Browse the corpus

Sjögren's syndrome (SS) is a chronic multisystem autoimmune disease with potential to cause substantial morbidity 1, 2. Primary SS is characterized by progressive destruction of the exocrine glands, with subsequent mucosal and conjunctival dryness 1. Although the precise cause of SS remains unknown, it is understood to be a complex and heterogeneous disease, with important contributions from both genetic and environmental factors 3, 4. It is likely that widespread clinical heterogeneity in SS reflects differences in underlying disease mechanisms, and current approaches to research and clinical care in SS are compromised by such phenotypic heterogeneity. Elucidation of how genetic and nongenetic factors contribute to disease heterogeneity should significantly affect how we diagnose, manage, and treat this complex disorder.

ences in underlying disease mechanisms, and current approaches to research and clinical care in SS are compromised by such phenotypic heterogeneity. Elucidation of how genetic and nongenetic factors contribute to disease heterogeneity should significantly affect how we diagnose, manage, and treat this complex disorder. A growing body of evidence has implicated epigenetic factors, in particular, altered patterns of DNA methylation, in models of autoimmune disease 5, 6. Furthermore, recent studies characterizing the DNA methylation profiles of naive CD4+ T cells, B cells, and salivary gland epithelial cells provide evidence of aberrant DNA methylation profiles in SS patients 7, 8, 9, 10. While it is unknown which differences, if any, reflect causal determinants of risk, it is likely that many of these patterns reflect subtle differences in subpopulation composition 11 downstream of true causal risk factors or disease processes. One of the most important compartments for analyzing immunoregulatory heterogeneity in SS is whole labial salivary glands (LSGs), a prominent target of the disease‐specific processes.

f these patterns reflect subtle differences in subpopulation composition 11 downstream of true causal risk factors or disease processes. One of the most important compartments for analyzing immunoregulatory heterogeneity in SS is whole labial salivary glands (LSGs), a prominent target of the disease‐specific processes. We report our findings of a genome‐wide study of DNA methylation in LSG tissue biopsied from 13 SS cases, 13 controls, and 2 subjects with intermediate phenotypes; all are women of genetically confirmed European descent who are participants in the Sjögren's International Collaborative Clinical Alliance (SICCA) Registry. We identified thousands of DNA methylation differences across the genome associated with case status, implicating immune‐related and cell lineage–specific pathways in disease pathogenesis. In addition to highlighting a large number of genes involved in general immune system processes (including known genetic risk loci associated with SS), we also observed enrichment for DNA methylation differences around specific transcription factor motifs. In total, our results demonstrate both widespread and targeted DNA methylation differences marking LSG‐specific immune processes in SS.

immune system processes (including known genetic risk loci associated with SS), we also observed enrichment for DNA methylation differences around specific transcription factor motifs. In total, our results demonstrate both widespread and targeted DNA methylation differences marking LSG‐specific immune processes in SS. SUBJECTS AND METHODS Study subjects and clinical evaluation Our study used samples of LSG tissue biopsied from 28 female subjects of European descent who were participants in the SICCA Registry (Table 1). As part of the enrollment into the SICCA Registry, subjects were evaluated for clinical criteria of SS at 1 or 2 time points; LSG tissue was biopsied during at least 1 of these visits, frozen, and stored using standard procedures. Table 1 Phenotype and covariates across study groupsa Cases (n = 13) Controls (n = 13) Noncases (n = 15) P (cases vs. controls) Focus score 3.4 ± 2 0.07 ± 0.13 0.09 ± 0.17 9.1 × 10−6 Ocular staining score 6.1 ± 2.8 1.2 ± 0.7 1.7 ± 1.9 1.5 × 10−5

SUBJECTS AND METHODS Study subjects and clinical evaluation Our study used samples of LSG tissue biopsied from 28 female subjects of European descent who were participants in the SICCA Registry (Table 1). As part of the enrollment into the SICCA Registry, subjects were evaluated for clinical criteria of SS at 1 or 2 time points; LSG tissue was biopsied during at least 1 of these visits, frozen, and stored using standard procedures. Table 1 Phenotype and covariates across study groupsa Cases (n = 13) Controls (n = 13) Noncases (n = 15) P (cases vs. controls) Focus score 3.4 ± 2 0.07 ± 0.13 0.09 ± 0.17 9.1 × 10−6 Ocular staining score 6.1 ± 2.8 1.2 ± 0.7 1.7 ± 1.9 1.5 × 10−5 SSA seropositive (indicator) 0.92 0 0 – SSB seropositive (indicator) 0.54 0 0 – Age 55 ± 13 53 ± 7.9 56 ± 10 0.84 Ancestry PC1 0.005 ± 0.003 −0.014 ± 0.026 −0.012 ± 0.024 0.035 a Noncases are a combined set of subjects with intermediate phenotype (n = 2) and controls (n = 13), who were included to emphasize the contrast in phenotype between the cases and the other subjects. Values are the mean ± SD of covariates, except for SSA/SSB seropositivity data, which are shown as indicator variables (1 = positive), and only the mean values (proportions) are reported for these phenotypes. P values were determined by Wilcoxon's rank sum test. PC1 = first principal component.

and the other subjects. Values are the mean ± SD of covariates, except for SSA/SSB seropositivity data, which are shown as indicator variables (1 = positive), and only the mean values (proportions) are reported for these phenotypes. P values were determined by Wilcoxon's rank sum test. PC1 = first principal component. Case–control status was evaluated according to the American College of Rheumatology (ACR) criteria for SS 12. Our study targeted cases with severe SS, requiring that cases meet all 3 of the following criteria: seropositivity (SSA and/or SSB autoantibodies), an ocular staining score (OSS) of ≥3 in at least 1 eye, and a focus score ≥1 (no subjects had a focus score of 1). Controls did not meet any of these criteria. Samples were designated as case or control based on clinical evaluation at the time of biopsy. Two of the study subjects met only the high OSS criterion at time of sample collection (Table 1) and are referred to herein as “intermediate phenotype” subjects. Neither cases nor controls were disqualified based on an additional systemic autoimmune disease diagnose (e.g., rheumatoid arthritis, Hashimoto's disease). Self‐reported medication data for the study participants are shown in Supplementary Table 1 (available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39792/abstract). Univariate testing to compare variable distributions in cases and controls was conducted using Fisher's exact test implemented in R. The Institutional Review Boards at the University of California, San Francisco and the University of California, Berkeley approved our study protocol.

com/doi/10.1002/art.39792/abstract). Univariate testing to compare variable distributions in cases and controls was conducted using Fisher's exact test implemented in R. The Institutional Review Boards at the University of California, San Francisco and the University of California, Berkeley approved our study protocol. Genotyping and principal components (PCs) analysis Prior to this study, the 28 SICCA subjects were genotyped using the HumanOmni2.5‐Quad BeadChip array (Illumina), as part of a genome‐wide association study (GWAS) 13. In addition to sample verification and other quality control assessments, these data were used to evaluate the genetic ancestry of the study subjects. EigenStrat analysis 14 was applied to genotypes from the full GWAS dataset in order to derive PCs reflecting global genetic variation. The 28 study subjects fell within 2 SD of the mean of the first 2 PCs in self‐identified Europeans; GWAS subjects within this range were deemed “European candidates.”

he study subjects. EigenStrat analysis 14 was applied to genotypes from the full GWAS dataset in order to derive PCs reflecting global genetic variation. The 28 study subjects fell within 2 SD of the mean of the first 2 PCs in self‐identified Europeans; GWAS subjects within this range were deemed “European candidates.” In order to examine the effects of intra‐European ancestry on LSG DNA methylation, we applied EigenStrat analysis to genotypes from all European candidates. The first 4 PCs were retained for downstream analysis. We saw no significant evidence of association between case status and age in our study population, and only weak association with the first ancestry PC (Table 1). Given the small study size, we chose not to adjust for these factors when comparing DNA methylation patterns between cases and controls, but rather to screen disease‐associated DNA methylation differences for any effects of ancestry PC1 and age.

opulation, and only weak association with the first ancestry PC (Table 1). Given the small study size, we chose not to adjust for these factors when comparing DNA methylation patterns between cases and controls, but rather to screen disease‐associated DNA methylation differences for any effects of ancestry PC1 and age. DNA methylotyping DNA methylation data were obtained for each sample using the Illumina 450K Infinium Methylation BeadChip (450K chip) platform. The 450K chip allows for high‐throughput interrogation of more than 450,000 highly informative CpG sites spanning ∼22,000 genes across the genome. The primary measure of DNA methylation at each CpG site is β, which is the ratio of the intensities of fluorescent signals from methylated and unmethylated alleles. Sample identity was verified by comparing genome‐wide genotypes to the genotypes derived from 35 single nucleotide polymorphism (SNP) probes on the 450K chip. Three of the DNA samples were subdivided into 2 intrabatch technical replicates, contributing to a total of 31 samples for subsequent DNA methylation analysis. Data normalization and filtering Our data preprocessing pipeline was implemented entirely in R 15 and used the methylumi data representation in Bioconductor 16, 17. We applied the normal‐exponential convolution method on out‐of‐band probe intensities (“noob”) to correct each sample for technical variation in background fluorescence 18. The 2 color channels on the 450K chip were normalized using the all‐sample mean normalization method, which is a natural extension of the Illumina GenomeStudio protocol 19.

onential convolution method on out‐of‐band probe intensities (“noob”) to correct each sample for technical variation in background fluorescence 18. The 2 color channels on the 450K chip were normalized using the all‐sample mean normalization method, which is a natural extension of the Illumina GenomeStudio protocol 19. The 450K chip includes 3,091 CpH (non‐CpG) probes and 65 SNP probes, all of which were removed prior to analysis. We also removed 16,177 cross‐reactive CpG probes 20. In order to avoid direct effects of genotype variation, we removed 1,213 CpG probes targeting variable SNPs genotyped in our study sample. We also considered the set of SNPs from the 1000 Genomes project that lie within the probe‐hybridizing sequence as tabulated by Chen et al 20. Using the University of California, Santa Cruz (UCSC) Genome Browser SNP138 track 21, we identified and removed 62,220 CpG probes neighboring SNPs. An additional 3,392 CpG probes were removed from the analysis due to high‐detection P values (P > 0.05) in 1 or more samples, as computed by Illumina's GenomeStudio software. A total of 404,353 CpG probes were therefore used for the primary analysis. After probe filtering, we corrected each sample for type I/II probe design bias using the beta‐mixture quantile normalization method 22.

o high‐detection P values (P > 0.05) in 1 or more samples, as computed by Illumina's GenomeStudio software. A total of 404,353 CpG probes were therefore used for the primary analysis. After probe filtering, we corrected each sample for type I/II probe design bias using the beta‐mixture quantile normalization method 22. Principal components analysis of DNA methylation data We computed PCs of the normalized β value matrix, centering and scaling per CpG. After averaging the PC values of replicates, the top 5 PCs were tested for association with several continuous covariates (focus score, mean OSS, age, and genetic ancestry PCs) and categorical covariates (SSA/SSB seropositivity and assay plate), applying Z tests (to Fisher‐transformed Spearman's rho) and Kruskal Wallis tests, respectively. Nonlinear adjustment for technical variation Despite our efforts to normalize the data using standard methods, the first PC of the β‐matrix clearly separated samples according to the assay plate (Supplementary Figures 1A and B, available at http://onlinelibrary.wiley.com/doi/10.1002/art.39792/abstract). We adjusted the data against proxies of known batch effects to remove technical bias (see Supplementary materials, available at http://onlinelibrary.wiley.com/doi/10.1002/art.39792/abstract).

to the assay plate (Supplementary Figures 1A and B, available at http://onlinelibrary.wiley.com/doi/10.1002/art.39792/abstract). We adjusted the data against proxies of known batch effects to remove technical bias (see Supplementary materials, available at http://onlinelibrary.wiley.com/doi/10.1002/art.39792/abstract). Single CpG site tests for differentially methylated positions and global DNA methylation analysis Wilcoxon's rank sum test was used to test each CpG β value for association with case–control status, followed by the Benjamini‐Yekutieli adjustment for multiple comparisons. The Benjamini‐Yekutieli adjustment is a more conservative version of the Benjamini‐Hochberg false discovery rate (FDR) procedure, which may be preferable when test statistics are correlated 23. Given the complex and often strong correlations between CpG methylation levels, we chose to use this more conservative FDR procedure. No thresholds were placed on mean or median β‐differences between cases and controls—specifically, we set no constraints on the magnitude of significant differences in methylation. We refer to disease‐associated CpGs (q < 0.01) as differentially methylated positions (DMPs). The β values for replicate samples were averaged prior to single CpG–site association tests.

ifferences between cases and controls—specifically, we set no constraints on the magnitude of significant differences in methylation. We refer to disease‐associated CpGs (q < 0.01) as differentially methylated positions (DMPs). The β values for replicate samples were averaged prior to single CpG–site association tests. Global DNA methylation was evaluated using 2 methods. First, Wilcoxon's rank sum tests were applied to evaluate differences in mean genome DNA methylation status between cases and controls; for each subject, mean genome DNA methylation is defined as the mean β value across all probes passing our stringent quality filtering. Second, we applied Fisher's exact test to determine whether the fraction of hypermethylated CpGs (as determined by the sign of the mean difference between cases and controls) varied significantly between DMPs and non‐DMPs. Identification of differentially methylated promoters CpGs were mapped to promoters (or, more generally, upstream regulatory regions) using the BEDTools suite 24. For each RefSeq entry in the UCSC RefGene track 21, we defined a promoter region as the genomic interval spanning 2,500 bp upstream and 500 bp downstream of the annotated transcription start site, similar to the definition described by Whitaker et al 25. RefSeq identifiers were mapped to gene symbols using the org.Hs.eg.db package in Bioconductor 26; all unmapped RefSeq entries were excluded from the analysis.

ic interval spanning 2,500 bp upstream and 500 bp downstream of the annotated transcription start site, similar to the definition described by Whitaker et al 25. RefSeq identifiers were mapped to gene symbols using the org.Hs.eg.db package in Bioconductor 26; all unmapped RefSeq entries were excluded from the analysis. We tested for differentially methylated promoters using hypergeometric tests for DMP enrichment, as described by Nakano et al 27. Enrichment P values were adjusted for multiple testing using the Benjamini‐Hochberg correction, with a q value threshold of 0.05. To avoid promoter‐specific bias, we excluded all CpGs that did not fall within promoters; enrichment tests were performed solely on promoter CpGs. Furthermore, to protect against biases associated with double‐counting CpGs sitting in the intersection of multiple loci, we excluded any CpGs mapping to 2 or more promoters. Gene set enrichment analysis After identifying the set of genes with significantly differentially methylated promoters, we considered whether this gene set is enriched for categories of biologic function or genomic position. Hypergeometric gene set enrichment analysis was used to test 2,666 gene sets from the Molecular Signature Database 28 for enrichment of differentially methylated promoters, including “hallmark” gene sets, positional gene sets, motif gene sets, and gene ontology gene sets, with a Benjamini‐Hochberg q value cutoff of 0.05.

pergeometric gene set enrichment analysis was used to test 2,666 gene sets from the Molecular Signature Database 28 for enrichment of differentially methylated promoters, including “hallmark” gene sets, positional gene sets, motif gene sets, and gene ontology gene sets, with a Benjamini‐Hochberg q value cutoff of 0.05. We further tested 2 candidate gene sets for enrichment of genes possessing differentially methylated promoters: 1) genes encoding the 50 transcripts showing the greatest fold‐change in LSG expression between SS cases and controls in the microarray study by Hjelmervik et al 29, and 2) genes highlighted in recent SS GWAS: GTF2I, TNFAIP3, IRF5, STAT4, IL12A, BLK, CXCR5, TNIP1, HLA–DRA, HLA–DQB1, HLA–DRB1, HLA–DPB1, and COL11A2 30, 31.

genes encoding the 50 transcripts showing the greatest fold‐change in LSG expression between SS cases and controls in the microarray study by Hjelmervik et al 29, and 2) genes highlighted in recent SS GWAS: GTF2I, TNFAIP3, IRF5, STAT4, IL12A, BLK, CXCR5, TNIP1, HLA–DRA, HLA–DQB1, HLA–DRB1, HLA–DPB1, and COL11A2 30, 31. CpG set enrichment analysis Although gene set enrichment analysis is a valuable tool for understanding the distribution of differentially methylated promoters, the DMPs on which this analysis is based are called at single‐basepair resolution; therefore, some information is lost when the analysis is applied to broad genomic regions such as promoters. This discrepancy can even lead to bias due to the variation in promoter coverage across the 450K chip platform; some promoters contain far more probed CpGs than others, giving us greater power to resolve extended differences in those regions. Some of this bias of differential power can be avoided by considering CpG sets rather than gene sets. For each of the differentially methylated gene sets identified in the gene set enrichment analysis, as well as the 2 candidate gene sets, a CpG set was also defined, containing all of the CpGs mapping to promoters of the corresponding gene set. DMP enrichment was performed using hypergeometric tests, as before, although CpGs mapping to multiple sets were included in this analysis. The CpG set enrichment analysis was adjusted for multiple testing, accounting for the 2,668 gene set enrichment tests used to select CpG sets. CpG sets with a Bonferroni‐adjusted P value less than 0.01 were considered enriched for DMPs.

, as before, although CpGs mapping to multiple sets were included in this analysis. The CpG set enrichment analysis was adjusted for multiple testing, accounting for the 2,668 gene set enrichment tests used to select CpG sets. CpG sets with a Bonferroni‐adjusted P value less than 0.01 were considered enriched for DMPs. Transcription factor motif enrichment analysis Given the intimate relationship between transcription‐factor binding and chromatin state, we considered whether disease‐associated DNA methylation changes colocalize with specific transcription factor binding motifs (TFBMs), using the Analysis of Motif Enrichment (AME) tool 32 to identify enriched TFBMs in the sequence surrounding disease‐associated DMPs. For each DMP, we extracted a window of the UCSC hg19 reference genome within 150 bp of the annotated CpG position. Overlapping intervals were merged, producing a set of DMP‐associated sequences. A “control” set of CpG‐neighboring sequences was generated using the same procedure applied to all non‐DMPs passing our quality filter.

MP, we extracted a window of the UCSC hg19 reference genome within 150 bp of the annotated CpG position. Overlapping intervals were merged, producing a set of DMP‐associated sequences. A “control” set of CpG‐neighboring sequences was generated using the same procedure applied to all non‐DMPs passing our quality filter. Using the AME, we tested DMP‐associated sequences for enrichment of 205 TFBMs from the JASPAR CORE 2014 vertebrates set 33, adjusting for sequence length and using the “control” set as a sequence control. AME was performed using 3 motif affinity options that use different scoring methods to evaluate motif matches: total number of matches above a threshold (“totalhits”), sum of motif scores (“sum”), and average motif score (“avg”). Default thresholds were used for all choices of motif affinity function, and observed enrichment was evaluated for statistical significance using Fisher's exact tests. Motifs were considered enriched if the corresponding Bonferroni‐adjusted P value fell below 0.01 (correcting for 615 tests) for any of the 3 affinity functions.

thresholds were used for all choices of motif affinity function, and observed enrichment was evaluated for statistical significance using Fisher's exact tests. Motifs were considered enriched if the corresponding Bonferroni‐adjusted P value fell below 0.01 (correcting for 615 tests) for any of the 3 affinity functions. RESULTS Different global methylation patterns in LSGs from SS cases and controls After adjusting for technical effects, none of the top 5 DNA methylation PCs (60% of variance) showed significant association with the sample batch (Supplementary Figures 1C and D). The first PC was strongly associated with the focus score (q = 2.1 × 10−5) and the mean OSS (q = 5.3 × 10−4) (Supplementary Table 2, available at http://onlinelibrary.wiley.com/doi/10.1002/art.39792/abstract), suggesting that this axis captures disease‐associated processes in the gland. Indeed, results of Wilcoxon's rank sum testing showed that the first PC of DNA methylation in LSG tissue was associated with case status (P = 1.3 × 10−5). Plots of the first 2 PCs place the 2 individuals of intermediate phenotype between the cases and the controls, consistent with their phenotype (Figure 1). Tests of association between case status and self‐reported medication use (Supplementary Table 1) showed that a confounding effect of medication was unlikely in the current study.

irst 2 PCs place the 2 individuals of intermediate phenotype between the cases and the controls, consistent with their phenotype (Figure 1). Tests of association between case status and self‐reported medication use (Supplementary Table 1) showed that a confounding effect of medication was unlikely in the current study. Figure 1 Principal components analysis of genome‐wide DNA methylation in all labial salivary gland tissue samples, including replicates. The first principal component (PC1) separates Sjögren's syndrome cases from controls, with samples from the 2 subjects with intermediate phenotype (ocular staining score ≥3 in at least 1 eye) between those of the cases and the controls. PC2 represents a spread of sample DNA methylation states orthogonal to the primary case–control contrast. This axis may represent biologic intrapatient heterogeneity. All study subjects were women of genetically confirmed European descent who were participants in the Sjögren's International Collaborative Clinical Alliance Registry. Global hypomethylation of LSGs in SS cases Thabet et al 8 previously reported global hypomethylation in cultured LSG epithelial cells from SS patients. We considered whether these differences could be detected in more heterogeneous LSG tissue samples. However, no significant differences in mean genome DNA methylation were observed across all CpGs (1.01‐fold hypermethylation in SS cases; P = 0.26).

obal hypomethylation in cultured LSG epithelial cells from SS patients. We considered whether these differences could be detected in more heterogeneous LSG tissue samples. However, no significant differences in mean genome DNA methylation were observed across all CpGs (1.01‐fold hypermethylation in SS cases; P = 0.26). Our epigenome‐wide association study identified 7,820 DMPs associated with SS case status. The median absolute β‐difference between cases and controls was 0.10 for DMPs, demonstrating that most SS‐associated DMPs identified in the current study showed modest‐to‐large differences in DNA methylation. Of the 7,820 DMPs tested, 5,699 (73%) were hypomethylated in cases. The set of DMPs contained far more hypomethylated CpGs than was expected by the distribution of non‐DMPs (P < 2.2 × 10−16 by Fisher's exact test) (Figure 2), suggesting that CpGs are generally more hypomethylated in whole LSG tissue from SS cases. Of the 7,820 DMPs tested, 338 (4%) were associated at P = 1.92 × 10−7 (q = 0.003). These top sites distinguished cases from controls in our study sample.

istribution of non‐DMPs (P < 2.2 × 10−16 by Fisher's exact test) (Figure 2), suggesting that CpGs are generally more hypomethylated in whole LSG tissue from SS cases. Of the 7,820 DMPs tested, 338 (4%) were associated at P = 1.92 × 10−7 (q = 0.003). These top sites distinguished cases from controls in our study sample. Figure 2 Global DNA methylation differences between labial salivary glands from the Sjögren's syndrome cases and the controls. A, Proportions of hyper‐ and hypomethylated CpGs with q values >0.01 by Wilcoxon's rank sum test. These represent a control set of CpGs, or non–differentially methylated positions (DMPs). Direction of methylation is determined by the sign of the difference in the mean β values between cases and controls. B, Proportions of hyper‐ and hypomethylated DMPs with q values <0.01 by Wilcoxon's rank sum test. DMPs are significantly enriched for hypomethylated sites as compared to non‐DMPs (P  < 2.2 × 10−16 by Fisher's exact test). Linear regression was used to model the associations between the DNA methylation level (logit transformed) for each of the 7,820 DMPs and the first PC of genetic ancestry or age at biopsy. No DMP was significantly associated with either factor at a Benjamini‐Hochberg FDR of 0.05. These 2 factors may affect DNA methylation levels of SS‐associated DMPs, but their average effects are too small to resolve in our study.

sformed) for each of the 7,820 DMPs and the first PC of genetic ancestry or age at biopsy. No DMP was significantly associated with either factor at a Benjamini‐Hochberg FDR of 0.05. These 2 factors may affect DNA methylation levels of SS‐associated DMPs, but their average effects are too small to resolve in our study. Differentially methylated promoters of various protein‐coding genes, microRNAs, and noncoding RNAs Differentially methylated promoter analysis identified 57 genes (Table 2 and Supplementary Table 3, available at http://onlinelibrary.wiley.com/doi/10.1002/art.39792/abstract). This list includes a large number of transcription factors (e.g., RUNX3 and SPI1) and known cell‐differentiation markers (e.g., TNFRSF13B, CCR6, BST2, BTLA, and CXCR5). In addition to coding genes, the list contains a number of RNA genes, including several antisense RNA genes (e.g., PSMB8‐AS1) and microRNAs (e.g., MIR339). The results could reflect differential regulation of neighboring coding genes or primary transcripts. Interestingly, 3 of the differentially methylated promoters are located within 1 interval of the major histocompatibility complex (MHC) genomic region: PSMB8, PSMB8‐AS1, and TAP1. Table 2 Top 25 DMPs in labial salivary glands from Sjögren's syndrome patientsa Upstream region DMP range Total DMPs Fold enrichment q value for enrichment % DMPs hypo. PSMB8‐AS1 Chr. 6: 32810001–32811253 11 38.3 1.4 × 10−11 100 CTSZ Chr. 20: 57582706–57583474 10 26.5 1.9 × 10−8 100 PTPRCAP Chr. 11: 67205096–67206434 8 35.3 1.2 × 10−7 100 LTA Chr. 6: 31539539–31540440 7 38.6 7.6 × 10−7 100

Table 2 Top 25 DMPs in labial salivary glands from Sjögren's syndrome patientsa Upstream region DMP range Total DMPs Fold enrichment q value for enrichment % DMPs hypo. PSMB8‐AS1 Chr. 6: 32810001–32811253 11 38.3 1.4 × 10−11 100 CTSZ Chr. 20: 57582706–57583474 10 26.5 1.9 × 10−8 100 PTPRCAP Chr. 11: 67205096–67206434 8 35.3 1.2 × 10−7 100 LTA Chr. 6: 31539539–31540440 7 38.6 7.6 × 10−7 100 MIR339 Chr. 7: 1062652–1064100 7 30.9 4.8 × 10−6 100 TNFRSF13B Chr. 17: 16875129–16875596 5 55.1 1.8 × 10−5 100 PSMB8 Chr. 6: 32813084–32815091 7 22 5.7 × 10−5 100 MTNR1A Chr. 4: 187476543–187476608 5 33.1 0.00053 0 MPEG1 Chr. 11: 58980157–58981095 4 52.9 0.00066 100 CCR6 Chr. 6: 167535909–167536184 5 27.6 0.0013 100 TAP1 Chr. 6: 32822565–32823941 4 44.1 0.0016 100 SSH3 Chr. 11: 67070233–67070967 5 23.6 0.0027 0 BST2 Chr. 19: 17516282–17518018 4 37.8 0.0029 100 PPFIA4 Chr. 1: 203019107–203020617 4 37.8 0.0029 100 AIM2 Chr. 1: 159046937–159047163 3 66.1 0.0036 100 BTLA Chr. 3: 112217973–112218761 3 66.1 0.0036 100 CXCR5 Chr. 11: 118754280–118763863 5 20.7 0.0036 100 FCRL3 Chr. 1: 157670328–157670869 4 33.1 0.0036 100 KCNQ1DN Chr. 11: 2890394–2890725 7 10.8 0.0036 0 LINC00926 Chr. 15: 57592040–57592438 3 66.1 0.0036 100 MIR3186 Chr. 17: 79419796–79420279 4 33.1 0.0036 100 MIR4269 Chr. 2: 240225062–240226201 3 66.1 0.0036 100 WDFY4 Chr. 10: 49892741–49893463 5 20.7 0.0036 100 RUNX3 Chr. 1: 25291472–25292225 7 10.1 0.0055 100

FCRL3 Chr. 1: 157670328–157670869 4 33.1 0.0036 100 KCNQ1DN Chr. 11: 2890394–2890725 7 10.8 0.0036 0 LINC00926 Chr. 15: 57592040–57592438 3 66.1 0.0036 100 MIR3186 Chr. 17: 79419796–79420279 4 33.1 0.0036 100 MIR4269 Chr. 2: 240225062–240226201 3 66.1 0.0036 100 WDFY4 Chr. 10: 49892741–49893463 5 20.7 0.0036 100 RUNX3 Chr. 1: 25291472–25292225 7 10.1 0.0055 100 FERMT3 Chr. 11: 63973846–63974153 4 26.5 0.0093 100 a Promoter enrichment results are shown for the most‐significant regions. The genomic interval for each differentially methylated position (DMP) range is given, as well as the total number of DMPs and the fold enrichment for DMPs in the region. Hypergeometric enrichment q values and hypomethylated (hypo.) fractions are also reported. Chr. = chromosome.

shown for the most‐significant regions. The genomic interval for each differentially methylated position (DMP) range is given, as well as the total number of DMPs and the fold enrichment for DMPs in the region. Hypergeometric enrichment q values and hypomethylated (hypo.) fractions are also reported. Chr. = chromosome. DMP‐enriched promoters of candidate gene sets The promoter enrichment results emphasized both the inflammation and tissue specificity of the observed DNA methylation differences. The set of differentially methylated promoters was found to be enriched for several gene ontology terms involving immune response and signal transduction. We also observed evidence of enrichment of genes known to contain transcription factor binding motifs for PU.1 and Ets‐2 (mouse orthologs of targets) in their promoters (Table 3), likely representing differences in cell composition and activity resulting from SS pathogenesis. Only a small number of these genes have been highlighted by SS GWAS (CXCR5 and BLK) 30 or are known to be differentially expressed at the transcription level in SS‐affected LSG tissue (ARHGAP25) 29; however, the promoter CpG sets corresponding to both of these candidate gene sets were significantly enriched for DMPs (Bonferroni‐adjusted P = 9.2 × 10−7 and 6.0 × 10−4, respectively). Table 3 Differentially methylated CpG sets in labial salivary glands from Sjögren's syndrome patientsa MSigDB gene set Differentially methylated promoters Adjusted P Immune response (GO:0006955) CCR6, BST2, AIM2, LCP2, CD79B, MADCAM1 2.9 × 10−8

DMP‐enriched promoters of candidate gene sets The promoter enrichment results emphasized both the inflammation and tissue specificity of the observed DNA methylation differences. The set of differentially methylated promoters was found to be enriched for several gene ontology terms involving immune response and signal transduction. We also observed evidence of enrichment of genes known to contain transcription factor binding motifs for PU.1 and Ets‐2 (mouse orthologs of targets) in their promoters (Table 3), likely representing differences in cell composition and activity resulting from SS pathogenesis. Only a small number of these genes have been highlighted by SS GWAS (CXCR5 and BLK) 30 or are known to be differentially expressed at the transcription level in SS‐affected LSG tissue (ARHGAP25) 29; however, the promoter CpG sets corresponding to both of these candidate gene sets were significantly enriched for DMPs (Bonferroni‐adjusted P = 9.2 × 10−7 and 6.0 × 10−4, respectively). Table 3 Differentially methylated CpG sets in labial salivary glands from Sjögren's syndrome patientsa MSigDB gene set Differentially methylated promoters Adjusted P Immune response (GO:0006955) CCR6, BST2, AIM2, LCP2, CD79B, MADCAM1 2.9 × 10−8 Intrinsic to plasma membrane (GO:0031226) TNFRSF13B, MTNR1A, CCR6, BST2, CXCR5, NCKAP1L, CD160, CD19, CD79B, IL12RB1 1.7 × 10−7 Genes with promoters containing Ets2 motif RYTTCCTG (M14654) PTPRCAP, TNFRSF13B, KCNQ1DN, RUNX3, FERMT3, LCP1, SPI1, SLAMF1, CD19, ERG, PIK3CG 3.6 × 10−7 Immune system process (GO:0002376) CCR6, BST2, AIM2, SPI1, LCP2, CD79B, MADCAM1 1.0 × 10−6

Intrinsic to plasma membrane (GO:0031226) TNFRSF13B, MTNR1A, CCR6, BST2, CXCR5, NCKAP1L, CD160, CD19, CD79B, IL12RB1 1.7 × 10−7 Genes with promoters containing Ets2 motif RYTTCCTG (M14654) PTPRCAP, TNFRSF13B, KCNQ1DN, RUNX3, FERMT3, LCP1, SPI1, SLAMF1, CD19, ERG, PIK3CG 3.6 × 10−7 Immune system process (GO:0002376) CCR6, BST2, AIM2, SPI1, LCP2, CD79B, MADCAM1 1.0 × 10−6 Cell surface receptor–linked signal transduction (GO:0007166) TNFRSF13B, MTNR1A, CXCR5, CD160, GNB3, LCP2, CD19, IL12RB1, PIK3CG 2.9 × 10−6 Genes with promoters containing PU.1 motif WGAGGAAG (M14376) PTPRCAP, LTA, NCKAP1L, LCP2, NR1H3, PIK3CG 4.7 × 10−5 Signal transduction (GO:0007165) LTA, TNFRSF13B, MTNR1A, CCR6, BST2, CXCR5, CD160, GNB3, BLK, LCP2, CD19, ERG, IL12RB1, KALRN, MADCAM1, PIK3CG 2.8 × 10−4 a These gene sets from the Molecular Signatures Database (MSigDB) were selected as candidates for CpG enrichment because they contained a significantly high fraction of differentially methylated promoters, as shown here. Bonferroni‐adjusted P values are reported for hypergeometric CpG set enrichment tests.

Signal transduction (GO:0007165) LTA, TNFRSF13B, MTNR1A, CCR6, BST2, CXCR5, CD160, GNB3, BLK, LCP2, CD19, ERG, IL12RB1, KALRN, MADCAM1, PIK3CG 2.8 × 10−4 a These gene sets from the Molecular Signatures Database (MSigDB) were selected as candidates for CpG enrichment because they contained a significantly high fraction of differentially methylated promoters, as shown here. Bonferroni‐adjusted P values are reported for hypergeometric CpG set enrichment tests. To further probe the meaning of the observed enrichment in differentially expressed genes, we assigned hypomethylation significance scores (score = sign[Δβ] × logP) to each CpG falling within the promoters of 42 genes reported as being highly differentially expressed in the microarray study by Hjelmervik et al 29. Regression analysis revealed that the average hypomethylation score across a promoter is positively associated with the extent of messenger RNA up‐regulation reported in SS‐affected tissue (Supplementary Figure 2, available at http://onlinelibrary.wiley.com/doi/10.1002/art.39792/abstract). The predictive power of differential methylation suggests that many DNA methylation differences in LSGs from SS cases are associated with the same upstream biologic factors driving differential transcription in SS.

sue (Supplementary Figure 2, available at http://onlinelibrary.wiley.com/doi/10.1002/art.39792/abstract). The predictive power of differential methylation suggests that many DNA methylation differences in LSGs from SS cases are associated with the same upstream biologic factors driving differential transcription in SS. Characteristic binding motifs neighboring SS‐associated DMPs The AME tool identified 3 enriched motifs in the immediate neighborhood of DMPs (Table 4 and Supplementary Figures 3B–D, available at http://onlinelibrary.wiley.com/doi/10.1002/art.39792/abstract). The most significant motif was annotated for TCF11/MafG 34, an antioxidant response element binding complex that is reported to play a role in proteasome regulation and stability 35. A second enriched motif was annotated for the STAT1/STAT2 heterodimer, targeting interferon‐stimulated response elements 36. The final motif is the conserved binding motif of PU‐box–binding transcription factor PU.1 37. Table 4 Differentially methylated position–associated motifs identified by analysis of motif enrichmenta JASPAR ID Annotated transcription factor complex Targets Adjusted P MA0089.1 TCF11/MAFG heterodimer Antioxidant response elements 5.2 × 10−5 MA0517.1 STAT2/STAT1 heterodimer IFN‐stimulated response elements 7.5 × 10−4 MA0080.3 PU.1 PU box 5.9 × 10−3 a P values were determined by Fisher's exact test, with Bonferroni adjustment for multiple testing. IFN = interferon.

JASPAR ID Annotated transcription factor complex Targets Adjusted P MA0089.1 TCF11/MAFG heterodimer Antioxidant response elements 5.2 × 10−5 MA0517.1 STAT2/STAT1 heterodimer IFN‐stimulated response elements 7.5 × 10−4 MA0080.3 PU.1 PU box 5.9 × 10−3 a P values were determined by Fisher's exact test, with Bonferroni adjustment for multiple testing. IFN = interferon. DISCUSSION Through whole‐genome DNA methylation profiling of a clinically well‐characterized sample of European women, we identified a strong signature of disease‐associated immune processes in LSG tissue. We observed evidence of hypomethylation at the whole‐tissue level in SS cases as compared to controls. Further, our findings showed that epigenetic states of inflammatory genes and immune‐cell markers are major contributors to DNA methylation differences that distinguish SS cases. While results from this observational study cannot establish a causal role for the observed DNA methylation patterns in the risk of SS, our DMP‐based gene set, CpG set, and transcription factor motif enrichment analyses all demonstrated that DNA methylation profiling in SS cases and controls provides unique insights into tissue‐specific differences involved in disease.

establish a causal role for the observed DNA methylation patterns in the risk of SS, our DMP‐based gene set, CpG set, and transcription factor motif enrichment analyses all demonstrated that DNA methylation profiling in SS cases and controls provides unique insights into tissue‐specific differences involved in disease. The most significant DMP enrichment observed in this study was in the promoter of PSMB8‐AS1, a long noncoding RNA neighboring the PSMB8 locus (aka PSMB5i or LMP7) in the MHC region. This antisense RNA is in a head‐to‐head configuration with PSMB8 (Supplementary Figure 4, available at http://onlinelibrary.wiley.com/doi/10.1002/art.39792/abstract). PSMB8, the promoter of which we have demonstrated to be hypomethylated in SS cases, encodes a subunit of the immunoproteasome that has been reported to be up‐regulated in the salivary glands of patients with SS 38. The greater proteasome regulatory network was further implicated by the enrichment of TCF11/MAFG motifs surrounding SS‐associated DMPs. While these differences in DNA methylation may be functionally related, there is no clear evidence of immunoproteasome regulation by the TCF11/MAFG complex 35.

of patients with SS 38. The greater proteasome regulatory network was further implicated by the enrichment of TCF11/MAFG motifs surrounding SS‐associated DMPs. While these differences in DNA methylation may be functionally related, there is no clear evidence of immunoproteasome regulation by the TCF11/MAFG complex 35. We have also presented evidence here for promoter hypomethylation of TAP1, neighboring both PSMB8 and PSMB9. Rare variants of TAP1 and extended HLA haplotypes are thought to confer disease risk in some SS patients 39. Given their specific roles in antigen presentation, most DMPs observed across these 3 neighboring loci are likely to be directly associated with an increased proportion of immune cells in the tissue. This “tissue‐heterogeneity interpretation” is further supported by the abundance of differentially methylated cell differentiation markers noted in our DMP enrichment analyses; this enrichment could indicate that many‐to‐most of the extended DNA methylation differences observed in this study are consequences of varying cell proportions in the gland tissue. As a deeper understanding of cell‐type–specific DNA methylation motifs in immune‐ and tissue‐specific cells becomes available, the patterns observed in target tissue may serve as clues to which cell types are driving recurring inflammation in SS patients.

re consequences of varying cell proportions in the gland tissue. As a deeper understanding of cell‐type–specific DNA methylation motifs in immune‐ and tissue‐specific cells becomes available, the patterns observed in target tissue may serve as clues to which cell types are driving recurring inflammation in SS patients. The transcription factor PU.1 was highlighted multiple times in the current study. Not only was extended hypomethylation observed in the promoter region of this gene, but there also appeared to be a spatial association between differential methylation patterns and PU.1 binding motifs, both at the promoter level (CpG set enrichment analysis) and at the nucleosome level (TFBM enrichment analysis). PU.1 is a known factor involved in B cell and macrophage differentiation, binding to the enhancers of many lineage‐specific genes 40, and it may directly recruit DNA methylation machinery to repress target genes 41. As such, differential proportions of immune cell types (i.e., B lymphoid versus myeloid lineage) may drive PU.1 target enrichments in inflamed tissue. In particular, the abundance of hypomethylated B cell and lymphoid markers, including CD19, CD79B, PTPRCAP, and TNFRSF13B, further supports this interpretation.

enes 41. As such, differential proportions of immune cell types (i.e., B lymphoid versus myeloid lineage) may drive PU.1 target enrichments in inflamed tissue. In particular, the abundance of hypomethylated B cell and lymphoid markers, including CD19, CD79B, PTPRCAP, and TNFRSF13B, further supports this interpretation. Thabet et al 8 report that disease‐associated gland up‐regulation of ICAM1/CD54 3, a gene critically involved in the processes of intercellular adhesion and trans‐endothelial migration, was associated with global hypomethylation of salivary gland epithelial cell genomes. The investigators hypothesized that global hypomethylation could be a regulatory mechanism upstream of increased expression 8. We found no evidence of differential methylation in or around the ICAM1 promoter, suggesting that other mechanisms are directly responsible. However, due to the heterogeneous nature of gland tissue used in the current study, both direct and indirect effects may be masked by cell proportion differences in tissue.

ion 8. We found no evidence of differential methylation in or around the ICAM1 promoter, suggesting that other mechanisms are directly responsible. However, due to the heterogeneous nature of gland tissue used in the current study, both direct and indirect effects may be masked by cell proportion differences in tissue. Promoter enrichment analysis highlighted a microRNA (miR‐339) that has been demonstrated to be a potential posttranscriptional regulator of ICAM1 42. Although this mechanism is intriguing, there exists little evidence to support it within the context of SS, beyond down‐regulation of miR‐339 reported in a microarray study of SS‐affected glands 43. Any mechanistic interpretation is further complicated by the hypomethylation observed in the upstream regulatory region, which would support up‐regulation of this gene product based on a simple model of DNA methylation–associated epigenetic regulation. Despite the unknown biologic role of the striking hypomethylation we identified at this microRNA locus, the proposed regulatory potential of miR‐339 makes it an attractive candidate for functional studies.

rt up‐regulation of this gene product based on a simple model of DNA methylation–associated epigenetic regulation. Despite the unknown biologic role of the striking hypomethylation we identified at this microRNA locus, the proposed regulatory potential of miR‐339 makes it an attractive candidate for functional studies. Recently, Imgenberg‐Kreuz et al 10 reported results from their study of DNA methylation in minor salivary gland biopsies from 15 primary SS cases and 13 controls in which they used the 450K platform. In addition to a parametric analysis approach, the authors used a conservative Bonferroni‐adjusted P value reporting criterion for DMPs. While a top hit in OAS2 (cg20870559) was successfully replicated in the current study, only 2 of the remaining 44 DMP hits reported by that study were replicated here: cg12560128 and cg16596716. Both study populations were small, and differences in phenotype or age may have contributed to the lack of replication of other findings. Enrichment analyses and more comprehensive analyses of extended patterns of DNA methylation may be better approaches to characterizing profiles associated with case status than single CpG–site testing.

ons were small, and differences in phenotype or age may have contributed to the lack of replication of other findings. Enrichment analyses and more comprehensive analyses of extended patterns of DNA methylation may be better approaches to characterizing profiles associated with case status than single CpG–site testing. Previous studies have defined a gene as being differentially methylated if it contains a number of DMPs exceeding a given threshold 25. One problem with this approach is that it is biased toward reporting genes with higher CpG coverage. Assuming that false‐positive results would be randomly distributed across the 450K chip, a gene with better coverage will have more false‐positive results. Coverage is also problematically associated with biologic function 44, but enrichment tests, such as the hypergeometric test, will take this coverage into account. Given the difficulties associated with interpreting single CpG–site results, we chose to emphasize enrichment results, at both the promoter and pathway levels.

erage is also problematically associated with biologic function 44, but enrichment tests, such as the hypergeometric test, will take this coverage into account. Given the difficulties associated with interpreting single CpG–site results, we chose to emphasize enrichment results, at both the promoter and pathway levels. One of the strengths of our study is its restriction to European women, which minimized potential confounding by genetic ancestry or sex. Both have been shown to influence DNA methylation profiles 45, 46, and thus, our current results may not be generalizable to other studies of non‐European or male populations. Importantly, sex differences in many immunologic parameters have been observed 47. As a result, epigenetic studies comparing male cases and controls might yield a different set of SS‐associated LSG DMPs. It is also possible that SS case subgroups (e.g., cases with specific extraglandular manifestations) exhibit different DNA methylation profiles. While the current study was not large enough to test these hypotheses, larger studies will be able to probe phenotype‐specific methylation patterns.

of SS‐associated LSG DMPs. It is also possible that SS case subgroups (e.g., cases with specific extraglandular manifestations) exhibit different DNA methylation profiles. While the current study was not large enough to test these hypotheses, larger studies will be able to probe phenotype‐specific methylation patterns. Studies of circulating blood cells are well poised to reveal novel mechanisms in disease etiology due to ease of sample collection and access to naive cell populations. However, disease‐associated changes observed in these cells likely reflect systemic aspects of the disease, rather than tissue‐specific disease states driven by local inflammation. Labial salivary gland biopsy is a minimally invasive procedure that provides investigators access to tissue targets of SS and may help to illuminate processes specific to a disease in progress. Furthermore, as a target tissue, these samples may prove more useful in characterizing disease phenotypes in patients with early evidence of SS symptoms. Insights from this study and larger studies may soon yield new epigenetic biomarkers for this complex and heterogeneous disease and may help to inform the development of novel treatment strategies in the future. AUTHOR CONTRIBUTIONS All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Criswell 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.

thors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Criswell 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. Study conception and design Cole, Baker, Barcellos, Criswell. Acquisition of data H. Quach, D. Quach, Criswell. Analysis and interpretation of data Cole, Taylor, Barcellos, Criswell. Supporting information Supplementary Table 1. Self‐reported Medication by Case‐Control Status Supplementary Table 2. Correlations between First and Second Principal Components of DNAm, Phenotype and Selected Covariates Supplementary Table 3. Remaining (Lower Significance) Differentially Methylated Promoters in LSGs of Sjögren's Syndrome Patients Click here for additional data file. Supplementary Figure 1. Principal component analysis on genome‐wide gland DNAm, before and after non‐linear adjustment. A. Before adjustment, the first principal component separates the three batches. B. Spearman Z‐test p‐values showing significance of correlation between PC1‐5 and categorical phenotypes and covariates: disease status, treatment plate, sentrix ID (methylation chip ID), SSA and SSB seropositivity. Red dashed line indicates Bonferroni significance threshold for 25 tests. C. Following adjustment, batches are mixed. D. Significance of correlation with batch is low across first five PCs, while preserving associations with phenotype. Click here for additional data file.

Supplementary Figure 1. Principal component analysis on genome‐wide gland DNAm, before and after non‐linear adjustment. A. Before adjustment, the first principal component separates the three batches. B. Spearman Z‐test p‐values showing significance of correlation between PC1‐5 and categorical phenotypes and covariates: disease status, treatment plate, sentrix ID (methylation chip ID), SSA and SSB seropositivity. Red dashed line indicates Bonferroni significance threshold for 25 tests. C. Following adjustment, batches are mixed. D. Significance of correlation with batch is low across first five PCs, while preserving associations with phenotype. Click here for additional data file. Supplementary Figure 2. Correlation between DNA methylation and expression disease‐associations from two studies. The x‐axis shows extent of mRNA down‐regulation from Hjelmervik et al. and the y‐axis shows the significance of DNA hyper‐methylation – from the present study ‐ for CpGs in the promoter of the corresponding gene. DMPs from the present study are highlighted in cyan. Click here for additional data file. Supplementary Figure 3. JASPAR motifs enriched in the neighborhood of SS‐associated DMPs in labial salivary gland tissue. A. Schematic of transcription factor binding motif enrichment analysis. B. MA0089.1, annotated for heterodimer of chicken orthologs of NFE2L1 and MAFG. C. MA0517.1, annotated for the heterodimer of STAT2 and STAT1. D. MA0080.3, annotated for the mouse ortholog of SPI1. Click here for additional data file.

Supplementary Figure 3. JASPAR motifs enriched in the neighborhood of SS‐associated DMPs in labial salivary gland tissue. A. Schematic of transcription factor binding motif enrichment analysis. B. MA0089.1, annotated for heterodimer of chicken orthologs of NFE2L1 and MAFG. C. MA0517.1, annotated for the heterodimer of STAT2 and STAT1. D. MA0080.3, annotated for the mouse ortholog of SPI1. Click here for additional data file. Supplementary Figure 4. Extended differential methylation in PSMB8‐AS1 promoter. A. Highlighted region shows region designated as PSB8‐AS1 promoter, sitting within the gene body of PSMB8. All SS‐associated DMPs (promoter and non‐promoter) are annotated in the top track, UCSC Genome Browser RefGene annotations in the middle, and all 450K CpG sites at the bottom. B. Evidence of an SS‐associated differentially hypo‐methylated region within the promoter of PSMB8‐AS1. Click here for additional data file.