Introduction

Genome-wide association studies (GWAS) and GWAS meta-analyses have recently led to the identification of numerous new susceptibility loci in MS [1–3]. STAT3 (encoding “signal transducer and activator of transcription 3”) was initially observed as a putative MS risk gene in a GWAS from Finland. The same group also found evidence for association between STAT3 and MS risk in other populations of European descent [3]. The strongest, genome-wide significantly associated single nucleotide polymorphism (SNP) in that study was rs744166 (odds ratio (OR) = 1.15, P = 2.8 × 10−10). Recently, association of MS and other SNPs in the STAT3 locus was reported in a GWAS meta-analysis [4]. Here, the strongest signal was observed with rs2293152 (OR = 0.82, P = 4.1 × 10−8). Interestingly, rs2293152 displays only weak linkage disequilibrium (LD) with the original SNP rs744166 (r 2 = 0.23 in the 1000 Genomes CEU sample). However, since these two studies used partly overlapping datasets, the association has yet to be confirmed in entirely independent samples. A fully independent study assessing the potential association between STAT3 and MS risk in a Spanish case–control sample did not detect evidence for association [5]. Therefore, additional independent evaluations in large datasets are needed to assess the potential role of polymorphisms in STAT3 and MS risk. It is noteworthy that STAT3, which encodes a nuclear transcription factor, is also a compelling MS candidate gene based on functional studies. These suggest a role in the differentiation and expansion of Th17 cells and in the development of experimental autoimmune encephalomyelitis, the rodent model of MS, in vivo (e.g., ref. [5, 6]). In this study, we have genotyped rs744166 and rs2293152 in STAT3 in one of the largest available case–control samples comprising 5,904 Caucasian individuals from Germany who were not included in any of the previous analyses.

Methods

This sample, collected as part of a German multicenter effort, comprised 2,932 MS cases and 2,972 control subjects (for demographic characteristics see Table 1). All subjects were of self-reported German Caucasian descent. Detailed medical history and/or examination excluded the presence of inflammatory diseases in all controls. All samples were collected with informed written consent and appropriate ethical approval at the respective sites.

Table 1 Demographic details of the German case–control sample
Full size table

SNPs rs744166 and rs2293152 in STAT3 were genotyped on 384-well microtiter plates using allelic discrimination assays based on TaqMan chemistry (Applied Biosystems) “C___3140282_10” and “C___3140302_1” following the manufacturer's instructions. Each plate contained an approximately equal proportion of cases and controls as well as 5% CEU HapMap control samples that were duplicated between plates. Genotyping and genotype calling were performed blind to the disease status. Power calculations were performed with the Genetic Power Calculator (http://pngu.mgh.harvard.edu/∼purcell/gpc/) assuming a disease prevalence of 0.1%.

All genetic analyses were performed in PLINK v1.07 (http://pngu.mgh.harvard.edu/purcell/plink/). Association statistics were based on an additive transmission model and adjusted for age and sex via logistic regression. Assessment of Hardy–Weinberg equilibrium (HWE) was performed using Pearson's χ 2 as implemented in PLINK. Statistical significance of the association results is expressed as one-tailed P values and 95% confidence intervals. LD between variants in the STAT3 region was assessed using the SNAP software (www.broadinstitute.org/mpg/snap/) based on CEU data from the 1000 Genomes Pilot 1 Project (http://www.1000genomes.org/). Functional annotation of SNPs was performed using the ENSEMBL database (http://www.ensembl.org/biomart/martview/).

Results and discussion

Overall, genotyping efficiencies were 99.2% and 99.3%, and accuracies were 100% and 99.8% (one inconsistency across 541 HapMap genotypes) for rs2293152 and rs744166, respectively. We had >99% and 97% power to detect association of rs2293152 and rs744166 at one-sided α = 0.05 assuming the originally reported ORs (0.82 and 1.15, respectively) as underlying effect sizes. Control genotype frequencies were distributed according to HWE (Table 2). We could confirm a nominally significant association between the G-allele at rs744166 in STAT3 and MS susceptibility in our dataset (OR, 1.09; 95% confidence interval (CI) 1.01–1.17; P (one tailed) = 0.016; see Table 2). The second tested SNP, rs2293152, only showed a nonsignificant trend of association with MS risk (OR = 0.95; 95% CI 0.88–1.02; P (one tailed) = 0.091, see Table 2), although the direction of effect was identical to that reported in the GWAS meta-analysis [4]. Logistic regression analysis after adjustment for rs744166 did not reveal a second, independent signal for rs2293152 (data not shown).

Table 2 Association data summary for STAT3 in the German case–control sample
Full size table

Here, we have independently assessed the potential association between MS risk and two common polymorphisms in the STAT3 gene in 5,904 subjects from Germany. Interestingly, the stronger and statistically more significant effect was observed with polymorphism rs744166, originally implicated by the Finnish GWAS [3], while rs2293152, which showed genome-wide significant evidence for association in the recent GWAS meta-analysis [4] yielded only a weak, statistically nonsignificant effect. In general, the effect sizes estimated in our sample were smaller than those described in the two previous studies (i.e., ORs of 1.15 and 0.82). The most likely explanation for the smaller ORs in our validation sample may be the well-characterized “winner's curse” phenomenon, i.e., an overestimation of the effect size in the GWAS datasets that led to the identification of STAT3 as an MS risk factor. Actually, the OR of 1.09 for rs744166 in our dataset is well in line with the effect size estimate for STAT3 (rs9891119: OR = 1.10, P = 4.6 × 10−7) as well as the median OR (1.11) of all MS susceptibility loci observed in the most recent and largest GWAS [2]. Not surprisingly, rs9891119 is in very strong LD with rs744166 (r 2 = 0.81 in the CEU sample of the 1000 Genomes Project), probably tagging the same underlying signal. Whereas in good agreement with the most recent and extensive association data, our results could still be affected by random error or some undetected bias, e.g., population stratification. While the latter possibility could not be directly assessed across the entire dataset owing to the small number of SNPs that have been genotyped in this sample to date, GWAS data are available for 760 of the controls included here with no evidence for admixture (L.B., unpublished data). This is in line with other reports suggesting only a very minor degree of population substructure in the German population (e.g., ref. [7]). Therefore, as all cases and controls were drawn from the same source population, undetected admixture has very likely not affected our results to an appreciable extent.

Finally, rs744166 and rs2293152 as well as rs9891119 are intronic SNPs that may merely tag the genetic variant(s) functionally responsible for the association of STAT3 with MS. However, none of these three SNPs is in LD (r 2 ≥ 0.2 based on the CEU sample of the 1000 Genomes Project) with any potentially functionally relevant variant in the coding region of STAT3. Further large fine-mapping studies and functional experiments are warranted to clarify the exact underlying molecular mechanisms. Thus far, our results point towards a potential pathogenic role of STAT3, a proinflammatory transcription factor, which is known to be involved in T cell differentiation and activity, particularly in interleukin 17 production and RORγt/Foxp3 balance [8].

Conclusion

We could confirm the previously reported association between common polymorphisms in STAT3 and MS risk in this large and independent sample of 5,904 subjects. Further studies are needed to clarify the exact molecular mechanisms underlying this association.