-DPL-cell lines and SU-DHL10 cells depend mainly on BCL2 and MCL1, respectively. We, therefore, focused on the biological effects of venetoclax especially in the two DH-DPL cell lines. {#f4-1041417} Venetoclax leads to dephosphorylation of BCL2 and downregulates MCL1 expression in DH-DPL cells ----------------------------------------------------------------------------------------------- We then examined alterations of protein levels after exposure to venetoclax. Western blot analysis showed that expression levels of BCL2 were unchanged at least 6 h after exposure to venetoclax in Karpas231 and OCI-Ly8 cells ([Figure 5A,B](#f5-1041417){ref-type="fig"}). In contrast, phosphorylation of BCL2 was clearly downregulated, and expression levels of BIM were also decreased 3 h after exposure to venetoclax in both DH-DPL cell lines ([Figure 5A,B](#f5-1041417){ref-type="fig"}). In contrast, although the total protein levels of BCL2 were slightly decreased, the proportion of phosphorylated BCL2 was increased 6 h after exposure to venetoclax in BJAB cells ([Figure 5C](#f5-1041417){ref-type="fig"}). Unexpectedly, MCL1 showed decreased expression levels in both Karpas231 and OCI-Ly8 cells within 3 h after exposure to venetoclax ([Figure 5A,B](#f5-1041417){ref-type="fig"}). This phenomenon was not observed in the BJAB and SU-DHL10 lines ([Figure 5C, D](#f5-1041417){ref-type="fig"}). Furthermore, the expression of protein phosphatase 2A (PP2A) B56α, which has a critical role in dephosphorylation of BCL2,^[@b33-1041417]^ was decreased within 3 h after exposure to venetoclax in both DH-DPL cell lines ([Figure 5A,5B](#f5-1041417){ref-type="fig"}). The expression levels of another regulatory subunit, PP2A B56δ,^[@b34-1041417]^ were unchanged in each cell line ([Figure 5A-D](#f5-1041417){ref-type="fig"}). These results confirmed that venetoclax clearly changes the biological behavior of BCL2 and MCL1 in the two DH-DPL cell lines. {#f5-1041417} Treatment with venetoclax augments the accumulation of PP2A B56α in BCL2 in DP-DHL cells ---------------------------------------------------------------------------------------- Immunoprecipitation and western blot analyses of Karpas231 and OCI-Ly8 cells showed that BCL2 was substantially associated with BIM at steady state ([Figure 6A](#f6-1041417){ref-type="fig"}). Likewise, BCL2 was bound to PP2A B56α ([Figure 6A](#f6-1041417){ref-type="fig"}). We also confirmed that PP2A B56α was bound to MCL1 in both DH-DPL cell lines (*data not shown*). We then evaluated alterations of these associations after exposure to venetoclax. Within 60 min after treatment with venetoclax, total protein levels of BIM bound to BCL2 were decreased similarly in Karpas231 and OCI-Ly8 cells ([Figure 6B](#f6-1041417){ref-type="fig"}). In contrast, the proportion of PP2A B56α bound to BCL2 was slightly increased ([Figure 6B](#f6-1041417){ref-type="fig"}). Although BIM protein levels decreased gradually following exposure to venetoclax, total levels of BCL2 bound to BIM were decreased at the early phase of venetoclax treatment in both DH-DPL cell lines ([Figure 6B](#f6-1041417){ref-type="fig"}). These results indicate that venetoclax not only disrupts the association between BCL2 and BIM, but also augments the accumulation of PP2A B56α in BCL2. {#f6-1041417} Exposure to 200 nM of venetoclax induces apoptosis in primary DH-DPL cells -------------------------------------------------------------------------- We also evaluated the apoptotic sensitivity of primary DH-DPL cells to venetoclax *in vitro*. The fresh bone marrow specimen was obtained from one patient with DH-DPL (UPN3) and exposed to 200 nM of venetoclax. Examination of the bone marrow revealed massive infiltration of blastoid cells with a mature B-cell phenotype \[terminal deoxynucleotidyl transferase (TdT)-negative and BCL6-positive\] ([Figure 7A](#f7-1041417){ref-type="fig"}). Immunohistochemistry confirmed that MYC, BCL2, and MCL1 were highly expressed in the neoplastic cells ([Figure 7A](#f7-1041417){ref-type="fig"}). Flow cytometry clearly detected the apoptotic change 24 h after exposure to venetoclax ([Figure 7B](#f7-1041417){ref-type="fig"}). Almost all the blastoid DH-DPL cells were positive for annexin V ([Figure 7B](#f7-1041417){ref-type="fig"}). Consistent with the results in Karpas231 and OCI-ly8 cells, the primary DH-DPL cells were highly susceptible to low concentrations of venetoclax. ![A low concentration of venetoclax induces apoptosis in primary double-hit and double-protein-expression lymphoma cells. (A) The bone marrow examination of one patient (UPN3) revealed massive infiltration by blastoid double-hit and double-protein-expression lymphoma (DH-DPL) cells with a mature B-cell phenotype terminal deoxynucleotidyl transferase (TdT)-negative and BCL6-positive \[Wright-Giemsa (WG) stain ×1000, hematoxylin-eosin (HE) stain ×600\]. Immunohistochemistry confirmed that the cells expressed high levels of MYC, BCL2, MCL1 and BIM (×600). (B) Flow cytometry clearly detected apoptotic changes 24 h after exposure to venetoclax dissolved in dimethyl sulfoxide (DMSO). More than 99% of blastoid DH-DPL cells were positive for annexin V.](1041417.fig7){#f7-1041417} Discussion ========== Since DH-DPL is a rapidly progressing disease and often refractory to even intensive therapies, patients with this lymphoma have dismal survival outcomes unlike those with any other type of DLBCL.^[@b2-1041417],[@b3-1041417],[@b11-1041417]^ Hence, it is urgently necessary to develop effective treatments for DH-DPL. Appropriate balances between pro-survival and pro-apoptotic BCL2 family proteins are usually disrupted in GCB-like DLBCL including DH-DPL, in which the overexpression of pro-survival proteins, including BCL2, MCL1, and BCL-xL, likely confers resistance to chemotherapeutic agents.^[@b10-1041417],[@b12-1041417],[@b15-1041417]^ In this study, we found that BCL2 seems to play crucial roles in anti-apoptotic activities in DH-DPL, and further verified that venetoclax is still a promising agent for this type of lymphoma. The present DH-HGBL cases, mostly corresponding to DH-DPL, showed lower levels of MCL1 expression, associated with resistance to venetoclax, compared with other GCB-like DLBCL cases. In addition, the frequent nuclear localization of MCL1 in non-DH DPL and other GCB-like DLBCL cases not only indicates the abundant expression, but also implies the presence of distinct functions from its anti-apoptotic activity in these groups.^[@b28-1041417],[@b35-1041417]^ Furthermore, our *in vitro* study confirmed that the inhibition of MCL1 by S63845 did not have any additive pro-apoptotic effect to that of venetoclax in the two DH-DPL cell lines, whereas S63845 restored the apoptotic sensitivity to low concentrations of venetoclax in the non-DH DPL-cell line BJAB. A previous study showed that overexpression of MCL1 is observed more frequently in ABC-like DLBCL than in GCB-like DLBCL.^[@b36-1041417]^ Our observations indicate that the anti-apoptotic activity of MCL1 for DH-DPL seems to be less limited than that for any other type of GCB-like DLBCL. DH-DPL seems to be dependent mainly on BCL2 for survival, and the pro-apoptotic action of venetoclax in DH-DPL cells may be less affected by intrinsic MCL1 expression. The BRD4 inhibitor JQ-1 is considered a promising drug for the treatment of MYC-driven lymphomas.^[@b16-1041417],[@b17-1041417]^ Although JQ-1 suppressed cell proliferation, this agent did not lead to adequate apoptosis in the two DH-DPL-cell lines. Since BET inhibitors can induce cell-cycle arrest,^[@b37-1041417]^ this raises concerns that JQ-1 might fail to exterminate malignant cells. Although the combination with BH3 mimetics is supposed to be a promising strategy, antagonistic effects have been observed.^[@b15-1041417]^ BET inhibitors are effective at inhibiting tumor expansion, but might be inappropriate for eradicating of residual lymphoma cells. The antitumor effect of venetoclax has mainly been discussed with regard to its physiological association with BCL2.^[@b21-1041417],[@b22-1041417],[@b24-1041417]^ In the present study, we demonstrated for the first time that venetoclax not only disrupts the association between BCL2 and BIM, but also modulates signal transduction related to BCL2 and MCL1 in DH-DPL cells. Phosphorylation at Ser70 and its mutant in the BCL2 loop domain has been shown to enhance binding to BIM and BAK, and confirmed to inhibit the effects of BH3 mimetics on its replacement of BIM in leukemia cells.^[@b27-1041417]^ Although protein levels of phosphorylated BCL2 are relatively high, venetoclax clearly led to dephosphorylation of BCL2 and effectively induced apoptosis in Karpas231 and OCI-Ly8 cells. In both DH-DPL cell lines, the dephosphorylation was probably mediated through accumulation of PP2A B56α in BCL2. The increased binding of PP2A B56α to BCL2 is likely to promote dephosphorylation of this protein. Consequently, a considerable part of phosphorylation at Ser70 seemed to be removed within 6 h after exposure to venetoclax. Dephosphorylation of BCL2 should cause increased sensitivity to venetoclax in DH-DPL cells. Therefore, phosphorylation at Ser70 in BCL2 at steady state is less likely to be a common mechanism of resistance to venetoclax in these DH-DPL cells. We further found that venetoclax altered the levels of expression of MCL1 in both DH-DPL cell lines. The stability of MCL1 is mainly controlled by its phosphorylation status.^[@b38-1041417],[@b39-1041417]^ Phosphorylated MCL1 can be an easy target of an E3 ubiquitin ligase complex and is mostly driven to proteasomal degradation.^[@b38-1041417],[@b39-1041417]^ PP2A dephosphorylates MCL1 and prevents interaction between MCL1 and the complex.^[@b38-1041417],[@b39-1041417]^ A previous study showed that PP2A inhibition dramatically decreased the protein levels of MCL1 within 2 h in Burkitt lymphoma cells.^[@b39-1041417]^ Although we confirmed that PP2A B56α is bound to MCL1 in both Karpas231 and OCI-Ly8 cells, the association was unchanged after exposure to venetoclax (*data not shown*). However, total amounts of PP2A B56α, which should be bound to MCL1, were clearly decreased 3 h after exposure to venetoclax. Therefore, a considerable part of MCL1 protein may be phosphorylated and disassembled within 3 h after exposure to venetoclax in the two DH-DPL cell lines. We further verified that 200 nM of venetoclax was sufficient to induce cell death even in primary DH-DPL cells, despite these cells showing relatively abundant MCL1 expression. Venetoclax achieves and maintains plasma exposure levels of approximately 4 μM at daily doses ranging from 400 mg to 1200 mg in patients with chronic lymphocytic leukemia or lymphomas.^[@b25-1041417],[@b40-1041417]^ Prolonged exposure to very low concentrations of venetoclax may raise new issues about resistance due to the expression of other BCL2 family proteins such as BFL1.^[@b14-1041417]^ However, we believe that venetoclax should kill primary DH-DPL cells at daily doses recommended in patients with chronic lymphocytic leukemia. In conclusion, DH-DPL cells seem to be dependent mainly on BCL2 for survival and relatively low concentrations of venetoclax effectively induced apoptosis regardless of MCL1 expression. Venetoclax not only disrupts the BCL2-BIM interaction, but also leads to dephosphorylation of BCL2 and further downregulates MCL1 expression, probably through modulation of PP2A B56α activity in DH-DPL cells. Although further investigation is needed for clinical application, targeting BCL2 with venetoclax is a promising strategy for the treatment of DH-DPL. Supplementary Material ====================== ###### Uchida et al. Supplementary Appendix ###### Disclosures and Contributions This study was not supported by any grant. We thank Dr. Masao Seto (Kurume University, Kurume, Japan) for providing OCI-Ly8 cells. We also thank Mr. Manabu Kubota for a technical assistant on immunohistochemistry. Check the online version for the most updated information on this article, online supplements, and information on authorship & disclosures: [www.haematologica.org/content/104/7/1417](http://www.haematologica.org/content/104/7/1417) [^1]: AU and YI contributed equally to this work.

Case presentation ================= A 46-year-old Caucasian woman presented to the gynaecology outpatient\'s department complaining of menorrhagia, bladder pressure and a feeling of heaviness in the pelvis. She had recently also started experiencing deep dyspareunia. She reported no bowel or constitutional symptoms. She was a housewife and was a teetotal, non-smoker. She had one child, born by vaginal delivery and her previous cervical smears were normal. She was healthy and had no significant medical, surgical or family history. Examination revealed an enlarged, fourteen-week size uterus. There were no adnexal masses felt. In view of her symptoms, she was counselled about the various treatment options for menorrhagia. She declined all conservative measures and was very keen on having a total abdominal hysterectomy. An ultrasound scan was requested and her operation was booked. Before diagnostic imaging could be performed, she attended hospital for her hysterectomy. Surgery was done through a Pfannenstiel incision, which revealed some haemorrhagic fluid in the abdomen. This was sent for cytology. The uterus was found to be uniformally enlarged, the fallopian tubes looked thickened and the ovaries appeared to have carcinomatous deposits on their surface. The upper abdomen was explored. The omentum was caked with tumour, a biopsy was taken and frozen section performed. This showed deposits of mucin with some signet ring cells -- suspicious of carcinoma of the bowel. The surgical consultant was called, at which time examination demonstrated a palpable tumour in the caecum. A decision was taken to proceed with total abdominal hysterectomy and bilateral salpingo-oophorectomy and to defer hemi-colectomy to a later date, until further investigations could be performed. Subsequent gastroscopy was normal but her colonoscopy was limited due to looping of the bowel. A computerized tomography scan of the chest, abdomen and pelvis revealed no mediastinal nodes or lung parenchymal deposits. Her upper abdomen was normal. However there was thickening of the right para-colic area and caecal pole consistent with local infiltration. Histology of her pelvic organs showed extensive infiltration of the cervix, basal endometrium, myometrium, ovaries and tubal tissues by poorly differentiated adenocarcinoma, with a signet ring pattern showing mucin production. This was identical to the omental biopsy. The impression was that these appeared to be metastases rather than intrinsic lesions. The patient was later referred for palliative chemotherapy as the tumour was felt to be incurable. She subsequently died of her disease some twelve months later. Discussion ========== Bowel cancers invade adjacent structures dependent on their site. Ascending colon tumours tend to invade the duodenum and liver, whilst rectal tumours are more likely to invade the genital organs \[[@B1]\]. In general, non genital malignancies rarely metastasize into the uterus. But should this occur, metastatic uterine cancer may cause uterine enlargement and thickening of the myometrium \[[@B2]\]. Thus mimicking a variety of gynaecological symptoms, which could suggest intrinsic uterine pathology. Our case report re-iterates the importance of considering bowel pathology in the differential diagnosis of all patients presenting with gynaecological symptoms. These days, it is vitally important to bear in mind that diagnostic imaging must be obtained prior to any surgical intervention, as we run the risk of missing serious pathology is we omit this step. In this case imaging prior to surgery may not have changed the eventual outcome, but it may have triggered further investigation, which could have led to the correct diagnosis prior to surgery. Consent ======= Written informed consent was obtained from the patient\'s next of kin for publication of this case report. Unfortunately the patient has died as a result of her condition. A copy of the written consent is available for review by the Editor-in-Chief of this journal. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= Both SD\'s and HS were integrally involved in the patient\'s management and diagnosis. All also contributed to writing of the case report. LI performed the literature search and final copy-editing the paper. All authors read and approved the final manuscript.

Introduction {#s1} ============ Working memory enables us to remember, manipulate, and reorganize information for short periods of time (Salthouse et al., [@B46]; Baddeley, [@B2], [@B3]). This executive memory process is fundamental for everyday life and is known to decline with advanced age (Li et al., [@B27]; Gazzaley et al., [@B16]). Age-related decline in working memory significantly contributes to loss of independence and decreased quality of life (Mograbi et al., [@B33]). However, there is currently a paucity of effective interventions for remediating working memory decline in older adults. Transcranial direct current stimulation (tDCS) demonstrates promise as a potential intervention in this domain, but knowledge of the mechanisms underlying tDCS effects on working memory function in older adults remains unexplored. Non-human primate studies have indicated the importance of the frontal lobes, particularly the dorsolateral prefrontal cortex (DLPFC), in working memory function (Goldman-Rakic, [@B19], [@B18]; Wang et al., [@B48]). In humans, structural and functional neuroimaging studies support the integral importance of frontal structures in working memory (Boisgueheneuc et al., [@B11]; Barbey et al., [@B4]). Reduced structural surface area of right frontal lobe regions in older adults is associated with lower working memory performance (Nissim et al., [@B35]). In addition, low performing older adults demonstrate significantly increased right lateralized blood oxygenation level dependent (BOLD) response compared to high performing older adults who demonstrate bilateral recruitment of frontal structures (i.e., neural compensation hypothesis; Cabeza et al., [@B12]). Moreover, functional connectivity of the working memory network decreases with older age (Hampson et al., [@B21]; Andrews-Hanna et al., [@B1]; Hampson et al., [@B22]; Keller et al., [@B24]). While there is growing evidence supporting the role of frontal structural and functional decline in age-related working memory performance, there is a lack of effective interventions aimed at treating this decline. Such interventions will be necessary for addressing public health concerns related to the increasing age of the world population and the increased prevalence of age-related cognitive decline in the population at large. Non-invasive brain stimulation techniques have the potential to impact cognitive processes and modify behavior by inducing changes in brain function (Gomes-Osman et al., [@B20]). tDCS is a safe and painless form of non-invasive brain stimulation that modulates the neuroplastic response of brain tissue and can impact behavior both during and after periods of stimulation through sub-threshold alteration of resting membrane potentials (Nitsche and Paulus, [@B36]; Monte-Silva et al., [@B34]; Pelletier and Cicchetti, [@B41]; Bikson et al., [@B8]; Woods et al., [@B52]; Bikson et al., [@B7]; Knotkova et al., [@B26]). Prior studies have demonstrated that tDCS can impact and enhance working memory in older adults (Jones et al., [@B23]; Woods et al., [@B53]). For example, Stephens and Berryhill ([@B47]) demonstrated that older adults who received active tDCS over sham (anode-F4, cathode-contralateral cheek) during a working memory training protocol experienced greater benefits on untrained tasks at post-training assessment (Stephens and Berryhill, [@B47]). Effects of tDCS combined with working memory training have shown to extend and increase training gains (Richmond et al., [@B45], [@B44]; Jones et al., [@B23]; Stephens and Berryhill, [@B47]). Recent research also suggests that the efficacy and transfer of tDCS paired with working memory training may be state dependent, with greater transfer occurring when training tasks are more difficult (Gill et al., [@B17]). These studies demonstrate the promise of tDCS as a potential intervention for age-related working memory decline. However, there is a gap in knowledge on the mechanism(s) of action of tDCS-based enhancement of working memory in the brain. No study to date has directly investigated the impact of tDCS on functional connectivity of the working memory network. Better understanding of the mechanism(s) of action for tDCS-based working memory enhancement is essential to harness and optimize tDCS as a treatment approach for cognitive decline in older adults. To address this gap in knowledge, we performed a mechanistic study designed to evaluate the acute and after-effects of tDCS on functional connectivity of working memory networks in older adults during N-Back performance (2- vs. 0-back). This study specifically aimed to query the direct impact of tDCS vs. sham on the working memory network, taking steps to stabilize test-retest behavioral and network change across N-back sessions and between conditions (active vs. sham) to focus our inquiry on direct mechanistic effects of an acute tDCS dose on the network. We hypothesized that active stimulation vs. sham would result in increased frontal connectivity during working memory performance in older adults. In addition, we hypothesized that changes in functional connectivity from active tDCS would be greater in the high-effort 2-Back vs. 0-Back conditions. Materials and Methods {#s2} ===================== We conducted a randomized double-blinded crossover within-subject study examining the acute effects of bilateral frontal tDCS (2 mA for 12 min; F3/F4) during an functional magnetic resonance imaging (fMRI) BOLD scan. For each stimulation condition (active vs. sham), participants performed three runs (baseline-active/sham, during-active/sham, post-active/sham) of an N-Back working memory task. Participants received each stimulation condition (active and sham) at two separate visits (randomized and counterbalanced), and thus, served as their own controls. [Figure 1](#F1){ref-type="fig"} illustrates the overall experimental design used in this study. {#F1} Participants {#s2-1} ------------ Sixteen older adults between the ages of 61--82 years old participated in this study \[n female = 6; mean age (SD) = 71.75 (7.29); mean education = 16.8 (1.92)\]. All participants completed a medical history questionnaire and cognitive assessments to meet study eligibility criteria. The Montreal Cognitive Assessment (MoCA) was used to screen for significant global cognitive deficits. Participants were enrolled in the study if they were MRI-compatible, able to receive electrical stimulation on the head, scored at least 20 or more on the MoCA (mean = 26.56 SD = 2.94), were right-hand dominant, and not taking certain medications potentially blocking tDCS excitability effects (i.e., GABAergic, glutamatergic, or sodium channel blockers; McLaren et al., [@B32]). Participants with a history of neurological disorders, seizures, psychiatric disorders, in active treatment for cancer, previous traumatic brain injury with loss of consciousness greater than 20 min, or any neurodegenerative disease were excluded from the study. The washout period between stimulation visits for each participant was 8.5 days on average (range = 6--16 days). The study protocol was in accordance with the Declaration of Helsinki and approved by University of Florida's Institutional Review Board. Participants provided written informed consent prior to any study procedures. Structural MRI Acquisition {#s2-2} -------------------------- T1-weighted MPRAGE structural MRI scans were obtained prior to BOLD fMRI using a 32-channel head coil (3T Philips MRI). Scan parameters included: repetition time (TR) = 7.0 ms; echo time (TE = 3.2 ms); flip angle = 8°; field of view (FOV) = 240 × 240 × 170 mm; voxel size = 1 mm^3^. fMRI Data Acquisition {#s2-3} --------------------- fMRI scans were acquired on a 3 Tesla Philips MRI scanner using a 32-channel head coil. Task-based fMRI data were collected using a single-shot EPI sequence (36 slices, no gap, TR = 2,000 ms, TE = 30 ms, flip angle = 80°, FOV = 224 × 224 × 126, voxel size = 3.5 mm^3^). Task stimuli were presented on a screen and reflected onto a mirror visible to participants. Participants were trained on the task prior to the MRI session. tDCS Parameters and Application {#s2-4} ------------------------------- Bilateral frontal tDCS was delivered inside the scanner at 2 mA for 12 min (30 s ramp up/down) during the active condition using an MRI-compatible tDCS device (neuroConn DC-Stimulator MR). The sham condition was identical except that the stimulation period lasted only 30 s. Prior studies that examined 1 mA intensity consistently report increased excitation under the anode and decreased excitation under the cathode electrode in the context of motor-evoked potential TMS paradigms in the motor cortex (Nitsche and Paulus, [@B36]; Nitsche et al., [@B37],[@B39], [@B38]). In contrast, increasing the intensity to 2 mA has been shown to produce net excitation under both anode and cathode electrodes (Batsikadze et al., [@B5]; Reato et al., [@B43]). Our montage (F3/F4) and parameter selection of 2 mA was chosen to broadly target bilateral frontal brain regions involved in working memory with the intent of producing net excitation under the anode and cathode during active stimulation. Head measurements were performed using the 10--20 International EEG System to identify F3 (cathode) and F4 (anode) locations over left and right DLPFC, respectively. A 5 mm thick electrical conductive paste (Ten20 paste) was applied directly on two 5 × 7 cm^2^ rubber electrodes prior to placing the electrodes onto the scalp before the scan (Khadka et al., [@B25]; Woods et al., [@B54]). A thin layer of paste was also applied to the scalp at the F3/F4 locations for at least 30 min before the scan to allow paste to saturate the skin/scalp and achieve target impedance levels (\<2 kOhms). The electrodes were held in place with a rubber Soterix Easy-Head strap. The device was turned on before the start of the second run of the N-Back to allow ramp time to complete prior to start of the fMRI scan. A blinded six-digit code was entered into the device to initiate the randomized stimulation condition for the session. No adverse events were reported during the study. N-Back Task {#s2-5} ----------- During the task period (36 min; 12 min per run), participants performed three runs of the N-Back task (baseline, during, post) for each stimulation condition (active and sham). The task paradigm for each run consisted of four blocks of 2-Back and four blocks of 0-Back presented in a randomized order. In total, 12 blocks of both 2-Back and 0-Back were performed at each visit. During the 2-Back task, participants viewed stimuli of uppercase letters one at a time on the screen with a crosshair (+) as an inter-stimulus interval between stimuli (see [Figure 2](#F2){ref-type="fig"} for example). The stimulus appeared for 1 s, and the cross hair for 3 s, providing a 4 s window to make a response. Participants were instructed to press a button with their right index finger when the current letter matched the letter that appeared two trials back, and press a different button with their right middle finger when stimuli did not match the 2-Back pattern. The 0-Back was identical to the 2-Back task; however, it lacked the 2-Back pattern component and was used as an attention-control task. For 0-Back, participants were instructed to press a button with their index finger only for the letter "X" and press a different button with their middle finger for any other letter. As this study aimed to mechanistically evaluate the acute impact of tDCS on working memory networks, participants were trained thoroughly on the N-Back task (2- and 0-Back) outside of the scanner to facilitate stabilization of test-retest reliability and comprehension of the task. The intent of this procedure was to facilitate stability of the activation pattern in the working memory network across subsequent runs and between MRI sessions, and isolate detected change in connectivity due to influences from tDCS alone, rather than from the combination of tDCS and test-retest learning. The training consisted of separate 0-Back and 2-Back practice sessions with immediate performance feedback. Afterwards, participants performed an identical version of the task as it appeared in scanner without any feedback. The training took approximately 20 min and was performed before the MRI scan at both active and sham visits. Participants were reminded of instructions while inside the scanner prior to starting the task. N-Back accuracy was analyzed as a percent accuracy score on the 2-Back and 0-Back task. Reaction time was first log-transformed to achieve a normal distribution and the average was taken for each task. Performance (accuracy and reaction time) was evaluated using repeated measures analysis of variance (ANOVA) with stimulation condition (active vs. sham) and time (baseline, during, post) as factors in the model. {#F2} Neuroimaging Pre-processing {#s2-6} --------------------------- Spatial and functional pre-processing was performed using the CONN Toolbox v.2017.f[^1^](#fn0001){ref-type="fn"} and SPM12[^2^](#fn0002){ref-type="fn"} running in MATLAB 2015b (The Mathworks Inc, Natick, MA, USA). Pre-processing steps were used to segment high-resolution T1-weighted anatomical volumes for each participant into gray matter, white matter, and cerebrospinal fluid, and normalize to Montreal Neurological Institute (MNI) space. Functional volumes underwent realignment, slice-timing correction, and normalization to MNI space using the normalized EPI template image in CONN and a spatial Gaussian smoothing kernel of 8 mm. The artifact detection toolbox (ART) was applied to detect any motion artifacts. Motion parameters from the realignment process were evaluated after pre-processing to identify outliers. Any volumes exceeded a threshold of \>3 mm (translation) and \>1° rotation were removed from the analysis. Participants included in the analysis did not exceed thresholds for movement. BOLD data was filtered at a bandpass of 0.008-Inf Hz to reduce low-frequency drift and noise effects. Noise correction was performed using the anatomical component-based noise correction (aCompCor) method (Behzadi et al., [@B6]) implemented in the CONN Toolbox and SPM12. This method extracted principal components from white matter and CSF time series, which were then added as confounds in the denoising step within the CONN Toolbox (Behzadi et al., [@B6]; Whitfield-Gabrieli and Nieto-Castanon, [@B51]; Demirakca et al., [@B14]). This step was used to reduce any physiological and subject movement effects from the time series of interest, and enhance sensitivity, specificity, and validity for subsequent functional connectivity analyses (Behzadi et al., [@B6]; Whitfield-Gabrieli and Nieto-Castanon, [@B51]; Demirakca et al., [@B14]; Fallon et al., [@B15]). Seed-to-Target Regions of Interest (ROIs) Analyses {#s2-7} -------------------------------------------------- Using the BOLD signal, we implemented the CONN Toolbox to estimate measures of functional connectivity (Whitfield-Gabrieli et al., [@B50]; Wei et al., [@B49]). Previously, the Owen et al. ([@B40]) meta-analysis identified significant activated brain regions during the fMRI N-Back task. We selected the 15 seed regions of interest from this article (Owen et al., [@B40]) *a priori* to represent the working memory network in our analyses. Spherical regions of interest (ROIs) were created using the WFU PickAtlas GUI in SPM12 (Maldjian et al., [@B29], [@B28]) and the MNI-coordinates from Owen et [@B40] ([@B40]; [Table 1](#T1){ref-type="table"}); each ROI was created using the volume (mm^3^) reported in the meta-analysis. The default setting of 10 mm radius was used if no volume was reported. It should be noted that the Owen et al. ([@B40]) meta-analyses contained overlap between a subset of ROIs. In each case, radius size varied between identified ROIs. While selecting a single representative ROI would reduce the number of multiple comparisons, potentially increasing the number of significant findings in the current study, we chose a conservative approach aimed at maintaining consistency with the Owen et al. ([@B40]) meta-analysis ROIs. In the case of overlap between left frontal pole and one of the two left dorsolateral prefrontal ROIs, the inclusion of these two overlapping ROIs also provides a degree of ability to examine whether tDCS effects within the frontal pole are related to the portion of the frontal pole consistent with DLPFC or the larger extent of the frontal pole ROI. After pre-processing, denoising, and first level analysis (generalized psychophysiological interaction---gPPI; bivariate regression), second level models for the 2-Back task were analyzed for each of the 15 ROIs with combinations of stimulation session (active or sham) and time (baseline, during, post) as conditions in each model. Target regions only included the spherical ROIs to reduce the number of comparisons and focus on nodes within the working memory network. Model contrasts were created within a single stimulation session (e.g., during-active \> baseline-active) for each seed to all targets. Contrasts with significant connectivity changes (FDR \< 0.05) were further analyzed between active vs. sham sessions via a *post hoc* paired sample *t*-test to quantify the differences between sessions (e.g., during-active \> baseline-active vs. during-sham \> baseline-sham).*T*-tests were FDR 0.05 corrected for multiple comparisons. The 0-Back was analyzed in an identical manner. ###### Regions of interest (ROIs), Montreal Neurological Institute (MNI) coordinates, radius for spherical ROIs. Region *x* *y* *z* Radius ------------------------------ -------- -------- -------- -------- LH DLPFC −37.75 50.19 13.6 6.2 −46.26 22.71 18.6 14.3 LH frontal pole −37.75 50.19 13.6 7.5 LH inferior parietal lobule −37.09 −47.7 45.58 10 LH lateral premotor −26.32 6.75 53.46 9 −45.96 3.1 38.47 10 LH ventrolateral PFC −31.36 21.11 0.58 10 Medial cerebellum 3.12 −69.09 −24.69 3 RH DLPFC 44.53 38.76 24.43 12.5 RH inferior Parietal Lobule 44.97 −45.49 41.73 12.46 RH lateral premotor 31.96 11.01 49.8 15.83 31.96 11.01 49.8 10 RH medial posterior parietal 12.77 −63.71 55.28 14.8 RH ventrolateral PFC 35.58 23.26 −3.01 10 Supplementary motor area −0.588 18.57 40.65 10 *LH, left hemisphere; RH, right hemisphere*. Analytical Approach {#s2-8} ------------------- Using the CONN Toolbox, we examined ROI-to-ROI seed-to-target functional connectivity for each seed to all other specified targets in the working memory network via bivariate regression (gPPI) for 2-Back and 0-Back separately. The N-Back blocks in each run were synchronized with the functional data to only capture the task period and not rest or instruction periods. Fisher-transformed bivariate regression coefficients (beta values) between two ROI BOLD time-series were used to identify significant increases or decreases in functional connectivity between the seed-to-target. Results {#s3} ======= Functional Connectivity During 2-Back {#s3-1} ------------------------------------- We examined the effects of active vs. sham stimulation (at two separate visits) across three time points (baseline-active/sham, during-active/sham, and post-active/sham) during the 2-Back working memory task. Seed-to-target analyses demonstrated selective modulation of functional connectivity within the working memory network on the 2-Back task during-active stimulation as shown in [Figures 3](#F3){ref-type="fig"}, [4](#F4){ref-type="fig"} and [Table 2](#T2){ref-type="table"}. {#F3} {ref-type="table"} for seed to target location names. Nodes colored in black represent the seed location. Nodes color coded in red or blue indicate the target location. Red signifies significantly increased connectivity (FDR \< 0.05). Blue signifies significantly decreased connectivity (FDR \< 0.05). A, anterior; P, posterior; R, right; L, left.](fnagi-11-00051-g0004){#F4} ###### Significant seed to target connectivity values and test statistics for each contrast. Contrast Condition Seed Target Beta T~(15)~ P-Unc P-FDR Outcome -------------------- ----------- --------------------- --------------------- ------- --------- ------- --------- ------------------------ During \> Baseline Active VLPFC LH DLPFC LH 0.18 3.59 0.002 0.045\* Increased connectivity During \> Baseline Sham Lateral premotor LH Frontal pole LH −0.32 −3.4 0.003 0.042\* Decreased connectivity During \> Baseline Sham Lateral premotor LH DLPFC LH −0.38 −3.28 0.005 0.042\* Decreased connectivity Post \> During Active DLPFC LH VLPFC LH −0.06 −3.55 0.002 0.049\* Decreased connectivity Post \> During Active IPL RH Lateral premotor LH −0.17 −3.84 0.001 0.027\* Decreased connectivity *\*p-FDR \< 0.05*. During \> Baseline {#s3-2} ------------------ ### Active {#s3-2-1} A significant increase in functional connectivity was observed during-active stimulation compared to baseline-active when seeding in the left VLPFC targeting left DLPFC (P-FDR = 0.045). On an individual subject level, we observed that 13/16 (81%) participants evidenced increases in connectivity. ### Sham {#s3-2-2} We found significantly decreased connectivity during-sham stimulation compared to baseline-sham when seeding in the left lateral premotor targeting the left frontal pole (P-FDR = 0.042) and the left DLPFC (P-FDR = 0.042). ### Active vs. Sham {#s3-2-3} Left ventrolateral PFC (VLPFC) to Left DLPFC: *post hoc* analyses comparing during-active \> baseline-active to during-sham \> baseline-sham demonstrated significantly greater connectivity values in the active condition (*t* = 2.65, *df* = 15, *p* = 0.01, P-FDR = 0.018). Left Lateral Premotor to Left Frontal Pole: *post hoc* analyses comparing during-active \> baseline-active to during-sham \> baseline-sham demonstrated significantly lower connectivity values in the sham condition (*t* = 2.77, *df* = 15, *p* = 0.01, P-FDR = 0.047). Left Lateral Premotor to Left DLPFC: *post hoc* analyses comparing during-active \> baseline-active to during-sham \> baseline-sham demonstrated significantly lower connectivity values in the sham condition (*t* = 2.68, *df* = 15, *p* = 0.01, P-FDR = 0.047). Post \> During {#s3-3} -------------- ### Active {#s3-3-1} There was a significant decrease in connectivity between left DLPFC targeting left VLPFC post-active compared to during-active stimulation (P-FDR = 0.049). In addition, connectivity between right inferior parietal lobule (seed) and left lateral premotor (target) was also significantly decreased (P-FDR = 0.027). ### Sham {#s3-3-2} No significant changes in connectivity were found when comparing post-sham to during-sham stimulation (P-FDR \> 0.05). ### Active vs. Sham {#s3-3-3} Left DLPFC to Left VLPFC: *post hoc* analyses comparing post-active \> during-active to post-sham \> during-sham did not demonstrate a significant difference between sham and active conditions (P-FDR \> 0.05). Right IPL to Left lateral premotor: analyses comparing post-active \> during-active to post-sham \> during-sham demonstrated that connectivity values were significantly lower in the active condition (*t* = −2.56, *df* = 15, *p* = 0.021, P-FDR = 0.048). Post \> Baseline {#s3-4} ---------------- ### Active {#s3-4-1} No significant changes in connectivity were found when comparing post-active stimulation to baseline-active stimulation (P-FDR \> 0.05). ### Sham {#s3-4-2} No significant changes in connectivity were found when comparing post-sham stimulation to baseline-sham (P-FDR \> 0.05). Control Contrast: Baseline-Active vs. Baseline-Sham {#s3-5} --------------------------------------------------- No significant changes in connectivity were found between baseline-active vs. baseline-sham stimulation (P-FDR \> 0.05). Functional Connectivity During 0-Back {#s3-6} ------------------------------------- The effects of active vs. sham stimulation were evaluated across the three time points during the 0-Back task in an identical manner as 2-Back. Seed-to-target analyses were performed seeding in each ROI targeting all other ROIs in the network. No significant connectivity changes were identified during 0-Back (P-FDR \> 0.05). In addition, there were no significant connectivity differences between baseline-active vs. baseline-sham (P-FDR \> 0.05). N-Back Behavior {#s3-7} --------------- Percent accuracy and reaction time on the N-Back task did not significantly differ between active and sham stimulation conditions at the various time points (*p* \> 0.05) as reported in [Table 3](#T3){ref-type="table"}. Evaluation of linear and non-linear fits for performance data demonstrated a trend towards faster reaction time for 2-Back targets (tests of within-subject contrast fit with quadratic equation; *F* = 2.146, *df* = 1.15, partial eta squared = 0.125, *p* = 0.16) and improved 2-Back target accuracy (fit with linear equation; *F* = 3.199, *df* = 1.15, partial eta squared = 0.176, *p* = 0.09) during active-stimulation using repeated measures ANOVA within-subject contrast (time × stimulation condition). Nonetheless, performance was not significantly different for accuracy and reaction time. ###### Mean percent task accuracy and standard error on 2-Back and 0-Back for each condition; mean reaction time and standard error in milliseconds for each condition. N-Back performance (mean, std. error) Baseline-Active During-Active Post-Active Baseline-Sham During-Sham Post-Sham --------------------------------------- ----------------- --------------- -------------- --------------- -------------- -------------- 2-Back accuracy 84% (4%) 81% (5%) 82%(6%) 85% (4%) 85% (3%) 79% (4%) 0-Back accuracy 89% (3%) 92% (4%) 84% (6%) 89% (4%) 89% (3%) 88% (3%) 2-Back reaction time 959.3 (57.2) 847.2 (63.6) 868.5 (39.0) 1015.2 (64.5) 955.5 (62.5) 886.6 (51.4) 0-Back reaction time 791.2 (55.1) 738.5 (60.6) 756.5 (55.5) 830.3 (74.6) 760.3 (62.7) 739.6 (56.9) Discussion {#s4} ========== To the best of our knowledge, this is the first study to investigate the mechanistic acute and after-effects of bilateral frontal tDCS on functional connectivity of the working memory network in older adults. Our results demonstrate the ability of in-scanner tDCS paired with an N-Back task to acutely modulate functional connectivity in the working memory network. We found increased functional connectivity within the working memory network (seeding left VLPFC targeting left DLPFC) while participants performed the 2-Back task during active stimulation compared to baseline. Moreover, the *post hoc* between-condition analysis demonstrated significantly greater connectivity values in the active condition (P-FDR = 0.018). In contrast, during sham stimulation compared to baseline, we observed a significant decrease in functional connectivity during the 2-Back task (seeding left lateral premotor to left frontal pole and left DLPFC). Importantly, the changes in functional connectivity that occurred during active stimulation only occurred while performing the more challenging 2-Back task, and not during the less challenging 0-Back task. Additionally, increased left frontal connectivity during active stimulation trended towards decreasing post-stimulation, and may suggest that functional connectivity returned to baseline values after stimulation was stopped. This is also potentially supported by a significant decrease in connectivity that was observed at post-active stimulation compared to during active stimulation when seeding left DLPFC targeting left VLPFC. Collectively, these data suggest the ability of acute bilateral F3/F4 tDCS at 2 mA to selectively modulate frontal functional connectivity of working memory related brain regions in older adults during stimulation. These results provide important insight into: (1) selective effects of tDCS on functional connectivity; (2) state-dependent effects of tDCS; (3) the importance of timing tDCS delivery; and (4) effects of 2 mA stimulation under the cathode electrode. Our results demonstrated an increase in connectivity between two frontal lobe structures (VLPFC and DLPFC) during-active tDCS. Sub-regions of the frontal lobe are essential for manipulation of verbal and spatial knowledge (Barbey et al., [@B4]). Functionally, the VLPFC has exhibited enhanced activity during various cognitive control processes such as selection, maintenance, and retrieval of goal directed information. The DLPFC has shown critical involvements during task delay periods and facilitates the ability to keep task-relevant information active, in addition to manipulating that information to accomplish the task. Together, the DLPFC and VLPFC demonstrate enhanced activity in cognitive control processes such as monitoring, manipulating, or processing goal directed information, all processes vital to working memory (D'Esposito et al., [@B13]; Petrides, [@B42]; Blumenfeld et al., [@B10]). Our finding of functional connectivity increase during-active stimulation on 2-Back, and not 0-Back is consistent with previous investigations of state-dependent effects of tDCS. Gill et al. ([@B17]) provided evidence that cognitive enhancement capacity of tDCS may depend on the nature of the task being performed during stimulation (Gill et al., [@B17]). A challenging and more cognitively demanding 3-Back task produced greater proficiency on a separate working memory task (Paced Auditory Serial Addition Task, PASAT). In contrast, the less cognitively demanding 1-Back task performed during stimulation did not result in an improvement on the PASAT. Our results support prior research that the effects of tDCS are state dependent on more cognitively demanding tasks (2-Back) vs. less demanding tasks (0-Back). Previous studies examining tDCS paired with cognitive training indicate the importance of timing of stimulation. Martin et al. ([@B31], [@B30]) demonstrated that gains from tDCS were significantly greater when stimulation was delivered during an adaptive N-Back cognitive training task vs. stimulation delivered prior to training (Martin et al., [@B31], [@B30]). These data suggest that optimal gains from tDCS may be achieved by "co-stimulating" neural networks through both behavior and electrical stimulation. Our finding of increased functional connectivity during active stimulation, and not at baseline or post stimulation, appears to be consistent with this notion, and further highlights the importance of timing and delivery of tDCS for optimizing brain-based effects. While absence of post-stimulation after effects on functional connectivity do not preclude behavioral effects of tDCS, these data might suggest that there is an optimal window for leveraging brain-based changes of tDCS during the period of stimulation. However, it is important to use caution when interpreting our results in this domain, as our participants only demonstrated trends for improved accuracy and reaction time in the current study. Previous studies in tDCS and the motor cortex have shown that 1 mA vs. 2 mA of intensity can have differing physiological responses under the anode vs. cathode electrodes by measuring changes in motor evoked potentials (MEPs). At 1 mA, an increase in excitability is typically seen under the anode electrode, whereas the cathode electrode demonstrates a decrease in excitability (Bindman et al., [@B9]; Nitsche and Paulus, [@B36]; Nitsche et al., [@B37],[@B39]). Batsikadze et al. ([@B5]) demonstrated that 2 mA tDCS can produce a net increase in excitability under both the anode and cathode electrodes (Batsikadze et al., [@B5]). At present, there are no MEP-like markers within the frontal lobes that indicate physiological change induced via tDCS. However, our finding of increased functional connectivity of left frontal regions during active stimulation may support prior motor cortex findings of increased excitability under the cathode electrode at 2 mA in frontal cortices. If the cathode electrode induced a decrease in excitability in underlying tissue, we might expect to see decreased connectivity, or no change, yet we saw an increase in functional connectivity lateralized to the left hemisphere where the cathode electrode was placed. Therefore, these data appear, at least in part, to support a growing body of research suggesting that 2 mA tDCS may result in increased excitability under the cathode electrode. When interpreting the results, limitations of the study should be considered. Artifacts in the MRI environment could alter imaging data; we took into consideration how introducing current into the MRI may cause artifacts. Our task paradigm randomized and switched between 2- and 0-Back for during-active stimulation. We did not identify any significant changes in functional connectivity while 0-Back was performed in during-active stimulation. The lack of change in during-active vs. during-sham stimulation on the 0-Back task provides further evidence that the observed connectivity increase was not caused by potential artifacts directly induced by current delivery. This suggests the connectivity changes that occurred during active stimulation on the 2-Back were a result of stimulation. This study did not demonstrate statistically significant behavioral improvements in the active vs. sham session, although a trend for significance and medium effect sizes were present. Our study trained participants prior to in-scanner performance to stabilize the behavioral and network performance across multiple runs, to better enable direct evaluation of acute impact of delivered current flow on functional connectivity in working memory networks. As expected, stable and high levels of performance likely minimized our ability to detect large changes in N-back performance. In addition, limited power due to small sample size likely played a role in these results. It is also important to consider the range of cognitive abilities of the selected sample. The current study used a liberal MoCA cutoff of 20, with an average of 26.5. Thus, the older adult sample in the current study included not only participants that could be classified as "normal healthy" older adults, but also some participants that may not be entirely cognitive normal. Nonetheless, participants were excluded for diagnosis of mild cognitive impairment or dementia. There are also nuances to performing fMRI in older adults that must be considered. The process of aging on the brain may uniquely impact adaptive network mechanisms. Thus, effects found in the current study may not extend to younger adults. Future studies replicated the current findings in younger adults would help to expand our understanding of tDCS mechanisms across the lifespan. As the population studied involves older adults, 25% or more of participants may have beta-amyloid present in the brain. As beta-amyloid can impact functional connectivity, this factor may play a role in our overall results and further stresses the need for follow-up studies in young adults and older adults with corresponding beta-amyloid positron emission tomography. Conclusion {#s5} ========== The underlying mechanism(s) of action of tDCS and its effects on brain function are not yet fully understood. Our results demonstrated that acute bilateral frontal tDCS selectively modulates functional brain connectivity in the working memory network of older adults. These data indicate the importance of timing of stimulation and the critical influence that state/cognitive effort may have during the period of stimulation. Further research pairing tDCS with unimodal and multimodal neuroimaging is needed to continue elucidating the mechanism(s) of tDCS brain-based effects on cognition. Data Availability {#s6} ================= The datasets generated for this study are available on request to the corresponding author. Author Contributions {#s7} ==================== NN, AO'S, AI, RT, LR, EP, RC and AW contributed to the manuscript. NN and AW performed data analyses. All authors approved the final version of the manuscript. Conflict of Interest Statement {#s8} ============================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. **Funding.** This work was supported in part by the National Institute of Aging/National Institutes of Health (K01AG050707, R01AG054077), the Evelyn F. McKnight Brain Research Foundation, and the University of Florida Center for Cognitive Aging and Memory. ^1^<https://www.nitrc.org/projects/conn> ^2^<https://www.fil.ion.ucl.ac.uk/spm/> [^1]: Edited by: P. Hemachandra Reddy, Texas Tech University Health Sciences Center, United States [^2]: Reviewed by: Milton Cesar Biagioni, New York University, United States; Rongqiao He, Institute of Biophysics (CAS), China

Background ========== Education level is one indicator of socioeconomic position which, in several countries including South Korea, is provided though death certificate data. Its validity determines the usefulness of death certificate data for exploring the association between socioeconomic position and mortality \[[@B1]\]. Although education is likely to be more accurately reported than measures such as income, some evidence from Western studies suggests inconsistencies can occur between education reported in health surveys and that subsequently recorded on death certificates \[[@B2]-[@B4]\]. In a study by Lerchen and colleagues, a considerable gap was found between education reported in health interview data and education reported in subsequent follow-up interviews from surviving spouses after death \[[@B5]\]. Contradictory results have also been shown in other studies, with some demonstrating a tendency for education to be reported at a higher level on death certificates than at baseline health survey, \[[@B2],[@B3]\] while other studies showed this was not true \[[@B1]\]. In addition, the reliability of reported education between interview survey and death certificate data may vary with decedent\'s gender, age, and educational level \[[@B3]\]. Education can be reported less accurately for female decedents since they are less likely than males to have surviving spouses and are more likely to die at an older age \[[@B3]\]. The overstatement of education was more common among those decedents who, according to health survey data, had a lower education level \[[@B3]\]. However, in a later study, Rosamond and colleagues reported that death certificate education level had substantial validity and that misclassification was similar in both genders \[[@B1]\]. Reports have shown that overstatement bias was more pronounced in older than younger decedents \[[@B2]\]. Apart from these studies, little has been published on the validity and reliability of education between health survey data and death certificate data. Education is known as a socioeconomic position indicator that is less likely to change over time than occupation and income \[[@B4]\]. As a result, misclassification of education may not vary with the length of time between acquisition of health survey and death certificate data. However, no studies have proven this. In addition, little is known about the misclassification of education by cause of death. Recently, several studies have been conducted to examine educational inequalities in mortality with use of census and death certificate data in South Korea \[[@B6]-[@B8]\]. For example, in prior studies we presented age-, gender-, and cause-specific mortality inequalities by education \[[@B6]\] and examined trends in mortality inequalities by education \[[@B7]\]. However, the possibility of numerator-denominator bias \[[@B9]\] could not be excluded in these studies since unlinked data were used \[[@B6],[@B7]\]. If education levels of the census were more inflated than those of death certificate data, exaggerated educational mortality differentials could be made. Son claimed that the reliability of education level between survey and death certificate data can be poor \[[@B10]\]. A prior Korean study examined the reliability of reported education level between survey and death certificate data \[[@B11]\]. However, due to the small number of deaths, a more detailed analysis by covariates such as cause of death and duration between health survey and death was not possible and the validity (sensitivity and specificity) of each educational category was not examined. Information on gender, age, and cause-specific level of validity is useful to assess any bias in the gender, age, and cause-specific magnitude of socioeconomic mortality inequalities measured with census and death certificated data. Therefore, the purpose of this study was to compare education level recorded on death certificates with that reported before death in a nationally representative cohort of participants in the National Health and Nutrition Examination Survey (NHANES) of South Korea. Specifically, we examined the difference in the magnitude of validity and reliability for educational attainment according to gender, age group, duration between health survey and death, educational category, and cause of death. Methods ======= The Institutional Review Board of the Asan Medical Center, Seoul, approved this study. The health survey data used in this study were the 1998 and 2001 NHANES conducted by the Korea Institute for Health and Social Affairs. Information was collected from a stratified multistage probability sample of South Korean households representing the civilian, non-institutionalized population \[[@B12]\]. Participation rates for the 1998 and 2001 NHANES were 89.8% and 77.3%, respectively. At a baseline visit, information on education level was obtained from an adult household member. For each family member, the highest education completed was reported. This data contained unique 13-digit personal identification numbers that were individually linked to mortality data from the Korean National Statistical Office. By law, all Korean deaths must be reported to the Korea National Statistical Office. We identified deaths that occurred in the years 1999--2005 for those in the 1998 NHANES and 2002--2005 for those in the 2001 NHANES. Duration of mortality follow-up was 7.1 years. Similar to that in NHANES, levels of education reported on death certificates were determined by the highest level of education completed. The information on education was based on reports from the decedent\'s kin. Covariates considered in our analyses were gender, age, duration between survey and death, and cause of death. Duration between health survey and death was grouped into categories of 3.4 years or less and 3.5--7.1 years to allocate samples equally into groups. Causes of death were coded according to the 10^th^revision of the *International Classification of Disease*(ICD-10) and grouped as cancer (C00--C97), cardiovascular disease (I00--I99), external causes (V01--Y98), and other. For our analysis, we first compared education level between NHANES data and death certificate data and calculated the proportion of concordant and discordant pairs. Then we determined sensitivity and specificity of education in death certificate using the standard method, \[[@B13]\] with education reported in NHANES considered the gold standard. Sensitivity and specificity by cause of death were not presented due to the small size of each education group. We also estimated the agreement rates of education level between NHANES data and death certificate data. Overstatement and understatement rates were calculated. Overstatement refers to the inflated education level recorded in death certificates compared with that in NHANES, while understatement means the reverse. Confidence intervals (CI) for sensitivity, specificity, agreement rates, overstatement rates and understatement rates were calculated by asymptotic method. For these analyses, education data from death certificates and NHANES were categorized into five levels (college or higher, high school, middle school, elementary school and no education) and three levels (college or higher, high and middle school, elementary school or less). This was consistent with categorizations used in prior two studies \[[@B6],[@B7]\]. Finally, after adjusting for all covariates, we estimated odds ratios with 95% CI for agreement in educational attainment when five and three educational groups were used. All statistical analyses were performed with SAS statistical software. Results ======= Unique personal identification numbers of 12,801 subjects aged 20+ (6,900 for the 1998 NHANES and 5,901 for the 2001 NHANES) were linked to mortality information. A total of 513 deaths were identified: 376 deaths for the 1998 NHANES and 137 deaths for the 2001 NHANES. Of these, 511 (99.6%) had education levels available from both the death certificate and NHANES. One decedent had no education information in the NHANES and another decedent\'s education was not recorded in the death certificate. Of the 511 participants included in our analyses, 57.9% (n = 296) were men and 61.1% (n = 312) were aged 65+. The mean age of subjects was 66.4 years (SD 14.4). Decedent age varied with education from 52.7 years (SD 16.3) for those with college education or higher to 75.8 years (SD 9.2) for those with no education. A marked gender difference in education was also found with 24.7% (n = 73) of men and 67.9% (n = 146) having no education recorded in the NHANES. While the proportion of those having college or higher education only reached 2.3% (n = 5) for women, the proportion for men was 8.8% (n = 26). Table [1](#T1){ref-type="table"} shows the cross-classification of education level by death certificate and the adult household member\'s report at baseline visit in the NHANES of South Korea. For 511 deaths, 150 (29.4%) discordant pairs were found. Inconsistencies were most common when education was reported as middle school in NHANES. Of 54 participants whose education was reported as middle school in NHANES, only 25 (46.3%) were recorded as having middle school education in death certificate data. ###### Comparison of education level between Korea National Health and Nutrition Examination Survey (NHANES) data and death certificate data among 513 decedents Education recorded on death certificate Education reported in the 1998 and 2001 NHANES ----------------------------------------- ------------------------------------------------ ------------ ------------ ------------- ------------- ----------- ----- College or higher 26 (83.9) 3 (4.9) 1 (1.9) 0 (0.0) 0 (0.0) 0 (0.0) 30 High school 3 (9.7) 42 (68.9) 13 (24.1) 3 (2.0) 3 (1.4) 0 (0.0) 64 Middle school 1 (3.2) 8 (13.1) 25 (46.3) 18 (12.2) 2 (0.9) 0 (0.0) 54 Elementary school 1 (3.2) 7 (11.5) 12 (22.2) 98 (66.7) 44 (20.1) 0 (0.0) 162 No education 0 (0.0) 0 (0.0) 3 (5.6) 28 (19.1) 170 (77.6) 1 (100.0) 202 Unknown 0 (0.0) 1 (1.6) 0 (0.0) 0 (0.0) 0 (0.0) 0 (0.0) 1 Total 31 (100.0) 61 (100.0) 54 (100.0) 147 (100.0) 219 (100.0) 1 (100.0) 513 Note: The denominator for the calculation is the number of deaths classified by level of education noted on the NHANES survey, so that the column percentages add to 100%. Table [2](#T2){ref-type="table"} presents the sensitivity and specificity of education in death certificate when education level derived from the health survey was considered as the gold standard. The sensitivity and specificity of the death certificate in classifying a decedent as having college or higher education were 0.84 (95% CI: 0.71--0.97) and 0.99 (95% CI: 0.98--1.00), respectively. However, for middle school education, the sensitivity was poor. While the level of sensitivity for high and middle school was greater in men than women, this pattern could not be generalized to other educational groups. A very low sensitivity (0.20) for middle school in women was based on the small sample number (n = 5). With the exception of high school, the sensitivity of education for ages 65+ was similar to that for ages 20--64. In addition, the magnitude of sensitivity and specificity did not vary with the length of time between acquisition of health survey and death certificate data. ###### Sensitivity, specificity and their 95% confidence intervals (CI)\* of education level recorded in death certificates using education reported in National Health and Nutrition Examination Survey (NHANES) of South Korea as the gold standard (N = 511) Education N Sensitivity (95% CI) N Specificity (95% CI) N Sensitivity (95% CI) N Specificity (95% CI) --------------------------- ------------------------------------------------------------- ---------------------------------------------------------- ----- ---------------------- ----- ---------------------- ----- ---------------------- All subjects College or higher 31 0.84 (0.71--0.97) 480 0.99 (0.98--1.00) High and middle school 114 0.77 (0.69--0.85) 397 0.92 (0.90--0.95) High school 60 0.70 (0.58--0.82) 451 0.95 (0.93--0.97) Middle school 54 0.46 (0.33--0.60) 457 0.94 (0.91--0.96) Elementary school or less 366 0.93 (0.90--0.96) 145 0.84 (0.78--0.90) Elementary school 147 0.67 (0.59--0.74) 364 0.82 (0.79--0.86) No education 219 0.78 (0.72--0.83) 292 0.89 (0.86--0.93) Men Women College or higher 26 0.81 (0.66--0.96) 270 0.99 (0.97--1.00) 5 1.00 (1.00--1.00) 210 1.00 (1.00--1.00) High and middle school 100 0.80 (0.72--0.88) 196 0.90 (0.86--0.94) 14 0.57 (0.31--0.83) 201 0.95 (0.92--0.98) High school 51 0.71 (0.58--0.83) 245 0.93 (0.89--0.96) 9 0.67 (0.36--0.97) 206 0.98 (0.96--1.00) Middle school 49 0.49 (0.35--0.63) 247 0.91 (0.88--0.95) 5 0.20 (0.00--0.55) 210 0.97 (0.94--0.99) Elementary school or less 170 0.91 (0.86--0.95) 126 0.87 (0.81--0.92) 196 0.95 (0.92--0.98) 19 0.68 (0.48--0.89) Elementary school 97 0.69 (0.60--0.78) 199 0.80 (0.75--0.86) 50 0.62 (0.49--0.75) 165 0.85 (0.79--0.90) No education 73 0.64 (0.53--0.75) 223 0.92 (0.88--0.96) 146 0.84 (0.78--0.90) 69 0.81 (0.72--0.90) Ages 20--64 Ages 65 College or higher 23 0.83 (0.67--0.98) 176 0.99 (0.97--1.00) 8 0.88 (0.65--1.00) 304 0.99 (0.98--1.00) High and middle school 82 0.77 (0.68--0.86) 117 0.85 (0.78--0.91) 32 0.78 (0.64--0.92) 280 0.96 (0.93--0.98) High school 44 0.80 (0.68--0.91) 155 0.90 (0.86--0.95) 16 0.44 (0.19--0.68) 296 0.98 (0.96--0.99) Middle school 38 0.45 (0.29--0.61) 161 0.91 (0.87--0.96) 16 0.50 (0.26--0.75) 296 0.95 (0.92--0.97) Elementary school or less 94 0.84 (0.77--0.91) 105 0.83 (0.76--0.90) 272 0.96 (0.94--0.98) 40 0.88 (0.77--0.98) Elementary school 67 0.69 (0.58--0.80) 132 0.80 (0.73--0.86) 80 0.65 (0.55--0.75) 232 0.84 (0.79--0.89) No education 27 0.52 (0.33--0.71) 172 0.94 (0.91--0.98) 192 0.81 (0.76--0.87) 120 0.83 (0.76--0.89) Duration between health survey and death, 3.4 years or less Duration between health survey and death, 3.5--7.1 years College or higher 6 1.00 (1.00--1.00) 249 1.00 (0.99--1.00) 25 0.80 (0.64--0.96) 231 0.99 (0.97--1.00) High and middle school 60 0.80 (0.70--0.90) 195 0.92 (0.89--0.96) 54 0.74 (0.62--0.86) 202 0.93 (0.89--0.96) High school 33 0.67 (0.51--0.83) 222 0.95 (0.92--0.98) 27 0.74 (0.58--0.91) 229 0.95 (0.92--0.98) Middle school 27 0.44 (0.26--0.63) 228 0.92 (0.89--0.96) 27 0.48 (0.29--0.67) 229 0.95 (0.92--0.98) Elementary school or less 189 0.92 (0.88--0.96) 66 0.83 (0.74--0.92) 177 0.94 (0.90--0.97) 79 0.85 (0.77--0.93) Elementary school 73 0.66 (0.55--0.77) 182 0.81 (0.75--0.86) 74 0.68 (0.57--0.78) 182 0.84 (0.79--0.89) No education 116 0.77 (0.69--0.84) 139 0.91 (0.86--0.95) 103 0.79 (0.71--0.87) 153 0.88 (0.83--0.93) \*Confidence intervals for sensitivity and specificity were calculated by asymptotic method. Table [3](#T3){ref-type="table"} shows the overall agreement rate, overstatement rate and understatement rates with 95% CI. When education level was categorized into five groups (college or higher, high school, middle school, elementary school, and no education), the overall agreement rate was 70.7% (95% CI: 66.8%--74.6%). The magnitude of agreement rate for education did not vary with gender, age, and duration between health survey and death. This was generally true for cause specific analyses. The overstatement rate tended to be greater than the understatement rate. The education level of 87 (17.0%) subjects was reported as greater in death certificate than NHANES data while 63 (12.3%) subjects were recorded with lower education in death certificate data than in NHANES data. This was true for both gender, age and causes of death except for external causes. Meanwhile, the understatement rate for external causes tended to be greater than the rates for other three broad causes. When educational level was grouped into five categories, understatement rate for external causes was 20% while understatement rates for other broad causes were about 10%. Based on logistic analysis when the educational level was grouped into three categories, odds ratio of understatement for external causes (reference group = other broad causes) was 2.85 (1.15--7.05). ###### Agreement rate (%), overstatement\* rate (%), understatement\* rate (%), and their 95% confidence intervals (CI)\* for education level by gender, age, and cause of death: Reliability between Korea National Health and Nutrition Survey (NHANES) data and death certificate data No. of subjects Agreement rate (95% CI) Overstatement rate (95% CI) Understatement rate (95% CI) -------------------------------------------------------------------------------------------------------------- ----------------- ------------------------- ----------------------------- ------------------------------ **Five educational groups (college or higher, high school, middle school, elementary school, no education)** All subjects 511 70.7 (66.8--74.6) 17.0 (13.7--20.3) 12.3 (9.5--15.1) Gender  Men 296 65.9 (60.5--71.3) 19.3 (14.8--23.8) 14.9 (10.8--19.0)  Women 215 77.2 (71.6--82.8) 14.0 (9.4--18.6) 8.8 (5.0--12.6) Age groups  20--64 199 65.8 (59.2--72.4) 17.6 (12.3--22.9) 16.6 (11.4--21.8)  65+ 312 73.7 (68.8--78.6) 16.7 (12.6--20.8) 9.6 (6.3--12.9) Duration between health survey and death  3.4 years or less 255 69.4 (63.8--75.1) 19.2 (14.4--24.1) 11.4 (7.5--15.3)  3.5--7.1 years 256 71.9 (66.4--77.4) 14.8 (10.5--19.2) 13.3 (9.1--17.4) Cause of death  Cancer 124 71.0 (63.0--79.0) 17.7 (11.0--24.4) 11.3 (5.7--16.9)  Cardiovascular disease 132 69.7 (61.9--77.5) 18.2 (11.6--24.8) 12.1 (6.5--17.7)  External causes 60 65.0 (52.9--77.1) 15.0 (6.0--24.0) 20.0 (9.9--30.1)  Other causes 195 72.8 (66.6--79.0) 16.4 (11.2--21.6) 10.8 (6.4--15.2) **Three educational groups (college or higher, high and middle school, elementary school or less)** All subjects 511 88.9 (86.2--91.6) 5.9 (3.9--7.9) 5.3 (3.4--7.2) Gender  Men 296 86.2 (82.3--90.1) 6.8 (3.9--9.7) 7.1 (4.2--10.0)  Women 215 92.6 (89.1--96.1) 4.7 (1.9--7.5) 2.8 (0.6--5.0) Age groups  20--64 199 80.9 (75.4--86.4) 8.5 (4.6--12.4) 10.6 (6.3--14.9)  65+ 312 93.9 (91.2--96.6) 4.2 (2.0--6.4) 1.9 (0.4--3.4) Duration between health survey and death  3.4 years or less 255 89.4 (85.6--93.2) 6.3 (3.3--9.2) 4.3 (1.8--6.8)  3.5--7.1 years 256 88.3 (84.3--92.2) 5.5 (2.7--8.3) 6.3 (3.3--9.2) Cause of death  Cancer 124 84.7 (78.4--91.0) 11.3 (5.7--16.9) 4.0 (0.6--7.4)  Cardiovascular disease 132 87.1 (81.4--92.8) 6.8 (2.5--11.1) 6.1 (2.0--10.2)  External causes 60 86.7 (78.1--95.3) 1.7 (0.0--5.0) 11.7 (3.6--19.8)  Other causes 195 93.3 (89.8--96.8) 3.1 (0.7--5.5) 3.6 (1.0--6.2) \*Overstatement refers to the inflated educational level recorded in death certificate compared with that in NHANES, while understatement means the reverse. Confidence intervals for agreement rate, overstatement rate, and understatement rate were calculated by asymptotic method. As presented in Table [3](#T3){ref-type="table"}, the magnitude of reliability increased as the number of educational groups decreased. When the education level was categorized into three groups (college or higher, high and middle school, elementary school or less), the overall agreement rates was 88.9% (95% CI: 86.2%--91.6%). For this categorization of education, overstatement and understatement rates diminished and there was no marked evidence of education inflation in death certificate data compared with health survey. Table [4](#T4){ref-type="table"} presents odds ratio (95% CI) of agreement when five and three educational groups were used to assess agreement. When five educational groups were used, the middle and elementary school education groups showed a significantly smaller likelihood of agreement after adjusting for gender, age (10-year age), duration between health survey and death, and cause of death. However, this was not true when assessing agreement with three educational groups. No significant difference in odds ratio was found for gender, age, duration between health survey and deaths, and cause of death, likely in part due to small sample size. ###### Odds ratio and 95% confidence interval (CI) of agreement in education level (N = 511) OR (95% CI) of agreement for five educational groups OR (95% CI) of agreement for three educational groups ------------------------------------------- ------------------------------------------------------ ------------------------------------------------------- Gender  Men 1.00 (reference) 1.00 (reference)  Women 1.47 (0.93--2.31) 1.27 (0.65--2.50) Age groups  10-year age (continuous) 1.10 (0.92--1.31) 1.20 (0.95--1.52) Duration between health survey and deaths  3.4 years or less 1.00 (reference) 1.00 (reference)  3.5--7.1 years 1.09 (0.74--1.63) 0.92 (0.52--1.65) Education in health survey  College or higher 1.00 (reference) 1.00 (reference)  High school 0.46 (0.15--1.41) 0.58 (0.20--1.73)  Middle school 0.17 (0.06--0.51) 0.58 (0.20--1.73)  Elementary school 0.33 (0.12--0.95) 1.59 (0.51--5.00)  No education 0.47 (0.15--1.45) 1.59 (0.51--5.00) Cause of death  Cancer 1.00 (reference) 1.00 (reference)  Circulatory disease 0.81 (0.46--1.42) 1.00 (0.48--2.11)  External causes 0.89 (0.44--1.78) 1.74 (0.68--4.47)  Other causes 0.91 (0.54--1.55) 2.00 (0.92--4.35) Note: all variables were entered into logistic models simultaneously. Discussion ========== Results of this study showed that the proportion of deaths without recorded education was very low, only one of 513 (0.2%) in death certificate, far less than prior US studies in which information on education for about 15%--27% of all deaths occurred was not available in the death certificate \[[@B1],[@B2]\]. According to death certificate data covering all South Koreans, the rate of failure to report the decedent\'s educational attainment was 0.3% for men and 0.4% for women aged 35--64 \[[@B6]\]. This may be due to the Korean death certification system where any vague or missing item on the death certificate is further clarified by the Korea National Statistical Office via telephone \[[@B6]\]. This low percentage of missing education information in South Korean death certificate data could provide a great potential to monitor socioeconomic inequalities in mortality. Direct comparison of the validity and reliability level found in this study with prior reports in other countries may be problematic because of differences in age distribution and educational categories used. Factors associated with participation to health survey may also affect the overall validity and reliability. Nevertheless, we may conclude that this study showed a similar or higher level of validity and reliability in education information than in Western studies. For example, the agreement rate for three educational groups (less than high school graduate, high school graduate, and greater than high school education) in a prior study was 84% \[[@B1]\]. Results of this study showed an agreement rate of 88.9% for the same categories. A prior study presented a sensitivity of 0.85 for college or higher education, \[[@B2]\] which is similar to what was found in this study (0.84). It should be noted that in prior studies, a substantial number of subjects with missing education information (15%--27%) were found but were eliminated from analyses that reported the agreement rate and sensitivity \[[@B1],[@B2]\]. Despite the relatively high level for overall validity and reliability, low sensitivity and agreement rates were found in this study for participants with middle school education. After adjusting for covariates, middle and elementary school education groups were less concordantly recorded in death certificate data. One possible explanation for this may be related to historical changes in the education system in South Korea. During Japanese colonial occupation (1910--1945) and the Korean War (1950--1953), South Korea could not establish a nationwide educational curriculum \[[@B14]\]. An education system divided by elementary school (6 years), middle school (3 years), and high school (3 years) began in 1954 when the Ministry of Education of Korea first introduced an official nationwide curriculum. Various types of educational attainment, which are difficult to group into the categories used in this study, would be inevitable among decedents whose childhood and adolescent education was achieved before the modern South Korean educational system was established. Difficulties in determining one\'s educational level have also remained in vital statistic records. For example, middle and high school levels were grouped as one category in early 1990s death certificate data \[[@B7]\]. However, when middle school and elementary school education were collapsed into high school and no education respectively, statistically significant differences in agreement of education did not appear. It was hypothesized that male decedents\' education would be more accurately reported by surviving female spouses than female decedents\' due to the gender gap in life expectancy \[[@B3]\]. However, this study showed that the agreement rate of education for women was generally greater than that for men. Statistically significant gender differences in odds of agreement were found for both educational categories (five and three groups) when only the gender variable was entered into logistic model (data not shown). In South Korea a woman\'s death is more frequently reported by her children, who may know more about her education than other informants. The widow\'s cognitive or communication problems may lower the agreement rate of male decedents\' education. However, the comparison of accuracy level between widow\'s reporting (for male decedents\' education) and children\'s reporting (for female decedents\' education) needs to be further explored. If a woman\'s informants were uncertain as to which education levels she achieved, it might be difficult to accurately deduce and report the actual education level. However, in this study reports of women\'s education in the NHANES were concentrated around the \"no education\" level (67.9%). The simplicity of a woman\'s education might help informants for death certification to report education information accurately. Based on logistic model, significantly greater odds ratios of agreement for women compared with men became insignificant after adjusting for education (data not shown). Although the magnitude of overall agreement rate for education did not vary with causes of death, understatement rate for external causes tended to be greater than the rates for other broad causes. Higher understatement rate for external causes was mainly due to stigmatized causes of death rather than other external causes such as transport accident. Understatement rate for suicide (n = 17) was 35.3% (n = 6) while the rate for transport accident (n = 24) was 8.3% (n = 2). Logistic analysis presented a greater odds ratio of understatement for suicide (data not shown). This suggests that a considerable understatement for education may occur in stigmatized causes of death. Overstatement and understatement of education by descendants may vary with regions with different cultural background. Results of this study showed a tendency for inflation of education when grouped into five categories but did not reveal any evidence of inflation when education was collapsed into three categories. These findings agree with prior US studies where inflation was found in six educational groups \[[@B3]\] but did not appear in collapsed educational groups \[[@B1]\]. Thus, when using unlinked data a more collapsed categorization of education would produce a more valid result on estimating mortality inequality. However, as the sensitivity of college or higher education was generally greater than that of no education and the pattern did not vary with covariates, it can be suggested that educational mortality differentials between no education and college or higher education in the previous study \[[@B6]\] were underestimated. Indeed, several longitudinal mortality follow-up studies presented a substantial difference in mortality by education in South Korea \[[@B15],[@B16]\]. Given that the level of agreement did not significantly vary with cause of death, cause-specific analysis may not result in biased results. However, a more collapsed categorization in education would be recommended especially when a more definitive conclusion regarding educational mortality inequality is required. Conclusion ========== The low percentage of missing education information in South Korean death certificate data could provide a great potential to monitor socioeconomic inequalities in mortality. Despite the relatively high level for overall validity and reliability, a more collapsed categorization of education would produce a more valid result on estimating mortality inequality when using unlinked data. List of abbreviations ===================== CI, confidence intervals; NHANES, National Health and Nutrition Examination Survey of South Korea; SD, standard deviation. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= YHK conceived and designed the study with advice from HRK and JWL. YHK analyzed and interpreted the data and drafted the paper. HRK and JWL interpreted the data and critically revised the draft of the paper. All authors read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2458/7/294/prepub>

Introduction {#Sec1} ============ Coronary heart disease is one of the leading causes of mortality in the world today (MacKay & Mensah, [@CR27]; Murray & Lopez, [@CR31]), and although new cardiac treatments have helped fight coronary heart disease in recent years, an estimated 1/3 of coronary heart disease risk factors remain elusive (Gudnason, [@CR22]). The addition of psychological factors to standard biomedical risk factors may enhance the prediction of patients at risk. Initial research on the Type A behavior pattern suggested that psychological factors were related to increased risk of heart attacks, but further investigations on Type A behavior were inconclusive (Rozanski et al., [@CR44]). Subsequently, researchers turned their focus towards isolated factors such as hostility, depression, anxiety, social isolation, and chronic stress (Matthews, [@CR29]; Rozanski et al., [@CR44], [@CR43]; Strike & Steptoe, [@CR51]) to document a relationship between psychological factors and poor cardiac prognosis (Rozanski et al., [@CR43]). Clustering of psychological factors within individuals enhances the risk of adverse health outcomes (Rozanski et al., [@CR43]; Strike & Steptoe, [@CR51]), and this clustering may partly be attributed to a specific vulnerability in the realm of personality (Dimsdale, [@CR16]). The distressed (Type D) personality construct was originally developed to identify cardiac patients who are vulnerable to emotional and interpersonal difficulties (Denollet, [@CR6]; Denollet et al., [@CR15]). Type D individuals tend to experience negative emotions (elevated score on negative affectivity) while not discussing them with others due to fear of rejection (elevated score on social inhibition) (Denollet et al., [@CR15]). Type D personality has been associated with poor quality of life and increased morbidity and mortality in cardiac patients (Denollet et al., 1996, [@CR11]; Kupper & Denollet, [@CR26]; Pedersen & Denollet, [@CR34]). The prevalence of Type D ranges from 28 to 32% across different cardiovascular conditions, including ischemic heart disease, chronic heart failure, and peripheral artery disease. The mortality risk incurred by Type D is 3-fold, with this risk being independent of disease severity, such as left ventricular dysfunction, and mood states such as anxiety and depression, and despite appropriate medical treatment (Pedersen & Denollet, [@CR34]). The mechanisms relating Type D personality with adverse prognosis in cardiac patients are generally not thought to derive from worse disease severity (de Jonge et al., [@CR5]; Nicholson et al., [@CR32]). Rather, negative health-related behaviors, such as smoking and poor treatment adherence (Kirkcaldy et al., [@CR25]; Pedersen et al., [@CR35]; Schiffer et al., [@CR47]), and dynamic physiological processes such as elevated cortisol levels (Molloy et al., [@CR30]; Whitehead et al., [@CR54]) and pro-inflammatory cytokines (Denollet et al., [@CR12]) have been suggested as possible contributing factors. Importantly, recent findings have casted doubt on the utility of using extent of coronary atherosclerosis as a surrogate means for inferring associations between psychological risk factors and adverse cardiovascular outcomes in cross-sectional data (Rozanski et al., [@CR45]). In the present study, we included assessment of extent of coronary artery disease to rule out the possibility of reverse causation, whereby disease severity can contribute to greater psychological distress and, in turn, may confound the assessment of Type D personality traits. In clinical and epidemiological research, Type D can be assessed with the standardized 14-item Type D Scale (DS14) that measures negative affectivity and social inhibition (7 items for each domain) (Denollet, [@CR7]). The DS14 scale has been validated in Belgian (Denollet, [@CR7]), Chinese (Yu et al., [@CR57]), Danish (Pedersen & Denollet, [@CR33]; Spindler et al., [@CR50]), Dutch (Denollet, [@CR7]), German (Grande et al., [@CR20]), Italian (Gremigni & Sommaruga, [@CR21]) and Ukrainian (Pedersen et al., [@CR37]) cardiac patients and healthy controls. However, only a few studies have examined how the Type D construct fits within the five-factor model of personality, and no study to date has tested how the social inhibition factor relates to emotional control. Hence, the objectives of the current study were (a) to evaluate the psychometric properties of the DS14 in Icelandic cardiac patients with a specific focus on the construct validity of Type D, (b) to examine whether assessment of Type D personality is confounded by worse disease severity in these patients and (c) to explore the association between Type D and health-related risk markers. Method {#Sec2} ====== Participants {#Sec3} ------------ This study included two cardiac patient samples. The first sample (cardiac sample I) consisted of 1,291 patients hospitalized for coronary angiography and/or percutaneous coronary intervention at Landspitali-university hospital in Reykjavik (May 2007--June 2008), the only hospital in Iceland where such operations are performed. These patients were approached when hospitalized to the coronary care unit, upon arrival to the emergency ward or by mail if they were on the waiting list for a coronary catheterization. Patients were eligible for participation only if they (a) underwent a coronary angiography or percutaneous coronary intervention during their current hospitalization; and (b) spoke and read Icelandic fluently. Forty-four patients were excluded because they either did not complete the DS14 (*n* = 34) or did not undergo coronary angiography (*n* = 10). The remaining 1,247 patients (875 men and 372 women) had a mean age of 64.8 years (SD 10.8), with women being significantly older than men (*M* = 63.3 (SD 11.0) vs. *M* = 68.2 (SD 9.5), *t*~(1,245)~ = 7.57, *P* \< 0.001). This patient sample was included in the study to (a) estimate the factor structure of the DS14 scale, and (b) examine whether the assessment of Type D personality is confounded by the severity of underlying coronary artery disease. The second sample (cardiac sample II) consisted of 161 patients from the coronary care unit, and from the heart failure clinic of the Landspitali-university hospital (February--March 2006 and November 2006--April 2007). This sample was included in the study to examine more extensively the validity of the Type D personality construct in Iceland, and how it is related to health-related risk markers. To this end, these patients completed additional measurements that were not administered in the larger cardiac sample I. Four patients were excluded from analysis due to incomplete questionnaire data. The final sample included 157 participants (118 males and 39 females) with an average age of 61.7 years (SD 11.3), and again women tended to be older than men (*M* = 60.2 (SD 11.1) vs. *M* = 66.4 (SD 11.0), *t*~(150)~ = 3.03, *P* \< 0.01). Baseline characteristics for the two participant samples are presented in Table [1](#Tab1){ref-type="table"}. Patients in cardiac sample I were older on average compared to patients in cardiac sample II (*t*~(1,397)~ = 3.24; *P* ≤ 0.001), but gender distribution was similar in the two samples ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \chi_{(1,n\,=\,1,404)}^{2} $$\end{document}$ = 1.68, *P* = 0.20). The majority of patients in cardiac sample I had coronary artery disease (55%) or had experienced one or more heart attacks (23%), while patients with a history of one or more heart attacks (41%) and heart failure (24%) were more prominent in cardiac sample II.Table 1Baseline characteristicsCardiac sample I (*n* = 1,247)Cardiac sample II (*n* = 157)*Age (years)*Mean (SD)64.8 (10.8)61.7 (11.3)*Gender*Males70% (875)75% (118)*Heart disease*Heart failure2% (22)24% (38)Pacemaker and cardiac arrhythmia7% (89)11% (17)≥1 heart attacks23% (290)41% (64)Coronary artery disease55% (678)10% (16)Hypertension7% (92)11% (17)No disease6% (73)0 (0%)Unknown0.2% (3)3% (5)Data are presented as percentages (*n*) unless otherwise specified The study protocol was approved by the medical ethics committee of the National Bioethics Committee in Iceland. The study was conducted to conform to the ethical tenets developed by the World Medical Association, as espoused in the Declaration of Helsinki. All patients provided written informed consent. The DS14 scale {#Sec4} -------------- The DS14 is a 14-item questionnaire comprised of two seven-item subscales (Denollet, [@CR7]), measuring the tendency to (a) experience negative emotions (negative affectivity) and (b) inhibit self-expression in social interactions (social inhibition). The answering format is on a five-point Likert scale, ranging from 0 (*false*) to 4 (*true*), with total scores ranging from 0 to 28 for each subscale. Items include "I am often irritated" (negative affectivity) and "I am a closed kind of person" (social inhibition). The original Dutch DS14 was translated into Icelandic by four researchers. They received aid from two fluent Dutch speakers who independently translated the DS14 items from Dutch to Icelandic; a translation group examined the two independent translations, and one final version was constructed. Subsequently, the final Icelandic version was back-translated and compared to the original Dutch version to ensure accuracy. Participants were defined as having a Type D personality if they scored ≥10 on both negative affectivity and social inhibition. This cut-off value has been used in previous research (Denollet, [@CR7]; Emons et al., [@CR17]), and is derived from the median split on negative affectivity and social inhibition scores of participants in those studies. A recent study using item-response theory has shown the cut-off ≥10 on both subscales to be the best to distinguish between Type D and non-Type D individuals, as all items had the highest measurement accuracy around that cut-off (Emons et al., [@CR17]). Construct validity {#Sec5} ------------------ To evaluate the construct validity of the Icelandic DS14 scale, the NEO-five-factor inventory (NEO-FFI) (Costa & McCrae, [@CR3]), emotional control questionnaire (ECQ) (Roger & Najarian, [@CR41]; Roger & Nesshoever, [@CR42]), hospital anxiety depression scale (HADS) (Zigmond & Snaith, [@CR58]) and perceived stress scale (PSS) (Cohen et al., [@CR1]) were administered in Cardiac sample II. The NEO-FFI is a 60-item self-report scale which assesses five broad personality traits from the five-factor model of personality, that is neuroticism (e.g. anxiety, impulsiveness, vulnerability), extraversion (e.g. sociability, activity, positive emotions), openness (e.g. fantasy, feelings, artistic), agreeableness (e.g. trust, straightforwardness, altruism) and conscientiousness (e.g. achievement striving, dutifulness, self-discipline) (Costa & McCrae, [@CR3]). The scale contains 12 statements for each trait, and respondents answer on a five-point Likert scale (ranging from *strongly disagree* (0) to *strongly agree* (4)) how each statement refers to them. The psychometric properties of the Icelandic version of the NEO-FFI are acceptable and the test--retest reliability and internal consistency deemed sufficient (Jónsson, [@CR24]), with Cronbach's alpha ranging from 0.71 to 0.88 for the five traits (Svansdóttir, [@CR53]). The emotional control questionnaire or ECQ measures how easily people express and control their emotions (Roger & Najarian, [@CR41]; Roger & Nesshoever, [@CR42]). The scale includes 56 items which are divided into four factors (emotional inhibition, aggression control, benign control and rehearsal), but in this study a shorter 20-item version measuring rehearsal and emotional inhibition only was used (Roger et al., [@CR40]). Rehearsal refers to the tendency of individuals to ruminate over emotionally distressing events while emotional inhibition assesses to what extent people express their emotions. The response format for each item ranges from *strongly disagree* (1) to *strongly agree* (4). The Icelandic version of this scale has adequate psychometric properties with Cronbach's alpha reliability coefficients of α = 0.83 for rehearsal and α = 0.74 for emotional inhibition (Ingibergsdóttir, [@CR23]). The HADS is a 14-item questionnaire that measures symptoms of depression and anxiety in physically ill people (Zigmond & Snaith, [@CR58]). The questionnaire contains seven statements for each mood status. Participants rate on a four-point scale (0--3) how well each statement refers to them. Total scores range from 0 to 21 for each domain. The Icelandic version of the HADS identifies symptoms of depression and anxiety sufficiently well (Schaaber et al., [@CR46]), with reliability estimates ranging from α = 0.78--0.86 for anxiety and α = 0.65--0.85 for depression (Smari et al., [@CR49]). The PSS or perceived stress scale is a 14-item measure of self-appraised stress (Cohen et al., [@CR1]). Items include for instance "In the last month, how often have you felt that you were unable to control the important things in your life?" The response format is on a five-point Likert scale ranging from *never* (0) to *very often* (4), and total scores range from 0 to 56. The scale has good psychometric properties (Cohen et al., [@CR1]; Cohen & Williamson, [@CR2]) and correlates with social anxiety and depression symptoms (Cohen et al., [@CR1]). The Icelandic version of PSS has comparable psychometric properties to the original language version (Davíðsdóttir & Bachman, [@CR4]) with Cronbach's alpha = 0.89 in a university student sample (Svansdóttir, [@CR53]). Disease severity {#Sec6} ---------------- Disease severity, defined by how many coronary arteries were affected by coronary artery disease (i.e. normal arteries, 1, 2, or 3 arteries affected, and main stem narrowing) was derived from results of the coronary angiography in cardiac sample I. Angiography results were inconclusive for one person, which was excluded from this analysis. Information on disease classification, categorized as hypertension, coronary artery disease, previous heart attacks, pacemaker/arrhythmias and heart failure, was obtained from medical staff and/or retrieved from medical records. Information concerning disease classification was missing for three patients in cardiac sample I (0.2%) and five patients in cardiac sample II (3.2%). Health-related risk markers {#Sec7} --------------------------- Participants in cardiac sample II provided information by self-report regarding certain health-related risk markers. These included (a) smoking status (*never, ex*-*smoker, current smoker*); (b) amount of smoking per day (*0*--*10 cigarettes, 10*--*20 cigarettes, 20*--*30 cigarettes, and* \>*30 cigarettes a day*); (c) duration of smoking (*0*--*5* *years, 5*--*10* *years, 10*--*20* *years,* \>*20* *years*); (d) previous mental problems, i.e. "Have you experienced any significant mental problems in the past?" (*yes, no*); and (e) psychopharmacological medication use, i.e. "Have you used one or more of the following medications for more than two weeks in the past 12 months: sleeping pills, anxiety-reducing medications, antidepressants and sedatives?" (*no, sleeping pills, anxiety-reducing medication, antidepressants, sedatives*). Of note, due to a low incidence rate for each medication category, answers were recoded post-hoc to a binary variable containing the following distinction: *no, I have not used any of these medications; yes, I have used one or more of these medications*. Statistical analysis {#Sec8} -------------------- Principal axis factor analysis with direct oblimin rotation (delta = 0) was used to explore the factor structure of the DS14, using the scree plot and criterion of eigenvalues \> 1 to determine the number of factors to extract. A confirmatory factor analysis of the scale was performed to confirm the two-factor structure of the scale, using structural equation modeling (SEM) and the maximum likelihood method in AMOS 17 (Analysis of Moment Structures, Chicago, IL, USA). In the construction of the model, the theoretical foundation of the scale was taken into account. As the negative affectivity and social inhibition subscales each cover three different facets of negative affectivity and social inhibition, respectively, error covariance was added to items representing each facet, i.e. for items measuring the negative affectivity facets dysphoria (items 4, 7 and 13), worry (items 2 and 12) and irritability (items 5 and 9), and for items measuring the social inhibition facets discomfort in social interactions (6, 8 and 14), reticence (10 and 11) and social poise (items 1 and 3). Goodness of fit indexes used in the analysis included the Chi-square, the Comparative fit index (CIF) and the Root mean square error of approximation (RMSEA). For Chi-square, a value ≥ 0.05 indicates good fit (agreement with the null hypotheses that residuals are minimal and the data fit the model well). The Chi-square is influenced by sample size, which can lead to inflated Chi-square values and thus statistical significance, indicating bad fit (Schumacker & Lomax, [@CR48]). For the CFI, values close to 1 indicate a very good fit and values above 0.90 or close to 0.95 good fit. The RMSEA index should be ≤0.05 to indicate good fit, but levels ≤0.08 are considered to indicate adequate fit. Internal consistency of the scale was assessed with Mean inter-item correlation and Cronbach's alpha. Validity of the DS14 was estimated by exploring the Pearson's correlation between the negative affectivity and social inhibition subscales and similar constructs, i.e. neuroticism and extraversion, emotional inhibition and rehearsal, anxiety and depression and perceived stress. A factor analysis of scale scores on the DS14 scale, NEO-FFI, ECQ, HADS and PSS was executed to verify that (a) negative affectivity, neuroticism and rehearsal, and (b) social inhibition, introversion and emotional inhibition measure related constructs, and to test how anxiety, depression and stress would relate to the negative affectivity and social inhibition factors. Differences in disease classification by Type D personality were assessed with Kendall's Tau-*c* calculations, but patients with arrhythmias and pacemakers were excluded from the analysis due to the different nature of their disease. The Kendall's Tau-*c* was also employed to estimate differences in disease severity by Type D personality in cardiac sample I, in both the entire sample and among patients who had established coronary artery disease. Finally, Type D and non-Type D patients in cardiac sample II were compared on smoking behavior, prevalence of previous mental problems, and medication use with Chi-square tests for nominal variables and Tau-*c* for ordinal variables. Association strength was estimated with Cohen's D calculations for quantitative variables and odds ratios for categorical variables. The SPSS 17 statistical software for Windows was used for all main analysis (Statistical Package for Social Sciences, Chicago, IL, USA). Results {#Sec9} ======= Factor structure of the DS14 {#Sec10} ---------------------------- A principal axis factor analysis (direct oblimin rotation, delta = 0) in a combined sample of cardiac patients (*n* = 1,404) indicated a two-factor solution, which explained 46% of variance in the patient sample. These two factors clearly reflected the negative affectivity and social inhibition subscales, with satisfactory factor loadings (ranging from 0.47 to 0.75) and good internal consistency (negative affectivity: Cronbach's alpha = 0.85 and Mean inter-item correlation = 0.45; social inhibition: Cronbach's alpha = 0.84, Mean inter-item correlation = 0.43) (Table [2](#Tab2){ref-type="table"}).Table 2Factor analysis and reliability of the DS14 scale in a combined cardiac sample (*n* = 1,404)FactorsIII*Negative affectivity items*2. I often make a fuss about unimportant things**0.61**0.104. I often feel unhappy**0.74**0.075. I am often irritated**0.73**0.047. I take a gloomy view of things**0.63**−0.149. I am often in a bad mood**0.58**−0.1312. I often worry about something**0.60**0.0213. I am often down in the dumps**0.73**−0.12Cronbach's alpha**0.85**Mean inter-item correlation**0.45***Social inhibition items*1. I make contact easily when I meet people0.14**0.72**3. I often talk to strangers0.16**0.61**6. I often feel inhibited in social interactions0.29−**0.50**8. I find it hard to start a conversation0.10−**0.75**10. I am a closed kind of person0.12−**0.66**11. I would rather keep other people at a distance0.16−**0.47**14. When socializing I don't find the right things to talk about0.15−**0.69**Cronbach's alpha**0.84**Mean inter-item correlation**0.43**The highest loadings on the corresponding factor, Cronbach's alpha and mean inter-item correlations are presented in bold A confirmatory factor analysis of the two-factor structure of the Icelandic DS14 in the same sample indicated a good to adequate model fit for the unconstrained model (χ^2^ = 435.63, *P* ≤ 0.001; CFI = 0.953 and RMSEA = 0.063, 90% CI 0.058--0.069). Standardized regressional weights of items to factor ranged from 0.52 to 0.79 for negative affectivity and 0.44--0.80 for social inhibition (Fig. [1](#Fig1){ref-type="fig"}).Fig. 1Standardized regression weights for the 2-factor model of the DS14, representing negative affectivity (*NA*) and social inhibition (SI) Construct validity {#Sec11} ------------------ The convergent and construct validity of the Icelandic DS14 scale was evaluated by examining correlations of negative affectivity and social inhibition with similar construct measurements in cardiac sample II. The negative affectivity subscale had a high positive correlation with neuroticism (*r* = 0.80) and rehearsal (*r* = 0.58), while social inhibition was negatively correlated with extraversion (*r* = −0.65) and positively with emotional inhibition (*r* = 0.50), which further supports the divergent validity of the Type D factors and their individual attributes. Negative affectivity had a high correlation with anxiety, depression and stress scores, indicating that it clearly measures increased negative affect. An axis factor analysis (direct oblimin rotation, delta = 0) of scale scores confirmed that the negative affectivity and social inhibition subscales were differentially related to the five-factor model of personality; negative affectivity (loading = 0.79), neuroticism (0.78) and rehearsal (0.64) loaded on a single negative affectivity/neuroticism factor. Social inhibition (−0.95), extraversion (0.57) and emotional inhibition (−0.44) loaded together on a separate inhibition factor. Neither DS14 subscale was related to agreeableness, conscientiousness or openness of the five-factor model of personality. Anxiety (−0.50), depression (−0.73) and stress (−0.60) loaded together on a single factor termed "psychological well-being", but anxiety also had a considerable loading on the negative affectivity/neuroticism factor (0.49) (Table [3](#Tab3){ref-type="table"}).Table 3Correlations and results of a factor analysis of scale scores for the DS14, NEO-FFI, ECQ and HADS subscales and PSS scaleCorrelationNegative affectivitySocial inhibitionPattern matrixCardiac sample II (*n* = 157)IIIIIIIVVNegative affectivity--0.47\***0.79**−0.14−0.09−0.070.04Social inhibition----0.11**−0.95**0.11−0.050.05Neuroticism0.80\*0.47\***0.78**−0.12−0.06−0.14−0.13Extraversion−0.48\*−0.65\*−0.09**0.57**0.12−0.020.17Agreeableness−0.33\*−0.21\*−0.28−0.030.07−0.350.17Conscientiousness−0.20\*−0.25\*−0.010.060.050.12**0.72**Openness−0.02−0.07−0.020.060.01**−0.45**−0.09Rehearsal0.58\*0.35\***0.64**−0.050.010.23−0.02Emotional inhibition0.25\*0.50\*−0.08**−0.44**−0.190.25−0.10Anxiety0.67\*0.26\*0.490.01**−0.50**−0.250.20Depression0.55\*0.35\*0.04−0.15**−0.73**−0.11−0.09Perceived stress0.38\*0.18\*0.040.05**−0.60**0.23−0.09The highest loadings on the corresponding factor are presented in bold*NEO*-*FFI* NEO-five-factor inventory, *ECQ* Emotional control questionnaire, *HADS* Hospital anxiety and depression scale, *PSS* Perceived stress scale\* *P* \< 0.001 Prevalence of Type D personality {#Sec12} -------------------------------- Average scores on negative affectivity and social inhibition were equivalent in the two patient samples (negative affectivity: *M* = 8.6 (SD 5.6) vs. *M* = 8.8 (SD 5.9), *t*~(1,402)~ = 0.46, *P* = 0.65; social inhibition: *M* = 9.3 (SD 5.8) vs. *M* = 9.3 (SD 6.1), *t*~(1,402)~ = 0.11, *P* = 0.91; for cardiac sample I and II, respectively). Using the cut-off ≥ 10 for both subscales (Denollet, [@CR7]; Emons et al., [@CR17]), 26% of patients in cardiac sample I and 29% of patients in cardiac sample II, were classified as Type D individuals. Confounding effect of disease severity {#Sec13} -------------------------------------- Assessment of Type D personality was not confounded by severity of underlying coronary artery disease in cardiac sample I, as estimated by number of arteries affected by coronary artery disease from the coronary angiography results (Tau-*c* = 0.010, *n* = 1,237, *P* = 0.72; Fig. [2](#Fig2){ref-type="fig"}). About 1/3 of both non-Type D and Type D patients had normal arteries or atheroma with no significant occlusions, and with those individuals excluded from the analysis, Type D personality was still not associated with worse disease severity (Tau-*c* = −0.001, *n* = 838, *P* = 0.98).Fig. 2Coronary artery disease severity, stratified by Type D personality Assessment of Type D personality was also not related to disease classification in cardiac sample I (Tau-*c* = −0.02, *n* = 1,155, *P* = 0.45) nor cardiac sample II (Tau-*c* = −0.15, *n* = 135, *P* = 0.068). In both cases, disease classification was categorized as: no disease, hypertension, coronary artery disease, ≥1 heart attacks and heart failure. Association with health-related risk markers {#Sec14} -------------------------------------------- As a final step, we explored the relationship of Type D personality with psychopharmacological medication use, previous mental problems and smoking in cardiac sample II. Type D patients reported more psychopharmacological medication use (Fig. [3](#Fig3){ref-type="fig"}). When asked about use of sleeping pills, anxiety-reducing medication, antidepressants and sedatives, half of the cardiac patients with a Type D personality (51%) reported having used one or more of these medications compared to 29% of their non-Type D counterparts ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \chi_{(1,n\,=\,154)}^{2} $$\end{document}$ = 6.79, *P* = 0.009; OR 2.59, 95% CI 1.25--5.34, *P* = 0.010). Prevalence of previous mental problems did however not differ between Type D (19%) and non-Type D (14%) patients ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \chi_{(1,n = 149)}^{2} $$\end{document}$ = 0.584, *P* = 0.45). Type D patients were significantly more likely to smoke as compared with non-Type D patients (Fig. [3](#Fig3){ref-type="fig"}); i.e., 22% versus 6% ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \chi_{(1,n = 156)}^{2} $$\end{document}$ = 8.35, *P* = 0.004; OR 4.25, 95% CI 1.50--12.00, *P* = 0.006). In patients with a history of smoking, no differences were found between Type Ds and non-Type Ds regarding how many cigarettes they smoked per day (Tau-*c* = 0.11, *n* = 115, *P* = 0.26). However, a trend towards a longer history of smoking was noted in Type Ds (Tau-*c* = 0.15, *n* = 120; *P* = 0.056), but 76% of Type D smokers (former or current) reported having smoked for 20 years or more compared to 59% of non-Type D smokers.Fig. 3Prevalence of psychopharmacological medication use and smoking, stratified by Type D personality Discussion {#Sec15} ========== The objectives of the current study were to evaluate the psychometric properties and construct validity of the Icelandic DS14 scale, to test whether Type D assessment is confounded by disease severity in Icelandic angiography patients, and to explore the relationship between Type D and health-related risk markers. The findings supported the two-factor structure of the Icelandic DS14, and its validity and reliability in this sample of Icelandic heart patients. Principal axis factor analysis revealed internally consistent negative affectivity and social inhibition factors, and a confirmatory factor analysis confirmed the two-factor structure of the original scale (Denollet, [@CR7]) in a large sample of Icelandic cardiac patients. The current results supported the convergent and divergent validity of the Type D construct in the Icelandic setting. An exploratory factor analysis of scale scores showed that negative affectivity, neuroticism and rehearsal loaded on the same factor, while social inhibition, extraversion, and emotional inhibition loaded together on another factor, supporting the construct validity of the two factors of the DS14 (Denollet, [@CR7]; Fruyt & Denollet, [@CR18]) whilst also strengthening its cross-cultural validity. Furthermore, negative affectivity correlated strongly with anxiety, depression and moderately with perceived stress, confirming the presence of increased negative mood states within the negative affectivity trait. In addition, social inhibition was clearly associated with emotional inhibition as measured by the emotional control scale. In a recent study, Grande et al. ([@CR19]) advocated more testing of the construct validity of the social inhibition dimension, especially since it is the combination of social inhibition with negative affectivity that seems to make Type D personality a stronger predictor of adverse cardiac events compared with other single-dimensional negative affect factors, such as depression. In the context of Type D personality, the inhibition of emotions in social interaction is believed to play a key part in the association with adverse cardiac prognosis, by modulating the effect negative emotions have on cardiac prognosis (Denollet et al., [@CR10]). Others have also linked social inhibition with social avoidance (Yu et al., [@CR57]), lack of social boldness (Grande et al., [@CR19]) and suppressed anger (Denollet et al., [@CR9]). The prevalence of Type D personality of twenty-six and twenty-nine percent in the cardiac samples was comparable to that found in European and Chinese samples (Denollet, [@CR7]; Grande et al., [@CR20]; Gremigni & Sommaruga, [@CR21]; Pedersen & Denollet, [@CR33]; Pedersen et al., [@CR37]; Spindler et al., [@CR50]; Yu et al., [@CR57]). Assessment of Type D personality was not confounded by disease severity, as estimated by the number of coronary arteries affected with coronary artery disease and/or presence of significant narrowing at the main stem. This finding is in accordance with previous results in coronary artery disease and congestive heart failure patients, where no association has been found between Type D personality and indicators of disease severity, such as multivessel disease (Martens et al., [@CR28]), left ventricular ejection fraction (Denollet & Brutsaert, [@CR8]; de Jonge et al. [@CR5]) and biomedical markers (i.e. brain natriuretic peptide) (Pelle et al., [@CR39]). Similarly, Type D personality was not related to disease classification in either of the cardiac samples. The majority of former findings have generally also revealed that Type D personality is stable across time, and does not seem to be affected by changes in mood status or severity of cardiac disease (Martens et al., [@CR28]). The lack of association between Type D personality and extent of coronary artery disease does not necessarily diminish the status of Type D personality as a predictor for adverse cardiac prognosis. Conversely, these findings may merely indicate that the mechanisms relating Type D personality with adverse prognosis do not stem from worse disease severity, but through other pathways. Furthermore, if disease severity were in fact the pathway through which Type D personality affects cardiac prognosis, then the association between Type D and prognosis should diminish in strength or disappear altogether when multivariate adjustments for disease severity markers are conducted. This has however not been the case in previous studies, as is evident in the recent review by Denollet et al. ([@CR13]). Mediating mechanisms linking Type D with adverse cardiac prognosis reside more likely in behavioral and physiological processes. Potential behavioral factors include unhealthy lifestyle behaviors (Williams et al., [@CR56]), more smoking (Pedersen et al., [@CR35]), poor treatment adherence (Rozanski et al., [@CR43]; Williams et al., [@CR55]) and inadequate consultation behavior (Schiffer et al., [@CR47]), while physiological and biological processes may include elevated cortisol (Molloy et al., [@CR30]; Whitehead et al., [@CR54]), pro-inflammatory cytokines (Denollet et al., [@CR12]) and cardiovascular stress reactivity (Denollet et al., [@CR13]) to name a few. Type D patients may thus be less likely to follow their doctors recommendations regarding medications or changing unhealthy lifestyle behaviors, and perhaps less efficient in presenting their symptoms to their doctor, due to their high social inhibition. Such factors could possibly explain why these patients develop or experience a more adverse prognosis compared to their non-Type D counterparts. A recent study by Rozanski et al. ([@CR45]) has also reported that psychological risk factors (depression, hostility, social support, perceived stress, job strain, and optimism) were not associated with the extent of coronary atherosclerosis. This further supports the lack of association between Type D and extent of coronary artery disease in the current study, as the Type D construct generally summarizes such psychological risk factors in the general negative emotional distress it encompasses (Suls & Bunde, [@CR52]). Finally, even as some researchers have disputed that the relation of psychological factors with cardiovascular prognosis is confounded by worse somatic health, findings from a recent study have indicated that the Type D personality construct is less confounded by somatic health compared with depression (de Jonge et al. [@CR5]). Type D personality had strong ties to health-related risk markers in cardiac patients. Although no association was found between Type D personality and prevalence of reported previous mental problems in the current study, psychopharmacological medication use was higher among Type D patients compared to their non-Type D counterparts, and a high correlation emerged between negative affectivity and anxiety and depression. Previously, researchers have also found that post-myocardial infarction patients with a Type D personality were significantly more likely to use benzodiazepines as compared to non-Type D patients (Denollet et al., [@CR14]). The lack of association with former mental problems seems contradictory with the high correlation noted between negative affectivity and anxiety and depression. The assessment of previous mental problems may not adequately portray the number of previously diagnosed mental problems, due to the simplistic one question format assessment. We also found a relationship between Type D personality and smoking among cardiac patients. Incidence of current smoking was higher in the Type D patient group, and there were some indications that Type D smokers had a longer history of smoking compared to non-Type D smokers. Previously, it has been reported that cardiac patients with a Type D personality may be more likely to smoke (Pedersen et al., [@CR36]), and that Type D individuals are less likely to engage in healthy lifestyle behaviors (Williams et al., [@CR56]). These findings suggest that cardiac patients with Type D personality may struggle more with the lifestyle changes recommended by doctors to decrease likelihood of further cardiac events. In addition, previous results have indicated that heart failure patients with Type D personality are more likely to show inadequate consultation behavior compared to non-Type D patients (Pelle et al., [@CR38]; Schiffer et al., [@CR47]), which implies that self-management and medical adherence in these patients may be impaired as well. Nevertheless, research results have indicated that the adverse effect of Type D on cardiac prognosis (Pedersen et al., [@CR36]) and poor health status (Pedersen et al., [@CR35]) remains significant despite statistical adjustment for smoking and other mechanisms that may mediate the relationship between Type D and health outcomes. More research needs to be conducted to clarify which mediating mechanisms are behind Type D's association with adverse prognosis in cardiac patients, and to determine whether health-behavior and/or poor medical adherence play a significant role. Certain limitations restrict the interpretation of the present findings. First of all, the participant samples were not randomly selected. Yet, cardiac sample I included consecutive patients nationwide in Iceland, which diminished greatly the risk of selection bias in that sample. Another limitation is the self-report of psychopharmacological medication use, previous mental health problems and smoking, and the unavailability of these measures from cardiac sample I. Overall, the results of the present study indicated that the Icelandic DS14 is a psychometrically sound assessment tool that can be readily applied in epidemiological and clinical research. The Type D personality construct was prevalent in Icelandic cardiac patients, not confounded by disease severity, and related to certain health-related risk markers in this clinical population. The present research was supported by Rannís, The Icelandic Centre for Research (Reykjavík, Iceland) by a grant to Dr. Hrobjartur D. Karlsson, the Landspitali-University Hospital Research found, Iceland, with a grant to the project, and by the Netherlands Organisation for Scientific Research (The Hague, The Netherlands) with a VICI grant (453-04-004) to Dr. Johan Denollet and a VENI grant (451-05-001) to Dr. Susanne S. Pedersen. Conflict of interest {#d28e1901} ==================== There is no conflict of interest related to this study. Open Access {#d28e1906} =========== This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.

1. Introduction =============== Airway stenosis is a potentially life-threatening condition, which may be caused by tumor invasion or benign diseases, such as, endotracheal intubation injury, endobronchial tuberculosis, and thoracic trauma. Most of these patients may be too sick to undergo definitive open surgery or may not be willing to undergo surgery under general anesthesia.^\[[@R1]\]^ Conservative treatment simply assists in alleviation of airway symptom, but show no curative effect on airway stenosis. Since first adoption of metal stents to treat bronchial obstruction, the airway stent has become an effective method for airway stenosis^\[[@R2]--[@R5]\]^ and has been widely used in clinical practice. Complications of airway stent include stent restenosis caused by tumor or granulation tissue, stent fracture, and stent migration.^\[[@R6]\]^ Stent removal may become necessary if these complications result in airway obstruction or injury. Endoscopic removal of airway stent has been reported in small series.^\[[@R7],[@R8]\]^ However, few studies have described the techniques and complications of fluoroscopic removal of airway stent. The safety and efficacy of fluoroscopic stent removal and the optimal safe duration of stent implantation have not been determined. We report clinical results in fluoroscopic removal of airway stents for airway stenosis over a 6-year period. 2. Materials and methods ======================== 2.1. Patients ------------- This study was approved by the Ethics Committee and Medical Records Management Section of our university. The written informed consents were obtained from all patients or their parents if the patients are less than 18 years old. We conducted a retrospective analysis of 67 patients who underwent fluoroscopic stent removal due to patient\'s refusal of surgery and general anesthesia from March 2011 to April 2017. All the patients were enrolled after chest computed tomography (CT) examination. The inclusion criteria included all patients with benign airway stenosis, and dyspnea had been significantly alleviated and needed planned removal. Following doctor\'s advice, they admitted to hospital for stents removal to avoid long-term complications, with no obvious symptom or sign (planned removal). Patients show complications after stenting such as intolerance of stenting, stent migration, excessive tissue hyperplasia, inadequate expansion and deformation, and strut fracture. New airway stents are needed to replace the original stents, such as bare stents placed under emergency situation will be removed and the covered airway stents will be replaced when the patient\'s condition is stable. The exclusion criteria included patients with advanced malignant tumors with a survival time of less than 3 months and no obvious stent complications; patients with severe visceral dysfunction who unable to withstand the stent removal procedure; patients refused to remove the stent (Fig.1). {#F1} 2.2. Technical details of fluoroscopic removal ---------------------------------------------- All patients underwent chest CT scans with/without bronchoscopy before stent removal (Fig. [2](#F2){ref-type="fig"}). Under conscious situation, balloon dilatation of airway stent was performed immediately after stenting due to insufficient expansion (n = 11) and before stent implantation or removal due to severe airway stenosis (n = 24). All self-expandable metal airway stents were designed and manufactured by Micro-Tech Co Ltd (Nanjing, China). Airway metal stent was woven with a temperature-memory nickel--titanium alloy wire, which was visible on fluoroscopy. In an emergency, we placed 6 bare stents to immediately relieve severe dyspnea and avoid asphyxia. Seventy covered stents were used in nonemergency situations. All airway stents were removed transorally under fluoroscopy and local anesthesia via lidocaine. Bronchoscopy was not performed during the removal, which might use after the operation if removal failed. The patient took the supine position and the head was faced toward the operator to prevent asphyxiation. A 5F vertebral artery catheter (Cook Corporation, Bloomington, IN) was advanced along a 0.035 inch stiff hydrophilic guide wire (Cook Corporation), then 2% lidocaine was injected through vertebral artery catheter for topical anesthesia. A 10-14F long sheath was inserted via the stiff hydrophilic guide wire, and an extractive hook was then inserted. The tip of hook was placed next to the airway stent to withdraw the airway stent (Fig. [3](#F3){ref-type="fig"}). The sputum aspiration tube was advanced for maintaining airway patency and sputum drainage. {#F2} {#F3} 2.3. Complications of stent removal and follow up ------------------------------------------------- All complications were scored and analyzed. Patients were followed up once a month within 3 months after stent placement, and every 2 months thereafter. If symptoms flared as needed, patients were asked to admit to hospital at any time. All patients underwent chest CT scans after stent placement, and bronchoscopy was used for some patients if necessary. Telephone follow up was used for patients did not admit to hospital. 2.4. Statistical analysis ------------------------- Data were expressed as mean ± standard error and analyzed using Student *t* test and analysis of variance. Qualitative data were expressed as the percentage, and analyzed by Fisher exact test. Patency rate was compared using the Log-rank (Mantel-Cox) Test (GraphPad Software, Inc San Diego, USA). Statistical significance was considered when *P* \< .05. 3. Results ========== 3.1. The patients' characteristics ---------------------------------- The 67 patients, 39 male and 28 female, ranged in age from 12 to 85 years, with a mean age of 51.1 ± 2.2 years. Seventy-six airway stents were implanted in 67 consecutive patients over a 6-year period, of which, 9 patients underwent second stent implantation after removal. The mean procedure time for stent implantation and removal were 26.8 ± 1.9 minutes and 23.3 ± 2.0 minutes, respectively. The mean fluoroscopic time for stent implantation and removal were 18.6 ± 1.1 minutes and 17.3 ± 1.2 minutes, respectively. Types of individual airway stents were shown in Table [1](#T1){ref-type="table"}. Causes of airway stenosis included tracheotomy in 16 cases, trachea cannula in 16 cases, tumor invasion in 15 cases, external pressure airway stenosis in 4 cases, inflammatory stenosis in 6 cases (4 tuberculosis and 2 pulmonary infection), airway stent restenosis in 5 cases, airway repair and reconstruction in 3 cases (2 trachea trauma and 1 airway hemangioma), and relapsing polychondritis in 2 cases. ###### The patients' characteristics.  Sixteen Y-shaped airway stents were placed in the trachea and bilateral main bronchi. A total of 55 tubular stents were placed in the middle and upper trachea (n = 47), the middle trachea (n = 4), and the middle and lower trachea (n = 4). Two L-shaped airway stents were placed in the trachea and left main bronchus. Their stents were placed in trachea-right main bronchus, right main bronchus and left main bronchus. Before stent placement, 66 patients had dyspnea in varying degrees, including 16 patients with cough and expectoration, and 2 patients with hemoptysis. Dyspnea classification was 3.6 ± 0.7 before stent placement according to the American Thoracic Association Dyspnea Classification Standard, which decreased significantly to 0.6 ± 1.4 after stent placement and dyspnea symptoms were significantly alleviated. 3.2. Indications for stent removal ---------------------------------- The indications for stent removal were shown in Table [2](#T2){ref-type="table"}, which included planned removal (n = 40), excessive granulation tissue (n = 15), intolerance of stenting (n = 6), inadequate expansion and deformation (n = 5), stent migration (n = 5), replacement of bare stent (n = 4), and strut fracture (n = 1). Six bare stents were emergency implanted for severe airway stenosis, of which, 3 stents were removed and replaced by covered stents, with a mean duration of 8.7 ± 3.2 days, and 1 stent was replaced by tracheal T tube. The remained 2 patients refused to replace the bare stents with covered stents, and the bare stents were removed due to severe stenosis after more than 3 months. Stent migration, intolerance of stenting, and inadequate expansion and deformation were the main indications for early stent removal. Excessive granulation tissue was the most common indication for later stent removal, with a mean duration of 89.9 ± 15.0 days. One case underwent stent removal due to strut fracture 105 days after stenting. ###### Indications for airway stent removal.  3.3. Complications of stent removal ----------------------------------- As shown in Table [3](#T3){ref-type="table"}, 74 of 76 airway stents were successfully removed from 67 consecutive patients, 5 stents showed strut fracture, only 2 stent showed retained struts after removal, for a technical success rate of 97.4% (74/76). Before stent removal, 29 patients had dyspnea in varying degrees, including 6 patients with cough and expectoration. Dyspnea classification was 1.4 ± 1.7 before stent removal, which decreased significantly to 0.8 ± 0.9 after stent removal and dyspnea symptoms were significantly alleviated. ###### Complications of stent removal.  Two patients died of complications (1 hemorrhage and 1 respiratory failure), resulting in a clinical success rate of 94.7% (72/76) and 30-day mortality of 2.6%. The 3 retained stent pieces were successfully removed by endoscope; the remaining 71 stents were removed in 1 piece. There were 17 complications of stent removal: death from massive bleeding (n = 1), restenosis requires stenting (n = 9), strut fracture and retained stent pieces (n = 5), and dyspnea requires mechanical ventilation (n = 2). The most common complications were restenosis caused by excessive granulation tissue and strut fracture, after a mean duration of 93.3 ± 14.5 days (Fig. [4](#F4){ref-type="fig"} and Table [4](#T4){ref-type="table"}). A 52-year male patient developed airway stenosis and esophageal airway fistula after resection of esophageal carcinoma, a Y type airway stent was implanted but had to remove immediately due to intolerance of stent. This patient died of spontaneous massive hemoptysis might cause by esophageal airway fistula 3 days after stent removal. In most of patients, mild hemoptysis was found, without need for medication. During stent removal, 4 patients showed dyspnea with decreased pulse oxygen, and 2 cases needed mechanical ventilation. The symptoms were relieved in 1 patient and then endotracheal tube was removed within 4 hours. A 43-year female patient with relapsing polychondritis showed bad respiratory condition before stent placement, the opening of the left upper lobe bronchus is partially blocked by the left branch of the second Y type stent. She died of respiratory failure 2 days after stent implant. {#F4} ###### Second stenting after removal of airway stent.  3.4. Second stenting after removal and follow up ------------------------------------------------ As shown in Table [4](#T4){ref-type="table"}, 20 airway stents, 15 straight airway stents, and 5 Y type airway stents, were implanted again after removal for replacement of bare stents (n = 4), or due to restenosis (n = 9), migration (n = 3), inadequate expansion (n = 3), and intolerance (n = 2). Chest CT and bronchoscopy were performed during post-removal admission to prevent long-term compilation. During follow up, 1 patient died of respiratory failure 1 month after removal, and 2 patients died of tumor progression 42 days and 6 months after removal. Two patients underwent tracheotomy and tracheal T tube implantation 1 to 2 years after stent removal. Survival rate follow up. The survival rates were 83.8%, 82.1%, and 82.1% for 0.5, 3, and 6 years (Fig. [5](#F5){ref-type="fig"}). {#F5} 4. Discussion ============= The airway stent has become an effective method for the treatment of airway stenosis,^\[[@R2]--[@R5]\]^ and has been widely used in clinical practice. However, long term implantation of airway stents may cause several complications, such as, strut fracture, stent migration, and excessive granulation hyperplasia. Stent removal is necessary to avoid these complications.^\[[@R9]\]^ At present, endoscopic removal of airway stent has been reported in small series,^\[[@R7],[@R8]\]^ which often performed under general anesthesia^\[[@R10]--[@R14]\]^ or local anesthesia in a few reports.^\[[@R15],[@R16]\]^ Fluoroscopic removal of airway stent is rarely reported for airway stenosis. In this study, 74 of 76 airway stents were successfully removed, with a technical success rate of 97.4%. Compared with stent removal with rigid alligator forceps, in which 73% of stents were removed piecemeal.^\[[@R10]\]^ In our study, only 2 stent showed retained struts after removal and 3 retained stent pieces were successfully removed by endoscope, the remaining 71 stents were removed in 1 piece. Our data indicate that fluoroscopic removal of airway stents is effective and feasible for airway stenosis. The details provided in previous reports varied widely and almost with no detailed discussions in complications. Our study is the few study to report techniques and complications of fluoroscopic removal of airway stents. Although the process of fluoroscopic removal is quite short in our study (less than 30 minutes), there are also risks during removal. These serious complications include bleeding, tissue laceration, airway obstruction, and even death.^\[[@R10],[@R17],[@R18]\]^ In our study, there were 2 deaths, 1 patient of airway stenosis and esophageal airway fistula died of massive hemoptysis 3 days after stent removal; the other patient of relapsing polychondritis died of respiratory failure 2 days after stent implant. The 2 patients might die of etiology rather than stenting procedure. In addition, the total incidence of complications was 23.0% (17/74), similar as that previously reported by using of bronchoscope.^\[[@R10],[@R11],[@R16],[@R17]\]^ Excessive granulation tissue may cause difficulty or failure of removal.^\[[@R15]\]^ As indwelling time increase, the incidence of granulation tissue hyperplasia and stent fracture increase.^\[[@R19]--[@R21]\]^ The endothelialization appear 2 months after stenting,^\[[@R18]\]^ stents are difficult to be removed after 10 months.^\[[@R10]\]^ So, 6 months^\[[@R22]\]^ or earlier^\[[@R23]\]^ is regarded as the optimal indwelling time for removal. However, the most suitable indwelling time for airway stenosis is still inconclusive. In our study, stents are recommended for removal within 3 months to avoid excessive granulation tissue. Silicon stents are more suitable for patients with benign disorders caused airway obstruction, especially young and little patients. Patients received stent placement of metallic stents rather than silicon stent to relieve dyspnea caused by severe airway stenosis due to their refusal of surgery and general anesthesia. In this study, covered stents were used for these patients to avoid excessive tissue growth into stent, and for the convenience of removing the stent. Radiologic placement and removal of metallic stents are suitable for old patients with an age more than 70; patient with severe co-morbidities; or patient\'s refusal of surgery and general anesthesia. Securing the airway is a major step and first step and priority in any interventional bronchology procedure. The placement or removal of the stent under general anesthesia appears to be safer than under local anesthesia in patients with airway stenosis. However, according to our clinical experience, patients with severe airway stenosis are at high risk for endotracheal intubation and anesthesia. In our center, emergency endotracheal intubation can be performed under fluoroscopy guidance and local anesthesia for patients with acute dyspnea or asphyxia. Besides, vocal cords are the narrowest part in airway and a stent can injure or even completely block the airway at vocal cords during stent removal procedure. According to our experience, the tip of hook was placed next to the airway stent and captured the stent firmly throughout the whole procedure of removal to avoid the drop of stent and life threatening complication of airway block. The stent should be pulling out slowly and carefully to minimize the injury to vocal cords and tracheal. There are several limitations of the study. The pathological changes and optimal removal duration of stent after placement need confirm in further animal study. In addition, surgical placement and removal of the metallic stents under general anesthesia with rigid bronchoscope remain gold standard and airway stents removal often requires a variety of technologies,^\[[@R24]\]^ we removed stent solely under radiology guidance. 5. Conclusions ============== Fluoroscopic removal of airway stent is technically feasible and effective. Stents are recommended for removal within 3 months for treating airway stenosis to avoid complications such as excessive granulation tissue formation in stent. Author contributions ==================== **Conceptualization:** Yonghua Bi, Gang Wu, Xinwei Han, Jianzhuang Ren. **Data curation:** Yonghua Bi, Zepeng Yu. **Formal analysis:** Yonghua Bi, Gang Wu, Zepeng Yu. **Investigation:** Yonghua Bi, Gang Wu, Zepeng Yu. **Methodology:** Gang Wu. **Project administration:** Gang Wu, Xinwei Han, Jianzhuang Ren. **Software:** Zepeng Yu. **Supervision:** Gang Wu, Xinwei Han, Jianzhuang Ren. **Validation:** Xinwei Han. **Writing -- original draft:** Yonghua Bi, Gang Wu, Zepeng Yu. **Writing -- review and editing:** Xinwei Han, Jianzhuang Ren. Abbreviation: CT = computed tomography. How to cite this article: Bi Y, Wu G, Yu Z, Han X, Ren J. Fluoroscopic removal of self-expandable metallic airway stent in patients with airway stenosis. *Medicine*. 2020;99:1(e18627). XH and JR contributed equally to this work and share co-corresponding authors. The authors have no conflicts of interest to disclose.

All relevant data are within the paper and its Supporting Information files. Introduction {#sec001} ============ During meiosis, the accurate segregation of chromosomes relies on the formation of crossovers (COs) and sister chromatid cohesion \[[@pbio.1002412.ref001]\]. COs are produced by homologous recombination (HR)-mediated repair of programmed DNA double strand breaks (DSBs). In *Caenorhabditis elegans*, the holocentric chromosome pairs usually undergo a single off-center CO, resulting in a cruciform and asymmetric bivalent \[[@pbio.1002412.ref002],[@pbio.1002412.ref003]\]. This is important to define the organization of kinetochore proteins, motor proteins, protein kinases, and the domains of cohesion release, which is necessary for the accurate segregation of homologs during anaphase I \[[@pbio.1002412.ref004]\]. During prophase I, several mechanisms guarantee the formation of COs as well as the repair of the additional recombination intermediates as non-crossovers (NCOs). First, to ensure CO formation, programmed DSBs are induced in excess compared to the actual number of COs, e.g., \[[@pbio.1002412.ref005]--[@pbio.1002412.ref007]\]. Second, a subset of the early recombination intermediates are stabilized and protected by pro-CO factors \[[@pbio.1002412.ref008]--[@pbio.1002412.ref010]\]. CO maturation is supported by the MutSγ complex (MSH-4/5), sumo/ubiquitin ligases ZHP-3/RNF212, and HEI10 and the cyclin-related protein COSA-1/CNTD1 \[[@pbio.1002412.ref011]--[@pbio.1002412.ref016]\]. In addition to efficient CO designation, it is essential that the resolution of joint molecules (JM) at CO-designated sites is biased toward the CO outcome. JMs result from second-strand capture and DNA synthesis after strand invasion \[[@pbio.1002412.ref017]\]. Depending on where the cut is made by structure-specific endonucleases (resolvases), CO or NCO products can be generated \[[@pbio.1002412.ref018]--[@pbio.1002412.ref021]\]. In *C*. *elegans*, at least two parallel pathways of a redundant resolvase system have been defined (MUS-81/SLX-1 and XPF-1/HIM-6) \[[@pbio.1002412.ref022]--[@pbio.1002412.ref024]\]. It is unclear how the bias in resolution toward either a CO or NCO is achieved. In yeast, the bulk of DSBs not destined to become COs are processed by synthesis-dependent strand annealing (SDSA) at an early stage \[[@pbio.1002412.ref025]\]. In *C*. *elegans*, a similar activity is attributed to RTEL-1 by D-loop displacement after strand invasion \[[@pbio.1002412.ref026],[@pbio.1002412.ref027]\]. NCOs can also be generated by the activity of the RTR complex, composed of a RecQ helicase (Bloom syndrome protein \[BLM\]), topoisomerase IIIα (TOP3), and the RecQ-mediated genome instability 1 and 2 (RMI1,2) scaffolding proteins \[[@pbio.1002412.ref028]--[@pbio.1002412.ref030]\]. This complex provides a major NCO activity during mitosis. Importantly, patients with mutated BLM helicase display genomic instability and a cancer predisposition, and they exhibit a cellular phenotype of elevated COs between sister chromatids, which potentially drives cancer by loss of heterozygosity \[[@pbio.1002412.ref031],[@pbio.1002412.ref032]\]. In vitro studies have shown that the RTR complex dismantles double Holliday junctions (dHJs). A branch migration step mediated by the RecQ helicase brings the two dHJs in close proximity to allow the topoisomerase to unhook the two DNA strands by decatenation, stimulated by the scaffolding proteins RMI1/2 \[[@pbio.1002412.ref033]\]. RMI1 was discovered as a protein associated with BLM containing protein complexes \[[@pbio.1002412.ref034]\]. It colocalizes with recombination foci containing BLM helicase \[[@pbio.1002412.ref035]\]. RMI1 depletion leads to destabilization of BLM and topoisomerase foci. Moreover, its depletion phenocopies the BLM mutant phenotype \[[@pbio.1002412.ref035]\]. Furthermore, RMI1 directly interacts with both BLM helicase and TOP3, shown with pulldown experiments using recombinant protein \[[@pbio.1002412.ref036]\]. Human RMI1 protein contains two oligonucleotide/oligosaccharide-binding (OB) fold domains, in which the N-terminal OB fold binds both the helicase and the topoisomerase \[[@pbio.1002412.ref037]\]. Structural and biochemical studies established that RMI1 cooperates with the topoisomerase in the strand passage reaction during decatenation. The topoisomerase-generated nick allows strand passage of the other strand, in which RMI1 plays a role in the regulation of opening and closure of the "gate" \[[@pbio.1002412.ref038],[@pbio.1002412.ref039]\]. An "anti-CO" role has been described for BLM orthologs in yeast, plants, mice, and *Drosophila* in meiosis \[[@pbio.1002412.ref025],[@pbio.1002412.ref040]--[@pbio.1002412.ref045]\]. A recent study showed that, in *C*. *elegans*, *him-6* is required to eliminate persistent MutSγ-independent recombination intermediates when an excess is induced by irradiation \[[@pbio.1002412.ref046]\]. Similar "anti-CO" functions have been attributed to other members of the complex: Rmi1 and/or Top3 in yeast and *Arabidopsis* \[[@pbio.1002412.ref047]--[@pbio.1002412.ref052]\]. In yeast, Top3 and Rmi1 cooperate with Sgs1 to prevent aberrant joint molecule (JM) accumulation, but they also act later, without Sgs1, to allow chromosome separation \[[@pbio.1002412.ref051],[@pbio.1002412.ref052]\]. Despite the widely conserved anti-CO function of the RTR complex, several studies suggest that it could also contribute a pro-CO activity; however, a direct involvement has not been demonstrated. In *C*. *elegans*, *him-6* (the BLM ortholog) is required to ensure that CO-designated recombination intermediates mature into interhomolog COs \[[@pbio.1002412.ref046],[@pbio.1002412.ref053],[@pbio.1002412.ref054]\]. In yeast, Sgs1 has been found colocalizing with the pro-CO factor Zip3 \[[@pbio.1002412.ref040]\]. Nevertheless, in *sgs1* mutants, the recombination rate is increased rather than decreased, as one would expect if it favored CO formation. More recently, the Sgs1/BLM helicase has been proposed to have an early role in promoting COs, acting to disassemble unprotected strand invasion intermediates to channel them back into CO pathways \[[@pbio.1002412.ref041],[@pbio.1002412.ref044],[@pbio.1002412.ref045],[@pbio.1002412.ref055],[@pbio.1002412.ref056]\]. Furthermore, in a situation in which JM resolution is blocked (*mms4 slx4 yen1*), Sgs1 can promote CO formation. Here, *sgs1* might generate JMs with a certain configuration that could be recognized by the major yeast CO pathway Exo1-MutLγ \[[@pbio.1002412.ref055]\]. In this study, we present the role of *C*. *elegans* RMH-1. We show that RMH-1 and HIM-6 localize at numerous recombination intermediates that will become either a CO or NCO. Consistent with its localization, RMH-1 plays genetically separable and antagonistic roles during meiosis. On one hand, *rmh-1* ensures CO formation at two levels: by promoting CO designation and by protecting CO-designated sites and promoting their maturation into chiasmata. Remarkably, RMH-1 and HIM-6 localize as closely juxtaposed doublets, suggesting that they may flank the two junctions of a dHJ CO intermediate. In addition, *rmh-1* prevents illegitimate connections between homologs redundantly with *smc-5*, an activity known from yeast meiosis to prevent multi-JMs \[[@pbio.1002412.ref057]--[@pbio.1002412.ref059]\]. Moreover, RMH-1 anti-CO activity is required to inhibit CO formation at chromosome centers, which would facilitate correct segregation of chromosomes. Results {#sec002} ======= Identification of *rmh-1* Mutants {#sec003} --------------------------------- The *C*. *elegans* genome encodes two RMI1 homologs: M01E11.3 and T07C12.12, named RMI1 homolog-1 (RMH-1) and -2 (RMH-2), respectively. These proteins contain three domains characteristic of RMI1 orthologs: an N-terminal domain of unknown function (DUF)1767 and two OB-fold domains ([Fig 1A](#pbio.1002412.g001){ref-type="fig"} and [S1A Fig](#pbio.1002412.s001){ref-type="supplementary-material"}). Point mutations in the recessive alleles *rmh-1(jf54*) and *rmh-1(tn309)* were isolated based on defects in meiotic chromosome segregation and embryonic lethality \[[@pbio.1002412.ref060]--[@pbio.1002412.ref062]\]. The mutations lead to protein truncations after the DUF1767 or OB1 domains, respectively ([Fig 1A′](#pbio.1002412.g001){ref-type="fig"} and [S1B--S1D Fig](#pbio.1002412.s001){ref-type="supplementary-material"}). In *rmh-1(jf92)* and *rmh-2(jf94)*, the ORFs were disrupted right after the start codon (see the [Experimental Procedures](#sec015){ref-type="sec"} section). {ref-type="supplementary-material"}). In the *tn309* allele, the G-to-A transition introduces a premature STOP codon at position aa 640, leading to the deletion of the OB2 domain (for more details, see the [Experimental Procedures](#sec015){ref-type="sec"} section). (B--E) Oocyte nuclei at the diakinesis stage of meiotic prophase. Each image shows the complete set of DAPI-stained chromosomes from a single nucleus. The wild-type (WT) and *rmh-2(jf94)* nuclei contain six bivalents (homolog pairs connected by chiasmata), whereas the *rmh-1* mutant nuclei contain a mixture of bivalents and univalents. (F) Quantification of the average number of DAPI-positive structures in diakinesis oocytes in the -1 position. WT *n* = 25, *rmh-1(jf92) n* = 21, *rmh-1(jf54) n* = 36, *rmh-1(tn309) n* = 30, and *rmh-2(jf94) n* = 18. (G--I) Quantification of embryonic hatch rates (G). Frequencies of male offspring (H) and larval arrest (I) among the progeny of WT and *rmh-1* and *rmh-2* mutant worms (*n* = 35--45 hermaphrodites per genotype). Data for F--I are represented as mean +/- SD; ns stands for not significant, differences are highlighted with stars (\* *p* \< 0.05, \*\* *p* \< 0.01, \*\*\* *p* \< 0.001. and \*\*\*\* *p* \< 0.0001). Scale bars, 2 μm.](pbio.1002412.g001){#pbio.1002412.g001} RMH-1 (But Not RMH-2) Is Required for Robust Chiasma Formation and Chromosome Segregation in Meiosis {#sec004} ---------------------------------------------------------------------------------------------------- To establish the respective requirements of *rmh-1* and *rmh-2*, we examined the mutants for embryonic lethality, larval arrest, and a high incidence of males (Him) phenotype among the progeny. An increased incidence of males in the progeny indicates X chromosome nondisjunction during meiosis, as *C*. *elegans* hermaphrodites have the genotype X/X and males X/O. The combination of increased embryonic lethality with a Him phenotype further suggests the occurrence of aneuploid eggs arising from meiotic missegregation of autosomes \[[@pbio.1002412.ref063]\]. Increased larval arrest is indicative of somatic defects. Analysis of *rmh-1* mutants revealed the cohort of phenotypes characteristic of a defect in meiotic recombination, as described above ([Fig 1G and 1H](#pbio.1002412.g001){ref-type="fig"}). Second, DAPI staining of oocytes at diakinesis, the last stage of meiotic prophase, revealed that a fraction of chromosome pairs was not connected by chiasmata in *rmh-1* mutants. Whereas in the wild type, six bivalents (chromosome pairs connected by chiasmata) can be seen in diakinesis oocytes, *rmh-1* mutant alleles showed an increase of DAPI-stained bodies, reflecting a mixture of bivalents and univalents (achiasmate chromosomes) ([Fig 1B--1E](#pbio.1002412.g001){ref-type="fig"}). In the -1 oocyte (the most mature oocyte, next to the spermatheca, the oocyte prior to fertilization), we observed on average 8.9 ± 1.0 DAPI bodies in *rmh-1(jf92)* and 9.3 ± 1.4 in *rmh-1(jf54)* ([Fig 1F](#pbio.1002412.g001){ref-type="fig"}). As the *jf54* allele behaved identically to the *jf92* insertion allele with regards to the hatch rate and Him and diakinesis phenotypes, and as we were unable to detect GFP::RMH-1 containing the *jf54* mutation ([S2B Fig](#pbio.1002412.s002){ref-type="supplementary-material"}), we infer that *rmh-1(jf54)* behaves like a null allele. Interestingly, fewer univalents were observed in *rmh-1(tn309)* (7.4 ± 1.1 DAPI bodies), suggesting that RMH-1 function is differently affected by this allele ([Fig 1F](#pbio.1002412.g001){ref-type="fig"}). To confirm the presence of univalents in *rmh-1* mutants (and not fragments due to absence of DNA repair), we measured the average volumes of DAPI bodies in diakinesis ([S3A and S3B Fig](#pbio.1002412.s003){ref-type="supplementary-material"} and [S1 Text](#pbio.1002412.s012){ref-type="supplementary-material"}) and confirmed both the presence of univalents and the absence of DNA fragments. In contrast to the *rmh-1* mutants, the *rmh-2(jf94)* single mutant did not exhibit any meiotic phenotypes. However, all *rmh* mutants showed increased levels of larval arrest (around 12% for *rmh-1* mutants and 7% for *rmh-2*) ([Fig 1I](#pbio.1002412.g001){ref-type="fig"}). Furthermore, the double mutant *rmh-1(jf54); rmh-2(jf94)* was inviable, suggesting a redundant function of the two genes in somatic tissue. To better understand which meiotic processes are impaired in *rmh-1(jf54)*, we analyzed chromosome pairing, synapsis, and induction and repair of DSBs. No defects in pairing and synapsis were observed ([S4 Fig](#pbio.1002412.s004){ref-type="supplementary-material"}). To study the induction and processing of DSBs, we assessed the appearance and disappearance of foci of the strand exchange protein RAD-51 ([Fig 2](#pbio.1002412.g002){ref-type="fig"}) \[[@pbio.1002412.ref064],[@pbio.1002412.ref065]\]. RAD-51 foci appeared with normal timing in the *rmh-1(jf54)* mutant, but accumulated at higher levels than in wild type and persisted at elevated levels through late pachytene. However, RAD-51 foci disappeared in diplotene, suggesting that repair of DSBs was delayed but did eventually occur. We also found that *rmh-1*-specific chromosomal defects depended on the induction of meiotic DSBs. In summary, meiotic recombination is aberrant in *rmh-1* mutants. {#pbio.1002412.g002} Dynamic Localization of RMH-1 in Distinct Foci during Pachytene Progression {#sec005} --------------------------------------------------------------------------- To assess RMH-1 localization in the gonad, we expressed GFP::RMH-1 using its endogenous regulatory elements as a functional fusion protein. Indeed, GFP::RMH-1 is capable of rescuing embryonic lethality and the Him phenotype when expressed in the null allele *jf92* ([S5A and S5B Fig](#pbio.1002412.s005){ref-type="supplementary-material"}). GFP::RMH-1 was found in foci throughout pachytene ([Fig 3A--3C](#pbio.1002412.g003){ref-type="fig"}). In early pachytene, RMH-1 was first detected as a nucleoplasmic haze, and faint foci progressively appeared on chromosomes. During mid pachytene, RMH-1 foci increased in number and intensity. Mid pachytene nuclei contained, on average, 15.2 foci ± 3.6, ranging from ten to 25, and, in late pachytene, the number of RMH-1 foci drastically reduced to 5.9 ± 1.7 ([Fig 3D](#pbio.1002412.g003){ref-type="fig"}). In late pachytene, the number of foci ranged from one to nine foci; however, the majority of nuclei contained six or seven foci ([Fig 3E](#pbio.1002412.g003){ref-type="fig"}). (This distribution is broader than the distribution of COSA-1 foci, visible as six signals in late pachytene. However, the mutant phenotype of *cosa-1* already suggests that it acts before the protein can be observed in the six prominent foci. In addition, we believe RMH-1 marks a stage of recombination, through which all mature CO intermediates have to pass). {#pbio.1002412.g003} Importantly, RMH-1 foci seem to increase in signal intensity (about 1.8-fold) between mid and late pachytene. This suggested that the protein might accumulate over time. RMH-1 localization in foci in pachytene is reminiscent of localization of markers of ongoing recombination events such as RAD-51 and CO-promoting factors MSH-5, ZHP-3, or COSA-1 \[[@pbio.1002412.ref012],[@pbio.1002412.ref064]--[@pbio.1002412.ref066]\]. Moreover, RMH-1 localization depends on meiotic DSBs, since foci were absent in a *spo-11* mutant ([S2A Fig](#pbio.1002412.s002){ref-type="supplementary-material"}). Co-immunostaining experiments indicated that RAD-51 foci appear and disappear earlier than RMH-1 ([S6 Fig](#pbio.1002412.s006){ref-type="supplementary-material"}). RAD-51 foci appear in transition zone, peak in mid pachytene, and disappear in late pachytene. GFP::RMH-1 foci appear in early pachytene, peak in mid pachytene, and disappear in diplotene. Moreover, when RAD-51 and RMH-1 are present in the same nucleus, they do not colocalize (only 4.0 +/- 4.0% RMH-1 foci coincide with RAD-51) ([Fig 3F and 3F′](#pbio.1002412.g003){ref-type="fig"}). This suggests that RAD-51 and RMH-1 mark different recombination intermediates and that RMH-1 may act after RAD-51 removal. In contrast with the lack of colocalization with RAD-51 foci, the six bright RMH-1 foci in late pachytene nuclei do coincide with the localization of pro-CO factors MSH-5, ZHP-3, and COSA-1 at the CO sites (90.1 +/- 10.7% RMH-1 foci colocalize with COSA-1 and MSH-5) ([Fig 3G, 3J and 3K](#pbio.1002412.g003){ref-type="fig"}) \[[@pbio.1002412.ref012],[@pbio.1002412.ref016],[@pbio.1002412.ref066]\]. Interestingly, RMH-1 foci disappear before COSA-1 foci, which persist until diplotene ([S6 Fig](#pbio.1002412.s006){ref-type="supplementary-material"}). RMH-1 foci also partially colocalize with the earlier MSH-5 foci present during mid pachytene; the brighter foci of each protein tend to colocalize, while the fainter ones do not ([Fig 3G--3I](#pbio.1002412.g003){ref-type="fig"}). Furthermore, we found that late-pachytene RMH-1 foci are severely reduced (to one focus on average) in *msh-5*, *zhp-3*, and *cosa-1* mutants, which fail to form CO-designated recombination intermediates. RMH-1 foci in mid pachytene are also less abundant and smaller in these mutants ([Fig 4A--4D](#pbio.1002412.g004){ref-type="fig"}, [S7D and S7E Fig](#pbio.1002412.s007){ref-type="supplementary-material"}), suggesting that *msh-5*, *zhp-3*, and *cosa-1* also contribute to stable localization of RMH-1 during earlier stages. The dynamic localization of RMH-1 through pachytene suggests that it may function at both CO and NCO repair events. {#pbio.1002412.g004} The mid pachytene number of RMH-1 foci, at least twice the number of eventual COs, is consistent with RMH-1 localizing both at CO sites and at sites that will ultimately become NCOs. This is consistent with our analysis of GFP::RMH-1 localization following induction of excessive DNA breaks by ionizing radiation (IR, 50Gy). We found that at 4 h post IR treatment, the number of RMH-1 foci in mid pachytene nuclei had risen significantly, suggesting that RMH-1 was loaded on many of the excess recombination intermediates produced by IR. To understand what would happen to those excess recombination intermediates positive for RMH-1, we analyzed gonads 8 h post IR treatment (after this time, the nuclei with the increased number of RMH-1 foci should have reached late pachytene). However, at this stage, the numbers were reduced to wild type numbers of foci in late pachytene nuclei ([S7A--S7C Fig](#pbio.1002412.s007){ref-type="supplementary-material"}). Thus, we conclude that RMH-1 marks recombination intermediates that can be increased transiently by irradiation but will be repaired as NCOs, as the final number of marked CO sites is not affected. Crossover Promoting Function(s) of RMH-1 {#sec006} ---------------------------------------- The robust accumulation of RMH-1 at CO sites, the dependence of its localization on pro-CO factors, and the presence of univalents in the *rmh-1* mutants together suggest that RMH-1 likely functions in promoting the CO outcome of initiated recombination events. Several lines of evidence provide further support for pro-CO role(s) for RMH-1. First, we tested whether *rmh-1* is required for normal localization of pro-CO factors by assessing the localization of GFP::COSA-1 in *rmh-1* mutants. This analysis revealed that only 75% of nuclei in *tn309* and 57% of nuclei in *jf54* had six COSA-1 foci (compared to 88% in wild-type) ([Fig 4E and 4F](#pbio.1002412.g004){ref-type="fig"}), indicating a deficit of CO-designated sites in *rmh-1* mutants. Furthermore, we observed several indications of a delay in pachytene progression in both *rmh-1* mutants. First, we assessed localization of ZHP-3, which normally localizes along the full length of the synaptonemal complex (SC) during mid pachytene, then retracts during late pachytene to a short stretch and, finally, to a single focus at the CO site in diplotene \[[@pbio.1002412.ref066]\]; we found that ZHP-3 retraction was delayed in *rmh-1(jf54)* ([S8A--S8D Fig](#pbio.1002412.s008){ref-type="supplementary-material"}). Second, we evaluated a readout of a checkpoint-like feedback mechanism that operates during *C*. *elegans* meiosis to coordinate meiotic progression with generation of CO-eligible recombination intermediates. A single chromosome pair that lacks a CO-eligible recombination intermediate can be detected and elicit a prophase progression delay, visible by the persistence of markers such as phospho-SUN-1 S8 on the nuclear envelope \[[@pbio.1002412.ref067]\]. SUN-1 S8Pi staining was prolonged in *rmh-1* mutants, indicating a delay in prophase progression ([S8E--S8H Fig](#pbio.1002412.s008){ref-type="supplementary-material"}), which is in line with delayed and/or impaired formation of CO-specific recombination intermediates. These data in combination with the observed reduction in COSA-1-marked CO-designated sites together indicate a role for RMH-1 in ensuring efficient CO designation. Although the number of CO-designated sites is reduced in *rmh-1* mutants, CO designation still occurs to a large extent, as 57% to 75% of nuclei contain six COSA-1 foci. Moreover, the reduction in number of COSA-1-marked CO sites is not sufficient to account for the number of univalents observed. To calculate the expected number of DAPI bodies in diakinesis nuclei based on our quantification of COSA-1 foci in *rmh-1* mutants, we assumed that pachytene nuclei containing six COSA-1 foci would lead to diakinesis nuclei with six bivalents, pachytene nuclei containing five COSA-1 foci would lead to diakinesis nuclei with seven DAPI bodies (five bivalents and two univalents), and so on. This approach yields an expectation of 6.7 DAPI bodies at diakinesis for *rmh-1(jf54)* and 6.6 for *rmh-1(tn309)*, values that are significantly lower than those observed in the -1 oocytes of these mutants ([Fig 1B and 1F](#pbio.1002412.g001){ref-type="fig"}). Thus, the number of univalents present at the end of meiotic prophase is higher than would be expected if all COSA-1-marked sites had successfully matured into COs. To understand this excess of univalents, we investigated the organization of chromosomes in diakinesis. In wild type, the presence of a CO precursor triggers domain restructuring of the ensuing bivalent such that the axial proteins HTP-1/2 and LAB-1 become concentrated along the long arm of the bivalent, and the synapsis protein SYP-1 and the Aurora B kinase (AIR-2) are found on the short arm \[[@pbio.1002412.ref068]--[@pbio.1002412.ref070]\]. In *spo-11* or *msh-5* mutants, short and long arm markers overlapped on the univalents. In *rmh-1* mutants, however, we observed long and short arm markers localized to reciprocal domains on univalents ([Fig 4G](#pbio.1002412.g004){ref-type="fig"}), suggesting that a CO site had been designated but failed to mature to actually link the homologs. Furthermore, quantification of DAPI bodies during progression through the diakinesis stage (-3, -2, and -1 oocyte) showed that they increased significantly ([Fig 4H--4J](#pbio.1002412.g004){ref-type="fig"}). We observed that in the -3 oocyte, the number of DAPI bodies in *rmh-1* mutants (6.9 ± 0.7 DAPI bodies for *jf54* and 6.8 ± 0.9 for *tn309*) was significantly increased compared to wild type. Interestingly, the observed number of DAPI bodies in -3 oocytes of the mutants corresponds to the calculated number of DAPI bodies based on COSA-1 foci analysis. However, in both mutants, the number of DAPI bodies strongly increased during oocyte maturation. In *rmh-1(jf54)*, we observed 9.3 ± 1.4 DAPI bodies in the -1 oocytes, while the effect was milder but still significant in *rmh-1(tn309)* (7.4 ± 1.1 in -1 oocytes). We conclude that, in *rmh-1* mutants, connections between homologs may sometimes be lost after CO designation. Genetic Interactions with Mutations Affecting Structure-Specific Endonucleases {#sec007} ------------------------------------------------------------------------------ Dissociation of bivalents that appeared to have undergone CO designation suggested that RMH-1 might function to enforce CO bias in the resolution of recombination intermediates at CO sites. This hypothesis prompted us to examine double mutants carrying *rmh-1(jf54)* in combination with mutations in *xpf-1*, *mus-81*, or *slx-1*, which encode structure-specific endonucleases proposed to function in two separate and partially redundant meiotic resolution pathways \[[@pbio.1002412.ref022]--[@pbio.1002412.ref024]\], and *him-18* (*Slx4 ortholog)*, which encodes a nuclease scaffolding protein \[[@pbio.1002412.ref071]\]. Dissociation of bivalents in *rmh-1(jf54)* was reduced by loss of *xpf-1*, *mus-81*, or *slx-*1 ([Fig 4K](#pbio.1002412.g004){ref-type="fig"}), consistent with persistence of recombination intermediates; however, it is unclear whether such putative persistent intermediates are located at CO-designated sites and/or at other sites on the chromosomes. Furthermore, we observed an elevated incidence of unresolved bivalents linked by DNA bridges ([Fig 4M and 4N](#pbio.1002412.g004){ref-type="fig"}) in these double mutants compared to the *xpf-1*, *mus-81*, or *slx-1* single mutants. Together with previous work showing that such linkages observed in nuclease-defective mutants are *spo-11*-dependent \[[@pbio.1002412.ref022]\], these results suggest that loss of RMH-1 function may result in the accumulation of excess and/or aberrant recombination intermediates that require these structure-specific endonucleases for their resolution. We also observed the occurrence of DNA bridges in the double *rmh-1(jf54); him-18* mutant, but, those linkages appeared with the same frequency as in the *him-18* single mutant. Furthermore, loss of *him-18* in *rmh-1(jf54)* did not appreciably suppress the dissociation of homologs ([Fig 4K](#pbio.1002412.g004){ref-type="fig"}), in contrast to the outcomes with the single-nuclease mutants. We can reconcile these seemingly paradoxical observations by postulating (1) that unscaffolded nucleases may be able to act on recombination intermediates produced in *rmh-1* mutants to remove excess connections between homologs, and (2) that the presence of multiple unresolved intermediates connecting homologs (in *rmh-1; xpf-1*, *rmh-1 mus-81*, or *rmh-1 slx-1* mutants) might obscure detection of discrete bridges. Interdependent and Separable Functions of RMH-1 and HIM-6 (BLM Helicase) {#sec008} ------------------------------------------------------------------------ As RMI1 was first identified as a component of a complex containing the RecQ helicase BLM and the topoisomerase TOP3, we used the yeast-two-hybrid assay to confirm interactions between RMH-1 and HIM-6 (the *C*. *elegans* BLM ortholog) and between RMH-1 and TOP-3 ([S9A Fig](#pbio.1002412.s009){ref-type="supplementary-material"}). Furthermore, as several phenotypes of the *rmh-1* mutants resemble phenotypes seen in *him-6* mutants ([S9B and S9C Fig](#pbio.1002412.s009){ref-type="supplementary-material"}) \[[@pbio.1002412.ref046],[@pbio.1002412.ref053]\], we investigated the cytological and functional inter-relationships between RMH-1 and HIM-6. To examine colocalization of RMH-1 and HIM-6, we used the CRISPR method to tag the *him-6* locus with a C-terminal human influenza hemagglutinin (HA) epitope. The tag did not compromise HIM-6 function, since neither an increase in embryonic inviability nor a Him phenotype was observed ([S5 Fig](#pbio.1002412.s005){ref-type="supplementary-material"}). HIM-6 is found in few foci in early pachytene. The number of foci increases in mid pachytene and decreases again in late pachytene, reminiscent of the RMH-1 pattern. In mid pachytene, RMH-1 and HIM-6 substantially colocalized, and, in late pachytene, RMH-1 strictly colocalized with HIM-6; however, additional HIM-6 foci were also present (96.6 +/- 4.1% RMH-1 foci colocalize with HIM-6 in mid pachytene and 96.3 +/- 6.7% in late pachytene; [Fig 5A--5C](#pbio.1002412.g005){ref-type="fig"}). {ref-type="fig"}. Difference in COSA-1 foci distribution was assessed by a Mann Whitney test: ns stands for not significant and \*\*\*\* *p* \< 0.0001. (G) Quantification of the number of DAPI positive structures in the -1 oocyte for WT; single mutants (*rmh-1(tn309)- him-6(ok412)- rmh-1(jf54)) and* the double *rmh-1(tn309)*; *him-6* and *rmh-1(jf54)*; *him-6* (*n* = 15--36 nuclei per genotype). Data are represented as mean +/- SEM with ns (not significant) and \*\*\*\* *p* \< 0.001. (H) In *him-6(ok412)*, RMH-1 is not detectable in MP (I,I′), but bright RMH-1 foci are present in LP (J,J′), albeit quantification (K) indicates that their numbers are reduced relative to WT (*p* \< 0.001). Data are represented as mean +/- SD. Scale bar 20 μm for gonads, 5 μm for pachytene nuclei insets, and 2.5 μm for single nuclei.](pbio.1002412.g005){#pbio.1002412.g005} To investigate the interdependence of localization, we assayed localization of HIM-6::HA in *rmh-1* mutants. In *tn309*, HIM-6 was completely absent in most of the gonads we imaged. In the rest, HIM-6 was barely detectable, and when foci were seen, they were not at CO sites in late pachytene ([Fig 5D](#pbio.1002412.g005){ref-type="fig"}). In *jf54*, HIM-6 foci were present in all gonads but seemed smaller and fainter in mid and late pachytene, suggesting that RMH-1 is required to stabilize and enrich HIM-6 in foci ([Fig 5E](#pbio.1002412.g005){ref-type="fig"}). Although HIM-6 is sometimes found in proximity to CO site markers, these do not exhibit robust colocalization in the absence of *rmh-1* ([Fig 5E](#pbio.1002412.g005){ref-type="fig"}). Assessment of the localization of GFP::RMH-1 in the *him-6(ok412)* mutant revealed that *him-6* is required for RMH-1 localization in mid pachytene but not in late pachytene ([Fig 5H--5J](#pbio.1002412.g005){ref-type="fig"}). This finding is consistent with the idea that RMH-1 may be present in genetically distinguishable complexes in mid and late pachytene stages. However, *him-6* is required for normal numbers of RMH-1 foci even in late pachytene ([Fig 5K](#pbio.1002412.g005){ref-type="fig"}). An important difference between *rmh-1* and *him-6* mutants is their impact on CO designation. A previous study showed that in the null allele of *him-6*, the abundance of COSA-1 foci was unaffected \[[@pbio.1002412.ref046]\]; thus, the *him-6* null mutant appears to be proficient for CO designation. On the other hand, both *rmh-1* alleles impair CO designation, as evidenced by an elevated frequency of nuclei containing fewer than six COSA-1 foci ([Fig 5F](#pbio.1002412.g005){ref-type="fig"}). Furthermore, the fact that CO designation is unaffected in a *him-6* null mutant but is impaired in an *rmh-1; him-6* double mutant implies that RMH-1 can assist in promoting CO designation in the absence of HIM-6. A deficit of bivalent connections is a shared feature of *rmh-1* and *him-6* mutants. The phenotype is evident but less pronounced in *him-6* or *rmh-1(tn309)* (*p* \< 0.05), while it is more prominent in *rmh-1(jf54)* (*p* \< 0.01). The number of DAPI bodies in *rmh-1(tn309); him-6(ok412)* and the respective single mutants was not different. Interestingly, however, we found that the deficit of connections between homologs in *rmh-1(jf54); him-6(ok412)* oocytes was reduced compared to the *rmh-1(jf54)* single mutant ([Fig 5G](#pbio.1002412.g005){ref-type="fig"}), despite the fact that COSA-1 foci were reduced in this double mutant. One possible explanation for this observation would be that the dissociation of bivalents observed in *rmh-1(jf54)* might be mediated by HIM-6. An alternative explanation is that the simultaneous loss of HIM-6 and RMH-1 may lead to persistence of unresolved recombination intermediates that maintain connections between the homologs. RMH-1 and HIM-6/ BLM Pachytene Foci Are Doublets/Elongated Structures {#sec009} --------------------------------------------------------------------- We investigated the foci structure of RMH-1 and HIM-6 foci using structured illumination microscopy (SIM). Both in mid pachytene and late pachytene, RMH-1 and HIM-6 signals were resolved into more complex structures ([Fig 6](#pbio.1002412.g006){ref-type="fig"}). In mid pachytene, we observed foci that clearly exhibit a doublet structure and foci that show an elongated shape ([Fig 6A and 6B](#pbio.1002412.g006){ref-type="fig"}). In late pachytene, RMH-1 and HIM-6 appear even more clearly as doublets ([Fig 6C--6E](#pbio.1002412.g006){ref-type="fig"}). In both pachytene stages, we hypothesize that RMH-1 and HIM-6 foci mark similar recombination intermediates. The doublet structure is likely easier to observe in late foci as RMH-1 protein accumulates during pachytene progression, leading to larger foci in late pachytene. We measured the average distance between the foci peak in the three-dimensional stacks as an average of 227 +/- 46 nm. Interestingly, in *Drosophila*, the size of a recombination nodule has been estimated to around 100 nm \[[@pbio.1002412.ref072]\]. It has also been shown that 1 kb of B-form DNA is 340 nm in length. We conclude that the structure we describe here is on an appropriate scale for flanking dHJs. We also observed that the late pachytene HIM-6 doublet structures flank a COSA-1 focus in the wild type ([Fig 6F](#pbio.1002412.g006){ref-type="fig"}). Interestingly, in the *rmh-1(jf54*) mutant, HIM-6 appears as single foci and not as elongated or doublet structures in late pachytene ([Fig 6G](#pbio.1002412.g006){ref-type="fig"}). This suggests that *rmh-1* is required to concentrate HIM-6 at CO sites and also to organize a complex structure surrounding late recombination intermediates. {#pbio.1002412.g006} SMC-5 and RMH-1 Act Synergistically to Eliminate Illegitimate Interhomolog Connections {#sec010} -------------------------------------------------------------------------------------- The mid pachytene localization of RMH-1 (and HIM-6) to numerous foci in excess of the eventual number of COs, together with the well-known role for the RTR complex in discouraging COs in other systems (see [Introduction](#sec001){ref-type="sec"}), led us to investigate the possibility that RMH-1 also functions at NCO sites during *C*. *elegans* meiosis. We focused on the possibility that RMH-1 might function in parallel with the SMC-5/6 complex, based on several considerations. First, several studies in budding yeast showed that in the absence of *Smc5* or *Smc6* during meiosis, COs form but joint molecules accumulate, leading to chromosome segregation defects \[[@pbio.1002412.ref057]--[@pbio.1002412.ref059]\]. Moreover, it was shown that Smc6 was required for the resolution of the joint molecules accumulating in absence of Sgs1, suggesting collaboration of pathways mediated through those two proteins \[[@pbio.1002412.ref059]\]. Second, a prior study showed that the *C*. *elegans* SMC-5/6 complex is important for meiotic DSB repair under conditions in which interhomolog CO formation is abrogated and/or in mutant situations when intersister repair is the only option \[[@pbio.1002412.ref073]\]. Moreover, the data of Bickel et al. are fully consistent with the SMC-5/6 complex participating in interhomolog NCO repair as well. Several lines of evidence support the hypothesis that RMH-1 functions in parallel with the SMC-5/6 complex to antagonize accumulation of aberrant interhomolog connections. First, we observed a significant increase in RMH-1 foci in mid pachytene nuclei in the *smc-5* mutant; 29.1% of nuclei contained more than 25 foci, never seen in wild type ([Fig 7A and 7B](#pbio.1002412.g007){ref-type="fig"}). However, late pachytene nuclei in the *smc-5* mutant had an average of six foci per nucleus, as in wild type ([Fig 7C](#pbio.1002412.g007){ref-type="fig"}), suggesting that an excess of RMH-1-associated recombination intermediates accumulate in the *smc-5* mutant, but they are ultimately taken care of in agreement with the wild type viability of the single mutant ([Fig 7D](#pbio.1002412.g007){ref-type="fig"}). Second, the *rmh-1(jf54); smc-5(ok2421)* double mutant showed both a synthetic decrease in embryonic hatch rates and increase in larval arrest ([Fig 7D and 7E](#pbio.1002412.g007){ref-type="fig"}), indicating a strong genetic interaction between *rmh-1* and *smc-5* in the soma but also likely during meiosis (a similar genetic interaction is reported between *him-6* and *smc-5* in \[[@pbio.1002412.ref074]\]). Third, quantification of the number of DAPI bodies in diakinesis oocytes in the *rmh-1; smc-5* double mutant revealed an average of 5.9 ± 0.3 DAPI bodies in the -1 oocyte ([Fig 7F](#pbio.1002412.g007){ref-type="fig"}), consistent with the presence of connections between the homologs. However, cytological examination of chromosome organization revealed that these interhomolog connections were aberrant. Consistent with the low viability of this double mutant, bivalent organization was dramatically defective: both long arms (indicated by LAB-1 staining) were always found together, precluding the wild-type cross-shaped appearance of the bivalent ([Fig 7G--7J](#pbio.1002412.g007){ref-type="fig"}). Such abnormal bivalents were already observed at a low frequency in each of the single mutants, but are now present in each diakinesis nucleus in the double mutant ([Fig 7K](#pbio.1002412.g007){ref-type="fig"}). This result strongly suggests that *rmh-1* and *smc-5* cooperate to eliminate and/or prevent accumulation of connections between homologs at NCO sites. Finally, we found that simultaneous loss of both RMH-1 and SMC-5 results in a high frequency of interhomolog connections even in the absence of the conserved meiotic CO factor ZHP-3. Whereas COs and chiasmata are eliminated in the *zhp-3* single mutant \[[@pbio.1002412.ref016]\], a small but significant reduction in DAPI bodies in -1 oocytes was observed in the *zhp-3; smc-5* mutant ([Fig 7L, 7M and 7O](#pbio.1002412.g007){ref-type="fig"}). This ability to create aberrant homolog connections in a *zhp-3* mutant is consistent with the finding that a mutation affecting the budding yeast Smc5/6 complex has the ability to create connections between homologous chromosomes in a mutant lacking Zip3 (the ZHP-3 ortholog) \[[@pbio.1002412.ref059]\]. Moreover, ectopic ZHP-3-independent connections occurred at very high frequency in a *zhp-3; rmh-1; smc-5* triple mutant, in which we observed an average of 7.3 DAPI bodies at diakinesis ([Fig 7N and 7O](#pbio.1002412.g007){ref-type="fig"}). Together, our data indicate that *smc-5* and *rmh-1* act in parallel to antagonize MutS gamma-independent inter-homolog connections, supporting the conclusion that RMH-1 also functions at NCO sites. {#pbio.1002412.g007} RMH-1 Discourages COs in the Center of Chromosomes {#sec011} -------------------------------------------------- To assess the role of RMH-1 at NCO sites in more detail, we analyzed the localization of COs along the chromosomes. We used deep sequencing to compare CO frequency and CO distribution between the wild type and *rmh-1* mutants (*jf54* and *tn309*). We took advantage of the *C*. *elegans* Hawaiian (Hw) strain, which contains a high frequency of SNPs compared to the Bristol strain (wild type). Thus, we generated strains containing *rmh-1* mutants in combination with a subset of introgressed "Hawaii chromosomes": chromosomes II and V for *rmh-1(jf54)* and chromosomes X, IV, and V for *rmh-1(tn309)*. (Introgression of all Hawaii chromosomes into the *rmh-1* mutants led to sterility). We sequenced all the parental strains (Bristol, Hw, and the "Hw introgressed *rmh-1* mutants") and used them as references. Recombinants between the Bristol strain and Hw were generated as described in [Fig 8A](#pbio.1002412.g008){ref-type="fig"}. We singled F2 animals and allowed them to self-fertilize for three to four generations and deep-sequenced DNA isolated from their progeny. Paired-end reads were mapped as described in the Experimental Procedures section. Bioinformatic analysis identified 166,928 homozygous unique SNPs in the Hw genome. To assess the chromosome composition in each recombinant, we used the homozygous unique SNPs of the Hw strain. We used SNPs located at least 400 bp apart to avoid having two or more SNPs per paired read. In the absence of recombination, two genotypes are expected in the offspring: fully homozygous (i.e., both homologous chromosomes come from the same parental strain) and fully heterozygous (each homolog comes from one parental strain). If recombination occurred, blocks of homozygosity and blocks of heterozygosity were detected. We identified the recombination break-points by analyzing the changes in heterozygosity along the chromosome (for more details, see the [Experimental Procedures](#sec015){ref-type="sec"} section). {ref-type="sec"} section. The relative distribution of COs in the center versus arm domains differed from the WT for *tn309* (*p* = 0.046, Chi^2^ test) and for *jf54*, (p = 0.062, Chi^2^ test).](pbio.1002412.g008){#pbio.1002412.g008} We did not detect a significant difference between the mutants and wild-type control in the total frequencies of crossovers ([Fig 8B](#pbio.1002412.g008){ref-type="fig"}). This observation contrasts with the reduction in COSA-1 foci observed in both mutants, and could potentially reflect sample size or the fact that viable progeny were used in the recombination assay; alternatively, it may reflect crossovers occurring at sites not marked by COSA-1 foci in the mutants. However, we found that the positions of the COs were shifted significantly toward the center of chromosomes in both mutants, in contrast to the preference observed in the wild type, in which the majority of COs occurred in the gene-sparse "arm" regions of the chromosomes ([Fig 8C and 8D](#pbio.1002412.g008){ref-type="fig"}). We conclude that *rmh-1* is required to discourage COs in the center of chromosomes. A shift of COs toward chromosome centers in the *rmh-1(jf54)* mutant was confirmed by using a PCR-based assay to genotype four SNPs along chromosome IV ([S10 Fig](#pbio.1002412.s010){ref-type="supplementary-material"}). A parallel study analyzed the recombination rate and CO position in the *him-6* mutant. Here, the entire brood was used (dead and living progeny) to determine recombination; similarly, a shift toward the chromosome centers was observed \[[@pbio.1002412.ref074]\]. Discussion {#sec012} ========== Genetically Separable Roles for RMH-1 at CO and NCO Sites {#sec013} --------------------------------------------------------- Although protein complexes involving RMI1 and/or BLM helicase were initially recognized based on anti-CO activities, recent work in several systems has led to a growing realization that BLM and RMI1 play multiple roles during meiotic recombination, functioning both in antagonizing and promoting COs (for review, see \[[@pbio.1002412.ref075]\] and [Introduction](#sec001){ref-type="sec"}). One current view for *Saccharomyces cerevisiae* is that the primary role of the RTR complex is as a "recombination intermediate chaperone" \[[@pbio.1002412.ref041],[@pbio.1002412.ref044],[@pbio.1002412.ref045],[@pbio.1002412.ref055],[@pbio.1002412.ref056]\]. Under this model, RTR plays an indirect role in promoting COs during yeast meiosis, either by generating recombination intermediates recognized and processed by the Exo1-MutLγ pathway or by disassembling aberrant recombination intermediates, thereby helping to channel early recombination intermediates into either a MutSγ-dependent CO pathway or into the SDSA pathway for generating NCOs. Yeast Sgs1 has also been implicated in promoting resolution of CO intermediates under circumstances in which JM resolution is impaired \[[@pbio.1002412.ref055]\]. A more prominent pro-CO role has been uncovered for HIM-6/BLM during *C*. *elegans* meiosis, in which loss of BLM function results in reduced COs \[[@pbio.1002412.ref053],[@pbio.1002412.ref054]\]. It was proposed that HIM-6 functions to enforce a CO-biased outcome of resolution of CO intermediates; however, it was not clear whether this proposed pro-CO role for HIM-6 was direct or indirect \[[@pbio.1002412.ref022],[@pbio.1002412.ref046]\]. Our combined genetic and cytological analysis provides a strong case that RMH-1 functions directly at both CO and NCO sites. For a summary, see [Fig 9](#pbio.1002412.g009){ref-type="fig"}. {#pbio.1002412.g009} First, RMH-1 and HIM-6 colocalize at CO designated sites (Figs [3G--3K](#pbio.1002412.g003){ref-type="fig"}, [5A--5C](#pbio.1002412.g005){ref-type="fig"} and [6F](#pbio.1002412.g006){ref-type="fig"}), and this late pachytene localization is dependent on conserved pro-CO factors ([Fig 4A--4D](#pbio.1002412.g004){ref-type="fig"}). Conversely, pro-CO factors can concentrate at presumptive CO sites in the absence of HIM-6 or RMH-1 (albeit less reliably in *rmh-1* mutants) consistent with RMH-1 and HIM-6 being recruited to CO designated sites to perform a late function in CO formation (Figs [4E, 4F](#pbio.1002412.g004){ref-type="fig"} and [5F](#pbio.1002412.g005){ref-type="fig"}). Moreover, SIM imaging of CO sites in late pachytene nuclei revealed that RMH-1 and HIM-6 colocalize in an elongated structure or closely juxtaposed doublets, potentially reflecting two separate populations of RTR complexes flanking the two junctions of a dHJ CO intermediate ([Fig 6](#pbio.1002412.g006){ref-type="fig"}). The "doublet" organization leads us to suggest that RMH-1 and HIM-6 localized at CO sites may support the CO outcome of recombination intermediate resolution either by promoting an optimal geometry for endonucleases (resolvases) to position DNA cleavage, or by protecting the structure from unscheduled action of endonucleases. Our analysis with the truncation allele *tn309* raises the possibility that HIM-6 itself may contribute to inappropriate resolution when RMH-1 is absent. Interestingly, RMH-1 and HIM-6 differ regarding the requirements for their recruitment at CO sites. Whereas RMH-1 is successfully recruited to most presumptive CO sites in *him-6* mutants ([Fig 5H--5K](#pbio.1002412.g005){ref-type="fig"}), HIM-6 does not stably concentrate at the sites of pro-CO factors in *rmh-1* mutants ([Fig 6G](#pbio.1002412.g006){ref-type="fig"}), raising the possibility that RMH-1 (or TOP-3 associated with RMH-1) may interact more directly with pro-CO factors at such sites. The absence of HIM-6 at CO sites in *rmh-1* mutants suggests that RMH-1 might provide a structural support to ensure the final processing of a mature recombination intermediate by limiting branch migration or the collapse of a putative dHJ and, thereby, influencing the outcome of resolution. We provide multiple lines of evidence that RMH-1 also localizes to NCO sites and that its function(s) in NCO repair events are distinct from its pro-CO role(s). First, localization of RMH-1 during early and mid pachytene has different genetic requirements than its late pachytene localization at CO-associated foci: Early/mid pachytene RMH-1 foci can still form (albeit at reduced levels) in mutants defective for pro-CO factors, but are strictly dependent on HIM-6, suggesting that the intermediates present at early and late foci differ in structure (Figs [4A--4D](#pbio.1002412.g004){ref-type="fig"} and [5H--5K](#pbio.1002412.g005){ref-type="fig"}). Furthermore, we show that such foci can occur at elevated levels when numbers of DSB are increased or when alternative pathways for DSB repair are abrogated ([Fig 7A](#pbio.1002412.g007){ref-type="fig"} and [S7A--S7C Fig](#pbio.1002412.s007){ref-type="supplementary-material"}). In mid pachytene, HIM-6 and RMH-1 proteins colocalize, but only a fraction colocalize with MSH-5 ([Fig 3G--3I](#pbio.1002412.g003){ref-type="fig"}). We hypothesize that foci positive for all three proteins correspond to CO-eligible recombination sites, while those positive for only HIM-6 and RMH-1 may be processed as NCOs (perhaps by a decatenation reaction). The occurrence of chromosome bridges in *rmh-1; resolvase* double mutants could indicate either the accumulation of recombination intermediates due to defective dissolution or a more direct collaboration between RMH-1 and resolvases to produce COs and/or NCOs ([Fig 4K--4M](#pbio.1002412.g004){ref-type="fig"}). Together, our data suggest direct but distinct roles for RMH-1 in both CO and NCO repair during *C*. *elegans* meiosis. Functions of RMH-1 at CO and NCO Sites Collaborate to Prepare Chromosomes for Segregation During the Meiotic Divisions {#sec014} ---------------------------------------------------------------------------------------------------------------------- Another important concept emphasized by our work analyzing the functions of RMH-1 during *C*. *elegans* meiosis is that a successful outcome of meiosis requires proper execution of the prescribed recombination events at both CO and NCO repair sites. Whereas COs serve as the basis of chiasmata that connect homologous chromosomes to enable them to orient toward opposite poles of the meiosis I spindle, efficient homolog segregation also depends on elimination of other recombination-based interactions between homologs that might impede their timely separation. Beyond ensuring CO formation, RMH-1 has (at least) two roles in preparing homologous chromosome for segregation at meiosis I. First, RMH-1 functions in parallel with SMC-5 in eliminating excess associations between homologs ([Fig 7](#pbio.1002412.g007){ref-type="fig"}). Importantly, we also demonstrate another major role for RMH-1 in ensuring that COs usually form in positions that favor reliable homolog segregation. In the absence of localized centromeres (due to the holocentric nature of *C*. *elegans* chromosomes), the position of the CO dictates multiple features of bivalent organization that are crucial for correct alignment and behavior of chromosomes \[[@pbio.1002412.ref069]\]. This defines distinct long arm and short arm domains where cohesion will be retained or released at the meiosis I division and delineates how kinetochore components, motor proteins, and cell-division protein kinases associate with the chromosomes. We found that, whereas COs usually occur at off-center positions during wild-type meiosis, COs occurred more frequently near the centers of the chromosomes in *rmh-1* mutants ([Fig 8](#pbio.1002412.g008){ref-type="fig"} and [S10 Fig](#pbio.1002412.s010){ref-type="supplementary-material"}). We speculate that an altered distribution of COs in an *rmh-1* mutant may reflect impaired execution at both CO and NCO repair sites. On the one hand, an impaired ability to impose CO-biased resolution at CO-designated sites results in intermediates at some of those sites being resolved as NCOs. On the other hand, impairment of a mechanism that would normally help dissociate intermediates to promote repair via SDSA may result in accumulation of DNA structures that require structure-specific endonucleases for their resolution, resulting in CO repair products at some sites that should have been NCOs. We have not observed a significant frequency of double COs in the mutants, suggesting that the misplaced COs are still subjected to interference. Interestingly, the SLX-1 resolvase is also required to suppress CO formation in the chromosome center \[[@pbio.1002412.ref024]\]. It is appealing to speculate that RMH-1 might cooperate with SLX-1 in a locally defined region to promote NCO resolution. We note that while CO positioning may be especially important in organisms like *C*. *elegans* that use COs to trigger differentiation of the bivalent, CO position is also very important during *Drosophila*, yeast, and human meiosis. in *Drosophila*, centromere proximal recombination events have been correlated to metaphase II nondisjunction \[[@pbio.1002412.ref076]\]. In yeast, centromere proximal COs have been shown to trigger premature separation of sister chromatids (PSSC). Interestingly, *sgs1* mutants exhibit reduced spore viability, potentially due to an excess of COs near the centromere leading to PSSC \[[@pbio.1002412.ref077]\]. In human meiosis, COs located very near to either telomeres or centromeres are associated with elevated risk of aneuploidies, e.g., \[[@pbio.1002412.ref078]--[@pbio.1002412.ref082]\]. Experimental Procedures {#sec015} ======================= Cultivation of worms followed standard protocols \[[@pbio.1002412.ref083]\]. All experiments were performed at 20°C. N2 Bristol was used as wild type. Experiments using GFP::RMH-1 always included the wild-type *rmh-1* gene, if not indicated otherwise. Strains are listed in [S1 Text](#pbio.1002412.s012){ref-type="supplementary-material"}. Raw data for the figures can be found in [S1 Table](#pbio.1002412.s011){ref-type="supplementary-material"}. Cytological Preparations and Immunofluorescence Analysis {#sec016} -------------------------------------------------------- Dissection and immunostaining was performed as described in \[[@pbio.1002412.ref067]\] with modifications when using SYP-1, LAB-1, and S10 phospho-histone 3: dissection and fixation was performed in 1x EBT instead of PBS 1X \[[@pbio.1002412.ref066]\]. Adults 24 h post L4 were dissected. Images are projections of stacks encompassing whole nuclei. For details on antibodies and microscopy, refer to [S1 Text](#pbio.1002412.s012){ref-type="supplementary-material"}. See also [S1 Text](#pbio.1002412.s012){ref-type="supplementary-material"} for *rmh-1* allele isolation and characterization, protein tagging (RMH-1, HIM-6::HA), generation of the *rmh-2* knockout, irradiation conditions, yeast two hybrid assays, photoconversion experiments, and recombination assays. Supporting Information {#sec017} ====================== ###### Identification of *rmh-1* (RMI homolog 1) mutants. \(A\) Alignment of the N- and C-terminal parts of RMI1 homologs: human (Hs), *C*. *elegans* (Ce), *Arabidopsis thaliana* (At), *Saccharomyces pombe* (Sp), and *Saccharomyces cerevisae* (Sc), with the DUF1767 domain (green frame) as described in \[[@pbio.1002412.ref047]\] and the OB domains (orange frame) as described in \[[@pbio.1002412.ref035]\]. (B) The mutation in *rmh-1(jf54)* affects the splice donor site of the first exon. Schematics of PCR primers used to analyze the structures of transcripts produced by the *rmh-1* locus. The spliced version corresponds to the wild type. In the unspliced version, intron 1 is present. Two different guanidines (three base pairs apart), present in exon 1, were used as cryptic splicing donor sites to substitute the original G mutated site in *rmh-1(jf54)*. Those versions led to deletions of intron 1 and a small portion of exon 1 (conserved amino acid in the DUF domain), as illustrated in (A). (C) Quantification of transcript abundance (four replicates). The percentage of transcripts represents the fraction of the total population of transcripts in one genotype (wild type or *jf54*). Data are represented as mean +/- SD (\*\*\* *p* \< 0.001 and \*\*\*\* *p* \< 0.0001). (D) Schematics of the protein products predicted for the different mutant alleles of *rmh-1*: *jf92*, *tn309*, and *jf54*. (TIF) ###### Click here for additional data file. ###### Requirements for GFP::RMH-1 localization. (A and A′) DSBs are required for RMH-1 localization throughout pachytene. In *spo-11(me44)*, which lacks meiotic DSBs, GFP:: RMH-1 is absent from DNA, although a few late pachytene nuclei contain one focus of protein (see inset). (B and C) Gonad from an *rmh-1(jf92)* worm expressing a *gfp*::*rmh-1* transgene containing the *rmh-1(jf54)* point mutation (B and B′) or the *gfp*::*rmh-1(tn309)* point mutation (C and C′). In both cases, GFP::RMH-1 is not detected in foci in the gonad. The fact that the transgene *gfp*::*rmh-1(tn309)* recapitulates the phenotype of the *tn309* allele (in a deletion background) proves expression of the transgene. Scale bars 20 μm for full gonad and 5 μm for insets. (TIF) ###### Click here for additional data file. ###### *rmh-1* mutants exhibit a mixture of bivalents and univalents in diakinesis. \(A\) Quantification of the average volume of a bivalent (assessed in the wild type) and a univalent (assessed in the *msh-5 (me23)* mutant in the -1 oocyte (*n* = 13 nuclei for both genotypes). Data are presented as mean +/- SD with *p* \< 0.0001. (B) Classification of the DAPI structures found in *rmh-1(jf54)* and *rmh-1(tn309)*. Classes (fragment \< univalent \< bivalent \< chromatin masses) were defined by the volume of the structure (*n* = 13 nuclei for wild type, *msh-5(me23)*, and *rmh-1(tn309)*, and *n* = 12 nuclei for *rmh-1(jf54)*). (TIF) ###### Click here for additional data file. ###### *rmh-1* is not required for pairing or synapsis. \(A\) Staining for HIM-8 (X chromosome pairing center binding protein) and (B) FISH with the 5S ribosomal locus (chromosome V) to follow pairing in *rmh-1(jf54)*. (D) Quantification of nuclei with a paired signal of HIM-8 in six equal zones in the gonad, as shown in (C) (*n* = 52--96 nuclei per zone for each genotype). Distribution of paired HIM-8 signals is not different between the wild type and mutant (Chi^2^ test). (E and G) Staining for SYP-1 protein (transverse filament of the SC) on wild type (E) or *rmh-1(jf54)* gonads (G). (F and H) Staining for HTP-3 (axial element of the SC) on wild type or *rmh-1(jf54)* gonads (H). Scale bars 5 μm. (TIF) ###### Click here for additional data file. ###### GFP::RMH-1 and HA::HIM-6 are functional transgenes. \(A\) Quantification of embryonic hatch rates and (B) the frequency of male offspring among the progeny of wild type, *rmh-1(jf92); gfp*::*rmh-1* (*n* = 35 hermaphrodites), *him-6*::*HA* (*n* = 45), and *gfp*::*rmh-1*; *him-6*::*HA* (*n* = 42). Data are represented as mean +/- SD (\* for *p* \< 0.05, \*\*\*\* for *p* \< 0.0001, and ns stands for not significant). (TIF) ###### Click here for additional data file. ###### Time course of recombination markers in prophase I. \(A\) Scheme of a *C*. *elegans* gonad showing the different meiotic stages: transition zone, pachytene (early, mid, and late), diplotene, and diakinesis. (B) Quantification of the average number of RAD-51, GFP::RMH-1, or GFP::COSA-1 foci per nucleus in the different meiotic zones of the gonad (*n* = 3 gonads). Data are represented as mean +/- SEM. (TIF) ###### Click here for additional data file. ###### Mid and late pachytene RMH-1 foci are differently regulated. \(A\) Mid pachytene (MP) nuclei (white square) contain more RMH-1 foci than the wild type, 4 h after 50 Gy irradiation. (B) Stacked bar graph showing the percentage of MP nuclei containing a defined number of RMH-1 foci: unirradiated wild type (*n* = 205) or after irradiation (*n* = 228). Distribution of GFP::RMH-1 differed between irradiated and unirradiated (Mann-Whitney test, *p* \< 0.0001). (C) Quantification of the average number of RMH-1 foci per nucleus in late pachytene in unirradiated wild type (*n* = 180) and 8 h post irradiation (*n* = 86). No difference is observed. Data are represented as mean +/- SD. (D and E) RMH-1 localization in *zhp-3(jf61)* or *msh-5(me23)* mutants. In both cases, RMH-1 foci are reduced and fainter in MP (insets) and absent in late pachytene (LP). Scale bars 20 μm for gonad and 5 μm for insets. (TIF) ###### Click here for additional data file. ###### Prophase progression is delayed in *rmh-1*. (A-D) Staining for SYP-1 and ZHP-3 on wild type (A) and *rmh-1(jf54)* (B). In late pachytene in wild type, ZHP-3 marks the putative CO sites (A′ and A′′), while in *rmh-1(jf54)*, ZHP-3 is still present on the chromatin following the SC staining as stretches (D--D′′). The asymmetric retraction of SYP-1 is also delayed (C--C′′). (E--G) The SUN-1 S8 Pi positive region is extended in *rmh-1*. (H) Quantification of the length of the gonad zone positive for SUN-1 S8 Pi, expressed in percentage of length positive for the marker with regard to the length of zone from meiotic entry to diplotene (*n* = 3 gonads per genotype). Data are represented as mean +/- SD with (\*\*\* for *p* \< 0.001 and \*\*\*\* for *p* \< 0.0001). Scale bars 20 μm for gonad and 5 μm for insets. (TIF) ###### Click here for additional data file. ###### Interplay between RMH-1 and HIM-6. \(A\) Yeast two-hybrid assays for protein interactions between RMH-1, HIM-6, and TOP-3. Full-length RMH-1 interacts with TOP-3 (row 1) and with HIM-6 (row 2), respectively. A truncation protein of RMH-1 containing only the N-terminal part with the DUF and OB1 domains (reminiscent of *tn309*) interacts with TOP-3 (row 5). Auto-activation of both bait and prey vectors were tested in lanes 6 to 10. Interactions were scored by growth on SC-Leu-Trp-His plates (left panel) and growth on SC-Leu-Trp plates (right panel). Quantification of the percentage of hatch rate (B) and males in offspring (C) for wild type (*n* = 45 hermaphrodites), single mutants (*rmh-1(tn309)* (*n* = 45)---*him-6(ok412)* (*n* = 35)---*rmh-1(jf54)* (*n* = 45)), and the double mutants *rmh-1(tn309); him-6* (*n* = 43) and *rmh-1(jf54); him-6* (*n* = 45). Data are represented as mean +/- SD with ns (not significant) and \* *p* \< 0.05, \*\* *p* \< 0.01, \*\*\* *p* \< 0.001, \*\*\*\* *p* \< 0.0001. (TIF) ###### Click here for additional data file. ###### COs are shifted toward the chromosome center in *rmh-1*. \(A\) Schematics of crosses to obtain the F2 individuals used for PCR-based SNP analysis. White insert indicates the wild type (Bristol) background, and black insert indicates the Hawaiian background. (B) Scheme of chromosome IV and location of the SNPs used during the recombination assay. SNP A is localized on the left arm, B is at the junction between left arm and the center, C is at the junction between the center and right arm, and D is on the right arm. (C) Recombination frequencies on chromosome IV left arm between SNP A and B for wild type (*n* = 94 individuals) and *rmh-1(jf54)* (*n* = 96 individuals). (D) Recombination frequencies on chromosome IV center between SNP B and C for wild type (*n* = 94 individuals) and *rmh-1(jf54)* (*n* = 96 individuals). COs were shifted to the centers of the chromosome in the mutant. (E) Recombination frequencies on chromosome IV right arm between SNPs C and D for wild type (*n* = 94 individuals) and *rmh-1(jf54)* (*n* = 96 individuals). (C--E) The column "theoretical" corresponds to the expected recombination frequency based on the published genetic distance (<http://www.wormbase.org>). Data are represented as percentage with ns for not significant, \* for *p* \< 0.05, and \*\* for *p* \< 0.01 (Chi^2^ test on raw data). (F) Graph illustrating the percentage of CO events along chromosome IV for wild type and *rmh-1(jf54)*. In the wild type, COs are concentrated to chromosome arms, whereas in the mutant, they are more concentrated at the center. (TIF) ###### Click here for additional data file. ###### Raw data. (XLSX) ###### Click here for additional data file. ###### Supporting information materials and methods. Detailed descriptions are provided for the strains used and generated in this study, antibodies and imaging methods, FISH procedure, identification of the *rmh-1* alleles, generation of *rmh-1*, *rmh-2* alleles and him6::HA strain by CRISPR method, the tagging of RMH-1, RT-PCR procedure, measurements of DAPI bodies in diakinesis, the Y2H assay, the recombination assays by PCR-based SNP mapping, and deep sequencing. (DOCX) ###### Click here for additional data file. The authors thank M. Velkova and J. Windisch for technical assistance and members of the Jantsch and Loidl laboratories for helpful discussions. We are indebted to A. Gartner and Y. Hong for sharing unpublished data and helpful discussion and P. Pasierbek for advice with imaging. We also thank K. Kelly and E. Tanner for contributions to mapping of the *tn309* allele, a generous gift of D. Greenstein and T. Furuta. Solexa deep sequencing was performed at the Vienna Biocenter Core Facility NGS Unit (<http://www.vbcf.ac.at>). Some strains were provided by the Caenorhabditis Genetics Center, which is funded by National Institutes of Health Office of Research Infrastructure Programs (P40 OD010440). We are grateful for reagents from the Bhalla, Colaiacovo, Dammermann, Gartner, Kraft, and Zetka labs. BLM : Bloom syndrome protein CO : crossover dHJ : double Holliday junction DSB : double strand break DUF : domain of unknown function HA : human influenza hemagglutinin Him : high incidence of males HR : homologous recombination Hw : Hawaiian IR : ionizing radiation JM : joint molecules NCO : non-crossover NGS : Next Generation Sequencing OB : oligonucleotide/oligosaccharide-binding PSSC : premature separation of sister chromatids RMH-1 : RMI1 homolog-1 RTR : RecQ helicase--topoisomeraseIII--Rmi1 SC : synaptonemal complex SDSA : synthesis-dependent strand annealing SIM : structured illumination microscopy RMH-2 : RMI1 homolog 2 [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: MJ, AW, AvH, MG, AMV, VJ. Performed the experiments: MJ, PH, AW, SM, MM, MRDS, LT, AT, CH. Analyzed the data: MJ, AW, AMV, PH, MG, AvH, VJ, LP. Wrote the paper: MJ, AMV, VJ.

1.. Introduction {#s1} ================ Helminths have plagued human beings even before recorded history. Today, it is estimated that roughly one-third of the almost 3 billion people who live on less than 2 USD per day in developing regions of sub-Saharan Africa, Asia and the Americas are infected with one or more species of helminths \[[@RSOB170031C1]\], exacerbating poverty, malnutrition and anaemia as well as causing delays in the physical and intellectual development of infants \[[@RSOB170031C2]\], the most susceptible population. The most common human helminthiases involve infection with intestinal helminths, such as ascariasis, trichuriasis and hookworms, followed by schistosomiasis. Moreover, helminth infections represent a significant economic and health burden to the global ruminant livestock industry, causing huge losses in meat and milk production. Also, resulting in a high mortality rate and the excessive use of anthelmintic treatments \[[@RSOB170031C3],[@RSOB170031C4]\], their effectiveness has been compromised in many geographical areas due to growing resistance \[[@RSOB170031C5]\]. Thus, there is an urgent need to develop new measures to control helminth infections, such as vaccinations \[[@RSOB170031C6]--[@RSOB170031C8]\]. We have previously demonstrated that the catalytic region of the enzyme serine/threonine phosphatase 2A (PP2A), which is highly conserved among a large number of nematode species \[[@RSOB170031C9]\], is capable of inducing partial levels of protection against infections by diverse parasitic nematodes when administered by the mucosal route, specifically intranasally, in mice and lambs. Mucosal immunity involves an intricate network of components of innate and adaptive immunity \[[@RSOB170031C10]\] that act independently of the systemic immune response \[[@RSOB170031C11]\]. In the mucosa, the antigens make contact with the local immune system both by antigen-presenting cells (APC) as well as 'specific microfold cells' (M) that are intercalated in the mucosa with follicles of mucosa-associated lymphoid tissue (MALT) in association with Peyer\'s patches (PPs). Mucosa covers both the aerodigestive and the urogenital tract, constituting a special immune system, as this mucosa is the entry pathway of almost 90% of infections in vertebrates. Despite being anatomically separate, mucosa acts synergistically; for example, intranasal immunization stimulates T or B cells to respond in the intestinal or the urogenital mucosa \[[@RSOB170031C12]\], thus being an alternative to intramuscular, intradermal or subcutaneous vaccination. Apart from the administration route, the induction of the immune response depends on the incorporation of an appropriate adjuvant that triggers the innate response and is capable of tolerating the transition from an innate response to an adaptive response. Some reviews on the use of adjuvants in the immunization through mucosa have recently been published \[[@RSOB170031C12]--[@RSOB170031C18]\]. By the use of a bioconjugation methodology, a number of soluble proteins, including ovalbumin (OVA), cytochrome *c*, Tamm Horsfall glycoprotein and HIV-1 gp120, have been attached to palmitic acid via the ɛ-amine groups of lysine \[[@RSOB170031C19]\]. The first work in which lipopeptides were used as adjuvants were performed by Hopp \[[@RSOB170031C20]\]. Other authors \[[@RSOB170031C21],[@RSOB170031C22]\] found that a lipopeptide of influenza triggered a cytotoxic CD8+ T lymphocyte (CTL) response. This finding enormously boosted the potential use of lipids bound to peptides as vaccines \[[@RSOB170031C23]\]. Conjugated peptides with tripalmitoyl-*S*-glycerylcysteinyl-seryl-serine (P3CSS) induced a CTL response similar to that achieved when live vaccines are used \[[@RSOB170031C24],[@RSOB170031C25]\]. This study tests the ability of recombinant PP2A subunit from the parasitic nematode *Angiostrongylus costaricensis* to induce cross-protective immune responses against a challenge *Trichuris muris* infection in mice when the subunit is administered intranasally and is formulated as a lipoconjugate with oleic-vinyl sulfone (OVS). Results demonstrate that this formulation induces a strong protective immunity in the infected and vaccinated animals. The rapid expulsion of the intestinal parasites in the PP2A-OVS vaccinated animals was found to be associated with the high production of IL-9 in the PPs. 2.. Results {#s2} =========== 2.1.. Synthesis of the oleic-vinyl sulfone {#s2a} ------------------------------------------ The oleic acid was transformed into the vinyl sulfone derivate as described by Morales-Sanfrutos *et al*. \[[@RSOB170031C26]\], in the three steps depicted in [figure 1](#RSOB170031F1){ref-type="fig"}. Compound (9Z)-*N*-\[2-(ethenylsulfonyl)ethoxy\]-ethyl\]-9-octadecenamide was obtained as a solid (279 mg, 81%). M. P. 61--62°C; nmax(film) cm^−1^: 3062, 2915, 2847, 1462, 1284, 1124, 908 and 730; ^1^H-NMR(CDCl~3~, 400 MHz): d 6.62 (dd, 1H, *J* = 16.6 and 9.8 Hz, CH=), 6.43 (d, 1H, *J* = 16.6 Hz, =CH~2~*trans*), 6.15 (d, 1H, *J* = 9.8 Hz, =CH~2~*cis*), 2.96 (t, 2H, *J* = 8.0 Hz), 1.77 (m, 2H), 1.46--1.20 (m, 30H), 0.88 (t, 3H, *J* = 6.7 Hz); ^13^C-NMR (CDCl~3~, 100 MHz): d 136.3, 130.4, 54.4, 32.0, 29.8, 29.7, 29.7, 29.7, 29.6, 29.4, 29.3, 29.1, 28.5, 22.8, 22.4, 14.2. HRMS (*m*/*z*) (FAB+) calcd. for C~20~H~40~O~2~SNa \[M + Na\]^+^: 367.2647; found: 367.2644. Figure 1.Recombinant peptide (rPP2A), oleic-vinyl sulfone (OVS) and lipopeptide (rPP2a-OVS). (*a*) The synthesis of (9Z)-*N*-\[2-(ethenylsulfonyl)ethoxy\]-ethyl\]-9-octadecenamide (OVS). (*b*) Multiple alignments (ClustalW2) of the sequence of the catalytic region of *Angiostrongylus costaricensis* PP2A and different *Trichuris* ssp. Ac, *Angiostrongylus costaricensis*; Ts, *Trichuris suis*; Tt, *Trichuris trichiura.* Black indicates the positions with 100% conservation, while grey represents a decline in conservation. (*c*) SDS-PAGE analysis of purified PP2A and western blot. Lane A, total protein transformed bacteria; lane B, PP2A after affinity chromatography with nickel-agarose column purification; lanes C and D, recognition by the immune serum against rPP2A. (*d*) Mass spectrometry of linked rPP2A to OVS. The arrow corresponds to rPP2A, and asterisks to fusions of 1, 2, 3, 4 and 5 OVSs linked to rPP2A. 2.2.. Lipopeptide: antigen and self-assembly with the oleic-vinyl sulfone {#s2b} ------------------------------------------------------------------------- After the sequencing and analysis of the recombinant peptide and search by Mascot, the recombinant protein was identified as a homologue of the catalytic subunit of the family serine/threonine phosphatase 2A, with a value of *E* = 2 × 10^−16^. The sequence is shown in [figure 1](#RSOB170031F1){ref-type="fig"}. The analysis of the nucleotide sequence of the recombinant protein (rPP2A) using the ClustalW algorithm (<http://www.ebi.ac.uk/Tools/msa/clustalw2/>) confirmed the high identity between the sequence of the catalytic region of recombinant serine/threonine phosphatase 2A from *A. costaricensis* (AcPP2Ar) and the catalytic region of the PP2A from *Trichuris suis* (TsPP2A) and *T. trichiura* (TtPP2A). The analysis of the recombinant protein by 12.5% SDS-PAGE electrophoresis showed that the molecular weight of the recombinant protein was 10 kDa, and the theoretical molecular mass was 9898.4 Da, with an isoelectric point value of pI 9.30 based on the sequence calculated by the ExPASy pI/Mw tool (<http://web.expasy.org/compute_pi/>). Western blot confirmed the identity of the purified protein ([figure 1](#RSOB170031F1){ref-type="fig"}). The recombinant protein was chemically bounded to OVS, yielding a lipopeptide with rPP2A coupled to 1 or 2 molecules of OVS with minor amounts of 3 and 4 OVS molecules and traces of 5 OVS molecules as revealed by mass spectrometry ([figure 1](#RSOB170031F1){ref-type="fig"}). Notably, the electron microscope revealed that the lipopeptide formed micelles with 46.66 ± 6.40 nm in diameter ([figure 2](#RSOB170031F2){ref-type="fig"}). [Figure 2](#RSOB170031F2){ref-type="fig"} shows that the purified rPP2A (not linked to the lipid vinyl sulfone) in a PBS suspension forms aggregates as the result of its low solubility in aqueous solution. Figure 2.Transmission electron microscopy images. (*a*) Micelles formed after the functionalization of the rPP2A with OVS. (*b*) Purified rPP2A in aqueous suspension. 2.3.. Evaluation of the anti-helminth activity {#s2c} ---------------------------------------------- To evaluate the possible adverse effects of the different intranasal inoculation treatments that the different groups of animals underwent, we weighed the mice throughout the experiment, comparing them to the IC group ([figure 3](#RSOB170031F3){ref-type="fig"}). The results indicate that there was no variation among the animals of the different groups, and the variability of the weights in the IC group was even higher than in the rest of the experimental groups. Although without statistical significance (*p* \> 0.05), at the end of the experiment, the mice immunized with rPP2A-OVS and rPP2A-BW increased slightly in weight when compared with the IC, probably because of the parasite loads, as discussed below ([figure 3](#RSOB170031F3){ref-type="fig"}). Figure 3.(*a*) Animal weight (average in grams) from the different immunized groups after the second immunization. No differences were obtained among the groups. (*b*) Number of eggs along time present in each animal group post-immunization. The number of eggs present in rPP2A-OVS immunized animals was significantly lower than in the control group and in those immunized with BW. (*c*) Number of worms found on average in each immunized group. The differences were statistically significant (*p* *\<* 0.001, \*\*\*) in a Tukey--Kramer test. White bars show the IC group; grey bars the BW immunization; dark grey bars rPP2A-BW immunization and back bars the rPP2A-OVS immunization. The average number of eggs expelled by the different groups of mice after immunization is represented in [figure 3](#RSOB170031F3){ref-type="fig"}, showing that one week post-immunization the immunized mice with the formulae that included rPP2A-BW and rPP2A-OVS reached percentages of 88.93% and 92.73% reduction of the number of eggs, respectively. Curiously, in the mice treated only with the BW at 7 days post-immunization a reduction of 7.67% was reached. At 12 days post-immunization, the groups immunized with rPP2A-BW and rPP2A-OVS reached minimum values, not being significant when compared with each other but highly significant compared with the IC and to the BW groups (Tukey--Kramer test, *p* *\<* 0.001). Two weeks post-immunization the number of eggs in mice treated with BW declined 16.13% compared with IC, plunging 99.85% in those treated with rPP2A-BW and 99.01% for those treated with rPP2A-OVS. This reduction not only remained steady but persisted until day 18 of the experiment. The number of worms in the control animals killed at 14 days post-infection (p.i), before beginning the immunization, was 83 ± 31.85, demonstrating that the infection was successful and that *T. muris* was correctly established. The fall in the number of worms compared with the control group at the end of the experiment (when the animals were killed) is shown in [figure 3](#RSOB170031F3){ref-type="fig"}, indicating that the animals immunized with rPP2A-OVS had 97.90% less worms than mice immunized with PP2A-BW (59.88%), and those treated only with BW (24.47%). 2.4.. Interleukin expression assessment by quantitative real-time PCR and evaluation of mucosal immune changes by confocal microscopy {#s2d} ------------------------------------------------------------------------------------------------------------------------------------- The evaluation of the immune response of the different groups of animals carried out as described in Material and methods by laser confocal microscopy in thin sections of the intestinal caecum. The levels of fluorescent areas of the mucine zones, antibody recognition against chemokines CCL20, CCL11 and OX40, neutrophils, the plasma-cell marker CD138 and Tuft cells were determined. Also, the expression levels of different interleukins were evaluated using quantitative real-time PCR, measuring the mRNA-production levels of these proteins both in the PPs and in the MLN. Analysis by confocal microscopy of the intestinal sections in different groups of mice was performed in order to evaluate the levels of mucus of the crypt in the colonic mucosa, using WGA lectin labelled with fluorescein. The results showed that the fluorescent areas inside the crypts were larger in animals treated with rPP2A-OVS than in the rest of the groups, and the intensity of the fluorescence was also stronger. These differences proved significant (Tukey--Kramer test, *p* \< 0.001) compared with the other groups. On the contrary, the values of the area in the IC group and in the group immunized with rPP2A-BW proved similar ([figure 4](#RSOB170031F4){ref-type="fig"}). Figure 4.The evaluation in thin sections of the intestine of the immune response of the different groups by laser confocal microscopy. (*a*) Levels of fluorescent areas of the mucus of the crypt in the colonic mucosa, using WGA lectin labelled with FITC. (*b*) Quantification of the mean values of fluorescence in the different treated groups, lectin labelled. Dark bars correspond to the means area measured plus s.d. of the values; grey bars correspond to the means of fluorescence intensity plus s.d. of the values. (*c*) CCL11 (red labelled with Alexa-Fluor 633) in the thin section of the intestinally treated mice. (*d*) Means of the number of fluorescent marks to CCL11 plus s.d. of the mean values of cells measured in an area of 125 µm^2^. (*e*) OX40 (green labelled with FITC) in the thin section of the intestinally treated mice. (*f*) Average values plus s.d. of fluorescent cells of OX40 in an area of 1000 µm^2^. (*g*) Confocal microscopy studies of plasma cells (CD138 in red with Alexa-Fluor 633) in thin intestinal sections. (*h*) Average number and s.d. of plasma cells measured in an area of 170 µm^2^. The nuclei were labelled with DAPI (blue). BW corresponds to control animals immunized with bacterial walls; rPP2A-BW corresponds to the animal group immunized with the recombinant peptide plus bacterial walls and rPP2A-OVS is the group immunized with the self-assembling lipopeptide. Tukey--Kramer test, *p* *\<* 0.001 (\*\*\*), *p* *\<* 0.01 (\*\*) and *p* *\<* 0.05 (\*). [Figure 5](#RSOB170031F5){ref-type="fig"} presents both the confocal microscopic images of the recognition levels of CCL20 (Alexa-Fluor 647) and the recognition by the specific antibody with the fluorochrome fluorescein isothiocyanate FITC against the neutrophils in an area of 2000 µm^2^ in the sections of intestine from animals of the different groups, taking into account the fluorescence intensity values and the fluorescence percentage occupied with respect to the total area measured. It was found that when the sections were treated with the antibody against the neutrophils, the number of fluorescent cells appearing in the mucosa surrounding the crypts and on the intestine borders was larger, whereas a green colour was appreciable along the borders of the mucosa, likely to be a consequence of the secretion products of these neutrophils. In the IC, BW and rPP2A-BW groups, fluorescence intensity did not show significant differences; by contrast, the animals treated with rPP2A-OVS registered highly significant differences (*p* \< 0.001) with respect to the other groups ([figure 5](#RSOB170031F5){ref-type="fig"}). Figure 5.Confocal microscopy studies on neutrophil cells and CCL20. (*a*) Confocal microscopy of recruitment of neutrophils (green with FITC) and CCL20 (red with Alexa-Fluor 647) in thin intestinal sections in IC and mice immunized with BW, rPP2A-BW and rPP2A-OVS. The nuclei were labelled with DAPI (blue). (*b*) Means plus s.d. of the fluorescence intensity values of neutrophils measured in an area of 2000 µm^2^. (*c*) Means plus s.d. of the fluorescence intensity values of CCL20 measured in the intestinal crypts. (*d*) Percentage of fluorescent area plus s.d. of the values of CCL20 labels measured in the intestinal crypts. Tukey--Kramer test, *p* *\<* 0.001 (\*\*\*), *p* *\<* 0.01 (\*\*) and *p* *\<* 0.05 (\*). The levels of recognition of CCL20, evaluated as the intensity of the red colour in the rPP2A-OVS group, showed significant intensity levels of fluorescence (*p* \< 0.001, \*\*\*) in comparison with the other groups, while the IC and rPP2A-BW groups were significantly increased with respect to the BW group ([figure 5](#RSOB170031F5){ref-type="fig"}). On the contrary, when the percentage of fluorescence was evaluated per surface area of the crypts, values proved significantly higher (*p* \< 0.001, \*\*\*) in the groups that included the peptide rPP2A (rPP2A-OVS and rPP2-BW) as opposed to the IC and BW groups, although the percentage of fluorescence per unit of surface area in the BW group was significant compared to that shown by IC ([figure 5](#RSOB170031F5){ref-type="fig"}). The levels of fluorescence found when studying eotaxin (CCL11) (Alexa-Fluor 633) showed ([figure 4](#RSOB170031F4){ref-type="fig"}) that the fluorescence appeared on the epithelium surrounding the crypts and the groups significantly differed from each other, especially the one treated with the lipopeptide and rPP2A, which registered the highest significance (*p* \< 0.001, \*\*\*) in the intensity of fluorescence compared with the rest of the groups ([figure 4](#RSOB170031F4){ref-type="fig"}). The fluorescence levels for OX40 (FITC) are presented in [figure 4](#RSOB170031F4){ref-type="fig"}*.* The signal appears preferentially on the brush border of the mucosa and in the epithelium that borders the crypts, and the intestines of the group rPP2A-BW had the highest expression levels, being non-significant in the other three groups ([figure 4](#RSOB170031F4){ref-type="fig"}). CD138 expression was highly specific for the plasma cells, given that this receptor was not expressed in the undifferentiated plasmablasts \[[@RSOB170031C27]\]. [Figure 4](#RSOB170031F4){ref-type="fig"} shows the location of the plasma cells CD138, near the lamina propria and inside the intestinal villi of the immunized mice. This marker indicated that the IC group significantly differed (*p* \< 0.001) in terms of the number of cells with respect to the other groups, but no differences were detected among the immunized groups ([figure 4](#RSOB170031F4){ref-type="fig"}). Regarding the count, the study of fluorescence intensity and size of Tuft cells using an antibody against Dcamkl1 and revealed with Alexa-Fluor 647 ([figure 6](#RSOB170031F6){ref-type="fig"}) showed that the number of cells per area studied (a square of 138 µm per side) did not vary between groups, nor did the average size of the cells recognized by the antibody ([figure 6](#RSOB170031F6){ref-type="fig"}). However, the fluorescence intensity per cell was highly significant (*p* \< 0.001) in the rPP2A-OVS group in comparison with the other groups. The significance of the rPP2A-BW and BW groups showed significant values with respect to the IC group. Figure 6.Confocal microscopy studies of Tuft cells. (*a*) Confocal microscopy studies of Tuft cells with anti-Dcamkl1 (red) in gut sections. (*b*) Means plus s.d. of the fluorescence intensity values of Tuft cells measured in an area of 138 µm^2^. (*c*) Number of cells (grey bars) and means plus s.d. of Tuft cells size (patterned bars). BW corresponds to control animal immunized with bacterial walls (BW); rPP2A-BW corresponds to the animal group immunized with the recombinant peptide plus bacterial walls, rPP2A-OVS is the group immunized with the self-assembling lipopeptide. Tukey--Kramer test, *p* *\<* 0.001 (\*\*\*), *p* *\<* 0.01 (\*\*) and *p* *\<* 0.05 (\*). The nuclei were labelled with DAPI (blue). The analysis of the expression levels of the different interleukins was also studied by quantitative real-time PCR. In MLN ([figure 7](#RSOB170031F7){ref-type="fig"}), a high IL-23 response was found, followed by IL-4 and IL-2, with a balanced value of IL-17 in the IC mice. Meanwhile, in the BW-intranasally immunized animals, there was an increase of G-CSF, IL-6 and IL-2 when compared with the other groups. However, in the groups immunized with rPP2A-BW, the predominant response was of IL-25 and IL-2. Also, in those immunized with rPP2A-OVS, a rise of IL-23 accompanied by IL-17, IL-9 and TNF-α was encountered. Figure 7.Determination by quantitative real-time PCR of the interleukin levels in the different immunized animal groups. (*a*) Study in mesenteric lymphatic nodules (MLN). (*b*) Study in Peyer\'s patches (PPs). The values are the means of the normalized expression values plus s.d. of these values. Tukey--Kramer test, *p* *\<* 0.001 (\*\*\*), *p* *\<* 0.01 (\*\*) and *p* *\<* 0.05 (\*). Grey bars, IC; black bars, animal groups immunized with bacterial walls (BW); red bars, animals immunized with rPP2A plus bacterial walls (rPP2A-BW); blue bars, animals immunized with the lipopeptide (rPP2A-OVS). In the PPs ([figure 7](#RSOB170031F7){ref-type="fig"}) of the animals from the IC group, a high IL-23 response was developed, followed (in order of expression) by IL-6, IL-17 and IL-21, with relatively moderate levels of IL-2, IFN-γ and G-CSF. The BW group developed a high IL-17 response accompanied by a medium level for IL-23, though this proved 3.5-fold lower than that registered in the control group. The Th2 responses such as IL-4 and IL-10 were in general very low with the exception of the rPP2A-BW group, in which the maximum response was recorded for IL-4. In addition, this group registered the maximum response for IFN-γ and intermediate for IL-23 and IL-25, followed by IL-9. However, in the rPP2A-OVS immunized mice, the response in the PPs was mainly IL-23 and IL-9, followed by IL-25, with low levels for IL-17. 3.. Discussion {#s3} ============== The results found in this study clearly demonstrate that a recombinant version of the catalytic subunit of the enzyme serine/threronine phosphatase 2A (PP2A) from the nematode *A. costaricencis* formulated as a lipopeptide conjugated with OVS can induce significant cross-protection against *T. muris* infection in mice. This confirms the high immunogenicity of the PP2A where the immunoprotective capability of the recombinant peptide was corroborated for *A. costaricensis* \[[@RSOB170031C28]\] as well as other nematodes such as *Haemonchus contortus* and *Teladorsagia circumcincta* \[[@RSOB170031C9]\]*.* The results presented reinforce the notion that conjugated lipopeptides are excellent adjuvants. Although the underlying molecular mechanism that triggers these compounds is not fully understood, it is attributed to the capability of aggregating to form micelles or particles, protecting the epitope from degradation by serum enzymes \[[@RSOB170031C29]--[@RSOB170031C31]\] or, perhaps, determining the type of antigenic presentation, among other possibilities. In an aqueous medium, the hydrophobic parts of lipids would remain inside, forming a lipophilic nucleus, while the more hydrophilic peptides would stay outside, thus permitting greater efficiency in the presentation of the APC \[[@RSOB170031C32],[@RSOB170031C33]\]. The OVS, without coupling to a protein, would bond by the high reactivity of the sulfone group to the amine groups of the proteins at the administration site, lacking activity as an adjuvant, as was determined in previous unpublished assays. The safety of the different antigen preparations administered in the assay is evident as no weight changes in the inoculated animals were observed. In terms of the effectiveness of the immunological activation against *T. muris,* the number of nematode eggs present in the faeces abruptly decreased from the seventh day after intranasal administration of the preparations containing the lipopeptide (rPP2A-OVS) or rPP2A plus the BW (rPP2A-BW). It was striking that the group treated only with BW and without the recombinant antigen had a lower average number of eggs in faeces than did control, although not as markedly as with the peptide formulae. Other authors have previously reported the positive effect of both LPS and BW as protectors against helminth infection \[[@RSOB170031C34]--[@RSOB170031C36]\]. These effects may be attributed to the immunological activation by the LPS present in the BW that act as TLR-agonists in a wide range of cell types \[[@RSOB170031C37],[@RSOB170031C38]\]. After their activation, the TLRs modulate the immune response, inducing the expression of the chemokine CCL20 \[[@RSOB170031C39],[@RSOB170031C40]\]. Intestinal epithelial cells, among others, express this chemokine in case of inflammation \[[@RSOB170031C41]--[@RSOB170031C43]\]. The evaluation of the sections of the intestine in the different groups of mice by immunohistochemistry revealed that those treated either with rPP2A-OVS or with rPP2A-BW reached high levels of fluorescence in the area studied, as well as a significant number of cells that expressed CCL20, compared with the IC and BW groups. The level of fluorescence for this chemokine in the IC group could be a consequence of the ability of *T. muris* capable to stimulate TLR4 \[[@RSOB170031C44]\]. Although *T. muris* is a parasite specific to mice, not all inbred strains of mice are susceptible \[[@RSOB170031C45]\]. After the ingestion of the embryonated eggs of the nematode, even at very low doses, acute colitis develops in all the mice strains. However, parasitism results only in the mice in which an inflammatory response was triggered, while in the hosts that developed a Th2 response (e.g. in BALB/c mice), worms were expelled approximately 20 days post-infestation \[[@RSOB170031C46]\]. In AKR/OlaHsd mice used in these experiments, infestation triggered a Th1-dependent inflammatory response in the PPs, and therefore a chronic colitis developed, indicative of the persistence of the adult worms in the large intestine due to the inability of the host to expel them 20 days post-infestation \[[@RSOB170031C47]--[@RSOB170031C49]\]. In this study, the IC mice developed high levels of inflammatory Th17-dependent interleukin expression, IL-17, IL-23 and IL-21, as well as the expression of IL-6 in PPs. However, the expression levels of IFN-γ, IL-2 and GM-CSF, though not reaching expression values of IL-6, appeared high. Nevertheless, in the MLN, a clear Th1 increase was found together with a rise in IL-2, IL-12 and IFN-γ, with an increase in IL-4 (Th2) but not IL-10 (Treg), and a less notable rise in some of the Th17 interleukins such as IL-23 and IL-17. This leads to the assumption that the Th17 inflammatory response in the PPs corresponds to a local type of Th17 response, while the one corresponding to the MLN would be a mixed Th1/Th17 systemic response, perhaps mediated by the expression of CCL20 in the intestinal epithelium cells. The first line of defence encountered by nematodes in their intestinal habitat is the mucosal barrier. The surfaces of the intestinal cells are coated by a gel that constitutes a mucus made up primarily of mucins, a series of glycoproteins that are secreted by the goblet epithelial cells, this constituting an essential component in the defence and elimination of gastrointestinal helminths \[[@RSOB170031C50]--[@RSOB170031C53]\]. Mucins are the major component of the mucus secreted by the goblet cells, but other molecules such as antibodies, defensins and lysozymes are also present, covering the entire intestinal epithelium \[[@RSOB170031C54],[@RSOB170031C55]\]. The layer of mucus keeps the intestinal surface covered with peptides that show a bactericide action ensuring its sterility, and also have an activity against infection by parasites, including nematodes, due to the presence of different bioactive factors \[[@RSOB170031C56],[@RSOB170031C57]\]. Furthermore, the physical barrier of the mucus may also interfere in both the feeding mechanisms as well as the mobility of the worms. This could explain why hyperplasia of the goblet cells occurs in nematode infections and the hyperplasia that might be induced by IL-9 \[[@RSOB170031C50],[@RSOB170031C58]\]. In our results, the highest levels of fluorescence intensity and area occupied by the mucus were found in the groups treated with lipopeptide (rPP2A-OVS) and rPP2A-BW in which worm reduction was most successful. The analysis of mucus production by WGA lectin demonstrated that these two formulae administered intranasally are capable of stimulating, by MALT, the mucus-producing goblet cells of the intestinal mucosa. As evidenced with the levels of eggs in faeces, the percentages of reduction of parasitism in the different groups ranged from 24.47% (BW treatment) to 98% (lipopeptide). The reduction after BW administration was lower than that observed for rPP2A-BW (59%), demonstrating the effectiveness of the stimulation of the response by the recombinant antigen. Also, when these BW were administered together with the recombinant antigen (rPP2A) the percentage of reduction in the number of worms would be comparable with levels found in other models of parasitic gastrointestinal nematodes of livestock when the same rPP2A-BW combination is used \[[@RSOB170031C59]\]. On the other hand, when rPP2A was linked with oleic-vinyl sulfone (rPP2A-OVS), the reduction rates of parasitism reached their highest levels, practically 98%. This group of animals registered the maximum fluorescence levels when glycoprotein production in the mucus was analysed and showed the highest expression levels of the chemokines CCL20 and CCL11, both linked to the process of cell attraction to the inflammation site and stimulation of goblet cells. In addition, it is known that the Th17 cells recruit neutrophils \[[@RSOB170031C60]\], and this could explain not only the increase in neutrophils that appear in the mice of the rPP2A-OVS group but also the high IL-17 expression levels. In addition to the goblet cells, other cells such as M-cells \[[@RSOB170031C61]\], Cup-cells \[[@RSOB170031C62]\] and Tuft cells \[[@RSOB170031C63]\] in the intestinal epithelium also regulate the immune response of the mucosa. Until very recently, the function of these cells was unknown, but they have been described as participating very actively in the immunological regulation against parasites. Tuft cells are differentiated from the rest of the mucosal cells for having a tubule-vesicular system and a plume of villi towards the luminal side, giving them the name 'Tuft'. In this group of tubules, microtubule-linked protein kinase Dclk1 is found, known as Dcamkl-1 \[[@RSOB170031C64]\], that allows these cells to be recognized by antibodies against this kinase. It is known that the number of Tuft cells increases with the presence of parasites \[[@RSOB170031C65]--[@RSOB170031C67]\]. In parasite-free animals, the number of Tuft cells of the intestinal epithelium ranges from 0.4 to 1% of the total of the epithelium cells \[[@RSOB170031C64]\]. In our experiments, the number of cells stained with anti-Dcamkl-1 did not vary among groups and neither did their size. A possible explanation is that in all the groups the infections occurred with a similar number of nematode larvae, and thus the parasitism undoubtedly boosted the number of Tuft cells, which in our case ranged between 13 and 20 in the study area (a square of 138 µm per side) for each of the different treatments. Despite the reduction in the number of worms found in the rPP2A groups, the parasite was not totally eradicated in any group. Consequently, we conclude that the decline in the number of Tuft cells could not have taken place without the absence of parasites in the intestinal mucosa \[[@RSOB170031C65]\]. However, the level of fluorescence, as indicative of the activity of the Tuft cells in the IC group, was significantly lower than in the other three groups, especially in comparison with the groups treated intranasally with the lipopeptide. This could indicate that the presence of the antigens, together with the BW or linked to the OVS, stimulated the activity of these cells. Intestinal bacteria also reportedly promote IL-25 (IL-17E) production through Tuft cells \[[@RSOB170031C68]\]. A similar response may involve the nasal mucosa, implying a stimulation of the Tuft cells mediated by the cooperation between the NALT-GALT systems by some of the stimulation factors such as chemokines. This could account for the high expression levels of IL-25 in the MLN of the mice inoculated with the preparations that carried the recombinant antigen (rPP2A-BW and rPP2A-OVS). Owyang *et al*. \[[@RSOB170031C69]\] found that IL-25 is expressed both in the caecum as well as in the MLN, and the animals resistant to *T. muris* infection present significantly higher levels in the caecum than in the MLN. The mice strain AKR/OlaHsd, susceptible to infections and used in this study, registered comparable IL-25 expression levels both in the MLN and the caecum. Tuft cells are currently considered the prime source of parasite-induced IL-25 production, and IL-25 promotes the production of IL-13 by innate lymphoid C2 (ILC2) cells. In our study, the group of mice that were vaccinated with rPP2A-BW showed higher levels of IL-25 expression in the MLN versus the expression levels in PPs. The IL-25 expression level in MLN from the mice treated with rPP2A-OVS did not correspond to the levels of IL-25 in the PPs, where the values of this group were similar to those of rPP2A-BW, although both were significantly higher than in the two other control groups (IC and BW). However, as noted above, expression levels differed from those found in the MLN, whereas IL-9 expression levels proved far higher in the PPs than in the MLN. Curiously, the group treated with the lipopeptide registered the highest expression levels of this interleukin in the PPs. It is known that T cells *in vitro* stimulated with TGF-β and IL-4 express IL-9 and produce high levels of mRNA for the IL-17RB receptor, the receptor for IL-25. In addition, the treatment of these T cells with IL-25 boosts the expression of IL-9, confirming the induction of this interleukin by IL-25 \[[@RSOB170031C70]\]. This might imply that the effect on the expulsion of the worms attributed to IL-25 could be potentially mediated by IL-9, able to stimulate the production of IL-13, as observed prior to the anti-parasite effect \[[@RSOB170031C71]\]. IL-9 is pleiotropic and, among other biological effects, reduces the expression of the claudin2 protein of the tight junction of the intestinal cells. Thus, IL-9 production by these cells could alter the function of the intestinal barrier \[[@RSOB170031C72]\], altering the intestinal flow towards the intestinal lumen \[[@RSOB170031C73]\]. Another role of this interleukin is the effect on intestinal contractibility, favouring the expulsion of the worms \[[@RSOB170031C74]\]. In summary, the results of the present study suggest that the treatment of the AKR/OlaHsd mice susceptible to infection by *T. muris*, by the nasal immunization with the recombinant peptide rPP2A as a lipopeptide bound to OVS and administered at the outset of the chronic phase of infection, is able to activate a combined Th17/Th9 response orchestrated by the cytokines IL-25, IL-17 and IL-9, restraining egg release by intestinal worms and resulting in accelerated worm expulsion. This strategy of immunization could be of great applicability not only in immunotherapy and immunoprophylaxis to control diseases caused by nematode parasites of the intestinal mucosa but also in those pathologies in which it is necessary to act at the level of the Th9 response in the mucosa. 4.. Material and methods {#s4} ======================== 4.1.. Antigen {#s4a} ------------- We used a recombinant peptide expressed by a CT2--2 clone corresponding to the PP2A catalytic region of *A. costaricensis* \[[@RSOB170031C28]\] as the antigen. Recombinant production was previously described by Fawzi *et al*. \[[@RSOB170031C9]\] and briefly described below. The protein expression in the recombinant bacteria was induced with 0.5 mM IPTG (isopropyl-β-[d]{.smallcaps}-thiogalactopyranoside). The recombinant protein was purified by affinity chromatography with HisTrap FF Crude column (GE Healthcare Life Sciences, 11-0004-58), previously equilibrated with binding buffer (6 M guanidine-HCl, 20 mM sodium phosphate, 500 mM NaCl, 5 mM imidazole, 1 mM β-mercaptoethanol, pH 7.4). The sample was loaded, and the column was washed with binding buffer and a gradient of imidazole from 20 to 60 mM. The recombinant protein was eluted with elution buffer (6 M guanidine-HCl, 20 mM sodium phosphate, 500 mM NaCl, 500 mM imidazole, 1 mM β-mercaptoethanol, pH 7.4). After dialysis, the protein concentration was determined by the Bradford method in denaturizing buffer (6 M guanidine-HCl). 4.2.. SDS-PAGE electrophoresis and western-blot analysis {#s4b} -------------------------------------------------------- A defined amount of 20 µg per lane of protein was subjected to 12.5% SDS-PAGE and stained with Coomassie Brilliant Blue. Unstained replicas were transferred to 0.2 µm PVDF membranes (Bio-Rad, 170-4156) using a Bio-Rad Trans-Blot Turbo TM (0.6 Å, 25 V, 40 min). Western blot analyses were carried out using sera obtained from mice immunized with PP2A \[[@RSOB170031C9]\] and diluted 1 : 150 with PBS-T (0.3% tween-20 in PBS). Membranes were washed with phosphate-buffered saline with 0.05% Tween 20 (PBS-T), incubated for 2 h at 37°C with 1 : 1000 polyclonal Goat anti-Mouse Immunoglobulins/HRP (Dako Cytomation-P0447) and developed using chemiluminescence (ClarityTM Western ECL Substrate, Bio-Rad, 170-5060). The protein band from the SDS-PAGE electrophoresis recognized by the sera was sequenced by fingerprinting and identified at the Servicio de Proteómica del Centro de Biología Molecular Severo Ochoa (CBMSO) in Madrid (Spain) in an Autoflex matrix-assisted laser-desorption ionization time-of-flight (MALDI-TOF) mass spectrometer (Bruker) equipped with a reflector following a previously described protocol \[[@RSOB170031C75]\]. 4.3.. Synthesis of alkyl vinyl sulfone {#s4c} -------------------------------------- The synthesis of (9Z)-*N*-\[2-(ethenylsulfonyl)ethoxy\]-ethyl\]-9-octadecenamide (OVS) has been previously described in detail \[[@RSOB170031C76]\] and is summarized in [figure 1](#RSOB170031F1){ref-type="fig"}. 4.4.. Coupling of the alkyl vinyl sulfone to rPP2A and mass determination {#s4d} ------------------------------------------------------------------------- The synthesized alkyl vinyl sulfone was solubilized in methanol and mixed with 0.125 mM carbonate buffer (pH 8.3) to a final concentration of 2 mg ml^−1^. The solution of the alkyl vinyl sulfone was mixed with the purified rPP2A (2 mg ml^−1^) in carbonate buffer (0.125 mM carbonate, pH 8.3) and kept at 4°C in an orbital shaker for 12 h. Thereafter, free reactive groups were blocked with a molar excess of glycine in carbonate buffer at room temperature for 4 h. The conjugates were dialysed for 12 h against a solution of ammonium acetate (0.1 M, pH 7) and then lyophilized. Mass spectrometer analyses were performed at Laboratorio de Espectrometría de Masas (SIdI), Faculty of Sciences, UAM, Madrid (Spain) in a Ultraflex III TOF/TOF instrument (Bruker) that uses an NdYAG laser (emission, 355 nm; accelerating voltage, 25 kV). The sample (0.5 mg of lipopeptide in 0.2 ml of 1,1,1,3,3,3 hexafluoruro-2-propanol) and matrix solution (10 mg of α-ciano-4-hydroxycinnamic acid (ACC) in 1 ml of acetonitrile, 0.1% TFA in water 3 : 1 v/v), were mixed in a proportion 1 : 20, and then applied to a metal sample plate for MS analysis. The lipopeptide mass was measured at approximately 1000 Da absorption laser intensity. 4.5.. Electron microscopy {#s4e} ------------------------- Electron microscopy was used to check whether the alkyl functionalized peptides were aggregated in micelles and to estimate their size. Hence, the lipopeptide was sonicated in a 0.1 M ammonium acetate solution and visualized in a transmission electron microscope (TEM) (Libra 120 PLUS de Carl Zeiss SMT) by depositing 5 µl of the suspension adsorbed directly onto a 300 mesh Cu grid covered with Formvar and stained with 1% uranyl acetate for 1 min \[[@RSOB170031C77]\]. To determine the size of the micelles, 100 of the observed particles were measured digitally using I[mage]{.smallcaps}J software (v. 1.48 e, Wayne Rasband, htpp:\\\\imagej\\nhi\\gov\\imagej). 4.6.. Animal experiments {#s4f} ------------------------ The inbred AKR/OlaHsd mouse strain was used in all experiments. This mouse strain is highly susceptible to *T. muris* infection and, following a primary infection, it allows the progression to a chronic stage with persistence of fecund adult worms due to activation of an inappropriate Th1-type response \[[@RSOB170031C78]\]. A total of 36 inbred female AKR/OlaHsd, aged six to eight weeks, were purchased from Harlan Laboratories UK Ltd (Blackthorn, Bicester, UK) and maintained under specific pathogen-free conditions at the VISAVET animal house facilities at Complutense University of Madrid with ad libitum access to food and water, regulated temperature and light/dark cycle conditions. For the experiments, the mice were reared in six randomly assigned groups and placed in the corresponding standard methacrylate cages for the duration of the experiments. All mice were weighed weekly along the time of the experiments. 4.7.. Parasites {#s4g} --------------- The Edinburgh strain of *T. muris* (isolated in 1961 \[[@RSOB170031C79]\] and since then maintained by periodical passage in outbred mice) was used in this study. The methods for the maintenance and infection were the same as previously described \[[@RSOB170031C80]\]. Each mouse was infected with approximately 200 embryonated eggs by oral gavage on day 0. 4.8.. Assessment of infection and immune response {#s4h} ------------------------------------------------- For an assessment of the level of infection as well as the local and systemic immune response elicited against infection, the mice were killed by an overdose with isoflurane, placed under aseptic conditions, and their large intestines removed and longitudinally opened. Intestinal L-3 larvae were released and collected as previously described \[[@RSOB170031C80]\], and then counted under a stereomicroscope. MLN and PPs samples were also taken, and placed in RNAlater buffer for further cytokine measurement. Likewise, the ending tip of the caecum from each animal was removed and placed in a solution of 0.5% glutaraldehyde and 2.5% paraformaldehyde buffered at pH 7.4 for further immunohistochemical studies. 4.9.. Immunization schedule and follow-up {#s4i} ----------------------------------------- In this study, the immunization schedule was designed on the basis of the demonstrated susceptible phenotype of the AKR/OlaHsd mouse strain regarding *T. muris* infection, thus allowing the establishment of a chronic infection as reported above \[[@RSOB170031C78]\]. Our rationale was to check whether the predominant Th1 response elicited during the first three weeks of infection leading to susceptibility could be reversed to resistance by applying a recombinant vaccine throughout the chronic stage of the infection able to trigger a predominant Th2 response and thus leading to expulsion of the established mature worms. For this, on day 49 p.i. the animals received the first immunization dose. Three freeze-dried vaccines---bacterial walls (BW) and free rPP2A protein plus (BW) at equivalent amounts to those given separately to the corresponding groups and rPP2A coupled to a lipid OVS---were suspended in 1% carboxymethylcellulose to reach a concentration of 0.8 µg ml^−1^. Four micrograms of each preparation were administered to mice by nasal instillation of 5 µl of the antigenic suspension applied into the right nostril using a micropipette coupled to a special fine long tip. An additional infected control (IC) group was administered with PBS (0.25 M plus the 1% carboxymethylcellulose). Boosters were applied two weeks later to all groups in the same way as the first administration vaccine dose. To monitor the effectiveness of the vaccine, faecal egg counts were performed weekly by the Stoll method, from the day of the first vaccination until the day of killing the mice. Briefly, the faeces released by each animal temporary isolated for 8 h in an individual cage were collected and homogenized in 0.1% NaOH (4 g/60 ml 0.1 N NaOH rate). Six aliquot fractions of the suspended material were placed on glass slide, and eggs were counted under an optical microscope at 4× magnification. The results were expressed as the number of eggs/gram of faecal samples. Sixty-three days after the first vaccination all animals were killed and processed for adult intestinal-worm collection and counting, as well as for immunochemical studies. 4.10.. Isolation of mRNAs, synthesis of cDNAs and quantitative real-time PCR {#s4j} ---------------------------------------------------------------------------- Total RNA was isolated from MLN and PPs preserved in RNAlater, an RNA stabilization reagent (Qiagen-76104) using the RNeasy Midi kit (Qiagen-74106). Once the RNA was isolated, the sample was digested with RNase-Free DNase Set (Qiagen-79254), in order to remove DNA contamination. For mRNA isolation, the Oligotex mRNA Midi kit (Qiagen-70042) was used. The quality of the purified mRNA was measured in an Experion automated electrophoresis system (Bio-Rad, Nazareth Eke, Belgium). Subsequently, cDNA was generated with the iScript cDNA Synthesis kit (Bio-Rad, 170-8891) in a CFX96 real-time system (Bio-Rad). Thus, a 20 µL final volume reaction containing 4 µl of 5× iScript reaction mix, 1 µl de iScript reverse transcriptase, and an mRNA volume between 100 fg and 1 µg, was performed. The retrotranscription conditions were 5 min at 25°C, 30 min at 42°C, 5 min at 85°C at the end the sampling, and conditions were maintained at 12°C. The concentration and quality of the cDNA was calculated spectrophotometrically in a Nanodrop (Nanodrop ND-1000, Thermo Scientific). The cDNA was diluted 1 : 10 and preserved at −80°C until use. Quantitative real-time PCR was performed on a thermocycler (CFX96 Real-Time System, Bio-Rad) in combination with primers specifically designated using the eprimer3 software (<http://www.bioinformatics.nl/cgi-bin/emboss/eprimer3>) and shown in electronic supplementary material, table S1. The cDNA was quantified using SsofastTM Eva Green Supermix (Bio-Rad, 172-5201), with cDNA equivalent to 50 ng mRNA. The cycling conditions were as follows: 95°C for 2 min, followed by 40 cycles of 95°C for 10 s, 55°C for 30 s and 60°C for 30 s. When the amplification was completed the samples were kept at 12°C. The primer concentration was optimized and dissociation curves were generated for each primer set to verify the amplification of a single PCR product. Expression of β-actine RNA \[[@RSOB170031C81]--[@RSOB170031C83]\] was used to normalize the expression of other genes quantified according to the ΔCT method, in which the β- actin: Interleukin ratio = 2 CT β-actin − CT interleukin. All of the assays were done in triplicate. 4.11.. Immunohistochemistry of mouse tissues {#s4k} -------------------------------------------- For histological evaluation, the glutaldehyde/paraformaldehyde-fixed tissues were thawed in PBS at laboratory temperature under permanent automatic rotation of the sample tube. Then, the samples were trimmed and embedded in paraffin wax. Sections of 8--10 µm were affixed to slides. The paraffin was removed by three dips of 15 min in xylene and an alcohol series (100%, 90%, 70%), and one rinse in water \[[@RSOB170031C84]\]. After paraffin removal and hydration, the slides were prepared for antigen retrieval by heating at 120°C for 10 min in an autoclave with 0.01 M citric acid at pH 6.0 \[[@RSOB170031C85],[@RSOB170031C86]\]. After the unmasking treatment of the antigens, as a means to prevent the non-specific binding of the antibodies, a treatment was made for 30 min with PBS containing 1% of albumin from chicken egg white (Sigma, A5503). The slides were treated with different specific antibodies, as shown in electronic supplementary material, table S2, at a concentration of 1 : 50 for 1 h at room temperature. Finally, cell nuclei were stained with a dilution 1 : 100 of 4′,6-Diamidino-2-phenylindole dihydrochloride (DAPI) (Sigma, D9542) for 10 min. They were subsequently preserved and mounted in a mounting medium (Prolong Antifade Kit, Molecular Probes) and examined under a Leica DMI6000 confocal laser microscope equipped with a filter system for FITC (mean wavelength 530 nm, maximum 490 nm). Electronic supplementary material, table S2 shows the origin and labelled antibodies and reagents for the immunocytochemistry by confocal studies. 4.12.. Statistical analysis {#s4l} --------------------------- The Tukey--Kramer multiple-comparisons test was used to estimate the significance of the difference between means. The results are expressed as mean values (standard deviation of the different groups at different times for each experiment performed). *p* \< 0.001 was considered to be highly significant (\*\*\*), *p* \< 0.01 was considered moderately significant (\*\*) and *p* \< 0.05 was considered a low level of significance (\*). GraphPad I[nstat]{.smallcaps} v. 3.05 software (GraphPad Software, Inc., La Jolla, CA, USA) was used for the statistical analysis. Supplementary Material ====================== ###### Table S1 from Self-adjuvanting C18 Lipid Vinylsulfone-PP2A vaccine: study of the induced immunomodulation against Trichuris muris infection by M. Gomez-Samblas, JJ. García-Rodríguez, M Trelis, D. Bernal, FJ Lopez-Jaramillo, F Santoyo-Gonzalez, S. Vilchez, AM Espino, F. Bolás-Fernández, A. Osuna; Table S2 from Self-adjuvanting C18 Lipid Vinylsulfone-PP2A vaccine: study of the induced immunomodulation against Trichuris muris infection by M. Gomez-Samblas, JJ. García-Rodríguez, M Trelis, D. Bernal, FJ Lopez-Jaramillo, F Santoyo-Gonzalez, S. Vilchez, AM Espino, F. Bolás-Fernández, A. Osuna Ethics {#s5} ====== All animal studies conformed to current Spanish legislation for animal welfare. The Ethics Commission at Complutense University of Madrid approved this study. Data accessibility {#s6} ================== The datasets supporting this article have been uploaded as part of the electronic supplementary material. Competing interests {#s7} =================== We declare we have no competing interests. Funding {#s8} ======= No funding has been received for this article. [^1]: Electronic supplementary material is available online at <https://dx.doi.org/10.6084/m9.figshare.c.3723946>.

INTRODUCTION ============ In multiple sequence alignments, information emerges as to which residues positions are important to the nature of a protein family, versus which vary freely but could be assigned undue importance during the scoring of pairwise alignments. From these multiple alignments, profile hidden Markov Models (HMMs) are built. These probabilistic models allow exquisitely sensitive searches for proteins related by homology to the aligned sequences. The TIGRFAMs database is a collection of these HMMs constructed with the purpose of letting automated annotation pipelines attach specific functional annotations to proteins encoded by newly sequenced microbial genomes. The HMM search produces evidence, and the logic of the annotation software exploits the evidence. But the HMM evidence itself is persistent, and based on fixed cutoff scores for consistency from use to use, and may be put to additional purposes. Genome Properties is a collection of rules for interpreting evidence, most in the form of HMM hits, to make judgments about the likely presence or absence of complex biological traits in an organism according to whether or not its genome encodes the components necessary for that trait. The process of systems reconstruction in Genome Properties provides guidance for protein family construction in TIGRFAMs, so the two databases develop in concert. TIGRFAMs ======== TIGRFAMs as an annotation-driving database ------------------------------------------ Publications that report specific protein characterization data typically describe one or two proven examples of the protein in question but stop short of providing any rule for recognizing all examples from all organisms. Subsequent biocuration at either specialized or comprehensive value-added databases ties articles to sequences, attaches meaningful functional names, and adds feature tables (signal peptides, active sites, modified sites, etc.) and other searchable content such as EC numbers and GO terms. But for only limited numbers of protein families has this curation matured into tools sufficient to identify and annotate all functionally equivalent sequences. Most proteins belong to large superfamilies whose memberships are heterogeneous in function, complicating such efforts. The right granularity for dividing a superfamily into subgroups by function seems to differ from case to case, and no one set of heuristics always works. New tools are needed. TIGRFAMs is built to address this need by serving as an *annotation-driving database*, a library of protein family definitions that becomes a tool set used by automated genome analysis pipelines. Each TIGRFAM as a proxy for a curatorial expert in that protein family ---------------------------------------------------------------------- The *TIGRFAMs* database provides manually reviewed definitions of protein families with the following characteristics to make them useful for automated genome annotation, pathway reconstruction, creation of phylogenetic profiles and any number of other computationally driven studies. First, all annotation is traced to its original source without ever relying on transitive annotations from public sequence databases. Second, every protein is considered in the context of any related protein families that differ in function, if they exist, with examination of molecular phylogenetic trees. Third, only sequences that can be assigned with high confidence are selected as exemplars for the seed alignment. Multiple sequence alignments are examined for misalignments, inconsistent domain architecture, altered active or binding sites, faulty gene models, long branches in phylogenetic trees that may suggest neofunctionalization, etc. Biocuration during model construction includes review of local synteny, metabolic context and phenotypic data. Fourth, each model is searched against several protein databases, including the CharProtDB collection of explicitly characterized proteins ([@gks1234-B1]), sequences with known structures in PDB and NCBI's non-redundant protein sequence collection, which pulls annotation from multiple sources, including value-added databases such as UniProt and archives of sequences as originally submitted. Each completed HMM is intended to serve as a proxy for an expert curator, emulating expertise developed at the time of model construction but operating at BLAST-like speeds ([@gks1234-B2]). TIGRFAMs models describe the level of their specificity ------------------------------------------------------- Each entry in TIGRFAMs carries a designation that describes how the set of proteins in the family vary in function. If all members of a protein family perform the same function, the family is designated *equivalog*. We previously introduced the term '*equivalog*' to describe a relationship of conserved function among homologs ([@gks1234-B3]), noting that the term '*ortholog*' is doubly wrong since orthology does not imply conserved function and laterally transferred genes (*xenologs*) may be equivalogs. The isology type informs automated annotation pipelines which of two different HMMs that matches the same region of the same protein gets priority. Each TIGRFAMs *equivalog* model provides a high-precedence instruction to automated annotation pipelines to transfer the protein name, enzyme commission (EC) number if present, gene symbol and Gene Ontology (GO) terms to any protein whose match to the HMM meets the score cutoff. An *equivalog* model assigns more specific annotations than a *subfamily* model, which in turn outranks a *domain* model. The general principle is that models built with narrower scope tend to outrank models with broader scope where they hit the same proteins. In many cases, a *subfamily* model is created in TIGRFAMs, with deliberately generic annotation attached, to prevent automated annotation pipelines from propagating an overly specific annotation from a more distant homolog according to evidence of lower rank. TIGRFAMs content complements the Pfam collection by emphasizing protein function over domain architecture --------------------------------------------------------------------------------------------------------- The TIGRFAMs database is designed to be used in conjunction with other sources of annotation, especially Pfam ([@gks1234-B4]). Both databases contain HMMs built with and searchable by the same freely available software package HMMER3, developed by Eddy ([@gks1234-B2]). Pfam models frequently describe domains of homologous sequence that occur as finite regions within larger proteins and are shared across sets of proteins that range widely in function ([@gks1234-B4]). A single protein often has multiple Pfam domains. A TIGRFAMs *equivalog* model, by contrast, typically identifies fewer proteins, covers a larger fraction of total sequence length, and provides more specific annotation that should be given higher precedence by automated annotation pipelines. For example, Pfam model PF04055 describes the radical SAM domain, found in over 40 000 recognizable member sequences in public databases. Many of these have additional domains, for a wide variety of domain architectures and even wider array of functions, most of which remain unknown. TIGRFAMs describes over 100 functionally distinct protein families that share the radical SAM domain, including methyltransferases that modify structural RNAs, cofactor biosynthesis enzymes performing complex rearrangements, lipid metabolism enzymes, peptide maturases for natural products biosynthesis and enzyme activases. These models resolve many subgroups of the radical SAM superfamily by function in a way that the domain decomposition provided by Pfam cannot. New TIGRFAMs work avoids construction of models that duplicate the scope and extent of existing Pfam models, but it will include construction of domain or repeat models for homology regions that have never before been described. TIGRFAMs has grown by over 40% since the previous database update paper ([@gks1234-B5]), from 3000 to 4284 models. Most new models describe less common proteins, so coverage by TIGRFAMs for the average microbial genome has increased only about 20%. Overall coverage of microbial genomes is variable, averaging about 33%. [Table 1](#gks1234-T1){ref-type="table"} shows illustrative levels of coverage by TIGRFAMs for five bacterial and two archaeal genomes. The table distinguishes hits in aggregate by all TIGRFAMs models from hits just to equivalog-type HMM hits, where the HMM matches spans nearly the full protein length and members of that family should share just one function. As expected, TIGRFAMs identifies more proteins in absolute terms in larger genomes, but matches a greater percentage of proteins for species with smaller genomes or from more intensively studied lineages. The current release, 13.0 (July 2012), is the fourth release to use HMMER3 instead of HMMER2. A web server page kindly provided by Janelia Farm, <http://hmmer.janelia.org/search/hmmscan>, provides high-speed searching of HMM libraries ([@gks1234-B6]), with links provided to TIGRFAMs pages for models meet the criterion of scoring better than the trusted cutoff. The TIGRFAMs home pages do not provide searching, but instead provide a link to the Janelia Farm hmmscan page. Table 1.TIGRFAMs HMM hit coverage for five bacterial and two archaeal genomesSpeciesNo. of proteinsNo. of matched, (%)No. of equivalog-level, (%)*Mycoplasma genitalium* G-37473260 (55)220 (47)*Streptococcus sp.* SK1401856672 (36)489 (26)*Haemophilus parainfluenzae* HK2622001935 (47)730 (36)*Burkholderia mallei* ATCC 2334450311379 (27)1009 (20)*Fusobacterium necrophorum* ATCC 513572060680 (33)506 (25)*Methanocaldococcus jannaschii* DSM 26611783584 (33)462 (26)*Haloferax volcanii* DS24074699 (17)415 (10) A project goal is to make negative results from searches informative -------------------------------------------------------------------- The goal during construction of each TIGRFAMs entry is to create a HMM with cutoff scores calibrated to collect comprehensively the set of all proteins that merit the annotations provided with the model, with vanishingly few false-positives but also very few false-negatives (other than proteins left incomplete through sequencing or assembly errors). The advantage of few false negatives becomes clear during the analysis of complete genomes. If six key protein families represent six essential steps in some biochemical pathway, and the six corresponding TIGRFAMs entries all report no match during the analysis of a complete microbial genome, then the repeated absence of evidence for that cohort of genes becomes fairly strong evidence that the cohort is in fact absent. TIGRFAMs may be used through representations in other resources --------------------------------------------------------------- TIGRFAMs is incorporated into InterPro ([@gks1234-B7]), a hierarchical collection composed of entries from multiple databases of protein family signatures. InterPro distributes an integrated software package that performs searches for all member databases at once, including TIGRFAMs. Similarly, the Conserved Domain Database (CDD) ([@gks1234-B8]) incorporates TIGRFAMs models, so domain searches through NCBI may show corresponding matches. Note, however, that CDD first converts TIGRFAMs models from HMMER3 to RPS-BLAST models, so the calibration of cutoff scores usually is lost in favor of simpler heuristics. A CDD entry built from a TIGRFAMs model, e.g. CDD:188483 derived from TIGR03968 ('mycofactocin system transcriptional regulator'), may attach a region feature to many proteins that we were careful to exclude when our model was constructed. A region feature showing TIGRFAMs information through a database cross-reference (db_xref) to a CDD entry is showing sequence homology that may be useful know, but true membership in the TIGRFAMs family should be verified based on actual HMM search scores. TIGRFAMs web pages ------------------ Web pages for TIGRFAMs was provided originally through the Comprehensive Microbial Resource ([@gks1234-B9],[@gks1234-B10]), or CMR. TIGRFAMs switched to an independent home page at [www.jcvi.org/tigrfams](www.jcvi.org/tigrfams) during the conversion from using HMMER2 to using HMMER3. Legacy data for TIGRFAMs releases 9.0 and earlier, and Genome Properties results relying on HMMER2-generated data for over 550 genomes, remain accessible through the CMR. [Figure 1](#gks1234-F1){ref-type="fig"} provides an illustration of a summary page for a single TIGRFAMs entry, TIGR04071, representing a family of precursor peptides that are post-translationally modified to become methanobactin Mb-OB3b (a copper chelator analogous to iron-chelating siderophores) or another member of the Mb-OB3b class ([@gks1234-B11]). Figure 1.Example of a TIGRFAMs web page. The TIGRFAMs homepage and all its web pages have five links located on the left navigational sidebar. These go to the TIGRFAMs Home page itself, a glossary of key terms used in TIGRFAMs, a complete listing of all models (4284 in release 13.0), the TIGRFAMs FTP Site, and a page with links to additional resources, including rapid searching of a protein sequence against HMM libraries. The page shown in the figure is summary page for a single TIGRFAMs entry, the Mb-OB3b-class methanobactin precursor family TIGR04071. Each summary page provides details about a single TIGRFAM, including cutoffs, HMM length, references and any assigned GO terms. The summary page also displays links to Genome Properties, if any, associated with that HMM. Users can view and download the HMM seed alignment through the HMM Seed Alignment link found on the left-hand side of the HMM Summary page. The TIGRFAMs complete listing page (not shown) displays accessions and functional names for all models, as well as length, isology type and EC number. GENOME PROPERTIES ================= Providing high-level views of attributes inferred from genome sequences ----------------------------------------------------------------------- Genome Properties is a collection of definitions for the higher-level attributes that may be ascribed to a species when a sufficient set of molecular markers are detected in its genome, or else reported jointly absent ([@gks1234-B12]). If all enzymes listed as essential markers of biotin biosynthesis are detected (by HMMs in the TIGRFAMs database), then the Genome Property 'biotin biosynthesis' is set by rule to 'YES'. Assertions made by Genome Properties are useful to summarize high-level traits of species biology from genome analysis, to understand metabolic context while trying to understand the roles of other proteins from the same species, and for comparative genomics based on the whole biological processes rather than single genes. Genome Properties entries describe subsystems --------------------------------------------- The linkage between TIGRFAMs and Genome Properties is paralleled in a kindred effort, the SEED from the Fellowship for the Interpretation of Genomes (FIG) ([@gks1234-B13]). FIG selected the term 'subsystem' to describe the biological role in aggregate that a set of marker proteins working in concert enable. We support usage of that term, using 'subsystem' here to refer in general to any emergent property in species biology that is understood more clearly by viewing the collection of component genes together rather than separately. We use the term 'Genome Property' for any subsystem described as an entry in the Genome Properties database. Metabolic reconstruction is based on evidence rather than annotation -------------------------------------------------------------------- Many sequences in public protein sequence databases are annotated incorrectly as biotin synthase (EC 2.8.1.6), to give one example, through errors of transitive annotation, in part because biotin synthase was one of the first radical SAM superfamily ([@gks1234-B14]) members sequenced, characterized and named ([@gks1234-B15]). In fact, misannotations that attach overly specific functions are a particularly abundant class of error ([@gks1234-B16]). Any metabolic reconstruction system that takes annotations at face value runs a risk of going badly astray. Genome Properties follows a principle that whether or not some genome encodes a subsystem should be determined by evidence sufficient to drive annotation, not by pre-existing annotations of uncertain provenance. Following this principle means that creating a new Genome Property often necessitates building new entries for TIGRFAMs database of HMMs, or else confirming that existing models in either TIGRFAMs or Pfam behave suitably. Genome Properties emphasizes non-pathway subsystems --------------------------------------------------- Genome Properties includes simple biochemical pathways, of course, but describes many additional types of subsystem. When Genome Properties describes a subsystem whose core is an enzymatic pathway, it often represents non-enzymatic proteins such as transcriptional regulators and molecular chaperones as additional components. Computation of a Genome Property may depend on a protein, a feature within a protein (such as a sorting signal), or a genetic element that does not even code for a protein, such as an array of CRISPR repeats or the selenocysteine tRNA. It describes physical complexes such as transporters and flagella. It describes systems in which one protein acts upon another, including secretion systems, protein-sorting systems, and post-translational modification systems. Methanobactins are generating considerable interest because neither their diversity nor their biosynthesis is well understood yet ([@gks1234-B17]). The very small size of the only known class of precursor peptide has meant missed gene calls (at least initially) in every species so far with an example of the system, originally just *Azospirillum* sp. B510 and *Methylosinus trichosporium* OB3b ([@gks1234-B11]), but now including *Gluconacetobacter oboediens* 174Bp2, *Gluconacetobacter* sp. SXCC-1, *Tistrella mobilis* KA081020-065, *Pseudomonas extremaustralis* and *Methylocystis* sp. SC2. Model TIGR04071 for the methanobactin precursor family belongs to a Genome Property, GenProp0962, which explains that the gene should occur in the context of two other, larger molecular markers, members of families TIGR04159 and TIGR04160. These two companion proteins are much easier to find when prospecting in newly sequenced genomes for new examples of methanobactin-like natural products. The architecture of Genome Properties ------------------------------------- Each Genome Property is described completely by records in a series of tables stored in a relational database. The tables are now made available for downloading by ftp through the TIGRFAMs/Genome Properties web pages, at [ftp.jcvi.org/pub/data/TIGRFAMs/GEN_PROP](ftp.jcvi.org/pub/data/TIGRFAMs/GEN_PROP). The table structure and meanings of all fields are described in the release notes. The logic of the table structure is discussed below. Each Genome Property has a basic definition in the *prop_def* table that includes a name (e.g. 'urease'), a unique accession (e.g. 'GenProp0051'), a paragraph-long description, and some ancillary information. Because the first subsystems described in Genome Properties were simple enzymatic pathways, components are referred to in relational database tables as 'steps.' Thus, the components that belong to a given Genome Property are enumerated in the *prop_step* table. Note that one enzymatic activity, a single entity in the typical pathway definition, will be represented by multiple components if the enzyme is comprised of multiple subunits. A Genome Property is judged complete, and may be assigned the state 'YES', if every component listed as required is found. Nearly every Genome Property has two or more components, largely because the comparative genomics of subsystem reconstruction is often essential for creating and establishing trust in the protein family definitions that a Genome Property requires. Property definitions include genes that are not necessarily essential --------------------------------------------------------------------- Some protein families occur exclusively as part of some subsystem, yet comparative genomics shows they are not always present, and not required for all instances of the subsystem to operate (although they may be required in some cases). Such a protein may be listed as a part of the system by entry into the *prop_step* table, but is marked as non-essential, meaning not core to the definition of the Genome Property and not used to compute its completeness. A protein that is absolutely required, but for which no reliable detection tool is available, similarly must be marked as non-essential to prevent the scoring system from calling the subsystem incomplete. Models may be unavailable because an activity has not yet been matched to any sequence, or a single known sequence example is not easily extended into a whole protein family definition, or the role tends to be filled by members of different proteins in different species (as is common for phosphatases and aminotransferases), or the function may be hard to assign based on full-length homology rather than select specificity-determining residues (as seems to occur with transporters). Genome Properties allows multiple lines of evidence for each step ----------------------------------------------------------------- Genome Properties defines types of evidence that will be treated as sufficient to show that a required component is encoded within a genome. In most cases, the evidence is a TIGRFAMs or Pfam HMM that scores above the model's cutoffs to some protein. Trusted cutoffs are used rather than gathering thresholds (these differ only for Pfam). For a given canonical protein function (e.g. glutamate--cysteine ligase, EC 6.3.2.2), several different known families may be known, and assignment to any one of these may constitute evidence. Thus, a separate linking table, *step_ev_link*, defines what evidence is sufficient to satisfy a step. In some cases, Genome Properties requires that there be at least one member encoded in a genome of some family of proteins, without implying that all members found from that family necessarily participate in the Genome Property in question. However, recording sets of such proteins during evaluations of Genome Properties across large numbers of genomes, and distinguishing those found in genomes encoding the other candidate markers from the rest supports data mining approaches that may lead to the construction of new protein families. Genome Properties allows an evidence type designated HMM-CLUST, meaning that a protein must be a member of the designated protein family but also within 3000 base pairs of another marker of the same system. The HMM-CLUST method may identify, for example, members of the radical SAM domain family (PF04055 in Pfam) found in close proximity to a precursor gene for post-translationally modified peptide. This co-clustering may mark a subfamily or equivalog group within radical SAM, which can then be ascribed a role in peptide modification. Such approaches let Genome Properties computation support both protein family development and discovery of new types of subsystems. Thresholds ---------- In the ideal case, a Genome Property has complete evidence, or no evidence, in nearly every genome. If most components are found, but not all, the YES-leaning state 'some evidence' may be assigned. But some proteins can play a role in any of several different properties. Finding such a marker, but no other, for a given property in question does not suggest the property is actually present. The enzyme selenide, water dikinase (SelD, TIGRFAMs entry TIGR00476), for example, is essential to at least three traits ([@gks1234-B18]): selenocysteine incorporation (GenProp0016), the selenouridine tRNA modification (GenProp0692) and post-translational activation of selenium-dependent molybdenum hydroxylases (GenProp0726). For any one of these systems, finding SelD only is very weak evidence. Similarly, a Genome Property may rely in part on an HMM that was available (perhaps from Pfam) but that hits a number of homologs beyond the set that actually carry the function of interest. Absence of any member of that family from a genome would be informative, so it is useful to require a hit as an additional constraint for recognizing a subsystem to be present. But finding a member of that family as the only evidence would be very weak evidence the entire subsystem is encoded. For these reasons, each Genome Property has a threshold value. If the number of components found does not exceed the threshold, evidence that the Genome Property on the whole is encoded by a genome should be considered weak, and the state 'not supported' will be assigned instead of 'some evidence.' Genome Properties is hierarchical --------------------------------- A component required for a Genome Property to be complete may itself be a Genome Property. Urea utilization (GenProp0814), for example, consists of a urea uptake system and a urea degradation pathway. For urea degradation, either of two pathways is sufficient, urease (GenProp0051) or the urea carboxylase/allophanate hydrolase pathway (GenProp0481). Whichever of the two is the more complete is used to score the urea utilization property. Genome Properties and TIGRFAMs usually are constructed in concert ----------------------------------------------------------------- Where a set of proteins cooperate to form an enzymatic pathway, or some other type of subsystem with a fixed set of required components, each successfully completed protein family definition gives contextual clues that help identify trusted exemplars for the remaining protein families. Some protein families are straightforward to construct because essentially every detectable homolog appears to perform the identical function. The very clear boundaries to such families provide information that can guide construction of additional protein families. It appears, for example, that every detectable homolog of the first described PqqA, a peptide whose role is to serve as the precursor of pyrroloquinoline quinone (PQQ), likewise serves as a PQQ precursor peptide. In contrast to PqqA, the PQQ biosynthesis enzyme PqqE belongs to radical SAM, a family so abundant that 1 genome may encode over 30 members, each different in function. Correctly separating all true PqqE from their functionally distinct homologs would be difficult except that the PqqA model (TIGR02107) assured that the PqqE model (TIGR02109) would be constructed with no false-positives among its seed members. The PqqE model, in turn, identifies all PQQ biosynthesis systems where the small (∼23 amino acid) PqqA peptide was missed because of faulty gene calling. Genome Properties made available for the current release, designated 3.0, number 628, a marked expansion over the ∼200 Genome Properties released at the time of the last published database description. The current total includes new subsystem definitions plus previously unreleased ones likely to benefit from additional development. GENOME PROPERTIES WEB PAGES =========================== Like TIGRFAMs, Genome Properties was originally hosted through the CMR ([@gks1234-B9],[@gks1234-B10]). At last update, the CMR contained over 500 microbial genomes. Genomes in the CMR show results from evaluating Genome Properties as implemented based on HMMER2 models, using TIGRFAMs models up through release 9.0. Genome Properties is now hosted independently of the CMR. A sample page, describing Genome Property GenProp0962 ('methanobactin biosynthesis, Mb-OB3b family'), is shown in [Figure 2](#gks1234-F2){ref-type="fig"}. The new Genome Properties pages provide a link to the legacy version of Genome Properties in the CMR, giving access to computed Genome Properties results in a comparative genomics browser environment. Figure 2.Example of a Genome Properties Definition Page. The Genome Properties home page provides three links on the left navigation sidebar: Genome Properties List, Top Level Genome Properties and Search CMR Genome Properties. The Genome Properties List shows all current Genome Properties along with their property type and property name. Clicking on a Genome Property will display the Genome Property Definition page. The Definition Page for GenProp0962 ('methanobactin biosynthesis, Mb-OB3b family') is illustrated. The Genome Property Definition page displays the property's name, description, parent/child/sibling properties, literature references and associated GO terms. The components of the property are listed with the step names, and whether or not the step is required (used to evaluate completeness). The evidence required to detect that a component has been found is described, usually the accession of a TIGRFAMs HMM with a working link to HMM Summary Page. Users can browse through the Genome Properties through the Top Level Genome Properties page linked on the home page. Users traverse through the properties by first selecting a top level property (e.g. Metabolism, Taxonomy). Associated parent, child and sibling properties are displayed for selection. The Search CMR Properties link directs users to the CMR genome property search. SPECIAL FOCUS AREAS FOR TIGRFAMS AND GENOME PROPERTIES ====================================================== TIGRFAMs focuses heavily, and Genome Properties almost exclusively, on subsystems encoded in bacterial and archaeal genomes. Successful construction of a TIGRFAMs equivalog-level model defines a molecular marker, and makes it possible to evaluate the presence/absence of that marker across the set of all complete microbial genomes. The result, a truth table of 1's and 0's for a large number of genomes, comprises a phylogenetic profile ([@gks1234-B19]). The profile serves as input to the Partial Phylogenetic Profiling algorithm ([@gks1234-B20]), which performs data mining to find (and guide construction of new HMMs for) additional protein families that belong to the same subsystem, even if the families in question have never before been defined and fall within large superfamilies. New models contribute to improved Genome Property definitions, more accurate comparative genomics, and iteratively improved descriptions of subsystems biology. Prokaryotic protein-sorting systems is an area of special focus for TIGRFAMs and Genome Properties, containing 37 such properties. Seventeen of these, including one involving a rhomboid family intramembrane serine protease ([@gks1234-B21]), one featuring a sorting signal unique to the deltaProteobacteria ([@gks1234-B22]), and all subclasses of exosortase-based and archaeosortase-based sorting/processing systems ([@gks1234-B23]), represent discoveries made via TIGRFAMs/Genome Properties development. TIGRFAMs and Genome Properties created the original nomenclature for CRISPR subtypes, defining eight variants (Ecoli, Dvulg, Ypest, Hmari, Tneap, Nmeni, Mtube and the RAMP module) ([@gks1234-B24]). More recently, a reanalysis using sensitive detection methods demonstrated remote homologies that grouped CRISPR systems into type I (where Ecoli is type I-E, Ypest is type I-F, etc), type II and type III ([@gks1234-B25]). Most existing TIGRFAMs models for CRISPR-associated protein families now show multilevel naming that reflects both schemes. Since our original report of the first eight subtypes, we have detected conserved architectures for additional CRISPR subtypes, and described them through TIGRFAMs and Genome Properties. These include the Aferr subtype (GenProp0670, the unique CRISPR system in *Acidithiobacillus ferrooxidans* ATCC 23270), the Pging subtype (GenProp0768, named after *Porphyromonas gingivalis* W83), the Myxan subtype (GenProp0922, named after *Myxococcus xanthus*), and the PreFran subtype (GenProp1061, named for genera *Prevotella and Francisella*). Lastly, development of these databases has introduced many Genome Properties in which a subfamily of radical SAM represents a key marker, and many examples of natural product biosynthesis systems in which a ribosomally translated precursor from a novel protein family is post-translationally modified. These classes overlap heavily for radical SAM proteins with a C-terminal SPASM domain (TIGR04085), where the acronym SPASM signifies roles in the maturation of Subtilosin, PQQ, Anaerobic Sulfatases and Mycofactocin ([@gks1234-B26]). CONCLUSION ========== TIGRFAMs and Genome Properties continue to be developed, in parallel, as annotation-driving databases for microbial genome analysis. Context provided by Genome Properties gives feedback for the process of making the fixed cutoffs of each HMM in TIGRFAMs as accurate as possible. The intended result is that classification of proteins into families with identified functions provides a trustworthy basis for many types of subsequent analysis. The most immediate of these is whether or not some genome in question encodes the necessary sets of components for each of over 600 different subsystems. AVAILABILITY ============ TIGRFAMs resources and Genome Properties resources are available through home pages at <http://www.jcvi.org/cgi-bin/tigrfams/index.cgi> and <http://www.jcvi.org/cgi-bin/genome-properties/index.cgi>, respectively. FUNDING ======= This project has been funded in whole or part with federal funds from the National Human Genome Research Institute \[R01 HG004881\] and from the National Institute of Allergy and Infectious Diseases, National Institutes of Health, Department of Health and Human Services \[N01-AI30071, HHSN272200900007C\]. Funding for open access charge: National Institute of Allergy and Infectious Diseases, National Institutes of Health, Department of Health and Human Services under contract numbers N01-AI30071 and/or HHSN272200900007C. *Conflict of interest statement*. None declared. [^1]: Present address: Malay K. Basu, Informatics, Department of Pathology, University of Alabama at Birmingham, 619 19th St. South WP220D, Birmingham, AL 35249, USA.

Suberin is among the most recalcitrant plant molecular structures in soils ([@b31-27_36]). Suberin forms a protective barrier in tissues such as woody stems, roots and underground storage organs which undergo secondary growth ([@b10-27_36]). This barrier controls the flux of water and also protects plant tissues against biotic diseases ([@b29-27_36]). Suberin is a biopolymer composed of polyaromatic and polyaliphatic domains linked by glycerol moieties ([@b3-27_36]). Microbial degradation of suberin is a process that is poorly characterized. Suberinases are poly-esterases that can depolymerize, at least partially, the lipidic polymer ([@b16-27_36]). Suberinases have been shown to be produced by some fungi belonging to the following genera: *Armillaria*, *Aspergillus*, *Coprinopsis* and *Fusarium*([@b16-27_36]). There is also evidence that some actinomycetes produce suberin-degrading esterases. Esterase activity is induced in *Thermoactinomyces vulgaris*([@b10-27_36]) and the plant pathogen *Streptomyces scabiei*([@b32-27_36]) in the presence of suberin. The genus *Streptomyces* belongs to the order of Actinomycetales, a division of Gram-positive bacteria that are characterized by a genome with a high G+C content. Their complex life cycle includes soil colonization by mycelial growth and terminates with sporulation. *Streptomyces* are known for producing a wide variety of biologically active secondary metabolites such as antibiotics; however, among the numerous species of the genus *Streptomyces*, a few have developed phytopathogenic traits, mainly relying on their ability to produce plant toxic secondary metabolites called thaxtomins ([@b2-27_36]). *Streptomyces scabiei* is the main causal agent of common scab, a severe disease which affects potato tubers and tap root crops ([@b18-27_36]). The cellular response of *S. scabiei* exposed to suberin was investigated using a proteomic differential display technique. It was revealed that its presence up-regulated proteins related to the stress response, glycolysis, morphological differentiation and secondary metabolism ([@b17-27_36]). The effect of suberin on differentiation was corroborated by cultivating *S. scabiei* in the presence or absence of suberin ([@b19-27_36]). Suberin strongly stimulates aerial mycelium development in *S. scabiei* which, in the presence of cellobiose, can lead to the production of the secondary metabolites, thaxtomins ([@b19-27_36]). Suberin is not the only biopolymer that influences differentiation and secondary metabolism in *Streptomyces*. For instance, chitin, the main polymer of insect cuticles and crustacean shells, modulates antibiotic biosynthesis and development in *Streptomyces coelicolor*([@b33-27_36]). Since soil-dwelling streptomycetes can hydrolyse complex natural biopolymers, it was suggested that these polymers play a determinant role in the production of their bioactive molecules ([@b33-27_36]). In the present paper, suberin is shown to affect development and secondary metabolite biosynthesis not only in *S. scabiei*, but also in the plant pathogen *S. acidiscabies*, as well as in saprophytic species such as *S. avermitilis*, *S. coelicolor* and *S. melanosporofaciens*. The molecular mechanisms responsible for the stimulation of differentiation are still unknown, but we demonstrate that suberin acts as a membrane and cell-wall perturbant in streptomycetes. Materials and Methods ===================== Suberin purification, bacterial cultivation and growth conditions ----------------------------------------------------------------- Suberin was obtained from potato peel (*Solanum tuberosum* 'Russet') and purified ([@b15-27_36]). Briefly, potato tubers were sliced and boiled for 20 min. The skin (suberin) was removed and flesh was roughly scraped away. The peel was then rinsed with tap water and residual flesh was digested overnight with cellulase (5 g L^−1^) and pectinase (1 g L^−1^) in 50 mM acetate buffer (pH 4.0). The peel was rinsed again with chloroform:methanol (2:1) and suberin purification was achieved using a Soxhlet extractor with chloroform as a solvent. Finally, suberin was dried and ground for 15 s in a coffee blender. *Streptomyces scabiei* strain EF-35 (HER1481) was initially isolated in Quebec (Canada) from scabby potato tubers ([@b9-27_36]) and was used in all assays. Strains *S. acidiscabies* ATCC 49003, *S. avermitilis* ATCC 31267, *S. coelicolor* M145 (ATCC BAA-471) and *S. melanosporofaciens* EF-76 (ATCC BAA-668) were also used in the morphology and mechanical lysis experiments. Unless otherwise specified, bacteria were cultivated in liquid medium as follows. 10^7^ to 10^8^ spores were inoculated in 50 mL tryptic soy broth (TSB) and grown in a rotary shaker (250 rpm) at 30°C for 48 h. Bacteria were then centrifuged (10 min at 3,450×*g*) and resuspended in 5 volumes of sterile 0.85% NaCl. Volumes of 5 mL of this suspension were used to inoculate flasks containing 200 mL minimal medium (0.5 g L^−1^ asparagine, 0.5 g L^−1^ K~2~HPO~4~, 0.2 g L^−1^ MgSO~4~ and 5 mg L^−1^ FeSO~4~--7H~2~O) supplemented with 1% (w/v) soluble starch and 0% (control medium: CM) or 0.1% (w/v) suberin (suberin medium: SM). In order to collect suberin-free bacterial samples at the end of the experiment, suberin was placed in *ca*. 4×4 cm cotton pouches (200 mg suberin per pouch). Control flasks contained empty cotton pouches. Bacteria were grown at 30°C with shaking (250 rpm). Morphology of bacterial colonies on solid medium ------------------------------------------------ Morphology of the five *Streptomyces* species tested in this study was determined as previously described ([@b19-27_36]). Briefly, 30--50 viable spores from each species were streaked on Petri dishes containing solidified (15 g L^−1^ agar) CM and SM. Petri dishes were incubated at 30°C for 5 d and representative colonies of each species and each treatment were photographed. Production of secondary metabolites ----------------------------------- The production of characteristic secondary metabolites, thaxtomin A for *S. scabiei* and *S. acidiscabies*, actinorhodin for *S. coelicolor* and geldanamycin for *S. melanosporofaciens*, was assessed. These four strains were grown in CM and SM; for the growth of *S. scabiei* and *S. acidiscabies*, a starch/cellobiose combination (0.5% each) ([@b19-27_36]) was used instead of 1% starch since cellobiose is required for the production of thaxtomin A ([@b12-27_36], [@b13-27_36]). After 4 d, bacterial cultures were centrifuged (10 min, 3,450×*g*) and supernatants were decanted for quantification of metabolites. Pellets were dried (24 h at 50°C) and weighed to determine bacterial growth. Thaxtomin A produced by *S. scabiei* and *S. acidiscabies* was purified as previously described ([@b11-27_36]) and quantified by HPLC Agilent 1260 Series (Agilent Technologies, Santa Clara, CA, USA) at 249 nm using a Zorbax SB-C18 column (Agilent Technologies). Abamectin (B~1a~) was extracted with ethyl acetate and quantified by HPLC at 246 nm ([@b20-27_36]). γ-Actinorhodin produced by *S. coelicolor* was quantified by spectrophotometry according to Kieser *et al.*([@b14-27_36]). Geldanamycin produced by *S. melanosporofaciens* was purified by chloroform extraction and quantified by HPLC at 306 nm ([@b4-27_36]). This experiment was carried out in triplicate. Cell wall morphology of *Streptomyces scabiei* ---------------------------------------------- The cell wall morphology of *S. scabiei* grown in the absence or presence of suberin was determined after 7 d of growth. Preparation of bacterial cells, the microscopy procedure and image analyses were carried out according to Miguélez *et al.*([@b25-27_36]). Samples were examined with a Philips EM201 (FEI Company, Hillsboro, OR, USA) electron microscope at 60 kV and photographed on an Eastman Fine Grain Positive film 5302 (Eastman Kodak, Rochester, NY, USA). High-contrast photographic negatives were digitized with a HP Scanjet 6300C slide scanner and images were analyzed using the Image-Pro Plus v.4.5 software (Media Cybernetics, Elizabeth, IN, USA). Bacteria sliced in the middle of the cell, *i.e.*, with a well-defined cell wall and a clearly visible DNA zone in the centre, were selected for analysis. Pixel intensity from zones randomly selected in cell walls (40 zones per treatment) was measured. Cell wall thickness was also determined (25 walls per treatment). Bacterial resistance to mechanical lysis ---------------------------------------- The five *Streptomyces* strains were grown for 7 d in liquid CM and SM (using cotton pouches) containing 2% starch. Bacteria were collected by centrifugation, rinsed with 10 mM Tris-HCl (pH 8.3) and centrifuged again. Supernatants were thoroughly discarded and 300 mg bacteria (fresh weight) were resuspended in 1 mL Tris-HCl buffer. Resistance to mechanical stress was assessed using 0.5 mL of this suspension with 250 mg glass beads (100 μm diameter) using a bead beater (FastPrep FP-120; Thermo Fisher Scientific, Waltham, MA, USA) for 45 s (speed 4.5 m s^−1^). Samples were chilled on ice, centrifuged and supernatants were filtered. Lysis efficiency was evaluated by quantifying protein concentration of supernatants using Bio-Rad protein assay (Bio-Rad Laboratories, Hercules, CA, USA). Measurement of bacterial membrane fluidity of *Streptomyces scabiei* -------------------------------------------------------------------- Membrane fluidity of *S. scabiei* EF-35 bacteria collected from control and suberin treatments was determined. After 1 d of growth in CM and SM, bacteria were centrifuged, washed with 0.85% NaCl and resuspended in 0.85% NaCl to a concentration of 12.5 g L^−1^ bacteria. Membrane fluidity was assessed by an anisotropy test, based on the incorporation of 1,6-diphenyl-1,3,5-hexatriene (DPH) into bacterial membranes. The method described by Shinitzky and Barenholz ([@b35-27_36]) was used with minor modifications. Twenty microliters of 1 mM DPH (prepared in acetone and kept in the dark) ([@b1-27_36]) was added to 10 mL bacterial suspension to obtain a final DPH concentration of 2 μM ([@b30-27_36]). Suspensions were incubated in the dark for 2 h at room temperature with mild shaking. Bacteria were then washed with one volume of 0.85% NaCl, resuspended with exactly 10 mL of 0.85% NaCl and kept on ice. Anisotropy tests were performed using a spectrofluorimetry system equipped with PTI polarizers. Fluorescence of the DPH probe was measured with an excitation wavelength of 355 nm and an emission wavelength of 425 nm ([@b22-27_36]), from 10°C to 40°C by 5°C increments. Data were analyzed with Felix 32 v.1.1 software (Photon Technology International, London, ON, Canada). In another experiment, membrane fluidity of *S. scabiei* EF-35 was determined over a 4-d period. Bacteria were grown in CM and SM (200 mL, four replicates per treatment) and 20 mL bacterial suspension was collected every day. Membrane fluidity was readily measured as described above at a temperature of 25°C. Determination of bacterial membrane fatty acid composition of *Streptomyces scabiei* ------------------------------------------------------------------------------------ Membrane fatty acid composition of *S. scabiei* EF-35 bacteria grown in CM and SM was examined. Bacteria were pelleted by centrifugation, washed with 0.85% NaCl and lyophilized. Bacterial membrane fatty acids were extracted and methylated as in Moss ([@b27-27_36]), with modifications. Saponification was performed on 150 mg dried cells with 1 mL of 15% NaOH in 50% ethanol. Suspensions were incubated for 30 min at 100°C; samples were then cooled and brought to pH 2.0 with 6N HCl. Methylation of fatty acids was achieved after adding 3 mL of 14% BF~3~ to methanol and by incubating at 80--85°C for 20 min. After cooling, methylated fatty acids were extracted twice with one volume of petroleum ether/hexane (1:1) and evaporated to 1 mL with a flow of N~2~. Extracts were then washed with 0.3 N NaOH solution ([@b26-27_36]). The organic phase was transferred to a new tube and the solvent was completely evaporated by N~2~ flow. Residual H~2~O was eliminated by the addition of 80--100 mg Na~2~SO~4~. Tubes were stored under N~2~ at −20°C until analysis. Dried fatty acids were then dissolved in 100 μL hexane and separated using a gas chromatograph HP6890 (Hewlett-Packard, Mississauga, ON) equipped with a capillary column RTX-1 of 30 m × 250 μm × 0.25 μm (Restek, Bellefonte, PA, USA). Fatty acids were identified by comparing with the commercial standard mixes Bacterial Acid Methyl Esters Mix (Matreya, Pleasant Gap, PA, USA) and Supelco 37-component FAME Mix (Sigma, St-Louis, MO, USA). Results ======= Suberin promotes secondary growth and production of secondary metabolites in *streptomycetes* --------------------------------------------------------------------------------------------- The composition of growth media noticeably influenced the morphology of the five *Streptomyces* strains tested in this experiment ([Fig. 1](#f1-27_36){ref-type="fig"}). As previously described ([@b19-27_36]), moderately hairy colonies were observed when *S. scabiei* EF-35 was grown on starch medium while suberin triggered the onset of secondary metabolism (formation of hairy colonies). Suberin also strongly stimulated the production of aerial hyphae in *S. acidiscabies* ATCC 49003, *S. coelicolor* M145 and *S. melanosporofaciens* EF-76 when compared to minimal starch medium (CM). In *S. avermitilis* ATCC 31267, suberin only moderately stimulated aerial growth, while suberin-deprived colonies were bald ([Fig. 1](#f1-27_36){ref-type="fig"}). In liquid minimal medium, the presence of suberin significantly stimulated the growth of all *Streptomyces* spp. tested ([Table 1](#t1-27_36){ref-type="table"}). Furthermore, the production of secondary metabolites typically synthesized by these strains was strongly promoted by the presence of the plant polymer in *S. scabiei*, *S. acidiscabies*, *S. coelicolor* and *S. melanosporofaciens*. In the absence of suberin, neither thaxtomin A nor geldanamycin was detected in media inoculated with *S. scabiei* and *S. melanosporofaciens*, respectively, while metabolite production was significantly limited in flasks inoculated with *S. acidiscabies* and *S. coelicolor* ([Table 1](#t1-27_36){ref-type="table"}). Substantial production of unidentified secondary metabolites was also detected from HPLC chromatograms of *S. acidiscabies* and *S. melanosporofaciens* grown in the presence of suberin ([Fig. S1](#s1-27_36){ref-type="supplementary-material"}). Abamectin production by *S. avermitilis* could not be detected in the presence or absence of suberin; however, three-dimensional HPLC profiles showed a strong increase in the production of various unidentified molecules in the presence of suberin ([Fig. S1](#s1-27_36){ref-type="supplementary-material"}). Suberin alters cell wall morphology ----------------------------------- The presence of suberin in growth medium induced morphological changes in *S. scabiei* EF-35. These modifications were clearly visible by electron microscopy after 7 d of growth. The cell walls of bacteria that had been exposed to suberin contained a high quantity of electron-dense material ([Fig. 2](#f2-27_36){ref-type="fig"}). Cell-wall density of bacteria grown in the presence of suberin (90.5±6.7 pixels) thus appeared significantly higher than cell-wall density of control bacteria (75.2±6.9 pixels; *P*\<0.0001, *t*-test). Image analyses also revealed that cell walls were thicker in suberin-treated bacteria (46.6±8.8 nm) than in control bacteria (36.6±6.7 nm; *P*\<0.0001, *t*-test). Bacteria grown in the presence of suberin showed higher resistance to mechanical lysis -------------------------------------------------------------------------------------- Protein content of supernatants obtained after mechanical lysis of bacterial suspensions was significantly higher for bacteria grown for 7 d in CM than in SM, in all *Streptomyces* strains tested ([Table 2](#t2-27_36){ref-type="table"}). An additional experiment was consequently conducted to assess resistance to mechanical lysis over a 7-d period. *S. scabiei* was grown in CM and SM and resistance to mechanical lysis was measured every day, as described above. The amount of protein released by mechanical treatment was similar for both experimental conditions after incubation periods of 1 and 2 d ([Fig. 3](#f3-27_36){ref-type="fig"}); however, from day 3 to day 7, resistance to mechanical lysis was significantly higher (*i.e.*, protein concentration was lower) in bacteria grown in the presence of suberin ([Fig. 3](#f3-27_36){ref-type="fig"}). Membrane fluidity and fatty acid composition -------------------------------------------- Anisotropy measurements performed with the DPH probe, 1 d after inoculation in minimal medium, revealed that membrane fluidity of *S. scabiei* was significantly higher (*i.e.*, anisotropy was lower) in bacteria grown in the presence of suberin than in control bacteria. This pattern was observed at all temperatures tested ([Fig. 4A](#f4-27_36){ref-type="fig"}). The greater membrane fluidity of bacteria grown in the presence of suberin was maintained over the 4 d time course performed at 25°C ([Fig. 4B](#f4-27_36){ref-type="fig"}). The analysis of fatty acid composition revealed that the membranes of *S. scabiei* contained a majority of two branched-chain (*iso*-16:0 and *anteiso*-15:0), an unsaturated (16:1 \[9\] *cis*, *i.e.*, palmitoleic acid) and a straight-chain (16:0, *i.e.*, palmitic acid) fatty acids ([Table 3](#t3-27_36){ref-type="table"}). Differences between bacteria grown in the absence or presence of suberin were observed after 1 d of growth in minimal medium. Suberin induced a higher proportion of total branched-chain fatty acids. The abundance of two of these, *iso*-16:0 and *anteiso*-17:0, increased significantly in the presence of suberin while the abundance of *iso*-15:0 and *anteiso*-15:0 remained unchanged; however, suberin induced a significantly lower proportion of unsaturated acids. No variation in the proportions of straight-chain fatty acids was observed ([Table 3](#t3-27_36){ref-type="table"}). Discussion ========== Suberin is a polymer recalcitrant to microbial degradation in nature. Unambiguous evidence for the presence of suberin in soil organic matter has been revealed by different groups ([@b24-27_36], [@b28-27_36], [@b31-27_36]). Only rare studies have investigated the biochemical mechanisms associated with suberin degradation ([@b16-27_36]). Nevertheless, some authors have suggested that actinomycetes might be involved in the degradation process ([@b10-27_36], [@b32-27_36]). All *Streptomyces* strains used in this study showed better growth in the presence of the polymer. This enhanced growth is probably not the effect of the utilization of suberin constituents as a carbon source (suberin concentration was low), but would rather result from an increase in membrane fluidity that facilitates the transport of nutrients and waste products ([@b5-27_36]). The fatty acid monomers associated with the suberin structure act as membrane perturbants of phospholipid vesicles ([@b8-27_36]). Here, suberin affected both the membrane composition and fluidity of living *S. scabiei* cells. The anisotropy test unequivocally showed that in the *S. scabiei* EF-35 membrane, fluidity was overall higher in bacteria grown in the presence of suberin. *Iso* and *anteiso* branched-chain fatty acids in the bacterial membrane generally contribute to its fluidity ([@b6-27_36], [@b34-27_36]) and the proportions of branched-chain fatty acids increased in bacterial cells grown in the presence of the polymer. On the other hand, unsaturated fatty acids also have the ability to increase membrane fluidity ([@b6-27_36], [@b7-27_36]), but their proportions decreased in suberin-treated membranes, showing that adjustment of membrane fluidity is a complex mechanism. This decrease in the proportions of unsaturated fatty acids may, however, represent a protection mechanism against phenolic compounds present in the suberin polymer. Denich *et al.*([@b6-27_36]) stated that saturated fatty acids help in preventing the access of phenol molecules to the membrane interior. The presence of suberin in the growth media of the five *Streptomyces* species tested conferred relative protection against mechanical stress, suggesting that this plant polymer triggers changes not only in the bacterial membrane, but also in the cell wall. This was demonstrated in *S. scabiei* since cell walls were thicker in bacteria grown in the presence of suberin. The cell wall also appeared to contain more electron-dense material. The increase in cell-wall thickness may explain the higher resistance of *S. scabiei* to mechanical lysis. Once again, the presence of phenols in suberin may be responsible for the higher thickness of cell walls associated with bacteria grown in the presence of suberin. It was suggested that a high number of peptidoglycan layers may effectively be caused by the exposure of bacterial cells to environmental stresses such as phenols ([@b21-27_36]). In the present study, potato suberin altered the development of five bacterial species of the genus *Streptomyces* (*S. scabiei*, *S. acidiscabies*, *S. avermitilis*, *S. coelicolor* and *S. melanosporofaciens*). The general morphological patterns of isolated colonies of the five *Streptomyces* species used here point toward the capacity of suberin to promote cellular differentiation. On solid minimal medium, colonies of the five species tested here presented a hairy morphotype in the presence of suberin (although this phenotype was only slightly visible in *S. avermitilis*), while in the absence of the biopolymer, colonies were bald, or to some extent hairy. The onset of morphological differentiation attributable to suberin concurs with the production of secondary metabolites whose presence, after 4 d of growth in liquid minimal medium, boosted the synthesis of characteristic phytotoxins and antibiotics by *S. scabiei*, *S. acidiscabies*, *S. coelicolor* and *S. melanosporofaciens*. Although no abamectin was produced by *S. avermitilis* under the tested conditions, the production of other unidentified metabolites was apparently stimulated by the presence of suberin. Abamectin is only one secondary metabolite potentially produced by *S. avermitilis* and other antibiotics may also be synthesized by this bacterium ([@b20-27_36]). It has been speculated that complex polymers could influence the development and production of bioactive molecules by soil-dwelling streptomycetes ([@b33-27_36]). For instance, *N*-acetylglucosamine and cellobiose, the main degradation products of chitin and cellulose, lock *S. coelicolor*([@b33-27_36]) and *S. scabiei*([@b19-27_36]) in the vegetative state. The effect of cellulose and cellooligosaccharides on morphological development has also been described in *S. griseus*([@b23-27_36]). Interestingly, it has been demonstrated that suberin counteracts the effect of cellobiose and could promote morphological differentiation, even in the presence of the disaccharide ([@b19-27_36]). In *S. coelicolor*, environmental stresses trigger cell morphological differentiation associated with secondary metabolism ([@b36-27_36]) and it was shown that suberin is perceived by streptomycetes as a stress factor ([@b17-27_36]). The data presented in this paper not only confirm previous observations suggesting that suberin triggers the onset of secondary metabolism ([@b17-27_36], [@b19-27_36]) but also bring to light new characteristics of the changes induced by the presence of suberin. When exposed to this plant polymer, streptomycetes undergo profound morphological modifications. While mechanisms linked to suberin biodegradation in streptomycetes are still largely unknown, determining which suberin constituents trigger differentiation is challenging. Characterization of the secretome of *Streptomyces* species grown in the presence of suberin is in progress. Supplementary Material ====================== This work was supported by the National Sciences and Engineering Research Council of Canada (NSERC) and the Fonds Québécois de Recherche en Nature et Technologie. AL gratefully acknowledges the receipt of a scholarship from NSERC. {#f1-27_36} {#f2-27_36} {#f3-27_36} {#f4-27_36} ###### Bacterial growth and production of typical secondary metabolites by five *Streptomyces* species grown for 4 d in MM in the absence or presence of suberin Dry mycelial weight (mg±SD) Metabolite production (μg mg DW^−1^±SD)[a](#tfn2-27_36){ref-type="table-fn"} ------------------------- ----------------------------- ------------------------------------------------------------------------------ ----------- ----------------- *S. scabiei* 28±4 90±3\*\*\* n.d. 3.61±0.14\*\*\* *S. acidiscabies* 60±3 88±2\*\*\* 0.05±0.00 1.44±0.42\*\* *S. avermitilis* 27±4 134±3\*\*\* n.d. n.d. *S. coelicolor* 38±1 64±4\*\*\* 0.35±0.20 1.22±0.45\* *S. melanosporofaciens* 41±4 69±6\*\* n.d. 1.32±0.79\*\*\* Values are the means of three replicates. Metabolites assayed were thaxtomin A for *S. scabiei* and *S. acidiscabies*, abamectin for *S. avermitilis*, γ-Actinorhodin for *S. coelicolor* and geldanamycin for *S. melanosporofaciens*. Values from suberin medium are significantly different from control at \*: *P*\<0.05, \*\*: *P*\<0.01 and \*\*\*: *P*\<0.001 (*t*-test). n.d.: not detected; detection limits were 0.05 μg, 0.05 μg and 0.1 μg of total thaxtomin A, abamectin and geldanamycin, respectively. ###### Proteins (mg mL^−1^±SD) released by mechanical lysis performed on five *Streptomyces* species grown for 7 d in the absence or presence of suberin *S. scabiei* *S. acidiscabies* *S. avermitilis* *S. coelicolor* *S. melanosporofaciens* ---------------------- -------------- ------------------- ------------------ ----------------- ------------------------- Control 0.26±0.05 0.75±0.06 0.71±0.06 1.05±0.10 2.23±0.11 Suberin 0.15±0.05 0.48±0.07 0.29±0.04 0.77±0.08 1.52±0.09 *P* value (*t*-test) 0.0005 0.0013 0.0001 0.0223 \<0.0001 Values are the means of four replicates. ###### Membrane fatty acid composition of *Streptomyces scabiei* EF-35 grown for 1 d in minimal medium in the absence or presence of suberin Fatty acids Control (%±SD) Suberin (%±SD) ------------------------------------------------ ---------------- ---------------- Branched-chain\*\* 51.0±3.4 55.2±0.6  *anteiso*-13:0 1.7±0.3 1.7±0.3  *iso*-13:0 2.1±0.1 2.1±0.1  *iso*-14:0 5.4±0.4 5.0±0.3  *iso*-15:0 4.2±0.4 4.22±0.23  *iso*-16:0\*\* 17.5±1.9 20.0±0.2  *anteiso*-15:0 11.3±0.3 12.2±0.5  *iso*-17:0 2.9±0.2 2.7±0.1  *anteiso*-17:0\*\* 6.0±0.4 7.2±0.0 Unsaturated\* 18.8±2.5 15.5±0.3  16:1 ([@b9-27_36])*cis*\* 16.4±3.1 13.0±0.1  *iso*-17:0 2.4±0.6 2.5±0.2 Straight-chain 30.2±0.9 29.3±1.4  14:0 1.2±0.3 1.1±0.3  14:0 3-OH 5.6±0.4 5.3±0.1  15:0 3.3±0.6 3.2±0.7  16:0 15.3±1.7 14.3±0.5  17:0 cyclopropane ([@b9-27_36], [@b10-27_36]) 3.7±1.1 4.2±0.4  18:0 1.1±0.1 1.2±0.0 Values are the means of six replicates (three replicates of two repeats). ANOVA; \*: *P*\<0.05, \*\*: *P*\<0.01.

1. Introduction {#sec1} =============== Multilevel inverters have generally functioned as an AC voltage synthesizer to create a quasi-sinusoidal voltage i.e., rough sinusoidal waveform with desired steps using several DC power supplies \[[@bib1]\]. Hence, high number of switching components has let the multilevel inverter to render a quasi-sinusoidal voltage with high quality \[[@bib2]\]. The first multilevel inverter prototype has been constructed and analyzed by Nabae et al. \[[@bib3]\]. Since that time, these structures have been captivating more industrial and academic consideration in the various medium voltages that such these multilevel structures have investigated in several literatures to elucidate the rating problem related to voltage and current of switching components \[[@bib4], [@bib5], [@bib6], [@bib7], [@bib8]\]. One of the strongpoints of these structures other than creating a pertinent quasi-sinusoidal waveform is their feasibility to occupy them in transmission and sub-transmission levels for such these reasons: reduction of SPL, PIV, dV/dt and EMI \[[@bib9], [@bib10], [@bib11]\]. One more special strongpoint of these structures is their capability in harmonic alleviation, avoiding the high frequency of semiconductors\' switching either reduction of the inverter\'s output power \[[@bib12], [@bib13], [@bib14]\]. Multilevel inverter is known as one of the most advanced and prominent power electronic structures. They have been widely applied in manifold compensators and energy conversion systems e.g., HVDC, FACTS, PV, UPS and Industrial drive systems. The conventional multilevel inverters have been essentially divided into three categories that are CHB, NPC, FC \[[@bib15], [@bib16], [@bib17]\]. It is worth mentioning that, the conventional multilevel inverters have borne high switching elements such as: switches, capacitors, diodes and the suchlike toward producing high step number. Over the past several years, many different structures and technologies based on IJBT and MOSFET semiconductors with have been introduced \[[@bib18], [@bib19], [@bib20], [@bib21]\]. Apart from the beneficial advantages of multilevel inverters, existence of a large number of semiconductor switches is their most important disadvantage that consequently will bring such detriments: complexity of the circuit and control strategy, increase of the structure\'s cost-price, decrease of the reliability, and maintenance of the structure. Therefore, it is essential that the number of switching components to be decreased. Across the world, the scholars are attempting and venturing to improve and introduce high performance multilevel inverters with reduced components so that the aforementioned problems related to the conventional multilevel inverters to be obviated e.g.: \[[@bib22], [@bib23], [@bib24], [@bib25], [@bib26]\]. Such the reduced multilevel topologies have generally competed with each other over the: low THD, and number of switching components as well as high-step of quasi-sinusoidal voltage, in which case, an inventive AMLI has been designed and introduced in this paper so that not only low number of components to be applied, but also high-step quasi-sinusoidal voltage to be attained. The proposed sub-AMLI has been designed based on the ladder capacitor clamped to H-bridge by means of the bi-directional switches. Based on an innovative geometric progression for DC power supplies of AMLI, it can operate as so-called odd-nary AMLI. In recent years, the power electronics experts have been endeavoring to enhance the conventional multilevel inverter structures and voltage quality. Therefore, undesirable voltage harmonics of multilevel inverters with DC power supplies can be reduced by the advantageous modulation strategies such as PWM and SVPWM \[[@bib27], [@bib28]\]. It should be considered, these strategies cannot accurately reduce the relevant harmonic components. Another strategy has been functioned in accordance with selection of the triggering angles to reduce the lower order harmonics viz. third, fifth, seventh from the quasi-sinusoidal voltage which is recognized by SHE \[[@bib29], [@bib30], [@bib31]\]. A principal problem as regards of such these strategies is acquiring the arithmetic solution of nonlinear transcendental equations which include the trigonometric clauses. These relationships can be also solved by iterative methods including Newton--Raphson \[[@bib32]\]. Unfortunately, its dependency on the initial estimation points may cause trapping in local optima. Therefore, this approach can\'t be considered to unravel the SHE problem with various triggering angles whereas have no initial estimation points. One more strategy has been operated via converting the transcendental equations into the polynomial equations \[[@bib33], [@bib34]\]. If the step number of voltage to be increased, the computational burden is highly intensified. As for the capability of heuristic algorithms in solving the complex objective functions, they have been widely used to figure out the optimal triggering angles. Toward this subject, PWM-based scission-style harmonic elimination has been designed and implemented to enhance voltage quality of AMLI. Hence, three prominent criterions related voltage quality i.e., desired fundamental component, elimination of third harmonic and Reduction of THD have been assigned as objective functions of optimization problem. Considering the multi-objective nature of the optimization problem, non-dominated sorting MOGWO has been implemented to solve the optimization problem, and then compared with NSGA-II. To sum up, the relevant analytical studies along with both the simulation and experimental results have transparently validated the performance of suggested AMLI. Also, the voltage quality problem using MOGWO-based scission-style harmonic elimination strategy has been well approved. 2. Suggested odd-nary multilevel inverter {#sec2} ========================================= A new AMLI structure is proposed aimed to provide high-step quasi-sinusoidal with respect to few switch numbers. Suggested AMLI can feasibly act as both the symmetric and statuses. According to the [Figure 1](#fig1){ref-type="fig"}, sub-AMLI is composed using DC power supplies, capacitors along with unidirectional and bidirectional semiconductor switches. By cascading of these sub-multilevel inverters the overall construction of the proposed AMLI is revealed. The relevant components utilized in suggested AMLI are given as follows:$$N_{\text{uni-directional}} = 4.N_{Source}$$$$N_{\text{bi-directional}} = \left( {k - 1} \right).N_{Source}$$$$N_{\text{Switch}} = N_{\text{uni-directional}} + 2N_{\text{bi-directional}} = 2\left( {k + 1} \right).N_{Source}$$$$N_{\text{capacitor}} = k.N_{Source}$$Figure 1(a). Proposed cascaded multilevel inverter. (b). Prototype\'s photograph of suggested AMLI.Figure 1 Having said that, Based on an innovative geometric progression for DC power supplies with different ratios a pertinent sinusoidal voltage with uniform step has been provided. Both symmetric and asymmetric operations of suggested structure have been explained in detail in as follows. To confirm the suggested AMLI concept, 25-level quasi-sinusoidal of [Figure 1 (a)](#fig1){ref-type="fig"} has been constructed and analyzed. Todd-nary AMLI can provide staircase sinusoidal voltage with maximum of 120 V on the output. For this case, a single-phase resistive-inductive load is considered as the test load with R = 170 ^Ω^ and L = 400 ^mH^. The prototype\'s photograph has been provided and also presented in [Figure 1](#fig1){ref-type="fig"}(b). The prototype AMLI has been constructed by BUP403 (600 V, 42A) IGBTs as switching devices, IRS2113 as IGBT driver, and BYW (56 V) as fast-recovery diodes. The IC ATMega 64 microcontroller by ATMEL Company has been utilized in order to create the triggering scheme." 2.1. Multilevel symmetrical status {#sec2.1} ---------------------------------- Based on the formation of bidirectional switches, capacitors and module numbers accompanied by particular switching operation the quasi-sinusoidal voltage with requires steps can be provided. In symmetrical operation, the values of DC power supplies are equal in symmetric status i.e., V~dc~. It is obvious that the steps of quasi-sinusoidal voltage will be increased by adding bidirectional semiconductor switches along with capacitors, and subsequently with cascading them. In accordance with the triggering of the switches\' gates as well as the arrangement of components, different voltage levels are created. Sub-structure can be designed based on essential components, while, one bidirectional switch along with two capacitors (k = 2) provides five levels, two bidirectional switches along with three capacitors (k = 3) provides seven levels and in the same way k-1 bidirectional switch along with k capacitors provides 2k+1 levels. For better understanding, the switching strategy considering: two capacitors for each sub-structure and two sub-structure modules i.e. k = 2 and n = 2 are tabulated in [Table 1](#tbl1){ref-type="table"}. The output voltage of each module accompanied by overall output voltage of multilevel inverter for this condition is given in [Figure 2 (a)](#fig2){ref-type="fig"}. Furthermore, the relevant experimental results are presented in [Figure 2 (b)](#fig2){ref-type="fig"}.Table 1Relevant triggering mode of suggested multilevel inverter as symmetrical operation.Table 1ON SwitchesV~out_symmetric~ON SwitchesV~out_symmetric~H~4~^1^, S~1~^1^, H~1~^2^, H~2~^2^ (H~1~^1^, H~2~^1^, H~4~^2^, S~1~^2^)1S~1~^1^, H~2~^1^, H~3~^2^, H~4~^2^ (H~3~^1^, H~4~^1^, H~2~^2^, S~1~^2^)-1H~4~^1^, H~1~^1^, H~1~^2^, H~2~^2^ (H~4~^1^, S~1~^1^, H~4~^2^, S~1~^2^)2H~3~^1^, H~2~^1^, H~3~^2^, H~4~^2^ (S~1~^1^, H~2~^1^, H~2~^2^, S~1~^2^)-2H~4~^1^, H~1~^1^, H~4~^2^, S~1~^2^ (H~4~^1^, S~1~^1^, H~1~^2^, H~4~^2^)3H~3~^1^, H~2~^1^, S~1~^2^, H~1~^2^ (S~1~^1^, H~2~^1^, H~2~^2^, H~3~^2^)-3H~4~^1^, H~1~^1^, H~1~^2^, H~4~^2^4H~3~^1^, H~2~^1^, H~2~^2^, H~1~^2^-4H~2~^1^, H~1~^1^, H~1~^2^, H~2~^2^0H~3~^1^, H~4~^1^, H~2~^2^, H~3~^2^0Figure 2(a). quasi-sinusoidal voltage in symmetrical operation with respect to k = 2 and n = 2. b. Experimental results of quasi-sinusoidal voltage in symmetrical operation with respect to k = 2 and n = 2.Figure 2 If each module has k capacitors, all capacitors\' values should be the same. The DC power supplies\' values in terms of capacitors value can be written by:$$V_{in,sub - multlvel} = \sum\limits_{j = 1}^{k}V_{c,j}$$where, $= V_{c,1 =}V_{c,2 =}...V_{c,k} = V_{c}$ Thus:$$V_{in,sub - multlvel} = k.V_{c} = V_{dc}$$ Summing the modules\' outputs, the maximum voltage of multilevel inverter is attained:$$V_{o,\max,symmetric} = \sum\limits_{i = 1}^{n}{V_{in,sub - multlvel,i} =}N_{Source}.k.V_{c} = N_{Source}.V_{dc}$$ Also, voltage step number can be presented by:$$M_{\text{Level,symmetric}} = 2.\left( {k.V_{o,\max,symmetric}} \right)/V_{dc} + 1$$Here, M~Level~/N~Source~ is introduced to accurately inspect the utilization performance of Switches and DC Sources to create the step level. It can be given as follows:$$\frac{M_{\text{Level,symmetric}}}{N_{Source}} = 2\left( {k + \frac{\left( {k + 1} \right)}{N_{Switch}}} \right)$$ 2.2. Multilevel asymmetrical status {#sec2.2} ----------------------------------- For providing a high-step quasi-sinusoidal voltage without adding any components, occupation of asymmetrical multilevel inverters isn\'t avoidable. In the asymmetric structure, the values of DC power supplies are non-equal which are increased in accordance with the particular geometric progression assumed. In general, such these structures are engaged as an alternative approach to minimize the output voltages THD with available electric power components. If each module has k capacitors, DC power supplies\' values in terms of capacitors value can be written by:$$V_{in,sub - multlvel,1} = \sum\limits_{j = 1}^{k}V_{c,j}$$where, $V_{c,1 =}V_{c,2 =}...V_{c,k} = V_{c1}$ Thus:$$V_{in,sub - multlvel} = k.V_{c1} = V_{dc}$$ Also, DC power supply value of the each module is determined by:$$V_{in,sub - multlvel,i} = \left\lbrack {2j + 1} \right\rbrack^{i - 1}V_{dc}$$where, j = 2, 3... k and i = 1, 2... n Summing the modules\' outputs, the maximum voltage of multilevel inverter is attained:$$V_{o,\max,asymmetric} = \sum\limits_{i = 1}^{n}{V_{in,sub - multlvel,i} =}\frac{\left( {\left( {2k + 1} \right)^{N_{Source}} - 1} \right)}{2}k.V_{c1} = \frac{\left( {\left( {2k + 1} \right)^{N_{Source}} - 1} \right)}{2}V_{dc}$$ Also, voltage step number can be presented by:$$M_{\text{Level,asymmetric}} = \frac{\left( {2V_{o,\max,asymmetric}} \right)}{V_{dc}} + 1$$ M~Level~/N~Source~ for asymmetric status can be given as follows:$$\frac{M_{\text{Level,asymmetric}}}{N_{Source}} = \frac{2}{N_{Switch}}.\left( {k + 1} \right) + \left( {2k + 1} \right)^{\frac{N_{Switch}}{2{({k + 1})}}}$$ The advantage of proposed AMLI is its high ability to act the same as "Quinary" with respect to one bidirectional switch along with two capacitors (k = 2), "Septenary" with respect to two bidirectional switches along with three capacitors (k = 3), "Nonary" with respect to three bidirectional switches along with four capacitors (k = 4), "Undenary" with respect to four bidirectional switches along with five capacitors (k = 5), and also in the same way the suggested Odd-nary AMLI has been attained. To have better understanding of aforesaid explanations, the triggering mode of switches considering: two capacitors for each module and two modules i.e. k = 2 and n = 2 are tabulated in [Table 2](#tbl2){ref-type="table"}. Furthermore, the output voltage of each sub-multilevel inverter accompanied by overall output voltage of AMLI is presented in [Figure 3](#fig3){ref-type="fig"}(a). Furthermore, the relevant experimental results are presented in [Figure 3](#fig3){ref-type="fig"}(b).Table 2Relevant triggering mode of suggested multilevel inverter as asymmetrical operation.Table 2ON SwitchesV~out_symmetric~ON SwitchesV~out_symmetric~S~1~^1^, H~4~^1^, H~1~^2^, H~2~^2^1S~1~^1^, H~2~^1^, H~3~^2^, H~4~^2^-1H~1~^1^, H~4~^1^, H~1~^2^, H~2~^2^2H~3~^1^, H~2~^1^, H~3~^2^, H~4~^2^-2H~3~^1^, H~2~^1^, S~1~^2^, H~4~^2^3H~1~^1^, H~4~^1^, S~1~^2^, H~2~^2^-3S~1~^1^, H~2~^1^, S~1~^2^, H~4~^2^4S~1~^1^, H~4~^1^, S~1~^2^, H~2~^2^-4H~1~^1^, H~2~^1^, S~1~^2^, H~4~^2^5H~3~^1^, H~4~^1^, S~1~^2^, H~2~^2^-5S~1~^1^, H~4~^1^, S~1~^2^, H~4~^2^6S~1~^1^, H~2~^1^, S~1~^2^, H~2~^2^-6H~1~^1^, H~4~^1^, S~1~^2^, H~4~^2^7H~3~^1^, H~2~^1^, S~1~^2^, H~2~^2^-7H~3~^1^, H~2~^1^, H~1~^2^, H~4~^2^8H~1~^1^, H~4~^1^, H~3~^2^, H~2~^2^-8S~1~^1^, H~2~^1^, H~1~^2^, H~4~^2^9S~1~^1^, H~4~^1^, H~3~^2^, H~2~^2^-9H~1~^1^, H~2~^1^, H~1~^2^, H~4~^2^10H~3~^1^, H~4~^1^, H~3~^2^, H~2~^2^-10S~1~^1^, H~4~^1^, H~1~^2^, H~4~^2^11S~1~^1^, H~2~^1^, H~3~^2^, H~2~^2^-11H~1~^1^, H~4~^1^, H~1~^2^, H~4~^2^12H~2~^1^, H~3~^1^, H~3~^2^, H~2~^2^-12H~1~^1^, H~2~^1^, H~1~^2^, H~2~^2^0H~3~^1^, H~4~^1^, H~3~^2^, H~4~^2^0Figure 3(a). quasi-sinusoidal voltage in Asymmetrical operation with respect to k = 2 and n = 2. (b). Experimental results quasi-sinusoidal voltage in symmetrical operation with respect to k = 2 and n = 2.Figure 3 2.3. Comparison of proposed multilevel inverter with other conventional multilevel inverters {#sec2.3} -------------------------------------------------------------------------------------------- The capability of proposed multilevel inverter has been compared with other conventional multilevel inverters. To validate its capability, it has been evaluated under both the symmetric and asymmetric operation and compared with the conventional and recently introduced symmetric and asymmetric structures. The relationships between switch numbers in CHB structure, proposed structure in \[[@bib35], [@bib36], [@bib37]\]: and \[[@bib38]\], and M~Level~/N~Source~ benchmark in symmetric status can be respectively given as follows:$$\frac{\text{M}_{\text{Level}}}{N_{Source}}\left\lbrack CHB \right\rbrack = 2 + \frac{1}{4N_{Switch}}$$$$\frac{\text{M}_{\text{Level}}}{N_{Source}}\left\lbrack 35 \right\rbrack = 2\left( {1 + \frac{1}{\left( {N_{Switch} - 2} \right)}} \right)$$$$\frac{\text{M}_{\text{Level}}}{N_{Source}}\left\lbrack 36 \right\rbrack = 2\left( {1 + \frac{1}{\left( {N_{Switch} - 4} \right)}} \right)$$$$\frac{\text{M}_{\text{Level}}}{N_{Source}}\left\lbrack 37 \right\rbrack = \frac{1}{3} + \frac{2}{N_{Switch}}$$$$\frac{\text{M}_{\text{Level}}}{N_{Source}}\left\lbrack 38 \right\rbrack = 2\left( {1 + \frac{3}{\left( {N_{Switch} - 4} \right)}} \right)$$ Also, the relationships between switch numbers in binary and trinary structures \[[@bib37], [@bib38], [@bib39]\]: and \[[@bib40]\], and M~Level~/N~Source~ benchmark in asymmetric status can be respectively given as follows:$$\frac{M_{\text{Level,asymmetric}}}{N_{Source}}\left\lbrack Binary \right\rbrack = \left( \frac{4}{N_{Switch}} \right)\left( {2^{\lbrack{\frac{N_{Switch}}{4} + 1}\rbrack} + 1} \right)$$$$\frac{M_{\text{Level,asymmetric}}}{N_{Source}}\left\lbrack Trinary \right\rbrack = \left( \frac{4}{N_{Switch}} \right)\left( 3^{\lbrack\frac{N_{Switch}}{4}\rbrack} \right)$$$$\frac{M_{\text{Level,asymmetric}}}{N_{Source}}\left\lbrack 37 \right\rbrack = \left( \frac{2}{N_{Switch}} \right)\left( 5^{\lbrack\frac{N_{Switch}}{6}\rbrack} \right)$$$$\frac{M_{\text{Level,asymmetric}}}{N_{Source}}\left\lbrack 38 \right\rbrack = 3 + \frac{2}{N_{Switch} - 4}$$$$\frac{M_{\text{Level,asymmetric}}}{N_{Source}}\left\lbrack 39 \right\rbrack = \left( \frac{2}{\left( {N_{Switch} - 4} \right)} \right)\left( {2^{\lbrack\frac{N_{Switch} - 2}{2}\rbrack} - 1} \right)$$$$\frac{M_{\text{Level,asymmetric}}}{N_{Source}}\left\lbrack 40 \right\rbrack = \left( \frac{3n - 1}{n\left( {N_{Switch} - 4} \right)} \right)\left( {2\left( {n + 1} \right)^{\lbrack\frac{N_{Switch} - 4}{3n - 1}\rbrack} - 1} \right)$$ The variation of M~Level~/N~Source~ benchmark versus switch number varieties of suggested structure and others for symmetrical and asymmetrical statuses are depicted in [Figure 4 (a)](#fig4){ref-type="fig"} and (b).Figure 4(a). Variation of M~Level~/N~Source~ versus of switch numbers of suggested structure and others for symmetric statuses. (b). Variation of M~Level~/N~Source~ versus of switch numbers of suggested structure and others for asymmetric statuses.Figure 4 The portrayed curves in [Figure 4 (a)](#fig4){ref-type="fig"} and (b) validate the high capability of suggested multilevel inverter aimed to create high-step quasi-sinusoidal voltage against the low embedded switches and DC power supplies. Furthermore, the number of required components for suggested multilevel inverter and others are presented in [Table 3](#tbl3){ref-type="table"}. As consequence, the significant problems related to multilevel inverters such as voltage quality, low THD, structure\'s cost price have been attained.Table 3The comparison of the suggested multilevel inverter and other topologies.Table 3ParametersCHBRef. \[[@bib35]\]BinaryTrinaryRef. \[[@bib36]\]Ref. \[[@bib37]\]Ref. \[[@bib38]\]Ref. \[[@bib39]\]Ref. \[[@bib40]\]Suggested InverterNo. of DC SuppliesNNNNNNNNNNNo. of Switches4N2(N+1)4N4N2(N+2)3N2(N+2)2(N+2)3N-12(K+1) per Sub-MLINo. of Steps2N + 12N + 12^(N+1)^-13^N^2N + 13^N^3N-22^(N+1)^-12(n+1)^k^-1(2K + 1)^N^No. of Gate Driver4N2(N+1)4N4N2(N+2)2N2(N+2)2(N+2)3N-1K+3 per Sub-MLINo. of ON State Switches2NN+12N2NN+23/2NN+2N+2(3N-1)/22N + 1 3. Voltage quality improvement scheme based on harmonic elimination Pulse Width Modulation {#sec3} ========================================================================================== 3.1. Harmonic elimination principle {#sec3.1} ----------------------------------- Increase of switches to unravel the voltage quality problem becomes paler because the multilevel inverter provides a quasi-sinusoidal voltage waveform with trivial THD. Toward this subject, voltage quality of the proposed structure has been improved taking into account of ten switches to earn three prominent criterions related voltage quality i.e., desired fundamental component, elimination of third harmonic and Reduction of THD. Hence, nine levels symmetric status of the suggested structure has been considered to better exhibit elimination and reduction of harmonic components. As well, the created quasi-sinusoidal voltage waveform can be extended in Fourier series:$$\upsilon_{out}\left( \alpha \right) = A_{0} + \sum\limits_{n = 1}^{\infty}{A_{n}.\cos\left( {n\alpha} \right)} + B_{n}.\sin\left( {n\alpha} \right)$$where,$$\left\{ \begin{array}{l} {A_{0} = \left( {2\pi} \right)^{- 1}{\int_{0}^{2\pi}{V_{out}\left( \alpha \right).d\alpha}}} \\ {A_{n} = \left( \pi \right)^{- 1}{\int_{0}^{2\pi}{V_{out}\left( \alpha \right).\cos\left( {n\alpha} \right).d\left( \alpha \right)}}} \\ {B_{n} = \left( \pi \right)^{- 1}{\int_{0}^{2\pi}{V_{out}\left( \alpha \right).\sin\left( {n\alpha} \right).d\alpha}}} \\ \end{array} \right.$$ Due to absence of the average value or DC component created by multilevel inverter A~0~ = 0. According to the general quasi-sinusoidal waveform of multilevel inverter presented in [Figure 5](#fig5){ref-type="fig"}\_a, due to being odd function along with quarter-wave symmetry, only sinusoidal part of Fourier series with presence of odd harmonics can be appeared:$$\upsilon_{out}\left( \alpha \right) = \sum\limits_{n = 1,3,5,...}^{\infty}{B_{n}.\sin\left( {n\alpha} \right)}$$Figure 5(a). General quasi-sinusoidal voltage of cascaded multilevel inverter. (b). General scission-style harmonic elimination in quarter cycle waveform of cascaded multilevel inverter.Figure 5 3.2. PWM-based harmonic elimination strategy {#sec3.2} -------------------------------------------- Harmonic elimination strategies have been designed and formulated to eliminate destructive harmonics from the quasi-sinusoidal voltage which is recognized as a contractual approach among the scholars and scientists. It has been essentially employed to decrease the THD from the output of inverter \[[@bib41], [@bib42]\]. Conventional HEPWM method i.e. non-scission HEPWM triggering has some prominent disadvantages. The first, to eliminate or reduce all the harmonics, a number of voltage levels must be produced by multilevel inverter which required more switches. The second, as the suggested symmetric inverter module has been triggered just twice in each quarter cycle, the perfect performance of these switches (high frequency switches) couldn\'t be used. Toward this subject, the scission-style harmonic elimination based on PWM that can modulate a number of angles within specific level provides quasi-sinusoidal voltage waveform with several scissions in each level as shown in [Figure 5 (a)](#fig5){ref-type="fig"}. The significant feature of "this strategy is eliminating further harmonics of low-step quasi-sinusoidal voltage i.e. low switch numbers. Thus, scission-style harmonic elimination owing to more triggering of switches in each quarter cycle, their performance will reach to fullest potential. Moreover, unequal DC power supplies can more augment the performance of this approach aimed to reduce lower order harmonics. Anyhow, the optimization problem has been formulated with various values of: DC power supplies, pulses width of quasi-sinusoidal waveform and triggering angles in each level which can be perceived with envisaging the [Figure 5 (b)](#fig5){ref-type="fig"}. Fourier series of steppes quasi-sinusoidal voltage with non-equal dc sources can be presented by (28): Where, s is the number of DC power supplies, and the *V*~*k*~*V*~*DC*~ is amount of the *kth* DC power supply. If all of DC power supplies have the equal value therefore *V*~*1*~*=V*~*2*~*= ...=V*~*S*~ *= 1*. Also, α~1~... α~m1~ are the triggering transitions in the first level, α~m1+1~...α~m1+m2~ are the triggering transitions in the second level, and the same α~m1+m2+...+1~ +...+ α~m1+m2+...+ms~ are the triggering transitions in the last level. The solution set will be attained using maintaining the value of fundamental component of output voltage in desirable value, *V*~*f*~, and all the harmonics except first set to be zero as presented by (31)--(33), and also the modulation index can be determined by Eqs. [(34)](#fd34){ref-type="disp-formula"} and [(35)](#fd35){ref-type="disp-formula"}:$$\upsilon_{out}\left( {\omega t} \right) = \sum\limits_{n = 1,3,5,...}^{s}\left\langle {\frac{4V_{f}}{n\pi}.\left\lbrack {\upsilon_{1}\sum\limits_{i = 1}^{m_{1}}{\left( {- 1} \right)^{i + 1}\ \cos\ n\alpha_{i} \pm \upsilon_{2}\sum\limits_{i = m_{1} + 1}^{m_{2}}{\left( {- 1} \right)^{i + 1}\ \cos\ n\alpha_{i}}} \pm ... \pm \upsilon_{s}\sum\limits_{i = m_{({s - 1})} + 1}^{m_{s}}{\left( {- 1} \right)^{i + 1}\ \cos\ n\alpha_{i}}} \right\rbrack} \right\rangle\sin\left( {n\omega t} \right)$$$$\hslash_{1} = \left\lbrack {\sum\limits_{i = 1}^{m_{1}}{\left( {- 1} \right)^{i - 1}\ \cos\ \alpha_{i} + \sum\limits_{i = m_{1} + 1}^{m_{2}}{\left( {- 1} \right)^{i - {({m_{1} + 1})}}\ \cos\ \alpha_{i}}} + ... + \sum\limits_{i = m_{({S - 1})} + 1}^{m_{S}}{\left( {- 1} \right)^{i - {({m_{({S - 1})} + 1})}}\ \cos\ \alpha_{i}}} \right\rbrack = M$$$$\hslash_{2} = \left\lbrack {\sum\limits_{i = 1}^{m_{1}}{\left( {- 1} \right)^{i - 1}\ \cos\ 5\alpha_{i} + \sum\limits_{i = m_{1} + 1}^{m_{2}}{\left( {- 1} \right)^{i - {({m_{1} + 1})}}\ \cos\ 5\alpha_{i}}} + ... + \sum\limits_{i = m_{({S - 1})} + 1}^{m_{S}}{\left( {- 1} \right)^{i - {({m_{({S - 1})} + 1})}}\ \cos\ 5\alpha_{i}}} \right\rbrack$$$$\hslash_{m} = \left\lbrack {\sum\limits_{i = 1}^{m_{1}}{\left( {- 1} \right)^{i - 1}\ \cos\left( {3m - 2} \right)\alpha_{i} + \sum\limits_{i = m_{1} + 1}^{m_{2}}{\left( {- 1} \right)^{i - {({m_{1} + 1})}}\ \cos\left( {3m - 2} \right)\alpha_{i}}} + ...}{+ \sum\limits_{i = m_{({S - 1})} + 1}^{m_{S}}{\left( {- 1} \right)^{i - {({m_{({S - 1})} + 1})}}\ \cos\left( {3m - 2} \right)\alpha_{i}}} \right\rbrack$$$$M = \frac{\pi V_{f}}{4V_{DC}},\mspace{9mu} 0 \leq M \leq S$$$$M_{i} = \frac{M}{S},\mspace{9mu} 0 \leq M_{i} \leq 1$$ 3.3. Multi-objective functions {#sec3.3} ------------------------------ To achieve both aforesaid prominent targets, objective functions are determined by (36), (37). The optimal problem will be solved by minimizing both aforesaid equations by MOGWO which will be explained next section.$$F_{1} = THD = V_{f}^{- 1}.\sqrt{\sum\limits_{i = 2}^{\infty}\hslash_{i}^{2}}$$$$F_{2} = \left| {\hslash_{1} - M} \right|$$$$F_{3} = \hslash_{3}$$ For appropriately carrying out the scission-style harmonic elimination scheme, it is impressive that an accurate sequencing of the triggering angles to be presented; i.e.:$$0 \leq \alpha_{1} < \alpha_{2}... < \alpha_{m_{1}}... < \alpha_{m_{S}} \leq \frac{\pi}{2}$$ Also, in order to augment this approach regarding to further harmonics, it can be taken into account that DC power supplies have non-similar magnitudes, i.e.:$$0 < V_{1} \neq V_{2} \neq ... \neq V_{s} \leq 1$$ And same the prior created restriction, different pulse-widths can be considered as one more criterions to augment the scission-style harmonic elimination scheme. Its relevant constraint is given as follows:$$0 < \alpha_{m_{1}} < \alpha_{m_{2}} < ... < \alpha_{m_{S}} < \frac{\pi}{2}$$ The multi-objective problem will be unravelled with consideration (39)--(41). 4. Description of non-dominated sorting multi objective grey wolf optimization & non-dominated Sorting Genetic Algorithm II {#sec4} =========================================================================================================================== Due to multi-objective nature of the harmonic elimination strategy, the optimization problem has been formulated based on MOGWO to precisely extract the triggering angles of IGBTs\' gates toward enhance the voltage quality of multilevel inverter. To apprise and confirm this optimization scheme, it has been compared with NSGA-II. Both the MOGWO and NSGA-II have been described in following subsections. 4.1. Non-dominated sorting multi objective grey wolf optimization {#sec4.1} ----------------------------------------------------------------- ### 4.1.1. Grey wolf optimization review {#sec4.1.1} GWO is a novel and advanced swarm based optimization algorithm inspired by imitating the headship hierarchy and hunting strategy of grey wolves. The performance of GWO is demonstrated more suitable than other prevalent optimization algorithm such as genetic algorithm and particle swarm optimization \[[@bib43]\]. Moreover, GWO is a high performance optimization technique with respect to its fast convergence (due to its search mechanism) and perceptible mathematical structure \[[@bib44]\]. Better and more precise, all wolves are divided into four groups according to their fitness. GWO search mechanism has been conducted by the best three wolves in any iteration. This mechanism upgrades the exploration so that all engaged wolves to be attracted, then the best three wolves, as a result of that the fast convergence will be obtained. ### 4.1.2. Multi-objective optimization {#sec4.1.2} Multi-objective optimization is functionally constructed for solution of a multi-dimensional problem. The optimization problem formulation as a multi-objective minimization problem can be given by \[[@bib45], [@bib46]\]:$$Minimize\quad F\left( \overset{\rightarrow}{x} \right) = f_{1}\left( \overset{\rightarrow}{x} \right),f_{2}\left( \overset{\rightarrow}{x} \right),...,f_{N}\left( \overset{\rightarrow}{x} \right)$$ Subject to:$$\begin{matrix} {g_{i}\left( \overset{\rightarrow}{x} \right) \leq 0,} & {\text{i} = 1\text{,}2\text{,...,m}} \\ {h_{i}\left( \overset{\rightarrow}{x} \right) = 0,} & {\text{i} = 1\text{,}2\text{,...,p}} \\ {L_{i} \leq x_{i} \leq U_{i},} & {\text{i} = 1\text{,}2\text{,...,n}} \\ \end{matrix}$$Where n, m, p, N are respectively: the number of variables, inequality constraints, equality constraints and objective functions. *g*~*i*~ and *h*~*i*~ are respectively the *i-th* inequality and equality constraints, and \[*L*~*i*~,*U*~*i*~\] indicates the *i-th* variable boundaries. Optimized single-objective results can simply be compared because of the unique objective function. Whereas, the extracted results from a multi-objective space may not be compared according to the relational operators because of multi-criteria objective functions. Hence, multi-objective problem has been unravelled via storing a set of the best solutions in a decision space. This space saves an archival record which is provided for the non-dominated solutions during the search process. The archival space has been initialized and iteratively updated, then, the best solutions are figured out as non-dominated set or Pareto optimal front (surfaces) \[[@bib47]\]. The mathematical description of Pareto dominance is presented to acquire minimum solution: **Statement 1.** Pareto Dominance: Assume the existence of two vectors such as: $\overset{\rightarrow}{\text{x}} = \left( \text{x}_{1}\text{,}\ \text{x}_{2}\text{,}\text{…}\text{,}\ \text{x}_{\text{k}} \right)$ and $\overset{\rightarrow}{\text{y}} = \left( \text{y}_{1}\text{,}\ \text{y}_{2}\text{,}\text{…}\text{,}\ \text{y}_{\text{k}} \right)$. Vector *x* dominates vector *y* i.e., *x*≻*y* if:$$\forall i \in \left\{ {1,2,...,k} \right\}\text{;}\left\lbrack {\text{f}\left( \text{x}_{\text{i}} \right) \geq \text{f}\left( \text{y}_{\text{i}} \right) \land \left\lbrack {\exists i \in 1,2,...,k:f\left( x_{i} \right)} \right\rbrack} \right\rbrack$$ Pareto front (surface) is stated by \[[@bib48]\]: **Statement 2.** Pareto Optimality: A solution $\overset{\rightarrow}{x}$ ε*X* is named Pareto optimal if:$$\left. \exists\overset{\rightarrow}{y} \in X \middle| F\left( \overset{\rightarrow}{y} \right) \succ F\left( \overset{\rightarrow}{x} \right) \right.$$ The non-dominated solutions set is named by Pareto optimal set which is stated by: **Statement 3.** Pareto optimal set: A Pareto-optimal solution set is named Pareto set by:$$P_{s} = \left\{ x,y \in X \middle| \exists F\left( y \right) \succ F\left( x \right) \right\}$$ A set including the relevant objective contents in Pareto optimal set is named Pareto optimal front (surface) which is states by: **Statement 4.** Pareto optimal front (surface): A set including the objective functions\' value with regard to Pareto solutions set:$$P_{f} = \left\{ F\left( x \right) \middle| x \in P_{s} \right\}$$ ### 4.1.3. Grey wolf optimization {#sec4.1.3} This algorithm is inspired by the gregarious headship and hunting strategy of grey wolves. The social hierarchy of wolves is mathematically designed with consideration of the best solution as α wolf. Also, the second and third appropriate solutions are respectively considered as β and $\delta$ wolves. Wolves that have are considered as $\omega$ wolves. Based on GWO, hunting or optimization has been carried out using α, β and $\delta$, withal the $\omega$ wolves have optimally pursued these three wolves. As for the [Figure 6](#fig6){ref-type="fig"}, the siege strategy of grey wolves during the hunting can be presented by:$$\overset{\rightarrow}{D} = \left| {\overset{\rightarrow}{C}.{\overset{\rightarrow}{X}}_{p}\left( t \right) - \overset{\rightarrow}{X}\left( t \right)} \right|$$$$\overset{\rightarrow}{X}\left( {t + 1} \right) = {\overset{\rightarrow}{X}}_{p}\left( t \right) - \overset{\rightarrow}{A}.\overset{\rightarrow}{D}$$where, t, ${\overset{\rightarrow}{X}}_{p}$ and $\overset{\rightarrow}{X}$ are respectively the current generation, the hunt situation vector, and the grey wolf situation vector. $\overset{\rightarrow}{A}$ and $\overset{\rightarrow}{C}$ are coefficient vectors.Figure 6Updating strategy of search operators by the weight.Figure 6 $\overset{\rightarrow}{A}$ and $\overset{\rightarrow}{C}$ are mathematically presented by:$$\overset{\rightarrow}{A} = 2\overset{\rightarrow}{a}.{\overset{\rightarrow}{r}}_{1} - \overset{\rightarrow}{a}$$$$\overset{\rightarrow}{C} = 2{\overset{\rightarrow}{r}}_{2}$$where, $\overset{\rightarrow}{a}$ is linearly subsided from 2 to 0 through the increase of iterations and *r*~*1*~ and *r*~*2*~ indicate the random vectors in \[0,1\]. The first three best solutions are preserved which are attained yet, then other search operators considering the omegas are compelled to improve their situation according to them. The following equations have been continually performed for each search operator during the optimization to construct the hunting mechanism and finding the desirable areas of the search space:$${\overset{\rightarrow}{D}}_{\alpha} = \left| {{\overset{\rightarrow}{C}}_{1}.{\overset{\rightarrow}{X}}_{\alpha} - \overset{\rightarrow}{X}} \right|$$$${\overset{\rightarrow}{D}}_{\beta} = \left| {{\overset{\rightarrow}{C}}_{2}.{\overset{\rightarrow}{X}}_{\beta} - \overset{\rightarrow}{X}} \right|$$$${\overset{\rightarrow}{D}}_{\delta} = \left| {{\overset{\rightarrow}{C}}_{3}.{\overset{\rightarrow}{X}}_{\delta} - \overset{\rightarrow}{X}} \right|$$$${\overset{\rightarrow}{X}}_{1} = {\overset{\rightarrow}{X}}_{\alpha} - {\overset{\rightarrow}{A}}_{1}.\left( {\overset{\rightarrow}{D}}_{\alpha} \right)$$$${\overset{\rightarrow}{X}}_{2} = {\overset{\rightarrow}{X}}_{\beta} - {\overset{\rightarrow}{A}}_{2}.\left( {\overset{\rightarrow}{D}}_{\beta} \right)$$$${\overset{\rightarrow}{X}}_{3} = {\overset{\rightarrow}{X}}_{\delta} - {\overset{\rightarrow}{A}}_{3}.\left( {\overset{\rightarrow}{D}}_{\beta} \right)$$$$\overset{\rightarrow}{X}\left( {t + 1} \right) = \frac{{\overset{\rightarrow}{X}}_{1} + {\overset{\rightarrow}{X}}_{2} + {\overset{\rightarrow}{X}}_{3}}{3}$$ The finding process is approved using A^→^ with random values upper than 1 or lower than -1 that compels the search operator to split from the hunt. C^→^ as accelerating factors in finding operations is randomly created in \[0, 2\], where in random weights has been created for prey to stochastically strengthen (C \> 1) or weaken (C \< 1) to determine the distance using hunt quality. The C parameter is being essentially created a random value to strengthen the finding mechanism through the iterations. It is important factor to avoid getting involved in the local optima minima, particularly in the latest iterations. \|A\|\<1 means that the optimization exploitation initiates, whereas A^→^ is randomly created in \[-1, 1\]. Individuals are diverged from the hunt as \|A^→^\|\>1 and converged towards the hunt as \|A^→^\|\<1. The following situation is located in any situation between its current situation and the hunt situation that boost the search procedure to converge to figured out situation of hunt generated by α, β, and δ solutions. The situation and α solution are come back so that the best solutions to be attained and the satisfied end state to delivered. Finally, the concise computational sages for MOGWO, as flowchart form can be comprehended in [Figure 7](#fig7){ref-type="fig"}.Figure 7Flowchart of multi objective grey wolf optimization.Figure 7 4.2. Non-dominated Sorting Genetic Algorithm-II \[[@bib49]\] {#sec4.2} ------------------------------------------------------------ ### 4.2.1. Non-dominated sorting {#sec4.2.1} ❖For each individual (p) in main population perform:➢Initialize *S*~*p*~*=Ø*. This set encompasses all the individuals which are dominated by p.➢Initialize *n*~*p*~*=0.* This is the number of individuals which dominate p.➢for each individual q in P✓if p dominated q then:•add q to the *S*~*p*~ i.e. *S*~*p*~ = *S*~*p*~ U {g}✓else if q dominates p then•increase the domination counter for p i.e. *n*~*p*~ *= n*~*p*~*+1*✓if *n*~*p*~*=0* i.e. no individuals dominate p therefore p belonging to the first front; adjust the rank of individual p to one i.e *p*~*rank*~*=1.* Improve the first front set via addition of p to front one i.e *F*~*1*~*=F*~*1*~ *U {p}*❖This is performed for all individuals in the main population.❖Initialize the front counter to one*. i=1*❖Subsequent is performed when the *ith* front is non-blank i.e. *F*~*i*~*=Ø*;➢*Q=Ø*; the set of storing the individuals for *(i+1) th* front.➢for each individual *p* in *front F*~*i*~✓for each individual q in *S*~*p*~ (*S*~*p*~ is the set of individuals dominated by *p*)•*n*~*q*~ *= n*~*q*~*-1*, decrease the domination counter for individual q.•if *n*~*q*~ *= 0 subsequently* there is no individuals in the following fronts would dominate *q.* Therefore, set *q*~*rank*~*=i+1.* Update the set *Q* with individual q i.e. *Q=Q U q.*➢Increase the front counter by one.➢Now the set *Q* is the next front and subsequently *F*~*i*~*=Q.* ### 4.2.2. Crowding distance {#sec4.2.2} ❖For each front *F*~*i*~*, n* is the number of individuals.➢For all the individuals, the initialized distance is zero i.e. *F*~*i*~*(d*~*j*~*)=0*, where *j* correlates to the *jth* individual in front *F*~*i*~*.*➢for each objective function m✓Sort the individuals in front *F*~*i*~ according to objective m i.e. *I=sort (F*~*i*~*, m).*✓The boundary values for each individual is defined infinite *F*~*i*~ i.e. *I(d*~*1*~*)=∞* and *I(d*~*n*~*)=∞*✓for *k=2 to (n-1):*$$I\left( d_{k} \right) = I\left( d_{k} \right) + \frac{I\left( k + 1 \right).m - I\left( k - 1 \right).m}{f_{m}^{\max} - f_{m}^{\min}}$$•*I(k).m* is *the* value of the *mth* objective function related to the *kth* individual in I. ### 4.2.3. Genetic operators {#sec4.2.3} #### 4.2.3.1. Simulated binary crossover {#sec4.2.3.1} Simulated binary crossover method is chosen which is presented as follows:$$c_{1,k} = \frac{1}{2}\left\lbrack {\left( {1 - \beta_{k}} \right)p_{1,k} + \left( {1 + \beta_{k}} \right)p_{2,k}} \right\rbrack$$$$c_{2,k} = \frac{1}{2}\left\lbrack {\left( {1 + \beta_{k}} \right)p_{1,k} + \left( {1 - \beta_{k}} \right)p_{2,k}} \right\rbrack$$where, *c*~*i,k*~ is the *ith* child with *kth* part, *p*~*i,k*~ is the chosen parent, and *β*~*k*~ (≥0) is randomly created.$$p\left( \beta \right) = \frac{1}{2}\left( {\eta_{c} + 1} \right)\beta^{\eta_{c}},\mspace{9mu} if\quad 0 \leq \beta \leq 1$$$$p\left( \beta \right) = \frac{1}{2}\left( {\eta_{c} + 1} \right)\frac{1}{\beta^{\eta_{c} + 2}},\mspace{9mu} if\quad 1 < \beta$$ The distribution index (*η*~*c*~) can be acquired using a uniform sampled random number u between (0, 1).$$\beta\left( u \right) = \left( {2u} \right)^{\frac{1}{({\eta + 1})}}$$$$\beta\left( u \right) = \frac{1}{\left\lbrack {2\left( {1 - u} \right)} \right\rbrack^{1/{({\eta + 1})}}}$$ #### 4.2.3.2. Polynomial mutation {#sec4.2.3.2} The polynomial mutation is used which is identified as follows:$$c_{k} = p_{k} + \left( {p_{k}^{u} - p_{k}^{l}} \right)\delta_{k}$$where, *c*~*k*~ and *p*~*k*~ are child and parent, respectively, also $p_{k}^{l}$ and $p_{k}^{u}$ are the lower and upper bounds on the parent components, respectively. *δ*~*k*~ is the small deviation which is defined as follows:$$\delta_{k} = \left( {2r_{k}} \right)^{\frac{1}{({\eta_{m} + 1})}} - 1,\mspace{9mu} if\quad r_{k} < 0.5$$$$\delta_{k} = 1 - \left\lbrack {2\left( {1 - r_{k}} \right)} \right\rbrack^{1/{({\eta_{m} + 1})}},\mspace{9mu} if\quad r_{k} \geq 0.5$$where, *r*~*k*~ is randomly chosen between (0, 1) and *η*~*m*~ is mutation distribution index. ### 4.2.4. Trade-off solution {#sec4.2.4} Choosing an appropriate trade-off solution from all non-inferior options depends on problem and determiner\'s distinction. Therefore, the ultimate solution will be revealed based on both the procedures of optimization and decision. For this study, a Fuzzy-based method is applied to choose the best tradeoff solution from the acquired Pareto set. The *jth* objective function of a solution in a Pareto set *f*~*i*~ is defined by a membership function *μ*~*j*~:$$\mu_{j}\left( x \right) = \begin{Bmatrix} 1 & {f_{j} \leq f_{j}^{\min}} \\ \frac{f_{j}^{\max} - f_{j}}{f_{j}^{\max} - f_{j}^{\min}} & {f_{j}^{\min} \leq f_{j} \leq f_{j}^{\max}} \\ 0 & {f_{j} \geq f_{j}^{\max}} \\ \end{Bmatrix}$$where, *f*~*j*~^*min*^ and *f*~*j*~^*max*^ indicate the minimum and maximum values of the *jth* objective function, respectively. For each solution *i*, the membership function *μ*^*i*^ can be presented by:$$\mu^{i} = \frac{\sum\limits_{j = 1}^{n}\mu_{j}^{i}}{\sum\limits_{i = 1}^{m}{\sum\limits_{j = 1}^{n}\mu_{j}^{i}}}$$ 5. Scission-style harmonic elimination analysis and results {#sec5} =========================================================== This approach has been designed to enhance the quality voltage of odd-nary AMLI. Owing to three prominent voltage quality criterions i.e., desired fundamental component, elimination of third harmonic and reduction of THD, the problem formulation has been constructed based on multi-objective algorithm. Toward this subject, MOGWO method is employed so that both the mentioned targets (minimum value of (36)--(38)) to be acquired. It is worth mentioning that, MOGWO based Harmonic Elimination not only figures out the triggering angles of scission points in each level, but also incorporates the non-equality of DC power supplies\' magnitude and also different pulse-widths. Hereof, the flexibility of this strategy has been heightened using three salient criterions aimed to more acquisition of voltage quality improvement. The quasi-sinusoidal voltage waveform and reference-carriers waveforms that are created according to the scission-style harmonic elimination scheme are presented in [Figure 8](#fig8){ref-type="fig"}(a) and (b), respectively. However, this strategy has been solved in accordance with three following schemes:Figure 8(a). The proposed scission-style harmonic elimination scheme based on: different scission in each level, non-equality of DC sources and different pulse-widths. (b). Reference and carrier waveforms based on scission-style harmonic elimination scheme.Figure 8 Due to heavy computational burden of this approach, aforesaid schemes have been severally carried out. 5.1. Scission-style harmonic elimination scheme considering different scissions {#sec5.1} ------------------------------------------------------------------------------- According to the relevant sequencing presented in (39), the search space of MOGWO is constricted for pre-defined area. Moreover, for scission-style harmonic elimination the angles should be accurately located between these levels. As can be seen in [Figure 5](#fig5){ref-type="fig"}(b) m~1~, m~2~ and m~s~ indicate the number of angles which are available in the first, the second and the latest level, respectively. The angle propagation index has been set as triggering angles to define a set angle in each step of quasi-sinusoidal voltage. This index is presented by m~1~/m~2~/.../m~s~; for example for angle propagation proportion of 6/4/7/5, it has indicated that first, second, third and fourth levels encompass 6, 4, 7 and 5 angles, respectively. Computational burden of scission-style harmonic elimination with presence of the angle propagation index and sequencing has been highly heightened. Also, the number of scission in each step acts as critical function to define the performance of this computation, and accordingly value of THD. Meantime, as the number of scissions (in each level) to be increased, subsequently the value of fundamental component will decrease, and reversely. Thus, the maintenance of fundamental component is one of the prominent issues in this field. Due to destructive effects of third harmonic on normal operation of system, it must be considered as another function. That\'s why MOGWO is engaged to obtain three aforementioned objective functions. The optimization problem has unravelled twenty four angles in each quarter cycle of quasi-sinusoidal waveform of proposed multilevel inverter. Also, the propagation proportion is taken to be 3/5/9/11 along with the modulation index (2.68 \< M \< 3.03). Following that the optimization processes, the Pareto solution surfaces of the optimization problem performed by MOGWO and NSGA-II are portrayed in [Figure 9](#fig9){ref-type="fig"}(a), and the trajectory curve of twenty eight triggering angle are depicted in and 9 (b). As can be seen, the optimization scheme based on MOGWO has provided more accurate and optimum result than NSGA-II. Also, the quasi-sinusoidal voltage and the relevant harmonic spectrum are presented in [Figure 10 (a) and (b)](#fig10){ref-type="fig"}, respectively. To strictly validate these results, their relevant experimental results are presented in in [Figure 10](#fig10){ref-type="fig"}(c) and (d). As consequence, the performance of scission-style harmonic elimination aimed at reduction of the THD, elimination of third harmonic and maintenance of fundamental component (1.7^p.u.^) is proved.Figure 9(a). Pareto solution surfaces of the optimization problem by MOGWO and NSGA-II. (b). Triggering angle trajectories versus values modulation index.Figure 9Figure 10(a). Simulation result of quasi-sinusoidal voltage. (b). Simulation FFT of quasi-sinusoidal voltage. (c). Experimental result of quasi-sinusoidal voltage. (d). Experimental FFT of quasi-sinusoidal voltage.Figure 10 5.2. Scission-style harmonic elimination scheme considering non-equal DC power supplies {#sec5.2} --------------------------------------------------------------------------------------- In accordance with the sequence of (40), the constraint of non-equality of DC power supplies is assigned. Same the previous section, the MOGWO has optimized the magnitude of DC power supplies in order to minimize three defined objective functions. Excluding the effects of triggering angle to carry out the Harmonic Elimination, variety of the DC power sources\' magnitude can be considered as an important criterion for optimization problem. The quasi-sinusoidal voltage and its harmonic spectrum related to the optimum solution are presented in [Figure 11 (a)](#fig11){ref-type="fig"} and (b), respectively. Furthermore, their relevant experimental results are presented in in [Figure 11 (c) and (d)](#fig11){ref-type="fig"}. These figure have clearly shown that third objective-functions have been well-reduced and also fundamental component is held on pre-defined value1.9 ^p.u.^.Figure 11(a). Simulation result of quasi-sinusoidal voltage. (b). Simulation FFT of quasi-sinusoidal voltage. (c). Experimental result of quasi-sinusoidal voltage. (d). Experimental FFT of quasi-sinusoidal voltage.Figure 11 5.3. Scission-style harmonic elimination scheme considering different pulse-width values {#sec5.3} ---------------------------------------------------------------------------------------- In this part, the width of all pulses of quasi-sinusoidal waveform is optimized by MOGWO so that three objective functions to be minimized. The performance of this item in the form of output voltage of inverter and harmonic spectrum are respectively shown in [Figure 12 (a)](#fig12){ref-type="fig"} and (b). Furthermore, their relevant experimental results are presented in in [Figure 12 (c) and (d)](#fig12){ref-type="fig"}. These figures have confirmed that, what accuracy of acquired results from two previous parts is also repeated for this analysis. The difference here is that: the magnitude of fundamental component in considered to be 1.8 ^p.u.^.Figure 12(a). Simulation result of quasi-sinusoidal voltage. (b). Simulation FFT of quasi-sinusoidal voltage. (c). Experimental result of quasi-sinusoidal voltage. (d). Experimental FFT of quasi-sinusoidal voltage.Figure 12 6. Conclusion {#sec6} ============= An innovational cascaded AMLI has been designed and analyzed so that a high-step quasi-sinusoidal voltage with low harmonic components to be acquired. To validate the suggested AMLI capability, it has been thoroughly evaluated and compared with the conventional and recently introduced symmetric and asymmetric multilevel inverters. Sub-AMLI has been designed using the ladder capacitors of clamped to the H-bridge by several bi-directional switches. A creative geometric progression has been formulated for DC power supplies of AMLI to function as odd-nary AMLI. Meanwhile, three defined objective functions related to voltage quality i.e., desired fundamental component, elimination of third harmonic and reduction of THD have been optimally minimized using suggested MOGWO-based scission-style harmonic elimination scheme. To confirm the optimization scheme based on MOGWO, it has been compared with this scheme based on NSGA-II. Finally, the relevant analytical analysis has been essentially confirmed that suggested structure can provide high-step quasi sinusoidal voltage as compared with others. Also, both the simulation and experiment results have clearly corroborated the performance of odd-nary AMLI. Declarations {#sec7} ============ Author contribution statement {#sec7.1} ----------------------------- Ali Darvish Falehi: Conceived and designed the experiments; Performed the experiments; Analyzed and interpreted the data; Contributed reagents, materials, analysis tools or data; Wrote the paper. Funding statement {#sec7.2} ----------------- This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Competing interest statement {#sec7.3} ---------------------------- The authors declare no conflict of interest. Additional information {#sec7.4} ---------------------- No additional information is available for this paper.

Introduction ============ Left atrial (LA) size as assessed by echocardiography has been shown to have prognostic value in patients with dilated cardiomyopathy (DCM). More recently cardiac magnetic resonance (CMR) imaging studies have reported that the presence of left ventricular mid-wall late gadolinium enhancement (LGE), a marker of myocardial fibrosis, also predicts adverse outcome in DCM. The biplane area-length method has been validated as an accurate method for assessment of LA volumes, which echocardiography may underestimate by up to 47% compared with CMR. Purpose ======= We sought to establish whether LA volume as assessed by CMR was associated with the presence of mid-wall LGE in DCM patients. Methods ======= DCM patients were studied between Jan 2006-Aug 2008. Patients with coronary artery disease were excluded. All patients underwent a comprehensive CMR study which included acquisition of LGE images in 2 separate phase-encoding directions to exclude artefact. LA area and length were traced at end-ventricular systole and diastole in both horizontal and vertical long axis images. Maximum (LA max) and minimum (LA min) LA volumes were then calculated using the biplane area-length method for ellipsoid bodies and indexed to BSA. A separate cardiologist who was blinded to the results of the LA area determined the presence of mid-wall LGE. Results ======= One hundred and ninety two patients (152 male; age 50.7 years ± 14.7 years (mean±SD); LV EF 40% ± 15%, were recruited to the study. The overall mean indexed LA max was 54.5 (95% CI: 51.7-57.5) and mean indexed LA min 29.8 (95%CI: 23.7-27.1). Patients with concurrent mid-wall LGE had significantly larger indexed LA max volumes than those without LGE (56.2 ml (25.6 ml - 60.9 ml) vs. 49.7 ml (45.5 ml -54.2 ml) (mean (95%CI), p = 0.03)). Patients with concurrent LGE also had significantly larger indexed LA min volumes than those without (31.9 ml (28.6 ml - 35.7 ml) vs. 24.3 ml (20.7 ml - 28.5 ml) respectively; p \< 0.01). The differences remained statistically significant when patients with atrial fibrillation were excluded from analysis. Conclusion ========== LA volumes were significantly greater in DCM patients with left ventricular mid-wall LGE compared to those with no LGE. The presence of mid-wall LGE may therefore be associated with increased left atrial pressures in DCM patients. Further studies are therefore required to evaluate whether LA size has incremental prognostic value to midwall LGE in DCM.

All relevant data are within the paper and its Supporting Information files. Introduction {#sec005} ============ Lung transplantation is an established therapeutic option for patients with various types of end-stage lung disease. Internationally, the annual number of bilateral lung transplants (BLTs) is greater than that of single lung transplants (SLTs) because of the better survival with the former procedure \[[@pone.0210975.ref001]\]. On the other hand, the current situation in Japan is completely different from other countries. The number of SLTs performed in Japan is almost the same as the number of BLTs. Miyoshi et al. reported that, given the severe donor shortage in Japan, SLT can be the first choice of surgical procedure type with acceptable outcomes if there is no contraindication to SLT \[[@pone.0210975.ref002]\]. The Toronto lung transplant group conducted cadaveric SLT for a patient with pulmonary fibrosis and achieved long-term survival as the first successful lung transplantation case in the world in 1983 \[[@pone.0210975.ref003]\]. Mal et al. successfully performed cadaveric SLT for a patient with chronic obstructive pulmonary disease (COPD)\[[@pone.0210975.ref004]\]. COPD had been considered as an indication for SLT after this successful case. However, overinflation started to be recognized as one of the most serious complications after SLT for COPD. In SLT recipients for COPD, the overinflation of the native lung is thought to occur due to increased compliance, whereas the transplanted lung graft on the other side has normal compliance. Pneumonectomy of the native lung \[[@pone.0210975.ref005]\], lung volume reduction surgery (LVRS) \[[@pone.0210975.ref006]\], and bronchoscopic lung volume reduction (BLVR) \[[@pone.0210975.ref007]\] were reported as therapeutic modalities for overinflation of the native lung. Lymphangioleiomyomatosis (LAM) is a rare cystic lung disease that develops primarily in women of childbearing age and is characterized by the proliferation of abnormal smooth muscle-like cells (LAM cells). LAM is associated with an obstructive disease pattern and reduced diffusion capacity of the lung for carbon monoxide (DLco) determined by a pulmonary function test, the main reason for which is the increased lung compliance \[[@pone.0210975.ref008]\]. Due to the different compliance between the native and the transplanted lungs, the native lung overinflation can be a serious complication after SLT for LAM, that is likely to occur after SLT for COPD. Liu et al. reported a case of pneumonectomy of an overinflated native lung after SLT for LAM that resulted in improved lung function \[[@pone.0210975.ref009]\]. CT scans and chest X-ray often show the mediastinum shifted to the transplanted lung side in recipients of SLT for LAM. Whereas native lung overinflation has been thought to happen in recipients of SLT for LAM because of its increased compliance, there is no study that has reported the details on the change of the native lung volume after SLT by three-dimensional computed tomography (3D-CT) volumetry. One of the advantages of 3D-CT volumetry is that we can quantitatively analyze the lung volume in a non-invasive manner and determine changes in the lung over time. We observe that some recipients of SLT for LAM deteriorate pulmonary function with findings of the mediastinum shifted to the transplanted lung side. It is speculated that the overinflation of the native lung affects the pulmonary function after SLT for LAM. However, there is currently no established way to objectively diagnose the overinflation of the native lung. The purpose of the present study was to evaluate the lung volume after SLT for LAM by 3D-CT volumetry and investigate the correlation between the native lung volume change and postoperative pulmonary function in the recipient. Patients and methods {#sec006} ==================== Patients {#sec007} -------- Between January 2006 and December 2015, 25 patients underwent SLT for LAM at Tohoku University Hospital. Eight of these patients were excluded from this study: 3 patients died of primary graft dysfunction in the perioperative period; 1 patients died of enteritis at 6 months after SLT; 1 patients developed deformation of the thorax and received retransplantation within 4 years after SLT; 3 patients developed chronic lung allograft dysfunction (CLAD) of, at least, BOS 0-p.\[[@pone.0210975.ref010]\] Thus, we analyzed the results of 17 patients who had undergone SLT and survived for more than 3 years. Pulmonary function test (PFT) data was available in 17 patients at post-transplant 1 year, 17 patients at 2 post-transplant years, 17 patients at 3 post-transplant years, 15 patients at post-transplant 4 years and 10 patients at post-transplant 5 years. In 9 patients, PFT data and low attenuation volume (LAV) analysis (described later) were available every year up to post-transplant 5 years. The Institutional Review Board of Tohoku University Hospital approved the study (approval No. 2017-1-021) and research was conducted in accordance with the 2000 Declaration of Helsinki. Written informed consent have been obtained from all participants. Image acquisition and measurement of lung volume by three-dimensional computed tomography volumetry {#sec008} --------------------------------------------------------------------------------------------------- CT images were obtained during a single respiratory pause at the end of maximum inspiratory effort using a multidetector row CT scanner (BrightSpeed Elite, GE Healthcare Japan Ltd, Tokyo, Japan). Whole lung scans were performed at a peak tube voltage of 120 kVp, with a variable mAs setting using an automatic exposure control system. CT data from 1-2-mm slices were used for the volumetric analysis. Each patient's CT data were transferred to a standalone workstation (Ziostation2, Ziosoft, Inc., Tokyo, Japan) and 3D models were reconstructed. Left and right lung volumes were calculated by summing the voxels of the CT value of the window level (WL), −469 Hounsfield units (HU); window width (WW), 684.8 HU; sharpness (SH), 0. The respiratory tract from the trachea to subsegmental bronchi was reconstructed and automatically excluded from the lung volumes. [Fig 1](#pone.0210975.g001){ref-type="fig"} shows pre- and post-transplant 3D-CT images of a case of SLT for LAM. We manually measured the lung volumes with the same values of WL, WW, and SH when they could not be measured automatically. In the case of manual measuring, two radiologists measured the lung volumes in a blinded manner and the mean value was used for the study. The data of PFT and 3D-CT volumetry at each time point after SLT were collected. {#pone.0210975.g001} Evaluation of overinflation of the native lung {#sec009} ---------------------------------------------- Currently, there is no standard method to evaluate the degree of overinflation of the native lung after SLT. We defined the ratio of the native lung volume to total lung volume (N/T ratio) as an indicator of overinflation of the native lung. The N/T ratio was calculated as follows: $$\text{N}/\text{T}\mspace{720mu}\text{ratio} = \text{native}\mspace{720mu}\text{lung}\mspace{720mu}\text{volume}/\text{total}\mspace{720mu}\text{lung}\mspace{720mu}\text{volume}\mspace{720mu} \times \mspace{720mu} 100\mspace{720mu}\left( \% \right)$$ In order to assess changes of the N/T ratio over time, we calculated the rate of change in the N/T ratio, which is standardized by the N/T ratio at 1 year after SLT. In the present study, we call this parameter, the 'rate of change in N/T ratio' which was calculated as follows: $$\text{Rate}\mspace{720mu}\text{of}\mspace{720mu}\text{change}\mspace{720mu}\text{in}\mspace{720mu}\text{N}/\text{T}\mspace{720mu}\text{ratio}\mspace{720mu}\left( \% \right) = \left\{ {\left( {\text{N}/\text{T}\mspace{720mu}\text{ratio}\mspace{720mu}\text{at}\mspace{720mu}\text{a}\mspace{720mu}\text{certain}\mspace{720mu}\text{year}\mspace{720mu}\text{after}\mspace{720mu}\text{SLT}} \right)/\left( {\text{N}/\text{T}\mspace{720mu}\text{ratio}\mspace{720mu}\text{at}\mspace{720mu} 1\mspace{720mu}\text{year}\mspace{720mu}\text{after}\mspace{720mu}\text{SLT}} \right)–1} \right\} \times 100.$$ Rate of change in forced expiratory volume in 1 second %predicted (%FEV1) was also calculated in the same way when we investigated the correlation between the rate of change in N/T and %FEV1. Measurement of low attenuation volume: Evaluation of emphysematous lesions {#sec010} -------------------------------------------------------------------------- LAV was calculated by summing the voxels with CT values of less than --950 HU. LAV was reported to be the volume of emphysematous lesions in COPD patients\[[@pone.0210975.ref011]\]. The percentage of LAV to total lung volume is considered to reflect the severity of COPD of the lung \[[@pone.0210975.ref011]\]. In the present study, in order to evaluate emphysematous lesions in the native lung, the percentage of LAV in the native lung was calculated as follows: $$\text{Native}\mspace{720mu}\text{lung}\mspace{720mu}\text{LAV}\% = \text{native}\mspace{720mu}\text{lung}\mspace{720mu}\text{LAV}/\text{native}\mspace{720mu}\text{lung}\mspace{720mu}\text{volume} \times 100.$$ $$\text{Rate}\mspace{720mu}\text{of}\mspace{720mu}\text{change}\mspace{720mu}\text{in}\mspace{720mu}\text{native}\mspace{720mu}\text{lung}\mspace{720mu}\text{LAV}\%{\mspace{360mu}\left( \% \right)} = \left\{ {\left( {\text{native}\mspace{720mu}\text{lung}\mspace{720mu}\text{LAV}\%\mspace{720mu}\text{at}\mspace{720mu}\text{a}\mspace{720mu}\text{certain}\mspace{720mu}\text{year}\mspace{720mu}\text{after}\mspace{720mu}\text{SLT}} \right)/\left( {\text{native}\mspace{720mu}\text{lung}\mspace{720mu}\text{LAV}\%\mspace{720mu}\text{at}\mspace{720mu} 1\mspace{720mu}\text{year}\mspace{720mu}\text{after}\mspace{720mu}\text{SLT}} \right)–1} \right\} \times 100.$$ Evaluation of size matching {#sec011} --------------------------- Under the current lung allocation system in Japan, lungs are allocated in accordance with size matching, which is calculated using predicted vital capacities (pred VCs) of the donor and the recipient. Pred VC and size matching were calculated as follows: $$\text{Pred}\mspace{720mu}\text{VC}\mspace{720mu}\text{for}\mspace{720mu}\text{male}\mspace{720mu}\left( \text{L} \right) = 0.045 \times \text{height}\mspace{720mu}\left( \text{cm} \right) - 0.023 \times \text{age} - 2.258$$ $$\text{Pred}\mspace{720mu}\text{VC}\mspace{720mu}\text{for}\mspace{720mu}\text{female}{\mspace{360mu}\left( \text{L} \right)} = 0.032 \times \text{height}\mspace{720mu}\left( \text{cm} \right) - 0.018 \times \text{age} - 1.178$$ $$\text{Size}\mspace{720mu}\text{matching}\mspace{720mu}\left( \% \right) = \text{pred}\mspace{720mu}\text{VC}\mspace{720mu}\text{of}\mspace{720mu}\text{donor}/\text{pred}\mspace{720mu}\text{VC}\mspace{720mu}\text{of}\mspace{720mu}\text{recipient} \times 100.$$ Postoperative assessment of pulmonary function {#sec012} ---------------------------------------------- Postoperative PFTs were performed using CHESTAC-8800 or 8900 (Chest Ltd., Tokyo, Japan) to measure VC, forced vital capacity (FVC), forced expiratory volume in 1 second (FEV1), DLco, and the DLco/alveolar volume (VA). The 6-minute walking distance (6MWD) (m) was measured according to the American Thoracic Society guidelines. Arterial blood gas (ABG) analysis was performed by RADIOMETER ABL 800 FLEX. Alveolar---arterial gradient (A-aDO~2~) was calculated as follows: $$\text{A} - \text{aDO}_{2} = 713 \times 0.21 - \text{PaCO}_{2}/0.8 - \text{PaO}_{2}.$$ Statistical analysis {#sec013} -------------------- All statistical analyses were performed using JMP, version 13, for MAC (SAS Institute Inc.; Cary, NC). All data are presented as the means ± SD or numbers of patients. The differences between groups were analyzed using Student's *t*-test for continuous variables. We used ANOVA for multi-group analysis and the paired t-test to analyze paired groups. Linea regression was used to describe associations between selected parameters. *P* \< 0.05 was considered significant. Results {#sec014} ======= Recipient and donor characteristics and size matching of the lung {#sec015} ----------------------------------------------------------------- [Table 1](#pone.0210975.t001){ref-type="table"} shows characteristics of the donors and the recipients, and the size matching of the lungs. As mentioned in methods, the recipients were allocated the lungs in accordance with size matching; therefore, height and age were similar between the recipients and the donors. 10.1371/journal.pone.0210975.t001 ###### Recipient and donor characteristics and size matching of the lung. {#pone.0210975.t001g} Variable Donor (n = 17) Recipient (n = 17) -------------------------------------------------------------------- ------------------------ -------------------- Sex (male/female) 3 (17.6%)/14 (82.4%) 0/17 Height (cm) 156.0 ± 8.6 156.7 ± 5.6 Age 44.4 ± 12.6 46.6 ± 7.7 Predicted vital capacity (ml)[\*](#t001fn002){ref-type="table-fn"} 2901.4 ± 496.4 2827.4 ± 212.9 Size matching of lung (%)[^†^](#t001fn003){ref-type="table-fn"} 108.2 ± 18.9 (87--147) Data are expressed as group mean ± standard deviation or number (%). \* Predicted vital capacity for male (L) = 0.045 × height (cm) − 0.023 × age − 2.258, predicted vital capacity for female (L) = 0.032 × height (cm) − 0.018 × age − 1.178. ^†^ Size matching (%) = predicted vital capacity of recipient / predicted vital capacity of donor × 100. Surgical and postoperative characteristics {#sec016} ------------------------------------------ Nine of 17 cases (52.9%) were right SLT. The duration of mechanical ventilation was 6.9 ± 6.5 days. The intensive care unit stay was 15.5 ± 9.6 days ([S1 Table](#pone.0210975.s001){ref-type="supplementary-material"}). Lung volume after SLT {#sec017} --------------------- The total lung volume, the native lung volume and the transplanted lung volume measured by 3D-CT volumetry are shown in [Fig 2A--2C](#pone.0210975.g002){ref-type="fig"}. In 3 of 17 cases (17.6%), we manually measured the lung volumes because they could not be measured automatically. The native lung volume and N/T ratio showed a tendency to increase; however, they did not differ significantly among the years. The transplanted lung volume did not show any difference ([Fig 2B and 2D](#pone.0210975.g002){ref-type="fig"}). On the other hand, the rate of change in the N/T ratio increased over time and reached a significant increase at 4 and 5 years after SLT ([Fig 3A](#pone.0210975.g003){ref-type="fig"}). The rate of change in the N/T ratio in each case is presented in [Fig 3B](#pone.0210975.g003){ref-type="fig"}. Seven of 9 cases that we were able to follow the lung volume and PFT available every year showed the increase of the rate of change in the N/T ratio over year ([Fig 3B](#pone.0210975.g003){ref-type="fig"}). {#pone.0210975.g002} {#pone.0210975.g003} Native lung LAV, transplanted lung LAV and native lung LAV% after SLT {#sec018} --------------------------------------------------------------------- The native and the transplanted lung LAV showed no significant change among the years ([Fig 4A](#pone.0210975.g004){ref-type="fig"}). The transplanted lung LAV at each year was much smaller than the native lung LAV ([Fig 4B](#pone.0210975.g004){ref-type="fig"}). The native lung LAV% showed no significant change over time ([Fig 4C](#pone.0210975.g004){ref-type="fig"}). The rate of change in the native lung LAV% showed no significant change over time ([Fig 5A](#pone.0210975.g005){ref-type="fig"}). The rate of change in the native lung LAV% in each case is presented in [Fig 5B](#pone.0210975.g005){ref-type="fig"}. Six of 9 cases that we were able to follow LAV and PFT available every year showed the increase of rate of change in the native lung LAV% over year ([Fig 5B](#pone.0210975.g005){ref-type="fig"}). The 6 patients also showed the increase of rate of change in the N/T ratio over year. {#pone.0210975.g004} {#pone.0210975.g005} PFT, 6MWD and ABG after SLT {#sec019} --------------------------- %FEV1 showed tendency to decrease; however, they did not differ significantly among the years. Other PFT parameters also showed no significant difference among the years ([S2 Table](#pone.0210975.s002){ref-type="supplementary-material"}). Similarly, 6MWD and all parameters of ABG analysis showed no significant difference among the years ([S2 Table](#pone.0210975.s002){ref-type="supplementary-material"}). Correlations between N/T ratio and PFT or ABG parameters or 6MWD after SLT {#sec020} -------------------------------------------------------------------------- [Fig 6A](#pone.0210975.g006){ref-type="fig"} shows the correlation between N/T ratio and %FEV1 at 1 year after SLT. There was no significant correlation between them. We investigated the correlations between the N/T ratio and other PFT or ABG parameters or 6MWD at 1 year; however, there was no significant correlation between them ([S1A](#pone.0210975.s003){ref-type="supplementary-material"}, [S2A](#pone.0210975.s004){ref-type="supplementary-material"} and [S3A](#pone.0210975.s005){ref-type="supplementary-material"} Figs). Similarly, we examined the correlations between the N/T ratio and PFT or ABG parameters or 6MWD at 5 years after SLT. There was no significant correlation. ([Fig 6B](#pone.0210975.g006){ref-type="fig"} shows the correlations between the N/T ratio and %FEV1 at 5 years after SLT. The rest of the data at 5 years are shown in [S1B](#pone.0210975.s003){ref-type="supplementary-material"}, [S2B](#pone.0210975.s004){ref-type="supplementary-material"} and [S3B](#pone.0210975.s005){ref-type="supplementary-material"} Figs). {#pone.0210975.g006} Correlations between rate of change in N/T ratio and that in PFT parameters after SLT {#sec021} ------------------------------------------------------------------------------------- As shown in [Fig 3A](#pone.0210975.g003){ref-type="fig"}, the rate of change in the N/T ratio showed a significant increase over time. Therefore, we investigated correlations between the rate of change in the N/T ratio and that in PFT parameters at 5 years after SLT. There was a significant negative correlation between the rate of change in the N/T ratio and that in %FEV1 ([Fig 7](#pone.0210975.g007){ref-type="fig"}). The rate of change in the N/T did not show a significant correlation with the rate of change in other PFT parameters: %FVC, %DLco, DLco/VA ([S4](#pone.0210975.s006){ref-type="supplementary-material"}--[S6](#pone.0210975.s008){ref-type="supplementary-material"} Figs). {#pone.0210975.g007} Correlations between rate of change in native lung LAV% and that in PFT parameters after SLT {#sec022} -------------------------------------------------------------------------------------------- We also investigated correlations between the rate of change in native lung LAV% and that in PFT parameters after SLT in the same way of change in N/T ratio mentioned above. There was no significant correlation between the rate of change in native lung LAV% and that in any PFT parameters ([S7](#pone.0210975.s009){ref-type="supplementary-material"}--[S10](#pone.0210975.s012){ref-type="supplementary-material"} Figs). Discussion {#sec023} ========== Overinflation of the native lung is one of the challenging complications after SLT for obstructive lung diseases, such as COPD. In such patients, the compliance of the native lung is much higher than that of the healthy lung, transplanted lung. As a result of the difference in the compliance between the native and transplanted lungs, overinflation can occur only in the native lung. We recognize overinflation of the native lung after SLT in COPD by the shifted mediastinum in CT and chest X-ray. LVRS \[[@pone.0210975.ref006]\] and BLVR \[[@pone.0210975.ref007]\] have been applied for this complication in SLT recipients for COPD. Some SLT recipients for LAM also show a shift of the mediastinum to the transplanted lung side. However, it still remains unclear if overinflation of the native lung really happens in such patients. Another question we have in daily practice is whether overinflation of the native lung affects the post-transplant pulmonary function of the SLT recipients for LAM. Nowadays, preoperative 3D-CT volumetry is utilized for lung cancer patients to predict the postoperative pulmonary function \[[@pone.0210975.ref012]\]. 3D-CT volumetry has also been commonly applied for the assessment of size matching in lung transplantation \[[@pone.0210975.ref013]--[@pone.0210975.ref015]\]. However, there is few studies that validated the usefulness of 3D-CT volumetry for the assessment of overinflation of the native lung after SLT. In the present study, we found that, whereas the mean of the native lung volume did not show a significant increase, the rate of change in the N/T ratio significantly increased over time after SLT for LAM. Generally, pulmonary function stabilizes at 1 year after lung transplantation. It was reported that the lung function of recipients improved after SLT and then plateaued by 1 year post-transplantation \[[@pone.0210975.ref016]\]. Therefore, when we evaluated the rate of change of PFT parameters and N/T ratio, we standardized them by the values at 1 year after SLT. We demonstrated that 8 of 10 SLT recipients for LAM have increased the rate of change in the N/T over year that will probably cause the overinflation of the native lung in the future ([Fig 7](#pone.0210975.g007){ref-type="fig"}). In order to assess the influence of the change in the N/T, we investigated the correlation between the N/T ratio and each PFT parameter at 1 and 5 years. We did not find any correlation between N/T ratio and each PFT parameter at either 1 or 5 years. On the other hand, there was a significant negative correlation between the rate of change in the N/T ratio and that in %FEV1 at 5 years after SLT. According to this result, if the N/T ratio during a long period---from postoperative 1 to 5 years---increases in a SLT recipient for LAM, %FEV1 will become worse. Indeed, [Fig 7](#pone.0210975.g007){ref-type="fig"} shows that 6 recipients with a positive rate of change in N/T ratios showed negative a rate of change in %FEV1. Whereas we revealed that the rate of change in the N/T affects the pulmonary function in recipients of SLTs for LAM, it remains unclear whether the overinflation of the native lung would really happen and finally cause end-stage respiratory failure in the future. We will have to follow up the recipients in the present study for a longer time in order to clarify wheter VRS or BLVR can be offered as an option for recipients with end-stage respiratory failure after SLT for LAM. It has been reported that by evaluating the low attenuation area (LAA (%)), we can predict mortality \[[@pone.0210975.ref017]\] and exacerbations \[[@pone.0210975.ref018]\] of emphysema in COPD patients, and by evaluating LAV, we can differentiate subtypes of COPD \[[@pone.0210975.ref019]\] and predict the incidence of complications after surgery for lung cancer \[[@pone.0210975.ref011]\]. LAA is also utilized for differential diagnosis including LAM and other diseases \[[@pone.0210975.ref020]\]. We measured LAV from 1 to 5 years after SLT to evaluate the change in the cystic region. As shown in [Fig 3A](#pone.0210975.g003){ref-type="fig"}, LAV did not differ significantly from 1 to 5 years. This demonstrates that the cystic regions of LAM did not progress in the recipients in the present study and the decrease of %FEV1 was not associated with the progression of LAM. This study has some limitations. First, the N/T ratio was originally defined in this study as an indicator of native lung overinflation. We are still not sure that this indicator really reflects the native lung overinflation. However, in previous reports, overinflation of the native lung was defined based on a qualitative assessment, such as radiological mediastinal shift, and so we had to set a novel quantitative indicator. We think that the recipient with increased rate of change in the N/T in the present study would have overinflated native lung and suffer from respiratory failure in the future. To support the clinical importance of the N/T ratio, the long-term observation and accumulation of cases will be necessary in the future. Second, this study is subject to the bias of a retrospective study because we excluded CLAD cases from this study. It may be difficult to differentiate disorders of the transplanted lung such as CLAD from the native lung overinflation when %FEV1 decreases after SLT for LAM. The ISHLT report in 2014 described 138 LAM patients who received SLT out of a cumulative number of 15,321 cases \[[@pone.0210975.ref001]\]. The present study is of marked significance as it evaluated 25 patients who received SLT for LAM in a single institute. In Japan, SLT is an important surgical procedure because of the chronic shortage of donors. However, this study is limited by the fact that it was a single-center, retrospective study involving a small group of patients. In conclusion, the rate of change in the N/T ratio occurs in some recipients of SLT for LAM during the long term and it may affect post-transplant pulmonary function. Supporting information {#sec024} ====================== ###### Surgical and postoperative characteristics. (DOCX) ###### Click here for additional data file. ###### Pulmonary function test, 6-minute walking distance and arterial blood gas analysis after single lung transplantation. (DOCX) ###### Click here for additional data file. ###### Correlations between native lung volume/total lung volume ratio and diffusion capacity of the lung for carbon monoxide %predicted. \(A\) There was no significant correlation between native lung volume/total lung volume (N/T) ratio and diffusion capacity of the lung for carbon monoxide %predicted (%DLco) at 1 year after single lung transplantation (SLT). (B) N/T ratio did not show a significant correlation with %DLco at 5 years after SLT. (TIFF) ###### Click here for additional data file. ###### Correlations between native lung volume/total lung volume ratio and alveolar--arterial gradient. \(A\) There was no significant correlation between native lung volume/total lung volume (N/T) ratio and alveolar---arterial gradient (A-aDO~2~) at 1 year after single lung transplantation (SLT). (B) N/T ratio did not show a significant correlation with A-aDO~2~ at 5 years after SLT. (TIFF) ###### Click here for additional data file. ###### Correlations between native lung volume/total lung volume ratio and 6-minute walking distance. \(A\) There was no significant correlation between native lung volume/total lung volume (N/T) ratio and 6-minute walking distance (6MWD) at 1 year after single lung transplantation (SLT). (B) N/T ratio did not show a significant correlation with 6MWD at 5 years after SLT. (TIFF) ###### Click here for additional data file. ###### Correlations between rate of change in native lung volume/total lung volume ratio and change in forced vital capacity %predicted. The rate of change in the native lung volume/total lung volume (N/T) did not show a significant correlation with the rate of change in forced vital capacity %predicted (%FVC) at 5 year after single lung transplantation (SLT). (TIFF) ###### Click here for additional data file. ###### Correlations between rate of change in native lung volume/total lung volume ratio and change in diffusion capacity of the lung for carbon monoxide %predicted. The rate of change in the native lung volume/total lung volume (N/T) did not show a significant correlation with the rate of change in diffusion capacity of the lung for carbon monoxide %predicted (%DLco) at 5 year after single lung transplantation (SLT). (TIFF) ###### Click here for additional data file. ###### Correlations between rate of change in native lung volume/total lung volume ratio and change in diffusion capacity of the lung for carbon monoxide/alveolar volume %predicted. The rate of change in the native lung volume/total lung volume (N/T) did not show a significant correlation with the rate of change in diffusion capacity of the lung for carbon monoxide/alveolar volume %predicted (%DLco/VA) at 5 year after single lung transplantation (SLT). (TIFF) ###### Click here for additional data file. ###### Correlations between rate of change in low attenuation volume and change in forced expiratory volume in 1 second %predicted. The rate of change in low attenuation volume (LAV) did not show a significant correlation with the rate of change in forced expiratory volume in 1 second %predicted (%FEV1) at 5 year after single lung transplantation (SLT). (TIFF) ###### Click here for additional data file. ###### Correlations between rate of change in low attenuation volume and change in forced vital capacity %predicted. The rate of change in low attenuation volume (LAV) did not show a significant correlation with the rate of change in forced vital capacity %predicted (%FVC) at 5 year after single lung transplantation (SLT). (TIFF) ###### Click here for additional data file. ###### Correlations between rate of change in low attenuation volume and change in diffusion capacity of the lung for carbon monoxide %predicted. The rate of change in low attenuation volume (LAV) did not show a significant correlation with the rate of change in diffusion capacity of the lung for carbon monoxide %predicted (%DLco) at 5 year after single lung transplantation (SLT). (TIFF) ###### Click here for additional data file. ###### Correlations between rate of change in low attenuation volume and change in diffusion capacity of the lung for carbon monoxide/alveolar volume %predicted. The rate of change in low attenuation volume (LAV) did not show a significant correlation with the rate of change in diffusion capacity of the lung for carbon monoxide/alveolar volume %predicted (%DLco/VA) at 5 year after single lung transplantation (SLT). (TIFF) ###### Click here for additional data file. ###### "Values-for-Tables-and-Figs". This file contains the full underlying data set for all Tables and Figs. (XLSX) ###### Click here for additional data file. The authors thank Katsunori Ono and other radiological technologists of Tohoku University Hospital for their assistance with the collection of data. The authors would like to express their gratitude to Brent Bell for assistance in editing this manuscript. [^1]: **Competing Interests:**The authors have declared that no competing interests exist.

Introduction {#S0001} ============ The health and progress of any society largely depends on the health of its women. Women experience many biological changes throughout their life that have a significant impact on their health, including pregnancy and lactation.[@CIT0001] Pregnancy changes women's lifestyle and they should at least begin to lead a healthy lifestyle and perform health-promoting behaviors during this critical period in order to avoid problems that could harm themselves or the embryo.[@CIT0002] Health-promoting behaviors during pregnancy reduce the likelihood of preterm delivery, the need for cesarean section and the risk of obesity and diabetes.[@CIT0003] Failure in observing such behaviors can lead to complications during pregnancy, such as bleeding and maternal infection, multiple admissions to intensive care units, low birth weight or early neonatal death.[@CIT0001] Since providing maternal and newborn health services is one of the priorities of health systems,[@CIT0004] various strategies have been considered by healthcare providers to ensure the health of pregnant women, which involves health-promoting behaviors and a healthy lifestyle.[@CIT0005] Given women's role in maintaining their family's health, these behaviors are important for all members of the society, especially women.[@CIT0006],[@CIT0007] According to statistics provided by the WHO, 60% of people's quality of life and health status depends on their own behaviors and lifestyle.[@CIT0008],[@CIT0009] According to the literature, "a health-promoting lifestyle is a multi-dimensional pattern of self-initiated feelings and behaviors aiming at ensuring individual's health, self-actualization, and self-accomplishment.[@CIT0010],[@CIT0011]" These behaviors include any measures taken to maintain and enhance the health of an individual or a group.[@CIT0012],[@CIT0013] Health-promoting behaviors should be further emphasized because the promotion of health in society is a dynamic process of empowering individuals to control their health based on first-grade preventive interventions and is focused on positive lifestyle changes.[@CIT0014],[@CIT0015] Pregnant women's lifestyle includes the way they work and rest, their type of nutrition, their manner of coping with stress or communicating with others and also prenatal care.[@CIT0008] The factors affecting health-promoting behaviors should be determined in pregnant women in order to promote their healthy behaviors.[@CIT0007] Multiple behavior change theories and models have been proposed by researchers due to the complexity of behaviors and the challenges of creating, maintaining and improving health-promoting behaviors.[@CIT0016] Some theories and models have identified the most important multi-faceted factors affecting a behavior and the relationship between these factors.[@CIT0017] Pender's health promotion model is one of the comprehensive models used for accomplishing health-promoting behaviors[@CIT0015] and has been recognized as a framework for explaining healthy lifestyle and health-promoting behaviors. This model serves as a guide in discovering individuals' complex biological-psychological processes in order to promote their health behaviors and explains how individuals make decisions about their health-promoting behaviors.[@CIT0011],[@CIT0018] The studies conducted by Pender et al[@CIT0019],[@CIT0020],[@CIT0021] have revealed the dimensions of health-promoting behaviors as spiritual growth, health responsibility, physical activity, nutrition, interpersonal relations, and stress management. Given the importance of women's status in society and their role in the overall development of the country, appropriate planning and policy-making for improving women's status and facilitating their progress should take account of their health status and identify its contributing factors.[@CIT0022] This study was conducted based on Pender's health promotion model to investigate the predictors of pregnant women's health-promoting lifestyles, since it is necessary to study and identify the factors affecting these behaviors separately for each group of the community.[@CIT0023] The results of this study might work as a basis for ensuring a higher quality of life in this vulnerable group of society. Methods {#S0002} ======= This descriptive research was designed to determine the predictors of a health-promoting lifestyle in pregnant women based on the constructs of Pender's health promotion model and examine the correlation between the model constructs and lifestyle. The sample included pregnant women in Yazd Province who were selected from 2018 to 2019. The sample size was obtained as 259 using the sample size equation, type-I error probability of α=0.05, test power of 1-β=0.90 and Pearson's correlation coefficient of r=0.20; ultimately, 300 individuals were included in the study to take account of a 15% probability of sample loss. After making the necessary arrangements, the samples were selected through simple random sampling from pregnant women in the second and third trimesters of their pregnancy, covered by one of the community health centers of Yazd Province. The pregnant women with problems such as high-risk pregnancy (bleeding, membrane rupture, preeclampsia or eclampsia and gestational diabetes with insulin injections) or pre-pregnancy physical problems were excluded from the study. Before completing the questionnaires, the participants were briefed on the research objectives and gave their written consent. The research was carried out in accordance with the Declaration of Helsinki. The participants were carefully monitored during their completion of the questionnaires. Data were collected using a demographic questionnaire, the Health-Promoting Lifestyle Profile II (HPLP-II) and a questionnaire based on Pender's model constructs. Demographic Questionnaire {#S0002-S2001} ------------------------- The demographic questionnaire inquired about variables such as age, the woman's education and employment status, the husband's education and employment status, gestational age and the woman's BMI. Health-Promoting Lifestyle Profile II (HPLP-II) {#S0002-S2002} ----------------------------------------------- Health-promoting behaviors were measured using the HPLP-II, designed by Pender et al (as described in ref.[@CIT0001]). This questionnaire consists of 52 items and measures health-promoting lifestyle behaviors within six dimensions: Physical activity (eight items), nutrition (nine), spiritual growth (nine), interpersonal relations (nine), stress management (eight) and health responsibility (nine items). The questionnaire items are scored based on a 4-point Likert scale: Never (1 point), sometimes (2 points), often (3 points) and always (4 points). The mean score of each dimension is calculated, and the total lifestyle score is then calculated by adding these scores. Higher scores indicate more favorable behaviors. The validity and reliability of this instrument have been confirmed in numerous studies.[@CIT0009],[@CIT0020],[@CIT0024],[@CIT0025] This questionnaire has also been translated in Iran, and the validity and reliability of the Persian version have been confirmed in several studies.[@CIT0026],[@CIT0027] Questionnaire Based on Pender's Model Constructs {#S0002-S2003} ------------------------------------------------ Pender's health promotion model was examined in this study according to Mohammadian's research,[@CIT0018] which was conducted on the constructs of perceived self-efficacy, affect, social support and benefits, and barriers. This study used the perceived self-efficacy construct of the Health Locus of Control Scale adapted by Smith et al. The Health Locus of Control Scale consists of eight items scored based on a 5-point Likert scale, ranging from strongly disagree (1 point) to strongly agree (5 points). The score ranges from 8 to 40 in this scale. Higher scores indicate the individual's greater ability to control the outcomes and consequences of health-related programs. The Cronbach's alpha coefficient calculated for this instrument by Smith et al was 0.84 (as described in ref. [@CIT0018]). The affect construct was evaluated using an assessment tool adapted by Watson et al. This assessment tool contains 20 items, including ten on positive affect and ten on negative affect, and is scored based on a 5-point Likert scale, ranging from never (1 point) to always (5 points). The score obtained for each of the positive and negative affects ranges from 10 to 50. Higher scores indicate better reported emotional states over the past 24 hours. The Cronbach's alpha coefficient reported by Watson et al was 0.94 for positive affect and 0.91 for negative affect.[@CIT0018] The social support construct was evaluated using an assessment tool adapted by Canty et al, including 12 items scored based on a 7-point Likert scale (strongly disagree =1 point, to strongly agree =7 points). The score obtained ranges from 12 to 84. Higher scores indicate greater support from friends, family and other key people. The Cronbach's alpha coefficient reported by Canty et al for this assessment tool was 0.91.[@CIT0018] The perceived barriers construct was evaluated using an assessment tool adapted by Becker et al, which has 18 items scored based on a 4-point Likert scale (never =1 point, to always =4 points). The score obtained for this scale ranges from 18 to 72. Higher scores indicate more barriers in health-promoting behaviors. The Cronbach's alpha coefficient reported by Becker et al for this assessment tool was 0.80.[@CIT0018] In this study, the perceived benefits construct was evaluated using an assessment tool adapted by Mohammadian et al, which consisted of 20 items scored based on a 7-point Likert scale (strongly disagree =1 point, to strongly agree =7 points). The score obtained ranges from 20 to 140 and higher scores indicate more benefits perceived for health-promoting behaviors.[@CIT0028] The content and face validity were evaluated quantitatively to determine the questionnaire's validity. For this purpose, the questionnaires were distributed among ten individuals, including four community health nursing professors, three maternal and child health specialists and three family health professionals. All the experts approved the items of the questionnaire. Cronbach's alpha was used as a measure of content reliability to evaluate the reliability of the questionnaires and was reported as 0.92 for the social support dimension, 0.74 for the self-efficacy dimension, 0.90 for the perceived barriers dimension, 0.84 for the affect dimension, 0.77 for the perceived benefits dimension and 0.85 for health-related lifestyle. The collected data were entered into SPSS-20. Initially, the normality of the data was tested using the Kolmogorov--Smirnov (K-S) test, and a normal distribution was obtained. Data were analyzed using descriptive and analytical statistics, including frequency distribution tables, mean, standard deviation, Pearson's correlation coefficient and the multiple linear regression model. The confidence interval was 0.95%. Before completing the questionnaires, the participants were fully briefed on the study objectives and were ensured that their information would remain confidential and that the results would only be used for research purposes and be published in general. Also, written consent was obtained from all the participants for taking part in the study. Findings {#S0003} ======== The mean age of the pregnant women was 32.85±6.85 years. A total of 27% of the women were illiterate or less educated, while the rest were educated; 66% were housewives and the rest were employed. [Table 1](#T0001){ref-type="table"} presents their other demographic details.Table 1The Frequency Distribution and Mean of Demographic Variables in the Participating Pregnant WomenVariableFrequency%Age\<20 years782620--30 years10735.7\>30 years11538.3EducationIlliterate8027Primary School3010Junior High School6522High School7552University5016.1Employment StatusHousewife20067Employed10033Husband 's EducationIlliterate258.3Primary School258.3Junior High School6020High School12541.7University6521.7Husband 's Employment StatusUnemployed5518.3Governmental Job4013.3Non-Governmental Job20568.3Gestational AgeSecond Trimester17558.4Third Trimester12541.6BMI\<18.5186.118.5--24.97123.725--29.910835.9\>3010334.3Income AdequacyYes13043.4No17056.6Health StatusGood7725.6Moderate15050Poor7324.4 [Table 2](#T0002){ref-type="table"} shows the mean and standard deviation of the pregnant women's scores in the health-promoting lifestyle dimensions and Pender's model constructs. As shown in the table, the mean score of health-promoting behaviors was 154.9±2.2. Also, among all the health-promoting lifestyle dimensions, the spiritual growth dimension had the highest score with a mean of 30±3.6 and the physical activity dimension the lowest score with the mean of 17±10.2. [Table 2](#T0002){ref-type="table"} presents information on the other dimensions of health-promoting lifestyle. Furthermore, among Pender's model constructs, perceived benefits had the highest score with a mean of 123.4±10.1 and self-efficacy the lowest score with a mean of 24±4.2. [Table 2](#T0002){ref-type="table"} presents information on the other constructs. According to the results of the regression analysis of the health-promoting lifestyle dimensions and Pender's model constructs ([Table 3](#T0003){ref-type="table"}), health-promoting lifestyle had a significant relationship with the perceived barriers construct, social support and perceived benefits (P\<0.05) and the regression coefficients of each of the three predictor variables (social support, perceived barriers and perceived benefits) indicated that all the three variables can significantly explain the variance in health-promoting lifestyles (P\<0.05). According to [Table 3](#T0003){ref-type="table"}, the correlation matrix of health-promoting lifestyle and Pender's model constructs using Pearson's correlation coefficient showed that health-promoting lifestyles were significantly and positively correlated with social support and perceived benefits and significantly and negatively with perceived barriers. Also, based on the results of the correlation coefficient test, there was a significant correlation between the dimensions of self-efficacy and support (r=0.28), self-efficacy and affect (r=−0.3), barriers and affect (r=0.28), perceived barriers and perceived benefits (r=0.14), perceived barriers and health-promoting lifestyle (r=0.45), social support and affect (r=0.15), social support and health-promoting lifestyle (r=0.24) and perceived benefits and health-promoting lifestyle (r=0.194); (P\<0.01; [Table 3](#T0003){ref-type="table"}).Table 2The Mean and Standard Deviation of the Scores on the Dimensions of Health-Promoting Lifestyle and the Constructs of Pender's Health Promotion ModelVariableMeanSDDimension of Health-Promoting LifestyleHealth Responsibility254.2Nutrition26.72.9Physical Activity17.810.2Interpersonal Relations28.73.4Stress Management29.62.2Spiritual Growth303.6Pender's Model ConstructPerceived Self-Efficacy244.2Perceived Benefits123.410.1Perceived Barriers37.158.2Positive and Negative Affect55.610.4Social Support61.413 Table 3The Correlation Matrix of the Health-Promoting Lifestyle Dimensions and Pender's Model ConstructsDimensionPerceived Self-EfficacyPerceived BarriersSocial SupportAffectPerceived BenefitsHealth-Promoting LifestylePerceived Self-Efficacy1Perceived Barriersr=−0.081Social Supportr=0.28\*r=−0.021Affectr=−0.30\*r=0.28\*r=0.15\*1Perceived Benefitsr=−0.02r=0.14\*r=0.10r=0.061Health-Promoting Lifestyler=0.095r=−0.45\*r=0.24\*r=−0.09r=0.194\*1[^1] The predictors of health-promoting lifestyle were determined using a multiple regression model, as shown in [Table 4](#T0004){ref-type="table"}. Among the constructs of Pender's model, social support (P=0.001), perceived barriers (P=0.001) and perceived benefits (P=0.001) had a significant effect on health-promoting lifestyle, such that the mean score of health-promoting lifestyle decreased by 1.28 units per one unit increase in the score of perceived barriers (assuming the other variables were constant; P=0.001). Also, the mean score of health-promoting lifestyle increased by 0.46 units (P=0.001) per one unit increase in social support (assuming the other variables remained constant), and the mean score of health-promoting lifestyle increased by 0.63 units (P=0.001) per one unit increase in perceived benefits (assuming the other variables remained constant). Table 4The Concurrent Effects of Pender's Model Constructs on the Dimensions of Health-Promoting Lifestyle in the Participating Pregnant WomenVariableBetaSDP-valueConstant Value102.316.3P\<0.001\*Perceived Self-Efficacy−0.130.27P=0.62Perceived Barriers−1.280.13P\<0.001\*Social Support0.460.088P\<0.001\*Affect−0.0590.11P=0.60Perceived Benefits0.630.10P\<0.001\*[^2] Discussion {#S0004} ========== This study was conducted to investigate the predictors of health-promoting lifestyles in pregnant women based on Pender's health promotion model. Based on the predictive constructs of Pender's model and their effect on health-promoting behaviors, the constructs of perceived benefits and social support, at higher levels, and the construct of perceived barriers, at lower levels, were most correlated with health-promoting behaviors, whereas perceived self-efficacy and affect had the least correlation with health-promoting behaviors. The results of other studies have also confirmed this finding.[@CIT0022],[@CIT0029],[@CIT0030],[@CIT0031] Pregnant women's greater awareness about the positive outcomes of health-promoting behaviors increases their intention to adopt these behaviors. Higher social support also leads to the greater adoption of health-promoting behaviors in pregnant women.[@CIT0001] In a study conducted by Thaewpia[@CIT0032] on older pregnant women, there was a significant correlation between health-promoting behaviors and education, perceived benefits, self-efficacy and social support. Research has shown that women with higher self-efficacy and perceived benefits scores are more likely to receive favorable social support and are therefore more likely to engage in health-promoting behaviors.[@CIT0009],[@CIT0020] A study by Lin et al,[@CIT0021] which had only considered the self-efficacy and perceived health constructs of Pender's model as predictors of health-promoting behaviors in pregnant women, revealed a significant and positive correlation between health-promoting behaviors and perceived self-efficacy. This finding is not in agreement with the results of the present study. In this study, perceived self-efficacy, perceived benefits and social support had a direct impact on health-promoting behaviors and perceived benefits. This finding is in agreement with the results of the present study. The positive effects of perceived benefits and social support on health-promoting behaviors have been confirmed by many other studies.[@CIT0014],[@CIT0033] In some studies, social support, perceived barriers and perceived benefits had the highest correlation with health-promoting behaviors,[@CIT0001],[@CIT0014] while in other studies, self-efficacy played a significant role in health-promoting behaviors.[@CIT0032],[@CIT0034] Research suggests that social support improves physical and mental health, and the lack of social support has a negative and adverse effect on health. Social support facilitates the support provided by others to adopt a behavior. In addition, social support can decrease negative psychosocial complications and consequences such as injuries and illness as well as restrictions and thus improve people's health and quality of life. Perceived benefits and perceived barriers are two constructs explaining behavior in some behavior change models; that is, they explain people's action based on whether there is a balance or imbalance between their perceived positive and negative forces on health behavior. People perform a behavior or avoid it based on their analysis of its benefits minus its barriers. Given the importance of this issue, the possibility of adopting a behavior is expected to increase when its perceived barriers are reduced and its perceived benefits are increased.[@CIT0017] Overall, it can be argued that if pregnant women have the desired social support, their self-efficacy in adopting a healthy lifestyle and their control over health behaviors will increase and they will become more likely to adopt a healthy lifestyle. Consequently, they will have fewer problems with pregnancy, experience a successful, risk-free or at least less-risky pregnancy, give birth to a healthy baby and suffer less complications, such as less preterm labor and preterm infants.[@CIT0001] One of the limitations of this study was that the participants were restricted to second- and third-trimester pregnant women due to the unavailability of accurate information on women in their first trimester of pregnancy. Therefore, if possible, lifestyle behaviors are recommended to be investigated from the onset of pregnancy, when hormonal changes begin, especially pregnancy symptoms. One of the strengths of the present study was the use of linear regression models to determine the degree and type of correlation between the constructs and lifestyle. Conclusion {#S0005} ========== According to the results of the present study on pregnant women, among all the constructs of Pender's health promotion model, social support and perceived benefits, and, to a less extent, perceived barriers, had the greatest effects on health-promoting behaviors. Given the effect of lifestyle on quality of life and health status, the results of this study can help health service policymakers and providers plan appropriate interventions for promoting the lifestyle of pregnant women. The researchers recommend further educational intervention studies based on Pender's health promotion model to promote healthy behaviors in pregnant women. This study was approved by the Research Council of Shahid Beheshti University of Medical Sciences with the ethics code IR.SBMU.RETECH.1397.27. Hereby, we wish to express our gratitude to the Vice Chancellor of Research and Technology of Shahid Beheshti University of Medical Sciences for funding this study. Disclosure {#S0006} ========== The authors report no conflicts of interest in this work. [^1]: **Note:** \*P\<0.05 is taken as the level of statistical significance. [^2]: **Notes:** Dependent variable: The mean score of the health-promoting lifestyle dimensions. \*P\<0.05 is taken as the level of statistical significance.

Related literature   {#sec1} ==================== For general background to the title compound, see: Rivera *et al.* (2010[@bb6]). For related structures, see: Rivera *et al.* (2010[@bb6], 2011[@bb5]); Silversides *et al.* (2006[@bb7]). Experimental   {#sec2} ============== {#sec2.1} ### Crystal data   {#sec2.1.1} C~10~H~10~N~4~*M* *~r~* = 186.2Orthorhombic,*a* = 7.6404 (3) Å*b* = 15.1703 (7) Å*c* = 15.9168 (7) Å*V* = 1844.87 (14) Å^3^*Z* = 8Cu *K*α radiationμ = 0.69 mm^−1^*T* = 120 K0.17 × 0.15 × 0.10 mm ### Data collection   {#sec2.1.2} Agilent Xcalibur (Atlas, Gemini ultra) diffractometerAbsorption correction: multi-scan (*CrysAlis PRO*; Agilent, 2012[@bb1]) *T* ~min~ = 0.68, *T* ~max~ = 19893 measured reflections1641 independent reflections1359 reflections with *I* \> 3σ(*I*)*R* ~int~ = 0.061 ### Refinement   {#sec2.1.3} *R*\[*F* ^2^ \> 2σ(*F* ^2^)\] = 0.039*wR*(*F* ^2^) = 0.104*S* = 1.551641 reflections133 parametersH atoms treated by a mixture of independent and constrained refinementΔρ~max~ = 0.16 e Å^−3^Δρ~min~ = −0.20 e Å^−3^ {#d5e449} Data collection: *CrysAlis PRO* (Agilent, 2012[@bb1]); cell refinement: *CrysAlis PRO*; data reduction: *CrysAlis PRO*; program(s) used to solve structure: *SUPERFLIP* (Palatinus & Chapuis, 2007[@bb3]); program(s) used to refine structure: *JANA2006* (Petříček *et al.*, 2006[@bb4]); molecular graphics: *DIAMOND* (Brandenburg & Putz, 2005[@bb2]); software used to prepare material for publication: *JANA2006*. Supplementary Material ====================== ###### Click here for additional data file. Crystal structure: contains datablock(s) global, I. DOI: [10.1107/S1600536812047538/bt6866sup1.cif](http://dx.doi.org/10.1107/S1600536812047538/bt6866sup1.cif) ###### Click here for additional data file. Structure factors: contains datablock(s) I. DOI: [10.1107/S1600536812047538/bt6866Isup2.hkl](http://dx.doi.org/10.1107/S1600536812047538/bt6866Isup2.hkl) ###### Click here for additional data file. Supplementary material file. DOI: [10.1107/S1600536812047538/bt6866Isup3.cml](http://dx.doi.org/10.1107/S1600536812047538/bt6866Isup3.cml) Additional supplementary materials: [crystallographic information](http://scripts.iucr.org/cgi-bin/sendsupfiles?bt6866&file=bt6866sup0.html&mime=text/html); [3D view](http://scripts.iucr.org/cgi-bin/sendcif?bt6866sup1&Qmime=cif); [checkCIF report](http://scripts.iucr.org/cgi-bin/paper?bt6866&checkcif=yes) Supplementary data and figures for this paper are available from the IUCr electronic archives (Reference: [BT6866](http://scripts.iucr.org/cgi-bin/sendsup?bt6866)). We acknowledge the Dirección de Investigaciones, Sede Bogotá (DIB) de la Universidad Nacional de Colombia, for financial support of this work, as well the Praemium Academiae project of the Academy of Sciences of the Czech Republic. LJ---C acknowledges the Vicerrectoría Académica de la Universidad Nacional de Colombia for a fellowship. Comment ======= The reaction of *N*^1^,*N*^2^-bis((1*H*-benzotriazol-1-yl)methyl)benzene-1,2-diamine with potassium cyanide in DMSO at room temperature gives the title compound in high yield. The two cyanomethyl groups are both *anti* with respect to the planar aromatic ring, no doubt in order to favour intermolecular N---H···N interactions. Intermolecular amine-to-amine and amine-to-cyano hydrogen bonding interactions established \"ladder-like\" parallel tapes in the crystalline solid (Fig. 2). The molecular structure and atom numbering scheme for the title compound are shown in Fig. 1. Its X-ray structure confirms the presence of intramolecular hydrogen bond between the amino groups \[N--H, 2.426 (18) Å\]. Furthermore the observed C3---N1 bond length \[1.4357 (17) Å\] is considerably elongated in relation to related structures \[1.404 (3) Å\] (Rivera *et al.*, 2010), \[1.3961 (18) Å\] (Silversides *et al.*, 2006) and is longer than the C4---N2 bond \[1.4016 (16) Å\]. Thus, these results indicate that the intramolecular interaction is responsible for the C3---N1 bond elongation. This is confirmed by the N1---C7 bond length \[1.4648 (18) Å\] that is longer than N2---C5 bond \[1.4457 (19) Å\]. Besides, the hybridization of both N1 and N2 atoms are slightly different. In fact, N1 has more *sp^2^* character \[Sα = 347.25°\] than N2 \[Sα = 330.39°\] if compared to *sp^3^*-hybridization \[Sα = 328.50°\] angle value. Other bond distances and bond angles are in good agreement with the standard values. Experimental {#experimental} ============ A mixture of potassium cyanide (0.260 g, 4.00 mmol) and *N*^1^,*N*^2^-bis((1*H*-benzotriazol-1-yl)methyl)benzene-1,2-diamine (0.370 g, 1.00 mmol) was stirred at room temperature in DMSO (5 ml) for 6 h. After quenching by addition of aqueous ammonium chloride a solid formed which was separated and extracted with ethyl acetate.The organic extract was sequentially washed with saturated Na~2~CO~3~ solution and water, then dried (Na~2~SO~4~), filtered, and concentrated to give a solid homogeneous by TLC (80% EtOAc/benzene) in 82% yield. Single crystals were obtained by recrystallization from chloroform/methanol by slow evaporation of the solvent at room temperaature, m.p. 399--400 K. Refinement {#refinement} ========== All hydrogen atoms were discernible in difference Fourier maps and could be refined to reasonable geometry, but according to common practice H atoms bonded to C were kept in ideal positions with C--H = 0.96 Å. The coordinates of the amino H atoms were refined. *U*~iso~(H) was set to 1.2*U*~eq~(C,N). All non-hydrogen atoms were refined using harmonic refinement. Figures ======= {#Fap1} {#Fap2} Crystal data {#tablewrapcrystaldatalong} ============ ------------------------- --------------------------------------- C~10~H~10~N~4~ *D*~x~ = 1.341 Mg m^−3^ *M~r~* = 186.2 Melting point: 399 K Orthorhombic, *Pbca* Cu *K*α radiation, λ = 1.5418 Å Hall symbol: -P 2ac 2ab Cell parameters from 4237 reflections *a* = 7.6404 (3) Å θ = 4.0--67.0° *b* = 15.1703 (7) Å µ = 0.69 mm^−1^ *c* = 15.9168 (7) Å *T* = 120 K *V* = 1844.87 (14) Å^3^ Polygon shape, white *Z* = 8 0.17 × 0.15 × 0.10 mm *F*(000) = 784 ------------------------- --------------------------------------- Data collection {#tablewrapdatacollectionlong} =============== ------------------------------------------------------------------- -------------------------------------- Agilent Xcalibur (Atlas, Gemini ultra) diffractometer 1641 independent reflections Radiation source: Enhance Ultra (Cu) X-ray Source 1359 reflections with *I* \> 3σ(*I*) Mirror monochromator *R*~int~ = 0.061 Detector resolution: 10.3784 pixels mm^-1^ θ~max~ = 67.2°, θ~min~ = 5.6° ω scans *h* = −7→9 Absorption correction: multi-scan (*CrysAlis PRO*; Agilent, 2012) *k* = −18→18 *T*~min~ = 0.68, *T*~max~ = 1 *l* = −18→16 9893 measured reflections ------------------------------------------------------------------- -------------------------------------- Refinement {#tablewraprefinementdatalong} ========== ------------------------------------- ------------------------------------------------------------------------------- Refinement on *F*^2^ 34 constraints *R*\[*F*^2^ \> 2σ(*F*^2^)\] = 0.039 H atoms treated by a mixture of independent and constrained refinement *wR*(*F*^2^) = 0.104 Weighting scheme based on measured s.u.\'s *w* = 1/(σ^2^(*I*) + 0.0016*I*^2^) *S* = 1.55 (Δ/σ)~max~ = 0.005 1641 reflections Δρ~max~ = 0.16 e Å^−3^ 133 parameters Δρ~min~ = −0.20 e Å^−3^ 0 restraints ------------------------------------- ------------------------------------------------------------------------------- Special details {#specialdetails} =============== ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Refinement. The refinement was carried out against all reflections. The conventional *R*-factor is always based on *F*. The goodness of fit as well as the weighted *R*-factor are based on *F* and *F*^2^ for refinement carried out on *F* and *F*^2^, respectively. The threshold expression is used only for calculating *R*-factors *etc*. and it is not relevant to the choice of reflections for refinement.The program used for refinement, Jana2006, uses the weighting scheme based on the experimental expectations, see \_refine_ls_weighting_details, that does not force *S* to be one. Therefore the values of *S* are usually larger than the ones from the *SHELX* program. ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Fractional atomic coordinates and isotropic or equivalent isotropic displacement parameters (Å^2^) {#tablewrapcoords} ================================================================================================== ------- -------------- -------------- --------------- -------------------- -- *x* *y* *z* *U*~iso~\*/*U*~eq~ N1 0.67776 (14) 0.46314 (7) 0.09105 (8) 0.0197 (3) N2 0.59009 (13) 0.37473 (8) −0.05397 (8) 0.0182 (3) N3 0.75959 (16) 0.31452 (9) 0.25065 (9) 0.0316 (4) N4 0.68740 (17) 0.42625 (9) −0.25896 (9) 0.0314 (4) C1 0.67383 (17) 0.35295 (9) 0.20428 (10) 0.0228 (4) C2 0.62292 (17) 0.38628 (9) −0.20642 (10) 0.0227 (4) C3 0.81190 (16) 0.41747 (8) 0.04529 (9) 0.0179 (4) C4 0.76583 (16) 0.37282 (8) −0.02878 (9) 0.0176 (4) C5 0.54231 (16) 0.33892 (9) −0.13496 (9) 0.0200 (4) C6 0.89448 (16) 0.32671 (9) −0.07229 (10) 0.0203 (4) C7 0.57084 (17) 0.40361 (9) 0.14220 (10) 0.0215 (4) C8 1.11225 (17) 0.37219 (9) 0.02841 (11) 0.0248 (4) C9 0.98466 (17) 0.41713 (9) 0.07282 (10) 0.0224 (4) C10 1.06689 (17) 0.32645 (10) −0.04320 (10) 0.0238 (4) H1c5 0.417265 0.339719 −0.140823 0.024\* H2c5 0.57431 0.277778 −0.137292 0.024\* H1c6 0.864639 0.294979 −0.122465 0.0244\* H1c7 0.508788 0.363609 0.106188 0.0258\* H2c7 0.481985 0.436862 0.17073 0.0258\* H1c8 1.23154 0.372856 0.047341 0.0298\* H1c9 1.015681 0.44837 0.123117 0.0268\* H1c10 1.154716 0.294105 −0.073369 0.0286\* H1 0.732 (2) 0.5050 (12) 0.1255 (12) 0.0237\* H2 0.536 (2) 0.4245 (12) −0.0403 (12) 0.0219\* ------- -------------- -------------- --------------- -------------------- -- Atomic displacement parameters (Å^2^) {#tablewrapadps} ===================================== ----- ------------ ------------ ------------ ------------- ------------- ------------- *U*^11^ *U*^22^ *U*^33^ *U*^12^ *U*^13^ *U*^23^ N1 0.0140 (5) 0.0231 (6) 0.0221 (7) 0.0001 (4) −0.0013 (5) −0.0013 (5) N2 0.0087 (5) 0.0257 (6) 0.0202 (7) 0.0023 (4) 0.0000 (5) −0.0013 (5) N3 0.0229 (6) 0.0399 (7) 0.0319 (8) −0.0047 (6) −0.0020 (6) 0.0091 (6) N4 0.0267 (6) 0.0389 (7) 0.0287 (8) 0.0053 (6) 0.0054 (6) 0.0065 (6) C1 0.0167 (7) 0.0287 (7) 0.0231 (8) −0.0035 (6) 0.0041 (6) −0.0004 (6) C2 0.0154 (6) 0.0286 (7) 0.0241 (9) 0.0052 (6) −0.0007 (6) −0.0013 (6) C3 0.0117 (6) 0.0210 (6) 0.0210 (8) −0.0002 (5) 0.0014 (5) 0.0041 (5) C4 0.0101 (6) 0.0217 (6) 0.0211 (8) −0.0006 (5) 0.0001 (5) 0.0045 (5) C5 0.0124 (6) 0.0268 (7) 0.0208 (8) −0.0011 (5) −0.0006 (6) 0.0008 (5) C6 0.0131 (6) 0.0249 (7) 0.0229 (8) 0.0011 (5) 0.0019 (6) 0.0012 (6) C7 0.0137 (6) 0.0285 (7) 0.0223 (8) 0.0013 (5) 0.0007 (6) −0.0007 (6) C8 0.0094 (6) 0.0300 (7) 0.0350 (9) −0.0019 (5) −0.0030 (6) 0.0081 (6) C9 0.0153 (6) 0.0262 (7) 0.0256 (9) −0.0038 (5) −0.0042 (6) 0.0051 (6) C10 0.0107 (6) 0.0284 (7) 0.0324 (9) 0.0028 (5) 0.0043 (6) 0.0064 (6) ----- ------------ ------------ ------------ ------------- ------------- ------------- Geometric parameters (Å, º) {#tablewrapgeomlong} =========================== ---------------- ------------- ------------------ ------------- N1---C3 1.4357 (17) C4---C6 1.3911 (19) N1---C7 1.4648 (18) C5---H1c5 0.96 N1---H1 0.935 (18) C5---H2c5 0.96 N2---C4 1.4016 (16) C6---C10 1.3963 (18) N2---C5 1.4457 (19) C6---H1c6 0.96 N2---H2 0.889 (18) C7---H1c7 0.96 N3---C1 1.146 (2) C7---H2c7 0.96 N4---C2 1.144 (2) C8---C9 1.384 (2) C1---C7 1.479 (2) C8---C10 1.379 (2) C2---C5 1.480 (2) C8---H1c8 0.96 C3---C4 1.404 (2) C9---H1c9 0.96 C3---C9 1.3908 (18) C10---H1c10 0.96 C3---N1---C7 112.49 (10) H1c5---C5---H2c5 105.18 C3---N1---H1 108.1 (10) C4---C6---C10 120.20 (14) C7---N1---H1 109.8 (11) C4---C6---H1c6 119.9 C4---N2---C5 119.29 (11) C10---C6---H1c6 119.9 C4---N2---H2 113.2 (11) N1---C7---C1 113.27 (11) C5---N2---H2 114.8 (12) N1---C7---H1c7 109.47 N3---C1---C7 177.27 (15) N1---C7---H2c7 109.47 N4---C2---C5 176.55 (16) C1---C7---H1c7 109.47 N1---C3---C4 118.67 (11) C1---C7---H2c7 109.47 N1---C3---C9 121.29 (13) H1c7---C7---H2c7 105.38 C4---C3---C9 120.04 (12) C9---C8---C10 119.55 (13) N2---C4---C3 118.05 (12) C9---C8---H1c8 120.22 N2---C4---C6 123.02 (13) C10---C8---H1c8 120.22 C3---C4---C6 118.90 (12) C3---C9---C8 120.63 (14) N2---C5---C2 113.45 (11) C3---C9---H1c9 119.69 N2---C5---H1c5 109.47 C8---C9---H1c9 119.69 N2---C5---H2c5 109.47 C6---C10---C8 120.65 (13) C2---C5---H1c5 109.47 C6---C10---H1c10 119.67 C2---C5---H2c5 109.47 C8---C10---H1c10 119.67 ---------------- ------------- ------------------ ------------- Hydrogen-bond geometry (Å, º) {#tablewraphbondslong} ============================= ------------------ ------------ ------------ ------------- --------------- *D*---H···*A* *D*---H H···*A* *D*···*A* *D*---H···*A* N1---H1···N4^i^ 0.935 (18) 2.202 (19) 3.0946 (19) 159.2 (16) N2---H2···N1 0.889 (18) 2.427 (18) 2.7524 (18) 102.0 (13) N2---H2···N1^ii^ 0.889 (18) 2.494 (17) 3.2536 (16) 143.8 (15) ------------------ ------------ ------------ ------------- --------------- Symmetry codes: (i) −*x*+3/2, −*y*+1, *z*+1/2; (ii) −*x*+1, −*y*+1, −*z*. ###### Hydrogen-bond geometry (Å, °) *D*---H⋯*A* *D*---H H⋯*A* *D*⋯*A* *D*---H⋯*A* ---------------- ------------ ------------ ------------- ------------- N1---H1⋯N4^i^ 0.935 (18) 2.202 (19) 3.0946 (19) 159.2 (16) N2---H2⋯N1 0.889 (18) 2.427 (18) 2.7524 (18) 102.0 (13) N2---H2⋯N1^ii^ 0.889 (18) 2.494 (17) 3.2536 (16) 143.8 (15) Symmetry codes: (i) ; (ii) .

Introduction {#Sec1} ============ Many fundamental questions concerning the mechanisms of self-renewal and differentiation of stem cells are addressed using *Drosophila* oogenesis as a model^[@CR1]^. *Drosophila* ovaries consist of ovarioles, chains of egg chambers connected to the germarium, which houses germline stem cells (GSCs). A microenvironment of somatic cells known as a niche regulates GSC state via different cell signaling pathways^[@CR1]--[@CR3]^. The ovarian niche includes terminal filament (TF) cells, cap cells (CCs), and escort cells (ECs). GSCs directly contact CCs and the most anterior ECs, which prevent GSC differentiation by secreting decapentaplegic (Dpp) and glass bottom boat (Gbb) protein ligands^[@CR4]--[@CR7]^. These ligands interact with GSC surface receptors and activate BMP signaling, which represses transcription of the *bam* gene required for GSC differentiation. After GSC division, one of the daughter cells retains its stem state, whereas the other one leaves the self-renewal niche and begins to differentiate into a cystoblast, which then divides and differentiates to form a cyst of germ cells surrounded by somatic follicle cells. A special marker of GSCs and cystoblasts is the spectrosome, a cytoplasmic body, which transforms into a branching structure called the fusome connecting the dividing germ cells. To initiate the differentiation of the cystoblast, BMP signaling must be decreased by different intrinsic and extrinsic mechanisms^[@CR8]^. The majority of ECs limit the spreading of BMP ligands and therefore promote differentiation of the cystoblasts and dividing cysts^[@CR9],[@CR10]^. Thus, the renewal somatic niche provides maintenance signals for GSCs, while a more posteriorly located differentiation niche, represented by ECs, is required for proper differentiation of GSC progeny. The piRNA (Piwi-interacting RNA) pathway controls expression of transposable elements (TEs) in both somatic and germ cells of *Drosophila* ovaries. Piwi proteins guided by small piRNAs (24--30 nt) recognize complementary RNA molecules leading to their degradation or the repression of transcription with the help of other proteins (for review see^[@CR11]^). The known molecular function of the piRNA pathway in the ovarian soma is the repression of a specific group of somatically active LTR retrotransposons^[@CR12]--[@CR16]^. The piRNA machinery in *Drosophila* ovarian somatic cells seems to be simpler than its counterpart in the germline. It operates via a single Piwi protein unlike the three proteins in germ cells and a substantial part of somatic piRNAs originates from a single source, the piRNA cluster *flamenco* (*flam*)^[@CR14],[@CR15],[@CR17]^ that is an extended 180 kb region of X-chromosome heterochromatin, filled by TE copies and their fragments^[@CR18]--[@CR20]^. The *flam* locus is responsible for the repression of at least three somatically expressed retrotransposons: *gypsy*, *ZAM* and *Idefix*^[@CR21]--[@CR24]^. Cleavage of *flam* transcripts into small RNA molecules occurs in cytoplasmic Yb bodies. The cytoplasmic piRNA biogenesis machinery in somatic cells includes the nuclease Zucchini (Zuc), the RNA helicase Armitage (Armi), the TUDOR domain-containing proteins fs(1)Yb (Yb) and Vreteno (Vret), and other components^[@CR16],[@CR25]--[@CR27]^. In the course of *flam* transcript cleavage, piRNAs are loaded into Piwi and then move into the nucleus, where mature piRNA-Piwi complexes recognize complementary TE transcripts and repress their transcription with the help of adaptors, which recruit histone modification proteins, such as H3K9 methyltransferase Eggless (Egg) and H3K4 demethylase dLSD1^[@CR28]--[@CR32]^. piRNA pathway mutations cause upregulation of TEs and lead to different oogenesis defects and sterility. Initially, two key components of the piRNA system, Piwi and Yb, have been shown to be required in somatic cells to prevent GSC loss^[@CR33],[@CR34]^. Later it was found that the lack of several components of the somatic piRNA pathway, including Piwi^[@CR35]--[@CR37]^, Vret^[@CR27]^, *flam*^[@CR23],[@CR38]^ and Egg^[@CR38],[@CR39]^ lead to the accumulation of undifferentiated germ cells in germaria, known as a germline tumor phenotype. The germ cell differentiation defects observed in piRNA pathway mutants are thought to be related to the dysfunction of ECs^[@CR36],[@CR37],[@CR39]^. Knockdowns of Piwi and Yb specifically in ECs induced large numbers of ectopic GSC-like cells^[@CR36],[@CR37]^. However, the underlying mechanisms are contradictory. Several papers noted an increased rate of somatic cell death in ovaries due to TE activation^[@CR27],[@CR38]^. Others have found that Piwi downregulates expression of the *dpp* gene in ECs^[@CR36],[@CR37]^ and that TE activation decreases the expression of Wnt4 ligand, which ensures EC function in germ cell differentiation^[@CR39]^. It has been shown also that *piwi* mutations disrupt the spatial position of gonadal intermingled cells (the EC progenitors) and germ cells in early development^[@CR36]^. Here we provide results indicating that the germ cell differentiation defects caused by somatic TE activation in *flam* mutants are due to a decrease of EC precursor population at the larval stage, whereas no EC death or additional decline of their production rate was observed in *flam* adult ovaries. We also found drastic oogenesis defects in *flam* mutants combined with mutations of genes encoding Chk2 (Checkpoint kinase 2) or ATM (ataxia telangiectasia-mutated) checkpoint kinases, contrary to known suppressor effect of *chk2* mutation on ovarian development caused by TE derepression in the germline^[@CR38],[@CR40]--[@CR43]^. These results indicate that the somatic cells of ovaries are especially sensitive to TE upregulation upon loss of the Chk2 DNA damage response pathway. Results {#Sec2} ======= The occurrence of germ cell differentiation defects caused by somatic TE activation correlates with a reduced number of ECs {#Sec3} --------------------------------------------------------------------------------------------------------------------------- To extend previous observations^[@CR27],[@CR37]--[@CR39]^ that activation of TEs in ovarian somatic cells leads to germ cell differentiation defects, we estimated the spectrosome-containing cell number in ovaries lacking various components of the somatic piRNA pathway, some of which have not been tested in this regard before. For this and most subsequent experiments, we analyzed ovaries of 7-day-old females to allow tumor phenotype to develop to a pronounced degree. α-spectrin immunostaining revealed a drastic increase in the number of spectrosome-containing cells upon somatic depletion of Asterix (Arx) (also known as GTSF1) (Fig. [1a,b](#Fig1){ref-type="fig"}), a nuclear Piwi cofactor^[@CR29],[@CR30]^. Depleting Armi, a cytoplasmic component of piRNA biogenesis machinery^[@CR16],[@CR25]^, in all somatic cells of ovaries (Fig. [1a,b](#Fig1){ref-type="fig"}) or only in ECs (Fig. [S1a](#MOESM1){ref-type="media"}) also caused germline tumors. Moreover, this phenotype was observed in ovaries lacking Zuc and Yb proteins (Fig. [S1b](#MOESM1){ref-type="media"}). *piwi*^*Nt*^ mutation causing TE derepression due to cytoplasmic Piwi localization^[@CR44]^ also led to the excess of spectrosome-containing germ cells (Fig. [S1c](#MOESM1){ref-type="media"}), whereas in agreement with our previous report^[@CR44]^ a GSC loss phenotype was rare in *piwi*^*Nt*^ ovaries in contrast to *piwi* null mutants (Fig. [S1d](#MOESM1){ref-type="media"}). Thus, our results together with previous findings^[@CR27],[@CR38],[@CR39]^, show that defects in germ cell differentiation are associated with the disruption of any component of the TE silencing pathway in somatic cells of ovaries.Figure 1The occurrence of germ cell differentiation defects caused by somatic TE activation correlates with a reduced number of ECs. (**a**) Examples of wild-type and tumorous germaria stained for α-spectrin (red) to detect spectrosomes and fusomes and for lamin (blue) to visualize cell nuclei. A wild-type germarium (upper panel) usually contains 2--3 GSCs and a few cystoblasts marked by round spectrosomes (s). GSC are located at the anterior end of the germarium in close proximity to somatic cap cells (cc). Dividing cysts carry branched fusome structures (f). Germaria of Arx (middle panel) and Armi (lower panel) knockdowns (KDs) driven by *traffic jam Gal4* (*tj-Gal4*) in ovarian somatic cells carry an excess of spectrosome-containing cells and lack fusomes. (**b**) Quantification of spectrosome-containing cells in 7-day-old females with KDs of piRNA pathway components in ovarian somatic cells. Each dot corresponds to a single germarium. The central mark indicates the median, and the bottom and top lines indicate the 25th and 75th percentiles, respectively. All tested KD germaria contain significantly more spectrosomes than control (Mann--Whitney U-test; \*p \< 0.00001). Effects of Piwi and Vret KDs corroborate previously reported result^[@CR27],[@CR35]--[@CR39]^ (**c**). Quantifi**c**ation of spectrosome-containing cells in 7-day-old *flam* mutants (Mann--Whitney U-test; \*p \< 0.00001). (**d**) Quantification of ECs in *flam* germaria using *PZ1444-lacZ* line. EC number per germarium is indicated (Mann--Whitney U-test; \*p \< 0.00001). (**e**) Immunostaining of *flam*^*KG*^/*Df* and control germaria for α-spectrin (red), *PZ1444-lacZ* (green) and lamin (blue). (**f**) Increase of spectrosome-containing cell number in mutant germaria containing a reduced number of ECs. Percentage of *flam*^*KG*^/*Df* germaria with more than 10 spectrosome-containing cells in groups of germaria with different number of ECs is shown, based on three replicates (n = 181). Mean +/− s.d. are indicated. (Student's t-test; \*\*p \< 0.05). Scale bars, 10 µm. Mutations affecting protein-coding genes could exert pleiotropic effects, if corresponding proteins have additional specific functions in oogenesis, unrelated to TE repression, as has been reported for Piwi^[@CR45],[@CR46]^. Therefore, to directly examine the influence of activated somatic TEs on germline differentiation, we focused on the studies of *flam* piRNA cluster mutants. In most experiments, we analyzed *flam*^*BG*^*/Df* and *flam*^*KG*^*/Df* mutants carrying P-element-induced mutations^[@CR17],[@CR23]^ and an X chromosome deletion (*Df*) covering the whole *flam* locus. Both mutants exhibit the derepression of *flam*-regulated somatic TEs (Fig. [S2b](#MOESM1){ref-type="media"}) and about one-half of mutant germaria show a prominent germline tumor phenotype (Figs. [1c](#Fig1){ref-type="fig"} and [S3a](#MOESM1){ref-type="media"}) and some other defects (Fig. [S3b](#MOESM1){ref-type="media"}). We found no or faint *bam-GFP* reporter^[@CR4]^ expression in germ cells constituting tumors in *flam*^*KG*^*/Df* germaria (Fig. [S3c](#MOESM1){ref-type="media"}), indicating an abnormally enhanced BMP-signaling, which may be caused by the failure of ECs to restrict Dpp spreading^[@CR9],[@CR10]^. The number of ECs visualized by immunostaining for *PZ1444 lacZ* reporter expression^[@CR47],[@CR48]^ was reduced about two-fold from an average of 29 ECs per germarium in the *flam*/+ control to 16 and 18 ECs in *flam*^*KG*^*/Df* and *flam*^*BG*^*/Df* mutants, respectively (Figs. [1d](#Fig1){ref-type="fig"} and [S3d](#MOESM1){ref-type="media"}). ECs of *flam* mutants lacked cellular processes that wrap up differentiating germ cells in wild-type ovaries (Fig. [S3e](#MOESM1){ref-type="media"}). The latter effect may be a consequence of defective germline differentiation according to literature. For example, it has been shown that *bam* mutation impedes the formation of EC processes^[@CR9]^. Although the decrease in the EC number was previously reported for *piwi* somatic knockdown^[@CR37]^, it was not clear whether EC reduction directly affects the germ cell differentiation. Since both the spectrosome and EC numbers substantially varied among individuals carrying *flam* mutations, we wondered how these parameters would be related within a single genotype. Simultaneous immunostaining of mutant ovaries with antibodies against α-spectrin and β-galactosidase (*PZ1444* reporter) (Fig. [1e](#Fig1){ref-type="fig"}) revealed that germline tumors were rarely detected in *flam* germaria containing more than 20 ECs, whereas germaria with a small number of ECs more often accumulated large numbers of spectrosome-containing cells (Fig. [1f](#Fig1){ref-type="fig"}). This result clearly shows a correlation between EC number reduction and germline tumor formation. Alternatively, the disruption of the differentiation niche may be caused by the loss of EC functional status, such as an abnormal or reduced production of signaling molecules. Impairment of different signaling pathways, including Wnt^[@CR35],[@CR49]--[@CR51]^, Rho^[@CR9],[@CR52]^, EGFR^[@CR53],[@CR54]^, Hh and Hpo/Yki^[@CR52]^, as well as the enhancement of BMP signaling in ECs may lead to the germline tumor phenotype. Specifically, loss of piRNA pathway in ECs has been shown to be associated with enhancement of Dpp expression^[@CR36],[@CR37]^ and a decrease in Wnt signaling^[@CR39]^. However, we observed no significant changes in the Wnt2 ligand mRNA expression, a two-fold decrease of the Wnt4 ligand and Frizzled3 (Fz3, target of Wnt pathway) mRNAs and a slight upregulation of Dpp in both *flam*^*KG*^*/Df* and *flam*^*BG*^*/Df* germaria compared to control siblings (Fig. [S4a](#MOESM1){ref-type="media"}). The two-fold decrease of *wnt4* expression is likely explained by the observed two-fold reduction of EC number in *flam* mutants (Fig. [1d](#Fig1){ref-type="fig"}), given that Wnt4 (but not Wnt2) is expressed only in ECs and is not detected in other cell types in the germarium^[@CR49]^. The observed Dpp upregulation in *flam* germaria (Fig. [S4a](#MOESM1){ref-type="media"}) can also be interpreted as a consequence of EC number reduction, because the antagonism between Wnt and BMP pathways in the ovarian somatic cells has been established^[@CR50],[@CR51]^. The Wnt4 target *tkv-lacZ* expression^[@CR49]^ was similar in ECs of *flam* and control ovaries indicating active Wnt signaling (Fig. [S4b](#MOESM1){ref-type="media"}). Secreted Wnt ligands are known to act in ECs in an autocrine manner^[@CR49],[@CR51]^ resulting in stabilization of a downstream effector protein β-catenin/Armadillo (Arm)^[@CR55]^. We failed to find any alteration of Arm protein level in *flam* germaria by Western blot (Fig. [S4c](#MOESM1){ref-type="media"}). Overexpression of Arm in *flam* mutants did not cause a decrease in spectrosome-containing germ cell number (Fig. [S4d](#MOESM1){ref-type="media"}). Similarly, expression of the constitutive Arm form (*UAS-Arm-S10*) driven by *c587-Gal4* in ECs did not rescue the germline tumor phenotype of *piwi* mutants (Fig. [S4e](#MOESM1){ref-type="media"}). As a whole, these results suggest that the germ cell differentiation defect in *flam* mutants is mediated rather by a decrease in the number of ECs, than by dysfunction of remaining ECs. EC number and the formation of germline tumor phenotype in flam mutants are determined at the larval stage {#Sec4} ---------------------------------------------------------------------------------------------------------- ECs are initially produced from the intermingled cells during larval and pupal development^[@CR56]--[@CR58]^. Then, in adult ovaries ECs exhibit slow turnover rates, though a fraction of ECs is renewed. Escort stem cells^[@CR59]^ or self-duplications of ECs^[@CR9],[@CR10]^ were previously suggested as a source of new ECs in the adult gonads. A recent study revealed that new ECs in the imago are produced by divisions of follicle stem cells^[@CR60]^. The reduction of EC number in *flam* germaria could be attributed to an increased rate of EC death, to defects of their renewal in adult ovaries or to a decline of EC production during earlier development. The TUNEL assay revealed less than 10% of *flam*/+ germaria containing at least one apoptotic EC. Unexpectedly, in *flam*^*KG*^*/Df* germaria the apoptotic ECs were even less frequently detected (Fig. [S5a--c](#MOESM1){ref-type="media"}). To examine the formation of new ECs in *flam* ovaries, we carried out immunostaining for phosphohistone H3 Ser10 (PH3) mitotic marker and EdU incorporation assay. Both methods failed to detect a significant number of newly formed ECs in the *flam*^*KG*^*/Df* and control flies (Fig. [S5d,e](#MOESM1){ref-type="media"}). Furthermore, most ovaries of females fed on EdU-containing food for three days did not contain EdU-positive ECs (Fig. [2a](#Fig2){ref-type="fig"}), suggesting that ECs in tested lines are mainly produced at earlier developmental stages. Importantly, we found about the same number of ECs in germaria of one-, four- or seven-day-old *flam* adults (Fig. [2b](#Fig2){ref-type="fig"}). Thus, the decrease of EC number is observed already in one-day-old *flam* mutants (Fig. [2b](#Fig2){ref-type="fig"}) and, therefore, is determined prior to the imago stage.Figure 2*flam* mutation leads to a reduction in EC number in larval development but not in adults. (**a**) *flam*^*KG*^/+ germarium of an adult female after feeding EdU for three days, stained for EdU (purple) and lamin (red). (**a'**) The same germarium with *PZ1444* immunostaining (green). ECs (indicated by green arrows) and CCs are EdU-negative. (**b**) Quantification of ECs in *flam*^*KG*^/+ (red dots) and *flam*^*KG*^/*Df* (green dots) flies at the age of 1, 4 and 7 days. The differences between samples of different ages of the same genotypes are not significant (n.s.) (Mann--Whitney U-test; p \> 0.1). (**c,d**) Germaria of females, obtained from larvae reared on EdU-containing food. White and green arrows indicate EdU-positive and EdU-negative ECs, respectively. (**e**,**e'**) An example of *flam*^*KG*^/*Df* germarium with large number of ECs, most of which are EdU-negative after larval EdU incorporation. Full Z-series projections are shown. (**f**,**f'**) *flam*^*KG*^/+ germarium. EdU-positive ECs are located at the more anterior region of the germarium compared to EdU-negative ECs. (**g**) Quantification of EdU-positive and EdU-negative ECs in *flam*^*KG*^/+ and *flam*^*KG*^/*Df* germaria after larval EdU feeding. Mean +/− s.d. are indicated, based on three replicates (Student's t test; \*p = 0.01; n.s. = not significant). Scale bars, 10 µm. To find out the developmental stage when ECs are lost, we reared larvae on EdU-containing food, then placed eclosed flies on standard food and analyzed 3-day-old fly ovaries. In this case, all dividing larval cells will contain EdU signals, which then will be diluted with each round of replication in pupae and adults. Expectedly, the follicle cells and most of the germ cells were EdU-negative. Conversely, strong EdU immunostaining was observed in CCs and TF cells (Fig. [2c--f](#Fig2){ref-type="fig"}), which are known to be formed in larvae and then do not divide or renew^[@CR56]--[@CR58],[@CR61]^. About 70% of ECs were also labeled by EdU in *flam*^*KG*^/+ germaria. Apparently, the EdU-positive ECs were formed as a result of a few divisions of parental cells marked by EdU incorporation at the larval stage, while the EdU-negative ECs were likely produced later in development or originated from more actively proliferating cells. Interestingly, EdU-positive ECs were usually located more anteriorly than EdU-negative ECs (Fig. [2c,f](#Fig2){ref-type="fig"}), which is consistent with the possible origin of the latter from follicle stem cells^[@CR60]^. In *flam*^*KG*^*/Df* germaria we observed a significant decrease of EdU-positive EC number compared to *flam*^*KG*^/+ sisters (Fig. [2d,e,g](#Fig2){ref-type="fig"}), which demonstrates decreased EC precursors formation in *flam* larvae. However, the number of EdU-negative ECs in *flam* mutant showed a large scatter of values (Fig. [2g](#Fig2){ref-type="fig"}). In some *flam* germaria the number of the EdU-negative ECs was even increased compared to control (as exemplified in Fig. [2e](#Fig2){ref-type="fig"}), suggesting that new ECs can be actively produced after the larval stage to compensate for the lack of EC precursors in earlier development. Primordial germ cells (PGCs) starting from mid-larval third instar stage are associated with intermingled cells that are EC progenitors. At this stage, all germ cells of the developing ovary are grouped together. Germaria formation occurs later in pupae, when TFs, CCs and their attached PGCs are separated into individual germaria units^[@CR56],[@CR62],[@CR63]^. If EC number and germline differentiation defects are determined during larval development of *flam* gonads, a correlation can be expected between phenotypes of germaria within the same ovary. Indeed, we found that the numbers of both ECs and spectrosomes were quite similar in *flam* germaria belonging to the same ovary but varied substantially between individual ovaries (Fig. [3](#Fig3){ref-type="fig"}). Thus, developmental events prior to the pupal stage predetermine the germ cell differentiation defects in *flam* mutants.Figure 3*flam* germaria belonging to the same ovary exhibit similar phenotypes. Quantification of spectrosome-containing cells (**a**) and ECs (**b**) per germarium in the same ovaries of *flam*^*KG*^/*Df* mutants. Germaria in one оvary are shown by grouped dots of the same color. Spectrosome numbers in germaria from tumorous ovaries (A and B) are significantly higher than their numbers in the non-tumorous ovaries (C-G) (Mann--Whitney U-test; \*p \< 0.05 for A and B vs C-G). *flam* mutation induces DNA breaks in somatic cells of larval ovaries {#Sec5} --------------------------------------------------------------------- The observed decline of EC precursor production in larval ovaries may be caused by the appearance of TE-induced DNA lesions in their genomes. To check this, we examined the presence of phosphorylated H2Av (γ-H2Av) histone, a commonly used DNA break marker^[@CR64]^, in larval somatic intermingled cells marked by Traffic jam (Tj) immunostaining^[@CR56],[@CR58]^. γ-H2Av dots were observed in 10--20% of Tj-positive cells in wild-type (*Batumi*) and *flam*^*KG*^/+ (Fig. [4a](#Fig4){ref-type="fig"}) third instar larval (L3) ovaries. In *flam*^*KG*^*/Df* L3 ovaries about 80% of Tj-positive cells contained γ-H2Av signals (Fig. [4b,c](#Fig4){ref-type="fig"}). γ-H2Av foci were also detected in Tj-negative somatic cells, including TF cells, as well as somatic apical (AP) and basal (BS) cells (Fig. [4b](#Fig4){ref-type="fig"}), which are known to be not incorporated into germaria^[@CR57]^. However, most PGCs surrounded by intermingled cells did not contain γ-H2Av foci in mutant ovaries (Fig. [4b](#Fig4){ref-type="fig"}). Thus, the *flam* mutation leads to DNA breaks in somatic, but not germline cells of the larval ovaries. Immunostaining with activated Caspase3 antibodies, as well as TUNEL assay detected an increase of somatic cell death in *flam*^*KG*^*/Df* larval ovaries (Fig. [S6](#MOESM1){ref-type="media"}). However, we cannot exclude that a reduction of EC number is partially caused by a decrease of division rate of EC precursors.Figure 4Intermingled cells in *flam* larval ovaries more often contain DNA breaks than ECs in adult ovaries. (**a**,**a'**) The *flam*^*KG*^/+ ovary of third instar larval stage stained for lamin (green), γ-H2Av DNA break marker (red) and Traffic jam (Tj, blue) showing intermingled cells (IC). Tj-negative cells include somatic apical (AP) and basal (BC) cells, TF, and Primordial germ cells (PGC). (**b**,**b**') *flam*^*KG*^*/Df* L3 ovaries accumulate γ-H2Av in most ICs and other somatic cells, including TF and AP, but not in PGCs. (**c**) Quantification of γ-H2Av-positive among Tj-positive cells in *flam*^*KG*^/+ and *flam*^*KG*^/*Df* larval ovaries. Mean + /− s.d. are indicated (Student's t test; \*p \< 1e-26). (**d**) *flam*^*KG*^/+ germarium of adult ovary stained for lamin (blue), γ-H2Av (red) and *PZ1444* EC marker (green). γ-H2Av signals are observed mainly in germ cells (GC, indicated by yellow arrows). (**e**,**e'**) In *flam*^*KG*^**/***Df* germaria γ-H2Av foci appear in most follicle cells (FC, white arrows) and only in some *PZ1444*-marked ECs (green arrows). (**f**) Quantification of γ-H2Av signals in FCs and ECs in ovaries of adults. Mean + /− s.d. are indicated (Student's t test; \*p \< 1e-9). Scale bars, 10 µm. Then we monitored γ-H2Av presence in the somatic cells of adult *flam* ovaries. In *flam*/+ germaria, as in wild-type, γ-H2Av signals were absent in ECs and CCs, but were detected in the meiotic germ cells and endocycling nurse cells (Fig. [4d](#Fig4){ref-type="fig"}) where DNA breaks are generated during normal development^[@CR65]--[@CR69]^. In the *flam*^*KG*^*/Df* germaria only about 20% of ECs contained γ-H2Av foci, whereas follicle cells were mostly γ-H2Av-positive (Fig. [4e,f](#Fig4){ref-type="fig"}). These observations indicate that DNA damage events occur in mature *flam* ECs less often than in their precursors, intermingled cells, at the larval stage and/or mature ECs have an enhanced capacity to repair DNA lesions. The absence of Chk2 or ATM checkpoint kinases enhanced oogenesis defects of *flam* mutants {#Sec6} ------------------------------------------------------------------------------------------ DNA damage is known to block cell proliferation through the activation of checkpoint kinases, which induce cell cycle arrest followed by apoptosis or DNA repair (for review see^[@CR70]^). *Drosophila* Chk2 encoded by the *Mnk/Loki* gene together with other checkpoint kinases is required for cell cycle arrest in response to DNA breaks in both somatic and germ cells^[@CR71]--[@CR74]^. Another function of Chk2 is p53 phosphorylation that activates transcription of genes involved in DNA repair and/or apoptosis pathways^[@CR75],[@CR76]^. To examine whether the *flam* mutant phenotype is mediated by the checkpoint response to TE-induced DNA breaks, we crossed the *mnk*^*p6*^ mutation (the well-characterized loss of function allele^[@CR42],[@CR43],[@CR71],[@CR77]^) into a *flam* mutant background. Although the *chk2* mutation was shown to partially rescue the germline differentiation defects induced by TE activation in germ cells^[@CR38],[@CR42],[@CR43]^, we unexpectedly observed its opposite effect in *flam* mutants. The *flam*^*KG*^*/Df*; *mnk*^*p6*^/*mnk*^*p6*^ double mutants had drastically more defective ovaries than *flam*^*KG*^*/Df*; *mnk*^*p6*^/+ individuals, whereas *flam*^*KG*^/+; *mnk*^*p6*^/*mnk*^*p6*^ ovaries displayed no visible morphological defects (Fig. [5a--d](#Fig5){ref-type="fig"}). The formation of germaria was abolished in most *flam*^*KG*^*/Df*; *mnk*^*p6*^/*mnk*^*p6*^ ovaries (Fig. [5c,d](#Fig5){ref-type="fig"}) and in some of them the number of Tj-positive ovarian somatic cells was highly reduced (Fig. [5d](#Fig5){ref-type="fig"}). Severe oogenesis defects were also observed when we combined *flam* with two different mutations in the *tefu* gene (Fig. [S7](#MOESM1){ref-type="media"}), encoding a *Drosophila* homolog of ATM kinase that is directly recruited and activated by DNA double-strand breaks, acting upstream of Chk2^[@CR70],[@CR78]^. We suggested that the observed catastrophic ovarian phenotypes can be induced by the death or dysfunction of ovarian somatic cells, including ECs due to their inability to repair TE-induced DNA lesions. Then, we checked whether DNA breaks are accumulated in ECs of *flam* mutants lacking checkpoint response. We found that somatic cell nuclei in rarely observed germaria-like structures of *flam*^*KG*^*/Df*; *mnk*^*p6*^/*mnk*^*p6*^ ovaries were dramatically enriched in γ-H2Av signals compared to both *flam*^*KG*^*/Df*; *mnk*^*p6*^/+ and *flam*^*KG*^/+; *mnk*^*p6*^/*mnk*^*p6*^ germaria (Fig. [5e--g](#Fig5){ref-type="fig"}). Somatic depletion of Mnk in the *flam*^*KG*^*/Df*, but not *flam*^*KG*^/+ background also caused accumulation of γ-H2Av in nearly all ECs (Fig. [5h,i](#Fig5){ref-type="fig"}) in contrast to about 20% of γ-H2Av-positive ECs observed in *flam*^*KG*^*/Df* ovaries (Fig. [4f](#Fig4){ref-type="fig"}). In addition, γ-H2Av foci were accumulated in ECs upon depletion of another checkpoint kinase, mei-41 (*Drosophila* homolog of ATR) (Fig. [5h,j](#Fig5){ref-type="fig"}), which, however, did not enhance oogenesis defects. Thus, Chk2, ATM and ATR kinases are involved in cellular response upon TE activation in ECs or their progenitors, but their specific molecular functions in this process warrant further examination.Figure 5c*hk2* mutation enhances ovarian defects in *flam* mutants. (**a**) *flam*^*KG*^/*Df*; *mnk*^*p6*^/+ germarium stained for lamin (blue), α-spectrin (red) and Tj (green) showing nuclei of CCs, ECs and FCs. (**b**) *flam*^*KG*^/+*; mnk*^*p6*^*/mnk*^*p6*^ germarium with no morphological defects. (**c,d**) Fragments of *flam*^*KG*^/*Df*; *mnk*^*p6*^/*mnk*^*p6*^ ovaries showing an impaired formation of germaria and ovarioles. Abnormal germaria-like structures (G) are filled with spectrosome-containing cells and lack Tj-positive somatic cells or lack both germ cells and ECs. Separate accumulations of Tj-positive somatic cells (S) are indicated. (**e**) Immunostaining of *flam*^*KG*^/+; *mnk*^*p6*^/*mnk*^*p6*^ germarium with Tj (green), γ-H2Av (red) and lamin (blue). γ-H2Av signals are observed in meiotic germ cells and in a few somatic cells. (**f**) *flam*^*KG*^*/Df*; *mnk*^*p6*^/+ germarium containing γ-H2Av dots in follicle cells and in some ECs. (**g**) *flam*^*KG*^*/Df*; *mnk*^*p6*^/*mnk*^*p6*^ germarium showing increased intensity of γ-H2Av signals in Tj-positive somatic cells. (**h**) Quantification of γ-H2Av-positive ECs in *flam*^*KG*^*/Df*; *tj-Gal4/Cy* (no KD, control), *flam*^*KG*^*/Df*; *tj-Gal4* \> *mnk KD* and *flam*^*KG*^*/Df*; *tj-Gal4* \> *mei-41 KD* ovaries. Mean +/− s.d. are indicated (Student's t test; \*p \< 0.001). **(i)** Immunostaining of *flam*^*KG*^*/Df*; *tj-Gal4* \>*mnk KD* germarium with γ-H2Av (red) and lamin (blue) and Tj-marked somatic cells (green) (**i')**. **(j,j')** Immunostaining of *flam*^*KG*^*/Df*; *tj-Gal4* \>*mei-41 KD* germarium with the same antibodies. (**k**) A working model of the occurrence of germ cell differentiation defects due to TE activation in ovarian somatic cells. Scale bars, 10 µm. Discussion {#Sec7} ========== Activation of TEs in ovarian somatic cells is known to compromise differentiation of germ cells^[@CR27],[@CR38],[@CR39]^. Here, we found that the accumulation of GSC-like cells caused by mutations in the somatic piRNA cluster, *flam*, is determined by an insufficient number of somatic ECs. We demonstrated that the decrease of EC production in *flam* mutants, as well as the formation of germline tumor phenotype, depend on the events which occur in larvae and possibly at earlier stages of development, but not in the adult ovaries (Fig. [5k](#Fig5){ref-type="fig"}). These observations are consistent with previous report showing that Piwi expression in intermingled cells during the larval L3 stage is required to restrict GSC number in adults^[@CR36]^. Of note, somatic piRNAs against TEs were found to be produced de novo in large amounts between embryogenesis and the L3 stage^[@CR79]^ that shows the piRNA pathway activity in somatic cells of larval ovaries. Our finding of abundant DNA breaks in intermingled cells of *flam* mutant (Fig. [4b,c](#Fig4){ref-type="fig"}) indicates that activated TEs affect the genome of EC precursors at the larval stage. Interestingly, the intermingled cells accumulated more DNA breaks than mature ECs (Fig. [4](#Fig4){ref-type="fig"}). The reason for this vulnerability of somatic niche to TE activation during larval development remains unclear. Mobilization of TEs in the germ cells was shown to initiate the checkpoint response. In developing oocytes, TE-induced DNA breaks trigger Chk2-dependent oocyte polarization abnormalities^[@CR40],[@CR41]^. Consistent with this, *chk2* mutation suppresses polarization defects in the oocytes of piRNA pathway mutants^[@CR40],[@CR41]^. Derepression of TEs in GSCs leads to Chk2-mediated arrest of cell cycle^[@CR42],[@CR43],[@CR80]^ and induction of p53 activity^[@CR81]^, which launches DNA repair or apoptosis. In particular, mutation of *aub* gene encoding a germline-specific piRNA-binding protein is phenotypically manifested as a decrease of GSC number and a delayed differentiation of cystoblasts^[@CR42],[@CR43]^. The *chk2* mutation partially rescues these defects^[@CR42],[@CR43]^. Transpositions of *P* element during hybrid dysgenesis also induce Chk2-dependent arrest of germ cell differentiation and selective apoptosis of some GSCs, whereas mutating Chk2 restores GSC self-renewal and normal looking germaria^[@CR38],[@CR80]^. However, in this case *chk2* mutants show strong γ-H2Av signals and death of some cells at all oogenesis stages and never restore fertility^[@CR80]^. Interestingly, GSCs in dysgenic females are able, over time, to acquire resistance to *P* element due to the piRNA amplification by ping-pong mechanism, whereas this adaptation does not occur in *chk2* mutants^[@CR80]^. As a result, the ovarian defects in older dysgenic females are enhanced by *chk2* mutation^[@CR80]^. Here, we for the first time examined the role of checkpoint response upon genomic stress caused by TE activation in somatic cells of ovaries, which lack ping-pong piRNA amplification^[@CR11],[@CR14],[@CR15]^. We found that the absence of Chk2 or ATM kinases in the *flam* mutant background leads to dramatically more severe oogenesis defects compared to those induced by the *flam* mutation alone (Figs. [5](#Fig5){ref-type="fig"} and [S7](#MOESM1){ref-type="media"}). Thus, in contrast to germ cells, the Chk2-dependent response to TE activation in somatic ovarian cells is critical for the preservation of normal ovarian structure. The observed phenotypes of *flam mnk* double mutants indicate the loss of somatic cells due to the accumulation of unrepaired DNA lesions (Fig. [5](#Fig5){ref-type="fig"}). Our results suggest that the primary function of the Chk2-mediated response in ovarian somatic cells is the induction of DNA repair (Fig. [5k](#Fig5){ref-type="fig"}). The canonical activation of DNA repair/apoptosis pathways following DNA damage requires Chk2-mediated phosphorylation of p53^[@CR75],[@CR76]^. However, p53 activity in *Drosophila* ovaries was shown to be restricted to GSCs and cystoblasts^[@CR81]^, suggesting that in ovarian somatic cells Chk2 induces DNA repair by an unknown mechanism, which is of interest for further research. Methods {#Sec8} ======= Drosophila stocks {#Sec9} ----------------- *Drosophila melanogaster* stocks were maintained under standard conditions at 25 °C. For analysis of *flam* mutations the following stocks were obtained from the Bloomington Drosophila Stock Center: *w*^1118^ *P{GT*1*}*^1^*, flam*^*BG02658*^ (\#13912, *flam*^*BG*^), *y*^*1*^ *P{SUPor-P}flam*^*KG00476*^ (BDSC \#16453, *flam*^*KG*^) and *Df(1)Exel6255*, *w*^1118^ *P{XP-U}Exel6255*/*FM7c* (BDSC \#7723, *flamDf*). To distinguish between *flam*^*KG*^*/Df* and *flam*^*KG*^/+ larvae we used the *y*^1^ *w*^*67c23*^ *Alr*^1^/*FM7i, P{w\[ + mC\] = ActGFP}JMR3* (BDSC \#25048) balancer and manually selected GFP-positive and GFP-negative larvae. To visualize ECs *flam* mutations were combined with the *PZ1444 lacZ* enhancer trap line^[@CR47],[@CR48]^. For *piwi*, we used *piwi*^2^ and *piwi*^3^ null mutations^[@CR34]^ and *piwi*^*Nt*^ mutation with disrupted Piwi nuclear localization^[@CR44]^. To analyze *piwi*^*Nt*^*/piwi*^*Nt*^ females we used a strain with a higher survival rate of homozygous flies due to a change of genetic background. The following UAS-RNAi stocks were obtained from Vienna Drosophila Resource Center (VDRC): piwi-RNAi (\#101658), vret-RNAi (\#34897, \#101134), armi-RNAi (\#103589), arx-RNAi (\#40480, \#40479), mei-41-RNAi (\#11251), Chk2-RNAi (\#110342). RNAi depletion or expression of proteins was induced by *UAS-Dicer tj-Gal4* driver active in most somatic ovarian cells^[@CR16],[@CR82]^ or *c587-Gal4* driver active in ECs and early follicle progenitors^[@CR9],[@CR37]^. Other fly stocks were the following: *Df(2 L)Prl* and *zuc*^*HM27*^ (from T. Schüpbach lab), *mnk*^*p6*^ (*lok*^*p6*^)^[@CR71]^ (from M. Simonelig lab), *tefu*^1^, *tefu*^*red3*1^ ^[@CR83]^, *fs(*1*)Yb*^*1*^ (*Yb*^*1*^), *fs(1)Yb*^[@CR72]^ (*Yb*^[@CR72]^)^[@CR33]^, *bamGFP*^[@CR4]^, *Batumi*, *P{w(+mC) = UAS-arm.S10}C, y(1) w(1118)* (BDSC \#4782, *UAS-arm.S10*), *y(1) w(1118); P{w(+mC) = UAS-arm.Exel}2* (BDSC \#8369, *UAS-arm*), *y(1) w(67c23)*; *P{w(+mC) = lacW}tkv(k16713)*/*CyO* (BDSC \#11191, *tkv-lacZ*), *y(1) w(\*); P{w(+mC) = UAS-mCD8::GFP.L}LL5* (BDSC \#5137, *UAS-mCD8::GFP*). Immunostaining {#Sec10} -------------- For spectrosome analysis ovaries from 7-day-old females were used, and for other purposes - as indicated in the text. We revealed that germline differentiation defects in *flam* and *piwi* mutants are more pronounced in the progeny of older parents and to standardize further analysis we used offspring from the parents less than three weeks old. Immunostaining was basically performed as described previously^[@CR84]^ with some modifications. Ovaries were manually isolated in PBT (PBS containing 0.01% Tween-20) at 4 °C, rinsed in PBS and fixed in 4% formaldehyde (in PBT) for 25 min at room temperature. Fixation was stopped by incubation with 0.25 M glycine (Sigma-Aldrich) for 5 min. Then ovaries were washed in PBS three times for 10 min at room temperature, permeabilized with PBTX (PBS with 0.1% Tween-20, 0.3% Triton X-100) for 10 min, blocked with PBTX containing 3% normal goat serum (NGS, Invitrogen) for 3 h, incubated with primary antibody in PBTX containing 3% NGS for 7 h at room temperature, or overnight at 4 °C, washed in PBTX three times for 10 min, incubated with secondary antibodies (1:1000) in PBTX containing 3% NGS for 7 h or overnight in a dark chamber, and then washed in PBTX three times for 10 min. Coverslips were mounted with a drop of SlowFade Gold Antifade reagent (Invitrogen) containing DAPI. The following primary antibodies were used: rabbit anti-lamin Dm0 (1:500, provided by P. Fisher^[@CR85]^), chicken anti-β-galactosidase (1:500, Abcam, ab9361), mouse anti-β-galactosidase (1:200, DSHB \#40-1а), rabbit anti-pS10H3 (1:200, Millipore \#MC463), rabbit anti-GFP (1:500, Abcam, ab290), mouse anti-α-spectrin (1:200, DSHB, 3А9), rabbit anti-γ-H2av (1:100, Rockland, anti-H2AvD pS137), rat anti-Vasa (1:100; DSHB), guinea pig anti-Tj (1:5000, a gift from Dorothea Godt), rabbit anti-Caspase-3 antibody (1:200; Abcam, ab13847). The following secondary antibodies (Invitrogen, Thermo Fisher Scientific) were used: anti-rat IgG Alexa Fluor 546; anti-rabbit IgG Alexa Fluor 488; anti-rabbit IgG Alexa Fluor 546; anti-rabbit IgG Alexa Fluor 633; anti-mouse IgG Alexa Fluor 488; anti-mouse IgG Alexa Fluor 633; anti-chicken IgG Alexa Fluor 633; anti-guinea pig IgG Alexa Fluor 488; anti-guinea pig IgG Alexa Fluor 633. Confocal microscopy was done using LSM 510 META system (Zeiss). TUNEL assay {#Sec11} ----------- TUNEL staining was performed using Click-iT™ Plus TUNEL Assay for *In Situ* Apoptosis Detection, Alexa Fluor™ 647 dye kit (\#C10619, Invitrigen, Thermo Fisher Scientific) according to the manufacturer's instructions. EdU incorporation assays {#Sec12} ------------------------ For the two-hour EdU labeling, the ovaries were incubated in Grace's medium containing 10 µM EdU for 2 hours at 25 °C. For the EdU *in vivo* incorporation assay, females were fed on food with yeast paste containing EdU (0.5 mM) for three days. For larval EdU assay, parental flies were placed on EdU-containing food (0.5 mM), where larvae developed. Then newly eclosed flies were placed on food without EdU and after 3 days the ovaries were dissected and analyzed. The ovaries from all these types of assays were fixed, permeabilized as described above, and processed for EdU label detection using the Click-iT™ reaction according to the manufacturer's instructions. Click-iT reaction was carried out in a cocktail containing Alexa Fluor 647 azide, triethylammonium salt (\#A10277, Invitrogen) and Reaction Buffer Kit (\#C10269, Invitrogen) 30 min in the dark at room temperature. Then ovaries were washed in PBTX and processed for immunostaining. Western blot {#Sec13} ------------ Ovarian lysates were fractionated by SDS-PAGE (10% acrylamide gel) and transferred to a PVDF membrane (Immobilon-P, Millipore). Blots were developed using alkaline phosphatase-conjugated secondary antibody (Sigma) and the Immun-Star AP detection system (Bio-Rad). The following primary antibodies were used: mouse anti-Arm (1:500, DSHB), and mouse anti-β-Actin (1:3000; Abcam, ab8224). RT-qPCR analysis {#Sec14} ---------------- Total RNA was isolated from manually dissected ovaries using Trizol reagent (Invitrogen, Thermo Fisher Scientific) and cleared of genomic DNA by DNA-free kit (Ambion). For analysis of signaling pathway genes in germaria, RNA was isolated from 0-1-day ovaries containing no late stage egg chambers. 1 μg of total RNA was used for the reverse transcription reaction with oligo(dT) primer and Superscript II reverse transcriptase (Invitrogen). The resulting cDNAs in at least three biological replicates were analyzed by RT-qPCR performed in MJ Mini thermal cycler (Bio-Rad) using SYBR Green chemistry (Applied Biosystems). The following primers were used for PCR: Gypsy for CTTCACGTTCTGCGAGCGGTCT, Gypsy rev CGCTCGAAGGTTACCAGGTAGGTTC, Zam for3 TCACATCCTTCCAGCAATCTTCAA, Zam rev3 TATTACAGTTTCTGACATTATTTCTTCGTG, MDG1 dir AACAGAAACGCCAGCAACAGC, MDG1 rev CGTTCCCATGTCCGTTGTGAT, Idefix for AACAAAATCGTGGCAGGAAG, Idefix rev TCCATTTTTCGCGTTTACTG, dpp for2 GGCTTCTACTCCTCGCAGTG, dpp rev2 TGCTTTTGCTAATGCTGTGC, wnt4 for5 ATGATCCTCACCCACCTGAG, wnt4 rev5 ACCTGACCAGCATTGTTTCC, wnt2 for CAATAACCGAGCAGGGAGAAC, wnt2 rev CATGAGTCTATCGCCAACCAG, fz3 for TCTGCTTCGTCCTGACACTG, fz3 rev CCTTGCTTGATTGTGGAACAC, Rp49_up ATGACCATCCGCCCAGCATAC, Rp49_rev2 GCTTAGCATATCGATCCGACTGG. Supplementary information ========================= {#Sec15} Supplementary information. **Publisher's note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information ========================= is available for this paper at 10.1038/s41598-020-57901-2. We thank T. Schüpbach, M. Simonelig and the Bloomington Drosophila Stock Center for the fly stocks, M. Siomi and P. Fisher for antibodies and Y. Shevelyov for helpful discussion on the manuscript. The work was carried out with the use of the equipment of the common use center «Center of Cell and Gene Technology», Institute of Molecular Genetics, RAS. The work was supported by the grant from Russian Foundation for Basic Research \[16-04-01524 for M.K.\] and by the Presidium of the Russian Academy of Sciences program Molecular and Cell Biology (for V. G.). M.K., O.S., V.G. and E.M. designed experiments; O.S., E.M., S.K., Y.A. and M.K. performed experiments; M.K. wrote the manuscript with support from V.G.; M.K. coordinated the project. All data generated or analyzed during this study are included in this published article (and its Supplementary Information Files). The authors declare no competing interests.

Background {#Sec1} ========== According with the definition of Abrams et al. \[[@CR1]\], which is accepted and adopted also by recent clinical and epidemiological studies \[[@CR2], [@CR3]\], urinary incontinence (UI) is the condition in which an involuntary loss of urine occurs, which is well identifiable and can create problems of a hygienic and social nature. Although the International Continence Society (ICS) withdrew the hygienic and social aspects of UI from the latest definition reported by the ICS Glossary, this study adopted the statement of meaning of Abrams et al. \[[@CR1]\], because it is broader and multidisciplinary and so compliant with the purpose of this research, i.e. to look at the UI and at its repercussions on family caregivers of older people with incontinence from a sociological perspective. UI is often a chronic, age-related disorder, frequently occurring in association with disability \[[@CR4], [@CR5]\]. The unprecedented and worldwide population ageing implies, therefore, that the prevalence of this phenomenon is likely to increase in the future. In Italy, for example, where 22.3% of the whole population is over 65 years old \[[@CR6]\], there are about 4,5 million people suffering from UI (i.e. 7.8% of the population) \[[@CR7]\]. This condition concerns between 15 and 35% of older people living in the community, and 30% of those in nursing homes and facilities \[[@CR7]\]. In Italy, as well as in Europe, older people with incontinence (OPI) are assisted mainly by family caregivers, often daughters, daughters-in-law and wives who deal with their older relatives' personal hygiene, change of absorbent devices and frequent washing of linen. This is not surprising, as family caregivers are often the main source of assistance in the informal care sector, representing the backbone of long-term care provision across Europe \[[@CR8]--[@CR11]\]. Several studies highlight that the physical and psychological burden of family caregivers of OPI is higher than that experienced by family caregivers of older people without incontinence \[[@CR12]--[@CR15]\], and that such a burden can be mitigated by social and healthcare services meeting family caregivers' needs \[[@CR13]\]. Literature shows also that when caregivers of OPI feel that they are supported in providing assistance, their perception of the burden of care may be alleviated \[[@CR16]\]. In Italy, services that explicitly target family caregivers are extremely limited, while those targeting OPI are more frequent. Among the latter, the main -- and often only -- service delivered by the Regional Health Systems (RHS) across the country is the free supply of absorbent products \[[@CR17]\] through provincial administrative structures called *Aree Vaste* (literally "Large Areas", occupying a position midway between the provincial and regional scale). The latter are responsible for the recognition of the disorder and of the granting of the free supply of absorbent products. In some regions, including the Marche region, each Large Area (LA) can decide for itself whether to distribute the products via pharmacies or deliver them to homes. As a result, methods of distribution of these products may differ within the same region. The number of absorbent products to be delivered each month and their type (model), in contrast, are decided by a regional law, which transposes and applies national legislation locally. For example, the Marche Region (Central Italy), where this study was carried out, at the time of this study data collection, allocated 60 pieces per month to individuals with incontinence, regardless of the severity of their disorder (Regional Law n. 1696/2012, which incorporated the law n. 135/2012, widely referred to as the "Spending Review"). Recently the Marche Region, in accordance with Law n. 716/2017, has doubled this amount, thus granting now 120 absorbent products per month (up to 150 for cases of serious incontinence), and delegating the LAs to decide the number of products to be granted. Other Italian Regional Health Systems (for example the Lombardy Region) have adopted the use of vouchers (i.e. a pre-paid receipt), but only as an experiment. International literature has taken into consideration the impact of the quantity and quality of absorbent products on the well-being of people with incontinence and their caregivers \[[@CR18]--[@CR20]\], and has examined the cost of incontinence management in terms of public expenditure \[[@CR21], [@CR22]\]. In Italy, two studies \[[@CR23], [@CR24]\] highlighted the negative impact of complex procedures and lengthy waiting times to receive absorbent products in an informal care setting. To date, however, there are no studies at national as well as international level that have examined the impact of the method of distribution of these products on family caregivers' perception of the support received. This study, conducted in the Marche Region between January and July 2016, seeks to identify the method of distribution of absorbent products that provides Italian family caregivers of OPI with the greatest support, and whether this can mitigate their burden. The research question which the study tried to answer was: do caregivers who receive the products at home feel more supported than those who collect them at the pharmacy? Methods {#Sec2} ======= Study design and participants {#Sec3} ----------------------------- To answer this research question, two groups of family caregivers, living in two LAs of the Marche region with two different systems for the supply of absorbent products were compared. The first group lived in Ancona (the region's capital city) and collected the products via pharmacies. The second group lived in Fermo (a small town in the south of the region's inland area), and received the products at home. The sample size was estimated in order to obtain the representativeness of the family caregivers of OPI residing in Ancona and in Fermo. In 2015, the older population of Ancona had 25,659 subjects \[[@CR25]\]. In the health district of Ancona in the same year, according to the data supplied by the Marche Region Health System, 1700 older people were entitled to receive absorbent aids, i.e. 7% of the total older population. Fermo counted 9251 older people \[[@CR25]\]. However, for Fermo the number of residents who had applied to receive absorbent aids was not available. The estimation of the sample, therefore, was based on available data for the regional capital. Consequently, assuming a prevalence of urinary incontinence in the older male population of between 11 and 34%, and that of between 5 and 69% in the older female population \[[@CR5]\], a sample of at least 100 subjects was estimated to be sufficiently representative of the OPI living in the Marche region. In order to ensure that the study complied with national ethical standards, the research protocol was submitted for approval to the Ethical Committee of the Marche Region, which deemed that ethical approval was not required as it was not a clinical trial. An ad hoc questionnaire was administered to 101 family caregivers of OPI, who were included in the study on the basis of the following criteria: they had to be caregivers of older people with a medium-to-severe incontinence level (without a catheter), who used at least 2 pads daily, and who were residents in the municipalities of Ancona and Fermo. Family caregivers were recruited from among all patients with UI admitted to the geriatric and neurology departments (which according to internal statistics had the highest annual OPI admissions) of two geriatric hospitals in Ancona and Fermo, between January and July 2016. Participants were recruited simultaneously in the two different geographical areas of the Marche region. Family caregivers were informed by the head nurse about the aims of the study and that they might be interviewed. They were then contacted in the ward by interviewers and informed about the aims and methods of the study, in compliance with national personal data protection legislation (Law no. 196 of 2003, recently coordinated with the 2018 General Data Protection Regulation). Only family caregivers who signed the informed consent form and agreed to take part in the study were interviewed, i.e. 89.4% of the caregivers contacted. Measurements {#Sec4} ------------ The questionnaire consisted of questions relating to socio-demographic (age, gender, living and working conditions, etc.) and health-related factors concerning both family caregivers (self-assessment) and OPI, who were assessed according to the Instrumental Activities of Daily Living (IADL) scale \[[@CR26]--[@CR28]\]. The variables generated by the self-assessment questionnaire for family caregivers and the IADL scale for OPI were used to analyse the main characteristics of both counterparts and the management of household assistance overall. A variable, "IADL lost", was built, indicating the sum of instrumental activities that the OPI in question can no longer perform. This variable was assigned a value of "\<= 9", if the older person was no longer able to perform 9 out of 10 activities, while it was assigned a value of "= 0" if he/she had lost the ability to perform all instrumental activities. The "chronic disease" variable is a binary (yes/no) variable. The variable "Number of absorbent products used in 24 hours" was broken down into three values: "\<= 2" if the older person changed 1 or 2 pads a day, "3--5" for 3 to 5 changes, and "\> 5" if the older person required more than 5 pads a day. The outcome variable of the study is the "Quality of perceived support", which was generated by the COPE Index "Quality of support" subscale, which has a reliability of α = 0.77--0.78 \[[@CR29], [@CR30]\]. Regarding the validity of this measurement, it is most highly correlated with the well-being and quality of life measures and with the Social Restriction Scale \[[@CR30], [@CR31]\]. The "Quality of support" subscale raw score ranged from 4 to 16. The score was dichotomised at a cut-off corresponding to the 50th percentile and was divided into two classes: a value between 4 and 10 indicated a low level of perceived support, while 11 and higher indicated a high level of perceived support. The 5-Well-Being Index \[[@CR32]\] is a short questionnaire used to assess the family caregivers' subjective psychological well-being. The raw score ranges from 0 to 25, where 0 represents the worst possible and 25 the best possible quality of life. A score below 13 indicates poor wellbeing, which is an indicator for testing for depression and an appropriate tool for screening this disorder \[[@CR33]\]. In light of this, the variable "poor wellbeing", identifying family caregivers at risk of depression, was assigned the value "1″ if the index was \< 13 or if at least one of the 5 items of the index had a score of 0 or 1. The variable was assigned the value "0″ if the index was \> 13. The variable indicating the opinion of caregivers on means of the pads distribution type they utilised, was followed by an open question to specify the reasons for the answer provided. Caregivers' preferences on how they would like to receive absorbent aids were also taken into consideration through a multiple-choice question. Finally, family caregivers were asked to give their opinion on the adoption of a voucher, i.e. a pre-paid receipt, which LAs would provide for people with incontinence for the free supply of absorbent products. The voucher-based system would allow users to choose their preferred type of absorbent products as long as the total cost of the products is below the expenditure threshold covered by the voucher. Respondents were asked to complete the following sentence: "The adoption of a voucher system for receiving absorbent products would be ..." with one of the following options: "positive", "indifferent", "negative", and were also asked to provide the reason for their choice through an open answer. Data analysis {#Sec5} ------------- Data are reported as mean (± SD) for continuous variables, and as absolute frequencies for categorical variables. Patients' characteristics, divided by geographical area of residence, were compared using the chi-square test for categorical variables, and the t-Student test for continuous variables. A probability value of \< 0.05 was considered statistically significant. Statistical analysis was performed using SPSS for Win V21.0 (SPSS Inc., Chicago, IL, USA). The association between "Quality of perceived support" and two geographical areas/methods of absorbent products delivery (i.e. Ancona and Fermo, respectively pharmacy distribution and home distribution) was analysed by means of a generalized linear model (GLM) by adjusting for the following confounding factors: family support (i.e. having a co-carer), income (i.e. being in employment and having any source of income) and living arrangements (i.e. cohabiting with the older person) \[[@CR34], [@CR35]\]. Results {#Sec6} ======= The 101 family caregivers of OPI enrolled in the study were asked to answer a questionnaire that included questions addressing both respondents and older people. The demographic characteristics of the sample of family caregivers and of the older relatives for whom they care are shown in Table [1](#Tab1){ref-type="table"}. The mean age of the family caregivers is 60.5 years; 83 respondents are females and 82 people are married. Almost two thirds of respondents (65 people) perceive their health as "bad" or "very bad". 51 respondents live with the assisted older person, in the same apartment, while 15 people live in the same building, but in different apartments. Only 8 family caregivers live 30 min away by car. 72 have secondary and higher education, while 12 have a bachelor's degree. 39 family caregivers are in paid work (26 full-time and 13 part-time job), 50 do not work (retired and unemployed) and 11 women are housewives. 89 caregivers provide their loved ones with emotional and psychological support, 79 provide physical and practical support, and 76 help the older relative with personal hygiene (multiple answers). 80 respondents of the 101 carry out activities directly connected to incontinence (i.e. cleaning, changing of absorbent aids and washing of linen) on a daily basis or at least twice a week. Table 1Characteristics of family caregivers to care recipientsFamily caregivers' characteristicsAll (*n* = 101) *Age (years)*60.5 (10.34) *Gender*  Male19  Female82 *Marital status*  Unmarried11  Married82  Widower4  Divorced/Separate4 *Health condition self-evaluation*  Very good/Good10  Nor good neither bad25  Bad/Very bad65 *Living condition*  With the cared for51  With spouse, cared for and children14  With cared for and migrant care worker4  With spouse, cared for and migrant care worker2 *Household proximity from the older person*  Same apartment51  In the same building but in different apartments15  At walking distance10  At 10 min by car bus or train16  At 30 min by car bus or train8 *Educational level*  None3  Primary School13  Secondary School36  High School36  Bachelor degree12 *Employment*  Full-time employed26  Part-time employed13  Retired43  Unemployed7  Housewife11 *Caregiving activities (from 3 times a week to every day)*  Emotional, psychological and social support89  Physical help79  Medicine administration78  Housework77  Personal hygiene76 *Frequency of caregiving activities related to the continence management*  Rarely/never20  Once/twice a week at least80Care recipients' characteristics *Age (years)*83.6 (9.35) *Gender*  Male42  Female59 *Number of pads per day*   \< 210  3--560   \> 531 *IADL lost*   \< =945  1056 *Chronic diseases*  No6  Yes95Data are mean (SD) or number of cases The mean age of OPI is 83.6 years. 43 are men and 59 are women. 10 respondents use up to two diapers a day, 60 use between 2 and 5, and 31 use more than 5 per day. 45 OPI have lost up to 9 out of 10 of the activities in the IADL and 56 have lost 10 IADL. 95 out of 101 older people suffer from a chronic illness. The sample was divided into two groups, according to the method of distribution of the absorbent products. The first group consisted of 47 caregivers (46.6% of the sample) who collected the continence products in pharmacies. We called "PHAD" (standing for "Pharmacy Distribution"). The second group consisted of 54 caregivers (53.4% of the sample) who received continence products at home, which we called "HOD" (for "Home Distribution"). Table [2](#Tab2){ref-type="table"} compares the main results for the two groups of caregivers. Table 2Family caregivers' level of support perceived, wellbeing and opinion on the pads delivery per method of distributionType of pads delivery systemPPHADs (*n* = 47)HODs (*n* = 54)Family caregivers' level of support perceived Low26 (65.0)15 (31.9)0.002 High14 (35.0)32 (68.1)Poor well-being No14 (29.8)25 (46.3)0.089 Yes33 (70.2)29 (53.7)Family caregivers' opinion about the modality of distribution of pads \"not suitable at all"5 (11.3)0 (0.0)\<  0.001 \"acceptable"25 (56.8)5 (9.6) \"completely suitable"14 (31.8)47 (90.4)Family caregivers' preferencesconcerning the method f distribution of continence products At home21 (47.7)0 (0.0)\<  0.001 Pharmacy17 (38.6)0 (0.0)  "It's ok"6 (13.6)54 (100.0)Family caregivers' opinion on voucher \"It is positive"35 (76.1)23 (43.4)\<  0.001 \"It does not make any difference"11 (23.9)20 (37.7) \"It is negative"0 (0)10 (18.9)Data are number of cases (percentage) Concerning the perceived support, 65% of caregivers in the PHAD group perceived a lower level of support (\<= 10) and 35% perceived a higher level of support (\> = 11). The results are virtually reversed among caregivers in the HOD group, 31.9% of whom perceived a lower level of support and 68.1% a higher level of support. The difference is statistically significant (*p* = 0.002). The association between perceived support level and distribution method remained significant even after correction for family support, income and living solution, according to the GLM. In addition, 70.2% of caregivers in the PHAD group reported "Poor well-being", as against only 53.7% in the HOD group. Moreover, concerning the respondents' opinions on the pads delivery method, 56.8% of caregivers in the PHAD group considered defined the distribution via pharmacies as "totally adequate", 31.8% considered it "adequate", and 11.3% "not at all adequate". The main reason for the latter response was difficulty for caregivers in leaving the workplace or the older people to go to the pharmacy. The responses changed significantly among caregivers in the HOD group, 90.4% of whom found home delivery as "totally adequate", 9.6% "adequate", and none found it "not at all adequate". Almost one half of caregivers in the PHAD group (47.7%) would prefer to receive products at home, while all caregivers in the HOD group were satisfied with home delivery and did not want to change. 76.1% of caregivers in the PHAD group were positive with regard to the introduction of a voucher system for the supply of absorbent products, as against only 43.4% of caregivers in the HOD group. 23.9% of caregivers in the PHAD group and 37.7% in the HOD group were indifferent. None in the PHAD group viewed the voucher system negatively, while 18.9% of those in the HOD group did. The difference is statistically significant (*p* \<  0.001). Discussion {#Sec7} ========== The aim of this study was to determine to what extent the method of distribution of absorbent products might affect the perception among OPI family caregivers of the support received and whether this, as a consequence, might influence the caregiving burden. In light of our descriptive data analysis, the characteristics of family caregivers and OPI are consistent with what is reported in the literature. Concerning family caregivers, our data confirms the prevalence of women, the poor health conditions experienced, and the heavy care burden imposed by managing incontinence \[[@CR8]--[@CR15]\]. As far as OPI are concerned, our results underline the very advanced age, the reduced autonomy, and high incidence of chronic diseases. In order to assess the impact of the method of distribution of absorbent products on the perception of support received among family caregivers of OPI, the study compared two groups of caregivers: the first collected the products from pharmacies (PHADs) and the second received them at home (HODs). The results show that there is a significant association between the specific absorbent products distribution system and family caregivers' perception of support received. Indeed, although PHADs and HODs received the same number of pads per month and of very similar quality, HODs perceived a greater level of support and reported higher levels of satisfaction with the service, which they did not wish to change. Moreover, the difference in the percentages of respondents reporting "Poor well-being" in the two groups (70.2% of PHADs vs 53.7% of HODs), albeit not statistically significant, should lead to additional research on the possible association between these two items (i.e. pads distribution system and caregivers' well-being). Indeed, people reporting "Poor well-being" rate might be at severe risk of depression \[[@CR32]\]. The results, therefore, show that the distribution of absorbent products to the recipients' homes better meets the needs of family caregivers involved in this study, making them feel better supported, with possible positive repercussions on their well-being. This study sheds light on the effects of the method of distribution of absorbent products on the caregiving burden. In doing so, it fills a gap in the existing literature, which is focused mainly on the impact of the quantity and quality of the continence devices on the management of the continence care \[[@CR18]--[@CR20]\] and on the public expenditure \[[@CR21], [@CR22]\]. Although a number of confounding factors were taken into account in the analysis (i.e. family support, living arrangements and income) \[[@CR34], [@CR35]\], social and cultural patterns, e.g. representations of elderly care, traditions and values system \[[@CR36]\], might have influenced the perception of the support received by family caregivers, which may be different in urban (i.e. Ancona) and rural (i.e. Fermo) areas. This study, due to its overall design, was not able to distinguish to what extent social and cultural factors might have influenced respondents' opinions. It is therefore recommended that future research addresses this issue, by including qualitative questions to identify the cultural drivers of individuals' preferences. In this respect, a mixed-method approach \[[@CR37], [@CR38]\] seems to be the most appropriate means of data collection for appropriately dealing with this issue. Such an approach might help take into account the multifactorial nature of caring for an older person with incontinence (such as multimorbidity, the need for constant watchfulness, the physical effort and psychological stress of family caregivers and so on), as well as the sensitivity of this kind of care, as it is closely associated with the patient's dignity and the risk of stigma and social exclusion. This study has only partially adopted this approach by adding some open questions. Nevertheless, in view of the above, it would be more appropriate for future studies to combine semi-structured interviews with the questionnaire. Despite these limitations, this is the first study in Italy to examine the effects of the distribution of absorbent products on family caregivers' perception of the support received, and its results could help to design more effective services for supporting them. The method of distribution of absorbent products is the meeting point between the citizen with a healthcare and the health service from which the response to such a need is expected. It should therefore be adapted to the needs of users, in order to ensure a more efficient match between health demand and supply, thus also improving family caregivers' well-being. Conclusions {#Sec8} =========== This study has analysed the level of support perceived by family caregivers of OPI in relation to the method of distribution of absorbent products, thus filling a gap in the literature available today. The results seem to suggest that the home distribution of absorbent products is of greater help for family caregivers of OPI living in the Marche region in Italy, regardless of working, housing and economic conditions. In light of these outcomes, any reform of the services supporting dependent OPI should therefore also take into account the preferences of their family caregivers concerning the method used to supply absorbent products, in order to provide a truly useful service. Such a reform should start from a systematic assessment of families' health and social needs. The health conditions of care recipients and family caregivers change over time, as do the formal support networks (e.g. public and private long-term care services) and the informal support networks (e.g. family members and close friends) on which they can count. Therefore, Italy's Large Areas and the Regional Health Systems should periodically distribute questionnaires to and conduct interviews with older people with incontinence and their family caregivers, in order to identify and monitor their needs and preferences not only regarding the quality and quantity of absorbent products (i.e. the "what" of the service), but also the method of distribution (i.e. the "how" of the service). Once the demand for health services has been assessed, the second step should be to trial new solutions for products distribution designed on the basis of the insights gathered, for example by doing a voucher-based system (where it is not yet in place) or a "personal budget", i.e. an amount of money granted by the RHS which people can use to buy absorbent products wherever they want (for example from supermarkets) with the option of home delivery. Indeed, given the enormous commitment required to family caregivers, any choice that saves them time and effort, including the method of distribution of absorbent products, can potentially mitigate the negative effects of care on their health and well-being. Allowing family caregivers to choose their preferred method of pad distribution would appear to be the best option in terms of the utility of the service. Just as the "one size does not fit all" principle was applied to the quality of absorbent products, it can similarly be said that "one pad distribution system does not fit all". GLM : General Linear Model HOD : Home Distribution IADL : Instrumental Activities of Daily Living LA : Large Area OPI : Older People with Incontinence PHAD : Pharmacy Distribution RHS : Regional Health System **Publisher's Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Special thanks to the head physicians and the nurses of geriatrics and neurology departments of INRCA geriatric hospitals in Ancona and Fermo for their cooperation during the recruitment phase. SS led the project and was responsible for conception, design, drafting and completion of the manuscript. PF made substantial contributions to the acquisition of data and analysis and contributed to drafting of the manuscript. GL made substantial contribution to the conception and design, was involved in drafting the manuscript and contributed critically for important intellectual content. SS, PF and GL gave final approval for the final version and took full responsibility for the content and agreed to be accountable for all aspects of the work, associated with accuracy and integrity. All authors have read and approved the final manuscript. This study was supported by Italian Ministry of Health Grant to Sara Santini and partially funded by Essity Italy. The latter did not play any role in designing the study nor in data collection, analysis and interpretation, nor in writing this paper. Data and material are available on request from the corresponding author. Marche Region Ethics Committee determined that ethics approval was not required for the study. Nevertheless, the procedures performed by the study and involving human participants were in accordance with the ethical standards of this Ethic Committee, the 1964 Helsinki Declaration and its amendments or comparable ethical standards, and with the national personal data protection legislation (Law no. 196 of 2003) recently coordinated with the 2018 General Data Protection Regulation. Written informed consent was obtained from all individual participants included in the study. Not applicable. The authors declare that they have no conflict of interest nor Essity Italy, which partially funded the study, as the company did not influence the study design nor the report of the results in any way.

Introduction {#s1} ============ Congenital nephrogenic diabetes insipidus (NDI) is a rare disease caused by genetic mutations in *AVPR2* or *AQP2* ([@r1]). *AVPR2* is located at the Xq28 locus, and it encodes arginine vasopressin receptor 2 (AVPR2). Mutations in *AVPR2* have been associated with X-linked NDI. *AQP2* is located at the 12q13.12 locus, and it encodes the water transporter aquaporin-2. Mutations in *AQP2* result in autosomal NDI. Here, we describe a male infant with a novel *AVPR2* variant who was referred to our hospital due to delayed head control. Case Report {#s2} =========== A 5-mo-old boy was referred to the Department of Child Neurology in Okayama University Hospital (Okayama, Japan) due to delayed head control. He was born to non-consanguineous parents. His birth weight was 3630 g (+ 1.39 SD) and birth length was 50.8 cm (+ 0.86 SD). He showed no signs of asphyxia. A recurrent fever of unknown origin was observed beginning at 2 mo of age. Poor weight gain was observed at 3 mo of age. Notably, he did not take any medications. At his first visit to our hospital, the patient's length was 62.6 cm (--1.55 SD) and weight was 6190 g (--1.82 SD). His heart rate was 128 bpm, and his body temperature was 36.8°C. He was dehydrated, and his head control was incomplete. Laboratory examination revealed that his sodium level and serum osmolarity were elevated, and his urinary osmolarity and specific gravity were noticeably low. Head magnetic resonance imaging scanning revealed that high posterior lobe intensity was absent in T1-weighted images. However, the pituitary stalk was intact, and no other abnormalities were identified. These findings were indicative of diabetes insipidus. Water deprivation test with vasopressin challenge test resulted in no urine osmolarity increase and no urine volume decrease. Moreover, the AVP level at admission was extremely high (114 pg/mL; reference range; 0.0--4.2 pg/mL). Based on these findings, we diagnosed the patient with congenital NDI. His urine volume decreased, his weight increased, and head control was achieved after initiation of hydrochlorothiazide. Assessment of the patient's family history revealed that his mother had polydipsia and polyuria from infancy until the present day ([Fig. 1A](#fig_001){ref-type="fig"}Fig. 1.A: Family tree of our patient (arrow). Black square and black circle indicate family members with polydipsia and polyuria, respectively. Arrow indicates our patient. B: Results of mutation analysis. Our patient showed hemizygous duplication of four base pairs (c.990_993 dup CAGC; single bold line indicates CAGC). His older brother did not show this duplication. His mother was heterozygous for the wild-type allele and the four-base pair duplication (dotted line).). She also had increased levels of sodium of unknown origin during each of her three pregnancies. Furthermore, the patient's maternal grandfather, maternal great-grandfather, and maternal aunt had histories of polyuria and polydipsia. We suspected that our patient had familial NDI and, thus, conducted a genetic analysis. Mutational Analysis {#s3} =================== Genetic analysis was approved by the ethical committee of Okayama University Hospital and conducted in accordance with the 1975 Declaration of Helsinki and subsequent amendments. Informed consent for genetic analysis was obtained from the patient's mother. We identified a hemizygous four-base pair duplication in exon 3 of *AVPR2* (c.990_993 dup CAGC, [Fig. 1B](#fig_001){ref-type="fig"}). This four-base pair duplication results in early termination (p.Val332Gln Fs26X). This variant was previously unreported, and it is not present in the 1000 Genomes Project databases (http://www.internationalgenome.org/1000-genomes-browers), the Human Genetic Variation Database (http://www.hgvd.genome.med.kyoto-u.ac.jp), or the Exome Aggregation Consortium Server (http://exac.broadinstitute.org). This variant is considered to be likely pathogenic based on ACMG criteria (PS1, PM2 and PM4). We suspected that the patient's mother was heterozygous for the wild-type allele and the four-base pair duplication. Thus, an amplicon obtained from the mother was cloned into a PCR^®^ 4-TOPO^®^ vector using the TOPO^®^ TA Cloning^®^ Kit for Sequencing (Thermo Fisher Scientific, Waltham, MA, USA), and each transformant was, subsequently, sequenced. The results confirmed heterozygosity for the wild-type allele and the four-base pair duplication. Since the patient's mother suspected that the patient's older brother had polydipsia, we also analyzed the patient's older brother, However, the patient's brother did not carry the four-base pair duplication. Notably, we did not analyze *AQP2* in our patient, and we did not perform genetic analysis for his father, older sister, maternal aunt, maternal grandfather, or maternal great-grandfather. Discussion {#s4} ========== We identified a novel *AVPR2* variant in a patient with familial congenital NDI. Duplication variants are relatively rare among X-linked NDI patients: insertion *AVPR2* variants were found in only about 5% of large cohort ([@r1]) and Japanese patients ([@r2]). There are five classes of loss of function mutations in *AVPR2* ([@r1]). Class I mutations result in unstable mRNA, which undergoes nonsense-mediated RNA decay. Class II mutations cause misfolding of receptors, and these mutant receptors remain in the endoplasmic reticulum. Class III mutations also cause receptor misfolding. Class III receptors reach the plasma membrane and react with AVP, but subsequent interactions with G proteins and cAMP production are impaired. Class IV mutations also result in misfolded receptors that reach the plasma membrane, but result in incomplete AVP binding. Class V mutations cause mis-sorting to incorrect cellular compartments. A duplication of four bases in exon 3 of *AVPR2* causes early termination. However, this is the final exon, which might avoid nonsense-mediated RNA decay ([@r3]). Residue 332 is located in an intracellular domain that does not bind to AVP. Therefore, the variant in our patient may lose its function via a non-class I mechanism. Functional analysis is needed to clarify the pathophysiology of this variant. Congenital NDI is generally identified by the presence of DI symptoms like polyuria, polydipsia, fever, and poor weight gain ([@r2]). The patient whose case is reported here was referred to the department of Child Neurology in our hospital because of delayed head control, poor weight gain, and fever. After treating his dehydration, he gained head control and his development proceeded normally. We consider delayed motor development in infancy to be an important clinical symptom of congenital NDI. The patient's mother had a heterozygous *AVPR2* variant. She demonstrated a high sodium level during each of her pregnancies, but her sodium levels were normal before pregnancy and after birth for all children. In gestational DI, urinary volume increases during pregnancy because of increased clearance of AVP by placental vasopressinase, a cystine aminopeptidase that strongly degrades AVP ([@r4]). Although inactivation of the X chromosome was not analyzed in the patient's mother, AVP resistance and degradation by vasopressinase, might have contributed to her high sodium levels during pregnancy. In conclusion, we identified a novel *AVPR2* variant in a patient with familial congenital NDI. Common symptoms of DI were observed, including polyuria, polydipsia, poor weight gain, and fever, and delayed motor development was also observed during infancy. Conflict of Interest {#s6} ==================== The authors declare that they have no conflict of interest. We thank the patient and his family for participating in this study.

Introduction {#j_bjmg-2019-0017_s_001} ============ Alexander disease (AxD) is a rare autosomal dominant leukodystrophy with three clinical subtypes: infantile, juvenile and adult. The neonatal forms were also described \[[@j_bjmg-2019-0017_ref_001], [@j_bjmg-2019-0017_ref_002], [@j_bjmg-2019-0017_ref_003], [@j_bjmg-2019-0017_ref_004], [@j_bjmg-2019-0017_ref_005]\]. Forms differ by age of symptoms occurrence and the clinical picture. Although recent data suggest considering only two subtypes: type I (infantile onset with lesions extending to the cerebral hemispheres); type II (adult onset with primary involvement of subtentorial structures). The infantile subgroup with early onset within 2 years of life is the most common. Dominant mutations in the glial fibrillary acidic protein (*GFAP*) gene in AxD cause dysfunction of astrocytes (a type III intermediate filament). The disease manifests with macrocephaly, retarded psychomotor development and seizures. Other common symptoms are: ataxia \[[@j_bjmg-2019-0017_ref_001],[@j_bjmg-2019-0017_ref_003],[@j_bjmg-2019-0017_ref_004],[@j_bjmg-2019-0017_ref_005]\], pyramidal signs \[[@j_bjmg-2019-0017_ref_003],[@j_bjmg-2019-0017_ref_004],[@j_bjmg-2019-0017_ref_005]\], including spasticity \[[@j_bjmg-2019-0017_ref_001],[@j_bjmg-2019-0017_ref_004]\]. Magnetic resonance imaging (MRI) shows demyelinating changes mainly in the frontal lobes. Genetic background is a mutation in the *GFAP* gene with variable clinical features. The authors present a case of infantile onset AxD with normal head circumference. **Case Presentation**. The authors present a boy with an unremarkable family history, a second child of non consanguineous parents, born vaginally at 41 weeks' gestation. The perinatal period was uneventful. He weighed 4280 g (50-90 percentile), his head circumference was 33.5 cm (3-10 percentile) and his birth length was 58 cm (\>97 percentile). The boy was rated 8 points in the Apgar scale. The parents were concerned about the child's delayed development: independent sitting at 9 month, first words at approximately 13 months, abnormal behavior, the boy rarely cried and had no interest in toys. Because of generalized hypotonia, the patient was rehabilitated since his 8th month of life. Ultrasound examination of the head performed at 12 months showed enlargement of the lateral and third ventricle. Brain MRI (at the age of 13 months) confirmed internal hydrocephalus and directed suspicion to an aqueductal stenosis. Furthermore, it additionally revealed spots of impaired myelination. Areas of diffuse hyperintensity of white matter in the frontal lobe, within the left lentiform nucleus and in left-sided posterior part of the internal capsule ([Figures 1](#j_bjmg-2019-0017_fig_001){ref-type="fig"}, [2](#j_bjmg-2019-0017_fig_002){ref-type="fig"} and [3](#j_bjmg-2019-0017_fig_003){ref-type="fig"}). Features of intracranial hypertension that might require neurosurgery intervention were not found. At 18 months, the boy was hospitalized due to a prolonged episode of left-sided seizures, with secondary generalization followed by a longer period of altered consciousness. At hospital admission, the head circumference was 49 cm (75-90 percentile). On neurological examination, meningeal symptoms were absent, hypotonia (L\>R) with preserved tendon reflexes, were observed. Computed tomography of the brain excluded intracranial hypertension, the size of the lateral ventricles was comparable with previous MRI result ([Figures 4](#j_bjmg-2019-0017_fig_004){ref-type="fig"} and [5](#j_bjmg-2019-0017_fig_005){ref-type="fig"}). During the few next days of hospitalization, clusters of epileptic spasms occurred. Frequent myoclonia were also observed concomitant with temperature up to 38.4 °C. The longest status epilepticus persisted for 8 hours. Because of exposure to Herpes virus, serum and cerebrospinal fluid examinations were performed, and Herpes infection was excluded. Electroencephalography revealed generalized paroxysmal activity, especially in temporal and occipital areas in the form of a symmetric hypsarrhythmia. Inadequate response to anticonvulsant treatment required many drug modifications and was coupled with developmental regression. Antiepileptic treatment (valproic acid and lamotrigine) managed to stop seizures for 10 months. {#j_bjmg-2019-0017_fig_001} {#j_bjmg-2019-0017_fig_002} {#j_bjmg-2019-0017_fig_003} {#j_bjmg-2019-0017_fig_004} {#j_bjmg-2019-0017_fig_005} The boy started sitting independently again, and at approximately 2 years of age, he started walking and speaking simple sentences, but the speech was unclear. Later, he was repeatedly hospitalized due to intensification of polymorphic seizures during infections. Neurological examination showed the head circumference within the range of 50-75 percentiles, hypotonia with preserved tendon reflexes, wide-based gait and ataxia. Ophthalmologic examination was normal. Based on performed metabolic work-up, organic acidurias, lysosomal storage disorders, peroxisomal disorders were excluded. Serum amino acids, acylcarnitine profile, ammonia and lactate level and thyroid function tests were within normal limits. Renal and liver function tests, hearing evaluation, ultrasonography of abdominal cavity did not show any abnormalities. Because of the presence of fever-induced seizures, mutations in the *SCN1A* gene were excluded. The neuroimaging findings were suggestive of AxD but macrocephaly did not occur during any period of his life. The head circumference has never exceeded 2 standard deviations. At the age of 5, molecular tests confirmed presence of a *de novo* heterozygous mutation (c.236G\>A) in the *GFAP* gene, which is responsible for AxD. In the parents' samples, the described mutation was not detected. At the present time, the boy is 7 years and 8 months old and he is proceeding with an intensive rehabilitation. The neurological picture slowly deteriorates, the child presents independent and unsteady gait, he cannot run or jump. Vocabulary is limited to simple words and speech is still unclear. He recognizes people, is friendly and open. There have been no episodes of epileptic seizures since he was 3 years old; it may be partially connected with the less number of infections, which acted as a trigger factor. Brain MRI has been repeated once a year or once in 2 years, and it is stable compared to previous examinations. The supratentorial ventricular system is dilated, not displaced and symmetrical. Hyperintensive signals are visible from white matter in the frontal lobe with fragmentary insulas and frontal parts of the external capsule in both hemispheres. Additionally, MRI showed atrophy of white matter of the occipital and parietal lobes. Focal points of incorrect signal also include head of caudate nucleus and frontal part of the left lentiform nucleus ([Figure 2](#j_bjmg-2019-0017_fig_002){ref-type="fig"}). Discussion {#j_bjmg-2019-0017_s_002} ========== Pathomechanism of the disease is connected with a toxic gain of function mechanism, in which the mutated *GFAP* gene accumulates within astrocytes and aggregates with several other proteins to form inclusion bodies, known as Rosenthal fibers. Based on experimental studies there is a suspicion that accumulation of *GFAP* may lead to microglia activation and T-cell infiltration \[[@j_bjmg-2019-0017_ref_006]\]. Van der Knaap *et al*. \[[@j_bjmg-2019-0017_ref_007]\] proposed specific MRI criteria for the diagnosis of AxD including extensive, symmetric white matter abnormalities with frontal preponderance, periventricular signal changes, basal ganglia and thalamic signal changes, brainstem lesions, and contrast enhancement of multiple areas throughout the brain. The diagnosis required fulfilling at least four of five criteria. The presented boy fulfilled four of the five criteria. The most commonly used criteria were: extensive, symmetrical, cerebral white matter abnormalities with frontal preponderance, either in the extent of white matter abnormalities, the degree of swelling, the degree of signal change, or the degree of tissue loss (white matter atrophy or cystic degeneration) in 14/15 of the cases described \[[@j_bjmg-2019-0017_ref_001], [@j_bjmg-2019-0017_ref_002], [@j_bjmg-2019-0017_ref_003], [@j_bjmg-2019-0017_ref_004], [@j_bjmg-2019-0017_ref_005]\]. One of the main goals of this study was to describe a patient with an AxD involving hydrocephalus without macrocephaly as well as to compare all stages of the disease with other cases in the literature. Another exceptional feature seen in our patient is late-onset epileptic spasms with hypsarrythmia. Characteristic symptoms for the infantile form are seizures, retarded psychomotor development and macrocephaly. Muscle tone abnormalities are also common symptoms. Generalized hypotonia appears in the presented case. Literature data show 15 cases of infantile form: six children had spasticity and four had hypotonia. Pyramidal symptoms, which are more common in juvenile forms, were found in the presented boy and in two other cases \[[@j_bjmg-2019-0017_ref_002]\]. Hydrocephaly, which was one of the first alarming symptoms, was described only in one analyzed case \[[@j_bjmg-2019-0017_ref_005]\]. Gorospe *et al*. \[[@j_bjmg-2019-0017_ref_008]\] reported 12 genetically confirmed cases of AxD including seven with infantile onset. All had megalencephaly at presentation \[[@j_bjmg-2019-0017_ref_001], [@j_bjmg-2019-0017_ref_002], [@j_bjmg-2019-0017_ref_003], [@j_bjmg-2019-0017_ref_004], [@j_bjmg-2019-0017_ref_005]\]. Li *et al*. \[[@j_bjmg-2019-0017_ref_009]\] studied 26 subjects with infantile AxD and found macrocephaly in 62.0% of cases. In their study seizures (92.0%) were the most common feature followed by cognitive defects (82.0%). Bulbar signs (62.0%), ataxia (58.0%) and spasticity (52.0%), were observed less often. Wilson *et al*. \[[@j_bjmg-2019-0017_ref_010]\] described the longest survivor with AxD, who was diagnosed with this disease at the age of 5 years and is still alive at the age of 38. The final AxD diagnosis was possible through results of genetic testing. Despite the fact that R79H amino acid substitution is the most common AxD variant, we did not find any case of another child with the c.236G\>A mutation and similar clinical outcome, improvement after treatment of seizures, in the literature \[[@j_bjmg-2019-0017_ref_011], [@j_bjmg-2019-0017_ref_012], [@j_bjmg-2019-0017_ref_013], [@j_bjmg-2019-0017_ref_014], [@j_bjmg-2019-0017_ref_015]\]. There is also lack of reliable phenotype-genotype correlation in other *GFAP* gene variants. Of these, the p.Arg239His mutation was described twice \[[@j_bjmg-2019-0017_ref_003],[@j_bjmg-2019-0017_ref_004]\], and in one case macrocephaly was not found \[[@j_bjmg-2019-0017_ref_004]\]. The p.Arg79His mutation was seen twice \[[@j_bjmg-2019-0017_ref_003]\] and in one case, the disease proceeded with normal head circumference \[[@j_bjmg-2019-0017_ref_004]\]. The mutation p.Arg79Cys was also recorded twice \[[@j_bjmg-2019-0017_ref_005]\]. Both cases were associated with hypo-tension and epileptic seizures. The R239L mutation was described twice \[[@j_bjmg-2019-0017_ref_001],[@j_bjmg-2019-0017_ref_002]\]. Despite some description of epileptic spasms in the infancy period \[[@j_bjmg-2019-0017_ref_016],[@j_bjmg-2019-0017_ref_017]\], the onset of this form of seizure in our patient was surprisingly late. The study was limited by the small number of publications that were compared with the gathered results. This demonstrates the importance of our work, because with more publications it would be possible to associate the type of mutation with the clinical image and the changes visible in brain MRI. **Conclusions**. Alexander disease leads to severe disability and a child's death. Sometimes, with atypical presentations or course, such as in our patient. The disease is incurable and rehabilitation is the only chance to improve functioning of a child. It is essential to limit seizures and infections occurrence which may accelerate progression of the disease. We sincerely thank Professor Mario van der Knaap (Department of Child Neurology, Amsterdam, The Netherlands; Department of Functional Genomics, Center for Neurogenomics and Cognitive Research, Amsterdam Neuroscience, VU University, Amsterdam, The Netherlands) for her collaboration and assistance with data processing, MRI consultations and valuable suggestions. **Conflict of Interest** Declaration of Interest. The authors report no conflicts of interest. The authors alone are responsible for the content and writing of this article.

Kin selection theory predicts that individuals should behave less aggressively (enhanced social tolerance) or more amicably toward closer kin than less related individuals ([@zoy082-B22], [@zoy082-B23]). In species that live in family groups, individuals may simultaneously gain indirect and direct benefits from helping kin. For example, meerkats *Suricata suricatta* are more likely to perform sentinel behavior when pups join foraging groups ([@zoy082-B57]) and the sentinels have a lower predation risk when they are guarding ([@zoy082-B15]). Moreover, helpers may gain kin selection benefits from raising related offspring, and these offspring might also boost the direct fitness of helpers in the future ([@zoy082-B22], [@zoy082-B23]; [@zoy082-B32]; [@zoy082-B55]). Nevertheless, agonistic behaviors commonly occur between relatives due to competition for resources ([@zoy082-B24]; [@zoy082-B61]). In many cases, animals exhibit their peak aggression during the breeding season (BS) (e.g., eastern broad-toothed field mouse, *Apodemus mystacinus*: [@zoy082-B63]), and they behave less amicably toward kin during the BS compared with other times of the year (e.g., black-tailed prairie dog, *Cynomys ludovicianus*: [@zoy082-B25]). Thus, how individuals treat conspecifics depends on their genetic relatedness, but also on the ecological context, which influences the benefits and costs of cooperative and competitive interactions. Evidence increasingly indicates that unrelated conspecifics that engage in frequent interactions due to their close spatial proximity may develop "kin-like" behavior ([@zoy082-B5]; [@zoy082-B42]; [@zoy082-B56]). Unrelated individuals may gain immediate shared benefits from mutualism or reciprocity ([@zoy082-B11], [@zoy082-B12]; [@zoy082-B40]). For instance, as the home range overlap increases, there is a higher probability of nest-sharing among unrelated female dusky-footed woodrats *Neotoma fuscipes* ([@zoy082-B28]), thereby suggesting that proximity can lead to social tolerance in this species. In addition, many territorial species are less aggressive toward known neighbors than they are toward strangers ([@zoy082-B39]; [@zoy082-B43]; [@zoy082-B62]), and individuals may save time and energy by averting a costly and unnecessary fight with familiar neighbors ([@zoy082-B20]; [@zoy082-B48]). Mongolian gerbils *Meriones unguiculatus* are geographically widespread in the typical steppe, desert steppe, or desert areas of northern China, Mongolia, and the Trans-Baikal region of Russia ([@zoy082-B69]). Mongolian gerbils live in social groups comprising 2*--*18 individuals throughout the year ([@zoy082-B37]), where each group occupies an exclusive territory and all group members share the burrow system ([@zoy082-B3]). The reproduction and recruitment of Mongolian gerbils occur mainly from March to August ([@zoy082-B35], [@zoy082-B37]), and they start to store food from September to October ([@zoy082-B3], [@zoy082-B4]). Thus, there are 2 distinct annual life-history stages in Mongolian gerbils: the BS (March to August) and food-hoarding season (FHS) (September to October). Previous studies have shown that the home-range size of a social group increases with the number of male gerbils in the group during the BS, whereas it is positively correlated with the number of female group members during the FHS ([@zoy082-B66]). In addition, field observations and genetic data have demonstrated that social groups are basically family groups ([@zoy082-B3], [@zoy082-B4]; [@zoy082-B67]), and that inter-group genetic distances and geographic distances are positively related ([@zoy082-B65]). Trespassing by neighbors and chases involving individuals from adjacent groups are frequently observed during the BS ([@zoy082-B3], 1989b), and females commonly copulate with neighboring males ([@zoy082-B1], [@zoy082-B2]; [@zoy082-B3]). Consequently, the gerbils that live in different social groups may be kin. These characteristics make the Mongolian gerbil a suitable model species for investigating the influence of kinship and spatial proximity on social behavior, which have not been thoroughly examined in this species. In this study, we conducted an experiment to test our hypothesis that the social behavioral traits of individuals living in different groups are affected by genetic relatedness and spatial distance (distance between burrow systems) in a natural population of Mongolian gerbils *Meriones unguiculatus.* We predicted that gerbils may exhibit higher rates of amicable interactions and allocate less time toward exploring genetically and spatially close individuals. We also tested whether these effects on social behavior differed between sexes and seasons. The social associations and home-range overlap frequently occur among individuals from adjacent social groups during the BS ([@zoy082-B4]; [@zoy082-B66]), and fewer intergroup social connections are established during the FHS than the BS ([@zoy082-B17]). Thus, we predicted that the effects of kinship and spatial proximity on social behavior might be stronger during the BS than the FHS. In addition, multiple breeding females can be found at the same time in natural populations of wild gerbils and reproductively active male gerbils mainly defend the territories ([@zoy082-B4]). Therefore, we predicted that these effects may be stronger in males but not in female gerbils given that females usually encounter relatively lower reproductive competition than males ([@zoy082-B14]). Materials and Methods ===================== Study site, trapping, and animal identification ----------------------------------------------- Our studies were conducted at Houhatai (42°23.613′N, 116°06.524′E, altitude 1300 m), which is located about 25 km north of Shangdu, Zhenglan Qi, Inner Mongolia, China. The study site is a typical steppe area. The average monthly temperature ranged from --13.9°C to 21°C and the annual total precipitation was 324.6 mm during 2014. Our trapping plot comprised a 2-ha (100 m × 200 m) grassland dominated by *Leymus chinensis*, *Artemisia sieversiana*, *Thalictrum petaloideum*, *Stellera chamaejasme*, *Klasea centauroides*, and *Aster altaicus*, which provided food or cover for gerbils. The sympatric small mammals comprised the Daurian ground squirrel *Spermophilus dauricus*, Daurian pika *Ochotona dauurica*, and striped dwarf hamster *Cricetulus barabensis.* The potential predators of Mongolian gerbils were the steppe polecat *Mustela eversmanii*, corsac fox *Vulpes corsac*, and some raptors such as the common kestrel *Falco tinnunculus*and upland buzzard*Buteo hemilasius.* No livestock grazed on the study site during our study period. Mark-recapture experiments were conducted from April 29 to October 24 at 2-week intervals in 2014, where each trapping session lasted for 3 consecutive days. We did not trap during the winter to avoid mortality. Mongolian gerbils were live-trapped using wire mesh live traps (28 cm × 13 cm × 10 cm) baited with fresh peanuts. To enhance the likelihood of successful trapping, we used a concentric circle trapping method ([@zoy082-B35]). The trap station was arranged in 3 or 4 concentric circles within a burrow system to cover most of the range. In total, around 380 traps were set each time to cover all of the trap stations. The traps were set at 05:00*--*06:00 h from May to August, and checked every 1*--*2 h until about 1100 h. The traps were then closed between 11:00 and 15:00 h to avoid trap mortality due to the heat, and trapping was resumed at 16:00 h and continued until 19:00 h. Trapping started between 06:30 and 07:30 h, and continued until 17:30 h during September and October. We also checked the traps every 1--2 h during this period. Gerbils were active during the trapping periods employed in this study ([@zoy082-B35]). All of the captured gerbils were toe clipped when initially captured to allow permanent identification. The clipped toes were preserved in 95% ethanol for subsequent genetic analyses. All of the burrow systems were marked and their coordinates were recorded with a tape. In the field, a typical Mongolian gerbil burrow system comprises entrance holes, feeding chambers, nest chambers, and tunnels ([@zoy082-B71]). The core area of the burrow system usually has about 10--20 entrances holes in the ground, which form a burrow entrance cluster ([@zoy082-B3]). There are areas with few or no entrances between the burrow systems, so it is easy to determine the distribution of the burrow systems based on the burrow entrance clusters ([@zoy082-B3], [@zoy082-B66]). Therefore, coordinates were used to calculate the distance between each 2 burrow systems according to the Pythagorean Theorem, where the distance from the center point of the core area of the burrow system was the average distance between all of the burrows in one social group and those of another social group. The distance between burrow systems was used as an index of the spatial distance between Mongolian gerbils living in different social groups. The following data were recorded for each captured gerbil: location, sex, body mass, and reproductive condition (male: testes scrotal or abdominal, ventral scent glands invisible, clear contour or large visible pores surrounded by secreted substance; female: vulva closed or open, pregnant, lactating) ([@zoy082-B35]). Age was estimated based on body mass and the developmental stage of the ventral sebaceous gland (juvenile: \<30 g and with no sign of ventral gland; subadult: 30--50 g, unless they had a ventral active gland wider than 4.2 mm; adult: \>50 g) ([@zoy082-B3]; [@zoy082-B35]). We treated gerbils captured in the same burrow system in 2 consecutive trapping sessions as members of the same social group ([@zoy082-B3]). After recording the data, all of the gerbils were taken to a tent beside the plot and only adults from different groups were selected for behavioral tests. The trapping and handling procedures for the Mongolian gerbils were approved by the Institutional Animal Use and Care Committee of the Institute of Zoology, Chinese Academy of Sciences (Ethical Inspection License No: IOZ13047). Procedures in behavioral tests ------------------------------ To test the social behavior in pairwise encounters, we used a neutral arena ([@zoy082-B49]), which has been employed widely in rodent behavioral studies ([@zoy082-B31]; [@zoy082-B59]; [@zoy082-B70]). Adult gerbils (\>50 g) of the same sex captured from different burrow systems were selected for pairwise encounters in behavioral tests and they were matched randomly. Social behavior was tested by staging paired encounters in a rectangular neutral arena (42.5 cm × 31 cm × 19 cm), where each dyad was tested only once during a specific season in our study. The arena was divided into 2 equal compartments using a removable opaque partition. Two individuals were placed on either side of the partition and allowed to acclimatize to the novel environment for 5 min. The barrier was then removed and interactions were observed for a 10-min period. A digital voice recorder was used to record any behavior, which was observed continuously by a specific observer. We terminated the tests immediately if one of the following occurred: 1) continuous fighting physically for almost 1 min at a time; or 2) an actual injury occurred. We measured the durations of the following 4 categories of behavior observed during each 10-min test in the arena: 1) investigative behavior, including sniffing the nose, body, and anal zone of the other gerbil; 2) neutral behavior where the gerbils remained more than 5 cm apart without exhibiting any agonistic behavior and they ignored each other; 3) amicable behavior defined as the pair of gerbils located less than 5 cm apart and they exhibited affiliative behaviors such as remaining side by side, one over the other, or grooming; and 4) agonistic behavior such as upright boxing, defense, and wrestling. All of these behaviors have been observed in natural and captive populations of Mongolian gerbils ([@zoy082-B27]). The arena was cleaned thoroughly with 75% ethanol between tests to remove any odors from the previously tested individuals. At the end of each trial, all of the gerbils were released back into the burrow systems from which they were captured. Gerbils were housed individually in plastic cages (30 cm × 15 cm × 20 cm) during the behavioral tests and adequate food was provided. Animals were kept in the tent for less than 3 h. Pairwise relatedness based on microsatellite markers ---------------------------------------------------- DNA was extracted from tissues using a TIANamp Genomic DNA Kit (TianGen Biotech Company Ltd, Beijing, China). Nine microsatellite loci (*Mung*µ1, *Mung*µ2, *Mung*µ3, *Mung*µ4, *Mung*µ5, *Mung*µ6, *Mung*µ7, *Mung*µ8, and *Mung*µ9) developed for Mongolian gerbils were used to estimate relatedness ([@zoy082-B46]). Polymerase chain reaction (PCR) was conducted in a 10-µL reaction mixture containing 0.5 ng genomic DNA, 5 µL Premix Taq (TianGen Biotech Company Ltd), and 0.6 µM of the forward (fluorescently labeled with 5^\#^-TAMARA, HEX, or FAM) and reverse primers. PCR was run under the following conditions: initial denaturation at 95°C for 5 min, followed by 30 cycles of denaturation at 94°C for 40 s, annealing at a specific temperature (*T*~a~) for 45 s, and extension at 72°C for 1 min, with a final extension at 72°C for 5 min. The specific annealing temperature (*T*~a~) for each locus was specified by [@zoy082-B46]. Successful PCR amplification was verified by examining samples on agarose gels. Amplified fragments were electrophoresed on an ABI 3730 XL capillary sequencer (Applied Biosystems, Foster City, CA) and the allele size was determined with GENEMAPPER 4.1 software (Applied Biosystems). Hardy--Weinberg equilibrium tests were conducted with GenePop 4.3 ([@zoy082-B54]). Micro-checker 2.0 ([@zoy082-B64]) was used to test for the probability of null alleles. We calculated genetic diversity metrics comprising the observed heterozygosity (*H*~o~), expected heterozygosity (*H*~e~), number of effective alleles (*N*~a~), and inbreeding coefficient (*F~IS~*) using GenAlEx 6.501 ([@zoy082-B51]). The pairwise genetic relatedness was estimated between test pairs using GenAlEx 6.501 ([@zoy082-B51]) with Lynch and Ritland's estimator ([@zoy082-B38]). The polymorphic information content (PIC) and discriminatory power (DP) of loci were estimated using Cervus 3.0.7 ([@zoy082-B30]). Statistical analysis -------------------- The behavioral data were examined using Shapiro--Wilk test to determine their normality and the results did not indicate normal distributions (*P *\<* *0.001). Thus, we transformed the data using the Box-Cox method ([@zoy082-B21]) in the MASS package ([@zoy082-B53]). We used linear mixed effects (LME) models to determine whether the durations of 4 social behaviors were affected by the pairwise relatedness, spatial distance, sex, and season. Relatedness, distance, sex, and season were treated as fixed effects, and pairwise encounter and month as random effects. The results were expressed as the mean ± standard error (*SE*). *P *\<* *0.05 was considered to indicate a statistically significant difference. LME models were analyzed using the lme4 package ([@zoy082-B8]). All statistical analyses were performed with R software ([@zoy082-B52]). Results ======= In total, 253 gerbils (137 females and 116 males) from 24 burrow systems were identified during the study period. We captured gerbils 1371 times during the BS and 609 times during the FHS. The distance between 2 burrow systems ranged from 11.6 m to 147.0 m with an average of 66.1 ± 1.6 m. We conducted 249 pairwise behavior tests and about 77.1% (192/249) pairwise encounters were observed aggression. Initially, the gerbils in the pairwise encounters generally sniffed each other (i.e., investigative behavior) or kept away and ignored each other (i.e., neutral behavior). After several investigative and neutral behaviors, the encounters became amicable or agonistic. Amicable and agonistic behaviors did not occur within 10 min in some tests ([Table 1](#zoy082-T1){ref-type="table"} , [Figure 1](#zoy082-F1){ref-type="fig"} ). The relatedness between pairs of individuals was relatively low ([Table 1](#zoy082-T1){ref-type="table"}). ###### Pairwise relatedness (Lynch and Ritland's genetic relatedness coefficients) and the duration (seconds) of 4 behaviors in encounters during the breeding season (BS) and food-hoarding season (FHS) Sex Season Relatedness Investigative behavior Neutral behavior Amicable behavior Agonistic behavior Sample size (No. of paired encounters) -------- -------- ---------------- ------------------------ ------------------ ------------------- -------------------- ---------------------------------------- Female BS 0.019 ± 0.011 33.75 ± 2.89 496.66 ± 11.46 23.76 ± 10.35 34.20 ± 4.97 92 Female FHS 0.028 ± 0.018 62.51 ± 12.46 396.68 ± 26.09 17.12 ± 11.99 114.51 ± 21.81 41 Male BS −0.005 ± 0.010 36.35 ± 4.27 462.17 ± 12.96 6.81 ± 6.10 81.42 ± 11.07 78 Male FHS 0.017 ± 0.009 44.18 ± 6.88 432.95 ± 22.86 0.00 ± 0.00 110.29 ± 22.02 38 Data represent the mean ± *SE.* {#zoy082-F1} Genetic analysis ---------------- The *Mung*µ8 allele was not amplified by PCR and it exhibited relatively low polymorphism, so we excluded it from subsequent analyses. *Mung*µ3 deviated from Hardy-Weinberg equilibrium after sequential Bonferroni corrections for multiple comparisons (*P *\<* *0.01). Thus, we removed this locus from any further estimations of relatedness. The number of alleles per locus ranged from 8 to 11, whereas the mean *H*~e~ = 0.783 ± 0.020 and mean *H*~o~ = 0.789 ± 0.030. The mean PIC was 0.756 ± 0.022, the mean DP for 7 loci was 0.927 ± 0.014, and mean *F~IS~* was --0.211 ± 0.020 ([Table 2](#zoy082-T2){ref-type="table"} ). We detected no evidence of null alleles. ###### Characteristics of microsatellite loci used to estimate relatedness in Mongolian gerbils Locus Allele size (base pairs) *N* ~a~ *H* ~o~ *H* ~e~ PIC DP *F~IS~* ---------- -------------------------- --------- --------- --------- ------- ------- --------- *Mung*µ1 212--232 10 0.720 0.764 0.734 0.917 −0.223 *Mung*µ2 107--129 10 0.829 0.848 0.822 0.959 −0.194 *Mung*µ3 125--149 9 0.683 0.688 0.683 -- −0.199 *Mung*µ4 204--220 8 0.646 0.659 0.646 0.847 −0.128 *Mung*µ5 161--185 11 0.805 0.808 0.780 0.936 −0.261 *Mung*µ6 113--131 10 0.829 0.802 0.770 0.933 −0.230 *Mung*µ7 162--184 11 0.927 0.856 0.833 0.962 −0.145 *Mung*µ9 119--205 9 0.793 0.812 0.780 0.941 −0.311 Na, number of alleles; Ho, observed heterozygosity; He, expected heterozygosity; PIC, polymorphic information content; DP, discriminatory power; *F~IS~*, inbreeding coefficient. Effects of relatedness and spatial distance on behavioral traits ---------------------------------------------------------------- The pairwise relatedness had significant effects on investigative, neutral, and amicable behavior. In particular, the durations of investigative behavior (*t *=* *2.869, *P *=* *0.0046) and amicable behavior (*t *=* *5.432, *P *\<* *0.0001) increased significantly, whereas the duration of neutral behavior (*t*= --2.734, *P *=* *0.0071) decreased significantly with increasing relatedness ([Figure 1](#zoy082-F1){ref-type="fig"}, [Table 3](#zoy082-T3){ref-type="table"} ). We also found a significant negative relationship between the durations of amicable behavior and spatial distance (*t*= --2.562, *P *=* *0.0111, [Table 3](#zoy082-T3){ref-type="table"}), and amicable behavior only occurred between individuals with a spatial distance of less than 40 m ([Figure 1](#zoy082-F1){ref-type="fig"}f). In addition, the interaction between relatedness and distance had significant negative effects on investigative behavior (*t*= --2.022, *P *=* *0.0475) and amicable behavior (*t*= --7.505, *P *\<* *0.0001), but positive effects on neutral behavior (*t *=* *4.764, *P *\<* *0.0001, [Table 3](#zoy082-T3){ref-type="table"}). However, pairwise relatedness (*P = *0.1248) or spatial distance (*P = *0.5687) had no significant effects on the intensity of agonistic behavior ([Figure 1g-h](#zoy082-F1){ref-type="fig"}, [Table 3](#zoy082-T3){ref-type="table"}). ###### LME models for testing the effects of pairwise relatedness (Lynch and Ritland's genetic relatedness coefficients) and spatial distance (distance between burrow systems), sex, seasonality, and their interaction on the duration of social behaviors in Mongolian gerbils Behavioral category Estimate SE *t* *P* -------------------------- ----------- ---------- -------- ------------------- Investigative behavior  Sex ¦ Male −0.9020 5.7218 −0.158 0.8750  Seasonality ¦ FHS 18.7815 9.2196 2.037 0.0718  Distance −0.1565 0.1363 −1.148 0.2523  Relatedness 83.0134 28.9383 2.869 **0.0046** ^†^  Sex×seasonality −22.1472 11.9224 −1.858 0.0654  Sex×distance −0.0271 0.2682 −0.101 0.9197  Sex×relatedness −18.4826 60.3405 −0.306 0.7595  Seasonality×distance −0.5388 0.2845 −1.894 0.0603  Seasonality×relatedness 201.9876 59.4921 3.395 **0.0007** ^†^  Distance×relatedness −2.8057 1.3875 −2.022 **0.0475** ^†^ Neutral behavior  Sex ¦ Male −12.6218 16.9520 −0.745 0.4577  Seasonality ¦ FHS −69.3533 28.1718 −2.462 **0.0385** ^†^  Distance 0.3481 0.4028 0.864 0.3962  Relatedness −232.7818 85.1354 −2.734 **0.0071** ^†^  Sex×seasonality 83.9565 34.8727 −2.408 **0.0188** ^†^  Sex×distance 0.1029 0.7908 0.130 0.8967  Sex×relatedness 224.6669 176.3941 1.274 0.2035  Seasonality×distance −0.4918 0.8408 −0.585 0.5717  Seasonality×relatedness −358.4613 175.9891 −2.037 **0.0425** ^†^  Distance×relatedness 18.5908 3.9025 4.764 **\< 0.0001** ^†^ Amicable behavior  Sex ¦ Male −8.8841 10.0254 −0.886 0.3760  Seasonality ¦ FHS −2.7259 6.9205 −0.394 0.6945  Distance −0.5665 0.2211 −2.562 **0.0111** ^†^  Relatedness 251.6024 46.3225 5.432 **\< 0.0001** ^†^  Sex×seasonality −5.7055 13.8146 −0.413 0.6797  Sex×distance 0.5058 0.4339 1.166 0.2444  Sex×relatedness −209.8939 90.9768 −2.307 **0.0235** ^†^  Seasonality×distance 0.0862 0.3388 0.255 0.7992  Seasonality×relatedness 7.5315 71.0561 0.106 0.9158  Distance×relatedness −14.3321 1.9096 −7.505 **\< 0.0001** ^†^ Agonistic behavior  Sex ¦ Male 26.8136 12.7975 2.097 **0.0379** ^†^  Seasonality ¦ FHS 59.4638 18.1132 3.283 **0.0123** ^†^  Distance 0.1795 0.3046 0.589 0.5687  Relatedness −100.1879 64.8515 −1.545 0.1248  Sex×seasonality −56.0029 26.8553 −2.085 **0.0410** ^†^  Sex×distance −0.1014 0.6008 −0.169 0.8664  Sex×relatedness −129.0281 135.2281 −0.954 0.3405  Seasonality×distance 1.1558 0.6306 1.833 0.0734  Seasonality×relatedness 142.8407 136.6413 1.045 0.2973  Distance×relatedness 1.3495 3.1414 0.430 0.6706 ^†^Significant *P*-values are indicated in bold. FHS, food-hoarding season. The effects of relatedness and spatial distance on social behavior exhibited no consistent patterns in different sexes or seasons. The interaction between relatedness and sex had significant negative effects on amicable behavior (*t*= --2.307, *P *=* *0.0235, [Table 3](#zoy082-T3){ref-type="table"}). The interaction between relatedness and seasonality had significant positive effects on investigative behavior (*t *=* *3.395, *P *=* *0.0007), but negative effects on neutral behavior (*t*= --2.037, *P *=* *0.0425, [Table 3](#zoy082-T3){ref-type="table"}). In addition, there were significant negative interaction effects of sex and seasonality on neutral (*t* = --2.408, *P *=* *0.0188) and agonistic behavior (*t* = --2.085, *P *=* *0.0410, [Table 3](#zoy082-T3){ref-type="table"}, [Figure 2](#zoy082-F2){ref-type="fig"} ). {#zoy082-F2} Discussion ========== As expected, the social behavior in staged encounters between Mongolian gerbils was influenced by relatedness, spatial distance, and their interactions. Gerbils spent significantly more time sniffing related individuals and behaved more amicably with increasing pairwise genetic relatedness. Thus, gerbils were more indifferent to each other as the pairwise relatedness decreased. However, the significant interaction effects between relatedness and spatial distance on investigative, neutral, and amicable behavior indicate that the effects of kinship on social behavior were constrained by space. In addition, we found that amicable behavior occurred only between individuals with a spatial distance of less than 40 m, which is approximately the active range of Mongolian gerbils ([@zoy082-B3]). Furthermore, amicable behavior was exhibited by both close relatives and distantly related gerbils. These results imply that familiarity may play a key role in social associations of Mongolian gerbils. It should be noted that a higher level of familiarity does not necessarily indicate closer spatial distance, but it implies a closer social distance based on frequent interactions. For example, female banner-tailed kangaroo rats *Dipodomys spectabilis* avoid inbreeding via the development of familiarity based on prior associations rather than by using spatial cues, even if males live in familiar spatial locations ([@zoy082-B68]). Thus, the relatively low inbreeding coefficient in gerbils may be related to a similar behavioral strategy for inbreeding avoidance, although females commonly copulate with neighboring males ([@zoy082-B1], [@zoy082-B2]). However, there were no significant effects of relatedness or spatial distance on agonistic behavior. We observed that 192 out of 249 pairs of encounters resulted in aggression, thereby confirming previous direct observations that most intergroup interactions between gerbils are aggressive ([@zoy082-B3]). Many studies have demonstrated that high relatedness increases the incidence of amicable behavior among relatives such as supportive behavior ([@zoy082-B60]) and cooperation ([@zoy082-B34]), and it can reduce the intensity of aggression ([@zoy082-B18]). However, we found no evidence that agonistic intergroup interactions between gerbils changed with variations in the genetic or spatial distance. [@zoy082-B58] also reported that the intensities of inter- and intra-family aggression in Mongolian gerbils were generally similar. This suggests that the effects of relatedness and familiarity on aggression may be limited in this species, which indicates a high level of competition for food or mate resources between intergroup individuals. We also found that the effects of relatedness and spatial distance on social behavior exhibited sexual or seasonal patterns. Analyses of the LME models showed that there were interaction effects between relatedness and seasonality on investigative and neutral behavior, thereby suggesting that the effects of kinship on investigative behavior are stronger during the FHS but stronger on neutral behavior during the BS. The negative interaction effects between sex and relatedness on amicable behavior suggest that male gerbils are less intimate with individuals of the same sex than females when the pairwise relatedness is low, but the intensity of the increase in intimacy was higher for males as their relatedness increased. These findings support the idea that individuals may alter their behavioral strategies to adapt to variations in the ecological context ([@zoy082-B44]; [@zoy082-B50]). Our results showed that both male and female gerbils were highly aggressive during the FHS. The food-hoarding period is known to be a crucial period for non-hibernating northern small mammals ([@zoy082-B33]; [@zoy082-B45]). Our previous field study showed that fewer intergroup social connections were established during the FHS than the BS ([@zoy082-B17]). Other previous studies of Mongolian gerbils have demonstrated that the home range size is positively related to group size ([@zoy082-B3]; [@zoy082-B66]), and the home ranges overlap less during the FHS than the BS ([@zoy082-B66]), thereby demonstrating that food defense is an important function of territorial behavior ([@zoy082-B3]; [@zoy082-B19]). Furthermore, food availability is a determinant of winter survival and of the social groups or population sizes for Mongolian gerbils in the following spring ([@zoy082-B36]). Similarly, highly aggressive behavior during the defense of territories is also found in American red squirrels *Tamiasciurus hudsonicus*, which are highly aggressive during both the breeding and non-BS while defending food stored in their individual territories ([@zoy082-B9]). Therefore, we consider that the high levels of aggression between intergroup gerbils during the FHS associated with an increase in territoriality were caused by food resource competition. Nevertheless, we found significant negative interaction effects between sex and seasonality on agonistic behavior, which indicates that although Mongolian gerbils exhibited enhanced aggression during the FHS, the intensities of the changes differed significantly between the sexes. Female gerbils exhibited significantly lower aggression during the BS than the FHS, whereas the intensity of aggression in males did not differ significantly between the 2 seasons. Moreover, female gerbils exhibited significantly lower aggression than males during the BS. In most groups of animals, the intensity of aggression is lower in females than males, and the secondary sexual characteristics are generally less developed ([@zoy082-B16]), which suggests that females usually experience relatively lower reproductive competition than males ([@zoy082-B14]). Females primarily fight to defend access to resources such as food and shelter according to Bateman's principle ([@zoy082-B6]; [@zoy082-B7]). In Mongolian gerbils, males have opportunities to copulate with visiting or resident females by defending an undisturbed area, but they are probably unable to prevent their female partners from leaving their territories to solicit copulations in adjacent territories ([@zoy082-B3]). In our study, males were more aggressive against out-group individuals of the same sex than females during the BS, which probably reflects the paternity benefits of repelling intruding male gerbils. This finding supports the suggestion that intergroup encounters may have very different fitness impacts on males and females, and that responses to intruders may reflect differences in the benefit and cost trade-offs between the sexes ([@zoy082-B41]; [@zoy082-B47]). A second reason why agonistic interactions are less frequent and intense in females than males during the BS is that the risks associated with escalated conflict are usually higher for females than males ([@zoy082-B10]; [@zoy082-B13]). For example, a fatal injury to a female may also lead to increased mortality for any dependent offspring (e.g., ring-tailed lemurs, *Lemur catta*: [@zoy082-B29]). In addition, even when the risk of injury is relatively low, there may be an energy trade-off between chasing and pup feeding ([@zoy082-B41]). Thus, lactating females may rarely engage in aggressive interactions (e.g., chacma baboons, *Papio ursinus*) ([@zoy082-B26]). In conclusion, our results demonstrate that the behavioral traits of Mongolian gerbils that live in different social groups are affected by kinship, spatial proximity, and their interaction. Furthermore, we showed that the effects of kinship and spatial proximity on social behavior exhibited sexual or seasonal patterns, thereby indicating the occurrence of context-dependent responses to out-group individuals in Mongolian gerbils. Further studies are required to investigate the social association patterns and to quantify their costs and benefits in order to advance our understanding of behavioral interactions in social rodents. We are grateful to all the members of the Animal Physiological Ecology Group for helpful discussions. We also thank Mr Bin Wu, Plant Protection Station of Taipusiqi, for help with the field work. We are grateful to Dr Michael Cant for his help revising the manuscript. This study was financially support from the National Natural Science Foundation of China (No. 31372211) for W.L. and from the Chinese Academy of Sciences (KSCX2-EW-N-005) for D-H.W.

Background ========== Cancer is fundamentally a disease of genomic origin. Alterations in genes and regulatory elements critical to cell cycle control lead to uncontrolled cell growth and proliferation, the common signature of all cancers. Such events can cause amplification or mutational activation of oncogenes \[[@B1],[@B2]\], deletion or mutation deactivation of tumor suppressor genes \[[@B3],[@B4]\], orientation of genes with incorrect regulatory regions \[[@B5]\], gene fusion products \[[@B6]\], etc. As cancers evolve, they accumulate a cascade of mutations, ranging in size from a single nucleotide change to the gain or loss of entire chromosomes \[[@B7]\]. Coupled with the subclonal heterogeneity that is a hallmark of solid tumors \[[@B8]-[@B10]\], obtaining a complete portrait of the genetic landscape of human cancer remains a significant challenge. Synergy between revolutionary genomic tools and advances in high-throughput computing has facilitated the development of a number of methods for detecting mutations. Chromosome banding and spectral karyotyping (SKY) \[[@B11]\] are low-resolution techniques used to detect large-scale chromosomal features. However, obtaining metaphase spreads for performing a karyotype is often difficult, especially when working with solid tumor biopsies and paraffin embedded, formalin fixed tissue. Fluorescence in situ hybridization (FISH) and its variants are a family of molecular cytogenetic techniques developed to correlate specific sequences to cytogenetic observations \[[@B12]\]. FISH offers higher resolution (compared to SKY) and has the advantage of not requiring metaphase spreads, but is limited by the fact that it requires a prior hypotheses about the locus of interest, making it unsuitable for discovery based research. Hybridization based microarray approaches, like SNP microarrays and array comparative genome hybridization (CGH), have been extensively used to detect large scale amplifications and deletions in tumor genomes \[[@B13]-[@B15]\], but are unable to detect changes where there is no net gain or loss of DNA, such as inversions and balanced translocations, which have been shown to be an important mechanism for oncogenic transformation \[[@B16]-[@B18]\]. Moreover, microarrays do not offer structural information, necessitating follow-up experiments to identify the breakpoints and sequence context of the aberration. Microarrays are also restricted to regions of the genome amenable to unique probe design, which precludes repeat-rich regions and novel insertions that are hotbeds of variation and mutation \[[@B19]-[@B22]\]. Most commercial microarrays (except custom designed, high-density arrays) lose sensitivity below \~50 kb, and variants, particularly insertions, in this size range have remained largely unexplored, especially in cancer genomes \[[@B23]\]. The advent of massively parallel, short read DNA sequencing- the 'second generation' sequencing technologies, and their application to cancer has also accelerated the pace of mutation discovery. Initially applied to targeted subsets of the genome, such as specific gene families (e.g.: all protein kinases, or 'kinome') \[[@B24]-[@B27]\], or all the coding sequences (the exome) \[[@B28]-[@B33]\], second-generation sequencing is increasingly being used to interrogate whole cancer genomes \[[@B34]-[@B40]\]. In theory, second-generation sequencing of whole genomes has the ability to discern the full range of genomic alterations. In practice, however, more than 90% of events discovered by these platforms are less than 1 kb, and are biased towards deletions rather than insertions \[[@B23],[@B41]\]. Second-generation sequencing instruments typically generate shorter reads with higher error rates from relatively short insert libraries, which present a significant computational and bioinformatic challenge in alignment and assembly \[[@B42]\]. Read-pair mapping approaches have successfully identified point mutations and indels in cancer \[[@B36],[@B38]-[@B40]\], but are limited by the insert size of the DNA library to detecting base substitutions and small indels \[[@B43]\] and are often confounded by repetitive regions of the genome. Further, accurate prediction of the exact breakpoints of an aberration depends on very tight size distribution of the DNA library, which can make library construction difficult \[[@B44]\]. Whole genome sequencing followed by *de novo* assembly might mitigate some of these issues, but current assembly algorithms tend to collapse homologous sequences, and consequently dramatically under-represent repeats and segmental duplications that are known to be critical mediators of genomic rearrangement \[[@B42]\]. There remains a pressing need for discovery-based systems that can provide a scalable, comprehensive view of the cancer genome in its entirety. In this study, we present Optical Mapping as one such system. Optical Mapping creates high-resolution ordered restriction maps of whole genomes through the analysis of ensembles of single molecule restriction maps. It has previously been used to map the genomes of microbes \[[@B45]-[@B48]\], plants \[[@B49]-[@B52]\] and mammals \[[@B53]-[@B57]\]. However, this is the first time it has been employed to analyze the genome of a solid tumor. Optical Mapping offers several unique advantages towards assembling the complex structure of a cancer genome. Genomic DNA isolated directly from cells is analyzed, thereby obviating any bias introduced by amplification or cloning steps. Moreover, because the DNA is of high-molecular weight (300 kb - \>500 kb), segmental duplications and other repeat-rich regions of the genome are revealed, and additionally, the structure and long-range context of any aberration are determined. Since the restriction maps are made from single DNA molecules, Optical Mapping effectively pieces together heterogeneous alterations, which is especially important for tumor genome analysis, as we demonstrate in oligodendroglioma. Oligodendrogliomas are frontal lobe tumors that are thought to arise from oligodendrocytes, supporting brain cells which provide myelination for neurons \[[@B58],[@B59]\]. The concerted loss of heterozygosity (LOH) of chromosome arms 1p and 19q, observed in 50-70% of patients, is a molecular signature of this malignancy \[[@B60]\]. The remarkably high prevalence of this molecular marker suggests that these regions harbor one or more tumor suppressor genes that might play an important role in the development of the tumor. Allelic losses of 1p/19q have been correlated with positive response to chemo- and radiotherapy and prolonged survival for patients with oligodendroglioma \[[@B61]\]. However, it remains unclear whether LOH of 1p/19q is a prognostic biomarker for a more indolent tumor subtype that has fewer unfavorable mutations overall, rather than predictive of treatment sensitivity \[[@B62],[@B63]\]. In fact, studies have shown that 1p/19q codeleted tumors have slower growth rates and are more responsive to treatment than tumors without the codeletion \[[@B64],[@B65]\]. In order to explore each of these possibilities, Optical Mapping was used to create physical maps from two individual oligodendroglioma tumor biopsies for the purpose of identifying and characterizing structural changes on a whole genome basis. Results and discussion ====================== Optical map construction ------------------------ We used the Optical Mapping (OM) system to explore the genomic landscape of a solid tumor. Optical Mapping creates high-resolution physical maps of genomes through the analysis of ensembles of single molecule ordered restriction maps. The tumor biopsies were disaggregated into single cells, then run through a Percoll gradient to enrich for cancer cells (methods). High molecular weight genomic DNA was extracted directly from these cells, stretched and immobilized in regular arrays on positively charged glass surfaces using a microfluidic device (Figure  [1](#F1){ref-type="fig"}A, details in methods) \[[@B66]\]. After deposition, the DNA was digested with the restriction enzyme SwaI. The surface-bound restriction fragments remained in register, and were stained with a fluorescent dye and imaged by automated fluorescent microscopy (Figure  [1](#F1){ref-type="fig"}B). Dedicated machine vision software calculated the size, in kilobase pairs (kb), of each fragment based on measurements of integrated fluorescent intensity, resulting in the high throughput, massively parallel generation of ordered restriction maps, or 'Rmaps', from individual genomic DNA molecules (Figure  [1](#F1){ref-type="fig"}C) \[[@B66]\]. The oligodendroglioma datasets comprise close to 700,000 such Rmaps, with an average size of greater than 400 kb (Additional file [1](#S1){ref-type="supplementary-material"}). The relative order and distance between successive restriction fragments in a single molecule optical map can be used to determine the precise location in the genome that gave rise to that molecule, by means of pair-wise alignment against an *in-silico* restriction map \[[@B67]\]. The HF087 tumor dataset comprised 235,026 aligned Rmaps, corresponding to 36.92 fold coverage of the human genome, while the HF1551 dataset comprised 167,012 aligned maps, representing 25.36 fold coverage of the human genome (Additional file [1](#S1){ref-type="supplementary-material"}). In the absence of a karyotype, our assessment of ploidy is based on optical map coverage and Affymetrix array analysis. Both these platforms, discussed in detail in subsequent sections, calculate chromosome copy number relative to normal, diploid genomes, and are in agreement that neither tumor sample is polyploid. They do, however, display aneuploidy, due to allelic losses of specific chromosomes/chromosome arms (1p, 19q, 13 in HF087 and 1p, 19q, 14, 21 in HF1551), so if anything the coverage is likely to be higher than what we reported. {#F1} The Rmaps that cluster together upon pair-wise alignment were then assembled into consensus optical maps and analyzed for presence of structural variants using the bioinformatics pipeline described in \[[@B56]\]. The final consensus map contigs span 96.73% and 93.92% of the human genome for tumors HF087 and HF1551, respectively. Optical map coverage analysis ----------------------------- ### ***Discernment of copy number variants*** Copy number was inferred from aligned coverage of Rmaps, prior to assembly, in a manner analogous to read-depth based methods for detecting copy number variants from second generation sequencing data (methods). Briefly, Rmaps were aligned to the *in silico* reference map, and then partitioned into discrete windows spanning each chromosome. These alignments were then compared to alignments of a reference data set (comprising of a number of normal genomes) that was used to 'normalize' the observed coverage. This was necessary because the number of Rmaps that align to a particular region of the genome depends, in part, on the density of restriction sites in that region, which varies from chromosome to chromosome (ranging from a low of 2.5 cuts/100 kb on chromosome 22 to a high of 9.25 cuts/100 kb on chromosome 4). A Hidden Markov Model (HMM) was then fitted to this data, and copy number changes were detected (Figure  [2](#F2){ref-type="fig"}A) \[[@B68],[@B69]\]. Optical map coverage analysis confirmed the allelic loss of chromosome arms 1p and 19q in HF087 and HF1551. The breakpoints appear to be very close to the centromere, consistent with the proposed mechanism of an unbalanced reciprocal translocation mediating the LOH event \[[@B70],[@B71]\]. Additionally, coverage analysis also detected allelic loss of chromosome 13 (HF087), 14 and 21 (HF1551), which are known to be rarer events associated with oligodendroglioma. {#F2} ### ***Solid tumor heterogeneity*** The genome wide optical map of HF1551 was created using DNA from two adjacent slices of the tumor: 446,933 (\~55%) Rmaps originated from slice 1 and 202,974 (\~45%) Rmaps from slice 2 (Figure  [2](#F2){ref-type="fig"}). Interestingly, when the Rmaps were partitioned according to the slice they originated from, and coverage analysis was performed separately, unique copy number profiles were obtained for each slice. In addition to allelic losses of 1p and 19q, slice 1 also had LOH of chromosomes 14 and 21, while slice 2 had evidence of LOH of 19p (Figure  [2](#F2){ref-type="fig"}B). Solid tumors are dynamic aggregates of continually evolving subclones, resulting in spatial and temporal genetic heterogeneity. Our findings suggest that the tumor slices used for Optical Mapping evolved from distinct cancer cell clones, and is congruent with recent evidence of branched evolutionary tumor growth \[[@B72]-[@B74]\]. Although assembly of whole genome maps on a per slice basis was not feasible due to insufficient depth of coverage, our results establishes proof-of-principle of Optical Mapping to interrogate tumor heterogeneity. Discovery of optical structural alterations ------------------------------------------- The optical consensus maps generated by map assembler were aligned to the *in silico* restriction map (generated from build 35 human reference sequence), and by comparison of the order and sizing of the 219,224 restriction fragments (fragments smaller than 0.4 kb in size were merged) between the experimental and the reference map. Such comparisons revealed structural variants in the experimental genome that were classified as four types: extra cuts (EC), where the optical consensus map displays a restriction site that was not predicted by the reference sequence; missing cuts (MC), where a cut that was predicted was not observed in the experimental map; insertions (INS), where the size of a fragment in the consensus map was significantly larger than its counterpart in the reference map; deletions (DEL), where a fragment in the experimental map was smaller than the corresponding reference fragment (or missing altogether); and finally, complex events (OTHER) involving multiple cut or size differences (methods). Approximately a third of the ECs and MCs represent small indels that are below the resolution of Optical Mapping (\~3 kb) \[[@B56]\]. Figure  [3](#F3){ref-type="fig"}C shows an example of each class of variant detected by Optical Mapping. {#F3} At first glance, it might appear that any one of these variants could be attributed to errors inherent in Optical Mapping. For instance, a missing cut could be due to incomplete digestion, an extra cut could result from spurious cutting by the restriction enzyme, or physical breakage of the DNA molecule, and uneven staining could lead to inaccurate estimation of fragment size. However, the high throughput advantage of Optical Mapping allows us to distinguish such random errors from legitimate genomic events. Any alteration in the optical consensus map was supported by multiple single molecule maps (Rmaps), each representing an independent observation at that locus. The Optical Mapping error models estimated the statistical significance of each structural variant, after taking into account the quality and quantity of the data \[[@B56]\]. A total of 1081 and 1085 differences were detected in HF087 and HF1551 respectively (Figure  [3](#F3){ref-type="fig"}A and B). The distribution of structural variants across the genome is uniform and the pattern is similar for both tumors (Figure  [3](#F3){ref-type="fig"}A). Variants range in size from single base differences to complex genomic events spanning hundreds of kilobases (Figure  [4](#F4){ref-type="fig"}). Approximately 800 single base changes were detected in each tumor, including point mutations (such as the example depicted in Figure  [4](#F4){ref-type="fig"}A), polymorphic SNPs where only one allele has a SwaI restriction site (referred to as snip-SNPs \[[@B75]\]), and small indels that create or remove a SwaI cut site but are below the detection limit of Optical Mapping. 179 indels with a median size of 6.6 kb were detected in each sample (Figure  [5](#F5){ref-type="fig"}). For comparison, the median size of indels reported in the Database of Genomic Variants is 2.3 kb (Figure  [5](#F5){ref-type="fig"}, inset). \~70 complex events were found in each tumor, including known polymorphic loci such as the major histocompatibility complex (MHC), giving us confidence that these results are not spurious. Optical Mapping also discerns balanced genomic events, where there is no net gain or loss of genomic sequence. A putative inversion spanning 352 kb of chromosome 7 was observed in HF087 which appears to disrupt the ZNF92 gene (Figure  [4](#F4){ref-type="fig"}C). Finally, the largest events detected by Optical Mapping include gains or losses of entire arms of chromosomes, for example, the allelic loss of chromosome 1 illustrated in Figure  [4](#F4){ref-type="fig"}D, and discussed in detail previously. Intersection counts with genes, segmental duplications, published SNPs (dbSNP build 135, <http://www.ncbi.nlm.nih.gov/projects/SNP/>) and published structural variants (Database of Genomic Variants, November 2010 release) are shown in Additional file [2](#S2){ref-type="supplementary-material"}. Comprehensive breakdown of the overlaps are shown in Additional file [3](#S3){ref-type="supplementary-material"}. {#F4} {#F5} Optical Mapping provides a comprehensive description of the vast and complex landscape of cancer genomes. The ability to study the genome in its entirety, including non-genic or repetitive regions using a single technology minimizes ascertainment bias. As detailed in subsequent sections, it is employed to generate a list of candidate cancer genes that is not hypothesis-limited, and elucidate their structure at sub-genic resolution. Validation of copy number and structural variants ------------------------------------------------- ### ***Experimental validation: SNP array*** The Affymetrix Genome Wide Human SNP 6.0 Array, which has probes for detection of both SNPs and copy number variants (CNVs), was used to validate our findings. Both platforms concurred on the LOH of chromosomes 1p, 19q and 13, but allelic loss of chromosome 14 in HF1551 was not detected by the Affymetrix array. The copy number profile generated by running the HMM algorithm on the maps from the first slice of tumor HF1551 was similar to that from the array, which suggests that the DNA originated from tumor sections that were in closer proximity. Many of the SNP probes on the Affymetrix chip correspond to SwaI snip-SNPs. Hence, the array data was used to validate ECs and MCs. We observed 100% (62/62 in HF087, 44/44 in HF1551) concordance between the SNP genotype and the SwaI cut pattern at all overlapping cut differences in both tumors (Additional file [4](#S4){ref-type="supplementary-material"}). The copy number variants detected by the array were also compared to Optical Mapping indels. Signal intensities from the chip were normalized by global median scaling, and copy number was assessed using several different algorithms (methods), relative to a reference model file generated from the 270 HapMap samples. Though the resolution of array CGH is much lower than Optical Mapping, we were able to validate 24 structural variants in tumor HF087 and 16 in tumor HF1551 (Additional file [5](#S5){ref-type="supplementary-material"}). ### ***Experimental validation: PCR*** The nature of many of the structural variants, being within repetitive portions of the genome, but detected by Optical Mapping unfortunately precludes their comprehensive validation by simple PCR techniques. Accordingly, we selected two variants that were amenable to PCR and overlapped genes that may offer insights into the chemo- and radio-sensitivity of oligodendroglioma. These loci were then PCR amplified, cloned and sequenced (methods). The optical map shows an EC in the PARK2 gene in HF1551 (Figure  [6](#F6){ref-type="fig"}A). PARK2 is a putative tumor suppressor, and mutations in this gene have been reported in multiple cancer types (detailed in 'candidate mutations' section of this document). An 848 bp amplicon spanning the predicted location of the EC was obtained (Figure  [6](#F6){ref-type="fig"}B), and Sanger sequencing proved that a G to T transversion resulted in the creation of a new SwaI restriction site (Figure  [6](#F6){ref-type="fig"}C). {#F6} We also validated an EC in tumor HF087 that occurred in the STMN2 gene (Figure  [7](#F7){ref-type="fig"}A). As discussed in subsequent sections, STMN2 regulates microtubule dynamics and is believed to be a target of beta-catenin/TCF signalling. We amplified a 1003 bp region around the putative mutation (Figure  [7](#F7){ref-type="fig"}B), and were able to validate the alteration via sequencing (Figure  [7](#F7){ref-type="fig"}C). {#F7} ### ***Comparative validation*** We also validated our findings by comparing them to two sources- Optical Mapping data from several normal genomes, and publicly available SNP and structural variant data. First, oligodendroglioma structural variants were compared against structural variants found by Optical Mapping of 6 other normal human genomes by our laboratory. This internal database includes: (three lymphoblast-derived cell lines and a complete hydatiform mole (dbVar study ID nstd49) \[[@B56]\], a lymphocyte-derived cell line (unpublished) and an early passage human embryonic stem cell line \[[@B55]\]). 80%-90% of oligodendroglioma variants were also detected in at least one of the normal human genomes (Additional file [6](#S6){ref-type="supplementary-material"}), suggesting that such loci are polymorphic, and affirming the veracity of our findings. Then, oligodendroglioma structural variants were compared against variants in the Database of Genomic Variants (DGV). The DGV is an extensive catalogue of structural variation in normal humans, currently holding 101,923 events detected by a variety of platforms. We observed the greatest concordance with variants found by fosmid-end sequencing (\~15%) and high density oligonucleotide array CGH (\~10%) (Additional file [6](#S6){ref-type="supplementary-material"}). Finally, cut differences detected by Optical Mapping were compared to published SNPs. Detailed breakdown of these intersections are shown in Additional file [3](#S3){ref-type="supplementary-material"}; parameters for comparisons are described in the Methods section. Candidate mutations ------------------- ### ***Separation of mutational and polymorphic OSAs*** The ultimate goal of our mapping efforts was to identify genes or genomic elements that maybe important to the biology of oligodendroglioma, with the caveat that such 'candidates' represent hypotheses requiring rigorous testing to establish their functional role in tumorigenesis. Distinguishing between structural polymorphisms and somatically acquired mutations is a key step towards accomplishing this goal. Unfortunately, matched normal DNA from the individuals whose tumors were optically mapped was not available. Instead, we adopted a stringent filtering scheme to remove putative polymorphisms and enrich for somatic mutations, based on comparisons to internal and publicly available data (described above). Parameters for these comparisons were determined based on the Optical Mapping error model and designed to be extremely parsimonious (methods). As a result of these operations, we arrived at a total of 21 somatic mutations (5 genes) in HF087 and 73 somatic mutations (21 genes) in HF1551.Since two mutations are seen in both tumors, 24 unique candidate cancer genes were identified in oligodendroglioma (Table  [1](#T1){ref-type="table"}). A few interesting candidate genes will be discussed in the next section. ###### Candidate cancer genes identified in oligodendroglioma **Gene symbol** **Gene name** **Entrez gene ID** **Location** **Tumor sample** ----------------- ----------------------------------------------------------------------------------------- -------------------- --------------------------- ------------------ ALMS1 Alstrom syndrome 1 7840 chr2,73612885,73837046 HF1551 APPL1 adaptor protein, phosphotyrosine interaction, PH domain and leucine zipper containing 1 26060 chr3,57261764,57307498 HF1551 ARHGAP10 Rho GTPase activating protein 10 79658 chr4,148653452,148993927 HF1551 CCDC91 coiled-coil domain containing 91 55297 chr12,28410132,28703099 HF1551 CECR2 cat eye syndrome chromosome region, candidate 2 27443 chr22,17956627,18033845 HF1551 DIAPH2 diaphanous homolog 2 (Drosophila) 1730 chrX,95939661,96724837 HF087 EFHC2 EF-hand domain (C-terminal) containing 2 80258 chrX,44007127,44202923 HF1551 EIF1 eukaryotic translation initiation factor 1 10209 chr17,39845126,39847898 HF1551 LASS3 LAG1 homolog, ceramide synthase 3 204219 chr15,100940599,101084925 HF1551 LOC339166 uncharacterized RNA coding gene 339166 chr17,5675553,5834016 HF1551 LRRN2 leucine rich repeat neuronal 2 10446 chr1,204586302,204654481 HF1551 MYOF myoferlin 26509 chr10,95066185,95242074 HF1551 NPAS3 neuronal PAS domain protein 3 64067 chr14,33408458,34273382 HF087, HF1551 OSBPL3 oxysterol binding protein-like 3 26031 chr7,24836163,25019760 HF087, HF1551 PARK2 Parkinson disease (autosomal recessive, juvenile) 2, parkin 5071 chr6,161768589,163148834 HF1551 PAX7 paired box 7 5081 chr1,18957499,19075360 HF1551 PHLDB2 pleckstrin homology-like domain, family B, member 2 90102 chr3,111451326,111695364 HF1551 PLEKHM3 pleckstrin homology domain containing, family M, member 3 389072 chr2,208686011,208890284 HF1551 PRKG1 protein kinase, cGMP-dependent, type I 5592 chr10,52750910,54058110 HF1551 SIPA1L3 signal-induced proliferation-associated 1 like 3 23094 chr19,38397867,38699008 HF087 STMN2 stathmin-like 2 11075 chr8,80523048,80578410 HF087 TACC2 transforming, acidic coiled-coil containing protein 2 10579 chr10,123748688,124014057 HF1551 TCEB3 transcription elongation factor B (SIII), polypeptide 3 6924 chr1,24069855,24088549 HF1551 ZFYVE26 zinc finger, FYVE domain containing 26 23503 chr14,68213236,68283306 HF1551 ### ***Candidates common to both HF087 and HF1551*** Two candidate genes, NPAS3 and OSBPL3, harbored mutations in both tumor samples (Figure  [8](#F8){ref-type="fig"}). NPAS3 (neuronal PAS domain protein 3) shows a complex event accompanied by a \~7 kb gain in the HF087 optical map, and a missing cut in the HF1551 optical map (Figure  [8](#F8){ref-type="fig"}A). This neuronally expressed basic helix-loop-helix transcription factor has been implicated in schizophrenia \[[@B76],[@B77]\]and bipolar disorder \[[@B77]\], and is frequently deleted or inactivated in many cancers. Recently, it has been demonstrated that NPAS3 exhibits features of a tumor suppressor which drives late progression of malignant astrocytomas, and is a negative prognostic marker for survival \[[@B78]\]. {#F8} Both tumor optical maps display cut differences in the OSBPL3 (oxysterol binding protein like-3) gene (Figure  [8](#F8){ref-type="fig"}B). This gene plays a vital role in cell adhesion, cytoskeletal organization and lipid metabolism \[[@B79]-[@B81]\]. It is highly expressed in B-cell associated malignancies \[[@B82],[@B83]\], where it is one of the common sites of retroviral integration \[[@B84]\]. An independent study that used exon sequencing to study oligodendroglioma also found somatic mutations in OSBPL3 \[[@B30]\]. ### ***Candidates observed in either HF087 or HF1551*** In the HF1551 optical map, we observe a point mutation that creates a SwaI restriction site in the PARK2 gene (Figure  [6](#F6){ref-type="fig"}). This gene encodes an E3 ubiquitin ligase, called *Parkin* that catalyzes the ubiquitination of a variety of target proteins for proteasome mediated degradation. Germline mutations in PARK2 have long been known to cause autosomal recessive juvenile Parkinson's disease \[[@B85]-[@B87]\]. More recently, PARK2 has been identified as a tumor suppressor gene in Glioblastoma multiforme, breast, ovary, lung, colorectal and liver cancers \[[@B28],[@B88]-[@B94]\]. It encompasses most of FRA6E, the third most active common fragile site in the human genome \[[@B95]\], and shares the characteristics of other tumor suppressors such as FHIT and WWOX, that also occur in fragile sites. PARK2 is frequently deleted or inactivated in cancer cell lines and primary tumors \[[@B88],[@B92]\], and concomitantly, *Parkin* expression is either significantly diminished or absent \[[@B89],[@B92]\]. Unlike classical tumor suppressors where biallelic inactivation is necessary for oncogenesis, heterozygous mutations in PARK2 are sufficient to confer a growth advantage during tumor development \[[@B88],[@B92]\]. Restoring *Parkin* expression in *Parkin*-deficient cell lines reduces their profileration *in vitro*\[[@B92]\], while injection of *Parkin*-deficient cells into immunocompromised mice generate tumors *in vivo*\[[@B91]\]. Interestingly, PARK2 also mediates chemosensitivity in breast cancer *via* microtubule dependent mechanism \[[@B93],[@B96]-[@B98]\]. STMN2 (stathmin-like 2) is another interesting candidate gene. We observe a point mutation in this gene in tumor HF087 (Figure  [7](#F7){ref-type="fig"}). STMN2 is a neuron specific member of the stathmin family of small regulatory phosphoproteins which control cell profileration and differentiation \[[@B99]\]. It is up-regulated in liver cancer and has been identified as a target of β-catenin/TCF-mediated transcription \[[@B100]\]. STMN2 sequesters soluble tubulin, forming a ternary complex, inhibits microtubule assembly and induces their disassembly \[[@B101]\]. Its highly similar, but more well-studied paralog STMN1, located on chromosome 1p, is known to sensitize cells to anti-microtubule drugs in glioma \[[@B102],[@B103]\], breast \[[@B104],[@B105]\] and prostate cancer \[[@B106]\].In light of recent studies demonstrating the synergistic epistasis between paralogous genes involved in essential cellular functions and its therapeutic implications \[[@B107],[@B108]\], we speculate that STMN1 and STMN2 might be functionally redundant, and inactivation of STMN2 might, in part, explain the treatment sensitivity of oligodendroglioma. In the HF1551 optical map, we see an extra cut in the gene ZFYVE26 (zinc finger, FYVE domain containing 26). Spastizin, the zinc finger protein encoded by ZFYVE26, causes the neurological disorder hereditary spastic paraplegia \[[@B109]\]. This gene binds to the tumor suppressor Beclin-1 and regulates cytokinesis \[[@B110],[@B111]\], and is recurrently mutated in breast cancer \[[@B29]\]. We detected a 485 kb inversion on 7q11.23 in tumor HF1551. Hemizygous deletions spanning a 1.5-1.8 MB region of this locus cause the neurodevelopmental disorder, Williams-Beuren (WB) syndrome \[[@B112]\]. However, inversion of this region is polymorphic, and is present in \~6% of the general population, and in \~25% of transmitting parents in WB families \[[@B113]-[@B115]\]. Given the disparity in size between the aberration detected by OM and reported instances of the WB inversion, it is possible that the event we observe arose de novo and is distinct from the 'canonical' inversion. To test this hypothesis, we ran several targeted assemblies on the WB region. The general strategy for this approach was to modify the reference map in-silico to reflect our hypothesized structure (Figure  [9](#F9){ref-type="fig"}A), and then use the iterative assembly framework described earlier to pull out individual restriction maps and generate an optical consensus map (methods). Since our map assembly pipeline was designed to provide the single most conservative answer, this approach is helpful in detecting large-scale aberrations that are significantly different from the reference sequence. Of the eight modified reference maps we started with, the one that reflected the canonical WB inversion/deletion did not grow, while the one that reflected the 485 kb event successfully generated a consensus map that spanned it and flanked contiguous regions on chromosome 7 (Figure  [9](#F9){ref-type="fig"}B). The optical consensus map also closes the putative sequence gap immediately to the right of the inversion, and in fact, approximately half of the sequence gaps in the reference genome (NCBI, build 35) are spanned by optical consensus maps. The inversion encompasses the genes GTF2IRD2, PMS2P5, WBSCR16, GTF2IRD2B and NCF1, and its breakpoints appear to disrupt the genes GTF2I and STAG3L2. In the absence of matched normal DNA, it is impossible to ascertain if the inversion we detected was inherited through the germline or somatically acquired, however, this is the first report, to the best of our knowledge, of inversions in the WB region in the context of cancer. {#F9} ### ***Candidates on 1p or 19q*** The concerted loss of chromosome arms 1p and 19q is a hallmark of oligodendroglioma. Seen in 50%-70% of tumors, it is believed that these regions harbor one or more tumor suppressor genes that play an important role in the development of this cancer. Hence, somatic mutations on these chromosome arms are particularly interesting. We found putative mutations on 2 genes residing on chromosome 1p (TCEB3, PAX7) and 1 gene on 19q (SIPA1L3). The roles of these genes in normal and disease states, and the structural variants we found in them are discussed briefly in the subsequent section. We observe a 6.3 kb deletion that potentially ablates the first exon of TCEB3 in tumor HF1551. TCEB3 (transcription elongation factor B, polypeptide 3) encodes the transcriptionally active subunit of the mammalian elongin complex \[[@B116],[@B117]\]. This elongation factor stimulates the rate of transcription by suppressing the transient pausing of RNA polymerase II on the DNA template \[[@B118]\]. TCEB3 is part of a multi-protein complex that functions as an elongin-based ubiquitin ligase \[[@B119]\], similar to the Von Hippel-Lindau (VHL) tumor suppressor complex, by mediating DNA damage induced ubiquitination and degradation of polymerase II \[[@B120]\]. Tumor HF1551 also has an insertion in the 1p-encoded gene PAX7 (paired box 7). The PAX genes encode a family of transcription factors that control development within the neural, myogenic and lymphoid lineages \[[@B121]\]. PAX7, in particular, is essential for survival, proliferation and migration of myogenic progenitor cells \[[@B122]\], and cell fate decisions in the developing nervous system \[[@B123]\]. PAX7 is the target of a recurrent gene fusion with the forkhead protein FKHR/FOXO1 that is found in \~15% of patients with alveolar rhabdomyosarcoma \[[@B121],[@B124]\]. The fusion transcript is much more abundant and transcriptionally active than wild type PAX7 \[[@B125]\], suggesting that the deregulation of PAX7 downstream target genes contribute to tumorigenesis. In the HF087 optical map, we observe a missing cut in the gene SIPA1L3 (signal induced proliferation associated 1 like 3) which is located on the long arm of chromosome 19. This gene encodes a Ras specific GTPase activating protein that is found at epithelial junctional complexes. These complexes play a crucial role in mechanical adhesion between epithelial cells to form cellular sheets and in the organization of actin cytoskeleton \[[@B126]\]. Somatic mutations in SIPA1l3 have been discovered in cancers of the brain \[[@B127],[@B128]\], prostate \[[@B129]\], breast \[[@B24]\], ovary \[[@B32]\], pancreas \[[@B130]\], colon \[[@B24]\], skin \[[@B131]\] and hematopoietic system \[[@B34]\], but a cohesive picture of the functional role that this gene plays in these diverse cancer types is yet to emerge. Taken together, the candidate genes discovered by Optical Mapping point to critical roles of transcriptional control and cytoskeletal organisation in the etiology of oligodendroglioma. ### ***Non-genic candidates*** Protein coding sequences comprise less than 2% of the human genome. The vast non-coding portion of the genome, once believed to be 'junk DNA', is rife with functional elements that orchestrate the gene expression program of cells. Recent evidence from the ENCODE (Encyclopedia of DNA Elements) consortium indicates that as much as 80.4% of the human genome encodes a defined product (for instance, a non-coding RNA) or displays a reproducible biochemical signature (for instance, a specific chromatin structure) \[[@B132]\]. Such signatures, either alone or in combinations, mark genomic sequences with important functions, such as promoters, enhancers, insulators and silencers \[[@B133]\]. The ENCODE data sheds some light on possible functional roles of Optical Mapping candidates that are not located within genes. A number of these candidates are actively transcribed, for instance an EC on chromosome 5 of HF1551 overlaps the transcribed pseudogene GUSBP9. Several non-genic variants occur within long intergenic non-coding RNAs (lincRNA) coding regions (Additional file [7](#S7){ref-type="supplementary-material"}). Both these classes of genomic elements provide an additional tier of gene regulation, and contribute significantly to the transcriptional landscape of human cancers \[[@B134],[@B135]\]. Several candidates also show interesting changes in their putative functions in cancer tissues. For example, we observe a MC on chromosome 2 (HF087) in a genomic region bearing a histone modification pattern characteristic of insulators in multiple different normal cell types, but the pattern changes to that of an enhancer in hepatocellular carcinoma. Optical Mapping provides a global view of the cancer genome, free from biases introduced by cloning, amplification or hybridization, and discovers structural variation and mutation on a scale ranging from kilobases to megabases. Moreover, since the platform uses high-molecular weight DNA as analyte, the long-range context and connectivity of each variant is preserved, potentiating meaningful interpretation of candidate genes. However, Optical Mapping does not provide single-base resolution. Point mutations or indels spanning a few base pairs, such as the events frequently observed in CIC and FUBP1 genes in 1p/19q codeleted oligodendrogliomas, are below the lower limit of resolution and would remain undetected (unless they create or destroy a SwaI restriction site). Biological significance of candidates identified by optical mapping ------------------------------------------------------------------- The aim of this study is to generate new hypotheses for oligodendroglioma genetics, and as such, functional studies are beyond the scope of this paper. However, by surveying publicly available data on the candidates discerned by Optical Mapping, we can gain some insight into the roles they might play in malignant transformation. Moving beyond the two tumors HF087 and HF1551, we wanted to take the candidate cancer genes and analyze them in the context of other genome-wide studies. Most somatic mutations in cancer cells arise due to genomic instability and do not contribute to tumorigenesis. However, mutations in genes that promote tumor development, so-called 'driver' mutations tend to be recurrent. To assess the extent of recurrence of our candidates across a large number of samples, we used the Catalog of Somatic Mutations in Cancer (COSMIC). The COSMIC database is a comprehensive archive of somatic mutations in human cancer, combining manually curated data from scientific literature and the output from the Cancer Genome Project \[[@B136],[@B137]\]. 10 genes from our list of candidates had mutation frequencies (number of unique mutated samples divided by the total number of unique samples) greater than 10%, with the top hits being MYOF (19.1%), CECR2 (18.8%) and ZFYVE26 (18.7%). Mutation frequencies for all the candidate genes are listed in Table  [2](#T2){ref-type="table"}. ###### Mutation frequencies of candidate cancer genes (COSMIC database) **Gene symbol** **No. of samples with mutations** **No. of unique samples** **Mutation frequency** ----------------- ----------------------------------- --------------------------- ------------------------ STMN2 3 517 0.580270793 EIF1 1 91 1.098901099 LRRN2 10 522 1.915708812 PAX7 15 748 2.005347594 PRKG1 21 545 3.853211009 ARHGAP10 9 229 3.930131004 PARK2 10 204 4.901960784 OSBPL3 11 206 5.339805825 APPL1 9 161 5.590062112 TCEB3 7 97 7.216494845 CCDC91 8 97 8.24742268 EFHC2 11 131 8.396946565 TACC2 30 312 9.615384615 NPAS3 13 114 11.40350877 CERS3/LASS3 12 104 11.53846154 SIPA1L3 12 102 11.76470588 DIAPH2 19 143 13.28671329 PLEKHM3 4 26 15.38461538 PHLDB2 17 108 15.74074074 ALMS1 28 165 16.96969697 ZFYVE26 35 187 18.71657754 CECR2 16 85 18.82352941 MYOF 22 115 19.13043478 LOC339166 0 0 \- Since cells can employ a number of mechanisms to compensate for loss or mutational inactivation of genes, a more direct way of assessing the functional role of a given candidate gene is to analyze changes in its pattern of expression between normal and disease states. Array based expression profiling of tumors HF087 and HF1551 was performed by Fine et. al. \[[@B138]\], and is publicly available through the NCBI GEO database \[[@B139]\], accession GSE4290. Differential expression analysis carried out using the EBarrays algorithm \[[@B140]\] shows that 5 genes (NPAS3, STMN2, ZFYVE26, PHLDB2 and PLEKHM3) from our list of 24 candidates is significantly up or down-regulated (p-value 1E-03 or less). A complete list of differentially expressed genes can be found in Additional file [8](#S8){ref-type="supplementary-material"}. To assay for functional effects in an even larger population of tumors, we queried for changes in expression of our candidate genes in REMBRANDT, a database of molecular data on brain tumors (National Cancer Institute, 2005, REMBRANDT home page <https://caintegrator.nci.nih.gov/rembrandt/>, accessed 13^th^August 2012). The results are reported for each gene as the number of oligodendroglioma samples in the database that are differential expression by at least two-fold (Additional file [9](#S9){ref-type="supplementary-material"}). All but 3 of our candidate genes were differentially expressed in at least 10 tumor samples. Congruent with the previous analysis, NPAS3, STMN2, ZFYVE26 and PHLDB2 are the most frequently deregulated candidate genes. Finally, we asked what biological processes and pathways are significantly enriched or depleted in our list of candidates. This can identify fundamental cellular mechanisms that contribute to cancer development. As a whole, these candidate genes are enriched for proteins involved in cytoskeletal organization (p-value = 0.00223, after correcting for multiple testing, Ontologizer \[[@B141]\]). Our candidate genes are also significantly enriched for microRNA binding targets (p-values between 0.0144-0.0198 after correcting for multiple testing, WebGestalt \[[@B142]\]). Approximately half of the over-represented sites have been associated with binding of cancer-related microRNAs \[[@B143]\], underscoring the importance of post-transcriptional control of expression in oligodendroglioma. While these results are not a direct indicator of aberrant function, this is a demonstration that Optical Mapping results can be expanded to clinical samples and used to create direct functional hypotheses. Conclusions =========== We have applied Optical Mapping to explore the genomic landscape of solid tumor oligodendroglioma. \~2100 discrete structural variants have been discovered, ranging in size from single base changes to loss of entire chromosomes. The structure of each alteration has been elucidated at sub-genic resolution, while retaining the long-range context of the event. 94 somatic mutations have been identified, 24 of which affect genes. These novel candidate cancer genes provide focused, testable hypotheses for follow-up functional investigation. We believe that Optical Mapping provides a comprehensive, high-resolution description of the complex and disperse genomes of solid tumors. Methods ======= Selection of tumors ------------------- The tumors used in this study originated from the tissue bank at the Hermelin Brain Tumor Center/Department of Neurosurgery, Henry Ford Hospital (provided by Dr. Oliver Bogler). Freshly resected tumors were snap frozen in liquid nitrogen in the operating room. Samples were sectioned in a guillotine in frozen condition, and adjacent pieces prepared for Optical Mapping and for re-review by a neuropathologist. The tumor samples selected for Optical Mapping had to meet two criteria. First, they needed to conform to the 1p/19q paradigm of treatment sensitivity. LOH status was assessed by quantitative PCR of microsatellite markers along chromosomes 1p and 19q (data not shown). Second, they needed to have a high proportion of cancer cells as opposed to normal cells. The percentage of tumor cells present in each biopsy was estimated by MIB-1 antibody staining of an adjacent section (data not shown). The MIB-1 antibody recognizes the Ki-67 antigen, which is a cell proliferation marker. For the most part, mitotic activity is absent in the adult brain, so the measurement of the Ki-67 cell proliferation marker can be used to judge tumor aggression and composition \[[@B144],[@B145]\]. The two samples chosen for this project, HF087 and HF1551, satisfied both these criteria. Table  [3](#T3){ref-type="table"} provides relevant clinical information for each tumor. ###### Clinical information on the tumors analyzed by Optical Mapping **Tumor sample** **Histology** **Histology code** **LOH status** **MIB Index (%)** **Patient age** **Patient sex** ------------------ ---------------------------- -------------------- ---------------- ------------------- ----------------- ----------------- HF087 Oligodendroglioma O II LOH 5-7% 65 Female HF1551 Atypical Oligodendroglioma O II LOH 4-13% 30 Female *O II:* Grade II; *LOH:* loss of heterozygosity. Extraction of high molecular weight DNA from solid tumor biopsies ----------------------------------------------------------------- The tumor was sectioned into 1--2 mm slices under sterile conditions in a cell culture hood. Each slice was treated with 0.8% type IV collagenase (Sigma-Aldrich, St. Louis, MO) in PBS (Phosphate buffered saline, Life Technologies, Carlsbad, CA) for 15 minutes at 37°C. The tumor tissue was mechanically disaggregated into a homogeneous suspension by repeated pipetting. The cells were pelleted by centrifugation at 1,000 RPM with a Beckman GS-6R centrifuge (Beckman Instruments, Fullerton, CA), and then resuspended in 1X HBS (Hanks Balanced Salts, Life Technologies, Carlsbad, CA) in order to lyse red blood cells. Cell debris and HBS were removed by centrifugation at 1,000 RPM. Finally, the pellet was rinsed three times with 35 mL of PBS, and resuspended in 0.5 mL of PBS. A three layer Percoll gradient was employed to enrich for cancer cells, and minimize stromal contamination \[[@B146]\]. First, a 100% solution was made by using 9 parts Percoll (Sigma-Aldrich, St. Louis, MO) and 1 part 10X HBS, which was subsequently diluted with PBS to prepare 10%, 30%, and 50% solutions. The gradient was prepared by layering 2 mL of 50% Percoll, 2 mL of 30% Percoll, and 1 mL of 10% Percoll in a 15 mL Falcon tube. The single cell suspension was then carefully layered on top, and the gradient was spun at 1,000 RPM for 10 minutes. Studies have shown that cellular debris and non-viable cells are unable to penetrate the 30% layer, while lymphocytes pelleted at the bottom of the tube. The 30% layer, containing viable cells, was carefully removed, rinsed three times with 10 mL of PBS and then resuspended in PBS at a final concentration of 1X10^7^ cells/mL. Next, this cell suspension was mixed 1:1 (v/v) with 1.6% low gelling temperature agarose, poured into a mold and cooled to 4°C so that the agarose solidified to for gel inserts (each \~100 μL in volume). The inserts were treated with 0.5 mg/mL proteinase K (Bioline USA, Taunton, MA), 100 mM EDTA pH 8.0 (Sigma-Aldrich, St. Louis, MO), 0.5% N-lauroylsarcosine (Sigma-Aldrich, St. Louis, MO) and incubated at 55°C overnight to lyse the tumor cells and degrade cellular proteins \[[@B45],[@B147]-[@B150]\]. Embedding cells in agarose inserts eliminates shear induced breakage of genomic DNA molecules upon lysis \[[@B151]\]. Prior to use, the gel inserts were rinsed in TE twice for 1 hour and then a third time overnight to remove the detergent and excess EDTA. DNA was electrophoretically extracted by applying a cycle of 100 V for 30 seconds and -100 V for 6 seconds. Generation of single molecule optical maps ------------------------------------------ Optical Mapping surfaces were prepared as described earlier \[[@B152]\]. Briefly, acid-cleaned glass coverslips (22 × 22 mm, Fisher\'s Finest, Fisher Scientific) were treated with a mixture of N-trimethoxylsilylpropyl-N,N,N-trimethylammonium chloride and vinyltrimethoxysilane (Gelest, Morrisville, PA) rendering a positive charge to the surface. Genomic DNA, mixed with a sizing standard, was elongated *via c*apillary flow in a microfluidic device \[[@B66]\], and immobilized by electrostatic interactions with the positively charged surface, creating arrays of stretched, biochemically accessible substrates. The surface was then washed with TE (10 mM Tris--HCl, 1 mM EDTA, pH8.0) twice, equilibrated with digestion buffer (NEB buffer 3), then incubated with the restriction endonuclease SwaI (New England Biolabs, Beverly, MA), which cleaves the genomic DNA at its cognate site. Since the elongated DNA molecule is under slight tension, upon cleavage its ends relax, creating a 1--2 micron gap, readily detected by microscopy. The resulting restriction fragments remain adsorbed to the surface, aided by a polyacrylamide overlay, and hence retain their order creating, in essence, a barcode from each genomic DNA molecule. Restriction fragments were then stained with the DNA intercalating dye YOYO-1 (0.2 μM in β-mercaptoethanol/TE, Life Technologies, Carlsbad, CA) and imaged by automated fluorescence microscopy. The images were collected on an Optical Mapping workstation, which consists of Zeiss 135M inverted microscope (Carl Zeiss, Thornwood, NY), illuminated by 488 nm argon ion laser (Spectra Physics, Santa Clara, CA) equipped with 63X oil immersion objective. Fully automated image acquisition software, referred to as Channel Collect \[[@B66]\], takes multiple overlapping images to span the entire length of each microchannel. The images were analyzed by custom machine vision software, called Pathfinder, which identifies DNA molecules on the surface and calculates the size of each restriction fragment based on integrated fluorescence intensity measurements relative to a sizing standard. Previous studies have shown that integrated fluorescence intensity scales with fragment mass, and is independent of stretch of the DNA molecule. The end result of these operations is the high throughput, massively parallel generation of single molecule ordered restriction maps, or optical maps, containing information about both the size and order of its restriction fragments. Pipeline for optical Map assembly and identification of structural variants --------------------------------------------------------------------------- The analytical framework for assembly of optical maps is analogous to sequence assembly. First, our pipeline automatically aligned optical maps against a SwaI restriction map created *in silico* from the human genome reference sequence (NCBI build 35) *via* SOMA (Software for Optical Map Alignment) using gapped global pair wise alignment \[[@B56],[@B67]\]. SOMA uses a scoring function that assigns penalties for differences in the optical map and the reference map, including missing or extra restriction sites, or differences in the size of the fragments that could represent insertions or deletions. The parameters of SOMA were set so that we are accurately aligning the molecule to the correct location, but loose enough for allow for a small number of differences that result from the mutations or polymorphisms present in the genome. The aligned maps were then partitioned into smaller bins (1 Mb windows spanning across each chromosome, with 500 kb overlap between adjacent windows) based on their location. The optical maps in each bin were assembled into optical consensus maps by a map assembler program, using a Bayesian inference algorithm \[[@B153]\]. Because some structural polymorphisms and mutations represent large-scale changes from the reference map, an iterative assembly process was used for the analysis of human data sets. The consensus map constructed in the previous step was used in place of the reference for seven more iterations of alignment and assembly, after which it was aligned to the reference sequence using SOMA. Using this strategy, Rmaps harboring major alterations that preclude alignment to the reference were gradually incorporated into the consensus map, extending it into regions that contain more complex rearrangements. Lastly, the pipeline automatically performed analysis that tabulated structural variants using the final consensus map to the reference (derived from NCBI build 35 of the human genome \[[@B56]\]) and identified five classes of differences: missing cuts, extra cuts, insertions, deletions, and 'other' (multiple cut and/or size differences) across each cancer genome. Each of these differences, which are largely structural variants, has to satisfy certain statistical and empirical criteria. These parameters have been detailed in Teague et al. \[[@B56]\]. The only difference being the indel calling threshold, which was increased to a 13% change relative to the reference, with a 4.5 kb minimum. Additionally, each structural variant was manually curated to ensure that the most conservative decision has been made at every locus. The genomic locations of the variants were converted to NCBI build 37 co-ordinates using the Batch Coordinate Conversion (liftover) tool from the University of California Santa Cruz Genome Browser \[[@B154]\] ( <http://genome.ucsc.edu>). Optical map coverage analysis ----------------------------- Variations in depth of coverage of optical maps aligned by SOMA across the genome can be used to detect copy number alterations. Intuitively, if a region of the tumor sample has increased (or decreased) copy number relative to the 'normal' reference genome, more (or less) maps will originate from it on an average. This is formalized as described. Pair-wise alignments of optical maps to an *in silico* reference were summarized by a single number (midpoint) representing location. These locations were modeled as realizations of a non-homogeneous Poisson process. The non-homogeneity arises from the fact that the likelihood of a map aligning to a genomic region depends on the density of restriction sites, and was accounted for using alignment data from a normal genome, which are used to define random intervals with counts that follow a negative binomial distribution. These counts were then modeled by a Hidden Markov Model, incorporating spatial dependence in the data and allowing more natural estimation of certain parameters \[[@B68],[@B69]\]. Affymetrix genome wide human SNP array 6.0 ------------------------------------------ DNA was prepared for hybridization using the Blood and Cell Culture Kit (Qiagen, Valencia, CA), starting from frozen cells (HF087), or tumor tissue (HF1551), disaggregated into single cells as described previously. The HF087 cells were derived from the same slice used for Optical Mapping. However, since the same was not available for HF1551, a slice adjacent to the one used for mapping was used. The DNA was digested with NspI and StyI restriction enzymes and ligated to adaptors that recognize the 4 bp overhangs. A generic primer that anneals to the adaptor sequence was then used to amplify adaptor-ligated DNA fragments, under PCR conditions optimized to preferentially amplify fragments in the 200 to 1,100 bp size range. The amplified DNA was then fragmented, labelled, and hybridized to a Genome-Wide Human SNP 6.0 Array (experiments were performed by the DNA Facility at the Carver College of Medicine, University of Iowa). Data analysis was performed using Genotyping Console 2.0 (Affymetrix, Santa Clara, CA). CNVs were called using either the Affymetrix algorithm (with default parameters) or five different algorithms (GLAD, Circular Binary Segmentation, Fused Lasso, Gaussian Model with Adaptive Penalty, Forward-Backward Fragment Annealing Segmentation) from CGHweb ( <http://compbio.med.harvard.edu/CGHweb/>) \[[@B155]\]. Only CNV calls made by two or more algorithms were considered for comparison. Parameters for comparing oligodendroglioma structural variants -------------------------------------------------------------- ### ***To other optical mapping datasets*** Only variants of the same type were compared to each other, e.g.: MCs from HF087 were compared to MCs from lymphoblast cell line GM15510. Intersection 'windows' were set based on the type of OSA (100 bp for MCs, 4200 bp for ECs and 0 bp for INS, DEL and OTHER) and are reflective of the error processes inherent to each type of event. ### ***To published SNPs and structural variants*** Published SNPs were compared against Optical Mapping cut differences using 100 bp or 3000 bp windows for MCs and ECs, respectively. Structural variants from the latest (November 2010) release of the Database of Genomic Variants \[[@B156]\] were divided into two categories on the basis of their size. Events smaller than 3 kb were compared to ECs and MCs, since \~1/3^rd^ of indels that are below the lower limit of detection for Optical Mapping manifest themselves as cut differences \[[@B56]\]. Events larger than 3 kb were compared to INS, DEL and OTHER variants using a 0 bp intersection window. PCR validation -------------- Template for PCR was prepared by whole genome amplification of tumor DNA using the REPLI-g Mini kit (Qiagen Inc., Valencia, CA) as per the protocol provided by the manufacturer. Pooled normal DNA from 6 individuals (Promega Corporation, Madison, WI) was used for control reactions. Primers were designed using freely available software Primer 3 Plus \[[@B157]\]. PCR reactions were performed using reagents from the Expand Long Template PCR System (Roche Applied Science, Indianapolis, IN) following the protocol supplied by the manufacturer. PCR reactions were digested with appropriate restriction enzymes to establish that the correct region had been amplified. The amplicon was then cloned in *E. coli* using the TOPO TA Cloning Kit (Invitrogen, Carlsbad, CA), plasmid DNA was purified using the Qiagen Plasmid Mini Kit (Qiagen Inc., Valencia, CA), and sequenced using Sanger biochemistry. Targeted assemblies on Williams-Beuren chromosomal region --------------------------------------------------------- The SwaI *in-silico* restriction map from the Williams-Beuren region on chromosome 14 was modified to reflect one of eight alterations: 4 possible inversions, each with unique start/end locations and spans (including the 'canonical' inversion), and 4 possible deletions, each with unique start/end locations and sizes (including the 'canonical' deletion). These modified *in-silico* maps were subjected to 8 rounds of iterative assembly, using the collection of HF1551 Rmaps, with the same parameters as the genome-wide assembly. The results were manually curated to rule out assembly errors. Ethics statement ---------------- This study was approved by the Institutional Review Board of the University of Wisconsin-Madison. Availability of supporting data ------------------------------- All structural variation calls and analysis are contained within the additional files. Abbreviations ============= OM: Optical mapping; Kb: kilobase pairs; EC: Extra cut; MC: Missing cut; INS: Insertion; DEL: Deletion; SNP: Single nucleotide polymorphism; CNV: Copy number variant; HMM: Hidden Markov Model; LOH: Loss of heterogeneity; CGH: Comparative genome hybridization; MHC: Major histocompatibility complex; DGV: Database of genomic variants; COSMIC: Catalog of somatic mutations in cancer; ENCODE: Encyclopedia of DNA elements. Competing interests =================== The authors declare that they have no competing interests. Authors' contributions ====================== MR carried out the experimental studies, analysis and manuscript writing. SG developed and applied new bioinformatic tools used for the study. SZ contributed to the data analysis. KP helped with the optical mapping and data interpretation. DS and MAN contributed new statistical tools. EE and CK assisted in the statistical analysis of structural variants. OB contributed samples and overall project guidance. DCS conceived the study, its design, manuscript writing and overall management of the project. All authors have read and approved the final manuscript. Supplementary Material ====================== ###### Additional file 1 **Basic attributes of oligodendroglioma datasets.** This spreadsheet describes the two optical maps in detail. Column A lists specific characteristics associated with each step of constructing the map. Columns B and C lists its values for tumor HF087 (column B) and HF1551 (column C). ###### Click here for file ###### Additional file 2 **Overlap (counts) between oligodendroglioma structural variants and genes, segmental duplications, SNPs, and variants reported by other investigators.** This table lists counts of structural variants from oligodendroglioma that intersect with genes, segmental duplications, SNPs, and events from the Database of Genomic Variants (DGV). Column A specifies the genomic element; columns B and H indicate total counts for HF087 and HF1551 respectively; Columns C-G and I-M shows overlap counts by variant class. Variants from the DGV are divided based on size into those over 3 kb and those under then 3 kb, then further by study. Variants less than 3 kb are compared to optical map EC and MCs, while those over 3 kb are compared to INS, DEL and OTHER. ###### Click here for file ###### Additional file 3 **Detailed description of intersections between oligodendroglioma structural variants and genes, SNPs, variants from the Database of Genomic Variants and other normal human optical maps.** This spreadsheet provides a detailed breakdown of the overlap between Each row in the spreadsheet shows an optical map difference (column A), it's location (columns B-D), and genes (column E), variants from the Database of Genomic Variants (column F, G), snip-SNPs (column H), and structural variants from other Optical Mapping datasets (columns I-N) that overlap with it. The number is parenthesis after each column header indicates the intersection window. ###### Click here for file ###### Additional file 4 **Experimental validation of oligodendroglioma ECs and MCs by SNP array.** This table lists cut differences found in HF087 and HF1551 (column A), their location (columns B-D), type (column E), the SNP genotype corresponding to it (column F) and whether it agrees with the optical map (column G). ###### Click here for file ###### Additional file 5 **Experimental validation of oligodendroglioma indels.** This table lists indels found in HF087 and HF1551 (column A), their location (columns B-D) and whether it is validated by a given copy number algorithm (columns E-J). ###### Click here for file ###### Additional file 6 **Oligodendroglioma structural variants and their intersection (counts) with variants detected in six other normal human optical maps.** This table lists counts of structural variants from oligodendroglioma that intersect with variants found in other normal human genomes that have been analyzed by Optical Mapping. Column A specifies the genomic element; columns B and H indicate total counts for HF087 and HF1551 respectively; Columns C-G and I-M shows overlap counts by variant class. Only variants of the same category are included in the comparison. ###### Click here for file ###### Additional file 7 **Non-genic candidates found in oligodendroglioma and functional elements from ENCODE that intersect them.** Column A lists identifiers for candidate loci that do not occur within a gene, their locations (columns B-D), overlapping transcripts found by GENCODE (column E), and predicted genomic state in different cell types (columns F-N). The cell types in red font are cancer cell lines. Different genomic states are color-coded as per ENCODE website, and is detailed in the key. ###### Click here for file ###### Additional file 8 **Differentially expressed genes in GEO dataset GSE4290, analyzed by Ebarrays, p = 1E-03.** Entrez gene identifer (column A), gene symbol (column B), gene name (column C), cellular location (column D) and molecular function (column E) of all differentially expressed genes in GEO dataset GSE4290. ###### Click here for file ###### Additional file 9 **Number of oligodendrogioma samples in REMBRANDT database where a given candidate gene is up or down regulated by at least 2 fold.** This table lists candidate genes identified through Optical Mapping (column A), and the number of oligodendroglioma samples in the REMBRANDT database where that gene is up or down regulated by at least two fold (column B). ###### Click here for file Acknowledgements ================ We thank Nick Shera for his early efforts during the data acquisition phase of this project and Ezra Lyon for his bioinformatic support. We also thank NHGRI and NCI for support (D.C.S.; R01- HG000225; R33CA111933); M.R would also like to acknowledge the Morgridge Biotechnology Fellowship for funding.

Introduction {#Sec1} ============ Breast cancer is the most common cancer diagnosed in women, with an estimated 1.7 million new cases per year worldwide^[@CR1]^. Molecular genetics studies have classified breast cancer into four major subtypes based on gene expression profiles: luminal A, luminal B, HER2-positive, and basal-like breast cancers^[@CR2]^. Similar breast cancer subtypes were confirmed by The Cancer Genome Atlas studies^[@CR3]^. Additionally, patient survival and response to therapies varies by breast cancer subtype^[@CR4]^. The majority of basal-like breast cancer tissue is characterized by the lack of hormone receptors and the absence of HER2 amplification; therefore, it is referred to as triple-negative breast cancer (TNBC)^[@CR5]^. TNBC is an aggressive breast cancer accounting for 15--20% of all breast cancer cases^[@CR6]^. TNBC is associated with a high mortality rate due in part to the highly metastatic nature of TNBC^[@CR7]^, as well as the lack of effective therapies^[@CR8]^. Therefore, understanding the mechanisms that drive TNBC progression and metastasis remains an important goal for effective management and treatment of this disease. The LIM domain protein family consists of 65 human proteins, characterized by the presence of the LIM domain^[@CR9]^. The LIM domain, a highly conserved cysteine-rich domain, participates in protein--protein interactions^[@CR10]^. A smaller subfamily in this category includes the LIM domain kinase 2 (LIMK2), which is an important regulator of growth and invasion of several cancers, including pancreatic cancer, bladder cancer, osteosarcoma, and glioblastoma^[@CR11]--[@CR15]^. Additionally, some studies have shown that LIMK2 expression and activity increases after treatment with anticancer drugs, and these studies have implicated LIMK2 in drug resistance^[@CR16]--[@CR18]^. Although LIMK2 has been implicated in several different cancer types, the role of LIMK2 in breast cancer is not fully understood. Furthermore, very few targets of LIMK2 have been identified. Of these, only cofilin 1 has been widely studied^[@CR19],[@CR20]^. Here, we show that LIMK2 is overexpressed in TNBC and is necessary for the metastatic progression of TNBC. Using an unbiased global phosphoproteomics approach of stable isotope labeling with amino acids in cell culture (SILAC), we identified 258 proteins whose phosphorylation was significantly reduced due to LIMK2 inhibition, including SRSF protein kinase 1 (SPRK1). We found that LIMK2 inhibition blocks SRPK1 phosphorylation and activity; therefore, LIMK2 imparts a distal metastasis promoting effect, in part via regulating SRPK1 function. Results {#Sec2} ======= LIMK2 is overexpressed in TNBC and is necessary for facilitating TNBC metastatic attributes {#Sec3} ------------------------------------------------------------------------------------------- To understand the role of LIMK2 in breast cancer, we first asked whether LIMK2 is overexpressed in breast cancer. To this end, we analyzed a breast cancer tissue microarray (\#BC081120e, US Biomax Inc.). This tissue array included 100 cases of invasive breast cancer and 10 adjacent normal breast tissues with information on ER, PR, and HER2 status (Table [S1](#MOESM2){ref-type="media"}). In immunohistochemical analysis, we found that LIMK2 expression was significantly higher in breast cancer samples, including those of TNBC, than in normal tissue (Fig. [S1a, b](#MOESM1){ref-type="media"}, Table [S1](#MOESM2){ref-type="media"}, Figs. [1a, b](#Fig1){ref-type="fig"}, and [S1c](#MOESM1){ref-type="media"}). Because TNBC is an aggressive breast cancer subtype that currently lacks effective therapeutic approaches, we focused our studies with LIMK2 on TNBC. To further establish the role of LIMK2 in TNBCs, we expanded our LIMK2 protein expression analyses. To this end, we analyzed three additional TNBC tissue microarrays for LIMK2 protein expression using immunohistochemistry (IHC): YTMA311 (*n* = 92), YTMA341 (*n* = 53), and YTMA347 (*n* = 54; Table [S2](#MOESM3){ref-type="media"}). Similar to the results of our initial analysis, we found that a subset of TNBC patient samples also expressed higher levels of LIMK2 (Fig. [1c, d](#Fig1){ref-type="fig"} and Table [S2](#MOESM3){ref-type="media"}). Consistent with our results, analysis of multiple publicly available breast cancer mRNA expression datasets also showed significantly higher LIMK2 mRNA expression in TNBC (ERBB2/ER/PR negative) compared to breast cancer with other biomarker status (Fig. [S1d--i](#MOESM1){ref-type="media"})^[@CR21]--[@CR26]^. Furthermore, by analyzing several publicly available breast cancer mRNA expression datasets, we found that increased LIMK2 expression was also associated with increased incidence of metastasis, recurrence, and death in breast cancer patients (Fig. [S2](#MOESM1){ref-type="media"}).Fig. 1LIMK2 is overexpressed in triple-negative breast tumors.**a** Representative images for LIMK2 expression analysis in normal breast tissue or triple-negative breast cancer (TNBC) samples in a tissue microarray (TMA) from US Biomax at ×20 magnification. Scale bar, 50 μm. **b** (Left) Relative fraction of TNBC samples from US Biomax TMA scored by LIMK2 staining intensity: 0, no staining (not shown); +1, weak; +2, moderate; or +3, high. (Right) Percentage of LIMK2-positive cells in TNBC samples from US Biomax TMA: 0--25%, 25--50%, 51--75%, or 76--100%. Unpaired *t*-test with Welch's correction was used to compare the LIMK2 expression between normal adjacent breast tissues and breast carcinoma. **c** Representative images for LIMK2 expression analysis in normal breast tissue or TNBC samples in three independent TNBC TMAs from Yale Tissue Microarray Facility (YTMA311, YTMA341, and YTMA347) at ×20 and ×40 magnification. Scale bar, 50 μm for ×20 and 25 μm for ×40. **d** (Left) Relative fraction of TNBC samples from three independent TNBC TMAs (YTMA311, YTMA341, and YTMA347) from Yale Tissue Microarray Facility scored by LIMK2 staining intensity: 0, no staining (not shown); +1, weak; +2, moderate; or +3, high. (Right) Percentage of LIMK2-positive cells in TNBC samples from three independent TNBC TMAs (YTMA311, YTMA341, and YTMA347) from Yale Tissue Microarray Facility: 0--25%, 25--50%, 51--75%, or 76--100%). \*\**P* \< 0.01. Next, based on the results of our IHC and guided by the previous work on the role of LIMK2 in metastasis^[@CR12],[@CR18],[@CR27]--[@CR29]^, we asked whether inhibition of LIMK2 attenuates the metastatic characteristics of TNBC cells. To test this, we used a small-molecule LIMK2 inhibitor, LX7101 (Lexicon Pharmaceutical), to determine whether pharmacological inhibition of LIMK2 altered the metastatic properties of breast cancer cells. LX7101 has also shown in vivo efficacy for treating glaucoma and is currently being tested in phase 1 clinical trials^[@CR30]^. To measure the effect of LX7101 on the metastatic attributes of TNBC cells, we treated TNBC cell lines MDA-MB-231 and BT-549 with LX7101 and assessed migration, invasiveness, actomyosin contractility, and extracellular matrix (ECM) degradation using three independent assays. As expected, LX7101 treatment resulted in reduced phosphorylation of the LIMK2 substrate cofilin (pSer3; Fig. [S3a](#MOESM1){ref-type="media"}), reduced migration (Fig. [2a, b](#Fig2){ref-type="fig"}), reduced invasion (Fig. [2c, d](#Fig2){ref-type="fig"}), increased disassembly of focal adhesions and associated stress fibers (Fig. [2e](#Fig2){ref-type="fig"}), and reduced gelatin degradation by TNBC cells compared with dimethyl sulfoxide (DMSO)-treated TNBC cells (Fig. [2f](#Fig2){ref-type="fig"}).Fig. 2LIMK2 facilitates metastatic attributes of triple-negative breast cancer (TNBC) cells.**a** MDA-MB-231 and BT-549 cells treated with the vehicle or LX7101 (5 μM) were analyzed using a wound healing assay. Representative images at the indicated times are shown. Scale bar, 250 µm. **b** Relative migration (%) calculated from the data presented in **a**. **c** MDA-MB-231 and BT-549 cells treated with the vehicle or LX7101 (5 μM) were analyzed using a 3D spheroid invasion assay. Representative images at the indicated times are shown. Scale bar, 250 µm. **d** Relative invasion (%) calculated from the data presented in **c**. **e** MDA-MB-231 and BT-549 cells were stained with phalloidin (red), vinculin (green), and DAPI (blue) after treatment with the vehicle or LX7101. Scale bar, 5 µm. **f** Extracellular matrix degradation capacity of MDA-MB-231 and BT-549 cells treated with the vehicle or LX7101 (5 μM) was analyzed using the gelatin degradation assay. Cells were stained with phalloidin (red) and DAPI (blue). Gelatin (green) degradation appears as black areas. Representative images are shown. Scale bar, 500 µm. **g** 3D spheroid invasion assay in MDA-MB-231 and BT-549 cells expressing *LIMK2* shRNA. Representative images at the indicated times are shown. Scale bar, 250 µm. **h**. Relative invasion (%) calculated from the data presented in **g**. Data are represented as the means ± SD. \*\**P* \< 0.01; \*\*\**P* \< 0.001. To further strengthen the results obtained using LX7101, we used another LIMK2 small-molecule inhibitor, TH-257, a type III allosteric LIMK2 inhibitor that targets LIMK2 DGF-out conformation in a non-ATP-competitive manner and that does not exhibit significant off-target activity by KINOMEscan assay at 1 μM^[@CR31]^. TH-257 also significantly inhibited the metastatic characteristics of TNBC cells in various cell-based assays (Fig. [S3b--g](#MOESM1){ref-type="media"}). Small-molecule inhibitors can inhibit more than one target, albeit at different specificities; therefore, to further confirm the results obtained using small-molecule inhibitors of LIMK2 (LX7101 and TH-257), we knocked down the expression of *LIMK2* in TNBC cells (MDA-MB-231 and BT-549) using short-hairpin RNAs and analyzed the effect of the *LIMK2* genetic knockdown on their ability to invade in a 3D invasion assay. In agreement with our results with LIMK2 inhibitors (LX7101 and TH-257), the genetic knockdown of *LIMK2* using shRNAs also resulted in significantly reduced invasion of TNBC cells (Figs. [2g, h](#Fig2){ref-type="fig"} and [S3h, i](#MOESM1){ref-type="media"}). Collectively, these results demonstrate that LIMK2 is overexpressed in TNBC cells and is necessary for maintaining metastatic attributes in TNBC cells. SILAC identifies proteins with reduced phosphorylation following LIMK2 inhibition {#Sec4} --------------------------------------------------------------------------------- Because LIMK2 is a serine/threonine kinase, we hypothesized that LIMK2 exerts its metastasis-promoting activities by regulating the phosphorylation of its downstream targets. However, downstream mediators of LIMK2 are largely unknown, and few downstream substrates have been identified^[@CR19],[@CR20],[@CR32]^. Therefore, to comprehensively identify proteins whose phosphorylation is reduced by LIMK2 inhibition, we took the unbiased phosphoproteomics approach of SILAC^[@CR33]^. For SILAC analysis, MDA-MB-231 cells were cultured in either light culture medium containing lysine and arginine labeled with light carbon (^12^C) and light nitrogen (^14^N), or cultured in heavy medium containing lysine and arginine labeled with heavy carbon (^13^C) and heavy nitrogen (^15^N). After five cell doublings, MDA-MB-231 cells in light medium were treated with DMSO, and cells in heavy medium were treated with LX7101 for 6 h (Fig. [3a](#Fig3){ref-type="fig"}). SILAC identified 258 proteins whose phosphorylation was significantly reduced due to LIMK2 inhibition (Table [S3](#MOESM4){ref-type="media"}). These proteins belong to several different biological pathways (Fig. [3b](#Fig3){ref-type="fig"}).Fig. 3SILAC identifies proteins whose phosphorylation is reduced due to LIMK2 inhibition.**a** Schematic representation of the major steps of SILAC analysis to identify phosphopeptides that are altered after treatment with LX7101 in MDA-MB-231 cells. **b** Functional classification, localization, and phosphorylation sites for a subset of proteins that show reduced phosphorylation after LX7101 treatment. We next analyzed the SILAC data to predict the preferred amino acid motif for LIMK2-induced phosphorylation using a newly developed method (R/Bioconductor package dagLogo). Using this approach, we identified serine or threonine followed by 2 prolines or serine as the LIMK2 preferred phosphorylation motif in our SILAC studies (Fig. [4a](#Fig4){ref-type="fig"}). We also analyzed the SILAC data using ingenuity pathway analysis (IPA) to functionally classify the proteins with reduced phosphorylation after LIMK2 inhibition. IPA analysis revealed proteins in several pathways including ones that regulate integrin signaling (Fig. [4b--d](#Fig4){ref-type="fig"} and Tables [S4](#MOESM5){ref-type="media"} and [S5](#MOESM6){ref-type="media"}), a result that is consistent with the important role of LIMK2 for stimulating metastatic attributes. In particular, we observed reduced phosphorylation of SRPK1 (Fig. [4d](#Fig4){ref-type="fig"}). SRPK1 promotes metastatic attributes in several cancer types, including breast cancer^[@CR32],[@CR34]--[@CR37]^. Thus, we focused our studies on investigating the role of SRPK1 in regulating the ability of LIMK2 to facilitate the metastatic attributes of TNBC cells.Fig. 4Ingenuity pathway analysis reveals key pathways altered due to LIMK2 inhibition.**a** LIMK2 phosphorylation site consensus sequence identified by dagLogo package. **b** Top 10 signaling pathways identified using IPA analysis on SILAC data. **c** Overlap of signaling network regulated by LIMK2 identified by IPA software. **d** Schematic diagram illustrating the SRPK1-regulated signaling network. SRPK1 is overexpressed in TNBC and its inhibition impairs metastatic attributes of TNBC cells {#Sec5} --------------------------------------------------------------------------------------------- SRPK1 was one of the proteins identified from SILAC analysis displaying reduced phosphorylation following LX7101 treatment. SRPK1 has been previously implicated in the regulation of breast cancer metastasis^[@CR34]^. To define the role of SRPK1 in breast cancer, we first asked whether SRPK1 is overexpressed in breast cancer. To this end, we assessed SRPK1 expression using the US BioMax tissue microarray (\#BC081120e) and found that SRPK1 expression was significantly higher in breast cancer samples, including TNBC samples, compared to normal adjacent tissue (Table [S6](#MOESM7){ref-type="media"} and Fig. [5a, b](#Fig5){ref-type="fig"}). To further establish the role of SRPK1 in TNBCs, we analyzed three additional TNBC tissue microarrays for SRPK1 protein expression using IHC: YTMA311 (*n* = 92), YTMA341 (*n* = 53), and YTMA347 (*n* = 54; Table [S7](#MOESM8){ref-type="media"}). In support of our initial SRPK1 expression analysis, we found that TNBC patient samples expressed higher levels of SRPK1 protein than did normal breast tissue (Fig. [5c, d](#Fig5){ref-type="fig"} and Table [S7](#MOESM8){ref-type="media"}). Consistent with our results, analysis of multiple, publicly available breast cancer mRNA expression datasets also showed significantly higher SRPK1 mRNA expression in TNBC (ERBB2/ER/PR negative) samples compared to non-TNBC breast cancers (Fig. [S4](#MOESM1){ref-type="media"}). Furthermore, by analyzing several publicly available datasets of breast cancer mRNA expression datasets, we found that increased SRPK1 expression correlated with an increased incidence of metastasis, recurrence, and death in breast cancer patients (Fig. [S5](#MOESM1){ref-type="media"}). Based on these collective findings, we decided to focus on SRPK1 for further studies.Fig. 5SRPK1 is overexpressed in triple-negative breast cancer tumors.**a** Representative images for SRPK1 expression analysis in normal breast tissue and triple-negative breast cancer (TNBC) samples on a tissue microarray (TMA) from US Biomax at 20× and 40× magnification. Scale bar, 50 μm for ×20 and 25 μm for ×40. **b** (Left) Relative fraction of TNBC samples from US Biomax TMA scored based on SRPK1 staining intensity: 0, no staining (not shown); +1, weak; +2, moderate; or +3, high. (Right) percentage of SRPK1-positive cells in TNBC samples from US Biomax TMA: 0--25%, 25--50%, 51--75%, or 76--100%. Unpaired *t*-test with Welch's correction was used to compare the SRPK1 expression between normal adjacent breast tissues and breast carcinoma. **c** Representative images for SRPK1 expression analysis in normal breast tissue or TNBC samples from three independent TNBC TMAs from the Yale Tissue Microarray Facility (YTMA311, YTMA341, and YTMA347) at ×20 magnification. Scale bar, 50 μm. **d** (Left) Relative fraction of TNBC samples from three independent TNBC TMAs (YTMA311, YTMA341, and YTMA347) from the Yale Tissue Microarray Facility scored by SRPK1 staining intensity: 0, no staining (not shown); +1, weak; +2, moderate; or +3, high. (Right) percentage of SRPK1-positive cells in TNBC samples from three independent TNBC TMAs (YTMA311, YTMA341, and YTMA347) from the Yale Tissue Microarray Facility: 0--25%, 25--50%, 51--75%, or 76--100%. \*\*\*\**P* \< 0.0001. SILAC analysis revealed five phosphorylation sites on SRPK1 (Ser7, Ser9, Ser51, Ser309, and Ser311) that showed reduced phosphorylation after LX7101 treatment (Fig. [6a](#Fig6){ref-type="fig"}). To confirm that LIMK2 inhibition also reduced SRPK1 phosphorylation in TNBC cells, we used immunoprecipitation experiments. MDA-MB-231 cells were treated with LX7101 and the proteins were immunoprecipitated with anti-SRPK1 antibodies. The phosphorylation status of immunoprecipitated SRPK1 was assessed by immunoblotting with a pan-specific anti-phosphoserine antibody. We found that treatment of MDA-MB-231 cells with LX7101 results in reduced serine phosphorylation of SRPK1 (Fig. [6b](#Fig6){ref-type="fig"}). To extend this analysis, we tested whether inhibition of LIMK2 or SRPK1 leads to reduced phosphorylation of downstream targets of SRPK1. We observed that inhibition of LIMK2 by LX7101 or of SRPK1 by a small-molecule inhibitor of SRPK1, SRPIN340, reduced the phosphorylation of the SRPK1 target protein Serine and Arginine Rich Splicing Factor 3 (SRSF3; Fig. [6c](#Fig6){ref-type="fig"}). Finally, we asked whether LIMK2 might directly phosphorylate SRPK1. In vitro kinase assays revealed that LIMK2 phosphorylates SRPK1 (Fig. [6d](#Fig6){ref-type="fig"}). These results suggest that the activity of SRPK1 is regulated by LIMK2; therefore, SRPK1 might function downstream of LIMK2 to affect the metastatic attributes of TNCB cells.Fig. 6SRPK1 is an important downstream mediator of LIMK2.**a** Schematic representation of SRPK1 protein with its identified phosphorylation sites. **b** MDA-MB-231 cells were treated with either LX7101 (5 μM) or vehicle control for 6 h. SRPK1 was immunoprecipitated from LX7101- and vehicle-treated-MDA-MB-231 cells. Immunoprecipitates were analyzed by immunoblotting with pan-phospho serine antibodies and beta-actin antibodies (loading control). **c** MDA-MB-231 cells were treated with LX7101 (5 μM), SRPK1 inhibitor SRPIN340 (10 μM), or vehicle control for 6 h. Expression of the indicated proteins was measured by immunoblotting. Beta-actin was used as a loading control. **d** LIMK2 in vitro kinase assays were performed in the presence and absence of ATP (1 mM). The samples were analyzed by immunoblotting for serine phosphorylation and total SRPK1. **e** TNBC cell lines (MDA-MB-231 and BT-549) expressing *SRPK1* shRNA were analyzed using a 3D spheroid invasion assay. Representative images taken at the indicated times are shown. Scale bar, 250 µm. **f** Relative invasion (%) calculated from the data presented in **e**. **g** Extracellular matrix degradation capacity of TNBC cell lines (MDA-MB-231 and BT-549) expressing *SRPK1* shRNA was analyzed using a gelatin degradation assay. Cells were stained with phalloidin (red) and DAPI (blue). Gelatin (green) degradation appears as black areas. Representative images are shown. Scale bar, 500 µm. Data are represented as the means ± SD. \*\*\**P* \< 0.001; \*\*\*\**P* \< 0.0001. To determine whether *SRPK1* knockdown mimics the phenotypes associated with LIMK2 inhibition, we knocked down the expression of SRPK1 in TNBC cell lines (MDA-MB-231, BT-549, MDA-MB-468; Fig. [S6a, b](#MOESM1){ref-type="media"}). SRPK1 knockdown cells were then analyzed using 3D invasion and ECM degradation assays. Our results showed that *SRPK1* knockdown impaired invasion in 3D collagen (Figs. [6e, f](#Fig6){ref-type="fig"} and [S6c, d](#MOESM1){ref-type="media"}) and led to decreased degradation of ECM (Figs. [6g](#Fig6){ref-type="fig"} and [S6e](#MOESM1){ref-type="media"}) in multiple TNBC cell lines. Finally, to corroborate the results obtained using *SRPK1* shRNAs, we used a small-molecule inhibitor of SRPK1, SRPIN340, on 3D invasion and ECM degradation in TNBC cells (MDA-MB-231, BT-549, and MDA-MB-468). SRPIN340 is an ATP-competitive serine-arginine-rich protein kinase inhibitor. Consistent with the results we obtained with *SRPK1* genetic knockdowns using shRNAs, the pharmacological inhibition of SRPK1 by SRPIN340 also resulted in reduced invasion (Fig. [7a, b](#Fig7){ref-type="fig"}) and led to decreased degradation of ECM (Fig. [7c](#Fig7){ref-type="fig"}) in multiple TNBC cell lines. Collectively, these results demonstrate that SPRK1 inhibition resulted in phenotypes similar to those observed with LIMK2 inhibition, and that SPRK1 inhibition reduced the metastatic properties of TNBC cells.Fig. 7SRPK1 is important for metastatic attributes of TNBC cells.**a** 3D spheroid invasion assay in TNBC cell lines (MDA-MB-231, BT-549, and MDA-MB-468) treated with SRPK1 inhibitor SRPIN340 (10 µM) or vehicle control. Representative images taken at the indicated times are shown. Scale bar, 250 µm. **b** Relative invasion (%) calculated from the data presented in **a**. **c** Extracellular matrix degradation capacity of TNBC cell lines (MDA-MB-231, BT-549, and MDA-MB-468) treated with SRPIN340 (10 µM) or vehicle control was analyzed using the gelatin degradation assay. Cells were stained with phalloidin (red) and DAPI (blue). Gelatin (green) degradation appears as a black area. Representative images are shown. Scale bar, 500 µm. Data are represented as the means ± SD. \*\**P* \< 0.01; \*\*\**P* \< 0.001. Pharmacological inhibition of LIMK2 inhibits metastatic progression in mice {#Sec6} --------------------------------------------------------------------------- Because we found that both genetic and pharmacological inhibition of LIMK2 blocked metastatic properties in cell culture-based assays, we asked whether LIMK2 inhibition using LX7101 could block the metastatic progression of TNBC cells in mice. To this end, we injected mice subcutaneously with MDA-MB-231 cells labeled with firefly luciferase. This was followed by treatment with either vehicle or the LIMK2 inhibitor LX7101. Throughout the experiment, tumor growth was measured each week. At the end of the experimental period, the animals were killed and the lungs were dissected and analyzed with bioluminescence imaging to measure spontaneous metastatic progression to the lungs. We found that LX7101 treatment did not inhibit tumor growth (Fig. [8a](#Fig8){ref-type="fig"}). However, mice treated with LX7101 showed spontaneous metastasis in only 1 out of 4 mice (25%), whereas vehicle-treated mice showed spontaneous metastasis in 3 out of 4 mice (75%; Fig. [8b, c](#Fig8){ref-type="fig"}). These results demonstrate the in vivo role of LIMK2 in distal metastatic progression and indicate that pharmacological inhibition of LIMK2 might be a useful therapeutic approach for preventing distal metastatic progression of TNBC.Fig. 8Pharmacological inhibition of SPRK1 or LIMK2 inhibits metastatic characteristics of TNBC.**a** Firefly luciferase--labeled MDA-MB-231 cells (MDA-MB-231-*F-Luc*) were injected subcutaneously into NSG mice (*n* = 4/group). From week 4, mice were treated orally with the vehicle (0.5% methyl cellulose in PBS) or 50 mg/kg LX7101 (in 0.5% methyl cellulose in PBS) every other day. Tumor volumes for mice were measured and plotted each week. Week 4 data includes four vehicle-treated mice and four drug-treated mice. **b** Imaged lungs of the mice killed at the end of the experiment presented in **a**. **c** Percentage incidence of metastasis is presented in indicated groups. Contingency analysis using Fisher's exact test shows statistically significant difference in the incidence of metastasis between vehicle or LX7101 treatment. **d** A model showing the mechanism by which LIMK2 facilitates metastatic progression of TNBC. The model indicates that LIMK2 is necessary for distal metastasis of TNBC and functions, in part, by phosphorylating and increased the kinase activity of SRPK1. Data in **a** is presented as the means ± SD. ns not significant *P*-value, \*\*\*\**P* \< 0.0001. Discussion {#Sec7} ========== Breast cancer remains the leading cause of cancer-related death in women. Among the breast cancer subtypes, TNBC remains a significant clinical challenge due to the relative ineffectiveness of current chemotherapeutic options and a paucity of actionable drug targets. In this study, we showed that LIMK2 kinase is overexpressed in TNBC and that both genetic and pharmacological inhibition of LIMK2 blocks TNBC metastatic progression due, in part, to its ability to phosphorylate SRPK1 and possibly other downstream targets (Fig. [8d](#Fig8){ref-type="fig"}). We found that LIMK2 was overexpressed in various breast cancer subtypes, including a substantial portion of TNBCs, using tissue microarrays. We corroborated this finding using publicly available datasets to show that LIMK2 expression was higher in TNBC tumors compared to other breast cancer subtypes and to show that LIMK2 overexpression predicted increased incidence of metastasis, disease recurrence, and death in breast cancer patients. These findings are consistent with those of a previous study that showed that elevated LIMK2 expression correlated with larger tumor size and higher histological grade and was a predictor of worse disease-free survival (DFS) and overall survival (OS)^[@CR38]^. Multivariable Cox regression analysis indicated elevated expression of LIMK2 was an independent prognostic factor for both DFS and OS^[@CR38]^. LIMK2 was previously implicated in promoting metastatic behavior in various cancer models. For example, a previous study showed knockdown of *LIMK2* reduced pancreatic cancer cell-induced angiogenesis^[@CR14]^. Another study showed that the selective LIMK2 inhibitor T56-LIMKi blocked pancreatic cancer xenograft growth in mice^[@CR39]^. Additionally, a colorectal cancer study found that the LIMK/cofilin pathway is associated with colorectal cancer progression and chemoresistance^[@CR18]^. Additionally, imbalance of LIMK1 and LIMK2 expression leads to human colorectal cancer progression and metastasis via promoting the nuclear translocation of β-catenin^[@CR11]^. We found that LIMK2 was required for maintaining the metastatic characteristics of TNBC cells because both genetic and pharmacological inhibition of LIMK2 inhibited their metastatic attributes. The finding that LIMK2 influences metastatic attributes but not tumor growth is intriguing, but not unprecedented. Other studies have also identified genes that primarily promote metastasis without affecting primary tumor growth^[@CR40],[@CR41]^, although the underlying mechanisms were not clear. Therefore, why some genes promote only tumor growth or metastasis remains an open question. Based on our results, we predict that LIMK2 promotes metastasis, but not tumor growth because the primary effect of LIMK2 is on phenotypes related to metastasis, such as invasion and migration. These characteristics are not required for tumor growth but are essential for metastatic progression. This idea is further supported by our mouse experiments, in which the treatment of MDA-MD-231 xenografts with the LIMK2 inhibitor LX7101 did not suppress primary tumor growth but did inhibit metastatic progression. Collectively, these results show that LIMK2 facilitates metastasis in TNBC. LIMK2 is a serine/threonine protein kinase for which only a limited number of downstream meditators have been identified^[@CR19],[@CR20],[@CR32]^. Prior studies have undertaken targeted approaches, rather than more comprehensive, unbiased approaches, like SILAC, to identify LIMK2 targets. Targeted approaches identified cofilin as a substrate of LIMK2. LIMK2 impairs the actin-severing activity of Cofilin by phosphorylating it at Serine 3, which causes actin stabilization^[@CR20],[@CR42]--[@CR44]^. Therefore, LIMK2 inhibition hyperactivates cofilin and increases actin turnover, leading to the disruption of actin stress fibers and their associated focal adhesions^[@CR20]^. Furthermore, LIMK2 phosphorylates the MT1-MMP protease on tyrosine 573, which enhances its association with cortactin, resulting in increased degradation of extracellular matrix and stimulation of invasion^[@CR28]^. In contrast to these prior studies, we applied the unbiased phosphoproteomics approach of SILAC to identify proteins with reduced phosphorylation following LIMK2 inhibition. We identified 258 proteins displaying significantly reduced phosphorylation levels. As expected, based on the known functions of LIMK2, many of the affected proteins were in pathways that regulate the actin cytoskeleton, integrin signaling, or cell polarity (PARVA, RRAS2, MPRIP, PAK2, ILKAP, and SCRIB). In addition, several functionally diverse protein groups, such as kinases (SRPK1 and RPS6KA4), phosphatases (PTPN12 and PTPN14), transcription factors (HSF1 and p53), or splicing factors (SRSF1 and SRSF6) were also identified. Furthermore, LIMK2 inhibition also affected mRNA processing (SRPK1, SRRM1/2, SRSF1, RBM15, and CLK3) and protein synthesis (EIF3G, RPS6, TSC2, RPTOR, and EIF4B) pathways. Collectively, these results show that LIMK2 inhibition led to decreased phosphorylation for a wide-array of proteins associated with many different biological processes. This, in turn, may affect the activity and/or stability of these proteins and thereby regulate the ability of LIMK2 to facilitate the metastatic attributes of TNBC cells. Taken together, we have created a comprehensive catalog of putative LIMK2 targets in TNBC cells. Further detailed characterization of these targets will likely reveal new functions of LIMK2 in TNBC, other subtypes of breast cancer, and other cancer types. Among the proteins identified in our SILAC analysis, we decided to further investigate SRPK1 as a mediator of LIMK2 function. SRPK1 is a serine/arginine protein kinase that specifically phosphorylates its substrates at serine residues in regions rich in arginine/serine dipeptides (known as RS domains) and is necessary for phosphorylation of SR splicing factors and the regulation of splicing^[@CR8],[@CR45]--[@CR47]^. Additionally, SRPK1 exerts metastasis-promoting activities in a wide variety of cancers, including breast cancer^[@CR32],[@CR34]--[@CR37]^. It is overexpressed in a variety of cancers and promotes tumor and metastatic growth^[@CR39],[@CR48]--[@CR50]^. Another recent study showed that SRPK1 maintains acute myeloid leukemia through effects on the isoform usage of epigenetic regulators including BRD4^[@CR51]^. We showed that like LIMK2, SRPK1 is overexpressed in several breast cancer subtypes compared to normal breast tissue. We also found that similar to LIMK2, SRPK1 was overexpressed in TNBC compared to other breast cancer subtypes, and SRPK1 overexpression in breast cancer predicted increased incidence of cancer metastasis, recurrence, and death. Our SILAC study identified reduced SRPK1 phosphorylation at five sites (Ser7, Ser9, Ser51, Ser309, and Ser311) after LIMK2 inhibition. Of these sites, Ser51 was previously shown to be phosphorylated by casein kinase 2^[@CR52]^. Additional experiments revealed that the genetic and pharmacological inhibition of SPRK1 phenocopied the effects of LIMK2 inhibition, indicating that this kinase, in addition to other LIMK2 substrates, could be an important downstream mediator of LIMK2 function. Because LIMK2 was overexpressed in TNBC and in vivo studies showed that its pharmacological inhibition can block the metastatic attributes of TNBC cells, we also tested the LIMK2 small-molecule inhibitor LX7101^[@CR30]^ in a mouse model of spontaneous TNBC metastasis. We found that the LIMK2 inhibitor LX7101 blocked metastatic progression without significantly affecting primary tumor growth, indicating that LIMK2 is important in the distal metastatic progression of TNBC cells. Collectively, our data link LIMK2 overexpression to TNBC metastasis and identify LIMK2 as a potential drug target for treating TNBC through its effects on SPRK1 and potentially other downstream targets identified by our SILAC studies. Materials and methods {#Sec8} ===================== Cell culture conditions and reagents {#Sec9} ------------------------------------ Short tandem repeat (STR) profile verified MDA-MB-231, BT-549, and MDA-MB-468 cell lines were obtained from the American Type Culture Collection (ATCC, Manassas, VA, USA) and maintained as recommended by the ATCC. MDA-MB-231 and MDA-MB-468 cells were grown in Dulbecco's Modified Eagle Medium (DMEM; Life Technologies, Thermo Fisher Scientific, Waltham, MA, USA), supplemented with 10% fetal bovine serum (FBS; Life Technologies, Thermo Fisher Scientific) and 1% penicillin/streptomycin (Life Technologies) under 5% CO~2~. BT-549 cells were grown in Roswell Park Memorial Institute (RPMI) 1640 medium (Life Technologies, Thermo Fisher Scientific) supplemented with 10% FBS and 1% penicillin/streptomycin in 5% CO~2~. Mycoplasma negative status for all cell lines was verified using MycoAlert mycoplasma detection kit (Lonza), and were routinely tested for the lack of mycoplasma contamination. Cell labeling and SILAC analysis {#Sec10} -------------------------------- Cells were seeded at 15% confluency in the respective complete medium (for MDA-MB-231: RPMI + 10% dialyzed FBS + 1% Pen-Strep). The labeling medium was deficient in lysine and arginine and supplemented with light- or heavy-labeled lysine (^13^C~6~ ^15^N~2~) and light- or heavy-labeled arginine (^13^C~6~ ^15^N~4~). Cells were subsequently cultured for at least five doublings in light or heavy medium and achieved over 95% labeling efficiency in our pilot experiments. After labeling, cells were treated for 6 hr with 5 μM LIMK2 inhibitor LX7101. After treatment, cells were trypsinized and counted to obtain a cell pellet of 2 × 10^7^ cells/condition and subjected to SILAC analysis using mass spectrometry. Additional details are presented in the [Supplemental Materials and Methods section](#MOESM1){ref-type="media"}. SILAC data analysis for identifying the preferred LIMK2 amino acid context for phosphorylation {#Sec11} ---------------------------------------------------------------------------------------------- To identify the LIMK2 phosphorylation consensus site from the SILAC data, we used a prediction algorithm developed in house. The motifs were generated by the R/Bioconductor package dagLogo (v.1.9.2). The background of the motifs was built from the human proteome retrieved via the R/Bioconductor package UniProt.ws (v.2.11.9). Lists of all quantified phosphopeptides for the MDA-MB-231 cell line are presented in Table [S5](#MOESM6){ref-type="media"}. The SILAC proteomics data have been submitted to PRIDE (<https://www.ebi.ac.uk/pride/archive/>). The data can be accessed via the PRIDE database using the accession number PXD008246. Immunohistochemistry {#Sec12} -------------------- Formalin-fixed, paraffin-embedded tissue microarray slides containing TNBC tissues were obtained from Yale Tissue Microarray Facility (Cat No. YTMA311 (*n* = 92), YTMA341 (*n* = 53), and YTMA347 (*n* = 54)) or from US Biomax, Inc. (Cat No. BC081120e). Briefly, following deparaffinization of slides, antigen retrieval was performed in citrate buffer (pH 6.0) at 97 °C for 20 min using the Lab Vision PT Module (Thermo Fisher Scientific). Endogenous peroxides were blocked using hydrogen peroxide, and proteins were blocked using 0.3% BSA. Slides were incubated in LIMK2 antibody (dilution 1:50) or SRPK1 antibody (dilution 1:500) followed by secondary anti-rabbit HRP-conjugated antibody (Dako, CA, USA). Slides were stained using the liquid DAB+substrate chromogen system (Dako) and counterstained using automation hematoxylin histological staining reagent (Dako). LIMK2 staining of tissue microarray slides was independently scored by three board-certified breast pathologists, Drs. Malini Harigopal, Ester Yoon, and Padmini Manrai, who were blinded to the slide identities. SPRK1 scoring was performed by Drs. Malini Harigopal and Kamalgeet Singh. The criteria for scoring IHC staining included percentage of positively stained cells and intensity of observed staining^[@CR53]^. For the purpose of our study, LIMK2 expression was recorded using an arbitrary scale based on the percentage of positive tumor cells: 0--25%; 26--50%; 51--75%, and 76--100%. To validate the specificity of the LIMK2 antibody used for IHC, immunofluorescence staining was performed on MDA-MB-231 cells expressing *LIMK2* shRNAs or a negative control shRNA. Antibody specificity was also validated separately by immunoblot analysis. Details of the antibodies used for IHC analyses are listed in Table [S8](#MOESM9){ref-type="media"}. Mouse tumorigenesis and spontaneous metastasis experiment with LX7101 treatment {#Sec13} ------------------------------------------------------------------------------- MDA-MB-231 cells stably expressing firefly luciferase under the control of a cytomegalovirus promoter were generated by cotransfection of the transposon vector piggyBac GFP-Luc and the helper plasmid Act-PBase as described previously^[@CR54]^. Cells with stable transposon integration were selected using blasticidin S (Thermo Fisher Scientific). MDA-MB-231-GFP-*F-Luc* cells (10 × 10^6^) in Matrigel (Corning; Cat No. 356237) were then injected subcutaneously into four female NSG mice aged 5--6 weeks per experimental group (Jackson Laboratory, Stock No. 005557). Tumor volume was measured every week. Tumor size was calculated using the following formula: length × width^2^ × 0.5. The vehicle (0.5% methyl cellulose in phosphate-buffered saline \[PBS\]) or LX7101 (50 mg/kg body weight) was administered by oral gavage every alternate day starting week 4 of injecting the cells (tumor volumes ∼80--100 mm^3^) until the end of experimental period. At the end of the experiment (4 weeks after the treatment), the mice were killed, and the lungs were imaged using the IVIS Spectrum in vivo imaging system (Perkin Elmer). All protocols for mouse experiments were approved by the Institutional Animal Care and Use Committee of the University of Alabama at Birmingham. Statistical analysis {#Sec14} -------------------- All of the experiments were conducted in triplicate. The results of the individual experiments are expressed as the means ± SD. For measuring statistical significance, the *P*-values were calculated using the two-tailed unpaired Student's *t* test in GraphPad Prism version 8.0 for Macintosh (GraphPad Software, San Diego, California, USA). Unpaired *t*-test with Welch's correction was applied to data with unequal variance. Contingency analysis was performed using Fisher's exact test in GraphPad Prism version 8.0 for Macintosh (GraphPad Software, San Diego, California, USA). Supplementary information ========================= {#Sec15} ###### Supplemental Information ###### Table S1 ###### Table S2 ###### Table S3 ###### Table S4 ###### Table S5 ###### Table S6 ###### Table S7 ###### Table S8 **Publisher's note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Parmanand Malvi, Radoslav Janostiak Supplementary information ========================= **Supplementary Information** accompanies this paper at (10.1038/s41389-020-00263-1). The authors declare that they have no conflict of interest.

O.S. & F.D. contributed equally to the manuscript. This article has been accepted for publication and undergone full peer review but has not been through the copyediting, typesetting, pagination and proofreading process, which may lead to differences between this version and the Version of Record. Please cite this article as doi: 10.1111/dth.13767.

{#sp1 .29} {#sp2 .30} {#sp3 .31} {#sp4 .32} {#sp5 .33} {#sp6 .34} {#sp7 .35} {#sp8 .36} {#sp9 .37} [^1]: A Paper read at a Meeting of the Bristol Medico-Chirurgical Society, held at the University of Bristol on January 13th, 1926.

Introduction {#s1} ============ The canonical Wnt/β-catenin signaling pathway plays pivotal roles not only in early embryogenesis but also in stem cell homeostasis and tumorigenesis [@pone.0063378-Clevers1]. The activation of Wnt signaling results in the stabilization of β-catenin through inhibition of glycogen synthase kinase 3 (GSK3), and β-catenin then translocates to the nucleus, where it serves as a coactivator for the Lef and Tcf family of DNA binding proteins in the formation of active transcriptional complexes at specific target genes [@pone.0063378-Behrens1]. Genetic studies have revealed that canonical Wnt/β-catenin signaling is essential for differentiation of the pluripotent epiblast into mesoderm in gastrulating mouse embryos [@pone.0063378-Liu1], [@pone.0063378-Huelsken1]. In contrast, the activation of Wnt/β-catenin signaling with compounds that inhibit GSK3 promotes propagation of mouse and human embryonic stem cells (ESCs) in the undifferentiated state [@pone.0063378-Ying1], [@pone.0063378-Sato1]. How the role of canonical Wnt signaling switches from maintenance of pluripotency in pluripotent stem cells to induction of mesoderm remains unknown, however. Mouse epiblast stem cells (EpiSCs), which are derived from the epiblast at embryonic day (E) 5.5 to E7.5, exhibit features of pluripotency and require Nodal-Activin and fibroblast growth factor (Fgf) signaling for maintenance of this characteristic. They are therefore thought to be more closely related to human ESCs than to mouse ESCs in this regard [@pone.0063378-Brons1], [@pone.0063378-Tesar1]. EpiSCs have little or no ability to give rise to chimeras when injected into blastocysts, suggesting that they are actually in a state of primed pluripotency, which represents a developmental state later than that of naïve mouse ESCs. A recent study showed that EpiSCs that express the E-cadherin gene (*Cdh1*) were able to contribute to embryos after blastocyst injection if they had been treated with the Rho kinase (ROCK) inhibitor Y27632 [@pone.0063378-Ohtsuka1]. Such expression and treatment may promote cell-cell adhesion and survival, and the injected EpiSCs may be reprogrammed to a naïve pluripotent state in the blastocyst and thus rendered capable of contributing to chimeric embryos. Although EpiSCs express the core determinants of pluripotency including Oct4, Nanog, and Sox2, many pluripotency-associated transcription factors for mouse ESCs are absent or present at low levels in EpiSCs [@pone.0063378-Brons1], [@pone.0063378-Tesar1], suggesting that these cells may possess a distinct regulatory circuitry for pluripotency. Rather than a naïve pluripotent ground state, EpiSCs appear to be in various heterogeneous states, likely reflecting their origin from a heterogeneous cell population in the epiblast at the egg-cylinder stage. They express early lineage markers for the primitive streak (PS) and mesoderm. The mixed developmental state of EpiSCs is thus considered a general hallmark of the primed state, which may restrict their pluripotency potential [@pone.0063378-Bernemann1], [@pone.0063378-Han1]. Similar heterogeneity is also evident in human ESCs, with different levels of canonical Wnt signaling possibly underlying variability in differentiation competence [@pone.0063378-Blauwkamp1], [@pone.0063378-Davidson1]. We have now investigated the role of canonical Wnt signaling in mouse EpiSCs with the use of small-molecule inhibitors and deletion of the β-catenin gene. Consistent with our previous observations in human ESCs [@pone.0063378-Sumi1], our data establish that canonical Wnt signaling impedes the self-renewal of EpiSCs with primed pluripotency as well as promotes mesoderm differentiation both in EpiSCs and in the postimplantation mouse embryo. Materials and Methods {#s2} ===================== Mice {#s2a} ---- ROSA26-*lacZ*, *β-catenin* ^fl/fl^, and UBC-*CreER* mouse strains (The Jackson Laboratory) were maintained on the B6/129 hybrid background. Mouse embryos were collected at E6.5, with noon of the day on which the vaginal plug was detected being designated E0.5. Mice harboring the UBC-*CreER* transgene were crossed with *β-catenin* ^fl/fl^ mice. Both UBC-*CreER* and *β-catenin* ^fl/fl^ alleles were identified by PCR analysis as described previously [@pone.0063378-Brault1]. Our study was approved by the Animal Care and Use Committee of Kyushu University (Permit Number: A24-038-1). EpiSC Culture {#s2b} ------------- All EpiSC lines used in the present study were newly established from E6.5 epiblasts of ICR, ROSA26-*lacZ*, or *β-catenin* ^fl/fl^ mice with a modified version of a method described previously [@pone.0063378-Tesar1], [@pone.0063378-Chenoweth1]. In brief, isolated epiblasts were cultured and propagated on a feeder layer in EpiSC medium \[Dulbecco\'s modified Eagle\'s medium (DMEM)-F12 containing 20% Knockout Serum Replacement (KSR, Invitrogen), 2 mM L-glutamine, 1 mM sodium pyruvate, 1× MEM nonessential amino acids, and 0.1 mM β-mercaptoethanol\] supplemented with Activin A (10 ng/ml, Peprotech), Fgf2 (5 ng/ml, Peprotech), and XAV939 (10 µM, Sigma-Aldrich). The effects of 20 µM CHIR99021 and 0.1 µM PD0325901 on the cells were examined. For the establishment of EpiSCs derived from single cells, isolated epiblasts were incubated with trypsin-EDTA (Nakarai) at 37°C for 10 minutes and then dissociated into single cells by passage through a pipette tip. The dissociated epiblast cells were seeded on a feeder layer in four-well plates with or without 10 µM Y27632 (Wako) and 25 µM XAV939 for the indicated times. Cells were fixed with 4% paraformaldehyde in phosphate-buffered saline (PBS) for staining of alkaline phosphatase activity with nitroblue tetrazolium. Quantitative RT-PCR Analysis {#s2c} ---------------------------- Total RNA was extracted and purified from EpiSCs with the use of Sepasol-RNA I Super G (Nakarai) and was subjected to RT with the use of a High Capacity cDNA Reverse Transcription Kit (Applied Biosystems). Real-time PCR analysis was performed with an ABI Prism 7500 Real-time PCR system and PowerSYBR Green PCR Master Mix (Applied Biosystems). The abundance of each mRNA was normalized by the amount of 18S rRNA and calculated by the standard curve method. Details of PCR primers are available on request. Immunoblot, Immunofluorescence, and Immunochemical Analyses {#s2d} ----------------------------------------------------------- Lysates of EpiSCs were prepared and subjected to SDS-polyacrylamide gel electrophoresis followed by immunoblot analysis as described previously [@pone.0063378-Sumi1]. For immunofluorescence analysis, cells seeded on cover glasses were fixed with 4% paraformaldehyde and permeabilized with 0.5% Triton X-100. They were then exposed to 5% bovine serum albumin (BSA) in PBS before consecutive incubations with primary antibodies and Alexa Fluor--conjugated secondary antibodies (Invitrogen). Cells were mounted on glass slides for examination with a confocal microscope (Leica TCS-SP5II). For immunochemical analysis, cells were fixed and permeabilized as for immunofluorescence analysis, incubated with 3% H~2~O~2~ for 10 minutes at room temperature, exposed to 5% BSA in PBS, and then incubated consecutively with primary antibodies and biotinylated secondary antibodies (Jackson Immunoresearch). They were then treated with ABC staining solution (Vector) and immersed in 3,3'-diaminobenzidine solution (Vector) for detection of immunoreactivity. Primary antibodies included those to Oct3/4, to E-cadherin, and to PCNA (Santa Cruz Biotechnology); those to β-catenin (BD Biosciences); those to Sox2 (Cell Signaling); and those to Nanog (Cosmo Bio). Cell Transplantation {#s2e} -------------------- Cell transplantation was performed as described previously [@pone.0063378-Lawson1]. EpiSCs derived from ROSA26-*lacZ* embryos were treated with Y27632 (10 µM) for 1 hour and then dissociated into single cells with trypsin-EDTA. ICR embryos at E6.5 and EpiSCs were handled with manipulators (Narishige) under an inverted microscope (Zeiss). EpiSCs (10 to 20 cells) were injected between the posterior epiblast and visceral endoderm layers of E6.5 embryos in DMEM supplemented with 10% fetal bovine serum. The injected embryos were then allowed to develop in whole-embryo culture. Cells expressing *lacZ* (those derived from EpiSCs) were visualized by fixation of embryos with 1% paraformaldehyde and 0.2% glutaraldehyde in PBS for 10 minutes followed by staining with 5-bromo-4-chloro-3-indolyl-β-D-galactopyranoside (X-gal, Roche). Whole-embryo Culture {#s2f} -------------------- Whole-embryo culture was performed as previously described [@pone.0063378-Oki1]. In brief, embryos at E6.5 were collected from pregnant ICR mice and transferred to DMEM supplemented with 10% fetal bovine serum. The embryos were then cultured under a humidified atmosphere of 5% CO~2~ at 37°C in DMEM supplemented with 75% rat serum either in four-well plates (for inhibitor experiments) or with rotation in 15-ml tubes (for cell transplantation experiments). In situ Hybridization {#s2g} --------------------- EpiSCs and embryos were fixed overnight at 4°C with 4% paraformaldehyde in PBS, dehydrated with a graded series of methanol solutions, and stored at --20°C prior to analysis. In situ hybridization was performed as described previously [@pone.0063378-Oki1]. The *Foxd3* probe plasmid was newly generated by cloning a 1410-bp fragment of the coding region of mouse *Foxd3* amplified by PCR into pBS (Stratagene). Results {#s3} ======= Wnt/β-catenin Signaling Promotes Epiblast Differentiation {#s3a} --------------------------------------------------------- The canonical Wnt/β-catenin signaling pathway plays a major role in maintenance of pluripotent mouse and human ESCs [@pone.0063378-Ying1], [@pone.0063378-Sato1], but it also promotes the differentiation of human ESCs toward mesoderm [@pone.0063378-Davidson1], [@pone.0063378-Sumi1]. To investigate the role of canonical Wnt/β-catenin signaling in primed mouse EpiSCs, which closely resemble human ESCs, we first examined the effects of activating such signaling in these cells. Stimulation of the canonical Wnt signaling pathway with CHIR99021, a small-molecule inhibitor of GSK3 [@pone.0063378-Ring1], in the absence of Activin and Fgf2 resulted in the rapid induction of *Brachyury* (*T*), *Mixl1*, and *Nanog* expression as well as in the repression of *Sox2* and *Foxd3* expression, indicative of the onset of mesoderm formation ([Fig. 1A and B](#pone-0063378-g001){ref-type="fig"}) [@pone.0063378-Pfister1]. To confirm the role of Wnt signaling in perigastrulation mouse embryos, we performed whole-embryo culture, which allows the direct chemical manipulation of embryos [@pone.0063378-Oki1]. Embryos at E6.5 were cultured with small-molecule inhibitors for 6 h and then subjected to whole-mount in situ hybridization analysis. In control embryos, formation of PS-mesoderm was apparent at the posterior-proximal side of the embryo proper, as revealed by the expression of *T* and *Nanog* in this region and by the low abundance of *Sox2* and *Foxd3* transcripts in the posterior epiblast ([Fig. 1C](#pone-0063378-g001){ref-type="fig"}). For embryos cultured with CHIR99021, however, the region of *T* and *Nanog* expression was expanded anteriorly, whereas *Sox2* and *Foxd3* expression had disappeared from the entire epiblast, indicative of the induction of PS-mesoderm throughout the epiblast ([Fig. 1C](#pone-0063378-g001){ref-type="fig"}). The expression of *Sox2* in the extraembryonic ectoderm was not affected by CHIR99021, suggestive of the operation of distinct mechanisms of transcriptional regulation in the embryonic and extraembryonic ectoderm. Conversely, inhibition of Wnt signaling with XAV939, a small-molecule inhibitor of Tankyrase [@pone.0063378-Huang1], resulted in suppression of the expression of *T* and *Nanog* as well as in maintenance of that of pluripotency factor genes such as *Oct4* (also known as *Pou5f1*), *Sox2*, and *Foxd3* in the entire epiblast ([Fig. 1C](#pone-0063378-g001){ref-type="fig"}). Collectively, these results supported the notion that canonical Wnt signaling induces the differentiation, rather than maintains the undifferentiated state, of EpiSCs and of the postimplantation epiblast. {#pone-0063378-g001} EpiSC Heterogeneity Caused by Wnt Signal Activation {#s3b} --------------------------------------------------- Mouse EpiSCs manifest various molecular features, including the expression of both pluripotency and PS-mesoderm markers [@pone.0063378-Brons1], [@pone.0063378-Tesar1], that are consistent with the properties of the epiblast of embryos at the perigastrulation stage. At this stage, Nodal/Smad2 and Wnt/β-catenin signaling pathways in embryos regulate the induction of the PS-mesoderm [@pone.0063378-Arnold1]. We hypothesized that inhibition of these signaling pathways might allow maintenance of EpiSCs in more undifferentiated state. Consistent with previous observations [@pone.0063378-Vallier1], however, inhibition of Nodal-Activin signaling in EpiSCs with the small-molecule inhibitor SB431542 resulted in a rapid loss of the undifferentiated state and in differentiation toward the neuroectoderm lineage as revealed by the expression of *Pax6* (data not shown). In contrast, inhibition of Wnt signaling with XAV939 improved the morphology of EpiSC colonies and extinguished differentiated cells usually found in EpiSC colonies cultured with standard medium (data not shown; also see below). Given that the expression profiles of EpiSCs harboring a transgene for green fluorescent protein (GFP) under the control of the *Oct4* regulatory region were found to differ between cell subpopulations positive or negative for GFP expression [@pone.0063378-Han1], we examined whether PS-mesoderm marker genes were uniformly expressed among EpiSCs in the absence (control) or presence of XAV939. In situ hybridization analysis revealed a patchy and speckled pattern of *T* expression and the partial loss of *Oct4* expression within control single EpiSC colonies, indicating that EpiSCs constitute a heterogeneous population as a result of spontaneous differentiation along the mesoderm and endoderm lineages during culture in the presence of Activin and Fgf2 ([Fig. 2A](#pone-0063378-g002){ref-type="fig"}). In contrast, inhibition of Wnt signaling with XAV939 resulted in the uniform expression of *Oct4*, *Sox2*, and *Nanog* as well as in the disappearance of *T* expression ([Fig. 2A](#pone-0063378-g002){ref-type="fig"}). In addition, expression of mesoderm and endoderm markers, including *Mixl1*, *Sox17*, *and Cer1*, was completely suppressed as a result of the inhibition of Wnt signaling. The expression of *Fgf5* and *Fgf8*, which is characteristic of the epiblast state, also exhibited a more homogeneous pattern in EpiSC colonies treated with XAV939 than in control colonies ([Fig. 2A](#pone-0063378-g002){ref-type="fig"}). Immunofluorescence analysis showed that β-catenin was localized to the nucleus in control EpiSCs but not in the cells treated with XAV939 ([Fig. 2B](#pone-0063378-g002){ref-type="fig"}). The levels of transcripts derived from pluripotency factor genes such as *Nanog*, *Sox2*, and *Oct4* in XAV939-treated EpiSCs were more than three to eight times those in control cells ([Fig. 2C](#pone-0063378-g002){ref-type="fig"}). These data indicated that canonical Wnt signaling is a determinant of EpiSC heterogeneity as a result of its induction of mesoderm and endoderm, and that inhibition of such signaling promotes the pluripotency of EpiSCs. {#pone-0063378-g002} Inhibition of Wnt Signaling Promotes the Establishment and Propagation of EpiSCs in the Undifferentiated State {#s3c} -------------------------------------------------------------------------------------------------------------- The finding of spontaneous Wnt signaling in EpiSCs prompted us to attempt the establishment of EpiSC lines comprising a more uniform cell population in the undifferentiated state through inhibition of Wnt signaling. Culture of primary epiblast explants under the standard condition with Activin and Fgf2 resulted in the formation of compact colonies of EpiSCs with a high nucleus-to-cytoplasm ratio ([Fig. 3A](#pone-0063378-g003){ref-type="fig"}). However, cells on the periphery of the colonies subsequently began to differentiate, giving rise to both alkaline phosphatase- and Oct4-negative cells. These differentiated cells were not observed when Wnt signaling was inhibited, although epiblast explants cultured with Fgf2 and XAV939 or with XAV939 alone could not expand continuously ([Fig. 3A](#pone-0063378-g003){ref-type="fig"}). Primary epiblast explants treated with the combination of Activin, Fgf2, and XAV939, however, were readily propagated for \>30 passages without obvious differentiation in that the expression of mesoderm and endoderm markers remained undetectable (data not shown). {#pone-0063378-g003} Given that EpiSCs, like human ESCs, do not survive well as single cells, we used the selective ROCK inhibitor Y27632 [@pone.0063378-Watanabe1] to establish EpiSC lines derived from single cells. Dissociation of individual E6.5 epiblasts to single cells with the use of trypsin-EDTA followed by culture of the cells for 5 days under the standard condition with Activin and Fgf2 did not result in the establishment of any EpiSC lines ([Fig. 3B--D](#pone-0063378-g003){ref-type="fig"}). Culture in the additional presence of Y27632 yielded many large colonies, but they were almost differentiated. Although XAV939-treated epiblast cells generated few undifferentiated colonies, culture in the presence of both XAV939 and Y27632 greatly increased the number of undifferentiated colonies ([Fig. 3B--D](#pone-0063378-g003){ref-type="fig"}). Given that the total cell number in the epiblast at around E6.5 is thought to be in the range of 250 to 660 [@pone.0063378-Snow1], we estimate that almost all epiblast cells produced undifferentiated stem cell colonies in the presence of these two inhibitors. Together, these results indicated that inhibition of Wnt signaling promotes EpiSC propagation in the undifferentiated state, and that Wnt signaling extinguishes the pluripotent stem cell identity of EpiSCs. β-Catenin is Dispensable for the Undifferentiated State of EpiSCs {#s3d} ----------------------------------------------------------------- To confirm further this notion, we took advantage of the availability of β-catenin--deficient EpiSCs and examined whether it was possible to maintain these cells in the undifferentiated state. EpiSC lines were established from E6.5 epiblasts of mice that were homozygous for a floxed allele of the β-catenin gene (*β-catenin* ^fl/fl^) and which also harbored a transgene (UBC*-CreER*) for a tamoxifen-activatable form of Cre recombinase. The cells were rendered null for *β*-*catenin* (*β*-*catenin* ^Δ/Δ^) by treatment with tamoxifen. Two independent cell lines (designated \#3 and \#4) were established, one of which (\#3) was subjected to further characterization ([Fig. 4](#pone-0063378-g004){ref-type="fig"}). Genotyping by PCR and immunoblot analyses confirmed the complete elimination of both the β-catenin gene and protein in *β*-*catenin* ^fl/fl^ cells in response to tamoxifen treatment ([Fig. 4A and B](#pone-0063378-g004){ref-type="fig"}). Whereas *β*-*catenin* ^fl/fl^ EpiSCs produced compact colonies typical of EpiSCs, the *β*-*catenin* ^Δ/Δ^ cells were loosely associated with each other ([Fig. 4D](#pone-0063378-g004){ref-type="fig"}), with the result that they were readily dissociated by treatment with collagenase (data not shown). In *β*-*catenin* ^fl/fl^ cells, E-cadherin accumulated at sites of cell-cell adhesion, where adherens junctions formed ([Fig. 4F](#pone-0063378-g004){ref-type="fig"}). In contrast, E-cadherin was diffusely distributed in *β*-*catenin* ^Δ/Δ^ cells, suggestive of an altered state of cell-cell adhesion ([Fig. 4F](#pone-0063378-g004){ref-type="fig"}) [@pone.0063378-Haegel1]. The abundance or distribution of Oct4 and Sox2 did not appear to differ between *β*-*catenin* ^fl/fl^ and *β*-*catenin* ^Δ/Δ^ cells, although the levels of *Oct4*, *Sox2*, and *Foxd3* transcripts in *β*-*catenin* ^Δ/Δ^ cells were two- to threefold those in *β*-*catenin* ^fl/fl^ cells ([Fig. 4B, C and F](#pone-0063378-g004){ref-type="fig"}). In addition, the expression of *T* was not observed in *β*-*catenin* ^Δ/Δ^ cells ([Fig. 4E](#pone-0063378-g004){ref-type="fig"}), suggesting that EpiSCs lacking β-catenin preserve the undifferentiated state. The levels of both Nanog protein and mRNA were down-regulated in *β-catenin* ^Δ/Δ^ cells compared with those in *β-catenin* ^fl/fl^ cells, however, and expression of the epiblast markers *Fgf5* and *Fgf8* was also diminished in the *β*-*catenin* ^Δ/Δ^ cells ([Fig. 4C, D and F](#pone-0063378-g004){ref-type="fig"}), suggesting that the *Fgf* genes are downstream targets of canonical Wnt signaling [@pone.0063378-Morkel1]. These changes in *β*-*catenin* ^Δ/Δ^ cells were not likely attributable to differentiation because *Oct4* and *Sox2* were stably expressed. To test whether the ablation of β-catenin in EpiSCs affects differentiation potential, we examined the effects of inducers of differentiation toward mesoderm or neuroectoderm. Consistent with previous results showing that *T* is a target of canonical Wnt signaling [@pone.0063378-Liu1], [@pone.0063378-Huelsken1], *β*-*catenin* ^Δ/Δ^ cells failed to up-regulate *T* expression even when cultured in the presence of CHIR99021 ([Fig. 4E](#pone-0063378-g004){ref-type="fig"}). Furthermore, whereas the MEK inhibitor PD0325901 [@pone.0063378-Greber1], [@pone.0063378-Yoo1] resulted in the formation of neuroectodermal cells positive for *Pax6* expression by *β*-*catenin* ^fl/fl^ cells, it had no such effect in *β*-*catenin* ^Δ/Δ^ cells ([Fig. 4E](#pone-0063378-g004){ref-type="fig"}). Cell-cell adhesion dependent on β-catenin has been shown to be necessary for neural differentiation of mouse ESCs [@pone.0063378-Lyashenko1], [@pone.0063378-Otero1]. Together, our data indicated that cell adhesion and transcriptional regulation mediated by β-catenin are not required for maintenance of self-renewal in EpiSCs, whereas β-catenin is required for differentiation toward mesoderm and neuroectoderm. {#pone-0063378-g004} Transplantation of EpiSCs Generates Chimeric Embryos {#s3e} ---------------------------------------------------- Finally, we examined the ability of XAV939-treated EpiSCs to differentiate in the developing mouse embryo. Given that EpiSCs were previously found not to contribute to chimeras after blastocyst injection [@pone.0063378-Brons1], [@pone.0063378-Tesar1], we performed cell transplantation in the gastrulating mouse embryo [@pone.0063378-Tam1], [@pone.0063378-Tam2]. EpiSCs expressing β-galactosidase (derived from ROSA26-*lacZ* embryos) were established with XAV939, treated with Y27632, and transplanted into the space between the epiblast and visceral endoderm of E6.5 embryos, which were then cultured for 2 days to allow development to the four- to six-somite stage. The embryos (*n* = 18) were then stained with the β-galactosidase substrate X-gal and examined for the distribution of EpiSC-derived cells. EpiSCs were found to contribute preferentially to the extraembryonic mesoderm including the allantoic bud (*n* = 17/18), yolk sac (*n* = 10/18), and amnion (*n* = 7/18) ([Fig. 5A--C](#pone-0063378-g005){ref-type="fig"}). Some cells also contributed to derivatives of endoderm (*n* = 11/18), mesoderm (*n* = 7/18), and ectoderm (*n* = 3/18) ([Fig. 5C--G](#pone-0063378-g005){ref-type="fig"}). Teratoma formation, a defining trait of pluripotency, was not observed in any of the EpiSC-grafted embryos. Given that primordial germ cells (PGCs) migrate from the extraembryonic mesoderm to the hindgut endoderm beginning around E7.5 [@pone.0063378-Ginsburg1], [@pone.0063378-Tsang1], we examined whether the transplanted EpiSCs were able to contribute to PGCs in the hindgut. Migrating PGCs at the base of the allantois and hindgut endoderm can be distinguished from other embryonic cells by their expression of the tissue-nonspecific form of alkaline phosphatase (TNAP) at the cell membrane [@pone.0063378-Ginsburg1], [@pone.0063378-Tsang1]. Double-staining for alkaline phosphatase and β-galactosidase activities revealed that some alkaline phosphatase--positive PGCs also exhibited β-galactosidase activity (*n* = 3/8 embryos) ([Fig. 5H](#pone-0063378-g005){ref-type="fig"}), indicative of the differentiation of EpiSCs into PGCs in vivo. Given that the EpiSCs were introduced into the space between the epiblast and visceral endoderm, and not into the epiblast layer, these cells might directly differentiate into various cell types without the gastrulation process in the epiblast. Together, these results indicated that EpiSCs have the developmental potential to form derivatives of all three embryonic germ layers as well as PGCs, and that they are able to contribute normally to the developing mouse embryo even after prolonged culture in the presence of an inhibitor of Wnt signaling. ![EpiSCs cultured with XAV939 contribute to chimeric embryos.\ Mouse embryos at E6.5 were injected with EpiSCs that harbor a *lacZ* transgene (ROSA26-*lacZ*) and which had been cultured in the presence of 10 µM XAV939 for at least 10 passages and exposed to 10 µM Y27632 for 1 hours before dissociation into single cells. The embryos were cultured until the early somite stage and then subjected to X-gal staining (green). The embryo in (H) was also stained for alkaline phosphatase activity (purple). LacZ-positive descendants of EpiSCs were detected in the allantoic bud \[asterisks in (A) and (C)\], yolk sac \[arrowhead in (A)\], amnion \[asterisk in (B)\], hindgut endoderm \[arrowheads in (C)\], somitic mesoderm \[arrowheads in (D)\], surface ectoderm \[arrowheads in (E)\], neural plate \[arrowheads in (F)\], myocardium \[arrowhead in (G)\], and migrating PGCs \[arrowheads in (H)\]. Embryos are oriented with anterior to the left (A, B, and E), posterior to the top (C, D, and H), or anterior to the top (F and G). A distal view of somites (D), enlarged view of the headfold (E), dorsal view of the neural plate (F), and ventral view of the heart primordium (G) are shown. Scale bars, 100 µm.](pone.0063378.g005){#pone-0063378-g005} Discussion {#s4} ========== We have shown that the canonical Wnt/β-catenin signaling pathway attenuates the undifferentiated state of EpiSCs and the epiblast of postimplantation mouse embryos. Importantly, expression of PS-mesoderm markers, which has been considered characteristic of EpiSCs, was found to be heterogeneous in these cells and to be extinguished by inhibition of Wnt signaling. These data indicate that the expression of PS-mesoderm markers is a consequence of spontaneous differentiation rather than a general hallmark of the primed pluripotent state. In mouse embryos, *Wnt3* is expressed in the posterior epiblast for induction of the PS, where mesoderm cells are generated [@pone.0063378-Liu1]. This fact suggests that EpiSCs in which canonical Wnt signaling has not been activated represent the ground state of primed pluripotency of the postimplantation epiblast. The heterogeneity caused by canonical Wnt signaling may be responsible in large part for the observed inefficiency in the direction of primed pluripotent stem cells toward specific cell types [@pone.0063378-Bernemann1], [@pone.0063378-Blauwkamp1], [@pone.0063378-Kim1]. The improvement in primed pluripotency observed in response to XAV939 treatment in the present study may also prove to be applicable not only to manipulation of human ESCs and induced pluripotent stem cells (iPSCs) of uniform quality but also to increasing the efficiency of somatic cell reprogramming in the establishment of human iPSCs. In mouse ESCs, the canonical Wnt signaling pathway functions to maintain pluripotency [@pone.0063378-Ying1], [@pone.0063378-Sato1]. On the other hand, β-catenin has been found to be dispensable for the pluripotency of these cells [@pone.0063378-Lyashenko1], [@pone.0063378-Wray1]. We have now shown that canonical Wnt signaling induces the differentiation of EpiSCs toward mesoderm, consistent with its function in mouse gastrulation [@pone.0063378-Liu1], [@pone.0063378-Huelsken1]. In human ESCs, the canonical Wnt signaling pathway induces differentiation rather than maintains pluripotency [@pone.0063378-Davidson1], [@pone.0063378-Sumi1]. The mechanisms by which canonical Wnt signaling promotes pluripotency in mouse ESCs and differentiation in human ESCs and mouse EpiSCs remain unclear. The modes of LIF/Stat3, BMP, Activin-Nodal, and Fgf/MAPK signaling have been found to differ between naïve and primed pluripotent cells [@pone.0063378-Nichols1]. These signaling pathways influence the functions of core transcription factors and other pluripotency-associated factors. The responses of naïve and primed pluripotent cells to canonical Wnt signaling may thus differ depending on the status of other signaling pathways that serve to maintain the core circuit of pluripotency. Although several downstream targets of canonical Wnt/β-catenin signaling have been identified [@pone.0063378-Cole1], [@pone.0063378-Pereira1], the target genes required for pluripotency and PS-mesoderm differentiation remain to be identified. Recent studies have shown that *Esrrb* is a pivotal target of β-catenin/Tcf3 as well as of Nanog in regulation of the naïve pluripotent state [@pone.0063378-Festuccia1], [@pone.0063378-Martello1], [@pone.0063378-Mesnard1]. However, it seems likely that the β-catenin/Tcf3/Esrrb pathway does not function in the primed pluripotent state, given that inhibition of GSK3 induces mesoderm differentiation ([Fig. 1](#pone-0063378-g001){ref-type="fig"}) rather than the expression of *Esrrb* in EpiSCs [@pone.0063378-Martello1]. We found that *Nanog*, *Fgf5*, and *Fgf8*, all of which are downstream target genes of Nodal signaling [@pone.0063378-Vallier1], [@pone.0063378-Mesnard1], are also regulated by β-catenin in EpiSCs. The expression of these genes was thus markedly diminished in *β-catenin* ^Δ/Δ^ cells even in the presence of Activin and Fgf2. The nuclear translocation of Smad2/3 did not differ between *β-catenin* ^fl/fl^ and *β-catenin* ^Δ/Δ^ cells (data not shown), indicating that Nodal signaling was not impaired by the loss of β-catenin. It is thus possible that β-catenin acts directly or cooperatively with Nodal signaling to regulate the expression of these genes in EpiSCs and the postimplantation epiblast. Given that XAV939 treatment conferred uniform expression of *Nanog*, *Fgf5*, and *Fgf8* in EpiSCs, a Wnt-independent function of β-catenin may be required for Nodal signaling to induce the expression of these genes. Evidence suggests that Nanog prevents neuroectoderm differentiation and supports the pluripotent state in EpiSCs and human ESCs [@pone.0063378-Vallier1]. Fgf signaling also inhibits neural commitment in EpiSCs and human ESCs [@pone.0063378-Greber1], [@pone.0063378-Greber2]. We were not able to induce expression of the neuroectoderm marker *Pax6* in *β-catenin* ^Δ/Δ^ EpiSCs by inhibiting Fgf/MAPK signaling, even though *Nanog* expression was down-regulated in these cells. These results suggest that β-catenin is essential for neural induction, and that Nanog is dispensable for inhibition of neural differentiation in *β-catenin* ^Δ/Δ^ cells. Consistent with this notion, Nanog was recently shown not to be required for establishment and maintenance of both naïve and primed pluripotency [@pone.0063378-Festuccia1], [@pone.0063378-Chambers1], [@pone.0063378-Osorno1]. The failure to induce neuroectoderm in *β-catenin* ^Δ/Δ^ EpiSCs is seemingly inconsistent with the fact that the neuroectoderm is formed in the anterior epiblast, which is devoid of canonical Wnt signaling. A function of β-catenin other than that in canonical Wnt signaling, such as in cell adhesion, may be required for the induction of neuroectoderm. We also found that the canonical Wnt/β-catenin signaling pathway negatively regulates *Sox2* and *Foxd3* expression in EpiSCs and gastrulating embryos. *Sox2* and *Foxd3* function in the maintenance of pluripotency in epiblast cells, with corresponding mutant embryos having been shown to die after implantation as a result of a loss of epiblast cells [@pone.0063378-Avilion1], [@pone.0063378-Hanna1]. Both genes also contribute to neural fate commitment [@pone.0063378-Dottori1], [@pone.0063378-Pevny1]. Consistent with this latter function, the expression of *Sox2* and *Foxd3* persists in the anterior epiblast during neuroectoderm differentiation, whereas the expression of these genes in the posterior epiblast is down-regulated during PS formation ([Fig. 1C](#pone-0063378-g001){ref-type="fig"}). Our results with embryo culture indicate that canonical Wnt signaling in the posterior epiblast attenuates the expression of *Sox2* and *Foxd3*. Expression of *Sox2* was previously shown to be increased in the posterior region of embryos lacking β-catenin [@pone.0063378-Lickert1]. Although it has not been examined, it is possible that Wnt3, a canonical Wnt ligand expressed at the perigastrulation stage, is responsible for this down-regulation of *Sox2* and *Foxd3* [@pone.0063378-Liu1]. Recent studies have shown that Oct4 and Nanog suppress the neuroectodermal fate and promote mesodermal differentiation, whereas Sox2 suppresses mesodermal fate, during the differentiation of mouse ESCs [@pone.0063378-Vallier1], [@pone.0063378-Teo1], [@pone.0063378-Thomson1]. Given that pluripotent factors participate in the differentiation of pluripotent stem cells, we therefore propose that germ layer specification during gastrulation may depend on transcriptional regulation of pluripotent factors by Wnt/β-catenin signaling. In summary, our data show that inhibition of canonical Wnt signaling confers a uniform pluripotent ground state on EpiSCs. Given the functional and molecular differences detected among human ESC lines as well as among human iPSC lines [@pone.0063378-Kim1], [@pone.0063378-Hong1], treatment with an inhibitor of canonical Wnt signaling might allow the isolation of primed pluripotent human ESCs and iPSCs of uniform and high quality. Further studies are warranted to identify the target genes of canonical Wnt signaling and to reveal the mechanisms by which such signaling elicits different responses in naïve and primed pluripotent stem cells. We thank H. Aoki, H. Suemori, M. Iwafuchi, S. Ohtsuka, and H. Niwa for technical advice on EpiSC culture; P. Tesar, A. Smith, and H. Niwa for providing EpiSCs; and the Research Support Center, Graduate School of Medical Sciences, Kyushu University, for technical support. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: TS SO. Performed the experiments: TS SO CM. Analyzed the data: TS SO KK CM. Contributed reagents/materials/analysis tools: TS SO KK. Wrote the paper: TS CM.

Following millennia of clinical medicine\'s efforts to embrace personalized, more effective and kinder treatments, arguably, it is the lessons garnered from patients with chronic myeloid leukemia (CML) in the chronic phase (CP) that appear to be moving us closer to the prospect of precision cancer medicine.^[@R1]^ The *BCR-ABL1* fusion gene encodes the oncoprotein BCR-ABL1 with a constitutive active tyrosine kinase activity that is the sole primary driver of the CP of CML. The discovery in 1996 that this kinase activity could be inhibited by imatinib mesylate (imatinib), a first-generation tyrosine kinase inhibitor (TKI) was pivotal. It established a new paradigm of targeted treatment for diverse cancers, in which relatively nonspecific and often toxic drugs are gradually being replaced by a cornucopia of safer, and in some cases, better targeted therapies and immunotherapies.^[@R2]^ Imatinib substantially and durably reduces the number of CML cells in the CP at a daily oral dose of 400 mg, and the life expectancy of most patients now approaches that of the general population.^[@R3]^ The greatest advance is in those patients who achieve a complete cytogenetic response within 2 years of starting imatinib leading to life spans indistinguishable from the general population.^[@R4]^ These impressive results with imatinib therapy have had profound effects on the natural history of CML and its prevalence. Current estimates suggest that in the United States, where about 5500 new cases are diagnosed annually, the prevalence may well increase to about 120,000 by 2020 and to about 200,000 by 2050.^[@R5]^ However, imatinib is far from perfect, with only approximately 60% of patients remaining on the standard daily dose of 400 mg after 6 years due to either lack of drug tolerance or drug resistance. Imatinib induces responses also in the more advanced phases of CML, but these responses are not durable. The 3 newer second-generation TKIs, dasatinib, nilotinib, and bosutinib, and the third-generation TKI, ponatinib, are all more potent than imatinib in in vitro assays. Current clinical experience suggests that patients treated with these newer TKIs achieve deeper and faster molecular responses than with standard-dose imatinib, but the precise benefits of such superior responses remain an enigma.^[@R6]^ Thus far, there is little evidence of a statistically significant improvement in overall survival with second-generation TKIs, though long-term follow-up confirmed a superior rate of freedom from progression compared with patients with less deep molecular responses at the same time points.^[@R7]^ It is possible, though not confirmed widely, that many of these patients will be able to discontinue therapy safely (treatment-free remission \[TFR\]) and effectively once they have been in a complete molecular remission (CMR) for about 24 months.^[@R8]^ Indeed, several studies, such as STIM, Euro-SKI, Australian CML-study, TWISTER, and other smaller studies, support the TFR concept.^[@R9],[@R10]^ Furthermore, the European Medicines Agency (EMA) recognized the significance of TFR, not just in terms of physical and financial toxicity for patients, but also for the society at large. It is possible, but no confirmed that the second- and third-generation TKI accord a greater potential in achieving TFR, and this remains the subject of ongoing studies. Recent expert statements provide a framework for consensus development, which should define the minimum acceptable *BCR-ABL1* transcript levels for TFR, precise definition of sustained CMR, and the impact of coexisting comorbidities, among others.^[@R11],[@R12]^ Randomized prospective studies have documented the occurrence of serious TKI-related cardiovascular events in CML patients with and without pre-existing cardiac conditions or risk factors, including adverse metabolic changes, diabetes mellitus, and lipid profile changes.^[@R13]--[@R17]^ Furthermore, meta-analyses and population-based studies clarify such risks as class effects or specific to certain TKIs.^[@R18]--[@R20]^ Clearly in efforts to effectively manage comorbidities and minimize treatment-related adverse events, in particular when commencing or switching to TKIs known to carry the highest risk for cardiovascular toxicity, robust recommendations for baseline and subsequent interval testing of indicators of vascular disease need to be in place. Additional tools, such as the Framingham risk model and the European Society of Cardiology-Score, and novel treatment approaches to suppress multiresistant CML subclones, such as "TKI rotation therapy," are being tested.^[@R21]--[@R23]^ Other areas of clinical importance include prediction of resistance to TKIs by assessing *BCR-ABL1* mutant subclone expansion, particularly in those who display 2 or more mutations in the *BCR-ABL1* kinase domain.^[@R24]^ The best treatment approaches for pediatric patients remain unclear. Recent work has confirmed the use of dasatinib as an effective treatment, with a safety profile similar to that seen in adults, though, interestingly, no examples of pleural or pericardial effusions or pulmonary arterial hypertension were noted.^[@R25]^ For patients with *BCR-ABL1*-negative myeloproliferative neoplasms (MPNs), clearly there was significant enthusiasm following the discovery of a gain-of-function mutation of JAK2 in 2005.^[@R26]^ Much has been learned about the genomic landscape and recent work, such as the importance of the order of acquisition of the mutations and their impact on the phenotype and clinical characteristics, including risks of thrombosis and prognosis, has now been garnered.^[@R27],[@R28]^ Recent revisions to the World Health Organization (WHO) classification and diagnostic criterion for all subtypes of MPNs, concomitantly with the introduction of genomic prognostic scoring systems, genetically inspired prognostic scoring system and mutation-enhanced international prognostic scoring systems for transplant-age patients (MIPSS70 and MIPSS70-plus) for patients with primary myelofibrosis (PMF), are noteworthy and impact clinical care.^[@R29]--[@R32]^ Though specific for patients with PMF, in the real world, these prognostic scoring systems are being used for all patients with myelofibrosis (MF), which include PMF, postpolycythemia vera MF (Post-PV-MF), and postessential thrombocythemia MF (Post-ET-MF). Ruxolitinib, a JAK1 and a type 1 JAK2 inhibitor with a short half-life, was licensed by the Food and Drug Administration in 2011 for patients with intermediate (with no specification for intermediate-1 or -2) and high-grade MF, and by the EMA in 2013 for patients with significant constitutional symptoms and splenomegaly, based on randomized trials.^[@R33],[@R34]^ Ruxolitinib does not appear to exert significant disease-modifying effects, with a minor effect on bone marrow fibrosis and *JAK2*^V617F^ allelic burden.^[@R35],[@R36]^ The ReTHINK trial attempted to assess the precise role of the drug in patients with high-risk MF without significant splenomegaly or symptoms, but had to be closed because of poor accrual.^[@R37]^ Another area of concern is the resistance to ruxolitinib, the mechanism of which remains poorly understood. The European LeukemiaNet (ELN) recently updated their treatment recommendations for all Philadelphia chromosome-negative MPNs,^[@R38]^ and included the WHO revisions, the newer prognostic scoring systems, and also the results of the randomized MAJIC trial.^[@R39],[@R40]^ The MAJIC trial is noteworthy of observing the lack of superiority of ruxolitinib compared to current second-line therapies for patients with ET. Ruxolitinib demonstrated some clinical efficacy in ET, but was only superior in terms of symptom control.^[@R40]^ The ELN expert consensus clinical management statements include the importance of health-related quality-of-life and the controversial iron-supplementation needs for some cases with PV following frequent phlebotomy. It also discusses the updates on hydroxyurea-resistant or -intolerant patients with PV and the benefit from the use of long-acting interferon alpha or ruxolitinib. Indeed, the current results of ropeginterferon alpha 2b versus hydroxyurea as frontline therapy for patients with PV demonstrate more complete hematologic and molecular responses following interferon treatment, compared with hydroxyurea after 2 years.^[@R41]^ The clinical development of next-generation JAK2 inhibitors has been difficult with many studies being discontinued due to the emergence of serious neurotoxicity, in particular Wernicke\'s encephalopathy.^[@R42]^ Fedratinib\'s development in MF was discontinued in 2013, and is now being re-evaluated.^[@R43]^ It was previously shown to be superior to placebo for control of splenomegaly and symptoms in patients with MF in the Jakarta I study, in addition to be active in the second-line setting for patients with MF who had previously been on ruxolitinib in the Jakarta II study.^[@R43],[@R44]^ Another JAK2 inhibitor, pacritinib, previously shown to be active in the PERSIST-1 and PERSIST-2 studies, is undergoing further development with refinement of optimal dosage.^[@R45],[@R46]^ And the recent phase 3 study results show pacritinib 200 mg twice daily to be significantly better than best available therapy, including ruxolitinib, for reducing splenomegaly and clinical symptoms in patients with MF and thrombocytopenia, for both previously untreated and those who had received prior ruxolitinib. Momelotinib, a JAK2 inhibitor, was tested in the SIMPLIFY-1 study in late 2017, but the trial failed to meet its primary endpoint.^[@R47],[@R48]^ In addition to JAK inhibition and interferons, many other investigational agents, either alone or in combination with ruxolitinib, are being tested. These include hedgehog, aurora kinase, SMAC, HDAC, and MDM2 inhibitors, in addition to the JAK2-allosteric inhibitors, such as LS104 and ON044580, which have a greater specificity for *JAK2*^*V617F*^ and are inhibitory in a non-ATP-competitive manner, and were recently reviewed.^[@R24]^ The success of targeted therapy for patients with CML in CP has been contingent upon *BCR-ABL1* being the founder lesion in every cell, and minimal genetic diversity. Resistance can be an issue, but many patients can achieve durable second and subsequent remissions, following a switch to an alternative TKI or an allogeneic stem cell transplant. By contrast, clinical progress in other subtypes of MPNs, which can demonstrate significant genetic diversity, has been qualified and limited to few patients. Interferons and ruxolitinib are useful in some patients with MF and PV. The authors thank the workshop faculty and Dr Alpa Parmar for their help and insights. **Citation:** Mughal TI, Saglio G, Van Etten RA. Chronic Myeloproliferative Neoplasms: Some Remaining Challenges. HemaSphere, 2018;1:1. <http://dx.doi.org/10.1097/HS9.0000000000000147> This article is based on the presentations and deliberations at the 12th post-ASH Workshop on BCR-ABL1-positive and -negative MPNs that took place on the 12th to 13th December 2017, in Atlanta, Georgia, immediately following the 59th American Society of Hematology Annual Meeting. Funding/support: None. Disclosure: TIM: Consultancy: Roche. Research funding: Alpine Oncology Foundation. GS: Consultancy: Bristol-Myers Squibb; Research funding: Novartis. RAVE: Scientific Advisory Boards: Bristol Myers-Squibb, Pfizer. Authorship contributions: TIM, GS, and RV designed the outline strategy of the manuscript, analyzed and interpreted the data, and wrote the manuscript.

On-demand or environmentally triggered disassembly of polymers is a widely sought-after goal, as such materials would be tremendously useful in a broad range of industries, including healthcare, cosmetics, agriculture, and electronics.^[@cit1],[@cit2]^ Despite this, few synthetic polymers degrade with high sensitivity in response to specific stimuli. Most current degradable materials are unresponsive to the often subtle changes found in biological systems or, in the case of photodegradable polymers, require long, intense irradiation that may not be biologically compatible. This limitation results from the fact that most of these materials convert one signalling event to only one chemical change, such as a single break in the polymer backbone^[@cit3]--[@cit5]^ or a change in hydrophobicity of one monomeric unit.^[@cit6],[@cit7]^ Self-immolative polymers can amplify responses to stimuli *via* head-to-tail depolymerization and have thus been developed to circumvent this limitation.^[@cit8]--[@cit10]^ However, most of these materials rely on slow intramolecular rearrangements to degrade their backbone,^[@cit11]--[@cit16]^ ultimately slowing down depolymerization. Alternatively, self-immolative polymers containing more labile bonds have also been developed,^[@cit17],[@cit18]^ but these bonds are likely not resilient enough to escape degradation in a physiological setting, even in the absence of the intended stimulus. The Phillips group has recently made substantial improvements to self-immolative polymers by creatively altering polymer backbones to maximize the effect of slow rearrangements^[@cit19],[@cit20]^ and minimize nonspecific degradation,^[@cit21]^ but there is still room to add to these strategies. Here, we have designed a polymer in which photocleavage unmasks acidic groups in the polymer backbone that then provide intramolecular assistance to ketal hydrolysis^[@cit22]^ so that minimal signal, in this case brief, low-power UV irradiation, triggers significant polymer degradation. This strategy should allow faster release with less irradiation than existing light-degradable polymers.^[@cit23]--[@cit25]^ Our design was inspired by the extensive literature on rates and mechanisms of ketal hydrolysis,^[@cit21],[@cit22]^ degradation rates of polyketals,^[@cit26]^ and disassembly of ketal-modified polymeric particles.^[@cit27]--[@cit31]^ Ketal hydrolysis rates are known to vary with hydrophilicity,^[@cit32],[@cit33]^ and water accessibility affects the kinetics of disassembly and degradation of polymeric particle assemblies containing ketals either within the backbone^[@cit34],[@cit35]^ or as pendant groups.^[@cit36]^ These findings inspired hydrophobic--hydrophilic switching mechanisms to exert further control over particle disassembly and/or degradation.^[@cit34]^ More recently, our group observed rapid degradation of a polyketal due to intramolecular assistance of acids^[@cit21],[@cit22],[@cit36],[@cit37]^ in a polymer designed as an MRI contrast agent.^[@cit37]^ The degradation occurred much more rapidly (in hours) than in comparable hydrophilic polymers (in days)^[@cit26]^ at the same buffered pH but containing no intramolecular acids. Here we employ the same concept to a light-degradable particle. We incorporate photoacids as pendant groups into a polyketal backbone ([Scheme 1](#sch1){ref-type="fig"}), from which we formulate particles. Cleavage of the photocage upon UV irradiation unmasks a carboxylic acid. This both releases acid groups in the vicinity of the backbone ketals (not necessarily adjacent along the backbone; polymer entanglement in a nanoparticle would juxtapose groups that would be distant from one another in dilute solution), and makes the polymer more hydrophilic, both of which facilitate ketal hydrolysis. {#sch1} To synthesize a polymer containing both ketal moieties and protected acid functions, we prepared two monomers for copolymerization ([Scheme 1](#sch1){ref-type="fig"}). Ketal monomer **5** was prepared by established methods.^[@cit26]^ To synthesize the monomer bearing a protected acid, **2** was esterified with alcohol **6** to form **3**. *ortho*-Nitrobenzyl alcohol **6** was chosen as a photolabile protecting group due to its commercial availability, relatively high tolerance to subsequent reactions, and its well-characterized photochemistry.^[@cit38],[@cit39]^ Though **6** has limitations as a photolabile group (low tissue penetration of UV light for drug delivery applications) it and related protecting groups have been used for cell studies^[@cit40]--[@cit42]^ and creative drug delivery methods in mammals.^[@cit43],[@cit44]^ Deprotection of the amines of **3** and treatment with acryloyl chloride gave **4**. Monomers **4** and **5** in equal proportions were copolymerized using a Michael addition with 1,3-propanedithiol to yield polymer **1** with weight average molecular weight (*M* ~w~) 13 900 Da and a polydispersity index (PDI) of 1.71 by gel permeation chromatography (GPC) relative to poly(methyl methacrylate) standards (Fig. S1, ESI[†](#fn1){ref-type="fn"}). The monomers were incorporated equally as seen by ^1^H NMR spectroscopy (Fig. S2, ESI[†](#fn1){ref-type="fn"}). Though it leads to relatively high PDI Michael addition proved to be an ideal means of polymerization due to its relatively mild conditions, a necessity to avoid degradation of the ketal. Polymer degradation was monitored using ^1^H NMR spectroscopy by following hydrolysis of the ketal to determine the degradation rate ([Fig. 1a and b](#fig1){ref-type="fig"}). Polymer **1** was dissolved in a 9 : 1 mixture of deuterated DMSO and deuterated phosphate buffer at pH 7.4 and phosphate solution at pH 5 and irradiated for times ranging from 0 to 20 minutes with UV light (1.35 mW cm^--2^). Irradiation and release of acids did not noticeably change the pH of either solution. Though the high proportion of organic solvent slows ketal hydrolysis by orders of magnitude,^[@cit45],[@cit46]^ DMSO was required to solubilize the polymer prior to irradiation. Following irradiation substantial amounts of the light-sensitive protecting groups still appeared intact; by ^1^H NMR only 50% of the acids were exposed even after 20 min of irradiation (Fig. S3A, ESI[†](#fn1){ref-type="fn"}). The samples were then monitored by ^1^H NMR spectroscopy at various time points throughout incubation at 37 °C. Although the ketal peak diminished and the acetone peak grew ([Fig. 1c](#fig1){ref-type="fig"}), the percentage of hydrolyzed ketal over time could not be accurately determined because of signal overlap. Ketal hydrolysis was instead followed by conversion of the methylene protons ([Fig. 1a](#fig1){ref-type="fig"}, protons A) vicinal to the ketal into protons vicinal to an alcohol. The initial rate of ketal hydrolysis was determined for each condition ([Fig. 1b](#fig1){ref-type="fig"}). {#fig1} The initial rate of hydrolysis at pH 7.4 increased with longer irradiation times, becoming four times faster after 20 minutes of irradiation than with no irradiation ([Fig. 1b](#fig1){ref-type="fig"}). Irradiation for only 5 min caused the pH 7.4 degradation kinetics to be 55% faster than the pH 5.0 degradation kinetics without irradiation. Comparable polymers containing the same ketal moiety have a half-life of roughly 1 h at pH 5 in solutions with a smaller proportion of organic solvents, suggesting that this polymer would degrade even more rapidly in biological settings.^[@cit26]^ A control polymer with benzyl protecting groups (removable by hydrogenation), polymer **9**, was synthesized (Fig. S9, ESI[†](#fn1){ref-type="fn"}) to ensure that degradation was accelerated by release of acids. No substantial difference in rate was observed between irradiated and untreated polymer **9**. In contrast, degradation was accelerated when roughly 50% of the acids of polymer **9** were exposed by hydrogenation (Fig. S11, ESI[†](#fn1){ref-type="fn"}). Polymer degradation was also assessed by GPC (Fig. S3, ESI[†](#fn1){ref-type="fn"}). The immediate shift to longer retention times observed upon irradiation of samples of polymer **1** is too rapid to indicate degradation. Instead, it likely results from a change in hydrophilicity caused by release of acids, increasing interactions with the column material. Shifts towards longer retention times in subsequent time points do support polymer degradation in support of the NMR spectroscopy experiments. To examine whether this degradation strategy allows rapid light-triggered release, we formulated nanoparticles of polymer **1** by single emulsion encapsulating the model payloads fluorescein diacetate (FDA) or Nile red (size = 193 ± 23 nm). We first examined light-triggered release by measuring fluorescence quenching of encapsulated Nile red. Nile red is fluorescent in the hydrophobic environment of nanoparticles, but its fluorescence is quenched in aqueous environments. Rapid fluorescence quenching was observed upon irradiation of particles suspended in pH 8.0 tris buffer ([Fig. 2a](#fig2){ref-type="fig"}). This quenching indicates substantial changes in morphology, allowing Nile red escape or entry of water into the particles. Particle degradation was assessed following irradiation and subsequent incubation at 37 °C by dynamic light scattering (DLS) with fixed attenuation. Upon UV irradiation, count rate decreased substantially and the PDI increased within 4 h, indicating substantial changes in particle morphology and possible degradation ([Fig. 2b](#fig2){ref-type="fig"}). Particles remained relatively stable in the absence of irradiation. The morphological changes were further examined by transmission electron microscopy (TEM) ([Fig. 2c](#fig2){ref-type="fig"}). After irradiation, subsequent incubation for 4 h, and drying particles appeared to disintegrate ([Fig. 2d](#fig2){ref-type="fig"}). {#fig2} To confirm payload release from nanoparticles, Raw 264.7 mouse macrophage cells were incubated with particles containing FDA ([Fig. 3a](#fig3){ref-type="fig"}) and irradiated for 5 min with UV light (10 mW cm^--2^) ([Fig. 3b](#fig3){ref-type="fig"}). This is a comparable power and shorter irradiation time than has been used with materials incorporating this photocage in cellular studies.^[@cit47],[@cit48]^ FDA is a non-fluorescent molecule hydrolyzed by intracellular esterases to form fluorescent fluorescein; only released FDA would encounter these esterases. UV irradiation led to high intensity fluorescence, while non-irradiated cells did not fluoresce appreciably ([Fig. 3c](#fig3){ref-type="fig"}). This demonstrates that nanoparticles composed of polymer **1** release cargo in the presence of cells under irradiation conditions that have minimal impact on cellular viability (the viability of cells irradiated with particles is confirmed by MTT assay (Fig. S8, ESI[†](#fn1){ref-type="fn"})). {#fig3} Finally, we assessed cellular compatibility by MTT assay in Raw 264.7 mouse macrophage cells after treatment with empty nanoparticles irradiated prior to treatment ([Fig. 4](#fig4){ref-type="fig"}), irradiated after incubation with cells (Fig. S8, ESI[†](#fn1){ref-type="fn"}), not irradiated, and polymer **1** (Fig. S7, ESI[†](#fn1){ref-type="fn"}). Neither nanoparticles nor polymer significantly impacted mitochondrial activity up to 200 μg mL^--1^, suggesting polymer **1**\'s potential for drug delivery. Particle degradation products also had less effect on cellular viability than intact nanoparticles ([Fig. 4](#fig4){ref-type="fig"}). {#fig4} Herein we have demonstrated that unmasking acids in a polymer backbone to accelerate the hydrolysis of ketals at neutral pH is a viable strategy to accelerate polymer and particle degradation. Rapid light-triggered release from polymer **1** nanoparticles demonstrates the potential of this strategy for triggered degradation in general; other chemical groups could be employed to confer responsiveness to other stimuli. The authors gratefully acknowledge the NIH New Innovator Award (DP 2OD006499) and KACST-UCSD Center for Excellence in Nanomedicine and Engineering for funding. NMR spectra were acquired at the UCSD Skaggs School of Pharmacy and Pharmaceutical Sciences NMR facility. The authors would also like to thank Jessica Moore, Minnie Chan, Amy Moore, Carl-Johan Carling and Brendan Duggan. [^1]: ‡Present address: ETH Zürich, Vladimir-Prelog-Weg 5, 8093 Zürich, Switzerland. [^2]: †Electronic supplementary information (ESI) available. See DOI: [10.1039/c5cc06143a](10.1039/c5cc06143a) Click here for additional data file.

Introduction {#s1} ============ Human mitochondrial diseases affecting the oxidative phosphorylation (OXPHOS) system can result from a large number of different mutations, both in the nuclear genome or in the maternally inherited mitochondrial DNA (mtDNA) [@pone.0008549-Schon1], [@pone.0008549-Smeitink1]. Environmental factors can also trigger or aggravate these diseases. The clinical phenotypes of mitochondrial diseases are highly variable [@pone.0008549-Smeitink1]. Although tissues most obviously dependent on bioenergy are commonly affected, notably heart and skeletal muscle, the central nervous system and sensory epithelia, the specific phenotypes are not understood. In general, genetic disorders of mitochondrial OXPHOS can be classified into those affecting a specific subunit of one of the four OXPHOS complexes to which mtDNA-encoded translation products contribute, (equivalent to *mit^−^* mutations in yeast) and those affecting the biosynthesis of many or all of the mtDNA-encoded polypeptides (equivalent to *syn^−^* mutants in yeast). The former class includes disorders caused by point mutations either in mtDNA-encoded polypeptides, such as the NARP syndrome [@pone.0008549-Schon2], or nuclear coded OXPHOS subunits[@pone.0008549-Triepels1]. The *syn^−^* class includes disorders such as MELAS or MERRF, caused by mutations in mitochondrial tRNA genes [@pone.0008549-Kirino1], as well as a diverse set of nuclear gene disorders caused by mutations in genes for the apparatus of mtDNA maintenance and expression. Examples of the latter include DNA polymerase γ [@pone.0008549-Copeland1], [@pone.0008549-Hakonen1], mitoribosomal proteins MRPS16 and MRPS22 [@pone.0008549-Miller1], [@pone.0008549-Saada1], and the SURF1 assembly factor for complex IV (cytochrome *c* oxidase) [@pone.0008549-Shoubridge1]. In both arthropods and vertebrates, mtDNA is a compactly organized circular molecule which encodes just 13 of the more than 75 polypeptides that comprise the five OXPHOS complexes of the inner mitochondrial membrane. In addition, it encodes the two rRNAs and 22 tRNAs necessary for their synthesis within mitochondria. Mitochondrial protein synthesis also requires approximately 100 or more nuclear-coded gene products that have to be transported into mitochondria. In addition, all of the proteins involved in the maintenance, replication and transcription of mtDNA, as well as the many chaperones involved in the assembly of the OXPHOS complexes and the proteins that influence the intracellular organization and distribution of mitochondria are encoded in the nucleus [@pone.0008549-OBrien1], [@pone.0008549-Smeitink2]. These nuclear encoded gene products include all of the protein components of the mitoribosome, which comprise an entirely different set than those present in cytosolic ribosomes [@pone.0008549-OBrien1]. Some of them have no counterparts in cytosolic or bacterial ribosomes, whereas others are phylogenetically conserved components of an ancient machinery of protein synthesis. One of these, the homologue of bacterial ribosomal protein S12, is a major component of the ribosomal decoding centre, and is of critical importance for translational accuracy. Mitoribosomal protein S12 (mRpS12) has been characterized in diverse taxa, including mammals [@pone.0008549-Shah1], [@pone.0008549-Toivonen1] and also *Drosophila* [@pone.0008549-Shah1], [@pone.0008549-Royden1], where it is encoded by the gene *technical knockout* (*tko*). It is well conserved in bacteria, as well as in the chloroplasts of higher plants and algae such as *Euglena*. The gene name in *Drosophila* reflects the so-called bang-sensitive phenotype of the canonical allele, *tko^25t^*, which suffers paralytic seizures induced by mechanical stress, This phenotype is shared with other mutants affecting mitochondrial bioenergy supply, e.g. in genes such as *sesB*, the adenine nucleotide translocase [@pone.0008549-Zhang1], or *knockdown*, citrate synthase [@pone.0008549-Fergestad1]. Null alleles of *tko* are larval-lethal, but the *tko^25t^* phenotype is relatively mild, and thus constitutes an animal model for mitochondrial disorders. In addition to seizure sensitivity, *tko^25t^* exhibits delayed larval development, antibiotic sensitivity, hearing impairment, locomotor hyporeactivity, and defective courtship [@pone.0008549-Toivonen2]. It carries a point mutation, L85H, at a conserved amino acid of mRpS12, which leads to the destabilization or defective assembly of the small mitoribosomal subunit [@pone.0008549-Toivonen1], [@pone.0008549-Toivonen2]. The resulting insufficiency of mitochondrial translational capacity entrains a substantially reduced activity of the major OXPHOS complexes to which the mtDNA-encoded polypeptides contribute, both in larvae and in adults, which is believed to underlie the developmental and behavioural phenotype [@pone.0008549-Toivonen2], [@pone.0008549-Toivonen3]. All aspects of the mutant phenotype are restored to wild-type by expression of a transgenic copy of the wild-type *tko* gene under the control of its natural promoter [@pone.0008549-Toivonen2]. The severity of the *tko^25t^* phenotype varies according to nuclear background [@pone.0008549-Toivonen2] and gene dosage [@pone.0008549-Kemppainen1], indicating that compensatory mechanisms can partially alleviate the effects of this stress. In order to gain insight into these compensatory mechanisms, and thus enhance our understanding of the global physiology of human mitochondrial disorders for which *tko^25t^* serves as a model [@pone.0008549-Smeitink2], we carried out a transcriptome-wide analysis, using the Affymetrix platform. We identified a number of genes for components of metabolic pathways systematically up- (or down-) regulated at the RNA level, induction of some specific stress-response genes, alterations in the expression of certain genes involved in development and reproduction which mirror the organism-level phenotype, and increased expression of a number of genes putatively involved in intra- and intercellular signalling which suggest pathways by which these changes might be effected. Based on our findings, and extrapolating from *Drosophila* to humans, we suggest that nutritional supplementation might be considered an appropriate strategy in the management of some types of mitochondrial OXPHOS disease. Results and Discussion {#s2} ====================== Identification of *tko^25t^*-Regulated Genes {#s2a} -------------------------------------------- Inbreeding under stressful conditions inevitably results in the selection of compensatory alleles of many genes. Previous analyses of *tko^25t^* indicated that inbred lines were, indeed, subject to partial suppression of the mutant phenotype [@pone.0008549-Toivonen2]. In order to avoid such issues, and thus determine the global effects on gene expression of the *tko^25t^* mutation in a truly unselected, 'wild-type' background, we outbred *tko^25t^* over more than 10 generations by back-crossing to each of two commonly used wild-type strains, Canton S and Oregon R. Subsequent to this backcrossing, *tko^25t^* was maintained in each background using a balancer chromosome. These stocks were then used to generate a *tko^25t^* mutant F1 generation, by crossing virgin Canton S *tko^25t^* homozygous mutant females with Oregon R *tko^25t^* males, as illustrated in [Figure 1](#pone-0008549-g001){ref-type="fig"}. For comparison, we generated otherwise isogenic wild-type F1 progeny by crossing virgin Canton S wild-type females with Oregon R wild-type males. ![Crossing scheme to generate maximally outbred *tko^25t^* mutant flies for analysis.\ Balanced stocks were used first to create homozygous females and hemizygous males of the two parental backgrounds, in order to include in the analysis any maternal effects of the mutation. Note that *tko* is an X-chromosomal gene. The initial outbreeding to create the balanced stocks restores a wild-type genetic background, but does not completely eliminate any potentially compensatory recessive alleles already in the wild-type backgrounds. To minimize the effects of any such alleles, the crossing scheme illustrated is both maximally wild-type and heterozygous, under which conditions we saw the most substantial accentuation of the mutant phenotype, compared with inbred *tko^25t^* lines [@pone.0008549-Jacobs1].](pone.0008549.g001){#pone-0008549-g001} Flies of both sexes were collected from three independent such crosses, their RNA extracted and used to synthesize cRNA probes for hybridization to separate oligonucleotide arrays as described in Experimental Procedures. Data analysis for each gene in the array compared the signal of each *tko^25t^* mutant mRNA with the signal of each wild-type mRNA from flies of the given sex, over which the statistical analysis was performed. First of all, the data were prefiltered according to their detection p-value, selecting those probe sets with a significant p-value (\<0.05) in their detection signal; this preliminary list comprised approximately 50% of the probe sets present in the array ([Table 1](#pone-0008549-t001){ref-type="table"}). Afterwards, MAS5 and RMA algorithms were performed using SAM software in order to select those probe sets with significant differences in gene expression. Approximately 7% and 3% of the probe sets were picked with a fold change higher that 1.5, in males and females respectively ([Table 1](#pone-0008549-t001){ref-type="table"}), although the false discovery rate (FDR) was very high (58% and 34% respectively). By increasing the stringency of the statistical analysis via extension of the cut off threshold (Δ-value) we reduced the FDR to less than 2.5%. After such filtering, 751 probe sets were identified as showing meaningful differences in expression between *tko^25t^* mutant and wild-type males, and 353 probe sets in females (respectively approximately 4% and 2% of the array). 10.1371/journal.pone.0008549.t001 ###### Number of selected probes during filtering and statistical analysis. {#pone-0008549-t001-1} number of probe sets \% of total [a](#nt101){ref-type="table-fn"} --------------------------------- -------------------------------------- ---------------------------------------------- -------- -------- [Pre-filtering]{.ul} detection p-value \<0.05 10110 9778 53% 52% [Filtering (SAM analysis)]{.ul} Fold change (R)\>1.5 1248[b](#nt102){ref-type="table-fn"} 662[c](#nt103){ref-type="table-fn"} 7% 3% both R\>1.5 and FDR\<5% 947 413 5% 2% ** both: R\>1.5 and FDR\<2.5%** **751** **353** **4%** **2%** \% of probe sets in the array, to nearest whole number. GeneChip® Drosophila Genome 2.0 Array contains 18952 probe sets. FDR = 58%. FDR = 34%. To gain an overview of the biological meaning of the differences in gene expression, probe sets corresponding to known genes were classified into different functional categories and pathways according to their gene ontology (molecular function, biological process and cellular component) as listed in [Table S1](#pone.0008549.s002){ref-type="supplementary-material"}. We then compared the output gene lists separately for male and female flies, which identified the categories systematically up- or downregulated in a sex-dependent and sex-independent manner ([Table 2](#pone-0008549-t002){ref-type="table"} and [Figure S1](#pone.0008549.s008){ref-type="supplementary-material"}). In mutant males, approximately the same number of genes was upregulated as downregulated. However, in mutant females upregulation was 3-times more prevalent than downregulation. About 25% of the genes upregulated in males were also upregulated in females, which corresponded to 38% of those upregulated in females. Nevertheless, most upregulated genes in the two sexes fell into the same functional categories or pathways ([Table S2](#pone.0008549.s003){ref-type="supplementary-material"}). 10.1371/journal.pone.0008549.t002 ###### Coherence of changes in gene expression by sex. {#pone-0008549-t002-2} Regulated genes[a](#nt104){ref-type="table-fn"} \% of regulated genes[b](#nt105){ref-type="table-fn"} \% of total genes[c](#nt106){ref-type="table-fn"} --------------------- ------------------------------------------------- ------------------------------------------------------- --------------------------------------------------- male up (total) 404 54% 2.1% male down (total) 347 46% 1.8% female up (total) 268 76% 1.4% female down (total) 85 24% 0.4% both sexes up 102 14% (m), 29% (f) 0.54% both sexes down 32 4% (m), 9% (f) 0.17% male up female down 3 4‰ (m), 8‰ (f) 0.02% male down female up 1 1‰ (m), 3‰ (f) 0.01% Number of genes regulated in the directional manner shown. For a graphical illustration see . \% of the genes regulated in that sex. \% of probe sets in the array. GeneChip® Drosophila Genome 2.0 Array contains 18952 probe sets. Only about 9% of genes downregulated in males were also downregulated in mutant females, which nevertheless represented some 37% of those downregulated in females [Table 2](#pone-0008549-t002){ref-type="table"}, [Figure S1](#pone.0008549.s008){ref-type="supplementary-material"}). However, many of the downregulated genes were already expressed in a sex-specific manner, and linked to reproduction. Setting these aside, the alterations to gene expression in *tko^25t^* mutant flies were qualitatively similar in the two sexes. Very few genes were oppositely regulated in the two sexes (less than 1%). In the following sections we discuss the changes, classified according to biological process. The most highly up- or downregulated genes are listed separately in [Table 3](#pone-0008549-t003){ref-type="table"}, with full details by functional category in [Table S3](#pone.0008549.s004){ref-type="supplementary-material"}. In a few indicative cases we validated the changes in expression using quantitative RT-PCR. In the following sections, the tissue-specificity of expression is based on [www.flyatlas.org](http://www.flyatlas.org), plus other data cited in Flybase. 10.1371/journal.pone.0008549.t003 ###### Genes showing largest alterations[a](#nt107){ref-type="table-fn"} in expression in *tko^25t^*. {#pone-0008549-t003-3} Gene[b](#nt108){ref-type="table-fn"} Function FC (male)[c](#nt109){ref-type="table-fn"} FC (female)[c](#nt109){ref-type="table-fn"} Chromosomal localization -------------------------------------------- ------------------------------------------------------------------- ------------------------------------------- --------------------------------------------- -------------------------- ----- ------------ [Bisexually upregulated genes]{.ul} *l(2)03659* Mdr-related ABC transporter, xenobiotic clearance 43.0 11.7 45D1 *Fbp1* Lsp receptor, aminoacid/nutrient transport 17.6 81 (33.7) 159 70D2 *Fbp2* Lsp receptor, aminoacid/nutrient transport 16.5 (17.9) 30B3 *CG31775* unknown function 12.2 (19.2) 35B5 *Obp99b* odorant-binding lipohilic protein 13.6 2 (17.0) 17 99B8 *CG2650* lipohilic hormone-binding protein 12.5 (18.0) 3B2 *CG17192* gut-specific triacylglycerol lipase 9.6 17.9 97D14 *Cyp6a23* cytochrome P450, xenobiotic metabolism 10.9 9.2 51D1 *Tequila* serine protease 12.2 5.8 66F4 *CG11659* long-chain fatty acyl-CoA synthetase 6.7 5 9.4 36 92B2 *Hsp22* heat-shock protein 11.9 25 4.2 6 67B2 *CG3819* endonuclease 5.7 10.4 75E6 *vav* actin filament organization 14.1 (1.8) 18B6 *CG5999* glucuronosyltransferase, xenobiotic metabolism 10.9 (3.9) 87C8 *Lsp1α* aminoacid/other nutrient transport 7.2 (6.9) 11A12 *Cyp4e3* cytochrome P450, xenobiotic metabolism 9.2 (4.8) 30C7 *nimC2* unknown function 6.6 (7.2) 34E5 *Lsp1β* aminoacid/other nutrient transport 6.3 (6.6) 21E2 *CG33346* endonuclease 5.4 7.2 98E1 *CG12057* unknown function 6.8 5.2 8C17 *CG15088* sodium-dependent aminoacid transporter 4.6 7.3 55E10 *Lsp1γ* aminoacid/other nutrient transport 6.0 (5.6) 61A6 *Peritrophin-15b* gut-specific, chitin metabolism 1.9 8.4 29C1 *CG11893* unknown function, protein-binding properties 6.1 (4.2) 96C9 *p24-2* intracellular protein transport 3.7 6.4 85E4 *Ugt86Dd* glucuronosyltransferase, xenobiotic metabolism 5.4 3.7 86D4 *CG5966* triacylglycerol lipase 5.5 2.5 5D1 *Uro* urate oxidase (1.9) 5.7 28C3 *Jon25Bi* gut-specific serine protease (1.9) 5.7 25B4 *Ser6* serine protease 5.8 1.7 19E5 *CG13947* unknown function 3.4 3.4 21E2 *GstE1* glutathione-S-transferase, xenobiotic metabolism/clearance 4.7 2.0 55C6 *Lsp2* aminoacid/other nutrient transport 3.8 (2.7) 68F5 *unc-115* actin-binding protein 3.4 3.1 85E4 *lectin-28C* galactose-binding lectin 4.0 2.5 28D2 *CG13905* unknown function 3.7 2.8 61D4 *Cyp6a8* cytochrome P450, xenobiotic metabolism 3.3 3.1 51D1 *Jon25Bii* gut-specific serine protease (1.8) 4.5 25B4 *CG12780* gram-negative bacterial binding 3.1 3.0 44D2 *CG11796* 4-hydroxyphenylpyruvate dioxygenase (aminoacid catabolism) 2.3 3.8 77C3 *CG1062*1 selenocysteine methyltransferase (aminoacid catabolism) 2.7 3.1 37B7 *CG31809* steroid dehydrogenase 4.1 1.7 36B2 *Pepck* PEP carboxykinase (GTP) 2 1.8 nt 4.0 2,5 55D3 *Fst* positive regulator of fatty acid β-oxidation in response to cold 2.0 3.8 85E2 *ImpL3* lactate dehydrogenase 2.6 1.5 3.1 4 65A11 *Hsp23* heat-shock protein 3.4 2.1 67B3 *CG30016* Malpighian tubule-specific steroid carrier 2.6 2.8 47C5 *Cpn* calciphotin, calcium-binding, involved in eye development 2.3 3.1 87B1 *CG10592* alkaline phosphatase, skeletal development 1.8 3.6 64D5 *εTry* gut-specific serine protease 2.7 2.6 47F4 *CG30104* nucleotide phosphatase 2.6 2.7 54B17 *Peritrophin-15a* gut-specific, chitin metabolism 2.2 3.0 29C1 *CG5767* unknown function 1.8 3.3 55B1--55B2 *CG3285* sugar transport 2.8 2.2 23E4 *CG8942* tenascin, Wnt-signalling receptor-related 2.1 2.9 34E5 *Cyp28a5* cytochrome P450, xenobiotic metabolism 2.6 2.4 34E5 *Arc1* cofactor for tRNA synthetase 2.5 2.5 50F6 [Bisexually downregulated genes]{.ul} *RFeSP* Rieske iron-sulfur protein, OXPHOS complex III (isoform A) −27.3 −33.1 22A3 *gkt* tyrosyl-DNA phosophodiesterase (DNA repair) −5.4 −5.2 23D3 *HDC20470* intergenic region −6.1 −4.5 82A4 *dro4* ion channel inhibitor with direct antimicrobial effect −4.3 −4.3 63D1 *CG17478* [d](#nt110){ref-type="table-fn"} ovary-specific unknown protein-binding protein (unlocalized gene) −1.7 −6.8 41C1--41C6 *CG10924* PEP carboxykinase (GTP) 1 −3.7 −(4.5) 55D1 *TotX* humoral stress response protein −5.0 −(2.0) 93A3 *phr* deoxyribodipyrimidine photo-lyase (DNA repair) −3.4 −2.8 43E18 *path* aminoacids transporter −2.1 −4.1 67B10 *CG11314* mesoderm development −3.2 −2.7 100A3 *CG10659* unknown function −2.9 −2.7 38B1 *Cyp6t1* cytochrome P450, xenobiotic metabolism −3.4 −2.0 20A1 [Sex-specifically regulated genes]{.ul} *Sdic* sperm-specific dynein intermediate chain 8.0 n.c. 19C1 *takeout* lipohilic hormone-binding protein, behavioural regulator −2.1 −5 n.c. −2 96C7 *Gld* glucose oxidase/dehydrogenase −3.1 n.c. 84D3 *osk* pole cell development −8.8 n.c. 85B7 *CG12200* unknown function, protein-binding properties n.c. 15.7 18C7 *LysX* gut-specific lysozyme n.c. 7.0 61F3 *CG15533* sphingomyelin phosphodiesterase n.c. 6.4 99F4 *PGRP-SC1b* defense against Gram-positive bacteria n.c. 4.6 44E2 α*-Est10* carboxyesterase n.c. −3.3 84D8--84D9 *bcn92* mitochondrial-targeted, unknown function 2.2 −1.7 2D4 *Rala* small GTPase 1.6 −1.9 3E5--3E6 *fit* female-specific of *tra* −3.0 1.9 93F14 [Regulated transposable elements]{.ul} *Transposon.82* transposon 22.3 10.2 \-\-- *Transposon.11* transposon 13.3 13.2 \-\-- *Transposon.27* transposon 8.9 7.4 \-\-- *Transposon.17* transposon 6.6 (2.5) \-\-- *Transposon.42* transposon 2.7 2.4 \-\-- *Transposon.30* transposon −2.3 −5.9 \-\-- *Transposon.2* transposon −2.5 −3.2 \-\-- *Transposon.22* transposon −5.6 −4.2 \-\-- *Transposon.3* transposon 1.7 −2.0 \-\-- Average of \>2-fold change, both sexes considered, except in regard to genes with proven or probable sex-specific functions, where \>2-fold change in only one sex was sufficient for inclusion in this list. For full list of alterations in gene expression, including details of relevant Affymetrix probe sets, see [Table S3](#pone.0008549.s004){ref-type="supplementary-material"}. Excluding genes normally expressed only in the opposite sex from that in which regulation was observed, or genes tightly inducible by a defined stress, e.g. bacterial infection, and which are normally expressed at a very low level in both sexes. Fold change, i.e. proportionate increase from wild-type (positive numbers) or to wild-type (negative numbers), in each sex. In parenthesis, those regulated genes that were unselected by the statistical analysis with the threshold that we used. Unaffected genes are denoted as no change (n.c.). Data shown alongside from the Affymetrix array experiment correspond to Q-RT-PCR analyses, where performed nt--not tested. Gene model currently withdrawn, probe set detects an ovary-specific transcript (flyatlas.org) at 41C1--41C6, but full genomic sequence not identified. Changes in Gene Expression Related to Metabolism {#s2b} ------------------------------------------------ We observed systematically altered expression of genes concerned with energy metabolism, indicating a remodeling of metabolic pathways in response to the stress of mitochondrial OXPHOS insufficiency in *tko^25t^* mutant flies. Specifically, genes involved in the cytosolic reoxidation of NADH and in anaplerotic reactions feeding the TCA cycle, including amino acid and fatty acid catabolism, were upregulated, whereas those involved in conflicting pathways, notably fatty acid biosynthesis and the first steps of gluconeogenesis, were downregulated ([Table S3-a](#pone.0008549.s004){ref-type="supplementary-material"}). Although most of the changes in gene expression were quantitatively modest (typically 2-fold) the inferred pattern of global effects on metabolism is similar to that seen in yeast mutants with OXPHOS defects, via the so-called retrograde signalling pathway [@pone.0008549-Traven1], [@pone.0008549-Liu1]. We now consider in turn each of these inferred metabolic shifts, and the specific genes involved. Metabolic Shunts and Anaplerotic Pathways {#s2c} ----------------------------------------- At least two upregulated genes provide metabolic shunts for the regeneration of NAD^+^ from NADH, namely *ImpL3* (lactate dehydrogenase, which converts pyruvate to lactate), and *CG31674* ([Table S3-a](#pone.0008549.s004){ref-type="supplementary-material"}), the *Drosophila* orthologue of human glyoxylate reductase, which yields glycolate from glyoxylate. Pyruvate and glyoxylate may be considered major intermediate products of carbohydrate and amino-acid catabolism, respectively. Their diversion to essentially useless waste products (lactate and glycolate), which brings about the regeneration of NAD^+^, implies that carbon skeletons for biosynthesis normally derived from pyruvate or glyoxylate must be provided from other sources. This suggests a rationale for the upregulation of anaplerotic pathways and lipid/fatty acid catabolism. Another may be that, complex I being the component of the electron-transfer chain (ETC) quantitatively most affected in *tko^25t^* [@pone.0008549-Toivonen2], breakdown of fatty acids by beta-oxidation partially circumvents the problem by feeding proportionately more electrons than pyruvate to the ETC at the level of complex III than complex I. Unexpectedly, three enzymes of glycolysis, along with some other enzymes of glucose catabolism and transport, were down-regulated, but only in males ([Table S3-a](#pone.0008549.s004){ref-type="supplementary-material"}). However, in every case the down-regulated gene is the testis-specific isoform, with at least one other, much more highly expressed 'ubiquitous' isoform unchanged. These changes in gene expression most likely represent downregulation of the testis, and of reproductive functions in general, rather than being connected with any general transformation of metabolism. The two isoforms of PEP carboxykinase (GTP) were reciprocally regulated. PEP carboxykinase (GTP) 1 (CG10924, downregulated in males along with pyruvate carboxylase) is the *Drosophila* orthologue of human PCK2, predicted to be mitochondrial and predominantly expressed in the larval fat body. PEP carboxykinase (GTP) 2 (Pepck, upregulated and widely expressed, closest homologue of human cytosolic PCK1 and also probably cytosolic despite its annotation [@pone.0008549-Sardiello1]) is presumed to be the major anaplerotic source of oxaloacetate. The net effect is thus to activate a metabolic switch to spare the TCA cycle under conditions where pyruvate is mainly diverted to lactate. Lipid Metabolism {#s2d} ---------------- The two mRNAs for metabolic enzymes most dramatically upregulated in *tko^25t^* (CG17192, midgut-specific triacylglycerol lipase, [Table S3-e](#pone.0008549.s004){ref-type="supplementary-material"}, and CG11659, long-chain fatty acyl-CoA synthetase, most highly expressed in the Malpighian tubule, [Table S3-a](#pone.0008549.s004){ref-type="supplementary-material"}) both participate in the primary mobilization of dietary lipids. Another upregulated triacylglycerol lipase, CG5966 ([Table S3-e](#pone.0008549.s004){ref-type="supplementary-material"}), is widely expressed. Some components of fatty acid beta-oxidation were upregulated ([Tables S3-a](#pone.0008549.s004){ref-type="supplementary-material"}, [S4-e](#pone.0008549.s005){ref-type="supplementary-material"}), such as beta-ketothiolase (*yip2*) and the ETF-ubiquinone oxidoreductase (*CG12140*), whereas enzymes of fatty acid biosynthesis were generally downregulated (see [Tables S3-a and S3-e](#pone.0008549.s004){ref-type="supplementary-material"}). However, many of these changes were only scored as statistically significant in one sex, since they were generally close to the filtering threshold of 1.5 fold. TCA Cycle and OXPHOS {#s2e} -------------------- Genes for TCA cycle components were generally not scored as changed in expression after statistical filtering, However, when we looked at them in the unselected data, ([Table S4-a](#pone.0008549.s005){ref-type="supplementary-material"}) there was a discernable pattern. Testis-specific isoforms were downregulated (in males), whereas many ubiquitously expressed isoforms were slightly upregulated (although this was usually below the filtering threshold and/or only in one sex). Perhaps surprisingly, only three (out of \>50) nuclear-coded OXPHOS genes showed altered expression ([Table S3-a](#pone.0008549.s004){ref-type="supplementary-material"}). *CG10320 and CG33493*, encoding two subunits of complex I, were modestly regulated, but oppositely, and this was significant only in males. However, one of two mRNAs for *RFeSP*, encoding the Rieske iron-sulfur protein subunit of complex III, was downregulated more than 20-fold in both sexes. *RFeSP* is an essential gene [@pone.0008549-Spradling1], which generates two variant polypeptides by alternative splicing. The RFeSP-PB variant is more extensively homologous with the yeast orthologue Rip1p ([Figure S2](#pone.0008549.s009){ref-type="supplementary-material"}), whereas RFeSP-PA carries an unrelated C-terminus lacking some of the highly conserved Rieske domain. The *tko^25t^*-downregulated RFeSP-PA mRNA is normally expressed at approximately 20% of the level of RFeSP-PB mRNA, but in a similar tissue pattern. One possibility is that the former serves a regulatory role, e.g. in complex III assembly, although the exact reason for it being so strongly downregulated in *tko^25t^* is unclear. Protein and Amino Acid Metabolism {#s2f} --------------------------------- Many proteases and peptidases were induced in *tko^25t^*, some of them sex-specifically. Many are likely to be involved in the primary breakdown of dietary protein, since they are close homologues of gut-specific mammalian serine proteases such as trypsin and chymotrypsin [@pone.0008549-Ross1] and are mainly expressed in the *Drosophila* gut, e.g. εTry [@pone.0008549-Wang1], *Ser6* and at least ten members of the Jonah-family [@pone.0008549-Carlson1]. Upregulation of other proteases may serve a scavenging or recycling function, although that of Tequila ([Table S3-n](#pone.0008549.s004){ref-type="supplementary-material"}), *Drosophila* homologue of neurotrypsin [@pone.0008549-Didelot1], and elsewhere implicated in learning and memory [@pone.0008549-Molinari1], may be more connected with chitin metabolism. Released amino acids may provide an alternative source of carbon skeletons for biosynthesis, replacing pyruvate, less of which is entering the TCA cycle. The sodium-dependent amino-acid transporter CG15088 ([Table S3-f](#pone.0008549.s004){ref-type="supplementary-material"}), as well as a number of enzymes of amino acid catabolism ([Table S3-c](#pone.0008549.s004){ref-type="supplementary-material"}), were upregulated in one or both sexes. Some enzymes annotated as being involved in amino acid biosynthesis were also upregulated, although their precise metabolic roles are unclear, as are those of many other enzymes, which may function in diverse pathways ([Table S3-e](#pone.0008549.s004){ref-type="supplementary-material"}). This complexity supports the idea that the mobilization of dietary protein mainly serves an anaplerotic rather than a purely catabolic role. Transport {#s2g} --------- The expression of genes for diverse transport functions was modified in *tko^25t^* flies ([Tables S3-a and S3-f](#pone.0008549.s004){ref-type="supplementary-material"}), most of which were increases. Focusing on the changes which were consistent between the sexes, and which were quantitatively the most dramatic, the major changes affect members of three families of sugar transporters, each expressed mainly in the Malpighian tubule and gut, hence implicated in dietary absorption, resorption or excretion. *CG3285*, *CG15406*, *CG7882* and 5 other upregulated genes are related to the yeast hexose transporter family (e.g. Hxt13p, Hxt2p) and to similar transporters in other eukaryotes and bacteria. *CG4726*, *CG8791* and three other upregulated genes belong to a family of sugar-phosphate transporters related to human SLC17A5, implicated in sialic acid storage disease [@pone.0008549-Verheijen1]. *CG2196* and *CG8957* (plus three genes upregulated sex-specifically) belong to a family of ion transporters most closely related to the human sodium/glucose cotransporter SLC5A12. Two testis-specific sugar transporters, *Glut3* and *CG17637*, were downregulated. Many other *tko^25t^*-upregulated transporter genes are expressed mainly or exclusively in the Malpighian tubule. *CG16727*, *CG8654* and *CG17752* (plus several other upregulated genes) belong to a superfamily of organic cation transporters, mammalian members of which are involved in diverse functions, including carnitine uptake [@pone.0008549-Nezu1] and excretion of xenobiotics [@pone.0008549-Tamai1]. *CG8323* ([Table S3-a](#pone.0008549.s004){ref-type="supplementary-material"}) encodes a member of the mitochondrial inner membrane carrier family. Its yeast orthologue Oac1p transports oxaloacetate, sulfate and hyposulfite. Enhanced capacity for oxaloacetate transport into mitochondria is consistent with the inferred anaplerotic function of Pepck upregulation. *CG18327, CG5805* and *Bmcp* are closely related members of the mitochondrial carrier super-family, most likely with overlapping substrate specificities and similar functions. The changes in expression of genes connected with metabolism and transport in *tko^25t^* have features in common with those associated with other stress conditions, notably nutritional restriction [@pone.0008549-Pletcher1] or starvation [@pone.0008549-Zinke1], as well as normal aging [@pone.0008549-Pletcher1]. Under dietary restriction, many transport processes and extracellular functions, as well as fat body-specific genes and peptidases are upregulated. Some of the specific changes are seen also in *tko^25t^* flies, and many similar pathways seem to be affected. Furthermore, like *tko^25t^*, starvation induces the expression of genes involved in fat breakdown, fatty acid activation and beta-oxidation, as well as *Pepck*. Conversely, growth on sugar-rich diet rich induces a reciprocal pattern of changes, with up-regulation of biosynthetic genes for sugar to fat conversion, increased fatty acid anabolism and lipid biosynthesis, and downregulation of lipases. These findings suggest a common pathway of nutritional stress, involving signalling via one or a few key metabolites, and affecting genes for metabolic functions via a global response mechanism. Responses to Mitochondrial Stress {#s2h} --------------------------------- Other changes in gene expression appear to be specific to *tko^25t^*, suggesting a more direct response to failing mitochondrial protein synthesis. As shown in [Table S3-b](#pone.0008549.s004){ref-type="supplementary-material"}, genes for 12 mitoribosomal proteins, as well as a number of proteins involved in the processing of the mitochondrial translation products, were upregulated in *tko^25t^*. These include the mitochondrial prohibitin 2, *l(2)03709*, the *Drosophila* orthologue of the Rca1p m-AAA metalloprotease subunit, the mitochondrial deformylase CG31373, as well as a number of genes involved in the synthesis, mitochondrial import and processing of cytosolically synthesized proteins, notably chaperones Hsp10 (CG11267) and Hsc70-5 ([Table S3-h](#pone.0008549.s004){ref-type="supplementary-material"}). Strikingly, most of these changes were seen only in males. One possible explanation might be that many of the genes for mitochondrial biosynthetic components are highly expressed in ovary, being important in oogenesis, so that increased expression in somatic cells due to mitochondrial stress is not detected in females using the thresholds we employed. The gene for the heat-shock protein Hsp22 was strongly induced (4--10 fold) in *tko^25t^* ([Table S3-h](#pone.0008549.s004){ref-type="supplementary-material"}). The gene is also upregulated during aging [@pone.0008549-Kurapati1] and by oxidative stress. Mutations affecting *Hsp22* expression impair locomotor activity and result in decreased lifespan [@pone.0008549-Morrow1], whereas over-expression promotes resistance to oxidative stress and increases lifespan [@pone.0008549-Morrow2]. Another heat-shock protein of the lens alpha crystallin-related superfamily, Hsp23, was more modestly upregulated. A second group of stress-response genes upregulated in *tko^25t^* encode glutathione-S-transferases. At least 38 such genes are found in the *Drosophila* genome, of which 6 were significantly upregulated in either or both sexes in *tko^25t^* flies, although some others of them were repressed. These enzymes are required for the processing of oxidative adducts, such as peroxidated lipids, and their induction could be considered a signature of oxidative stress Three of them were previously found to be upregulated also by oxidative stress and/or in aging [@pone.0008549-Landis1]. Mostly the upregulated genes of this class are larval, tubule or gut specific, whereas the downregulated members are testis or head specific. One mitochondrial enzyme involved in iron-sulfur cluster assembly, cysteine desulfhydrase (CG12264, the *Drosophila* orthologue of yeast Nfs1p) was also upregulated. However, none of 25 genes arbitrarily selected from the NCBI GEO database, showing at least twofold induction by paraquat in *Drosophila* heads, were found to be also upregulated in *tko^25t^*. This indicates clearly that the pattern of changes in gene expression induced by severe oxidative stress is quite different from that seen in *tko^25t^*. It offers no support to the suggestion, based on other studies, that OXPHOS deficiency results systematically in ROS overproduction, and that the ensuing oxidative stress could constitute a common pathway of pathogenesis of mitochondrial dysfunction. Glutathione-S-transferases are also considered to be physiologically important for the processing of xenobiotics for detoxification and excretion. Many other differences in gene expression in *tko^25t^*, whether scored as stress-related responses ([Table S3-h](#pone.0008549.s004){ref-type="supplementary-material"}), transport functions ([Table S3-f](#pone.0008549.s004){ref-type="supplementary-material"}) or endosomal-related ([Table S3-i](#pone.0008549.s004){ref-type="supplementary-material"}), could serve this same purpose. Glucuronosyltransferases such as UGt86Dd and CG5999, P450 cytochromes such as Cyp6a8, Cyp4e3 and Cyp6a23, and transporters of xenobiotics related to the mammalian multidrug resistance (Mdr) family, such as l(2)03659, are amongst the most highly induced genes. A set of induced lysosomal class II (degradative) alpha-mannosidases (CG9466, CG9463 and CG9468, [Table S3-i](#pone.0008549.s004){ref-type="supplementary-material"}) is likely also be involved in xenobiotic clearance. Increased mobilization of potentially harmful or non-metabolizable compounds may be a secondary effect of increased primary mobilization and absorption of dietary components. The induction of the above proteins would constitute a line of defense against such compounds, leading to their exclusion, activation, conjugation, and eventual excretion. Another possible detoxification enzyme induced in *tko^25t^*, CG30022 ([Table S3-h](#pone.0008549.s004){ref-type="supplementary-material"}), is a mitochondrial member of the glyoxylase II superfamily and the *Drosophila* orthologue of *ETHE1*, the ethylmalonic encephalopathy disease-gene [@pone.0008549-Tiranti1], whose manifestations include cytochrome *c* oxidase deficiency in muscle. The physiological role of ETHE1 is in mitochondrial sulfide detoxification [@pone.0008549-Tiranti2]. Other members of the superfamily are implicated in clearance of metabolic by-products in cells with high glycolytic rates [@pone.0008549-Thornalley1]. Several genes related to the innate immune response were upregulated. To test whether this reflects a response to possible chronic infection of the *tko^25t^* stock with *Wolbachia*, which might also contribute to the abnormal reproductive behaviour of *tko^25t^* males [@pone.0008549-Bandi1] we performed PCR using *Wolbachia*-specific 16S rDNA primer pairs. This failed to detect any evidence of infection ([Figure 2](#pone-0008549-g002){ref-type="fig"}). A second possibility is that the induction of antimicrobial defense genes is due to activation of a common signalling pathway involved in stress responses. A third possibility is that metabolic stress may render the organism more susceptible to infection, and priming of key defense mechanisms may be advantageous to survival under such conditions (or could be a response to the actual proliferation of commensal bacteria). Most antimicrobial peptides are tightly inducible by pathogen challenge [@pone.0008549-Tzou1]--[@pone.0008549-Levashina1]. Their low level of expression in wild-type flies means that many of the largest changes (such as 8-fold upregulation of *Defensin* in males, 2-fold in females) did not pass the statistical thresholds. Some (e.g. dro4) were also down-regulated. {#pone-0008549-g002} Two *tko^25t^*-upregulated stress-related genes, *adipose* and *Frost* ([Table S3-h](#pone.0008549.s004){ref-type="supplementary-material"}), are involved in metabolic responses to other environmental stresses [@pone.0008549-Clark1], [@pone.0008549-Goto1]. The action of the adipose protein appears to be in the storage, transport and metabolization of triacylglycerol, [@pone.0008549-Hder1] which fits the general pattern of induction of genes involved in the dietary mobilization of fats. Mutants suffer hypertrophy of the fat body, due to the accumulation of fat droplets [@pone.0008549-Teague1], although the protein is widely expressed. Fst is expressed mainly in gut and tubule and its metabolic functions are not clear. It is also modulated by bacterial infection of the gut [@pone.0008549-Buchon1]. CG17734, upregulated in males (also in females by 40%, hence missed in filtering) is a homologue of the mammalian, mitochondrially localized HIG1 (hypoxia-inducible gene) family, which protects cells from apoptosis under conditions of hypoxia or glucose deprivation [@pone.0008549-Wang2]. Effects on DNA and RNA Metabolism {#s2i} --------------------------------- Systematic effects of *tko^25t^* on nucleic acid metabolism were relatively few ([Table S3-d](#pone.0008549.s004){ref-type="supplementary-material"}). There was a pronounced upregulation of gut-specific DNA endonucleases CG3819, CG33346 and CG6839 (females only), possibly indicative of the mobilization of additional dietary components. Two (out of the many) genes connected with DNA repair were downregulated in both sexes. *gkt* (*glaikit*, *Tdp1*) is annotated as tyrosyl-DNA phosphodiesterase, an enzyme involved in the resolution of 'dead-end' complexes\' between DNA and topoisomerase I [@pone.0008549-Pouliot1], as well as in other DNA repair pathways [@pone.0008549-Nitiss1]. However, *gkt* mutants also show neural phenotypes associated with deranged epithelial cell polarity [@pone.0008549-Dunlop1], proposed to be due to the lack of phospholipase activity of the gene product. *phr* (*photorepair*) is responsible for the repair of UV light-induced cyclobutane-type pyrimidine dimers [@pone.0008549-Yasui1]. The significance of these changes is unclear. Cell Cycle Regulation, Development, and Cell-Death {#s2j} -------------------------------------------------- Logically the developmental delay and reproductive phenotypes of *tko^25t^* should be reflected in subtle alterations in the expression of developmentally regulated genes, in particular those connected with metamorphosis, organogenesis, and reproduction. Many changes in gene expression indeed fell under these headings ([Tables S3-g](#pone.0008549.s004){ref-type="supplementary-material"}, [S3-j](#pone.0008549.s004){ref-type="supplementary-material"}, [S3-k](#pone.0008549.s004){ref-type="supplementary-material"} and [S4-b](#pone.0008549.s005){ref-type="supplementary-material"}). However, they are hard to interpret unambiguously, since the vast majority affected only one sex, and only in rather few cases were multiple genes contributing to a single organ, differentiation programme or physiological process clearly coregulated. There was substantial upregulation in *tko^25t^* of a set of genes expressed in the larval fat body, equivalent to the mammalian liver, and encoding the major larval serum proteins Lsp2 (3--4 fold) and all three subunits of Lsp1 (6--7 fold, [Tables S3-g](#pone.0008549.s004){ref-type="supplementary-material"}, [S4-b](#pone.0008549.s005){ref-type="supplementary-material"}). Upregulation was similar in males and females, but seems not to have passed statistical filtering in females, due to the very low expression in wild-type adults [@pone.0008549-Benes1]. The upregulation of *Fbp1* and *Fbp2*, considered as receptors for the larval serum proteins, was even more substantial (up to 2 orders of magnitude), and that of *Fbp1* was verified in both sexes by Q-RT-PCR. The upregulation of these genes may represent the persistence of larval gene expression connected with developmental delay. Their expression in *tko^25t^* adults is at least an order of magnitude less than in wild-type L3 larvae, when the genes are most highly expressed. These various proteins have been proposed to play roles in wound healing, nutrient transport, oxygen diffusion and immunity. Their stage-specific regulation [@pone.0008549-Deutsch1], [@pone.0008549-Burmester1] suggests that they provide a nutrient storage system during metamorphosis, involving resorption of the serum proteins into the fat body at L3 stage [@pone.0008549-Burmester1]. The circulating serum proteins may also serve a more general nutrient transport function in larvae and adults. Their upregulation in *tko^25t^*, as under dietary restriction [@pone.0008549-Pletcher1], might contribute to more efficient absorption of dietary components or clearance of xenobiotics, involving their transport to the fat body for detoxification and eventual excretion. Some unrelated serum protein genes were also upregulated, including *fat-spondin, Idgf5*, two monooxigenases and one endopeptidase (*CG3505*), implicated in clotting [@pone.0008549-Karlsson1] and cuticle formation ([Table S3-k](#pone.0008549.s004){ref-type="supplementary-material"}). Upregulation of genes connected with skeletogenesis [@pone.0008549-Arbeitman1], [@pone.0008549-Gagou1] ([Table S3-k](#pone.0008549.s004){ref-type="supplementary-material"}), including constituents of the cuticle, proteins involved in chitin metabolism, and several alkaline phosphatases, some of them gut-specific, may again be a signature of delayed development. Changes in the expression of some genes normally expressed only at very early developmental stages can probably be disregarded as quantitatively trivial ([Table S3-k](#pone.0008549.s004){ref-type="supplementary-material"}), although the repression of genes involved in sense organ development such as *Optix, mirr, phl, sdk, Magi, ana, Oseg1,Tig* and *nompB* (generally significant only in males), could be related to the sensory deficit seen in *tko^25t^* flies. The apoptosis and autophagy pathways were generally unaffected, perhaps surprising given the fact that nutrient deprivation induces autophagosome upregulation in many organisms [@pone.0008549-Boya1], [@pone.0008549-Lum1]. One obvious explanation would be that increased mobilization of dietary resources is sufficient to overcome the metabolic consequences of the mutation. Changes affecting cell division and cytoskeletal functions (e.g. *unc-115, vav*, [Table S3-j](#pone.0008549.s004){ref-type="supplementary-material"}) are hard to interpret. The impacted pathways are again similar to those induced by dietary restriction [@pone.0008549-Pletcher1], although many of the affected genes are different. The downregulation of histones and of genes involved in the cell division apparatus may be an indicator of decreased cell division in the germline. For example, the downregulated gene *piwi* ([Table S3-m](#pone.0008549.s004){ref-type="supplementary-material"}) promotes cell proliferation and differentiation in the female germline [@pone.0008549-Szakmary1]. The implied decrease in female gametogenesis was confirmed by measuring oviposition of *tko^25t^* females outbred into the Oregon R background, when mated to wild-type males. The total number of eggs laid by *tko^25t^* females over 5 days was approximately half the number laid by wild-type females ([Figure 3a](#pone-0008549-g003){ref-type="fig"}). {#pone-0008549-g003} Strikingly, about two-thirds of the many genes of unknown function downregulated specifically in males ([Table S3-o](#pone.0008549.s004){ref-type="supplementary-material"}) are expressed specifically in testis. This fits with the idea that mitochondrial dysfunction, perceived as a nutritional limitation, provokes a general shift of resources away from reproduction towards maintenance. Curiously, however, the same does not appear to be true of females, where rather few downregulated genes are ovary-specific. Most of the functionally unidentified genes downregulated only in females ([Table S3-o](#pone.0008549.s004){ref-type="supplementary-material"}) show widespread expression, whereas unidentified genes upregulated only in females are mainly gut-specific. Expression of Other Sex-Specific Genes {#s2k} -------------------------------------- Many of the genes regulated differently between the sexes in *tko^25t^* ([Tables 3](#pone-0008549-t003){ref-type="table"}, [S3-m](#pone.0008549.s004){ref-type="supplementary-material"}, [S4-c](#pone.0008549.s005){ref-type="supplementary-material"}) are already expressed sex-specifically, being putatively involved in gametogenesis, sex determination or reproductive behaviour. However, some appeared to be downregulated in the sex where they are not usually expressed, which can be considered to have little or no physiological meaning. In the sex determination pathway ([Figure S3](#pone.0008549.s010){ref-type="supplementary-material"}), both sexes showed evidence of feminization ([Table S3-m](#pone.0008549.s004){ref-type="supplementary-material"}): the female-specific *doublesex* transcript (dsx^F^) was upregulated in females, as was *fit* (*female-specific independent of transformer*), which may itself be regulated by the dsx^F^ product. Downregulation of *fit* in males is probably inconsequential, since it is normally expressed in males only at a low level. However, males also showed downregulation of several genes implicated in regulation of male-specific functions, notably *takeout* (*to*) and *sxe2* (*male sex-specific enzyme 2*). The product of *takeout* belongs to an insect-specific family of lipohilic ligand-binding proteins, the best characterized member of which is the juvenile hormone binding protein of *Manduca sexta*. They appear to influence various behaviours and developmental events, and are expressed in response to diverse signals. *takeout* expression shows cycling under the control of circadian regulators and is upregulated by starvation, to which *to^1^* mutant flies are hypersensitive [@pone.0008549-SarovBlat1]. *takeout* mRNA is expressed in a highly localized manner in structures within the gut and the antennae [@pone.0008549-SarovBlat1], as well as male-specifically in the adult fat-body. The protein is widely distributed, though its presence in the hemolymph is male-specific [@pone.0008549-Lazareva1]. It has been proposed to regulate both feeding and reproductive behaviour [@pone.0008549-Meunier1], and in males promotes (and is required for) courtship [@pone.0008549-Lazareva1], [@pone.0008549-Dauwalder1]. Its basal expression level is influenced by *dsx* [@pone.0008549-Dauwalder1] and by *fruitless* (*fru*), and the *to^1^* mutation results in behavioural feminization either in an outbred genetic background or in *fru* heterozygotes [@pone.0008549-Dauwalder1]. *takeout* downregulation thus provides a plausible explanation for the male courtship defect of *tko^25t^* males, which we verified in flies outbred into the Oregon R background ([Figure 3b](#pone-0008549-g003){ref-type="fig"}). In wild-type flies the circadian cycling of *takeout* mRNA exhibits a peak-to-trough ratio of approximately 5, and the decrease in *takeout* mRNA levels seen in *tko^25t^* males ([Figure 4](#pone-0008549-g004){ref-type="fig"}), may simply reflect a loss of this cycling. Since starvation induces *takeout* expression in flies not synchronized in a light-dark cycle [@pone.0008549-SarovBlat1], a disturbance in circadian cycling of *takeout* might stimulate food-seeking behaviour whilst suppressing male courtship. This makes biological sense, delaying reproduction under conditions of limited food resources. {ref-type="sec"}. Means±SD of three sample runs of each of three biological replicates are shown. Significance at the *p* value shown was computed using a *t*-test. See also [Table 3](#pone-0008549-t003){ref-type="table"}.](pone.0008549.g004){#pone-0008549-g004} *CG2650*, upregulated 12-fold in males ([Table S3-n](#pone.0008549.s004){ref-type="supplementary-material"}), and also in females (though excluded by statistical filtering), encodes an RNA highly expressed in the last stages of pupation under circadian control, and localized to the cuticle of the newly eclosed adults [@pone.0008549-Lorenz1]. Its level decays rapidly following eclosion, hence the upregulation in *tko^25t^* may be considered a further example of developmental delay. *CG2650* is a member of the same gene family as *takeout*. The 3′ untranslated portion of the *CG2650* mRNA overlaps that of the circadian regulator *per* by 60 nt, suggesting possible mutual regulation by RNA interference. One possibility is that *CG2650* upregulation disrupts expression of *per*, leading to the male-specific downregulation of *takeout*; another is that temporally altered expression of these putative hormone-binding proteins is induced by persistence of juvenile hormone. The odorant-binding protein gene *Obp99b* (*tsx*), normally expressed more highly in males than in females under the regulation of *dsx* [@pone.0008549-Fujii1], was highly upregulated in both sexes: although excluded by statistical filtering in females, Q-RT-PCR showed a much larger elevation in females than in males. This may represent an additional mechanism to attenuate reproduction under unfavorable conditions, since ectopic *Obp99b* expression in females negatively regulates receptivity [@pone.0008549-Fujii1], [@pone.0008549-Wolfner1]. Signaling {#s2l} --------- In this study we found several types of genes to be regulated in a systematic way, indicative of a signalling pathway that senses mitochondrial stress and generates specific transcriptional readouts (summarized in [Figure 5](#pone-0008549-g005){ref-type="fig"}). Few known molecules involved in signalling were themselves transcriptionally responsive in *tko^25t^*. To widen the search for relevant signalling pathways we looked also at their reported interaction partners and those of other plausible candidates. ![Summary of major alterations to gene expression and their proposed effects in *tko^25t^* flies.\ (a, b) Proposed metabolic effects, based on differences in gene expression affecting nutrition and metabolism between (a) wild-type and (b) *tko^25t^* flies. In wild-type flies glucose is metabolized via PEP to pyruvate, which is then fed to the TCA cycle mainly via the pyruvate dehydrogenase complex generating acetyl-CoA, with a small amount converted to oxaloacetate to replenish the TCA cycle intermediates as needed, maintaining a supply of carbon skeletons for biosynthesis. Surplus NADH is reoxidized via the ETC (complexes I, III and IV), generating potentially most of the cell\'s ATP needs at complex V. In *tko^25t^* flies, the maximal activity of the ETC complexes is only 10--20% that of wild-type flies [@pone.0008549-Toivonen2]. For simplicity, its greatly decreased contribution to NADH oxidation and ATP generation is omitted altogether in panel (b). Instead, the bulk of ATP must be supplied by glycolysis, with NADH reoxidation dependent on lactate dehydrogenase and similar shunts. Because pyruvate is, under such conditions, mainly shunted to lactate, the TCA cycle must be supplied from other sources, via the mobilization of dietary lipids, generating acetyl-CoA, PEP carboxykinase (I) diverting a small amount of PEP to oxaloacetate, and the mobilization of dietary protein and amino acid catabolism supplying these and other TCA cycle intermediates, as well as biosynthetic reactions directly. The modifications to metabolism in *tko^25t^* flies are accompanied (c) by altered expression of genes connected with nutrient breakdown, absorption and transport, plus xenobiotic handling, affecting mainly the gut, Malpighian tubule and fat body. In addition, there is downregulation or delayed expression of genes connected with gametogenesis and skeletogenesis, and, notably in males, altered expression of genes controlling circadian and courtship behaviour, interpretable as a biological response to poor nutritional conditions.](pone.0008549.g005){#pone-0008549-g005} The serine-threonine protein kinase *Akt1*, known to exhibit a plethora of developmental and physiological signaling functions [@pone.0008549-Hou1], [@pone.0008549-Kandel1], including responses to nutritional conditions [@pone.0008549-Kandel1], [@pone.0008549-Scanga1], was upregulated too modestly to pass statistical filtering. Of 161 gene entries for Akt1 in the DroID interactions database ([www.droidb.org](http://www.droidb.org)), only 9 were significantly *tko^25t^*-responsive, close to random expectation. Apart from the putative spliceosomal component CG13900, downregulated in females only, and nucleosome positioning protein Nlp, downregulated in both sexes, all were regulated in males only, but in either direction. Although CG13900 is also downregulated in larvae under starvation [@pone.0008549-Zinke1], the evidence that Akt1 is involved in signaling of mitochondrial stress and metabolic adaptation in *tko^25t^* is, at this point, purely speculative. *shaggy* (*sgg*), upregulated in *tko^25t^* males, encodes a serine-threonine protein kinase with a widespread expression pattern and involvement in many metabolic, behavioural and developmental processes [@pone.0008549-Ruel1], some of them relevant to the *tko^25t^* phenotype, including sense organ specification [@pone.0008549-Kanuka1] and circadian and courtship behaviour [@pone.0008549-Martinek1], [@pone.0008549-Wolf1], [@pone.0008549-Mackay1]. Its mammalian homologue, GSK3 [@pone.0008549-Kaufmann1] plays a key role in cellular signalling cascades, and interacts with PKB (homologue of *Akt1*) [@pone.0008549-Cross1], [@pone.0008549-Shaw1]. In *Drosophila*, the phosphorylation of tim by sgg promotes its translocation to the nucleus, driving the circadian pacemaker [@pone.0008549-Martinek1], and suggesting a pathway for the regulation of *takeout*, which is a known target of *tim*. A further interactor of *tim*, the circadian regulatory transcription factor *Clk*, was 2-fold upregulated in *tko^25t^* females, although this failed statistical filtering ([Table S4-d](#pone.0008549.s005){ref-type="supplementary-material"}). I-2 ([Table S3-l](#pone.0008549.s004){ref-type="supplementary-material"}), downregulated (but only significant in females), is a widely expressed protein phosphatase inhibitor. Inhibition of protein phosphorylation produces pleiotropic developmental phenotypes via the antagonization of proliferative signals [@pone.0008549-Bennett1], but there are no known links with nutritional or stress-response pathways. Two (of 14) reported interaction partners of I-2, Pp1-13C ([Table S3-l](#pone.0008549.s004){ref-type="supplementary-material"} and testis-specific) and sgg were regulated, but only in males, and in opposite directions. *Tsp42Ed*, a gene encoding a protein of the tetraspanin family was found to be upregulated in *tko^25t^*, but this also was significant only in females Tetraspanins are expressed in distinct tissue and developmental patterns and are functionally diverse, having roles in cell migration, signaling, cell fusion and adhesion [@pone.0008549-Todres1]. Tsp42Ed is highly expressed in tubule, gut and fat body, but its specific role in signaling, if any, is unknown. In yeast, the retrograde pathway depends upon three gene products without convincing structural homologues in metazoans. Rtg3p and Rtg1p are zinc-finger transcription factors distantly related to Mitf and Usf, but having quite different functions. The sensor protein Rtg2p has homologues in fungi but not beyond, and its ligand remains unidentified. It is distantly related to bacterial polyphosphatases, which are stress markers, especially of amino acid starvation [@pone.0008549-Stacpoole1]. Responses to amino acid starvation in both yeast and higher eukaryote involve the [T]{.ul}arget [O]{.ul}f [R]{.ul}apamycin (TOR) pathway [@pone.0008549-DeVirgilio1], with Akt as a downstream target. In yeast, the RTG and TOR pathways interact. No components of the *Drosophila* TOR complexes (*Tor*, *raptor*, *rictor*, *Sin1*, *CG3004*/*Lst8*,) were *tko^25t^*-responsive, nor were any of their known or predicted targets (*S6K*, *Thor*). We looked also at their interaction partners, but here again only 12 new targets of *tko^25t^*-regulation were identified, out of \>200 such proteins, and all such cases were predictions from yeast rather than genetically or biochemically proven interactions. Only one of them, CG10592, was significantly (up)regulated in both sexes. However, this and 5 others represent a coherent set of proteins involved in skeletogenesis. Neither CG9809 (Spargel), the *Drosophila* orthologue of PGC1α, the proposed global regulator of mitochondrial biogenesis in mammals [@pone.0008549-Lin1], nor its interaction partners CG15323 and CG7800, were *tko^25^* ^t^-responsive. However, 9 out of 142 predicted or known interaction partners of *SNF1A*, the *Drosophila* homologue of *AMPK*, implicated in the regulation of lipid metabolism in mammals, showed altered expression, and the overall readout of changes in fatty acid metabolism and Pepck (CG10924) [@pone.0008549-Lochhead1] suggests its possible involvement. Four of the 9 regulated genes are members of the cytochrome P450 superfamily, some of which are known to be regulated via AMPK in mammals [@pone.0008549-Rencurel1], [@pone.0008549-Rencurel2]. SNF1A itself was unaffected. Another possible candidate for involvement in retrograde signalling in *tko^25^* ^t^ is CG17734, a homologue of HIG1 [@pone.0008549-Wang3], which upregulates glycolysis and anaerobic metabolism in mammals, and which was induced almost 2-fold in males ([Table S3-h](#pone.0008549.s004){ref-type="supplementary-material"}). To identify possible hormonal signals mediating downstream metabolic responses, we searched the gene list for putative peptide hormones, peptidases and enzymes which could be involved in ecdysteroid or sesquiterpenoid metabolism, including those without clearly established physiological roles ([Table S5](#pone.0008549.s006){ref-type="supplementary-material"}). We considered only those which responded in both sexes, and were expressed specifically in the larval fat body and/or adult head (where the fat body is located). Excluding metabolic enzymes with no known roles in steroid or sesquiterpenoid biosynthesis, we narrowed the gene list to three candidates, none of which is compelling. Two of them encode proteins related to antimicrobial responses. *dro4* (drosomycin-4, 4-fold downregulated) is one of a cluster of seven putative antifungal peptide genes [@pone.0008549-Bischoff1], [@pone.0008549-DeGregorio1], also expressed at a substantial level under standard conditions. PGRP-SD (2-fold upregulated in males is involved in the recognition of gram-positive bacteria [@pone.0008549-Bischoff1]. One possibility is that it is targeted against bacteria which produce mitochondrial toxins, such as *Streptomyces antibioticus*, which secretes antimycin A. Upregulation of PGRP-SD may thus be directed against an unseen pathogen. TotX is a member of the Turandot family of humoral peptides [@pone.0008549-Ekengren1], induced by various stresses, and believed to mediate repair processes. It is strongly induced by bacterial infection or by paraquat [@pone.0008549-Ekengren1]. However, TotX is downregulated in *tko^25t^*, and the change was not scored as significant in females. In summary, despite some intriguing circumstantial evidence, we found no convincing data to support the involvement of any known intracellular or humoral pathway in mediating responses to mitochondrial stress in *tko^25t^*. Clearly a different experimental approach will be needed to reveal any such pathways and their components. Other Regulated Genes {#s2m} --------------------- Expression of a small number of genes implicated in behaviour, in addition to those directly involved in courtship, was found to be altered in *tko^25t^* ([Tables S3-n](#pone.0008549.s004){ref-type="supplementary-material"} and [S4-d](#pone.0008549.s005){ref-type="supplementary-material"}). These include *no extended memory* (*nemy*, mitochondrial glutaminase), whose upregulation might be part of the retrograde response to maintain glutamate levels anaplerotically, and *Rhythmically expressed gene 2* (*Reg-2*) a haloacid dehydrogenase, perhaps also with a metabolic function. Apart from *Obp99b*, the expression of several odorant-binding protein genes [@pone.0008549-Graham1] was modified sex-specifically, although the genes remain functionally unclassified in regard to ligands or downstream effects. Some may play roles in pheromone perception and mating behaviour, since many of the same genes were differently expressed in flies selected for fast *versus* slow latency in mating [@pone.0008549-Mackay1]. Some other genes picked up in the latter screen were also affected by *tko^25t^*, including *sgg*, *Clk*, *tim*, *Dat*, *elav* and *Hsp27* [@pone.0008549-Mackay1]. One possible explanation for the regulation of genes without obvious biological sense is that such genes may be coexpressed with nearby genes whose regulation is of functional importance, with which they share *cis*-regulatory elements. To investigate whether the altered expression of any genes in *tko^25t^* might be attributable to such an effect, we compared the chromosomal locations of the genes listed in [Table 3](#pone-0008549-t003){ref-type="table"}. Other than co-regulated, adjacent members of the same gene family (Hsp22/Hsp23, Jon25Bi/ii, Peritrophin-15a/b), plus *unc-115* and *p24-2*, which are adjacent at band 85E4 and possibly involved in the same cellular process, we saw no plausible examples of such opportunistic regulation. The list of most highly regulated genes ([Table 3](#pone-0008549-t003){ref-type="table"}) includes 11 genes with no annotated function. Of these, nimC2 (larval fat body-specific) is almost certainly involved in phagocytosis, like its homologues Eater, Draper and nimC1 [@pone.0008549-Kurucz1]. The putative gene product of *CG31775*, also larval fat body-specific, is a bizarre short polypeptide with oligoglutamine, glycine and alanine repeats. These make it impossible to identify homologues by conventional search programmes and it is far from certain that it is ever translated. *CG13947* (salivary gland and crop-specific) also has an internally repetitive structure and is practically impossible to parse for similar reasons. *CG12057* and *CG5767* (upregulated) and *CG10659* (downregulated) are all midgut-specific. The first two have homologues only in the Drosophilids, and their function is completely unknown, but CG10569, with homologus in diverse taxa, has an acyl-CoA N-acyltransferase domain, although its substrate(s) are unidentified. CG13905 (tubule-specific) is found in other insects and is induced by other stresses, notably by pathogens. CG12200, CG17478 (both ovary specific) are annotated as having protein-binding properties. The apparent regulation of the latter in males can be discounted. CG12200 (strongly upregulated) is a RING finger protein with weak homologues in many taxa and may be assumed to be involved in protein metabolism. CG11893, another upregulated gene suggested to have protein-binding properties and possible protein kinase function, is very weakly expressed in many tissues and its function cannot be extrapolated. Lastly, *bcn92* (*CG3717*), annotated as mitochondrially targeted, and a rare example of a protein significantly regulated between the sexes in a reciprocal manner (up in males, down in females) is related to proteins involved in iron-sulfur biogenesis (Isd11p in yeast) and complex I subunit NDUA6, and is probably the orthologue of the former. Finally, the *tko^25t^*-regulated gene list includes 12 transposable elements, 266 functionally unclassified genes (117 upregulated, 137 downregulated), as well as 44 Affymetrix probe-sets currently assigned to intergenic regions ([Table S3-o](#pone.0008549.s004){ref-type="supplementary-material"}). Only around 10% of them appeared to be similarly regulated in the two sexes. Congruence with Gene Expression in Other Models of Mitochondrial Disease {#s2n} ------------------------------------------------------------------------ Mitochondrial diseases in humans are physiologically and biochemically diverse, and *tko^25t^* should be considered a strict model only for those cases where there is a generalized defect of mitochondrial protein synthesis. However, this includes the common disorders associated with mtDNA deletions or tRNA point mutations, as well as an increasing number of nuclear genome-determined disorders. Studies of gene expression in *Drosophila* essentially track changes in the whole organism. This may bias the results in favour of genes expressed prominently in major organs, although regulation specific to the needs of a single tissue is less likely to be mistaken for a general response pathway. The latter is a particular problem when studying individual mammalian cell-lines, whose developmental origin or tumorigenic nature may bias the set of affected genes. For example, a recent analysis of gene expression changes associated with OXPHOS dysfunction in the 143B osteosarcoma cell background [@pone.0008549-vanWaveren1] identified a number of targets involved in invasiveness, notably upregulation of MMP1 and downregulation of its inhibitors. Although revealing an unexpected role for OXPHOS dysfunction in bone cancer, the wider relevance to OXPHOS dysfunction is questionable. Conversely, substantial changes in expression of tissue-specific genes in *tko^25t^*, e.g. those expressed in gut or Malpighian tubule, are almost certainly of global physiological significance, but would be invisible in cultured cell studies, unless a wide range of cell-types was studied. In order to minimize such problems, Alemi *et al.* [@pone.0008549-Alemi1] studied a variety of different cell types bearing pathological mtDNA deletions and looked for coherent patterns of altered gene expression. Consistent changes occurring in at least 4 of the 7 cell-types studied were considered meaningful. However, since only one *in vivo* tissue was included, the results remain strongly biased in favour of changes seen in cultured (cancer) cells, which typically revert to a highly glycolytic metabolism, relying on OXPHOS mainly for specific biosynthetic pathways such as pyrimidine biosynthesis, where the ETC performs a redox role rather than acts a source of ATP. Despite this, the genes most highly responsive to mitochondrial dysfunction were different from those identified in the 143B cell study [@pone.0008549-vanWaveren1], apart from the upregulation of the key autophagy gene ATG12, syntaxin-16 and of the Hsp70 family member HSPA4. In this light, it is not surprising that there is little congruence between our gene list and those of previous authors, or between those of previous authors when compared with each other. It must be kept in mind that the physiology of the whole organism is a co-operative enterprise between different organs with complementary functions, whose responses to mitochondrial dysfunction may also be reciprocal, generating a superficially contradictory picture when gene expression in the whole organism is tracked. We analysed the behaviour in the *tko^25t^* model ([Table S6-a](#pone.0008549.s007){ref-type="supplementary-material"}) of the closest *Drosophila* homologues of the 222 genes scored as showing consistently altered expression in 143B cells under conditions of OXPHOS deficiency [@pone.0008549-vanWaveren1]. Seven showed significantly altered expression in *tko^25t^*, close to random expectation, most of them only in one sex, or in the opposite direction to that seen in 143B cells. However, fewer than half of the total gene list has a convincing orthologue in *Drosophila*, which means that some of the congruent changes could be meaningful. Two of them showed indisputable orthology and similar regulation in all 143B cell models plus both sexes in *tko^25t^*: *CG3961* (*ACSL1* in human) and *unc-115* (*ABLIM1* in human), both of which were upregulated (although for *CG3961* this was significant only in males). *ACSL1* (*CG3961*, long-chain fatty acyl-CoA synthetase) is the key enzyme of fatty acid mobilization. Its upregulation in OXPHOS-deficient 143B cells may serve the same anaplerotic function we hypothesize in *tko^25t^*. *ABLIM1* (*unc-115*) encodes a cytoskeletal actin-binding protein [@pone.0008549-Roof1]. Its exact physiological functions are unclear. Its homologue in *C. elegans* regulates the formation of lamellipodia and filopodia in axonal guidance, and one isoform performs a similar role in *Drosophila*. However, *unc-115* is widely expressed, and may play a more general role in downstream signalling from Rac. One of its (few) predicted interaction partners, *drk*, is a signal transduction molecule involved in development of the peripheral nervous system [@pone.0008549-Okabe1], which may be relevant to the sensory defect of *tko^25t^*. Comparing gene expression alterations in *tko^25t^* with the study of mtDNA deletions in different cell types [@pone.0008549-Alemi1] we saw almost no overlap ([Table S6-b](#pone.0008549.s007){ref-type="supplementary-material"}). Many OXPHOS genes were downregulated in the deletion study, but few or none in *tko^25t^* or 143B cells. Unlike the deletion study, we saw no systematic downregulation of the ubiquitin/proteasome system, nor any general upregulation of vesicle transport, although we did see a strong upregulation of p24-2, involved in this process and expressed widely [@pone.0008549-Boltz1] but most prominently in the salivary gland. Apart from the modest downregulation of the NDUFB3 subunit of complex I in males (*CG10320*), and upregulation of a subunit of the vacuolar ATPase, the only congruent observation was upregulation of *GLS* (*nemy* in *Drosophila*), interpreted as part of a general enhancement in amino acid catabolism as in our own study. The involvement of glutaminase is intriguing, since glutamate has been postulated as a key intermediary metabolite sustained by the retrograde response in yeast [@pone.0008549-Liu1]. In the three studies there are distinct alterations in protein metabolism: the induction of gut-specific proteases and enzymes of amino acid catabolism in *tko^25^*, proteasome downregulation in cells with mtDNA deletions, and the activation of autophagy in OXPHOS-deficient 143B cells. All three may constitute a response to mitochondrial dysfunction perceived as amino acid starvation, although their physiology differs. Comparing our findings with the one substantial study of gene expression in an *in vivo* mammalian model of mitochondrial dysfunction [@pone.0008549-Hansson1] reveals both similarities and also some striking differences ([Table S6-c](#pone.0008549.s007){ref-type="supplementary-material"}). The latter study compared gene expression in hearts of mice in which a tissue-specific mtDNA depletion was induced by heart-specific knockout of the *Tfam* gene. In the intermediate and late stages of the resulting cardiac-disease there was a consistent shift in the pattern of expression of genes connected with metabolism. However, although many of the same genes and pathways were affected as in *tko^25t^* (TCA cycle and fatty acid catabolism), mostly they were oppositely regulated, and the change in expression in *tko^25t^* was so modest that it was rarely scored as significant in our filtering ([Tables S4-a](#pone.0008549.s005){ref-type="supplementary-material"}, [S4-e](#pone.0008549.s005){ref-type="supplementary-material"}). Some glycolytic enzymes were also upregulated in *Tfam*-knockout hearts, whereas we saw an increase only in LDH and other shunts for regeneration of NAD^+^. Similar to our findings, no substantial or systematic differences were seen in the expression of nuclear genes for OXPHOS subunits, whereas some genes involved in mitochondrial biogenesis were upregulated, including the mitochondrial chaperones Hspa9 and Hsp1 (Hsc70-5, CG11267) and a set of mitochondrial ribosomal proteins partially overlapping with those picked up in our own study. The most prominently upregulated gene in the *Tfam*-knockout heart, the mitochondrial Lon protease Lonp1 (*CG8978* in *Drosophila*), was unaffected in *tko^25t^*. A more detailed comparison with the *Tfam* knockout study is not possible, since Hansson et al. [@pone.0008549-Hansson1] did not publish their data relating to other genes than those discussed here. Signalling pathways implicated in the response to mitochondrial dysfunction appeared to differ in all four studies. The presence of mtDNA deletions induced up-regulation of the AMP-activated protein kinase (PRKAA) and its targets [@pone.0008549-vanWaveren1], [@pone.0008549-Prigione1], with cellular ATP depletion postulated as the primary signal. In 143B cells study about 20% of all genes affected were connected with signaling, although no single pathway was implicated. Many genes responding to *Tfam*-knockout are regulated by PPARα [@pone.0008549-Hansson1]. In *Drosophila*, we found only weak evidence implicating any signalling pathway, although Akt1/sgg and AMPK seem the best candidates. The differences in the *Tfam* and *tko^25t^* models may reflect the specific physiology of the heart which, in *Drosophila*, as the dorsal vessel [@pone.0008549-Molina1], [@pone.0008549-Tao1], accounts for only a tiny fraction of the total tissue mass. Another difference is that *Tfam* knockout progressively abolished OXPHOS function whereas it is only downregulated by *tko^25t^* [@pone.0008549-Toivonen2]. Finally, the changes in the *Tfam*-knockout heart might reflect tissue remodeling, rather than a shift in gene expression of a specific cell-type, as suggested by the gross anatomical changes seen in the failing heart: *tko^25t^* causes no major anatomical abnormalities. Implications Regarding Management of Mitochondrial Disease {#s2o} ---------------------------------------------------------- The primary response to mitochondrial OXPHOS dysfunction in widely different organisms (yeast, *Drosophila*, mammalian tissues and cell-lines) seems to be a remodeling of gene expression connected with metabolism. Surprisingly, many changes appear to be directed more at maintaining redox balance and the pool of carbon skeletons needed for biosynthesis, rather than at boosting ATP production. This contrasts with the more traditional view of OXPHOS as a machine for highly efficient ATP synthesis [@pone.0008549-Lehninger1] and emphasizes the long-standing observation that glycolysis is a preferred pathway in highly proliferative cells even when oxygen is present [@pone.0008549-Warburg1]. Upregulation of the HIG1 homologue CG17734 in *tko^25t^* may further enhance this effect. The overall response resembles one expected in cases of starvation. One way to mitigate the effects of pathological OXPHOS inhibition might thus be to provide nutritional supplements that alleviate the deficiency, in which case it is vital to determine exactly which metabolites are most affected. It may be, for example, that manipulating the balance between different types of nutrient might have more effect than changing their overall caloric amount. A second aspect is the use of alternate pathways for NADH re-oxidation which, in *tko^25t^*, as in mitochondrial disease patients, involves an upregulation of the shunt to lactate. The use of dichloroacetate (DCA), an activator of PDC, as a therapy for lactic acidosis [@pone.0008549-Stacpoole2] may thus be questioned, if the lactate shunt is essential for maintaining redox homeostasis. Whilst promoting a drop in circulating lactate levels, DCA does not appear to affect the major clinical features of disease [@pone.0008549-Stacpoole2]. Diverting pyruvate back into the TCA cycle will not alleviate the block on NADH reoxidation, and may even be harmful [@pone.0008549-Kaufmann1], if entraining increased mitochondrial ROS production. Alternative strategies for reoxidizing NADH, such as the use of a metabolic by-pass [@pone.0008549-Yagi1], might be more fruitful. In plants, a major target of the retrograde response is the induction of the alternative oxidase (AOX) and the non-proton-pumping NADH dehydrogenases, which provide a substitute pathway to maintain redox balance and mitochondrial intermediary metabolism under conditions of OXPHOS inhibition. Such conditions occur naturally in plants, due to the fact that, in the light, photosynthesis maintains ATP at a high level, which limits mitochondrial OXPHOS by coupling. Expression of the by-pass genes unblocks NADH reoxidation and the operation of the TCA cycle to serve the needs of biosynthesis. Increased intramitochondrial ROS production has been postulated as a primary inducing signal of retrograde responses in plants [@pone.0008549-Rhoads1]. A ketogenic diet has also been proposed to alleviate some symptoms of mitochondrial diseases [@pone.0008549-Stacpoole2], [@pone.0008549-Yagi1], especially where oxidative stress is a key pathological mechanism [@pone.0008549-Zhao1]. In our study we did not find evidence for chronic oxidative stress being a generalized consequence of OXPHOS dysfunction in *tko^25t^*. However, dietary supplementation by specific fatty acids and/or specific amino acids might bring benefits for the reasons already discussed. The elimination of carbohydrates from the diet might, conversely, have deleterious or even catastrophic consequences, since glucose as a fuel seems even more essential under conditions of OXPHOS limitation. We would suggest that supplementation and modulation, rather than complete elimination of glucose might be more effective. Experimental Approaches to Retrograde Signalling {#s2p} ------------------------------------------------ In plants, the signalling molecules and pathways involved in retrograde regulation have been studied using an AOX-luciferase reporter gene in *Arabidopsis* cell culture [@pone.0008549-Zarkovic1]. Over 100 independent mutants were isolated, altered both in global and more stress-specific responses. Although retrograde regulation may be more complex in plants, given the presence of two semi-autonomous organelles, some of the underlying signalling machinery may be common to plants and animals. The identification of some specific target genes in the present study offers the prospect of using a similar reporter strategy, combined with the genetic resources of *Drosophila*, to identify both the cell-autonomous and hormonal signalling pathways of retrograde regulation in a metazoan, with obvious possible relevance to finding new drug targets and other therapies for mitochondrial disease. It will also be important to check that the widespread transcriptional regulation documented in the present study is reflected also at the protein level. Other target genes in the affected pathways, as well as components of the signaling machinery that brings about the adaptive changes, may turn out to be regulated translationally or post-translationally, rather than at the RNA level, to unravel this will require an extensive further analysis using proteomic tools. Materials and Methods {#s3} ===================== *Drosophila* Stocks, Maintenance, and Crosses {#s3a} --------------------------------------------- Wild type and *tko^25t^* mutant flies in both Canton S and Oregon R backgrounds were maintained in standard oatmeal and molasses medium containing 1.5% (w/v) sucrose (Merck), 3% glucose (Sigma), 3.5% Instant Dry Baker\'s Yeast (European), 1.5% Maize flour (Oriola), 1% Wheat germ (Oriola), 1% Soya flour (Oriola), 1% agar (Oriola) and 3% Lyle\'s black treacle (Tate & Lyle, UK), to which was added 0.1% Nipagin M (Sigma) and 0.5% (v/v) propionic acid (JT Baker). Crosses were set up as indicated in [Figure 1](#pone-0008549-g001){ref-type="fig"}. RNA Extraction {#s3b} -------------- Three independent groups of 30 *tko^25t^* virgin females and another 3 independent groups of 30 virgin wild-type females were collected and aged to 1 day prior to RNA extraction. In parallel, 3 independent groups of 40 *tko^25t^* males and another 3 of wild-type males were collected and aged to 1 day. Each group of flies was homogenized independently in 1 ml of Trizol (Invitrogen) and RNA was extracted with chloroform and precipitated with 2-propanol by centrifugation at 12000 *g~max~* for 10 min at 4°C. After washing with 75% EtOH, RNA was resuspended in DEPC-treated water and cleaned using the RNeasy® MinElute™ Cleanup kit (Qiagen). RNA concentration and purity were checked spectrophotometrically and RNA quality was verified electrophoretically. RNA was stored at −20°C in RNase-free water. Probe Synthesis and Hybridization {#s3c} --------------------------------- A cRNA probe was synthesized and fragmented from each RNA sample using Affymetrix® protocols and kits provided by Qiagen. Each cRNA probe was then hybridized to independent GeneChip® Drosophila Genome 2.0 Arrays (Affymetrix). Hybridization was performed overnight in the GeneChip® Hybridization Oven 640 at 45°C with shaking at 60 rpm; washes and staining of the probe were performed in the GeneChip® fluidics station using buffers and protocols provided by the manufacturer; and the final high-quality scan of the arrays was performed with the GeneChip® Scanner 3000. The full system was controlled by the GeneChip® Operating Software (GCOS), and both the equipment and the software were provided by Affymetrix®. Statistical Analysis {#s3d} -------------------- Data extraction, cell intensity calculation and computational analysis were performed using GeneChip® Operating Software and Expression Console Release Software from Affymetrix. For each pair-wise comparative analysis, the signal value (MAS5 and RMA algorithms, at linear scale) of each probe was compared as follows: wild-type to *tko^25t^* in the both sexes. The statistics was performed with Microsoft® Office Excel 2003. Firstly, the data were prefiltered according to their detection p-value, using those signal-probes with p-value\<0.05. Secondly, SAM (Significance Analysis of Microarrays, Stanford Tools) was performed as described by Tusher *et al.* [@pone.0008549-Tusher1] using s0 = 20 and minimum full change (R) = 1.5 as fixed parameters, and Δ-value so that False Discovery Rate (FDR)\<2.5%. Finally, regulated probes in both male and female *tko^25t^*--wild type analysis were selected and classified according to their gene ontology. The data have been deposited with the NCBI GEO database (Series record GSE10169). Quantitative RT-PCR Analysis of mRNA {#s3e} ------------------------------------ RNA was reverse transcribed using random hexamers and M-Mul V Reverse Transcriptase (Fermentas). Real-time PCR was run in triplicate on a Light Cycler (Roche) using the SYBR Green PCR kit (Qiagen) and primers as detailed in [File S1](#pone.0008549.s001){ref-type="supplementary-material"}. Parallel reactions for *RpL32* (rp49) were used as a standardization control. The raw fluorescence data were extracted using Light Cycler data collection software version 3.5 (Roche) and analyzed according to manufacturer\'s instructions for baseline adjustment and noise reduction. Courtship and Egg-Laying Analyses {#s3f} --------------------------------- Males and virgin females were collected after eclosion, placed in separate vials (10 animals/vial) and stored over five days, with transfer to fresh food-containing vials every second day. For courtship analysis, single males and females were placed together in a food-containing vial, and the proportion of successful copulation was observed over 1 h. At least 15 females were observed per experiment. To assay oviposition, single mated females were transferred daily to fresh food-containing vials and the number of eggs laid by individual females was counted in each 24 h period. Between 15 and 20 mated females were included in each experiment. Each experiment was repeated three times. PCR Test for *Wolbachia* Infection {#s3g} ---------------------------------- Total DNA was extracted from different *Drosophila* stocks (see [Figure 2](#pone-0008549-g002){ref-type="fig"}) as previously [@pone.0008549-Toivonen2] and amplified with the following primer pairs and amplification conditions: Wolbachia 16S rDNA-F, 5′- TTGTAGCCTGCTATGGTATAACT-3′, Wolbachia 16S rDNA-R, 5′- GAATAGGTATGATTTTCATGT-3′, denaturation 95°C 1 min, annealing 52°C 1 min, extension 72°C 1 min, 35 cycles; mitochondrial 12S rDNA-F, 5′- TTTGGCGGTATTTTAGTAT - 3′, mitochondrial 12S rDNA-R, 5′- CTTAAATATAAGCTACACCTTGATC - 3′, denaturation 95°C 45 s, annealing 56°C 45 s, extension 72°C 1 min, 35 cycles, after which samples were mixed and analysed by 1% agarose gel electrophoresis. Supporting Information {#s4} ====================== ###### Supplementary text including additional information on materials and methods. (0.01 MB PDF) ###### Click here for additional data file. ###### Gene ontology categories used to classify the selected genes according to their biological process, molecular function, and cellular component. (0.03 MB XLS) ###### Click here for additional data file. ###### Regulated genes in *tko^25t^* mutant flies. (0.03 MB XLS) ###### Click here for additional data file. ###### Changes in gene expression according to functional category. (0.17 MB XLS) ###### Click here for additional data file. ###### Fold-change (calculated as significance analysis of microarrays) of unselected genes. (0.04 MB XLS) ###### Click here for additional data file. ###### *tko^25t^*-regulated genes with possible roles in hormonal signalling of mitochondrial dysfunction. (0.02 MB XLS) ###### Click here for additional data file. ###### Changes in expression in *tko^25t^* of homologues of genes modulated in other oxidative phosphorylation disease models. (0.14 MB XLS) ###### Click here for additional data file. ###### Congruence in patterns of changes in gene expression in *tko^25t^* flies. In each field are denoted the number of changes in the stated directions at the intermediate filtering stringency condition (see [Tables 1](#pone-0008549-t001){ref-type="table"} and [2](#pone-0008549-t002){ref-type="table"}). The relative numbers of changes in each category are shown in the boxes, and denoted by color intensity, from red to pale yellow. A higher proportion of down-regulated genes than up-regulated genes are altered sex-specifically, but many of these are already expressed in a sex-specific manner. (0.04 MB PDF) ###### Click here for additional data file. ###### Aligned amino acid sequences of the Rieske iron-sulfur protein variants. The two *Drosophila melanogaster* variant proteins, RFeSP-PA (NCBI database accession number AAF51353, blue) and RFeSP-PB (NCBI database accession number AAF51354, red) are shown aligned with the sequence of *Saccharomyces cerevisiae* Rip1p (NCBI database accession number NP_010890, black), using the one-letter amino acid code. Identical amino acids are boxed in pale blue. RFeSP-PB is homologous with Rip1p throughout its length, whereas the carboxy-terminal one-third of RFe-SP-PA is unrelated. The point of divergence with RFeSP-PB is arrowed, and the Rieske domain (<http://www.ncbi.nlm.nih.gov/Structure/cdd/cddsrv.cgi?uid=58540>) is underlined. (0.00 MB PDF) ###### Click here for additional data file. ###### Sex determination hierarchy in *Drosophila*. (0.01 MB PDF) ###### Click here for additional data file. We thank Tea Tuomela and Outi Kurronen for technical assistance. **Competing Interests:**The authors have declared that no competing interests exist. **Funding:**This work was funded by Academy of Finland 118654, 116056, and 119553; Sigrid Juselius Foundation; Tampere University Hospital Medical Research Fund 9F021; European Union LSHM-CT-2004-503116; and European Research Council 232738. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. [^1]: Conceived and designed the experiments: DJMFA KMCO HTJ. Performed the experiments: DJMFA SC EK. Analyzed the data: DJMFA SC EK. Contributed reagents/materials/analysis tools: KMCO. Wrote the paper: DJMFA KMCO HTJ. [^2]: Current address: Centro Andaluz de Biología del Desarrollo, Universidad Pablo Olavide, Seville, Spain

Introduction ============ The mammary gland is an interesting model for studying gene expression since this organ experiences different cycles of differentiation and regression during adult life. In eukaryotes, gene expression is a complex process that involves DNA methylation, chromatin modification, imprinting, and interfering RNA ([@b45-gmb-36-465]). Although Fire and Mello received the Nobel Prize for their discovery of double-strand interfering RNA in *Caenorhabditis elegans* ([@b17-gmb-36-465]), the first miRNA (Lin 4) was in fact discovered years before by two groups and published simultaneously in 1993 ([@b39-gmb-36-465]; [@b60-gmb-36-465]). In 2000, a second miRNA (Lin 7) was identified in the same nematode ([@b53-gmb-36-465]) and soon after in many other species ([@b50-gmb-36-465]). By 2001, it was clear that miRNAs represent a class of small conserved RNAs ([@b33-gmb-36-465]; [@b34-gmb-36-465]; [@b38-gmb-36-465]). Three main categories of interfering RNAs have been recognized: microRNA (miRNA), short interfering RNA (siRNA) and piwi interactive RNA (piRNA). These three RNAs are of similar size (∼23 nucleotides in length) but have sequence-specific inhibitory functions ([@b10-gmb-36-465]; [@b4-gmb-36-465]). siRNAs are excised from long, fully complementary double-stranded RNAs ([@b57-gmb-36-465]; [@b4-gmb-36-465]) and were initially considered to protect the host genome from foreign nucleic acids (such as viruses, transposons and transgenes) but that view changed with the discovery of abundantly expressed endogenous siRNAs in animal cells ([@b21-gmb-36-465]). piRNAs are small RNAs that bind the piwi subfamily of argonaute proteins and protect the germline ([@b3-gmb-36-465]) and embryonic stem cells ([@b47-gmb-36-465]) from invasive transposable elements. miRNAs regulate gene expression at a post-transcriptional level either by causing RNA degradation or by blocking translation through base-pairing with complementary sequences within mRNA. Since partial miRNA complementarities are enough to target an mRNA, each miRNA has the ability to regulate a large number of genes ([@b28-gmb-36-465]). Another feature is that a conserved miRNA can regulate different genetic pathways and developmental processes in various organisms ([@b3-gmb-36-465]). In recent years, many miRNAs have been identified in plants and animals by experimental and computational approaches. Adequate characterization of the actions of miRNAs in the mammary gland in different physiological states could provide new insights into the regulation of gene expression. The goals of this review are to outline and integrate the experimental and computational studies on miRNA in healthy and mastitic mammary gland and to address the stability of miRNA in milk and possible inter-species transmission. Biogenesis of miRNA =================== miRNAs are DNA-derived RNA transcripts that are not translated into proteins. miRNA biosynthesis is a complex process that involves several steps. The canonical pathway for miRNA biogenesis requires two RNaseIII enzymes, Drosha and Dicer. In this process, miRNAs are initially transcribed by RNA polymerase II as long transcript hairpin-shaped units. As with mRNA, these molecules are spliced, capped and polyadenylated to produce primary miRNA (pri-miRNA). This precursor is processed by the enzyme Drosha to yield pre-miRNA that is then transported to the cytoplasm where it is cleaved by Dicer to the mature length. The functional strand of mature miRNA is loaded onto an argonaute (Ago) protein, the main constituent of the RNA-induced silencing complex (RISC), while the other strand is degraded ([@b40-gmb-36-465]; [@b16-gmb-36-465]; [@b56-gmb-36-465]) ([Figure 1](#f1-gmb-36-465){ref-type="fig"}). [@b13-gmb-36-465] used Dicer- and Drosha-knock-out mice to confirm the requirement of both enzymes for this canonical miRNA biosynthesis. However, not all miRNAs are formed by this pathway. It is now known that there are variations in many of the steps involved and that this can influence the biogenesis results ([@b49-gmb-36-465]; [@b61-gmb-36-465]). Deep sequencing technology has revealed marked variability in miRNA biogenesis and has shown that many different sequences can share the same miRNA precursor. Identification of miRNA in mammary gland and milk ================================================= miRNAs have been identified in cells and in fluids such as saliva, amniotic fluid, blood, urine and milk. In fluids, miRNAs are enclosed in exosomes ([@b66-gmb-36-465]). The number of miRNAs reported for mammary gland varies considerably ([Table 1](#t1-gmb-36-465){ref-type="table"}). [@b12-gmb-36-465] used high-throughput sequencing technology to search for expressed small RNA in cow colostrum and milk. After classifying the molecules based on size, they found a similar number of reads in both fluids (1,594,965 in colostrum vs 1,418,136 in milk). A search in the miRBase (a database of published miRNA sequences and annotation) resulted in the identification of 230 and 213 known miRNAs in colostrum and mature milk, respectively. Among the differently expressed miRNAs, 108 were up-regulated and only eight were down-regulated in colostrum compared to mature milk. These results indicate a dynamic gene expression during lactation. [@b26-gmb-36-465] identified a smaller number of miRNAs than [@b12-gmb-36-465], with 100 being unique for colostrum and 53 for mature milk. It should be noted that whereas [@b12-gmb-36-465] prepared RNA from a larger amount of milk and purified small RNA by PAGE, [@b26-gmb-36-465] started with a smaller amount of milk and used microarray analysis. [@b44-gmb-36-465] reported 11,964,909 and 15,968,116 clean reads from mammary gland tissues in the lactating and non-lactating periods, respectively. After aligning the reads against the *Bos taurus* genome and miRBase, 885 pre-miRNAs were identified and encoded for 921 miRNAs; ∼60% of these miRNAs were expressed during the lactation and non-lactation period; of these, 248 miRNAs were known, 57 were conserved and 239 were new identifications. In human milk, [@b66-gmb-36-465] identified exosomes containing pre-miRNA from four libraries. Four of the top ten miRNAs (30b, 182, 200a and 148a) were related to different aspects of the immune system, with the most abundant of them being miRNA 148a, which is also expressed in bovine milk ([@b12-gmb-36-465]). In a sheep mammary gland library constructed from early pregnancy tissue, [@b18-gmb-36-465] identified 54 sequences already described in the miRBase; two of these miRNAs (27e and 36e) were identified for the first time in mammals and only miRNA 379 had previously been shown to occur in sheep. In goats, [@b29-gmb-36-465] characterized miRNAs from a pool of five Laoshan breed animals. A total of 18,031,615 read sequences were obtained after discarding ∼2.6% that did not meet the control criteria (*i.e.*, they were of low quality, had contaminants formed by adapter-adapter ligation, and contained reads without insert tags). Of these reads, 9,093,530 had a perfect match to the *Ovis aries* genome and 305,711 were new sequences. Furthermore, 290 conserved miRNAs and 38 novel miRNAs were identified, and this total of 338 miRNAs was very similar to that reported by [@b43-gmb-36-465] for dairy goats (441 miRNAs). The biological material (mammary gland tissue or milk), the amount of sample used, and the species and breed examined can influence the results. Moreover, differences in the quality control criteria used to classify the sequencing products (raw and clean reads) and discrepancies associated with library normalization may also contribute to variations in the number of miRNAs reported. Finally, some degree of variation may be introduced by the mode of sample preparation. [@b37-gmb-36-465] observed no changes in the 3′ end of a 22-nucleotide-long synthetic RNA introduced into RNA samples, but a fraction of the synthetic sequences were truncated at the 5′ end. These authors concluded that the changes in the 5′ end may have been caused by premature termination during the production of synthetic RNA. This finding suggests that at least part of the variation reported for end-region sequences may be linked to sample processing prior to sequence analysis. An additional source of diversity is that many miRNAs may vary from the published reference sequences. To address the latter phenomenon, [@b47-gmb-36-465] proposed the terminology "isomers" to refer to sets of miRNAs that show similarity in their sequences. These authors suggested that isomer variability could be related to variation in the cleavage positions for the enzymes Dicer or Drosha within the pre-miRNA hairpin and showed results in which the variability among isomers influenced the differential expression of miRNA. Functions of miRNA in mammary gland and milk ============================================ An important question in assessing miRNA function is whether the miRNAs present in milk are derived from blood or are specific for mammary gland. To address this issue, [@b12-gmb-36-465] compared the miRNA profile of milk with that of serum from healthy cows and found that the total number of miRNAs in milk was about two-fold higher than in serum; they also identified 47 miRNAs unique to milk. Human breast milk also has a different pattern of miRNA expression compared to blood plasma ([@b32-gmb-36-465]). These results clearly indicate that mammary alveolar cells express their own miRNAs. Another question is whether the pattern of miRNA expression in mammary gland is constant throughout the lactation period. Of the 1,692,810 reads described by [@b44-gmb-36-465], 34% were expressed only in the dry period compared to the peak period of milk production. Moreover, analysis of the expression patterns of 173 differentially expressed miRNAs showed that 165 were down-regulated during peak lactation compared to the dry period. Among the sequences reported by [@b44-gmb-36-465], 56 showed significant differences in expression between lactating and non-lactating cows, as assessed using the IDEG6 package ([@b54-gmb-36-465]); of these, nine were expressed only in lactating animals and six in non-lactating animals. However, 48 of these were confirmed by deep sequencing ([@b44-gmb-36-465]), indicating that deep sequencing may be more sensitive and reliable than microarray analysis in identifying differentially expressed miRNAs. Together, these findings indicate that the pattern of miRNA expression varies according to the animals physiological state. To examine the expression of specific miRNAs associated with cellular proliferation, metabolism and the innate immune response during lactation, [@b59-gmb-36-465] assessed the expression of 13 miRNAs in cows during the dry period (30 d prepartum), fresh period (7 d postpartum) and early lactation (30 d postpartum). Twelve of the miRNAs identified (miRNAs 10a, 15b, 16, 21, 33b, 145, 146b, 155, 181a, 205, 221 and 223) were down-regulated in the dry period compared to during lactation. The exception was miRNA 31, which showed greater expression in early lactation compared to the dry period. Under normal conditions and using bioinformatic assays and biological experiments, [@b63-gmb-36-465] demonstrated that miRNA 31 up-regulated IL-2 (interleukin 2) expression by reducing the levels of the cytokines upstream kinase suppressor, KSR2 (kinase suppressor of ras 2). Interleukins are present in human milk ([@b9-gmb-36-465]) and have an important role in modulating the offsprings immunological system ([@b7-gmb-36-465]). These findings indicate that miRNA 31 may have an indirect immunological role in the neonate. Based on microarray analysis, [@b18-gmb-36-465] identified three major patterns of miRNA expression in sheep. In pattern 1, expression was down-regulated during pregnancy, in pattern 2, miRNA expression was induced during pregnancy, and in pattern 3, miRNA expression was induced during lactation. The authors selected one miRNA from each pattern to confirm their expression by RT-qPCR in four animals per pattern. miRNA 21, which is expressed in alveolar epithelial cells, was up-regulated in non-pregnant sheep and at the beginning of pregnancy. This expression profile was attributed to a role for miRNA 21 in adipogenic differentiation. In this regard, [@b31-gmb-36-465] showed that the activity of miRNA 21 in adipogenic tissue was mediated through TGF-β signaling. In contrast, miRNA 205 was expressed mainly in the basement membrane of normal mammary ducts and lobules during the first half of pregnancy and miRNA 200 was expressed in epithelial cells throughout pregnancy but was up-regulated at the end of pregnancy and lactation. Bioinformatic analysis using the program RNA hybrid identified a miRNA 15a target sequence on the growth hormone receptor ([@b41-gmb-36-465]). However, this miRNA was not identified in a sequence search by [@b12-gmb-36-465]. To confirm their finding, [@b41-gmb-36-465] transfected this small RNA in mammary epithelial cells and observed a reduction in growth hormone receptor (GHR) transcription and in the expression of β casein. Growth hormone is the most relevant galactopoietic hormone in ruminants ([@b8-gmb-36-465]) and triggers casein expression ([@b55-gmb-36-465]). These results therefore indicated that miRNA 15a indirectly decreases milk production by blocking the expression of growth hormone receptor and thus identified a novel regulatory mechanism for GHR. Some miRNAs have been suggested to have immunosuppressive roles. *In silico* analyses of two members of the miRNA 30 family (miRNAs 30a-5p and 30d-5p) predicted binding sites in several suppressors of cytokine signaling inhibitors of the JAK/STAT pathway that regulate IL-10 transcription ([@b20-gmb-36-465]). This miRNA has also been implicated in the formation of the adipose pad in mammary gland. Le [@b22-gmb-36-465] constructed transgenic mice over-expressing miRNA 30b and observed (based on histological analysis) that these animals had acinar structures with abnormally small lumens. Even when there were no differences in the concentration of major milk proteins, the number of lipid droplets was smaller and did not show the spherical shape seen in the wild type. Microarray analyses of animals that did not express miRNA 30b showed that 164 genes were up-regulated and 56 genes were down-regulated. All of the up-regulated genes were associated with tissue development, except for seven that were involved in the inflammatory response. It is remarkable that the blockade of just a single miRNA altered the expression levels of 222 genes. Further studies are needed to understand the biological roles of most of the reported miRNAs. After identifying miRNAs in mammary gland by sequencing procedures or computational searches, it is important to validate the results in functional experiments and to study their expression pattern in physiological and pathological conditions ([@b25-gmb-36-465]). The purpose of expression studies is to compare patterns between groups, *e.g.*, disease *vs.* healthy and lactation *vs.* dry. The subsequent variation observed between experimental groups should reflect differences in expression between the groups and not be attributable to other sources of variation such as sampling methods, stabilization procedures and extraction methods. A normalizer or internal control should be used. This control is generally an RNA that exhibits invariant expression across all samples, is expressed along with the target miRNA in the cells of interest, and demonstrates equivalent storage stability and efficiency of extraction and quantification as the target miRNA of interest ([@b58-gmb-36-465]; [@b51-gmb-36-465]). Selecting an optimal normalizer (an aspect that is frequently undervalued) may help to avoid inconsistent results. [@b22-gmb-36-465] proposed a set of miRNAs for porcine milk studies. These authors compared the expression of six porcine milk miRNAs from different lactation periods and proposed three of them (miRNAs 17, 107 and 103) as internal controls because they were stabled throughout the periods studied (1 h and 3, 7, 14, 21 and 28 days postpartum). In cow, [@b26-gmb-36-465] normalized the samples using a synthetic cel-miRNA 39. [@b48-gmb-36-465] normalized the data using miRNA 320 and miRNA U6. The usefulness of the latter miRNA has been questioned because of its stability in serum ([@b11-gmb-36-465]). For the next generation of sequencing technology, normalization will be challenging because different sequencing experiments may generate quite different total numbers of reads ([@b42-gmb-36-465]). The total count normalization is inadequate for data generated by new generation sequencing technologies ([@b19-gmb-36-465]). To overcome this limitation various statistical models have been proposed for the normalization of data. Although there is still no general agreement about the most adequate internal controls and the best methods for normalizing data, it is nevertheless essential to establish criteria for selecting which controls should be used for each species and tissue and which method should be used to normalize the data in order to decrease the false discovery rate. Functions of miRNA in mastitis ============================== Mastitis, or inflammation of the mammary gland, is one of the most prevalent and costly diseases in dairy animals (for specific reviews see [@b23-gmb-36-465]; [@b2-gmb-36-465]; [@b15-gmb-36-465]). Intra-mammary infection occurs when bacteria cross the teat sphincter and reach the alveolar lumen after passing through the teat and gland cisterns. The first response of the immune system is a neutrophil influx via chemotaxis to establish the inflammatory process ([@b1-gmb-36-465]). Understanding the molecular mechanisms involved in mastitis would be helpful in developing new strategies to prevent and treat this condition. *Streptococcus uberis* is one of the major etiological agents of mastitis. This Gram-positive bacterium can cause contagious or environmental mastitis ([@b52-gmb-36-465]). Two recent studies examined the miRNA expression pattern in *S. uberis*-induced mastitis: one focused on *in vivo* infection and miRNA expression at 12 h post-infection while the other examined the expression pattern at different times in cultured mammary epithelial cells. Both studies provided interesting insights into the rapid and diverse response triggered after infection. In the first of these studies, [@b48-gmb-36-465] examined the expression pattern of 14 miRNAs in mammary gland 12 h after a challenge with *S. uberis*. The resulting data plus the microarray gene expression patterns of 2,102 genes were used in bioinformatic analyses to identify miRNA targets and the biological pathways involved. The results showed down-regulation of miRNAs 15b, 16a, 31, 145 and 181a, and up-regulation of miRNA 223 in mastitis compared with healthy control animals ([Figure 2, panel A](#f2-gmb-36-465){ref-type="fig"}). The target genes identified were mostly associated with immunological regulation, metabolic processes and cellular proliferation/differentiation. The change in miRNA 16a expression was associated with the up-regulation of some interleukins (IL-6, IL-8 and IL-10). The authors suggested that miRNA 16a might control the level of key inflammatory components in bovine mammary gland and could play a role in regulating the response to mastitis. Mastitis caused the down-regulation of miRNA 181a which has a role in the immune system (such as an increase in toll-like receptor and B cell receptor signaling). However, the decrease in miRNA 181a was the opposite of that observed in mice with an acute inflammatory response ([@b62-gmb-36-465]). In the latter case, the levels of miRNA 181a increased within 2 h after the induction of inflammation and remained high for up to 6 h post-treatment; however, 12 h later the expression was lower than in control animals. The discrepancy between these two studies may be related to the sampling intervals since [@b48-gmb-36-465] obtained the mammary biopsies 20 h after the bacterial challenge. However, differences among species should also be considered. The only miRNA that was up-regulated in udder mastitis was miRNA 223, which can inhibit several cellular signaling pathways mediated via the down-regulation of IGF1R (insulin growth factor 1 receptor). Negative modulation of the immune response may be required to avoid damage to the host tissue. In the second study, [@b35-gmb-36-465] identified 15 miRNAs that showed altered expression in cultured bovine mammary epithelial cells challenged with *S. uberis.* No changes were observed in the first hour after inoculation but at 2 h, miRNA 29e and miRNA 708 were up-regulated. This was followed by the up-regulation of miRNA 7b and miRNA 98 at 4 h post-challenge. At 6 h, 12 miRNAs were up-regulated (miRNAs 7b, 7d, 7e, 24-2, 128-1, 128-2, 185, 200c, 210, 494, 652 and 2342). Down-regulation was first observed only at 4 h and involved miRNAs 29b-2, 193a and 130a. Two miRNAs (29b-2 and 130a) were down-regulated at 4 h and 6 h ([Figure 2, Panel B](#f2-gmb-36-465){ref-type="fig"}). Although a large number of 5′ isomers were identified, they were expressed at a low rate. The prediction of target genes showed that only the miRNAs down-regulated at 4 h and 6 h post-inoculation were significantly enriched in genes with a role in innate immunity. [Table 2](#t2-gmb-36-465){ref-type="table"} summarizes the different miRNAs reported in mammary gland and milk and their level of expression in cow and sheep. The mammary alveolus cell is a three-dimensional structure in which the extracellular matrix plays an active role in epithelial function such as casein expression ([@b30-gmb-36-465]). Therefore, the time intervals studied and the biological model used (*in vivo* vs. cell culture) could account for the divergent miRNA expression reported in these studies. Nevertheless, the results clearly suggest that *S. uberis* can coordinate different processes by regulating target miRNAs. Another aspect to emphasize is the variability in miRNA expression and their rapid and dynamic temporal expression patterns that can modulate the ability of mammary epithelial cells to mobilize the innate immune system. miRNA and biotechnology ======================= Lactoglobulin (LGB) is the major whey protein in ruminants whereas human milk contains no LGB ([@b6-gmb-36-465]). This difference in milk composition accounts for some of the milk allergy problems in infants. LGB intolerance in infants has been known since 1965 ([@b14-gmb-36-465]). One strategy to reduce the allergenic potential of milk is to produce LGB-free milk. In pigs, [@b46-gmb-36-465] reported the effectiveness of short hairpin RNAs and artificial miRNAs in blocking LGB. [@b27-gmb-36-465] designed a successful strategy to block LGB synthesis by constructing a transgenic cow that expressed a tandem miRNAs construct (miRNA 6 and miRNA 4) against bovine LGB (see [Figure 3](#f3-gmb-36-465){ref-type="fig"}). By controlling the artificial miRNAs through a lactation-specific-promoter it is possible to express miRNAs only during the lactation period when LGB is being produced. The analysis of milk samples by SDS-PAGE and HPLC showed that milk from miRNA 4 and 6 calves contained no LGB, but there was a strong, compensatory effect on the levels of other milk proteins. Consequently, artificial miRNAs could be an alternative for abolishing the production of specific proteins ([@b27-gmb-36-465]). Stability of miRNA in milk ========================== One of the interesting features of miRNAs is their stability. miRNAs are resistant to acidic conditions, digestion by RNAse, incubation at room temperature and various freeze/thaw cycles ([@b24-gmb-36-465]; [@b32-gmb-36-465]; [@b22-gmb-36-465]; [@b26-gmb-36-465]; [@b66-gmb-36-465]). In milk, this resistance to degradation is explained by the fact that miRNAs are contained in exosomes or microvesicles. Treatment with Triton-X, a detergent that disrupts lipid membranes, results in the degradation of miRNAs by RNAse ([@b65-gmb-36-465]). The resistance to acidic conditions ensures passage through the stomach and absorption into the bloodstream, and this in turn allows the exchange of genetic information between mother and offspring. An important recent finding is the isolation of miRNAs from industrially processed foods for infants ([@b26-gmb-36-465]). This raises new questions about the possible role that cow miRNAs could have in the end-consumer, especially children. Could cross-species miRNAs have an epigenetic role? In this regard, the study by [@b64-gmb-36-465] have provided interesting results regarding cross-kingdom transmission and regulation since these authors isolated a plant miRNA (miRNA168a) from the blood of mice fed with rice. They also demonstrated that this miRNA could bind to low density lipoprotein receptor adapter 1 mRNA in liver and that this binding led to a decrease in low density proteins in mouse plasma. This plant-derived miRNA was also identified in people consuming a rice-based diet. Even though plant and animal miRNAs have undergone independent evolutionary adaptations and are unrelated ([@b5-gmb-36-465]), these findings show that plant miRNA can still exert a cross-kingdom effect. The identification of cross-kingdom miRNA transmission raises significant questions. To what extent can miRNAs present in the diet regulate mammalian genes? How many miRNAs are incorporated through a normal diet? What are the effects of these miRNAs in health and disease? Milk provides a variety of bioactive components such as lactoferrin, defensin and immunoglobulin. In this regard, miRNAs represent yet another group of molecules transported by milk that could influence the immunological system in neonates. Conclusion ========== The identification of small non-coding RNAs has provided new insights into cell regulation and intercellular communication. Studies of the mammary gland in lactation and mastitis show that the temporal expression of miRNAs regulates the innate immune system. The discovery of cross-species and cross-kingdom miRNA regulation provides a basis for new research into the regulatory mechanisms involved and the impact of epigenetic regulation. As stated by [@b64-gmb-36-465]: "like vitamins, minerals and other essential nutrients derived from food sources, plant (or milk) miRNAs may serve as a novel functional component of food and make a critical contribution to maintaining and shaping animal body structure and function". We are in an exciting time of new discoveries and a new understanding of gene regulation and epigenetic effects. Future research will improve our understanding of the role of miRNAs in health and disease and their importance as food resources. Associate Editor: Carlos F.M. Menck {#f1-gmb-36-465} ![Diagram of the mammary epithelial cell response to infection by *Streptococcus uberis in vivo* and *in vitro*. (A) Mammary alveolus showing (a) epithelial cells, (b) myoepithelial cells, (c) basement membrane, (d) extracellular matrix and (e) capillary. In the alveolar lumen: (f) bacterial infiltration and (g) neutrophil infiltration. The accompanying box shows that 12 h after inoculation with *S. uberis* there was a decrease in the expression of miRNAs 15b, 16a, 31, 145 and 181a and an increase in miRNA 223 to modulate the inflammatory response ([@b48-gmb-36-465]). (B) Mammary epithelial cells inoculated with *S. uberis*. No changes were observed at 1 h but there was an increase in miRNAs 29e and 708 at 2 h, an increase in miRNAs 7b and 98 and a decrease in miRNAs 29b-2, 193 and 130a at 4 h and, finally, an increase in 12 miRNAs (7b, 7d, 7e, 200c, 210, 24-2, 128-2, 128-1, 185, 652, 494 and 2342) concomitantly with a decrease in miRNA 29b2, 29e, 29c, 100 and 130a at 6 h ([@b35-gmb-36-465]).](gmb-36-465-g002){#f2-gmb-36-465} ![*Bos taurus* lactoglobulin B (LGB) mRNA sequence (GenBank accession number BC108213.1) showing the positions of miRNA6 and miRNA4 that targeted LGB, as designed by [@b27-gmb-36-465]. The numbers refer to the nucleotide positions.](gmb-36-465-g003){#f3-gmb-36-465} ###### Total number of miRNAs reported in colostrum, milk and mammary gland in cow, goat and sheep. Number of miRNAs Methods Reference ------- ------------------ --------- ----------- --------------------------------- ------------------- Cow 230 213 Solexa deep sequencing analysis [@b12-gmb-36-465] Cow 921 Solexa deep sequencing analysis [@b44-gmb-36-465] Goat 100 53 Microarray-real time PCR [@b26-gmb-36-465] Goat 328 Solexa deep sequencing analysis [@b29-gmb-36-465] Goat 180 441 Solexa deep sequencing analysis [@b43-gmb-36-465] Sheep 101 cDNA sequencing [@b18-gmb-36-465] ###### Summary of miRNA expression in mammary gland, mammary epithelial cell lines and milk in cow and sheep. miRNA Tissue or biological fluid Species Expression level Reference ------------- ---------------------------- ----------- --------------------------------------------------------------------------- --------------------------------------------------------- miRNA 7b Milk Cow Elevated in mastitis [@b35-gmb-36-465] miRNA 7d Milk Cow Elevated in mastitis [@b35-gmb-36-465] miRNA 7e Milk Cow Elevated in mastitis [@b35-gmb-36-465] miRNA 10a MGT Cow Elevated in lactation [@b59-gmb-36-465] miRNA 15a MECL Cow Elevated in transfected cells [@b41-gmb-36-465] miRNA 15b MGT Cow Elevated in lactation and reduced in mastitis [@b59-gmb-36-465]; [@b48-gmb-36-465] miRNA 16 MGT Cow Elevated in lactation [@b59-gmb-36-465] miRNA 16a MGT Cow Reduced in mastitis [@b48-gmb-36-465] miRNA 21 MGT Cow/sheep Elevated in cow during lactation and in sheep during early pregnancy [@b18-gmb-36-465]; [@b59-gmb-36-465] miRNA 24-2 Milk Cow Elevated in mastitis [@b35-gmb-36-465] miRNA 29-b2 Milk Cow Reduced in mastitis [@b35-gmb-36-465] miRNA 29c Milk Cow Reduced in mastitis [@b35-gmb-36-465] miRNA 29e Milk Cow Elevated and then reduced in mastitis [@b35-gmb-36-465] miRNA 31 MGT/milk Cow Reduced in lactation and mastitis [@b59-gmb-36-465]; [@b48-gmb-36-465]; [@b35-gmb-36-465] miRNA 33b MGT Cow Elevated in lactation [@b59-gmb-36-465] miRNA 98 Milk Cow Elevated in mastitis [@b35-gmb-36-465] miRNA 100 Milk Cow Reduced in lactation [@b35-gmb-36-465] miRNA 128-1 Milk Cow Elevated in lactation [@b35-gmb-36-465] miRNA 128-2 Milk Cow Elevated in lactation [@b35-gmb-36-465] miRNA 130a MGT Cow Reduced in mastitis [@b35-gmb-36-465] miRNA145 MGT Cow Elevated in lactation and reduced in mastitis [@b59-gmb-36-465]; [@b48-gmb-36-465] miRNA 146b MGT Cow Elevated in lactation [@b59-gmb-36-465] miRNA 148a Milk Cow Elevated in lactation [@b12-gmb-36-465] miRNA 155 MGT Cow Elevated in lactation [@b59-gmb-36-465] miRNA 181a MGT Cow Elevated in lactation and reduced in mastitis [@b59-gmb-36-465]; [@b48-gmb-36-465] miRNA 185 Milk Cow Elevated in mastitis [@b35-gmb-36-465] miRNA 193a MGT Cow Reduced in mastitis [@b35-gmb-36-465] miRNA 200 MGT Sheep Elevated in lactation [@b18-gmb-36-465] miRNA 200c MGT/milk Cow Elevated in mastitis [@b35-gmb-36-465] miRNA 205 MGT Cow/sheep Elevated in cow lactation and mastitis and second half of sheep pregnancy [@b18-gmb-36-465]; [@b59-gmb-36-465] miRNA 210 MGT/milk Cow Elevated in mastitis [@b35-gmb-36-465] miRNA 221 MGT Cow Elevated in lactation [@b59-gmb-36-465] miRNA 223 MGT Cow Elevated in lactation and mastitis [@b59-gmb-36-465]; [@b48-gmb-36-465] miRNA 494 Milk Cow Elevated in mastitis [@b35-gmb-36-465] miRNA 652 Milk Cow Elevated in mastitis [@b35-gmb-36-465] miRNA 708 Milk Cow Elevated in mastitis [@b35-gmb-36-465] miRNA 2342 Milk Cow Elevated in mastitis [@b35-gmb-36-465] ECL: mammary epithelial cell line, MGT: mammary gland tissue.

{#sp1 .478} {#sp2 .479} {#sp3 .480}

1. Introduction {#sec1-behavsci-10-00114} =============== Healthy parent--child interactions provide opportunities for children to accrue rich social experiences for development \[[@B1-behavsci-10-00114],[@B2-behavsci-10-00114],[@B3-behavsci-10-00114],[@B4-behavsci-10-00114],[@B5-behavsci-10-00114],[@B6-behavsci-10-00114]\]. These dyadic exchanges are bidirectionally influenced by nuanced patterns of emotional transactions contributed by both the parent and the child \[[@B1-behavsci-10-00114],[@B2-behavsci-10-00114],[@B3-behavsci-10-00114],[@B4-behavsci-10-00114],[@B6-behavsci-10-00114],[@B7-behavsci-10-00114],[@B8-behavsci-10-00114]\] and are influenced by psychological \[[@B9-behavsci-10-00114]\] and biological and factors \[[@B10-behavsci-10-00114],[@B11-behavsci-10-00114]\]. The quality of parent--child interactions hinges on dyadic partners' emotional availability to each other \[[@B12-behavsci-10-00114]\]; that is, their emotional connectedness and ability to mutually discern and respond to each other's needs \[[@B12-behavsci-10-00114],[@B13-behavsci-10-00114],[@B14-behavsci-10-00114]\]. When evaluated from both the parental and child aspects of the relationship, emotionally available dyads are better able to reciprocate each others' socio-emotional cues in a congruent manner \[[@B15-behavsci-10-00114],[@B16-behavsci-10-00114],[@B17-behavsci-10-00114]\]. Play is an essential activity that stimulates parent--child interactions, especially among preschool-aged children who are beginning to exercise their newfound autonomy and motor capacities \[[@B18-behavsci-10-00114]\]. Maternal and paternal interactions differ in ways that can be observed in such play situations. Through the lens of emotional availability, Bergmann and colleagues \[[@B19-behavsci-10-00114]\] showed that mothers engage in more sensitive interactions compared to fathers---an observation that was postulated to be driven by maternal predominance in caregiving functions \[[@B16-behavsci-10-00114],[@B20-behavsci-10-00114]\]. Parental gender differences have also emerged in the structuring domain during play. For instance, while mothers tend to employ greater scaffolding and didactic strategies, fathers lean more towards physical play and tend to engage with their child like age-mates \[[@B14-behavsci-10-00114],[@B21-behavsci-10-00114],[@B22-behavsci-10-00114]\]. Although fathers customarily occupy the "playmate" role in child-rearing, father--child play has been shown to be essential for development, having been previously associated with children's attachment security \[[@B23-behavsci-10-00114],[@B24-behavsci-10-00114]\], socio-cognitive development and emotional regulation \[[@B25-behavsci-10-00114],[@B26-behavsci-10-00114]\]. Parent's attachment styles are influenced by the first relationship one has with their parents as an infant, which further influences subsequent caregiving behaviours that one may exhibit in the future. Based on attachment theory, early experiences of attachment as an infant provides exposure to social relationships that help form the internal working model of attachment, which is used as a reference for future social relationships \[[@B27-behavsci-10-00114]\]. In this way, early experiences of social relationships with a parent would shape future relationships; in particular, relationships where one becomes the caregiver. An adult's parental attachment has been predictive of subsequent parenting behaviours \[[@B28-behavsci-10-00114],[@B29-behavsci-10-00114]\]. Parental attachment, as examined by the Adult Attachment Interview, are classified into dismissive, preoccupied and autonomous \[[@B30-behavsci-10-00114],[@B31-behavsci-10-00114]\]. One's adult attachment then corresponds to subsequent parent-infant attachment, as assessed by the strange situation procedure, where mothers with autonomous adult attachment are more likely to have infants with secure attachments and mothers with preoccupied adult attachment are more likely to have infants with resistant attachments \[[@B31-behavsci-10-00114]\]. Parental attachment, in this way, would influence future parenting practices that may be captured by parental bonding history. The prevailing literature supports the view that parental^g^ bonding history (parental^g^ denotes the intergenerational influence of past parental bonding history on parents' current caregiving behaviours), which is evaluated from the parental care and overprotection that the parent received as a child, influence parent--child interactions in the subsequent generation \[[@B32-behavsci-10-00114],[@B33-behavsci-10-00114],[@B34-behavsci-10-00114]\]. Intergenerational transmission of parenting occurs when parenting practices, behaviours and attitudes of one generation are either directly or indirectly influenced by the previous generation \[[@B35-behavsci-10-00114]\]. Longitudinal studies investigating parenting behaviours across generations have revealed consistent continuities in parenting from one generation to the next \[[@B36-behavsci-10-00114]\]. Despite modest associations, intergenerational transmission of positive and negative parenting has been observed in various social and economic contexts \[[@B37-behavsci-10-00114],[@B38-behavsci-10-00114],[@B39-behavsci-10-00114],[@B40-behavsci-10-00114],[@B41-behavsci-10-00114],[@B42-behavsci-10-00114]\]. Moreover, suboptimal parental bonding in a parent^g^--child relationship has been posited to be intergenerationally transmitted through less emotionally available parent--child interactions in the subsequent generation, where insecurely attached mothers exhibit less maternal sensitivity when interacting with their own children \[[@B43-behavsci-10-00114]\], while their children tend to display less responsiveness towards them \[[@B44-behavsci-10-00114]\]. Parental attachment has also been shown to be influenced by parenting stress, where mothers who perceived a lower-quality rearing from their parents were more likely to experience greater parenting stress when they become parents as compared to mothers who perceived higher-quality rearing from their parents \[[@B45-behavsci-10-00114]\]. Besides factors that reside in the past, current parenting stress inevitably affects parents' interactions with their children too. Parenting stress reflects a state in which parenting demands exceed the coping resources that the parent possesses \[[@B46-behavsci-10-00114]\]. Higher levels of parenting stress have been consistently linked to maladaptive parenting behaviours \[[@B47-behavsci-10-00114],[@B48-behavsci-10-00114]\] which are characterised by less emotionally available parenting \[[@B49-behavsci-10-00114],[@B50-behavsci-10-00114]\]. For instance, mothers who experienced higher levels of parenting stress have been found to display less sensitivity \[[@B51-behavsci-10-00114]\], while exhibiting greater hostility and intrusiveness \[[@B52-behavsci-10-00114],[@B53-behavsci-10-00114]\]. Parenting stress has also been associated with decreased brain-to-brain synchrony in parts of the brain that are implicated in inferring and understanding others' mental states and social cognition, thereby adversely influencing mother--child attunement \[[@B7-behavsci-10-00114]\]. Compounding these negative effects, children of stressed and less emotionally available mothers showed less involvement and responsiveness when interacting with their mothers \[[@B52-behavsci-10-00114]\]. These studies point to the pivotal role of both past and present experiences in determining dyadic emotional availability. Emotional availability is a construct used to assess the quality of interaction between an adult and a child. Both members of the dyad are considered as interdependent agents that contribute to their bidirectional social exchanges \[[@B12-behavsci-10-00114]\]. The Emotional Availability Scale (EAS) operationalises emotional availability into six multidimensional domains of behaviours, four of which are parental scales (i.e., sensitivity, structuring, non-intrusiveness, non-hostility) and two of which are child scales (i.e., responsiveness and involvement \[[@B12-behavsci-10-00114],[@B15-behavsci-10-00114]\]. Adult sensitivity measures the extent to which the adult responds to the child in a timely and appropriate manner, which reflects his/her emotional attunement to the child. Adult structuring captures the parent's ability to provide suitable guidance and set limits when necessary while taking the child's autonomy into consideration. Adult non-intrusiveness assesses the parent's proclivity to support the child's age-appropriate autonomy by refraining from overstimulating, interfering with and controlling the interaction. The final parental scale, adult non-hostility, evaluates the absence of antagonistic responses displayed by the parent towards the child. These behaviours may take the form of disgruntled facial expressions and body language, disparaging comments and mockery and an inimical tone of voice. Covert hostility may manifest as annoyance, impatience and indifference during the interaction. Meanwhile, the child responsiveness scale characterises whether the child is genuinely eager to reciprocate the parent's attempt at engaging with him/her, whereas the child involvement scale reflects the child's inclination to initiate social exchanges and include the parent in the interaction. The EAS has been applied to numerous naturalistic contexts, including play situations (e.g., \[[@B16-behavsci-10-00114],[@B21-behavsci-10-00114]\]) due to its property of being able to measure bilateral components of parent--child interaction. Taken together, the burgeoning literature has provided compelling evidence for the use of EAS in evaluating maternal and paternal emotional availability during play. From the studies above, it can be seen that both parental^g^ bonding history and currently experienced parenting stress have an impact on the emotional availability of the parent when interacting with their child. Depending on the quality of parental^g^ bonding and parental stress, these changing parent-related variables will produce different levels of emotional availability during parent--child interaction. Identification of how specific patterns of parental^g^ bonding and parenting stress affect emotional availability will be clinically significant in improving the quality of the parent--child relationship \[[@B54-behavsci-10-00114]\], and provide a more nuanced approach to parenting interventions that have previously viewed parents as a relatively homogeneous group based on their life stages \[[@B55-behavsci-10-00114]\]. Psychotherapies that focus on improving parent--child interactions can take into account the prevailing relationships between parental^g^ bonding, parenting stress and emotional availability, and make use of specific strategies that target these factors to produce more efficacious outcomes \[[@B56-behavsci-10-00114]\]. However, despite the considerable literature on parental^g^ bonding history and current parenting stress (e.g., \[[@B43-behavsci-10-00114],[@B44-behavsci-10-00114],[@B47-behavsci-10-00114],[@B48-behavsci-10-00114],[@B52-behavsci-10-00114],[@B53-behavsci-10-00114]\]), studies examining their simultaneous effects in both father--child and mother--child dyads is lacking. Only one study so far examined parental^g^ bonding history and current parenting stress, where parents who reported optimal parental^g^ bonding history also reported the lowest level of current parenting stress \[[@B56-behavsci-10-00114]\]. Therefore, the present study aims to investigate how parental^g^ bonding history and current parenting stress influence dyadic emotional availability observed in father--child and mother--child pairs during a typical play situation. We embarked on this study with the following hypothesis to be tested: given the robust link between parenting stress and maladaptive parenting behaviours (e.g., \[[@B52-behavsci-10-00114],[@B53-behavsci-10-00114],[@B57-behavsci-10-00114]\]), we predicted that parenting stress would also interact with poor parental^g^ bonding history. However, the direction of effect of this interaction on dyadic emotional availability remains exploratory. 2. Methods {#sec2-behavsci-10-00114} ========== 2.1. Participants {#sec2dot1-behavsci-10-00114} ----------------- In total, 29 father--child dyads (18 boys, 11 girls; father's age = 38.07 years, child's age = 42.21 months) and 36 mother--child dyads (21 boys, 15 girls; mother's age = 34.75 years, child's age = 41.72 months) participated in this study. Participants were recruited through online platforms, such as Facebook groups and forums. The following criteria had to be met before participants were deemed eligible for the study: (1) adult participants must be at least 21 years old at the time of recruitment; (2) child participants must be between 24 and 48 months at the time of recruitment; (3) adult and child participants need to be residing in the same household in Singapore; (4) adult participants must be the biological parents of child participants; (5) all participants must not suffer from any cognitive impairments, hearing or visual impairments or major diseases that would prevent them from understanding and responding to the experimental tasks. Prior to the commencement of the study, informed consent was obtained from all participants, where parents would provide consent for their children. Participants were remunerated upon completion of the study. The study was conducted in accordance with the Declaration of Helsinki, and the protocol was approved by the Institutional Review Board of the Nanyang Technological University (IRB 2018-06-016). 2.2. Questionnaires {#sec2dot2-behavsci-10-00114} ------------------- Prior to attending the laboratory sessions, adult participants were requested to complete a series of online questionnaires which consisted of basic demographic questions (e.g., birthdate), the Parental Bonding Instrument (PBI; \[[@B58-behavsci-10-00114]\]) and the Parenting Stress Index--Short Form (PSI; \[[@B59-behavsci-10-00114]\]). *Parental Bonding Instrument (PBI)*. The PBI is a self-reported questionnaire used to assess an adult's perception of their parent's parenting attitudes and behaviours when they were younger \[[@B58-behavsci-10-00114]\]. It is completed retrospectively, meaning adults above the age of 16 would complete the scale in terms of how they remember their parents during their first 16 years of life. This 25-item scale measures two constructs: care (12 items) and overprotection (13 items). Examples of items on the care scale include, "Spoke to me in a warm and friendly voice," and "Seemed emotionally cold to me"; examples of items of the overprotection scale include, "Tried to control everything I did," and "Let me decide things for myself." This measure has both a maternal and paternal component, to be completed for each parent respectively. Responses are measured in a 4-point scale, ranging from 0 ("very likely") to 3 ("very unlikely"). High scores in the care sale reflect parental affection and warmth, whereas high scores in overprotection scale reflect parental control and prevention of autonomy \[[@B60-behavsci-10-00114],[@B61-behavsci-10-00114]\], with test re-test coefficients scores remaining significant at *p* \< 0.001 level for both care and overprotection scales over 20 years \[[@B62-behavsci-10-00114]\]. The PBI has also been found to have good validity, demonstrate high convergent validity scores with the Emotional Warmth, Rejection and Protection (EMBU) scale and have sufficient construct validity \[[@B62-behavsci-10-00114]\]. Although self-reported responses may be subjective, this construct is posited to be a valid indicator of actual parenting experiences, due to its close corroborations with siblings' and parents' own parenting style reports \[[@B63-behavsci-10-00114],[@B64-behavsci-10-00114]\]. *Parenting Stress Index-Short Form (PSI)*. The PSI is a self-reported questionnaire assessing the amount of stress associated with parenting \[[@B59-behavsci-10-00114]\]. This 36-item measure assessed parenting stress on three 12-item subscales: parental distress, parent--child dysfunctional interaction and difficult child, and a total parenting stress index score. Parental distress assesses the extent to which parents feel competent, restricted, supported and/or depressed in their parenting roles. Parent-child dysfunctional interaction refers to the extent to which parents are satisfied with their children and their interactions with their children. Difficult child refers to the extent of difficulty in taking care of their child. Total score is an indication of the overall stress a parent is experiencing. Responses are recorded in a 5-point Likert scale, ranging from 1 ("strongly disagree") to 5 ("strongly agree"). High scores indicate higher levels of stress. PSI has been shown to be as reliable and valid as its long form \[[@B65-behavsci-10-00114]\]. It is also known to have acceptable reliability of 0.70 for parental distress, parent-child dysfunctional interaction and difficult child, and 0.85 for total score and acceptable validity \[[@B66-behavsci-10-00114]\]. 2.3. Procedure {#sec2dot3-behavsci-10-00114} -------------- The present study consisted of a home-based online questionnaire followed by a play session conducted in a standard laboratory. In the first part of the study, participants who were assessed to be eligible for the study after pre-screening would complete the online questionnaire. Each parent--child dyad was given a unique participant code which parents were required to use to identify their responses. After completion, an appointment was set for each dyad to come down to the laboratory for the second part of the study. In the second part of the study, adult and child participants were briefed about the procedures relating to the play session. They were informed that they were being video recorded and that they may withdraw from the study at any point in time. Informed consent was obtained from the participants before they were brought to an adjacent room for the experiment. Parents were instructed to engage in a free play session with their child for a duration of 10 minutes. During the play session, the parent and child sat next to each other and a standard set of toys was provided on a table in front of them \[[@B67-behavsci-10-00114]\]. The following items were made available to the dyad: a cake and tea set, a cash register set, a doll, a toy car, two balls, a set of building blocks and three age-appropriate books. The positions of all items on the table were standardised across participants. A bell was provided for the parent in the event that they would like to terminate the play session. At the start of the session, the researchers started recording the video and left the room, only to return at the end of the 10-minute mark. Participants were then debriefed and remuneration was provided. The dataset generated for this publication are available on the Data Repository of the Nanyang Technological University at the following page <https://doi.org/10.21979/N9/IZQPBI> \[[@B68-behavsci-10-00114]\]. 2.4. Play Coding {#sec2dot4-behavsci-10-00114} ---------------- *Emotional Availability Scale (EAS)*. The EAS measure assesses the emotional quality of a parent--child relationship \[[@B69-behavsci-10-00114]\]. This is done by rating the emotional availability of parent--child interactions through four global adult scales (i.e., adult sensitivity, adult structuring, adult non-intrusiveness, adult non-hostility) and two global child scales (i.e., child responsiveness and child involvement). Two independent coders were trained on EAS scoring prior to coding the recorded videos. They rated each of the six dimensions on a Likert scale ranging from 1 to 7. The "irr" package in RStudio \[[@B70-behavsci-10-00114]\] was used to calculate inter-rater agreement between the two coders. An inter-rater agreement of at least 80% was achieved for each EAS subscale. In the event where scores between the two coders were different, the two coders deconflicted their ratings and agreed upon a final score. The EAS has also been shown to be highly reliable, with satisfactory construct validity \[[@B15-behavsci-10-00114]\]. 2.5. Analytical Plan {#sec2dot5-behavsci-10-00114} -------------------- ### 2.5.1. Descriptive Analyses {#sec2dot5dot1-behavsci-10-00114} The mean and standard deviations of all EAS, PBI and PSI scores, along with the age of the participants, were reported. ### 2.5.2. Preliminary Analyses {#sec2dot5dot2-behavsci-10-00114} Preliminary analysis of variance (ANOVA) analyses were conducted to examine the effects of the child's sex and age and the parent's age on EAS, PSI and PBI scores in pooled father--child and mother--child samples. Inter-correlation coefficients between different subscales of EAS, PSI and PBI were reported. ### 2.5.3. Inferential Analyses {#sec2dot5dot3-behavsci-10-00114} To test the separate and interacting effects of parenting stress and parental^g^ bonding history on emotional availability in a parent--child interaction, analysis of variance (ANOVA) with EAS scores as the dependent variable, and PSI and PBI as the independent variable, were conducted on the pooled sample consisting both mother--child and father--child dyads. Parental gender was included as a factor in all models (e.g., non-hostility \~ parenting distress \* maternal^g^ overprotection \* parental gnder). Since all combinations of PBI and PSI scales were being tested, this resulted in 16 models for each EAS scale. As such, Bonferroni correction was applied, such that the new alpha was 0.05/16 = 0.00313. All statistical analyses were conducted on R studio (version 1.0.153, R-core 3.4.2). 3. Results {#sec3-behavsci-10-00114} ========== 3.1. Descriptive Results {#sec3dot1-behavsci-10-00114} ------------------------ Descriptive data for the pooled sample, along with data on father--child and mother--child samples, are reported in [Table 1](#behavsci-10-00114-t001){ref-type="table"}. 3.2. Preliminary Results {#sec3dot2-behavsci-10-00114} ------------------------ Preliminary analyses of the pooled father--child and mother--child samples did not show a significant main effect of child's sex, child's age or parent's age on any of the dependent and independent variables. Inter-correlation coefficients between the different subscales of EAS (i.e., adult sensitivity, adult structuring, adult non-intrusiveness, adult non-hostility, child responsiveness, child involvement), the subscales of PBI (i.e., matercal care, maternal overprotection, paternal care, paternal overprotection) and PSI (parental distress, difficult child, parent--child dysfunctional interaction and total parenting stress) are reported in [Table 2](#behavsci-10-00114-t002){ref-type="table"}. 3.3. Inferential Results {#sec3dot3-behavsci-10-00114} ------------------------ Inferential analyses of the pooled father--child and mother--child samples showed a significant main effect of parental gender on adult non-intrusiveness across all non-intrusiveness models: parental distress by maternal^g^ care model (F(1,54) = 18.156; *p* = 8.2 × 10^−5^), parental distress by maternal^g^ overprotection model (F(1,54) = 16.004; *p* = 0.000194), parental distress by paternal^g^ care model (F(1,54) = 14.938; *p* = 0.0003), parental distress by paternal^g^ overprotection model (F(1,54) = 14.019; *p* = 0.000441), parent--child dysfunctional interaction by maternal^g^ care model (F(1,54) = 19.504; *p* = 4.86 × 10^−5^), parent--child dysfunctional interaction by maternal^g^ overprotection model (F(1,54) = 17.365; *p* = 0.000112), parent--child dysfunctional interaction by paternal^g^ care model (F(1,54) = 15.595; *p* = 0.000229), parent--child dysfunctional interaction by paternal^g^ overprotection model (F(1,54) = 13.753; *p* = 0.000493), difficult child by maternal^g^ care model (F(1,54) = 19.865; *p* = 4.23e-05), difficult child by maternal^g^ overprotection model (F(1,54) = 17.010; *p* = 0.000129), difficult child by paternal^g^ care model (F(1,54) = 16.145; *p* = 0.000183), difficult child by paternal^g^ overprotection model (F(1,54) = 14.847; *p* = 0.000312), total parenting stress by maternal^g^ care model (F(1,54) = 18.912; *p* = 6.11 × 10^−5^), total parenting stress by maternal^g^ overprotection model (F(1,54) = 17.319; *p* = 0.000114), total parenting stress by paternal^g^ care model (F(1,54) = 15.615; *p* = 0.000227), total parenting stress by paternal^g^ overprotection model (F(1,54) = 13.800; *p* = 0.000484). Post-hoc *t*-test analyses demonstrated that fathers showed significantly greater non-intrusiveness compared to mothers (t(59.94) = 4.26; *p* \< 0.001; [Figure 1](#behavsci-10-00114-f001){ref-type="fig"}). ANOVA analyses also elicited a significant interaction between difficult child score and maternal^g^ overprotection on adult non-hostility (F(1,54) = 9.963; *p* = 0.00261). Regression slope tests of maternal^g^ overprotection, averaged across difficult child scores, showed a significant difference between mean − 1 SD, mean and mean + 1 SD values of maternal^g^ overprotection (t ratio = 3.494; *p* = 0.0009; [Figure 2](#behavsci-10-00114-f002){ref-type="fig"}). From [Figure 2](#behavsci-10-00114-f002){ref-type="fig"}, differences in maternal^g^ overprotection could be observed at lower values of difficult child score. A contrast between mean − 1 SD and mean + 1 SD maternal^g^ overprotection score yielded a significant difference when difficult child score = 0, further verifying this observation (t ratio = 3.824; df = 58; estimate = 2.75; *p* = 0.0003). When difficult child scores are low, higher maternal^g^ overprotection would result in greater adult non-hostility compared to lower maternal^g^ overprotection. 4. Discussion {#sec4-behavsci-10-00114} ============= This study seeks to investigate the link between parenting stress and parental^g^ bonding history (i.e., parental care and overprotection that a parent received as a child) on the emotional availability in a parent--child interaction. We embarked on this study with one hypothesis: that parenting stress would interact with parental^g^ bonding history to influence emotional availability. The hypothesis was supported. In the pooled mother--child and father--child sample, greater adult non-hostility, a component of emotional availability, was observed when parents report low parenting stress on the difficult child subscale, but only for parents who perceived higher maternal^g^ overprotection. These results suggest that greater perceived parental^g^ overprotection, when combined with lower parenting stress, predict enhanced emotional availability. Findings from the pooled mother--child and father--child samples revealed that parents displayed less hostile parenting behaviours when they believe their child to be manageable, but only if they experienced greater maternal overprotection in their childhood. This phenomenon is largely contradictory to the literature on the perception of maternal overprotection, which has been shown to have suboptimal outcomes for the child \[[@B71-behavsci-10-00114]\]. This may be attributed to cultural differences, where maternal overprotection may be more positively received in Asian families. In a study on immigrant Chinese mothers' parenting styles, it was found that authoritative parenting was associated with fewer adjustment problems in children \[[@B72-behavsci-10-00114]\]. Stright and Yeo \[[@B73-behavsci-10-00114]\], in studying Singaporean children's perception of their mother's parenting styles, found that their perception of maternal use of psychological control was positively associated with perception of maternal warmth. Moreover, perception of greater parental control as a function of order-keeping (i.e., parental organisation) was found to be positively correlated with greater independence and self-esteem in children \[[@B74-behavsci-10-00114]\]. Maternaloverprotection, in this way, may have captured maternal^g^ control in childhood that was serving as a way to enforce well-defined limits for children's behaviours, allowing for children to practise autonomy through controlling their own behaviours. Additionally, maternal control in Asian mothers is typically expressed in situations involving difficult child behaviours, which is in accordance with the use of parental behavioural control to encourage impulse control and proper conduct in children \[[@B75-behavsci-10-00114],[@B76-behavsci-10-00114]\]. Perception of maternal control may then be associated with greater capabilities in stress management by fostering a sense of competence and autonomy in their child. It is possible that through this, parents who perceived greater parental^g^ overprotection when younger may have developed stress coping mechanisms that were beneficial in their parenting roles, which would explain the lower difficult child and parenting stress scores, as observed in this study. Travis and Combs-Orme \[[@B77-behavsci-10-00114]\] also found similar results, in that mothers who received poor parenting in childhood (i.e., reported high maternal^g^ overprotection) were able to report lower levels of parenting stress and developed greater resilience to life stressors. They also reported low levels of dissatisfaction with their parent--child interaction and low levels of difficulty in managing their child, much like positive-adaptive mothers (mothers who recollected warm and non-controlling parenting in childhood) in the study. In this way, mothers who perceived greater maternal^g^ overprotection may have found their child manageable, and thus display fewer hostile behaviours. Conversely, results also indicate that lower adult non-hostility scores were observed when greater maternal overprotection scores interacted with greater difficult child scores. This suggests that the advantage of parental non-hostility diminishes when parents perceive both greater maternal overprotection and greater stress in child management (as denoted by higher difficult child scores). High difficult child scores indicate a lower perceived ability to cope with parenting stress that stems from strenuous child management \[[@B59-behavsci-10-00114]\]. Although maternal overprotection may have been protective in parents with low difficult child scores (i.e., greater non-hostility observed in parenting behaviours), it is possible that maternal overprotection loses that effectiveness in parents who find their children difficult to manage in the first place. For example, children with difficult temperaments have been shown to covary with greater maternal intrusiveness \[[@B78-behavsci-10-00114]\], lower maternal sensitivity \[[@B79-behavsci-10-00114]\] and greater parenting stress \[[@B80-behavsci-10-00114]\]. When parental coping resources are overwhelmed in this way, parents then rely on prior experiences (i.e., parenting practices received when younger). Indeed, studies on intergenerational transmission of parenting showed that parents who perceived greater maternal^g^ control when younger are more likely to repeat similar power-assertive discipline methods when they become parents, especially so if they perceived their children as difficult to manage \[[@B37-behavsci-10-00114],[@B81-behavsci-10-00114]\]. Through this, parenting stress associated with child management is positively associated with maternal^g^ overprotection parents received when they were younger, such that parents with higher difficult child stress display lower non-hostility scores as compared to parents with lower difficult child stress. Additionally, it should be noted that non-hostility as measured in the emotional availability scale is scored based on the absence of hostile responses, whether they are covert or openly hostile. It is an indicator of background anger where the adult's hostility does not need to be directed at the child \[[@B15-behavsci-10-00114]\], in which case, displays of dissatisfaction, impatience, anger, etc., are also included in the scoring of hostile behaviours during a parent--child interaction. Non-hostility scores have also been found to be strongly correlated with maternal stress \[[@B52-behavsci-10-00114]\]. Indeed, mothers who are more attuned to their children's emotional cues and expressions are better able to respond to their children's needs and respond appropriately, whereas mothers who perceive their children to be harder to understand have a higher tendency to be less emotionally present in their interactions with their child, which may be displayed as speaking in flat tones and less ideal response timing \[[@B53-behavsci-10-00114],[@B82-behavsci-10-00114]\]. Put together, these studies illustrate how maternal^g^ overprotection may influence or interact with current experience of parenting stress (particularly in child management), which together have the combined influence on a parent's degree of hostility in parenting behaviours. As with all studies, interpretation of the results obtained has to be considered in the context of its limitations. Access to actual parenting behaviours that parents in this study received in childhood are not available. This means that responses on parental^g^ bonding history are retrospective and vulnerable to interpretive biases. Recollections of parenting received in childhood may be influenced by a participant's present status at the time of responding (e.g., \[[@B83-behavsci-10-00114]\]). This may have had an influence on the way participants recalled parenting in childhood, which might explain the contrary results obtained with regard to maternal^g^ and paternal^g^ overprotection, current participants' emotional availability and parenting stress. In addition, the exact mechanisms behind the results obtained are also uncertain because the results of this study are correlated in nature. It is possible that the parental^g^ bonding history may have a different relationship with current parenting behaviours and parenting stress. For example, perception of poor paternal^g^ parenting has been hypothesised to lead to better paternal parenting in subsequent generations based on the reworking hypothesis \[[@B84-behavsci-10-00114]\]. Men who reported receiving less warm parenting from their fathers in childhood are more likely to report experiencing the most parenting stress and rated themselves worse on paternal ability. This, in turn, was associated with greater time spent in engaging their children in verbal stimulation and physical play. This is an avenue that merits further investigation, for example, by conducting a longitudinal study in examining actual parenting behaviours as received by parents when they were children and its effects on current parenting behaviours. Future studies may also explore the use of the PBI to investigate cultural differences in the use and understanding of parental^g^ overprotection and control. As seen in this paper, parental^g^ overprotection may not necessarily have negative implications in the local context. The finding that parental^g^ overprotection, rather than parental^g^ care, has a relationship with the frequency of hostile and sensitive parenting behaviours, requires further investigation. It is likely that a highly warm and highly overprotective parental^g^ bonding history may allow for a more positive socio-emotional development in children. Parenting in childhood that is high in warmth and control has also been shown to be associated with children's greater ability to regulate their behaviour and attention \[[@B72-behavsci-10-00114]\]. Alternatively, it is also possible that a highly overprotective parental^g^ bonding history served as a reminder for current parents to do better, allowing for the development of more emotionally available parenting. Additionally, the implications of this study fill a gap in the literature in understanding how parental^g^ bonding history may interact with current parenting stress to influence emotional availability in dyadic interactions. It highlights possible areas of misconception that need to be addressed with regard to the effects of the father's involvement in parenting and provide evidence on the differential effects of gender on the parent--child relationship. These results provide a backdrop for future studies to examine the associations between current parenting behaviours and experiences and past parental^g^ experiences. Understanding these directional relationships, and the parenting attitudes that may have developed from parental^g^ bonding history, can be the next steps to take in elucidating parenting behaviours in Asian societies. Finally, taking the results and implications discussed above, the findings of this study may be able to shed light on clinical applications on parenting skills and parent--child relationships in a culturally-specific context. It must be noted that parents from cultures where parental overprotection (or specifically in this study, maternal overprotection) is perceived more positively \[[@B73-behavsci-10-00114]\] may not benefit from parenting strategies that focus on cultivating child autonomy and independence at the expense of decreasing parental protection, as it disrupts the larger cultural influences of encouraging greater parental protection and involvement \[[@B85-behavsci-10-00114]\]. Instead, a more nuanced approach in teaching parenting skills that have to do with setting developmentally appropriate limits without undermining the child's potential to develop autonomy and independence \[[@B86-behavsci-10-00114]\] would be recommended. 5. Conclusions {#sec5-behavsci-10-00114} ============== To conclude, current parenting behaviours may be influenced by both past and present circumstances. This means that when examining the development of parenting behaviours, both parenting received as a child (i.e., parental^g^ bonding history) and current parenting stress experienced should be taken into account. When parenting behaviours are assessed in terms of dyadic emotional availability, the bidirectional relationship between the parent and child are taken into account. Use of parental^g^ bonding history and parenting stress to understand parenting behaviours as assessed by the emotional availability scale in this study hints not only at the development of parenting behaviours, but also at how past parenting received and current parenting stress influence current parent--child interaction. When seen in this way, parental^g^ bonding history and parenting stress have long lasting effects on parent--child interactions and relationships. Alternatively, it may be that the parent--child relationship may have had an influence on the way parenting stress and current parenting behaviours are demonstrated and experienced. Reciprocal interactions from child to parent have a significant influence on how parenting stress is experienced, as evident in the influence of the difficult child subscale on the interaction between parental^g^ bonding history and current parenting behaviours. All in all, this study showed that in studying current parenting behaviours, several factors may need to be taken into account: past parenting received, current parenting stress and relationships with one's parent and one's child. We would like to extend our appreciation to Michelle Neoh, Jezebel Chin Syuen Chong, Justin Randall Durnford, Siti Syazana Biti Abdul Halim and Anais Ang for their assistance in the project. **Sample Availability**: The dataset is available on the Data Repository of the Nanyang Technological University at the following page <https://doi.org/10.21979/N9/IZQPBI> \[[@B68-behavsci-10-00114]\]. Conceptualisation, A.A. and A.W.T.W.; analysis, A.A.; investigation, A.A., A.W.T.W., M.L., J.P.M.B. and G.E.; data curation, A.A. and G.G.; writing---original draft preparation, A.W.T.W. and A.A.; writing---review and editing, G.E., P.S., A.A. and A.W.T.W.; visualisation, A.A., A.W.T.W. and G.G.; funding acquisition, A.A., P.S. and G.E. All authors have read and agreed to the published version of the manuscript. This work was supported by the the Singapore's Children Society (AA), the 2015 NAP Start-up Grant M4081597 (GE) from Nanyang Technological University Singapore, the Ministry of Education Tier-1 Grant RG55/18 2018-T1-001-172 (GE), the Ministry of Education Tier-1 Grant RG55/15 (PS) and the Singapore Ministry of Education Social Science Research Thematic Grant (MOE2016-SSRTG-017, PS). The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results. {#behavsci-10-00114-f001} {#behavsci-10-00114-f002} behavsci-10-00114-t001_Table 1 ###### Descriptive data for pooled father--child and mother--child samples (Parents), for the father--child sample (Father), and for the mother--child sample (Mother). Notes. *M* (*SD*) = mean (standard deviations). Parents Father Mother ---------------------------------------- -------------- --------------- --------------- Age 36.34 (4.27) 38.07 (3.67) 34.82 (4.22) Child's Age 42.10 (5.67) 42.21 (5.25) 42.00 (6.09) Parental Bonding Instrument (PBI) Maternal^g^ Care 23.40 (7.53) 25.21 (6.88) 21.82 (7.83) Maternal^g^ Overprotection 12.90 (5.74) 13.10 (4.65) 12.73 (6.62) Paternal^g^ Care 17.16 (8.84) 15.07 (8.35) 19.00 (8.97) Paternal^g^ Overprotection 11.73 (6.83) 9.31 (5.53) 13.85 (7.22) Parental Stress Index 87.05 (8.85) 87.93 (18.52) 86.27 (19.39) Parenting Distress 31.87 (9.24) 32.76 (9.92) 31.09 (8.68) Parent-Child Dysfunctional Interaction 23.34 (6.42) 23.41 (5.77) 23.27 (7.03) Difficult Child Score 31.84 (7.91) 31.76 (8.13) 31.91 (7.84) Emotional Availability Scales (EAS) Adult Sensitivity 5.32 (0.77) 5.38 (0.83) 5.27 (0.73) Adult Structuring 5.21 (0.70) 5.24 (0.65) 5.18 (0.76) Adult Non-Intrusiveness 5.13 (0.77) 5.52 (0.62) 4.79 ( 0.73) Adult Non-Hostility 5.82 (0.71) 5.95 (0.66) 5.71 (0.75) Child Responsiveness 5.43 (0.81) 5.57 (0.78) 5.30 (0.84) Child Involvement 5.35 (0.98) 5.62 (0.70) 5.12 (1.13) behavsci-10-00114-t002_Table 2 ###### Inter-correlation coefficients between different subscales of the Parenting Stress Index (PSI), the Parental Bonding Instrument (PBI) and the Emotional Availability Scale (EAS). Notes. *PD* = parental distress, *PCDI* = parent-child dysfunctional interaction, *DC* = difficult child, *Total PSI* = total parenting stress score, *Care~M~* = maternal care, *Overprotect~M~* = maternal overprotection, *Care~P~* = paternal care, *Overprotect~P~* = paternal overprotection, *Sensitivity* = adult sensitivity, *Structuring* = adult structuring, *Non-Intrusive* = adult non-intrusiveness, *Non-Hostile* = adult non-hostility, *Responsive* = child responsiveness, *Involved* = child involvement. \* *p* ≤ 0.05, \*\*\* *p*≤ 0.001 (Bonferroni corrected). PD PCDI DC Total PSI Care~M~ Overprotect~M~ Care~P~ Overprotect~P~ Sensitive Structuring Non-Intrusive Non-Hostile Responsive Involved ---------------- ------------- ------------- ------------- ----------- --------- ---------------- --------- ---------------- ------------- ------------- --------------- ------------- ------------- ---------- PD \- \- \- \- \- \- \- \- \- \- \- \- \- \- PCDI 0.48 \* \- \- \- \- \- \- \- \- \- \- \- \- \- DC 0.32 0.61 \*\*\* \- \- \- \- \- \- \- \- \- \- \- \- Total PSI 0.79 \*\*\* 0.83 \*\*\* 0.78 \*\*\* \- \- \- \- \- \- \- \- \- \- \- Care~M~ −0.27 −0.24 −0.05 −0.24 \- \- \- \- \- \- \- \- \- \- Overprotect~M~ 0.26 0.48 0.27 0.40 −0.10 \- \- \- \- \- \- \- \- \- Care~P~ −0.18 −0.04 0.07 −0.08 0.38 0.01 \- \- \- \- \- \- \- \- Overprotect~P~ 0.12 0.43 0.21 0.30 −0.19 0.80 \*\*\* 0.16 \- \- \- \- \- \- \- Sensitive −0.01 −0.13 −0.09 −0.09 −0.09 0.11 −0.09 −0.01 \- \- \- \- \- \- Structuring −0.05 −0.21 −0.13 −0.15 0.02 0.12 −0.01 0.04 0.72 \*\*\* \- \- \- \- \- Non-Intrusive 0.21 0.16 0.12 0.21 −0.09 0.11 −0.14 −0.12 0.33 0.18 \- \- \- \- Non-Hostile −0.03 −0.26 −0.11 −0.15 0.06 0.17 −0.14 0.00 0.76 \*\*\* 0.61 \*\*\* 0.20 \- \- \- Responsive 0.15 0.04 −0.01 0.08 −0.10 0.26 −0.09 0.05 0.58 \* 0.47 0.53 0.36 \- \- Involved 0.08 −0.03 0.00 0.03 −0.03 0.17 0.04 −0.05 0.39 0.32 0.46 0.22 0.76 \*\*\* \- [^1]: These authors contributed equally to this work.

1. Introduction {#sec1-jfb-09-00008} =============== Building an electronic device out of single molecules represents the ultimate miniaturization of active electronic components. Not only could such molecular-scale devices conceivably increase circuit density by thousands of folds compared to today's state of the art silicon-based technologies, but the unique geometrical properties of molecules also add overwhelming freedom and flexibility for the design of functions not possible with conventional semiconductors. Since the initial proposal of this attractive idea by Aviram and Ratner in 1974 \[[@B1-jfb-09-00008]\], a large body of research has been stimulated to probe charge transport (CT) through individual molecules both theoretically and experimentally \[[@B2-jfb-09-00008],[@B3-jfb-09-00008],[@B4-jfb-09-00008],[@B5-jfb-09-00008],[@B6-jfb-09-00008],[@B7-jfb-09-00008],[@B8-jfb-09-00008],[@B9-jfb-09-00008],[@B10-jfb-09-00008],[@B11-jfb-09-00008]\]. Thanks to rapid development of experimental techniques in recent years \[[@B3-jfb-09-00008],[@B4-jfb-09-00008],[@B12-jfb-09-00008],[@B13-jfb-09-00008]\], researchers are now able to reliably and repeatedly wire an individual molecule between two metallic electrodes and study their CT properties. Recent experimental advances include the demonstration of conductance switching \[[@B11-jfb-09-00008],[@B14-jfb-09-00008],[@B15-jfb-09-00008],[@B16-jfb-09-00008]\], rectification \[[@B9-jfb-09-00008],[@B17-jfb-09-00008],[@B18-jfb-09-00008],[@B19-jfb-09-00008]\], negative differential conductance \[[@B20-jfb-09-00008],[@B21-jfb-09-00008]\], and other promising phenomena beyond simple electron transport, such as quantum interference \[[@B22-jfb-09-00008],[@B23-jfb-09-00008],[@B24-jfb-09-00008]\], thermoelectricity \[[@B25-jfb-09-00008],[@B26-jfb-09-00008]\], optoelectronics \[[@B27-jfb-09-00008],[@B28-jfb-09-00008],[@B29-jfb-09-00008]\], spintronics \[[@B30-jfb-09-00008],[@B31-jfb-09-00008]\] and Peltier cooling \[[@B32-jfb-09-00008]\]. DNA, known as the reservoir for genetic information, has recently caught enormous attention in the field of molecular electronics. This is mainly because the overlapping electronic orbitals of the stacked DNA bases together with the ability to program both sequence and length in a nearly endless number of combinations make DNA an excellent one-dimensional system for CT studies ([Figure 1](#jfb-09-00008-f001){ref-type="fig"}). In addition, DNA's ability to form error-free self-assembled nanostructures, such as DNA origami \[[@B33-jfb-09-00008],[@B34-jfb-09-00008],[@B35-jfb-09-00008]\], with no need for expensive microfabrication technologies renders it a leading candidate to be exploited as a template for assembling nanoscale integrated circuits. In the meantime, probing CT through DNA offers opportunity to answer questions associated with biological processes, such as oxidative damage of DNA \[[@B36-jfb-09-00008],[@B37-jfb-09-00008]\]. These unique properties of DNA hold huge potential for the development and application of future real-life molecular electronic devices and biosensors ([Figure 2](#jfb-09-00008-f002){ref-type="fig"}), and it has therefore attracted many researchers, applying a variety of techniques to DNA CT studies, in the past two decades. Investigation of the possibility of electric conduction in DNA was first raised by Eley and Spivery in 1962 \[[@B38-jfb-09-00008]\]. In the following years, Barton and coworkers \[[@B39-jfb-09-00008]\] pioneered electron transfer through DNA using photoelectron transfer systems where electron acceptor and donor molecules were introduced into a short DNA sequence. In their studies, CT through DNA were measured by injecting electrons from one transition metal intercalated site along a DNA duplex chain and collecting the outputting current from another site that is a few base pairs (bps) away. Details about this method can be found in several excellent reviews (see refs. \[[@B39-jfb-09-00008],[@B40-jfb-09-00008],[@B41-jfb-09-00008],[@B42-jfb-09-00008]\]). However, understanding charge transport through a structure where an individual DNA molecule is wired to two metallic electrodes represents the most fundamental step towards achieving the goal of building DNA-based molecular circuits. The experimental attempts to measure the electrical conductance of such nanoscopic structure started from the 1990s. However, confusion emerged in the early days. Due to poor control and understanding of experimental conditions, the experimental results were all over the map. A wide range of CT properties of DNA have been reported, including semiconductor \[[@B7-jfb-09-00008]\], superconductor \[[@B43-jfb-09-00008]\], and insulator \[[@B44-jfb-09-00008],[@B45-jfb-09-00008]\] by different groups using different experimental approaches. Most of these results were perhaps mainly due to different measurement conditions and methods which made precise interpretation of DNA's conformation difficult and sometimes even induce contaminations \[[@B40-jfb-09-00008]\]. Thanks to recent development of single-molecule break junction techniques and other detailed measurements, concensus has been reached that a short DNA molecule is a one-dimensional conductor and can be used as molecular wires since the frontier orbitals of the bases, the highest occupied molecular orbital (HOMO) and the lowest unoccupied molecular orbital (LUMO), are π-electrons residing orbitals perpendicular to the molecular plane \[[@B40-jfb-09-00008],[@B46-jfb-09-00008],[@B47-jfb-09-00008]\]. Substantial progress has been made to fabricate DNA-based single-molecule junctions and characterize their CT properties in the past two decades. In this review, we aim to provide an overview of recent experimental advances in probing charge transport through various DNA molecules. Special focus will be given to single-molecule conductance and *I--V* characteristics of DNA. We also introduce works that probed other interesting physical properties of DNA, such as piezoelectricity, thermoelectricity and spintronics. Potential future directions and outstanding challenges will also be discussed. 2. Experimental Approaches towards DNA Single-Molecule Conductance {#sec2-jfb-09-00008} ================================================================== To determine the electrical conductance of an individual molecule such as DNA, it is necessary to wire the target molecule to two probing electrodes in a reliable and repeatable manner. However, this had remained very challenging until the introduction of scanning probe microscopy (SPM) techniques, i.e., scanning tunneling microscopy (STM) and atomic force microscopy (AFM), into the field of molecular electronics. SPM techniques have been therefore regarded as the major driving force for molecular electronics because it has made great contribution to the fabrication and modulation of single-molecular junctions and will continue to promote advances of molecular electronics in the future. In 2003, Xu and Tao \[[@B4-jfb-09-00008]\] demonstrated the STM break junction (STM-BJ) method which for the first time enabled repeatable formation of metal-molecule-metal junctions in a very reliable fashion ([Figure 3](#jfb-09-00008-f003){ref-type="fig"}a). In this method, a metal STM tip, the movement of which is precisely controlled by a piezoelectric transducer, is repeatedly driven in and out of contact with the substrate electrode adsorbed with sample molecules. The molecules have chance to form metal-molecule-metal junctions by bridging both the tip and the substrate electrodes when the tip is brought close enough to the substrate. The tip is then withdrawn away from the substrate to break the metal-molecule-metal junctions. By accurately controlling the movement of the tip, the number of bridged molecules can be changed and a single molecule junction can form during the tip retraction process. By collecting thousands of conductance vs. tip displacement traces during tip retraction, a conductance histogram can be then constructed to reveal well-defined peaks at integer multiples of a fundamental conductance value. This histogram-based method enables one to identify the conductance of a single molecule. Nowadays, this STM-BJ technique has turned out to be the most widely-used experimental platform to measure the conductance of various single molecules, including DNA \[[@B48-jfb-09-00008],[@B49-jfb-09-00008],[@B50-jfb-09-00008]\]. As an alternative, the conductive probe AFM break junction (CPAFM-BJ) method, a technique closely related the STM-BJ, uses a metal-coated AFM tip as the tip electrode and involves a laser-reflection controlled force signal detector, which enables measurements of conductance and force in parallel ([Figure 3](#jfb-09-00008-f003){ref-type="fig"}b) \[[@B52-jfb-09-00008],[@B53-jfb-09-00008]\]. Unlike STM-BJ which uses current to control tip-positioning, CPAFM-BJ uses force to control tip-positioning. While a similar conductance histogram can be constructed to determine single-molecule conductance, a corresponding force histogram can also be plotted to identify the force features associated with the formation and rupture of molecular junctions. Besides, by parking a metal-coated AFM tip on top of a self-assembled monolayer (SAM) of target molecules controlled with a setpoint contact force, CPAFM method allows one to measure current through a monolayer of hundreds of molecules \[[@B12-jfb-09-00008],[@B54-jfb-09-00008],[@B55-jfb-09-00008]\]. In addition, with the capability to image surface profile of the sample, STM- and AFM-based techniques are also used to target molecules that are custom-modified on the substrate surface, for instance, nanoparticle capped DNA molecules \[[@B51-jfb-09-00008]\] as illustrated in [Figure 3](#jfb-09-00008-f003){ref-type="fig"}c. This can be achieved by imaging the sample surface prior to the electrical measurements. Another method to construct metal-molecule-metal junctions is called mechanically controlled break-junction (MCBJ) technique. This method was first used by Reed et al. \[[@B3-jfb-09-00008]\] to measure the properties of single molecules in 1997. MCBJ uses notched metal wire fixed onto an elastic substrate \[[@B56-jfb-09-00008],[@B57-jfb-09-00008]\]. A diagram of the MCBJ is shown in [Figure 3](#jfb-09-00008-f003){ref-type="fig"}d. The substrate is usually covered with an insulator, and the metal wire is mechanically broken by bending the substrate with a pushing rod from the bottom side of the substrate. A single-molecule junction is formed when only one molecule is left in the gap between the two terminals of the broken metal wire. This method has been widely-adopted in molecular electronics because it allows one to build a three-terminal molecular transistor by introducing a gate electrode into the two-terminal metal-molecule-metal system \[[@B25-jfb-09-00008],[@B50-jfb-09-00008],[@B57-jfb-09-00008]\]. Instead of using metal electrodes, carbon-based electrode materials, such as single-walled carbon nanotube (SWCNT) \[[@B11-jfb-09-00008],[@B13-jfb-09-00008],[@B58-jfb-09-00008]\] and graphene \[[@B59-jfb-09-00008],[@B60-jfb-09-00008]\], have also been exploited to sandwich single molecules via creating nanoscale gap and chemically connecting target molecules to carbon electrodes. Guo et al. \[[@B13-jfb-09-00008]\] demonstrated the first measurement of charge transport through a SWCNT-DNA-SWCNT junction in 2008 ([Figure 3](#jfb-09-00008-f003){ref-type="fig"}e). In their work, a nanogap with a controllable size can be fabricated through a combination of the electrical breakdown and the electron beam-induced decomposition (EBID). A wide gap was firstly formed by passing a high-density current through a SWCNT followed by exposure to organic vapor. Then an electron beam with an appropriate energy was used to induce the EBID to regulate the gap size of the CNTs. To bind with CNT electrodes, DNA was modified with amino groups at the two terminals to form an amido group with the carboxylic group on the gapped CNTs. The above-mentioned techniques for creating metal-molecule-metal junctions have been extensively adopted to probe CT properties of individual DNA molecules, including single-molecule conductance and *I--V* characteristics of DNA. These techniques have been reviewed in several outstanding review papers \[[@B49-jfb-09-00008],[@B50-jfb-09-00008],[@B56-jfb-09-00008],[@B61-jfb-09-00008],[@B62-jfb-09-00008]\], and will not be revisited in detail here. 3. Charge Transport through Native DNA Molecules {#sec3-jfb-09-00008} ================================================ CT through native DNA has been the main stream of experimental investigations in DNA-based molecular electronics in the past two decades. This section gives an overview of some of the most representative experimental works that have advanced our understanding of DNA CT and paved pathways for future studies. 3.1. Environmental Effect: Wet vs. Dry {#sec3dot1-jfb-09-00008} -------------------------------------- The experimental discrepancies in DNA conductivity measurements \[[@B7-jfb-09-00008],[@B45-jfb-09-00008],[@B46-jfb-09-00008]\] at the intersection of the century were thorny enough to trigger additional investigations of the experimental conditions where DNA conductivity was measured. The variances in the experimental results formed the basis of numerous theoretical calculations which aimed at answering questions such as what is the role of water molecules and counter ions surrounding DNA strands \[[@B40-jfb-09-00008],[@B63-jfb-09-00008],[@B64-jfb-09-00008],[@B65-jfb-09-00008]\]. In 2000, Tran et al. \[[@B45-jfb-09-00008]\] measured the contactless ac conductance of λ-DNA in a buffer solution, which showed that the conductance of the DNA in solution was an order of magnitude higher than the conductance in the dry state. This, in principle, is consistent with multiple theoretical predictions \[[@B63-jfb-09-00008],[@B64-jfb-09-00008]\] which suggested that DNA has better base pair (bp) stacking in wet conditions as it maintains B-DNA compared to that in dry conditions where it is in A-DNA form. In addition, it was also predicted that positive protons or metal counterions are necessary to neutralize and stabilize DNA. Water plays a crucial role as the hydrophobic forces make DNA form a double helix, and the polarity of the water molecules helps screen DNA's charges \[[@B63-jfb-09-00008]\]. It has been suggested that better base-pair stacking and coupling and stable DNA conformation are key to DNA conductivity and it is preferable for them to be formed and maintained in aqueous conditions. To meet the quest for high conductance of DNA, later experiments were therefore mostly performed in aqueous solutions containing metal counter ions. 3.2. Single Strand and Mismatched DNA {#sec3dot2-jfb-09-00008} ------------------------------------- Single strand DNA: many DNA molecules studied in the field of molecular electronics consist of two self-complementary single strand (ss) DNA. It is, therefore, important to first understand CT through ssDNA because it helps to not only interrogate the possibility of using ssDNA as a molecular wire but also understand the potential effect of uncoupled ssDNA on CT measurements focusing on double strand (ds) DNA. In 2005, Cohen et al. \[[@B51-jfb-09-00008]\] studied charge transport through a monolayer formed by both double strand (ds) DNA and single strand (ss) DNA using CPAFM method ([Figure 4](#jfb-09-00008-f004){ref-type="fig"}). In their work, it was found that negligible current was observed for ssDNA monolayer, which indicated that single strand DNA monolayer is nearly insulating as it lacks the π-electron stacking structure of ds DNA which serves as the major pathway for charge migration. It was later confirmed in other STM-BJ measurements that ssDNA has a conductance several orders of magnitude lower than that of the corresponding dsDNA \[[@B9-jfb-09-00008]\]. Similar results were reported using STM I(t) method \[[@B66-jfb-09-00008]\]. Nowadays, ssDNA is usually considered as an insulator and detrimental to building molecular circuits despite its versatile functions in facilitating DNA self-assembled nanostructures. Mismatched DNA: Base pair (bp) mismatch can occur in dsDNA as the consequence of errors in replication or mutations in base-stacks, which have been considered the major cause of genetic diseases such as cancer \[[@B67-jfb-09-00008],[@B68-jfb-09-00008]\]. While biologists invest tremendous efforts to repair mismatched bps, experimentalists in molecular electronics consider it a crucial step to study the influence of bp mismatch on the conductance of DNA before real use of DNA as a molecular wire. In 2005, Hihath et al. \[[@B69-jfb-09-00008]\] measured the conductance of a series of dsDNA molecules with different type of mismatched bps using STM-BJ technique, as shown in [Figure 5](#jfb-09-00008-f005){ref-type="fig"}a. The conductance measurement results ([Figure 5](#jfb-09-00008-f005){ref-type="fig"}b) showed that the alteration of a single base in the stack can either increase or decrease the conductivity of the dsDNA helix, depending on the type of the mismatched. For example, while switching a T-A bp to T-G bp only increased the conductance of the DNA by 50%, changing a well-matched C-G bp to C-A can decrease its conductance by over an order of magnitude. Both the electronic states of each base and the structural instability due to a mutated base were suggested to be responsible for the conductance change in the molecule. It should be noted that the conductance change reported in this work may vary if the sequence and length of the studied DNA are altered. However, the quantifiable change in conductance of the molecule caused by a single-base mutation offer opportunity to electronically identify bp mutation in DNA. Given both ssDNA and mismatched bp in dsDNA could potentially reduce the conductance of DNA dramatically and introduce complexity into the system, research efforts in recent years have been more concentrated on well-matched DNA molecules with different sequences and bp modifications. 3.3. Sequence- and Length-Dependent CT through DNA {#sec3dot3-jfb-09-00008} -------------------------------------------------- Charge transport through double helical DNA and related charge transfer processes have received increasing interest over recent decades because of its intriguing transport properties that differentiate DNA from other molecules studied in the field of molecular electronics. For most molecules, when wired between metal electrodes, their CT is governed by either coherent tunneling, where the conductance of the molecule decreases exponentially with molecular length, or incoherent hopping, where the resistance increases linearly with length \[[@B70-jfb-09-00008]\]. However, CT through DNA has shown strong length- and sequence-dependent transport phenomena where coherent tunneling \[[@B42-jfb-09-00008],[@B47-jfb-09-00008]\], incoherent hopping \[[@B42-jfb-09-00008],[@B47-jfb-09-00008],[@B71-jfb-09-00008]\] and intermediate hopping-tunneling \[[@B72-jfb-09-00008]\] could all be the dominant transport regimes depending on the specific sequence chosen. Catalyzed by this unique trait of DNA, numerous experimental investigations have been performed using different approaches. In what follows, we will discuss some of the representative experimental discoveries that shape the current understanding of CT through DNA. Although it was indeed supported by many photoinduced charge transfer measurements \[[@B73-jfb-09-00008],[@B74-jfb-09-00008],[@B75-jfb-09-00008]\] in the early 2000s that long range transport through DNA was mediated by multistep hopping reaction involving positive charges (holes) moving between guanines (Gs), the DNA bases with lowest ionization energy, detailed experimental investigation yielded exceptions when specific DNA sequence was chosen. In 2001, Giese et al. \[[@B76-jfb-09-00008]\] experimentally studied the rate of charge transfer through G bases of DNA, separated by adenine-thymine (A:T)*n* bridges of various lengths, in double strands. As shown in [Figure 6](#jfb-09-00008-f006){ref-type="fig"}, they found that the rate of charge transfer between two guanine bases decreases exponentially with increasing separation only if the guanines are separated by no more than three (A:T) base pairs; if more bridging base pairs are present, the transfer rates exhibit only a weak distance dependence. This distinct change in the distance dependence of the rate of charge transfer through DNA to a shift from coherent superexchange charge transfer (tunneling) at short distances was attributed to a process mediated by thermally induced hopping of charges between adenine bases (A-hopping) at long distances. This work highlighted that a change in charge transfer mechanism can occur in a distance-dependent fashion. To gain deeper insight into the results obtained in the previous charge transfer measurements, researchers also interrogated the single-molecule conductance of various DNA sequences and lengths using the emerging STM-BJ technique. In 2004, Xu et al. \[[@B47-jfb-09-00008]\] measured the single-molecule conductance of a series of dsDNA molecules containing different number of alternating (AT)*~m~* in the middle of alternating (GC)*~n~* bps in aqueous solutions. The experimental results are shown in [Figure 7](#jfb-09-00008-f007){ref-type="fig"}a--d. They found that for (GC)*~n~* sequences, the conductance is inversely proportional to the length (greater than eight base pairs) ([Figure 7](#jfb-09-00008-f007){ref-type="fig"}d). Contradictory to previous findings where short range transport through molecules was governed by tunneling \[[@B42-jfb-09-00008]\], the conductance of short DNAs with (GC)*~n~* sequences decreases very slowly with the length. When inserting (A:T)*~m~* into GC-rich domains, it decreases exponentially with the length of A:T base pairs with a decay constant of 0.43 Å^−1^ ([Figure 7](#jfb-09-00008-f007){ref-type="fig"}c). It was highlighted that this decay constant of the studied DNAs is much lower than alkanes and peptides, indicating that DNA is much more conductive than alkanes and peptides. This work demonstrated different conduction mechanisms for different sequences even when the DNA is short. The results reported in this work were later confirmed in other measurements. Li et al. \[[@B77-jfb-09-00008]\] recently reported conductance measurements of DNA molecules with similar sequences but longer length. In their work, the resistance (inverse of conductance) of DNAs with A(CG)*~n~*T sequences showed a linear length-dependence behavior ([Figure 7](#jfb-09-00008-f007){ref-type="fig"}e), consistent with the results in Xu's work. In contrast, the resistance length-dependence of ACGC(AT)*~m~*GCGT/ACGC(AT)*~m~*AGCGT is significantly different. For *m* = 0, 1 and 2, the resistance increases more rapidly with length and can be best fitted with an exponential function. Within this range of molecular length, the results agree well with those in Xu's work. For *m* \> 2, however, the resistance becomes weakly length dependent, indicating a transition in the charge transport mechanism. This observation further verified the results reported in the DNA charge transfer rate measurements by Giese et al. ([Figure 6](#jfb-09-00008-f006){ref-type="fig"}). To briefly summarize at this point, it has been accepted now that the conduction mechanism of native dsDNA is dominated by tunneling (exponential dependence of conductance on length) when the G-C bps are separated by three or fewer AT bps. If the number of A-T bps is increased, diffusive hopping (linear dependence of conductance on length) becomes the main transport mechanism. However, for (GC)*~n~* sequences, the conductance of DNA is usually governed by thermally activated hopping with G bases as the hopping sites. However, it was recently found that the hopping model no longer held when the alternating G is changed to stacked G sequences. This was reported in a recent study by Xiang et al. \[[@B72-jfb-09-00008]\] In their work, single-molecule conductance of DNA duplexes with alternating G and stacked G sequences in different lengths were measured using STM-BJ method ([Figure 8](#jfb-09-00008-f008){ref-type="fig"}). It was found that in good agreement with previous results, the resistance of alternating G (black dot in [Figure 8](#jfb-09-00008-f008){ref-type="fig"}) showed a linear growth with the increase of molecular length and the results can be described with the hopping model. In contrast, the resistance of the stacked G sequences showed a surprising periodic oscillation superimposed on the linear length dependence. With the assist of theoretical simulations based on Buttiker theory, this oscillation phenomenon is attributed to a partially coherent and partially hopping regime of charge transport. The calculations revealed that the HOMOs in the stacked G are delocalized over several G bases, supporting the observation of an intermediate coherent tunneling and incoherent hopping charge transport mechanism. A more detailed follow-up theoretical investigation \[[@B78-jfb-09-00008]\] further suggested that coherence dominates charge transport for odd-n sequences, whereas incoherent hopping dominates even-n sequences. The CT in odd-length sequences cannot be dominated by incoherent transport because polaron formation in incoherent CT will stabilize the polaron states and remove the mid-gap levels, which would destroy the energy distinction between odd- and even-length sequences. As such, resistance oscillations would disappear. When *n* ≥5, the block delocalization is expected to be disrupted by thermal fluctuations, oscillations are expected to be weak and the resistance is expected to become Ohmic. 3.4. Structure-Dependent Transport in DNA {#sec3dot4-jfb-09-00008} ----------------------------------------- DNA molecules have proven to exhibit surprising conformational versatility, while retaining remarkable precision and uniformity \[[@B79-jfb-09-00008]\]. It can adopt different conformations in solution depending on the sequence and environment. The best characterized form of DNA is the right-handed, B-form, helix, which is normally adopted by dsDNA in physiological conditions \[[@B80-jfb-09-00008]\]. We note that the conformation of DNA molecules described earlier in the present review is in B-form. DNA can also adopt other structures, including a right-handed but more compact helix known as the A-form and a left-handed helix known as Z-form. It has been reported while the presence of ethanol in addition to alkaline metal ions in solution may cause a right-handed internal switch from B- to A-DNA \[[@B81-jfb-09-00008],[@B82-jfb-09-00008]\], increasing the concentration of alkaline metal ions in the solution could shift the right-handed B-DNA to a left-handed Z-DNA \[[@B83-jfb-09-00008],[@B84-jfb-09-00008],[@B85-jfb-09-00008]\]. Apart from the fact that the conformational transition of DNA from one to another is of biological and medical significance \[[@B86-jfb-09-00008],[@B87-jfb-09-00008],[@B88-jfb-09-00008]\], understanding its influence in CT properties has remained a central question in molecular electronics as it has been suggested as one of the underlying reasons for the large variation in early DNA conductance measurements. Recently, researchers explored both the influence of DNA's structural transition from B- to Z-form and from B- to A-form on its conductance using STM-BJ method. B-Z transition: using STM-BJ method, Wang et al. \[[@B79-jfb-09-00008]\] studied the conductance change of a 8 bp poly(GC)~4~ DNA duplex induced by a structural transition from B- to Z-form ([Figure 9](#jfb-09-00008-f009){ref-type="fig"}a). In their work, the B- to Z-DNA transition was facilitated by increasing the concentration of Mg^2+^ ions in the buffer solution from 0 M to 4 M. The structural change was monitored and confirmed by circular dichlorism (CD) measurements ([Figure 9](#jfb-09-00008-f009){ref-type="fig"}c). Through measuring the single-molecule conductance of DNA, they found that the conductance of DNA molecule decreased by two orders of magnitude as DNA conformation transfers from B- to Z-form ([Figure 9](#jfb-09-00008-f009){ref-type="fig"}b). The cause of this conductance reduction is primarily attributed to the structural change-induced breaking of π-π orbital stacking between neighboring bps which may come from a few sources, including the rise between adjacent bps by 14%, axial angle twist from 36° in B-DNA to −30° in Z-DNA and flipping of G bases by almost 180°, which have a detrimental effect on distribution of effective orbitals for CT. Therefore, DNA in Z-form is not suggested for molecular electronics applications. Interestingly, this work also showed that when three A-T bps were bridged in the middle of the (GC)~4~ sequence, such B-Z structural transition effect was not present and therefore no significant change in DNA conductance was observed even when Mg^2+^ concentration in the solution increased from 0 M to 4 M. It was also highlighted that the B-Z transition can be monitored simply by measuring the conductance of the DNAs even when traditional method such as CD fails ([Figure 9](#jfb-09-00008-f009){ref-type="fig"}d). B-A transition: change in conductance of DNA induced by a B-A structural transition was experimentally explored by Artes et al. \[[@B89-jfb-09-00008]\] It should be noted that before this work the base alignment in A-form was expected to be worse than that in B-form, which led to the prediction that A-form is insulating \[[@B63-jfb-09-00008]\]. However, the results of this work showed that the conductance of DNA duplexes with a GC rich sequence of 5′-CCCGCGCGCCC-3′ increases by approximately one order of magnitude when its conformation changed from B-form to A-form ([Figure 10](#jfb-09-00008-f010){ref-type="fig"}). The B- to A-form transition was achieved by changing the solution from phosphate buffer to 80% ethanol. This conductance increase by one order of magnitude was also observed in DNA molecules with longer lengths. They also found that this large conductance increase is fully reversible, and by controlling the environment, the conductance can be repeatedly switched between the two values. However, length-dependent conductance studies of the two conformations suggest that neither tunneling nor simple hopping dominate the charge transport processes in these guanine-rich sequences. ab initio electronic structure calculations coupled with calculations of the electronic density of states of the two conformations revealed that the HOMO, the dominant transport orbital, is distributed over 70% of molecular length in the A-form, but only 50% in the B-form and that the A-form transmission function is orders higher than B-form transmission, therefore leading to a higher conductance of A-form DNA. By directly demonstrating the conformational change plays a significant role in the transport properties, the results of these studies also help rationalize the large dispersion of DNA conductivity found in the literature. Effect of stretching: one important structural perturbation dsDNA duplex may undergo is being stretched, leading to a dramatic structural change. Indeed, it has been shown that DNA undergoes a structural transition when mechanically stretched \[[@B90-jfb-09-00008],[@B91-jfb-09-00008],[@B92-jfb-09-00008]\]. This has been attributed to either a reversible configuration change from native form (B-DNA) to stretched form (S-DNA) or irreversible force-induced melting \[[@B93-jfb-09-00008],[@B94-jfb-09-00008],[@B95-jfb-09-00008],[@B96-jfb-09-00008],[@B97-jfb-09-00008]\]. As charge transport in dsDNA molecules is mediated by the π-π stacking interactions between neighboring base pairs, mechanically stretching DNA is believed to seriously disrupt the π-π stacking interactions and therefore leads to a large change in CT of DNA \[[@B98-jfb-09-00008]\]. Bruot et al. \[[@B98-jfb-09-00008]\] recently reported their experimental study of the effect of mechanical stretching on DNA conductance ([Figure 11](#jfb-09-00008-f011){ref-type="fig"}). Using STM-BJ method, they explored the effect of the stretching transition on dsDNA conductance by analyzing the evolution of single-molecule conductance during electrode separation. dsDNA molecules (5′-A(CG)~N~-3′, N = 2\~12) with lengths varying from 6 bps (\~2 nm) to 26 bps (\~9 nm) were studied in aqueous environment. It was found that although the single-molecule conductance of studied DNA molecules showed linear length dependence ([Figure 11](#jfb-09-00008-f011){ref-type="fig"}c), consistent with previous results, CT of DNA is highly sensitive to mechanical stretching ([Figure 11](#jfb-09-00008-f011){ref-type="fig"}b), showing an abrupt decrease in conductance at surprisingly short stretching distances, which is weakly dependent on the molecular length ([Figure 11](#jfb-09-00008-f011){ref-type="fig"}d). These unexpected observations are attributed to a force-induced melting mechanism \[[@B99-jfb-09-00008]\] and are consistent with de Gennes' DNA ladder model \[[@B100-jfb-09-00008]\] for dsDNA mechanics. Based on de Gennes's ladder model, as shown in [Figure 11](#jfb-09-00008-f011){ref-type="fig"}a, the abrupt conductance decrease is caused by the breaking of the hydrogen bonds of the base pairs near the ends by the external shear force (\~3 pN) which decrease the coupling between the dsDNA hopping sites and the electrodes. It is worth noting that Bruto et al. reported a new molecule--electrode linker based on a hairpin-like design which could overcome this force-induced melting at the end of single DNA molecules during stretching \[[@B101-jfb-09-00008]\], opening up the possibility to construct DNA electromechanical devices. *G4-DNA*: Guanine-quadruplex (G4) DNA, an important derivative of dsDNA, has also drawn considerable attention in molecular electronics over the past decade \[[@B102-jfb-09-00008],[@B103-jfb-09-00008],[@B104-jfb-09-00008]\]. This is mainly because G4-DNA has several advantageous structural traits over dsDNA. First, given that short dsDNA with G-C rich sequence has been reported to be more conductive than A-T sequence \[[@B47-jfb-09-00008],[@B105-jfb-09-00008]\], the stacking of several G-quartets in G4-DNA is predicted to yield high conductance \[[@B106-jfb-09-00008]\]. Second, unlike the relatively flexible conformation of dsDNA, the quadruple helical conformation ensures that the G4-DNA is rather stable under physiological conditions and indicates higher stiffness and stronger resistance to surface forces than the dsDNA \[[@B102-jfb-09-00008],[@B107-jfb-09-00008]\]. Third, G4-DNA is an excellent prototype to study self-assembling properties at the supramolecular scale and the design of biomimetic systems \[[@B107-jfb-09-00008],[@B108-jfb-09-00008]\]. Due to these superior features, G4-DNA has been considered a promising candidate as a molecular wire. Direct measurement of charge transport through G4-DNA has been carried out very recently. In 2014, Livshits et al. \[[@B102-jfb-09-00008]\] reported the *I--V* measurement of a long G4-DNA using an experimental setup depicted in [Figure 12](#jfb-09-00008-f012){ref-type="fig"}a. In their work, a gold electrode was evaporated using stencil lithography on top of a long biotin-avidin (BA)--G4-DNA (\~250 nm) which had been pre-adsorbed onto a flat mica surface. By contacting the molecule with a metalized AFM tip, *I--V* curves were collected at different locations along the long G4-DNA chain. Negligible current below ±3--4 V followed by a rapid rise in the current was observed in all *I--V* curves, as shown in [Figure 12](#jfb-09-00008-f012){ref-type="fig"}b. In addition, a weak and non-trivial dependence of the *I--V* data, especially the current rise, on the distance between the measured point and the border of the electrode ([Figure 12](#jfb-09-00008-f012){ref-type="fig"}c) was observed. The obvious asymmetry of the *I--V* curves was attributed to asymmetric coupling at the metal-molecule interfaces (the contact at the evaporated electrode is much better and more reproducible than the tip-molecule contact). Conducted in a very controllable manner, their experiments showed that G4-DNA can transport significant amount of current over long distances when deposited on a hard substrate, which is the more relevant configuration for solid-state devices, and the long-range transport through G4-DNA is mediated by thermally activated hopping over stacked G-tetrad blocks. Single-molecule conductance of short G4-DNA has also been performed recently using single-molecule break junction methods. Using the MCBJ approach, Liu et al. measured the conductance of a short G4-DNA (5′-T\*G3\[TTAGGG\]3T\*)-3′, T\* is modified thymine to couple the molecule to the electrodes) which was covalently bonded to Au electrodes through thiol-Au interaction \[[@B104-jfb-09-00008]\]. In their study, a pronounced and flat conductance plateau with a length of around 2 nm was observed ([Figure 13](#jfb-09-00008-f013){ref-type="fig"}b, bottom-right panel). It was noticed the plateau resistance remains almost constant while the G-quadruplex conformation is present. *I--V* measurements performed at the molecular plateau region showed a highly non-linear, with a typical S shaped *I--V* curve of G4-DNA. Although detailed transport mechanism was unclear, this work also showed that G4-DNA can transport considerable current at reasonable voltages in the range of 1 V. More importantly, contrast to dsDNA, the resistance of the G4-DNA molecule is quite independent of the elongation of the molecule. Liu et al. recently studied the possibility of using SWCNT-G4-DNA-SWCNT junction as a protein-detection device \[[@B109-jfb-09-00008]\]. Their experimental setup involved connecting a 15-mer thrombin DNA aptamer with thymine 7 (T7) linkers on both the 3′ and 5′ termini to nanogapped SWCNTs, as shown in [Figure 13](#jfb-09-00008-f013){ref-type="fig"}c. Under a constant source-drain voltage of −15 mV, the current vs. gate voltage signal was constantly monitored when the G4-DNA aptamer was connected between the SWCNTs before and after thrombin treatment. It was found that the current of the SWCNT-G4-aptamer-SWCNT junction increased dramatically upon treatment of thrombin and this change is reversible over several cycles by alternating treatment with thrombin (2.6 fM) and guanidine HCl (6 M). To understand the drastic conductance change induced by G4-DNA interacting with thrombin, they hypothesized that DNA--thrombin interactions do not distort the G4 conformation, but instead rigidify the G4 conformation and promote tight packing, thus enhancing the conductance of the G4-DNA. Apart from measuring CT through G4-DNA, their work demonstrated how G4-DNA can be used to interface with biological processes for the development of bio-detection devices. Collectively, experimental studies discussed in this section have delivered an important message that CT through DNA is highly sensitive to the structure of DNA, and it therefore requires better understanding of the correlation between DNA's conformation and its conductance before DNA can be used in future molecular electronic devices. 4. Charge Transport through Modified DNA Molecules {#sec4-jfb-09-00008} ================================================== The ability to effectively tune the electronic structure of native DNA molecules is key to satisfying the quest for high conductance and versatile functionalities of DNA. To achieve this, experimentalists have attempted to modify native dsDNA molecules using multiple approaches, such as methylation \[[@B110-jfb-09-00008],[@B111-jfb-09-00008]\], metallo-base pair \[[@B112-jfb-09-00008]\], and small molecule intercalation \[[@B9-jfb-09-00008],[@B113-jfb-09-00008]\]. In this section, various modifications used in molecular electronics experiments and their effect on CT through DNA will be discussed. 4.1. DNA Methylation {#sec4dot1-jfb-09-00008} -------------------- Among DNA modifications, DNA methylation constitutes an essential epigenetic mechanism for various important biological processes such as embryonic development, replications, and aging \[[@B110-jfb-09-00008],[@B114-jfb-09-00008],[@B115-jfb-09-00008]\]. Beyond its significant role as a biomarker closely associated with human health and disease, DNA methylation has also been regarded as one possible method to tune the CT properties of native DNA as it is anticipated to affect the π-conjugation of the pyrimidine (or purine) bases considering the electron-donating character of methyl substitutions \[[@B110-jfb-09-00008]\]. In 2011, Tsutsui et al. \[[@B110-jfb-09-00008]\] reported their experimental investigation of the effects of methylation on the conductance of deoxycytidine-5′-monophosphate. The studied DNA 4-mers have sequences of 3′-mCGmCG-5′ and 3′-GGGG-5′, where cytosine bases were methylated, as shown in [Figure 14](#jfb-09-00008-f014){ref-type="fig"}a. The conductance measurements were carried out using MCBJ method under a constant bias of 0.4 V. It is worth noting that the measurements were performed transverse to a ssDNA rather than longitudinally as in most of the other papers discussed. Upon statistically collecting the electrical signal when a DNA oligomer was trapped between Au electrodes, they found that methyl substitution contributes to increase the tunneling conductance of deoxycytidines ([Figure 14](#jfb-09-00008-f014){ref-type="fig"}b). This conductance increase was attributed to a shift of the highest occupied molecular orbital level closer to the electrode Fermi level by methylation. This result also suggested a possible use of the transverse electron transport method for a methylation level analysis. However, Hihath et al. \[[@B111-jfb-09-00008]\] reported a different observation when studying the effect of the methylation of cytosine bases on DNA conductance. Using the STM-BJ method, they measured the conductance of a poly(GC)~4~ DNA duplex before and after methylation of its cytosine bases ([Figure 14](#jfb-09-00008-f014){ref-type="fig"}c). The measurement results showed the methylated DNA has a lower conductance than the its native counterpart even though the methylated DNA is more stable and has cytosine bases with a lower energy gap. They reasoned that since charge transport through GC rich sequence is dominated by hopping through guanine bases, methylation of cytosine bases will have little effect on the energy levels of guanine itself and it may not imply a better coupling between bases in the DNA stack. The difference between the work by Tsutsui et al. and the work by Hihath et al. could be reasonable because in Tsutsui's work, charge transports in the lateral direction across the base pair, which is perpendicular to the axis of DNA. It is, therefore, more sensitive to the methylation. However, accurate understanding of these experimental discrepancies will require more detailed investigation of methylation status of individual cytosine bases within the stack of the specific DNA sequence. However, these studies suggested that DNA methylation has a clear impact on the CT properties of DNA. 4.2. Metallo-DNA {#sec4dot2-jfb-09-00008} ---------------- Metals are the carriers of conducting electrons which are particularly desired in the field of molecular electronics. It was therefore envisioned that doping the interior of oligonucleotide duplexes with metal ions may yield hybrid materials with enhanced conductivity or other interesting electronic effects \[[@B116-jfb-09-00008]\]. Toward this direction, extensive research efforts have been put to both covalently attach metal complexes to native DNA and create "metal-base pair" by replacing the hydrogen-bonded Watson--Crick base pairs with metal--ligand interactions inside the DNA double helix \[[@B117-jfb-09-00008]\]. We refer readers to other excellent reviews that cover the related topics (see refs. \[[@B116-jfb-09-00008],[@B117-jfb-09-00008]\]). The possibility of using metal-DNA as electronic device was first investigated theoretically: Lee and co-workers presented a model system for a field-effect transistor based on metal-DNA \[[@B118-jfb-09-00008]\]. They proposed that a gate voltage applied perpendicular to the helix axis will cause displacement of the metal ions (an ionic Stark effect) and induce differences in the site energies which will consequently modulate the hopping rate. However, Joseph et al. \[[@B119-jfb-09-00008]\] studied the effect of a T--Hg--T base pair (see below) on the long-distance radical cation hopping properties but found no significant effect of this metal-base pair on the charge transport. Since then, more efforts were invested to explore the optimal conditions for metal ions to couple with natural DNA base pairs. Due to technical difficulties, it remains challenging to study metal-DNA in a metal-DNA-metal single-molecule junction system. However, taking advantage of the SWCNT junction system, Liu et al. recently reported current measurement of metallo-DNA duplex by chemically sandwiching it between two SWCNT electrodes ([Figure 15](#jfb-09-00008-f015){ref-type="fig"}) \[[@B112-jfb-09-00008]\]. A stable metal-mediated base pair in the presence of a Cu^2+^ ion (H--Cu^2+^--H), a motif geometrically similar to the hydrogen-bonded natural base pairs, were formed at the middle of their studied DNA sequence, as shown in [Figure 15](#jfb-09-00008-f015){ref-type="fig"}b. It can also be removed after ethylenediaminetetraacetic acid (EDTA) treatment. Under a constant source-drain bias, results of the current vs. gate voltage measurements showed a sharp increase in current upon metal-base pair formation. Similar phenomena were observed when multiple metal ions were incorporated. They proposed that H--Cu^2+^--H base pairs incorporated parallel to the neighboring natural base pairs could rigidify π-stacking between DNA base pairs and mediate the electronic coupling for hole transfer, thus favoring DNA charge transport, as compared to the ligand-containing metal-free DNAs \[[@B112-jfb-09-00008]\]. This work showed the possibility to enhance the electrical conductance of DNA by rational arrangement of multiple metal ions inside the core of the DNA base-pair stack and will encourage more future investigations toward metallo-DNA molecules bridging nanodevices. 4.3. DNA-Small Molecule Complex {#sec4dot3-jfb-09-00008} ------------------------------- It has been shown that CT properties of π-stacked aromatic molecules can be modulated when donor or acceptor molecules were present. For example, rectification behavior was observed for the stack composed of electron donor and acceptor molecules (triphenylene and naphthalenediimide, respectively), whereas such behavior was absent in the stack composed solely of donor molecules \[[@B120-jfb-09-00008]\]. Since π-π stacking interactions mediate the transport in both the aromatic molecule and DNA systems, the electron transport properties of DNA could thus be analogously tuned if the π-π stacking between DNA bases could be modulated. One efficient way to achieve such modulation is by intercalating small π-conjugated molecules. Experimental investigations toward this direction were recently reported. In 2016, Guo et al. \[[@B9-jfb-09-00008]\] studied the effect of intercalating coralyne molecule ([Figure 16](#jfb-09-00008-f016){ref-type="fig"}a) to a custom-designed DNA duplex (5′-CGCGAAACGCG-3′) on its CT properties. They found that two coralyne molecules can stably intercalate into one DNA duplex, forming a DNA-coralyne complex structure which yields strong diode-like *I--V* behavior. This work will be detailed in the next section. Similarly, Harashima et al. \[[@B113-jfb-09-00008]\] recently investigated how ethidium bromide (EB) or Hoechst 33258 (HOE), affects the electron transport of a single DNA molecule by means of the STM-BJ technique ([Figure 16](#jfb-09-00008-f016){ref-type="fig"}b). While EB has been known to intercalate between stacked DNA base pairs \[[@B121-jfb-09-00008]\], HOE binds to the minor groove of the DNA helix and has limited effects on the helix structure \[[@B122-jfb-09-00008]\]. They found that the conductance of DNA can be enhanced by over four folds upon EB binding due to the decreased gap between the HOMO of the DNA and the Fermi level of the electrode. It was also found that the single-molecule conductance remains almost unaffected by a groove-binding molecule, HOE, because of the absence of the interaction between the ligand and stacked bases. Results of these studies strongly suggest tuning π-π stacking of DNA bases via intercalating π-conjugated molecules as a very effective way of achieving promising functions of DNA. 5. Toward DNA Molecular Diode and Transistor {#sec5-jfb-09-00008} ============================================ The goal of molecular electronics is to functionally incorporate molecular components in electronic devices. The key to achieving this goal is to search for molecular candidates that mimic the electronic behavior of conventional semiconductors, such as diode (rectifier), which facilitates current flow in one (forward) bias direction, and transistor, current flow of which can be controlled with a gate voltage. To date, the diode-like (rectification) behavior has been observed in molecular junction devices involving small organic molecules, the structures of which usually comprise electron-donor and electron-acceptor groups \[[@B123-jfb-09-00008],[@B124-jfb-09-00008],[@B125-jfb-09-00008],[@B126-jfb-09-00008]\]. The cause of such rectification behavior has been attributed to two main factors, asymmetric electronic structures of the molecular core \[[@B124-jfb-09-00008],[@B127-jfb-09-00008]\] and asymmetric coupling at the two molecule-electrode contact interfaces \[[@B128-jfb-09-00008],[@B129-jfb-09-00008]\]. For details of this topic, we refer readers to several excellent reviews (see refs. \[[@B50-jfb-09-00008],[@B127-jfb-09-00008],[@B128-jfb-09-00008]\]). However, research using DNA as a molecular diode has been lagging behind due to poor understanding of its structure-property relation. There have been few experimental attempts until very recently. The first successful demonstration of a DNA-based molecular rectifier was reported by Guo et al. in their recent publication \[[@B9-jfb-09-00008]\]. They created a DNA-based rectifier by intercalating two coralyne molecules into specifically designed duplex DNA molecule (5′-CGCGAAACGCG-3′) ([Figure 17](#jfb-09-00008-f017){ref-type="fig"}). The three mismatched A-A bps at the middle of the sequence facilitated the intercalation of coralyne molecules into the DNA, which turned out to significantly modulate the electronic structure of the treated DNA ([Figure 17](#jfb-09-00008-f017){ref-type="fig"}a,b). Electrical measurements of single DNA-coralyne complex molecular junctions using STM-BJ method revealed rectifying *I--V* characteristics with a high rectification of around 15 at 1.1 V ([Figure 17](#jfb-09-00008-f017){ref-type="fig"}c,d). It was noted that this is a completely counterintuitive finding considering the apparent structural symmetry of the DNA-coralyne complex. Therefore, an unprecedented transport mechanism for molecular rectification was proposed. In the new mechanism, the rectification behavior was caused by the coralyne-induced local spatial asymmetry of the distribution of electron states along the DNA chain, which results in an asymmetric change in the transmission function associated with the HOMO-1 level of the molecule. This work offered a new strategy for engineering molecular electronic elements by exploiting DNA-small molecule interaction and will ignite new interest in DNA-based functional molecular electronic devices. Creating a molecular field-effect transistor (FET) requires incorporating the third gating electrode into a two-terminal metal-molecule-metal junction system. A gate electrode can be fabricated aside a two-terminal molecular junction using the MCBJ and electromigration break-junction setups (EBJ) \[[@B25-jfb-09-00008],[@B50-jfb-09-00008],[@B57-jfb-09-00008]\]. However, apart from the fact that these methods are limited in practice by low device yield, \[[@B50-jfb-09-00008]\] measurements using these setups are often conducted in air or vacuum, which is detrimental for achieving high conductance in DNA molecules. To satisfy the requirement of aqueous environment for measurements involving DNA, STM-BJ setup proves to be ideal because it allows one to avoid leaking current between the metallic source and drain electrodes induced by counter ions in the buffer solution by coating a layer of insulating material, such as apiezon wax, onto the STM tip electrode \[[@B9-jfb-09-00008],[@B47-jfb-09-00008],[@B72-jfb-09-00008]\]. More importantly, the promising electrochemical (EC) gating method can be incorporated into STM-BJ setup to achieve efficient gate coupling that manipulates the energy alignment and the molecular redox processes for a single-molecule junction. Using EC gating method ([Figure 18](#jfb-09-00008-f018){ref-type="fig"}a), Xiang et al. \[[@B130-jfb-09-00008]\] recently reported their conductance measurements of a modified DNA duplex molecule where one base was replaced with a redox group for optimal EC control ([Figure 18](#jfb-09-00008-f018){ref-type="fig"}b). By applying an EC gate voltage to the molecule, they showed that switching the redox group between the oxidized and reduced states leads to reversible switching of the DNA conductance between two discrete levels (high and low) ([Figure 18](#jfb-09-00008-f018){ref-type="fig"}c,d). Their theoretical calculation shows that the conductance switching arises from a change in the molecular energy alignment associated with the redox state switching \[[@B130-jfb-09-00008]\]. Their work successfully demonstrates the possibility of switching DNA conductance between two levels with an EC gate. Despite of the high complexity of molecular junctions involving DNA, the unique ability of DNA to self-assemble holds huge potential for building DNA nanochips and biosensors, which is not possible with other molecules. Combined with deeper insights gained in recent years, the promising experimental results introduced in this section have paved way for more future investigations in creating functional CT out of DNA. 6. Beyond Simple Charge Transport in DNA {#sec6-jfb-09-00008} ======================================== The desire to create functional molecular devices has pushed the frontiers of both measurement capabilities and our fundamental understanding of varied physical phenomena at the single-molecule level, including mechanics, thermoelectrics, optoelectronics and spintronics in addition to electronic transport characterizations \[[@B48-jfb-09-00008]\]. Atomic precision of single molecule devices is beyond what is achievable with many other nanomaterials. Therefore, metal-molecule-metal junction represents a powerful platform to explore a variety of physical properties beyond electron transport. This section will discuss interesting properties beyond CT achieved in DNA-based molecular junctions. 6.1. DNA Spintronics {#sec6dot1-jfb-09-00008} -------------------- The fact that electron can travel through chiral molecules in a spin selective manner raised the intriguing possibility that DNA CT is affected by the inherent spin of the electrons passing through it. It has been found that DNA duplex can act as a spin filter for spin selective transmission of electrons \[[@B131-jfb-09-00008]\]. This was demonstrated in photoemission spectroscopy of DNA monolayers on Au \[[@B132-jfb-09-00008]\] and in electrochemical charge transfer measurements on DNA monolayer films \[[@B133-jfb-09-00008]\]. Xie et al. \[[@B134-jfb-09-00008]\] recently reported a magnetoresistance measurement of DNA using a CAFM-BJ setup. In their work, a platinum-coated conductive AFM tip measures the current flowing from a nickel substrate through the DNA to a gold nanoparticle ([Figure 19](#jfb-09-00008-f019){ref-type="fig"}a). *I--V* characteristics were obtained for DNA duplex molecules with different lengths under two magnetic field polarities (up and down) (left column in [Figure 19](#jfb-09-00008-f019){ref-type="fig"}b). They found that the *I--V* curves are symmetrical with respect to zero bias, but *I--V* curve for Ni depends strongly on the magnetic field orientation. The magnetic field of the permanent magnet has little effect on the current flow through the DNA, but the magnetic field does affect the spin alignment in the Ni substrate. The lower traces in [Figure 19](#jfb-09-00008-f019){ref-type="fig"}b show a control experiment in which the Ni electrode was replaced by an Au electrode, which is non-ferromagnetic, and no magnetic field effect was observed. Since the dI/dV curves (right column of [Figure 19](#jfb-09-00008-f019){ref-type="fig"}b) reflect the density of states for each system qualitatively, carrier density can then be used extract an effective tunneling barrier height, as shown in [Figure 19](#jfb-09-00008-f019){ref-type="fig"}c. The difference in spin selectivity for the different DNA lengths enlarges because the effective barrier associated with the unfavorable spin is smaller for the shorter oligomers. Hence, electrons with this unfavorable spin produce higher conductance. Due to larger effective barriers for longer DNA molecules, the current associated with the unfavorable spin is blocked. 6.2. DNA Piezoresistivity {#sec6dot2-jfb-09-00008} ------------------------- Mechanical force-induced change in the resistivity of materials is known as piezoresistivity. This important property of materials has also been observed in single molecule junction devices \[[@B135-jfb-09-00008],[@B136-jfb-09-00008],[@B137-jfb-09-00008]\]. These behaviors arise primarily from molecule--electrode coupling effects, as opposed to distortions within the molecule causing changes of the molecular electronic states \[[@B138-jfb-09-00008]\]. Studying piezoresistivity effect in more complex molecules such as DNA and the role that distortions of the nucleic acid units play in CT is of persistent interest. In 2015, Bruot et al. \[[@B138-jfb-09-00008]\] demonstrated the first measurement of piezoelectricity in DNA molecular junctions ([Figure 20](#jfb-09-00008-f020){ref-type="fig"}). To measure conductance and piezoresistivity of single DNA molecules, they used a tip modulation STM-BJ technique where tip retraction was stopped and a 0.02 nm sinusoidal modulation (1 kHz frequency) was applied to the tip position along the axis of the tip when a molecular plateau was detected. To describe piezoresistance (α) in single DNA molecules, the amplitude of conductance at the modulation frequency was normalized by the molecular conductance (red curve in [Figure 20](#jfb-09-00008-f020){ref-type="fig"}b). A two-dimensional piezoresistance vs. molecular conductance histogram can be constructed ([Figure 20](#jfb-09-00008-f020){ref-type="fig"}c). They found that the piezoresistivity is larger for sequences with intra-strand purine stacking (G~N~C~N~) than for inter-strand purine sequences ((G-C)~N~). By investigating the length dependence of both piezoresistance and conductance, they also found that DNA piezoresistance is determined by the sensitive dependence of the electronic coupling between neighboring bases, along with the bridge site energy on mechanical force. Molecular orbital calculations showed that the piezoresistivity of DNA is caused by force-induced changes in the π-π electronic coupling between neighboring bases, and in the activation energy of hole hopping. 6.3. DNA Thermoelectricity {#sec6dot3-jfb-09-00008} -------------------------- Thermoelectric effect, a basic property of materials, is the direct conversion of temperature difference to electric voltage and can be described by an important parameter, Seebeck coefficient. This effect in nanometer-long molecules is expected to be distinctively different from that in bulk materials \[[@B139-jfb-09-00008]\]. The first measurement of Seebeck coefficient (S) of single molecules was reported by Reddy et al. \[[@B26-jfb-09-00008]\]. They showed that beyond potential energy-conversion applications, the sign of Seebeck coefficient (S) of molecular junctions can indicate the nature of charge carrier (electron vs. hole) and the relative position of electrode fermi level to the frontier molecular orbitals. Stimulated by this work, increasing experimental efforts in investigating the thermoelectric effect in molecules have been devoted over the past decade. Detailed measurement techniques and the value of S for different molecules were reviewed in two recent articles (see ref. \[[@B140-jfb-09-00008],[@B141-jfb-09-00008]\]), and will not be revisited here. Thermoelectric effect of dsDNA molecules was recently studied by Li et al. \[[@B77-jfb-09-00008]\] using STM-BJ setup. In their experiments, the STM tip was held at 295 K while the substrate was cooled from 295 to 275 K ([Figure 21](#jfb-09-00008-f021){ref-type="fig"}a). By studying multiple sequences and lengths, they measured the S of DNA in both tunneling and hopping regime. It was found that for hopping dominant sequences (5′-A(CG)*~n~*T-3′), the value of S is small (\~1 μV/K) and weakly dependent on the molecular length (black dots in [Figure 21](#jfb-09-00008-f021){ref-type="fig"}b). However, when inserting (AT)*~n~* blocks into the CG sequence, it changes both the conductance and S substantially. As shown in [Figure 21](#jfb-09-00008-f021){ref-type="fig"}b, when the number of AT block is less than four, it acts as a tunneling barrier and the S is large compared to the CG sequence and has a linear dependence in molecular length, ranging from 5 μV/K to 8 μV/K. When AT block is longer than four bps, the charge transport mechanism changes from tunneling to hopping, and S drops to smaller values (\~2 μV/K) and is weakly dependent in length. For all the sequences studied, S reveals a positive sign, indicating HOMO-dominant transport. The results demonstrated in this work implies that DNA thermoelectricity may be tuned by its length and sequence. 7. Conclusions and Outlook {#sec7-jfb-09-00008} ========================== The precision control with atomic accuracy over sub-nanometer distances is the ultimate limitation of electronic device miniaturization and is unachievable with the state of the art silicon-based technologies \[[@B50-jfb-09-00008]\]. Therefore, molecular electronics holds the greatest promise for overcoming the bottleneck of semiconductor-based technologies. The remarkable experimental and theoretical research on molecular electronics over the past two decades has exhibited promising results and significant achievement, among which DNA molecules have been playing an increasingly important role. The experiments performed on DNA-based molecular electronic devices in the past 15 years have delivered several important messages to the field that: first, short dsDNA duplex is a conductor and can be considered as a molecular wire while ssDNA and long dsDNA molecules are detrimental to facilitate CT; second, even for short DNA molecules, its CT can be dominated by completely different mechanisms depending on the specific sequence chosen, i.e., largely electron tunneling regime for AT rich sequences and hole hopping regime for GC rich sequences; third, CT through DNA is highly sensitive to its secondary structure and the surrounding ionic environment, which also means CT of DNA can be tuned by properly modulating its structure and measurement conditions; last, native dsDNA can be modified by various means, including metallization, methylation, doping, and intercalation of small molecules, which could tailor the electronic structure of DNA toward functional CT. In this review, we have also discussed recent experimental progress in DNA-based single-molecule electronics which suggest that due to its structural flexibility, diversity and programmability, DNA has indeed demonstrated many unique properties that are not possible in other molecules, such as sequence- and length-dependent transport, structural transition, strong interaction with metal ions and small molecules, and spin filter effect, These superior properties of DNA are believed to offer us unprecedented opportunities for the design and fabrication of molecular-scale devices and biosensors. However, challenges exist in several aspects. First, thorough control of DNA's structure in different environment is necessary for both narrowing down the data distribution in CT measurements and creating long-lasting and highly stable DNA molecular devices. Second, more investigations toward functional transport properties of DNA are required both experimentally and theoretically. Special efforts should be given to improve the performance of DNA devices, such as the rectification ratio of DNA molecular rectifier and on/off ratio of DNA transistor. Third, there remains plenty of room to modify native DNA in various forms, which will require close collaboration between chemists, biologists, physics and engineers. Last, there still exists open questions in the basic field of CT in DNA. The complexity of large biomolecules like DNA makes accurate simulation of the experimental conditions difficult. More precise theoretical models and calculation algorithms that better mimic the actual experimental measurements need to be developed. Beyond single molecule transport measurements, rapid development of the emerging DNA origami technologies has been refreshing our understanding of DNA's remarkable self-assembling ability. To date, both random shaped two-dimensional (2D) DNA nanopatterns \[[@B34-jfb-09-00008],[@B142-jfb-09-00008]\] and three-dimensional (3D) DNA nanoblocks \[[@B143-jfb-09-00008],[@B144-jfb-09-00008]\] can be "mass-produced" up to a scale of hundreds of nanometers in a controllable manner. There is enough reason to believe that these exciting DNA nanotechnologies will soon be applied in conjunction with single-molecule electronics to scale up DNA-based molecular circuits ([Figure 22](#jfb-09-00008-f022){ref-type="fig"}). As DNA-based molecular electronics continue to grow, one can foresee that DNA will offer a solution to many of the hurdles that need to be overcome in further scaling down electronic circuits. I thank Bingqian Xu for helpful advice and NSF for support. The author declares no conflict of interest. {#jfb-09-00008-f001} {#jfb-09-00008-f002} ![Single-molecule break-junction techniques (SMBJs). (**a**) Scanning tunneling microscope break-junction (STM-BJ) technique; (**b**) Conductive probe atomic force microscope break-junction (CPAFM-BJ) technique; (**c**) CPAFM-BJ involving nanoparticle capped DNA molecule. Reprinted with permission from ref. \[[@B51-jfb-09-00008]\]. Copyright (2005) National Academy of Sciences, USA; (**d**) Mechanically controlled break-junction (MCBJ) technique; (**e**) SWCNT-molecule-SWCNT molecular junction. Reprinted with permission from ref. \[[@B13-jfb-09-00008]\]. Copyright (2008) from Nature publishing group.](jfb-09-00008-g003){#jfb-09-00008-f003} ![(**a**) Schematic of CPAFM-based method with a nanoparticle capped DNA molecule sandwiched between AFM tip and substrate; (**b**) Measured *I--V* curves of double strand DNA was performed on a metal particle without pressing on it; (**c**) *I--V* curve measured on the ssDNA monolayer without pressing it. Negligible current was observed here, indicating that the ssDNA monolayer is insulating. Reprinted with permission from ref. \[[@B51-jfb-09-00008]\]. Copyright (2005) National Academy of Science, USA.](jfb-09-00008-g004){#jfb-09-00008-f004} ![Effect of mismatched base pair on DNA conductance. (**a**) Mismatched DNA sequences studied in the work by Hihath et al.; (**b**) Measured single-molecule conductance of DNA molecules presented in (**a**). Reprinted with permission from ref. \[[@B69-jfb-09-00008]\]. Copyright (2005) National Academy of Science, USA.](jfb-09-00008-g005){#jfb-09-00008-f005} ![Measurement of charge transfer rate by Giese et al.: log(*P*~GGG~/*P*~G~) against the number *n* of the A:T base pairs The steep line corresponds to the coherent tunneling charge transfer. Reprinted with permission from ref. \[[@B76-jfb-09-00008]\]. Copyright (2001) Nature Publishing Group.](jfb-09-00008-g006){#jfb-09-00008-f006} ![Sequence and length dependent conductance (G) of single DNA molecules. Panel (**a**,**b**) show the example conductance histogram and conductance vs. distance traces, respectively; (**c**) ln(G/G~0~) vs. molecular length plot for DNA with ACGC(AT)*~m~*GCGT sequences; (**d**) G vs. 1/length plot for DNA with (GC)*~n~* sequences; (**e**) Resistance of A(CG)*~n~*T DNA as a function of number of base pairs; (**f**) Log R vs. number of base pair plot for DNA with ACGC(AT)*~m~*GCGT/ACGC(AT)*~m~*AGCGT sequences. (**a**--**d**) are reprinted with permission from ref. \[[@B47-jfb-09-00008]\]. Copyright (2004) American Chemical Society. (**e**--**f**) are reprinted with permission from ref. \[[@B77-jfb-09-00008]\]. Copyright (2016) Nature Publishing Group.](jfb-09-00008-g007){#jfb-09-00008-f007} ![Single-molecule conductance measurements of DNA with alternating G (A(CG)*~n~*T) and stacked G (AC*~n~*G*~n~*T) sequences. Reprinted with permission from ref. \[[@B72-jfb-09-00008]\]. Copyright (2015) Nature Publishing Group.](jfb-09-00008-g008){#jfb-09-00008-f008} ![Conductance measurements of poly(GC)~4~ DNA under a structural transition from B- to Z-form. (**a**) Schematic of STM-BJ method; (**b**) Example conductance histograms and traces of B-DNA (upper) and Z-DNA (lower) in 1 M MgCl~2~ solution; (**c**) CD spectra measured in 0--2 M MgCl~2~ solutions; (**d**) Transition degree vs. log MgCl~2~ concentration plot, showing the entire transition process monitored by conductance measurement. Reprinted with permission from ref. \[[@B79-jfb-09-00008]\]. Copyright (2014) Royal Society of Chemistry.](jfb-09-00008-g009){#jfb-09-00008-f009} ![Conductance measurements of DNA under B-A structural transition (**a**) Schematic of B- and A-DNA molecular junction; (**b**) Example conductance traces of B-DNA (blue) and A-DNA (green); (**c**) Example conductance histograms of B-DNA (blue) and A-DNA (green); (**d**) CD spectra of B-DNA (blue) and A-DNA (green) sample. Reprinted with permission from ref. \[[@B89-jfb-09-00008]\]. Copyright (2015) Nature Publishing Group.](jfb-09-00008-g010){#jfb-09-00008-f010} ![Stretching effect of DNA conductance. (**a**) Illustration of the evolution of dsDNA duplex during stretching; (**b**) Example conductance trace for dsDNA sequence 5′-S(CG)~2~T-3′; (**c**) Resistance vs. molecular length plot; (**d**) Stretching length vs. molecular length plot. Reprinted with permission from ref. \[[@B98-jfb-09-00008]\]. Copyright (2015) American Chemical Society.](jfb-09-00008-g011){#jfb-09-00008-f011} ![(**a**) biotin-avidin (BA)--G4-DNA scheme showing an oriented ordered stack of tetrads and AFM image showing a typical measurement scenario: a gold electrode with a sharp edge is on the left and molecules are clearly visible on the mica to the right; one molecule (at the bottom) is protruding from under the edge of the metal electrode; (**b**) *I--V* measurements taken at the positions indicated by colored dots on the molecule shown in (**a**); (**c**) Distance dependence of the current measured at a bias of 5 V for three different molecules (plotted in different colors). Reprinted with permission from ref. \[[@B102-jfb-09-00008]\]. Copyright (2014) Nature Publishing Group.](jfb-09-00008-g012){#jfb-09-00008-f012} ![(**a**) Schematic illustration of G4-DNA measured by MCBJ setup; (**b**) *I--V* curves measured at different at different stages of G4-DNA defolding. (**a**,**b**) are reprinted with permission from ref. \[[@B104-jfb-09-00008]\] Copyright (2010) Wiley--VCH Verlag GmbH & Co., KGaA. (**c**) Schematic representation of the CNT-G4-DNA-CNT sensing setup for protein detection; (**d**) Current vs. gate voltage curves measured at a constant source-drain voltage of −15 mV before and after thrombin treatments, showing reversible current change at two discrete levels; (**c**,**d**) are reprinted with permission from ref. \[[@B109-jfb-09-00008]\]. Copyright (2011) Wiley--VCH Verlag GmbH & Co. KGaA.](jfb-09-00008-g013){#jfb-09-00008-f013} ![(**a**) Structure of a DNA oligomer with sequences of GGGG and mCmCmCmC; (**b**) Current signal (**upper**) and histogram (**lower**) of the DNA. (**a**,**b**) are reprinted with permission from ref. \[[@B110-jfb-09-00008]\]. Copyright (2011) American Chemical Society (**c**) **Left**: DNA sequence and structure of cytosine and 5-methylcytosine; **Right**: conductance histograms for native DNA (grey) and methylated DNA (black). (Reprinted with permission from ref. \[[@B111-jfb-09-00008]\]. Copyright (2012) IOP Publishing.](jfb-09-00008-g014){#jfb-09-00008-f014} ![(**a**) The molecular structure of Cu^2+^ mediated base pair; (**b**) Components of metallo-DNA-bridged SWCNT devices; (**c**) Current vs. gate voltage signals of a device reconnected with metallo-DNA (blue); (**d**) Current vs. gate voltage under periodic treatment of Cu^2+^ and EDTA. Reprinted with permission from ref. \[[@B112-jfb-09-00008]\]. Copyright (2011) Wiley--VCH Verlag GmbH & Co. KGaA.](jfb-09-00008-g015){#jfb-09-00008-f015} ![(**a**) Sequence of dsDNA, structure of coralyne molecule and their interaction revealed by UV-vis spectra. Reprinted with permission from ref. \[[@B9-jfb-09-00008]\]. Copyright (2016) Nature Publishing Group. (**b**) Conductance measurements of dsDNA molecules with EB and HOE intercalated between the base pairs and into the grooves, respectively. Reprinted with permission from ref. \[[@B113-jfb-09-00008]\]. Copyright (2017) Royal Society of Chemistry.](jfb-09-00008-g016){#jfb-09-00008-f016} ![Demonstration of first DNA-based molecular diode. (**a**) STM-BJ setup; (**b**) Schematic of DNA-coralyne complex molecular junction; (**c**,**d**) show the *I--V* curves and rectification ratio curves of native DNA (blue) and DNA-coralyne complex (red); (**e**) Transmission function of DNA-coralyne complex molecular junction under different biases; (**f**) Transmission function of HOMO (solid lines) and HOMO-1 (dashed lines) levels of DNA-coralyne complex molecular junction under 1.1 V and −1.1 V. Reprinted with permission from ref. \[[@B9-jfb-09-00008]\]. Copyright (2016) Nature Publishing Group.](jfb-09-00008-g017){#jfb-09-00008-f017} ![(**a**) Illustration of the STM-BJ-based electrochemical (EC) gating setup; (**b**) Redox modified DNA (Aq-DNA), where a base was replaced with an anthraquinone (Aq) moiety (highlighted in blue) at the 3′ end of a DNA strand; (**c**) Conductance histograms of Aq-DNA with the gate voltage set above (0.085 V in I), at (−0.002 V in II) and below (−0.078 V in III) the redox potential; (**d**) Conductance values at different gate voltages showing two discrete conductance states. Reprinted with permission from ref. \[[@B130-jfb-09-00008]\]. Copyright (2017) Nature Publishing Group.](jfb-09-00008-g018){#jfb-09-00008-f018} ![(**a**) CAFM-BJ setup for spin-dependent CT in DNA; (**b**) *I--V* curves of DNA with different lengths measured under two magnetic field polarities (up: red and down: blue); (**c**) Estimated effective barrier for DNA with different lengths. Reprinted with permission from ref. \[[@B134-jfb-09-00008]\]. Copyright (2011) American Chemical Society.](jfb-09-00008-g019){#jfb-09-00008-f019} ![(**a**) Schematic diagram of STM-BJ with a modulating tip; (**b**) Upper: low-frequency component of the current collected from the single DNA junction; lower: the piezoresistance (α) in DNA (red curve) and conductance modulation due to tip modulation (blue curve); (**c**) α vs. conductance histogram for G-C sequence; α vs. molecular length for (**d**) (G--C)~N~ and (**e**) (G~N~--C~N~) sequences, respectively. Reprinted with permission from ref. \[[@B138-jfb-09-00008]\]. Copyright (2015) Nature Publishing Group.](jfb-09-00008-g020){#jfb-09-00008-f020} ![(**a**) Illustration of a DNA molecule bridged between STM tip (kept at 295 K) and substrate (cold); (**b**) The measured Seebeck coefficient (S) of DNA molecules with different sequences and lengths. Reprinted with permission from ref. \[[@B77-jfb-09-00008]\]. Copyright (2016) Nature Publishing Group.](jfb-09-00008-g021){#jfb-09-00008-f021} ![Possible routes to develop future integrated DNA circuits by combining DNA origami technologies (upper row) and appropriate modifications of the electronic structure of individual DNA molecule (lower row). The middle frame of the upper row is reprinted with permission from ref. \[[@B34-jfb-09-00008]\]. Copyright (2017) Nature Publishing Group. The right frame of the upper row is reprinted with permission from ref. \[[@B143-jfb-09-00008]\]. Copyright (2017) Nature Publishing Group. The middle and right frames of the lower row are reprinted with permission from ref. \[[@B116-jfb-09-00008]\]. Copyright (2010) ELSEVIER.](jfb-09-00008-g022){#jfb-09-00008-f022}

INTRODUCTION {#Sec1} ============ The potential influence of excipients on the performance of solid oral dosage forms is a topic of great interest in terms of pharmaceutical Quality by Design (QbD). Further to their intended use, excipients may affect the properties and performance of final dosage forms leading to batch inconsistencies, altered bioavailability and bioinequivalence of products ([@CR1]--[@CR3]). Excipient variability (changes in material properties) and variation (changes in amount) constitute an additional obstacle to robust manufacturing, as changes in excipient material properties can undermine the critical quality attributes of the final product ([@CR4]). Moreover, excipient inertness is questionable as their presence can influence oral drug absorption ([@CR5],[@CR6]). Understanding the biopharmaceutical risks of excipient presence will pave the way for manufacturing products with robust product performance. Binders are typically used in solid dosage form manufacturing to promote adequate mechanical strength of granules or tablets. Hypromellose, also called hydroxypropyl methylcellulose (HPMC), is a polymeric binder used in wet granulation ([@CR7]). HPMC is a water soluble non-ionic cellulosic polymer substituted with methoxy and hydroxypropyl groups (Fig. [1](#Fig1){ref-type="fig"}). In addition to its binding properties, HPMC is extensively used as a release controlling excipient. This dual functionality depends on excipient level (2% *w*/*w* to 5% *w*/*w* as binder and 10% *w*/*w* to 80% *w*/*w* as release modifier) and viscosity type (high viscosity grades are typically used to control drug release) ([@CR8]). The effectiveness of HPMC can be influenced by its swelling and gelling properties which can delay drug dissolution or release ([@CR9],[@CR10]) or by the presence of other excipient types in a formulation ([@CR11],[@CR12]).Fig. 1Chemical structure of Hypromellose (ChemDraw Professional 15.0) Molecular properties (molecular weight, degree of substitution, substitution pattern), particle properties (particle size distribution) and excipient level have been identified as critical material attributes affecting excipient and product performance ([@CR4]). Molecular weight and excipient level directly relate to the formation of a viscous HPMC gel layer. HPMC brands of high molecular weight swell faster and form a thicker viscous layer compared to low molecular weight brands ([@CR9],[@CR13],[@CR14]). The formation of thick viscous layers when increasing HPMC molecular weight is attributed to the slow rate of polymer erosion ([@CR9]). Real-time surface dissolution UV imaging demonstrated the complex polymeric network formed by high molecular weight HPMC brands and the susceptibility of low molecular weight HPMC brands to erosion ([@CR15]). Due to the proportionality of HPMC molecular weight and viscosity of its aqueous solutions, high molecular weight excipient brands correspond to high viscosity HPMC types. The formation and thickness of the gel layer depend also on the level of HPMC in solid dosage forms. Increasing the level of HPMC results in a concentrated and viscous gel layer either due to increased chain entanglement ([@CR16]) or slow polymer erosion rate ([@CR17]). A minimum excipient concentration (referred to as excipient percolation threshold) under which HPMC cannot form an effective viscous layer able to control drug release was reported ([@CR18]). The percolation thresholds of different viscosity type HPMC brands were similar (20% *v*/*v* corresponding to 20% *w*/*w*) and showed consistent controlled drug release (verapamil HCl, United States Pharmacopeia (USP 2) apparatus, 50 rpm, 900 mL, phosphate buffer pH 7.5) from extended release tablets containing HPMC ([@CR18]). Use of confocal laser scanning microscopy revealed that the early gel layer formation depends on the level of HPMC. Hydrophilic matrices containing 5% *w*/*w* and 10% *w*/*w* of HPMC formed a thin and heterogeneous gel layer initially (5--15 min), which could not be maintained due to the increased water uptake accelerating polymer erosion ([@CR19]). Several biopharmaceutical factors affecting the impact of HPMC on product performance have been identified. The performance of HPMC depends on the composition and properties of the dissolution medium. Salts ([@CR20]), sugars ([@CR21]) and food components ([@CR22]) interact with the polymeric chains of HPMC and affect the formation of the gel layer. Moreover, bile salts were found to affect the thermal transition (gel formation) of cellulosic polymers. The hydrophobic parts of bile salts adsorb onto the hydrophobic regions of polymers and increase the excipient transition temperature upon heating ([@CR23],[@CR24]). The pH of the dissolution medium can also affect the swelling/gelling properties of HPMC due to changes in the transport behaviour of water into the polymer, despite the non-ionic nature of HPMC. Rapid polymer hydration and larger excipient swelling were observed in media of basic (24% increase in polymer diameter after 200 min at pH = 6) compared to acidic pH (15% increase in polymer diameter after 200 min at pH = 2) \[pH defined as basic and acidic according to the physiological pH range\]. The authors highlighted the biopharmaceutical consequence of this polymeric response as faster drug release should be expected in the acidic gastric compared to the basic intestinal compartment ([@CR25]). The aim of this study was to assess the biopharmaceutical impact and criticality of HPMC variability and variation (when used as a binder in immediate release formulations) on drug apparent solubility. HPMC variability and variation were addressed by selecting three HPMC brands of different viscosity type using two different excipient levels (low: 2% *w*/*w*, high: 5% *w*/*w*). The biopharmaceutical implications of HPMC variability on drug apparent solubility were examined by choosing compounds of a wide range of physicochemical properties (drug ionization, drug lipophilicity, drug aqueous solubility) and media (compendial and biorelevant) simulating the gastrointestinal conditions. Multivariate data analysis (partial least squares (PLS)) and roadmaps were used to identify the critical role of certain variables (drug properties, excipient presence, medium characteristics) on the impact of HPMC on drug apparent solubility. MATERIALS AND METHODS {#Sec2} ===================== Materials {#Sec3} --------- APIs: sulfamethoxazole (SMX) and paracetamol (PRC) were obtained from Fisher Scientific (UK). Furosemide (FRS), itraconazole (ITZ) and dipyridamole (DPL) were obtained from VWR (UK). Ibuprofen (IBU), carbamazepine (CBZ) and metformin (MTF) were obtained from Fagron (UK). Excipients: HPMC-Sigma was obtained from Sigma Aldrich (UK). Pharmacoat 606 and Pharmacoat 615 were obtained from Shinetsu (Japan). Chemicals: acetic acid (\> 99.7%), hydrochloric acid 36.5--38%, high-performance liquid chromatography (HPLC) grade methanol, HPLC grade acetonitrile, dichloromethane, pepsin (from porcine) were obtained from Sigma-Aldrich (UK). Maleic acid, sodium chloride, sodium hydroxide, potassium phosphate monobasic, sodium dihydrogen orthophosphate dihydrate, disodium hydrogen orthophosphate dihydrate, potassium dihydrogen orthophosphate, anhydrous sodium sulfate, HPLC grade trifluoroacetic acid were obtained from Fisher Scientific (UK). Sodium taurocholate (Prodotti Chimici Alimentari S.P.A., Italy), egg lecithin--Lipoid EPCS (Lipoid GmbH, Germany), were obtained from the sources specified. Water was ultra-pure (Milli-Q) laboratory grade. Filters: Whatman® 13 mm cellulose nitrate filters 0.45 μm pore size and polytetrafluoroethylene (PTFE) 13 mm filter 0.45 μm pore size were purchased from Fisher Scientific (UK). Instrumentation {#Sec4} --------------- Fisherbrand waterbath (Fisher Scientific, UK), Sartorius BP 210 D balance (Sartorius Ltd., UK), Buchi R114 Rotavapor (Buchi, Switzerland), Mettler Toledo SevenCompact S210 pH meter (Mettler Toledo, Switzerland), Vortex-Genie 2 vortex mixer (Scientific Industries Inc., USA), Agilent Technologies 1100 series HPLC system (quaternary pump (G1311A), autosampler (G1313A), thermostatted column compartment (G1316A), diode array detector (G1329A) and Chemstation software (Agilent Technologies, USA). Methods {#Sec5} ------- ### Compounds Selected for Solubility Experiments {#Sec6} Compound selection was based on properties affecting oral drug absorption, namely ionization, lipophilicity and aqueous solubility ([@CR26]). The compounds were chosen to cover a wide range of ionization (low ionized: F~(ion)~ \< 50%, highly ionized: F~(ion)~ \> 50%), drug lipophilicity (based on the drugs' partition coefficient (log P): − 1.5 \< log *P* \< 6.5) and aqueous solubility (based on the compound's BCS (Biopharmaceutical Classification System) classification (high: BCS class I and III; low: BCS class II and IV)) ([@CR27]). The compounds used for the solubility experiments and their physicochemical properties (drug ionization, drug lipophilicity, drug aqueous solubility) are presented in Table [I](#Tab1){ref-type="table"}.Table IPhysicochemical Properties and Structure of the Compounds Used for the Solubility Experiments (ChemDraw Professional 15)DrugIonizationLipophilicity (log P)\*Solubility\*\*Neutral Drugs{#d29e534}Neutral (pKa=9.38) ([@CR28])0.20 ([@CR28])High ([@CR28]){#d29e557}Neutral (pKa=15)^a^2.45 ([@CR29])Low ([@CR30])Weak acids{#d29e582}Weak acid (pKa=3.8) ([@CR31])2.29 ([@CR31])Low ([@CR31]){#d29e605}Weak acid (pKa=4.5) ([@CR32])4.00 ([@CR33])Low ([@CR32])Weak Bases{#d29e631}Weak base (pKa=2.8) ([@CR34])-1.43 ([@CR35])High ([@CR34]){#d29e654}Weak base (pKa=6.2) ([@CR36])2.74 ([@CR37])Low ([@CR38]){#d29e677}Weak base (pKa=4.5) ([@CR39])6.20 ([@CR39])Low ([@CR40])Ampholytes{#d29e703}Ampholyte \[Weak base: pKa~1~=1.7/Weak acid:. pKa~2~=5.6\] ([@CR41])0.89^a^Low ([@CR42])\*Experimental values, \*\* based on the compound's BCS (Biopharmaceutical Classification System) classification (high: BCS Class I and III; low: BCS Class II and IV) ([@CR27]), ^a^Source: DrugBank ### Media Prepared for Solubility Experiments {#Sec7} Compendial media (0.1 N HCl pH 1, phosphate buffer pH 6.8) were prepared according to the method described in the European Pharmacopoeia ([@CR43]). Fasted State Simulated Gastric Fluid (FaSSGF) and Fasted State Simulated Intestinal Fluid (FaSSIF-V2) were prepared as described by Jantratid *et al.* ([@CR44]). ### Design of Experiments Used for Solubility Experiments {#Sec8} A full-factorial design of experiments (DoE) was performed to determine the number of necessary experiments using StatGraphics Centurion XVII (Statpoint Technologies Inc., USA). As the composition of the studied media (pH, presence of bile salts) will affect drug solubility, two models for the DoE were constructed to discriminate between the effects of excipients on drug solubility in compendial (model 1) and biorelevant conditions (model 2). The examined factors were (i) compound (Table [I](#Tab1){ref-type="table"}), (ii) excipient brand (Pharmacoat 606, Pharmacoat 615, HPMC-Sigma), (iii) excipient level (low, high), (iv) medium (gastric, intestinal). The response was the impact of each excipient on drug solubility \[expressed as the relative increase or decrease in presence compared to absence of excipient (section "Treatment of *In Vitro* Solubility Data")\]. A total of 96 × 3 experiments were determined for each model. A total of 16 × 3 additional experiments for each model were conducted to determine drug solubility in the corresponding media in the absence of excipient. These experiments were not included in the DoE as drug solubility in excipient absence was measured only for the calculation of relative excipient effects on drug apparent solubility. ### Solubility Studies {#Sec9} Drug solubility studies in the absence and presence of excipient were performed in triplicate using the shake-flask method ([@CR45]). Drug excess amount and 2% *w*/*w* or 5% *w*/*w* of each excipient were weighed and placed in centrifuge tubes. For poorly soluble drugs, the amount of excipient was determined considering an average of 500 mg tablet weight ([@CR46]), which resulted in 9% *w*/*w* (10 mg of excipient and 100 mg of drug; low level) and 20% *w*/*w* (25 mg of excipient and 100 mg of drug; high level) of excipient in the total volume of the physical mixture. For highly soluble drugs, as higher drug excess amount was used to ensure saturation, the excipient amount was increased in order to keep the same % *w*/*w* of excipient in the total volume of the physical mixture as per the poorly soluble drugs. The physical mixtures were vortexed for 3 min. Five millilitres of each medium was added in the tubes and the samples were placed in a shaking water bath (37 °C, 200 strokes per minute (spm)). At 0.5, 4 and 24 h (for PRC, SMX, CBZ, DPL, IBU) and 24 h (for MTF, FRS, ITZ), 500 μL was sampled and filtered through PTFE filters (or cellulose nitrate filters for the cases of IBU and CBZ). Filter adsorption studies were prior performed in triplicate for each drug. No adsorption issues onto the filters used were observed for the studied drugs. Filtered samples were further diluted (if needed) with the corresponding medium and analysed by HPLC (Supplementary Table [I](#MOESM1){ref-type="media"}). Analytical procedures for drug quantification in the samples were modifications of already published methods. Drug quantification was made based on calibration curves. Standards were formulated from concentrated stock solutions consisting of drug dissolved in MeOH. As drug solubility can be influenced by the changes in the pH of solutions by presence of dissolved drug ([@CR38]), the pH of samples after the completion of each experiment was measured. Drug solubility was calculated based on the sample drug concentration measured. For neutral drugs, weak acids in acidic media and weak bases in basic media, the experimental drug solubilities determined the intrinsic solubility values. For weak acids and weak bases in basic and acidic media, respectively, the experimental drug solubilities determined drug solubility of the ionized molecules. The drug solubility measured was considered as the apparent drug solubility (dynamic solubility), as experimental points over a period of time were not available for the whole set of drugs to ensure that equilibrium solubility has been reached in 24 h for all the studied compounds. ### Treatment of *In Vitro* Solubility Data {#Sec10} The relative effect (RE) of each excipient on drug apparent solubility was calculated based on Eq. 1:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ RE=\frac{\left(S- Sr\right)}{Sr}\times 100 $$\end{document}$$where *S* and *Sr* denote the drug solubility in presence and absence (reference solubility) of excipient at 0.5, 4 and 24 h. REs of excipients on drug solubility \> 25% or \< −20% were considered as significant change in drug apparent solubility to assess excipient criticality (this range was selected as a similar range is set in order to assess differences in drug exposure after oral administration, i.e. in bioequivalence studies) ([@CR47]). Box plots depicting the impact of excipients on drug solubility at 24 h for all the studied compounds or as a function of time (0.5, 4 and 24 h) for CBZ and DPL were constructed using Spotfire 7.10.1 (TIBCO software Inc., USA). The classification gradient maps portraying the impact of the studied brands on drug solubility at 24 h as a function of drug aqueous solubility were generated using SigmaPlot 13.0 (Systat Software Inc., USA). For the construction of 3D mesh plots depicting the impact of excipients on drug solubility at 24 h as a function of time and drug ionization, solubility data for PRC, SMX, IBU, DPL were smoothed via the negative exponential technique to allow better visualization using SigmaPlot 13.0 (Systat Software Inc., USA). In cases where drug intrinsic solubility was not determined experimentally (SMX and DPL in compendial and biorelevant media), the theoretical intrinsic solubility was calculated using the solubility-pH equations (Eqs. 2--5) ([@CR48]):$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \log S=\log {S}_o+\log \left({10}^{- pKa+ pH}+1\right)\ \mathrm{for}\ \mathrm{weak}\ \mathrm{acids} $$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \log S=\log {S}_o+\log \left({10}^{pKa- pH}+1\right)\ \mathrm{for}\ \mathrm{weak}\ \mathrm{bases} $$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \log S=\log {S}_o+\log \left({10}^{+p{Ka}_2+{pKa}_1-2 pH}+{10}^{pKa_2- pH}+1\right)\ \mathrm{for}\ \mathrm{diprotic}\ \mathrm{bases} $$\end{document}$$$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \log S=\log {S}_o+\log \left({10}^{+{pKa}_1- pH}+{10}^{-{pKa}_2+ pH}+1\right)\ \mathrm{for}\ \mathrm{ampholytes} $$\end{document}$$where *S* and *S*~*o*~ indicate the drug solubility at the given pH and the intrinsic solubility, respectively. Deviations in the determination of drug solubility from the aforementioned simplified equations are expected in cases of drug self-association or solubilization in biorelevant media ([@CR48]). The final pH and experimental solubility values of the ionized drug in basic (for weak acids) or acidic media (for weak bases) were used for the calculation of the theoretical intrinsic solubility. Theoretical pH--solubility profiles in the physiological pH range were constructed to assess if changes in the pH of the medium could justify differences in drug solubility by excipient presence. The final pH and intrinsic solubility values (experimental or theoretical) were used for the construction of the theoretical pH--solubility profiles in the physiological pH range based on Eqs. 2--5. ### Multivariate Analysis of Solubility Data {#Sec11} Excipient REs on drug apparent solubility were correlated to drug physicochemical properties (drug ionization, drug lipophilicity, drug aqueous solubility), excipient critical material attributes (viscosity, level) and medium characteristics (gastric, intestinal) by partial least squares (PLS) regression using the XLSTAT software (Microsoft, USA). Two models for the REs of excipients on drug apparent solubility in compendial media (model 1) and biorelevant media (model 2) were constructed. The evaluated variables for both models were categorized according to their type as categorical (expressing a category or type) and numerical (measurements with numerical meaning). Categorical variables included (i) drug solubility (low, high), (ii) aminic group (absence, presence), (iii) excipient level (low, high), (iv) medium (gastric, intestinal) while numerical parameters included (i) theoretical % of drug ionized (F~ion~; calculated based on the Henderson--Hasselbalch equation at the pH of each medium), (ii) drug lipophilicity (log P), (iii) excipient viscosity (cP). Excipient REs on drug solubility at 24 h were used as the response. The selected interaction terms included each excipient property combined with each drug physicochemical property (drug ionization, drug lipophilicity, drug aqueous solubility) and medium characteristics (gastric, intestinal). Observation diagnostics were performed prior to model analysis to identify outliers in the data set. The distance of each observation to the model in the Y-plane (DmodY) tool based on PLS residuals was used. Plots of standardized DmodY vs each observation were generated and any observation exceeding the maximum tolerance volume in Y (D~crit(Y)~) was considered an outlier ([@CR49],[@CR50]). Exclusion of outliers was based on (i) deviating cases (positive REs) in solubility caused by a shift in the pH of the solution, (ii) observations resulting in high variability (coefficient of variation (CV%) \> 20%) within the triplicate samples (one value from the triplicate could be excluded as the outlier analysis could detect these values). PLS models generated with and without outlier exclusion (data not shown) confirmed that outlier exclusion did not alter the interpretation of results but only enhanced the predictive ability of the regression model. The generated models were assessed in terms of goodness of fit (*R*^2^) and goodness of prediction (*Q*^2^). High values of *R*^2^ and *Q*^2^ with a difference not greater than 0.2--0.3 were indications of successful models ([@CR51]). The number of PLS components (lines on the X-space which best approximate and correlate with the Y-vector) was based on minimum predictive residual sum of squares (PRESS) ([@CR51]). From the available components, the one at which *Q*^2^ reached its maximum value was selected ([@CR49]). Standardized coefficients were used to show the direction (positive or negative) and extent of each variable on the response. The significance of the variables was assessed by the variable influence on projection (VIP) value. VIP values \> 0.8 were considered as moderately influential in the model while VIP values \> 1 were considered the most influential in the model ([@CR51]). A 95% confidence interval was used. ### Roadmap Design {#Sec12} The biopharmaceutical implications of HPMC variability on drug apparent solubility were depicted with the use of roadmap designs. These were constructed by combining the impact of excipients on drug solubility at 24 h from the solubility studies to excipient (viscosity) and drug (drug ionization, drug lipophilicity, drug aqueous solubility) physicochemical properties. Drug categorization was based on drug aqueous solubility and drug lipophilicity (Table [I](#Tab1){ref-type="table"}) and drug ionization (low ionized: F~(ion)~ \< 50%, highly ionized: F~(ion)~ \> 50%). Reference range criteria of − 20% to 25% ([@CR47]) on the REs of HPMC on drug solubility were set for the risk assessment of the excipient effects on drug solubility. REs of excipients on drug apparent solubility outside these values (REs \< − 20% or REs \> 25%) were considered to be potentially significant for oral drug performance. RESULTS AND DISCUSSION {#Sec13} ====================== Properties of HPMC {#Sec14} ------------------ Based on product specifications and certificates of analysis, differences in terms of viscosity (cP) for the studied HPMC brands were identified. Pharmacoat 606 (6 cP) ([@CR52]) and Pharmacoat 615 (15 cP) ([@CR52]) are low viscosity brands compared to HPMC-Sigma (3500--5600 cP) ([@CR53]). The studied polymers belong to the same category of substitution type 2910 (28--30% of methoxy and 7--12% hydroxypropoxyl content). Solubility Studies {#Sec15} ------------------ ### Impact of the Studied HPMC Brands on Drug Apparent Solubility {#Sec16} The reference solubility values at 24 h of the studied compounds in compendial and biorelevant media are summarized in Table [II](#Tab2){ref-type="table"}. For neutral drugs, the reference solubility did not depend on the pH of the studied media (as expected) but on the presence of solubilizing components, as increased reference solubilities were observed in biorelevant compared to compendial media due to the presence of bile salts ([@CR54]). Different reference solubility values within acidic and basic media were observed for weak acids (FRS, IBU) and weak bases (DPL, ITZ) due to their ionization pattern, except from the case of MTF as drug was fully ionized in all the studied conditions ([@CR34]). SMX (weak base and weak acid in acidic and basic media, respectively) was ionized in both acidic and basic media due to its two pKa constants. For weak acids in acidic media (drugs in the unionized state), slight differences in drug solubility were observed between compendial and biorelevant media (IBU: 43 μg/mL in 0.1 N HCl pH 1 vs 44 μg/mL in FaSSGF, FRS: 14 μg/mL in 0.1 N HCl pH 1 vs 15 μg/mL in FaSSGF). For weak bases, drug solubility values in the low ionization drug state were higher in biorelevant as compared to compendial media (DPL: 5 μg/mL in phosphate buffer pH 6.8 vs 13 μg/mL in FaSSIF-V2) which is explained by the presence of bile salts ([@CR54]) or the higher % of drug ionized in FaSSIF-V2 (34% in FaSSIF-V2 vs 20.1% in phosphate buffer pH 6.8). For weak acids/weak bases in media where drugs are highly ionized, drug solubility values were higher in compendial (0.1 N HCl pH 1, phosphate buffer pH 6.8) compared to biorelevant media (FaSSGF pH 1.6, FaSSIF-V2 pH 6.5), as pH has a greater impact on drug solubility compared to bile salts ([@CR55]). The effects of the HPMC brands on the solubility of the studied compounds in compendial and biorelevant media are presented in Fig. [2](#Fig2){ref-type="fig"}.Table IIReference Solubility Values (μg/mL) of the Studied Drugs at 24 h in Compendial and Biorelevant Media (mean ± SD, *n* = 3)Compendial mediaBiorelevant mediaDrug0.1 N HCl pH 1Phosphate buffer pH 6.8FaSSGFFaSSIF-V2MTF3.1 × 10^5^ (± 0.3 × 10^5^)3.1 × 10^5^ (± 0.2 × 10^5^)3.4 × 10^5^ (± 0.8 × 10^5^)4.3 × 10^5^ (± 0.4 × 10^5^)PRC1.6 × 10^4^ (± 0.1 × 10^4^)1.5 × 10^4^ (± 0.1 × 10^4^)1.7 × 10^4^ (± 0.2 × 10^4^)1.7 × 10^4^ (± 0.1 × 10^4^)SMX1.6 × 10^3^ (± 0.1 × 10^3^)3.7 × 10^3^ (± 0.1 × 10^3^)862 (± 21)1.3 × 10^3^ (± 0.1 × 10^3^)FRS14 (± 2)3.4 × 10^3^ (± 1.4 × 10^2^)15 (± 1)1.6 × 10^3^ (± 3.0 × 10^2^)CBZ265 (± 6)227 (± 9)368 (± 1)280 (± 7)DPL1.3 × 10^4^ (± 9.1 × 10^2^)5 (± 1)8.6 × 10^3^ (± 2.0 × 10^2^)13 (± 1)IBU43 (± 3)5.5 × 10^3^ (± 6.7 × 10^2^)44 (± 5)1.5 × 10^3^ (± 5.8)ITZ11 (± 1)--^a^1.2 (± 0.2)0.05 (± 0.01)^a^Below limit of detection of the analytical method*MTF* metformin, *PRC* paracetamol, *SMX* sulfamethoxazole, *FRS* furosemide, *CBZ* carbamazepine, *DPL* dipyridamole, *IBU* ibuprofen, *ITZ* itraconazoleFig. 2Box plots of the relative effects (%) of the HPMC brands on drug solubility at 24 h in **a** compendial and **b** biorelevant media. The excipient brands are shown as: (i) Pharmacoat 606 (green colour), (ii) Pharmacoat 615 (blue colour) and (iii) HPMC-Sigma (red colour). Light and dark colours correspond to low and high excipient level, respectively. Mean (white line), median (black diamond), *n* = 3 Neutral Drugs {#Sec17} ------------- Cases of significant decrease in drug solubility at 24 h by the studied HPMC brands were observed for PRC (Pharmacoat 606: − 36% \< REs \< − 20%, Pharmacoat 615: − 32% \< REs \< − 22%, HPMC-Sigma: − 30% \< REs \< − 22%). Differences in the apparent solubility of neutral drugs in HPMC presence cannot be attributed to changes in the pH of the media (Fig. [3](#Fig3){ref-type="fig"}). The pronounced reduction in drug solubility at 24 h by HPMC is attributed to the slow drug dissolution and/or drug solubilization as excipients with slow dissolution rate may shield drug particles from the dissolution medium ([@CR56]) or the formation of a viscous excipient layer on the powder surface ([@CR57]). The aforementioned mechanisms indicate that the HPMC will likely affect the kinetic processes of drug dissolution rather than affecting the true thermodynamic drug solubility. Significant increase in drug apparent solubility by the studied HPMC brands was observed for CBZ in compendial media (Pharmacoat 606: 32% \< REs \< 50%, Pharmacoat 615: 28% \< REs \< 33%, HPMC-Sigma: 50% \< REs \< 60%) and in FaSSIF-V2 (REs of approximately 60% for all the studied brands). Solubility data of CBZ at 0.5, 4 and 24 h in absence and presence of the studied HPMC brands in compendial and biorelevant media are presented in Fig. [4a](#Fig4){ref-type="fig"}. The solubility of pure CBZ decreased through time in compendial media (350 μg/mL and 250 μg/mL at 0.5 h and 24 h, respectively), potentially due to drug precipitation ([@CR58]) or due to the solution-mediated phase transformation of CBZ from the anhydrate to the dihydrate form ([@CR59]). The decrease in CBZ apparent solubility was slower in biorelevant media (FaSSGF: 387 μg/mL to 367 μg/mL at 0.5 h and 24 h, respectively, FaSSIF-V2: 307 μg/mL to 280 μg/mL at 0.5 h and 24 h, respectively). Reduction in CBZ apparent solubility is not observed in presence of HPMC, as potentially dissolved polymer particles may enhance drug solubilization ([@CR60]). Inhibition of the CBZ solution-mediated phase transformation of by hydrogen bonding ([@CR61]) or by excipient adsorption in the formed nuclei/crystals delaying the rate of nucleation and crystal growth ([@CR59]) could also explain the fact that CBZ apparent solubility was not reduced in excipient presence.Fig. 3Theoretical pH--solubility profiles of the studied drugs in compendial and biorelevant media and experimental drug solubility values in absence (black colour) and presence of excipients ((i) Pharmacoat 606 (green colour), (ii) Pharmacoat 615 (blue colour), (iii) HPMC-Sigma (red colour)). Dashed lines indicate drug intrinsic solubilityFig. 4Box plots of **a** CBZ solubility (μg/mL) and **b** DPL solubility (μg/mL) in absence (black colour) and presence of the studied HPMC brands ((i) Pharmacoat 606 (green colour), (ii) Pharmacoat 615 (blue colour) and (iii) HPMC-Sigma (red colour)) in compendial and biorelevant media. Light and dark colours correspond to low and high excipient level, respectively (mean, *n* = 3) Weak Acids {#Sec18} ---------- Cases of significant reduction are observed in weak acidic compound solubility at 24 h, especially in basic conditions (Pharmacoat 606: − 65% \< REs \< − 20%, Pharmacoat 615: − 40% \< REs \< − 20%, HPMC-Sigma: − 80% \< REs \< − 20%) (Fig. [2](#Fig2){ref-type="fig"}). In the cases of significant reduction in drug apparent solubility by HPMC presence, small changes in the pH of the media occurred in absence or presence of excipient (± 0.1 pH units) in acidic conditions while the pH of the media in absence or presence of excipient decreased (0.2--0.7 pH units) in basic conditions (potentially due to the dissociation of weak acids). The experimental solubility values in excipient presence were lower compared to theoretical ones (expected by the change in the pH of the media) (Fig. [3](#Fig3){ref-type="fig"}). As the pronounced reduction in drug solubility at 24 h by HPMC cannot be explained by the shift in the pH of the media, it is attributed to the delay in drug dissolution and/or drug solubilization, as explained previously in the case of neutral drugs. Significant increase in drug apparent solubility was only observed for FRS in presence of both levels of HPMC-Sigma in phosphate buffer pH 6.8 (REs of 53% and 41% for the low and high level, respectively). Differences in the pH of the medium in presence (reduction of 0.3 pH units) and absence of excipient (reduction of 0.2 pH units) were observed. Evaluation of the theoretical pH--solubility profile (Fig. [3](#Fig3){ref-type="fig"}) reveals that drug solubility in absence of excipient does not correspond to the theoretical drug solubility (expected according to the change in the pH of the medium). In presence of HPMC-Sigma, the experimental and theoretical solubility values are similar, therefore the increase in drug solubility at 24 h in presence of HPMC-Sigma is attributed to the change of the pH and not to a potential drug-excipient interaction (further investigations are needed to explain the nature of this change, as shifts in the pH of the media by the non-ionic HPMC are not expected). Weak Bases {#Sec19} ---------- For the majority of cases, significant reduction is observed in weak basic compound solubility at 24 h by HPMC presence (− 50% \< REs \< − 20% for all the studied HPMC brands) (Fig. [2](#Fig2){ref-type="fig"}). The impact of pH on the solubility of MTF, DPL and ITZ cannot be evaluated due to *in situ* salt formation between the drugs and counterions of the medium ([@CR62]) (Fig. [3](#Fig3){ref-type="fig"}). For SMX in acidic media, the changes in the pH of the media observed in presence of HPMC were small (± 0.1 pH units). The differences in SMX solubility in HPMC presence cannot be attributed to the pH change of the media as the experimental solubility values in excipient presence were lower compared to theoretical ones (Fig. [3](#Fig3){ref-type="fig"}). As per neutral drugs and weak acids, presence of HPMC affected the kinetic processes of drug dissolution/solubilization leading to reduced drug apparent (rather than equilibrium) solubility. Significant increase in drug solubility at 24 h was observed for DPL in phosphate buffer pH 6.8 (38% \< REs \< 58%) and ITZ in FaSSIF-V2 (25% \< REs \< 61%) in presence of all the studied HPMC brands. Solubility data of DPL at 0.5, 4 and 24 h in absence and presence of HPMC in phosphate buffer pH 6.8 and FaSSIF-V2 are presented in Fig. [4b](#Fig4){ref-type="fig"}. In phosphate buffer pH 6.8, increased DPL apparent solubility in presence of HPMC is observed even at early time points (0.5 and 4 h) and can be attributed to intermolecular interactions between the amine group of DPL and the hydroxyl group of HPMC which improve drug solubilization ([@CR63]). Presence of HPMC did not affect the DPL apparent solubility in FaSSIF-V2 potentially due to the enhanced drug solubilization by the bile salts. ITZ forms a supersaturated solution in FaSSIF-V2 due to the micellar solubilization effect of bile salts and slowly precipitates with time ([@CR64]). The increase in ITZ apparent solubility in HPMC presence in FaSSIF-V2 can be attributed to the inhibition of drug precipitation by HPMC due to potential interaction of the neutral aminic group of ITZ with the hydroxyl groups of HPMC. The solubility data showed increased variability in the cases where HPMC presence significantly affected drug solubility (MTF: CV% \> 40%, PRC or highly ionized poorly soluble drugs: 10% \< CV% \< 30%). As working with physical mixtures may yield high standard deviations due to the heterogeneous dispersion of the constituents ([@CR65],[@CR66]), the increased variability can be attributed to the heterogeneous saturation of powder surface with excipient particles. Impact of Excipients on Drug Apparent Solubility Based on Drug Physicochemical Properties {#Sec20} ----------------------------------------------------------------------------------------- The effects of the studied HPMC brands on the apparent solubility of neutral drugs, weak acids and weak bases as a function of drug ionization and drug lipophilicity in compendial and biorelevant media are presented in Fig. [5](#Fig5){ref-type="fig"}. For neutral drugs, the different effects of the HPMC brands on drug solubility at 24 h (decrease and increase for the solubility values of PRC and CBZ, respectively) are attributed to the differences in drug lipophilicity between the two compounds (Table [I](#Tab1){ref-type="table"}). For weak acids and weak bases, significant decrease in drug apparent solubility was observed in media (compendial or biorelevant) where drugs are highly ionized (excluding the cases of increased drug solubility attributed to the change of the pH of the medium). A trend between the impact of excipients on the 24-h drug solubility and drug lipophilicity was found for weak bases in biorelevant media where drugs are unionized, as increased drug solubility was observed in HPMC presence for highly lipophilic drugs. The decrease in drug apparent solubility by HPMC presence was more pronounced with increasing level of drug ionization and/or decreasing drug lipophilicity (Fig. [5](#Fig5){ref-type="fig"}**)** which can be explained by the presence of an increased number of excipient molecules on the powder surface ([@CR56],[@CR67]). Reduced drug mobility and slower drug diffusion have been reported for charged molecules due to interactions with the polymeric chains of HPMC ([@CR57]) and may also have contributed to the decrease in the apparent solubility of ionized molecules by HPMC. The classification gradient map depicting the effects of the studied HPMC brands on drug solubility at 24 h as a function of decreasing drug aqueous solubility in compendial and biorelevant media is presented in Fig. [6a](#Fig6){ref-type="fig"}. The map confirms that the decrease in drug apparent solubility in presence of HPMC was more pronounced for highly soluble drugs which are inherently less lipophilic while for poorly soluble drugs, the reduction in drug solubility at 24 h was more pronounced in media where drugs were highly ionized.Fig. 5Relative effects (%) of the studied HPMC brands on drug solubility at 24 h as a function of drug ionization (%) and drug lipophilicity (log P) in **a** compendial and **b** biorelevant media. The excipient brands are shown as: (i) Pharmacoat 606 (green colour), (ii) Pharmacoat 615 (blue colour), (iii) HPMC-Sigma (red colour). Light and dark colours correspond to low and high excipient levels, respectively. Media representing gastric and intestinal conditions are presented as (i) acidic media (circles) and (ii) basic media (squares)Fig. 6**a** Classification gradient maps of the relative effects of the HPMC brands on the solubility of highly and poorly soluble compounds at 24 h. Y-axes are set in an increasing excipient viscosity type and excipient level order. The x-axes are set in a decreasing drug aqueous solubility order (red colours for highly soluble and blue colours for poorly soluble drugs). **b** Relative effects of (i) Pharmacoat 606 (green colour), (ii) Pharmacoat 615 (blue colour) and (iii) HPMC-Sigma (red colour) on drug solubility as a function of drug ionization (%) and time (h). Light and dark colours correspond to low and high excipient level, respectively ### Impact of Excipients on Drug Solubility Based on Excipient Properties {#Sec21} The effects of the studied HPMC brands on drug apparent solubility as a function of drug ionization and time for drugs were solubility data at early time points were available (PRC, SMX, CBZ, DPL and IBU) are presented in Fig. [6b](#Fig6){ref-type="fig"}. Solubility data of CBZ in compendial and biorelevant media and of DPL in phosphate buffer pH 6.8 were not included in this visualization, as significant increase in drug solubility was observed. The decrease in drug apparent solubility by HPMC presence was more pronounced at early time points. When HPMC is used at low levels (\< 10% *w*/*w*) in tablet formulations, a heterogeneous viscous HPMC layer is formed initially and cannot be maintained due to water penetration ([@CR19]). Therefore, the higher reduction in drug apparent solubility at early compared to late time points may be explained by the disruption of the gel layer by water molecules. The high viscosity brand (HPMC-Sigma) resulted in higher reduction in drug apparent solubility at early time points compared to the low viscosity brands (Pharmacoat 606, Pharmacoat 615) which can be explained by the formation of a stronger gel layer by the high HPMC viscosity types compared to the low ones ([@CR9],[@CR13],[@CR14]). The time-dependent impact of HPMC on drug solubility reveals that the excipient affects the kinetic processes of drug dissolution and mostly influences the drug apparent rather than the true equilibrium drug solubility. Increasing the level of all the studied HPMC brands resulted in higher reduction in drug apparent solubility compared to the low excipient level, as the increased number of chain entanglements, when increasing excipient level, leads to the formation of strong gel layer ([@CR16]) which further delays drug dissolution and/or drug solubilization. Multivariate Data Analysis {#Sec22} -------------------------- The standardized coefficients of the variables in compendial and biorelevant media are presented in Fig. [7](#Fig7){ref-type="fig"}. The two models showed a good predictive power and fit (compendial media: 2 principal components, *Q*^2^ = 0.7 and *R*^2^ = 0.7, biorelevant media: 1 principal component, *Q*^2^ = 0.6 and *R*^2^ = 0.6). The statistical analysis reveals that the impact of HPMC on drug apparent solubility strongly depends on the physicochemical properties of the studied brands. Amine group (compendial media: positive effect, VIP = 3.2; biorelevant media: positive effect, VIP = 2.9) was a significant variable in both models. The variable aminic group term indicates that pronounced increase in drug apparent solubility in HPMC presence is expected for drugs containing a neutral aminic group due to a potential drug-HPMC interaction which enhances drug solubilization. Pronounced reduction in drug apparent solubility is anticipated for highly ionized drugs, potentially due to the slower drug mobility and diffusivity by the presence of the viscous HPMC layer ([@CR57]) as indicated by the significance of drug ionization (compendial media: negative effect, VIP = 2.2; biorelevant media: negative effect, VIP = 2.3) in both models. In biorelevant media, drug lipophilicity (positive effect, VIP = 1.3) and drug solubility (negative effect, VIP = 1.3) were also significant variables in the statistical models. Both variables indicate that pronounced reduction is expected for highly soluble/less lipophilic drugs as the enhanced drug solubilization caused by the presence of bile salts can result in saturation of the powder surface by HPMC particles. The statistical analysis demonstrates that, when HPMC is used as a binder, excipient variability and variation are not significant parameters for the impact of HPMC on drug solubility.Fig. 7Standardized coefficients of the studied variables (and interaction terms) in compendial (blue colour) and biorelevant (red colour) media. Asterisk denotes coefficients of VIP \> 1 (mean, −SE) Road Map of Excipient Effects on Drug Apparent Solubility {#Sec23} --------------------------------------------------------- The road map categorizing the excipient REs on drug apparent solubility according to excipient (HPMC) and drug properties is presented in Fig. [8](#Fig8){ref-type="fig"} (cases where increased drug solubility was caused by a potential shift in the pH of the medium were not considered). The impact of the studied HPMC brands on drug solubility at 24 h depends on the physicochemical properties of the studied compounds. Significant changes (decrease) in drug apparent solubility by HPMC brands of different viscosity type are anticipated for highly soluble drugs, irrespective of drug ionization state (low or highly ionized). For poorly soluble drugs, the impact of the studied HPMC brands on drug apparent solubility depends on the drug ionization state, as significant decrease in drug solubility at 24 h is expected when poorly soluble drugs are highly ionized (irrespective of drug lipophilicity). For poorly soluble/low ionized, HPMC is not found to affect drug apparent solubility, apart from drugs containing a neutral aminic group and for which presence of HPMC may result in significant increase in drug solubility. The generated roadmap reveals that presence of HPMC may be challenging for oral drug performance, as differences in drug apparent solubility caused by the excipient may complicate oral drug absorption and bioavailability.Fig. 8Road map of the effects of the studied excipients (low viscosity: Pharmacoat 606, Pharmacoat 615, high viscosity: HPMC-Sigma) on drug solubility. Red boxes and green boxes indicate significant and insignificant changes in drug solubility by excipient presence, respectively CONCLUSIONS {#Sec24} =========== Excipient presence, variability and variation can affect product performance and present challenges for oral drug bioavailability. HPMC is a commonly used binder in immediate release formulations but can compromise drug dissolution and/or drug solubilization as it forms a viscous layer around particles. In this work, the impact of HPMC viscosity type on drug apparent solubility was assessed in a biopharmaceutical perspective. Solubility studies showed that presence of the studied HPMC brands can significantly change drug apparent solubility, but its magnitude depends on drug physicochemical properties. Significant reduction in the 24-h solubility of highly soluble or highly ionized drugs was observed potentially due to the formation of a viscous layer by HPMC. For poorly soluble drugs containing a neutral aminic group, increase in drug apparent solubility was observed in HPMC presence attributed to the enhanced drug solubilization by the polymeric HPMC chains. Increasing HPMC viscosity and/or level resulted in pronounced decrease in drug apparent solubility especially at early time points, as at the studied excipient levels, the formation of a viscous layer is disrupted by water penetration. The use of multivariate data analysis and the construction of roadmaps revealed that the effects of HPMC on solid oral dosage form performance needs to be studied to better understand the potential impact of these effects on product performance. It is concluded that HPMC presence and variability may present challenges for oral drug absorption and that expanding current knowledge to other excipient types would be beneficial for the successful implementation of excipient variability on QbD approaches. Electronic supplementary material ================================= {#Sec25} ESM 1(DOCX 17 kb) **Publisher's Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Part of this work has been previously included in a poster at the AAPS Annual Meeting in San Diego, USA, November 2017. The AstraZeneca and the University of Bath provided funding the current project.

Background {#Sec1} ========== As a member of the nuclear receptor family, the androgen receptor (AR) plays an important role in breast cancer, and has been identified as a biomarker for a specific molecular subtype of breast cancer. The AR gene expression profile can be used for further classifying receptor-negative tumors as molecular apocrine breast cancer (MABC) \[[@CR1]\]. MDA-MB-453 breast cancer cells have been classified as molecular apocrine by gene profiling studies \[[@CR2]\]. Our previous study has stated that patients with MABC develop distant metastases earlier and have poor prognosis \[[@CR3]\]. As MABC is characterized by increased androgen signaling, the malignant potential of MABC may partly be because of the AR. A study has shown that when the AR in MDA-MB-453 breast cancer cells is knocked down, the cell colony formation rate is significantly decreased, which verifies the fact that the AR regulates the biological behavior of MABC \[[@CR4]\]. Both ligand binding and translocation from the cytoplasm to the nucleus play important roles in the function of the AR \[[@CR5]\]. However, the specific mechanism of cytoplasmic and nuclear translocation of the AR has not been clarified in MABC. In the absence of ligand, the AR remains in a non-active state, which forms a protein complex with heat shock proteins (HSPs) and other co-chaperones in prostate cancer, while the AR becomes active when ligand binds, and it translocates to the nucleus as a transcriptional regulator \[[@CR6]\]. Based on this result, HSPs play crucial roles in the process of AR activation. As a member of the molecular chaperones, HSP27 forms a chaperoning oligomer which can regulate multiple cellular survival and signaling pathways \[[@CR6]\]. However, whether HSP27 combines with the AR during its translocation from the cytoplasm to the nucleus in MABC cells remains unclear. Furthermore, in the process of AR translocation to the nucleus, the rapid phosphorylation of HSP27 is regulated by ligand binding \[[@CR6], [@CR7]\]. There are three serine residues, serine 15, 78, and 82, reported as the sites of human HSP27 phosphorylation \[[@CR8], [@CR9]\]. However, which one of these residues plays an indispensable role for AR cytoplasmic and nuclear translocation still remains unknown. To detect the specific mechanism of AR cytoplasmic and nuclear translocation in MABC cells, androgen and siRNAs specific for *HSP27* were used to analyze the location of the AR and HSP27 in vitro, and their effects on the tumorigenic capacity of MABC cells in vivo. The results of this study could determine the mechanism of the AR in regulating the malignant potential of MABC. Additionally, it aims to explore potential therapeutic targets for patients with MABC. Methods {#Sec2} ======= Cell lines and culture {#Sec3} ---------------------- Because the MDA-MB-453 cell line is classified as a model of MABC \[[@CR2]\], and MCF7 cells as nonMABC cell line \[[@CR10]\], we obtained them from American Type Culture Collection (ATCC, USA) for this study. The MDA-MB-453 cells were cultured in L15 medium (Gibco, USA), containing 10% fetal bovine serum (FBS, Gibco, USA) and 1% penicillin/streptomycin (Life Technologies, USA). MCF7 cells were cultured in DMEM medium (Gibco, USA) which contained 10% FBS and 1% penicillin/streptomycin. Both cell lines were incubated at 37 °C in 5% CO~2~. Plasmids and transfection {#Sec4} ------------------------- The *HSP27* siRNAs and control plasmids were constructed by Genechem (China). Three target sequences for the *HSP27* siRNAs were studied, which included siRNA\#4892-1: 5′-CTGTGAGGACTGTGGATAA-3′, siRNA\#4893-1: 5′-CCCAGCAAATCCCTCTCTA-3′ and siRNA\#4894-2: 5′-GGCAAGTTCCAGGCATTT-3′. The deletion of HSP27 phosphorylation sites (Ser15, Ser78 and Ser82; CS-I0586-Lv201-01, CS-I0586-Lv201-02, and CS-I0586-Lv201-03) were carried out by GeneCopoeia (China, Additional file [1](#MOESM1){ref-type="media"}). The plasmids were amplified in *E. coli*, and then extracted by using the Endotoxin-free Plasmid Size Kit (TIANGEN, China). Cell transfections were performed as follows: firstly, cells were seeded in 6-well plates at a density of 1.0 × 10^4^ cells per well overnight. Subsequently, 2 μg of the constructed plasmids were added to MEM (Gibco, USA), respectively, and incubated for 5 min at 37 °C. Further, the FuGENE Transfection Reagent (Promega, USA) was added and mixed. After incubating for 15 min, the complex solution was added to the cells, and replaced with complete medium 8 h later. The reactions were incubated for 48 h. Quantitative real-time PCR (qPCR) {#Sec5} --------------------------------- The RNAs used in this study were extracted using Trizol reagent (Takara, Japan). Reverse transcription was carried out using the SuperScript RT kit (Takara, Japan). qPCRs were performed according to the manufacturer's protocol using the SYBR Green PCR kit (Toyobo, Japan). The transcript level of *GAPDH* was adopted as an internal control, and the primers used were as follows: *GAPDH*: 5′-GGAAGGTGAAGGTCGGAGTC-3′ and 5′-GTCTTCTGGGTGGCAGTGAT-3′; *AR*: 5′-GGAATTCCTGTGCATGAAA-3′ and 5′-CGAAGTTCATCAAAGAATT-3′; *HSP27*: 5′- GCGTGTCCCTGGATGTCAAC-3′ and 5′-TGTATTTCCGCGTGAAGCAC-3′. Each sample was assayed in triplicate. Western blot analysis {#Sec6} --------------------- Total proteins were extracted with RIPA buffer (Thermo Scientific, USA) and 1 mM PMSF. Cytoplasmic and nuclear subcellular fractionation was performed according to the manufacturer's instructions using the Nuclear and Cytoplasmic Isolation Kit (KeyGEN, China). All proteins were separated on 10% SDS-PAGE (Invitrogen, USA) gels, transferred onto PVDF membranes (Millipore, USA), and then blocked using 5% skim milk. The proteins were detected by incubating the following primary antibodies: anti-AR (AR441; Abcam, USA), anti-estrogen receptor (ER; D8H8; CST, USA), anti-progesterone receptor (PR; 6A1; CST, USA), anti-HSP27 (G3.1, Abcam, USA), anti-HSP27 (phospho S15) (EP2293Y, Abcam, USA), anti-HSP27 (phospho S78) (Y175, Abcam, USA), anti-HSP27 (phospho S82) (EPR7278, Abcam, USA), anti-β-actin (8H10D10, CST, USA), and anti-Histone H3 (D18C8, CST, USA) overnight; and incubated with horseradish peroxidase-labeled anti-rabbit or anti-mouse IgG, followed by detection using the ECL detection kit (Solarbio, China). Each sample was analyzed in triplicate. Co-immunoprecipitation(Co-IP) and western blot {#Sec7} ---------------------------------------------- Co-IP of the AR and HSP27 was carried out according to the manufacturer's protocol using the Pierce Co-IP kit (Thermo Scientific, USA). Briefly, AR or HSP27 antibody (10 μL) was first incubated with the AminoLink Plus coupling resin. The antibody-coupled resin was incubated with protein lysates overnight. Subsequently, the resin was washed, followed by elution of the protein complexes, which bound to the AR or HSP27 antibody. Subsequently, the samples were detected by western blot using the HSP27 or AR antibody as described previously. Each sample was assayed in triplicate. Cell counter kit-8 (CCK8) cell proliferation and clonogenic assays {#Sec8} ------------------------------------------------------------------ Cells (DHT treatment or *HSP27* knock-down) were suspended and seeded in 96-well plates at 5000 cells per well and incubated for 24 h. Further, 10 μL of CCK8 (KeyGEN, China) solution was added into each well on day 1, 2, 3, 4 and 5. After 1--4 h, the absorbance of each well was measured at 450 nm using a microplate reader. Clonogenic assays were carried out using 6-well plates with 1000 cells per well. After 15 days, cells were collected and stained with crystal violet, and then the number of cell colonies was counted. Each sample was assayed in triplicate. Immunofluorescence (IF) assay {#Sec9} ----------------------------- Cells were seeded in 24-well plates. After 24 h, cells were fixed with 4% paraformaldehyde, permeabilized with 0.2% Triton X-100, blocked with bovine serum albumin, incubated with the primary antibodies and fluorescein-labeled secondary antibody, and then DAPI (Thermo Scientific, USA) was used to stain the nucleus. Images were visualized and analyzed using a fluorescence microscope. Each sample was analyzed in triplicate. In vivo experiments {#Sec10} ------------------- Eighteen female BALB/c-nude mice (4--6 weeks old, 18--20 g) were purchased and randomly divided into three groups: MDA-MB-453 cells with DHT treatment, *HSP27* knock-down, and the control group. Cells (2 × 10^6^) were inoculated subcutaneously in the groin. Care and procedures of the mice were provided by the Institution of Animal Use and Care Committee of Tianjin Medical University Cancer Institute and Hospital. All mice were sacrificed until 55 days. Tumor volumes were calculated as previously reported \[[@CR11]\]. All the tumors, livers, and lungs were paraffin-embedded and stained with hematoxylin-eosin (HE). HSP27 and Ki67 expression of mouse tumors were analyzed by immunohistochemistry as previously reported \[[@CR3]\]. In order to detect apoptotic cells, the terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) assay was carried out according to the manufacturer's instructions, and the kit was obtained from KeyGEN (China). Statistical analysis {#Sec11} -------------------- Statistical analyses were performed using SPSS 19.0 software. The data were recorded as means ± standard deviation from at least three independent experiments, and analyzed by one-way ANOVAs and T-tests. *P* \< 0.05 was considered as statistically significant. Results {#Sec12} ======= Determination of cells and DHT working conditions {#Sec13} ------------------------------------------------- Two breast cancer cell lines were used in this study, MDA-MB-453 and MCF7, which are representative of MABC and nonMABC, respectively. Cells were tested by western blot to determine the expression levels of ER, PR, and AR. MDA-MB-453 cells demonstrated no detectable levels of ER or PR, but high levels of AR, while all of these proteins could be detected in MCF7 cells. However, the level of AR protein was 1.73-fold higher in MDA-MB-453 than in MCF7 cells (Fig. [1a](#Fig1){ref-type="fig"}).Fig. 1DHT affected the proliferation ability of breast cancer cells. Western blot analyzed the expression of ER, PR, and AR in MDA-MB-453 and MCF7 cells, and quantified by the bar graph (**a**). CCK8 assay determined the working conditions of DHT (**b**). The proliferation ability of MDA-MB-453 and MCF7 cells treated with or without DHT was measured by CCK8 (**c**) and clonogenic (**d**) assays. \**P* \< 0.05 To determine whether dihydrotestosterone (DHT) could affect the proliferation of breast cancer cells, MCF7 cells were cultured in 96-well plates in medium containing 10% dextran-coated charcoal-treated FBS (DCC-FBS) incubated with or without different concentrations of DHT (10^− 6^, 10^− 7^, and 10^− 8^ M) for 24, 48, or 72 h. CCK8 assays showed that the effects of DHT were dose and time dependent, as the growth inhibitory effects could work best at a concentration of 10^− 8^ M and simultaneously for 48 h (Fig. [1b](#Fig1){ref-type="fig"}). Therefore, a 10^− 8^ M concentration for 48 h was considered as the optimal condition for DHT in the following studies. Effects of androgen on MABC cell proliferation {#Sec14} ---------------------------------------------- The effect of DHT on MDA-MB-453 cell proliferation when exposed to DHT for 1--5 days was examined with the CCK8 assay, and the MCF7 cell line was used as a control. A significant anti-proliferative effect by DHT on MCF7 cells could be observed at the fourth day. However, a significant promotion of cell proliferation in MDA-MB-453 cells by DHT treatment was detected at the third day (Fig. [1c](#Fig1){ref-type="fig"}). Clonogenic assays also showed that the number of colonies formed by MDA-MB-453 cells treated with DHT increased dramatically compared to the control cells (221% vs. 100%, respectively), while it decreased in MCF7 cells with DHT treatment compared to the control cells (56% vs. 100%, respectively; Fig. [1d](#Fig1){ref-type="fig"}). Therefore, proliferation of MDA-MB-453 and MCF7 cells showed opposite responses to DHT treatment. Effects of androgen on AR and HSP27 expression and localization {#Sec15} --------------------------------------------------------------- To determine whether androgen could affect the expression of the AR and HSP27, western blot was carried out, and the results showed a significant increase in the levels of AR protein in both MDA-MB-453 and MCF7 cells treated with DHT. However, DHT had no effect on the expression of HSP27. As HSP27 is phosphorylated on three serine residues (Ser15, Ser78, and Ser82), the phosphorylated forms of HSP27 after DHT treatment were analyzed. Interestingly, the expression of Ser82 significantly increased, while the others remained unchanged (Fig. [2a](#Fig2){ref-type="fig"}). The changes in *AR* and *HSP27* transcript levels in these two cell lines treated with or without DHT were consistent with the changes in protein level. However, the mRNA expression of *AR* was significantly increased while *HSP27*expression remained steady-state (Fig. [2b](#Fig2){ref-type="fig"}).Fig. 2DHT affected the expression and location of AR and HSP27. The effect of DHT on the expression of AR, HSP27, and phosphorylated forms of HSP27 at serine 15, 78 and 82 was analyzed by western blot (**a**). The effect of DHT on the expression of AR and HSP27 mRNAs was measured by qPCR (**b**). The expression and location of AR and HSP27 in the cytoplasm and nucleus after MDA-MB-453 and MCF7 cells treated with or without DHT were analyzed by western blot (**c**) and immunofluorescence (**d**) assays. \**P* \< 0.05 Furthermore, the protein levels of the AR and HSP27 in the cytoplasm and nucleus after subcellular fractionation were analyzed. In MDA-MB-453 cells, the levels of both the AR and HSP27 were significantly increased in the nucleus and decreased in the cytoplasm by DHT treatment, while only the level of the AR increased in the nucleus and decreased in the cytoplasm in MCF7 cells (Fig. [2c](#Fig2){ref-type="fig"}). In addition, IF assays illustrated that both the AR and HSP27 were localized in the nucleus of MDA-MB-453 cells with DHT treatment compared to the cytoplasm in control cells. However, only the AR was localized in the nucleus of MCF7 cells after DHT treatment (Fig. [2d](#Fig2){ref-type="fig"}). Mechanisms of DHT-induced AR and HSP27 cytoplasmic/nuclear translocation {#Sec16} ------------------------------------------------------------------------ To further confirm the mechanisms of DHT-induced AR and HSP27 translocation from the cytoplasm to the nucleus, the interaction between HSP27 and AR was analyzed by Co-IP methods. Firstly, total proteins were extracted from MDA-MB-453 and MCF7 cells treated with DHT. The results indicated that the AR could be detected in HSP27 immunoprecipitated complexes, and vice versa (Fig. [3a](#Fig3){ref-type="fig"}). Subsequently, a cytoplasmic and nuclear subcellular fractionation was performed. Analysis showed that only after DHT treatment, the AR-HSP27 complex could be detected in both the cytoplasm and nucleus of MDA-MB-453 cells (Fig. [3b](#Fig3){ref-type="fig"}), and in the cytoplasm of MCF7 cells (Fig. [3c](#Fig3){ref-type="fig"}).Fig. 3DHT induced the formation of AR-HSP27 complex. Co-IP for AR and HSP27 in the total proteins extracted from MDA-MB-453 and MCF7 cells treated with or without DHT (**a**). Co-IP for AR and HSP27 in the proteins extracted from the cytoplasm and the nucleus of MDA-MB-453 (**b**) and MCF7 (**c**) cells treated with or without DHT Effects of HSP27 on MABC cell proliferation {#Sec17} ------------------------------------------- Cells transfected with siRNAs specific for *HSP27* were examined to explore the effect of HSP27 on AR expression. qPCR and western blot assays determined that *HSP27* siRNA\#4893--1 could clearly decrease the expression of *HSP27* compared with the others (Fig. [4a](#Fig4){ref-type="fig"} and [b](#Fig4){ref-type="fig"}). CCK8 assays showed that *HSP27* knock-down decreased the proliferative ability of MDA-MB-453 cells compared to DHT treated cells. This effect could partly be prevented by DHT treatment. The proliferative ability of MCF7 cells was also decreased by *HSP27* knock-down, which was similar to DHT treatment, and was more obvious when cells were treated with both *HSP27* siRNAs and DHT (Fig. [4c](#Fig4){ref-type="fig"}). Clonogenic assays indicated that for MDA-MB-453 cells, the number of colonies in the *HSP27* knock-down group showed a significant decrease compared to the DHT-treated group (27% vs. 100%, respectively). When cells knocked-down for *HSP27* were treated with DHT again, the number of colonies increased (75%). For MCF7 cells, *HSP27* knock-down decreased the number of colonies, almost to the same extent as with DHT (90% vs. 100%, respectively), and the number was much lower when cells were treated with both *HSP27* siRNAs and DHT (62%, Fig. [4d](#Fig4){ref-type="fig"}).Fig. 4HSP27 affected the expression and location of AR. Verification of the siRNA specific for HSP27 at three sites (4894-2, 4893-1, 4892-1) was measured by qPCR (**a**) and western blot (**b**). The proliferation ability of *HSP27* knock-down cells treated with or without DHT was measured by CCK8 (**c**) and clonogenic (**d**) assays. The effect of *HSP27* knock-down on AR expression in MDA-MB-453 and MCF7 cells was analyzed by western blot (**e**) and qPCR (**f**). Western blot analyzed the level of AR expression in the cytoplasm and nucleus after cells transfected with siRNA specific for HSP27 in MDA-MB-453 (**g**) and MCF7 (**h**) cells. \**P* \< 0.05 Effects of HSP27 on AR expression and localization {#Sec18} -------------------------------------------------- The protein level of the AR in *HSP27* knock-down cells was detected by western blot, and the results showed that the down-regulation of HSP27 decreased the AR protein level in these cells (Fig. [4e](#Fig4){ref-type="fig"}). However, *HSP27* knock-down had no influence on the mRNA levels of *AR* (Fig. [4f](#Fig4){ref-type="fig"}). In addition, *HSP27* knock-down decreased the level of AR translocation into the nucleus in both MDA-MB-453 and MCF7 cells (Fig. [4g](#Fig4){ref-type="fig"} and [h](#Fig4){ref-type="fig"}). The critical sites of HSP27 phosphorylation for AR cytoplasmic/nuclear translocation {#Sec19} ------------------------------------------------------------------------------------ Cells transfected with HSP27 serine 15 deleted, HSP27 serine 78 deleted, and HSP27 serine 82 deleted constructs, respectively, were examined to confirm which site would play a critical role in AR cytoplasmic and nuclear translocation. Western blot was performed to verify the effects of the deletions (Fig. [5a](#Fig5){ref-type="fig"}). Cytoplasmic and nuclear subcellular fractionations from cells transfected with or without the specific plasmids and treated with DHT were analyzed. Compared to the control group, the level of the AR decreased in the nucleus and increased in the cytoplasm in cells transfected with HSP27 serine 82 deleted, while no significant differences were found in cells transfected with HSP27 serine 15 deleted or HSP27 serine 78 deleted and the control group (Fig. [5b](#Fig5){ref-type="fig"}). IF assays further verified that after DHT treatment, both the AR and HSP27 were localized to the cytoplasm of MDA-MB-453 and MCF7 cells transfected with HSP27 serine 82 deleted, compared to the nucleus of cells in the other three groups (Fig. [5c](#Fig5){ref-type="fig"}).Fig. 5The critical sites of HSP27 phosphorylation for AR cytoplasmic/nuclear translocation. Verification of deleting the residues of HSP27 phosphorylation sites (serine 15, 78, and 82) was measured by western blot (**a**). The expression and location of AR and HSP27 in the cytoplasm and nucleus after HSP27 phosphorylation sites deleted and treated with DHT were analyzed by western blot (**b**) and immunofluorescence (**c**) assays Effects of androgen and HSP27 on the tumorigenic capacity of MABC cells {#Sec20} ----------------------------------------------------------------------- MDA-MB-453 cells (2 × 10^6^) were divided into three groups and injected into the mammary fat pad of mice. DHT treatment accelerated tumor growth significantly, while *HSP27* knock-down decreased the tumor formation rate compared to the control group (83.3% vs. 100%) and inhibited tumor volume significantly (Fig. [6a](#Fig6){ref-type="fig"} and [b](#Fig6){ref-type="fig"}).Fig. 6HSP27 and DHT affected the tumor formation and metastasis in vivo using MDA-MB-453 cells. The effect of DHT and *HSP27* knock-down on xenograft tumors growth and formation was described by tumor volume curve (**a**). Images of tumors removed from the mice (**b**). HE staining, immunohistochemistry staining for HSP27 and Ki67, and TUNEL staining were carried out on xenograft tumors (**c**). Representative images of liver with cancer cells invasion (**d**) In sections with hematoxylin-eosin (HE) staining, no obvious difference was found in the tissue structure and cell morphology of xenograft tumors generated from these three groups; however, necrosis could be found in the *HSP27* knock-down group. Tumors generated from cells with DHT treatment expressed significantly higher Ki67 labeling, while tumors with *HSP27* knock-down expressed lower Ki67 compared to the control group. Apoptosis of cancer cells was evaluated by TUNEL staining. Tumors generated from the DHT treatment group showed the smallest proportion of TUNEL-positive cells, while the *HSP27* knock-down group showed the largest proportion compared to the control group (Fig. [6c](#Fig6){ref-type="fig"}). In addition, no metastases were found in the lungs among all these three groups; however, interestingly, there were two mice that developed liver metastases in the DHT treatment group (Fig. [6d](#Fig6){ref-type="fig"}). Discussion {#Sec21} ========== Breast cancer is a heterogeneous tumor that can be reflected in different aspects, such as the classical histopathological features and the more modern molecular classification. MABC is defined as ER-negative, PR-negative, and AR-positive \[[@CR12]\]; furthermore, with a poor prognosis by a series of clinical samples \[[@CR3]\]. In this study, we further explored the mechanisms by which the AR regulates the malignancy of MABC*.* The results showed that HSP27 phosphorylation induced by androgen played a vital role in the process of AR translocation from the cytoplasm to the nucleus, which could affect the proliferation of tumor cells and tumorigenic capacity. As a model of MABC, the proliferation of MDA-MB-453 cells is regulated by androgen in an AR-dependent manner \[[@CR13]--[@CR15]\]. In MABC cells, HSP27 was highly expressed and played a pivotal role in cell proliferation. In accordance with previous studies \[[@CR2], [@CR14], [@CR16]\], we found that androgen could stimulate the proliferation of MDA-MB-453 cells. In contrast to the promotion effect on MABC cells, androgen has a predominantly inhibitory proliferative effect on MCF7 cells \[[@CR17]\], a model of luminal breast cancer \[[@CR10]\]. However, the effects of androgen on MDA-MB-453 cells could be partly rescued by down-regulation of HSP27, suggesting that knocking-down *HSP27* could reduce the androgen promotion effect on MDA-MB-453 cells. Additionally, HSP27 can decrease the toxic effects of oxidized proteins and reduce reactive oxygen species, which results in inhibiting cancer cell apoptosis \[[@CR18]--[@CR20]\]. These results suggest that androgen and HSP27 may interact with each other and co-regulate the proliferative ability of MABC cells. HSP27 is a member of the HSPs family and has been stated to regulate the functions of the AR, such as AR cytoplasmic/nuclear translocation and AR transactivation \[[@CR6], [@CR21]\]. Studies confirm that gain and loss of function of HSP27 are strongly related to the expression of the AR in myogenic cells and prostate cancer cells \[[@CR22], [@CR23]\]. We also found that *HSP27* knock-down significantly decreased the level of AR protein, but no difference could be found in the level of AR mRNA, which suggests that HSP27 might only regulate the protein translation of AR in MABC cells. However, Stope et al. \[[@CR23]\] stated that the decrease in the AR protein level is paralleled with a down-regulation in AR mRNA levels. This difference might be owing to the different cancer cells used for the research; however, the related mechanism should be further analyzed. In malignant tumor cells, the expression of HSP27 is obviously high \[[@CR24], [@CR25]\]. The main mechanism by which the AR and HSP27 regulate the malignant behavior of MABC may rely on the the androgen-triggered phosphorylation of HSP27 that accompanies the AR tanslocation from the cytoplasm to the nucleus where the AR interacts with androgen response elements to promote its genomic activity. HSP27 is directly phosphorylated at three serine residues via the p38 mitogen-activated protein kinase (MAPK) pathway, which affects its subcellular distribution \[[@CR26]--[@CR28]\]. The phosphorylation sites at serine 78 and 82 are identified as the predominant residues of HSP27, and serine 78 can be substituted by asparagine in some animals, but serine 82 is conserved throughout the animals \[[@CR9]\]. In MABC cells, deleting residue serine 82 of the HSP27 phosphorylation sites induced a significant decrease in the expression levels of AR in the nucleus compared to the control group, which indicated that HSP27 phosphorylation at serine 82 might be the critical site for AR cytoplasmic and nuclear translocation. Several studies have confirmed that the positive expression of AR is significantly associated with poor survival, and increased mortality in AR-positive triple negative breast cancer (TNBC) \[[@CR3], [@CR29]--[@CR31]\]. HSP27 is reported to be associated with a decreased survival in breast cancer \[[@CR32]\]. In accordance with clinical research, we found that DHT treatment increased xenograft tumor volume and distant metastasis, while HSP27 knock-down inhibited tumor growth greatly, which indicated that the AR and HSP27 might interact with each other and co-influence the development of MABC. Studies have stated that AR antagonists are able to induce breast cancer cell apoptosis and decrease tumor proliferation significantly in TNBC cell lines \[[@CR33], [@CR34]\]. In addition, the HSP second generation antisense oligonucleotide targeting HSP27 can increase drug efficacy in pancreatic and prostate cancer xenograft models \[[@CR35]--[@CR37]\]. Based on these results, although no valid endocrine therapy is suggested for MABC, the combination of AR antagonists and HSP27 inhibitors could be a potential therapeutic regimen. Of course, more xenograft models and clinical trials should be carried out to confirm this hypothesis. Conclusions {#Sec22} =========== In conclusion, this study confirmed that activation of AR and HSP27 by androgen could increase the proliferative ability of MABC cells and the growth of xenograft tumors. HSP27 phosphorylation on residue serine 82, induced by androgen, played a critical role in the process of AR translocation from the cytoplasm to the nucleus. The AR and HSP27 could form a protein complex, which was the main factor in AR regulating the malignancy behavior of tumor cells, and could present a new therapeutic regimen in clinical therapies of MABC. Additional file =============== {#Sec23} Additional file 1:The sequence of the deleted HSP27 phosphorylation sites. (DOCX 13 kb) AR : Androgen receptor CCK8 : Cell counter kit-8 Co-IP : Co-immunoprecipitation DCC-FBS : Dextran-coated charcoal-treated FBS DHT : Dihydrotestosterone ER : Estrogen receptor HE : Hematoxylin-eosin HSPs : Heat shock proteins IF : Immunofluorescence MABC : Molecular apocrine breast cancer MAPK : Mitogen-activated protein kinase PR : Progesterone receptor TNBC : Triple negative breast cancer TUNEL : Terminal deoxynucleotidyl transferase dUTP nick end labeling **Electronic supplementary material** The online version of this article (10.1186/s13046-018-0762-y) contains supplementary material, which is available to authorized users. Funding {#FPar1} ======= This work was financially supported by National Science Foundation of China (Grant number: 81172532). Availability of data and materials {#FPar2} ================================== Please contact the corresponding author for all data requests. YN and XZL conceived and designed the experiments. XZL and JJL performed most of the experiments. CYF performed the statistical analysis. LC and GMX cultured the cells. FL, SLW and JJ performed some animal experiments. XZL, CYF and YN wrote the manuscript. All authors read and approved the final manuscript. Ethics approval and consent to participate {#FPar3} ========================================== For the in vivo tumor growth study, 18 mice were used. All studies were performed according to the American Association for the Accreditation of Laboratory Animal Care guidelines for humane treatment of animals and adhered to national and international standards. And it was approved by the Institution of Animal Use and Care Committee of Tianjin Medical University Cancer Institute and Hospital. Competing interests {#FPar4} =================== The authors declare that they have no competing interests. Publisher's Note {#FPar5} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Introduction {#s0001} ============ Rising healthcare costs and the quadruple burden of disease demand critical scrutiny of healthcare organisation and practice globally. It is estimated that by 2020 chronic diseases will account for 73% of all deaths and 60% of the global burden of disease.^[@CIT0001]^ The growing burden of chronic diseases is risking the sustainability of healthcare systems. In 2009, the World Health Assembly placed people and person-centredness at the centre of health care.^[@CIT0002]^ Efforts to strengthen health systems and improve health outcomes should focus not only on technical and structural aspects but also the user experience and process of care by measuring features known to be essential for cost-effective personal health care. The South African (SA) healthcare system has entered a period of major reform that includes primary healthcare (PHC) re-engineering and implementation of a national health insurance (NHI). The SA government has committed itself to improving PHC, the stated orientation of SA district health services.^[@CIT0003]^ After 21 years of democratic governance, increased PHC funding and other measures to undo the legacies of apartheid policies and practices, there is widespread concern that desired health outcomes are not being achieved and that gross inequities in health status and access to services continue.^[@CIT0004],[@CIT0005]^ In Western Cape (one of nine provinces), there are initiatives to improve the quality of primary care in line with the Provincial Department of Health's Vision 2030. It includes a commitment to shift the focus from illness to wellness and from patient care to person-centred care and engaged leadership.^[@CIT0006]^ This shift reflects significant changes in thinking and will require reorientation of staff, clinical practice, organisational management and resource allocation. Effect should also be given to the right of users to participate individually and collectively in the planning and implementation of their health care as espoused in 1978 in the Declaration of Alma-Ata, for example, through PHC stakeholder partnerships.^[@CIT0007]^ Better health outcomes, reduced costs and reduced inequity are among the results of an integrated, publically funded health system that ensures user access to elements known to be essential for cost-effective PHC, and on which performance can be measured.^[@CIT0008],[@CIT0009],[@CIT0010],[@CIT0011]^ The Primary Care Assessment Tool (PCAT) is a validated instrument^[@CIT0012]^ used in developing and developed contexts internationally to determine the extent to which PHC is aligned with the evidence for cost-effective care based on users' access to, and utilisation in their care, of these essential features (domains) -- namely first contact access (to a primary care practitioner); ongoing care (relational continuity); comprehensive care; coordinated care; family- and community-orientated care and cultural competence. These are also known as the principles of family medicine which are embedded in primary care and family physician training in South Africa. Together with implementation of the NHI and provincial interventions, they provide SA with a significant opportunity to align PHC with evidence in favour of comprehensive, person-centred primary care. A baseline measure of PHC performance and organisation on these domains could help to guide reforms and other interventions prior to them taking root, and to monitor impact. This study reflects a partnership between service providers (Western Cape Provincial District Health Services) and a tertiary education institution (Division of Family Medicine, University of Cape Town) to support efforts to improve primary care in the Western Cape Province. The main purpose of the study was to obtain a baseline measure of performance and organisation at comprehensive primary care facilities (PCFs) by determining users' experience and managers' and practitioners' assessments of primary care. By surveying users, providers (practitioners) and managers, the study aimed to determine gaps between users' experience of primary care and desired performance on the universally accepted essential primary care domains and to identify those that need strengthening and aligning with evidence-based care and with provincial and national health plans. Study objectives included determining the demographic profile of primary care users in urban and rural districts; measuring performance on 11 primary care domains and subdomains, and a total primary care score in selected health districts in Western Cape; describing user experiences of care by demographic variables; describing and comparing primary care user, practitioner and manager assessments (scores); comparing domain, subdomain and total primary care scores between districts; and reporting the main findings to PHC the user, practitioner and manager stakeholders. In 2011, Cape Town had a population of approximately 3 810 000; 73.0% were uninsured and dependent upon the public sector, and the Cape Winelands district (CWD) had a population of approximately 768 300; 77.0% were uninsured and dependent on the public sector. The large uninsured proportion of the population skews the quadruple burden of disease towards the public sector, that is, infectious diseases (including child mortality and malnutrition), degenerative and chronic diseases, injuries and HIV-related diseases. Utilisation rates and annual patient visits at PCFs were 3.7/10 415 052 (Cape Town) and 3.4/1 978 282 (CWD).^[@CIT0013]^ The investigators conducted the first South African PCAT study (unpublished) in two Cape Town metropolitan sub-districts (eight urban PCFs) in 2011. This second study was conducted in urban and rural PCFs -- in the remaining six of the eight sub-districts in Cape Town (Southern = Western, Eastern = Khayelitsha and Northern = Tygerberg) and four of the five sub-districts in the CWD. All PCFs studied are comprehensive PHC facilities in the provincial health department. Following the experience gained and lessons learned in the 2011 study in which the expanded (E) forms of the original USA PCAT questionnaires were used -- adult expanded (AE), provider (practitioner) expanded (PE) and facility manager expanded (FE) -- a cross-cultural adaptation and validation of the original expanded version was conducted and the expanded South African ZA PCAT produced for use in this study. The validation method, content and domain definitions of the ZA PCAT are described in an earlier paper.^[@CIT0014]^ In summary, a combination of the face validation, Delphi and nominal group technique methods was used with two expert panels to achieve consensus on the relevance of PCAT domains and their items (domain questions) for use in SA. Consensus in favour of inclusion was achieved for all 9 domains. One new domain, the PHC team, was added. Of the original 95 items, 3 achieved \< 70% agreement and were excluded; 19 new items were added; a few items needed rephrasing for local comprehension; the demographic section was adapted for local socio-economic conditions; and isiXhosa and Afrikaans translations of the ZA PCAT were developed. Method {#s0002} ====== This was a multilevel cross-sectional study of primary care users, practitioners and managers in six of eight urban and/or peri-urban and four of five rural sub-districts in Cape Town and CWD, respectively. The remaining urban sub-districts were studied in a pilot study in 2011. The 4 rural sub-districts were considered by CWD management as sufficiently representative of the whole district; including all five rural districts in the province was beyond the capacity of our study budget. In each Cape Town substructure, PCFs were stratified into large, medium and small clusters as determined by user visits per month to ensure representation for size; opening hours (24 h and 8 h); and the 3 main languages spoken, that is, that demographic diversity was reflected across the metro sample. One PCF was selected from each stratum in collaboration with district managers. In the CWD, the largest PCF in each of the 4 sub-districts was included because they best reflected user and staff diversity, and the range of PHC services offered in CWD PCFs, that is, using cluster sampling, 13 PCFs covering 6 urban and/or peri-urban and 4 rural sub-districts were included in the sample. The outcome measures are performance scores on 11 key elements (subdomains) of primary care and total primary care score. The user sample size calculation was based on primary care measures derived from a previous PCAT study (2011) with an estimated mean total primary care score between 2 PCFs of 2.5 and 2.9 with a standard deviation of 0.8. The minimum sample size required per PCF was 85 (α = 0.05 and a power = 90%). The total number of users that were interviewed in the 13 PCFs was 1432. The PCF with the smallest and the largest sample size was 97 and 123 users, respectively. All full-time doctors and nurse practitioners (140) working in the 13 PCFs and all PCF managers (87) working in the 10 sub-districts represented were invited to participate. Participant selection {#s20003} --------------------- For users, a systematic sampling method was used in each PCF. From the 2011 study, we estimated that each interviewer would conduct seven interviews per day. We aimed to survey approx. 20 users per day in each PCF so that approximately 100 interviews would be completed over 5 consecutive days from Monday to Friday. Using the average number of patients seen per day at each PCF (obtained from PCF records), the sampling interval (*n*th admission folder) was calculated by dividing the total number of daily admissions by 21 (3 interviewers x 7 interviews = 21). Counting from the top of the pile of admissions folders, every *n*th folder was assessed for eligibility using the inclusion and exclusion criteria. Only patients (users) ≥ 18 years of age with a minimum of three previous visits to their respective PCFs were included Responses to the ZA PCAT AE require patients' experience of primary care over time; to be eligible respondents had to have visited the PCF at least three times prior to the day of the interview. Where a patient was not eligible or did not consent, the next folder was selected and so on. Fieldworker management and data quality control was by an experienced research assistant -- a member of the fieldworker training team -- who was on-site throughout the user data collection. Only full-time PCF practitioners and managers were included in the practitioner and manager samples; locums, interns and practitioners doing their community service were excluded. The user survey was conducted by trained fieldworkers using the ZA PCAT AE. Fieldworkers selected for the study (CVs were scrutinised and interviews conducted) had a minimum of 12 years of schooling and previous research fieldwork training including data collection and research ethics. Each was fluent in at least two of the three official languages spoken in Western Cape -- English, isiXhosa and Afrikaans. Following selection, the fieldworkers were trained to administer the ZA PCAT in a 3-day training workshop based on the original authors' training manual.^[@CIT0015]^ The manual content and training method were adapted and aligned with the cross-culturally validated ZA PCAT by the research investigator team. The practitioner ZA PCAT (PE) was self-administered -- not necessarily on the same day -- after it was explained by an investigator at respective PCF clinical practitioner meetings. In the metro, the manager ZA PCAT (FE) was administered by appointment by the trained study investigators. In order to avoid the cost of travelling to individual appointments in the rural sub-districts -- a round trip of 240 km -- rural PCF managers self-administered the ZA PCAT in sub-district manager groups after their monthly management meetings. An investigator explained the PCAT before completion and was available to respond to any queries. Information meetings were held prior to the start of the study to brief and obtain cooperation of PCF staff, sub-district and district managers, and district directors. This included site visits to determine how best to adhere to the study protocol and yet keep disruption to clinic operations to a minimum. These meetings paved the way for meetings with stakeholders to report the main findings. Summary posters were placed in PCF waiting rooms and presentations made to user-represented PCF committees; hard copy reports and presentations were presented at PCF staff meetings (practitioners and managers) and sub-district and district manager meetings which included district directors. Verbal presentations and hard copy reports were also given to provincial health department senior management. Data analysis {#s20004} ------------- The method of data scoring, analysis and formulation of results followed the steps in the PCAT manuals for the three expanded user, practitioner and manager PCAT versions (AE, PE and FE, respectively). These are obtained from Johns Hopkins Primary Care Policy Center.^[@CIT0015]^ Data from each of the three informant groups were analysed separately. The PCAT Likert scale responses and analysis are the same for user, practitioner and manager questionnaires. Responses are scored on a 1--4 scale with 1 indicating 'definitely not', 2 indicating 'probably not', 3 indicating 'probably', 4 indicating 'definitely'. A fifth 'not sure/don't remember' response option is scored as 2 (except for the comprehensive services domain where 'not sure/don't remember' is scored as 0). The PCAT methodology calculates the score for each subdomain by summing the scores of the items in that subdomain (after reverse coding of items where required by the data analysis method) divided by the number of items to produce a mean score. Questionnaire data were entered into EpiData^[@CIT0016]^ and exported to Stata version 12.0 for statistical analysis.^[@CIT0017]^ The internal consistency of the scores for users was examined using Cronbach's alpha coefficient. The Shapiro-Wilk test indicated that the PCAT scores were not normally distributed; hence, we constructed a binary variable. A score ≥ 3 is considered 'acceptable to good performance' and \< 3 as 'poor performance'. Multivariate binomial regression analysis^[@CIT0018]^ was used to estimate the prevalence ratios (PRs) and assess the association between user's primary care score and socio-demographic characteristics (independent variables). For all analyses, a *p*-value of less than 0.05 and a 95% confidence interval that did not span unity were considered the thresholds of statistical significance. Results {#s0005} ======= The ZA PCAT AE was administered to 1439 users (acceptance rate 90%). Only 7 user questionnaires (AE) were unsuitable for analysis due to missing or incomplete data; 1432 user questionnaires were analysed. Of the approximately 140 eligible practitioners in the PCFs studied, 100 completed the ZA PCAT PE (acceptance rate 71%). Of the 87 eligible PCF managers and deputy managers in the 10 sub-districts studied, 64 completed the ZA PCAT FE (acceptance rate 74%). All practitioner (100) and manager (64) questionnaires were complete and analysed. [Table 1](#T0001){ref-type="table"} summarises the main user characteristics: 68.9% are female; 62.9% attended but only 9.9% completed high school and 2.7% had further education; 22.8% live in informal dwellings; and 35.3% are employed. The age distribution of users, 18--40 (34.8%), 40--54 (32.6%) and 55 + (32.6%), was similar in proportion. It also shows that 60.5% of the users perceived their health status as 'good'. ###### User characteristics. Variables Number \% --------------------------------------------------------- -------- ------ **Cape Town metro sub-districts**  Southern-Western 351 24.5  Northern-Tygerberg 307 21.4  Eastern-Khayelitsha 339 23.7 Cape Winelands (Rural) 435 30.4 **Gender**[†](#TFN0001){ref-type="table-fn"}  Male 443 30.9  Female 987 68.9 **Age-group**  \< 40 498 34.8  40--54 467 32.6  55+ 467 32.6 **Health Status**[†](#TFN0001){ref-type="table-fn"}  Poor 562 39.3  Good 866 60.5 **Employment**  No 927 64.7  Yes 505 35.3 **Educational level**[†](#TFN0001){ref-type="table-fn"}  No schooling or some primary schooling 525 36.7  Secondary (and higher) 901 62.9 **Type of dwelling**  Informal 326 22.8  Formal 1,106 77.2 **Language**[†](#TFN0001){ref-type="table-fn"}  English 244 17.0  Afrikaans 611 42.7  Xhosa 491 34.3  Other 83 5.8 Missing values for the following variables in brackets: Gender (2); Health Status (4); Language (3); Educational level (6). [Table 2](#T0002){ref-type="table"} shows the descriptive results of user scores by PCAT domain. The means reflect positive experiences in seven of the 11 primary care subdomains. The mean scores for first contact access, comprehensiveness (services provided), family centredness and community-orientated are below 3. Cronbach's alpha estimates show acceptable reliability for 8 out of the 11 primary care elements (range 0.7--0.9). Coefficients for first contact utilisation (0.4) and coordination (information systems) (0.2) are far below the minimal acceptable score (Cronbach's alpha = 0.7) to be reliable as a coherent domain.^[@CIT0012],[@CIT0019]^ The Cronbach's alpha coefficient for family centredness (0.6) is marginally below 0.7. Cronbach's alpha levels were far below for first-contact utilisation and coordination information systems, indicating that the homogeneity of variances among items within the scale was low. ###### Distribution of USER scores by PCAT subdomain (2013). Subdomains Number of items Variable Cronbach's alpha ---------------------------------------- ----------------- ---------- ------------------ ---------- ----- First contact -- utilisation 3 3.1 0.6 1.0--4.0 0.4 First contact -- access 17 2.5 0.4 1.3--3.6 0.7 Ongoing care 15 3.0 0.5 1.2--4.0 0.7 Coordination 10 3.2 0.7 1.4--4.0 0.8 Coordination (information systems) 3 3.2 0.7 1.0--4.0 0.2 Comprehensiveness (services available) 28 3.1 0.6 1.1--4.0 0.9 Comprehensiveness (services provided) 12 2.7 0.6 0.6--4.0 0.7 Family-centredness 3 2.8 1.0 1.0--4.0 0.6 Community orientation 6 2.3 0.8 1.0--4.0 0.8 Culturally competent 5 3.4 0.8 1.0--4.0 0.8 Primary healthcare team 7 3.4 0.6 1.3--4.0 0.7 PCAT, Primary Care Assessment Tool. [Table 3](#T0003){ref-type="table"} (graphically presented in [Figure 1](#F0001){ref-type="fig"}) shows the proportion of users, providers and managers who rated each domain as 'acceptable to good' performance. 11.5% of users scored performance on access as 'acceptable to good'; 20.8% scored community orientation and 39.9% scored comprehensive services provided 'acceptable to good'. The remaining subdomains were scored as acceptable to good by at least 50.0% of patients. Among the providers (doctor and clinical nurse practitioners), the lowest 'acceptable to good' score was for access (33.3%) and the highest for comprehensive services available (100.0%) and primary healthcare team (98.0%). Managers scored access (13.5%) and family centredness (45.6%) lowest; and comprehensive services available (90.6%) and primary healthcare team (85.9%) highest. The total primary care scores of provider and manager are similar but significantly higher than users' total score. {#F0001} ###### Proportion of users (AE), providers (PE) and managers (FE) who scored subdomains ≥ 3 (i.e. rated performance as'good'). Subdomains AE (%) FE (%) PE (%) ---------------------------------------- ---------- ---------- ---------- First contact -- Access 11.5 13.5 33.3 Ongoing care 62.2 53.1 62.0 Coordination 64.8 78.1 75.0 Coordination (information systems) 83.3 79.7 80.0 Comprehensiveness (services available) 62.4 90.6 100.0 Comprehensiveness (services provided) 39.9 65.3 72.0 Family-centredness 52.7 45.6 77.0 Community orientation 20.8 67.2 81.0 Culturally competent 73.5 50.0 69.0 Primary healthcare team 76.1 85.9 98.0 **Total primary care score** **50.2** **82.8** **88.0** [Table 4](#T0004){ref-type="table"} (graphically presented in [Figure 2](#F0002){ref-type="fig"}) shows the proportion of users who rated each domain as 'acceptable to good' performance by urban and rural regions. Urban users scored first contact access and cultural competence significantly higher than rural users, whereas rural users scored coordination, comprehensiveness (services available), comprehensiveness (services provided) and family-centredness significantly higher. {#F0002} ###### Proportion of users who scored subdomains ≥ 3 (i.e. rated performance as 'good') by metro and rural regions. Subdomains Metro (%) Rural (%) *p*-value ---------------------------------------- ----------- ----------- ------------ First contact -- utilisation 90.8 92.4 0.312 First contact -- access 15.2 3.0 \< 0.001\* Ongoing care 63.1 60.2 0.305 Coordination 59.9 76.9 0.015\* Coordination (information systems) 82.9 84.1 0.579 Comprehensiveness (services available) 60.1 67.6 0.007\* Comprehensiveness (services provided) 38.1 43.9 0.039\* Family-centredness 50.3 58.2 0.006\* Community orientation 22.0 18.2 0.103 Culturally competent 78.1 62.8 \< 0.001\* Primary healthcare team 75.2 78.2 0.231 Primary care score 52.4 45.3 0.014\* Multivariate analysis: The socio-demographic variables ([Table 1](#T0001){ref-type="table"}) that were considered as possible predictors of user's total primary care score were included in the binomial regression model as independent variables. The users' total primary score, with a range of 1.7--3.7, was entered as binary dependent variable (a score ≥ 3 is considered 'acceptable to good performance'. The results show that users who resided in Northern--Tygerberg and Eastern--Khayelitsha sub-districts; users ≥ 40 years; and users who rated their health status as good were significantly associated with a positive total primary care score (≥ 3 good): $$\begin{array}{l} {\text{N/T }PR = 1.5\lbrack P = 0.002\, CI:1.2 - 1.8\rbrack;\text{E/K }PR = 1.6\lbrack P < 0.001\, CI:1.3 - 2.1\rbrack;} \\ {\geq 40\text{ years }PR = 1.2\lbrack P = 0.017\, CI:1.0 - 1.5\rbrack;\text{ Good health status}} \\ {PR = 1.4\lbrack P < 0.001\, CI:1.2 - 1.6\rbrack} \\ \end{array}$$ Users were not surveyed for their responses to the waiting room posters. Only two functional PCF committees were identified for presentation of the main findings. They identified with the user findings; responses were positive and encouraging and showed a willingness to work with PCF staff to improve care. Manager and practitioner responses to reports are noted below. Discussion {#s0006} ========== The demographic findings ([Table 1](#T0001){ref-type="table"}) show that public sector PHC facilities serve largely female users (68.9%); that more than a third (36.7%) of users have no formal schooling or only some primary school education; and that 23.0% live in informal housing in contrast to national figures (15.8% and 12.9%, respectively).^[@CIT0020]^ Only 35.3% (female 34.3%; male 37.7%) reported any form of employment - whether part-time, full-time, informal or formal, that is, 64.7% were unemployed -- by definition a larger proportion than 'narrow' unemployment defined as the proportion of unemployed actively seeking work -- 24.3% nationally.^[@CIT0021]^ Three socio-demographic factors ([Table 1](#T0001){ref-type="table"}) emerged as predictors of total primary care score: users' area of residence; age (≥ 40 years); and self-reported health status, that is, users who rated their health status as 'good'. These were significantly associated with a positive total primary care score (≥ 3 = good). Although analysis of the demographic data suggests that sub-districts with higher unemployment and a lower education level are associated with higher total scores, it is not a conclusive finding in this study. A longer standing relationship with a PCF as the chosen provider may account for age as a predictor. Older users are more likely to have chosen a PCF over other options and therefore more likely to be satisfied. Users who rated their health as 'good' also rated primary care performance as better than those who rated their health as poor. A similar finding is reported in a Korean study where higher quality primary care was found to be associated with good self-rated health status.^[@CIT0022]^ In a Brazilian study, users' demographic and socio-economic factors were not predictors of primary care score.^[@CIT0023]^ While socio-demographic factors such as unemployment, educational level and type of dwelling have not been shown to be predictors of primary care score in this study, they may nevertheless add to the complexity and challenge of providing primary care to communities where these are prevalent.^[@CIT0024]^ Smith and Haggerty note that poor literacy is an independent health risk; high-literacy users may be less dependent on health service interventions and more able to act on providers' advice.^[@CIT0025]^ Most users (60.5%) perceived their health status to be good to excellent despite 75.8% reporting that they attended for a special medical problem; 56.0% had a physical, mental or emotional problem lasting or likely to last longer than 1 year. It is surprising that similar proportions who attended and who were not attending for a chronic condition rated their health as good (59.6% and 63.0%, respectively). It is unlikely that this self-rated 'good' health indicates good disease control in the majority of these patients; local chronic disease care audits generally reflect poor control. A recent study in 10 PCFs including facilities in our study confirms previous chronic disease audits -- that the average proportion of users attending these facilities for chronic disease care is high (82.0%) and the quality of care and control for diabetes and hypertension is poor.^[@CIT0026]^ The apparent discrepancy between perceived good health and evidence of poor disease control may be due to users' interpretation of health and illness, that is, health is 'good' when not feeling ill or if function is not significantly limited. This may be particularly so in largely silent yet prevalent conditions such as hypertension and diabetes. The self-rated health status may be a function of a low expectation of heath care and disease control, that users were interviewed on site, and users' association with respective PCFs. In the context of a quadruple disease burden, poor disease control, a 65.0% comorbidity^[@CIT0027]^ and proportionally lower male user attendance, a falsely founded perception of good health has important implications for primary care practitioners, managers and initiatives aimed at improving the quality of care. Such a perception underscores the importance of good relational continuity necessary to build long-term therapeutic partnerships for effective management of chronic disease. When comparing urban and rural districts' user scores (Cape Town Metro and Cape Winelands, respectively), the patterns created by plotting the scores on the radar graph ([Figure 1](#F0001){ref-type="fig"}) are similar even though there are significant differences between their domain and total primary care scores ([Table 4](#T0004){ref-type="table"}). The similar patterns suggest similar strengths and weaknesses in performance on the essential features measured -- expected given standard provincial PHC packages and treatment guidelines, management training and protocols, and comparable clinician training and practice. However when comparing user, provider and manager scores ([Table 3](#T0003){ref-type="table"}; [Figure 1](#F0001){ref-type="fig"}), the patterns are different; differences between the three key stakeholders are greater than the differences when comparing user scores by region (not shown). Users scored comprehensive services available, comprehensive services provided and community orientation, significantly lower than providers and managers. In general, the providers (doctors and CNPs) scored PHC performance on most domains higher than both managers and users. Managers' scores are closer to those of users' experience ([Table 3](#T0003){ref-type="table"}). Practitioners' scores are optimistic relative to managers and users -- a cause for concern, practitioners being the frontline providers of care. It is hoped that these findings will increase awareness among managers and practitioners of the gaps between the user experience and perceived performance; and encourage a search for ways to reduce the gaps between current and desired performance, for example, by developing and implementing interventions aimed at improving staff adherence to evidence-based care. The validity of comparing user (AE), practitioner (PE) and manager (FE) domain scores may be debatable given their different roles and perspectives. However, a co-author of the original PCAT agrees[^1^](#FN0001){ref-type="fn"} they can be compared.^[@CIT0028]^ We can only speculate on reasons for the optimistic practitioner scores relative to users and managers. Much of what the PCAT measures are process transactions which take place during user consultations with practitioners - the heart of personal primary care. The relatively optimistic practitioner scores may reflect concern at being judged by low scores. Viewed together, practitioner and manager scores for total primary care are optimistic relative to users' scores, that is, users' experience of primary care is significantly different to what managers and providers think they are delivering. Further research is required to explain this; the findings could help to identify and implement interventions to improve practitioner and manager performance on essential domains of primary care. Joint consideration of the disparate performance scores by these key stakeholders can serve as an opportunity to build a primary care stakeholder partnership that includes generating, implementing and monitoring appropriate interventions aimed at improving performance thereby improving health outcomes. *Access* involves the extent to which primary care services are accessible to users when needed. Domain items include opening times, waiting times and staff attitudes, that is, both structural and process factors. Relative to other scores, performance on first contact access was rated as poor by users, practitioners and managers (11.5%, 33.5% and 13.5%, respectively) suggesting agreement that access needs attention. Of note is that when measured on a scale of 1--4, a Canadian study reported a mean score of 2.21 (3 being acceptable) for first-contact access^[@CIT0029]^ -- a finding similar to our 2.5 ([Table 3](#T0003){ref-type="table"}). Stakeholder agreement on poor access contrasts with the high user score for first contact (utilisation), that is, high user affiliation with PCFs in urban (90.8%) and rural (92.4%) regions ([Table 4](#T0004){ref-type="table"}) -- a finding supported by a high proportion of users reporting an association with their respective primary care facilities of \> 5 years. High user utilisation and affiliation contrasts with the lower user score (62.2%) for ongoing care ([Table 3](#T0003){ref-type="table"}) and suggests missed opportunities to strengthen relational continuity and to build strong therapeutic user-practitioner and user-facility relationships by practicing good continuity of care. This is an important finding given the influence of continuity on other domain performance (discussed below). *Ongoing (continuing) care* refers to the use of a regular source of care over time that is not limited to certain types of healthcare needs, that is, health care is provided regardless of the presence or absence of disease. It includes building a user--practitioner relationship based on trust and practitioner knowledge of patients and their families. Although 62.2% of the users rated ongoing care overall as good (as did the practitioners), it is generally accepted among managers and practitioners in the districts studied (confirmed at our report-back meetings with practitioners and managers) that relational continuity of care is poor and that the service is not structured to encourage nor support relational continuity. The higher than expected user score for ongoing care in this study could be due to a low user expectation. Here too, practitioner scores are relatively optimistic regarding how well and how much they know about their patients. Sub-analysis of the 15 items in this domain may help to explain these discrepancies. Detailed analysis of the ongoing care domain was not an objective of this study; however, it deserves more attention here. Relational continuity is considered the most important element (principle) of primary care.^[@CIT0030]^ It is embedded in this 15-item domain (e.g. item D1: *Do you see the same practitioner at each primary care visit?* No = 70.0%). Two unpublished audits of continuity conducted in Cape Town sub-districts (also included in this study) reported poor continuity of care. When continuity was defined as seeing the same practitioner for at least 2/3 (66.0%) of the consultations, it was present in 21.4% (95% CI: 13.4--31.3), whereas 92% (95% CI: 84.6--96.8) of the respondents preferred seeing the same doctor at each consultation.^[@CIT0031]^ A large body of evidence summarised in a review by Haggerty et al.^[@CIT0032]^ shows that poor continuity results in fragmented care and poor and costly outcomes. The authors note that policy reports and charters call for enhancement of continuity in healthcare delivery, for example, the Ljubljana Charter on Reforming Health Care urges that heath reforms reinforce continuity.^[@CIT0033]^ It is difficult to envisage other components of ongoing care receiving the necessary attention in the absence of continuity -- such as users being known and understood by their practitioners. Involvement of consecutive practitioners in an individual's primary care leads to reduced accountability -- labelled the 'collusion of anonymity' by Balint.^[@CIT0034]^ Such practice is likely to be aggravated where there is a high patient-to-practitioner ratio, a feature of the public sector where practitioners generally see large numbers of patients per day. Under such circumstances comprehensive care also suffers. A study examining the quality of chronic disease care found that models of primary care with more than four family physicians and high patient ratios performed less well on PCAT domains than those with fewer family physicians and lower patient--practitioner ratios; and that community health centres in Canada and the USA performed better than office-based family physicians and hospital outpatient clinics.^[@CIT0035]^ In a systematic review examining the relationship between (sustained) continuity of care and the quality of care, continuity was found to be associated with patient satisfaction, decreased hospitalisations and emergency department visits, and better acceptance of preventive services particularly in the care of chronic conditions.^[@CIT0036]^ Continuity, when practiced by primary care professionals in a regulated health system is associated with better health outcomes and lower costs than in fragmented, unregulated systems -- such as market driven systems where patients initiate visits to medical specialists.^[@CIT0008]^ It is worth noting that continuity is increasingly being highlighted in SA national and provincial health policy documents such as the Western Cape's Vision 2030.^[@CIT0037]^ Primary care practice should be aligned with such policy initiatives. *Coordination* (subdomains: coordination of information systems and coordination of information) links healthcare events and services, and requires mechanisms to communicate and incorporate information into patient care plans. It includes the responsibility and obligation to transfer information to and receive it from other resources involved in the patient's care (coordination of information systems), and to develop and implement an appropriate plan for healthcare management and disease prevention. All three stakeholders scored performance on coordination of information as good -- both rural and urban -- reflecting longstanding public sector record keeping and referral practice. Access to secondary and tertiary care (gatekeeping) is by referral only from a PCF practitioner. There is usually exchange of at least some patient information between PCFs and referral hospitals using paper-based or electronic referral letters. While misplaced or misfiled user records are not uncommon and fragments care, records are generally available at visits. User access to their records is also good; in many PCFs patients carry their own records between service points, for example, from the practitioner to pharmacist. In contrast, performance on the sharing of test results with users shows a significant difference between users on the one hand and managers and practitioners on the other. Relative to users' experience, managers and practitioners are optimistic about their performance. The lower user score may indicate that results, for example, of special investigations are not communicated in ways they understand. This needs further research. Effective information sharing is an essential ingredient for a therapeutic user--provider partnership. We are unable to explain the difference between the urban and rural scores. *Comprehensive care* (subdomains: comprehensive services available and comprehensive services provided) provides a range of essential personal health services that promote and preserve health as well as services for illness and disability, and arranges access to services elsewhere for uncommon or special needs. The importance and impact of comprehensive care is unique to primary care compared with other clinical disciplines. Comprehensive care provided includes the opportunistic provision of information and screening for health promotion, disease prevention and early detection guided by epidemiology and the health profile of user communities. Users, practitioners and managers scored comprehensive care provided lower than comprehensive services available. User scores (39.9% and 62.4%, respectively) were considerably lower than practitioner and manager scores. The scores suggest that while services are available at PCFs, they are not applied at an acceptable level of performance; and that practitioners and managers are optimistic about their performance. A study involving 14 PCFs in Cape Town (including PCFs in this study) reported a missed opportunity rate of 25.0% -- 46.0% (depending on a strict or loose definition) for reproductive and mental health^[@CIT0038]^ -- a finding supported by more recent studies showing poor chronic disease control and staff adherence to policy.^[@CIT0026],[@CIT0039]^ A Quebec study involving 100 PCFs across urban and rural settings found that even among users who had regular family physicians and reported experiencing high relational continuity, only 56.0% reported having age- and sex-appropriate health promotion and preventive issues addressed; 38.0% of those without family physicians (e.g. those attending walk-in clinics instead) had these addressed.^[@CIT0029]^ The relationship between comprehensiveness and continuity was noted above. While relational continuity does not guarantee good performance on comprehensive care (a narrow, disease-oriented approach is still possible), the evidence that continuity is independently associated with improved outcomes^[@CIT0032]^ suggests that comprehensive care is a function of continuity and likely to improve with improved continuity. *Community orientation* recognises the primary care needs of a defined (practice) population. The effective delivery of services to individuals and communities is based on an understanding of community needs and the integration of a population perspective in the provision of primary care. Primary care providers contribute to and participate in community assessment, health surveillance, monitoring and evaluation. When reporting our findings to stakeholders, PCF managers were surprised by the low user score for community orientation (20.8%) and felt that the community may not be aware of community-based services (CBS) such as those rendered by a third party on behalf of the provincial health department. The score gap between users and managers and providers for community orientation was a consistent finding across all sub-districts and PCFs -- urban and rural). This is an important finding given that community-orientated primary care is a key feature of PHC and the SA Health Department's PHC re-engineering initiative. This may in part be due to absent or inadequate community involvement and messaging about existing CBS, that is, the low user score may indicate inadequate communication of services offered to users and user bodies. While there has been a significant shift to CBS, elements of a disease orientation nevertheless remain (e.g. TB and HIV home-based care and chronic disease adherence support groups) in contrast to the comprehensive PHC approach described in reviews of former local CHW programmes^[@CIT0040],[@CIT0041]^ and being practised in recently established ward-based out-reach teams (WBOTs) in wards in Tswane and Soweto -- projects of the Universities of Pretoria and Witwatersrand, respectively, and the Gauteng Health Department. The Brazilian comprehensive healthcare team-based family health programme (on which the WBOT is modelled) is associated with a higher total primary care score as well as higher scores for comprehensiveness, family and community orientation when compared with the traditional health system.^[@CIT0023]^ *Family-centred care* considers the impact of the family on the genesis and prevention of ill health, as well as the response to both medical and psycho-social interventions. It recognises and incorporates knowledge of the family context (e.g. resources, risk factors and social factors) into the planning and provision of primary care. As noted above, practitioner scores are optimistic (77%) relative to users and managers (52% and 45%, respectively). 'Thinking family' is the distinguishing feature of family-centred primary care. Family-centred thinking is an approach to routine primary care of individuals that considers the family or household as an integral part of information gathering, clinical reasoning and patient care.^[@CIT0042]^ There is considerable evidence to show that family-centred care improves health outcomes.^[@CIT0043]^ Narrowing the gap between user experience and provider-rated performance can therefore be expected to yield better outcomes. *Cultural competence* incorporates cultural references into the provision of primary care. Culturally competent services are acceptable to people in the community who may be distinguished by common values, language, heritage and beliefs about health and disease. It implies that their views are determined and incorporated into decision-involving policies, priorities and plans related to the delivery of healthcare services. Managers' performance scores on cultural competence are considerably lower (50.0%) than users and practitioners (73.5% and 69.0%, respectively). The user score may reflect staff demographics such as language and ethnicity increasingly approximating those of the users at many PCFs. The lower manager score could be due to items included in the manager PCAT that determine whether specific interventions to address diversity and transformation are part of continuing staff development, suggesting managers would like more attention given to these. The oversight and governance roles of managers would include having to deal with problematical interactions between staff and users; they may therefore be more aware of service shortcomings. Cultural competency is a complex and dynamic concept.^[@CIT0044],[@CIT0045]^ The PCAT definition which determines the items in this domain, for example, language competency and sensitivity to traditional heath beliefs and practices, may be limited relative to the complexities in SA society that have to be negotiated daily and driven by historical, sociopolitical, ethnic diversity and other factors. *PHC team* (composition) determines which members of the team (other than practitioners) are present, that is, whether users have on-site or nearby access to these for at least some days of the week. Managers (score 85.9%), given their leadership and management roles, are best placed to provide information on PHC team composition at their respective PCFs. Users (score 76.1%) who do not need the services of team members other than nurses and doctors may be less aware of them. The lower user score may also reflect less than satisfactory inclusion of other team members by doctors and CNPs -- supporting the contention above that the opportunities to use the services available in a comprehensive approach to care are frequently missed. This notwithstanding, the PHC team domain may be of limited use in the ZA PCAT; the results merely indicate that on the whole, core human resources are available at PCFs. Such information can easily be obtained from managers and PCF records. Assessing PHC team functioning and effectiveness would be more useful in such an audit; its absence is a shortcoming of the tool. (A validated 7-item instrument^[@CIT0046]^ that assesses team effectiveness^[@CIT0046]^ since being included in the provider (PE) and manager (FE) ZA PCAT is being tested.) At the report-back sessions involving district managers we observed a desire to understand the user experience in more detail in order to identify ways to improve PHC, suggesting that appropriate interventions are likely to be well received. This is especially important for chronic disease care given its contribution to the growing disease burden and poor outcomes despite the high costs of care in local PCFs.^[@CIT0026],[@CIT0027]^ It augurs well for other important initiatives, for example, to improve comprehensive care by improving health promotion and disease prevention, and reducing missed opportunities in primary care. Study limitations {#s20007} ----------------- Our facility-based sample excluded non-users of PCFs, including those in poor health too unwell to attend. Among the non-user group there is likely to be a subgroup of former users who for one or more reasons -- including service dissatisfaction -- no longer attend but have important information on the user experience. We chose a facility-based sample to identify actual PCF users in order to audit public sector primary care especially given the considerable effort and resources committed to improve public sector care -- such as the National Department of Health's PHC re-engineering initiative.^[@CIT0047]^ A community survey would have required a much larger sample and budget to achieve the calculated user sample size for the facilities studied. Sampling users on-site also made a rapid appraisal of current primary care performance, possible and affordable in a resource-constrained setting. Facility-based sampling was used in a Brazilian PCAT study for similar reasons.^[@CIT0023]^ The PCF-based sample is likely to result in a selection bias towards positive scores and better health status. Sampling each PCF over 1 working week (5 days) may not represent the user experience during other weeks of the year given changing operational and seasonal effects. However, user responses were determined by the past experience of primary care which would have reduced the impact of operational and seasonal effects and therefore on the results. Users' responses depend on their knowledge of the service and recall of past experience; an element of recall bias is therefore expected. This may have been tempered by respondents' length of association with their clinic (94.0% for 3 years or more; 59.6% for 5 years or more). On-site surveys are known to be less reliable; respondents are less likely to report negative views when their perceptions and experiences of a service are **e**licited on-site in face-to-face interviews.^[@CIT0048],[@CIT0049]^ Measures to reduce the impact included assuring respondents that the survey was anonymous; that identifiable and personal information was not required; that their responses would not affect usual care; and that the survey aimed to improve services based on users' collective experience of care. Healthcare users are encouraged to report poor health service delivery by using complaints boxes in health facilities and a dedicated telephone line. Although the PCAT is not a satisfaction survey, 10.0% of the respondents spontaneously offered details of negative experiences and/or ideas to improve the service -- recorded on a blank page by the interviewers -- suggesting they were comfortable responding to the questions. Haggerty et al., in a review of instruments assessing primary care performance, note that users are less likely to assess performance negatively if they are unable to pinpoint the cause of their negative experiences. Assessments therefore tend to be skewed towards the adequate to excellent range; negative assessments are more likely to be true negatives, that is, have a higher specificity than positive assessments. The authors suggest that it may be more accurate to report the 'percentage of less-than-positive scores' to avoid positives scores 'masking' negative scores.^[@CIT0050]^ As noted above, self-reported 'health status' is a subjective assessment. Differences between perceived health status and the audit findings may reflect a limitation of the PCAT. However, while health status can be expected to influence users' experience and therefore users' scores, this does not diminish their experience and the validity of their scores. Differences between user, provider and manager scores may be due to different interpretations of subdomain items (questions) -- a concern raised by managers and staff during report-back sessions. While item content and phrasing for managers and practitioners are necessarily determined by their respective roles, subdomain definitions (Appendix 1) are same for users, managers and practitioners. Any limitation notwithstanding, the significant difference in performance scores between users on the one hand and practitioners and managers on the other -- for example, for community-orientated primary care especially following increased emphasis on CBS for chronic disease care -- is concerning. Optimistic scores may lead to a false sense of doing well when in fact improvement is needed. It is noteworthy that during our report-back sessions, district managers generally accepted users' scores as reflecting their (users') experience of primary care and low scores as indicative of subdomains needing attention. Caution is needed when applying the study results to other PCFs and districts in the province. Although sub-district managers had a say in selecting PCFs in the Cape Town metro, it was within strata and where options were comparable. The districts sampled are the responsibility of one provincial health authority; PCFs operate in similar contexts with similar staff training, operational plans and constraints. Similarly, caution should be exercised before generalising to other provinces given the resource disparities between provinces. Conclusion {#s0008} ========== These are the first PCAT audit results using the cross-culturally adapted ZA PCAT (expanded form) and the first PCAT study in Africa. Primary care elements shown universally to be essential for cost-effective primary care were audited; elements that need strengthening -- especially with respect to users' experience - were identified. The results also suggest a need to reduce, or at least explain, gaps between the users' experience of primary care and practitioners' and managers' assessment of the care they provide. They highlight a need for better alignment with international best practice as well national and provincial health policies that increasingly promote the importance of the user experience of care in health sector reform. The findings were acknowledged by PCF managers -- and district-level managers present -- at the report-back sessions; managers did not find users' account of their experience surprising. While caution should be exercised before generalising the study findings, the results are in keeping with much of what is already known about primary care in South Africa -- as noted in the introduction to this paper. We believe the results provide an important baseline measure of urban and rural PHC performance and organisation in the Western Cape Province -- needed to determine the impact of imminent national and current local reforms. Notwithstanding the study limitations, the results have potential to guide the implementation of reforms. Recommendations {#s0009} =============== An audit is of little use if the information generated is not used to identify and guide interventions -- in this case directed at strengthening PHC. Rather than generating a list of recommendations covering the range of domains, we mention two recommendations: building PHC stakeholder partnerships and user registration with a PCF -- which we consider necessary for more specific interventions to succeed. Finding common ground among stakeholders by discussing and interpreting results together could form the basis of strong PHC stakeholder partnerships in the communities served to strengthen efforts to achieve health sector reform goals embodied in National Health Department initiatives -- such as PHC re-engineering and the 'ideal clinic' -- and the Western Cape Provincial 2030 Plan. In keeping PHC and the Alma-Ata Declaration, communities should be recognised as key stakeholders in partnerships that generate, prioritise, select, implement and monitor interventions -- for example, using the quality improvement cycle in a participatory action approach and the ZA PCAT to re-audit performance. User registration with a local PCF is not a feature of public sector primary care in South Africa where users are free to use PCFs wherever they wish. One-third of public sector users regularly seek primary care from more than one provider including the private sector.^[@CIT0031]^ Limited access to 8-h facilities and long waiting times are among the contributing factors. However, user acceptance of registration with one PCF -- especially among the indigent who have few choices -- is likely to be a challenge. South Africans were forced by law to live and use services in designated areas prior to 1994. Uptake will need strong user--provider partnerships, trust and improved services. The high user--facility association (noted under the *access* domain above) is encouraging and suggests a likelihood of some success if user registration is properly encouraged. The PCAT subdomains are synonymous with the well-known principles of family practice; the findings therefore suggest an important role in reform for primary care (family) physicians specifically trained to practise continuing, comprehensive, person-centred, family and community-orientated primary care in a range of contexts. These results should be considered when reviewing primary care practitioner and manager training. They should also be considered in PHC policy formulation and the research agenda. We thank the primary care users, managers and practitioners for their willingness to provide the information needed; the district directors for ready access to the services; Mr Deon September (data management, fieldworker supervision and general research assistance) and Delena Fredericks (study administration and management); they made a significant contribution to this study which was funded by the European Union. Competing interests {#s20010} =================== The authors declare that they have no financial or personal relationships which may have inappropriately influenced them in writing this article. Authors' contributions {#s20011} ====================== G.B. is the principal investigator and leader of the PCAT project and assumed primary responsibility for writing the paper. A-R.S. provided all the statistical advice, conducted the data analysis, generated the summary tables and graphs, and guided reporting of statistical elements in the method and results sections. C.lG. and S.B. were centrally involved in the data collection and management, advised on the content and structure of the paper, ensuring accurate reporting of the method and results sections. N.M. assisted with data collection and management and contributed to general project oversight, coordination and management. D.H. assisted with the overall support for the study. **Research Project no:** RP 033/2013 **How to cite this article:** Bresick G, Sayed A, le Grange C, Bhagwan S, Manga N, Hellenberg D. Western Cape Primary Care Assessment Tool (PCAT) study: Measuring primary care organisation and performance in the Western Cape Province, South Africa (2013). Afr J Prm Health Care Fam Med. 2016;8(1), a1057. <http://dx.doi.org/10.4102/phcfm.v8i1.1057> Shi, L. 'It is OK to compare patients' perceptions of experience with those of providers and managers provided the sample size is adequate'. E-mail to <Graham.Bresick@uct.ac.za> 8th January 2012.

Introduction ============ Dietary guidance is consistent in recommending greater consumption of fruit and vegetables to promote health. Indeed, the 2015 Dietary Guidelines Advisory Committee report noted that greater fruit and vegetable intake was the only characteristic of dietary patterns that was consistently identified in their report in every conclusion statement across health outcomes ([@b1]). Although the report does not recommend specific types of fruit, there has been a growing body of evidence that the phytochemical composition of berry fruit may differentiate them from other fruits and underlie some of their putative benefits. Recent advances in analytical methods have improved the characterization of polyphenols in berry fruit and subsequently the data in food-composition and metabolomics databases that are essential for observational studies. Furthermore, the development of a standard reference material (SRM)[^14^](#fn4){ref-type="fn"} and matched placebos for use in clinical trials has provided an important and innovative component for the design and conduct of new randomized clinical trials. This review, prepared from the proceedings of the Cranberry Health Research Conference held in conjunction with the Berry Health Benefits Symposium in Madison, Wisconsin, 12--15 October 2015, focuses particularly on advances in the field during the last 5 y with regard to the gut microbiota and cardiometabolic health. Cranberries and the Gut Microbiota ================================== ### Molecular mechanisms. Much of the attention regarding the impact of cranberries on the gut microbiota has been directed to studies of the effect of cranberry extracts or juice on uropathogens and urinary tract infections (UTIs) ([@b2], [@b3]). However, this focus has expanded to encompass a broader range of the cranberry's antimicrobial, antifungal, and antiviral actions against *Helicobacter pylori* ([@b4]--[@b6]), *Streptococcus mutans* ([@b7]), *Porphyromonas gingivalis* ([@b8]), *Staphylococcus aureus* ([@b9]), *Pseudomonas aeruginosa* ([@b10]), *Cryptococcus neoformans* ([@b6]), *Haemophilus influenzae* ([@b11]), *Candida albicans* ([@b12], [@b13]), and extraintestinal pathogenic *Escherichia coli* (ExPEC) ([@b14]). Cranberry constituents, particularly the proanthocyanidins, flavonols, and hydroxycinnamic acids, may act against these pathogens by preventing bacterial adhesion and coaggregation, decreasing biofilm formation and/or reducing inflammation rather than via bactericidal activity. This expanding body of research includes in vitro, ex vivo, and animal studies that have suggested potential clinical effects and have helped to elucidate mechanisms of action as well as human studies that have shown physiologic effects ([@b3]--[@b5], [@b14]). The antimicrobial properties of cranberry proanthocyanidins have been generally associated with their degree of polymerization (DP) and ratio of A- to B-type linkages. For example, by using an in vitro broth microdilution assay for growth inhibition of several yeast species, treatment of cultures with cranberry fractions of varying composition showed that cranberry proanthocyanidin fractions with a larger DP were found to be more effective than those with a smaller DP at inhibiting the growth of *Candida* spp. ([@b12]). In comparing primarily A-type proanthocyanidins from cranberries with primarily B-type proanthocyanidins from apples, Feliciano et al. ([@b15]) found that, although both increased agglutination and reduced epithelial cell invasion by ExPEC, the strongest effects were associated with a higher percentage of A-type linkages. This observation is consistent with other research that showed that A-type proanthocyanidins interact most strongly with bacterial virulence factors and more effectively decrease bacterial motility ([@b16], [@b17]). ### Microbiota biofilm. The prevention of biofilm formation, an early step in the development of infection, through interference in the coaggregation of bacteria is a well-documented antimicrobial mechanism of cranberry proanthocyanidins. The extensively hydroxylated structure of proanthocyanidins encourages intermolecular hydrogen bonding, allowing smaller molecules to aggregate and interact with receptors on cell surfaces. Thus, many studies of high-molecular-weight nondialyzable material from cranberry juice concentrate reveal potent antiadhesion activity with microbial species, including those found in the oral cavity, stomach, small intestine, and colon ([@b6], [@b11], [@b14], [@b18], [@b19]). However, although purified cranberry proanthocyanidins are more effective in some antimicrobial assays than are crude or mixed extracts, several studies suggest that other compounds in cranberry possess antibacterial properties that alone or in combination with proanthocyanidins may enhance overall protection against infection. For example, Pinzón-Arango et al. ([@b20]) exposed *E. coli* to cranberry juice cocktail (CJC) or cranberry proanthocyanidins over 48 h and found that the proanthocyanidins reduced whereas the CJC completely eliminated biofilm formation. Candidate CJC constituents may include nonphenolic compounds such as isoprenoids like ursolic acid and xyloglucans, hemicellulose oligosaccharides found in high-molecular-weight nondialyzable fractions ([@b21]). Hotchkiss et al. ([@b22]) found that arabinoxyloglucans isolated from pectinase-treated cranberry hulls prevented the adhesion of *E. coli* strains to bladder and colonic epithelial cells in vitro. Bacterial adhesion to cells and other surfaces involves basic physical forces such as electrostatic and steric interactions, van der Waals forces, and surface charge, as well as both specific and nonspecific interactions of surface proteins and carbohydrates such as glucans, adhesins, and sugar-specific lectins ([@b23]--[@b25]). Using atomic force microscopy, Liu et al. ([@b26]) found that exposure to cranberry juice decreased the adhesion forces of P-fimbriated *E. coli* (HB101pDC1) and altered the conformation and length of the P-fimbriae. Pinzón-Arango et al. ([@b24]) found that these fimbrial changes were reversible, even for cultures grown in the presence of cranberry juice. de Llano et al. ([@b27]) showed the efficacy of colonic metabolites of cranberry polyphenols, including hydroxylated benzoic and phenylacetic acids, in inhibiting the adhesion and biofilm formation of uropathogenic *E. coli* to bladder epithelial cells, a relation that underscores the critical need to elucidate the role of the gut microbiota in transforming cranberry polyphenols to bioactive and bioavailable compounds. ### Gut microbiota metabolism and function. The gut microbiota is now appreciated as a critical factor in nutrition and health, influencing the bioavailability and metabolism of food components and affecting body systems, including brain and immune functions. The integrity of the gut mucosal barrier is essential for maintaining a chemical and physical barrier against food, environmental antigens, and microbes ([@b28], [@b29]). Goblet cells migrate up the villi after differentiating from crypt stem cells and turn over with the epithelial layer every 3--5 d. Goblet cells secrete mucins, particularly mucin 2 (Muc-2), that contribute substantially to the maintenance of mucosal integrity ([@b30]). Mucin secretion is regulated by a complex network of cholinergic stimulation and T-helper 2 (Th2) cytokines IL-4 and IL-13 ([@b31]--[@b35]). Dysfunction of the gut barrier and dysbiosis have been associated with typical Western diets high in saturated fat and low in fiber and phytochemicals, patterns that may lead to increased permeability of bacterial LPS and a pathogen-associated molecular pattern that stimulates innate immune responses in macrophages, neutrophils, endothelial cells, and adipocytes. LPS plays a role in acute infection-related inflammatory responses and is found in blood and tissues with both postprandial and chronic inflammation ([@b36]--[@b39]). With the use of mice (CEABAC10) that express human carcinoembryonic antigen-related cell adhesion molecules (CEACAMs), Martinez-Medina et al. ([@b37]) found that a high-fat, high-sugar diet increased intestinal permeability and TNF-α secretion, which resulted in a greater ability of adherent-invasive *E. coli* (AIEC) to colonize gut mucosa and induce inflammation. This diet also induced gut barrier dysfunction reflected by reduced levels of Muc-2 mRNA, increased permeability of 4-kDa fluorescein isothiocyanate-dextran, and decreased numbers of goblet cells. It is worth noting that AIEC may contribute substantially to the etiology of Crohn disease, an inflammatory bowel disease in which CEACAM6 is overexpressed on the apical surface of ileum epithelium ([@b40]). Furthermore, variant AIEC type 1 pili adhere to CEACAM6, a key step in the colonization of the ileum and chronic inflammation present in Crohn disease. In addition, after feeding mice a high-fat, high-sugar diet, Anhê et al. ([@b41]) reported that the addition of a cranberry extract attenuated the consequent chronic inflammation associated with gut barrier dysfunction, including reductions in plasma LPS, cyclooxygenase-2, and TNF-α. Furthermore, the ratio of NF-κB to inhibitor κB was significantly lower in the jejunal tissue of the mice fed cranberry extract relative to the mice fed the high-fat, high-sugar diet. Also suggesting the capacity of cranberry polyphenols to reduce intestinal oxidative stress and inflammation, in vitro experiments with Caco-2/15 intestinal cells by Denis et al. ([@b42]) revealed positive but differential effects of low-, medium-, and high-molecular-mass polyphenols from cranberries on oxidative stress, proinflammatory cytokines, NF-κB activation, and nuclear factor E2-related factor 2 (Nrf2) downregulation, as well as PPAR-γ coactivator 1α. Interestingly, the effects of high-fat, high-sugar diets on gut barrier function in mice are similar to those observed in animal models of parenteral nutrition and elemental enteral nutrition (EEN) ([@b43], [@b44]). EEN induces dysfunction of gut-associated lymphoid tissue, including decreased lymphocytes in Peyer's patch and reduced tissue Th2 cytokines, and suppresses mucosal barrier function when compared with normal nutrition ([@b43], [@b45]--[@b48]). The addition of cranberry proanthocyanidins to EEN was found to increase ileal tissue IL-4 and IL-13 concentrations, goblet cell number and size, and the secretion of intestinal Muc-2, attenuating the impairment of the mucosal barrier integrity after EEN alone ([@b44]). Pierre et al. ([@b43]) reported that the addition of cranberry proanthocyanidins significantly supported other indexes of gut-associated lymphoid tissue function impaired by EEN in mice, indicated in part by decreased Peyer's patch lymphocytes and lower concentrations of tissue Th2 cytokines. Cranberry proanthocyanidins also helped to restore the EEN-induced decreases in polymeric Ig receptor, a transport protein involved in enterocyte transcytosis of secretory IgA (sIgA) from B cells in the lamina propria into the intestinal lumen. EEN decreases in luminal concentrations of sIgA were attenuated by cranberry proanthocyanidins; intestinal sIgA opsonizes bacterial antigens such as the virulence factors of pathogenic *E. coli*, rendering them less viable and more susceptible to killing by lymphocytes. The addition of cranberry proanthocyanidins also significantly prevented EEN-induced decreases in tissue IL-4 and phosphorylated signal transducers and activators of transcription 6 (STAT6). Clinical studies are necessary to determine whether the results from these mouse models can be translated to the capacity of cranberry phytochemicals to reduce diet-induced intestinal inflammation in humans. Interestingly, there is limited evidence suggesting an effect of cranberry on systemic immune function in humans, which may be partly mediated via gut metabolism of cranberry polyphenols. For example, a randomized, double-blind placebo-controlled study documented increased ex vivo proliferation of γδ-T cells, immune cells located within the epithelium of the gastrointestinal and reproductive tracts, after the consumption of a cranberry beverage for 10 wk ([@b49]). ### ExPEC in the gut. Although ExPEC generally do not cause acute enteric disease, their colonization in the gut increases the risk of subsequent extraintestinal infection, including UTIs, septicemia, surgical wound infections, and neonatal meningitis ([@b50], [@b51]). ExPEC attach to and invade epithelial cells through adhesins expressed on type I pili (protein FimH) and P fimbriae (fimbrial adhesin PapG) and persist inside the host cell in vacuoles where they may evade immune detection. ExPEC, uropathogenic *E. coli*, and the AIEC associated with Crohn disease have similar virulence factors and are within the same *E. coli* phylogroups (B2 and D) ([@b40]). These phylogroups differ from enteropathogenic *E. coli* and Shiga toxin--producing *E. coli*, such as O157:H7, because enteropathogenic *E. coli* and Shiga toxin--producing *E. coli* cause acute intestinal disease and produce attaching and effacing lesions of the intestinal epithelium. Gut colonization by ExPEC is a likely cause of a chronic inflammatory state because ExPEC may evade immune detection and colonize enterocytes. The continuous presence of *E. coli* LPS in the gut mucosa may cause chronic intestinal inflammation. Although ExPEC have a meaningful impact on public health via their consequences on morbidity and mortality, they have not received concordant attention because they have been highly susceptible to antibiotics. However, 20--45% of ExPEC have become resistant to first-line antibiotics such as cephalosporins, fluoroquinolones, and trimethoprim-sulfamethoxazole ([@b52], [@b53]). Thus, it is becoming critical to appreciate and further investigate the potential role for dietary bioactive components in reducing such infections. Recently, Feliciano et al. ([@b15]) showed that A-type proanthocyanidins have greater bioactivity than B-type proanthocyanidins for increasing ExPEC agglutination and decreasing their invasion (and subsequent colonization) of gut epithelial cells, an important observation for the elucidation of the effect of cranberry proanthocyanidins on UTIs. As suggested by Feliciano et al. ([@b15]) and other studies described above, decreasing intestinal colonization and associated inflammation may be achieved by usual serving sizes of cranberry juice without the requirement for absorption of its constituent proanthocyanidins into the circulation or their appearance in the urine. It is important to note that recent randomized clinical trials have confirmed and extended the body of evidence showing cranberry's bacterial antiadhesion activity in urine ex vivo ([@b3], [@b54]), its capacity to reduce the recurrence of UTIs ([@b55]), and its therapeutic efficacy in preventing UTIs in gynecologic surgery patients after catheter removal ([@b56]). Nonetheless, additional research that uses similarly relevant ex vivo and in vivo models can be used to substantiate the structure-function relation of A-type proanthocyanidins to intestinal and extraintestinal infections and to develop preventive and therapeutic strategies against increasingly antibiotic-resistant classes of pathogens ([@b57], [@b58]). Such an effort could be advanced by the availability of a cranberry SRM as discussed below. Cranberries and Cardiometabolic Health ====================================== A limited but growing number of clinical research studies ([@b59]--[@b72]) have focused on cardiometabolic health ([**Tables 1**](#tbl1){ref-type="table"} and [**2**](#tbl2){ref-type="table"}). The most commonly examined risk factors for cardiometabolic conditions in these studies have included serum lipid profiles, blood pressure (BP), endothelial function, glucoregulation, and a variety of biomarkers of inflammation and oxidative stress. Although the results of this research have generally been promising, a clear and consistent picture of this emerging area is confounded by sometimes marked differences in the cranberry products (cranberry juices, dried cranberries, and cranberry extracts) and doses used, as well as the characteristics of the study populations ([@b2], [@b73]). Although few animal model studies have examined this topic, Kim et al. ([@b74]--[@b76]) reported that 5% cranberry powder added to atherogenic diets with or without intraperitoneal LPS administration produced positive effects on serum lipids, proinflammatory cytokines, oxidative stress, and antioxidant capacity in rodents. ###### Summary of randomized placebo-controlled trials on the cardiometabolic effects of cranberry[^1^](#tblfn1){ref-type="table-fn"} Study, year (ref) Intervention Population Age,[^2^](#tblfn2){ref-type="table-fn"} y *n* Study design Duration, wk Dose/d Polyphenol content Outcomes --------------------------------- ------------------- -------------------------------- ------------------------------------------- ----- -------------- -------------- ------------ -------------------------------------- ----------------------------------------------------------------------------------------------------------------------------------------- Lee et al., 2008 ([@b59]) Cranberry extract Adults with T2D 65 ± 1 30 Parallel 12 500 mg × 3 Not reported ↓: TC, LDL-C; NC: FBG, HbA1c, SBP, DBP, TGs, HDL-C, oxLDL-C, insulin, HOMA-IR, CRP Shidfar et al., 2012 ([@b60]) CJ Adults with T2D 55 ± 9 58 Parallel 12 240 mL Not reported ↓: FBG, apoB; ↑: paraoxonase-1 activity; NC: apo A-1, Lp(a) Novotny et al., 2015 ([@b61]) LC CJ Healthy adults 50 ± 11 56 Parallel 8 480 mL 346 mg TPs, 21 mg ACNs, 235 mg PACs ↓: DBP, FBG, HOMA-IR, TGs, CRP; NC: SBP, insulin, HOMA-β, TC, LDL-C, HDL-C, apo A-1, apo A-2, apoB, ICAM, VCAM Basu et al., 2011 ([@b62]) LC CJ, 27% Adults with metabolic syndrome 52 ± 8 31 Parallel 8 480 mL 458 mg TPs, 25 mg ACNs ↓: oxLDL-C, malondialdehyde, 4-hydroxynonenal; NC: SBP, DPB, FBG, TC, LDL-C, HDL-C, VLDL-C, TGs, IL-6, CRP, plasma antioxidant capacity Dohadwala et al., 2011 ([@b63]) CJ, 54% Adults with CAD 63 ± 9 44 Crossover 4 480 mL 835 mg TPs, 94 mg ACNs ↓: HDL-C, carotid femoral PWV; NC: SBP, DPB, FBG, insulin, HOMA-IR, TC, LDL-C, TGs, carotid radial PWV, CRP, ICAM-1 Flammer et al., 2013 ([@b64]) LC CJ, 54% Adults with CVD risk factors 49 ± 16 69 Parallel 16 460 mL 800 mg TPs, 69 mg ACNs, 1224 mg PACs NC: SBP, DBP, TC, HDL-C, TGs, AIX, pulse pressure, heart rate, reactive hyperemia index, CRP, ICAM, VCAM, IL-6, TNF-α, oxLDL-C Ruel et al., 2013 ([@b65]) LC CJ, 27% Healthy overweight men 45 ± 10 35 Crossover 4 500 mL 400 mg TPs, 21 mg ACNs NC: SBP, DBP, mean arterial pressure, heart rate, AIX, global endothelial function, NOx, uric acid, oxLDL-C, ICAM-1, VCAM-1, E-selectin ACN, anthocyanin; AIX, augmentation index; CAD, coronary artery disease; CJ, cranberry juice; CRP, C-reactive protein; CVD, cardiovascular disease; DBP, diastolic blood pressure; FBG, fasting blood glucose; HbA1c, glycated hemoglobin; HDL-C, HDL cholesterol; HOMA-β, homeostatic model assessment of β cell function; ICAM, intercellular adhesion molecule; Lp(a), lipoprotein a; LC, low-calorie; LDL-C, LDL cholesterol; NC, no significant change; NOx, nitrate/nitrite; oxLDL-C, oxidized LDL cholesterol; PAC, proanthocyanidin; PWV, pulse-wave velocity; ref, reference; SBP, systolic blood pressure; TC, total cholesterol; TP, total polyphenol; T2D, type 2 diabetes; VCAM, vascular cell adhesion molecule; VLDL-C, VLDL cholesterol; ↓, significantly different decrease from placebo group; ↑, significantly different increase from placebo group. Values are means ± SDs. ###### Summary of open-label trials on the effects of cranberry on cardiometabolic markers[^1^](#tblfn3){ref-type="table-fn"} Study, year (ref) Intervention Population Age, y *n* Study design Duration Dose/d Polyphenol content Outcomes ------------------------------ --------------------- -------------------------------- --------------------------------------------------------------------- ----- ------------------------------------------- -------------------------- --------------------- ------------------------------------------ -------------------------------------------------------------------------------------------------------------- Ruel et al., 2006 ([@b66]) LC CJ, 27% Healthy sedentary men 51 ± 10[^2^](#tblfn4){ref-type="table-fn"} 30 Crossover 4 wk/dose 125, 250, 500 mL 100 mg TPs, 5 mg ACNs, 74 mg PACs/125 mL ↓: TC:HDL-C, antioxidant capacity, NOx; ↑: HDL-C; NC: TC, LDL-C, TGs, apo A-I, apoB Ruel et al., 2008 ([@b67]) LC CJ, 27% Healthy sedentary men 51 ± 10 30 Crossover 4 wk/dose 125, 250, 500 mL 100 mg TPs, 5 mg ACNs, 74 mg PACs/125 mL ↓: SBP, HDL-C, oxLDL-C, ICAM-1, VCAM-1; NC: DBP, heart rate, TC, LDL-C, TGs, apoB, E-selectin Ruel et al., 2009 ([@b68]) LC CJ, 27% Healthy sedentary men 51 ± 10 30 Crossover 4 wk/dose 125, 250, 500 mL 100 mg TPs, 5 mg ACNs, 74 mg PACs/125 mL ↓: SBP, MMP-9, NOx; NC: DBP, mean arterial BP Ruel et al., 2005 ([@b69]) LC CJ, 27% Healthy men 38 ± 8 21 Single-arm intervention; no control group 14 d 7 mL/kg body weight Not reported ↓: oxLDL-C; ↑: antioxidant capacity; NC:SBP, DBP, TC, LDL-C, HDL-C, TC:HDL-C, TGs, apoB, LDL-C particle size Wilson et al., 2008 ([@b70]) CJ, 27%; LC CJ, 27% Adults with T2D and obesity 65 ± 2 12 Crossover Single dose Not reported Not reported for dosage given CJ vs. LC CJ---↑: glucose at 30 and 60 min, insulin at 60 min; NC: glucose, insulin at 120 min Wilson et al., 2010 ([@b71]) SDC, LC SDC, RC, WB Adults with T2D 62 ± 2 13 Crossover 30, 60, and 120 min/dose 40 g each Phenolic and PAC profiles reported WB, SDC, LC SDC vs. RC---↑: glucose at 30, 60, 120 min; insulin at 30, 60 min; glucose AUC; insulin AUC Simão et al., 2013 ([@b72]) LC CJ Adults with metabolic syndrome 48.5 (control), 51.0 (cranberry)[^3^](#tblfn5){ref-type="table-fn"} 56 Parallel; no-intervention control 60 d 700 mL 364 mg TPs, 231 mg PACs ↑: Adiponectin; ↓: lipoperoxidation, protein oxidation; NC: CRP, IL-1, IL-6, TNF-α, folic acid, homocysteine ACN, anthocyanin; BP, blood pressure; CJ, cranberry juice; CRP, C-reactive protein; DBP, diastolic blood pressure; HDL-C, HDL cholesterol; ICAM, intercellular adhesion molecule; LC, low-calorie; LDL-C, LDL cholesterol; MMP, metalloproteinase; NC, no significant change; NOx, nitrites/nitrates; oxLDL-C, oxidized LDL cholesterol; PAC, proanthocyanidin; RC, raw cranberries; ref, reference; SBP, systolic blood pressure; SDC, sweetened dried cranberries; TC, total cholesterol; TP, total polyphenol; T2D, type 2 diabetes; VCAM, vascular cell adhesion molecule; WB, white bread; ↓, significantly different decrease from baseline, between groups, and/or across increasing doses; ↑, significantly different increase from baseline, between groups, and/or across increasing doses. Mean ± SD (all such values). Median ages in control and treatment groups. ### Lipid profile. Early reports by Ruel et al. ([@b66]--[@b68]) found that interventions with low-calorie cranberry juice were associated with increases in plasma HDL cholesterol as well as with reductions in plasma oxidized LDL cholesterol, adhesion molecules, and matrix metalloproteinase 9. Lee et al. ([@b59]) showed a reduction in both LDL cholesterol and total cholesterol in a trial in 30 patients with type 2 diabetes (T2D) who consumed cranberry extract supplements daily for 12 wk. In a double-blind, placebo-controlled trial, Shidfar et al. ([@b60]) reported that 58 men with T2D who consumed 1 cup cranberry juice/d for 12 wk experienced decreases in apoB and increases in apo A-1 and paraoxonase-1, although data on LDL, HDL, and total cholesterol were not reported. In an 8-wk randomized clinical trial of low-calorie cranberry juice consumption by 56 healthy adults, Novotny et al. ([@b61]) found that TGs were significantly decreased in the cranberry group whereas other elements of the lipid profile were unchanged. ### BP. Previous studies of the effect of cranberry juice on BP suggested a potential benefit on BP ([@b67], [@b69]). More recent studies also examined changes in BP after cranberry intake ([@b59], [@b61]--[@b65]). The durations of these studies ranged from 1 to 4 mo and tested intakes of total polyphenols ranging from 346 to 835 mg/d; and study populations were heterogeneous, including subjects with obesity, metabolic syndrome, T2D, coronary artery disease (CAD), and risk factors for cardiovascular disease (CVD), as well as healthy volunteers. With daily doses of CJC increasing every 4 wk from 0--125 to 250--500 mL, systolic BP decreased by 3 mm Hg with the 500-mL intervention compared with baseline in obese men ([@b67]). Of the more recent studies, only the study performed in healthy individuals and with the lowest dose of polyphenols showed an improvement in BP, with a reduction of 4.7 mm Hg in diastolic BP achieved after 8 wk of daily supplementation ([@b61]). ### Endothelial function. Endothelial dysfunction, often characterized by a decrease in nitric oxide production and impaired flow-mediated vasodilation (FMD), is a critical factor underlying the development and progression of atherosclerosis ([@b77]). In a randomized controlled trial with a crossover design, Dohadwala et al. ([@b63]) found that daily supplementation with cranberry juice for 4 wk did not improve FMD or peripheral artery tonometry in 44 patients with CAD, although an uncontrolled pilot study in a subset of the same population showed a modest improvement in FMD 4 h after an acute dose of cranberry juice. In a 4-wk trial with a cranberry juice drink, Flammer et al. ([@b64]) found no significant changes in peripheral artery tonometry in individuals with endothelial dysfunction and other CVD risk factors. Further research on the effect of cranberries on measures of vascular reactivity is required in healthy individuals examining both the dose-response and time course of the intervention. Recently, in a clinical study of 10 healthy adults, Feliciano et al. ([@b78]) identified and quantified by ultra-performance liquid chromatography/quadrupole-time-of-flight mass spectrometry analysis a total of 60 cranberry-derived phenolic metabolites in plasma and urine after the acute ingestion of cranberry juice containing 787 mg polyphenols. These metabolites included sulfates of pyrogallol, valerolactone, benzoic acids, phenylacetic acids, and glucuronides of flavonols, as well as sulfates and glucuronides of cinnamic acids. Their concentrations ranged from in the low nanomolars to the high micromolars depending on the compound. Among these 60 phenolic metabolites, 12 were found to be independent predictors of time- (0--6 h) dependent increases in FMD after an acute dose (range: 409--1909 mg total polyphenols) ([@b79]). These results indicate that cranberry polyphenols can acutely increase endothelial function in healthy individuals. Arterial stiffness, commonly assessed by pulse-wave velocity or the augmentation index (a measure of the enhancement of central aortic pressure by a reflected pulse wave), is an established risk factor for CVD ([@b80]--[@b82]). In their randomized clinical trial of patients with CAD, Dohadwala et al. ([@b63]) found that a 4-wk intervention with cranberry juice significantly reduced the carotid-femoral pulse-wave velocity. However, no changes in the augmentation index were observed by Ruel et al. ([@b65]) after 4 wk of supplementation with cranberry juice in 35 volunteers presenting with obesity and other cardiovascular risk factors. ### Glucoregulation. Berry fruit polyphenols have been shown by in vitro experiments and animal models to inhibit carbohydrate digestion and glucose absorption in the intestine, stimulate insulin secretion from β cells in the pancreas, regulate glucose release from the liver, and activate insulin receptors and glucose uptake in insulin-sensitive tissues ([@b83]--[@b85]). Emerging clinical evidence suggests that dietary modification to increase polyphenol intakes from whole-food sources can lead to improved glycemic control in T2D ([@b86], [@b87]). While exploring the antidiabetic effects of a cranberry extract in high-fat, high-sugar--fed mice, Anhê et al. ([@b41]) found a decrease in glucose-induced hyperinsulinemia and improved insulin sensitivity along with a reduction in weight gain and visceral obesity. As noted above, along with these improvements, the cranberry extract also altered the gut microbiome by increasing mucin-degrading bacteria. In light of the evolving link between the gut microbiota and diabetes, these findings provide an important connection between the studies documenting the effects of cranberry on gut barrier function and the potential to reverse the dysbiosis and metabolic inflammation underlying diabetes ([@b88]). Human studies testing low-calorie cranberry juice and unsweetened dried cranberries were shown to produce favorable acute postprandial glycemic responses in adults with T2D ([@b70], [@b71]). However, the limited number of longer-term studies in patients with T2D generated discordant outcomes. Shidfar et al. ([@b60]) reported that daily cranberry juice consumed for 12 wk by 58 male patients with T2D induced significant decreases in fasting blood glucose when compared with the placebo group. In contrast, in a trial of 30 patients with T2D, Lee et al. ([@b59]) found no impact of daily supplementation with cranberry extracts for 12 wk on fasting blood glucose or glycated hemoglobin. In a diet therapy intervention in 27 adults with T2D, Chambers and Camire ([@b89]) found no significant effect of treatment with cranberry extract on measures of glycemia. However, in 12 healthy volunteers in a randomized crossover trial, Törrönen et al. ([@b90]) found that a berry puree containing cranberries was able to delay the postprandial plasma response to sucrose. Clinical trials in patients with metabolic syndrome have suggested some benefit associated with cranberry intervention but not specifically on outcomes of glycemic control ([@b62], [@b72], [@b91]). ### Biomarkers of inflammation. In their analysis of observational data collected from NHANES, Duffey and Sutherland ([@b92], [@b93]) found inverse associations of popular polyphenol-containing beverages, such as cranberry juice, with obesity and inflammation. In their 8-wk randomized clinical trial, Novotny et al. ([@b61]) showed a reduction in C-reactive protein (CRP) after daily consumption of cranberry juice. In a clinical trial of 56 subjects with metabolic syndrome, Simão et al. ([@b72]) reported that daily intake of low-calorie cranberry juice for 8 wk had no significant effect on proinflammatory cytokines IL-1, IL-6, and TNF-α but did reduce biomarkers of lipid peroxidation and advanced oxidation protein products. Similarly, in an 8-wk study of 36 patients with metabolic syndrome, Basu et al. ([@b62]) observed increases in plasma biomarkers of antioxidant capacity and decreases in lipid peroxidation after the daily consumption of low-calorie cranberry juice. These results are consistent with in vitro experiments showing that cranberry polyphenols decreased the generation of reactive oxygen species and lipid peroxides and increased glutathione peroxidase activity and phospho-c-Jun N-terminal kinase ([@b94]). Cranberries and Health: Knowledge Gaps ====================================== The recent growing body of research on cranberries and health is a part of the emerging evidence from in vitro, animal model, and human studies of plant polyphenols as protective dietary agents that act both directly and indirectly via their metabolites and/or interactions with the gut microbiota. Improvements in these research approaches, particularly in analytical methods as diverse as MS and gene-sequencing methods for microbial communities, are making important contributions to our understanding of polyphenol mechanisms and functions. However, an important need in cranberry health research is the consistent use of a fully characterized SRM to help promote the generation of more readily comparable and replicable research protocols. SRMs have been available for some other polyphenol-rich foods, including highbush blueberries ([@b95]) and California table grapes ([@b96]), for in vitro, animal, and human studies. Although SRMs for cranberries have been developed by the National Institute of Standards and Technology and the Office of Dietary Supplements of the NIH, these materials are intended for the validation of analytical methods and quality assurance for in-house control materials. Furthermore, these SRMs are not accompanied by matching placebos for use in research studies. Wide availability of cranberry SRMs in sufficient quantities to conduct in vivo human health research remained lacking until recently. Because concerns for study accuracy and quality were raised because of the diversity of cranberry products available commercially and in research protocols ([@b97]), The Cranberry Institute undertook the development of a cranberry reference material in 2014 to ensure the authenticity and consistency of cranberry products used in research on human health. The first question to be answered was whether the SRM would be developed from whole fruit or one of the many processed forms in which cranberry is consumed. The impact of processing, particularly juicing, on the phytochemical content and profile of fresh cranberries has been characterized ([@b98], [@b99]). Grace et al. ([@b100]) compared fresh and freeze-dried cranberries to cranberry-containing commercial products including juices (from concentrate and not from concentrate), sweetened dried cranberries, and cranberry sauces (homemade and commercially canned). Cranberry skins and flesh were cross-compared for anthocyanin and proanthocyanidin content. Proanthocyanidins were typically higher in skins than in flesh with the exception of the proanthocyanidin A-2 dimer. Anthocyanin and proanthocyanidin concentrations were lower in juice reconstituted from concentrate. In general, the retention of proanthocyanidins in processed cranberries was found to be robust, whereas anthocyanins were sensitive to degradation. Grace et al. ([@b101]) explored ways to better concentrate and stabilize cranberry bioactive compounds via complexing concentrated juices with proteins isolated from soy, hemp, peanuts, and peas for formulating both beverage and solid-food products. By using an in vitro model to simulate digestion, Ribnicky et al. ([@b102]) were able to show that protein complexes with blueberry polyphenols remained more intact and bioaccessible than the free bioactive compounds. To retain and encourage the study of the complete phytochemical profile of cranberry, the SRM is a freeze-dried whole-cranberry powder (FWCP). It is produced from a blend of cranberry varieties grown in Wisconsin and approximating the proportion available in the marketplace (i.e., 56% Stevens plus 11% each of Ben Lear, Grygleski, Pilgrim, and HyRed varieties for the first batch produced in 2015). The berries are individually frozen after harvest, freeze-dried, and ground into powder form. Silicon dioxide (3% total volume of powder) is added as an anticaking agent. The production process is fully documented from harvest to storage. Each 50 g (0.5 cup) of whole cranberries produces ∼4.5 g FWCP. Complete specifications for each nutrient and phytochemical ingredient were prepared by using a series of assays, including matrix-assisted laser desorption/ionization time-of-flight MS for authentication of proanthocyanidins ([@b103], [@b104]), 4-(dimethylamino)cinnamaldehyde assay for quantification of soluble proanthocyanidins ([@b57], [@b103], [@b105]), and butanol-hydrochloric acid for quantification of insoluble proanthocyanidins as well as characterization of efficacy via an established in vitro antiadhesion assay and microbiological testing. Accurate quantification of proanthocyanidins for health research is essential but also problematic because proanthocyanidins are complex polydispersed hetero-oligomers ([@b57]). Previously, the procyanidin A2 (ProA2) dimer was recommended as the standard of choice for proanthocyanidin analysis in the 4-(dimethylamino)cinnamaldehyde assay because cranberry proanthocyanidins contain ≥1 "A-type" interflavan bonds ([@b106]). However, current evidence shows that the use of the ProA2 dimer as a standard for quantification of complex proanthocyanidin oligomers results in a serious underestimation of proanthocyanidins ([@b107]). To address this problem, a cranberry proanthocyanidin standard (c-PAC), reflective of the structural heterogeneity of proanthocyanidins found in fresh cranberry (i.e., DP, hydroxylation pattern, and ratio of A- to B-type interflavan bonds), was developed. The use of the c-PAC to quantify proanthocyanidin content in FWCP resulted in values that were 3.6 times those determined by ProA2. Thus, adoption of this c-PAC standard reflects an improvement over the use of ProA2 for the accurate quantification of cranberry proanthocyanidins ([@b105]). Because these findings were only recently published, the soluble proanthocyanidin content of the FWCP is reported as both c-PAC and ProA2 equivalents, allowing researchers time to adopt the new methodology. The c-PAC was also used to quantify the FWCP insoluble proanthocyanidins by the butanol-hydrochloric acid method. The polyphenol content of the FWCP includes the following: 28.35 mg total polyphenols (gallic acid equivalents)/g, 31.20 mg total soluble proanthocyanidins (c-PACs)/g, 8.77 mg soluble proanthocyanidins (ProA2)/g, 10.38 mg insoluble proanthocyanidins (c-PACs)/g, 5.98 mg anthocyanins (cyanidin-3-galactoside equivalents)/g, 9.01 mg flavonols (quercetin-3-rhamnoside equivalents)/g, and 1.81-mg hydroxycinnamic acids (caffeic acid equivalents)/g ([@b108]). The FWCP processing and packaging facilities are compliant with FDA regulations. A suitable placebo was created from a blend of maltodextrin, citric acid, artificial flavoring, fructose, and food-grade coloring agents ([@b109]). Calcium silicate is added to the FWCP and placebo as a flow agent. The use of the FWCP should help overcome some of the critical limitations associated with past studies that used uncharacterized or only partly characterized cranberry foods or extracts. Recipes for the administration of FWCP and placebo in human studies have been developed and are made freely available to researchers. Like other studies of whole foods, it is recommended that protocols that use the FWCP not apply this material directly to target tissues, with some possible exceptions such as oral and gastrointestinal cells. In vitro and ex vivo research approaches should consider the use of metabolite(s) on the basis of their likely bioavailability to these tissues, an approach not often followed in early studies of polyphenol-rich foods and extracts. The design of clinical trials that use the FWCP should also be informed by human bioavailability data generated from studies of other cranberry foods and extracts ([**Table 3**](#tbl3){ref-type="table"}) ([@b110]--[@b114]), although some consideration should be directed to results from animal models ([@b115]). However, because the product matrices and pharmacokinetic characteristics of these other products will undoubtedly differ, new studies on the absorption, metabolism, and elimination of the bioactive compounds in the FWCP must be undertaken. Although the availability of the FWCP as an SRM for clinical research may help ensure the consistency and full characterization of the cranberry intervention, the need to perform reasonable dose-response and time-course studies for each health-related outcome remains an important priority, as does the need to develop biomarkers of compliance to the intervention. ###### Summary of human trials investigating the bioavailability and pharmacokinetic variables of cranberry[^1^](#tblfn6){ref-type="table-fn"} Study, year (ref) Population *n* Cranberry product and dose Compounds studied Variables studied Timing of measurements --------------------------------- ------------------------------- ----- ---------------------------------------------------------- -------------------------------------------------- ------------------------------------------------------------------- -------------------------------------------------------------------- Feliciano et al., 2016 ([@b78]) Healthy young men 10 450 mL CJ Phenolic metabolites Plasma AUC, Cmax, Tmax, % urinary recovery Plasma: 1, 2, 4, 6, 8, and 24 h; urine: 0--8 h, 8--24 h Zhang and Zuo, 2004 ([@b110]) Healthy adults 1 1800 mL, 27% CJ Flavonoids, phenolic acids, benzoic acids Plasma and urine concentrations 0, 45, and 270 min Milbury et al., 2010 ([@b111]) Adults aged 62 ± 8 y with CAD 15 480 mL 54% CJ (835 mg TPs, 94.47 mg ACNs) ACNs Plasma AUC, Cmax, Tmax, T1/2 % urinary recovery 0--4 h Iswaldi et al., 2013 ([@b112]) Adults aged 25--40 y 4 0.6 mL/kg cranberry syrup Polyphenols, phase I and II phenolic metabolites Urine concentration Urine: 0, 2, 4, and 6 h McKay et al., 2015 ([@b113]) Adults aged ≥50 y 10 54% CJ Flavonoids, phenolic acids, PACs Plasma AUC, Cmax, Tmax, antioxidant capacity, urine concentration Plasma: 0.25, 0.5, 1--6, and 10 h; urine: 12, 14, 16, 18, and 20 h Walsh et al., 2016 ([@b114]) Healthy women aged 20--30 y 5 237 mL cranberry beverage (140 mg PACs); weekly for 7 wk PACs Urine concentration 24 h ACN, anthocyanin, CAD, coronary artery disease; CJ, cranberry juice; Cmax, maximal plasma concentration; PAC, proanthocyanidin; ref, reference; T1/2, biological half-life; Tmax, time to maximal plasma concentration. Summary ======= Cranberry juice, dried cranberries, and various cranberry extracts have been shown via in vitro, animal model, and human studies to possess an array of biochemical and physiologic activities mediated by their phytochemical constituents. Although the greatest research focus has been reasonably placed on their rich content of polyphenols, emerging evidence of their actions on the gut microbiota and cardiometabolic functions suggests that attention is also warranted on their synergy with cranberry phenolic acids, isoprenoids, and oligosaccharides. Acting in high concentrations within the gastrointestinal lumen, these cranberry compounds may act to quench reactive oxygen species, modulate inflammatory pathways, adhere to carbohydrates and proteins on bacterial surfaces, exert prebiotic effects, and alter the dynamic cross-talk between intestinal epithelial cells and the gut microbiota. These actions may underlie not only the antimicrobial effects of cranberries but their role in the complex pathogenesis of UTIs and inflammatory bowel diseases. The importance of these relations beyond the gastrointestinal tract has grown substantially with the recognition of the broad role that the gut microbiota plays in regulating energy homeostasis, glucose and lipid metabolism, and systemic inflammation, all factors associated with the maintenance of cardiometabolic health. Further substantiating the actions and mechanisms of cranberry constituents can best be accomplished by taking advantage of recent advances in cranberry research. For example, efforts to identify biomarkers of compliance to clinical protocols, as well as their relation to physiologic and health outcomes, may evolve from improved understanding of cranberry constituents (e.g., the specific nature of proanthocyanidin interflavan bonds and DP, as well as a more robust phytochemical profile) and the numerous bioactive catabolites arising from the biotransformation of cranberry constituents by the gut microbiota and phase I, II, and III metabolism pathways. Furthermore, a greater degree of accuracy, consistency, and quality of new studies has become possible with the availability of a fully characterized FWCP and matched placebo as SRMs. JBB and CDT outlined and coedited this article in addition to their contribution to the writing; JBB, AB, CGK, MAL, CCN, JAN, JDR, AR-M, and CDT contributed to the writing and review of the manuscript. All authors read and approved the final manuscript. We appreciate the excellent moderation of the Cranberry Health Research Conference presentations and panel discussions by Amy Howell (Rutgers University), Christina Khoo (Ocean Spray Cranberries, Inc), and the active participation by the panelists: C-Y Oliver Chen (Tufts University), André Marette (Université Laval), Yeonwha Park (University of Massachusetts at Amherst), Navindra Seeram (University of Rhode Island), David Sela (University of Massachusetts at Amherst), and Christine Wu (University of Illinois at Chicago). Abbreviations used: AIEC, adherent-invasive *Escherichia coli*; BP, blood pressure; CAD, coronary artery disease; CEACAM, carcinoembryonic antigen-related cell adhesion molecule; CJC, cranberry juice cocktail; c-PAC, cranberry proanthocyanidin standard; CRP, C-reactive protein; CVD, cardiovascular disease; DP, degree of polymerization; EEN, elemental enteral nutrition; ExPEC, extraintestinal pathogenic *Escherichia coli*; FimH, protein FimH; FMD, flow-mediated vasodilation; FWCP, freeze-dried whole-cranberry powder; Muc-2, mucin 2; Nrf2, nuclear factor E2-related factor 2; PapG, fimbrial adhesin PapG; ProA2, procyanidin A2; sIgA, secretory IgA; SRM, standard reference material; STAT6, signal transducers and activators of transcription 6; Th2, T-helper 2; T2D, type 2 diabetes; UTI, urinary tract infection. [^1]: Published in a supplement to *Advances in Nutrition*. Presented at the Cranberry Health Research Conference, held in Madison, Wisconsin, 12 October 2015 and sponsored by the Cranberry Institute (CI), the US Cranberry Marketing Committee (CMC), and the American Cranberry Growers Association. The Supplement Coordinator for this supplement was Cheryl D Toner. Supplement Coordinator disclosure: Cheryl D Toner is the contracted Health Research Coordinator for the CI and a consultant to the CI and the CMC. Publication costs for this supplement were defrayed in part by the payment of page charges. This publication must therefore be hereby marked "advertisement" in accordance with 18 USC section 1734 solely to indicate this fact. The opinions expressed in this publication are those of the author(s) and are not attributable to the sponsors or the publisher, Editor, or Editorial Board of *Advances in Nutrition*. [^2]: Author disclosures: The Cranberry Institute (CI) provided travel expense reimbursement to all authors and provided an honorarium to each author except for JA Novotny and CD Toner. CD Toner is a consultant to the CI and the CMC and formerly to the Juice Products Association. JB Blumberg is a member of the Scientific Advisory Board of the CI and the CMC and has received research support from the CI and Ocean Spray Cranberries, Inc. CG Krueger, JD Reed, and A Rodriguez-Mateos have received research support from the CI. JA Novotny has received research support from Ocean Spray Cranberries, Inc. [^3]: This is a free access article, distributed under terms (<http://www.nutrition.org/publications/guidelines-and-policies/license/>) that permit unrestricted noncommercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

INTRODUCTION {#sec0005} ============ Cancers of the kidney and renal pelvis account for approximately 3.5% of all new cancer cases in the United States and are responsible for 3.9 deaths per 100,000 individuals per year. Most arise in the kidney, and are commonly described as renal cell carcinoma (RCC). The majority (65%) of RCCs present as localized (stage I) disease, and approximately 16% are diagnosed as *de novo* metastatic disease (stage IV). The prognosis of RCC is highly dependent on the stage at diagnosis with 92% of individuals with localized disease alive at 5 years, while only 11.7% of individuals with metastatic disease survive 5 years \[[@ref001]\]. In addition, the histological subtype of RCC impacts prognosis and treatment with clear cell RCC (ccRCC) being the most common subtype, accounting for over 70% of all RCC \[[@ref003]\]. The most common genetic event in ccRCC carcinogenesis is loss of the von Hippel Lindau (VHL) gene, a key tumor suppressor on chromosome 3p25 with one of its main functions being to downregulate hypoxia inducible factor 1 alpha (HIF1*α*) and 2 alpha (HIF2*α*) via VHL ubiquitinating HIF leading to proteasomal degradation \[[@ref004]\]. HIF1*α* and HIF2*α* are crucial pro-angiogenic transcription factors with multiple downstream angiogenic and metabolic targets including vascular endothelial growth factor (VEGF), platelet derived growth factor (PDGF), fibroblast growth factor (FGF), and GLUT1, factors which aid in renal cell carcinogenesis initiation and progression \[[@ref004]\]. Thus the standard of care treatment options for first-line therapy in metastatic ccRCC target dysregulated angiogenesis and metabolism, though currently we lack synthetic lethal systemic treatment strategies for RCC \[[@ref003]\]. As is the case with many cancer subtypes, the treatment of localized RCC is primarily surgical; and, there are currently no standard perioperative neoadjuvant or adjuvant systemic treatment approaches for localized RCC \[[@ref003]\]. However, unlike most other cancer subtypes, surgery in the form of radical or partial nephrectomy is also part of a multidisciplinary treatment approach for metastatic RCC in select patients with good performance status. Nephrectomy has demonstrated a survival benefit in prospective trials and retrospective reviews of patients with metastatic RCC in both the earlier immune therapy era, as well as the current tyrosine kinase inhibitor (TKI) era \[[@ref007]\]. However, studies have consistently shown that nephrectomy provides little to no benefit for individuals with poor prognosis and/or poor performance status \[[@ref007]\]. In the metastatic setting, systemic therapy in the form of an anti-angiogenic tyrosine kinase inhibitor is typically initiated following nephrectomy with the choice and timing of systemic therapy in relation to surgery being somewhat case dependent, and remains an area of active research \[[@ref011]\]. Retrospective studies have shown benefit to nephrectomy and targeted therapy versus targeted therapy alone \[[@ref010]\]. While the safety and efficacy of presurgical targeted anti-angiogenic therapy has been evaluated in retrospective and Phase I/II studies, we currently lack any Phase III randomized trial data to support or refute its use as a standard of care. Large randomized trials comparing cytoreductive nephrectomy with tyrosine kinase inhibitors (both presurgical and postsurgical) versus tyrosine kinase inhibitors alone are ongoing \[NCT00930033 (CARMENA)\] or have been closed prematurely \[NCT01099423 (SURTIME)\], and current guidelines recommend cytoreductive nephrectomy for patients with good performance status and low burden of metastatic disease \[[@ref003]\]. In this systematic review, we will evaluate and summarize published literature and data on the use of presurgical and postsurgical systemic therapy in the setting of metastatic RCC. METHODS {#sec0010} ======= We performed a search of Medline and PubMed from inception through 01/2017 under the direction of a medical librarian for prospective clinical trials as well as retrospective studies and retrospective reviews related to perioperative systemic therapy in metastatic RCC. The search was restricted to English-language articles only. Key search terms included kidney cancer, renal cell carcinoma, metastatic, presurgical, adjuvant, neoadjuvant, perioperative, systemic therapy, immune therapy, tyrosine kinase inhibitors, nephrectomy, metastasectomy, safety, efficacy, randomized control trial. Input from clinical experts in RCC was also obtained to assess for any publications or studies missed by the search. RESULTS {#sec0015} ======= Initial search yielded over approximately 740 articles for evaluation. We then filtered for randomized control trials, retrospective studies, systematic reviews/meta-analyses specifically related to presurgical and/or perioperative therapy in metastatic renal cell carcinoma and also included input from experts in the field \[[@ref013]\]. PRESURGICAL THERAPY IN METASTATIC RCC {#sec0020} ===================================== Presurgical therapy in the metastatic RCC setting can serve to identify those patients who are not actively progressing or decompensating, to better select those individuals for nephrectomy. In addition, a presurgical systemic therapy approach as outlined in [Fig. 1](#kca-1-kca170009-g001){ref-type="fig"} can give insight into the biology of a patient's tumor that can aid in future systemic treatment planning and allow for earlier systemic therapy than waiting until post-surgery. Molecular profiling of tumor tissue obtained prior to initiation of presurgical systemic therapy and then again at time of surgery can help identify predictive biomarkers of sensitivity and resistance, as well as detect molecular changes associated with clinical outcomes \[[@ref023]\]. However, a pre-surgical treatment strategy necessitates interruption of systemic therapy in the perioperative period, which may increase likelihood of tumor progression. In addition, prior studies have shown that pre-surgical antiangiogenic therapy may lead to an increase in post-surgical complications due delayed wound healing, though other studies show no evidence of excessive post-surgical morbidity or mortality \[[@ref016]\]. {#kca-1-kca170009-g001} An early pilot study of pre-surgical interferon-alpha based immunotherapy in patients with metastatic ccRCC and a primary tumor *in situ* showed that those patients who did not progress on immunotherapy and were able to undergo nephrectomy had better overall survival 11.5 months versus 3 months) than those who progressed while on immunotherapy. In addition, there was no increased surgical morbidity in patients who received presurgical immunotherapy. This pilot study highlights how pre-surgical therapy can better select patients for nephrectomy and that individuals with progressive disease are considerably less likely to benefit from nephrectomy \[[@ref021]\]. Following the introduction of anti-angiogenic and targeted therapy agents for RCC, a phase II study of presurgical bevacizumab in treatment-naïve patients with metastatic ccRCC was performed to test if 1) presurgical bevacizumab was safe and if 2) presurgical anti-angiogenic therapy could select patients who would benefit from cytoreductive nephrectomy \[[@ref015]\]. This single-arm pilot study had primary endpoints of time to disease progression and safety. The study included 23 patients who received bevacizumab plus erlotinib and 27 patients who received bevacizumab alone, with 82% of the study population having intermediate-risk features by Memorial Sloan Kettering Cancer Center (MSKCC) criteria. Median progression free survival (PFS) was 11 months, which met predetermined outcome measures for improvement in PFS attributable to bevacizumab. The 50 patients included had an overall response rate (ORR) of 12% and median overall survival (OS) of 25.4 months. Ultimately, 84% of the study population underwent nephrectomy with no intraoperative complications attributable to the study drug(s); however, 20.9% of those individuals who underwent nephrectomy had delayed wound healing at post-operative week 4. Thus this study showed that outcomes of patients treated pre-surgically with bevacizumab were similar to outcomes in groups treated post-surgically, but also suggests that at least this particular agent contributes to delayed wound healing when compared to retrospective historical controls. Since single-agent bevacizumab is not considered a first line option for metastatic ccRCC and is rarely used even in second line setting in current clinical practice, this study provides general guidance on the use of antiangiogenic therapy in the presurgical setting but does not directly reflect current treatment of RCC. A prospective pilot study of sorafenib, a multitargeted TKI, was performed in a preoperative population with stage II or higher renal masses, including 17 patients with localized disease and 13 with metastatic disease, which also included non-clear cell histologies \[[@ref025]\]. 28/30 of these patients were evaluable for response to sorafenib prior to surgery with 2/28 having a partial response (PR) and 26/28 with stable disease (SD). Interestingly thirty patients were able to proceed with nephrectomy, and there were no surgical complications related to sorafenib administration. Sunitinib is a multiple receptor TKI that is widely used in the setting of metastatic RCC. In a randomized phase III trial of sunitinib versus interferon-alfa for metastatic RCC, patients had a median OS of greater than 2 years in the sunitinib group with a median PFS of 11 months \[[@ref029]\]. Two prospective single-arm phase II studies of presurgical sunitinib prior to planned nephrectomy have been published with a variable number of cycles given prior to surgery \[[@ref016]\]. These studies of pre-surgical sunitinib included a combined total of 66 patients with metastatic ccRCC that were treatment naïve, with 32% having MSKCC poor risk disease. The majority of the cohort (73%) achieved clinical benefit. The median PFS for the intermediate-risk patients was 8 months in the combined cohort, with the poor-risk patients having a PFS of 6.0 months. The OS for the combined cohort was 15.2 months, with the intermediate-risk group having a significantly longer OS than the poor-risk group (26.0 months vs 9.0 months) which is consistent with prior studies with other treatment modalities. Seventy-one percent of the cohort went on to nephrectomy; and, patients in these studies were off systemic therapy perioperatively for approximately 29 days, with a significant minority of assessable patients (36%) progressing during or shortly after this required treatment break. In addition 13% of individuals undergoing nephrectomy had complication of delayed wound healing. Not surprisingly, progression in the primary tumor *in situ* or in metastatic sites at the time of planned surgery portended for shorter OS. A recent single-arm, phase II study of pazopanib- a tyrosine kinase inhibitor which is a standard of care option in first line treatment of metastatic RCC- in the presurgical setting prior to cytoreductive nephrectomy has shown a favorable safety and efficacy profile \[[@ref020]\]. In this study, 104 patients with metastatic ccRCC that were treatment naïve received pazopanib 800 mg daily prior to undergoing cytoreductive nephrectomy (CN), with the study's primary end point being clinical or radiographic response prior to surgery. One hundred patients were included in the final analysis with 84% of the study population achieving a clinical benefit defined as tumor size reduction or stable disease prior to nephrectomy with a 14.4% median reduction in the size of the primary kidney tumor. The majority of the study population (71%) achieved only stable disease, thus pazopanib did not facilitate for better surgical outcomes in this patient population as a whole. In addition patients with poor risk disease as defined by MSKCC criteria did not benefit from nephrectomy and had very poor outcomes with an OS of 5.7 months. Ultimately, 63% of the study population underwent nephrectomy with a surgical complication rate of 22%, with the most common complications being bleeding and delayed wound healing. The PFS and OS for the study population were 7.1 months and 22.7 months, respectively, which are similar to studies of pazopanib in the metastatic ccRCC setting following nephrectomy \[[@ref031]\]. Each of these presurgical studies not only provided clinical information on patient outcomes in the metastatic ccRCC setting ([Table 1](#kca-1-kca170009-t001){ref-type="table"}), but also allowed for tissue based molecular analyses to elucidate underlying mechanisms of drug sensitivity and resistance, as well as providing a more complete picture of the biology driving tumor growth and progression. A retrospective analysis of RNA microarray and reverse phase protein array profiling of 37 tumors from patients treated in the presurgical bevacizumab plus or minus erlotinib study \[[@ref015]\] showed worse OS and PFS in those patients with upregulation of the phosphoinositide 3-kinase (PI3K) pathway and cell-cycle related pathways \[[@ref023]\]. ###### General characteristics and outcomes of presurgical systemic therapy studies in mRCC Study Type Presurgical Systemic Patients^a^ Best Completed Overall Survival ------------------------------ ------------ ---------------------- --------------- ------------ ----------- ------------------ Bex et al. \[[@ref017]\] Phase I/II Sunitinib 16 CR^b^ 0/16 11 (69) 11.5 PR^c^ 2/16 SD^d^ 9/16 Jonasch et al. \[[@ref015]\] Phase II Bevacizumab 50 CR 1/50 42 (84) 25.4 +/--Erlotinib PR 5/50 SD 29/50 Cowey et al. \[[@ref025]\] Phase I/II Sorafenib 30 CR 0/30 30 (100) N/A^e^ 17 Local PR 2/30 13 Metastatic SD 26/30 Powles et al. \[[@ref016]\] Phase II Sunitinib 66 CR 0/66 47 (71) 15.2 PR 13/66 SD 35/66 Powles et al. \[[@ref020]\] Phase II Pazopanib 104 CR 0/100 63 (61) 22.7 PR 13/100 SD 71/100 ^a^studies included patients with metastatic disease only unless otherwise specified; ^b^CR = complete response; ^c^PR = partial response; ^d^SD = stable disease; ^e^N/A = not available. A subsequent retrospective analysis of tissue obtained from 41 untreated primary RCCs, 42 bevacizumab treated RCCs, and 39 sunitinib-treated RCCs showed that the tumors from patients pretreated with antiangiogenic therapy had increased infiltration of CD4+ and CD8+ T lymphocytes \[[@ref024]\]. Those tumors from pretreated patients showed higher infiltration of CD4+ FOXP3+ regulatory T cells and higher expression of programmed death-ligand 1 (PD-L1), and both of these immunosuppressive features were correlated with T-cell infiltration and worse patient survival \[[@ref024]\]. The effect of presurgical TKI therapy on ten predetermined biomarkers was assessed in sequential tissue pooled from patients who were pretreated with VEGF TKIs from three phase II studies, including a matched untreated renal biopsy specimen and subsequent treated primary \[[@ref033]\]. This study showed TKI treatment resulted in a reduction in vessel density; reduction in phospho-S6K, PD-L1, and FOXP3; and, an increased expression of cytoplasmic FGF-2 and MET receptor in vessels- this highlights the heterogeneous, dynamic molecular changes occurring with TKI therapy \[[@ref033]\]. In another tissue-based molecular profiling study of sequential tumor samples from primary tumors obtained from patients with metastatic ccRCC pretreated with sunitinib versus untreated control samples, the investigators assessed for intratumoral heterogeneity as defined by morphologic and molecular (including DNA, mRNA, and protein) variance within a single tumor \[[@ref022]\]. This study found that morphologic intratumoral grade heterogeneity was present to a higher degree in sunitinib-treated versus untreated samples and that driver-mutation gene signatures and protein expression were increasingly variable in treated samples. The phase II trial of pazopanib in the presurgical setting in metastatic RCC assessed antibody-based biomarkers of PD-L1, C-MET, HIF1*α*, VEGFR2, and VHL in sequential tissue specimens from pre and post-pazopanib treatment \[[@ref020]\]. They found in comparison of pre and post pazopanib specimens a statistically significant decrease in CD8 expression and reduction in expression of VHL and C-MET with increased PD-L1 expression. No biomarkers in this study were associated with response to therapy. Thus these pilot and early-phase studies indicate that the approach of pre-surgical targeted therapy is generally safe, can provide disease control albeit predominantly in intermediate-risk cases, and can select patients with poor-risk or actively progressing disease that will not benefit from nephrectomy. In addition molecular studies on tissue from patients that receive presurgical systemic therapy can give insight into underlying biology, resistance mechanisms, as well as predictive and prognostic biomarkers. Immune checkpoint therapy is currently standard of care in the second line setting for metastatic ccRCC \[[@ref003]\] following the phase III study showing nivolumab, an anti-programmed cell death 1 (PD-1) receptor antibody, had an OS benefit compared to everolimus in metastatic ccRCC in second line setting \[[@ref034]\]. While prior combinations of anti-angiogenic agents and cytokine-based therapies did not show benefit and/or were too toxic, the relatively favorable side effects profile of modern immune checkpoint therapies have led to multiple ongoing trials involving the combination of these immune directed agents with anti-angiogenics \[[@ref035]\]. Preclinical studies have shown a signal for increased efficacy of modern immunotherapy in the neoadjuvant setting compared to adjuvant setting due to increased, sustained activation of T cell antitumor activity with an effector/memory phenotype \[[@ref036]\]. The mechanisms by which an intact tumor may serve to prime an antitumor immune response in the setting of neoadjuvant or presurgical therapy are numerous, but a general schematic is that the generation of neoantigens from dying tumor cells can lead to an quantitative expansion of targeted tumor-specific T cells as well as a sustained qualitative improvement in immune memory even after the tumor is resected. In this setting, the intact tumor is necessary to expand the anti-tumor T cell army, as well as to train these T cells to systemically surveil the body for tumor cells even after the primary tumor or metastasis is removed \[[@ref036]\]. Thus, early phase clinical studies are underway to evaluate the safety and efficacy of anti- PD-1 receptor antibody in the presurgical setting both as single-agent therapy \[[@ref037]\] and in combination with anti-VEGF therapy or with ipilimumab, an anti-CTLA4 antibody \[[@ref038]\]. A phase I study of nivolumab in the perioperative setting for patients with metastatic RCC for which nephrectomy is planned is ongoing (ADAPTer), with plans to recruit nineteen patients and requires sequential tissue specimens pre- and post-nivolumab therapy \[[@ref037]\]. In addition, studies are ongoing to combine radiation with immune therapy in presurgical and metastatic settings with the idea that radiation can further drive neoantigen production and take advantage of potential abscopal effects of local radiation to lead to immune-mediated regression of distant disease \[[@ref039]\]. With a strong preclinical rationale combined with the potential for long-term sustained responses following surgery and predictive immune-related biomarkers, presurgical immunotherapy both alone and in combination with other therapies in metastatic RCC is very promising. POSTSURGICAL THERAPY IN METASTATIC RCC {#sec0025} ====================================== The current guidelines for metastatic RCC recommend cytoreductive nephrectomy in patients with good performance status and low-volume metastatic disease, which is based on the prospectively confirmed survival benefit of nephrectomy in the era of systemic therapy with interferon alfa \[[@ref007]\]. Multiple trials of first-line and subsequent line targeted inhibitors have been performed for metastatic clear cell and non-clear cell histologies, with a significant number of patients in those studies having undergone prior cytoreductive nephrectomy at time of initiation of systemic therapy \[[@ref029]\]. However, there are no randomized trials completed and reported to date that specifically address perioperative or postsurgical systemic therapy in the setting of nephrectomy versus systemic therapy alone. Multiple randomized control trials of TKIs, including sunitinib and pazopanib, alone versus TKI following nephrectomy are ongoing (CARMENA and TARIBO) \[[@ref045]\]. In addition, multiple prospective and retrospective studies suggest that patients with isolated surgically resectable metastatic disease obtain clinical benefit from metastasectomy, even in the era of targeted therapy, though randomized data are lacking \[[@ref019]\]. Thus in current practice, almost all patients seeking treatment for metastatic RCC will receive systemic therapy at some point in their treatment course, and those with good performance status and who are amenable to surgery would be advised to undergo upfront nephrectomy, and possibly metastasectomy in appropriately selected cases. CONCLUSIONS {#sec0030} =========== Metastatic RCC requires a multidisciplinary, team-based treatment approach that incorporates surgical resection and systemic therapy. Cytoreductive nephrectomy in patients with metastatic RCC has been shown to prolong survival in selected patients with good prognosis and performance status; however, large randomized control trial data are still lacking in this space. This review highlights important retrospective and early-phase prospective studies on the use and timing of systemic therapy, including targeted anti-angiogenic agents and immune therapy, prior to cytoreductive nephrectomy. Published data thus far support the efficacy and safety of a pre-surgical systemic therapy approach using standard of care TKIs including sunitinib or pazopanib, though randomized data are still lacking. Given the rapid proliferation of effective immune checkpoint therapy options for metastatic RCC, the community's increasing familiarity with their use and management of their side effects in the second-line setting, and the strong biologic rationale, it is likely that immune checkpoint therapy will move to the front-line and presurgical settings for metastatic RCC in the near future. The molecular alterations and potential biomarkers for selecting those patients who would benefit most from nephrectomy and specific systemic therapies are evolving but still not well understood, which highlights the necessity and benefit to sequential tumor sampling in neoadjuvant and presurgical clinical trials. Data from current ongoing perioperative trials with TKIs together with immune checkpoint blocking agents will better inform these questions of best choice and timing of systemic treatment in metastatic RCC.

Introduction {#Sec1} ============ Taxol is a diterpenoid natural product, had been originally isolated and chemically identified from the bark of Pacific yew *Taxus brevifolia*^[@CR1]^. It has been approved by FDA in 1992 as a broad-spectrum drug for treatment of refractory ovarian, breast, melanoma, pancreatic and lung cancers. The anticancer activity of Taxol elaborated from its unique affinity to bind with tubulin of tumor cells, stabilizing microtubules from depolymerization, arresting the cellular division at G2-M phase, causing cell apoptosis^[@CR2]^. With the vast therapeutic applications of Taxol derivatives, their global demand continues to increase by about 6--10% annually, thus, the Taxol sources are the ongoing challenge for sustaining its affordability as commercial drugs^[@CR3]^. Exploring the endophytic fungi from Taxol producing and non-producing plants, with their fast growth, cost effectiveness and ability to produce Taxol independent on plant source, in addition to the feasibility of nutritional and molecular manipulation, raised the hope for using fungi for Taxol industrial production^[@CR3]^. More than 200 fungal endophytes were identified from *T*. *baccata* and about 10% of this population has the potentiality to produce Taxol^[@CR4],[@CR5]^. The biodiversity of endophytic fungi and their Taxol yield have been extensively reviewed^[@CR6],[@CR7]^. The putative Taxol biosynthetic pathway in fungi requires about 19 enzymatic steps starting from the precursor geranylgeranyl diphosphate (GGPP) that cyclized into taxa-4(5),11(12)-diene by taxadiene synthase followed by hydroxylation of taxadiene nucleus by cytochrome P450-monooxygenases (taxadiene 5α-hydroxylase and taxane 10β-hydroxylase)^[@CR3],[@CR8],[@CR9]^. The metabolic pathways for synthesis of Taxol and melanin from acetyl-CoA precursor has been summarized (Fig. [S1](#MOESM1){ref-type="media"}). Several studies claimed the traditional medicinal plants of ethnopharmacological relevance could be a fertile source for various therapeutic compounds^[@CR10]^. Among these plants, species of Podocarpaceae were nominated for their antibacterial, antifungal and anticancer activities^[@CR7],[@CR11],[@CR12]^. From our previous study, twenty-four endophytic fungal isolates were recovered from *Podocarpus gracilior* and screened for Taxol biosynthetic potency, among them, *Aspergillus terreus* EFB108 was reported as a potent Taxol producer^[@CR7]^. However, the attenuation of Taxol biosynthetic machinery of *A*. *terreus* with subculturing and aging was the technical challenge that limits the ongoing application of this fungal isolate as industrial platform for Taxol production. Similar results were reported the silencing of expression of Taxol encoding genes with multiple subculturing of the endophytic fungi^[@CR13]--[@CR15]^, for examples, *Taxomyces andreanae*^[@CR13]^, *Tubercularia* sp^[@CR16]^ and *Periconia* sp^[@CR17]^. Interestingly, with addition of surface sterilized leaves of *P*. *gracilior*, the biosynthetic potency of Taxol by *A*. *terreus* had been restored completely comparing to the original isolate^[@CR7]^. So, re-triggering of the biosynthetic gene cluster of Taxol by *A*. *terreus* could be due to the intimate physical interaction with the endogenous microbiome of *P*. *gracilior*. Thus, the objective of this work was to evaluate the Taxol yield of *A*. *terreus* "endophyte of *Podocarpus gracilior*" in response to storage, multiple subculturing and potency to restore their Taxol biosynthetic machinery upon addition of *P*. *gracilior* leaves. And to unravel the molecular and metabolic identity of Taxol biosynthetic machineries of *A*. *terreus* associated with the attenuation and restoration of Taxol biosynthesis upon addition of *P*. *gracilior* leaves. Materials and Methods {#Sec2} ===================== Aspergillus terreus growth conditions, Taxol extraction and chemical identification {#Sec3} ----------------------------------------------------------------------------------- *Aspergillus terreus* EFB108 (MF377552), an endophyte of *Podocarpus gracilior*, had the highest Taxol producing potency according to our previous study^[@CR7]^. The fungal isolate was grown on MID media^[@CR18]^ with slight modifications. Briefly, the medium contains (g/l): xylose 27 g, asparagine 6 g, yeast extract 0.5 g, soytone 1.0 g, Ca(NO~3~)~2~ 0.2 g, KNO~3~ 0.08 g, KCl 0.06 g, MgSO~4~ 0.36 g, NaH~2~PO~4.~ H~2~O 0.02 g, FeCl~3~ 0.002 g, MnSO~4~ 0.005, ZnSO~4~. 7H~2~O 0.003 g, H~3~BO~3~ 0.014 g and KI 0.0001 g, dissolved in distilled water of initial pH 8.0. Two plugs of 6-day-old *A*. *terreus* were inoculated to 50 mL medium/250 mL Erlenmeyer flask, incubated for 20 days at 30 °C. The cultures were filtered with sterile cheesecloth, the filtrates were amended with 0.03% sodium bicarbonate to precipitate fatty acids, and Taxol was extracted with double volume of dichloromethane (DCM). The organic phase was collected, evaporated to dryness, and the residues were re-dissolved in 2 mL methanol. Taxol was identified by thin layer chromatography (TLC) using 1 mm (20 × 20 cm) pre-coated silica gel plates 60 F~254~ (Merck KGaA, Darmstadt, Germany), with the solvent system methylene chloride/methanol/dimethyl formamide (90:9:1, v/v/v)^[@CR19]^. Taxol was detected by UV illumination at 254 nm, then the plates were sprayed with 1% acidic vanillin, gentle heating, a dark gray spot was developed after 24 hours, allocating the Taxol spots^[@CR20]^ comparing to authentic Taxol (Cat. \# T7402). The putative spots of silica containing Taxol were scraped-off, dissolved in methylene chloride, the purity and concentration of Taxol were analyzed by HPLC (Agilent Technology, G1315D) of C18 reverse phase column (Cat.\# 959963-902) with isocratic mobile phase of methanol/acetonitrile/water (25:35:40, v/v/v) at flow rate 1.0 mL/min for 20 min^[@CR21]^. The fractions were scanned from 200 to 500 nm by photodiode array detector (DAD), their identity and concentration were confirmed from the retention time and peak area at 227 nm comparing to authentic Taxol. The chemical structure of extracted Taxol was confirmed from the ^1^H and ^13^C NMR spectra (JEOL, ECA-500II, 500 MHz NMR) comparing to authentic Taxol^[@CR7]^. Effect of subculturing and storage time of A. terreus on its Taxol productivity {#Sec4} ------------------------------------------------------------------------------- The effect of subculturing of *A*. *terreus*, till the 10^th^ generation, on its Taxol productivity was assessed. First culture, is the originally recovered isolate from *P*. *gracilior*, that subsequently transferred into potato dextrose agar (PDA) plates, incubated at 30 °C for 10 days, and then repeated the subculturing till the 10^th^ subcultural generation. The isolate from each generation was centrally inoculated to PDA plates, incubated at 30 °C, and the conidial and mycelial pigmentation were inspected daily. The fungal isolates from each generation were grown on modified MID media^[@CR7]^, incubated under standard conditions for 20 days, then Taxol was extracted, purified and quantified as described above. The mycelial fungal pellets were washed by sterile saline and kept at −20 °C for further biochemical analyses. The effect of storage time of *A*. *terreus* on their Taxol biosynthetic stability has been evaluated. The original cultures of *A*. *terreus* were stored as slope cultures on MID media^[@CR18]^ at 4 °C, the Taxol productivities were evaluated after 1, 2, 3, 4, 5 and 6 months of storage. After culture incubation, the Taxol was extracted and quantified by the standard assay as mentioned above. As well as, the mycelial melanin and acetyl-CoA were extracted from each culture in parallel with Taxol estimation, as below. Estimation of melanin and acetyl-CoA {#Sec5} ------------------------------------ Melanin was extracted from the mycelial culture of *A*. *terreus*^[@CR22]^. Briefly, 0.5 g of mycelial biomass was homogenized in 2 M NaOH (pH 10.5), incubated for 48 h, centrifuged at 5000 rpm for 20 min, the pH of supernatant was acidified to 2.5 with 2 M HCl. The mixture was incubated overnight, centrifuged at 5000 rpm for 20 min, supernatant was decanted, and the precipitate was collected, acid hydrolyzed with 6 M HCl at 100 °C for 2 h. The mixture was treated with ethylacetate, the precipitate was dissolved in 2 M NaOH, centrifuged at 5000 rpm for 15 min, the supernatant was collected in pre-weighed tube, acidified with 6 M HCl, dried at room temperature. The purified melanin was dissolved in 1 mL borate buffer (pH 8.0, 100 mM) and measured at 459 nm^[@CR23]^ regarding to authentic melanin. The concentrations of acetyl-CoA were assessed for the different cultures of *A*. *terreus*. The fungal biomass was harvested and pulverized (10 g) in liquid nitrogen, dispensed in 10 mL of 100 mM HEPES buffer (pH 7.5) containing 1 mM DTT, thoroughly vortex for 5 min. The homogenate was centrifuged at 10000 rpm for 5 min at 4 °C, and the supernatant was used for acetyl-CoA determination^[@CR24]^. Acetyl- CoA was determined by HPLC (Agilent Technol, G1315D, C18 reverse phase column Cat. \# 959963-902) with mobile phase 100 mM monosodium phosphate, 75 mM sodium acetate and 6% acetonitrile, the pH was adjusted to 4.6 with phosphoric acid. The flow rate of mobile phase was 40 µl, and detection wavelength at λ~259~ nm^[@CR25]^. As well as, the concentration of acetyl-CoA was determined by Acetyl-Coenzyme A assay Kit (Cat. \# MAK039, Sigma-Aldrich, Louis, MO, USA) according to the manufacturer's instructions. The amount of acetyl-CoA was calculated based on the peak area and retention time of the authentic sample (Cat\#. A2056). Activities of rate-limiting enzymes of Taxol biosynthesis {#Sec6} --------------------------------------------------------- The activities of rate-limiting enzymes of Taxol biosynthesis; taxadiene synthase, 10-deacetyl-baccatin III-*O*-acetyltransferase and baccatin III 13-*O*-(3-amino-3-phenylpropanoyl) transferase were assessed. For extraction of these enzymes, the collected fungal biomass (20 g) were washed, pulverized in liquid nitrogen, and dispensed in 50 mL of 30 mM, pH 8.0 HEPES buffer with 5 mM sodium ascorbate, 5 mM DTT, 5 mM Na~2~S~2~O~5,~ 15 mM MgCl~2,~ 10% (v/v) glycerol and 1% (w/v) polyvinylpolypyrrolidone^[@CR26]^. The mixtures were shacked thoroughly for 15 min, centrifuged at 8000 rpm for 15 min at 4 °C, the supernatant was used as source of crude enzymes. The activities of the putative enzymes for Taxol biosynthesis were estimated as follow; ### Taxadiene synthase (TDS) activity {#Sec7} The activity of TDS was determined as adopted by^[@CR26]^ with minor modifications. The reaction mixture contains 50 mM GGPP, 5 mM MgCl~2~ in 50 mM Tris-HCl pH 8.0 and 500 µl of the enzyme extract in 2 mL reaction volume. The reaction was incubated at 32 °C for 1 h, terminated by 50 µl of EDTA (0.5 M, pH 8.0). The GGPP level was determined by TLC using silica gel plates 60 F~254~ (Merck KGaA, Darmstadt, Germany) with developing solvent system propan-2-ol, ammonia and water (9:3:1) regarding to the authentic GGPP (Cat\#. G6025) after visualization by vapor iodine^[@CR27]^. The intensity of GGPP spots were determined by ImageJ software^[@CR28]^. One unit of TDS activity was expressed by the amount of enzyme consuming 1 µmol of GGPP per min under the standard assay conditions. ### 10-Deacetylbaccatin III-*O*-acetyltransferase (DBAT) activity {#Sec8} The activity of DBAT was determined^[@CR29]^. The reaction mixture contains 20 mM 10-deacetylbaccatin (10-DAB), 20 mM acetyl-CoA dissolved in 50 mM Tris-HCl (pH 8.0), and 500 µl of enzyme extract in 2 mL total volume. After incubation at 30 °C for 1 h, the mixture was extracted with equal volume of chloroform, the organic layer was evaporated, and the residue was dissolved in 0.5 mL ethanol. The released baccatin III was quantified by HPLC with solvent system; water: acetonitrile (1:9 v/v) at flow rate 1 mL/min^[@CR30]^. The concentration of baccatin III was determined from the peaks area comparing to authentic sample. One unit of DBAT was expressed by amount of enzyme releasing 1 µmol of baccatin III under standard assay conditions. ### Baccatin III-13-*O*-(3-amino-3-phenylpyropanyoyl) transferase (BAPT) activity {#Sec9} The BAPT activity was estimated^[@CR31]^ as follows; the mixture contains 10 mM acetyl-CoA, 10 mM baccatin III in Tris-HCl pH 8.0 and 500 µl of enzyme extract in 1 mL reaction volume. The pH was increased to 9.0 with sodium bicarbonate, followed by addition of 10 µM benzoyl chloride, incubated for 1 h to induce *N*-benzoylation^[@CR32]^. The concentration of synthesized 2-deoxytaxol was determined by HPLC. One unit was expressed by the amount of enzyme releasing 1 µmol of 2-deoxytaxol at standard conditions. RNA Isolation, cDNA Synthesis, real-time PCR analysis {#Sec10} ----------------------------------------------------- The molecular expression of Taxol biosynthesis rate limiting genes such as *tds*, *dbat* and *bapt* by *A*. *terreus* were assessed. The fungal mycelial pellets were pulverized to a fine powder and the total RNA was extracted using IQeasy^TM^ plus Plant Mini Kit (Cat\#. 17491, iNtRON Biotech. Korea). cDNA was synthesized by SuperScript III First Strand Synthesis Kit (Invitrogen, USA) with oligo-dT primes. For qRT-PCR analyses, the reaction mixtures contain cDNA, forward and reverse primers, with Topreal^TM^ One-Step RT qPCR Kit (Cat\#. RT432S) according the manufacturer's instructions using the real-time PCR machine (Agilent Technology, Stratagene Mx3005P). The primers for RT-qPCR molecular expression analysis ion of *tds*, *dbat* and *bapt* were listed in Table [1](#Tab1){ref-type="table"}. The qPCR was programmed to initial denaturation at 95 °C for 3 min, followed by 40 cycles of 95 °C for 15 s, 55 °C for 30 s (annealing), 72 °C for 1 min (extension). Melting curve analyses were performed at 55--95 °C. Triplicates of each sample were conducted. Data were normalized to the constitutively expressed actin *actaA* gene of *A*. *terreus* as endogenous control. The expression folds of the target genes were calculated from standard curve of relative quantification^[@CR33]^. Statistical comparisons were conducted using Student's *t*-test, and the p-value ≤ 0.05 were considered significant. Data are presented as fold of change between the control fungal cultures and cultures amended with leaves of *P*. *gracilior*, comparing to negative controls of *P*. *gracilior*.Table 1Oligonucleotides primers for PCR amplification of fragments of Taxol encoding genes.PrimerGene IDPrimers (F, R) 5′-3′qTDSJQ618974.15′-CCAGGTAGTTTCGTGTCCAA-3′; 5′-GTTGCCTGAACAGGTGAGTA-3′qDBATKR047791.15′-CACCACTTTCGAAGGGATACT-3′; 5′-TCACGATTGTTGACGTGAAATG-3′qBAPTKC959480.15′-TGAGGACCTCCATCTCTTCAT-3′; 5′-TACACATTCGCTCCCACAAC-3′pbcRXM654111.15′-GAATGGCGTAGGACTGTTG-3′; 5′-CGTATCCATAAACTCGGTAATC-3′*acta*AXM749985.15′- GACTGGTTTGGCAATTGATG- 3′; 5′- GCATCAGTGATCTCACGCTT- 3′ Effect of *P*. *gracilior* leaves on restoring the Taxol biosynthesis by *A*. *terreus* {#Sec11} --------------------------------------------------------------------------------------- The surface sterilized leaves of *P*. *gracilior* had a significant effect on inducing the Taxol yield by *A*. *terreus*^[@CR7]^. The fungal isolate was grown on modified MID media, incubated at 30 °C for 10 days, then the cultures were amended with different concentration (0.5, 1.0, 3.o and 5.0%) of surface sterilized leaves of *P*. *gracilior*, and the cultures were re-incubated for 20 days under standard conditions. Negative control media of *P*. *gracilior* leaves at the same concentrations without fungal spores were used. After incubation, the cultures were filtered, Taxol was estimated, and activity and molecular expression of Taxol encoding genes were analyzed as described above. Proteomics analyses of *A*. *terreus* in response to addition *P*. *gracilior* leaves {#Sec12} ------------------------------------------------------------------------------------- The proteomic analyses of *A*. *terreus* in response to supplementation with *P*. *gracilior* leaves were assessed. After incubation of each fungal cultures, the total intracellular proteins were extracted, and quantified^[@CR34]^. Equal concentrations of extracted proteins from each fungal culture were electrophoresed using 12% SDS-PAGE^[@CR35]^. The prominent over-induced protein bands in response to addition of *P*. *gracilior* leaves were identified by MALDI-TOF/TOF analysis (Bruker Daltonics) at the Proteomics Research Lab, Medical Research Center, Alexandria University, Egypt. The target gel bands were excised from the gel, washed with 50 µl washing buffer (10 mM NH~4~HCO~3~ and 50% ACN), followed by addition of 100 µl ACN to shrink the gel and 25 µl 20 mM NH~4~HCO~3~ to rehydrate the gel~,~ then the drying by vacuum centrifuge^[@CR36],[@CR37]^. The gel was reduced by 50 µl reduction solution (10 mM DTT/10 mM NH~4~HCO~3~) for 15 min at 56 °C, then gel alkylation (55 mM IAA/10 mL NH~4~HCO~3~) for 20 min in dark, then gel washing and drying prior to in-gel trypsinization (20 ng/μl of trypsin in 10 mM NH4HCO3)^[@CR37]^. The peptides were extracted with 50% ACN and 1% TFA, analyzed by MALDI-TOF/TOF. A linear gradient of acetonitrile (ACN) (5.0--60%), with flow rate 250 μl min^−1^ was used for eluting peptides from the column to the mass spectrometer. MS/MS data were acquired independently in which the MS1 data were acquired for 250 ms at m/z 400--1250 and at m/z 50--2000 Da. The raw MS/MS data files were extracted, and the peptides were analyzed and identified by Protein Pilot 4.0 (ABSCIEX)^[@CR38]^ normalizing to the proteome of *Aspergillus terreus*^[@CR39]^. The identification criteria included at least five peptide fragment ions per protein with E-values \< 0.05. Compliance with Ethical Standards {#Sec13} --------------------------------- This article does not contain any studies with human or animal subjects performed by any of the authors. Results {#Sec14} ======= Metabolic stability of *A*. *terreus* for Taxol production with subculturing and storage {#Sec15} ---------------------------------------------------------------------------------------- The productivity of Taxol by *A*. *terreus* till the 10^th^ subcultural generations have been investigated. The culture of *A*. *terreus* grown on PDA, were incubated at standard conditions for 20 days, and the visual appearance of the fungal mycelial pigmentation was daily inspected and photographed. The cultural mycelial biomass was collected and kept at −20 °C, while, the filtrate was used for Taxol extraction and quantification. From the results (Fig. [1](#Fig1){ref-type="fig"}), Taxol productivity of *A*. *terreus* was sequentially decreased, accompanied with visual fading of the mycelial melanin pigmentation with the multiple subculturing. The Taxol yield for the original culture of *A*. *terreus* grown on modified MID media was 268 µg/l, while their yields for the 5^th^, 7^th^ and 10^th^ generations were 133.4, 94.7 and 78.2 µg/L, respectively, as revealed from the TLC and HPLC chromatograms. Thus, by the 6^th^ subculture, *A*. *terreus* lost about 60% of its metabolic potency for Taxol production comparing to the original isolate. In parallel with the reduction on Taxol productivity by *A*. *terreus*, a visual fading of their mycelial pigmentation was observed with the fungal subculturing. The mycelial melanin was quantitively estimated, intriguingly, the yield of melanin of *A*. *terreus* was reduced sequentially with fungal subculturing, by the 5^th^ generation, the yield of melanin was reduced by about 30%, comparing to the original culture, that was strongly correlated with the total yield of Taxol. Overall, the fungus retains about 50% of its Taxol productivity by the 7^th^ subculture. The metabolic pathways for synthesis of these two metabolites are competing for the same biosynthetic precursors "acetyl-CoA" (Fig. [S1](#MOESM1){ref-type="media"}). So, the decreasing yields of both Taxol and melanin could be ascribed to decreasing on influx of acetyl-CoA to these metabolic pathways, through overall silencing of the different machineries generating pyruvate and acetyl-CoA precursors.Figure 1Taxol productivity and mycelial pigmentation of *A*. *terreus* with the multiple subculturing. (**A**) Morphological features of *A*. *terreus* EFB108 sub-cultured on Czapek's agar (upper panel) and MID medium (lower panel) till the 10^th^ subcultures. (**B**) TLC chromatogram of extracted Taxol. (**C**) HPLC concentration of Taxol. (**D**) Relative concentrations of melanin extracted from the different subcultures of *A*. *terreus*. (**E**) Taxol yield of *A*. *terreus* with the fungal storage as slope cultures at 4 °C for 8 months. The influence of storage time of the original isolate *A*. *terreus* on its Taxol productivity was assessed. The original isolate was stored as slope culture at 4 °C, and its Taxol productivity was checked monthly along 8 months. From the obtained results (Fig. [1F](#Fig1){ref-type="fig"}), the yield of Taxol was sequentially decreased with the storage time. By the 6^th^ month of storage, the yield of Taxol by *A*. *terreus* was reduced by about 3.5 folds, comparing to the original culture. The mechanism of metabolic changes of *A*. *terreus* for Taxol biosynthetic potency with the storage time remains equivocal, however, it might be attributed to the aging of cells or re-programing of the cellular machinery with the storage time. So, the metabolic attenuation of *A*. *terreus* for Taxol biosynthetic potency could be not only due to the multiple subculturing but also to the storage time negating the carrying-over of Taxol residues from the host plant. Activity and molecular expression of the rate-limiting enzymes of Taxol biosynthesis {#Sec16} ------------------------------------------------------------------------------------ To validate the biosynthetic stability of Taxol by *A*. *terreus*, the rate-limiting enzymes of Taxol biosynthesis such as geranylgeranyl-pyrophosphate synthase (GGPPS), taxadiene synthase (TDS), 10-deacetylbaccatin III-*O*-acetyltransferase (DBAT) and baccatin III 13-*O*-(3-amino-3-phenylpropanoyl) transferase (BAPT) were assessed. After cultural incubation, the intracellular proteins were extracted, and the activity of GGPPS, TDS, DBAT and BAPT were estimated. From the results (Fig. [2](#Fig2){ref-type="fig"}), the activity of GGPPS, TDS, DBAT and BAPT were subsequently suppressed with the fungal subculturing comparing to the original culture, correlating with the decreasing on Taxol yield. The activity of these enzymes by *A*. *terreus* were reduced by about 50% by the 5^th^ fungal cultural generation regarding to the original culture. However, these enzymes retain less than 10% of their initial activities by the 10^th^ fungal cultural generation. Interestingly, the activities of Taxol rate-limiting enzymes by *A*. *terreus* being consistent with the Taxol yield for the different cultural generations. In addition to the activity of putative enzymes and quantitative analyses of Taxol, visual inspection of TDS activity on TLC was assessed for the different fungal subcultures (1^st^-10^th^). The TDS activity was assessed, the residual concentration of GGPP substrate was determined by TLC, visualized by vapor iodine solution^[@CR27]^. From the results (Fig. [2E](#Fig2){ref-type="fig"}), the intensity of residual GGPP spots, as determined by ImageJ software, was strongly increased by about 5 folds with the progression of fungal subculturing suggesting the attenuation of TDS activity, in contrary to the original culture.Figure 2Activity and molecular expression of Taxol rate-limiting enzymes (GGPPS, TDS, DBAT and BAPT) by *A*. *terreus* in response to frequent subculturing. After cultural incubation, the intracellular proteins from the fungal biomass were extracted, and the activity of Taxol biosynthetic enzymes such as GGPPS (**A**), TDS (the onset figure shows the relative activity of TDS with the fungal subcultures) (**B**), DBAT (**C**) and BAPT (**D**), were estimated. (**E**) TLC profile of the residual GGPP of TDS reaction (20 µl) of *A*. *terreus* (from 1^st^ till 10^th^ subculture) visualized by vapor iodine solution. Quantitative RT-PCR analyses for expression of TDS, DBAT and BAPT genes with subcultures of *A*. *terreus* (**F**) using one-step real time kit for qPCR analysis as in Materials and Methods. The molecular expression of Taxol encoding genes were determined by qPCR using cDNA as template, normalizing to actin gene, using the listed primers in Table [1](#Tab1){ref-type="table"}. From the results (Fig. [2F](#Fig2){ref-type="fig"}), the expression of *A*. *terreus* Taxol biosynthetic genes *tds*, *dbat* and *bapt* were sequentially reduced with the fungal subculturing, normalizing to expression of *actinA* as house-keeping gene. Interestingly, the molecular expression of *tds*, *dbat* and *bapt* was conceivably reduced with the fungal subculturing, that correlates with the metabolic yield of Taxol and activities of their rate-limiting enzymes. By the 5^th^ and 10^th^ subculture, the expression of *tds*, *dbat* and *bapt* was reduced by about 15 and 35 folds, respectively, comparing to the original culture. Taken together, the results of Taxol yield by chromatographic approaches, activities of rate-limiting enzymes and molecular expression by qPCR analyses, approve the attenuation of the biosynthetic machinery of Taxol by *A*. *terreus* with the fungal subculturing. These results have been frequently reported for a plethora of secondary metabolites production by fungi with growing as monoaxenic cultures, under standard laboratory conditions^[@CR15]^. Restoring the Taxol biosynthetic potency of *A*. *terreus* upon addition of *P*. *gracilior* leaves {#Sec17} --------------------------------------------------------------------------------------------------- To evaluate the ability of *A*. *terreus* to restore its Taxol biosynthetic machinery, the 7^th^ fungal subcultures were grown on Taxol production medium for 10 days, then amended with sterilized leaves of *P*. *gracilior* and re-incubated under standard conditions. The 7^th^ subculture of *A*. *terreus* was selected as it retains 50% of Taxol biosynthetic potency comparing to original one. Taxol was extracted, and molecular expression of its biosynthetic genes were estimated. Practically, the biosynthetic potency of Taxol by *A*. *terreus* has been gradually restored upon addition of surface sterilized leaves of *P*. *gracilior* comparing to the positive (7^th^ subculture of *A*. *terreus*) and negative control (surface sterilized leaves without fungus) (Fig. [3A](#Fig3){ref-type="fig"}). The yield of Taxol by *A*. *terreus* was significantly increased by about 3.5 folds (395 µg/L) with addition of 3% surface sterilized leaves of *P*. *gracilior* to the fungal culture, comparing to positive controls (7^th^ culture of *A*. *terreus*) (124 µg/L). The Taxol yield by *A*. *terreus* cultures amended with 0.5% and 1.0% sterilized leaves of *P*. *gracilior* was increased by 2 folds (210 µg/L) and 3.2 folds (315 µg/L), respectively, comparing to the positive control fungus as shown from the TLC and HPLC profiles (Fig. [3A,C](#Fig3){ref-type="fig"}). From the HPLC profile, there is no detectable Taxol by the negative control *P*. *gracilior* leaves.Figure 3Effect of surface sterilized *P*. *gracilior* leaves on restoring the Taxol productivity of *A*. *terreus*. The 7^th^ fungal subculture was grown on MID medium, incubated for 10 days, then amended with surface sterilized leaves of *P*. *gracilior* at 0.5, 1, 3 and 5% (w/v), incubated for 20 days, Taxol was extracted and quantified by HPLC (**A**). The activities of Taxol biosynthesis rate-limiting enzymes, TDS, DBAT and BAPT (**B**) were assessed. Molecular expression of TDS, DBAT and BAPT genes by qRT-PCR analysis (**C**). To validate the chromatographic results, the activity of Taxol rate-limiting biosynthetic enzymes "TDS, DBAT and BAPT" were determined. From the results, the activities of these enzymes by *A*. *terreus* were strongly increased upon addition of *P*. *gracilior* leaves, that highly correlated with the yield of Taxol by HPLC and TLC. The activities of TDS, DBAT and BAPT by *A*. *terreus* were increased by about 2.0--3.0 folds upon addition of 3% sterilized leaves of *P*. *gracilior* comparing to the positive control fungal cultures (Fig. [3B](#Fig3){ref-type="fig"}). The activity of TDS was 4.9 ± 0.1, 5.5 ± 0.1, 6.7 ± 0.2 and 5.9 ± 0.2 µmol/min/mg for the 7^th^ *A*. *terreus* cultures amended with 0.5, 1.0, 3.0 and 5.0% of *P*. *gracilior* leaves, respectively, comparing to positive control cultures of *A*. *terreus* (without *P*. *gracilior*) (2.8 ± 0.1 µmol/min/mg). Consistently, the activities of DBAT and BAPT were compatible with the activity of TDS and Taxol yield in response to addition of *P*. *gracilior* leaves, ensuring the restoration of the whole enzymatic system for Taxol synthesis with addition of *P*. *gracilior* leaves. The activity of TDS for the 7^th^ subculture of *A*. *terreus* amended with *P*. *gracilior* leaves regarding to the residual levels of GGPP as substrate was authenticated from visual inspection of TLC sprayed with iodine vapor (Fig. [3D](#Fig3){ref-type="fig"}). From the TLC chromatogram, the intensity of residual GGPP spots was reduced gradually with addition of *P*. *gracilior* leaves, comparing to control (7^th^ culture of *A*. *terreus* without *P*. *gracilior* leaves), assuming the higher activity of TDS. At 3% *P*. *gracilior* leaves, the intensity of GGPP as results of TDS activity of *A*. *terreus* was reduced by about 3 times, ensuring the higher activity of TDS upon plant leaves addition comparing to control. Practically, the metabolic productivity of *A*. *terreus* for Taxol has been restored completely upon supplementation of the fungal culture with surface sterilized leaves of *P*. *gracilior* as authenticated from the chromatographic assessment of Taxol and activities of their rate-limiting enzymes. Further molecular expression analyses to the putative Taxol biosynthetic genes have been investigated by qPCR, using the total cDNA as template, and specific primers as listed in Table [1](#Tab1){ref-type="table"}. The expression folds were normalized to the actin *actaA* gene, as house-keeping gene, and 7^th^ subculture of *A*. *terreus* without *P*. *gracilior* leaves as positive control. From the molecular expression results (Fig. [3E](#Fig3){ref-type="fig"}), the expression of Taxol-biosynthetic genes *tds*, *dbat*, *bapt* of *A*. *terreus* was increased sequentially with incorporation of 3% *P*. *gracilior* leaves by about 4--5 folds comparing to the control culture of *A*. *terreus* (Zero plant leaves). While, the expression of these genes was increased by about 2.5--3.0 upon supplementation of 5% *P*. *gracilior* leaves to the cultures of *A*. *terreus*, comparing to control. From the metabolic yield of Taxol and activities of its biosynthetic enzymes, it is clearly shown the supplementation with surface sterilized leaves of *P*. *gracilior* restore the molecular machinery of Taxol production by *A*. *terreus*. Intracellular levels of *A*. *terreus* acetyl-CoA in response to subculturing and addition of *P*. *gracilior* {#Sec18} -------------------------------------------------------------------------------------------------------------- From the above results, the metabolic correlation of Taxol and melanin biosynthesis by *A*. *terreus* and their attenuation with the multiple fungal subculturing and storage, have been authenticated from the chromatographic, spectroscopic and molecular analyses. Since the two metabolic pathways are acetyl-CoA- dependent as precursors (Fig. [S1](#MOESM1){ref-type="media"}), thus, we hypothesized that this attenuation of Taxol and melanin productivity could be due to diminishing on the influx of acetyl CoA, in addition to silencing of the molecular biosynthetic machinery of polyketides and terpenoids. To validate this hypothesis, the concentration of acetyl-CoA as crucial intermediate of Taxol and melanin biosynthesis by *A*. *terreus* in response to multiple subculturing and addition of sterilized *P*. *gracilior* leaves was determined. From the HPLC profiles (Fig. [4A](#Fig4){ref-type="fig"}), the intracellular levels of acetyl-CoA of *A*. *terreus* was reduced consecutively with the fungal subculturing. At 3^rd^, 5^th^ and 7^th^ subcultural generation of *A*. *terreus*, the concentration of acetyl-CoA was 620.2, 515 and 450.9 ppm, respectively, comparing to the original culture of *A*. *terreus* (870.1 ppm). Thus, by the 7^th^ and 9^th^ subculture of *A*. *terreus*, the intracellular concentration of acetyl-CoA was reduced by 2 and 3 folds, respectively, comparing to the original culture. The gradual decreasing on the acetyl-CoA levels of *A*. *terreus* with the multiple subculturing are highly correlated with the attenuation on their Taxol yield.Figure 4Concentration of *A*. *terreus* Acetyl-CoA in response to subculturing and amendment with various concentrations of *P*. *gracilior* leaves. The cultures of *A*. *terreus* were incubated under standard conditions, then, the fungal biomass was collected, and the intracellular acetyl-CoA were determined. Concentrations of *A*. *terreus* acetyl-CoA in response to subculturing (**A**). The 7^th^ subculture of A. terreus was grown for 10 days on standard conditions, then the cultures were amended with *P*. *gracilior* leaves (zero-5% w/v), continue incubation under standard conditions till 20 days, then Acetyl-CoA was extracted and quantified (**B**). The impact of addition of *P*. *gracilior* leaves on restoring the biosynthetic potency of acetyl-CoA by *A*. *terreus* has been evaluated. Interestingly, the biosynthetic potency of acetyl-CoA has been resumed completely upon addition of 3% surface sterilized leaves of *P*. *gracilior* (Fig. [4C,D](#Fig4){ref-type="fig"}). Upon addition of 3% *P*. *gracilior* leaves, the intracellular levels of the 7^th^ subculture of *A*. *terreus* acetyl-CoA was increased by about 2.8 and 5.4 folds regarding to the original culture and 7^th^ culture without *P*. *gracilior* leaves, respectively. Negative controls of *P*. *gracilior* leaves without fungal inocula has been used to baseline the calculated concentrations of acetyl-CoA of samples. Practically, the resuming on intracellular levels of acetyl-CoA was typically linked to the restoring of Taxol yield by *A*. *terreus* in response to subculturing and addition of *P*. *gracilior* leaves, suggesting the influx of acetyl-CoA is the profound factor limiting the anabolism of acetyl-CoA dependent metabolites especially Taxol. Protein profiling of *A*. *terreus* in response to subculturing and addition of *P*. *gracilior* leaves {#Sec19} ------------------------------------------------------------------------------------------------------- The protein profile of *A*. *terreus* in response to frequent subculturing and addition of *P*. *gracilior* surface sterilized leaves was evaluated using SDS-PAGE. Equal amounts (25 µg/mL) of the isolated intracellular proteins were loaded to gel wells, after gel running, CBB staining, the protein banding patterns were visually inspected, and the intensity of emerged protein bands were analyzed by GelQuant software normalizing to the original culture of *A*. *terreus*. Negative control of the surface sterilized leaves of *P*. *gracilior* without *A*. *terreus* were used, under the same conditions. From the protein banding pattern (Fig. [S2](#MOESM1){ref-type="media"}), substantial changes in the protein profiling were observed with the subsequent culturing of the fungus as well as with the supplementation of *P*. *gracilior* leaves. Visually, the intensity of proteins especially of 74.4, 68.2, 37.9, 24.1 and 22.2 kDa were substantially decreased with the frequent subculturing, connecting to decreasing on Taxol yield, regarding to the original fugal culture (Fig. [S2](#MOESM1){ref-type="media"}). The intensity of proteineous components of 74.4 and 22.2 kDa were reduced by \~ 50% from the 3^rd^ to 10^th^ subculture. While intensities of the protein bands of 68.2, 37.1 and 24 kDa were reduced exponentially with the fungal subculturing, retaining about 15--20% of their original intensities. Since the expression of these proteins were dramatically attenuated with the fungal subculturing, it could be anticipated that, these proteins are associated with the biosynthetic metabolic of Taxol. Fascinatingly, the intensities of proteins 74.4 kDa, 68.2 kDa and 37.1 kDa were completely restored corresponding to proteins from of the original culture that correlated with the restoring of Taxol biosynthesis, raising the hypothesis that these proteins might be implemented on Taxol biosynthesis. While the protein of 24.1 and 22.2 kDa, that had been exponentially decreased with the fungal subculturing, did not affected by incorporation of *P*. *gracilior* leaves negating the implication of these proteins on Taxol biosynthesis. Thus, correlation of restoring the Taxol biosynthesis with the re-expression of the attenuated proteins (74.4, 68.2 and 37.1 kDa), strongly evidenced the implication of these proteins on Taxol biosynthesis. Particularly, these proteins are generally corresponding to the predicted protein masses implemented in Taxol biosynthesis. For more validation of these results, the restored protein compartments (74.4, 68.2 and 37.1 kDa) were sequence and identified. Proteomic analysis of *A*. *terreus* in response to subculturing and addition of *P*. *gracilior* leaves {#Sec20} -------------------------------------------------------------------------------------------------------- From the SDS-PAGE analysis, the expression of proteins of 74.4, 68.2 and 37.1 kDa were the most sensitive proteins for downregulation with fungal subculturing and restoration with addition of *P*. *gracilior* leaves, that practically correlated with the attenuation and resuming of Taxol productivity by *A*. *terreus* (Fig. [5A](#Fig5){ref-type="fig"}). Thus, these proteins were excised, digested with trypsin, sequenced and identified by MALDI-TOF/TOF. For annotation, the recovered peptides were searched in the proteome of *A*. *terreus* and in the NCBI non-redundant protein database. The protein band of 74.4 kDa displayed a higher similarity to ribosome biogenesis protein YTM (Fig. [5B](#Fig5){ref-type="fig"}), with score identity 58% and expected value 0.056 and matches 11. From the conserved domain analysis, this protein had two domains the major one belongs to WD40 superfamily (390 amino acid) and minor one belongs to NLE superfamily (90 amino acids) (Fig. [5C](#Fig5){ref-type="fig"}). The recovered peptide sequences were non-redundantly blast searched on the protein database of NCBI, displaying two biological functions, ribosome biogenesis and microtubule-associated proteins (Fig. [5D](#Fig5){ref-type="fig"}). The ribosome biogenesis peptide sequence from *A*. *terreus* has been deposited into database with UniProt SPIN200017187. However, with the blast searching of recovered peptide sequences within the proteome of *A*. *terreus*, the results displayed a similarity with proteins of multiple biological functions including ribosome biogenesis, chromatin assembly factor, pre-mRNA splicing factor, periodic tryptophan, WD-repeat protein, rRNA processing proteins, in addition to various conserved hypothetical proteins (Fig. [5E](#Fig5){ref-type="fig"}).Figure 5Proteomic analysis of *A*. *terreus* in response to subculturing and addition of *P*. *gracilior* leaves. (**A**) SDS-PAGE profile of *A*. *terreus*, Lane 1 is the Ladder, lane 2 is the original culture, lane 3 is 3^rd^ subculture, lane 4 is the 7^th^ subculture, lane 5 is the 7^th^ subculture with 0.5% P. gracilior leaves, lane 6 is 7^th^ subculture with 1.0% *P*. *gracilior* leaves, 7^th^ subculture with 3% *P*. *gracilior*, lane 8 7^th^ subculture with 5% *P*. *gracilior* leaves. (**B**) Protein sequence of the over-induced protein bands in response to 3% *P*. *gracilior* with molecular mass 74.4 kDa leaves. (**C**) Graphical abstract of the conserved domains of the target peptide from the database. (**D**) BLAST pairwise alignment of the target sequence with non-redundant proteins. (**E**) BLAST pairwise alignment of the target sequence with the proteome of *A*. *terreus* with MEGA7 software portal package. Discussion {#Sec21} ========== Taxol is one of the most clinically valuable worldwide natural product of anticancer activity, since its discovery from the bark of Pacific yew tree, and FDA approval in 1985. Since the discovery of Taxol producing endophytic fungi^[@CR40]^, a plethora of researches reporting the same biosynthetic potency by various fungal endophytes as reviewed by our studies^[@CR37]^. However, the implementation of fungal endophytes for industrial production of Taxol has not seen the light till date, crucially due to the bewildering loss of Taxol productivity with the fungal subculturing and storage^[@CR14],[@CR15]^. *A*. *terreus*, an endophyte of *P*. *gracilior*, has been identified as promising Taxol producer, among the recovered fungal endophyte^[@CR7]^, however, a strong attenuation for Taxol productivity with fungal subculturing and storage was reported. Intriguingly, the biosynthetic potency of Taxol by *A*. *terreus* has been restored completely upon addition of surface sterilized leaves of *P*. *gracilior*, in contrary to the autoclaved parts or methanolic extract of *P*. *gracilior* that have no effect on fungal Taxol biosynthesis^[@CR7]^. So, re-triggering the biosynthetic genes of Taxol by *A*. *terreus* could be due to the intimate physical interaction of endogenous endophytes of *P*. *gracilior* with the fungus. Thus, the objective of this work was to explore the molecular expression of Taxol biosynthetic genes, connected with attenuation and restoring of Taxol yield in response fungal subculturing and storage. The stability of Taxol productivity by *A*. *terreus* through ten successive cultural generations was investigated. The productivity of Taxol was sequentially decreased with successive subculturing, in parallel to a visual fading to the conidial pigmentation. By the 7^th^ subculture of *A*. *terreus*, the yield of Taxol and mycelial melanin concentration were decreased by \>4--5 folds comparing to the original culture. This attenuation of fungal Taxol yield could be due to downregulation/silencing of Taxol encoding genes in axenic culture of *A*. *terreus* with frequent subculturing, due to lack of elicitors from the host plant or its microbiome^[@CR14],[@CR15]^. Consistently, successive subculturing of *Taxomyces andreanae*^[@CR13]^, *Tubercularia* sp^[@CR16]^ and *Periconia* sp^[@CR41]^ strongly attenuated their Taxol productivity. Plethora of endophytic fungi growing as monoaxenic cultures under standard conditions^[@CR42]^ losses their target metabolites biosynthetic potency, since their encoding genes are clustered on the fungal genome, and due to dilution/absence of inducing signals, their overall biosynthetic machinery were attenuated^[@CR3],[@CR43],[@CR44]^. Similarly, attenuation of secondary metabolites productivity of endophytic fungi with the multiple subculturing and storage were frequently reported^[@CR45],[@CR46]^ that could be due to lack of elicitors carried-over from plant and surrounding microenvironment. Since the fungal endophytes are normally exist within a niche of diverse communities of microbiome that contain a multitude of variegated interspecies crosstalk, so such multiplexed interaction are indispensable for inducing the plethora of natural products^[@CR47],[@CR48]^. As well as, the metabolites by *in vitro* growing fungal endophytes are correlated to the host plant physiology, defense and resistance to biotic and abiotic stresses^[@CR15]^. The yield of Taxol has been correlated with the activity and molecular expression of the rate-limiting enzymes of Taxol biosynthesis "GGPPS, TDS, DBAT and BAPT". The activities of TDS, DBAT and BAPT by *A*. *terreus* were strongly reduced with the fungal subculturing that might be due to the reduction on acetyl-CoA influx and/or autonomous silencing of expression of the genes encoding these enzymes or their transcriptional factors. Attenuation of Taxol productivity was matched to the fading of melanin pigmentation that could be due to decreasing of acetyl-CoA influx which is the main precursors of Taxol and melanin biosynthetic pathways. Supporting to this hypothesis, production of secondary metabolites by *A*. *flavus*, *A*. *parasiticus* and *A*. *nidulans* "endophytes of maize" are dependent on acetyl-CoA derived from oxidation of fatty acids in the kernel^[@CR49]^ and supplementation of oleic acid increases their productivity to sterigmatocystine and aflatoxin *via* inducing the biogenesis of fungal peroxisome^[@CR50]^. So, the concentration of conidial/mycelial melanin was reduced concomitantly with the fungal subculturing. Taken together, the reduction of both Taxol and melanin yield was strongly correlated, that might be due to reduction on acetyl-CoA influx with the fungal subculturing. This assumption was validated from the intracellular concentration of acetyl-CoA for the fungal subcultures. Intriguingly, under the standard conditions, the yield of intracellular acetyl-CoA was profoundly reduced with fungal subculturing. Thus, the suppression of Taxol and melanin biosynthesis with the fungal subcultures could be due to the loss of influx of acetyl-CoA. The influence of *P*. *gracilior* leaves on restoring the Taxol biosynthetic machinery of *A*. *terreus* has been evaluated. Remarkably, the machinery of Taxol biosynthesis by the 7^th^ *A*. *terreus* subculture has been completely restored with the implementation of *P*. *gracilior* surface sterilized leaves. Parallel to increasing on Taxol yield, the molecular expression of rate-limiting enzymes of Taxol biosynthesis have been completely restored. Thus, the downregulated gene expression of Taxol biosynthesis of *A*. *terreus* with subculturing, could be induced with signals from plants tissues or from surrounding microbiome. The activation of gene cluster encoding the fungal secondary metabolites could be due to the signals from plants such as homoserine and asparagine, similarly to activation of virulence genes of *Nectria hematococca*^[@CR51]^. Consistently, the expression of gene cluster of lolitrem in *Neotyphodium lolii*, endophyte of Ryegrass, had been dramatically increased *in planta* approving the presence of intrinsic gene inducing signals^[@CR52]^. Coincidently, needle extracts of *Taxus* sp significantly increased the Taxol yield by various endophytes^[@CR53]^. Elicitation of Taxol biosynthesis of *Paraconiothyrium* with the extracts of *Taxus* plant has been reported due to the presence of salicylic and benzoic acids, as well as, *via* cocultivation with *Alternaria* and *Phomopsis*^[@CR54]^. So, fungal metabolism can be greatly affected by the metabolites of other fungi, fungal-fungal interaction, with diffusion of some metabolites from *P*. *gracilior* mycobiome, could be the signals for expression of Taxol biosynthetic gene cluster of *A*. *terreus*. Triggering the expression of Taxol biosynthetic genes of endophytic fungi upon cultivation with non-Taxol endophyte producers have been studied extensively. The yield of Taxol by *F*. *mairei* was reported to be increased by about 38-fold upon cocultivation with *T*. *chinensis*^[@CR41]^. Similar scenario for induction of expression of Taxol encoding genes was reported to *A*. *proliferans* upon cocultivation with *Streptomyces olivaceoviridis*^[@CR55]^. The intimate physical fungal-actinomycetes interaction leads to activation of fungal secondary metabolites encoding genes "polyketide synthase"^[@CR48],[@CR56]^. From the SDS-PAGE profile, the expression of proteins of 74.4, 68.2 and 37.1 kDa were the most sensitive for downregulation with fungal subculturing and restoration by addition of *P*. *gracilior* leaves, that practically correlated with attenuation and resuming of *A*. *terreus* Taxol productivity. The protein band of 74.4 kDa was excised and their identified by MALDI-TOF/TOF proteomic analysis. This peptide sequence displayed 58% similarity with the ribosome biogenesis protein YTM as the result of searching with proteome of *A*. *terreus*. This protein belongs to WD40 protein superfamily^[@CR57]^. This protein had two domains, the major one belongs to WD40 superfamily (390 amino acid) and minor one belongs to NLE superfamily (90 amino acids). The domain WD40 in eukaryotes have multiple functions as regulatory modules of signal transduction, pre-mRNA processing and cytoskeleton assembly, protein-protein and protein-DNA interaction platform (Xu and Min, 2011). WD40 domain has conserved serine/glycine--histidine (SH) dipeptides at *N*-terminus and tryptophan--aspartate (WD) dipeptides at *C*-terminal motif  ^[@CR58]^. With non-redundant blast search on the protein database, the recovered peptide sequences displaying two biological functions, ribosome biogenesis and microtubule-associated proteins. However, with the blast search of the recovered peptides within the proteome of *A*. *terreus*, the results displayed a similarity with proteins of multiple biological functions including ribosome biogenesis, chromatin assembly factor, pre-mRNA splicing factor, periodic tryptophan, WD-repeat protein, rRNA processing proteins, in addition to conserved hypothetical proteins^[@CR59]^. In conclusion, attenuation/downregulation of the biosynthetic machinery of Taxol by *A*. *terreus* with the storage and subsequent subculturing are the main limitation for fungal usage as industrial platform. From the metabolic analysis, the attenuation of Taxol yield with the fungal multiple-culturing was assessed to be due to the reduction on main influx of acetyl-CoA, as revealed from the Taxol and melanin yields or due to silencing of ribosome biogenesis proteins that belong to WD40 protein superfamily, that had indirect effect on triggering the Taxol biosynthetic machinery of *A*. *terreus*. It could be deduced, the restoration of Taxol biosynthetic machinery by *A*. *terreus* is associated with resuming of expression of ribosome biogenesis protein YTM, due to the cross communication, interspecies crosstalk, signals production from the microbiome of *P*. *gracilior*. Thus, further ongoing studies to evaluate the overexpression of ribosome biogenesis protein on *A*. *terreus* to obtain a genetically engineered *A*. *terreus* with metabolic stability for sustainable Taxol production, have been raised. Supplementary information ========================= {#Sec22} Supplementary Data **Publisher's note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information ========================= **Supplementary information** accompanies this paper at 10.1038/s41598-019-47816-y. We appreciate the financial support from the Academy of Scientific Research and Technology (ASRT), Egypt to ASEA. A.S.A.E. designed the research plane. N.Z.M., and M.A.Y. performed the experiment. S.S., L.S. and A.A.S. contributing in data analysis. G.S.A., A.S.A.E. and M.Z.S. wrote and edit the manuscript. All authors have read and approved the final manuscript. The authors declare no competing interests.

Background ========== Interleukin-6 (IL-6) is a pleiotropic cytokine involved in several brain diseases as a detrimental factor playing a causal or exacerbating role in neuroinflammation and neurodegeneration \[[@B1]-[@B7]\]. Elevated levels of IL-6 are typical for brains from animal models or humans suffering from multiple sclerosis, Alzheimer\'s disease, Parkinson\'s disease, lethal sepsis, meningitis and stroke \[[@B8]-[@B12]\]. Additionally, long-term exposure of neurons or astrocytes to IL-6 as well as over-activation of IL-6 signaling by IL-6/sIL-6R fusion protein lead to a robust induction of neuroinflammatory response and to neuronal death \[[@B6],[@B13]-[@B16]\]. Therefore, suppression of IL-6 signaling or of IL-6 expression itself is believed to represent a powerful strategy for the treatment or prevention of neuroinflammation and subsequent neurodegeneration. This is supported by diminished neuroinflammation induced by spinal cord injury after infusion of a monoclonal antibody against IL-6 receptor \[[@B17]\]. Furthermore, the potency of drugs to inhibit IL-6 expression *in vitro*and *in vivo*correlates with their anti-neuroinflammatory and neuroprotective properties \[[@B18]-[@B22]\]. Astrocytes, the main glial cell type of the brain, respond in general to multiple kinds of acute and chronic brain insults with a reaction known as astrogliosis \[[@B23]\]. This reactive astrogliosis involves morphological, structural and biochemical features including thickened cellular processes, increased expression of glial fibrillary protein and the induction of pro-inflammatory cytokines including IL-6 \[[@B24],[@B25]\]. Different types of signaling molecules are able to trigger the astrocytic IL-6 mRNA expression via distinct intracellular signaling pathways \[[@B26]\]. For example, lipopolysaccharide (LPS) activates the IL-1 receptor-associated kinase (IRAK)-dependent pathway including IκB kinase and nuclear factor κB (NF-κB) \[[@B27]\]. Another potent group of IL-6 inducers are cytokines such as tumor necrosis factor α, interleukin-1β, oncostatin M (OSM) and leukaemia inhibitory factor (LIF) \[[@B28]-[@B30]\]. Interestingly, OSM and LIF belong together with IL-6 to the same cytokine family. These IL-6-type cytokines are characterized by using of glycoprotein gp130 to induce gene expression via JAK/STAT (Janus kinase/signal transducer and activator of transcription) and MAPK (mitogen-activated protein kinase) cascades in a NF-κB-dependent manner \[[@B31],[@B32]\]. Thus, blocking of such pathological IL-6-driven gene expression by low molecular weight inhibitors provides a possible strategy for targeting the onset or further propagation of astrogliosis and, subsequently, secondary neuronal cell death. In the present study, the time- and dose-dependent stimulation of IL-6 expression by OSM was characterized in human U343 glioma cells. Subsequently, our compound libraries were screened for inhibitory effects on OSM-induced IL-6 expression. We identified bioactive compounds belonging to the chemical class of heteroarylketones (HAK). These HAK compounds were able to suppress the LPS-induced IL-6 expression in primary mouse and rat astrocytes as well as in an acute septic shock mouse model *in vivo*. Finally, the underlying molecular mechanism of HAK compounds interfering with key signaling molecules of OSM-induced signal transduction cascade was analyzed. We demonstrate a selective suppression by HAK compounds of the OSM-mediated phosphorylation of STAT3 at serine 727, which affects STAT3 binding to the NF-κB subunit p65. Methods ======= Primary cultures of murine astrocytes ------------------------------------- According to Löffler \[[@B33]\], astrocyte-rich primary cell cultures were started with brains of newborn mice and rats and were maintained in Dulbecco\'s modified Eagle\'s medium (DMEM) for 33 days at 37°C in a humified atmosphere with 95% air/5% CO~2~. Cell culture ------------ The human glioma cell line U343 \[[@B34]\] was maintained in DMEM containing 10% fetal bovine serum and 60 μg/ml gentamycin (Invitrogen, Darmstadt, Germany) at 37°C in a 10% CO~2~atmosphere. Lipopolysaccharide (LPS)-induced acute septic shock model --------------------------------------------------------- A total number of 20 C57/B6 mice at the age of 5 months with an initial body weight of 22 g to 30 g was analyzed in this study. Mice were randomly assigned to 2 groups of control mice (injection of saline or LPS; n = 6 each) and 1 group of LPS-treated mice co-injected with compound HAK-2 (n = 8). Animals were housed at the Experimental Animal Core Facility of the University of Leipzig. The study was approved by the Regierungspräsidium Leipzig, License \# TVV 28/07 on November 14, 2007. To induce septic shock and acute inflammation, 14 mice were injected intraperitoneally (i.p.) with 1 mg LPS (Serotyp 055:B5)/kg body weight. Saline injection i.p. was used as control treatment (n = 6). Compound HAK-2 was co-injected i.p. to 8 LPS-treated mice at a concentration of 10 mg/kg body weight. Two hours post treatment mice were sacrificed by CO~2~inhalation and tissue samples from hippocampus and cortex as well as plasma were prepared as described later. Mouse brain samples for qRT-PCR were prepared from 6 - 8 animals per group. After removal of the brain, the tissue samples were flushed shortly with ice cold saline and placed briefly on filter paper. Cortex and hippocampus were prepared, weighted and snap frozen in liquid nitrogen. Samples were stored at -80°C until RNA isolation. Frozen tissue samples were homogenized in RNA lysis buffer (Macherey and Nagel, Düren, Germany) using Precellys ceramic beads (diameter of 1.4 mm; Peqlab, Erlangen, Germany). At the end of the experiment blood samples were collected in ice-cooled tubes and centrifuged within the next 20 min (10 min, 1500 × g, 4°C). Plasma samples were aliquoted into fractions of 50 μl, shock frozen and stored at -80°C until analysis. Quantitative real-time PCR -------------------------- Total RNA of cultured cells and homogenized mouse brain tissue samples was isolated using RNA isolation kit \"NucleoSpin RNA II\" from Macherey and Nagel. RNA quantity was measured by spectrophotometrical quantification (NanoDrop 2000, Peqlab, Erlangen, Germany). Total RNA (0.5 μg) was transcribed to cDNA using Superscript III and oligo-dT primer (Invitrogen, Darmstadt, Germany). Quantitative real-time PCR was performed with QuantiFast SYBR Green PCR Master mix (Qiagen, Hilden, Germany) using Rotorgene 3000 system (Corbett, Sydney, Australia). Gene expression was normalized to the expression of three reference genes glyceraldehyde 3-phosphate dehydrogenase (GAPDH), glucose-6-phosphate dehydrogenase (G6PDH) and hypoxanthine-guanine phosphoribosyltransferase (HPRT). Primers for PREP (5\'-TGAGCAGTGTCCCATCAGAG-3\'; 5\'- CATCTTCGCTGAACGCATAA-3\') and IL-6 (5\'-AAAGGACGTCACATTGCA-3\'; 5\'-GCCTCAGACATCTCCAGTCC-3\') were designed using the Primer3 Software (<http://frodo.wi.mit.edu/>). Data were analyzed by Rotorgene 3000 Software version 6.0 by comparative quantitation. The appropriate size of PCR products was confirmed by gel-electrophoresis stained with ethidium bromide. ELISA ----- Measurements of IL-6 in conditioned medium and murine plasma samples were done by means of human- or murine-specific IL-6 Cytoset kits (Invitrogen, Darmstadt, Germany) according to manufacturer\'s protocols. Preparation of whole-cell extracts for western blotting and immunoprecipitation ------------------------------------------------------------------------------- Human U343 cells were lysed with cell lysis buffer (Invitrogen, Darmstadt, Germany) supplemented with 0.2 mg/ml sodium orthovanadate (Sigma, Taufkirchen, Germany), protease inhibitor mix complete mini (Roche, Mannheim, Germany) and 1 mM AEBSF (Sigma, Taufkirchen, Germany) for 30 min on ice. Lysates were centrifuged for 15 min at 15,000 × g and 4°C. Protein content in supernatants was quantified by Bradford assay (Sigma, Taufkirchen, Germany) according to the manufacturer\'s protocol and stored at -20°C. Western blotting ---------------- Western blotting was performed with 30 μg protein of whole-cell extracts, mixed with 4 x SDS sample loading buffer (Invitrogen, Darmstadt, Germany) and denatured for 10 min at 85°C. Cell extracts separated by 4 - 12% Novex Bis-Tris Mini Gel system (Invitrogen, Darmstadt, Germany) were transferred to Roti-NC nitrocellulose membranes (Roth, Karlsruhe, Germany). Membranes were probed with primary antibodies against STAT3 (\#9132) and P-STAT3^S727^(\#9134) from Cell Signaling (Frankfurt/M., Germany) as well as with anti-αtubulin (Sigma, Taufkirchen, Germany) to confirm equal loading and blotting of protein samples. Proteins were visualized using HRP-conjugated secondary antibodies (1:4000, cell signaling, Frankfurt/M., Germany) and the SuperSignal West Pico system (Thermo Fisher Scientific, Karlsruhe, Germany). Small interfering RNA (siRNA) ----------------------------- Human U343 cells (0.25 × 10^6^) were seeded in 24-well plates (Greiner, Frickenhausen, Germany) and transfected immediately with 2.5 μl A. dest. (mock), 100 μM non-target control (NTC) or PREP-specific siRNA ON-TARGETplus SMARTpool (\#L-006006-00, Dharmacon, Schwerte, Germany) using DharmaFECT 1 siRNA transfection reagent (Dharmacon, Schwerte, Germany) according to the manufacturer\'s protocol. After 48 h adherent cells were transfected a second time under identical conditions for further 24 h and subsequently stimulated with OSM (100 ng/ml) for additional 6 h. For IL-6 specific ELISA 5 - 40 μl of conditioned media were utilized. To analyze IL-6 mRNA expression by qRT-PCR, total RNA was isolated and reversely transcribed as described above. Immunocytochemistry ------------------- Human U343 cells (1 × 10^5^cells/well) were grown on cover slips in 24-well plates (Greiner, Frickenhausen, Germany) for 24 h. After the time of treatment indicated cells were fixed in ice-cold methanol for 10 min on ice, and then incubated with rabbit anti-phospho-STAT3 antibody (\#9134, Cell Signaling, Frankfurt/M., Germany) overnight at 8°C. Subsequently, cells were incubated with goat anti-rabbit IgG Cy2-conjugated secondary antibody (Dianova, Hamburg, Germany) at room temperature for 45 min. Finally, cover slips were mounted on microscope slides and approximately 250 cells/sample were evaluated densitometrically by fluorescence microscopy (Zeiss, Jena, Germany) and MetaMorph Image Analysis Software (Universal Imaging Corporation, USA). Immunoprecipitation ------------------- Cell lysates from 4 × 10^6^U343 cells/sample were obtained as described above. From each sample 200 μg of total protein were immunoprecipitated with 2 μg rabbit anti-p65 antibody (Santa Cruz Biotechnology; Heidelberg, Germany) or A. dest. (control) overnight at 4°C. Immunoprecipitated samples were incubated with 20 μl Dynabeads Protein G (Thermo Fisher Scientific, Karlsruhe, Germany) for 1 h at 4°C. Beads were washed 3 times with 1 ml PBS. Preparation of samples for SDS-PAGE analysis was done by means of dilution with SDS-PAGE sample buffer (Invitrogen, Darmstadt, Germany), followed by denaturation at 90°C for 10 min. SDS-PAGE analysis were performed as described before. PREP enzymatic activity assay ----------------------------- The activity of human recombinant prolyl endopeptidase (PREP) was determined photometrically using the substrate Z-Gly-Pro-pNA (Bachem, Weil am Rhein, Germany) usually at a concentration of 150 μM. As assay buffer 50 mM HEPES buffer pH 7.6, containing 200 mM NaCl, 1 mM EDTA, and 1 mM dithiothreitol was used. Release of pNA was monitored continuously at 405 nm for 15 min at 30°C in a 96-well plate reader (Spectrafluor, Tecan, Crailsheim, Germany). PREP activity was calculated from the slope of the time product curve with the help of a pNA standard \[[@B35]\]. For determination of Ki values three different substrate concentrations (62.5 μM; 125 μM and 250 μM) and seven different inhibitor concentrations were analyzed. Substrate concentrations were selected to be at the Km value of Z-Gly-Pro-pNA (125 μM) as well as half and twice the Km. Data were fitted by non-linear regression to the competitive inhibitor equation (Graphit 5.0, Erithacus software). Statistical analyses -------------------- Values are expressed as mean ± SD. Standard unpaired t test was used for analyses of statistically significance. Differences between treatments were considered significant when p \< 0.05. Results ======= Oncostatin M-mediated release of IL-6 in human U343 glioma cells ---------------------------------------------------------------- Beside myocytes and adipocytes, glial cells are representing the most prominent source of the cytokine IL-6 in mammals and play an important role in neuroinflammatory processes \[[@B23],[@B36]\]. Therefore, the human glioma cell line U343 was selected for screening of IL-6-lowering effects of our in-house compound libraries. Preceding experiments with a set of stimuli known from literature to induce IL-6 expression in astroytes \[[@B30],[@B37]\], identified Oncostatin M (OSM) as a robust inductor of IL-6 protein release in our experimental setup (data not shown). The dose- and time-dependent stimulation of IL-6 expression by OSM in U343 cells is characterized in figure 1. To analyze dose dependence of IL-6 release, U343 cells were treated for 24 hours with various concentrations of OSM followed by measurement of IL-6 protein concentrations in the conditioned medium by a specific ELISA. OSM induced the release of IL-6 in a dose-dependent manner with an EC~50~of 70.5 ± 22.68 ng/ml (Figure [1A](#F1){ref-type="fig"}). Therefore, all further investigations were performed with 100 ng/ml OSM. The highest dose of 200 ng/ml OSM led to 30-fold increase of IL-6 accumulation in the conditioned media in comparison to vehicle-treated cells. To analyze the time-course of OSM-induced IL-6 expression, U343 cells were incubated with OSM for different periods of time as indicated in figure [1B](#F1){ref-type="fig"}. Amounts of IL-6 mRNA and protein were subsequently quantified by qRT-PCR and by ELISA, respectively. Time course studies revealed that IL-6 mRNA displays a biphasic induction pattern with peak synthesis at 1 h (6.5-fold) and 16 h (5.5-fold) post stimulation. Significant induction of IL-6 protein was detected in the conditioned medium as early as 3 h post stimulation and reaching a maximum at 24 h. In contrast to the mRNA expression profile, IL-6 protein release did not show a biphasic pattern. {#F1} Identification of compounds reducing OSM-induced IL-6 release in human U343 glioma cells ---------------------------------------------------------------------------------------- Using the characterized cell culture model of OSM-induced IL-6 expression, our in-house compound libraries were screened for potent IL-6 expression inhibitors. Human U343 glioma cells were treated with 100 ng/ml OSM for 24 h. The analysis by IL-6 ELISA identified a set of structurally related compounds as potent inhibitors of IL-6 secretion (Table [1](#T1){ref-type="table"}). Interestingly, all bioactive compounds identified belong to the class of heteroarylketones (HAK) and differ from each other at residues R to R\'\' and P~1~, P~1~\' and P~2~, respectively (Figure [2](#F2){ref-type="fig"}). HAKs with proline in position P~1~are known from the literature as inhibitors of prolyl endopeptidase (PREP), a proline-specific serine protease \[[@B38]\]. Determination of the Ki values proved that compounds with proline in position P~1~are highly potent inhibitors of PREP. The remaining compounds with a substituted moiety in that region showed very poor PREP inhibition (Table [1](#T1){ref-type="table"}). ###### Effects of HAK compounds on PREP inhibition and on OSM-stimulated IL-6 secretion from U343 cells Compound Z-P~2~-P~1~-P~1~\' Remaining IL-6 expression \[% of control\] PREP assay (Ki \[M\] ---------- -------------------- -------------------------------------------- ---------------------- HAK-1 Z-Phe-Pro-BTh **15.9 ± 1.87** **4.6 × 10^-9^** HAK-2 Z-Ala-Pro-BTh **18.7 ± 3.87** **3.8 × 10^-11^** HAK-3 Z-Phe-Leu-BTh **22.1 ±5.91** \> 1.0 × 10^-6^ HAK-4 Z-Phe-Ala-BTh **35.1 ± 6.75** \> 1.0 × 10^-5^ HAK-5 Z-Phe-Pro-Th **35.2 ± 12.92** **1.1 × 10^-10^** HAK-6 Z-Lys-Pro-BTh **38.8 ± 5.26** **4.1 × 10^-9^** HAK-7 Z-Asp-Pro-BTh **52.8 ± 3.91** **1.2 × 10^-8^** HAK-8 Z-Arg-Pro-BTh 96.8 ± 10.69 **3.2 × 10^-9^** HAK-9 Z-Phe-Phe-BTh 182.8 ± 25.41 **\> 1.0 × 10^-5^** IL-6 values are expressed as mean ± SD. IL-6 release of vehicle-treated cells was set to 100% (n = 2). Seven out of 9 HAK compounds suppressed significantly the OSM-induced IL-6 release. There was no intimate correlation between significant IL-6 reduction (**bold**) and the potency of HAK compounds to inhibit activity of recombinant prolyl endopeptidase (**bold**), presented as Ki values. All amino acids were used in L-configuration. Z, benzyloxycarbonyl, BTh, benzothiazole, Th, thiazole. {#F2} While PREP inhibitor compounds HAK-1, HAK-2, HAK-5, HAK-6 and HAK-7 significantly reduced OSM-induced IL-6 secretion, there was no intimate correlation between the extent of PREP inhibition and the potency to suppress the IL-6 expression for different HAK compounds (Table [1](#T1){ref-type="table"}). For example, compound HAK-8 is a potent PREP inhibitor but does not reduce OSM-stimulated IL-6 secretion (Table [1](#T1){ref-type="table"}). On the other hand, compounds HAK-3 and HAK-4 are poor PREP inhibitors but significantly reduced OSM-stimulated IL-6 secretion (Table [1](#T1){ref-type="table"}). This indicates that HAKs reduce IL-6 secretion independent from their PREP-inhibiting activity. In contrast to PREP inhibition, the proline residue at position P~1~can be replaced by other amino acid residues like alanine or leucine without loss the bioactivity to reduce IL-6 expression. To clarify the role of PREP in the regulation of IL-6 expression, PREP was knocked down by siRNA technique in U343 cells. The remaining mRNA expression level of PREP was lower than 15% in comparison to mock and to non-target control (NTC) siRNA sample (Figure [3A](#F3){ref-type="fig"}). Interestingly, 6 h after onset of OSM stimulation, a 2-fold higher PREP mRNA level was obtained in non OSM-treated cells compared to OSM-stimulated NTC and mock samples. The biological basis of PREP up-regulation under these experimental conditions is not known, but could involve similar mechanisms contributing to induction of PREP in glial cells in experimental animals \[[@B39]\]. Compound HAK-2 did not have an effect on this phenomenon, allowing to investigate the effect of siRNA-mediated PREP knock down on OSM-stimulated IL-6 expression. Contrary to compound HAK-2, neither IL-6 mRNA level nor IL-6 protein level in the conditioned medium was significantly reduced after specific knock down of PREP (Figures [3B](#F3){ref-type="fig"} and [3C](#F3){ref-type="fig"}). This result strongly indicates that PREP is not involved in regulation of IL-6 expression by HAKs. Therefore, we conclude that HAKs exert their effects on IL-6 expression independent from PREP inhibition by modulating at least a second molecular target. {#F3} Effect of HAK compounds on OSM-induced IL-6 mRNA expression ----------------------------------------------------------- To reveal whether bioactivity of HAK compounds is based on suppression of IL-6 protein biosynthesis or on interference with IL-6 mRNA expression OSM-treated U343 cells were incubated with 20 μM of compound HAK-2 for different periods of time. Time course analyses revealed a strong inhibition of the OSM-induced IL-6 mRNA expression by compound HAK-2 (Figure [4](#F4){ref-type="fig"}). Notably, only the second peak in IL-6 mRNA synthesis at 6 h post stimulation was affected, whereas the first peak 1 h post stimulation was insensitive to HAK-2 treatment. Additional experiments demonstrated suppression of OSM-induced IL-6 expression in U343 cells by HAK compounds even after delayed onset of treatment 6 h after OSM stimulation (data not shown). Therefore, it is very likely that the relevant molecular target of HAK compounds is involved in the OSM-induced signal transduction process not earlier than 6 h after onset of the stimulation. {#F4} Furthermore, IL-6 mRNA decay experiments were performed with actinomycin D, a transcription arresting agent, to study whether the strong inhibition of IL-6 mRNA expression by HAK compounds was based on modified mRNA stability. No difference in mRNA stability was observed between treated and non-treated cells (data not shown), demonstrating that the HAK compound-mediated suppression of IL-6 mRNA is most likely due to inhibition of transcription rather than modified mRNA stability. Suppression of LPS-induced IL-6 release by HAK compounds in primary murine astrocytes ------------------------------------------------------------------------------------- To analyze whether inhibition of OSM-induced IL-6 expression is a cell line-specific effect or a common feature of HAK compounds and valid in general, primary murine astrocytes were treated with HAK compounds. In contrast to human U343 glioma cells, OSM treatment did not lead to an increased IL-6 expression in mouse and rat primary astrocytes (data not shown). However, LPS significantly induced IL-6 release into the conditioned medium of mouse and rat astrocyte cultures (Figure [5](#F5){ref-type="fig"}). {#F5} Primary murine astrocytes stimulated with 1 μg/ml LPS were co-treated with compounds HAK-2 and HAK-5 at a concentration of 20 μM each for 24 h. The observed effect of compounds HAK-2 and HAK-5 on LPS stimulation was similar to that of OSM-induced IL-6 expression in human U343 glioma cells (Figures [5A](#F5){ref-type="fig"} and [5B](#F5){ref-type="fig"}). In comparison to untreated samples LPS-induced IL-6 expression was reduced by 60% in mouse and 50% in rat astrocytes by both HAK compounds. Thus, suppression of both, LPS- and OSM-induced IL-6 expression in different cell types by structurally related compounds is another indication for strong potency of HAK compounds to target neuroinflammatory processes. Potent inhibition of IL-6 upregulation by compound HAK-2 *in vivo* ------------------------------------------------------------------ Based on the results obtained from primary murine astrocytes, we reasoned that HAK compounds might also suppress elevated IL-6 expression *in vivo*. Therefore, bioactivity of the compound HAK-2 was analyzed in the LPS-induced mouse septic shock model. In preparation of this *in vivo*study, selected HAK compounds were characterized in detail concerning brain bioavailability. It was demonstrated by quantitative analysis of mouse plasma and brain samples using LC-MS/MS that the compounds are bioavailable and able to pass the blood brain barrier (data not shown). For further *in vivo*investigations compound HAK-2 with a logBB of -0.22 was selected. Intraperitoneal injection of 1 mg/kg LPS into C57/B6 mice resulted in an acute elevation of IL-6 concentration in plasma, hippocampus and cortex 2 h post administration (Figures [6A-C](#F6){ref-type="fig"}). The plasma level of IL-6 protein was significantly reduced by 55% in mice treated with 5 mg/kg HAK-2 in parallel as compared to LPS alone (Figure [6C](#F6){ref-type="fig"}). Similar effects of HAK-2 treatment were observed for induced IL-6 mRNA in hippocampus (60%) and cortex (53%) (Figures [6A](#F6){ref-type="fig"} and [6B](#F6){ref-type="fig"}). These data clearly show the strong potency of HAK compounds to modify IL-6 expression *in vivo*. {#F6} Effect of HAK compounds on OSM-mediated phosphorylation of signal transducer and activator of transcription 3 (STAT3) and extracellular signal-regulated kinase 1 (Erk1) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------ The activation of the OSM signaling cascade resulting in the stimulation of IL-6 expression is known to involve intracellular phosphorylation events \[[@B31],[@B40],[@B41]\]. Therefore, effects of HAK compounds on the OSM-induced phosphorylation of signal transducer and activator of transcription 3 (STAT3) and extracellular signal-regulated kinase 1 (Erk1) were investigated. Since HAK compounds were shown to be bioactive 3 to 6 h post stimulation, U343 cells were incubated with OSM for 6 h. In contrast to non-stimulated control, OSM induced phosphorylation of Erk1 as well as STAT3 6 h post stimulation. Interestingly, HAK compounds suppressed STAT3 phosphorylation at serine 727 (pSTAT3^S727^) (Figure [7](#F7){ref-type="fig"}), but neither phosphorylation of pSTAT3^Y705^nor pErk1/2^T202/Y204^(data not shown). Compound HAK-8 showed a significantly lower effect on pSTAT3^S727^phosphorylation. This observation correlates well with the missing effect of compound HAK-8 on IL-6 expression as shown in Table [1](#T1){ref-type="table"}. HAK compound-specific suppression of OSM-induced pSTAT3^S727^was confirmed by immunocytochemistry (Figures [7B-C](#F7){ref-type="fig"}). U343 cells cultivated on cover slips were treated identically as described before. In figure [7A](#F7){ref-type="fig"} an example of HAK compound efficiency to suppress nuclear pSTAT3^S727^phosphorylation 6 h post OSM stimulation is shown. Nuclear localization was confirmed by DAPI staining (data not shown). In figure [7C](#F7){ref-type="fig"} densitometric results of at least 3 independent experiments are summarized. While compounds HAK1-7 significantly suppressed the OSM-mediated phosphorylation of pSTAT3^S727^, compound HAK-8 did not. Results from western blot analyses and immunocytochemistry are strongly correlating with each other. Thus, it appears that the IL-6 reducing bioactivity of HAK compounds is most likely based on suppression of STAT3 phosphorylation at serine 727. {#F7} STAT3 and NF-κB subunit p65 are forming a OSM-dependent complex which is sensitive to HAK compounds --------------------------------------------------------------------------------------------------- Normally, STAT3 is activated by phosphorylation at tyrosine 705, which induces dimerization, nuclear translocation and DNA binding. In contrast to other transcription factors like NF-κB, Creb and c/EBPb, STAT3 can not bind directly to the IL-6 promoter, because there are no STAT3 binding elements present. It is known from literature that several forms of interactions and cross talks between NF-κB and STAT3 exist. For instance, recent studies have shown a physical interaction between STAT3 and NF-κB \[[@B42],[@B43]\]. The importance of pSTAT3^S727^for protein interactions are under discussion \[[@B42],[@B44]\]. However, there is no information so far on the role of pSTAT3^S727^for the interaction with NF-κB. To characterize a possible OSM-induced and p^S727^-dependent complex formation between STAT3 and NF-κB, we performed co-immunoprecipitation of p65 followed by western blotting for STAT3. In this experiment, U343 glioma cells were stimulated with OSM for 6 h in the absence or presence of compound HAK-2. The immunoprecipitation analysis showed that STAT3 and p65 were co-immunoprecipitated in a OSM-dependent manner (Figure [8](#F8){ref-type="fig"}). Furthermore, treatment of human U343 glioma cells with compound HAK-2 led to a clear reduction of p65-mediated STAT3 co-immunoprecipitation. Moreover, these data confirm OSM- and phosphorylation-dependent complex formation between STAT3 and p65, which is sensitive to HAK compounds. {#F8} Discussion ========== Neuropathological situations with extended astroglial activation are associated with increased levels of pro-inflammatory mediators including IL-6. *In vitro*and *in vivo*studies have demonstrated that IL-6 plays a pivotal role in the initiation of neuroinflammatory cascades and in secondary neuronal cell death \[[@B6],[@B45]\]. Thus, prevention of the neuroinflammatory response to primary lesions has a neuroprotective potential. The present study was performed to identify new small molecular weight inhibitors acting on the pathway that results in IL-6 expression. For screening of our in-house compound libraries the human glioblastoma cell line U343 was used, because glioblastoma cell lines were shown to respond with increased IL-6 expression to different neuroinflammatory stimuli like LPS \[[@B27]\], Substance P \[[@B46]\], tumor necrosis factor α, interleukin-1β, leukemia-inhibitory factor and OSM \[[@B30]\]. Analysis of conditioned media revealed, that in our experimental setup only OSM treatment significantly induced the expression of IL-6 in human U343 glioma cells. This result is consistent with published data, showing that U343 cells express the OSM-receptor components LIFR and OSMRβ as well as the common signal transducer gp130 \[[@B47]\]. Furthermore, the OSM-mediated activation of signal components of the Jak/STAT- and MAPK-pathways was described for U343 and U373 glioma cells, respectively \[[@B32]\]. We observed a biphasic induction pattern of OSM-induced IL-6 mRNA expression, which was described earlier also for human U373 astroglioma cells \[[@B32]\]. The time-course is characterized by a first strong, rapid and transient IL-6 mRNA expression peak at 1 h followed by a second one at 6 h with a less strong, but prolonged induction. The same type of expression pattern was observed for tissue factor mRNA in OSM-treated smooth muscle cells \[[@B48]\]. Thus, biphasic induction seems to be an OSM-specific feature with general relevance for OSM action. All potent inhibitors of IL-6 secretion identified in the compound library screen (see Table [1](#T1){ref-type="table"}) belong to the chemical class of HAK and are structurally related to inhibitors of PREP \[[@B38]\]. This observation is in line with the hypothesis, that PREP is involved in regulation of intracellular protein transport and secretion \[[@B49]\]. However, there was no correlation between PREP siRNA- (knock-down) and pharmacological inhibition of PREP on one hand and the potency of these compounds to suppress the OSM-induced IL-6 expression on the other. Furthermore, our data on the temporal profile of IL-6 suppression suggest that the bioactivity of HAK compounds is most likely based on interference with IL-6 mRNA synthesis but not on disturbed intracellular transport and secretion mechanisms. Therefore, PREP can be excluded as the IL-6 relevant molecular target of HAKs and HAK compounds appear to interact with at least one further molecular target. Interestingly, only the second IL-6 mRNA peak was affected by HAKs indicating that the molecular target of HAK compounds is involved 3 to 6 h post OSM stimulation at earliest. Theoretically, the biological target of HAKs can pre-exist in untreated cells or be induced by OSM treatment and subsequently incorporated in signaling pathways. Notably, in an experiment analyzing the puromycin sensitivity of OSM-induced IL-6 mRNA expression, it was demonstrated that OSM induces the protein synthesis of signaling molecules essential for the second IL-6 mRNA expression peak \[[@B32]\]. Whereas puromycin completely abrogated the second IL-6 expression peak it showed no effect on the first OSM-induced IL-6 mRNA peak. This demonstrates a requirement for de novo protein synthesis exclusively for the second IL-6 expression peak of this biphasic response signaling. The relationship between HAK-mediated suppression of OSM-induced IL-6 release and the effect of HAK compounds exclusively on the second mRNA peak suggests that more than 75% of secreted IL-6 is based on the second phase of OSM-induced IL-6 mRNA expression. Thus, the mRNA induced in the first phase appears to have regulatory functions rather than acting as a template in protein synthesis. Such a regulatory role of mRNA molecules was recently described by Poliseno et. al. \[[@B50]\] showing that mRNA molecules from pseudogenes or long non-coding RNAs can act as competitive endogenous RNAs sequestering microRNA molecules. To elucidate whether the HAK-mediated suppression of OSM-induced IL-6 expression is cell line-specific or valid in general, experiments with primary murine astrocytes were performed. In contrast to human U343 glioblastoma OSM did not induce IL-6 expression in mouse and rat primary astrocytes. However, LPS, known to act as a powerful stimuli of cytokines \[[@B51]\], significantly increased IL-6 expression in primary murine astrocytes. Co-treatment with HAK compounds markedly suppressed levels of OSM-stimulated IL-6 expression in both rat and mouse astrocytes. These data demonstrate that the anti-inflammatory bioactivity of HAKs is not limited to a single OSM-based cell culture model but also valid for a series of pathophysiological conditions contributing to neuroinflammation and neurodegeneration. We were also interested to reveal whether HAK compounds are bioactive under inflammatory conditions *in vivo*. For this study, compound HAK-2 was selected based on its beneficial features concerning toxicity, bioavailability and blood brain barrier passage. In accordance with the data obtained from primary murine astrocytes, compound HAK-2 significantly suppressed LPS-induced IL-6 levels in brain- and plasma-derived samples from septic mice. This result strongly indicates the anti-inflammatory potency of HAK compounds *in vivo*for possible treatment of central nervous system diseases. To get more information about the underlying molecular mechanism of HAK bioactivity, the signal-transduction pathways involved in OSM-mediated IL-6 expression were dissected in more detail. Interestingly, LPS- and OSM-induced signal pathways are based on the same molecular mechanism such as STAT3 or NF-κB activation \[[@B31],[@B52]\], indicating that HAK compounds may target a common cellular event. Two major signaling cascades the JAK/STAT as well as the MAPK pathways are switched on by binding of OSM to the receptor heterodimers OSMR/gp130 or LIFR/gp130 \[[@B53]\]. Subsequent activation of signal tyrosine kinases of the JAK family leads to phosphorylation of pivotal signal molecules such as STAT3 and Erk1 and 2 respectively \[[@B31],[@B32]\]. The essential role of receptor subunits as well as of downstream signaling molecules as STAT3, Erk1 and p65 for OSM-triggered IL-6 expression in U343 cells was confirmed by siRNA-based knock-down experiments (data not shown). Furthermore, Erk1/2 and STAT3 were phosphorylated 6 h post OSM treatment, which was identified as the critical time point for the HAK bioactivity. Immunoblotting and immunofluorescence experiments revealed that neither OSM-induced pErk1/2^T202/Y204^phosphorylation nor pSTAT3^Y705^phosphorylation were modified by HAK compounds. However, HAK treatment led to a significant reduction of OSM-stimulated pSTAT3^S727^phosphorylation. Importantly, the HAK-based inhibition profiles for IL-6 expression and pSTAT3^S727^phosphorylation are strongly correlating with each other. Thus, suppression of OSM-induced phosphorylation of pSTAT3^S727^is most likely the relevant molecular mechanism of the HAK compound bioactivity to suppress IL-6 expression. In contrast to pSTAT3^Y705^, which is essential for dimerization, nuclear translocation and DNA binding \[[@B54],[@B55]\], the physiological role of pSTAT3^S727^is discussed controversially \[[@B56]-[@B59]\]. Depending on the specific promoter and/or the cellular context pSTAT3^S727^can influence transcriptional activity of target genes \[[@B60],[@B61]\]. However, in the case of the IL-6 promoter, where activated NF-κB binds directly to DNA, no cis-regulatory elements for STAT3 binding were identified so far \[[@B62]-[@B64]\]. Based on these observations, we hypothesize that pSTAT3^S727^may regulate IL-6 gene expression by an alternative pathway. It is known that STAT3 is complexed with transcription factors such as c-Jun, c-Fos, forkhead and endothelial cell-derived zinc finger protein, respectively \[[@B65]-[@B68]\]. Furthermore, it was shown that physical interaction of the STAT3 DNA-binding domain with the NF-κB subunit p65 led to a reduced promoter activity of inducible nitric oxide synthase gene \[[@B69]\]. Together, these findings strongly suggest that physical interaction between STAT3 and p65 may result in a functional coupling important for the STAT3-dependent regulation of p65 responsive genes. Indeed, we here demonstrated by co-immunoprecipitation that p65 and STAT3 interact with each other in an OSM-dependent manner. Noteworthy, the OSM-stimulated STAT3 and p65 complex formation is quite sensitive against treatment with HAK compounds. This supports our hypothesis and indicates for the first time a regulatory function for pSTAT3^S727^in OSM-triggered STAT3/NF-κB interaction. In summary, HAK compounds selectively target the expression of genes with promoters co-regulated by pSTAT3^S727^-dependent signaling. Based on this mechanism, kinases phosphorylating STAT3 at serine 727 such as MAPKs, mTOR, NLK and PKCδ may represent direct molecular targets of HAK compounds. Thus, further studies are required to identify the precise molecular mechanisms and the neuroinflammatory-related genes sensitive to HAK-treatment. This will enable the therapeutic development of HAK compounds for treatment of neurological diseases including Alzheimer\'s disease, multiple sclerosis, Parkinson\'s disease and traumatic brain injury. Conclusions =========== In the present study, we identified and characterized for the first time HAK compounds as potent inhibitors of OSM- and LPS-induced IL-6 expression *in vitro*and *in vivo*. The molecular trigger of HAK bioactivity is most-likely a selective suppression of STAT3 phosphorylation at serine 727. Pathological upregulation of astrocytic IL-6 expression is known to play a pivotal role in onset and progression of neurological diseases including Alzheimer\'s disease, multiple sclerosis, Parkinson\'s disease and traumatic brain injury. In conclusion, we propose that HAK compounds represent a new potent class of drugs for treatment or prevention of neuroinflammation and subsequent neurodegeneration. Abbreviations ============= **HAK:**heteroarylketone; **IL-6:**interleukin-6; **LPS:**lipopolysaccharide; **MAPK:**mitogen-activated protein kinase; **NF-κB:**nuclear factor kappa B; **OSM:**Oncostatin M; **PREP:**prolyl endopeptidase; **STAT3:**signal transducer and activator of transcription. Competing interests =================== IS, CE, AJN, KM, AK and AM are employees of Probiodrug AG. HUD serves as CSO of Probiodrug AG. Authors\' contributions ======================= IS initiated the project and wrote the paper. IS, AJN, CE, UZ, KM, AM and SR performed the experiments and analyzed the data. AK assisted with the design and interpretation of qRT-PCR experiments. AK, HUD and SR participated in study design and coordination as well as drafted and revised the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ The authors would like to thank Susann Wendler, Heiderose Mosdzen, Mercedes Scharfe and Dr. Livia Böhme for superior technical assistance. We are also indebted to Diane Meitzner for LC-MS/MS analysis and to Dr. David Ruiz-Carrillo for providing recombinant prolyl endopeptidase. We also wish to acknowledge the expert assistance by Dr. Torsten Hoffmann concerning PREP activity assay. This work was supported by a grant from the European Commission FP7 Programme of Health (NEUROPRO HEALTH-F2-2008-223077).

1. Introduction {#s0005} =============== All live cells have a stable transmembrane potential (TMP) voltage differential across the cell membrane when the cell is at rest. This is the result of the accumulated ion concentrations within the cell compared to that outside the membrane. Electrically active cells, such as neurons, muscles, and pancreatic beta cells, are called excitable because they can produce an action potential due to a short-lived rapid depolarization of the TMP before returning to the higher resting state. These cells achieve this by expressing fast-acting voltage gated ion channels that allow the very rapid exchange of ions across the cellular membrane. While calcium and chloride ions make distinct contributions, it is particularly the sodium (Na+) and potassium (K+) which contribute most to this ion concentration potential. Even cells that do not generate action potentials need to maintain a TMP in order to enable secondary active transport of metabolites. Non-excitable cells use slower membrane ion exchange pumps or transporters to move charges across the membrane, often in conjunction with the transport of a metabolite. Most of these transporters require a concentration gradient, a supply of energy, and are considerably slower than the fast-acting ion channels of the excitable cells. Alterations to membrane potential are carefully modulated by ion channels within the cell membrane to maintain homeostasis. Subjecting cells to pacing waveforms alters the TMP since it results in a manipulation of these charged ions. We offer evidence that electrically altering the TMP can then have profound effects on cellular physiology in terms of both metabolism and function [@bib1]. 2. Materials and methods {#s0010} ======================== 2.1. Cells {#s0015} ---------- Early passage human primary fibroblast cells (CRL-2703, ATCC; Manassas, VA) were subcultured in a 37 °C, high humidity incubator with 5% CO~2~ and maintained in Iscove\'s Modified Dulbecco\'s Medium (IMDM, Thermofisher Scientific; Waltham, MA) supplemented with 10% fetal bovine serum (FBS; ATCC), 25 mM HEPES (Thermofisher Scientific) and an antibiotic-antimycotic solution (ATCC). Cells were passaged at approximately 80% confluence and checked weekly or biweekly to be mycoplasma-free. Twenty-four hours prior to pacing, cells were seeded at a density of \~3.0×10^5^ cells per cm^2^ for attachment in designated pacing chambers. 2.2. Electrodes and pacing chambers {#s0020} ----------------------------------- Pacing chambers were T-25 cell culture flasks, 6-well culture plates, and 3-D printed plastic forms with electrodes attached with similar spacing. Non-reactive carbon rods (4.0 mm diameter, Frey Scientific; Appleton, WI) or platinum wires (0.5 mm diameter, WPI; Sarasota, FL) were inserted 4 cm apart. There were no apparent variations of results in experiments based on the material composition of the electrodes. The chambers were tilted so that the seeded cells attached and were localized close to the active electrode, defined as the polarity of the defined waveform when measured on an oscilloscope. The other electrode served as the 'non-active' reference electrode. 2.3. Pacing treatment {#s0025} --------------------- An external cardiac pacer (PACE Medical, Inc.; Waltham, MA) set at 5 V (V) magnitude and 1.8 ms (mSec) corresponding to an output of 10 mA (mA) at a rate of 1.7 Hz was used as the input signal to a Slave Stimulator (Model 71006, Rivertek Medical Systems; Minneapolis, MN). This circuit produced the actual monophasic anodal and cathodal, and biphasic (anodal followed by cathodal) pulsed square waveforms used for stimulation. The monophasic waveforms selected for these experiments were ±5.0 Vx1.8 ms. The biphasic waveform was defined as a +2.5 V×0.9 ms anodal pulse immediately followed by a −2.5 V×0.9 ms cathodal pulse. Selected waveforms were verified on the electrodes before each experiment using digital oscilloscopes with probes placed within the electrolyte near the active electrode. All pacing treatments were continuous for times stated with the cells incubated in the conditions described. Electrolyte for cells paced less than 24 h was the defined media; cells paced 24 h or longer included 5% FBS. 2.4. Transmembrane potential assay {#s0030} ---------------------------------- After pacing treatment, cells were suspended in phenol red-free IMDM with 2% FBS and 25 nM DiOC6(3) and allowed to equilibrate to room temperature prior to cytometric analysis. The cationic potentiometric carbocyanine dye, 3,3′-dihexyloxacarbocyanine iodide (DiOC6(3), Thermofisher Scientific) was used to measure TMP on a flow cytometer (Beckton Dickenson Cantos, Japan) with BD FACSDiva software (Ver. 8.0). The positively charged dye is taken up by the cell in proportion to the negative charge on the membrane. Using 488 nm excitation, a minimum of 25 k resulting cellular mean florescence events on the FITC channel (MFCH) were averaged for each treatment. The TMP thus recorded is directly related to the MFCH on a linear histogram. MFCH from non-paced cells treated with valinomycin (30 nM, Sigma-Aldrich; St. Louis, MO) defined the hyperpolarization setpoint (−90 mV, arbitrary); and gramicidin (100 nM, Sigma-Aldrich) the depolarization setpoint (0 mV, arbitrary). Controls were included on each experiment. 2.5. Proliferation assay {#s0035} ------------------------ Cells were checked for respiration using 3-(4,5-dimethylthiazol-2-yl)-2, 5-diphenyltetrazolium bromide (MTT Assay; Thermofisher Scientific) following manufacturer\'s protocol. Immediately after pacing, the electrolyte was discarded and replaced with whole phenol red-free IMDM with 1.2 nM MTT. After incubation, all but 25% of the media was discarded and 50% (v/v) dimethyl sulfoxide (DMSO, Sigma-Aldrich) was added to the remaining media to dissolve the resulting formazan crystals. Optical density for the resulting DMSO solutions was read at 540 nm. Cell cytotoxicity was estimated by comparison to absorbance levels generated by non-paced cells of similar number and treatments. 2.6. Trypan blue vital staining {#s0040} ------------------------------- To assess cellular death, cells paced for 3 h were subjected to Trypan Blue (0.1% Thermofisher) exclusion staining assay. The monolayer was gently rocked for 15 min in the cell culture incubator then washed in PBS for digital imaging under 10x magnification. The monolayer was then lysed with 1% sodium dodecyl sulfate (SDS) and the cleared lysate also read for optical density at 590 nm. 2.7. Electrotaxis {#s0045} ----------------- Prior to pacing, cells were directly passaged onto a 0.5% gelatin substratum with appropriate whole media, and incubated overnight. They were then paced for 96 h with the electrolyte changed daily. Digital imaging with 10x magnification was used to capture the results near the electrodes. 2.8. ATP assay {#s0050} -------------- After pacing treatments, cell pellets were suspended in PBS supplemented with a protease inhibitor cocktail (Sigma-Aldrich) and flash frozen in liquid nitrogen. After sonication, protein concentrations of lysates were determined by Bradford assay (Sigma-Aldrich) and samples normalized to the lowest concentration value. Lysates were deproteinized with trichloroacetic acid precipitation and subjected to ATP assay (ABCam; Cambridge, MA) utilizing the phosphorylation of glycerol to produce a colorimetric product quantitated by absorbance at 570 nm. 2.9. Mitochondrial imaging {#s0055} -------------------------- Cells were attached to poly-[l]{.smallcaps}-lysine treated cover coverslips and paced for 1 h or 24 h with non-paced controls. Cells were incubated and stained with 100 nM MitoTracker Red CMXRos (Thermofisher Scientific) before being fixed with 3.7% paraformaldehyde (Sigma-Aldrich). Fluorescence imaging (ex. 579 nm/em. 599 nm) was captured on a Zeiss AxioObserver Z1, 63X oil, confocal microscope using ZEN Blue Pro 2.0 program to produce the images. 2.10. Statistical analyses {#s0060} -------------------------- Statistical analyses were performed on GraphPad Prism (ver. 4.0, GraphPad Software Inc.; La Jolla, CA). Results are given as means±Standard Error of the Mean (SEM). ANOVA\'s were performed with Bonferroni\'s or Dunnett\'s Multiple Comparison Post Testing. P-values were considered significant if p\<0.05 compared to the similar non-paced control values. The respective symbols \* indicate p\<0.05, \*\* p\<0.01, and \*\*\* p\<0.001 in comparison to the unpaced cells. 3. Results {#s0065} ========== Paced cells showed altered global TMPs based on the polarity of the active electrode. The TMP of the non-paced cells was determined to be −83.2 mV±0.5 mV. Anodal pacing resulted in hyperpolarization with TMP increasing to −91.3 mV±1.0 mV. Cathodal paced cells exhibited an expected hypopolarization with TMP of −69.2 mV±0.3 mV. The symmetric biphasic waveform resulted in a TMP of −69.0 mV±0.5 mV similar to that of the cathodal paced cells. In the case of the cathodal cells, after removing the stimulus, multiple chambers were rested in an incubator and TMP was measured hourly. Non-paced cells were used to establish the TMP baseline, and assays were compared to the 0 h time point. The cells remained in a depolarized state for several hours before equilibrating back to near the resting baseline. Pacing for longer periods of time did not appear to affect the magnitude of the altered potential however did produce elongation of the cells and migration towards the anodal polarity ([Fig. 1](#f0005){ref-type="fig"}).Fig. 1Transmembrane potential measured with a cationic fluorescent dye after one hour of each treatment in panel A. Valinomycin caused hyperpolarization and gramicidin depolarization. Anodal pacing produced increased potential and cathodal and biphasic pacing showed less potential than unpaced cells. In panel B, the reduction of polarization caused by cathodal pacing required several hours to return to the baseline unpaced state.Fig. 1 A Trypan blue exclusion assay showed no staining in any of the cells as noted by both absorption fluorescence and phase contrast microscopy, indicating no dead cells or cells with ruptured membranes. All the electrically stimulated cultures showed inhibition of expansion of the cultures when compared to non-paced control cells by MTT data and duplicate hemocytometer counts although anodal and biphasic paced cells exhibited less inhibition of growth than the cathodal pulse (anodal vs non-paced 92.9%; cathodal vs non-paced 70.0%; biphasic vs non-paced 80.2%) ([Fig. 2](#f0010){ref-type="fig"}).Fig. 2In panel A, manual cell counts after 24 h of pacing showed inhibition of cell culture expansion with cathodal and biphasic waveforms, significant at p values less than 0.01 level. In panel B, MTT assay showed less cellular staining with any of the waveforms compared to non-paced at p values less than 0.01 (anodal and biphasic) and p value less than 0.05 level with cathodal.Fig. 2 ATP production, as another surrogate of metabolic activity, was measured in mono- and biphasic paced cells and in non-paced cells after 10 min, 3 h, and 24 h treatments using a colorimetric assay utilizing phosphorylation of glycerol to produce a colorimetric product quantitated by optical density at 570 nm. Cathodal and biphasic stimulated cells showed statistically significant increased ATP levels after 10 min of treatment while the ATP level in the anodal paced cells was similar to that of the non-paced controls (non-paced 0.53 ng ATP vs anodal 0.66 ng; vs cathodal 1.12 ng; vs biphasic 1.09 ng). After 3 h of treatment, the increased available ATP was still significantly higher than non-paced cells. At 24 h treatments, ATP levels in the cathodal and biphasic paced cells returned to non-paced control levels however, levels in the anodal paced cells were now significantly less than all the others (non-paced 0.53 ng ATP vs anodal 0.41 ng; vs cathodal 0.60 ng; vs biphasic 0.56 ng) ([Fig. 3](#f0015){ref-type="fig"}).Fig. 3Cellular ATP levels measured colorimetrically after ten minutes, three hours, and 24 h of pacing in panels A, B, and C respectively. In panel A, ATP levels were increased by cathodal and biphasic pacing at p values less than 0.001, in panel B increased at p values less than 0.01 and 0.05 respectively, and in panel C decreased by anodal pacing at a p value less than 0.05.Fig. 3 Together, the previously noted MTT and these ATP results suggest alterations in mitochondrial function when cells are influenced with pulsed electrical fields. These mitochondrial images at 24 h are A=non-paced, B=Anodal paced, and C=Cathodal paced, and D=Biphasic paced. The size bars indicate 20 µm of length. The arrows indicate mitochondria with increased bright staining and roundness indicating increased mitochondrial membrane potential in the case of anodal and biphasic (anodal/cathodal) stimulation ([Fig. 4](#f0020){ref-type="fig"}).Fig. 4Mitochondrial membrane potential measured with fluorescent dye after 24 h of treatment in panel A- unpaced, panel B- anodal, panel C- cathodal, and panel D- biphasic. White arrows in the anodal and biphasic paced cells show increased mitochondrial membrane potential. The white horizontal size bars indicate 20 µm of length.Fig. 4 4. Discussion {#s0070} ============= There is increasing evidence that there are histologic changes in the myocardium and impaired contractility [@bib2], and unfavorable morbidity and mortality outcomes in patients paced with the standard monophasic cathodal waveform [@bib3], [@bib4]. Previous work in the literature has shown improved acute hemodynamics in a number of species paced with anodal and biphasic (anodal/cathodal) stimulation [@bib5], [@bib6], [@bib7], [@bib8]. In Sheep with induced myocardial infarctions paced for three months with a biphasic (anodal/cathodal) waveform, statistically significant reduction of left ventricular volumes and increased percent fractional shortening has been noted [@bib9]. In addition, deliberate acute pacing of the heart with an anodal waveform in Humans has shown improved hemodynamics [@bib10]. Clearly we show that ATP levels are increased by electrical stimulation, and in fact may be drastically underestimated due to influences of the creatine phosphokinase buffering system. As we are dealing both with glycolysis and the Kreb\'s cycle, the absolute amounts measured can vary both by its buffering when being synthesized and rate-limiting when needed. The amounts of reaction products produced can vary by the number of viable cells or by the amount of NADH/NADPH present in those cells. Our findings suggest possible underlying mechanisms which may play a role in producing these observations. We conclude that anodal pacing, or a waveform including an anodal element, clearly alters mitochondrial morphology and function, however it is not clear whether this relates to dysfunction or might actually produce beneficial effects in the healing of areas of tissue damage. Specific pacing waveforms increase transmembrane potential and these effects last several hours after stopping pacing, cause migration of cells towards the anodal polarity, and appear not to cause cellular damage. Appendix A. Transparency document {#s0080} ================================= Supplementary materialSupplementary material. Supplementary materialSupplementary material. Supplementary materialSupplementary material. Supplementary data associated with this article can be found in the online version at [http://dx.doi.org/10.1016/j.bbrep.2016.09.004](http://doi:10.1016/j.bbrep.2016.09.004){#ir0005}.

All data are hosted by the Harvard Dataverse (<https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/HZDQ70>). Data are also hosted at IRRI (<http://ricestat.irri.org/fhsd/php/survey.php>) under Project ID: 98. Introduction {#sec001} ============ Rice *(Oryza sativa* L.*)* is one of the most important food crops as it is consumed by more than half of the world's population \[[@pone.0150345.ref001]\]. Despite its importance, the international rice market is considered a "thin" market; it is highly segmented because rice consumers have very specific preferences \[[@pone.0150345.ref002]\]. The definition of "premium-quality" rice is largely dependent on the socioeconomic context of consumers, with data suggesting that even lower income classes are increasingly conscious of food quality \[[@pone.0150345.ref003]--[@pone.0150345.ref006]\]. Rice quality is judged based on attributes, which could be classified several ways. Product characteristics could either be *intrinsic*, such as taste, texture, or color; or *extrinsic* to the product, such as packaging, brand, or label. Another attribute classification distinguishes between *search*, *experience*, and *credence* attributes. Search attributes are available for product evaluation before purchase, such as price, appearance, brand, and packaging. Experience attributes can be evaluated only upon product experience, thus after purchase or product use---examples are taste, texture, ease of cooking, and swelling capacity. Credence attributes are attributes that consumers cannot evaluate or verify themselves. Instead, they rely on people or institutions, such as government controls or industry claims. Attributes relating to production, processing, and product contents are typical examples of the credence-type attributes \[[@pone.0150345.ref007]\]. In this paper, we will focus on *intrinsic search* and *experience* attributes, such as visual and physicochemical grain properties. It is argued that measuring such properties objectively is difficult \[[@pone.0150345.ref008]\] but relatively high-throughput routine methods have been developed to conduct measurements of a number of rice quality parameters. Visual characteristics of rice grains are important *search* attributes that affect consumers' purchasing decisions and hence are used as some of the first selection criteria in varietal improvement programs \[[@pone.0150345.ref009]--[@pone.0150345.ref011]\]. Grain size is mainly based on the length. On the other hand, grain shape is based on length-to-width ratio \[[@pone.0150345.ref010]\]. The classification of rice samples based on size and shape is not standardized across different countries and different markets \[[@pone.0150345.ref012]--[@pone.0150345.ref014]\]. The routine classification system used by the International Rice Research Institute (IRRI) breeding programs is as follows: short (≤ 5.50 mm), medium/intermediate (5.51--6.60 mm), long (6.61--7.50 mm), and very long (\> 7.50 mm). The grain shapes of rice, likewise, can be described based on the routine value ranges used in IRRI: bold (≤ 2.0), medium (2.1--3.0), and slender (\> 3.0) \[[@pone.0150345.ref014]\]. Chalky areas in rice grains---those opaque white parts of the grain---are deemed, generally, to represent poor quality in many rice market segments and thus these grains fetch lower market prices \[[@pone.0150345.ref015]\]. Grains are classified based on the proportion of the grain that is chalky: none (0%), small (\< 10%), medium (10--20%), and large (\> 20%) \[[@pone.0150345.ref014],[@pone.0150345.ref016]\]. Traditionally, rice grain dimensions were measured using photographic enlargers and transparent rulers \[[@pone.0150345.ref009]\] while visual scoring by an experienced technician was conducted to determine chalkiness in rice grains \[[@pone.0150345.ref010]\]. Using manual ways of measuring grain dimensions is laborious and time-consuming while visual assessment of chalk has some degree of subjectivity and does not indicate where the chalky portion is in the grain \[[@pone.0150345.ref017]\]. *Experience* attributes, such as cooking and organoleptic properties of rice, affect a consumer's repeat purchasing behavior. Three parameters deemed most important in gauging the cooking and eating quality of a rice variety are: apparent amylose content (AAC), gel consistency (GC), and gelatinization temperature (GT). As AAC increases, cooked rice grains tend to be increasingly harder \[[@pone.0150345.ref018]\]. Colorimetry with iodine \[[@pone.0150345.ref019]--[@pone.0150345.ref021]\] remains the method of choice for measuring AAC despite its limitations \[[@pone.0150345.ref018],[@pone.0150345.ref022]--[@pone.0150345.ref024]\] and the development of new methodologies summarized by \[[@pone.0150345.ref025]\]. Based on AAC, rice can be grouped into five arbitrarily set classes: waxy (0--2%), very low (3--9%), low (10--19%), intermediate (20--25%), and high (\> 25%) \[[@pone.0150345.ref026]\] although a more recent study suggests that these AAC classes can further be subdivided \[[@pone.0150345.ref027]\]. There are cases in which rice materials of the same AAC class are very distinct in hardness. In these cases, GC is used as a complementary test for degree of hardness upon retrogradation. The methods for measuring GC, or the hardness of rice upon cooling after being cooked, are still in routine use today mainly for rice breeding programs focused on intermediate- and high-AAC materials \[[@pone.0150345.ref027],[@pone.0150345.ref028]\] while a method has been developed for glutinous rice \[[@pone.0150345.ref029]\]. Rice can be classified into three groups based on GC: hard and very flaky (≤ 40 mm), medium and flaky (41--60 mm), and soft (\> 61 mm) \[[@pone.0150345.ref010]\]. On the other hand, GT is associated with the cooking time of rice \[[@pone.0150345.ref030],[@pone.0150345.ref031]\]. Rice can be classified based on GT: low (\< 70°C), intermediate (70--74°C), and high (\> 74°C) \[[@pone.0150345.ref032]\]. The alkali spreading test \[[@pone.0150345.ref033]\] is a high-throughput assay for GT but this entails some subjectivity since the scores are based on perceptions of the analyst. Scores indicate GT classes, not direct information regarding the GT of the rice sample. Throughout the world, consumer preferences are far from homogeneous. Various market segments can be distinguished between continents, regions, countries, and even between socio-economic groups \[[@pone.0150345.ref007],[@pone.0150345.ref034],[@pone.0150345.ref035]\]. Grain quality experts in 23 countries have identified the top three popular rice varieties in their countries and, for some countries, at various sub-country levels; the most commonly assessed cooking and eating properties of these varieties have been reported \[[@pone.0150345.ref027]\]. Consumers may not be able to articulate the reasons behind their preferences or describe what they like or dislike in food items but they show appreciation or the value they attach to food in other ways \[[@pone.0150345.ref036]\] such as a willingness to pay higher prices for rice with certain quality attributes. Price differences between rice samples of different quality classes indicate that grain quality attributes must be contributing to the price of rice. Determining the implicit contribution of the various grain quality attributes to the market price of rice varieties can be done through hedonic pricing analysis \[[@pone.0150345.ref037]\]. Hedonic pricing regressions have been quite popular in the economics literature, being applied to various food commodities such as wine \[[@pone.0150345.ref038],[@pone.0150345.ref039]\], tea \[[@pone.0150345.ref040]\], apples \[[@pone.0150345.ref041]\], and breakfast cereals \[[@pone.0150345.ref042]\]. In all these commodities, the products could be grouped into quality classes or varieties \[[@pone.0150345.ref043]\]. The hedonic pricing model has also been applied to study the effects of extrinsic and intrinsic quality attributes of rice to market prices \[[@pone.0150345.ref044]--[@pone.0150345.ref047]\]; with the results suggesting that varietal improvement programs should not be limited only to yield-enhancing traits. Furthermore, dissemination and adoption of new varieties need to be supported. A recent study in Central Luzon found that out of the 200 modern rice varieties released, fewer than 10 varieties are being used by farmers in the Central Luzon area \[[@pone.0150345.ref048]\]. Previous hedonic studies involving intrinsic rice quality parameters \[[@pone.0150345.ref044],[@pone.0150345.ref046],[@pone.0150345.ref047]\] have two main limitations. First, they did not investigate how homogeneity in some physical characteristics influences rice prices. Homogeneity in physical characteristics---such as length and width---of the rice sample being purchased may play a major role in consumers' willingness-to-pay for rice. Rice varieties are often mixed at various stages of harvest and post-harvest activities (i.e. harvesting, threshing, drying, and milling), which results in heterogeneous grain quality. Second, rice quality data obtained in these studies, specifically degree of chalkiness and GT, were measured through semi-quantitative means: scores were provided based on experienced technicians' evaluations. Techniques that provide quantitative data now exist and can potentially improve hedonic pricing models. Machine vision technology, such as digital imaging systems, is available for monitoring quantifiable attributes of post-harvest quality of plant and animal products such as size, shape, and degree of chalkiness, as in the case of rice \[[@pone.0150345.ref017],[@pone.0150345.ref049]--[@pone.0150345.ref051]\]. Differential scanning calorimetry (DSC) is an alternative method (to the alkali spreading test) for characterizing GT; it monitors thermal transitions and provides the temperature range at which the crystalline starch structures irreversibly melt in the presence of plasticizing water \[[@pone.0150345.ref052],[@pone.0150345.ref053]\]. Paste viscosity, measured using a Rapid Visco-Analyzer (RVA), is another indicator of cooking and eating quality in rice \[[@pone.0150345.ref054],[@pone.0150345.ref055]\]. Different parts of the viscosity curve have been associated with GT, with AAC, and with texture \[[@pone.0150345.ref055]--[@pone.0150345.ref057]\]. Previous literature on consumer preferences in the Philippines has mainly focused on big urban consumption zones \[[@pone.0150345.ref044],[@pone.0150345.ref047],[@pone.0150345.ref058]\]. Since the Philippines is a net importer, consumer preferences in urban areas close to the port tend to be dominated by imported rice characteristics, which are not necessarily satisfied by domestic supply. In order to focus on consumer preferences for the characteristics of rice varieties which are currently produced domestically, we need to move away from these highly urbanized zones. Since we are also not interested in preferences of consumers who are producers themselves, we need to look for concentrated consumption zones close to production zones, i.e. rural towns. Therefore, in this study, we determine how the implicit market value of intrinsic search and experience quality attributes of rice contribute to the total market price of rice in two rural towns in the Philippines: Famy (14.4333°N, 121.4500°E) and Sta Maria (14.4700°N, 121.4261°E). In doing so, the contributions of this study are: (1) to determine consumer preferences in rural areas, which have largely been de-prioritized in consumer preference studies; (2) to apply quantitative methodologies in measuring GT and chalkiness; and (3) to identify the household willingness-to-pay for various characteristics of rice by income groups. Materials and Methods {#sec002} ===================== Ethics statement {#sec003} ---------------- There is no Institutional Review Board at IRRI at this time. However, this work has been reviewed and approved by the division head of the Social Sciences Division through an internal review process. All participants in this study gave informed oral consent prior to the survey interview and had the option to terminate the interview at any point. No minors were directly interviewed during this study. All datasets collected by the Social Sciences Division of IRRI are ultimately uploaded and made available for public use. However, it is our policy to first make datasets anonymous prior to uploading. This is done by removing all identifying information within the dataset, including: name, email, telephone number, street address, and gps coordinates. These measures are done with the approval of the Chief Information Officer of the International Rice Research Institute. Survey and sample collection {#sec004} ---------------------------- Rice consumer respondents (n = 128) were selected in the adjacent towns of Famy and Sta Maria, two northernmost towns in Laguna province in the Philippines, situated at 93 and 86 km from the capital city, Manila, respectively. These towns are both rural areas. To select the respondents, high-resolution imagery from Google Earth and a global positioning system (GPS) were used. The target population was delimited in the town proper because houses are more closely situated there. Households 20 meters apart were marked based on GPS coordinates. Then, 100 GPS coordinates were randomly selected. In cases when the GPS coordinates pointed to vacant lots or roads, respondents were obtained from the house nearest the GPS point. All respondents gave oral informed consent to be surveyed. Respondents were interviewed to determine their socioeconomic profiles ([Table 1](#pone.0150345.t001){ref-type="table"}). A sample of uncooked milled rice (300 g) consumed by these respondents was then collected in exchange for 1 kg of premium milled rice. At the time of collection, respondents also reported the price paid per kilogram of rice. The milled rice samples from the respondents were sent to the Grain Quality and Nutrition Center of the IRRI for physicochemical analyses. For the statistical analysis presented in this paper, households were classified based on Philippine National Statistics Office (NSO) income classes \[[@pone.0150345.ref059]\] but with a slight modification by merging the three lowest NSO-reported income classes into a single category as the low household income class used for this study. For the final analysis, only three income classes remain: low household income (\< 2,431.91 USD per annum), medium household income (between 2,431.91 and 6,079.77 USD per annum), and high household income (\>6,079.77 USD per annum). The exchange rate at the time of the study was 1 USD = 41.12 PHP 10.1371/journal.pone.0150345.t001 ###### Characteristics of respondents by income group and by location (Famy and Sta Maria, Laguna, Philippines). {#pone.0150345.t001g} Income Group Location ---------------------------------- ------------------- ------------------- ---------------------- --------------------- --------------------- --------------------- Household size 4 (1) 5 (2) 5 (2) 5 (2) 5 (2) 5 (2) Annual household income (USD) 1,512.79 (657.56) 3,923.47 (991.44) 10,305.86 (4,836.09) 5,366.85 (4,335.58) 4,910.46 (4,829.18) 5,145.79 (4,568.85) Age of rice purchaser (yrs) 43 (15) 46 (15) 43 (13) 43 (15) 45 (14) 44 (15) Educ. of rice purchaser (yrs) 9 (3) 10 (3) 11 (3) 10 (3) 10 (3) 10 (3) Rice consumption per capita (kg) 211 (91) 214 (107) 200 (86) 230 (102) 186 (83) 209 (95) Rice price (USD/kg) 0.78 (0.05) 0.80 (0.05) 0.83 (0.07) 0.80 (0.05) 0.80 (0.07) 0.80 (0.05) Sample Size 41 47 40 66 62 128 Note: Figures presented here are means and standard deviations (in parentheses). The conversion rate at the time of the study (September to December 2012) was 1 USD = 41.12 PHP. Grain quality analyses {#sec005} ---------------------- Milled grains underwent assessment of physical traits (grain dimensions, proportion of head rice in milled rice, and chalkiness) and then an test portion of each sample was ground into fine flour (100-mesh) using a Udy Cyclone Sample Mill (model 3010--30, Fort Collins, CO). Reverse osmosis (RO) water and reagent-grade chemicals were used for the chemical analyses. Physical traits (length, width, and degree of chalkiness) of the milled rice grains were determined using the Cervitec^™^ 1625 Grain Inspector (FOSS, Denmark). Grain shape was calculated based on the length-to-width ratio of the grains. The proportion of head rice (%) in the milled rice was determined by measuring the amount of grains that are 75% intact after a test portion (100 g) of milled rice was sorted using a shaking sieve; the rest are broken grains (%). The measurement of AAC was conducted following the Routine Method of ISO 6647 \[[@pone.0150345.ref060]\], and calculated based on a standard curve generated using the iodine-binding capacities of a set of standard rice varieties. The AACs of these standards were determined as described in the Reference Method of ISO 6647 \[[@pone.0150345.ref061]\]. Absorbance was measured at 620 nm using a San^++^ Automated Wet Chemistry Analyzer (Skalar Analytical B.V., Breda, The Netherlands) equipped with an SA1100 sampler. Data were collected and analyzed using the Skalar FlowAccess^™^ V3 software. Gel consistency was determined according to a previously published protocol \[[@pone.0150345.ref028]\]. Gelatinization temperature was measured by DSC (Q100, TA Instruments, New Castle, DE, USA). Flour (4 mg) and RO water (8 μL) were placed in an aluminum pan, which was then sealed hermetically. An empty hermetically sealed pan served as the reference. The temperature was raised from 35°C to 120°C at 10°C min^-1^. Thermal transitions were recorded and analyzed using the TA Universal Analysis 2000 software. The peak of each resulting endotherm was reported as the GT. Viscosity curves for the rice samples were generated using the RVA following the profile detailed in the AACC Method 61--02 \[[@pone.0150345.ref062]\]. Several points in the viscosity curves were recorded: peak (PV), trough (TV), and final (FV) viscosities; derived values from these points were calculated: breakdown (BD, the difference between PV and TV), lift-off (LO, the difference between FV and TV), and setback (SB, the difference between FV and PV) \[[@pone.0150345.ref063]\]. The time of PV (peak time) and pasting temperature were also obtained. Statistical analysis {#sec006} -------------------- ### Across income groups {#sec007} One-way analysis of variance (ANOVA) was conducted for hypothesis testing for grain quality parameters that passed the assumptions of ANOVA (average length, variability in length, PV, TV, FV, and LO). For the other parameters, hypotheses were tested using Kruskal-Wallis rank sum test and means were compared using the Mann-Whitney-U post-hoc test with Bonferroni correction. ### Between towns {#sec008} The Z-test was used to compare samples from the two towns in the quality parameters whose data were normally distributed. For parameters whose data were not normally distributed, the Mann-Whitney (Wilcoxon rank sum) test was used. These analyses were conducted using R (version 3.2.0, released 2015). Establishing the hedonic price model {#sec009} ------------------------------------ Hedonic pricing regressions are based on Lancaster's "characteristics theory of value," which states that any good can be described in terms of its attributes or characteristics \[[@pone.0150345.ref064]\]. The price consumers are willing to pay for a good at a given time is therefore assumed to depend on the attributes of the good or commodity. When buying rice, consumers face a choice of several search attributes, some are visible and others are not, but those attributes are embedded in the product. Each of these attributes contributes to the final price paid on the market. The socioeconomic status of consumers and their preferences will determine which products they will buy among a given set of quality-differentiated rice types available in the market and hence which prices they are willing to pay for those products. The rice types purchased and prices paid by consumers can be interpreted as their revealed preferences and willingness-to-pay (WTP) \[[@pone.0150345.ref065]\]. Thus, besides the physical and chemical characteristics of rice, we consider that consumers' socioeconomic status also determines the type of rice they choose in the market and the price they are willing to pay. We therefore specify our hedonic pricing model as follows: $$P_{i} = \beta\mathbf{x}_{i} + \rho\mathbf{z}_{i} + \rho\mathbf{k}_{\mathbf{i}} + \eta D + \varepsilon_{i}$$ where, *P* represents the price paid for rice by consumer *i*, **x** is a vector of physical attributes characterizing rice purchased by consumer *i*, **z** is a vector of chemical characteristics embedded in rice purchased by consumer *i*, **k** is a vector of socioeconomic characteristics describing consumer *i*, *D* is a location dummy assumed to capture region-specific factors, and ε is the error term of the model. The hedonic model expressed in [Eq 1](#pone.0150345.e001){ref-type="disp-formula"} is estimated in log-log functional form using Ordinary Least Squares (OLS). The model was first estimated as a pooled sample, then per income group, and with income interaction terms on physical and chemical rice characteristics and one socio-economic factor (per capita rice consumption). The physical characteristics considered in the model are: the proportion of broken rice in the sample and chalkiness. The chemical characteristics are: GC, AAC, and GT. The consumer characteristics are: rice consumption per capita, age, gender, and education of the rice purchaser. Traditionally, the estimated coefficients from the hedonic regression are interpreted as consumers' WTP for a given attribute of the good or commodity. A positive sign indicates that consumers are willing to pay a price premium for the attribute, while a negative sign reveals that consumers discount the attribute. Studies on consumer preferences typically use consumer and expert surveys or interviews \[[@pone.0150345.ref027],[@pone.0150345.ref066]\], quality evaluations of samples coming from national programs \[[@pone.0150345.ref058],[@pone.0150345.ref067]\], or consumer product preference and acceptability tests using specific sets of rice samples \[[@pone.0150345.ref068]\]. Such approaches, although effective in determining stated consumer preferences and the characteristics of those rice varieties, do not indicate buyers' WTP based on quality attributes. To reveal realistic contributions of grain quality to market prices in rice that people actually buy, it is best to base hedonic regression models on information about samples obtained from consumers. This method of obtaining sample from surveyed households was previously employed by Abansi et al. \[[@pone.0150345.ref044]\]. In that study, rice samples obtained from surveyed households were analyzed for physicochemical characteristics to compare consumer preferences between urban and rural consumers \[[@pone.0150345.ref044]\]. Additionally, consumer preferences were investigated across different income strata. Results and Discussion {#sec010} ====================== Socioeconomic characteristics of the respondents {#sec011} ------------------------------------------------ Among the 128 respondents surveyed ([Table 1](#pone.0150345.t001){ref-type="table"}) 66 came from Famy and 62 from Sta Maria. The 128 respondents were also grouped by income classes: 41 were classified as low-income, 47 as middle-income, and 40 as high-income. The respondents in Famy and in Sta Maria had similar socioeconomic characteristics and, on average, bought rice with the same price (0.80 USD/kg, [Table 1](#pone.0150345.t001){ref-type="table"}). The majority (80%) of the rice purchasers who participated in the survey have studied for 8--12 years, with age averaging in the 40s across the three income classes ([Table 1](#pone.0150345.t001){ref-type="table"}). Household sizes of respondents, on average, were four household members for the low-income class and five members in both middle- and high-income classes, and in both towns. The annual rice consumption of the respondents in these households was mainly 100--250 kg per capita. The averages of per capita consumption of rice by location and by income class ([Table 1](#pone.0150345.t001){ref-type="table"}) were higher than the per capita consumption determined in the 1990s for rural consumer groups \[[@pone.0150345.ref069]\] and the national average in 2008--2009 \[[@pone.0150345.ref070]\]. Most of the respondents earned less than 4,863.81 USD annually, with the number of respondents decreasing with an increase in income. Physical characteristics of raw grains of the rice samples {#sec012} ---------------------------------------------------------- Based on the IRRI classification system for grain size and shape \[[@pone.0150345.ref014]\], the respondents across the different income groups in the towns had a revealed preference for rice with long and slender grain shape (Tables [2](#pone.0150345.t002){ref-type="table"} and [3](#pone.0150345.t003){ref-type="table"}). 10.1371/journal.pone.0150345.t002 ###### Physical, cooking, and eating quality indicators of rice samples obtained from respondents by income class. {#pone.0150345.t002g} Income Class -------------------------- --------------------------------------------------- ---------------------------------------------------------------------------------------- --------------------------------------------------- --------- ------------------ Grain length (mm) 6.67 (0.08) 6.67 (0.09) 6.69 (0.08) 0.35^A^ 6.68 (0.08) CV in length (%) 4.76 (0.47) 4.66 (0.36) 4.86 (0.41) 0.09^A^ 4.76 (0.41) Width (mm) 2.08 (0.06) 2.10 (0.06) 2.10 (0.05) 0.30^B^ 2.09 (0.06) CV in width (%) 7.94 (0.82) 7.77 (0.02) 7.87 (0.78) 0.52^B^ 7.85 (0.75) Ratio of length/width 3.20 (0.11) 3.18 (0.11) 3.19 (0.08) 0.43^B^ 3.19 (0.10) Chalkiness (%) 17.00 (8.00) 19.00 (8.00) 17.00 (7.00) 0.13^B^ 18.00 (7.00) Head rice (%) 56.75 (8.85) 55.56 (11.13) 58.49 (14.41) 0.70^B^ 56.86 (11.61) AAC (%) 24.01 (1.98) 23.54 (2.32) 24.12 (1.46) 0.78^B^ 23.87 (1.98) GT (°C) 77.37 (1.07) 77.04 (1.70) 76.76 (2.41) 0.32^B^ 77.06 (1.80) GC (mm) 50.90 (12.10) 54.20 (13.50) 51.60 (14.80) 0.49^B^ 52.30 (13.50) PV (cP) 2924.00 (317.00) 2922.00 (290.00) 2809.00 (325.00) 0.16^A^ 2887.00 (312.00) TV (cP) 1638.00 (256.00) 1637.00 (229.00) 1617.00 (287.00) 0.92^A^ 1631.00 (255.00) BD (cP) 1287.00 (228.00) 1285.00 (335.00) 1192.00 (222.00) 0.10^B^ 1257.00 (272.00) FV (cP) 3900.00 (517.00) 3850.00 (480.00) 3908.00 (613.00) 0.86^A^ 3884.00 (533.00) SB (cP) 975.00 (422.00) 928.00 (578.00) 1099.00 (497.00) 0.49^B^ 997.00 (508.00) Peak time (min) 5.56[^a^](#t002fn001){ref-type="table-fn"} (0.09) 5.5[^a^](#t002fn001){ref-type="table-fn"}[^b^](#t002fn002){ref-type="table-fn"} (0.09) 5.52[^b^](#t002fn002){ref-type="table-fn"} (0.09) 0.03^B^ 5.54 (0.09) Pasting temperature (°C) 75.46 (0.96) 75.10 (1.64) 75.38 (1.31) 0.58^B^ 75.30 (1.35) LO (cP) 2262.00 (298.00) 2213.00 (303.00) 2291.00 (364.00) 0.52^A^ 2253.00 (321.00) ^a^ Figures presented here are means and standard deviations (in parentheses). For peak time, a different lowercase letter beside each mean indicates that the means are significantly different (α = 0.05). ^b^ The conversion rate at the time of the study (November 2012) was 1 USD = 41.12 PHP. ^c^ The letter beside the p-value indicates the test statistic used: (A) ANOVA (F-statistic), (B) Kruskal-Wallis rank sum test (χ^2^). 10.1371/journal.pone.0150345.t003 ###### Comparison of physical, cooking, and eating quality indicators of rice samples obtained from respondents by location (Famy and Sta Maria, Laguna, Philippines) using the Z-test. {#pone.0150345.t003g} Location ------------------- ------------------- ------------------- ------- Grain length (mm) 6.66 (0.08) 6.70 (0.08) -2.55 CV in length (%) 4.72 (0.43) 4.80 (0.41) -1.05 PV (cP) 2,839.00 (307.00) 2,939.00 (311.00) -1.82 TV (cP) 1,613.00 (248.00) 1,649.00 (263.00) -0.79 FV (cP) 3,872.00 (530.00) 3,897.00 (540.00) -0.26 Note: Values presented are means and standard deviations (in parentheses). At p \< 0.05, attributes with Z \< -1.96 or Z \> 1.96 are significantly different between the locations. Results indicated that the widths and the shapes of rice samples obtained from respondents in Famy and in Sta Maria were not significantly different across income classes ([Table 2](#pone.0150345.t002){ref-type="table"}) and between towns ([Table 4](#pone.0150345.t004){ref-type="table"}). The lengths of the rice grains were not significantly different across income classes ([Table 2](#pone.0150345.t002){ref-type="table"}) but there was a small but significant difference in the length between the grains consumed in the two towns: Famy respondents purchased, on average, slightly shorter grains than Sta Maria respondents ([Table 3](#pone.0150345.t003){ref-type="table"}), but the difference was not large enough to put the grains into different quality classes. The variability in length and in width of rice grains across the different income classes and between towns were not significantly different as well (Tables [2](#pone.0150345.t002){ref-type="table"}--[4](#pone.0150345.t004){ref-type="table"}). The data indicated a preference for long and slender grains. This similarity may be associated with the proximity of the two towns. It is possible that the markets in these towns have the same set of rice suppliers, hence leading to the same grains being sold. It is also possible that this preference for long and slender grain is stable due to similarities to grain dimensions of IR64 in the late 1980s \[[@pone.0150345.ref071]\], a benchmark of rice grain quality in the Philippines for millers, traders, and consumers \[[@pone.0150345.ref072]\]; and to recently reported expert opinions on grain dimensions of highly preferred Filipino rice varieties \[[@pone.0150345.ref027]\]. 10.1371/journal.pone.0150345.t004 ###### Comparison of physical, cooking, and eating quality indicators of rice samples obtained from respondents by location (Famy and Sta Maria, Laguna, Philippines) using the Mann-Whitney (Wilcoxon rank sum) test. {#pone.0150345.t004g} Location ----------------------- ---------- ------- ------ Width (mm) 2.09 2.09 0.98 CV in width (%) 7.79 7.92 0.26 Ratio of length/width 3.19 3.2 0.35 Chalkiness (%) 18 18 0.75 Head rice (%) 56.71 57.01 0.99 AAC (%) 23.66 24.1 0.06 GT (°C) 77.09 77.04 0.39 GC (mm) 53.65 51.06 0.21 BD (cP) 1226 1290 0.24 SB (cP) 1033 958 0.9 LO (cP) 2,259 2,248 0.55 Peak time (min) 5.52 5.56 0 Pasting temp (°C) 75.27 75.38 0.31 Note: Values presented are means. ^a^ For comparison between locations, attributes with p \< 0.05 are significantly different. Chalky areas in rice grains are caused by loose packing or incomplete filling of starch granules \[[@pone.0150345.ref015],[@pone.0150345.ref031],[@pone.0150345.ref073]\]. It effectively weakens the grain \[[@pone.0150345.ref031],[@pone.0150345.ref074]\], leading to elevated incidence of breakage during the milling process and to reduced head rice yield \[[@pone.0150345.ref075]\]. Across the income classes and between towns, the degree of chalkiness was not significantly different, with the grains having medium chalkiness, on average (Tables [2](#pone.0150345.t002){ref-type="table"} and [4](#pone.0150345.t004){ref-type="table"}). These findings indicate that the respondents in this study have similar preferences in terms of chalkiness in grains. Perhaps, the respondents did not mind having opaque spots on the raw rice grains as long as grains are not broken. After all, chalkiness does not directly affect the cooking and eating experience of rice \[[@pone.0150345.ref016]\]. There were non-significant differences in the proportions of head rice across the different income groups ([Table 2](#pone.0150345.t002){ref-type="table"}) and between the two towns ([Table 4](#pone.0150345.t004){ref-type="table"}), with respondents having submitted samples with 57% head rice, on average. Based on the proportion of head rice in the respondents' samples, it appears that the respondents were willing to pay for milled rice that fell below premium standards set by the Philippines' National Food Authority (NFA) \[[@pone.0150345.ref076]\]; perhaps, premium grade milled rice was not available in the markets surveyed in this study. However, the data reported here (Tables [2](#pone.0150345.t002){ref-type="table"} and [4](#pone.0150345.t004){ref-type="table"}) indicate improvement from previously reported rice mill yields in the Philippines \[[@pone.0150345.ref077]\], suggesting that post-harvest processing conditions and processing facilities have improved. The motivation of breeders and post-harvest practitioners to improve head rice recovery could stem from consumers' possible association between good taste and the wholeness of the rice grain. It is possible that the respondents' choices of rice were constrained by their purchasing power as head rice has been reported to be correlated with market rice prices \[[@pone.0150345.ref078]\]. Cooking and eating properties of the rice samples {#sec013} ------------------------------------------------- Amylose is one of two starch polymers in rice. Amylose content is believed to be one of the best single indicators of the texture, particularly of the hardness, of rice samples \[[@pone.0150345.ref024]\]; hence, it plays a critical role in selection decisions in rice breeding programs \[[@pone.0150345.ref023]\]. In this study, there were no significant differences in AACs in rice consumed across the different income classes ([Table 2](#pone.0150345.t002){ref-type="table"}) and between towns ([Table 4](#pone.0150345.t004){ref-type="table"}), with the respondents consuming rice with intermediate AAC, on average. These findings agree with previously reported Filipino consumer preferences \[[@pone.0150345.ref027],[@pone.0150345.ref034]\]. On the other hand, the rice samples from the respondents were, on average, of the medium GC class, in agreement with characteristics of popular Philippine rice varieties reported by Calingacion et al. \[[@pone.0150345.ref027]\]. The respondents' textural preferences, according to GC values, were similar across income classes ([Table 2](#pone.0150345.t002){ref-type="table"}) and between towns ([Table 4](#pone.0150345.t004){ref-type="table"}). Amylose has been implicated in affecting GC \[[@pone.0150345.ref053]\], which predicts texture of cooked rice; however, rice texture is also reportedly influenced by proteins and lipids \[[@pone.0150345.ref063],[@pone.0150345.ref079]--[@pone.0150345.ref082]\]. Amylopectin is the other polymer of starch in rice grains. During gelatinization, the crystalline lamellae of amylopectin melt; the temperature range at which this happens---referred to as GT---depends on the distributions of chain-lengths within the amylopectin semi-crystalline cluster \[[@pone.0150345.ref030],[@pone.0150345.ref083]--[@pone.0150345.ref084]\]. Respondents across income classes ([Table 2](#pone.0150345.t002){ref-type="table"}) and between towns ([Table 4](#pone.0150345.t004){ref-type="table"}) similarly preferred rice with high GT. This result contrasts reports that indicated that the preferred GT class in the Philippines is low to intermediate \[[@pone.0150345.ref027],[@pone.0150345.ref085]\]. However, DSC could give a different value from what is obtained from alkali-based methods \[[@pone.0150345.ref030]\], which have been used for GT characterization in the contrasting studies. Juliano et al. had used the alkali turbidimetric assay \[[@pone.0150345.ref085]\] while Calingacion et al. obtained ASV values from the experts who participated in the survey, only conducting DSC for samples of unknown GT (the authors did not specify which samples were subjected to DSC or had reported ASV data) \[[@pone.0150345.ref027]\]. The rice submitted by the respondents had statistically similar viscosity values across income groups ([Table 2](#pone.0150345.t002){ref-type="table"}) and between towns (Tables [3](#pone.0150345.t003){ref-type="table"} and [4](#pone.0150345.t004){ref-type="table"}). However, there was a small but significant difference in the amount of time the rice samples needed to reach PV among the different income groups. The low-income group had rice samples with slightly longer peak times than samples obtained from the high-income group ([Table 2](#pone.0150345.t002){ref-type="table"}). The difference in peak times indicates slightly different swelling behaviors between the rice from the low-income group and the high-income group; its impact during cooking, however, might be too small to be distinguished as GT is similar across income classes ([Table 2](#pone.0150345.t002){ref-type="table"}). The hedonic price model {#sec014} ----------------------- This study uses a log-log functional form in all preliminary models ([Table 5](#pone.0150345.t005){ref-type="table"}) as well as the final model ([Table 6](#pone.0150345.t006){ref-type="table"}). One advantage of this functional form is that estimated coefficients can be interpreted as elasticities. The original hedonic model that was investigated can be found in column (1) of [Table 5](#pone.0150345.t005){ref-type="table"}. In this model, per capita income was found to be significant; different income levels are likely to have different hedonic price models. Therefore, we split the same regression in three income classes (columns (2) to (4)) and observe that slopes of explanatory variables such as percent broken and GC indeed differ across income strata ([Fig 1](#pone.0150345.g001){ref-type="fig"}). Unfortunately, this method also has the disadvantage that sample sizes for income groups are smaller and degrees of freedom become limiting due to the inclusion of several explanatory variables on rice grain quality as well as socio-economic factors. Because of this limitation, income classes are interacted with grain quality characteristics and one socio-economic factor in [Table 6](#pone.0150345.t006){ref-type="table"}. Other socio-economic factors were found to have similar effects among income classes. Hence, we were able to exploit the full, pooled sample and significantly increase the explanatory power of the original model (column (1) in [Table 5](#pone.0150345.t005){ref-type="table"}) from R-squared = 39% to 50% ([Table 6](#pone.0150345.t006){ref-type="table"}). 10.1371/journal.pone.0150345.t005 ###### Preliminary regression results for hedonic price models for rice. {#pone.0150345.t005g} \(1\) \(2\) \(3\) \(4\) ----------------------------- ------------------------ ---------------------- -------------------- ------------------------ Percent broken --0.0695\*\*\*(0.0140) --0.0450\*\*(0.0208) --0.0582\*(0.0331) --0.0031(0.0419) GC 0.0394\*(0.0223) 0.0595(0.0452) --0.0238(0.0348) 0.0033(0.0430) AAC 0.0954(0.0656) 0.2875(0.1944) 0.0569(0.1037) 0.0431(0.0932) GT 0.2191(0.2245) 0.1964(0.3593) 0.9785\*(0.5190) --0.5421(0.5689) Small Chalkiness 0.0121(0.0178) 0.1419\*\*(0.0515) --0.0291(0.0360) --0.0061(0.0208) Per capita income 0.0159\*\*(0.0063) 0.0458(0.0308) --0.0058(0.0365) 0.0118(0.0132) Per capita rice consumption 0.0275\*\*(0.0116) 0.0507\*\*(0.0234) --0.0083(0.0187) 0.0518\*\*(0.0205) Household size 0.0202(0.0130) --0.0163(0.0340) --0.0045(0.0416) 0.0425\*(0.0237) Age of rice purchaser 0.0338\*\*(0.0151) --0.0475(0.0416) 0.0405\*(0.0238) 0.0066(0.0210) Educ. of rice purchaser 0.0098(0.0137) --0.0892\*(0.0482) 0.0446\*\*(0.0204) --0.0095(0.0152) Gender of rice purchaser --0.0015(0.0124) 0.0132(0.0265) 0.0114(0.0174) --0.0189(0.0207) Location --0.0216\*\*(0.0109) 0.0319(0.0238) --0.0256(0.0177) --0.0497\*\*\*(0.0173) Intercept 1.9591\*\*(0.9331) 1.5375(1.7302) --0.7861(1.9570) 5.3386\*\*(2.5060) Observations 127 40 46 41 R-squared 0.39 0.70 0.47 0.42 Note: Standard errors in parentheses \'\*\*\*\', \'\*\*\', \'\*\' significant at 1, 5, and 10%. 10.1371/journal.pone.0150345.t006 ###### Hedonic price function interacting income classes. {#pone.0150345.t006g} Interaction with ----------------------------- ------------------------ ----------------------- ------------------- Percent broken --0.0560\*\*\*(0.0185) --0.0078 (0.0398) 0.0395 (0.0538) GC 0.0776\*\*(0.0354) --0.1055\*\* (0.0508) --0.0760 (0.0601) AAC 0.1585(0.1583) --0.1315 (0.1917) --0.1447 (0.1925) GT --0.1008(0.3051) 1.0539\* (0.5894) --0.6394 (0.7458) Small Chalkiness 0.0631(0.0392) --0.0858 (0.0541) --0.0637 (0.0456) Per capita rice consumption 0.0592\*\*\*(0.0191) --0.0664\*\* (0.0259) --0.0449 (0.0290) Middle income dummy --3.3724(2.4746) Low income dummy 3.5979(3.2838) Household size --0.0035(0.0132) Age of rice purchaser 0.0230(0.0160) Education of rice purchaser 0.0084(0.0132) Gender of rice purchaser 0.0029(0.0127) Location --0.0225\*(0.0114) Intercept 2.9427\*\*(1.4672) Observations 127 R-squared 0.50 Note: Standard errors in parentheses \'\*\*\*\', \'\*\*\', \'\*\' significant at 1, 5, and 10%. {#pone.0150345.g001} The main advantages of the model presented in [Table 6](#pone.0150345.t006){ref-type="table"} are that this model maintains the original sample size and still shows the income effect on the revealed preferences of consumers. Results of this model suggest that only the high-income class significantly discounts broken grains. However, this is likely the result of high-income consumers purchasing rice from a wider range of quality classes, which is confirmed by the higher variability in the amount of percent broken in the rice samples obtained from this income class ([Fig 1](#pone.0150345.g001){ref-type="fig"}). Middle- and low-income consumers in this study did not purchase premium rice with low amounts of percent broken. Results also indicate that soft rice is preferred by high-income consumers with GC significant at five percent. Conversely, middle-income consumers discount rice with higher GC. This study also revealed that per capita rice consumption significantly affected WTP. However, the sign changes between high- and middle-income classes which indicate that high-income consumers spend more per kg of rice as their consumption increases and middle-income consumers spend less per kg of rice as their consumption increases. The results of high-income consumers' preference for soft rice, as measured by GC values, is in agreement with previous studies \[[@pone.0150345.ref086]\]. Implicit price {#sec015} -------------- Marginal implicit prices are calculated as the product of the mean rice price from the collected samples and the mean beta coefficients of the physical and chemical characteristics divided by the mean of the explanatory variables of the collected rice samples. These implicit prices were estimated based on [Table 5](#pone.0150345.t005){ref-type="table"} for the whole sample as well as by income classes. The data show that amount of percent broken grains was significant in almost all income classes. Although the magnitude of the implicit price is rather small for percent broken, it is important to note that there is a large range in percent broken values. Samples in this study ranged from 2.4% broken to 66.4% broken; the level of percent broken can change the price of rice by more than 0.08 USD (9.7%) throughout the total range. Gel consistency was found to be significant at the 10% level for the non-interacted variable with an implicit price of approximately 0.08 USD/kg for every 10% change and a total range of 550%. Also significant at the 10% level was GT for middle-income consumers. The implicit price for these consumers was 0.01 USD/kg but the range of values for GT in the middle-income was only 7.43°C. As such, GT could affect the price of rice by as much as 0.08 USD/kg (9.4%). Conclusion {#sec016} ========== This study was conducted to measure the contribution of grain quality attributes to the market value of rice in two rural towns in the Philippines. Unlike previous hedonic studies which considered semi-quantitative scores for measuring of GT and chalkiness, this study included measurement of GT based on thermal transitions in DSC and of chalkiness based on computerized image analyses. These techniques are more accurate and reliable than routinely used assays (alkali spreading test for GT and visual scores for chalkiness). Moreover, in addition to commonly used physicochemical data, this study also employed an interaction term for income classes to reveal income effects in these factors on rice price. The results of this study indicate that consumers' response to grain quality characteristics changes over income classes. Generally speaking, low-income consumers appear to have less pronounced preferences for rice based on physical and chemical characteristics. Or more likely, these consumers do not have the economic power to express their preferences. Additionally, the absence of preference may result from homogeneity in the rice consumed by this income class. High-income consumers have the largest variability in rice grain quality attributes and concurrently appear to have the most pronounced preferences among consumers. High-income consumers also spend more money per kg as their consumption increases, while the opposite has been observed for the middle-income class. These results provide important insights into value chain upgrading as the Philippines is currently struggling to reduce imports, increase rice self-sufficiency and raise income of poor farmers. Greater head rice recovery, for example, is consistently emphasized as a priority trait by consumers in the Philippines since the 1980s \[[@pone.0150345.ref044],[@pone.0150345.ref047],[@pone.0150345.ref058]\]. Our hedonic analysis revealed that greater percentage of broken rice grains (i.e. lower head rice recovery rate) is still discounted by consumers in medium and high income classes today. Therefore, in order to enable Filipino farmers to access those market segments with local rice, more investment will be needed in upgrading of pre- and post-milling operations (separation of varieties, sorting, and grading) as well as through rice breeding \[[@pone.0150345.ref087]\]. More generally, our findings can be used by rice breeders for setting priorities and incorporating grain quality improvements in varietal development, along with agronomic and stress-tolerance traits. Issues of grain quality and postharvest losses are likely to become more pronounced in the future as heat stress can reduce milled rice yields by as much as 13.8% for every 1°C increase in the average growing season temperature \[[@pone.0150345.ref088]\] and annual mean temperatures in all areas of the Philippines are expected to increase by as much as 1.1°C by 2020 and 2.2°C by 2050 compared to baseline temperatures from 1971--2000 \[[@pone.0150345.ref089]\]. Supporting Information {#sec017} ====================== ###### Distribution of chalkiness, grain length, grain width, percentage of head rice from the milled grain, and length-width ratio of the grain length of raw milled rice grains submitted by respondents from Famy and Sta Maria. (TIFF) ###### Click here for additional data file. ###### Distribution of AAC, GT, and GC of the rice samples obtained from the respondents in Famy and Sta Maria. (TIFF) ###### Click here for additional data file. [^1]: **Competing Interests:**Although one author \[OV\] is currently affiliated with a multilateral development finance institution, his contributions to this study were carried out while employed by the International Rice Research Institute and his current affiliation with the Asian Development Bank is in no way a competing interest for the study. This does not alter the authors\' adherence to PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: VOP OV. Performed the experiments: RPC OV. Analyzed the data: RPC VOP JM MD. Contributed reagents/materials/analysis tools: RPC VOP JM OV MD. Wrote the paper: RPC VOP JM OV MD. [^3]: ‡These authors also contributed equally to this work.

All relevant data are within the manuscript and its Supporting Information files. Introduction {#sec001} ============ Nest site selection is a key component of successful breeding in seabirds \[[@pone.0212668.ref001]\]. Optimal conditions at a breeding site are shaped by multiple factors, including both abiotic and biotic components as well as their interactions \[[@pone.0212668.ref002]\], operating at various scales. At macro- and meso-scales birds may search for a site with appropriate climate characteristics (e.g. time sufficient to complete the breeding season), nest site topography (e.g. appropriate site for nest building/anchoring), and biotic conditions (e.g. location of foraging areas at a cost-effective distance from the nesting site, low predation pressure). At a micro-scale birds may search for more specific climate and topographic conditions that influence nest site availability at a given time, such as nest temperature and wind exposure. Colonial seabirds often nest in coastal habitats, where, once established, colonies remain occupied in a given location for a long period of time. Seabirds typically exhibit a high site fidelity and philopatry \[[@pone.0212668.ref003],[@pone.0212668.ref004]\]. Even when conditions are temporarily unfavourable, (e.g. low food availability, high predation pressure etc.) seabirds often return to the same breeding location in the following seasons \[[@pone.0212668.ref005]\]. On the other hand, when unfavourable conditions persist for a longer time, demography of local populations can be affected. Colony size may then carry information about current environment conditions \[[@pone.0212668.ref006],[@pone.0212668.ref007]\]. Recognizing habitat preferences of seabirds, and their population size are important steps in the process of conservation management. In this study, we focus on the little auk (or dovekie, *Alle alle*), a small planktivorous seabird breeding exclusively in the High Arctic ([Fig 1](#pone.0212668.g001){ref-type="fig"}). Due to its abundance (37--40 million pairs globally) \[[@pone.0212668.ref008]\], the little auk is an important component of the High Arctic ecosystem. By transporting large amounts of organic matter from sea to land, fertilizing the nutrient-deprived Arctic tundra \[[@pone.0212668.ref009],[@pone.0212668.ref010]\], the little auk plays a role of an ecosystem engineer, transforming terrestrial ecosystems across the High Arctic \[[@pone.0212668.ref011]\]. The population of the little auk in the High Arctic may soon undergo changes due to the ongoing climate change \[[@pone.0212668.ref012]\]. In fact, modelling of the species' future distribution under scenarios of 1°C and 2°C sea surface temperature increase predicts losses of suitable foraging habitat for the majority of colonies on Svalbard, and, as a result, declines in the local populations \[[@pone.0212668.ref012]\]. Considering the ornithogenic fertilization effect of little auk colonies, the birds' retreat may have serious negative implications for marine and terrestrial Arctic ecosystems. Until now, the population size of the little auk and factors potentially affecting it are poorly recognized. Due to nesting in burrows, usually on steep coastal slopes, the little auk is a challenging species to accurately measure its population size. There are only few studies focused on the little auk colonies distribution and size (e.g. \[[@pone.0212668.ref013]--[@pone.0212668.ref016]\]). Consequently, current estimates are typically imprecise and incomplete, with habitat preferences having never been systematically examined. {ref-type="table"}. Background (bathymetry and contour map): NOAA Satellite Maps.](pone.0212668.g001){#pone.0212668.g001} The aim of our study was firstly to verify location and size of the little auk colonies on the W and NW coast of the Svalbard Archipelago (Norway). The second aim was to determine what factors are associated with colony presence/absence and size. For this purpose, we considered the following predefined factors: a) distance between the colony and main foraging grounds; b) altitude c) aspect of the slope d) angle of the slope, e) solar radiation, and f) rock type. We expected birds to nest at a cost-effective distance from the foraging grounds, with the largest colonies located closer to the main foraging grounds. Considering shadow effect of neighbouring mountains, affecting snow melting in the spring, birds should prefer southern aspects ensuring higher solar radiation and providing a faster release of nesting chambers. Considering slope stability, we expected little auks to breed on moderate slopes (20--30°), reducing risk of rolling boulders, but also avoiding flat topographies, which impede the species predator response behaviour. We expected that little auks would avoid nesting in sedimentary rock scree, built by small rocks, with fewer suitable nesting chambers and less stable slopes than locations covered with bigger rocks. Materials and methods {#sec002} ===================== The study was conducted under permission of the Governor of Svalbard. Colony surveys {#sec003} -------------- To establish location and size of little auk colonies, surveys were conducted along the N and the W coast of Spitsbergen (the biggest island in Svalbard Archipelago, Norway) during the breeding seasons between 2009 and 2015 ([Table 1](#pone.0212668.t001){ref-type="table"} and [Fig 2](#pone.0212668.g002){ref-type="fig"}). The choice of sites surveyed was based on known breeding locations (<http://svalbardkartet.npolar.no/>) and logistic constraints. At each visited site, being a potential little auk habitat \[[@pone.0212668.ref017]\], we carefully searched for evidence of bird presence (visible/audible adults and/or chicks, flocks of birds flying in circles above the scree, apparent faeces deposition etc.). Occupied colony patches were then examined to establish colony borders, based on bird presence. Patch (polygon) coordinates were marked along the border of the patch; the number of nodes depended on the complexity of local topography. A laser range-finder (TruPulse 360, USA) connected with a GPS receiver was used for setting coordinate points. Photographs (1--5 images) were taken in order to measure rock diameter within patches, using a 1 m levelling rod for a scale. This allowed for the later estimation of local nest site density. 143 colony patches were surveyed in total. {#pone.0212668.g002} 10.1371/journal.pone.0212668.t001 ###### A summary of surveyed little auk colonies and at-sea surveys performed during this study. {#pone.0212668.t001g} ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Year Month Area ------------------ -------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- *Colony surveys* 2009 July Hornsund (Ariekammen), Billefjorden (Petuniabukta) 2010 July Hornsund (Revdalen, Gulliksenfjellet) 2011 July Hornsund (Revdalen, Lechbotnen, Steinvikdalen, Treskellen, Burgerbukta, Gåshamna) 2012 June Bellsund, Wedel Jarlsberg Land (Dunderdalen) 2013 July-August Isfjorden (Bjørndalen), Prins Karls Forland, Kongsfjorden, Krossfjorden, Magdalenefjorden, Smeerenburgfjorden, Nordaustlandet (Depotodden), Sjuøyane 2015 June-July Van Keulenfjord, Alkepynten, Blomstrandhalvøya, Krossfjorden, Danskøya, Amsterdamøya, Fuglefjorden, Liefdefjord, Andøyane, Hinlopenstretet, Nordaustlandet (Zeipelodden) *At-sea surveys* 2007 06.07--05.08 \ Greenland Sea (transect central point locations in [Fig 2](#pone.0212668.g002){ref-type="fig"}) 2009 24.07--06.08 2010 22.07--03.08 ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Colony patch characteristics {#sec004} ---------------------------- For each colony patch, the following features were assessed: total area (in m2), number of nests and five environmental traits: elevation (in m a. s. l.), slope (in degrees, 0--90°), aspect (degrees, 0--360°), solar radiation in May (kWh), and geomorphology (rock types). To establish the total area of patches, patch coordinates were transformed into polygons using ArcGIS 10.3 (ESRI, Redlands, USA). Rectified aerial images (Norwegian Polar Institute) of the studied area were used for cross-reference and checking the accuracy of a polygon positioning. The size of each colony patch was calculated using the "Calculate geometry" tool in ArcGIS 10.3. The size was corrected for the slope (measured in grades) using the mean angle of the slope of each colony patch and formula: *c* = *a* × cos(*α*) ^-1^, where *a* is the original (flat) area of the polygon, *α* is the mean angle of the slope of the polygon and *c* is the new, corrected area value of the colony patch \[[@pone.0212668.ref018]\]. To identify the number of nests in patches, every patch area was multiplied by the local nest density. The local nest density was based on a linear model relating mean rock size and nest densities known from the literature \[[@pone.0212668.ref013], [@pone.0212668.ref019], [@pone.0212668.ref020]\] and this study (rock diameters obtained from scaled photographs taken *in situ* ([Table 2](#pone.0212668.t002){ref-type="table"}). An estimation of a nest density based on a stone size gives comparable results to other methods such as a video surveillance method \[[@pone.0212668.ref021]\]. The stone size-based estimation is less invasive than other nest searching methods which require long-lasting visits to colonies, and furthermore, direct nest counts are challenging because nests are typically situated deep below stones \[[@pone.0212668.ref021]\]. The estimates based on this method are not free form uncertainties, nonetheless it offers a higher degree of accuracy than knowledge available in that matter up to this day. 10.1371/journal.pone.0212668.t002 ###### Estimated density of nests (with 95% CrI in parentheses) and rock diameters in the surveyed little auk colonies. The density values marked as "this study" are the estimates of mean densities from the nest density model (see section "Colony patch characteristics" above and [S1 Table](#pone.0212668.s001){ref-type="supplementary-material"}). {#pone.0212668.t002g} ------------------------------------------------------------------------------------------------------------------------------------------------- Region Average nest density\ Average rock diameter\ Rock diameter (RD) and nest density (ND) data source \[nest/m^2^\] \[m\] ----------------------------- ----------------------- ------------------------ ------------------------------------------------------------------ Hornsund (Gulliksenfjellet) 1.99 (1.29--2.72) 0.61 RD, ND--this study Hornsund (Fugleberget) 1.28 (0.90--1.68) 0.39 RD, ND--this study N-W Spitsbergen 1.51 (1.03--1.99) 0.45 RD, ND---this study Bellsund 1.44 (0.99--1.89) 0.43 RD, ND---this study Isfjorden (Bjørndalen) 0.58 (0.21--0.97) 0.25 RD---literature (13) Kongsfjorden 1.0 \- RD data not available,\ ND---averaged from Hornsund and Isfjorden nest densities Sjuøyane 0.2 \- RD---data not available, ND---half of the Isfjorden nest density ------------------------------------------------------------------------------------------------------------------------------------------------- We built a simple linear model: $\hat{D}$ = *a* + *b* × (log(*RS*)), where $\hat{D}$ is the nest density (number of nests / 1 m^2^), and *RS* is the mean rock size (in cm) per plot. It explained more variation (*R*^*2*^ = 0.70) than a straight linear model (*R*^*2*^ = 0.61) or any other relationship. This model had a curvilinear relationship between a rock size and a nest density and allowed to estimate nest densities (along with their uncertainty) across the range of observed rock sizes except for very small ones (smaller than 10 cm in diameter), which were, however, not recorded in the studied colonies. To make inference valid with our small sample size (7 data points), we carried out a Bayesian analysis and fitted this model using MCMCglmm library \[[@pone.0212668.ref022]\] in R 3.2 (R Core Team 2015). We ran two Markov chain Monte Carlo simulations, with standard settings (13,000 iterations, a burn-in of 3,000 and a thinning rate of 10, resulting in a posterior sample of 1,000 per chain) which were fully sufficient for this small data set. Convergence was perfect as assessed visually, and with Gelman-Rubin-Brooks statistics ($\hat{R}$, 1.003 for both coefficients) computed in the coda library (\[[@pone.0212668.ref023]\]; [S1 Table](#pone.0212668.s001){ref-type="supplementary-material"} and [S1 Fig](#pone.0212668.s003){ref-type="supplementary-material"}). We then calculated expected mean nest densities and their 95% credible intervals from the posterior samples (presented in [Table 2](#pone.0212668.t002){ref-type="table"} as nest densities obtained in this study). In two cases (Kongsfjorden and Sjuøyane) exact data on the rock diameter was not available, therefore the average nest density in these cases was based on values from other colonies of similar rock structure and known rock diameter. For a single colony in Kongsfjorden we decided to conservatively apply density of 1 nest/m^2^ to avoid overestimation. This density equals averaged value of nest denisties from Hornsund and Isfjorden. The colonies at Sjuøyane have similar morphology (rock diameter) to the Isfjorden colonies, therefore we assumed that the nest density should not exceed 0.2 nest/m^2^, what is equal to the lowest nest density documented in Isfjorden by Isaksen and Bakken \[[@pone.0212668.ref013]\]. Additionally, since the colony on Amsterdamøya is placed on a vertical cliff, it was impossible to estimate the nest density using the same method as for the other colonies. Thus, the number of pairs for this colony was estimated based on observations of birds flying in the neighbourhood of the cliff. Based on literature considering colony attendance pattern of little auks during the time of the visit in the colony--mid chick rearing period \[[@pone.0212668.ref024]\], we assumed that we observed 50% of nesting adults ([Table 3](#pone.0212668.t003){ref-type="table"}). 10.1371/journal.pone.0212668.t003 ###### The surveyed colonies, and predicted number of little auk pairs. CI--confidence interval. {#pone.0212668.t003g} Region Area No of colonies Predicted no. of nests per m^2^ Predicted number of nests 5% CI 95% CI -------------------- ----------------- ---------------- ------------------------------------------ --------------------------- ------------- ------------- Bellsund Ingeborgfjellet 15 1.44 35 814 24 665 47 165 Hornsund Ariekammen 19 1.99 18 064 11 669 24 704 Fugleberget 2 1.28 76 80 5 370 10 039 Hyttevika 9 1.99 370 530 239 350 506 715 Lechbotnen 7 1.99 30 893 19 956 42 248 Revdalen 6 1.99 29 615 19 130 40 500 Rotjesfjellet 16 1.99 68 596 44 311 93 808 Torbjørnsenfjellet 6 1.99 66 513 42 965 90 960 Isfjorden Bjørndalen 1 0.58 135 49 227 Kongsfjorden Kongsfjorden 1 1.00 3 890 3 890 3 890 N-W Spitsbergen Amsterdamøya 1 ---[\*](#t003fn001){ref-type="table-fn"} 500 500 500 Fuglesangen 19 1.51 28 710 19 658 37 906 Hamburgbukuta 11 1.51 36 515 25 002 48 210 Magdalenefjorden 18 1.51 18 089 12 386 23 883 Nilsenfjellet 7 1.51 8 164 5 590 10 778 Sjuøyane Sjuøyane 5 0.20 4 821 4 821 4 821 Total **728 529** **479 312** **986 352** *\** little auks nest there on vertical cliffs; the number of pairs was estimated based on bird counts close to the cliffs Environmental traits were established in ArcGIS 10.3 using the Spatial Analyst toolbox based on Digital Elevation Models (DEMs, Norwegian Polar Institute, University of Silesia) in 20 m resolution. Solar radiation (Solar Radiation Tool) was calculcated separately for each month between May and August (little auk presence in colonies), and expressed as a sum of the radiation for these four months. Slope angle was calculated with the Slope Tool, aspect with the Aspect Tool and elevation was retrieved directly from the DEMs. Geomorphology data (type of rocks) were obtained from a geological map by the Norwegian Polar Institute \[[@pone.0212668.ref025]\]. Main foraging areas of little auks {#sec005} ---------------------------------- To identify foraging hotspots, at-sea surveys of foraging little auks were undertaken during the chick-rearing period in 2007, 2009 and 2010. The surveys were performed on board of s/y 'Oceania' research vessel (Institute of Oceanology, Polish Academy of Sciences) along the West coast of Spitsbergen in the shelf and off-shelf region of the Greenland Sea (77°--79° N, 0°-16° E). In total, 615 survey transects of different lengths (weather conditions and sail program dependent) were performed to cover potential foraging areas of birds from the examined colonies ([Fig 2](#pone.0212668.g002){ref-type="fig"} and [Table 1](#pone.0212668.t001){ref-type="table"}). Birds were counted while sailing during calm to moderate sea conditions (0--4 force Beaufort scale according to World Meteorological Organization). Counts were performed from the vessel's bridge within a range of 300 meters (i.e. from the bow of the ship to 90 degrees on the port and starboard side of the ship). All birds spotted in the established range were counted (both sitting on the water and flying). To identify the foraging hotspots, data from the three seasons of at-sea surveys was combined. The number of birds observed per 1 km^2^ was calculated by dividing the number of little auks observed along a transect by the transect area in km^2^ (transect length × 300 m). The foraging hotspots were established using the Natural Neighbour Interpolation tool in ArcGIS 10.3, with the hotspot being considered as the highest local concentrations of the birds at sea (the highest numbers of little auks observed per km^2^). The distance from the center of the colony to the foraging hotspots was calculated in ArcGIS 10.3 ([Fig 3](#pone.0212668.g003){ref-type="fig"}). {#pone.0212668.g003} Data analysis {#sec006} ============= Factors associated with colony presence/absence {#sec007} ----------------------------------------------- To investigate factors associated with colony presence/absence, we used true colony patches (n = 143) and randomly generated absence data (sites without colony patches) in ArcGIS software using the Create Random Points tool (n = 194 polygons). The absence locations were placed within 7 km from the coastline in order to cover all available potentially suitable nesting terrain. The interior (more than 7 km inland on average) was not taken into account because it is mostly covered by glaciers and ice sheets and there are no records of nesting little auks that far inland. The absence locations were placed also within 55 km distance from the foraging hotspots (twice the mean distance from the colony patch to the nearest foraging hotspot in the main concentration of little auks on the West coast of Spitsbergen: Hornsund). The size of all random polygons was the same (2 900 m^2^) and was set to the mean size of the measured colony patches. The random polygons were then validated visually based on the aerial images. The polygons located on glaciers, flat rocky islets (unlikely location of little auk colonies) or at known colony patches position were excluded leaving 194 polygons meeting all the above crtieria. For the purpose of occurrence probability modelling, mean values of environmental traits calculated from the DEM raster cells (20 m size) falling into polygons were used to produce the colony-level (polygon-level) variables. All spatial traits were ln(x+1) transformed. To explain a variation in colony presence/absence, generalized linear additive models (GAMs) with binary response and logit link function at the level of a 'breeding site' (true colony or unoccupied random polygon) were used \[[@pone.0212668.ref026]\]. With smoothing functions, GAMs are capable of capturing highly non-linear patterns between the response and predictors, as was expected and evident for two (slope and aspect) out of five environmental traits investigated here. Our global model included slope and aspect modelled with the thin plate regression splines, while solar radiation, elevation and distance to foraging hotspots were treated as linear (on the logit scale). All continuous predictors were uncorrelated, and scaled to help convergence. No collinearity was detected. GAM fitting was performed using *mgcv* library \[[@pone.0212668.ref027]\] in R (R Core Team 2015). 32 models were fitted in total with the *MuMIn* library \[[@pone.0212668.ref028]\], including the null model; models covered all possible combinations of the five predictors. Only the top four high-ranking models received substantial support with cumulative AIC weight of 0.988 ([S2 Table](#pone.0212668.s002){ref-type="supplementary-material"}). Relationships estimated from the four top-supported models were very similar ([S2 Fig](#pone.0212668.s004){ref-type="supplementary-material"}), thus model-averaging was unnecessary. We therefore present estimates and relationships from the top-supported model \[[@pone.0212668.ref029]\]. We also assessed the importance of predictors using the relative variable importance (RVI) concept, where RVIs are summed weights of models in which a given predictor occurs \[[@pone.0212668.ref030]\]. Factors associated with the size of existing colonies {#sec008} ----------------------------------------------------- To model environmental variables asssociated with a number of nests in a colony patch, a Conditional Inference Tree (CIT) was used. CIT is a non-parametric class of regression trees, examining the relationship between multiple explanatory variables and a single response variable using a recursive binary-partitioning process. Model outputs produce an 'inverted tree', in which the root at the top contains all observations, which is divided into two branches at the node. The aim of splitting the data at each step is to establish groups that had a between-variation as large, and a within-variation as small, as possible. The node provides information about the explanatory variable name and its significance. Branches are further split into two subsequent nodes and so on \[[@pone.0212668.ref031]\]. CIT uses a machine learning algorithm to determine when splitting is no longer valid using a statistically-determined stopping criterion and an a priori *p* value. This is a non-parametric class of regression tree, robust to typical regression problems such as over-fitting, collinearity, and bias with regard to the types of explanatory variables used \[[@pone.0212668.ref032]\]. We performed CIT analysis with five quantitative (distance from the foraging hotspots, elevation, slope, aspect, May solar radiation \[mean values per colony polygon\]), and one factorial (rock type) predictors (the rock type was not dependent from the rock size). This analysis was performed using *partykit* library \[[@pone.0212668.ref032]\] in R (R Core Team 2015). Results {#sec009} ======= Colony distribution and size {#sec010} ---------------------------- The little auk colonies in W and NW Spitsbergen are concentrated mainly in three areas--Hornsund (S Spitsbergen), Bellsund (central Spitsbergen) and Magdanefjorden region (NE Spitsbergen) ([Fig 4](#pone.0212668.g004){ref-type="fig"}). The 143 visited colonies covered a total area of 0.415 km^2^, and accounted for an estimated number of 728 529 (5--95% CI 479 312--986 352) nests, with estimated nest density between 0.2 and 1.99 nest/m^2^, depending on a site ([Table 3](#pone.0212668.t003){ref-type="table"}). An average colony patch covered 2 900 (SD = 7 859 m^2^), and hosted 5 094 nests on average ([Table 3](#pone.0212668.t003){ref-type="table"}). {#pone.0212668.g004} Factors associated with colony presence/absence {#sec011} ----------------------------------------------- Among 32 models fitted to the data, the top four models clearly outcompeted the remaining ones, with 99% of cumulative AIC weights ([S2 Table](#pone.0212668.s002){ref-type="supplementary-material"}). Estimated relationships were very similar ([S2 Fig](#pone.0212668.s004){ref-type="supplementary-material"}). The most important predictors were slope, elevation and solar radiation (RVI = 0.99 for all three), while aspect and distance to foraging hotspots were less important (RVI of 0.64 and 0.48, respectively). The top-supported GAM model explained 64% of the variance in the data. Birds showed a clear preference towards the steeper available slopes and avoidance of flat areas (spectrum of available slopes: 0--54°, average slope within the colonies is 28°). The probability of colony occurrence was significantly associated with elevation, with most true colonies located at low elevations. Solar radiation was also found to affect the probability of colony occurrence positively, with higher probabilities at locations with higher solar radiation. The effect of aspect was not significant due to broad confidence intervals around the estimate. However, the highest average probability of occurrence was estimated for intermediate values of aspect, that is, at southern directions (southern aspect falls within 157.5°--202.5°) ([Fig 5](#pone.0212668.g005){ref-type="fig"} and [Table 4](#pone.0212668.t004){ref-type="table"}). {#pone.0212668.g005} 10.1371/journal.pone.0212668.t004 ###### Parameter estimates of the best-supported GAM model explaining variation in colony presence/absence. Effective degrees of freedom (*edf*) is a measure of wiggliness of a smoother (a linear relationship on the logit scale has *edf* of 1). {#pone.0212668.t004g} ------------------------------------------------------ -------------- ------------ -------------- --------- **Effect** **Estimate** **SE** **Z** **p** Intercept 0.409 0.275 1.490 0.136 Altitude -3.083 0.595 -5.185 \<0.001 Solar radiation[\*](#t004fn001){ref-type="table-fn"} 1.183 0.376 3.148 \<0.001 **Smoothers** ***edf*** **Ref df** **χ**^**2**^ **p** Slope 4.528 5.517 57.751 \<0.001 Aspect 2.157 2.764 4.303 0.334 ------------------------------------------------------ -------------- ------------ -------------- --------- \*---solar radiation was included as a sum for May-August. Factors associated with colony size {#sec012} ----------------------------------- A Conditional Inference Tree characterizing little auk colony size (estimated number of breeding pairs) showed that among the studied quantitative (altitude, slope, aspect, average solar radiation in May, distance to the nearest foraging hotspot \[mean values per polygon\]) and qualitative (type of rock) predictors, only one variable: rock type, characterized significantly (p \< 0.001) colony size ([Fig 6](#pone.0212668.g006){ref-type="fig"}). Colonies located in areas with prevalence of amphibolite and (AMP) and quartzite (QUA) were larger (Node 2; mean = 17 466, 11 colonies) than those located in areas with prevalence of other types of rocks (Node 3; mean = 1 134 breeding pairs, 132 colonies). ![Conditional Inference Tree characterizing factors related to the size of the little auk colonies in W and NW coast of Spitsbergen (number of pairs) based on quantitative (altitude, slope, aspect, average solar insolation in May, distance to the nearest foraging hotspot \[mean values per polygon\]) and quantitative (type of rocks) predictors.\ Encircled variable is significantly (p \< 0.001) related to the response variable (colony size). The p value shown in an encircled node represents the test of independence between the variable (type of rocks) and the response variable (colony size). N in terminal nodes indicates the number of colonies corresponding to specific type of rocks. Histograms in terminal nodes (nodes 2, 3) depict the size of colonies in particular groups differing in type of rocks. Boxplots show the median (band inside the box), the first (25%) and third (75%) quartile (box), the lowest and the highest values within 1.5 interquartile range (whiskers) and outliers (circles). Rock type codes: AMP--amphibolite, QUA- quarzite, GNE--gneiss, GRA--granite, MAR---marble, MIG--migmatite, PHY--phyllite, SIL--silt.](pone.0212668.g006){#pone.0212668.g006} Discussion {#sec013} ========== Our study confirms the importance of the W and NW coast of Spitsbergen as the little auk breeding grounds, with the greatest breeding aggregations concentrated in three regions: Hornsund (WS Spitsbergen), Bellsund (central Spitsbergen) and Magdanefjorden (NW Spitsbergen). This result is not suprising as all these areas have been recognized as breeding hotspots for the species (Norwegian Polar Institute, [www.svalbardkartet.npolar.no](http://www.svalbardkartet.npolar.no/)). Nevertheless, we found differences when comparing the current distribution of colonies with the distribution reported historically \[[@pone.0212668.ref033]\]. At 14 sites where little auks have been reported to breed, we did not find evidence of their current presence. As there are no other seabird species exhibiting the same nesting preferences within the breeding range of little auks (burrows on mountain slopes), nest site competition is an unlikely factor driving the presence/absence of little auks. This discrepancy may instead be attributed to altering habitat properties, such as plant growth or rock erosion, limiting access to the nest burrows \[[@pone.0212668.ref034]\]. Plant overgrowth of slopes has been considered a likely reason driving the breeding site extinction in two Pacific alcids in Alaska \[[@pone.0212668.ref035],[@pone.0212668.ref036]\]. However, as the discrepancy only concerned small colonies (\<100 pairs) any conclusions about demography in those locations would be speculative, with small colonies typically prone to extinction solely by chance, without any observable changes in the environment \[[@pone.0212668.ref037],[@pone.0212668.ref038]\]. Estimation of the little auk population size, regardless of geographical scale, is challenging due to nesting in deep rock crevices, often in inaccessible areas. Our method of combining documented nest densities with the measurements of rock sizes from the colonies is not free from caveats, stemming mostly from necessary simplifications, primarily because densities of nests may vary within any colony patch (every colony has its own confidence intervals, however it was not included in the analysis), depending on the local topography, rock sizes and other factors. However, even with these limitations, it is the first estimate relying on available empirical data on rock size, providing a formal measure of uncertainty of the estimate which makes it more reliable than any other documentation in this topic up to this date. We estimated the Svalbard population at 728 529 (5--95% CI 479 312--986 352) pairs, while previous estimates reported over 1 million pairs \[[@pone.0212668.ref013]\]. On the other hand, the whole Barents Sea region population (including Svalbard) has been estimated to 580 000 pairs by other authors \[[@pone.0212668.ref039]\]. With such an enormous range of estimates of the population size it is difficult to assess whether, firstly, any temporal changes have taken place, and, secondly, to predict future numbers. Nevertheless, considering our results, and the most recent estimates of the global population (37--40 million pairs, \[[@pone.0212668.ref008]\]), the Svalbard population of little auk accommodates approximately 1.9% (95% CI: 1.2%--2.7%) of the global population ([Table 5](#pone.0212668.t005){ref-type="table"}). Even if the value *per se* may not seem large, particularly when compared with Greenland that hosts the core of the global population (\~89%), it still represents an extremely important site for the species, which, in the Svalbard Archipelago is one of the most important components of the marine and terrestrial ecosystem. This is because the little auk is the most numerous seabird in the Svalbard area \[[@pone.0212668.ref017],[@pone.0212668.ref040]\] and acts as an ecosystem engineer transporting marine-derived nutrients from sea to land, transforming terrestrial ecosystems and further affecting benthic coastal communities by a nutrient runoff from colony areas \[[@pone.0212668.ref010],[@pone.0212668.ref041],[@pone.0212668.ref042]\]. Given this, if recently proposed scenarios predicting noticeable decrease of the little auks in most Svalbard populations \[[@pone.0212668.ref012]\] occur, considerable ecosystem changes in various Svalbard regions are expected in future. 10.1371/journal.pone.0212668.t005 ###### Breeding population estimates for the little auks according to available literature and this study (non-additive). {#pone.0212668.t005g} -------------------------------------------------------------------------------------------------------------------------------------------------- Region No. of breeding pairs \% of world population Reference ------------------------------------------- ------------------------ ------------------------ ---------------------------------------------------- ***Svalbard*** Svalbard (the whole archipelago) \>1 000 000 \>3.7--4.0 \[[@pone.0212668.ref016]\] Nordaust-Svalbard 1 190 \[[@pone.0212668.ref043]\] Bjørnøya 10 000 \[[@pone.0212668.ref044]\] Nordvest-Svalbard Nasjonalpark, Forlandet 1 500 000 \[[@pone.0212668.ref045]\] Sør-Spitsbergen National Park (Hornsund) 20 100 \[[@pone.0212668.ref046]\] Hornsund 200 000 \[[@pone.0212668.ref047]\] Sør-Spitsbergen National Park (Hornsund) 886 000 \[[@pone.0212668.ref013],[@pone.0212668.ref016]\] Edgeøya 450 \[[@pone.0212668.ref048]\] \ **Spitsbergen--this study, mean values** Bellsund 35 814 This study Hornsund 591 892 This study NW Spitsbergen 91 978 This study Isfjorden (Bjørndalen) Kongsfjorden\ 135\ This study\ Sjuøyane 3 890\ This study\ 4 821 This study the whole Spitsbergen 728 529 1.8--2.0 This study ***East Barents Sea*** Whole Barents Sea region \>1 300 000 3.2--3.5 \[[@pone.0212668.ref033]\] Whole Barents Sea region 580 000 1.4--1.5 \[[@pone.0212668.ref039]\] Novaya Zemlya 11 000 \[[@pone.0212668.ref049]\] Novaya Zemlya 50 000 \[[@pone.0212668.ref050]\] Novaya Zemlya 30 000--50 000 \[[@pone.0212668.ref051]--[@pone.0212668.ref053]\] Severnaya Zemlya 10 000--80 000 \[[@pone.0212668.ref054]\] Franz Joseph Land 250 000 \[[@pone.0212668.ref055]\] Franz Joseph Land 30 000--50 000 \[[@pone.0212668.ref017]\] ***Jan Mayen*** 10 000--20 000 \[[@pone.0212668.ref050]\] ***Greenland*** Whole Greenland 23 500 000--38 500 000 63.5--96.2 Scoresby Sund (E Greenland) \>3 500 000 \[[@pone.0212668.ref015]\] Thule (W Greenland) 20 000 000--33 000 000 \[[@pone.0212668.ref056],[@pone.0212668.ref057]\] Northumberland Island\ 8 834 919 \[[@pone.0212668.ref014]\] (Thule, W Greenland) World population estimate 37 000 000--40 000 000 100 \[[@pone.0212668.ref008]\] -------------------------------------------------------------------------------------------------------------------------------------------------- As we expected, the little auk colonies along the NW Spitsbergen coast showed a non-random distribution. The probability of colony occurrence was significantly associated with elevation, and was the highest at low and moderate altitudes and moderate at high altitudes. This was likely to be related to appropriate rock or substrate sizes being distributed mainly at low and moderate elevations \[[@pone.0212668.ref016],[@pone.0212668.ref017],[@pone.0212668.ref033]\]. Solar radiation was the next important factor associated with presence/absence of little auk breeding colonies on Svalbard. The colonies located in southern Spitsbergen were exposed to higher solar radiation values than those in NW Spitsbergen. This parameter may be easily translated to overall air temperature, particularly important for birds in spring, when it affects the timing of snow melt, which in turn affects the onset of the breeding season \[[@pone.0212668.ref058]\]. For example, median hatching date in Isfjorden (central W Spitsbergen) is usually \>1 week earlier than in Hornsund (SW Spitsbergen \[[@pone.0212668.ref059]\]) due to an earlier nest chamber availability facilitated by snow melt. Slope aspect exhibited very broad confidence intervals (possibly due to a restricted range of aspect values in true colonies) and, although did not appear significant in the statistical sense, the highest probability of colony occurrence on average fell within W-SW aspects. Such a nest exposure was in line with expectations: favouring faster snow melting during spring, and acting in a similar way to that of solar radiation, may affect the probability of the little auk colony occurrence. The probability of colony occurrence was significantly associated with slope. Moderate slopes provide relatively stable ground, where rocks and stones sliding is limited. Moderate slopes are more likely to provide more nest chambers than steeper slopes or vertical cliffs. Moderate slopes also facilitate easier take-off, which may be particularly important for short-winged seabirds, like little auks, with an unfavourable ratio of wing area to body mass \[[@pone.0212668.ref017]\]. In this context, it was not surprising that the average value of slope recorded in the little auks colonies (28.4°) was similar to values reported for other alcids; the least and crested auklets (*Aethia pusilla*, *Aethia cristatella*) nesting in Pacific colonies \[23° \[[@pone.0212668.ref060]\]\]. Generally, seabird colony size and location depend to some degree on local food availability \[[@pone.0212668.ref061],[@pone.0212668.ref062]\]. The waters off the W coast of Spitsbergen are considered to be optimal foraging grounds for little auks \[[@pone.0212668.ref063],[@pone.0212668.ref064]\]. This is either because of the strong influence of cold Arctic currents running along the E coast of Spitsbergen and then along the W coast (Hornsund, Bellsund; [S3](#pone.0212668.s005){ref-type="supplementary-material"} and [S4](#pone.0212668.s006){ref-type="supplementary-material"} Figs) or a relatively close distance to the marginal ice zone (Magdalenefjorden; [S5 Fig](#pone.0212668.s007){ref-type="supplementary-material"}) offering a high availability of energy-rich zooplankton \[[@pone.0212668.ref065]--[@pone.0212668.ref067]\]. However, distance to foraging hotspots was not significantly related to either colony presence/absence probability or colony size, which is best explained by flexibility of little auks to forage in suboptimal or more distant feeding areas under unfavourable foraging conditions (e.g. \[[@pone.0212668.ref012],[@pone.0212668.ref062]\]). In other words, it is possible, that despite considerable differences in location of the foraging hotspots in relation to the studied breeding colonies, these grounds are still within the foraging range of the little auk. Our study revealed that local population size is associated with rock type. This finding may result from the patchy distribution of appropriate rock substrate \[[@pone.0212668.ref025]\]. Worth highlighting is, that the rock type is not dependent from the rock size. Regarding colonies found on sedimentary rocks in Isfjorden and Bellsund, a morphological difference can be observed. The average size of the rocks in Bellsund was approximately two times bigger compared to Isfjorden. In Bellsund the colonies were dense with \>35 000 pairs of little auks, while the BjØrndalen (Isfjorden) colony was very small. This may be a result of mass wasting and mud slides that are more pronounced in this area than in other places, preventing the birds from establishing stable colonies that persist for many years. Mass wasting and rock erosion are known to be more pronounced in sedimentary rocks due to their porosity \[[@pone.0212668.ref025],[@pone.0212668.ref068]\]. However, colonies in Bellsund comprise the third largest little auk concentration in Svalbard. The silts that build the colonies in Bellsund provide much more stable grounds than the sandstones in Isfjorden. Rocks in Bellsund are also larger and offer more burrow systems suitable for nesting than in Isfjorden. We conclude therefore, that the fast erosion of sandstones may limit the Isfjorden population. The largest little auk colonies were found on metamorphic rocks (quartzite and amphibolite). The colonies in Hornsund contain mostly phyllites and quartzite, whereas in NW Spitsbergen they contain migmatite and granite (Bellsund and Isfjorden are built almost exclusively of sedimentary rocks). Our findings show that the birds form larger colonies on metamorphic rocks, probably due to the stability of the slopes and higher resistance to erosion (therefore, bigger rocks on the slope providing more nesting chambers) \[[@pone.0212668.ref068]\]. Little auks are gregarious and require colonial nesting for successful breeding, so they likely prefer locations with rock types where it is possible to establish large, dense colonies \[[@pone.0212668.ref017],[@pone.0212668.ref034]\]. However, since different rock types are not evenly distributed along the W coast of Spitsbergen, the birds do not have the whole spectrum of rocks to choose from at every location. Therefore, conclusions about a preference towards a certain rock type should be treated with caution. Conclusions {#sec014} =========== With 728 529 (5--95% CI 479 312--986 352) breeding pairs in total, the W and the NW Spitsbergen populations of little auks represent the most important seabird aggregations in the Svalbard Archipelago. Spatial distribution of the colonies along Spitsbergen coast is not random but rather determined by elevation, solar radiation and slope. The size of the particular colonies is also associated with geomorphological factors such as rock type but not with the distance to foraging hotspots. Knowledge of breeding population distribution and size of this Arctic endemic species is crucial for current population status assessments and any future conservation management. Supporting information {#sec015} ====================== ###### Parameters estimated for the relationship between mean rock size and nest density using a Bayesian linear model ($\hat{D}$ = a + b \* (log(*RS*)). (DOCX) ###### Click here for additional data file. ###### Generalized additive linear models fitted to data to explain variation in little auk colony presence. '+' indicates whether a given term is included in the model. Models are ranked according to AIC value (lower AIC = bigger support). All models contain an intercept (omitted in the table for clarity). (DOCX) ###### Click here for additional data file. ###### Relationship between mean rock size and nest density according to the linear model applied in this work. Line--mean, grey area-- 95% CrI, points--original observations. The bigger symbol denotes two observations. (TIF) ###### Click here for additional data file. ###### Relationships between predictors and the probability of little auk colony occurrence in Spitsbergen according to top four models. Slope and aspect were modelled with smoothers, remaining relationships are linear on logit scale. The relationship from the top-supported model is shown with bold, remaining three top-supported models with thin curves. Note that not all predictors are present in all four models. (TIF) ###### Click here for additional data file. ###### Detailed map with the colony distribution in Hornsund (65 colonies), southern Spitsbergen. Background (contour map): NOAA Satellite Maps. (TIF) ###### Click here for additional data file. ###### Detailed map with the colony distribution in Bellsund (15 colonies), central Spitbsergen. Background (contour map): NOAA Satellite Maps. (TIF) ###### Click here for additional data file. ###### Detailed map with the colony distribution in NW Spitsbergen (56 colonies). Background (contour map): NOAA Satellite Maps. (TIF) ###### Click here for additional data file. ###### Mean size of the colonies (number of pairs) located in different types of rocks. The number of colonies within specific groups given above the columns. Yellow columns correspond to the second node in Conditional Inference Tree ([Fig 6](#pone.0212668.g006){ref-type="fig"}). Whiskers---standard deviation. (TIF) ###### Click here for additional data file. ###### Spreadsheet with colonies (presence: 1) and absence locations (absence: 2) and environmental factors mean values. (XLSX) ###### Click here for additional data file. ###### GIS Shapefile with randomly selected polygons without colonies (absence). (ZIP) ###### Click here for additional data file. ###### GIS Shapefile with colonies. (ZIP) ###### Click here for additional data file. ###### GIS Shapefile with foraging hotspots centroids. (ZIP) ###### Click here for additional data file. The study was conducted under permission of the Governor of Svalbard. We are grateful to Professor Lech Stempniewicz for valuable comments to the manuscript. We are grateful to Adam Nawrot for help during field studies. We appreciate logistic support from Polish Polar Station in Hornsund. We are indebted to the captain and crew of the R/V Oceania (IOPAS) and for their assistance during cruises. We acknowledge the use of imagery from the NOAA Satellite Maps application: [www.nesdis.noaa.gov](http://www.nesdis.noaa.gov/). We appreciate the English proofreading made by George Day and Daniel Schönberger. We also thank Reviewers whose comments improved the manuscript. [^1]: **Competing Interests:**The authors have declared that no competing interests exist.

INTRODUCTION ============ Rhabdomyolysis is the breakdown of skeletal muscle that can rapidly progress to acute renal failure or death. It is important to make a rapid diagnosis and initiate treatment for rhabdomyolysis in order to decrease morbidity and mortality. To date there are no reports in the emergency medicine literature on the use of point-of-care ultrasound in the diagnosis of rhabdomyolysis. This unique case describes a patient who presented to the emergency department (ED) with localized musculoskeletal pain. Using ultrasound, the patient was quickly diagnosed and treated for rhabdomyolysis prior to confirmation with an elevated serum creatine phosphokinase (CPK). When coupled with a high index of suspicion, ultrasound can be one of the most portable, readily available, low-cost, and minimally invasive techniques for making a rapid diagnosis of rhabdomyolysis in the ED. CASE REPORT =========== A 24-year-old male presented to the ED with a two-day history of bilateral arm pain. The pain was constant, located primarily to the biceps region of his upper arms. His pain began shortly after weight lifting. Past medical history included bipolar disorder and polysubstance abuse, including recent use of cocaine, marijuana, and a synthetic marijuana known as "spice." On arrival to the ED the patient had a temperature of 97.9 F, blood pressure of 167/88, heart rate of 122, and respirations of 16 per minute. Physical findings included bilateral biceps swelling and fullness with diffuse tenderness to the musculature. There was no external evidence of trauma. The patient's upper extremities were neurovascularly intact with no evidence of paresthesias, weakness, or pallor. He was noted to display paranoid behavior without features of acute psychosis. A point-of-care musculoskeletal ultrasound was performed by the emergency physician to evaluate for a possible muscle or biceps-tendon tear. The sonogram showed areas of both increased and decreased echogenicity of the biceps muscle, as well as disorganized muscle fibers with surroundings areas of fluid ([Figure 1](#f1-wjem-17-801){ref-type="fig"}, [Video 1](#f3-wjem-17-801){ref-type="fig"}). There was preservation of the muscle boundary and the biceps tendon was intact. A presumptive diagnosis of rhabdomyolysis was made pending laboratory testing and the patient was started on intravenous (IV) fluids. The patient's lab work in the ED was notable for a creatine phosphokinase (CPK) of 83,000 U/L. AST/ALT were 813/169 IU/L respectively, total bilirubin of 1.5 mg/dl, and bicarbonate of 16 mEq/L. Urine results included amber color and "large" hemoglobin with 3--5 red blood cells per high powered field. A complete blood count, complete metabolic profile, and urinalysis were otherwise unremarkable. A urine drug screen was positive for marijuana. The patient was hospitalized and treated with aggressive IV fluids. His CPK peaked at 124,000 and subsequently improved daily thereafter. His liver enzymes improved as well. A hepatitis panel led to a new diagnosis of hepatitis C, though liver enzyme elevation was thought to be primarily related to his acute rhabdomyolysis. Renal function remained normal throughout his stay. His hospital course was complicated by paranoia, psychosis, and aggression toward hospital staff. He was subsequently placed under an involuntary psychiatric hold and required security intervention and sedation. He was transferred to the local mental health facility on hospital day five for further psychiatric care after his rhabdomyolysis resolved. DISCUSSION ========== Rhabdomyolysis is a syndrome of skeletal injury that can rapidly progress to acute renal failure or even death.[@b1-wjem-17-801],[@b2-wjem-17-801] It is imperative that it be diagnosed and treated appropriately to reduce the lasting effects of the disease.[@b1-wjem-17-801] Rhabdomyolysis results from a myriad of conditions including trauma, illicit drugs, medications, infection, excessive exercise, immobilization and psychiatric condition.[@b2-wjem-17-801] The symptoms of rhabdomyolysis are as variable as the causes: from a patient with a traumatic crush injury, who has pigmented urine and renal failure, to one with no significant history of trauma who may simply present with fatigue, nausea, fever or muscle weakness.[@b1-wjem-17-801],[@b2-wjem-17-801] For these reasons, diagnosing rhabdomyolysis cannot be based on history and physical alone and generally requires a serum CPK that is greater than five times the normal limit, without evidence of brain or cardiac injury. [@b3-wjem-17-801] Once a diagnosis of rhabdomyolysis is suspected, additional workup is needed. An electrocardiogram should be performed to screen for conduction abnormalities caused by hyperkalemia including: peaked T waves, prolonged PR interval and a widened QRS complex.[@b2-wjem-17-801] A metabolic panel should be performed to assess renal function, calcium, potassium and phosphorous levels.[@b1-wjem-17-801] CBC and coagulation studies should be done to monitor for evidence of DIC.[@b1-wjem-17-801] Imaging may be required in patients who present with localized pain. Ultrasound and magnetic resonance imaging (MRI) can both be used to determine the extent of muscular damage that exists. On ultrasound normal skeletal muscle has a relatively hypoechoic echotexture with clearly demarcated linear hyperechoic strands of fibroadipose septa[@b4-wjem-17-801] ([Figure 2](#f2-wjem-17-801){ref-type="fig"}). In rhabdomyolysis the findings are variable but may include hyperechoic areas of muscle,[@b5-wjem-17-801],[@b6-wjem-17-801] which is likely due to hypercontractile muscle fibers in the acute phase of muscle injury,[@b4-wjem-17-801] hypoechoic areas of muscle,[@b4-wjem-17-801],[@b7-wjem-17-801] which is believed to be caused by edema and inflammation of the muscle,[@b4-wjem-17-801],[@b7-wjem-17-801] increased muscle thickness, and fluid within the surrounding the muscles.[@b4-wjem-17-801] Areas of locally disorganized fascicular architecture[@b8-wjem-17-801] may also be appreciated but generally the muscle boundary itself remains intact unless an associated tear is present.[@b8-wjem-17-801] The area of locally disorganized fasicular architecture is thought to represent necrosis of the muscle.[@b4-wjem-17-801],[@b7-wjem-17-801] Abscesses and hematomas may also appear anechoic and hypoechoic and should be considered on the differential diagnosis.[@b5-wjem-17-801] However, on examination neither of these conditions correlates with the clinical picture. With both an abscess and hematoma one would expect a localized fluid collection on ultrasound in conjunction with physical exam findings of infection and trauma respectively. Additionally, there would not be a significantly elevated serum CPK or increase in creatinine level associated with these diagnoses. Clinicians performing soft tissue ultrasound should be familiar with the sonographic changes associated with other common musculoskeletal pathologies such as muscle tears, strains, and contusions in order to avoid diagnostic error. Complete muscle tears are associated with avulsion and retraction of the injured muscle segment along with an adjacent anechoic or hypoechoic hematoma.[@b9-wjem-17-801] Partial tears, strains, and contusions have focal hypoechoic areas within the muscle fibers themselves, representing localized edema and hemorrhagic changes that have disrupted the normal fibroadipose pattern.[@b9-wjem-17-801] Unlike in rhabdomyolysis, the sonographic findings seen with tears, strain injuries, and contusions are limited to a focal area of injury and are not diffusely present throughout the involved muscle. MRI is excellent at demonstrating rhabdomyolysis with T2-weighted MRI images of rhabdomyolysis generally show increased signal intensity.[@b8-wjem-17-801] Unfortunately performing an MRI is associated with greater cost and long testing times. In addition, MRI is not readily available in most EDs. The initial treatment of rhabdomyolysis involves early and aggressive IV fluid resuscitation.[@b2-wjem-17-801] The patient should be admitted and monitored for life-threatening complications such as acute renal failure and hyperkalemia. In this case, the patient presented with bilateral upper extremity pain and swelling and the ultrasound was used to narrow the differential of possible musculoskeletal pathologies. A presumptive diagnosis of rhabdomyolysis was made and later confirmed by serum CPK levels. If a patient has sonographic findings suggestive of rhabdomyolysis, treatment can be initiated immediately while laboratory testing is still in progress. The increasing availability of point-of-care ultrasound in most EDs makes it an ideal imaging modality for the initial evaluation of the patient in whom rhabdomyolysis is considered. LIMITATION ========== There are no follow-up ultrasounds available to show this patient's return to normal muscles. *Section Editor:* Rick A. McPheeters, DO Full text available through open access at <http://escholarship.org/uc/uciem_westjem> *Conflicts of Interest:* By the *West*JEM article submission agreement, all authors are required to disclose all affiliations, funding sources and financial or management relationships that could be perceived as potential sources of bias. The authors disclosed none. {#f1-wjem-17-801} {#f2-wjem-17-801} ###### This ultrasound video is of the patient's left bicep muscle. It shows disorganized muscle fibers with surrounding areas of anechoic fluid. The biceps muscle displays areas of both increased and decreased echogenicity. There is preservation of the muscle boundaries within the echogenic fascial planes.

INTRODUCTION {#sec1-1} ============ L-carnitine is an essential factor for the membrane transport of acyl-CoA compounds, particularly for the intramitochondrial transport of long-chain fatty acids \[[@ref1]\]. L-carnitine also helps to remove by-products of fatty acid metabolism and other toxic compounds from the cells \[[@ref2]\]. The liver and kidney represent the main sources of endogenous carnitine synthesis \[[@ref3]\]. Also among the homeostatic processes controlling the endogenous L-carnitine pool in humans, the kidney has a vital role through extensive and adaptive tubular reabsorbtion \[[@ref4]\]. Kidney disease can lead to disturbances in L-carnitine homeostasis, and carnitine deficiency may occur in hemodialysis patients. Bellinghieri et al reported that hemodialysis may promote excessive losses of L-carnitine \[[@ref5]\]. Oxidation of fatty acids and lipid metabolism are severely affected in carnitine defi-ciency \[[@ref6]\]. Aberrant fatty acid metabolism has been associated with the pro-motion of free radical production, insulin resistance and cellular apoptosis \[[@ref7]\]. Oxidative stress, by definition, is a biochemical condition in which oxidant species overwhelm antioxidant defense ultimately leading to a given biological damage \[[@ref8],[@ref9]\]. Galli F, et al indi-cated the evidence for the presence of oxidative stress in hemodialysis patients \[[@ref10]\]. In these patients multiple factors can lead to a high susceptibility to oxidative stress \[[@ref11],[@ref12]\]. One of the most investigated biological effects of the oxidative stress in hemodialysis patients is lipid peroxidation \[[@ref13]\]. In the view of the above, the aim of this study was to determine the hemodialysis ef-fectiveness on the change rate of serum L-carnitine and lipid peroxidation. MATERIALS AND METHODS {#sec1-2} ===================== Chemicals {#sec2-1} --------- Thiobarbituric acid (TBA), Buthylated Hydroxy Toluene (BHT), and other reagents are purchased from Sigma (Deisenhofen, Germany) or Merck (Darmstadt, Germany). L-carnitine was assayed by the ready to use kit from ROCHE. Subjects {#sec2-2} -------- 27 patients with chronic renal failure on hemodialysis were included in the study. All patients were dialyzed 3 times/wk (4 hrs sessions). The duration of dialysis was 6-12 months. The age of patients was 24-48 yrs. The patients were not taking any antioxidants. Diabetic patients were excluded from the study. The demographic information was written in a check-list form. gave The informed consent was obtained from all hemodialysis subjects included in the study. Venous blood samples were taken into EDTA-tubes (1 g/l) after an overnight fast immediately before and after the dialysis session. Plasma was separated by centrifugation at 3000 g for 5 min at 4**°** C. Methods {#sec2-3} ------- Malondialdehye (MDA), as an indicator of lipid peroxidation was measured colorimetrically with standard TBA method at 532 nm wavelenght \[[@ref14],[@ref15]\]. L-carnitine concentration was measured with enzymatic UV-method (Genesis, 340 nm). The reference range for L-carnitine in humans was 3.85 ± 0.82 mmol/L \[[@ref16]\]. Statistical analysis {#sec2-4} -------------------- All results were presented as mean ±SD. Levene test, paired t-test, independent t-test were used for rejection or acceptance of the assumption. Pearson's correlation coefficient was calculated. RESULTS {#sec1-3} ======= The weight mean of serum L-carnitine, before and after hemodialysis was 7.67±3.6 and 2.07±1.6 mg/l, respectively. In this study 55.6% of patients suffered from carnitine deficiency (P\<0.001). Pearson's correlation coefficient indicated the relationship between the value of serum L-carnitine before and after hemodialysis (r=0.4, P=0.045). The significant difference between the value of serum L-carnitine before and after hemodialysis was observed (P\<0.001) ([Table 1](#table001){ref-type="table"}). Pre-hemodialysis and post-hemodialysis serum MDA was 4.17±1.24 µmol/l and 4.98±1.2, respectively. Pearson correlation coefficient indicated that there is a strong relationship between serum MDA before and after hemodialysis (r=0.96, P\<0.01) ([Table 2](#table002){ref-type="table"}). The relationship was found between average of serum L-carnitine and serum MDA before and after hemodialysis (r=0.82; p\<0.001; r=0.75; p\<0.001). The confidence interval of percentage of change in average of L-carnitine and MDA is shown in Fig.1. DISCUSSION {#sec1-4} ========== Existing evidence showed that despite advances in dialysis therapy, a high percentage of patients on maintenance dialysis therapy, suffered from complication of hemodialysis \[[@ref17]\]. The study of Evans et al showed, that the average of predialysis serum L-carnitine concentration was 19.5 µmol/l, where the post dialysis level was 5.6 µmol/l \[[@ref4]\]. Also, Alhomida's research indicated that the average of serum L-carnitine before and after hemodialysis was 42±6.3 µmol/l and 17.1±6.3 µmol/l, respectively \[[@ref18]\]. The results of this study were consistent with the studies of others and showed a decrease of L-carnitine after hemodialysis. L-carnitine is a small water-soluble molecule, therefore it is freely dialyzed because of a molecular weight gradient; the acyl-carnitine moieties are less likely to be filtered by the membrane than free carnitine \[[@ref4],[@ref19]\]. Accumulation of acyl-carnitine is belived to contribute directly to arrythmogenesis \[[@ref20]\]. This accumulation also alters mitochondrial membrane permeability and has been suggested to promote apoptosis. Altered membrane permeability has been implicated to modify the activity of various hormone receptors, including insulin receptors \[[@ref21]\]. The susceptibility for oxidative stress is mainly correlated with MDA concentration \[[@ref22],[@ref23]\]. It has been reported that the hemodialysis procedure altered lipid peroxidation. Ozden et al found that MDA was elevated in post-hemodialysis (1.39±0.38 nmol/ml) in comparison to prehemodalysis state (0.83±0.22 nmol/ml) \[[@ref24]\], but Nand and Surri showed that MDA was decreased after hemodialysis (pre hemodialysis 2.96±0.89 µmol/l and post hemodialysis 2.32±0.84 µmol/l) \[[@ref25]\]. In this study we observed an increase of MDA, which possibly was the result of free radical reactions during the hemodialysis procedure. The absence of a complete correction of the uremic toxicity together with the untoward effect of the dialysis, malnutrition and the progressive worsening of the clinical condition can lead to a high susceptibility to oxidative stress \[[@ref26],[@ref27]\]. Increased lipid peroxidation in hemodialysis patients is largely a result of anemia of kidney failure, therefore anemia management can improve this condition \[[@ref28],[@ref29]\]. Chronic heart failure in hemodialysis patients is of high prevalence, it was indicated that lipid peroxidation and carnitine deficiency are main risk factors in some patients \[[@ref30]\]. In this study 55.6% of patients suffered from carnitine deficiency before hemodialysis and serum L-carnitine decreased markedly after hemodialysis (P\<0.001). In these patients, oxidative stress is further exacerbated by hemodialysis, as evidenced by increased lipid peroxidation (P\<0.001). Multiple pathogenic factors are responsible for intra-dialysis muscle cramping. Carnitine deficiency is a potential cause of intra-dialysis muscle cramping \[[@ref31],[@ref32]\]. Also oxidative damage may play a role in the pathogenesis of skeletal myopathy in hemodialysis patients \[[@ref33]\]. Oxidative stress has been proposed to play a role in many diseases states, including cardiovascular and infectious diseases \[[@ref34],[@ref35]\], cancer \[[@ref36]\], diabetes \[[@ref37]\] and neurodegenerative diseases \[[@ref38]\]. Therefore oxidative stress management and supplementation with L-carnitine, parallel to other therapeutic agents, may improve condition of hemodialysis patients. {#fig001} ###### Mean and SD of L-carnitine concentration in females and males. Parameter Sex Number Mean±SD Weight mean P-value ------------------------ --------- ---------- ----------- ------------- --------------------------------------------- L-carnitine Males 17 8.96±3.66 7.67±3.6 0.053 (before hemodialysis) Females 10 5.93±2.88 L-carnitine Males 17 2.66±1.67 2.07±1.6 0.01[\*](#table1-tfn1){ref-type="table-fn"} (after hemodialysis) Females 10 1.06±0.85 L-carnitine difference Males 17 6.3±3.69 \-- 0.045 Females 10 4.87±2.6 \* difference between males and females; statistical tests: independent t-test and Levene test ###### Mean and SD of MDA concentration in females and males Parameter Sex Number Mean±SD Weight mean P-value ----------------------- -------- --------- ----------- ------------- ---------------------------------------------- MDA Male 17 3.75±0.91 4.17±1.24 0.048[\*](#table2-tfn1){ref-type="table-fn"} (before hemodialysis) Female 10 4.87±1.45 MDA Male 17 4.63±1.11 4.98±1.2 0.05[\*](#table2-tfn1){ref-type="table-fn"} (after hemodialysis) Female 10 5.57±1.16 MDA difference Male 17 0.88±0.28 \-- 0.05 Female 10 0.7±0.4 \* difference between males and females; statistical tests: independent t-test and Levene test

1. Introduction {#sec1-biosensors-10-00039} =============== Capillary-driven flow microfluidics is a type of microfluidics which allows fluid movements into microchannels via capillary forces and does not require internal/external expensive and complicated fluid management mechanisms \[[@B1-biosensors-10-00039]\]. This type of microfluidics has the potential to provide rapid, inexpensive and simple clinical assays in short times and near the patient. The reagents can be pre-coated on to the surface of the channels, and results are obtained via a portable detector/smartphone by merely dipping the device into the patient sample, after a short incubation. Therefore, this type of system can lead to the development of the ever-promised simple point-of-care (POC) devices sought after by the scientific community for decades \[[@B2-biosensors-10-00039],[@B3-biosensors-10-00039]\]. Several capillary-driven flow devices have been developed and tested for POC diagnostics applications, such as that of Ramalingam et al. \[[@B4-biosensors-10-00039]\], who developed a numerical model to study capillary-driven flow in capillaries for polymerase chain reaction (PCR). The authors numerically modelled and tested polydimethylsiloxane (PDMS) microchips and validated the flow inside the capillaries by tracking the fluid meniscus. Capillary-driven flow devices were also fabricated \[[@B5-biosensors-10-00039],[@B6-biosensors-10-00039]\] to measure the viscosity of a fluid based on capillary action in microchannels. For example, Lee et al. \[[@B5-biosensors-10-00039]\] developed a capillary-driven flow microfluidics device to measure zebrafish blood viscosity in microchannels. The method allowed the validation of the Newtonian fluid behavior and dynamic viscosity of blood, requiring a minute amount of blood from zebrafish. In another development, PMMA was also used to develop a capillary flow device for nucleic acid biosensing applications using 500 µm-wide microfluidic channels consisting of sealed reagent-loaded pads \[[@B7-biosensors-10-00039]\]. Furthermore, capillary-driven flow microfluidics has also been developed for the measurement of biomedically relevant biomarkers \[[@B8-biosensors-10-00039],[@B9-biosensors-10-00039],[@B10-biosensors-10-00039],[@B11-biosensors-10-00039],[@B12-biosensors-10-00039],[@B13-biosensors-10-00039]\]. The dynamics of open microfluidic channels has also been studied via 3D printing of microchannels for rapid prototyping and mass fabrication options \[[@B14-biosensors-10-00039]\]. Additionally, capillary flow has also been used for blood plasma separation in microfluidic channels, as reported by Madadi et al. \[[@B13-biosensors-10-00039]\], who developed a capillary flow device to separate plasma using 5 µL of sample to obtain 0.1 µL of plasma for diagnostics applications. Delamarche's group \[[@B9-biosensors-10-00039],[@B15-biosensors-10-00039],[@B16-biosensors-10-00039]\] has widely developed plasma separation devices combined with immunodiagnostics devices, such as a system to detect C-reactive protein (CRP), which was quantified by using 5 µL of human serum extracted from a blood sample and 3.6 nL of a reagent solution deposited on the chip \[[@B15-biosensors-10-00039]\]. This type of device has been further developed for multiple biomedical applications such as portable bead-based and immunodiagnostics assays, with the possibility of detection via smartphones or handheld devices \[[@B16-biosensors-10-00039]\]. Glass/hydrophilic capillaries have also been used to drive flow via capillary action, as described by Lapierre et al. \[[@B17-biosensors-10-00039]\], who used bare glass capillaries to collect blood samples. In contrast, fluoropolymer microcapillaries (FEP) have been coated with reagents to render their surface hydrophilic and draw up blood or aqueous samples in a minute fraction of time \[[@B18-biosensors-10-00039],[@B19-biosensors-10-00039],[@B20-biosensors-10-00039],[@B21-biosensors-10-00039],[@B22-biosensors-10-00039]\]. Pivetal et al. \[[@B19-biosensors-10-00039]\] coated FEP capillaries with polyvinyl alcohol (PVA) to convert their surface from hydrophobic to hydrophilic and attached reagents or antibodies on the surface for the detection of protein biomarkers. The reagents reacted with the biomarkers, generating a colour or fluorescent signals which were detected under a microscope attached to a camera. Similarly, FEP microcapillaries were used to assess prostate-specific antigen (PSA) by an enzymatic reaction \[[@B20-biosensors-10-00039],[@B21-biosensors-10-00039]\] and cytokines \[[@B22-biosensors-10-00039]\] within microcapillaries. The FEP microcapillaries were multiplexed in parallel by injecting solutions into 10 capillaries at the same time, i.e., a PSA standard solution, detection antibodies, the enzyme complex, washing solutions and the enzymatic substrates, which were injected in the capillaries simultaneously. The FEP microcapillaries were placed vertically in a blood sample to draw up liquid for ABO blood typing \[[@B18-biosensors-10-00039]\]. As the liquid rose into the capillaries, the reagents were released into the sample and reacted with biomarkers to produce a colour or a fluorescent signal, which was then detected by microscope or portable/smartphone systems. Capillary-driven flow microfluidics have a great potential as POC diagnostics, for instance, for the prevention of antimicrobial resistance in healthcare \[[@B2-biosensors-10-00039],[@B18-biosensors-10-00039]\]. Here, we developed a capillary-driven flow device which is simple to operate and allows loading multiple samples in a single device. In this study, the design, fabrication and working principle of the capillary-driven flow device are illustrated along with applications consisting in β-lactamase and cortisol activity assays. 2. Experimental {#sec2-biosensors-10-00039} =============== The microchip was designed using the CAD software (SolidWorks, Dassault Systemes), and 8 cm-long microchannels were micromachined on a PMMA piece (1.2 mm of thickness) with dimensions of 0.25 × 0.20 mm (width × depth). Another PMMA piece with similar dimensions was cut for sealing the microchannels. Both pieces were sonicated in a water bath to completely remove the debris from the microchannels or surface and were cleaned with isopropyl alcohol (IPA). The microchannels were air-dried and placed on a glass Petri dish, with the microchannels facing upward. a mixture of ethanol and acetone (50:50) was poured on the PMMA pieces and left for 30 s. The microchip was fabricated by combining the two PMMA pieces for 30 s (the solvent-treated surfaces of both PMMA pieces were bound together producing microchannels), air-drying the microchannels to completely remove the excess of mixture left in the microchannels and pressing with a metal piece (3--4 kg weight) for 5 min. [Figure 1](#biosensors-10-00039-f001){ref-type="fig"} shows a photograph of the fabricated microchip. The microchip was then placed in an oxygen plasma cleaner (Diener electronics) for 8 min at a power of 90 watts. Then, 2% PVA (146000--186000) was prepared in deionized water (DI) water and used to fill the microchannels at room temperature. The solution was removed from the microchannels after 10 min, and the microchannels were dried by compressed air and left in an oven for 15 min at 60--65 °C. A second solution consisting of 5% PVA, 5 mM glutaraldehyde and 5 mM hydrochloric acid was freshly prepared and injected into the microchannels at room temperature. After 10 min, the solution was removed via compressed air, and the microchip was left for 30 min at room temperature. The microchip coated with PVA was directly used for food dye calibrations. For reagents cross-linking, the microchannels were filled with reagent solutions (3,3′,5,5′-tetramethylbenzidine (TMB) or nitrocefin) at room temperature and left for 30 min. The reagents were then removed from the microchannels via compressed air and used straightaway for the experiments. [Figure 1](#biosensors-10-00039-f001){ref-type="fig"}b shows the schematic of the working principle of the chip-and-dip device. Firstly, the device is fabricated with microchannels and coated with reagents as discussed above. Secondly, the device is dipped into the sample, which allows the sample to be loaded into the microchannels via capillary action. Thirdly, the device is left in incubation to allow the reaction between the analytes of interest and the reagents to occur and produce a colourimetric signal. Finally, the device is placed under a smartphone, and a photo is captured, which is further analyzed via image processing. The simplicity of the chip-and-dip operation procedure makes the device potentially useful for clinicians in point-of-care settings. 3. Results and Discussion {#sec3-biosensors-10-00039} ========================= 3.1. Characterization of the PVA Coating in Microchannels {#sec3dot1-biosensors-10-00039} --------------------------------------------------------- To test the surface hydrophilicity of PVA-coated microchannels, coated and uncoated microchannels were dipped into a red food dye (East End Foods plc, UK) to observe the liquid rise in the microchannels. The liquid rose only in PVA-coated microchannels, as shown in [Figure 2](#biosensors-10-00039-f002){ref-type="fig"}a. Snapshots of the liquid rising in 15 microchannels were taken at multiple times (0, 0.5, 1.0 and 2.0 s), and the speed of the liquid rising in each microchannel was measured. [Figure 2](#biosensors-10-00039-f002){ref-type="fig"}b shows the speed of the liquid rising in the microchannels (12 mm/s), with a variability in the 15 microchannels of less than 5% (%RSD). The microchannels were filled within 4 s (4.5 cm-long), as shown in time-lapse photos in [Figure 2](#biosensors-10-00039-f002){ref-type="fig"}b. The variation in height and velocity was the highest at 0.5 s, which can be attributed to the slow and abrupt loading of the microchannels at the start of the dipping experiment. The velocity initially increased up to 1.0 s and then decreased. As the microchannels filled, the resistance to the liquid movement increased, and the capillary flow decreased \[[@B1-biosensors-10-00039]\]. The variations in the speed measurements can be attributed to differences in the PVA coating, position of the microchannels in the liquid (difference in fluid pressure at the inlet) and reagents cross-linking inside the microchannels. 3.2. Characterization of Reagent Cross-Linking inside the Microchannels {#sec3dot2-biosensors-10-00039} ----------------------------------------------------------------------- A solution of TMB (ThermoScientific, 1 Step Ultra TMB ELISA) was loaded into the microchannels (coated with PVA and cross-linked with glutaraldehyde as previously described) and left for 30 min at room temperature. The microchip was then dipped into a horseradish peroxidase-cortisol (HRP-cortisol) solution (United immunoassay, Cortisol Conjugate (HRP)), and the blue colour was observed ([Figure 3](#biosensors-10-00039-f003){ref-type="fig"}a,b). The reagents were used without any further dilution. HPR conjugated to cortisol catalyzes the conversion of TMB, a chromogenic substrate, into a blue-green product. The blue colour intensity increased with time, as shown in [Figure 3](#biosensors-10-00039-f003){ref-type="fig"}a, and the colour intensity levels variability was found to be 2% (%RSD) between the 15 microchannels and 7% (%RSD) between the microchips (*n* = 3). [Figure 3](#biosensors-10-00039-f003){ref-type="fig"}b shows the colour intensity change at different times (0.5, 1.0, 1.5, 2.5, 3.5, 4.5, 5.5 and 6.5 min) for the 15 microchannels in a single microchip. Furthermore, the presence of β-lactamase, an indicator of antimicrobial resistance \[[@B23-biosensors-10-00039],[@B24-biosensors-10-00039]\] was tested by dipping nitrocefin-coated microchannels into a β-lactamase solution (Abcam, recombinant *Escherichia coli* β-lactamase protein) and DI water samples. Nitrocefin (0.25 mg/mL in dimethyl sulfoxide) was loaded into the microchannels and left for 30 min at room temperature. The microchip was then dipped into a solution of β-lactamase (0.2 mg/mL) prepared in phosphate-buffered saline (PBS) at a pH of 7.2. Nitrocefin converted β-lactamase into a red product within 2 min, while the water samples remained yellow ([Figure 4](#biosensors-10-00039-f004){ref-type="fig"}a,b). The red colour intensity increased with time ([Figure 4](#biosensors-10-00039-f004){ref-type="fig"}a), and the colour intensity levels variability was found to be 5% (%RSD) between the 15 microchannels and 9% (%RSD) between the microchips (*n* = 3). [Figure 4](#biosensors-10-00039-f004){ref-type="fig"}c shows the graph of colour intensity change at different times (3, 6, 12, 18, 24, 30, 60 and 120 s) for the 15 microchannels in a single microchip. A smartphone (iPhone XR) was used to capture images of the microchip under external light, which could add variations in the light signal of the different microchannels. Smartphones can have a variety of optical components, and the quality of an image can be different, which can also add variability to the colour measurements. These variations can be reduced by placing the microchips in a light-controlled box with even light or a white background in a black box and running multiple controls/standards in a single device to generate a calibration curve \[[@B21-biosensors-10-00039]\]. In this study, we only focused on the reproducibility of the reaction in between microchannels, and the colour was read by image analysis (ImageJ, NIH) while considering the area between the microchannels as background \[[@B25-biosensors-10-00039]\]. However, negative controls and multiple concentrations of analytes should be loaded in the microchannels for a sensitive and quantitative detection of analytes, which we plan to report in future publications. The chip-and-dip device can be fabricated with multiple sample-loading options and/or with microchannels coated with multiple reagents to capture analytes from a single sample. The mass production of chip-and-dip devices could allow for cheaper analyses at the point-of-care and can be combined with smartphones or portable detectors to provide results near the patients. These microchips can also be coated with reagents and stored away for the detection of various analytes in the clinical settings, such as the determination of minimum inhibitory concentrations (MIC) in microbial assays or antimicrobial susceptibility testing (AST). Our chip-and-dip platform technology is ideally suited to analyse bodily fluids such as urine, tissue dialysate, milk or fisheries water but not blood. This is mainly due to the colorimetric assays used for β-lactamase detection in the microchannels. However, it can be used to analyse clinical blood samples if it is combined with membranes that allow the collection of only serum into the microchannels. This will be further explored in our future studies. 4. Conclusions {#sec4-biosensors-10-00039} ============== The chip-and-dip device offers a simple and pump-free method to collect samples into microchannels, making it possible to perform antimicrobial resistance (AMR) testing at the point-of-care. The simple-to-operate nature of this device, its robustness and ease of operational procedures, combined with the portable/handheld detectors which facilitate the analyses can lead towards the development of point-of-care diagnostics for AMR testing in healthcare. The chip-and-dip device design, fabrication and working principle have been illustrated along with specific applications consisting of cortisol and β-lactamase assays. The chip-and-dip device works on the principle of capillary-driven flow microfluidics and allows detection by multiple microchannels in a single microchip via smartphone imaging/portable detectors. Compared to other types of devices such as dipsticks and paper microfluidic devices, our device is fabricated with cheaper materials, coated with minute amounts of reagents and offers multiplexity on a single microchip. The sample is loaded into the microchannels via capillary force, which eliminates the requirement of external/internal fluidic mechanisms or controls. The current device has the potential to be used for detecting β-lactamase-based antimicrobial resistance. Moreover, the multiplexing capability of the device can allow the quantification of specific analytes in a single microchip, which is currently being explored in our lab. This research was performed as part of the DOSA Project (Diagnostics for One Health and User Driven Solutions for AMR, [www.dosa-diagnostics.org](www.dosa-diagnostics.org). DOSA is funded by UK Research and Innovation/Economic Social Science Research Council, Newton Fund, and the Government of India's Department of Biotechnology. Conceptualization, S.-u.H. and X.Z.; methodology, S.-u.H.; validation, S.-u.H.; formal analysis, S.-u.H.; investigation, S.-u.H.; resources, X.Z.; writing---original draft preparation, S.-u.H.; writing---review and editing, S.-u.H. and X.Z.; supervision, X.Z.; project administration, X.Z. All authors have read and agreed to the published version of the manuscript. This research was funded by the Economic Social Science Research Council, grant number ES/S000208/1. The authors declare no conflict of interest. {#biosensors-10-00039-f001} {#biosensors-10-00039-f002} {#biosensors-10-00039-f003} {#biosensors-10-00039-f004}

Background {#s1} ========== Developing a drug from first-in-human studies to Food and Drug Administration (FDA) approval is a lengthy and costly process ([@CIT0006]). In an effort to streamline therapeutic development, recent efforts have sought to repurpose clinically-available drugs by detecting heretofore unrecognized or off-target effects. Psychiatry is replete with examples of such repurposing, such as the application of anticonvulsants to treat bipolar disorder or antihistamines for psychosis. Unlike development of novel agents, repurposed agents are already vetted for safety and tolerability and often have significant prescriber experience and postmarketing safety data. However, this inversion of the therapeutic pipeline, finding a disease for a drug instead of a drug for a disease, presents new scientific challenges and opportunities. Progress in chemical screening programs allows rapid screening of large libraries of compounds acting on a specific target ([@CIT0005]). While this capacity to screen compounds for specific activities has proven highly scalable, it has been difficult to create low-cost and high-throughput approaches to reliably estimate brain penetrance. Central nervous system (CNS) availability is an obvious but critical next step in the development of CNS-active drugs. While rodent model systems allow pharmacokinetic estimates of blood-brain-barrier permeability, they are imperfect models of human pharmacokinetics and are relatively costly ([@CIT0001]). Alternatively, sophisticated *in silico* models have been developed to predict brain penetration on the basis of physical properties such as molecular weight and lipophilicity, but such models are also limited and may not be accessible to clinical investigators ([@CIT0002]; [@CIT0012]). Given the desirability of repurposing and difficulty of modeling human pharmacokinetics, side effect profiles may represent *in vivo* bioassays of drug targets. A 2008 report illustrated that compounds with similar adverse effect profiles may target similar pathways *in vivo* ([@CIT0004]), and that a pathway similarity conveys data not fully captured by similarity of chemical structure or drug indication. This approach has also been utilized successfully to predict postmarketing adverse effects ([@CIT0013]). The present study builds on this work, using adverse effect profile similarities to develop a predictor of brain penetration, rather than drug target per se, for FDA-approved compounds. The performance of the tool was compared to models relying solely on physical properties, using a curated list of compounds known to be brain-penetrant or -impenetrant. Finally, a web-based tool was created to allow clinical investigators to readily access this predictor and investigate prototypical as well as custom-designed sets of drug comparators. Methods {#s2} ======= A straightforward methodology for generating side effect profile similarity scores was initially described by [@CIT0004]. Using a similar approach, [@CIT0013] derived similarity scores (<http://people.dbmi.columbia.edu/tatonetti/resources.html>, accessed 7/23/2013) based on both FDA labels and postmarketing reports; the latter often detects adverse effects not initially noted in randomized trials. A panel of prototypical CNS-penetrant agents was selected manually by the authors to represent a broad range of medication types within and across classes; this included the antidepressants fluoxetine, nortriptyline, and mirtazapine, as well as the antipsychotics olanzapine, risperidone, and haloperidol. For any given drug, the similarity score was estimated as the mean similarity z-score for the comparator drugs, then re-scaled to yield a z-score for that comparator group. Where a drug is included in the comparator panel, its z-score is estimated from the other drugs. Physical property data for each compound was downloaded from the online database PubChem (<http://pubchem.ncbi.nlm.nih.gov/>, accessed 10/12/2013), including molecular weight and LogP. The latter is a standard estimate of lipophilicity, which has been shown to correlate with blood-brain barrier penetration ([@CIT0012]). To benchmark the models created here, a curated list of known brain-penetrant and non-penetrant drugs was drawn from a larger list of molecules previously described ([@CIT0008]). The total list included 108 brain-penetrant and 56 non-penetrant drugs for which pairwise similarity data was available. This list was also used to generate comparison panels to the 6-drug curated panel, by randomly selecting 6 brain-penetrant medications. First, the mean similarity scores were compared between the known-penetrant and impenetrant drugs using student's *t*-tests. To estimate the improvement in discrimination when adverse effect-based measures were added to existing (physical property-based) measures, nested logistic regression models were fitted both with and without similarity measures. Model discrimination was estimated and compared in terms of the area under the receiver operating characteristic curve (AUC). In light of the limitations of the AUC for comparing model improvement, we also estimated the net reclassification improvement ([@CIT0009]) using the incrisk command ([@CIT0007]). Finally, a web-based calculator and visualization tool was developed that implements the approach described here. The calculator accepts a panel for comparison and screens the remaining drugs for similarity to this panel. The output is a scatterplot showing the difference of a drug's mean similarity to all other drugs and mean similarity to the panel drugs versus difference between mean z-score of panel drugs within each drug and across all drugs. This provides readily-differentiated groups of agents more and less similar to the comparator panel despite the highly-variable similarities (e.g. some drugs are similar to most drugs whereas others appear nearly unique by side effects). Additionally, an upper bound on *p* -value is reported by means of comparison between randomly permuted panels and the user supplied or selected panel. Results {#s3} ======= A total of 632 drugs were examined for similarity to the antidepressant + antipsychotic panel. Mean similarity to the panel was 0.78 (standard deviation 0.51) for the known brain-penetrant drugs (excluding the drugs in the panel) and 0.50 (standard deviation 0.40) for the known non-penetrant drugs; *t* = -3.50, *p* \< 0.001. Two nested logistic regression models were compared: one including molecular weight and LogP alone, and the other added similarity. Similarity remained significantly associated with CNS penetration in the adjusted model ([Table 1](#T1){ref-type="table"}). Including the similarity score significantly improved the predictions, whether measured by model fit \[likelihood ratio chi2(1df) = 9.79, *p* = 0.002\], area under receiver operating characteristic (ROC) curve \[[Figure 1](#F1){ref-type="fig"}; 0.65 (standard error 0.05) versus 0.74 (standard error 0.04), ch2(1df) = 5.28, *p* = 0.02\], or net reclassification improvement \[0.55, 95% confidence interval 0.115--0.798\]. Similar results were obtained when panels of randomly-selected brain penetrant comparator medications were utilized: for 10 randomly-selected panels, the mean area under ROC curve was 0.72 (standard error 0.04). ###### Prediction model for brain penetration, incorporating adverse-effect similarities. Odds Ratio 95% Confidence Interval *p*-value ------------------ ------------ ------------------------- ----------- ------- Molecular weight 0.994 0.990 0.998 0.004 LogP 1.246 1.006 1.543 0.044 Similarity score 3.346 1.491 7.509 0.003 {#F1} [Figure 2](#F2){ref-type="fig"} illustrates the input and output screens of the web-based calculator, available at <http://repoman.mghcedd.org>. After a comparator panel is selected, similarities are calculated for all medications; the resulting scatterplot of all medications, scored against the comparator panel, can be updated to depict various physical properties and similarity scores. The calculator allows use of standard reference panels, as well as custom panels entered by the user. The medications with the greatest similarities to the reference panel are also presented by decreasing similarity. ###### (A and B) Screen illustrations of the web-based similarity calculator for brain penetration.   For example, with the 6-medication reference panel, nearly all of the most similar medications are known CNS-active medications. Among the non-CNS-indicated medications are gatifloxacin, an antibiotic subsequently withdrawn from the US market for other reasons, which was associated with emergence of psychosis ([@CIT0010], [@CIT0011]), and montelukast, which has been implicated as a potential contributor to pediatric psychiatric adverse events ([@CIT0003]). Discussion {#s4} ========== These results indicate that the similarity between two or more drugs in an adverse-effect profile may be used to predict the likelihood that an FDA-approved medication crosses the blood-brain barrier. This bioassay captures properties of a drug not fully explained by traditional physical properties often used in *in silico* modeling ([@CIT0002]; [@CIT0012]). This work relies on previous reports ([@CIT0004]; [@CIT0013]) that described the methodology for deriving adverse effect profiles from pre- and postmarketing data, then utilized similarities between pairs to suggest common targets. While those studies noted the potential usefulness in repurposing compounds, they did not describe a specific method for applying and visualizing such data. The present study is an effort to facilitate investigation of drugs for repurposing by clinical researchers who may not have access to sophisticated *in silico* models or pharmacokinetic resources required to characterize brain penetration. Several important limitations should be noted. First, by relying on adverse effects, this algorithm will yield results biased towards drugs with less clean adverse effect profiles. Thus, a perfect drug which acts in the brain but yields no brain-specific adverse effects will likely not be prioritized. (Such drugs are to be hoped for, but whether they exist is unclear.) Similarly, depending on the panel of drugs selected, it is possible that scores will point not only to brain penetration, but to mechanisms of action. So, for example, seeding with a single tricyclic antidepressant prioritizes other tricyclic antidepressants. To counter this tendency, the default panel investigated here draws on drugs from multiple classes and targets. Additionally, defining the visualization in terms of differences from panels and drugs means an element of nominal within-drug scaling is introduced. Despite these limitations, the present report suggests that adverse effects represent a useful proxy for brain penetration, and further validates the general proposal in earlier reports that adverse effects may be used for repurposing. Furthermore the tool created suggests that research of this nature can be readily translated into forms that place minimal demands on potential downstream researchers. By combining intuitive data with rudimentary statistics in the form of a readily-accessible calculator, the intention is to encourage clinical investigators to consider repurposing drugs which may have off- or on-target effects useful for psychiatric indications. Supplemental materials ====================== Supplemental materials: website ([http://repoman.mghcedd.org](http://ijnp.oxfordjournals.org/lookup/suppl/doi:10.1093/ijnp/pyu078/-/DC1)) Statement of Interest ===================== Dr Perlis is a member of scientific advisory boards or has received consulting fees from Genomind, Healthrageous, Pamlab, Perfect Health, Pfizer, Proteus Biomedical, and RIDVentures. He has received research support from Proteus Biomedical, and royalties from Concordant Rater Systems (now UBC). Dr Perlis is supported by NIMH MH086026 and the Stanley Center for Psychiatric Research at the Broad Institute. Dr McCoy is a participant in NIMH 5R25MH094612.

Erratum to: Clin Transl Oncol (2015) 17:996--1004 DOI 10.1007/s12094-015-1456-y {#Sec1} =============================================================================== The affiliation of the author A. Custodio was rendered wrongly. The correct affiliation is: Medical Oncology Department, Hospital Universitario La Paz, Madrid, Spain. The authors regret this error and apologize for any inconvenience caused. The online version of the original article can be found under doi:10.1007/s12094-015-1456-y.

{#sp1 .288} {#sp2 .289}

Introduction {#Sec1} ============ Excessive use of antibiotics in general and in veterinary medicine has resulted in the contamination of the natural environment with these substances (EEA [@CR20]). Of all the veterinary antibiotics (VAs), only 20% are used to treat infections; the other 80% are used to prevent diseases or to promote animal growth (US FDA [@CR65]). The dosage of VAs varies (from 3 to 220 g kg^−1^) depending on animal size and the type of antibiotic used (Health Canada [@CR31]). After being administered, VAs are largely (from 10 to 90%) eliminated from the body to the environment either in an unchanged form or as active metabolites (Elmund et al. [@CR21]; Halling-Sørensen et al. [@CR30]). Antibiotics have been found in animal faeces and in soil in concentrations ranging from 0.01 to 1420 mg kg^−1^ (Pan and Chu [@CR50]). In aquaculture, pharmaceuticals are added to water in fish ponds, and their concentrations in water and lake deposits are high (Li et al. [@CR47]). An increase in the concentration of VAs and their metabolites in the environment can have a negative impact on land organisms (Pan and Chu [@CR50]). The most common antibiotics used in veterinary medicine include tetracyclines, fluoroquinolones, macrolides and sulphonamides (EMA [@CR23]). Fluoroquinolones (FQs) are a new, synthetic group of antibiotics with strong antibacterial action and a broad spectrum of anti-pathogenic activity. FQs are used for the treatment of infections of the urinary tract, the gastrointestinal system, the respiratory tract and the skin in medicine (EMA [@CR24]) and in intensive breeding of cattle, poultry, pigs, fish, as well as dogs and cats (EMEA [@CR22]). Fluoroquinolones are highly stable, due to which they are not wholly metabolised in the body. Their residues can be eliminated, and they can find their way to wastewater treatment plants or to the environment (Watkinson et al. [@CR68]). The toxicity of FQs towards microorganisms limits the processes of antibiotic biodegradation in water and in soil (Robinson et al. [@CR55]). They have been found in crude wastewater (2573 ng L^−1^), secondary wastewater (1013 ng L^−1^) and in sludge (18.4 mg kg^−1^) (Jia et al. [@CR35]). Manure used as fertiliser (faeces of chickens, pigs and cows) is the main source of contamination of soil with antibiotics (Li et al. [@CR48]). Di- and trivalent cations (Ca^2+^, Mg^2+^ and Al^3+^) form stable complexes with fluoroquinolones. The chelates are strongly adsorbed in soil, and they cannot be leached out of it (Turel et al. [@CR64]; Djurdjević et al. [@CR17]). The concentration of FQs in the soil fertilised with manure ranges from 0.01 to 0.4 mg kg^−1^ of soil, and it remains at this level for several months (Karcı and Balcıoğlu [@CR37]). Boxall et al. ([@CR10]) proved the stability of enrofloxacin in soil; its half-life was longer than 152 days. Furthermore, Golet et al. ([@CR26]) applied ciprofloxacin and norloxacin to soil at a concentration ranging from 0.29 to 0.40 mg kg^−1^; after nearly 2 years in the soil, there was still 93 and 75% of the drugs, respectively. Fluoroquinolones are divided into four groups (generations) in which antibiotics differ by the structure and the spectrum of action (Soni [@CR60]). Third-generation FQs include the [l]{.smallcaps}-isomer of ofloxacin---levofloxacin (LEV). LEV is characterised by good infiltration of target tissues and is used to control intracellular pathogens. The drug is eliminated from the body through the kidneys, usually as the active antibiotic. LEV is metabolised to a small extent and its inactive metabolites, i.e. demethyllevofloxacin and levofloxacin *N*-oxide, account only for \< 5% of the dose applied (Levaquin® [@CR46]). After being eliminated, the antibiotic is introduced to soil, surface and ground waters and to lake deposits (Tonga et al. [@CR62]). Boxall et al. ([@CR10]) proved that antibiotics are absorbed from soil by roots of carrot and lettuce. The consequences of the presence of fluoroquinolones in the environment have not been fully revealed, but they are known to be toxic to plants (Frade et al. [@CR25]). Since levofloxacin effectively enters inside the cell and is used to control intracellular pathogens, changes caused by its toxic action will become apparent at the cellular level earlier than anywhere else. When evaluating the environmental risk caused by xenobiotics, the majority of studies have focused on toxic chemicals. However, biochemical changes in plants can be detected earlier than injuries, or the considerable decrease in plant growth and viability. Therefore, antibiotic toxicity to lupin was evaluated in this study with the use of standard endpoints recommended in guidelines, which include biomass parameters. Lupin is a common fodder crop (Lefroy and Rydberg [@CR45]), and biochemical contaminants accumulated in it can be transferred to the food chains comprising farm animals. Moreover, it was recently proposed as a component of functional foods and a health-promoting plant (Pilvi et al. [@CR53]). Considering this, it is important to understand the reactions of this plant to environmental pollutants. Furthermore, new endpoints were calculated in this study which had not been examined before, namely, oxidative stress enzymes (guaiacol peroxidase, catalase), distribution of reactive oxygen species (ROS) in cytoplasm and protein contents and profiles. Material and methods {#Sec2} ==================== Germination and growth of seedlings {#Sec3} ----------------------------------- Seeds of yellow lupin (*Lupinus luteus* L.), Dukat cultivar, were used in the study. The seeds germinated for 7, 9 and 12 days on PHYTOTOXKIT test plates (MicroBioTests Inc., Mariakerke (Gent), Belgium). Before germination, the seeds were surface-sterilised by immersing them in 1% sodium hypochlorite for 2 min; the seeds were then washed in distilled water for 3--4 min. Next, 90 mL of soil was added to the plates and they were then watered with 27 mL of water (control) or 27 mL of levofloxacin solutions at different concentrations (Sigma-Aldrich): 0.01, 0.05, 0.1, 0.5, 1, 1.5, 2 and 2.5 mM (i.e. 0.7, 3.5, 7, 35, 70, 105, 140, 175 mg kg^−1^ of soil). The soil was covered with Whatman no. 1 filter paper on which seeds were placed. The biotest was carried out for 7, 9 and 12 days at room temperature. Roots and shoots of the seedlings were measured with the ImageJ tool software. Fresh and dry weights were determined. Activity of guaiacol peroxidase {#Sec4} ------------------------------- Plant extracts were prepared on ice. Roots, shoots and leaves (500 mg) of the seedlings which had grown for 7, 9 and 12 days in soil with an addition of water or levofloxacin were homogenised for 30 min in a buffer solution (0.1 M Tris-HCl, Sigma-Aldrich; 8.75% polivinylpyrrolidone, Sigma-Aldrich; 0.1 M KCl, 0.28% Triton X-100, Sigma-Aldrich). The samples were centrifuged for 30 min at 4000×*g* at 4 °C. The supernatant was purified with membrane syringe filters with a pore diameter of 0.45 μm. Protein in the samples was assayed by the Lowry method (Lowry et al. [@CR49]). The activity of peroxidase was determined spectrophotometrically (CECLI, CE2021 2000 Series) in the reaction mixture containing 100 μL of 1% guaiacol (Sigma-Aldrich), 2 mL 0.1 M KH~2~PO~4~ (Chempur), 50 μL of the supernatant and 20 μL 0.06% H~2~O~2~ (Chempur). The absorbance growth rate was measured at room temperature at the wavelength of 470 nm. One unit corresponds to oxidation of 1 μM H~2~O~2~ for 1 min. Activity of catalase {#Sec5} -------------------- Plant extracts were prepared on ice. Roots, shoots and leaves (500 mg) of the seedlings which had grown for 7, 9 and 12 days in soil with an addition of water or levofloxacin were homogenised in phosphate buffer solution pH 7, which contained 10 g L^−1^ polivinylpyrrolidone (PVP, Sigma-Aldrich), 0.2 mM EDTA (Sigma-Aldrich) and 10 mL L^−1^ Triton X-100 (Sigma-Aldrich). The samples were centrifuged for 20 min at 12000×*g* at 4 °C. The supernatant was then carefully separated from the sediment. Protein in the samples was assayed by Lowry's method (Lowry et al. [@CR49]). The activity of catalase was determined spectrophotometrically (CECLI, CE2021 2000 Series) with a reaction mixture containing 50 mM phosphate buffer, pH 7 and 15 mM H~2~O~2~. The absorbance was measured for 10 min at room temperature at the wavelength of 240 nm. One unit corresponds to oxidation of 1 μM H~2~O~2~ for 1 min. Reactive oxygen species {#Sec6} ----------------------- Seeds of yellow lupin were germinated in soil watered with deionised water. After 7 days, the seedlings were transferred to new, sterile PHYTOTOXKIT test plates filled with control soil and with a 2.5 mM solution of levofloxacin. After 24 and 48 h, roots of the seedlings were incubated in the dark for 30 min in 0.1 M PBS at pH 7.4 (phosphate-buffered saline), with 10 μM 2′,7′- dichlorodihydrofluorescein diacetate (H~2~DCF-DA). Subsequently, they were washed several times with a 0.1 M PBS solution at pH 7.4. 2′,7′-Dichlorodihydrofluorescein diacetate is a lipophilic and non-fluorescent compound which freely infiltrates cell interiors. H~2~DCF-DA is degraded in live cells by intracellular esterases to lipophobic dichlorofluorescein (DCF), which emits a fluorescent signal. DCF reacts with reactive oxygen species (ROS), as well as with reactive nitrogen species and with lipid peroxides. The fluorescence of DCF was measured with a Leica TCS SP5 confocal scanning laser microscope with LAS software (Leica Application Suite 2.0.2 build 2038). Oxidised dichlorofluorescein fluoresces at the absorption maximum of 525 nm. Isolation of protein and 2D electrophoresis-PAGE {#Sec7} ------------------------------------------------ Plant extracts were prepared on ice. The roots of yellow lupin seedlings (300 mg) grown for 12 days in soil with an addition of water and 2.5 mM of solution of levofloxacin were homogenised in a buffer which contained 7.4 M of urea, 2.1 M of thiourea, 62 mM of CHAPS (Sigma-Aldrich), 8 mM of DTT (Sigma-Aldrich), 0.24% of Triton X-100 (Sigma-Aldrich), 16.5 mM of Trizma base (Sigma-Aldrich), 21 mM of Trizma hydrochloride (Sigma-Aldrich), 1% of ampholytes pH 3--10 (Bio-Rad) and a cocktail of proteases (Sigma-Aldrich) (Badowiec et al. [@CR3]). The samples were shaken for 45 min and subsequently centrifuged for 10 min at 18000×*g* at 4 °C. The supernatant was frozen at −80 °C and then lyophilised. The lyophilisate was purified with ReadyPrep™ 2-D Cleanup Kit (Bio-Rad). The isolated proteins were dissolved in a buffer for rehydration which contained 7 M of urea, 2 M of thiourea, 2% of CHAPS, 0.5% of ampholytes pH 3--10, 80 mM DTT and 0.002% of bromophenol blue. Two-way electrophoresis was carried out in two stages. First, rehydration and isoelectric focusing was carried out with a PROTEAN IEF Cell (Bio-Rad). A protein solution (70 μg of protein) was added to a well in the focusing plate, IPG strips (ReadyStrip™ 7 cm, pH 3--10; Bio-Rad) were placed and a 12-h active rehydration was carried out. After rehydration was completed, isoelectric focusing was carried out (conditions: preparatory stage S1---250 V / 15 min; voltage increase stage S2---4000 V / 2 h 30 min; isoelectric focusing S3---rapidly increasing voltage up to 20,000 V). After isoelectric focusing was completed, IPG strips were equalised in buffer I (6 M urea, 2% SDS, 0.375 M Tris-HCl with pH 8.8, 20% glycerol, 130 mM DTT; Bio-Rad) and buffer II (6 M urea, 2% SDS, 0,375 M Tris-HCl with pH 8.8, 20% glycerol, 135 mM iodoacetamide; Bio-Rad) for 10 min. The second stage involved separation of proteins in 10% polyacrylamide gels (7.0 cm × 10.0 cm) in an electrophoresis apparatus (Mini PROTEAN Tetra System; Bio-Rad). Electrophoresis was carried out for 1 h, with 45 min in alternating current: 90 V (15 min) and 120 V. After the electrophoresis was completed, gels were dyed in colloidal solution of Comassie Brilliant Blue G-250 (Sigma-Aldrich). The gels were visualised with a Gel Doc EZ Imager scanner (Bio-Rad). The protein maps were analysed in the PDQuest™ Basic program (2-D Gel Analysis Software; Bio-Rad). Digesting proteins and LC-MS-MS/MS analysis {#Sec8} ------------------------------------------- Spots with proteins were carefully cut out from the polyacrylamide gels. After reduction and blocking the reduced bridges, a mixture of peptides was obtained with the use of trypsin. The mixture was separated by liquid chromatography (LC), and the mass of peptides and their fragments was determined in a mass spectrometer (Orbitrap LC-MS-MS/MS; Thermo). The NCBI and UnitProt databases were searched for MS/MS spectra. The proteins were identified based on PMF (peptide mass fingerprint) with the MASCOT search engine. Statistical analysis {#Sec9} -------------------- One-way analysis of variance (ANOVA) followed by Tukey's comparison post hoc test (*p* ≤ 0.01) was applied to evaluate differences between controls and treatments. Effective concentrations (EC~*x*~) calculations (Alexander et al. [@CR1]) were carried out for inhibition of shoot and root growth, dry and fresh weight and enzyme activity, and the results were expressed in mM. EC~10~, EC~25~, EC~50~ and EC~90~ express concentrations of the antibiotic resulting in 10, 25, 50 and up to 90% of the toxic effect. The effective concentration (EC) was computed from the following equation: $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {\mathrm{EC}}_x\kern0.5em =\kern0.5em C\kern0.5em +\kern0.5em \frac{\left(A\kern0.5em -\kern0.5em 50\%\max \kern0.5em \mathrm{response}\right)\kern0.5em \times \kern0.5em X}{y} $$\end{document}$$ where *x* = *D* ~(max\ concentration)~, *C* ~(min\ concentration)~ = difference of concentration, *y* = *A* ~(max\ response)~, *B* ~(min\ response)~ = difference of response. The EC values expressed in μM were transformed to the measured toxicity units (TU): TU~10,\ 25,\ 50,\ 90~ = (1/EC~10,\ 25,\ 50,\ 90~) × 100. Results {#Sec10} ======= The toxicity of levofloxacin (LEV) towards seedlings of yellow lupin (*L. luteus* L. cv. Dukat) was examined. The seedlings grew for 7, 9 and 12 days on control soil (water) and on soil contaminated with LEV at various concentrations (0.01, 0.05, 0.1, 0.5, 1, 1.5, 2, 2.5 mM, i.e. 0.7, 3.5, 7, 35, 70, 105, 140, 175 mg kg^−1^ of soil). The following morphological parameters of the seedlings were evaluated: length of roots and shoots, seedling fresh and dry weight (Fig. [1](#Fig1){ref-type="fig"}) and the physiological and biochemical parameters, including activity of enzymes (catalase and peroxidase), protein profile and location of free radicals.Fig. 1Roots (**a**) and shoots length (mm) (**b**), fresh (mg) (**c**) and dry mass (%) (**d**) of *Lupinus luteus* L. after 7 (■), 9 (■) and 12 (□) days on soil supplemented with different levofloxacin concentrations (**c** control; 0.01, 0.05, 0.1, 0.5, 1, 1.5, 2, 2.5 mM). Means with the same letter are not significantly different from each other (Tukey's test, *p* ≤ 0.01) Morphological parameters {#Sec11} ------------------------ Lupin seed germination, manifested by protrusion of the radicle through the seed coat, occurred at 1--1.5 days after sowing. The longest roots were observed in seedlings which had grown for 7, 9 and 12 days on control soil; they were 78, 80 and 81 mm long (Fig. [1](#Fig1){ref-type="fig"}a), respectively. The roots of 7-, 9- and 12-day-old seedlings which had grown in soil with 0.1 mM of solution of levofloxacin were on average slightly shorter than the roots of the control seedlings by 3, 4 and 4%, respectively. With the highest LEV concentration applied (2.5 mM), inhibition of root growth was much stronger---58, 59 and 60% compared to the roots of the control seedlings in 7-, 9- and 12-day-old seedlings, respectively (Fig. [1](#Fig1){ref-type="fig"}a). The response of shoots to LEV was comparable to the reaction of roots. In control seedlings grown for 7, 9 and 12 days, shoot lengths were 53, 68, 71 mm long, respectively (Fig. [1](#Fig1){ref-type="fig"}b). The smallest concentration of LEV (0.01 mM) reduced the shoot length in 12-day-old seedlings by 5% (Fig. [1](#Fig1){ref-type="fig"}b), and the highest LEV concentration (2.5 mM) resulted in a 49% reduction of shoot length of the 12-day-old seedlings. The EC~50~ indexes for the root and shoot elongation were 0.7 and 1.3 mM (Table [1](#Tab1){ref-type="table"}), respectively.Table 1Effect of levofloxacin on growth and enzyme activity parameters in 7, 9 and 12 days in yellow lupin assaysEC~10~EC~25~EC~50~EC~90~LengthRootShootRootShootRootShootRootShoot7 days0.150.0090.250.230.651.231.191.89 days0.120.0070.20.10.651.231.72.1512 days0.120.0080.20.20.671.261.72.23Fresh weightSeedlings7 days1.241.51.682.09 days1.121.371.672.012 days1.031.381.692.3Dry weightSeedlings7 days0.650.71.782.379 days0.731.51.72.312 days1.181.61.82.3Catalase activityRootsShootsLeavesRootsShootsLeavesRootsShootsLeavesRootsShootsLeaves7 days0.130.540.0020.250.680.0051.520.80.0692.32.261.99 days0.150.580.0030.270.70.0831.150.830.12.352.252.212 days0.110.560.0020.20.690.0070.490.860.22.322.232.31Peroxidase activityRootsShootsLeavesRootsShootsLeavesRootsShootsLeavesRootsShootsLeaves7 days0.90.0080.041.50.70.71.81.61.62.3229 days0.80.0090.011.50.750.71.81.651.62.32.132.1512 days0.70.10.0091.30.080.651.781.721.632.32.22.3EC~10~, EC~25~, EC~50~ and EC~90~ values are expressed in mM The fresh mass of seedlings decreased with an increasing concentration of levofloxacin. The weight of seedlings grown for 12 days in soil with 1.5 and 2.5 mM LEV solutions was smaller than the control seedlings by 18 and 62%, respectively (Fig. [1](#Fig1){ref-type="fig"}c). The opposite pattern was observed in the case of seedling dry weight, i.e. it increased with increasing concentrations of LEV solutions (Fig. [1](#Fig1){ref-type="fig"}d). The dry mass of 12-day-old seedlings grown in soil contaminated with a 2.5 mM was greater than the control seedlings by 58%. Units of toxicity for dry and fresh weight of biomass ranged from 0.04 to 0.1, depending on LEV concentration and seedling age (Table [2](#Tab2){ref-type="table"}).Table 2Toxicity units (TU~10,\ 25,\ 50,\ 90~ = (1/EC~10,\ 25,\ 50,\ 90~) × 100) calculated on the basis of levofloxacin concentration for 12 daysTU~10~TU~25~TU~50~TU~90~LengthRootsShootsRootsShootsRootsShootsRootsShoots0.7712.50.460.570.150.080.070.05Fresh weightSeedlings0.090.070.060.05Dry weightSeedlings0.120.080.060.04Catalase activityRootsShootsLeavesRootsShootsLeavesRootsShootsLeavesRootsShootsLeaves0.770.18430.420.143.20.090.120.810.040.040.05Peroxidase activityRootsShootsLeavesRootsShootsLeavesRootsShootsLeavesRootsShootsLeaves0.132.65.10.070.200.150.060.060.060.040.050.05The value EC~10,\ 25,\ 50,\ 90~ is expressed in μM Peroxidase and catalase activity {#Sec12} -------------------------------- The activity of guaiacol peroxidase (Fig. [2](#Fig2){ref-type="fig"}a--c) and catalase (Fig. [2](#Fig2){ref-type="fig"}d--f) in roots, leaves and shoots of seedlings grown in control soil (water) and in soil contaminated with levofloxacin.Fig. 2Activity of guaiacol peroxidase in roots (**a**), shoots (**b**) and leaves (**c**) and activity of catalase in roots (**d**), shoots (**e**) and leaves (**f**) (*U* one unit of enzyme activity corresponds to the oxidation of 1 mM H~2~O~2~ for 1 min) of *Lupinus luteus* L. after 7 (■), 9 (■) and 12 (□) days on soil supplemented with different levofloxacin concentrations (**c** control; 0.01, 0.05, 0.1, 0.5, 1, 1.5, 2.5 mM). Means with the same letter are not significantly different from each other (Tukey's test, *p* ≤ 0.01) The highest activity of peroxidase was recorded in roots of seedlings grown in soil with 2.5 mM of LEV (Fig. [2](#Fig2){ref-type="fig"}a). The activity of this enzyme in 7-, 9- and 12-day-old seedlings was nine times higher than in the control seedlings. The toxicity indexes (TU) 10 for the activity of the enzyme were very high: 2.6 and 5.1 for shoots and leaves, respectively (Table [2](#Tab2){ref-type="table"}). The enzyme activity in leaves and shoots of 12-day-old seedlings grown in soil with the lowest LEV concentration (0.01 mM) was 15 and 35% compared to the leaves and shoots of the control seedlings, respectively (Fig. [2](#Fig2){ref-type="fig"}b, c). The activity of peroxidase in leaves and shoots of the seedlings grown in soil with the highest concentration of LEV (2.5 mM) increased 4- and 8-fold, respectively. EC~50~ for the activity of peroxidase in roots, shoots and leaves of 12-day-old seedlings was 1.78, 1.72 and 1.63 mM, respectively (Table [1](#Tab1){ref-type="table"}). The catalase activity in 12-day-old lupin seedlings grew with increasing concentrations of LEV (Fig. [2](#Fig2){ref-type="fig"}d--f). In the case of leaves (but not roots and shoots), this rise was clearly visible even with the lowest LEV concentration. The largest increase in the enzyme activity (75, 72 and 73%) was observed in the shoots of 7-, 9- and 12-day-old seedlings grown in soil with an addition of 2.5 mM solution of LEV. EC~50~ determined for catalase was 1.1, 0.8 and 0.1 (Table [1](#Tab1){ref-type="table"}) for roots, shoots and leaves, respectively. The toxicity index for catalase was highest for leaves (Table [2](#Tab2){ref-type="table"}). Location of reactive oxygen species {#Sec13} ----------------------------------- The distribution of reactive oxygen species (ROS) in roots is presented in Fig. [3](#Fig3){ref-type="fig"}. The photographs were created by superimposing all photos taken at different depths of the preparation (snapshot). The reactive oxygen species in roots of 7-day-old seedlings which had grown for 24 h in soil contaminated with 2.5 mM LEV was seen to increase rapidly compared to roots of control seedlings (Fig. [3](#Fig3){ref-type="fig"}, 2A--2C). The intensity of fluorescence in the elongation zone was uniform in the material under study. At the beginning of the analyses period (24 h), ROS occurred within the whole interior of root cells. Extended exposure to the stressor (48 h) resulted in saturation of fluorescence in the roots under study (Fig. [3](#Fig3){ref-type="fig"} 2A'--2C'). Fluorescence in root cells was no more uniform, and ROS were found to accumulate along the cell wall. Fluorescence was also observed in the control roots, but its intensity was very low (Fig. [3](#Fig3){ref-type="fig"} 1A--1C, 1A'--C').Fig. 3The control roots of 7-day-old (photo 1A, 1B and 1C) and 8-day-old (photo 1A', 1B' and 1C') seedlings and the roots of 7-day-old seedlings that have grown by 24 h (photo 2A, 2B and 2C) and 48 h (photo 2A', 2B' and 2C') in the soil with the addition of 2.5 mM concentration of levofloxacin. The results obtained were created by ROS location shooting performance at different depths preparation in the *Z* axis (Series Z) by way of a confocal microscope scan capture. *A*, *A'* the fluorescence image; *B*, *B'* without fluorescence image; *C*, *C'* overlapping images 2D-PAGE electrophoresis {#Sec14} ----------------------- The effect of selected concentrations of levofloxacin on the yellow lupin proteome was analysed in this study. The 2D maps obtained which present the protein content in 12-day-old roots of seedlings which grew in control soil (water) and in soil with addition of 2.5 mM of LEV solution revealed differences in the amount and content of the isolated proteins (Fig. [4](#Fig4){ref-type="fig"}a, b). There were 205 and 261 proteins, respectively, found in gels made from control roots and from those growing in soil with an addition of levofloxacin. The maps created with control roots in the mass range from 25 to 37 kDa revealed the most abundant presence of proteins (79) of all the analysed samples (Fig. [5](#Fig5){ref-type="fig"}a). The number of proteins with molecular weights smaller than 20 kDa and larger than 75 kDa was the smallest. The number of proteins visualised in the control map at pH 6 was 75, which accounted for 35% of all the proteins obtained (Fig. [5](#Fig5){ref-type="fig"}b). The smallest number of proteins was found in the control gel in the area with pH 3. A 2D map made for roots of seedlings grown in soil contaminated with 2.5 mM LEV showed that proteins with a molecular weight between 25 and 37 kDa accounted for 37% of all proteins (Fig. [5](#Fig5){ref-type="fig"}c). Proteins with a molecular weight above 75 kDa accounted for only 4% of all proteins. The largest number of proteins (100) in the analysed gel was found at pH 6 (Fig. [5](#Fig5){ref-type="fig"}d). Three proteins (1% of all proteins obtained) were visualised at pH 9.Fig. 42D electrophoresis of proteins of root lupin (*Lupinus luteus* L.) without (**a**) and with (**b**) the levofloxacin at the concentration of 2.5 mM of soil. Protein separation was conducted at pH 3--10 Fig. 5Pie chart showing the molecular mass distribution (kDa) of proteins from control roots (**a**) and the levofloxacin affected roots (**b**) at the concentration of 2.5 mM of soil and the pH range of proteins from control roots (**c**) and the levofloxacin-affected roots (**d**) at the concentration of 2.5 mM of soil Identification of antibiotic-responsive proteins {#Sec15} ------------------------------------------------ The toxicity of LEV was determined in regard to protein expression in 12-day-old roots of yellow lupin. Identification of 24 proteins was carried out, which may be involved in developing yellow lupin resistance to LEV toxicity. The results are presented in Table [3](#Tab3){ref-type="table"}, in which proteins were categorised into groups depending on their function. The following functional groups of proteins were identified: defence and response to stress, growth, energy and carbohydrate metabolism, detoxication, oxygen supply, signal transduction, translation control and protein pump. The largest group was that of proteins responsible for defence and plant reaction to environmental stress. The quantities of the most members of this group (five out of nine) increased as a result of levofloxacin treatment. Among the other members of this group, there were both those which decreased (nos. 3, 19) and those which remained constant (16, 6). Seven proteins (protein nos. 3, 7, 12, 18, 19, 20, 23) were found to show increased expression in control roots, and synthesis of eight proteins (nos. 1, 2, 5, 10, 13, 14, 15, 17) decreased (Fig. [6](#Fig6){ref-type="fig"}a). The expression of the remaining proteins (nos. 4, 6, 8, 9, 11, 16, 21, 22, 24) did not change (Fig. [6](#Fig6){ref-type="fig"}a--c; Table [3](#Tab3){ref-type="table"}).Table 3Proteins identified by LC-MS-MS/MS analysesSpot no.Protein nameSpeciesProtein IDFunctionMascot score1Full = protein LlR18A*Lupinus luteus*gi 1730080Defence response2302Class III chitinase*Lupinus albus*gi 2853142Defence and stress responses152314-3-3-like protein A*Glycine max*gi 351720808Metabolic regulation, stress responses11264Full = 60S acidic ribosomal protein P0*Lupinus luteus*gi 1710585Translation control mechanisms, protein turnover4895Ascorbate peroxidase*Glycine max*gi 4406539Antioxidant, defence and stress responses2766Germin-like protein 3*Glycine max*gi 196122010Development and defence1217Expansin-B3-like precursor*Glycine max*gi 387528019Development1348Stem 28 kDa glycoprotein precursor, putative*Ricinus communis*gi 255549796Vegetative storage protein, acid phosphatase activity1239Malate dehydrogenase 1, mitochondrial, partial*Glycine soja*gi 7343118888Energy metabolism36310Aspartate aminotransferase P1*Lupinus angusitifolius*gi 168324Carbon and nitrogen metabolism34311UDP-[d]{.smallcaps}-glucuronate carboxy-lyase*Pisum sativum*gi 13591616Energy metabolism23212Fructose-bisphosphate aldolase, cytoplasmic isozyme*Glycine soja*gi 734419305Energy metabolism90513Full = alcohol dehydrogenase 1*Pisum sativum*gi 113361Protection against hypoxic metabolism58714Enolase*Lupinus luteus*gi 6996529Carbohydrate and energy metabolism, stress responses117215Leghemoglobin reductase-like*Glycine max*gi 356565179Metabolism, provide oxygen104216Extensin peroxidase*Lupinus albus*gi 27448342Development, stress responses21617Heat shock cognate protein 80-like*Glycine max*gi 356552478Stress responses, defence265218Glyceraldehyde-3-phosphate dehydrogenase*Lupinus albus*gi 62816190Carbohydrate and energy metabolism88219Lipoxygenase*Pisum sativum*gi 2459611Metabolism, stress responses, defence, signal transduction7120Nucleotide-binding subunit of vacuolar ATPase*Arabidopsis thaliana*gi 166627Proton pomp, ions and metabolite transport9521F1 ATPase*Pisum sativum*gi 2116558Energy metabolism11322Actin*Oryza sativa*gi 20329Development, resistance, elongation and differentiation of the cell10723Pyruvate dehydrogenase E1 component subunit alpha, mitochondrial-like*Glycine max*gi 356526868Carbohydrate and energy metabolism44424Peptidyl-tRNA hydrolase 2, mitochondrial*Beta vulgaris*gi 731320994Transaction factor, apoptosis175 Fig. 6The average intensity of specific protein spots (PDQuest, Bio-Rad) for control roots (□) and levofloxacin (2.5 mM of soil)-treated roots (■). **a** Spot numbers (compare Table [3](#Tab3){ref-type="table"} ). **b** 3D profile of specific spots' optical intensities: 1---full = protein LlR18A, 2---class III chitinase, 3---14-3-3-like protein A, 5---ascorbate peroxidase, 13---full = alcohol dehydrogenase 1, 18---glyceraldehyde-3-phosphate dehydrogenase, 19---lipoxygenase, 20---nucleotide-binding subunit of vacuolar ATPase. **c** Plain view of selected spots on 2D gels Discussion {#Sec16} ========== Improvement of soil quality by fertilisation with manure is the main cause of contamination of agricultural land with antibiotics. VAs introduced to soil accumulate in it. They have been found in arable land in concentrations ranging from several milligrams to several grams per kilogram of soil under study (Du and Liu [@CR18]). Plants grown in soil contaminated with antibiotics take them up and accumulate them in various organs. Hu et al. ([@CR33]) showed antibiotics to be present at the highest concentration in leaves, followed by shoots and roots. They also pointed out that growth and development of plants affects the distribution of antibiotics in tissues. VAs absorbed by plants may inhibit photosynthesis, growth of shoots and roots, reduce the leaf area and increase the production of ROS (Di Marco et al. [@CR15]). EC~25~ and EC~50~ determined for the growth of roots of yellow lupin in this study were 0.2 and 0.67 mM (Table [1](#Tab1){ref-type="table"}), respectively. Hillis et al. ([@CR32]) found EC~25~ and EC~50~ for the length of *Lactuca sativa*, *Medicago sativa* and *Daucus carota* roots to range from 39 μg L^−1^ (i.e. 1.08 × 10^−5^ mM) to over 10,000 μg L^−1^ (i.e. 0.00277 mM). The toxicity of LEV towards shoots was higher and ranged from 0.17 to over 1.24 mM for EC~25~ and EC~50~, respectively. Our study has also shown LEV to be more toxic towards roots than towards shoots (Tables [1](#Tab1){ref-type="table"} and [2](#Tab2){ref-type="table"}). However, the LEV toxicity determined in this study was much lower than that reported by Hillis et al. ([@CR32]). This probably resulted from the differences in the way LEV was applied. We applied it directly to the reference soil in a PHYTOTOXKIT. It must be emphasised that soil can limit sorption of antibiotics, thereby reducing their reactivity and biological availability. Moreover, the content of organic carbon and clay in soil, the ionic strength and the pH can change the sorption mechanisms (Wegst-Uhrich et al. [@CR69]). There are also strong sorptive interactions between drugs and soil, which may prevent antibiotic release and dispersion. These properties determine fate of the drugs in the environment and also limit antibiotic bioavailability (Hu et al. [@CR33]). Moreover, absorption of antibiotics by plants is affected by the type of VAs, plant species and their root system. The uptake of pharmaceutics from the environment by roots and their biological activity *in planta* depends on the *n*-octanol-water partition coefficient of the VAs (Seo et al. [@CR58]). The toxicity of LEV may be suggestive of poorly developed stress tolerance mechanisms in seedlings. Antibiotics absorbed by plants induce oxidative stress (Di Marco et al. [@CR15]) whose level can be determined by establishing the content of ROS (O~2~ **∙** ^**−**^, ∙OH, H~2~O~2~, ^1^O~2~) in plant cells. Reactions catalysed by oxidative stress enzymes create the main mechanism of scavenging and neutralising ROS (Wang et al. [@CR67]). Adverse environmental factors may trigger an oxidative burst---an intracellular increase in the ROS level (Sofo et al. [@CR59]). On the other hand, a change in the activity of the oxidative stress enzymes may indicate increased production of ROS in cells (Wen et al. [@CR70]). The activity of two oxidoreductases, guaiacol peroxidase (GPX) and catalase (CAT), are upregulated in lupin seedlings grown in soil contaminated with levofloxacin (Fig. [2](#Fig2){ref-type="fig"}), and the rate of this stimulation depends on antibiotic concentration in soil (Fig. [2](#Fig2){ref-type="fig"}a--f). There are no literature data on levofloxacin EC~50~ for GPX and CAT (Table [1](#Tab1){ref-type="table"}). According to our findings, catalase is a better toxicity biomarker than peroxidase. However, it must be emphasised that the activity of GPX and CAT in 12-day-old roots of lupin increased nine and four times, respectively, compared to the activity of GPX and CAT in roots of control seedlings (Fig. [2](#Fig2){ref-type="fig"}). An increase in reductase activity in tissues indicates that the amount of ROS in levofloxacin affected cell increases (Sofo et al. [@CR59]). A study carried out by Gujrathi and Linden ([@CR29]) revealed an association between oxidative stress and increased production of antioxidative enzymes in roots. Contamination of soil with tetracycline at low concentrations (from 0.5 to 10 mg L^−1^, i.e. from 1.1 × 10^−3^ to 2 × 10^−2^ mM or 25 to 300 mg L^−1^) results in a small increase in the activity of GPX and CAT in wheat roots. An increase in tetracycline concentration from 25 to 300 mg L^−1^(i.e. 6 × 10^−2^ to 0.7 mM) increases the activity of antioxidative enzymes by as much as 629.7 and 528.5% for CAT and GPX, respectively (Xie et al. [@CR71]). A similar increase in the activity of the enzymes was observed in 12-day-old roots of lupin when soil was contaminated with a 2 mM (140 mg kg^−1^ of soil) solution of levofloxacin (GPX---663.1% and CAT---278.2%; Fig. [2](#Fig2){ref-type="fig"}). Di Marco et al. ([@CR15]) also reported an increase in the antioxidative enzyme activity in plants which grew in soil contaminated with antibiotics. The toxicity coefficients determined for catalase and peroxidase indicate that the toxicity units grow significantly when soil is contaminated with LEV at low concentrations (Tables [1](#Tab1){ref-type="table"} and [2](#Tab2){ref-type="table"}). The metabolic responses of seedlings are much more rapid than the morphological changes. Under normal environmental conditions, the production and neutralisation of oxygen metabolites, potentially toxic, is balanced and the amounts of these molecules are strictly limited. At low levels, non-toxic to cells, reactive oxygen species can be used as signal molecules (Baxter et al. [@CR4]). An increase in the ROS level may result in DNA, protein and lipid damage (Krishnamurthy and Rathinasabapathi [@CR39]) and, ultimately, in programmed cell death (PCD) (Petrov et al. [@CR52]). Visualisation (Fig. [3](#Fig3){ref-type="fig"}) of ROS demonstrated the toxicity of levofloxacin (2.5 mM) and the presence of oxidative stress in lupin roots. The oxidative burst is visible in cells as early as after 24 h of plant growth in soil contaminated with the antibiotic. Microscopic photographs show that ROS are first distributed fairly evenly in the cellular space (Fig. [3](#Fig3){ref-type="fig"}); however, after 2 days they are mainly localised along the cell wall. Non-specific metabolic disturbances are observed in root cells. The effect of drugs (diclofenac, paracetamol) on stress induction in cells has been demonstrated earlier by Kummerová et al. ([@CR40]). Reactive oxygen species and reactive nitrogen species (RNS) have been detected in stressed *Lemna minor*, especially in the tips and elongation zone of roots. Visualisation of reactive oxygen species in lupin roots 48 h after LEV was added to soil shows that they had accumulated near the cell membrane and cell wall (Fig. [3](#Fig3){ref-type="fig"}). ROS produced in plants under stress conditions cause direct or indirect oxidative modification of proteins (Das and Roychoudhury [@CR14]). Proteomic studies comparing plant proteins in extreme and normal conditions confirm the toxicity of levofloxacin. Depending on the tolerance of plants to abiotic stress, protective proteins can accumulate in them with different intensity (Rodziewicz et al. [@CR56]). Changes of gene expression and protein accumulation are among the first defence reactions of plants to abiotic stress (Gong et al. [@CR27]). This study shows the effect of levofloxacin on changes in the lupin root proteome (Figs. [4](#Fig4){ref-type="fig"}, [5](#Fig5){ref-type="fig"} and [6](#Fig6){ref-type="fig"}). Among the 24 identified proteins (Table [3](#Tab3){ref-type="table"}), there were nine directly related to the defence and response of plants to abiotic stress (Fig. [4](#Fig4){ref-type="fig"}a, b). The 2D maps show the accumulation of protective proteins and the ones related to stress responses; these were full = protein LlR18A, class III chitinase, ascorbate peroxidase, germin-like protein 3, extensin peroxidase and heat shock cognate protein 80-like (Table [3](#Tab3){ref-type="table"}). The identified proteins serve roughly the same function, however, through different mechanisms. Chitinases are PR (pathogenesis-related) proteins whose role was initially identified only as plant protection against pathogens (viruses, fungi, bacteria) and pests (Grover [@CR28]; Laribi-Habchi et al. [@CR43]). Shrimp shell chitinase and chitin, as well as chitinase produced by *Bacillus licheniformis* (ChiA), have similar activity (Benhabiles et al. [@CR5], [@CR6]; Laribi-Habchi et al. [@CR44]). Due to its antimicrobial properties, chitosan has been applied as a fruit preservative (Benhabiles et al. [@CR7], [@CR8]). Moreover, chitin and its derivatives show cytotoxic action towards THP-1 (human monocytic leukaemia cell line), Hep2 (human larynx carcinoma cell line) and Rd cells (human embryo rhabdomyosarcoma cell line) (Salah et al. [@CR57]; Bouhenna et al. [@CR9]). Decrease in molecular mass of chitin decreases its affinity towards chitosan (Dziril et al. [@CR19]) and enhances its anticancer activity (Salah et al. [@CR57]). It has been shown that synthesis of chitinases is also induced under abiotic stress conditions, such as drought, salinity, the presence of heavy metals (Kasprzewska [@CR38]) and osmotic stress (Grover [@CR28]). Kadouche et al. ([@CR36]) have demonstrated that chitosan can facilitate the ecological removal of heavy metals. When applied in a suspension containing also hydroxyapatite and metals, chitosan acts as flocculant and accelerates the sedimentation process. The accumulation of class III chitinase (IF3) in roots of seedlings which grow in soil contaminated with LEV proves the involvement of a plant defence reaction (Fig. [6](#Fig6){ref-type="fig"}a--c). An increase in the expression of the class III chitinase gene was also shown in organs of *Arabidopsis thaliana*, which were exposed to environmental stresses, especially salt stress (Takenaka et al. [@CR61]).The level of ROS increases in plants exposed to environmental stress and can cause cell damage. The glutathione-ascorbate cycle plays a crucial role in antioxidative defence mechanisms, in which ascorbate peroxidase catalyses transformation of H~2~O~2~ to H~2~O (Caverzan et al. [@CR12]). Chai et al. ([@CR13]) observed an increase in the level of ascorbate peroxidase in *A. thaliana* exposed to abiotic stress*.* A similar accumulation of APX takes place in roots of lupin grown in soil contaminated with levofloxacin (Fig. [6](#Fig6){ref-type="fig"}; Table [3](#Tab3){ref-type="table"}). A glycoprotein called germin-like protein 3 (GLP) is involved in the defence reaction against the toxic action of H~2~O~2~. The activity of GLPs is associated with enzymatic activity of OXO and superoxide dismutase (production of H~2~O~2~), ADP glucose pyrophosphatase/phosphodiesterase (AGPPase) and serine protease inhibitors (Wang et al. [@CR66]). Cell wall proteins, such as extensin proteins, take part in plant response to various stressors (Lamport et al. [@CR42]). Extensins can mediate a reaction of H~2~O~2~ binding by peroxidase (Brownleader et al. [@CR11]). Increasing the content of H~2~O~2~ in cells induces the binding of extensins to peroxidases (Prxs---class III plant peroxidases), \>which produce an insoluble network (EPs). EPs perform a defensive function by reinforcing the cell wall (Almagro et al. [@CR2]). There is a molecular chaperon, heat shock protein 80-like (hsc80), which is responsible for maintaining the proper structure and regulation of specific target proteins. The hsc80 protein plays a key role in the regulation of the cell cycle and signal transduction. Moreover, it is indirectly responsible for maintaining cellular homeostasis during plant response to stress (Tran et al. [@CR63]). Accumulation of the hsc80 protein in roots may indicate that some additional processes are triggered which are necessary to protect cells from the stressor. Soil supplementation with levofloxacin reduces the accumulation of 14-3-3-like protein A in lupin roots. 14-3-3 Proteins play a role in a plant's response to stress by regulating the action of many proteins involved in signal transduction (Roberts et al. [@CR54]). However, regulation of nitrogen and carbon metabolism in plants is the main role of 14-3-3-like proteins (Diaz et al. [@CR16]). Reduction of protein synthesis may be a sign of sensitivity and intolerance of lupin seedlings to soil contamination with antibiotics (Figs. [4](#Fig4){ref-type="fig"}, [5](#Fig5){ref-type="fig"} and [6](#Fig6){ref-type="fig"}). A slight decrease in the accumulation of the 14-3-3 proteins was observed by Laino et al. ([@CR41]) and Hurkman et al. ([@CR34]) in wheat subjected to thermal stress. Synthesis of some proteins responsible for the regulation of energy metabolism was reduced. The activity of the glycolytic enzyme, glyceraldehyde-3-phosphate dehydrogenase, lipoxygenase, pyruvate dehydrogenase, leghemoglobin reductase-like and F1 ATPase was found to decrease. At the same time, studies revealed an accumulation of enolase, which is a protein responsible for carbohydrate metabolism. Enolase takes part in the process of glycolysis and catalyses the transformation of 2-phosphoglycerate (2PEP) into phosphoenolpyruvate (PEP) (dehydration) (Pan et al. [@CR51]). Conclusion {#Sec17} ========== The wide range of changes in protein synthesis in lupin roots indicate the diversity of plant metabolic reactions affected by soil contamination with antibiotics and suggest lupin is a plant with a rather high sensitivity to such stress. The contents and distribution of ROS suggest the antibiotic-induced impairment of redox reaction network, particularly in the vicinity of the cell membrane and cell wall. Guaiacol peroxidase and catalase are sensitive indicators of levofloxacin toxicity. On a morphological level, the levofloxacin soil contamination may result in over 50% inhibition of root and shoot growth and a similar reduction in seedling fresh mass. On the other hand, the dry mass of levofloxacin-affected seedlings increases by over 50%. The values of toxicity indexes and EC~50~ for levofloxacin induced disturbances in lupin are given. CAT : Catalase DCF : Dichlorofluorescein FQ : Fluoroquinolone GPX : Guaiacol peroxidase LEV : Levofloxacin ROS : Reactive oxygen species TU : Toxicity indexes VA : Veterinary antibiotics This project was supported by the National Science Centre, Poland, N N 305 275440. Conflict of interest {#FPar1} ==================== The authors declare that they have no competing interests. Ethical statement {#FPar2} ================= Our work complies with the ethical rules applicable for this journal. [^1]: Responsible editor: Yi-ping Chen

Overview of cancers =================== Bladder cancer -------------- Bladder cancer refers to any of several types of malignant growths of the urinary bladder, about 90--95% of which are transitional cell carcinomas (TCC); the remaining are squamous cell carcinomas and adenocarcinomas ([@bib50]). Every year in the United Kingdom almost 10 200 people are diagnosed with bladder cancer, with \>4800 deaths, accounting for around 1 in 20 of all cancer registrations and 1 in 30 cancer deaths ([@bib12]). In GB, the age-standardised incidence rates increased throughout the 1970s and 1980s to reach a peak in the late 1980s, although the numbers of deaths have remained steady in recent years ([@bib50]). In most European countries, including England and Wales, bladder cancer is at least three times less frequent in women than in men, which has been seen as partly due to different smoking habits and also an indication for an occupational origin ([@bib39]; [@bib47]). Patients with superficial non-penetrating tumours have an excellent prognosis with 5-year survival rates between 80 and 90%, whereas patients with muscle-invasive tumours have 5-year survival rates of \<50%. Population-based bladder cancer survival rates have changed very little between the late 1980s and the late 1990s, with men having a persistent 6--10% survival advantage ([@bib53]). Many studies have suggested ∼40 potentially high-risk occupations ([@bib56]). Despite this, the relationship between many of these occupations and bladder cancer risk is unclear, with evidence of a strong association for a few occupations: aromatic amine manufacturing workers, dyestuffs workers and dye users, painters, leather workers, aluminium workers and truck drivers ([@bib56]). Tobacco smoking and occupational exposure to aromatic amines (AAs) are two established environmental risk factors for bladder cancer, and controlling exposure to these has been an important contributor to the reduction in mortality, particularly among men ([@bib48]). Up to 40% of all male and 10% of female cases might be ascribable to smoking ([@bib12]); the International Agency for Research on Cancer (IARC) has suggested that the proportion of cases attributable to prolonged smoking in most countries is of the order of 50% in men and 25% in women ([@bib30]). The relative risks (RR) are around 2- to 4-fold ([@bib51]; [@bib68]; [@bib12]). Kidney cancer ------------- This can refer to cancer of the renal cells only (renal cell carcinoma; RCC) or can include the less-common cancers of the renal pelvis, ureter and other non-bladder urinary organs such as the urethra (transitional cell carcinoma; TCC). Kidney TCCs are closely associated with bladder cancers ([@bib50]). Although the incidence of kidney cancers has increased in Caucasian populations, RCC has increased more rapidly than TCC. More men are affected than women, and most cases occur in the age range of 50--70 years ([@bib50]). The five-year RCC survival rate is currently about 50%, but if detected early enough may be \>80%. Incidence, mortality and survival rates for kidney cancer in Great Britain are in the midpoint of the global range ([@bib50]). Cigarette smoking is the most well-established risk factor associated with RCC and particularly TCC, but other risk factors, which include body weight, diet, pre-existing kidney disease and genetic predisposition, have been identified ([@bib40]). Although kidney cancer is not generally considered to be occupationally associated, occupational agents/exposure scenarios associated with RCC include asbestos, trichloroethylene (TCE), tetrachloroethylene, polycyclic aromatic hydrocarbons (PAH), diesel engine exhaust (DEE), heavy metals (e.g., cadmium, lead), polychlorinated biphenyls, coke production, oil refining and gasoline/diesel delivery ([@bib22]). Occupational associations for transitional cell carcinomas of the kidney resemble those for bladder cancer ([@bib40]) and include links to dyes or employment in leather and shoe manufacturing. Cancers of the renal pelvis and ureter have also been associated with exposure to coal/coke, natural gas and mineral oils, or employment in dry cleaning, iron and steel, chemical and petroleum refining industries ([@bib36]). Methods ======= Occupational risk factors ------------------------- ### Group 1 and 2A human carcinogens The agents that the IARC has classified as either definite (Group 1) or probable (Group 2A) human carcinogens for urinary cancers are summarised in [Table 1](#tbl1){ref-type="table"}. The IARC has not identified any agent that is common to both kidney and bladder cancers. Causes for kidney cancer include work in coke production (Group 1) and exposure to TCE (Group 2A). For bladder cancer, these include mineral oils (Group 1), magenta manufacture (Group 1), auramine manufacture (Group 1), aluminium production (Group 1), work as a painter (Group 1), work in the rubber industry (Group 1), boot and shoe manufacture/repair (Group 1), exposure to AAs (Group 1/2A), PAHs (Group 1/2A), DEE (Group 2A), work as hairdressers and barbers (Group 2A), intermediates in plastic and rubber manufacturing (Group 2A) and petroleum refining (Group 2A). Choice of studies providing risk estimates for urinary tract cancers -------------------------------------------------------------------- Detailed reviews of occupational risk factor studies identified for urinary tract cancer are provided in the relevant Health and Safety Executive (HSE) technical reports ([@bib21], [@bib22]). Occupational exposures considered for bladder cancer ---------------------------------------------------- ### Aluminium production The predominant exposure in aluminium manufacture is to PAHs, and thus those working in this industry will be considered among the estimates for PAHs. Although high risks of bladder cancer have been reported for those working in the aluminium industry, the causative agent(s) is unknown or unproven ([@bib31]). Most of these workers were concurrently exposed to aluminium dust, or fumes containing other known carcinogens such as tobacco smoke or PAHs. The PAH exposures mostly originate from the evaporation of carbon electrode materials used in the electrolysis process ([@bib26], [@bib27], [@bib28]). For a more comprehensive summary of the studies undertaken, refer to the relevant HSE technical report ([@bib21]). Studies generally report that the incidence of bladder cancer is increased in this industry, but since 1980 there has been a downward trend in incidence. Only studies in Canada demonstrate an exposure--response relationship. In a UK case--control study of 80 urothelial cancer cases, there was an almost doubling of the risk among those involved in aluminium refining/smelting ([@bib61]). ### Aromatic amines Human exposure to AAs has been associated with an increased risk of bladder cancer, but their use has continued because of their industrial and commercial value. Well-established occupational causes of bladder cancer include the AAs 2-naphthylamine (*β*-naphthylamine), benzidine, 4-aminobiphenyl and chloraphazine ([@bib31]; [@bib69]; [@bib55]; [@bib11]). Aromatic amines have been used as antioxidants in the production of rubber and in cutting oils, as intermediates in azo dye manufacturing and as pesticides. They are also a common contaminant in several working environments, including the chemical and mechanical industries and aluminium transformation, and are widely used in the textile industry. Occupational exposures to AAs may explain up to 25% of bladder cancers in some Western countries ([@bib70]; [@bib69]). ### Risk estimates for occupational exposure to AAs and bladder cancer For the present study, a series of risk estimates were used for different groups of workers. These were selected from the study by [@bib61] who investigated occupational exposure to AA and estimated the risk of developing urothelial cancer on the basis of a hospital-based case--control study in the West Midlands in the United Kingdom. Smoking-adjusted relative risks (RR) of \>2.0 were obtained for seven occupations, including dyestuff manufacture (RR=2.61, 95% CI=0.98--7.00), leather work (RR=2.51, 95% CI=1.44--4.35), cable manufacturing (RR=2.46, 95% CI=1.20--5.04) and textile printing and dyeing (RR=2.32, 95% CI=0.98--5.45). Sorahan\'s RR estimates for the manufacture of rubber products (RR=1.89, 95% CI=1.34--2.66), plastics (RR=1.73, 95% CI=1.17--2.55) and organic chemicals (RR=1.70, 95% CI=1.05--2.76) are used for industrial exposure to 4,4′-methylenebis(2-chloroaniline) (MOCA). The authors also provided an estimate of RR for medical and nursing occupations (RR=1.62, 95% CI=1.03--2.55), as well as for laboratory technicians (RR=1.05, 95% CI=0.60--1.86), which have been used for lower-level exposure groups. In a review of studies, [@bib69] reported considerable increased risks of bladder cancer in workers exposed to 2-naphthylamine, benzidine and 4-aminobiphenyl, but also stated that some of these studies were poorly designed and/or based on very small numbers ([@bib70]; [@bib69]). A review of 11 European case--control studies also concluded that about 5--10% of bladder cancers in men could be attributed to occupational exposures, including, but not specifically, AAs ([@bib37]). Studies of *ortho*-toluidine and aniline have demonstrated elevated risks associated with bladder cancer. The use of 2-naphthylamine, benzidine and other carcinogenic arylamines has now been banned ([@bib44]; [@bib65]). ### Auramine and magenta manufacture These exposures have been included in the general estimation of AAs and dyestuffs. Although IARC has classified the manufacture of auramine as Group 1, the responsible carcinogens are not known ([@bib31]). The IARC summarises one epidemiological study that suggests auramine manufacture as an occupational bladder cancer risk ([@bib24]). [@bib9] also observed a significant risk among dyestuff workers in the United Kingdom coming into contact with auramine. [@bib31] also concluded that the manufacture of magenta entails exposures that are carcinogenic, but that overall there is inadequate evidence for human carcinogenicity of this dye. The manufacturing of magenta II involves the use of *ortho*-toluidine, formaldehyde, nitrotoluene, and for magenta the use of aniline, *ortho*- and *para*-toluidines and their hydrochlorides, nitrobenzene and *ortho*-nitrotoluene. In two studies of manufacturers from the United Kingdom and Italy, the risk of bladder cancer was highly significant ([@bib9]; [@bib52]); in both cases there was evidence of *ortho*-toluidine and 4,4′-methylene-bis(2-methyl aniline) exposure, implicating these compounds in the increased rate of bladder cancer mortality observed. ### Diesel engine exhaust An effect of DEE on the occurrence of bladder cancer is plausible because metabolites of PAH are present in DEE and are concentrated in the urine and may interact with the urothelium of the bladder ([@bib57]). Exposure to DEE occurs in many occupational settings, and levels of PAHs in this fume are highest in emissions from heavy-duty diesel engines and lower (and comparable) in emissions from light-duty diesel engines and from petrol engines without catalytic converters ([@bib5]). Professional drivers, mechanics and people working in other related professions are exposed to elevated levels of emissions from combustion engines ([@bib19]). ### Risk estimates for occupational exposure to DEE and bladder cancer For this current study, a suitable RR was calculated by the research team as 1.24 (95% CI=1.10--1. 41) for the 'high-exposed\' group using estimates from studies reviewed by [@bib6]. The RR was estimated using a random-effects model based on an overall inverse-variance-weighted average of all RRs from the studies based on cancer incidence, but excluding overlapping exposure categories. [@bib6] reviewed 35 European and North American epidemiological studies published between 1977 and 1998 that examined the risk for bladder cancer and exposure to DEE among highly exposed workers (e.g., railroad workers, garage maintenance workers, truck drivers and drivers and operators of heavy machines in ground and road construction). They classified exposure to DEE using a job-exposure matrix (JEM) or an experts\' assessment of individual occupational histories. Most of the cohort studies included did not control for smoking, whereas a majority of the case--control studies did. The studies based on routinely collected data were assumed to have been adjusted for smoking. An overall meta-analysis of the data was not carried out by [@bib6] because the results were heterogeneous owing to the fact that the studies used different definitions of exposure. Although the review did not offer an overall summary RR, because of this heterogeneity the value they calculated for all studies (RR=1.18, 95% CI=1.08--1.28) was in line with their observation of an overall RR in the range 1.1--1.3. For low-exposure groups, an RR (based on a fixed effects model) of 1.03 (95% CI=0.84--1.26) was estimated. This was based on a reassessment of 6 of 10 studies examined by [@bib6], for which they were able to classify exposure as 'low\' on the basis of JEM. They noted that among these 10 studies although there were a few positive results (three above RR=1.1) most were close to unity. ### Hairdressers and barbers Elevated risk of bladder cancer in association with occupational exposure to hair dyes in hairdressers, barbers, beauticians and cosmetologists has been widely reported; for a wider review of these studies, refer to the relevant HSE report ([@bib21]). Hairdressers have used a wide range of chemical products, including hair colourants and bleaches, shampoos and conditioners (e.g., primarily AAs, aminophenols and hydrogen peroxide; nitro-substituted AAs, aminophenols, aminoanthraquinones and azo dyes; and metal salts). However, the individual chemicals used have varied over time, and only permanent and semi-permanent colourants are now used to a significant extent by hairdressers ([@bib33]). ### Risk estimates for employment as hairdressers and barbers and bladder cancer In the present study, a standardised incidence ratio (SIR) of 1.22 (95% CI=0.98--1. 51) was used for men, and an SIR of 1.09 (95% CI=0.81--1.43) for women. These figures were based on a study by [@bib13] who followed up a large cohort of \>45 000 male and female Swedish hairdressers recruited via the national census. These estimates were not adjusted for smoking. The study observed an increased risk among men and women irrespective of the census period over 39 years. The study reported the highest SIR of 2.56 for urinary bladder cancer in male hairdressers working in 1960 and followed up from 1960 to 1969. However, this risk decreased to 1.25 when these hairdressers were followed up for the whole period of 1960--1998. This reduction may have been because of the reduced use of brilliantine in male hair ([@bib20]; [@bib60]). Other Scandinavian studies of hairdressers have shown a statistically significant increased risk for bladder cancer in Sweden, Norway, Finland and Denmark ([@bib60]). ### Mineral oils Exposure to mineral oils, and in particular shale oil, is of historical interest, but metalworking fluids (MWF) appear to be the predominant route for exposure to mineral oils during the study period of interest. A comprehensive and systematic NIOSH review examined the association between MWF exposure and bladder cancer, and concluded that this association is well supported by studies from different geographical locations employing different study designs, all of which controlled for smoking ([@bib8]; [@bib42]). ### Risk estimates for occupational exposure to mineral oils and bladder cancer In the present study, the research team used papers from a review by [@bib66] describing the relationship between mineral oil and cancer to calculate a risk estimate of 1.39 (95% CI=1.20--1.61), assuming a random-effects model owing to significant heterogeneity across the different study results. This RR was used for those in situations of high exposure to MWF. The studies examined by [@bib66] reported substantial dermal and inhalation exposure for occupations such as metal machining, print press operating, and cotton and jute spinning. The overall RR for bladder cancer related to exposure to mineral oils was taken as a weighted average across these case--control and population-based studies reviewed, and reflected incidence and mortality due to bladder and other urinary tract organs. Owing to the absence of sufficient exposure--response data for mineral oils, an RR of 1 was assigned to the lower and background exposure categories of workers in the Labour Force Survey (LFS) data set. ### Occupation as a painter Many chemicals are used in paint products such as pigments, extenders, binders, solvents and additives. Painters are commonly exposed by inhalation to solvents and other volatile paint components; inhalation of less volatile and non-volatile compounds is common during spray painting. ### Risk estimates for employment as a painter and bladder cancer For this study, an overall RR of 1.17 (95% CI=1.11--1.27) was used to estimate the attributable fraction (AF) for work in this occupation. This overall estimate was obtained from a study by [@bib7] who systematically reviewed all epidemiological studies on bladder cancer in painters published after the [@bib32] monograph (this followed changes to paint technology and the nature of the exposures). This study examined cohort and case--control investigations from Europe and the United States, including one from the United Kingdom, up until 2004. Several RR estimates were obtained, including a pooled RR of 1.10 (95% CI=1.03--1.18) for four cohort studies, a pooled RR of 1.23 (95% CI=1.11--1.37) for mortality estimates and a pooled RR of 1.35 (95% CI=1.19--1.53) from 14 case--control studies, which included a pooled analysis of another 11 case--control studies. Other studies have lent weight to this evidence of an association between bladder cancer and painting as an occupation. [@bib10] undertook a meta-analysis of published papers (1966--1998) from Europe, North America, New Zealand and China, and derived a combined standardised mortality ratio (SMR) of 1.30 (95% CI=1.14--1.50) for bladder cancer, based on 17 follow-up studies of painters. Occupational cohort studies also provided a combined SMR of 1.26 (95% CI=0.98--1.62). In the United Kingdom, a hospital-based case--control study of urothelial cancers by [@bib61] obtained a significant RR of 1.91 (95% CI=1.41--2.91) in occupations specified as manufacturing paints or in the professional use of paints. ### Polycyclic aromatic hydrocarbons Polycyclic aromatic hydrocarbons are formed by the incomplete combustion of carbon-containing fuels such as wood, coal, diesel, petrol, fat or tobacco. Polycyclic aromatic hydrocarbons are produced in a number of occupational settings including coal gasification, coke production, coal-tar distillation, chimney sweeping, coal tar and pitches, carbon black manufacture, carbon and graphite electrode manufacture, creosotes and others ([@bib26], [@bib27], [@bib28], [@bib29]). Higher risks have been reported for specific categories of painters, metal, textile and electrical workers, miners, transport operators, excavating-machine operators and also for non-industrial workers such as concierges and janitors ([@bib37]). Industries entailing a high risk included salt mining, manufacture of carpets, paints, plastics and industrial chemicals. A number of studies have documented an increased risk among workers exposed to petrochemicals and combustion products in different industries, suggesting an association with PAHs, and to their nitro-derivatives as well as DEE ([@bib49]; [@bib55]). ### Risk estimates for occupational exposure to PAHs and bladder cancer For the present study, an overall RR of 1.4 (95% CI=1.2--1.7) was based on 26 studies and was applied to those in 'high-exposed\' groups in the manufacturing industry. This overall RR was obtained from [@bib5] who reviewed the cancer risk from occupational and environmental exposure to PAHs, in aluminium production, coal gasification, coke production, iron and steel foundry work, DEE exposure, and workers exposed to coal tars and related products (i.e., tar distillation, shale oil extraction, creosote exposure, carbon black manufacture, carbon and graphite electrode manufacture, chimney sweeps and calcium carbide production). Results from all these sectors, with the exception of DEE exposure (for which AFs are calculated separately) and coke production (for which no evidence was found for a raised risk of bladder cancer), were used to calculate an inverse-variance-weighted combined estimate of RR. The RR for the 'low-exposed\' group was set to 1 on the basis of combined OR estimates from population-based case--control studies covered in the same review [@bib5]. These included lower relative risk estimates of 0.9 (95% CI=0.8--1.1) for a large Montreal case--control study, and 1.2 (95% CI=1.1--1.4) for a range of other smaller studies (both used a random-effects model). Exposures to DEE and driving, and to mineral oils (cutting fluids), were again excluded from this current analysis. ### Rubber industry In 1982, IARC classified work in the rubber industry as Group 1([@bib25]), with numerous studies providing strong evidence that workers in the rubber industry had elevated risks for bladder cancer ([@bib54]; [@bib71]; [@bib38]; [@bib11]), and continued to be at risk even among former workers ([@bib18]). ### Risk estimates for employment in the rubber industry and bladder cancer More recent evidence suggests that the risk for bladder cancer ceased in the rubber industry in GB after 1950, and thus no AF has been calculated. For example, in a recently defined UK cohort of workers employed between 1982 and 1991, \>8000 workers from 41 factories were followed up through 2004 ([@bib16]). For both men and women, the incidence was nonsignificantly raised, and among men mortality was raised nonsignificantly. An analysis of 6500 male workers at a UK tyre factory who worked between 1946 and 1960 also found a nonsignificant excess in mortality ([@bib67]). Mortality studies have been undertaken by the British Rubber Manufacturer\'s Association ([@bib45]; [@bib46]; [@bib62]) and by HSE ([@bib3]). Detailed studies have also been undertaken because of inadvertent exposure of some workers to 2-naphthylamine (used as an antioxidant) ([@bib21]). In the most recent of these studies, men employed in the period 1945--1949 were compared with those first employed after January 1950 (when 2-naphthylamine was removed) and were followed up until 1995. Overall, bladder cancer incidence significantly increased (standardised risk ratio (SRR)=1.23, 95% CI=1.02--1.48), especially for those employed in the period 1945--1949 (SRR=1.71, 95% CI=1.30--1.48) compared with those first employed after 1950 (SRR=1.02, 95% CI=0.72--1.39). Occupational exposures considered for kidney cancer --------------------------------------------------- ### Coke production Coke production has been classified as an IARC Group 1 carcinogenic occupation ([@bib28]) related to the exposure to PAHs. However, as there are few reports of a direct association between PAH exposure and kidney cancer (unlike for bladder cancer), it is not possible to derive a relative risk estimate for PAH as a causal factor of this cancer ([@bib5]). ### Risk estimates for coke production and kidney cancer The most appropriate study (considering cohort size and period of follow-up) for coke-oven workers in Britain was provided by [@bib23] who found risk estimates of 1.16. for coke-oven workers employed in 6 British steel industry plants (2790 men) and a deficit of risk of 0.16 for coke-oven workers at 13 National Smokeless Fuels plants in Britain (3883 men). If these two are combined, an overall SMR of 0.58 is obtained, showing no excess of kidney cancer. As this combined SMR was \<1, no AF calculation was carried out for coke-oven workers. Another study by [@bib14] reported an SMR of 2.52 for kidney cancer but in a much smaller cohort of 610 coke-oven workers. ### Trichloroethylene Trichloroethylene is not widely used today, but has been used as a metal degreasant/solvent in a range of manufacturing industries including rubber, textile and paint manufacture, and as a dry cleaning solvent ([@bib34]). ### Risk estimates for occupational exposure to TCE and kidney cancer For the present study, the relative risk estimate for 'higher-exposure\' situations was 1.2 (95% CI=0.8--1.7), based on an average SMR calculated by [@bib72]. [@bib72] provided a robust methodology to evaluate 20 cohort and 40 case--control studies of TCE exposure. They divided these cohort studies into three tiers based on the specificity of the exposure information and consideration of confounding influences from other exposures. For those situations considered as 'low exposures\', the RR has been set to 1 based on a study by [@bib41] that provided a low-exposure SMR of 0.47 (95% CI=0.01--2.62). The majority of cohort studies of TCE exposure report small, usually nonsignificant, elevations or deficits in kidney cancer associated with TCE exposure ([@bib21]). As all the studies include exposure to a combination of organic solvents, of which TCE may be a component, it is difficult to determine whether TCE is the causal factor, suggesting that exposure is more likely to be to organic solvents rather than to TCE specifically. Renal cell carcinoma is usually associated with occupational exposures to TCE ([@bib34]), but links have been made with exposure to TCC; thus, the risk estimate can be used for attributable risk estimation for all kidney cancers. Estimation of numbers ever exposed ---------------------------------- The data sources, major industry sectors and jobs for estimation of numbers ever exposed over the REP, defined as the period during which exposure occurred that was relevant to the development of the cancer in the target year 2005, are given in [Table 1](#tbl1){ref-type="table"}. For bladder cancer, AFs were estimated for occupational groups as a whole for painters and hairdressers/barbers. Coal tar and pitches, aluminium production, coal gasification, coke production and petroleum refining have all been included within PAHs. Auramine and magenta manufacture have been included under AAs. Exposure to boot and shoe manufacture/repair was considered only up to 1962 and was included with AAs. For the rubber industry, the risk was confined to before 1950 in the United Kingdom. The following occupations were designated as jobs with known exposure to soluble MWFs in large droplet form: press and machine tool setters, other centre lathe turners, machine tool operators, machine tool setter operators, press stamping and automatic machine operators, and toolmakers tool fitters markers out. In addition, there were a number of occupations assigned a background exposure, including foremen of metal polishers, shot blasters and fettlers dressers; metal polishers; fettlers, dressers and shot blasters. For PAHs, workers involved in the manufacture of industrial chemicals, miscellaneous products of petroleum and coal, as well as other non-metallic mineral products, and workers in the iron and basic steel and non-ferrous metal basic industries were assigned a high-exposure level. For DEE, workers in the metal ore and other mining industries, construction, land transport and services allied to transport were assigned to the high-exposure level. For TCE, high-exposure-category industries were taken as those manufacturing finished metal products where TCE was likely to have been used as a metal degreasant, as well as the textile industry for similar reasons. The CARcinogen EXposure Database (CAREX) records only 117 workers as being exposed to TCE in clothing manufacture, and thus it was assumed that 99% of these workers were men. For exposed service workers, it was assumed that 25% were men, based on numbers of drycleaners/launderers reported in the UK LFS 1979--2003 (19% male workers in 1979, 25% in 1991 and increasing to 38% in 2003). Over the current burden period of 1956--1996, the number of workers employed in coke production industries decreased from ∼20,000 to \<500. Results ======= Owing to assumptions made about cancer latency and working age range, only cancers in ages 25 years and above in 2005/2004 could be attributable to occupation. In the present study, a latency period of at least 10 years and up to 50 years has been assumed for urinary tract cancers. For bladder cancer, AFs have been calculated for exposure to mineral oil, AAs, PAHs (in coal tar and pitches, aluminium production, coal gasification, coke production and petroleum refining), DEE and for occupation as painters, and hairdressers and barbers. For the rubber industry, the risk for bladder cancer was confined to before 1950 in the United Kingdom; therefore, no AF was calculated. An AF has been calculated for kidney cancer for TCE exposure only. [Table 2](#tbl2){ref-type="table"} provides a summary of the attributable deaths and registrations in Britain for 2005 and 2004 and shows the separate estimates for men and women, respectively. For all exposure scenarios combined, the estimated overall AF for bladder cancer was 5.28% (95% CI=3.43--7.72%). This resulted in a total of 245 attributable deaths (95% CI=159--358) and 550 attributable registrations (95% CI=357--795). For all exposure scenarios combined, the estimated overall AF for kidney cancer was 0.04% (95% CI=0.00--0.15%). This resulted in a total of one attributable death (95% CI=0--5) and three (95% CI=0--10) attributable registrations. Exposures affecting bladder cancer ---------------------------------- There were 101 654 men and 94 170 women 'ever exposed\' to AAs over the REP. The overall AF for bladder cancer and exposure to AAs was estimated as 0.67% (95% CI=0.30--1.49%), with 31 (95% CI=14--69) deaths and 66 (95% CI=30--147) registrations, respectively. Male workers in manufacture of wire and cable, as well as textile finishing, industries were most at risk, whereas women workers in the dry cleaning industry were most at risk. There were an estimated 1 636 322 men and 426 949 women 'ever exposed\' to DEE over the REP. The overall total AF for bladder cancer and exposure to DEE was estimated as 1.00% (95% CI=0.17--2.03%), with 47 (95% CI=8--94) deaths and 106 (95% CI=18--214) registrations, respectively. Among men, workers in construction and land transport were at most risk. There were in total 4,426,581 men and 466,252 women 'ever exposed\' to mineral oils over the 40-year relevant period. The overall total AF for bladder cancer and exposure to mineral oils was estimated as 2.81% (95% CI=1.47--4.31%), with 131 (95% CI=68--200) attributable deaths and 296 (95% CI=155--452) registrations. Metal workers were at most risk (231 male and 21 female registrations), especially machine tool operators (157 men and 15 women). There were an estimated 334 339 men and 188 252 women 'ever exposed\' to PAHs. Work with coal tar and pitches, aluminium production, coal gasification, coke production and petroleum refining were all included under PAHs. The overall total AF for bladder cancer and exposure to PAHs was estimated as 0.07% (95% CI=0.03--0.11%), with three (95% CI=1--5) deaths and seven (95% CI=3--11) registrations, respectively. More than half of the total registrations occurred among workers in the iron and steel basic industries. For work as a painter, 1 118 813 men and 130 630 women were estimated to have 'ever worked\' in the occupation over the REP. The overall total AF for bladder cancer and work as a painter was estimated as 0.67% (95% CI=0.44--0.91%), with 31 (95% CI=20--42) deaths and 71 (95% CI=47--97) registrations, respectively. There were an estimated 96 041 male and 631 937 female hairdressers and barbers over the REP. The overall total AF for bladder cancer and work as a female hairdresser or barber was estimated as 0.16% (95% CI=0.00--0.63%), with 8 (95% CI=0--29) deaths and 15 (95% CI=0--56) registrations, respectively. Exposures affecting kidney cancer --------------------------------- Over the 40-year exposure period (1956--1996), a total of 43 861 male and 42 288 female workers were 'ever exposed\' to TCE, with the majority (∼85%) working in manufacturing industries. For women, there was a roughly equal split between manufacturing and services. Most exposures (\>95% for men and women) were classified as 'high\'. For kidney cancer, the estimated total AF was 0.04% (95% CI=0.00--0.15), with one (95% CI=0--5) attributable death and three (95% CI=0--10) attributable registrations. Attributable deaths/registrations corresponded to exposure in the manufacturing industries for both sexes. Discussion ========== Our estimate of the AF figure of 7.06% for all bladder cancer in men is lower than that of the 10% estimated by [@bib15] in their original critical review of literature, as well as the AF estimated by [@bib43] at 14.2% and the estimates for US white and non-white men of 21--27% ([@bib58], [@bib59]). However, our overall estimate of male and female bladder cancer of 5.3% is higher than the 2% obtained by [@bib17] for Nordic countries, and approximately the same as the 5.5% obtained for France by [@bib4]. It is also within the ranges observed in the qualitative reviews by [@bib70], [@bib35] and [@bib63], and a review of occupational studies in Italy ([@bib70]; [@bib69]; [@bib2]). Overall, the AF for kidney cancer obtained in this study was 0.04% (men and women), which is significantly lower than the figure quoted by [@bib43]; that is, 4.7% for men and 0.8% for women. It is possible that the overall AFs for kidney and bladder cancer reported here might be underestimated because certain agents and exposure circumstances were not considered. During 2009, IARC updated the critical review of Group 1 carcinogens as part of the 100th monograph. During this review process, the various committees reclassified arsenic as a Group 1 carcinogen for bladder cancer ([@bib64]), and soots and coal tars as Group 2A carcinogen ([@bib1]). Arsenic was also classified as a Group 2A carcinogen for kidney cancer, as was cadmium. The lack of consistency in occupational findings for both cancers may partly be explained by small sample size, recall bias, misclassification and low levels of exposures, inadequate adjustment for confounding factors and short duration of follow-up in some studies. British Occupational Cancer Burden Study Group ============================================== Lesley Rushton (PI)^\*,1^, Sanjeev Bagga^3^, Ruth Bevan^3^, Terry Brown^3^, John W Cherrie^4^, Gareth S Evans^2^, Lea Fortunato^1^, Phillip Holmes^3^, Sally J Hutchings^1^, Rebecca Slack^5^, Martie Van Tongeren^4^ and Charlotte Young^2^. ^1^Department of Epidemiology and Biostatistics, School of Public Health and MRC-HPA Centre for Environment and Health, Imperial College London, St Mary\'s Campus, Norfolk Place, London W2 3PG, UK; ^2^Health and Safety Laboratory, Harpur Hill, Buxton, Derbyshire SK17 9JN, UK; ^3^Institute of Environment and Health, Cranfield Health, Cranfield University, Cranfield MK43 0AL, UK; ^4^Institute of Occupational Medicine, Research Avenue North, Riccarton, Edinburgh EH14 4AP, UK; ^5^School of Geography, University of Leeds, Leeds LS2 9JT, UK. See [Appendix](#app1){ref-type="app"} for the members of the British Occupational Cancer Burden Study Group. The authors declare no conflict of interest. ###### Occupational agents, groups of agents, mixtures and exposure circumstances classified by the IARC monographs, Vols 1--88, into Groups 1 and 2A, which have the kidney and/or bladder as the target organ Agents, mixture, circumstance Main industry, use Evidence of carcinogenicity in humans Source of data for estimation of numbers ever exposed over REP Comments ---------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------ ---------------------------------------------------------------- ----------------------------------------------------------------------- **Group 1: Carcinogenic to humans** **Agents, groups of agents** Aromatic amine dyes 4-aminobiphenyl Benzidine 2-naphthylamine Production: dyestuffs and pigment manufacture Bladder *strong* CoE and LFS   Coal tars and pitches Production of refined chemicals and coal tar products (patent-fuel); coke production; coal gasification; aluminium production; foundries; road paving and construction (roofers and slaters) Bladder *suggestive* CAREX Included with PAHs Polyaromatic hydrocarbons Benzo(a)pyrene Work involving combustion of organic matter; foundries; steel mills; fire-fighters; vehicle; mechanics Bladder *suggestive* CAREX   Mineral oils, untreated and mildly treated Production; used as lubricant by metal workers, machinists, engineers; printing industry (ink formulation); used in cosmetics, medicinal and pharmaceutical preparations Bladder *suggestive* LFS             **Exposure circumstances** Coke production Coal-tar fumes Kidney *suggestive* Bladder *suggestive* LFS AF not calculated for kidney -- RR\<1; included with PAHs for bladder Aluminium production Pitch volatiles; aromatic amines Bladder *strong* CAREX Included with PAHs Auramine manufacture 2-naphthylamine; auramine; other chemicals; pigments Bladder *strong* CAREX Included with aromatic amines Magenta manufacture Magenta; ortho-toluidine; 4,4′-methylene bis(2-methylaniline); orthonitrotoluene Bladder *strong* CAREX Included with aromatic amines Rubber industry Aromatic amines; solvents Bladder *strong* CoE and LFS Risk confined to pre-1950 in GB Boot and shoe manufacture and repair Leather dust; benzene and other solvents Bladder *suggestive* CAREX Exposure up to 1962 included with aromatic amines Coal gasification Coal tar; coal-tar fumes; PAHs Bladder *strong* CAREX Included with PAHs Painters   Bladder LFS             **Group 2A: Probably carcinogenic to humans** **Agents, groups of agents** Trichloroethylene Production; dry cleaning; metal degreasing Renal *suggestive* CAREX LFS Included with PAHs -- benzo(a)pyrene above Polyaromatic hydrocarbons Dibenz(a,h)anthracene Cyclopenta(c,d)pyrene Dibenzo(a,l)pyrene Work involving combustion of organic matter; foundries; steel mills; fire-fighters; vehicle mechanics Bladder *suggestive* CAREX   Diesel engine exhaust Railroad, professional drivers; dock workers; mechanics Bladder *suggestive* CAREX   Intermediates in plastics and rubber manufacturing 4,4′-methylene bis(2-chloroaniline) Styrene-7,8-oxide Production; curing agent for roofing and wood sealing Production; styrene glycol production; perfume preparation; reactive diluent in epoxy resin formulations; as chemical intermediate for cosmetics, surface coating, and agricultural and biological chemicals; used for treatment of fibres and textiles; in fabricated rubber products Bladder *suggestive* CoE and LFS   Aromatic amine dyes Benzidine-based dyes 4-chloro-*ortho*-toluidine *Ortho*-toluidine Production; used in textile, paper, leather, rubber, plastics, printing, paint, and lacquer industries Dye and pigment manufacture; textile industry Production; manufacture of dyestuffs, pigments, optical brightener, pharmaceuticals, and pesticides; rubber vulcanising; clinical laboratory reagent; cleaners and janitors Bladder *suggestive* CAREX Included in aromatic amine           **Exposure circumstances** Hairdressers and barbers Dyes (aromatic amines, amino-phenols with hydrogen peroxide); solvents, propellants; aerosols Bladder *suggestive* LFS   Petroleum refining PAHs Bladder *suggestive* LFS Included with PAHs Abbreviations: CAREX=CARcinogen EXposure Database; CoE=Census of Employment; IARC=International Agency for Research on Cancer; LFS=Labour Force Survey; REP=relevant exposure period. ###### Urinary cancer burden estimation results for men and women Agent Number of men ever exposed Number of women ever exposed Proportion of men ever exposed Proportion of women ever exposed AF men (95% CI) AF women (95% CI) Attributable deaths (men) (95% CI) Attributable deaths (women) (95% CI) Attributable registrations (men) (95% CI) Attributable registrations (women) (95% CI) --------------------------------------- ---------------------------- ------------------------------ -------------------------------- ---------------------------------- ------------------------- ------------------------- ------------------------------------ -------------------------------------- ------------------------------------------- --------------------------------------------- **Bladder cancer** Aromatic amines 101,654 94,170 0.0052 0.0045 0.0070 (0.0034--0.0151) 0.0060 (0.0024--0.0144) 21 (10--46) 10 (4--23) 49 (24--106) 17 (7--41) Diesel engine exhaust 1,636,322 426,949 0.0843 0.0203 0.0145 (0.0026--0.0283) 0.0017 (0.0000--0.0054) 44 (8--86) 3 (0--9) 102 (18--198) 5 (0--15) Hairdressers/barbers 96,041 631,937 0.0050 0.0301 0.0011 (0.0000--0.0025) 0.0027 (0.0000--0.0134) 3 (0--8) 4 (0--21) 8 (0--18) 8 (0--38) Mineral oils 4,426,581 466,252 0.2228 0.0222 0.0392 (0.0205--0.0598) 0.0073 (0.0038--0.0113) 119 (62--182) 12 (6--18) 275 (144--420) 21 (11--32) PAHs 334,339 188,252 0.0172 0.0090 0.0008 (0.0004--0.0013) 0.0004 (0.0002--0.0007) 2 (1--4) 1 (0--1) 6 (3--9) 1 (1--2) Painters 1,118,813 130,630 0.0577 0.0062 0.0097 (0.0064--0.0132) 0.0011 (0.0007--0.0014) 30 (19--40) 2 (1--2) 68 (45--92) 3 (2--4) Totals**[a](#t2-fn2){ref-type="fn"}**         0.0706 (0.0457--0.0975) 0.0189 (0.0128--0.0386) 215 (139--296) 30 (21--62) 496 (321--684) 54 (37--110) **Kidney cancer** Trichloroethylene 43,861 42,288 0.0023 0.0020 0.0004 (0.0000--0.0016) 0.0004 (0.0000--0.0014) 1 (0--3) 1 (0--2) 2 (0--7) 1 (0--4) Abbreviations: AF=attributable fraction; CI=confidence interval; PAHs=polycyclic aromatic hydrocarbons. Totals are the product sums and are not therefore equal to the sums of the separate estimates of attributable fraction, deaths and registrations for each agent. The difference is especially notable where the constituent AFs are large.

The authors confirm that all data underlying the findings are fully available without restriction. Raw data are not suitable for public deposition as they contain identifying human information. Data are available upon request from the corresponding author. Introduction {#s1} ============ In Western countries multiple sclerosis (MS) is second only to trauma as cause of chronic neurological disability in young adults. Around 15% of MS sufferers have a progressive course from the outset (primary progressive MS); a further 35% develop progressive disease after a variable period with relapsing-remitting course (secondary progressive MS) [@pone.0109679-Kremenchutzky1]. Although new therapies can attenuate disease course, people with primary or secondary progressive MS lack effective treatment options [@pone.0109679-Humphries1]. Reduced mobility is one of the commonest and most visible impairments of people with progressive MS, but concurrent compromise of other neurological functions, such as cognition, swallowing, and speech, are present in various combinations and impact each other [@pone.0109679-BenZacharia1]--[@pone.0109679-Solari1]. Recent studies indicate that life expectancy of people with MS in general is reduced by about a decade [@pone.0109679-Runia1]. However highly disabled patients may live many years [@pone.0109679-Campbell1], enduring co-morbidities and complications such as aspiration pneumonia, urinary tract infections, complications of falls and fractures, and sepsis secondary to pressure ulcers, all of which are major causes of death [@pone.0109679-Higginson1]. MS also impinges on the physical and psychological well-being of patients\' significant others [@pone.0109679-Giordano1]. The long disease trajectory allows time for family members to prepare and adjust to their carer roles, which differ from those in cancer and other disabling neurological conditions such as stroke or amyotrophic lateral sclerosis. However, as time passes, carer ageing and co-morbidities add to the complexity and burden of the disease. Although robust evidence supporting treatment decisions in advanced MS is lacking, recent guidelines suggest shifting to a palliative approach as the disease progresses [@pone.0109679-National1]. Palliative care has been defined as: "The active total care of patients whose disease is not responsive to curative treatment. Management of pain and other symptoms, and of psychological, social and spiritual problems is paramount. The goal of palliative care is the achievement of the best quality of life for patients and their families." (<http://www.who.int/cancer/palliative/definition/en/accessed> 1 April 2014). Consequent European government initiatives, including Italian legislation of 2010 (<http://www.normativasanitaria.it/jsp/dettaglio.jsp?id=32922> accessed 1 April 2014) aim to develop and improve palliative services for non-cancer patients including those with neurodegenerative diseases [@pone.0109679-Kristjanson1]--[@pone.0109679-Galushko1]. In this context, we have developed a new home-based intervention for people with severe MS and their carers called the Palliative Network for Severely Affected Adults with MS in Italy (PeNSAMI). We were inspired by two models: the phased approach for the development and evaluation of complex interventions proposed by the Medical Research Council [@pone.0109679-Craig1]; and the approach envisaging interaction between neurology, rehabilitation and palliative care services as an effective way of managing patients with neurodegenerative disorders [@pone.0109679-TurnerStokes1]. In developing PeNSAMI we specifically aimed to involve patients, caregivers and health professionals (HPs) in the development of the intervention; and take specific cultural, socioeconomic, and healthcare contexts across Italy into account [@pone.0109679-Stjernwrd1]. From our systematic review of the literature (see below) we identified four publications [@pone.0109679-Golla1], [@pone.0109679-Galushko1], [@pone.0109679-Edmonds2], [@pone.0109679-Edmonds3] and one unpublished study [@pone.0109679-VeroneseSOliver1] which employed qualitative methodologies to assess MS patient needs, and were conducted in the UK, Germany and Italy over the last seven years. The UK [@pone.0109679-Edmonds2], [@pone.0109679-Edmonds3] and Italian [@pone.0109679-VeroneseSOliver1] studies assisted the construction of a home palliative care service which was compared with standard care in phase II randomized controlled trials (RCTs). The UK RCT showed that a home palliative care service improved symptoms management, reduced caregiver burden and also reduced the use of primary and acute hospital services over the short-term [@pone.0109679-Higginson2]. The Italian RCT (Ne-Pal) included people with severe MS, Parkinson\'s disease and related disorders, or amyotrophic lateral sclerosis. They found that a home palliative care intervention produced a clinically and statistically significant improvement in patients\' health-related quality of life and in four physical symptoms (pain, breathlessness, sleep, and bowel symptoms) [@pone.0109679-VeroneseSOliver1]. Nevertheless it may not be straightforward to transfer interventions of this sort to different contexts and health systems [@pone.0109679-Stjernwrd1], [@pone.0109679-Voltz1]. The UK RCT [@pone.0109679-Higginson2] was included in a recently published systematic review on the effectiveness and cost-effectiveness of home palliative care services for adults with advanced illness and their caregivers. The review found that home palliative care increased the chance of dying at home and reduced symptom burden especially for patients with cancer, without impacting on caregiver grief. However data on neurodegenerative conditions were few and heterogeneous [@pone.0109679-Gomes1]. Rehabilitation plays a major role in providing long-term MS care and support, often over many years, especially in the progressive phase of the disease: a systematic review on the effectiveness of multidisciplinary rehabilitation for adults with MS revealed improved activity and participation [@pone.0109679-Khan1]. In caring for people with long-term neurological conditions such as MS, intensity of involvement and interaction between neurology, rehabilitation and palliative care depends on disease phase: as the patient\'s condition becomes more advanced, rehabilitation and palliative care approaches often overlap (neuropalliative rehabilitation) [@pone.0109679-TurnerStokes1], and it is expected that exchange of competences and co-ordination between these specialities will improve care quality. The present study reports the unmet needs perceived by severely affected MS adults, their carers, and HPs experienced in caring for them, as determined by the qualitative research method. The study is part of the PeNSAMI project which included a systematic review of the literature and the construction of a home-based palliative program to be tested in a multicenter setting spanning the three main geographic areas of Italy. Methods {#s2} ======= Data collection {#s2a} --------------- We involved three key players: people with severe MS, their carers, and HPs experienced in the care of people with severe MS. Our objective was to explore their views, experiences and needs, and hence theorize the intervention areas. For people with severe MS a home-based personal semi-structured interview methodology was felt to be most appropriate. For carers and HPs focus groups (FG) were held to promote interactions and exchange of ideas within a flexible structure [@pone.0109679-Denzin1]. Three FGs (one in each area of Italy) with carers of adults with severe MS, and two FGs with HPs (in Genoa and Rome) were conducted. Ethics statement {#s2b} ---------------- All study patients gave written or oral witnessed consent to participate in a personal audio-recorded interview. All carers and HPs gave written consent to participate in audio-recorded FGs. The protocol and consent procedures were approved by the Ethics Committee of: Foundation IRCCS Neurological Institute C. Besta, Milan; AISM Liguria Region Rehabilitation Service, Genoa; Foundation IRCCS S. Lucia Rehabilitation Hospital, Rome; University "G. d\'Annunzio" of Chieti-Pescara, Chieti; and University Hospital Policlinico Vittorio Emanuele, Catania (all in Italy). Procedures {#s2c} ---------- Five psychologists (all women, median age 35 \[29--47\], one at each participating center) conducted the personal semi-structured interviews in the homes of MS patients. Before the interview, data on patient\'s personal circumstances and general and clinical condition were available to interviewers via the study case report form compiled by the patient\'s caring neurologist. Patients had no relationship with the interviewer prior to the study. The psychologists were trained (one-day) to conduct the interview by two researchers (CB and EB) experienced in qualitative research in the fields of oncology and palliative care. A prototype interview guide was devised beforehand, based on the literature on severe MS [@pone.0109679-Campbell1], [@pone.0109679-Higginson1], [@pone.0109679-Golla1], [@pone.0109679-Galushko1], [@pone.0109679-Edmonds2]--[@pone.0109679-VeroneseSOliver1], [@pone.0109679-Voltz1] and palliative care [@pone.0109679-Barazzetti1]. The guide was modified during the study based on the results emerging from the qualitative analysis. Each interview was audio-recorded and transcribed. Interview duration depended on the content of the patient\'s contribution and his/her willingness/ability to continue. Only exceptionally was a carer present (when the patient requested or appeared ill-at-ease in carer absence). Each FG was planned to include 6--10 participants, plus 2 moderators. One moderator (EB, facilitator) had no previous contact with FG participants; her task was to engage participants, promote exchanges, modulate conflicts, and ensure that all the topics were adequately covered, while allowing time for exploration of any pertinent issues arising. The co-moderator (AS or EP) took notes, noted relevant non-verbal communication, assisted with logistics, and oversaw the audio recording. The FG guides were devised beforehand and modified as required. FG reports were sent to all participants for approval. Sampling and recruitment {#s2d} ------------------------ Participants were selected using a purposeful sampling technique [@pone.0109679-Cote1]. Those recruited for personal interviews were adults with definite MS [@pone.0109679-Polman1], primary or secondary progressive form, Expanded Disability Status Scale \[EDSS\] [@pone.0109679-Kurtzke1] ≥8.0, who had a carer. Institutionalized patients, and those with severe cognitive compromise or impairment precluding communication were excluded. The intention was to recruit patients who varied in terms of age, gender, and intensity of care. A minimum of 10 interviews was planned; sampling ended when data saturation was achieved and no new themes emerged [@pone.0109679-Denzin1]. Participants in carer FGs were recruited from relatives, friends, next of kin, and "key decision makers" usually designated by the patient. Each FG had to include at least three carers of MS patients with severe cognitive compromise or inability to communicate. Participants in HP FGs were recruited from physicians, psychologists, nurses, social workers, and therapists, all involved in caring for patients with severe MS. Analysis {#s2e} -------- Grounded theory (constructivist approach) [@pone.0109679-Charmaz1], guided the analysis of the interview and FG transcripts. We aimed to identify patients\' unmet needs, and theorize broad areas of intervention to meet those needs. Two researchers (CB and EB) independently codified the raw transcripts. After joint discussion of interpretations, the researchers developed a coding framework. The constant comparative technique was used to identify emerging themes (open coding) [@pone.0109679-Glaser1]. Second-level (axial) coding was undertaken to group similar phenomena into categories [@pone.0109679-Strauss1]. Extensive notes were taken to trace researchers\' thinking and guide conceptualization [@pone.0109679-Miles1]. After the principal categories had been established, third level (selective) coding was undertaken [@pone.0109679-Charmaz1], [@pone.0109679-Bluff1] to produce domains (at the highest level of abstraction). The characteristics of patients to be recruited for subsequent interviews were decided based on the coding framework reported above; interviews were stopped when data saturation occurred (no new codes emerged). Results {#s3} ======= Participant characteristics {#s3a} --------------------------- Between October 2012 and May 2013, 22 interviews were conducted at the five centers. [Table 1](#pone-0109679-t001){ref-type="table"} summarizes the characteristics of participants. Participant identification codes and background information are shown in [Table S1](#pone.0109679.s001){ref-type="supplementary-material"}. The personal interviews lasted on average 51 minutes (range 23 --102), and in two (9%) a paid carer also participated. No patient refused to participate. 10.1371/journal.pone.0109679.t001 ###### Characteristics of the 22 people with severe multiple sclerosis who participated in the personal semi-structured interviews. {#pone-0109679-t001-1} Characteristic Sub-characteristic N (%) ----------------------------------------------------------------- ----------------------------------------- -------------------- Women 14 (64%) Age (years)[1](#nt102){ref-type="table-fn"} 58.7, 9.3 (41--77) Disease course Primary progressive 6 (27%) Secondary progressive 16 (73%) Time since MS diagnosis (years)[1](#nt102){ref-type="table-fn"} 22, 7.2 (14--39) EDSS score[2](#nt103){ref-type="table-fn"} 9.0 (8.0--9.5) PEG 2 (9%) Barthel Index[2](#nt103){ref-type="table-fn"} 5 (0--55) Marital status Single 6 (27%) Married 13 (59%) Divorced 2 (9%) Widow 1 (5%) Occupation Employed 2 (9%) Home employment 1 (5%) Not working 19 (86%) Living with Family 9 (41%) Spouse 6 (27%) Mother (81, 85, 87 years) 3 (14%) Alone 2 (9%) Paid caregiver 2 (9%) Followed by MS center 11 (50%) Rehabilitation hospital 5 (23%) Rehabilitation service, community based 4 (18%) Neurologist 1 (5%) Family doctor 1 (5%) MS is multiple sclerosis. EDSS is Expanded Disability Status Scale. PEG is percutaneous endoscopic gastrostomy. Mean, SD (range). Median (range). The three carer FGs involved 30 participants ([Table 2](#pone-0109679-t002){ref-type="table"}), lasted on average 126 minutes (range 120--140) and were conducted in Genoa, Rome and Catania. Carers were varied in terms of age and relation to the patient. Most were full-time caregivers, and six cared for MS patients with severe cognitive compromise. For many carers, participation in the FG was itself a challenge since a substitute carer was difficult to find, particularly for those from Northern and Central Italy; whereas finding a substitute (generally a relative) in the South was considerably easier. FGs were highly informative and emotionally rich, revealing many of the facets and challenges associated with the highly demanding caregiver role. 10.1371/journal.pone.0109679.t002 ###### Characteristics of the 30 carers who participated in the focus group meetings. {#pone-0109679-t002-2} Characteristic Sub-characteristic N (%) --------------------------------------------------- -------------------- --------------------- Women 16 (53%) Age (years)[1](#nt104){ref-type="table-fn"} 59.2, 15.1 (24--91) Relation to patient Spouse 20 (66%) Parent 5 (17%) Offspring 3 (10%) Other relative 2 (7%) Carer of patient with severe cognitive compromise 6 (20%) Occupation Home employment 7 (23%) Employed 10 (33%) Retired 13 (44%) Caregiving for \>10 years 25 (83%) 5--10 years 3 (10%) \<5 years 2 (7%) Time spent caregiving Full-time 18 (60%) Part of the day 11 (37%) Part of the week 1 (3%) Mean, SD (range). The two HP FGs, one in Genoa and one in Rome, involved 18 participants ([Table 3](#pone-0109679-t003){ref-type="table"}) and lasted 130 and 140 minutes. All the main professions involved in the care of people with severe MS were represented, although only one family doctor participated, in spite of efforts to find more. Nurses were the most common professionals (n = 4), followed by psychologists and physiotherapists (n = 3). Most participants had over 10 years of experience in MS. None come from the area of palliative care. All HPs contributed actively and were enthusiastic about the research aims. 10.1371/journal.pone.0109679.t003 ###### Characteristics of the 18 health professionals who participated in the focus group meetings. {#pone-0109679-t003-3} Characteristic Sub-characteristic N (%) ---------------------------------------------------- ------------------------ --------------------- Women 11 (62%) Age (years)[1](#nt106){ref-type="table-fn"} 43.5, 10.2 (26--59) Profession Nurse 4 (22%) Physiotherapist 3 (16%) Psychologist 3 (16%) Physiatrist 2 (12%) Social worker 2 (12%) Neurologist 1 (5%) Occupational therapist 1 (5%) Speech therapist 1 (5%) Family doctor 1 (5%) MS experience (years) \>10 11 (61%) 5--10 4 (22%) \<5 3 (17%) No of MS patients followed in last 3 months \>30 6 (34%) 10--30 9 (50%) \<10 3 (16%) No of severe MS patients followed in last 3 months \>20 1 (5%) 10--20 9 (50%) \<10 8 (45%) MS is multiple sclerosis. Mean, SD (range). Expressed needs {#s3b} --------------- Forty-eight themes grouped into 14 categories and four domains emerged ([Table 4](#pone-0109679-t004){ref-type="table"}). 10.1371/journal.pone.0109679.t004 ###### Codification of needs into themes, categories and domains, as referred by people with severe multiple sclerosis (**¥**), their carers (Δ) and health professionals (▴). {#pone-0109679-t004-4} Domain Category Theme Referred by ------------------------------------ ------------------------------------------ ------------------------------------------------------------------------------------------------- ------------- ***"Managing everyday life"*** **Symptoms management** Symptoms control ¥ Physiotherapy ¥ Δ Devices/aids ¥ **Personal care/hygiene** Management ¥ Δ Reducing uneasiness, embarrassment ¥ Δ▴ Professional for care ¥ Δ **Activities of daily living** Reduce caregiver burden ¥ Δ▴ More and better qualified professionals ¥ Δ▴ Indoor mobility/housing adaptations/communication aids ¥ Δ▴ **Outdoor mobility and transport** Ability to get out ¥ Δ Equipment (elevators, stair-lifts, ramps, etc) ¥ Δ Aids that meet patient requirements ¥ Δ▴ ***"Psychosocial"*** **Relationships/communication** Preservation of family/social relationships ¥ Δ▴ Not being alone at home ¥ Δ Relationships with other MS patients ¥ Δ Affection/empathy ¥ Δ▴ Use of computer and social media ¥ Reduce stigmatization ¥▴ **Leisure/holidays** Cultural activities ¥ Δ▴ Getting out for leisure ¥ Δ▴ Pets ¥ **Psychological well-being/social role** Dealing with fear of the (patient\'s) future ¥ Δ Being accepted ¥ Δ▴ Preserve "normality" ¥ Δ▴ Preserve respect ¥ Δ▴ Preserve biographical continuity ¥ Δ▴ Preserve family/social role and decisional autonomy ¥ Δ▴ Being useful (to others) ¥ Employment (of patients and caregivers) ¥ Δ▴ Preserve hope with research and MS cure ¥Δ▴ Psychological therapy ¥ Δ▴ ***"Organization"*** **Information** Entitlement to services and facilities ¥ Δ Competent persons/service ¥ ▴ **Access to services** Reduce bureaucracy ¥ Δ▴ Help with affronting bureaucracy ¥ Δ▴ Timely and sufficient delivery of aids/consumables ¥ Δ Services suited to people with severe MS ¥ Δ▴ **Co-ordination of services** Person to refer to (case manager) Δ▴ Networking ¥ Δ▴ **Competent professionals** Competent professionals ¥ Δ▴ ***"Health and social policies"*** **Rights** Criteria for benefit entitlement ¥ Δ Accessibility (reduction of architectural barriers) ¥ Δ▴ Equity (across Italian areas) ¥ Δ Resource allocation, investment ¥ Δ▴ **Culture** Reduce stigmatization/discrimination ¥ Δ Respect for disabled rights/facilities ¥ Δ **Patient organizations** Influence public health policies (lobbying) Δ▴ Promote initiatives (educational/cultural/leisure) and facilitate patient participation in them ¥ Δ *Managing everyday life --* This domain comprised four categories, all of which were important for the daily life of the patient-caregiver dyad. Physiotherapy was an important theme: *Movement, I move little \[...\], but a bit more physiotherapy during the week would be nice, yes that would be helpful* \[Patient GIAD Genoa\]. *Improving their quality of life would help them avoid degeneration... patients are sometimes abandoned as there is no cure for disease progression... so we must provide more and better physical therapy to maintain muscle tone* \[Carer, FG Rome\]. Most of the symptoms reported by patients revolved around bowel/bladder functions ([Figure 1](#pone-0109679-g001){ref-type="fig"}). Consequently, personal care and hygiene were of great concern to the dyad: *What we do most is cleaning up and things. These are what tire me and my caregiver the most. She has to turn me over... she says I\'ve become heavier...* \[Patient ALBU Genoa\]. ![Histogram of symptoms (n = 145) reported by the 22 people with severe multiple sclerosis during the personal interviews.\ Blue bars identify symptoms of the Palliative Outcome Scale-Symptoms-Multiple Sclerosis [@pone.0109679-Sleeman1]. The "other symptoms" category includes: shortness of breath (n = 2), peripheral edema (n = 2), thermoregulation problems (n = 2), and loss of smell (n = 1). Stars identify symptoms (n = 48, 33%) pertaining to the perineal area.](pone.0109679.g001){#pone-0109679-g001} *I can\'t bring myself to 'go' in the diaper. I don\'t like being wet. I can\'t even go in the bedpan. I prefer the bathroom, but then they have to lift me. And I\'m heavy, I know. But I try to help using my arms -- more than that I can\'t manage*... \[Patient MASA Milan\]. *As regards my personal hygiene I am completely dependent... I have diapers and have to be changed... I can\'t do anything for myself. The diapers make me feel bad... I wish there was something else I could wear... like catheter or something invisible...* \[Patient ASCO Chieti\]. *I shower my daughter while she\'s in bed, there is no other way to do it*... \[Carer, FG Genoa\]. *I realize that I\'m more trouble when I have a shower or a wash, or they have to change me. I always need help... That really kills me. I\'ve stopped going out because I\'m afraid of doing it in my pants... I\'m always scared of sudden incontinence and creating a stink* \[Patient GRRA Catania\]. Managing these activities had strong implications, particularly for couples and family carers: *No, I can\'t wash myself, I\'ve had to swallow my shame... I even had to get my son to do my bidet... That\'s hard...* \[Patient TEDR Chieti\]. *When I was working I would have to absent myself to help my wife go for a pee... I had to live with that... she didn\'t want anyone else to help her*... \[Carer, FG Catania\]. Home adaptations and mobility aids to enable leaving the house were a major concern; reasons for getting out were of secondary importance: *I need someone \[to help\]... at least then I would be a bit more independent... because leaving the house only 3 times a month isn\'t enough... I\'m in jail... even prisoners are let out every day... Me no... and you see \[how small\] the house is...* \[Patient ALSA Rome\]. *What I really miss is getting out... no-one has time... It\'s better if I don\'t say how long it\'s been since I\'ve been out... months. It\'s really really difficult* \[Patient ASCO Chieti\]. *My wife never leaves the house... she sleeps until the caregiver comes, mostly she watches television, eats and sleeps* \[Carer, FG Genoa\]. *Patients often ask us to help them get out, but they can\'t to tell us where they want to go... they just want to get out* \[HP, FG Rome\]. All players spoke of the need to alleviate the burden of the caregiver: *I need someone who can give my son \[the caregiver\] a break... It seems that I\'m a burden, they tell me I\'m not but* *...* \[Patient GRRA Catania\]. *It would help if caregiver had a bit of breathing space... time to get out, but \[caregiver\] doesn\'t have this and feels abandoned* \[Carer, FG Rome\]. *I think it\'s important for us to be able to get away for 5 minutes, although none of us readily admits it* \[Carer, FG Catania\]. *Psychosocial --* This was the largest domain, with 19 themes grouped into three categories: relationships/communication, leisure/holidays, and psychological wellbeing/social role. All players recognized the need to preserve relationships, and preserve the personal and social roles of the dyad ([Table 4](#pone-0109679-t004){ref-type="table"}). Fear for the patient\'s future was a prominent theme, particularly for parent carers: *My mum is very old so... there is no future \[for me\]. I\'m concerned that she\'s old, she\'s 85* \[Patient ALBU Genoa\]. *On Thursdays and Sundays it\'s my mother who has to look after me, even though she needs someone to help her... she\'s quite badly off... she\'s 87* \[Patient MACI Rome\]. *You worry about... \[the patient\]... thinking about when you\'re no longer here... I\'ve made sacrifices to make sure my husband is OK if anything happens to me* *...* \[Carer, FG Genoa\]. *I\'m fearful about when my husband and I are no longer here... we\'re both old... our relatives are all gone... what will happen to my son then?* \[Carer, FG Genoa\]. Social isolation and loneliness, as a consequence of disease worsening, had a negative effect on patient quality of life, particularly when carers were absent or dysfunctional: *I only want friendship, a chat. They don\'t have to stay for hours... 10 minutes would be enough... I just want someone to make me talk, make me laugh!* \[Patient MASA Milan\]. *Organization --* This domain included nine themes grouped into four categories: information, access to services, coordination of services, and competent professionals. Organizational deficits and (obstructive) bureaucracy were experienced by all. Health and social services were described as scarce and difficult to access/obtain. The rehabilitation facilities that were available did not adequately meet the needs of people with severe MS. Patient aids and assistive devices emerged as a crucial issue with several facets: criteria and procedures for obtaining them; suitability for a given patient; difficulties patients have in accepting or familiarizing themselves with them. As was the case with health/social policies domain (see below), differences between different areas of Italy were enormous. Obtaining information on patient rights and knowing what patients and their families were entitled to was another major problem for caregivers, whose impression was that practically no policies for the social protection for carers were in force or being implemented. *By law we should have been reimbursed for the stair-lift we installed, but we never got anything... The money has always run out*... \[Patient MAGH Chieti\]. *We had to pay for the electric bed ourselves, even though they should have paid for it. I was cheated out of € 2100. I presented all the invoices but was never reimbursed, I don\'t know what to do about it* \[Patient PASC Catania\]. *We\'d like to be informed about new regulations as they come out, and what we can apply for... we have to buy everything ourselves. They say that according to the law, we should get help, but it\'s not easy to get it* \[Carer, FG Catania\]. *I\'d like more information about what \[devices and aids\] are available; we\'ve only just found out about the existence of a baclofen pump*... \[Carer, FG Genoa\]. *What we are entitled to should be made public... available to everyone... not kept hidden*... \[Carer, FG Genoa\]. *You have to wait for ages on the phone, and then they can\'t give you an appointment for months or years... I used to get a check-up every six months, and have a blood test more easily. \[Now there\'s\] too much bureaucracy!* \[Patient MAGH Chieti\]. *The wheelchair is no longer comfortable... it\'s too tight. I have to keep asking them \[my offspring\] to make me more comfortable. At the clinic they say I\'ll have to wait 6 years for a new wheelchair* \[Patient GRRA Catania\]. *Health and social policies -* This domain consisted of nine themes grouped into three categories -- rights, culture and patient organizations. All players referred to the need to reduce environmental barriers: *The sidewalks have been taken over by cars, you can\'t walk the sidewalk \[you have to go round\]. You go up, you go down, there are holes... you have to walk in road...* \[Carer, FG Rome\] *We should set up strong \[patients\'\] association, that looks after these questions on behalf of all of us with these problems* \[Carer, FG Rome\]. *We need a standard level of services in all districts; \[services\] shouldn\'t change depending on where you happen to live* \[Carer, FG Rome\]. The following citations refer to cultural issues: *We had to take \[my condo\] to court before we could install a stair-lift. They said it brought bad luck and looked bad. After we\'d finally installed it, it got vandalized... What satisfaction when some residents asked us if they could use the stair-lift!* \[Carer, FG Catania\]. *When they come with a car they squeeze you in... you really feel you\'re disabled. If they come with a van, everyone knows \[it\'s a disabled person\]. It shouldn\'t be like that* \[Patient ROPE Chieti\] Death and sexuality/intimacy {#s3c} ---------------------------- *Death -* Patients never spontaneously broached death or end-of-life issues, which did not emerge as a primary need. The theme of death always had to be raised by interviewers, in accord with the interview guide. *\[Do you ever think about death?\] Sometimes... \[What do you think about?\] That perhaps it will be a liberation* \[Patient ALSA Rome\]. Many patients expressed a wish not to outlive their carer (see psychosocial domain citations above, e.g. Patient ALBU Genoa). Suicide and euthanasia were not mentioned by patients, who never presented openings for these issues to be explored by interviewers. It also emerged that patients had given little thought to advance care directives or end-of-life decisions. *Sexuality and intimacy* - Some patients (mostly women) viewed loss of sexuality as a consequence of disease worsening: *You\'re invisible when you\'re disabled, you\'re not seen as normal... and \[a sex life\] is one of the many things you miss. But you\'ve got to accept it. I suffered \[when I lost my sexuality\] just as I suffered when I lost my ability to walk. But I adapted, concentrating on other things. In that way I overcame it.* \[Patient FRLI Rome\]. Some patients reported the need to adjust to functional limitations (including sexual functioning, incontinence, mobility compromise, and spasticity): *Some couples ask for help on how to manage intercourse, and these are couples that have maintained a form of intimacy* \[HP, FG Genoa\]. *Sometimes paid caregivers meet the sexual needs of male patients, or of the partners of women with MS* \[HP, FG Rome\]. Sexuality was strongly influenced by relational and patient-carer dynamics: *I haven\'t had sex for almost three years*. *I went to the doctor but what I need is a woman who wants to have sex... \[My wife\] has always been rather reluctant...* \[Patient GIAD Genoa\]. *My sex life* was so *short. When my MS became* serious *and I was no longer able to have intercourse, I cancelled men... denied everything... become completely asexual. Then thanks to a person I recovered some desire... and the desire has become quite strong... in short, now I\'ve re-found myself* \[Patient MACI Rome\]. For partner carers, caregiving often extinguished intimacy as a result of the infantilization of the patient (see personal care and hygiene category above): *For young patients the disease stops them becoming adults, they remain at the stage of a child who is governed by the parents. But it can be like this even for adults*, *who once had independent and autonomous lives, but as the disease worsens, start being cared for by their elderly parents, and this brings the patient back to being a total child, who must silence \[his/her\]* *sexuality* \[HP, FG Genoa\]. The fact that the issue of sexuality never emerged during carer FGs may be due to the presence of other, more fundamental needs. However, addressing a more homogeneous population (e.g. partner carers) or using a different setting (e.g. individual interviews with carers) might have been more appropriate for allowing such an intimate theme to emerge. In the HP FGs, sexuality emerged as difficult theme to approach with their patients: *Few patients speak about their sexual problems, so I do ask about them during consultations* \[HP, FG Rome\]. *Sex therapy is a taboo, it remains submerged. In critically ill patients it is not relevant, although it is in young patients without cognitive impairment* \[HP, FG Rome\]. Carer and HP needs {#s3d} ------------------ During the FGs, important needs specific to carers and HPs emerged. Carers complained strongly about lack of time to themselves, and economic consequences for a family with a severely disabled person: *I can\'t get my wife out of the house. I get her into the lift, but there are six steps to negotiate so she changes her mind and doesn\'t want to leave. These are the problems. Sometimes I pay people to keep her company. Our offspring do not live close. I pay a lady to come. I pay a nurse to wash her in the morning. That\'s what I do. The \[health services\] don\'t send anyone, but by paying you get help. I worked to enjoy my retirement with my wife, but now that money is being spent \[caring for her\]. What else can I do?* \[Carer, FG Genoa\] *Even those who have the money to pay, in the end they run out of savings because there are so many expenses* \[Carer, FG Genoa\]. Social relations and leisure time deprivation: *My old life has vanished. It\'s just me, my husband, and his disease. That\'s all. The disease stands between me and my husband. I used to work, now my life is dedicated to him. This is all very depressing and I don\'t know how long I can go on* \[Carer, FG Genoa\]. *I\'m there day and night, but I think it\'s important to escape every now and then* \[Carer, FG Catania\]. Lack of time to attend medical appointments/procedures: *I need an operation on my shoulder, but who would take care of my husband? Then I would need to convalesce for two months... How am I going to solve this? I\'ve phoned everywhere but no one will help because he\'s not self-sufficient* \[Carer, FG Rome\]. Carers were sometimes reluctant to leave their loved ones in the hands of someone else, who they felt may not be able to cope adequately. In addition, because attention was mainly focused on the patient, carers felt the need for their problems to be considered: they were all but forgotten and completely ignored by health policy makers. *It\'s like having a baby, you go out for half an hour to do the shopping or something... but you always have this dread... you\'re never serene... you\'re always worried that something might happen while you\'re not there. Even if you take an hour off once a month, you never feel free* \[Carer, FG Rome\]. *You\'re afraid of getting sick, even of just a catching cold. I\'ve experienced it: there are things that no one else can do, it doesn\'t matter how good the paid caregiver is, it\'s you who\'ve known \[the patient\] since the beginning of the illness, only you know the little things you can do, that need to be done. You can\'t get sick. You can\'t even die, because then what would happen?* \[Carer, FG Rome\]. Many HPs reported difficulty in establishing a relationship with their MS patients: *Certain patients have an attitude that is conflictual or demanding... and this is a challenge for the professional* \[HP, FG Genoa\], or with the patient and his/her family: *When psychological distress disrupts the family, the operator must be even better at figuring out how to get around it... is very complicated* \[HP, FG Genoa\]. Sometimes carers tended to impose themselves between the patient and the HP, further inhibiting the patient: *\[Sometimes\] the caregiver needs more care than patient, because caregivers become another disease \[burden\] for their patients, sometimes they are harmful... they create a strong bond of dependency with the patient* \[HP, FG Genoa\]. Disproportionate expectations was another HP issue: *Patients expect us to work miracles or \[give them\] new treatments. It\'s difficult to make them see that we work for micro-objectives* \[HP, FG Rome\]. HPs strongly voiced the need for support in their demanding work with patients and their families: *We need psychological support for the difficult task we face, especially for home visits when family members \[often\] take advantage of our visit to let off steam* \[HP, FG Genoa\]. Intervention areas {#s3e} ------------------ In the last phase of the analysis, interventions to meet identified needs in the following seven areas were theorized: medical care (of physical symptoms due to MS and comorbidities); rehabilitation/retraining (physiotherapy and other interventions to improve/preserve patient functioning); psychosocial interventions (to improve psychological wellbeing and alleviate disease burden); HP skills (particularly communicational and relational); domestic support (for personal care/hygiene and other home-related issues); administrative (to improve accessibility to and coordination of services); and public health policies (pertaining to this population and addressing its specific needs). There was marked interdependence between these intervention areas, and the area of public health policies embraced all others ([Figure 2](#pone-0109679-g002){ref-type="fig"}). {#pone-0109679-g002} Discussion {#s4} ========== The unmet needs reported by people with severe MS and their carers transcended purely medical issues to embrace psychosocial aspects. All participants voiced a strong need for qualified personnel and care coordination in the day-to-day home care and management of these patients. Personal care (particularly perineal hygiene) was viewed as crucial. Other priorities were mobility, the need for a supportive (social) network, and preservation of the patient\'s role within the family and the community. Coping with disability, rather than preparing for dying, emerged a major concern of patients, their carers, and HPs. As also found by Galushko et al. [@pone.0109679-Galushko1] our patients, particularly the most disabled, had difficulty in articulating and deliberating on their needs, even though we purposely excluded those with severe cognitive compromise. By contrast, carer FGs were rich in content and emotion, an indication that carers took seriously their responsibility as sole interpreter and provider of their loved ones\' needs. However, these responsibilities had a profound impact on carer lives, particularly for parent carers and those in small or less affluent families. Feelings common to carers and patients were regret for the disruption of the original patient-carer relation, sense of isolation, and fear for the future. A strength of our study is that a variety of stakeholders from the three geographic areas of Italy participated. Thus our results reflect a spectrum of perspectives and the social and cultural diversities that exist in Italy. Availability of information about services, access to services, and appropriateness of services, varied with geography and were conspicuously worse in the South than the North. In general, families were larger in Central and Southern Italy so caregiving could be shared; in the North, by contrast, even participation in the carer FG often proved difficult. In the South however, cultural barriers and stigmatization could be as strong as physical and architectural obstacles (see for example the last two citations in the *health and social policies* paragraph of the Results). Study limitations include the fact that issues of sexuality and intimacy were rarely raised by patients and never by their carers, although they were broached by HPs. This could be because carer FGs -- whose participants were of variable age and relation to the patient -- were unsuited for the discussion for such personal issues. Furthermore, although we invited family doctors to our HP FGs, only one attended. Similarly, although we wanted at least three carers of patients with severe cognitive compromise in each carer FG, we managed to recruit only two (however no additional themes emerged from these participants). Another limitation is that complementary and alternative medications were not part of the interview guide. Many of the needs we identified were also reported by the qualitative studies of Edmonds et al. [@pone.0109679-Edmonds2]; [@pone.0109679-Edmonds3] and Galushko et al. [@pone.0109679-Galushko1], particularly the need to improve the continuity and coordination of care. Our study and the UK [@pone.0109679-Edmonds2]; [@pone.0109679-Edmonds3] study recruited MS patients of similar severity, while the German study [@pone.0109679-Galushko1] had few severely disabled patients. Our study differs from the others [@pone.0109679-Galushko1], [@pone.0109679-Edmonds2], [@pone.0109679-Edmonds3] in that it was multicentric, and did not originate from the palliative care field, but the MS field (Italian MS Society and MS clinicians and researchers, including those with expertise in MS rehabilitation). Notably, interviewers were trained by researchers who had oncology and palliative care backgrounds. Nonetheless, the marked interdependence of the seven intervention areas we theorized is in line with the multifaceted approach of palliative care. Our results were used to guide the construction of our home-based palliative program, which addresses each of the intervention areas identified in our analysis except the all-important area of public health policy. Our next step will be to assess the effectiveness of the home-based palliative program in a randomized controlled trial (ISRCTN73082124) and a qualitative study nested in the trial. Supporting Information {#s5} ====================== ###### **Background characteristics of the 22 interviewed people with multiple sclerosis.** (DOC) ###### Click here for additional data file. We are indebted to the patients, carers and health professionals who participated in the study. We thank Drs. Maren Galushko and Heidrun Golla, from the Department of Palliative Medicine, University Hospital, Cologne, Germany, for their valuable suggestions and comments on the interview guide. We also thank Don Ward for help with the English. **PeNSAMI project investigators.** **Steering Committee:** R Amadeo, A Giordano, M Ponzio, MG Grasso, A Lugaresi, F Patti, G Martino, L Palmisano, S Veronese, P Zaratin, MA Battaglia, A Solari. **Data Management and Analysis Committee:** A Giordano, M Ponzio (statistician), G Ferrari, A Solari. **Independent Data and Safety Monitoring Committee:** DJ Oliver: *Wisdom Hospice, University of Kent, Rochester, Kent, UK;* E Pucci: *Neurology Unit, Ospedale Provinciale di Macerata, Macerata;* L Tesio: *Department of Biomedical Sciences for Health, University of Milan, Milan.* **Qualitative Analysis Panel:** E Bianchi, E Pietrolongo, A Solari, A Giordano, I Rossi, S Cilia, M Giuntoli, C Borreani. **Literature Review Panel:** MG Grasso, L Palmisano, A Fittipaldo, A Giordano. **Intervention Panel:** C Cugno, R Causarano, P Morino: *'Ex Convento delle Oblate' Hospice, Local Health Unit of Florence, Florence*, S Veronese. **Centers and Investigators:** *AISM Liguria Region Rehabilitation Service, Genoa*: ML Lopes de Carvalho, M Giuntoli, R Motta; MA Battaglia; *Antea Association Onlus (charity), Rome*: G Casale, MC Stefanelli; *F.A.R.O. Foundation Onlus (charity), Turin*: S Veronese, C Cugno; *Foundation IRCCS Istituto Nazionale per la Cura dei Tumori, Milan*: C Borreani, E Bianchi; *Foundation IRCCS Neurological Institute C Besta, Milan*: A Solari, P Confalonieri, A Giovannetti, V Torri Clerici, E Rossetti, A Totis, A Campanella, A Giordano, F Martini, A Fittipaldo, G Ferrari, R Mantegazza; *Foundation IRCCS S Lucia Rehabilitation Hospital, Rome*: MG Grasso, I Rossi, A Clemenzi, E Troisi, A Pompa, G Morone, S Passarelli, A Fusco; *Istituto Superiore di Sanità, Rome*: L Palmisano; *Italian MS Foundation (FISM) and Association (AISM), Genoa*: R Amadeo, P Zaratin, M Ponzio, G Martino, MA Battaglia; *Niguarda Ca\' Granda Hospital, Milan*: R Causarano, D Da Col, B Lissoni; *University G d\'Annunzio, Chieti-Pescara, Chieti:* A Lugaresi, E Pietrolongo, M Onofrj; *University Hospital Policlinico Vittorio Emanuele, Catania:* F Patti, S Cilia, C Leone; V Cascio, V Cimino, G Occhipinti, A Pappalardo, C Cavallaro, F Zagari. [^1]: **Competing Interests:**I have read the journal\'s policy and the authors of this manuscript have the following competing interests: EP has received travel grants from Bayer Schering, Biogen Idec, Merck Serono, Novartis, Sanofi Aventis and Teva and also received travel and research grants from the Fondazione Italiana Sclerosi Multipla. PC has been a board member of Biogen-idec, received travel grants from Sanofi Aventis, Biogen Dompé AG and Merk Serono. AL has been a Bayer Schering, Biogen Idec, Merck Serono and Genzyme advisory board member. She received travel grants and honoraria from Bayer Schering, Biogen Idec, Merck Serono, Novartis, Sanofi Aventis and Teva and research grants from Bayer Schering, Biogen Idec, Merck Serono, Novartis, Sanofi Aventis and Teva. She has also received travel and research grants from the Associazione Italiana Sclerosi Multipla and was a consultant of "Fondazione Cesare Serono". FP received honoraria for speaking activities from Bayer Schering, Biogen Idec, Merck Serono, Novartis, and Sanofi Aventis. He has served as advisory board member of the following companies: Bayer Schering, Biogen Idec, Merck Serono, and Novartis. MGG has received research funding from Merck Serono and consulting and speaking fees from Biogen Idec. AS has been a board member of Biogen Idec and Novartis, and has received speaker honoraria from Genzyme, Merck Serono and the Fondazione Serono. The authors confirm that this declaration does not alter the authors\' adherence to all PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: CB AG LLC LP PZ MAB AS. Performed the experiments: EB EP IR SC MG PC AL FP MGG LLC. Analyzed the data: CB EB. Wrote the paper: CB EB EP IR SC MG AG PC AL FP MGG LLC LP PZ MAB AS. [^3]: ¶ Membership of the PeNSAMI Project is provided in the Acknowledgments.

This article has been accepted for publication and undergone full peer review but has not been through the copyediting, typesetting, pagination and proofreading process, which may lead to differences between this version and the Version of Record. Please cite this article as doi:10.1111/anae.15220

Introduction {#sec1-1} ============ Ununited fractures of clavicle are occasionally seen in adults, but are rarely found in children\[[@ref1][@ref2]\]. Post-traumatic non-union of the clavicle is a rare complication in adulthood with a frequency of about 1%. This condition is also exceptional in children, despite the frequency of clavicular fracture at any given age\[[@ref3][@ref4]\]. Clavicular fractures usually occur at the junction of the medial two third with the lateral one third of the bone and usually heal by conservative treatment within three weeks. Surgery is required in about one in 100 cases in which there is remaining deformity. Case Report {#sec1-2} =========== {#sec2-1} ### Patient one {#sec3-1} An 8-year-old male child presented with a right-sided clavicular fracture for one year. There was a history of fall while playing and the patient sustained injury to the right shoulder, which led to the fracture of the clavicle. Both swelling and severe pain were present. The patient had received treatment from an orthopedic surgeon with a figure 8 bandage for three weeks; however, the pain was not relieved. At one-year follow-up, the upper part of the chest, the bony prominence in the clavicular region and the overlying skin appeared normal. On palpation, there was bony protuberance at the site of the injury. There were no signs or symptoms of neurological deficit. All of the arterial pulses were present in the affected limb (i.e., axillary, brachial and radial). Movements of the shoulder were within normal range. On auscultation, no bruit was heard at the site of injury. Diagnosis was made following a radiograph of the right shoulder, which showed old, non-union of the clavicle at the junction of the medial two third with the outer one third. There was rounding of both ends without any callus formation ([Fig. 1](#F1){ref-type="fig"}). The patient was treated conservatively with analgesics and shoulder exercise as the movements of the shoulder were within normal range. At one-year follow-up, the child was doing well. {#F1} ### Patient two {#sec3-2} A 26-year-old male patient presented with a history of fracture of the right clavicle six years ago. He received treatment from an orthopedic surgeon in Patna with figure 8 bandages and a shoulder sling for three weeks. The patient now complained of pain in the right shoulder with limitation of movement for one year. On local examination, there was no swelling or deformity and shoulder joint movements were within normal limits, up to 90%. There had been restriction of the movements of the shoulder joint, up to 10 degrees on internal and external rotation. All pulses were felt normally. A radiograph of the shoulder joint revealed an old, non-united fracture of the mid one third of the clavicle on the right side with no callus formation ([Fig. 2](#F2){ref-type="fig"}). The patient was treated conservatively with analgesics and physiotherapy exercises. At a six-month follow-up visit, the patient was asymptomatic. {#F2} Discussion {#sec1-3} ========== Ununited fractures of the clavicle are rare\[[@ref1][@ref2]\]. The non-union rate has been reported to be between 0.1% and 15%\[[@ref5][@ref6]\]. Clavicular non-union is rarely asymptomatic and often results in disability from pain at the site of non-union, altered shoulder mechanics, or a compression lesion involving the underlying brachial plexus or vascular structures\[[@ref5]\]. Fractures of the clavicle are usually in the medial two third of the bone, which may result from a fall and subsequent outstretched hand during the fall. The lateral fragment is displaced forward and downward by the weight of the limb, while the medial fragment is held at a higher level by the sternocleidomastoid muscle. The essential treatment is to support the weight of the limb by a sling tied over the opposite shoulder. The fractures are almost always clinically united within three weeks. About one in 100 fractures of the clavicle require primary surgical treatment. Rarely, a fragment may be displaced backward and endanger the subclavian vessels. Sir Robert Peel, who established the police force of Great Britain, died of a fractured clavicle which ruptured the subclavian vein. Peel was attended by Sir Benjamin Brodie who wrote, "The hemorrhage itself was the consequence of the subclavian vein having been lacerated by splinters of the fractured bone"\[[@ref3]\]. Cosmetically, it is best to treat clavicular fractures conservatively. If deformity persists at the bony ends of the clavicle after several months, surgical smoothening of these ends is indicated by a short incision in the line of the skin creases. This causes less deformity than the scarring resulting from more extensive operative procedures that may be required for primary open reduction with internal fixation. In addition, major surgical procedures may carry the risk of additional surgery for non-union\[[@ref4]\]. The majority of clavicular fractures can be effectively treated non-surgically\[[@ref7]\]. The non-union rate of fractures of the lateral end of the clavicle can rise to 37% when a nonsurgical treatment protocol is initially adopted. Reported results for the nonsurgical treatment of fractures of the clavicle have been uniformly positive; a combined series of over 3000 fractures showed a rate of non-union of 0.4%. Occult fracture has been well documented in the hip and the scaphoid and failure to recognize this type of fracture could lead to serious consequences. While clavicular fracture is often viewed as benign, it is important for patients to be aware that any fracture may impact expected time of recovery. In addition, complications such as non-union do occur and inadequate initial immobilization is a common cause\[[@ref5][@ref8]\]. Patients who have suffered a clavicular fracture often recover well in spite of the risk of non-union; fatal complications that may occur following vascular injuries are extremely rare. Fracture location and the type of immobilization have little effect on the final result or prognosis. Functional outcome is mainly determined by associated systemic and critical trauma. Conclusion {#sec1-4} ========== Careful attention should be paid when obtaining a detailed history and physical examinations, as traumatic arthritis at either clavicular joint may mimic the symptoms of non-union. The explicable evidence of osseous non-union on radiographs may be minor and may not correlate with the clinical symptoms. A patient with an atrophic pattern of non-union may become asymptomatic with time. Surgeons should be cautious when operating on the non-union merely due to its presence, although asymptomatic. If a surgical procedure is planned, possible outcomes should be communicated to the patient, including the possibility of additional surgery, if required. This study was completed in the Dr. Kundan Lal Hospital, Ahmedgarh-148021, District: Sangrur (Punjab), India.

Introduction {#s1} ============ Microbial biomass in marine sediments accounts for 0.18−3.6% of Earth's total biomass (4.1 petagram carbon), and their community composition is highly diverse due to variation in oxygen concentrations in the overlying water, sediment carbon content, and sediment depth [@pone.0096219-Kallmeyer1], [@pone.0096219-Orcutt1]. Sediment bacteria fulfill an important role in organic matter remineralization [@pone.0096219-Arnosti1], [@pone.0096219-Boetius1] and nutrient cycling [@pone.0096219-Alongi1], and are an integral component of food-webs, particularly those that are detritus-based [@pone.0096219-Cammen1], [@pone.0096219-vanOevelen1]. Sediment bacterial communities are more diverse than planktonic communities, and respond actively to environmental conditions of their habitat [@pone.0096219-Lozupone1]--[@pone.0096219-Nemergut1]. Studies on the role of bacteria in sediment biogeochemistry particularly require a quantitative assessment of both bacterial biomass and community composition. Nevertheless, studies on estimates of bacterial biomass, community composition, and diversity are constrained by the methodological limitation that over 99% of the total bacterial population cannot be cultivated by traditional culture techniques [@pone.0096219-Amann1]--[@pone.0096219-GontangE1]. In the past few decades, the rise of culture-independent techniques (molecular approach and chemical analysis-based method) has allowed us to further reveal sediment microbial ecology. Molecular approaches that are highly suited for high resolution description of bacteria communities in marine sediment are rDNA clone libraries, denaturing gradient gel electrophoresis (DGGE), and terminal restriction fragment length polymorphism (T-RFLP). In recent years, the advent of high-throughput sequencing technologies (e.g. pyrosequencing and Illumina) has greatly enhance the knowledge on bacterial community structure [@pone.0096219-Logares1], [@pone.0096219-Bolhuis1]. As most powerful quantitative molecular approaches, the Q-PCR approach has been widely applied to quantify gene copy number as a proxy of bacterial abundance [@pone.0096219-Inagaki1]--[@pone.0096219-Smith1], and the FISH technique has been used for visualizing and quantifying bacterial cells in sediments [@pone.0096219-LlobetBrossa1], [@pone.0096219-Boetius2]. Both quantitative approaches are distinctly suitable for targeting specific phylogenetic groups but less suitable for analysis of the full bacterial community, because quantitative application for analysis of all bacterial groups requires the use of many target-specific primers and probes and also need to optimize its protocol for each target group. Moreover, the PCR-based approaches cannot eliminate methodological biases, and nucleic acid extraction from sediment samples has inherent biases, for instance extraction efficiency from sample and bacterial species [@pone.0096219-Smith2]. Despite the superiority of molecular approaches for the analysis of bacterial community structure, also phospholipid-derived fatty acid (PLFA) analysis [@pone.0096219-Findlay1], [@pone.0096219-Boschker1] and the quinone profiling method [@pone.0096219-Urakawa1], [@pone.0096219-Kunihiro1] have been successfully used as a chemotaxonomic analytical-based method to quantify bacterial biomass and to profile the bacterial community composition in marine sediments. PLFAs are essential membrane lipids of microbial cells and therefore proxies for bacterial biomass. Microorganisms contain numerous PLFAs with some being "general" and unspecific, while others are more specific and found in higher abundance in some microbial groups [@pone.0096219-Boschker1], [@pone.0096219-Kaur1]. Respiratory quinones (RQ, including ubiquinone \[UQ\] and menaquinone \[MK\]), and photosynthetic quinones (including phylloquinone \[K1\] and plastquinone \[PQ\]) are lipid coenzymes used for electron transfer in microbial cell membranes. A bacterial phylum has generally only one dominant molecular species of respiratory quinone (e.g. [@pone.0096219-Collins1]). The main advantage of both chemotaxonomic methods is that there are established and standardized quantitative extraction protocols available [@pone.0096219-Boschker2], [@pone.0096219-Hu1] that allow rapid and quantitative extraction from various types of sediment samples. Therefore, the lipid analysis is easily applicable to quantify bacterial biomass for a wide range of marine sediments without optimizing the extraction method. In fact, concentrations of bacteria-specific PLFAs have been used as a proxy for total bacterial biomass in various marine sediments (e.g. [@pone.0096219-Middelburg1]), and the total RQ concentration has been found to correlate with bacterial biomass in soil [@pone.0096219-Saitou1], with the total bacterial cell count in various environments [@pone.0096219-Hiraishi1], and with the bacterial cell volume in lake water [@pone.0096219-Takasu1]. Moreover, analysis of PLFA- and quinone profiles has been widely utilized as a valuable tool for showing differences between samples and also community shift during experimental/monitoring periods (e.g. [@pone.0096219-Polymenakou1], [@pone.0096219-Kunihiro2]). Cluster analysis for characterizing bacterial community structure based on dissimilarity- or similarity value matrix of PLFA- and quinone profile for marine sediments showed a similar clustering pattern with that of the molecular techniques [@pone.0096219-Polymenakou1], [@pone.0096219-Kunihiro2]. One major disadvantage of both lipid analyses for studies of microbial ecology is the lower phylogenetic resolution for identifying bacterial groups than molecular approach. Thus, the lipid analysis has been combined with molecular approach as a means of overcoming the limitation of low phylogenetic resolution [@pone.0096219-Urakawa1], [@pone.0096219-Kunihiro2], [@pone.0096219-Polymenakou2]. A major advantage of the PLFA technique is that it can be combined with carbon stable isotope analysis to identify active bacterial groups and to trace carbon flows in both benthic and pelagic food webs via bacteria and other microbial groups, such as microalgae, to higher trophic levels (e.g. [@pone.0096219-Boschker1]). In addition, the quinone profiling method is also possible to identify active bacterial groups by combination with carbon radioisotope labelling [@pone.0096219-Saitou2]. For marine sediments, PLFAs have been widely used as quantitative bacterial biomarkers [@pone.0096219-Middelburg1], but applications of quinones as biomarkers for sedimentary bacteria are very few [@pone.0096219-Urakawa1], [@pone.0096219-Kunihiro1]. In addition to their potential as a proxy for bacterial biomass, the ratio between quinones and PLFAs may also provide a proxy for the level of activity of the bacterial community [@pone.0096219-Hedrick1], because PLFAs are structural biomass components while quinone concentrations are related to biomass and respiratory activity as they are part of electron transport chains (e.g. [@pone.0096219-Nowicka1]). The ratio between total RQ and total PLFA was firstly applied in estuarine and deep sea sediments by Hedrick and White [@pone.0096219-Hedrick1]. Until now, very few studies have applied this potential proxy in other aquatic systems [@pone.0096219-Villanueva1], [@pone.0096219-Peacocka1]. In this study we compared and evaluated PLFAs and quinones as quantitative bacterial biomarkers for bacterial biomass and community structure in marine sediments. We also explored whether the ratio between these two bacterial biomarkers could be used as a potential proxy for bacterial activity, by examining the concentrations of PLFAs and quinones and quantity and quality of organic matter in a wide range of marine sediments from the intertidal zone to the deep sea. Materials and Methods {#s2} ===================== Study Areas and Sampling Procedure {#s2a} ---------------------------------- Samples were collected from a wide variety of sediments, ranging from intertidal to coastal, shelf and deep-sea sediments (*see* [Table 1](#pone-0096219-t001){ref-type="table"} for the sites and sampling depth details and [Table S1](#pone.0096219.s002){ref-type="supplementary-material"}). Samples selected for this study came from previous published and unpublished studies were either PLFA or quinone analysis had already been performed and we completed the data set by additional analyses. No specific permissions for all sampling were required for these locations. Intertidal sediments were collected from 3 locations (Oude Bietenhaven, Zandkreek, and Rattekaai) in the Oosterschelde, a marine embayment in the SW of the Netherlands, and one location (Kapellebank) in the nearby Scheldt Estuary. Sediment was sampled manually at low tide using cores (30 cm height and 6 cm in diameter) and cores were sliced. Another tidal flat location was sampled for long-term incubations in the laboratory. In short, lab incubation sediments were collected from the surface (0∼2 cm) of a tidal flat (Biezelingse Ham) in the Scheldt Estuary, homogenized, and incubated for up to 261 days *in vitro* with regular sampling in a similar manner of the experiment that is described in [@pone.0096219-Veuger1]. North Sea sediments were collected from three stations in November 2010. Stations NS-1 and -3, located close to the Dutch coast and on the Dogger Bank respectively, are non-depositional areas, while station NS-2, situated on the Oyster Ground, is a semi-depositional area. Sediment was sampled with cores by multi-corer (Octopus type). 10.1371/journal.pone.0096219.t001 ###### Sample codes and characteristics. {#pone-0096219-t001-1} Site Code Water depth (m) Sediment depth[\*](#nt101){ref-type="table-fn"} (cm) *n* [\*\*](#nt102){ref-type="table-fn"} Analysis ---------------------------- --------- ----------------- ------------------------------------------------------ ----------------------------------------- ---------- --- **Dutch intertidal (DI):** Oude bietenhaven DI-N-OB \- 0--2 2 n n Zandkreek DI-N-Z \- 0--5 2 n n Rattekaai DI-N-R \- 0--1.5 2 n n Kapellebank DI-N-K \- 0--2 2 n n Lab incubations DI-L \- 0--2 1--12 y n **North Sea (NS):** Station 1 NS-1 12 0--1 1 y n Station 2 NS-2 45 0--9 1--6 y n Station 3 NS-3 27 0--9 1--6 y n **Japanese coast (JC):** Natural JC-N 6--83 0--1 or 0--2 1--9 y y Fish farm JC-FF 30--75 0--2 1--14 y y **Arabian Sea (AS):** Station 1 AS-1 989 0--2 1 y n Station 2 AS-2 1700 0--2 1 y n **Galicia Bank** GB 1900 0--1 1 y n \*Total sampled depth range; \*\**n*, sample number; \*\*\*OC: organic carbon content, \*\*\*\*DI: degradation index. Additional information is shown in [Table S1](#pone.0096219.s002){ref-type="supplementary-material"}. Japanese natural coastal sediments were collected from nine different bays and embayments in the Seto Inland Sea using a Smith-McIntyre Grab sampler or an Ekman-Berge grab sampler, and then subsampled by collecting surface sediment samples from late September to early October 2008 and from early May to early June 2009. Japanese fish farm sediments were collected from 14 stations located in and around fish farming areas in the north part of Sukumo Bay, located in Sikoku, Japan in the same manner as the coastal sediments collected from the Seto Inland Sea. Deep sea sediments were collected from the Arabian Sea and the Atlantic Ocean. Samples from the Arabian Sea were obtained from two stations, with one station (AS-1) being situated within the oxygen minimum zone (OMZ) (i.e. \<9 µM O~2~ in the overlying water) and the other station (AS-2) below the OMZ (i.e. oxic bottom water) in January 2010 [@pone.0096219-Pozzato1]. Sediment from the Atlantic Ocean was sampled at the Galicia Bank off the coast of Spain in September--October 2008 [@pone.0096219-Pozzato2]. All sediment samples were either directly stored frozen (−20°C) or freeze-dried and subsequently stored frozen (−20°C) until extraction and analysis of PLFAs, quinones, and organic carbon. Chemotaxonomic Markers of PLFAs and RQs in Different Groups of Bacteria {#s2b} ----------------------------------------------------------------------- Important chemotaxonomic PLFA and RQ markers for bacteria are listed in [Table 2](#pone-0096219-t002){ref-type="table"}. In this study, we defined the sum of saturated fatty acid (SFAs, C~13~--C~18~), branched fatty acids (BFAs), and mono-unsaturated fatty acids (MUFAs, ≤C~19~) as total bacterial PLFA. In addition, there are various other bacteria-*specific* PLFAs, for instance i17∶1ω7 is for a marker for the genus *Desulfovibrio* [@pone.0096219-Taylor1], but these compounds are typically present in low concentrations, which precluded analysis of these compounds in most sample sets in the present study. 10.1371/journal.pone.0096219.t002 ###### Major fingerprints of PLFA and quinone as a marker for different bacterial groups in this study. {#pone-0096219-t002-2} Biomarker Proteobacteria Bacteroidetes Actinobacteria ---------------------------------------------------- ------------------------------------------------------ ------------------------------------------------ ---------------------------------------------------------------------------------- ----------------------------------------------------- ---------------------------------------------- ---------------------------------------------------------------------------- ---------------------------------------------------------------------------------- **PLFA** [a](#nt106){ref-type="table-fn"} SFA (C~12~--C~19~)[c](#nt108){ref-type="table-fn"} G G G G G G G i14∶0 i15∶0 ++M +++ ++ i16∶0 \+ +++ i17∶0 ++M \+ \+ a15∶0 +M +M \+ a17∶0 +M ++ 10Me16∶0 ++M +M 10Me17∶0 +M 10Me18∶0 ++M cy17∶0 \+[d](#nt109){ref-type="table-fn"} \+ \+ +++M \+ cy19∶0 \+ \+ +M 16∶1ω7c G G G G G G G 18∶1ω9c G G G G G G G 18∶1ω7c +++ \+ +++ ++M +++ **Ref. no.** [@pone.0096219-Martens1]--[@pone.0096219-Romanenko1] [@pone.0096219-Oyaizu1]--[@pone.0096219-Lim1] [@pone.0096219-Vancanneyti1], [@pone.0096219-Jean1] [@pone.0096219-Taylor1], [@pone.0096219-Londry1] [@pone.0096219-Smith3], [@pone.0096219-Kim1] [@pone.0096219-Oyaizu1], [@pone.0096219-OSullivan1], [@pone.0096219-Kaur2] [@pone.0096219-Kroppenstedt2] **Quinone** [b](#nt107){ref-type="table-fn"} UQ-8 ++++\* ++++M ++++M UQ-9 ++++\* ++++M UQ-10 ++++M MK-*n* (*n*≤8) ++++\* ++++\* ++++M ++++M (MK-6) ++++M ++++\* MK-*n* (*n*≥9) ++++\* ++++M MK-*n*(H*~x~*) ++++\* ++++M **Ref. no.** [@pone.0096219-Martens1]--[@pone.0096219-Romanenko1] [@pone.0096219-Lim1], [@pone.0096219-Knittel1] [@pone.0096219-Jean1], [@pone.0096219-AkagawaMatsushita1], [@pone.0096219-Shin1] [@pone.0096219-Collins3], [@pone.0096219-Devereux1] [@pone.0096219-Lancaster1] [@pone.0096219-Oyaizu1], [@pone.0096219-Kaur2], [@pone.0096219-Nakagawa1] [@pone.0096219-Kroppenstedt2], [@pone.0096219-Yamada1], [@pone.0096219-Athalye1] In this study, we refer to the different quinones with the following abbreviations: ubiquinone - UQ-*n*; and menaquinone - MK-*n*. The number (*n*) indicates that of the isoprene unit in the side chain of the quinone. Partially hydrogenated MKs were expressed as MK-*n*(H*~x~*), where *x* indicates the number of hydrogen atoms saturating the side chain. PLFA data were modified mainly from [@pone.0096219-Boschker1], [@pone.0096219-Kaur1], [@pone.0096219-Boschker3], [@pone.0096219-Ratledge1]. Quinone data were modified mainly from [@pone.0096219-Collins2], [@pone.0096219-Yokota1]--[@pone.0096219-Fujie1]. Saturated fatty acids. +, 1--5%; ++, 5--15%; +++, 15--40%; ++++, \>40% of total PLFA pool or total quinone pool; \*, present in few species; G, a maker found in a broad range of bacteria and algae, and M, a marker can be used specifically as an indicator for specific bacterial group with the phylum. PLFA Extraction and Analysis {#s2c} ---------------------------- PLFAs were extracted from freeze-dried sediment (∼4 g) and analyzed as described in [@pone.0096219-Boschker2]. In short, total lipids were extracted from the sample in chloroform--methanol--water (1∶2∶0.8, v/v) using a modified Bligh and Dyer method and fractionated on silicic acid into different polarity classes. The methanol fraction, containing phospholipids, was derivatized using mild alkaline methanolysis to yield fatty acid methyl esters (FAMEs), which were recovered by hexane extraction. FAME concentrations were determined by gas-chromatography-combustion-isotope ratio mass spectrometry (GC-c-IRMS) for all samples except for Japanese samples that were analysed by gas chromatography-flame ionization detection (GC-FID). The concentrations obtained by both methods are comparable from our previous experience (*r^2^* = 0.99, unpubishied data). Identification of individual FAME was based on comparison of retention times with known reference standards. Quinone Extraction and Analysis {#s2d} ------------------------------- Quinones were extracted from freeze-dried or frozen sediment (∼6 g) as described previously [@pone.0096219-Kunihiro1], [@pone.0096219-Hu1]. The types and concentrations of each quinone were determined using a HPLC equipped with an ODS column (Eclipse Plus C18, 3.0 (I.D.) ×150 mm, pore size 3.5 µm, Agilent technologies) and a photodiode array detector (SPD-M20A, Shimadzu: for the Japanese samples, and Waters 996 for the samples from the Dutch intertidal zone, North Sea, and the deep sea). A mixture of 18% isopropyl ether in methanol was used as the mobile phase at a flow rate of 0.5 mL min^−1^. The quinone molecular species were identified by the linear relationship between the logarithm of the retention times of quinones and the number of their isoprene units, using the identification-supporting sheet, which is available upon request from T. K, based on the equivalent number of isoprene units (ENIU) of quinone components as described by [@pone.0096219-Tamaoka1]. Details on the analytical conditions have been described by [@pone.0096219-Takasu1]. Organic Carbon Content {#s2e} ---------------------- For determination of the organic carbon (OC) content, sediment samples were first freeze-dried or dried at 60°C in an oven overnight, acidified to remove carbonate, and further vacuum-dried. The OC content of the sediment was determined with an elemental analyzer (NA-1500n, Fisons, Rodano-Milan, Italy: for the Japanese sediments and FlashEA 1112, Thermo Electron, Bremen, Germany: for the sediments of other samples). HAA Extractions, Analysis and Calculation of Degradation Index {#s2f} -------------------------------------------------------------- For the Japanese sediments, concentrations of hydrolysable amino acids (HAAs) were analyzed as described in [@pone.0096219-Dauwe1] and used to calculate the degradation index (DI), a proxy for the quality, or "freshness", of the organic matter in the sediment. Briefly, samples (∼1 g) of freeze-dried sediment were washed with 2 M HCl and Milli-Q water and then hydrolyzed in 6 M HCl at 110°C for 24 h. After neutralization by 1 M NaOH, amino acids were derivatized with *o*-phthaldialdehyde (OPA) [@pone.0096219-Lindroth1] prior to injection to reverse-phase high-pressure liquid chromatography (HPLC). Amino acid concentrations were measured by HPLC and further details on the analytical conditions have been described by [@pone.0096219-Dauwe1]. The DI was calculated following [@pone.0096219-Dauwe1]:where var*~i~*, AVGvar*~i~*, STDvar*~i~*, and fac.coef*~i~* are the mol%, mean, standard deviation and factor coefficient of amino acid *i*, respectively. The factor coefficient was described in [@pone.0096219-Vandewiele1]. Cluster Analysis of the Pattern of Differences Among Samples in Individual PLFAs and RQs {#s2g} ---------------------------------------------------------------------------------------- We conducted a cluster analysis to identify groups of similar bacterial PLFA and RQ patterns. We first normalized the mole fraction of bacterial PLFA and RQ (Z*~j,i~*), because this analysis depends on the absolute values of the data, using the following normalization equations [@pone.0096219-Kreyszig1]: With: where *P~j,i~* is the mole fraction of bacterial PLFA or RQ component *j* and sample *i*, *N* is the number of samples, and and *S~j~* are the average value and the standard deviation of the mole fraction of bacterial PLFA or RQ among samples, respectively. After normalization, the average value and the standard deviation *S~k~* are shown respectively as 0 and 1, where *k* is the normalized component of bacterial PLFA or RQ. We used both\>1 mol% of component to the bacterial PLFA (without general bacterial compounds (SFAs (C~12~−C~19~), 16∶1ω7c, and 18∶1ω9c), MUFAs (≥C~20~), and PUFAs) or RQ profile, \>30% of coefficient of variance of compound among all samples for this data analysis, and reconstructed profiles, because general and minor components interfere with the result of this analysis. As results, the cluster analysis was conducted based on the mole fraction of 12 bacterial PLFAs and 16 RQ molecular species among all samples (*see* "Cluster analyses of bacterial PLFA and RQ profiles"). The normalized values were used to produce a cluster dendrogram based on the Euclidean distance matrix, and the dendrogram was constructed using Ward's method with the graphing program KyPlot version 5.0 (KyensLab Inc., Tokyo, Japan). Cluster Analysis Based on the Full Profiles of the Bacterial PLFAs and RQ {#s2h} ------------------------------------------------------------------------- We conducted another cluster analysis to compare sample discrimination and its resolution based on bacterial PLFA or RQ profiles. A dissimilarity index (*D*) of profile was calculated using the following equation [@pone.0096219-Hiraishi2].where *n* is the number of PLFA or RQ component. In the PLFA profiles, *f~ki~* and *f~kj~* are the mole fractions of the *k* PLFA component for the *i* and *j* samples, respectively. In RQ profiles, *f~ki~* and *f~kj~* are the mole fractions of the *k* RQ component for the *i* and *j* samples, respectively (*f~ki~*, *f~kj~*\>1 mol%; Σ*f~ki~* = Σ*f~kj~* = 100 mol%). Cluster analysis was performed with the program KyPlot version 5.0 based on the *D* distance matrix and a dendrogram was constructed using the between-groups linkage method. Values≤0.1 of *D* of RQs are not recognized as different RQ profiles according to the analytical precision based on the duplicate analytical results including extraction and measurement process (97% statistical reliability) [@pone.0096219-Hu2]. For the PLFA analysis, we determined the threshold value, 0.13, in the same manner as the value of the quinone profiling method (*see* [@pone.0096219-Hu2]) using 12 duplicate results of the incubation sediment samples ([Fig. S1](#pone.0096219.s001){ref-type="supplementary-material"}). Statistical Analysis {#s2i} -------------------- Spearman's rank correlations (*r~s~*) were used to show the relationships among bacterial PLFA concentration, RQ concentration and organic carbon content and the relationships between OC content and DI. Pearson's correlation coefficients (*r*) were used to show the relationships between OC content and RQ/bacterial PLFA ratio and between DI and RQ/bacterial PLFA ratio. Analysis with Spearman's rank correlation and Pearson's correlation coefficient was performed using the statistical program PASW Statistics for Windows version 18J (IBM Japan, Tokyo, Japan). Mantel tests were used to test the significance of the correlation between dissimilarity matrices based on bacterial PLFA or RQ profiles, using the R package [@pone.0096219-R1]. Results {#s3} ======= PLFA and Quinone Concentrations {#s3a} ------------------------------- Total bacterial PLFA concentrations (i.e. the sum of SFAs \[C~13~--C~18~\], BFAs, and MUFAs \[≤C~19~\]) in the sediment varied over three orders of magnitude (range 1.2--834 nmol gdw^−1^) with lowest values for Japanese natural coastal sediment (JC-N-9) and highest values for Dutch intertidal natural sediment ([Fig. 1](#pone-0096219-g001){ref-type="fig"} and [Table 3](#pone-0096219-t003){ref-type="table"}). Total RQ concentrations in the sediment ranged from 0.01 to 28 nmol gdw^−1^ with lowest values for the Galicia bank (GB) and highest values for Japanese fish farm sediment, and were one to two orders of magnitude lower than the bacterial PLFA concentrations ([Fig. 1](#pone-0096219-g001){ref-type="fig"} and [Table 3](#pone-0096219-t003){ref-type="table"}). RQ concentration showed a positive log-log correlation with the bacterial PLFA concentration for the full dataset as well as within the individual sample sets ([Fig. 1](#pone-0096219-g001){ref-type="fig"} and [Table 4](#pone-0096219-t004){ref-type="table"}). However, there were clear differences in slopes of the fits for the individual sample sets with the highest slope for the Japanese fish farm sediments (1.487) (i.e. relatively rich in RQs) and the lowest slope for the Dutch intertidal incubation sediments (0.716) (i.e. relatively rich in PLFAs) ([Table 4](#pone-0096219-t004){ref-type="table"}). Two deep sea samples from the Arabian Sea (AS-2) and GB were relatively far from the overall trend line with relatively high PLFA concentrations and low RQ concentrations ([Fig. 1](#pone-0096219-g001){ref-type="fig"}). {#pone-0096219-g001} 10.1371/journal.pone.0096219.t003 ###### Concentration of bacterial PLFAs, respiratory quinones (RQ) and organic carbon (OC) in marine sediments in this study. {#pone-0096219-t003-3} Bacterial PLFA (nmol gdw^−1^) RQ (nmol gdw^−1^) OC (mg-C gdw^−1^) ---------------------------- ------------------------------- ------------------- ------------------- ----------- ------------------------------------- ----------- **All samples** 1.17−834 67.0±122 0.01−28.0 1.22±3.75 0.37−60.4 8.8±11.2 **Dutch intertidal (DI):** Natural 27.2−834 202±280 0.03−3.4 0.84±1.14 \-[\*](#nt110){ref-type="table-fn"} \- Lab incubations 11.5−89.3 43.0±26.2 0.10−0.36 0.23±0.11 3.6−16.3 9.8±6.2 **North Sea (NS):** 3.84−10.3 6.05±2.2 0.01−0.05 0.03±0.01 0.44−3.0 1.6±1.2 **Japanese coast (JC):** Natural coast 1.17−202 63.3±30.2 0.01−5.8 1.26±1.8 0.37−23.7 10.1±8.6 Fish farm 9.55−295 68.8±72.9 0.18−28.0 3.54±7.19 1.7−49.6 10.5±11.7 \*Not determined. 10.1371/journal.pone.0096219.t004 ###### Log/log power regressions and Spearman's rank coefficients between the bacterial PLFA (nmol gdw^−1^) and respiratory quinone (RQ) concentrations (nmol gdw^−1^), and between organic carbon (mg-C gdw^−1^) and the bacterial PLFA and quinone concentration of individual sample set. {#pone-0096219-t004-4} Bacterial PLFA (*x*) OC (*x*) OC (*x*) ---------------------------- ---------------------- ----------- --------------------- ----------- --------------------- ----------- **All samples** *y = *0.0049×^1.151^ 0.823\*\* *y = *6.326×^0.882^ 0.946\*\* *y = *0.039×^1.148^ 0.809\*\* **Dutch intertidal (DI):** Natural *y = *0.0096×^0.774^ 0.643 − − − − Lab incubations *y = *0.0156×^0.716^ 0.825\*\* *y = *5.991×^0.861^ 0.781\*\* *y = *0.045×^0.720^ 0.982\*\* **North Sea (NS):** *y = *0.0059×^0.899^ 0.624\* *y = *5.650×^0.138^ 0.253 *y = *0.028×^0.115^ 0.263 **Japanese coast (JC):** Natural coast *y = *0.0094×^1.122^ 0.983\*\* *y = *6.027×^1.011^ 1.000\*\* *y = *0.069×^1.149^ 0.983\*\* Fish farm *y = *0.0043×^1.487^ 0.952\*\* *y = *5.962×^1.022^ 0.880\*\* *y = *0.056×^1.570^ 0.847\*\* Levels of significance are \**P*\<0.05, \*\**P*\<0.01. Relationship Between Organic Carbon Contents and Bacterial PLFAs and RQs {#s3b} ------------------------------------------------------------------------ The sediment organic carbon (OC) content ranged from 0.4 to 60 mg gdw^−1^ (mean 8.8±11, mg gdw^−1^ *n* = 51) over more than two orders of magnitude in all samples ([Table 3](#pone-0096219-t003){ref-type="table"}). A positive power correlation between the OC contents and the bacterial PLFA concentrations was observed ([Fig. 2a](#pone-0096219-g002){ref-type="fig"} and [Table 4](#pone-0096219-t004){ref-type="table"}). This correlation was similar for all sample sets, except for the Dutch intertidal incubation and North Sea samples ([Table 4](#pone-0096219-t004){ref-type="table"}). Given the correlation, it was not surprising to find that the correlation between the OC contents and the RQ concentrations was also positive ([Fig. 2b](#pone-0096219-g002){ref-type="fig"} and [Table 4](#pone-0096219-t004){ref-type="table"}). {#pone-0096219-g002} RQ/PLFA Ratios {#s3c} -------------- We used a ratio based on mole concentration between total RQ and total bacterial PLFA (RQ/bacterial PLFA). The ratios of RQ/bacterial PLFA ranged from 0.0007 to 0.095 with lowest value for the deep sea sediment (GB) and highest values for Japanese fish farm (JC-FF-13) ([Fig. 3a](#pone-0096219-g003){ref-type="fig"}). Strong positive log-log correlations between the OC contents and the RQ/bacterial PLFA ratios of the Japanese fish farm samples were observed (*r* = 0.888, *P*\<0.01) ([Fig. 3a](#pone-0096219-g003){ref-type="fig"}), while ratios for the other sample sets, except deep sea samples, showed no significant correlation with OC content. {#pone-0096219-g003} Degradation Index Values for Japanese Sediments {#s3d} ----------------------------------------------- DI values for all Japanese samples were ranged strongly from −1.1 to −0.2 ([Fig. 3b](#pone-0096219-g003){ref-type="fig"}) with more negative values indicating more degraded (refractory) material. The DI value was positively correlated with the OC content (*r~s~* = 0.738, *P*\<0.05 for Japanese natural coast and *r~s~* = 0.702, *P*\<0.01 for Japanese fish farm), meaning the OC in the sediments with the highest OC content was relatively fresh (labile). A positive linear correlation between DI and the RQ/bacterial PLFA ratios only for the Japanese fish farm sediments was observed (*r* = 0.751, *P*\<0.01) ([Fig. 3b](#pone-0096219-g003){ref-type="fig"}), whereas there was no significant correlation for Japanese natural coastal sediments (*r* = 0.665, *P* = 0.072). Note that the positive relationship for Japanese fish farm sediments was due to the sample from Stn. JC-FF-13 (without the plot of Stn. JC-FF-13, *r* = 0.469, *P* = 0.106). Relative Composition of PLFA and Quinone Pools {#s3e} ---------------------------------------------- The composition of the bacterial PLFA (general \[SFAs (≤ C~19~), 16∶1ω7c, and 18∶1ω9c\] + specific) in the sediment showed less variation as compared to the composition of RQ ([Fig. 4](#pone-0096219-g004){ref-type="fig"}). The three dominant PLFAs, 16∶0, 16∶1ω7c and 18∶1ω7c, were present generally in almost all the samples ([Fig. 4a](#pone-0096219-g004){ref-type="fig"}). SFAs (≤ C~19~), 16∶1ω7c, and 18∶1ω9c as a *general* marker for bacteria accounted for 57 mol% of the total bacterial PLFA pool in all samples (range 47--71 mol%). Bacteria-*specific* PLFAs showed variation in the full dataset. Together, i15∶0, a15∶0, 10Me16∶0, and 18∶1ω7c as a *specific* marker for bacterial groups accounted for average 25 mol% (range 11--46 mol%) of the total bacterial PLFA pool in all samples ([Fig. 4a](#pone-0096219-g004){ref-type="fig"}). {#pone-0096219-g004} In general, the relative composition of RQ varied more strongly ([Fig. 4b](#pone-0096219-g004){ref-type="fig"}). The most obvious difference is seen between the deep-sea and the other (coastal and estuarine) sediments. Almost all coastal sediments except Japanese fish farm sediments were dominated by PQ-9 and UQ-8, while two deep sea sediments (AS-2 and GB) were dominated by MK-8(H~2~) and MK-8. Japanese fish farm sediments were dominated by UQ-10 and UQ-8. Together, UQ-8, -9, and -10 accounted for 45 mol% (range 9--83 mol%) of the total RQ pool in all samples. PQ-9 and K1, which are derived from photosynthetic organisms, were observed in not only coastal area, but also in the oxygen minimum deep-sea sediment (AS-1) ([Fig. 4b](#pone-0096219-g004){ref-type="fig"}). Cluster Analysis of the Pattern of Differences Among Samples in Individual PLFAs and RQs {#s3f} ---------------------------------------------------------------------------------------- The differences in the bacterial PLFA and RQ profiles for the different sample sets ([Fig. 4](#pone-0096219-g004){ref-type="fig"}) were further clarified by two cluster analyses. The first analysis was performed to investigate the co-variation in the relative abundance of the individual bacteria-*specific* PLFAs (sum of BFAs and MUFAs (≤C~19~) except 16∶1ω7c and 18∶1ω9c) and RQs ([Fig. 5](#pone-0096219-g005){ref-type="fig"}). When different compounds cluster closely, this indicates that these compounds are probably derived from the same bacterial groups. Two main clusters (cluster-1 and -2) were observed that were further divided into two sub-clusters (cluster-1a, -1b, 2a, and -2b) ([Fig. 5](#pone-0096219-g005){ref-type="fig"}). These five clusters were characterized by a relatively high mole fraction of group-specific bacterial PLFAs and RQs among all samples. It is noteworthy that UQs were present in cluster-1, whereas almost all partially saturated MKs were in cluster-2. {#pone-0096219-g005} Cluster Analyses of Bacterial PLFA and RQ Profiles {#s3g} -------------------------------------------------- The second cluster analysis was conducted to investigate the differences in bacterial community structure of the different sediments based on the bacterial PLFA and RQ profiles separately in order to compare the chemotaxonomic resolution of these two methods ([Fig. 6](#pone-0096219-g006){ref-type="fig"}). The bacterial PLFA profiles clearly separated into three main groups (group PI, PII, and PIII in [Fig. 6a](#pone-0096219-g006){ref-type="fig"}). Group PI comprised all Dutch intertidal natural sediments, while all other samples were included in group PII. The only exception here is a single Japanese natural coastal sediment sample (JC-N-9) that formed a separate cluster (PIII). Further differentiation involved division of group PII into six different groups (group PI-1∼6 in [Fig. 6a](#pone-0096219-g006){ref-type="fig"}) based on the threshold value of 0.13 (representing the observed level of dissimilarity between replicate samples, *see* [Fig. S1](#pone.0096219.s001){ref-type="supplementary-material"}). The RQ profiles were divided clearly into four main groups (group QI, QII, QIII, and QIV in [Fig. 6b](#pone-0096219-g006){ref-type="fig"}). Within these main groups, almost all sample sets were distinguished as separate groups based on the threshold value of 0.1 for sample discrimination of different RQ profiles [@pone.0096219-Hu2] ([Fig. 6b](#pone-0096219-g006){ref-type="fig"}). The general sample classification of the different sediments between bacterial PLFAs versus RQs based on the dissimilarity index was significantly correlated (using 10,000 randomizations, Mantel's coefficient *r* = 0.435, *P* = 0.0001). {ref-type="table"}). Parentheses in the abbreviation indicate the depth layer at the sampling station.](pone.0096219.g006){#pone-0096219-g006} Discussion {#s4} ========== Analysis of lipid biomarkers is a powerful tool for quantification of bacterial abundance and community structure. While PLFAs have been widely utilized as quantitative bacterial biomarkers in marine sediment [@pone.0096219-Findlay1], [@pone.0096219-Boschker1], applications of the quinone profiling method to marine sediments are still very few [@pone.0096219-Urakawa1], [@pone.0096219-Kunihiro1]. In this study, we analyzed concentrations of PLFAs and RQs in a broad range of marine sediments to investigate and compare their application as indicators of bacterial biomass and community composition. PLFAs and RQs as Bacterial Biomass Indicators {#s4a} --------------------------------------------- We found a strong correlation between total concentrations of the bacterial biomarkers PLFAs and RQs across several orders of magnitude, both for the individual sample sets and for the whole dataset ([Fig. 1](#pone-0096219-g001){ref-type="fig"}). PLFA concentrations have been used frequently as a measure of bacterial biomass in seawater and marine sediments (e.g. [@pone.0096219-Findlay1], [@pone.0096219-Boschker1], [@pone.0096219-Dijkman1]), because PLFA concentrations are relatively constant in bacterial biomass and PLFAs degrade rapidly upon death of the source organism, meaning that they are specific for *living* bacterial biomass [@pone.0096219-Boschker1]. The strong correlation across several orders of magnitude between PLFAs and RQs indicates that RQs also provide an estimate of *living* bacterial biomass in sediment. This allowed us to determine the conversion from RQ concentration (nmol gdw^−1^) to bacterial biomass (mg C gdw^−1^; biomass = 0.192 RQ^0.586^, *r~s~* = 0.853, *P*\<0.001, *n* = 59). The equation is based on the correlation between the RQ concentration and summed concentrations of four bacteria-specific PLFAs (i14∶0, i15∶0, a15∶0 and i16∶0) calculated by the equation detailed in [@pone.0096219-Middelburg1] using the conversion factors from [@pone.0096219-Evrard1]. Interestingly, this relationship is not linear, which is probably due to metabolic variation in quinone concentrations in bacterial cells in different environments. Previous studies have already demonstrated that total RQ concentrations correlated very well with microbial biomass carbon in soil (measured by a fumigation-extraction method, *r* = 0.96, [@pone.0096219-Saitou1]), with the total bacterial cell count in various environments (*r* = 0.98, [@pone.0096219-Hiraishi1]), and with bacterial cell volume in lake water (*r* = 0.98, [@pone.0096219-Takasu1]). Our study is the first to demonstrate the good correlation between concentrations of RQ versus PLFAs as a compound-specific biomarker for bacteria in marine sediment. Thus, these results indicate that RQ concentration can be utilized as a proxy for bacterial biomass in sediment samples. Despite the overall strong correlation between PLFAs and RQs, a more detailed look at [Fig. 1](#pone-0096219-g001){ref-type="fig"} reveals that there is residual variation to be explained. Firstly, the range in RQ concentrations was around one order of magnitude higher than that of the bacterial PLFA concentrations in the full sample set. Secondly, the slopes of the fits for the individual sample sets were different ([Fig. 1](#pone-0096219-g001){ref-type="fig"} and [Table 4](#pone-0096219-t004){ref-type="table"}), which implies that the different sediments contained bacterial communities with different RQ/PLFA ratios. We consider two possible explanations for this varying RQ/PLFA ratio. The first explanation is inherent group-specific differences in the RQ/PLFA ratios of the different groups of bacteria contributing to the overall bacterial community. This can be related, for example, to the type of energetic metabolism of the bacteria (e.g. [@pone.0096219-Collins2]). The second explanation concerns the activity of the bacteria. While bacterial PLFA concentrations (being a structural biomass component) are relatively stable under different conditions [@pone.0096219-Guckert1], [@pone.0096219-BrinchIversen1], concentrations of RQs in bacterial cells can also depend on the metabolic activity due to growth phase [@pone.0096219-Polglase1], substrate utilization [@pone.0096219-Hu3], and redox state [@pone.0096219-Mannheim1]. The strong PLFA versus RQ correlation over a broad range of sediments suggests that RQ concentration reflects mainly bacterial biomass but that may have an additional component related to the activity/metabolism of bacteria. If RQ concentrations are also dependent on the activity of the bacteria, RQ concentrations relative to PLFA concentrations (the RQ/bacterial PLFA ratio) may also depend on the quality and quantity of the OM in the sediment as these two factors directly influence bacterial activity [@pone.0096219-Polymenakou2], [@pone.0096219-Hoppe1], [@pone.0096219-Mayor1]. We investigated this relationship through assessment of the correlation between the RQ/bacterial PLFA ratio versus OM *quantity* (total OC content) for all samples and OM *quality* (i.e. DI, the amino acid-based degradation index) for the Japanese sediments ([Fig. 3b](#pone-0096219-g003){ref-type="fig"}). The absence of a clear correlation between OC content and the RQ/bacterial PLFA ratio for the full dataset ([Fig. 3a](#pone-0096219-g003){ref-type="fig"}) indicates that this ratio was not influenced by OM *quantity*. In addition, we also investigated the relationship between the RQ/bacterial PLFA ratio versus OM *quality* for the Japanese samples. The OM *quality* was determined by the degradation index (DI), which is based on the relative composition of hydrolysable amino acids in the sediment [@pone.0096219-Dauwe1]. This index provides an indication of the quality (or 'freshness') of the organic matter in the sediment with most negative values indicating relatively low quality (or 'refractory') material. Despite the wide range of observed DI values (−1.10 to −0.02), which indicates substantial variation in OM *quality* between samples for both the natural and fish farm sediments ([Fig. 3b](#pone-0096219-g003){ref-type="fig"}), there was no correlation between DI values and the RQ/bacterial PLFA ratio for the natural sediments and only a weak correlation for the fish farm sediments ([Fig. 3b](#pone-0096219-g003){ref-type="fig"}). Overall, our results indicate that there was no strong control of bacterial activity on the RQ/PLFA ratio by both quantity and reactivity of the OC pool. Still, the RQ/bacterial PLFA ratios for Japanese fish farm sediments were clearly higher than the natural sediments. According to previous studies, the ratio between total RQ and total PLFA concentration has been used to indicate mainly two aspects: a presence of aerobic bacteria and facultative heterotrophic bacteria and a respiratory activity in comparison with fermentation processes [@pone.0096219-Hedrick1], [@pone.0096219-Villanueva1], [@pone.0096219-Peacocka1]. Further investigation of the RQ/PLFA ratio, combined with a study on bacterial metabolism in marine sediments, is needed to explain the observed residual variation and the role of these two aspects. Linking PLFA and Quinone Biomarkers {#s4b} ----------------------------------- The cluster analysis as shown in [Fig. 5](#pone-0096219-g005){ref-type="fig"} was conducted to investigate the co-variation between the bacterial PLFAs and RQs in the different sediments and their association with specific bacterial groups. The cluster analysis showed two main groups: cluster-1 comprised UQs, which are specific for Gram-negative Proteobacteria (*see* [Table 2](#pone-0096219-t002){ref-type="table"}). Cluster-2 comprised almost all partially saturated MKs, which are predominantly present in Gram-positive bacteria, thereby indicating that cluster-2 was dominated by Gram-positive bacteria ([Fig. 5](#pone-0096219-g005){ref-type="fig"} and [Table 2](#pone-0096219-t002){ref-type="table"}). Based on the taxonomic assignment of PLFAs and RQs in [Table 2](#pone-0096219-t002){ref-type="table"}, the subcluster can be analyzed in more detail. Subcluster-1a comprised UQ-10, indicating that this cluster was dominated by members of the class Alphaproteobacteria. Subcluster-1b was characterized by the presence of MK-6, i15∶0, 10Me-16∶0, and cy17∶0, indicating that this cluster was predominance of the class Deltaproteobacteria. Subcluster-1c was characterized by UQ-8 and 18∶1ω7c, indicating that it comprised mainly members of the class Gamma- and Beta-proteobacteria. Betaproteobacteria are well known to be a minor group in marine sediment [@pone.0096219-Polymenakou2], therefore, Subcluster-1c must have been dominated by mainly members of the class Gammaproteobacteria. Cluster-2a was characterized by MK-10, MK-9(H~8~), and i17∶0, indicating that this cluster is relatively rich in members of the Actinobacteria. Subcluster-2b was characterized by MK-8, MK-9, and a15∶0, indicating that this cluster comprised members of the Bacteroidetes. Our study is the first to demonstrate a general agreement in the chemotaxonomic classification based on bacterial PLFAs versus RQs. This strengthens the use of these biomarkers for characterization of the sediment bacterial community. Although taxonomic resolution of both analyses is limited to identify phylogenetic groups of bacteria (low phylogenetic resolution), the value of this approach can be in combination with stable isotope probing to allow researchers to trace the flow of elements within communities [@pone.0096219-Boschker1]. Bacterial Communities of the Different Sediments {#s4c} ------------------------------------------------ The cluster analyses as shown in [Fig. 6](#pone-0096219-g006){ref-type="fig"} were performed to investigate the resolution of the two types of bacterial biomarkers and their ability to distinguish between bacterial communities from different sediments. In general, the bacterial PLFA- and RQ profiles revealed a similar classification pattern in bacterial community differences in our wide range of marine sediments ([Fig. 6](#pone-0096219-g006){ref-type="fig"}). However, there is a clear difference in the resolution of both methods with sample classification based on the RQ profile distinguishing 37 groups, whereas classification based on of the bacterial PLFA profile distinguishes only 13 groups. In other words, the level of dissimilarity between RQ profiles was substantially higher than the level of dissimilarity between the PLFA profiles. The higher sample discrimination in the RQ profile can be explained by the higher specificity of the RQs for specific bacterial groups (*see* [Table 2](#pone-0096219-t002){ref-type="table"}) as well as the more pronounced differences between RQ profiles of different bacterial groups (that are typically dominated by one RQ while PLFA profiles typically comprised 5∼17 PLFAs) [@pone.0096219-Hedrick1], [@pone.0096219-Haack1], [@pone.0096219-Zelles1]. In addition to this general trend, there were also some notable differences in the classification of the deep sea sediments (AS-1, AS-2, and GB), Japanese natural coastal sediment (JC-N-9), Japanese fish farm sediment (JC-FF-10 and JC-FF-13), North Sea sediment (NS-2), and Dutch natural intertidal sediment (DI-N-K) ([Fig. 6](#pone-0096219-g006){ref-type="fig"}). Two groups of RQs that were particularly important for the higher level of dissimilarity in RQ profiles are UQ-*n* and the partially saturated MKs. UQs are good markers for Alpha-, Beta-, and Gamma-proteobacteria, whereas as for PLFAs, the only specific marker for proteobacteria is 18∶1ω7c (*see* [Table 2](#pone-0096219-t002){ref-type="table"}). Partially saturated MKs, which exist in Actinobacteria, Deltaproteobacteria, and Epsilonproteobacteria, showed most variation between sample sets, especially, MK-8(H~2~), MK-9(H~2~), MK-7(H~4~), MK-9(H~4~) and MK-9(H~8~) which were present in more than 29 samples (15.4±9.2 mol% in the total RQ pool). Actinobacteria generally show a larger variation in MKs than PLFAs (e.g. [@pone.0096219-Kroppenstedt1]). Thus, higher sample discrimination in RQ profile could be due to presence of compounds originated from members of mainly Proteobacteria (UQ-*n*) and Actinobacteria (MK-*n*(H*x*)). Overall, we demonstrated that both concentration of bacterial PLFAs and RQs are good indicators for bacterial biomass and RQ profile can discriminate community difference more clearly than the bacterial PLFA profile. Thus, the combination between PLFA analysis (in combination with the stable isotope probing (SIP) technique) and quinone profiling method is a good strategy for studies on the role of bacteria in sediment biogeochemistry. These methods, and their applications, can be further expanded with the development of a method for stable isotope analysis of quinones so that quinones can also be applied in combination with stable isotope probing. Supporting Information {#s5} ====================== ###### **Analytical precision of total bacterial PLFA pools using dissimilarity values from the two PLFA pools resulted from duplicate analyses.** (TIF) ###### Click here for additional data file. ###### Sample codes and characteristics. (DOCX) ###### Click here for additional data file. We are grateful to Dr. Laura Villanueva and Prof. Arata Katayama for their constructive comments on this work, to Prof. Akira Hiraishi for his critical comments on Table 2, and to Pieter van Rijswijk, Marco Houtekamer, Peter van Breugel, and Cobie van Zetten for their analytical support. We thank Dr. Katsutoshi Ito, Dr. Hideki Hamaoka, Hidejiro Onishi, Dr. Jun-ya Shibata, and Dr. Atsushi Sogabe for their assistance to collect samples in Japan, and Prof. Hiroaki Tsutsumi for the opportunity to use the facilities of the Prefectural University of Kumamoto. We also thank to Dr. Todd W. Miller for critically reading the manuscript. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: TK BV DvO. Performed the experiments: TK BV DVC LP MlG KM MK KO. Analyzed the data: TK BV DVC LP MlG. Wrote the paper: TK BV HTSB DvO.

Introduction {#Sec1} ============ Left atrial and ventricular volumes and ejection fraction are important diagnostic and prognostic parameters, widely used in daily practice. Indication for cardiac surgery in patients with valvular abnormalities such as bicuspid aortic valve disease, relies heavily on accurate left ventricular (LV) volume and function assessment. For many years, two-dimensional echocardiography (2DE) has been the most widely used modality for LV volumetric assessments; however, it is based on geometric assumptions which cause inaccuracy and the reproducibility remains suboptimal. Three-dimensional echocardiography (3DE) has largely overcome these drawbacks as it has the ability to visualize cardiac structures from any perspective, entailing an accurate quantitative and more reproducible evaluation of cardiac chambers. However, the use of 3DE in daily clinical practice has been hampered, because there is a learning curve for data acquisition, and 3D data analysis can be a time-consuming process, moreover there is need for experienced observers. A fully automated and user-interference free algorithm may improve the feasibility of 3DE in daily clinical practice. Such a method has now been proposed in the new 'Heart Model' Software (LHM), which promises a rapid and accurate automated quantification of left atrial (LAV) and LV volumes and ejection fraction (LVEF). The feasibility and accuracy of the new "Heart Model" was recently compared to cardiac magnetic resonance imaging by Tsang et al. who concluded that this technique is strongly correlated with CMR, with a high reproducibility and short analysis time \[[@CR1], [@CR2]\]. However, it has not been reported whether this also applies to patients with valvular heart disease where high reproducibility and feasibility is very important. Therefore, this study aims to assess feasibility and reproducibility of the 'Heart model' software in a prospective cohort study by comparing the results between echo and CT using conventional 2D, 2D-xPlane, 3D transthoracic echocardiography TTE (3DE) techniques in patients with a bicuspid aortic valve (BAV) with moderate to severe aortic valve stenosis or regurgitation. Methods {#Sec2} ======= Patients selection {#Sec3} ------------------ Patients with a BAV who visited the outpatient clinic for regular follow-up between October 2014 and March 2016 were consecutively included into this prospective study. All patients underwent the full study protocol on the same day. The study protocol consisted of physical examination, electrocardiography (ECG), 2D and 3D echocardiography and a cardiac CT scan. The study was approved by the medical ethical committee of the Erasmus medical center and informed consent was given by all patients who participated in the study. Two protocols were used reference methods were used to compare the LFM data: in the first protocol 3DE was used as a reference standard, in the second protocol the LHM was compared to CT. Echocardiography {#Sec4} ---------------- ### Image acquisition {#Sec5} Two experienced sonographers (J.S.M., W.B.V.) performed a standard two-dimensional transthoracic echocardiogram (2DE). All studies were acquired in the left lateral decubitus position, in harmonic imaging using an EPIQ7 ultrasound system (Philips Medical Systems, Best, the Netherlands) equipped with a x5--1 matrix-array transducer (composed of 3040 elements with 1--5 MHz) A non-foreshortened apical four-chamber view (A4C) and two-chamber view (A2C) were recorded with manual or electronic rotation (iRotate) followed by a focused LV, A4C and A2C image. From the focused LV-A4C a true perpendicular image view (A2C) was acquired with xPlane mode, in order to retrieve both views from the same heart-beat. This was repeated with the focus on the true long axis of the LA \[[@CR3]\]. Real-time 3D-TTE was performed immediately after the 2D-TTE with the same ultrasound unit and transducer. A four-or six-beat full volume dataset of the LV and LA was acquired from the apical window during a single breath hold. Two extra datasets were acquired from the A4C view in the dedicated 'Heart Model' acquisition mode. ### Image analysis {#Sec6} Analysis was performed by A.T.H. and J.S.M. All measurements were blinded to patient specific information. Before measurements were performed, the quality of each dataset was evaluated by both observers. Patients were excluded from further analysis in cases of poor image quality (e.g. poor endocardial visualization). #### 2D echocardiography {#Sec7} LV end diastolic volume (LVEDV), LV end systolic volume (LVESV) and LVEF were calculate using the Simpsons bi-plane method of disk summation, as stated in the guidelines, from the standard A4C and A2C and apical xPlane images \[[@CR4]\]. LAV was calculated using the area length method. #### 3D echocardiography {#Sec8} Manual: LV volumes and LVEF were measured using commercially available software (QLab-3DQ, Philips medical systems). The user aligned the multiplanar view to obtain the true long axis of the LV in the A4C and A2C view. Landmarks were placed on the mitral annulus and apex. The endocardial border was traced automatically and adjusted manually where needed. For the LA volume, the true long-axis was aligned using the multiplanar mode in the end-systolic frame and the contour traced as mentioned above. The 3D dataset was scored feasible when the entire cardiac contour could be traced. Automatic: Offline fully automatic analysis of the datasets was performed using the Q Lab advanced 'Heart Model' analysis software. This software has previously described in detail before by Tsang et al. \[[@CR1], [@CR2]\]. In brief, this software detects the endocardial surfaces by using an adaptive algorithm. This identifies a global end-diastolic shape which it uses in combination with motion detection to determine an end-systolic cavity \[[@CR1], [@CR2]\]. The program combines information from a database of 1000 3D TTE studies and its endocardial surface detection to model the LA and LV. Afterwards it matches features from the known datasets to the current patient for which it needs, much like manual measurements, a minimum of approximately 14 or 15 LV segments. The final model (Fig. [1](#Fig1){ref-type="fig"}) is displayed with the possibility to manually edit the contours if the user deems this necessary. To better estimate the added value over existing 3D techniques we chose not to manually edit the contours. Fig. 1Example of a model created by the LHM. The model as generated by the LHM of left atrial (LA and left ventricular (LV) volumes Cardiac computed tomography {#Sec9} --------------------------- ### Data acquisition {#Sec10} Acquisition was performed on a 3rd generation dual-source CT (Somatom Force, Siemens Healthcare, Forchheim, Germany). Retrospective ECG gated spiral acquisition was used, with a mean dose length product (DLP) of (estimated effective dose 5 mSv, using k = 0.017) 362 mGy-cm kVp was modulated to patient size. Besides patient size and the selected kV the mA was modulated to the heartrate to provide a high mA pulse during 1--40% of the RR interval. The pitch was adapted to increase proportionally with higher heartrates. Reconstructions were made with a medium smooth kernel, with a slice thickness of 1.5 mm at an increment of 0.4 mm at every 5% of the RR interval. ### Image processing {#Sec11} Images were analyzed semi-automatically using Syngo Via software (vb 10., Siemens, Forcheim Germany). Image analysis was performed by AC, with 1 year of cardiac CT experience. All cardiac phases were analyzed the software automatically detects the ED and ES phase, this was manually changed if needed. The endo and epicardial contours are automatically placed by the software and manually adjusted were need. The papillary muscles and if present trabeculations were included into the LV lumen. The basal plane was selected perpendicular to the short axis at the level of the mitral valve. Care was used to make sure the basal plane was on the same level in the ED and ES phase. Statistics {#Sec12} ---------- The IBM SPSS^®^ statistics 21.0 software was used to analyze the data. Continuous variables were presented as mean ± standard deviation (SD) or as median with an interquartile range. Categorical variables were presented as frequencies and percentages. We tested for normality by calculating Z-values of skewness and kurtosis, using the Shapiro--Wilk test and by visually assessing the data. For comparison of normally distributed continuous variables between two groups the student's *t* test was used. To quantify correlations the Pearson or Spearman correlation test was applied. Intra-observer and inter-observer agreement between two investigators (A.T.H, J.S.M.) were assessed by repeated analysis of the same images in a third of the dataset a month after initial analysis at the same images and blinded to the initial results. The limits of agreement between two measurements were determined as the mean of the differences (bias) ± 1.96 SD and presented in a Bland--Altman plot \[[@CR5]\]. Additionally, the coefficient of variation (COV) was provided to compare the dispersion of two variables. The COV was defined as either the SD of the differences of two measurements divided by the mean of their means. The statistical tests were two sided and a p-value below 0.05 was considered significant. Results {#Sec13} ======= Fifty-three patients with a BAV were included; baseline characteristics are presented in Table [1](#Tab1){ref-type="table"}. Eight patients also had Turner syndrome. Four patients previously underwent aortic coarctation resection and three patients underwent a balloon dilatation of their stenotic aortic valve. One patient underwent closure of a type II atrial septal defect. Table 1Baseline characteristics of the study population (n = 37)ParameterMedian (IQR)Baseline Men, n (%)25 (68) Age (years)35.2 (23) Height (cm)178 (26) Weight (kg)72 (24) BMI (kg/m^2^)23.9 (3.0) SBP (mmHg)122 (20) DPB (mmHg)80 (19) 3DE LVEF \< 50%, n (%)16 (43)Mitral valve E-wave (m/s)0.70 (0.2) A-wave (m/s)0.50 (0.2) E/A-ratio1.2 (0.9) DT (ms)209 (67) E' septal, cm/s7.8 (3.1) Ee'-ratio9.0 (3.6)Aortic valve BAV (n = 32)  No AoI, n= (%)5 (16)  Mild AoI, n= (%)21 (66)  Moderate AoI, n= (%)5 (16)  Severe AoI, n= (%)1 (3)  Peak velocity (m/s)2.65 (1.6)  VT (cm)52.8 (45)  Gradient (m/s)28 (33) TS\* (n = 5)  No AoI, n= (%)4 (80)  Mild AoI, n= (%)1 (20)  Moderate, AoI n= (%)0 (0)  Severe AoI, n= (%)0 (0)  Peak velocity (m/s)1.4 (0.6)  VTI (cm)28.4 (12)  Gradient (m/s)8 (6.5)Data are expressed as median and inter quartile range (IQR) or as 'n=, (%)' for the variables 'gender', 3DE LVEF\<50% and 'aortic insufficiency'*AoI* aortic insufficiency, *BMI* body mass index, *SBP* systolic blood pressure, *DBP* diastolic blood pressure, *DT* deceleration time\*Subgroup of patients with Turner syndrome (TS) and a bicuspid aortic valve (BAV) In all 53 patients, 2DE measurements could be performed and functional echo parameters were measured, as presented in Table [1](#Tab1){ref-type="table"}. In four patients, no 3D dataset could be acquired. Additionally, 12 patients were excluded based on poor imaging quality. Left ventricle {#Sec14} -------------- ### Protocol 1: LHM versus conventional echocardiography {#Sec15} There was a good correlation between the LHM and the manual bi-and xPlane measurements for LVEDV, LVESV, and LVEF. In Table [2](#Tab2){ref-type="table"} the measurements of LV volumes and function and LA maximal volume are presented for all different methods. Figure [2](#Fig2){ref-type="fig"}a, b show the Bland--Altman plots for EF by 2DE bi-plane and xPlane compared to the LHM. The results of the agreement analysis between the LHM and the measurements based on the 2DE are shown in Table [2](#Tab2){ref-type="table"}. For clarity purposes, we did not include mutual correlations between the different echo modalities; however, Bi-plane and xPlane had a high correlation for LVEDV (r = 0.977), LVESV (r = 0.978) and LVEF (r = 0.702). Additionally, both methods correlated strongly with conventional 3D as expected. Table 2Correlations between LHM and four different methods of volumetric chamber quantificationMethodPhaseMean, SDPearson's r ^†^BiasLower LOAUpper LOALHM (n = 37)EDV (ml)146 ± 48ESV (ml)77 ± 29EF (%)47 ± 5LA (ml)61 ± 19LHM 2nd dataset (n = 12)EDV (ml)143 ± 610.99\*\*−0.8−97ESV (ml)78 ± 340.99\*\*1−79EF (%)45 ± 40.58\*−1.5−74LA (ml)57 ± 190.98\*\*−1−972DE Bi-plane (n = 37)EDV (ml)145 ± 540.93\*\*0.06−4141ESV (ml)77 ± 310.88\*\*0.2−29.129.4EF (%)47 ± 80.63\*\*0.5−11.812.8LA (ml)43 ± 160.61\*\*17−13.647.42DE xPlane (n = 37)EDV (ml)143 ± 520.94\*\*2−3236.4ESV (ml)77 ± 300.88\*\*0.8−27.128.8EF (%)47 ± 80.43\*\*0.6−1314.2LA (ml)42 ± 160.69\*\*18−8.845.23D (n = 37)EDV (ml)143 ± 500.97\*\*2−2327.6ESV (ml)71 ± 260.91\*\*6−16.929.8EF (%)50 ± 70.51\*\*−3−15.59.8LA (ml)53 ± 190.81\*\*9−13.132.1CT (n = 37)EDV (ml)185 ± 630.88\*\*−42−102.318.9ESV (ml)67 ± 240.81\*\*10−24.143.6EF (%)64 ± 50.2416−28.3−4.7Data are presented as mean and SD*EDV* end-diastolic volume,*ESV* end-systolic volume,*EF* ejection fraction,*LA*left atrium,*LOA* limit of agreement,*COV*coefficient of variation^†^Compared with the LHM, \*\*p\<0.01, \*p\<0.05. A negative mean implies a smaller value was given by the LHM Fig. 2Bland--Altman plots demonstrating inter-modality agreement of EF and LAV in 2DE biplane (*panel* **a** and **e**), 2DE xPlane (*panel* **b** and **f**), 3DE (*panel* **c** and **g**) compared to the LHM. Ejection fraction measured by bi-plane 2DE (**a**), by xPlane 2DE (**b**) and by 3DE (**c**). Left atrial volume measured by bi-plane 2DE (**e**), by xPlane 2DE (**f**) and by 3DE (**g**). The*solid lines* depict the mean difference of the two measurements; the*dashed lines* depict the limits of agreement.*COV* coefficient of variation The LHM correlated strongly with the 3D LV measurements as shown in Table [2](#Tab2){ref-type="table"}. The LHM seems to estimate slightly larger LVEDV and LVESV and a smaller LVEF compared to manual 3D (Table [2](#Tab2){ref-type="table"}). Figure [2](#Fig2){ref-type="fig"}c shows the Bland--Altman plot for EF measured by 3DE. 16 patients (30%) had to be excluded due to poor imaging quality, which is considered acceptable in a routine clinical setting for 3DE. ### Protocol 2: LHM versus CT {#Sec16} In the second protocol, where CT was used as a golden standard method, agreement was found for EDV and ESV (r = 0.88 and r = 0.81); mean differences were 42 ml (23%) and 10 ml (15%) respectively (p \< 0.001 and p = 0.002). However, compared to CT the LHM seems to estimate smaller LVEDV (Table [2](#Tab2){ref-type="table"}) and consequently also produces a smaller LVEF (mean difference 16%, p \< 0.001) as shown by the positive difference in EF between CT and the LHM in Fig. [3](#Fig3){ref-type="fig"}. The correlations for Bi-plane, xPlane and conventional 3D with CT for LVEDV, LVESV and LVEF (Fig. [4](#Fig4){ref-type="fig"}) were comparable; a relatively small LVEDV and therefore a lower LVEF compared to CT. Three patients did not undergo a CT scan for personal reasons, those three patients are excluded from analysis. Fig. 3Bland--Altman plots demonstrating inter-modality agreement between CT and 3DE and LHM. Agreement in end-diastolic (EDV) and end-systolic volume (ESV) and ejection fraction (EF) comparing CT versus conventional 3D echocardiography (*top row*) and CT versus the LHM (*bottom row*) Fig. 4Bland--Altman plots demonstrating inter-modality agreement between CT and 3DE and LHM. Agreement in end-diastolic (EDV) and end-systolic volume (ESV) and ejection fraction (EF) comparing CT versus 2D bi-plane echocardiography (*top row*) and CT versus xPlane echocardiography (*bottom row*) ### Inter- and intra-observer variability {#Sec17} The LHM has no intra-, or inter-observer variation as the model produced the exact same measurements when the same dataset is used (mean difference: 0 ± 0, p \< 0.001). In addition, when the measurements were repeated on a second acquisition of the same patients (n = 12) there was some variation but generally there was good agreement reflected by a small bias and narrow limits of agreement (Table [2](#Tab2){ref-type="table"}). The inter-observer variability for the LVEDV and LVESV on conventional 3DE was calculated (r = 0.909, p \< 0.001 and r = 0.862, p \< 0.001) in 14 (38%) patients. Left atrial volume {#Sec18} ------------------ The LAV was estimated consistently significantly larger by the LHM (Table [2](#Tab2){ref-type="table"}) compared with on both 2DE and 3DE as is shown in Fig. [2](#Fig2){ref-type="fig"}e--g. This 'overestimation' was the most evident comparing to the 2D methods and less pronounced compared with manual 3D measurements. The correlation between 2D Bi-plane and 2D xPlane measurements of the LA volume was very high (r = 0.946, p \< 0.001). When comparing the LAV measured on 2D with the LA volume measured on 3D, xPlane outperforms Bi-plane (r = 0.632, p \< 0.001 and r = 0.551, p = 0.002 respectively). Discussion {#Sec19} ========== The main findings of this study can be summarized as follows: Automated chamber quantification is feasible in patients with bicuspid aortic valve disease in a routine clinical setting.The 'Heart Model' provides accurate automatic measurements of LVEDV, LVESV and LVEF and LAV compared to 2D and 3D echocardiography.There is no inter or intra-observer variability and very little 'inter-dataset' variability. Left ventricular assessment {#Sec20} --------------------------- In daily clinical practice LV function is routinely estimated by bi-plane Simpson method of disk-summation; however, this requires sufficient experience and has limitations in accuracy and reproducibility, especially due to the geometric assumptions of the shape of the ventricular or atrial cavity inherent to 2DE \[[@CR4]\]. Moreover, the lack of a third dimension is generally considered to result in high inter-measurement variability and limits endocardial visualization, predominantly of apical lateral segments. Foreshortening of the LV often performed in an attempt to alleviate this shortcoming causes reduced accuracy and reproducibility \[[@CR6]\]. Volumetric quantification from 3D data sets allows frame-by-frame detection of endocardial surface and does not require manual image plane positioning or geometric assumption \[[@CR7]\] and has furthermore been shown to have a higher reproducibility than 2DE \[[@CR8]--[@CR10]\]. The bias of CT and 3D echocardiography is approximately as low as the bias of MRI when estimating LVEF \[[@CR11]\]. Additionally, previous studies showed that for LVEF CT has the best correlation with MRI. In this light it is remarkable that in our study the correlation with CT is weak, as the EDV seems to be systematically estimated to be larger by CT than by the LHM or other echo modalities \[[@CR11], [@CR12]\]. This systematic underestimation may in part reflect an inter observer variability, as more of the trabeculations were included in the LV cavity for CT, entailing a larger EDV. Also, the trabeculations in this population may be more pronounced than in healthy subjects, adding to the observed difference. Left heart chamber quantification by 3DE is hampered by several factors, mainly the ease with which these techniques can be used in daily clinical practice as a degree of experience is required. However, especially in the growing and aging population of patients with valvular and congenital heart disease, where patients regularly undergo extensive echocardiographic evaluation, user-independent and non-invasive follow-up imaging is much needed. Added value of LHM {#Sec21} ------------------ Accurate and reproducible measurements of left chamber volumes are very important in clinical practice, as they correlate with prognosis and determine treatment strategies. Moreover, in order to test new therapies, changes in these parameters must be accurate to demonstrate the efficacy of medical therapy or intervention. Our results indicate that intra-observer, inter-observer, and test--retest reproducibility of this method are very high and even exceptionally for intra-observer variability (0%) since this algorithm has no human interaction (i.e., phase selection and contouring are automatic). We had to exclude 16 patients (30%) in total due to imaging quality limitations which is comparable to conventional 3DE in a routine clinical setting. The main concern with 3D echocardiography is the accessibility in terms of time and skill required to produce accurate and reproducible results. The algorithm described in this study promises to improve on these points. This study demonstrates a high feasibility and accuracy in a population of patients with BAV disease, a population where fast, reliable, user-independent and non-invasive follow-up imaging by echocardiography is imperative. Left atrial volume {#Sec22} ------------------ Another remarkable finding is the relatively large LA volume estimated by the LHM compared to 2DE measurements. First, we suggested that this could be explained by the inclusion of the pulmonary vein orifice into the left atrial cavity by the algorithm. However, when carefully re-evaluating the LA contours, most seemed to adequately follow the left atrial walls. Another explanation could be that it is actually an underestimation of the 2DE measurements. This can partly be explained by the use of the length area formula which assumes the LA to be ellipsoid, which is evidently not always the case \[[@CR4]\]. The correlation of LA volume measurements with 3D TTE was better. Still, inherent flaws of 2D imaging in combination with LA anatomy could contribute to this discrepancy. In the standard 2DE the LA measurements are performed on the A4C and A2C views which are focused on the true long axis of LV. The true long axis LA may not be in the same plane as the LV and therefore may appear foreshortened on the apical four chamber view. In xPlane a special focus view on the true LA long axis was acquired which better correlated with the LHM. Limitations {#Sec23} ----------- The golden standard for quantitative volumetric heart chamber assessment is cardiac magnetic resonance imaging (MRI). Unfortunately, no cine cardiac images were available in this study. Therefore, no comparison with MRI could be made; consequently, we used volumetric data from CT as an additional source of validation. Additionally, we feel that EDV by 2DE has been slightly underestimated and therefore 3DE is best used as a reference modality. We were strict when it came to image quality therefore we had to exclude 12 patients from analysis. However, a sufficiently large cohort could be analyzed. Conclusion {#Sec24} ========== Automated chamber quantification is feasible and accurate in patients with bicuspid aortic valve disease in a routine clinical setting. The 'Heart Model' provides accurate automatic measurements of LVEDV, LVESV and LVEF and LAV and has a high reproducibility between dataset and no inter or intra-observer variability. It does however seem to underestimate end-diastolic volume compared to CT. Funding {#FPar1} ======= This study was funded by a grant of the Dutch Heart Foundation (The Hague, The Netherlands, Grant No. 2013T093). The Dutch Heart Foundation had no influence on study design; in the collection, analysis and interpretation of data; in the writing of the report; or the decision to submit the article for publication. Conflict of interest {#FPar2} ==================== All authors declare that they have no conflict of interest.

Introduction {#s001} ============ D[espite neonatal polycythemia]{.smallcaps} being more common in infants born to mothers with gestational diabetes mellitus (GDM) than in infants in the general population---to our knowledge---no published study has assessed the hematopoietic stem and progenitor cell (HSPC) population in the cord blood of neonates born to mothers with GDM. Even in infants born to women with only diet-controlled GDM, neonatal hematocrit values above 60% are occurring at a higher than 10% rate \[[@B1]\], and the prevalence of neonatal polycythemia in insulin-requiring GDM and type 1 diabetic pregnancies may even exceed the 30% and 40%, respectively \[[@B2],[@B3]\]. The pathophysiology of this neonatal complication of maternal diabetes has been studied focusing on later stages of hematopoiesis. Authors suggested decades ago that the increased human umbilical plasma erythropoietin (Epo) levels were secondary due to fetal hyperinsulinemia \[[@B4]\]; however, no data were available about the cord hematopoietic stem cell (HSC) population in GDM before our study. In addition, the *DPP-4-CXCL12* axis has been recently identified as an important regulator of HSC mobilization in the adult human bone marrow in response to ischemic stimuli \[[@B5],[@B6]\]. We recently reported that the serum DPP-4 enzymatic activity was decreased in the cord blood of neonates born to mothers with GDM \[[@B7]\], and therefore, it could be hypothesized that the cord HSC/HSPC population may be altered in GDM. Materials and Methods {#s002} ===================== Research design {#s003} --------------- This study has been conducted in a single Hungarian center in the framework of the New Horizons Initiative of the EFSD after receiving approval from the appropriate ethical bodies. Altogether, 90 pregnant women signed the informed consent following the 75 g oral glucose tolerance test (OGTT) from the 24th to 28th gestational week: cord blood sampling at delivery with HSPC population assessment was possible from umbilical cord blood samples of neonates born to mothers with GDM (*n* = 45) and nondiabetic controls (*n* = 42), and three individuals were not included in the final analysis due to damaged samples. Patients and cord blood collection {#s004} ---------------------------------- Altogether, 45 pregnant women with GDM as well as 42 control pregnant women were enrolled immediately after their routine OGTT between the 24th and 28th gestational week on a voluntary basis after signing the informed consent. We designated a control group as individuals who carry a pregnancy and were neither diagnosed with GDM nor with overt diabetes between the 24th and 28th gestational week at OGTT or later throughout the course of their pregnancy. The diagnosis of GDM has been established according to the Hungarian national recommendations (modified 1999 WHO recommendation---GDM: 75 g OGTT at 24--28 gw: FPG ≥6.1 mM, 120′ PG ≥7.8 mM). Individuals diagnosed with overt diabetes (classified as diabetes in the pregnancy category in 2013, WHO recommendation, that is, FPG ≥7.0 mM, 120′ PG ≥11.1 mM) were excluded from the analysis. Cord blood samples were collected from the umbilical vein using the cord clamping technique as soon as it was possible after birth. The most important clinical maternal data were recorded (including the age at delivery, 75 g OGTT 0′ and 120′ min values between the 24th and 28th gestational week, prepregnancy body mass index (BMI), weight gain during pregnancy, and third trimester HbA1c value in the GDM group). The birth weight percentiles were assessed using a downloaded program on fetal weight equations as a published global reference for birth weight percentiles using 3,340 and 3,500 g as the Hungarian population mean birth weight values at 40 weeks for female and male neonates, respectively \[[@B8]\]. Flow cytometric analysis {#s005} ------------------------ EDTA-treated cord blood samples were analyzed by flow cytometry (Beckman Coulter Navios flow cytometer, Kaluza 1.2 software). Leukocytes were gated and differentiated from debris according to their forward and side scatter. BD Via-Probe (BD Biosciences, San Jose, CA) cell viability solution (7-AAD labeled) was used to differentiate living from dead cells. Cell surface CD45 (label: FITC, isotype: IgG1, κ, Clone: HI30) and CD34 (label: PE-Cy7, isotype: IgG1, κ, Clone: 581) were stained with specific fluorescent antibodies (BioLegend, San Diego, CA). Residual red blood cells (RBC) were filtered out from the leukocytes according to their absent CD45 expression. Identification of stem and progenitor cell populations {#s006} ------------------------------------------------------ We identified CD34^+^ cells within the gate of the nucleated cells. Circulating HSPCs were defined according to the International Society of Hematotherapy and Graft Engineering (ISHAGE) criteria as CD34^+^ HPCs within the CD45^dim^ blast gate, indicating that they express CD45 at a lower intensity than matured leukocytes \[[@B9],[@B10]\]. Our gating strategy is indicated on [Fig. 1](#f1){ref-type="fig"}. {#f1} Statistical analysis {#s007} -------------------- Statistica software (version 11; StatSoft, Tulsa, OK) was used for statistical analysis. Kolmogorov--Smirnov test was used to assess normality. As data distributions were non-normal, the Mann--Whitney *U* test (MWU) was used to compare the data of the healthy control and GDM study populations on the primary outcomes (proportions of hematopoietic stem and precursor cells). The Kruskal--Wallis ANOVA test was used to compare multiple samples. Student\'s *t*-test was used to compare means of clinical data ([Table 1](#T1){ref-type="table"}) due to the normal demographic data distributions in the GDM and control populations according to the Kolmogorov-Smirnov normality test. For paired comparisons, we also performed power analysis using the mean values, the sample size numbers, and the population sigma values. ###### [Clinical Characteristics of the Pregnant Populations Studied]{.smallcaps}   *Plasma glucose values at 75 g OGTT (24--28 gw, mM)*         -------------------- ------------------------------------------------------ ------------------- --------------------- ---------------------- ---------------------- ------------------------------------------ GDM (*n* = 45) 5.07 (4.85--5.29) 8.9 (8.58--9.21) 27.9 (25.55--30.16) 35.07 (33.69--36.43) 7.58 (5.43--9.72) 5.24 (5.08--5.4) \[33.9 (32.15--35.64)\] Control (*n* = 42) 4.62 (4.42--4.83) 5.76 (5.41--6.11) 23.9 (22.74--25.01) 31.66 (29.9--33.42) 14.16 (12.62--15.69) NA *P* value 0.0036 \<0.0001 0.0024 0.0025 \<0.0001   BMI, body mass index; 95% CI, 95% confidence interval; GDM, gestational diabetes mellitus; OGTT, oral glucose tolerance test. Results {#s008} ======= Clinical data {#s009} ------------- Prepregnancy BMI and the age at delivery were significantly higher in the GDM group compared to the controls. The most important clinical data of the pregnant population are summarized in [Table 1](#T1){ref-type="table"}. Neonatal birth weight categories according to gestational age and the neonatal gender distributions in the study groups are indicated in [Table 2](#T2){ref-type="table"}. ###### [The Number of Neonates with Large, Appropriate, or Small Birth Weight for Gestational Age Born to Mothers with GDM or to Control Mothers]{.smallcaps} *Distribution of neonatal birth weight percentiles \[*n *(%)\]* *GDM* *Control* ----------------------------------------------------------------- --------------------------- --------------------------- Entire neonatal study population (gender distribution) *n* = 45 (F = 21; M = 24) *n* = 42 (F = 19; M = 23) *LGA* 13 (28.9) 9 (21.4) *AGA* 31 (68.9) 31 (73.8) *SGA* 1 (2.2) 2 (4.8) AGA, appropriate for gestational age; [F]{.smallcaps}, female; LGA, large for gestational age; M, male; SGA, small for gestational age. Proportions of hematopoietic stem and precursor cells {#s010} ----------------------------------------------------- The proportion of CD34^+^CD45^dim^ cells (HSPCs) among the nucleated cells was significantly (MWU *P* \< 0.05, statistical power = 60.8%) higher in the cord blood samples of neonates born to mothers with GDM (median 0.38% \[SD^2^ 0.07\]) compared to neonates born to nondiabetic mothers (0.32% \[SD^2^ 0.03\]). Effect size ( = 0.497) calculation was also performed according to Cohen. We found a significant difference (Kruskal--Wallis test *P* \< 0.05) in the proportion of HSPCs among the nucleated cord blood cell population among different study groups according to the maternal GDM diagnosis and treatment types (control: median 0.32% \[SD^2^ 0.03\], GDM requiring only diet 0.37% \[SD^2^ 0.07\], and GDM on insulin therapy 0.45% \[SD^2^ 0.05\]) ([Fig. 2](#f2){ref-type="fig"}). The Newman--Keuls post hoc test also demonstrated a significant difference between the control and the GDM on insulin study groups (*P* \< 0.05). {#f2} Although routine neonatal full blood count, hemoglobin level, and serum bilirubin measurements were not feasible in all participating neonates as per guidelines and due to technical difficulties, we detected only a borderline correlation between neonatal hemoglobin and bilirubin levels. Discussion {#s011} ========== This line of research was motived by major prior clinical findings: a long standing observation from the everyday clinical praxis proved that neonatal polycythemia is a feared complication of diabetic pregnancies occurring with a high enough prevalence to attract clinical attention (over 30% and 40% in insulin-requiring GDM or type 1 diabetic pregnancies, respectively). Polycythemia in the infants of diabetic mothers may be associated with serious clinical complications, including the hyperviscosity syndrome, thrombosis, respiratory distress with vascular congestion, heart failure, and cardiomegaly due to increased afterload and neurologic complications, such as stroke. The ∼20%--25% increase in the proportion of circulating HSPCs (CD34^+^CD45^dim^) among all nucleated cells in the cord blood of neonates born to mothers with GDM compared to nondiabetic women may be related to the altered fetal stem cell mobilization in GDM pregnancy or possibly due to the insulin therapy. Neonatal polycythemia was previously explained by the fetal hypoxia-induced increase in Epo production that was closely related to the control of maternal antepartum hyperglycemia (ie, maternal HbA1c levels during the last month of pregnancy in type 1 diabetic women) \[[@B11]\]. Although previously Epo was described to stimulate RBC production downstream in the path when precursor cells already lost their CD34 marker, it has been recently suggested that high systemic levels of Epo may reprogram HSPCs inducing an increased output toward an erythroid fate in vivo \[[@B12]\]. In a prior study, Azouna et al. found that the number of CD34^+^ cells was higher in the cord blood of neonates with higher birth weight (\>3.5 kg) and born to older mothers \[[@B13]\]. This finding might be in concordance with our findings; due to that, both of the parameters they indentified are clinically associated with GDM, higher maternal age is a risk, and higher birth weight is a possible consequence of GDM. Therefore, our study might provide a plausible common clinical path for their findings (ie, development of GDM); despite that the proportion of neonates in the large for gestational age group was only numerically higher in the GDM group (statistically nonsignificant), the maternal age was significantly higher in the GDM study group. In addition, a decreased cord serum DPP-4 activity of neonates born to women with GDM compared to the activity in samples of neonates born to nondiabetic women was recently reported \[[@B7]\]. This---theoretically---might implicate an altered cleavage of other chemokines, growth factors, including *CXCL12*, and may possibly result in an altered mobilization of HSPCs through binding to *CXCR4* \[[@B14]\], as this receptor is also present on human CD34^+^ cells from the umbilical cord blood \[[@B15]\]. Therefore, we may speculate that the *DPP-4-CXCL12-CXCR4* axis could also be involved in linking the fetal intrauterine metabolic environment and the stem cell population distributions to certain neonatal complications of the GDM pregnancy, possibly including but not limited to the neonatal polycythemia. Limitations {#s012} =========== The results presented here could not gain the 80% statistical power with the limited number of enrolled individuals and the calculated effect size of 0.5, considered as a "medium" relative size also indicates that the sample sizes should be further expanded and the current results may therefore only be interpreted as preliminary \[[@B18]\]. This was a hypothesis generating pilot study; therefore, no prior sample size calculation was performed. Although our results are coherent with the current knowledge about the regulatory mechanisms involved in HSPC mobilization, due to the limited sample sizes and the distribution of data we may not entirely exclude the occurrence of significances by chance alone. The reported data only include cord stem/progenitor cell proportions, and due to the unknown total cell counts, these results represent only one aspect of cell populations that is also characterized by absolute cell numbers. The assessment of more matured precursor forms (ie, the progenitor cell population) with additional markers (eg, when cells already lack CD133 positivity) is currently missing, yet the results are expected to be more pronounced in the view that those progenitor cells are closer to the matured RBC that are proven to be different in amount in the neonates born to mothers with GDM. To assess stem cell mobilization in a dynamic manner, both the DPP-4 activity and also the CXCL12 levels should have been measured at tissue levels (eg, bone marrow, placenta, and in the cord blood) to provide evidence for the altered CXCL12 gradient between the bone marrow and the periphery. In our study, we could not present any data about the potential alteration of the CXCL12 gradient between the stem cell niche of the neonatal bone marrow and the periphery; therefore, the discussion in this regard is hypothetical. Since in a prior report \[[@B13]\] higher CD34^+^ cell counts were found in the cord blood of neonates born to mothers with higher age, maternal age cannot be excluded as a confounding factor in this analysis. Acknowledgment {#s013} ============== This study has been funded by the European Foundation for the Study of Diabetes (EFSD) New Horizons Initiative. Author Disclosure Statement {#s014} =========================== No competing financial interests exist. [^1]: These authors contributed equally to this work.

INTRODUCTION {#s1} ============ BREAST CANCER {#s2} ============= At present, one in eight women in the United States will develop breast cancer \[[@R1]\]. Recent advances in breast cancer detection and treatment have decreased the mortality rate of breast cancer \[[@R1]\] but the success of treatment relies largely on detection of the disease at early stages \[[@R1]\]. A lack of knowledge regarding the molecular mechanisms underlying breast tumor progression to invasive and then metastatic disease limits the ability to treat advanced disease. The identification of factors that promote metastasis is essential for the development of new breast cancer therapies and a further reduction in breast cancer mortality \[[@R2]\]. Inflammatory breast cancer (IBC) is a highly aggressive, angioinvasive, and a highly metastatic form of breast cancer. IBC is associated with a high incidence of early nodal and systemic spread. Systemic chemotherapy, adjuvant therapy, surgery and radiation have not improved disease prognosis or overall survival of IBC patients \[[@R3], [@R4]\]. TUMOR MICROENVIRONMENT {#s3} ====================== The tumor microenvironment (TME) is the cellular environment in which the tumor exists and provides its niche for growth, progression, survival and evolution. The normal microenvironment provides crucial signaling to maintain appropriate tissue architecture, inhibit cell growth and suppress the malignant phenotype, and acts as a barrier to tumorigenesis \[[@R5]\]. Incorrect signals from TME can destabilize tissue homeostasis, initiate, promote, and push normal cells to malignant phenotype \[[@R5]\]. The TME is highly complex and dynamic, and includes blood vessels, immune cells, inflammatory cells, and components of the extracellular-matrix (ECM). TME is also rich in various cytokines, chemokines, ECM proteins, growth and angiogenic factors involved in autocrine but also paracrine cell signaling. The field of TME has expanded our understanding of cancer as more than a single factor driven disease. Rather, cancer biology involves complex reciprocal interaction and co-evolution between cancer cells and host stromal cells, interplay of soluble growth factors and chemokines as the key mediators involved in oncogenic signaling pathways of tumors. The list of anti-cancer therapies and clinical trials based on the role of the microenvironment in various types of cancers has grown long and it includes endostatin, Bevacizumab (Avastin), MK-2416, Anastroazole, Bay 43-9006, DX2400, Celecoxib, and PG545 etc. \[[@R5]\]. The majority of these are multikinase inhibitors blocking tumor cell growth pathways such as BRAF, Bcr-Abl, c-Kit, vascular endothelial growth factor receptor-1 (VEGFR-1), VEGFR-2, VEGFR-3, PDGFR and colony-stimulating factor-1 receptor \[[@R5]\]. Proteomic landscape of the breast cancer TME includes diverse factors involved in tumor growth, proliferation, metastasis, vascularity, evading cell death pathways and host immune system. In our previous study we demonstrated that inflammatory and invasive breast cancer TME is rich in OPG, chemokines such as urokinase-type plasminogen activator receptor (uPAR), Oncostatin M (OSM), IL-6 and GRO-α \[[@R6]\]. Here, we focus only on osteoprotegerin (OPG) as one of the factors present in the TME of inflammatory and invasive breast cancer cell lines, and discuss how it multitasks various functions to drive tumorigenesis. OSTEOPROTEGERIN (OPG) {#s4} ===================== OPG was first identified by sequence homology to the tumor necrosis factor receptor (TNFR) family during a rat intestine cDNA sequencing project and was named based on its function (Latin: os bone, protegere to protect) \[[@R7]\]. OPG was independently discovered and alternatively named osteoclastogenesis inhibitory factor (OCIF) \[[@R8]\], TR1 \[[@R9]\], and follicular DC-derived receptor-1 (FDCR-1) \[[@R10]\], which are found to be identical to OPG. The human OPG gene is located at chromosome 8q23-24 and is composed of 401 amino acids \[[@R11]\]. The OPG protein comprises 401 amino acids of which 21 form a signal peptide \[[@R12]\] (Figure [1](#F1){ref-type="fig"}). Human and murine OPG consist of four cysteine rich pseudo repeats located in the N-terminal, two death domains, and a heparin-binding site located in the C-terminus and a 21 amino acid signal peptide \[[@R11]\]. Signal peptide is cleaved to generate a mature form of 380 amino acids (Figure [1](#F1){ref-type="fig"}). At the N terminus, there are four domains (D1-D4), which have cysteine-rich TNF receptor homologous motifs \[[@R12]\]. These motifs are required and are sufficient for binding to its major target, the receptor activator of nuclear factor (NFΚB) ligand (RANKL), and for inhibiting osteoclastic differentiation and activation \[[@R12]\]. At the C terminus, there are tandem death-domain homologous regions (D5 and D6) followed by a heparin-binding site (D7) \[[@R12]\] (Figure [1](#F1){ref-type="fig"}). More specifically at position 400, there is a cysteine required for homodimerization of the molecule (Figure [1](#F1){ref-type="fig"}). OPG is a secreted protein with no transmembrane or cytoplasmic domain, and is produced as a monomer (55-62 kDa) that undergoes homodimerization \[[@R11], [@R12]\]. OPG is secreted as a disulfide-linked homodimeric glycoprotein with four or five potential glycosylation sites, generating a mature form of OPG of 110-120 kDa \[[@R12]\]. The dimeric form of the protein exhibits a greater affinity for RANKL and a higher heparin-binding capacity than the monomeric form. The heparin binding site (D7) of OPG facilitates its binding to cell membrane-associated heparan sulfates. Heparan sulfates are expressed on the cell surface as heparan sulfate proteoglycans (HSPGs), and are involved in cell signaling, controlling cell behavior, actin cytoskeleton regulation, cell adhesion, and cell migration \[[@R13]\]. For biological use, D7, D5, and D6 are removed, and the remaining amino acids 22-194 OPG peptide is fused to the Fc domain of human IgG1 (OPG-Fc), which maintains the potent dimeric nature of full-length OPG and exhibits a significantly increased circulating half-life \[[@R12], [@R14]\]. OPG is highly expressed in the adult lung, heart, kidney, liver, thymus, lymph nodes and bone marrow \[[@R11]\]. OPG is syn­thesized by several cells including stromal cells, osteo­blasts, vascular smooth muscle cells, B lymphocytes, and articular chondrocytes \[[@R11]\]. {#F1} OPG AND BREAST CANCER {#s5} ===================== One of the first studies to characterize OPG revealed its expression in two human breast cancer cell lines, MDA-MB-436 and MCF-7 \[[@R15]\]. Further studies have confirmed the expression of OPG in breast cancer cell lines and tissues. The MCF-7, MDA-MB-231 and T47D human breast cancer cells lines were tested for OPG mRNA expression by real time PCR alongside 12 primary breast tumor samples \[[@R16]\]. OPG is expressed in 40% of breast cancers but not in normal breast tissue, and the TRAIL sensitive breast cancer cell line MDA-MB-436 produces sufficient levels of OPG to inhibit TRAIL-induced apoptosis *in vitro* \[[@R17]\]. OPG mRNA expression is upregulated in human breast cancer cell lines and tumor samples. The expression pattern of OPG was examined by immunohistochemistry in 400 invasive breast cancer tissue samples \[[@R18]\]. It was found that 40% of the invasive breast tumors expressed OPG with expression confined to tumor cells. It should be noted that the expression of OPG by tumor cells is not limited to the cancer cell expressing it but it has tremendous paracrine effect on the neighboring healthy cells. Our study \[[@R6]\] for the first time, revealed that OPG is secreted and expressed at very high levels from the SUM1315MO2 invasive breast cancer cell line, as well as the SUM149PT and SUM190PT inflammatory breast cancer cell lines. OPG was secreted at a concentration of 500 pg/ml and 1100 pg/ml from SUM1315MO2 and SUM149PT, respectively \[[@R6]\]. Our study \[[@R6]\] demonstrated specific OPG staining in inflammatory breast cancer patient tumor sections apart from invasive breast cancer tumor sections which has been reported previously \[[@R18]\]. Heavy secretion of OPG led us to hypothesize that OPG directly or indirectly might be involved in inducing various oncogenic factors and thus contributing to the severity of the disease. Breast cancer tissue is a heterogeneous system, mainly composed of tumor epithelial cells and stromal cells. The crosstalk between malignant and nonmalignant cells takes place in the breast TME at the primary site. This interaction, which can be *via* cytokines and chemokines, plays a major role in the various steps of breast cancer progression \[[@R19]--[@R23]\]. Therefore, the bidirectional crosstalk between the breast cancer cells and the tumor microenvironment components including immune cells, mesenchymal stem cells (MSCs), tumor associated fibroblasts (TAFs), fibroblasts, tumor vasculature, and extracellular matrix play vital roles in shaping tumor progression, aggressiveness and the bulk of the disease. This interaction, mainly driven by soluble secreted factors allows tumor cells to modify the stroma *via* tissue remodeling and gene expression and vice versa \[[@R22]\]. Therefore, defining the nature of the signals exchanged between the tumor microenvironment and the tumor cells should provide insights into how breast cancer develops and progresses, and may help to reveal therapeutic modalities based on intercepting the tumor-stroma crosstalk. Recruitment of the different components, including the MSCs and TAFs, is mediated by binding of different chemical messengers such as the TRAIL, TNF-α, TGF-β, chemokine (C-C motif) ligands, IL6, and platelet derived growth factor (PDGF) to their respective receptors. The MSCs are known to increase tumor progression, which is mediated by different cytokines such as OPG, TRAIL, IL-6, IL-8, CCL-5, CCL-2, and RANKL. These factors are known to increase the breast cancer development by regulating patient survival, bypassing apoptosis, invasion, migration and tumor angiogenesis \[[@R24]\]. OPG, TRAIL, and RANKL are significantly higher in tumor epithelial cells from patients with breast cancer than in epithelial cells of non-neoplastic breast tissues \[[@R25]\]. Moreover, the expression of OPG, TRAIL, RANKL, and RANK was significantly higher in spindle-shaped stromal cells from patients with breast cancer than in non-neoplastic tissue stromal cells \[[@R25]\]. ROLE OF OPG IN METASTASIS {#s6} ========================= Metastasis is an intricate multistep process, which is closely associated with worse prognosis of patients with tumors, and spread of the tumor. For example, bone metastasis is considered lethal and is an active area of research for potential therapeutic developments targeting bone cancer metastases. Bone metastases occur in more than 70% of breast cancer patients and cause severe skeletal complications such as fractures, spinal cord compression, bone pain, and hypercalcemia \[[@R26]\]. Patients with estrogen receptor-positive (ER+) breast cancer constitute a major clinical population who are at risk for bone metastases \[[@R26]\]. More than 50% of primary breast cancer cells express OPG and RANK, while RANKL could be detected only in 14-60% \[[@R25]\]. OPG is one of the important players of the OPG/ receptor activator of nuclear factor kappa-B (RANK)/RANK ligand (RANKL) triad (Figure [2](#F2){ref-type="fig"}), and is a secreted member of the TNFR superfamily of proteins \[[@R27]\]. OPG functions as an en­dogenous antagonist receptor that prevents the biological effects of RANKL, both membranous and soluble, and thus acts as an inhibitor of bone remodeling and resorption. Unlike RANK and RANKL, OPG does not have a transmembrane domain or cytoplasmic domain \[[@R28]\]. OPG is thus a decoy receptor, which interferes with the osteoclastic RANKL/RANK signaling to prevent bone loss \[[@R29]\]. ![Schematic diagram depicting the diverse signaling pathways that are triggered or modulated by OPG\ OPG can bind to various cell surface receptors such as Syndecan-1, membranous RANKL, and αVβ5 integrin in order to trigger different cell survival pathways particularly, ERK, PI-3K/Akt pathway which results in expression of various cell survival favorable genes. Simultaneously, stable β-catenin of the canonical Wnt/β-catenin pathway, which is turned on in the majority of cancer cells, translocate to nucleus and turns on OPG gene expression. The OPG gets secreted out of the cell in the microenvironment, and can exert its paracrine and autocrine effects. The released OPG can affect the nearby healthy cells thus driving them towards tumorigenesis. The released OPG can also exert autocrine effects on the cancer cells by binding to cell surface receptors. The binding of OPG turns on the vicious cell survival signaling in cancer cells thus favoring their growth, proliferation, survival, angiogenesis \[[@R6]\].](oncotarget-07-42777-g002){#F2} The four cysteine rich pseudo repeats form an elongated structure and binds to one of the grooves of the active RANKL therefore preventing RANKL/RANK interaction and hence osteoclastogenesis. OPG production is modulated by sev­eral cytokines, vitamins, estrogens and other molecules, thereby modulating osteoclastogenesis and bone resorp­tion. OPG production is induced by 1α, 25-dihydroxyvi­tamin D3, estrogens, pro-inflammatory cytokines such as interleukin-1 (IL-1) and TNF-α as well as transforming growth factor-β (TGF-β), whereas parathyroid hormone (PTH) and glucocorticoids inhibit OPG production. OPG expression is modulated by biphosphonates in osteoblasts \[[@R30], [@R31]\]. Another study has also highlighted the possibility of bone derived OPG to increase survival of breast cancer cells that reach the bone microenvironment as part of the metastatic process thus promoting the breast cancer mediated bone osteolysis \[[@R32]\]. Aggressive breast cancer malignancies metastasize to bone and are associated with dysregulation of the RANK/RANKL/OPG pathway and can increase the RANKL/OPG ratio, which would favor excessive osteolysis \[[@R33]\]. OPG blocks the maturation of bone resorbing osteoclast cells in the bone microenvironment thus preventing bone resorption. OPG does not have a transmembrane domain or cytoplasmic domain \[[@R11]\]. A high affinity anti-RANKL monoclonal antibody denosumab (AMG162; bone antiresorptive drug) targeting the osteoblast/cancer cell interphase has been developed to prevent bone loss in prostate and breast cancer bone metastases by slowing down bone turnover, hence prostate and breast cancer growth on the skeleton \[[@R29]\]. Clinical phase III study showed that even patients with breast cancer without bone metastases but with reduced bone density due to treatment with aromatase inhibitors benefit from treatment with denosumab \[[@R34]\]. The treatment significantly increased bone density after 1-2 years compared with the placebo treated control group \[[@R34]\]. AMGN-0007 is a recombinant OPG construct and suppressed bone resorption when administered to multiple myeloma (MM) and breast cancer patients during a phase I clinical trial \[[@R35]\]. Knocking down OPG expression in triple-negative breast cancer cells led to a significant reduction in metastasis in the chick embryo metastasis model. A reduction in metastasis was observed from both a primary tumor and by intravenous injection of tumor cells, suggesting a direct impact of OPG on metastasis \[[@R2]\]. The discovery that OPG is a potent inhibitor of osteoclast activity and maturation initiated research into the possibility of using this molecule as a therapeutic agent for the treatment of a variety of conditions associated with increased bone resorption, including tumor-induced bone disease \[[@R36]\]. OPG AS A SURVIVAL FACTOR {#s7} ======================== Discovery of OPG\'s ability to bind and inhibit the activity of TRAIL (TNF related apoptosis inducing ligand) opened an altogether new avenue for research suggesting that OPG production may provide cells with a survival advantage (Figure [2](#F2){ref-type="fig"}) \[[@R37]\]. TRAIL is produced in tumors by invading monocytes, inducing apoptosis in neoplastic cells sensitive to this cytokine. It has been shown in MDA-MB 436 and MDA-MB 231 cells that OPG produced by breast cancer cells enhances tumor cell survival by inhibiting TRAIL-induced apoptosis \[[@R18]\]. *In vitro* studies using a number of different tumor types have supported this hypothesis \[[@R18], [@R32], [@R38]--[@R40]\] but an undisputed functional link between OPG and cell survival in cancer has been not established to date. Resistance to apoptosis being a hallmark of cancer also correlates with aggressiveness of the tumor and poor prognosis. Wnt/ß-catenin, a major pathway of cell proliferation and growth has been shown to drive OPG expression thus leading to cell survival in colon cancer (Figure [2](#F2){ref-type="fig"}) \[[@R41]\]. In addition, the OPG produced by endothelial cells may increase survival through binding to TRAIL, thus helping the tumor cells to proliferate and increase metastatic bulk (Figure [2](#F2){ref-type="fig"}). One of the study highlighted the association of OPG expression in endothelial cells with increased tumor grade. The study also reported negative correlation between endothelial OPG and ER status of the tumors. OPG has been shown to be able to act as an autocrine survival factor for both breast and prostate tumor cells *in vitro* and this could also be a mechanism used by endothelial cells potentially favored by conditions within high grade tumors \[[@R42]\]. Apart from inducing angiogenesis and bypassing apoptosis by quenching TRAIL ligand, our study \[[@R6]\] for the first time demonstrated that OPG has the potential of reprogramming healthy human mammary epithelial spheres (HMEC), driving them towards tumorigenesis thus mimicking the scenario in breast cancer spheres (Figure [2](#F2){ref-type="fig"}). The addition of OPG to the normal HMEC growth media induced proliferation in the HMEC spheres. Another study \[[@R43]\] showed that intra-tibial tumors from the MCF-7 cells overexpressing OPG had an increase in cells staining positive for the proliferation marker Ki67. These observations together suggest that OPG can induce proliferation. Apart from a drastic increase in proliferation, few morphological changes were also induced in the control spheres. Aneuploidy, a hallmark of cancer, has been proposed to initiate tumorigenesis and is a remarkably common characteristic of tumor cells \[[@R44]\]. The increase in proliferation in HMEC spheres can be corroborated with the onset of aneuploidy markers such as IAK-1, Bub-1 and BubR1 \[[@R6]\]. Our goal was to decipher the cellular mechanisms of how OPG modulates and reprograms the normal mammary epithelial cells to a tumorigenic state thus suggesting promising avenues for treating IBC as well as highly invasive breast cancer with new therapeutic targets \[[@R6]\]. Our study also highlighted how OPG upregulates the phophorylated survival kinases such as Erk, Akt, GSK3β and p65 in HMEC spheres which contends the increased proliferation as seen in HMEC spheres in the presence of OPG (Figure [2](#F2){ref-type="fig"}) \[[@R6]\]. With DNA copy number variations (CNVs) in cancer cells having prognostic impact, it opens several avenues for therapeutic treatment and translational research. OPG selectively amplified the DNA copy numbers of AKT1, AURK1, EGFR, MYC and PAK1, CDK4 and downregulated tumor suppressive CDKN2A, PTEN and TOP2A genes (Table [1](#T1){ref-type="table"}) \[[@R6]\]. OPG has been reported to exert its effects *via* OPG receptors, such as type II membrane forms of RANKL \[[@R45], [@R46]\], TRAIL \[[@R47]\] and heparan sulfate containing proteoglycans, such as syndecan-1 (Figure [2](#F2){ref-type="fig"}) \[[@R13], [@R48]\]. Interestingly, our study for the first time showed how one of the crucial breast cancer stem cell markers CD24 was upregulated in HMEC spheres in the presence of OPG which also supports the sustained proliferation in control spheres. Recent studies have demonstrated OPG expression in breast cancer tissue samples and in a large cohort of invasive breast cancers (*n* = 400), 40% of samples showed OPG expression that was confined to tumor cells \[[@R18]\]. Interestingly, the gain in OPG gene copies was observed in 182 out of 934 tumors when the TCGA-2013 human breast invasive carcinoma data set was analyzed through the cBioPortal website \[[@R2]\]. The presence of an OPG copy number gain is a significant predictor of decreased overall survival or poorer prognosis in this cohort \[[@R2]\]. ###### Comparison of copy number variations in breast cancer cell lines, inflammatory breast cancer tissue from patient, and HMEC sphere cultures grown for long time in the presence of OPG rich microenvironment Gene SUM149PT SUM1315M02 Patient sample HMEC sp w/500pg/ml rhOPG --------------------------------- ---------- ------------ ---------------- -------------------------- AKT1 (serine-threonine kinase) ✓ ✓ ✓ ✓ AURKA (Aurora-A kinase) ✓ ✓ ✓ ✓ CDK4 (cell cycle progression) ✓ ✓ ✓ ✓ EGFR (surface receptor) ✓ ✓ ✓ ✓ ERBB2 (RTK;oncogene) ✓ ✓ ✓ ✓ MYC (regulator of transcripton) ✓ ✓ ✓ ✓ PAK 1 (RhoGTPases) ✓ ✓ ✓ ✓ PTEN (tumor sup.) X X X X CDKN2A (CDK inhibitor) X X X X RB1 (tumor sup.) X X X X TOP2A X X X X Red check marks denote the remarkably high gene expression of AKT1, AURKA, CDK4, EGFR1, ERBB2, PAK1 and MYC copy number in the DNA prepared from sphere cultures of SUM149PT, SUM1315MO2, inflammatory breast cancer tissue from patients, and HMEC spheres cultured for long time in the presence of human recombinant OPG (500pg/ml). Green crosses depict the downregulation of tumor suppressors (RB1 and PTEN), cycle regulator CDKN2A, and DNA repair enzyme topoisomerase DNA II alpha TOP2A in SUM149PT, SUM1315MO2, inflammatory breast cancer tissue from patients, and HMEC spheres cultured for long time in the presence of human recombinant OPG (500pg/ml). OPG AND ANGIOGENESIS {#s8} ==================== Angiogenesis is an essential step for breast cancer progression and dissemination. The development of new blood vessels in a tumor setting (angiogenesis) is conducted by numerous physiological and pathological stimuli. These stimuli can be various cytokines, chemokines, and growth factors. Molecular players of angiogenesis have been characterized since the early years of angiogenic studies, and one of the most prominent stimulating growing factors is certainly the vascular endothelial growth factor (VEGF) family. The most prominent member of this family, VEGF, is the foremost controller of physiological and pathological angiogenesis. An increasing number of reports now consider that OPG also has a function in other biological systems, including the vasculature \[[@R49]--[@R51]\]. It is highly possible that OPG produced either by the endothelial cells themselves or by the surrounding tumor cells, may result in increased endothelial cell survival and differentiation to form blood vessels, thereby facilitating tumor growth. OPG has been implicated in microvascular endothelial cell proliferation, migration, and induction of angiogenesis (Figure [2](#F2){ref-type="fig"}) \[[@R52]\]. Involvement of the heparin-binding domain D7 in OPG\'s proangiogenic activity has been suggested by McGonigle et al. 2009 \[[@R53]\]. OPG may interact with endothelial colony forming cells (ECFCs) through its binding to HSPG\'s, syndecan-1 (Figure [2](#F2){ref-type="fig"}), thereby exerting an anti-adhesive effect and promoting ECFC migration through a SDF-1/CXCR4 dependent pathway (as well as the MAPK and the Akt cascades), finally facilitating their recruitment to sites of neoangiogenesis \[[@R54], [@R55]\]. Our study \[[@R6]\] also corroborates the previous findings \[[@R42], [@R54], [@R56]\] highlighting OPG\'s important role in angiogenesis and neovasculogenesis in endothelial tube formation in an *in vitro* model of angiogenesis (Figure [2](#F2){ref-type="fig"}). When we used 500pg/ml and 1100pg/ml rhOPG, the length of the neovascular tubes were increased along with the number of branch points when compared to control media without rhOPG \[[@R6]\]. Quantitatively, conditioned media from SUM149PT and SUM1315MO2 adherent cell conditioned medium induced \~ 5-fold and 2.5 fold more branch points/field, respectively than HMEC medium \[[@R6]\]. However, OPG-depleted SUM149PT and SUM1315MO2 conditioned media reduced node formation by 36 and 51%, respectively \[[@R6]\]. OPG AND ITS BINDING PARTNERS IN BREAST CANCER CELLS {#s9} =================================================== In our quest to understand more about the binding partners of OPG we performed mass spectrometry analysis. OPG pulled down proteins controlling cell proliferation (nucleolin, IQGAP1/3, proliferation inducing gene 32, proliferation inducing gene 44), proteins involved in transport and fusion (valosin and ATP binding cassette), and proteins involved in gene expression (Leucine rich PPR, DHX9 or DEAH, t-RNA synthetase) (Figure [3](#F3){ref-type="fig"}) (unpublished results). OPG pulled down many cytoskeleton elements such as myosin, filamin, keratin, ankyrin, vinculin, and α-actinin (Figure [3](#F3){ref-type="fig"}) (unpublished results). {#F3} Interestingly, immunoprecipitation of breast cancer cell extracts by OPG antibody revealed a major band at a molecular mass of 110 kDa (unpublished results). Mass spectrometry analysis revealed it to be the protein nucleolin (Figure [3](#F3){ref-type="fig"}) (unpublished results). Nucleolin is a multifunctional shuttling protein present in the nucleus, cytoplasm, and on the surface of some types of cells. Nucleolin is a major constituent of nucleoli in exponentially growing cells and functions in the organization of nucleolar chromatin, packaging of pre-rRNA, rDNA transcription, and ribosome assembly by shuttling between the nucleus and the cytoplasm \[[@R57]--[@R59]\]. Expression of nucleolin on the cell surface has been reported in HeLa cells \[[@R60]\], lymphoblastoid T cells \[[@R60]\], breast carcinoma cells \[[@R61], [@R62]\], lung \[[@R63]\], and laryngeal epithelial cells \[[@R64]\], and hepatocarcinoma cells \[[@R68]\]. Nucleolin is expressed on the surface of endothelial cells in angiogenic blood vessels \[[@R65]\]. The interaction between nucleolin and OPG in breast cancer cells adds another layer of complexity to how OPG could be manipulating functions at the nuclear levels, and these studies are ongoing in our lab. The observation of cytoskeleton elements being pulled down validates with findings that have been reported previously \[[@R43]\]. Studies have investigated the mechanism whereby OPG could promote angiogenic behavior of endothelial cells. Human dermal microvascular endothelial cells (HuDMECs) treated with OPG were elongated with extensive actin networks compared to untreated cells \[[@R43]\]. This was due to cytoskeletal reorganization and cell spreading which was mediated by Focal Adhesion Kinase (FAK) as phosphorylation of FAK at tyrosine 397 was induced by treatment with OPG \[[@R43]\]. Valosin-containing proteins; also known as p97 or VCP, is an abundant ATPase which has the ability to use the energy derived from ATP hydrolysis to unfold client proteins \[[@R65]\]. VCP engages in a range of cellular processes such as ER-associated degradation (ERAD) that mediates the extraction of misfolded proteins across the ER membrane and their delivery to the proteasome. The ATPase valosin-containing protein (VCP; p97) is an essential regulator of protein degradation in multiple pathways and has emerged as a target for cancer therapy. VCP inhibition affects protein synthesis, eukaryotic initiation factor 2α (eIF2α) and mechanistic target of rapamycin complex 1 (mTORC1), and attenuates global protein synthesis. VCP inhibitors perturb intracellular amino acid levels, activated eukaryotic translation initiation factor 2α kinase 4 (EIF2AK4), and enhance cellular dependence on amino acid supplies, consistent with a failure of amino acid homeostasis \[[@R65]\]. Thus, depletion of VCP triggers cancer cell death in part through inadequate control of protein synthesis and amino acid metabolism, and allows it to have implications for the development of anti-cancer therapies \[[@R65], [@R66]\]. OPG\'s interaction with VCP in the context of invasive and inflammatory breast cancer opens up many interesting avenues to research the therapeutic implications. OPG interestingly pulled down lipid metabolic enzyme, fatty acid synthase (FASN), which is a key enzyme of the fatty acid biosynthetic pathway in breast cancer cells (Figure [3](#F3){ref-type="fig"}) (unpublished results). FASN controls the process of producing de novo fatty acids from carbohydrate and amino acid-derived carbon sources \[[@R67]\]. FASN is a multifunctional polypeptide enzyme that produces saturated fatty acids, uses one acetyl-CoA and sequentially adds seven malonyl-CoA molecules to produce the 16-carbon saturated palmitic acid \[[@R67]\]. Overexpression of FASN has been strongly associated with many cancer types because it plays important metabolic roles in molecular pathways regulating cancer cell proliferation and tumor development \[[@R67]\]. Reduction in FASN enzyme activity by chemical inhibitors including orlistat, cerulenin and triclosan have been reported to remarkably decrease progression in various cancer cell types \[[@R68]\]. FASN is an attractive therapeutic target as it regulates neoplastic transformation, metastasis as well as angiogenic pathways manipulating tumor vascularity and cell proliferation \[[@R69]\]. It also serves as a potential diagnostic and prognostic biomarker as it is secreted in the blood of patients with breast, prostate, colon and ovarian cancers compared with normal healthy subjects \[[@R69]\]. In our recent study we demonstrated that compared to HMEC, SUM149PT and SUM1315MO2 breast cancer cells express increased level of FASN \[[@R70]\]. We have several interesting findings in the context of paracrine signaling in the OPG rich breast cancer microenvironment that drives carcinogenesis via inducing and sustaining inflammatory cycloxygenase-2 (COX-2) and lipogenic FASN in an invasive breast cancer setting (Figure [3](#F3){ref-type="fig"}) (unpublished results). Our lab findings reveal the synergistic anti-proliferative role of the COX-2 inhibitor celecoxib and FASN blocker C75 in aggressive metastatic breast cancer cells (Figure [3](#F3){ref-type="fig"}) (unpublished results). OPG IN CANCERS AND OTHER DISEASES {#s10} ================================= OPG has been reported to have connection with Prostate cancer, which is one of the malignancies that have great avidity to bone as advanced prostate cancer commonly metastasizes to bone leading to osteoblastic and osteolytic lesions \[[@R71]\]. Prostate cancer cells secrete OPG and *in vitro* OPG can protect the tumor cells from apoptosis *via* its ability to inhibit TRAIL and the apoptotic mechanisms it activates \[[@R18], [@R38], [@R40]\]. In contrast, *in vivo* OPG was shown to inhibit the survival of prostate cancer cells in bone \[[@R72], [@R73]\]. A potential role for serum OPG as a marker of early relapse in prostate cancer is suggested by the study of Eaton et al. 2004 \[[@R74]\]. Here, they compared serum OPG levels in 104 prostate cancer patients, ten cases of benign prostatic hyperplasia, and ten healthy young men \[[@R74]\]. The prostate cancer patients were divided into several groups: untreated patients with (i) organ confined or (ii) locally advanced disease, (iii) patients with advanced disease responding to androgen ablation and (iv) patients with early signs of disease progression. Serum OPG was found to increase in patients who progressed following androgen ablation, and this increase was detectable prior to elevation of the classical marker prostate specific antigen (PSA). The authors \[[@R74]\] suggest that serum OPG may not simply be a marker of advanced disease, but indicate changes in tumor cell survival and growth. The proinflammatory cytokine TNF-α has been demonstrated to drive OPG production in a variety of cell types and prostate cancers \[[@R75]\]. One of the major clinical features of multiple myeloma is the development of osteolytic bone disease and, over recent years, there is increasing evidence to suggest that the dysregulation of the RANK/RANKL/OPG system is important in the pathogenesis of myeloma bone disease. One study \[[@R76]\] has shown a subset of melanomas constitutively produce soluble OPG. OPG expression is regulated by the presence or absence of TNFR1 on melanomas, and melanomas drive OPG production in the absence of accessory cells through autocrine signaling *via* surface expression of mTNF. OPG levels are dramatically elevated in some colon carcinoma cells and that its secretion is potently upregulated by inflammatory cytokines such as TNF-α and IL-1 \[[@R77]\]. Increased serum OPG levels have been reported in patients with colorectal cancer (CRC)\[[@R41]\]. Interestingly, overexpression of OPG at the invasive tumor front plays a critical role in the initiation of progression and metastasis of CRC and high OPG level is a novel marker for recurrence after curative surgery for CRC \[[@R78]\]. The study also suggests OPG\'s potential to be used as a reliable biomarker to decide whether to use adjuvant chemotherapy in patients with CRC after surgery \[[@R78]\]. Ito et al. \[[@R79]\] examined the expression of OPG by gastric carcinoma cell lines and material from 103 cases of primary gastric carcinomas by gene expression analysis and immunohistochemistry, and related OPG expression to clinicopathological information such as tumor stage, depth of invasion, presence of lymph node metastasis and prognosis. They report a significant correlation between OPG expression and depth of tumor invasion, nodal metastasis and tumor stage, with strong OPG expression more frequent in stages III and IV than stages I and II. Mizutani et al. 2004 \[[@R80]\] demonstrated that serum OPG concentration is correlated with both tumor stage and tumor grade and that elevated serum OPG levels are predictive of early recurrence in patients with bladder carcinoma. These findings suggest that serum OPG concentration may have utility as a prognostic parameter in this setting. OPG overexpression has also been related to poor prognosis for pancreatic cancer (PaC) and is a key modulator of metastasis and resistance to TRAIL-induced apoptosis \[[@R81]\]. Shi et al. 2014 showed increased OPG expression in PaC tissues compared with normal pancreas, and in PaC tumors \[[@R81]\]. OPG overexpression was associated with new-onset PaC-diabetes mellitus (PaC-DM) \[[@R81]\]. Patients with new-onset PaC-DM had a higher serum OPG level than those without diabetes mellitus \[[@R81]\]. These findings suggest that in at least a portion of PaC tissues, islet cells would face high OPG stress, and thus be more susceptible to damage than normal human islets. Taken together these studies also provide the rationale to consider OPG as a diabetogenic factor and most importantly as a candidate target for new approaches to retard the development of PaC-DM \[[@R81]\]. Toffoli et al. 2011 \[[@R82]\] found that OPG induces morphological alterations and reduction of islet function in mouse pancreatic islets. Serum OPG level may be used as a diagnostic tool and a prognostic variable for patients with muscle invasive bladder cancer. Future trials are required to elucidate its therapeutic role in such patients \[[@R83]\]. OPG is also involved in many other diseases unrelated to cancer. OPG was recently defined as an important cardiovascular (CV) marker in the general population with arterial stiffness, vascular calcification, and carotid intima media thickness \[[@R84]\]. OPG constitutes a novel biomarker with prognostic significance in patients with severe malaria. In addition, further studies are required to determine whether OPG plays a role in modulating malaria pathogenesis \[[@R85]\]. Circulating OPG levels are increased in patients with acute coronary syndrome \[[@R86]\] and enhanced expression has been found within symptomatic carotid plaques \[[@R87]\]. Elevated OPG levels have also been associated with the degree of coronary calcification in the general population as a marker of coronary atherosclerosis \[[@R88]\]. OPG levels were higher in obese than in normal subjects and HDL-C was also associated with OPG levels in obese women \[[@R89]\]. OPG has been reported to predict survival in patients with heart failure after acute myocardial infarction \[[@R90]\] to predict heart failure hospitalization and mortality in patients with acute coronary syndrome \[[@R86]\] and to be associated with long-term mortality in patients with ischemic stroke \[[@R91]\]. OPG AND GENETIC POLYMORPHISM {#s11} ============================ The occurrence of single nucleotide polymorphisms (SNPs) in the OPG gene in association with breast cancer has been examined. Ney et al. 2013 \[[@R92]\] for the first time reported a significant association between the SNP rs3102735 (5′ near promoter region containing the minor allele C as well as for the homo- and heterozygous genotype with the minor allele C) of the OPG gene and the susceptibility of breast cancer in Caucasian populations. A 1.5-fold increased risk of breast cancer was associated with SNP rs3102735 \[[@R92]\]. Another study reported that the C allele of the OPG SNP rs2073618 and the T allele of the OPG SNP rs2073617 occurred more frequently in breast cancer patients \[[@R93]\]. The variant C allele of 950 T/C in the OPG promoter has been reported to play a major role as a genetic safe guard against progression in patients with prostate cancer \[[@R94]\]. The analysis of OPG gene variants C950T (promoter) and C1181G (exon 1) revealed that the presence of polymorphic 1181G/950T alleles and 950 TT/1181 GG genotypes may play a role in the development of bone disease \[[@R95]\]. Apart from breast cancer, genetic variations in form of polymorphisms in OPG and RANKL have also been associated with bone fractures in premenopausal patients with systemic lupus erythematosis (SLE) \[[@R96]\]. OPG/A163G polymorphism has been suggested to contribute to the genetic regulation of bone mineral density or bone turnover markers in Slovak population and thus could increase or decrease osteoporosis risk \[[@R97]\]. 163A/G (rs3102735) and 950T/C (rs2073617) polymorphisms of OPG have been evaluated in patients with pre-eclampsia (60 cases of early-onset severe pre-eclampsia and 91 cases of late-onset pre-eclampsia) \[[@R98]\]. Kwan et al., 2014 used a meta-analysis of GWAS from FIVE studies comprising \> 10 000 individuals from European and Asian origin, and identified two genome-wide significant loci (8q23-q24.1 and 17q11.2) and one locus on chromosome 14 associated with OPG levels with near genome-wide significance \[[@R99]\]. In conclusion, they discovered that variants \> 100 kb upstream of the gene encoding OPG are associated with variation in circulating OPG levels and identified another new significant locus on chromosome 17q11.2 as well as a suggestive locus on chromosome 14q21.2 associated with the trait \[[@R99]\]. They estimated that over half of the heritability of age-adjusted OPG levels could be explained by all SNPs studied \[[@R99]\]. PERSPECTIVES {#s12} ============ Despite the wealth of literature on OPG, there are many questions still unresolved, including the exact role of OPG in bone metastasis of breast cancer and the therapeutic potential of targeting OPG. Besides roles in cancer, OPG plays role(s) in bone metabolism, endometriosis, periodontal disease, thyroid disease and coronary heart disease \[[@R11]\]. Though OPG can modulate breast tumor growth and progression, future studies required to fully determine the mechanism of effect and overall outcomes of the different types of interactions required to determine whether strategies to block OPG signaling would be effective in blocking the development of primary breast tumors \[[@R43]\]. At this point OPG can no longer be considered solely in the bone microenvironment in breast cancer and caution must be exercised in the development of systemic treatment strategies aimed at increasing OPG levels. Localized delivery of OPG to the bone may be more appropriate to inhibit osteolysis linked metastatic breast cancer \[[@R43]\]. Conditionally replicating adenoviruses (CRAds) have been used as anticancer agents designed to infect and lyse tumor cells. In a murine model of osteolytic bone metastases of breast cancer, the CRAd armed with shortened OPG (sOPG)-Fc reduced tumor burden in the bone and inhibited osteoclast formation more effectively than an unarmed CRAd suggesting the role of OPG \[[@R17]\]. Breast cancer development in BRCA1/2 mutation carriers is a consequence of autonomous and nonautonomous cell factors, which serve as excellent targets for cancer prevention. BRCA-mutation carriers were reported to have lower mean values of free serum OPG in particular in BRCA1-mutation carriers compared with controls \[[@R100]\]. OPG may become a new biomarker and a target of treatment for patients with colorectal cancer since many studies have revealed the clinicopathologic significance of OPG expression by using clinical tissue samples from patients \[[@R78]\]. The effects of sustained expression of OPG using a recombinant adeno-associated viral (rAAV) vector in a mouse model of osteolytic breast cancer has clearly indicated the potential of rAAV-OPG therapy for reducing morbidity and mortality in breast cancer patients with osteolytic bone damage \[[@R101]\]. With all this development in the understanding of OPG, there might be more therapeutic avenues to manipulate OPG to predict and manage aggressive forms of breast cancer. This study was supported by Rosalind Franklin University of Medicine and Science start-up funds and RFUMS-Advocate Lutheran General Hospital grant to NSW. We thank Keith Philibert for critically reading the manuscript. **CONFLICTS OF INTEREST** No potential conflicts of interest exist. TME : tumor microenvironment ECM : extracellular matrix OPG : osteoprotegerin TNFR : tumor necrosis factor receptor TRAIL : TNF related apoptosis inducing ligand IBC : Inflammatory breast cancer RANKL : receptor activator of nuclear factor (NFΚB) ligand OCIF : osteoclastogenesis inhibitory factor FDCR-1 : follicular DC-derived receptor-1 MSCs : mesenchymal stem cells TAFs : tumor associated fibroblasts PDGF : platelet derived growth factor IL-1 : interleukin-1 IL-6 : interleukin-6 IL-8 : interleukin-8 CCL-5 : Chemokine (C-C motif) ligand 5 TNF-α : tumor necrosis factor- α TGF-β : transforming growth factor-β PTH : parathyroid hormone PSA : prostate specific antigen HSPGs : heparin sulfate proteoglycans HuDMECs : Human dermal microvascular endothelial cells FAK : Focal Adhesion Kinase VEGF : vascular endothelial growth factor PaC : pancreatic cancer Pa-DM : PaC-diabetes mellitus FASN : Fatty acid synthase IQGAP1/3 : IQ Motif Containing GTPase Activating Protein 1/3 CRAds : Conditionally replicating adenoviruses CRC : colorectal cancer ECFC : endothelial colony forming cells AURK1/IAK1 : Aurora A Kinase EGFR : epidermal growth factor receptor MYC and PAK1 CDK4 : Cyclin-Dependent Kinase 4 CDKN2A : Cyclin-Dependent Kinase Inhibitor 2A PTEN : Phosphatase and tensin homolog TOP2A : topoisomerase (DNA) II Alpha