Genome-wide Association Study in a High-Risk Isolate for Multiple Sclerosis Reveals Associated Variants in STAT3 Gene

Abstract

Genetic risk for multiple sclerosis (MS) is thought to involve both common and rare risk alleles. Recent GWAS and subsequent meta-analysis have established the critical role of the HLA locus and identified new common variants associated to MS. These variants have small odds ratios (ORs) and explain only a fraction of the genetic risk. To expose potentially rare, high-impact alleles, we conducted a GWAS of 68 distantly related cases and 136 controls from a high-risk internal isolate of Finland with increased prevalence and familial occurrence of MS. The top 27 loci with p < 10⁻⁴ were tested in 711 cases and 1029 controls from Finland, and the top two findings were validated in 3859 cases and 9110 controls from more heterogeneous populations. SNP (rs744166) within the STAT3 gene was associated to MS (p = 2.75 × 10⁻¹⁰, OR 0.87, confidence interval 0.83–0.91). The protective haplotype for MS in STAT3 is a risk allele for Crohn disease, implying that STAT3 represents a shared risk locus for at least two autoimmune diseases. This study also demonstrates the potential of special isolated populations in search for variants contributing to complex traits.

Main Text

Multiple sclerosis (MS) (MIM #126200) is a complex inflammatory disease of the central nervous system with presumed autoimmune etiology. Both environmental and genetic factors are thought to contribute to the development of MS,^1–3 and the genetic risk factors likely include both common and rare risk alleles. Recent GWAS and subsequent meta-analysis have established the critical role of the HLA locus^4–6 and identified new MS loci: IL2RA (MIM ^∗147730),⁷ IL7R (MIM ^∗146661),^7–9 CLEC16A (MIM ^∗611303),^7,10–13 CD58 (MIM ^∗153420),^11,12,14 TNFRSF1A (MIM ^∗191190),¹⁵ IRF8 (MIM ^∗601565),¹⁵ and TYK2 (MIM ^∗176941).^12,16,17 These associated variants, except for TYK2, are common, have small odds ratios (ORs), and explain only a fraction of the genetic risk.

The population history of Finland and the province of Southern Ostrobothnia (SO), an internal isolate with increased prevalence of MS,^18–22 is compatible with a founder effect.^22–24 Previous studies in Finnish MS families originating from this high-risk subisolate have demonstrated linkage and association to the HLA locus (HLA-DRB1 [MIM ^∗142857]),^25–27 17q22-24,^25,28,29 and 5p14-p12.^25,30–32 Therefore, we hypothesized that some variants predisposing to MS have either become enriched in SO or can be more easily detected against a homogenous background with a genome-wide, high-density SNP screen. We looked for shared alleles enriched in cases, as well as potential extended homozygous regions and copy number variations (CNVs) enriched in MS cases.

We included in our GWAS 72 cases with either both parents from the high-risk isolate or one parent from the isolate and positive family history of MS and genotyped them with the Illumina HumanHap300 chip. Extensive genealogical research revealed that the majority of the cases could be traced to two large interrelated pedigrees (see Figure S1 available online). A total of 2206 population-based controls were genotyped with either Illumina HumanHap300 chip or with Illumina HumanHap610-quad chip. We excluded samples and SNPs with <95% success rates, leaving 72 cases and 2196 controls for the subsequent analyses, and selected only SNPs present on both Illumina platforms (297,343 SNPs) for analyses. Gender check was performed with X chromosomal SNPs, and no discrepancies between the observed and expected gender were noted. Identity-by-descent (IBD) analysis was performed to study possible close cryptic relatedness between individuals and to identify possible samples with excess relatedness, suggestive of sample contamination. We then performed identity-by-state (IBS) and multidimensional scaling analyses: four cases were initially considered as isolate samples clustered outside the isolate sample set and were excluded from subsequent analyses (Figure S2). We selected the two closest IBS-matched controls for each case, and the final GWAS set (isolate GWAS) consisted of 68 cases and 136 controls. The genomic inflation factor suggested no major inflation (λ 1.078) and a fairly well-matched case-control set, which was also confirmed by quantile-quantile plot analysis of single SNP association results (Figure S3). Because we had parental birthplace data for both the cases and the majority of controls (n = 2174), we could further verify that all cases and 125 of the 136 selected controls had at least one parent born in Southern Ostrobothnia, and of these, 64 cases and 90 controls had both parents born in Southern Ostrobothnia. We have recently shown a correlation between geographical origin of samples (based on parental birthplace information) and genome-wide SNP data in the Finnish population.²³ Thus, IBS matching of cases and controls combined with genealogical information should minimize the risk of population substructure in our study set. All patient samples were collected with informed consent, and the study design and the Finnish sample collection have been approved by the Helsinki University Hospital Ethical Committee of Ophthalmology, Otorhinolaryngology, Neurology and Neurosurgery (permit 192/E9/02).

Taking advantage of the distant relatedness in the subisolate, we conducted homozygosity analyses with PLINK.³³ First we searched for extended regions of homozygosity (ROHs), the signature features of isolated populations, enriched in MS cases to identify loci that could influence MS susceptibility in a recessive manner. ROHs with at least 50 consecutive SNPs and a minimum length of 500 kb were identified in each individual. On average, we identified 149 (standard deviation: 12 in cases, 10 in controls) ROHs per individual with an average length of 1030 kb (500 kb–31.3 Mb) in cases and 1018 kb (500 kb–49.6 Mb) in controls. We then evaluated which overlapping homozygous regions were enriched in cases by permuting the group (case-control) labels 10,000 times. The analysis revealed three putative regions with empirical p < 10⁻³: 1q42.12 (242 kb, 24 SNPs, p = 3 × 10⁻⁴), 2q24.3 (512 kb, 39 SNPs, p = 8 × 10⁻⁵), and 12q24.33 (573 kb, 48 SNPs, p = 3 × 10⁻⁴) (Table S1 and Figure S4). Although the cases and controls are matched on the basis of their genome-wide IBS sharing and are augmented by parental birthplace information, the permutation-based approach is susceptible to population substructure, and obtained p values should be interpreted with caution. Excess homozygous sharing was observed with the same haplotype for 13% (9 of MS cases) and 7% (10 of controls) for 1q42.12 and 37% (25 of MS cases) and 20% (27 of controls) for 2q24.3. For the 12q24 region, we observed multiple different haplotypes (Table S1). These regions have not been previously implicated in MS except for suggestive linkage in 12q23-24,³⁴ and their putative role in MS susceptibility requires further validation. Haplotype sharing outside of the isolate in the population control samples (n = 2194) was similar to the GWAS internal isolate control population (frequencies 5.8% for 1q42.12 and 20% for 2q24.3 haplotypes). This indicates that the homozygous haplotypes have been enriched in the subisolate MS cases, but not in the isolate controls, although the IBD analysis showed the isolate controls to be as related to each other as the isolate MS cases (Table S2).

The Illumina HumanHap300 platform has relatively sparse coverage and is void of probes in the most common CNV regions but could be suitable for detecting rare, large CNVs, potentially enriched in the internal isolate population. We used the QuantiSNP software³⁵ for CNV detection (GC content correction option, restricted to CNVs with log Bayesian factor > 10 and length ≥ 3 SNPs) and verified these results visually with Bead Studio 3.3. All CNVs in centromeric regions were excluded. We identified altogether 106 CNV regions in 68 cases (Table S3); all but 6 of the 106 CNVs have been previously reported. Furthermore, all novel CNVs were found in only one case each. Hypothesizing that genes mapping next to the 106 CNVs identified in cases could belong to a common pathway involved in MS etiology, we used Ingenuity Pathway Analysis to search for connecting pathways. One pathway potentially regulating oligodendrocyte differentiation and myelin sheet formation^36–41 involving NRG3 (MIM ^∗605533), ERBB4 (MIM ^∗600543), DLG2 (MIM ^∗603583), UTRN (MIM ^∗128240), and LARGE (MIM ^∗603590) (all CNVs previously reported) was identified (Figure S5), but CNV deletions in these genes were observed to have similar frequency in MS cases compared to controls with Fisher's exact test (ERBB4: 11% of cases and 12% of controls, p = 0.388; NRG3: 4% of cases and 4% of controls, p = 0.90; and DLG2: 1% of cases and 0% of controls, p = 0.404) when genotyped in an independent set of 703 cases and 1051 controls with an in-house-developed PCR-based fragment analysis method.⁴²

Southern Ostrobothnia is an old isolate, and thus the expected shared haplotypes are of modest length. We therefore performed single SNP standard χ² allelic association analysis with PLINK.³³ Because of the limited power, we analyzed all 27 loci (28 of the 37 initial SNPs) showing nominal association in the GWAS analysis (p < 10⁻⁴; Table S4) in a larger independent Finnish sample set of 711 cases and 1029 controls, of which 83 MS cases and 365 controls were from the isolate (Table 1). Population-stratified Cochran-Mantel-Haenszel (CMH) association analysis provided evidence for three SNPs: rs3135338 in the HLA region (p = 1.6 × 10⁻²⁵), rs744166 in first intron of STAT3 (MIM ^∗102582) in chromosome 17q21.1 (p = 0.0012), and rs1364194 in chromosome 16 (p = 0.0047) (Table S4). The non-HLA SNPs were then analyzed in an international sample of 3859 MS cases and 9110 controls from six different populations (Table 1). The combined evidence for association to STAT3 (rs744166) was significant (p = 2.75 × 10⁻¹⁰ and OR 0.87 [95% confidence interval (CI) 0.83–0.91]) (Figure 1; Table S5). The Breslow-Day analysis of heterogeneity of odds ratios revealed no significant heterogeneity (p = 0.34). When the combined replication data set was analyzed by logistic regression for additive, dominant, and recessive models with study set as a covariate in the analyses, the statistically most significant p value was obtained for the additive model (Table S6). We obtained no additional support for the chromosome 16q region.

Table 1.

Description of Sample Sets in GWAS and Replication Studies

Study Population	Number of MS Cases	Number of Controls	Genotyping Platform
Southern Ostrobothnia (SO) isolate GWASa	68	136	Illumina HumanHap300 or Illumina Human610-quad
Finland SO replicationb	83	365	Sequenom iPlex Gold
Finlandc	628	668	Sequenom iPlex Gold
Norwayd	607	816	Sequenom iPlex Gold
Denmarke	628	1074	Sequenom iPlex Gold
Gene MSA Switzerlandf	253	208	Illumina HumanHap300
Gene MSA Netherlandsf	230	232	Illumina HumanHap300
Gene MSA USf	486	431	Illumina HumanHap300
IMSGC UKg	453	2950	Affymetrix 500K
IMSGC USg	342	1679	Affymetrix 500K
BWHg	860	1720	Affymetrix 6.0
Total	4638	10,279

All samples have been diagnosed with clinically definite MS according to either Poser's or McDonald's criteria.

^aMS samples included in the genome-wide analysis originated from the MS high-risk isolate located on the western coast of Finland (Southern Ostrobothnia, SO) with ∼2-fold prevalence and higher familial clustering of MS compared to other regions of Finland. Most of the cases were distantly related, and no closer than second-degree relatives were included (Table S2). The cases were genotyped with Illumina HumanHap300 at The Broad Institute of MIT and Harvard. Control samples (n = 136) were selected by utilizing identical-by-state (IBS)-sharing and parental birthplace information from a pool of population-based controls (total n = 2206) genotyped either with Illumina HumanHap300-duo chips at the Institute for Molecular Medicine Finland (FIMM) Technology Centre or with Illumina Human610-quad chips at the Sanger Institute.

^bThe MS cases have at least one parent born within SO, and anonymous population controls were collected from the Central Hospital of Seinajoki in Southern Ostrobothnia. The samples were genotyped in the FIMM Technology Centre.

^cFinnish MS patients (excluding samples from the SO region) from various regions (Tampere, Helsinki, Kuopio, Oulu) and anonymous population controls collected from Kuopio and Helsinki University Hospitals. The samples were genotyped in the FIMM Technology Centre.

^dNorwegian samples have been described in more detail in Lorentzen et al.⁵⁴ The samples were genotyped in the FIMM Technology Centre.

^eThe Danish nationwide study set cases have been diagnosed with clinically definite MS according to the McDonald criteria. The controls are healthy blood donors and hospital workers residing in the same region as patients. Experimental protocols (KF 01314 009) were approved by the local ethics board, and informed consent was obtained from all participants. The samples were genotyped in the FIMM Technology Centre.

^fStudy sample from the Gene MSA consortium is also a part of the recently published meta-analysis¹⁵ and is described in detail elsewhere by De Jager et al.¹⁵ and Baranzini et al.⁴⁶

^gStudy sample from a recently published meta-analysis is described in detail elsewhere¹⁵ and was kindly provided by De Jager et al.¹⁵

Figure 1.

Population-Specific Association for the STAT3 rs744166 A Allele

The rs744166 A allele that tags a putative MS protective haplotype associated to MS and shows consistent reduced risk in all studied populations with available genotypes. The results are presented here by study set. Each line represents one study set showing the name of the set, A allele frequencies for cases and controls, ORs, p value for association, and graphic illustration of the odds ratio (square, size relative to study set size) with the 95% confidence intervals for the odds ratio (thin lines).

Evaluation of the STAT3 linkage disequilibrium (LD) block that contains the associated SNP rs744166 in Hapmap2 (build 23a) samples⁴³ with Haploview 4.0⁴⁴ showed that rs744166 non-risk-associated A allele completely tags the most common haplotype in Southern Utah residents of European descent (CEU) (56%), Han Chinese from Beijing (CHB 65%), and Tokyo Japanese (JPT 57%), but the G allele is present on four different haplotypes (Table S7). In the Yoruban population from Nigeria (YRI), the A allele is present on four different haplotype backgrounds, and the most common A haplotype in CEU, CHB, and JPT populations has the frequency of 7% in the YRI population. We speculate that this notable enrichment of a single haplotype in non-African populations might suggest positive selection of the putative MS protective haplotype outside Africa, although this locus did not reach genome-wide significance in an analysis of signs of recent positive selection.⁴⁵ The rs744166 A allele also shows changes in frequency distribution in the Human Genome Diversity Panel (Figure S6).⁴⁵ The LD block carrying the haplotype is 54 kb in length in the CEU population and contains the beginning of STAT3 and its immediate promoter region (Figure 2).

Figure 2.

Description of the Associated LD Region in STAT3

The associated SNP and haplotypes are in a 54 kb LD block covering the beginning and immediate 5′ region of the STAT3 gene. The associated rs744166 SNP is marked with a red arrow, and the other two SNPs, rs6503695 and rs957970, used in the haplotype analysis are marked with yellow arrows. The SNP marked with the blue arrow (rs2293152) was listed among the 100 top SNPs suggestively associated to MS in a previous meta-analysis.¹⁵

We tagged the haplotypes with three SNPs (rs744166, rs6503695, and rs957970) with Haploview 4.0 tagging option. These SNPs were genotyped in the Finnish sample set, and the data for the same SNPs were available from four other populations from a recent meta-analysis.^7,15,46 We phased the haplotypes with PLINK and performed a CMH analysis with populations as clusters. We could define both a putative predisposing haplotype (30.9% in MS, 27.1% in controls, OR 1.18, 95% CI 1.11–1.27) with CMH p = 1.29 × 10⁻⁶ and a tentative protective haplotype (55.0% MS, 58.7% controls, OR 0.86, 95% CI 0.81–0.91) (Figure 1) with CMH combined p = 1.19 × 10⁻⁶ (Table 2; Tables S7 and S8). The Breslow-Day test revealed no significant heterogeneity of odds ratios (p = 0.271 and p = 0.301, respectively). Further studies, including resequencing, will be needed to identify the true affecting variants segregating in one or both of these haplotypes.

Table 2.

Summary of the 54 kb STAT3 Haplotype Data Showing One Putative Predisposing and One Putative Protective Haplotype

Haplotype	Frequency MS (n = 3255)	Frequency Control (n = 8133)	P Value	OR	95% CI
CGGa	0.309	0.271	1.29 × 10−6	1.18	(1.11–1.27)
TAAb	0.550	0.587	1.19 × 10−6	0.86	(0.81–0.91)
TGG	0.082	0.080	0.439	1.06	(0.94–1.17)
CGA	0.057	0.059	0.591	0.98	(0.85–1.10)

The haplotypes were constructed with SNPs rs6503695, rs744166, and rs957970 and phased with PLINK. Only phased haplotypes with posterior probability of 1 were included in the analysis. Each haplotype was analyzed separately and showed no evidence for heterogeneity of odds ratios between populations in the Breslow-Day test, which allowed us to combine the haplotype results with CMH. The analysis included a total of 3255 MS cases and 8133 controls from Finnish, BWH, IMSGC UK, IMSGC US, Gene MSA US, Gene MSA CH, and Gene MSA NL sample sets. The results for individual populations are provided in Table S5.

^aThe predisposing haplotype CGG is significantly overrepresented in the MS cases.

^bThe protective haplotype TAA is significantly underrepresented in the MS cases.

STAT3 codes for a transcription factor that is involved in multiple pathways and functions, including the Jak-STAT pathway, neuron axonal guidance, apoptosis, activation of immune responses, and Th17 cell differentiation.⁴⁷ Interestingly, the A allele of rs744166 tagging the MS-protective haplotype is associated with Crohn disease,⁴⁸ and mutations in STAT3 are known to cause hyperimmunoglobulin E recurrent infection syndrome (HIES [MIM #147060]),^49,50 a rare autosomal-dominant disorder characterized by elevated immunoglobulin E levels and inflammation. Additionally, mouse studies have shown that targeted deletion of Stat3 in CD4+ T cells prevents the development of experimental autoimmune encephalomyelitis (EAE), the rodent model of MS,⁵¹ and that T_reg-specific ablation of Stat3 resulted in the development of a fatal intestinal inflammation due to unstrained T_H17 response.⁵² Recent meta-analysis of GWAS in MS listed STAT3 as one of the genes with a suggestive role in at least two autoimmune disorders¹⁵ but failed to replicate the initial STAT3 association. The failure to replicate the initial association was probably due to selecting the most significantly associated regional SNP (rs2293152), which resides just outside of the rs744166 containing LD region and has only limited LD with the rs744166 (r² 0.35 in HapMap2 CEU population), for the replication analysis (Figure 2). These observations support a wider role for STAT3 in autoimmunity and adds this gene to the growing list of MS-susceptibility genes with validated or substantial evidence for association in at least two inflammatory diseases.^48–50 All of these together suggest a significant role of this locus in immune system and autoimmune disease pathogenesis.

Most of the currently validated (IL2RA, IL7R, CD58, CLEC16A, IRF8, TNFRSF1A, TYK2)^{7,9,12–17,53} and suggested (C7 [MIM ^∗217070], CD6 [MIM ^∗186720], IL12A [MIM ^∗161560], OLIG3 [MIM ^∗609323]–TNFAIP3 [MIM ^∗191163], PTGER4 [MIM ^∗601586], RGS1 [MIM ^∗600323])^15,30 non-HLA MS susceptibility loci have known functions in the immune system and particularly in T cells. Although their independent ORs are modest, their combined effect might be larger, and a large-scale international study would be required to estimate their combined effect toward disease predisposition. The present study demonstrates the power of the founder population study design to complement large-scale GWAS in identifying genes and pathways of general significance, not only rare high-impact alleles.

Web Resources

The URLs for data presented herein are as follows:

• PLINK: Whole Genome Association Analysis Toolset, http://pngu.mgh.harvard.edu/∼purcell/plink/

• Online Mendelian Inheritance in Man (OMIM), http://www.ncbi.nlm.nih.gov/Omim/

• Ingenuity Pathway Analysis (IPA) Software, http://www.ingenuity.com

• Database of Genomic Variants (DGV), http://projects.tcag.ca/variation/

• International HapMap Project, http://www.hapmap.org/

• The Human Genome Diversity Project (HGDP) Selection Browser, http://hgdp.uchicago.edu/