Chromosome 7q Region Associated with Female Rheumatoid Arthritis in a British Population Fails to Replicate in a North American Case-Control Series

Abstract

Objective

The single nucleotide polymorphism (SNP) rs11761231 on chromosome 7q has been reported as a sexually dimorphic marker for rheumatoid arthritis susceptibility in a British population. We sought to replicate this finding and better characterize susceptibility alleles in the region in a North American population.

Methods

DNA from two North American collections of RA patients and controls (1605 cases and 2640 controls) was genotyped for rs11761231 and 16 additional chromosome 7q tag SNPs using Sequenom iPlex assays. Association tests were performed for each collection and also separately contrasting male cases versus male controls and female cases versus female controls. Principal components analysis (EIGENSTRAT) was used to determine association with RA before and after adjusting for population stratification in the subset of the samples (772 cases and 1213 controls) with whole genome SNP data.

Results

We failed to replicate association of the 7q region with rheumatoid arthritis. Initially, rs11761231 showed evidence for association with RA in the NARAC collection (p=0.0076) and rs11765576 showed association with RA in both the NARAC (p = 0.019) and RA replication (p = 0.0013) collections. These markers also exhibited sexual differentiation. However, in the whole genome subset, neither SNP showed significant association with RA after correction for population stratification.

Conclusion

While two SNPs on chromosome 7q appeared to be associated with RA in a North American cohort, the significance of this finding did not withstand correction for population substructure. Our results emphasize the need to carefully account for population structure to avoid false positive disease associations.

It has long been recognized that gender is a major risk factor for the development of autoimmune disease and that females are at an increased risk for these conditions. In the case of rheumatoid arthritis (RA), the female to male ratio is approximately 3:1(1). Furthermore, males and females with RA also tend to have different disease characteristics(2). Despite the suspicion that genetic factors underlie these observations, the genes that have been implicated in autoimmunity have yet to explain the female predominance of RA or any other autoimmune disorder.

Genetic studies have identified a number of loci that are associated with RA susceptibility including HLA-DRB1(3), PTPN22(4), and more recently STAT4(5), TRAF1/C5(6), and a region near TNFAIP3 on chromosome 6q23 (7, 8). Among the findings in their whole genome association (WGA) study of 7 diseases(9), the Wellcome Trust Case Control Consortium (WTCCC) identified a single SNP rs11761231 on chromosome 7q, which exhibited sexual dimorphism showing a statistically significant association with RA susceptibility only in females (p = 6.8 × 10^-8 in females and p = 0.68 in males). The authors suggested that this variant in RA may represent one of the first sex-differentiated genetic effects in human autoimmune disease.

In the present study, we sought to confirm the association of rs11761231 with RA in females by genotyping North American RA patients and healthy controls. Furthermore, to better characterize the region surrounding this SNP, we also genotyped 16 additional SNPs to determine if other nearby markers may have similar or stronger disease association.

Materials and Methods

Subjects

DNA from European ancestry North American RA patients and unrelated controls was obtained from two previously reported case control collections, the NARAC series and the RA replication series (5). The NARAC cases included one affected member from each family of European descent from the NARAC collection of affected sibling pairs collected throughout North America (10). The replication series cases were self described Caucasians obtained through the Wichita Rheumatic Disease Data Bank (Witchita, Kansas) (11) the National Inception Cohort of Rheumatoid Arthritis Patients (nationwide USA), (12), and the Study of New Onset Rheumatoid Arthritis (North America) (13). All the cases met the ACR criteria for the diagnosis of RA. The NARAC cases were 100% long-standing disease, 81.7% rheumatoid factor positive, 80.5% anti-cyclic-citrullinated peptide positive, 80.9% shared epitope positive, 80.1% female, and 94.2% had hand erosions on X-rays read by a single radiologist. The replication series cases were 72.2% female, 98.5% anti-cyclic-citrullinated peptide positive, 48.5% long-standing disease, and had a mean age of onset of 49.7 ± 14.1 years. The controls for these collections were population controls, not evaluated for disease, from the New York Cancer Project (14). They were selected from 20,000 samples collected from donors from New York City and the surrounding area. The controls selected were between 30 and 60 years old, Caucasian by self report, and where possible, they were matched to cases by self reported grandparental country of origin and age by birth decade. In total, the samples analyzed included 1605 independent RA cases and 2640 independent European-ancestry population controls. We also performed a subset analysis limited to samples (772 cases and 1213 controls) for which whole genome genotype data were available (6). Informed consent was obtained from every subject, and approval of the local institutional review board was secured at every recruitment site prior to the start of enrollment.

Selection of SNPs

In addition to rs11761231, we used the tag SNP Picker utility available at the International HapMap Consortium website (www.hapmap.org), which uses the Tagger algorithm(15), to select additional tag SNPs. These 18 tag SNPs capture with pairwise r² greater than 0.8, all HapMap variants with greater than 5% minor allele frequency (MAF) located in close proximity to and within the EST DA600502, which contains the reported SNP, rs11761231.

Genotyping

Multiplex SNP assays were designed using Sequenom RealSNP software (www.realsnp.com); two SNPs, rs2909480 and rs12536699, failed assay design. The remaining 17 SNPs were genotyped by the iPLEX Gold protocol and the genotypes determined by SpectroTyper software (Sequenom, Inc, San Diego, California). Calls were evaluated and edited by cluster analysis performed with the SpectroTyper software. Deidentified cases and controls were analyzed together. Genotyping accuracy was evaluated by two methods. In a set of 154 controls genotyped for 17 SNPs in duplicate (duplicates on different assay plates), the genotype concordance rate was 99.9%. Additionally, four of the 7q region SNPs were genotyped in the recently reported WGAS (6). A concordance rate of 98.7% was found in the 772 WGAS cases and 1213 controls genotyped on both the Sequenom and Illumina platforms.

Statistical Analysis

After genotyping, SNP markers were evaluated for significant deviation from Hardy-Weinberg equilibrium or low minor allele frequencies. We planned to exclude markers with disequilibrium p<0.005 in controls or minor allele frequency <0.01 to avoid errors in genotyping or insufficient power, but all markers genotyped passed these measures. SNPs were then analyzed for association by comparison of the minor allele frequency in cases and controls with significance determined by a chi squared test; this was also done separately for males and females. Linkage disequilibrium patterns in the 7q region were determined using Haploview version 4.0 (16). A principal component (PC)-based method, EIGENSTRAT (17), was used to adjust for population structure. The EIGENSTRAT analysis did not include the following intervals due to strong linkage disequilibrium: chromosome 6 (24-36 Mb), 8 (8-12 Mb) and 17 (40-43 Mb). The genomic control inflation factor (λ_gc) stabilized at 1.06 at PC6 and all adjusted Chi square tests are shown for PC6 corrected for the residual λ_gc.

Results

7q Linkage Disequilibrium Structure

Despite the close proximity of the 17 SNPs genotyped, linkage disequilibrium in the 908 kb region was poor. The marker rs11761231 had a D′ value of > 0.9 with only one of the other 16 markers (rs7795093), however the r² value was only 0.32 (See Figure 1).

SNP Association with RA in NARAC and RA Replication Collections

Of the seventeen SNPs genotyped, two showed evidence of association with RA. The first, rs11761231, replicated WTCCC's findings of significant association with RA and sexual dimorphism in the NARAC collection (see Table 1). However, this SNP was not associated with disease in the RA replication collection. Another marker, rs11765576, showed evidence for association with RA in both the NARAC and RA Replication collections and also showed sex-differentiation. These markers were independent with D′ = 0.11 and r² = 0. In comparing the two collections, we noted that the minor allele frequencies, especially for rs11761231, were quite different between the control groups suggesting that population stratification may be contributing to these results (See Table 1).

Table 1.

Sex-differentiated analysis of two RA case-control series. Minor allele frequencies (MAF) and association test results for two chromosome 7q SNPs genotyped amongst males and females. These data comprise the complete case-control collections previously described (5). Bold type indicates p < 0.05.

SNP	Sample collection	Sex	# cases, # controls	Allele frequency (cases)	Allele frequency (controls)	X2	p-value
rs11761231	NARAC	M + F	607, 1315	0.315	0.361	7.19	0.0073
		M	120, 329	0.338	0.347	0.06	0.8022
		F	486, 986	0.309	0.366	8.57	0.0034
	RA Replication	M + F	998, 1325	0.364	0.338	3.15	0.0760
		M	275, 613	0.386	0.335	4.05	0.0442
		F	722, 677	0.356	0.343	0.49	0.4849
	Pooled	M + F	1605, 2640	0.346	0.350	0.13	0.7182
		M	395, 942	0.371	0.339	2.36	0.1247
		F	1208, 1663	0.337	0.356	2.16	0.1418
rs11765576	NARAC	M + F	607, 1315	0.410	0.374	4.32	0.0378
		M	120, 329	0.397	0.392	0.03	0.8739
		F	486, 986	0.414	0.368	5.51	0.0189
	RA Replication	M + F	998, 1325	0.409	0.362	10.37	0.0013
		M	275, 613	0.405	0.368	2.11	0.1463
		F	722, 677	0.411	0.356	8.56	0.0034
	Pooled	M + F	1605, 2640	0.410	0.368	13.97	0.0002
		M	395, 942	0.402	0.376	1.58	0.2092
		F	1208, 1663	0.412	0.363	13.55	0.0002

SNP Association with RA after Correction for Population Structure

In order to determine the effect of population structure on our results, we analyzed the subset of samples for which whole genome SNP data were available, both before and after correction for stratification using EIGENSTRAT (17). Before correction for stratification, this subset showed significant evidence of association with RA and a sex-differentiated effect for both rs11761231 and rs11765576. However, after accounting for population differences by correcting the association results using principal components analysis, none of associations remained statistically significant (See Table 2).

Table 2.

RA-associated allele frequencies and association tests of chromosome 7q SNPs genotyped amongst the subset of samples with whole genome SNP data analyzed with and without EIGENSTRAT adjustment for population stratification. Bold type indicates p < 0.05.

SNP	Sex	# cases, # controls	Allele frequency (cases)	Allele frequency (controls)	OR (95% CI)	Unadjusted		Adjusted for population substructure
						X2	p-value	X2	p-value
rs11761231	M + F	772, 1213	0.325	0.361	1.17 (1.02 to 1.35)	4.33	0.038	2.42	0.12
	M	92, 293	0.331	0.349	1.08 (0.76 to 1.56)	0.49	0.49	0.094	0.78
	F	680, 920	0.324	0.365	1.20 (1.03 to 1.39)	4.46	0.035	2.76	0.097
rs11765576	M + F	772, 1213	0.414	0.373	1.18 (1.04 to 1.35)	6.32	0.012	0.64	0.42
	M	92, 293	0.421	0.394	1.12 (0.80 to 1.58)	0.25	0.62	0.12	0.73
	F	680, 920	0.413	0.367	1.21 (1.05 to 1.41)	6.87	0.0088	1.76	0.18

Discussion

This study demonstrates some of the difficulties associated with genetic disease association studies. The 7q region identified by the WTCCC is an intriguing region for a number of reasons(9). Even without sex-differentiation, the rs11761231 p-value in the WTCCC study suggested an association with RA. When sex was considered, the fact that the association with this SNP was strong in females and not detected in males, made an even more intriguing argument for this variant in RA, a female predominant disease. These data, coupled with our observation that this SNP is located within a novel EST derived from human RA synoviocytes, motivated us to attempt to replicate the association and to further evaluate additional variants from this genomic region.

Our genotyping of North American RA cases and controls initially gave supportive evidence for association of rs11761231, the SNP identified by the WTCCC as a marker for a sexually dimorphic risk factor for RA. The evidence was strongest in females of the NARAC case control collection. Furthermore, we identified a second SNP, rs11765576, which also had stronger evidence for association in females than in males.

Interestingly, the minor allele frequencies of some of the SNPs in the region varied between the collections; even between the two control groups (see Table 1). We therefore sought to determine if population admixture or population substructure differences affected the associations we observed. By limiting our analysis to the roughly half of individuals from the NARAC and RA Replication cohorts that had also been genotyped as part of the NARAC/EIRA RA WGA scan (6) and using EIGENSTRAT software, we attempted to identify whether such unseen nuances existed in our case and control population structures. Six PCs were necessary to control for substructure in this data set. For these analyses there was no further drop-off in the chi squared test statistic for any of the SNPs examined after PC6 (PC7-PC10). Most of the drop-off was in PC1 and PC2, suggesting this is at least in part due to European ancestry north/south differences(18). Correction for population substructure is greatest for markers in which allele frequency differences correlate with one or more of the PCs identified by the genome-wide data. Markers without these sub-population allele frequency differences are relatively unaffected by these corrections. The RA-associated STAT4 SNP, rs7574865, is an example of a SNP for which the correction for population substructure does not reduce the evidence for association (5). We found, however, that after accounting for population stratification using principal components, the associations between RA and SNPs on 7q32 were no longer significant in our population.

In this study, RA cases and population controls were derived from the genetically diverse white North American population. The cases were recruited from centers across North America, whereas the controls were recruited only from New York. However, because both the case and control populations are derived from individuals of diverse European genetic backgrounds, it is unlikely that matching for geographic location would substantially reduce the genetic diversity or the chance for stratification. Our results add to mounting evidence that in genetic association studies in North American and other genetically complex populations, stratification should be expected, even when rigorous methods are used to match cases and controls. To properly interpret association results in these populations it is therefore necessary to apply methods that detect and correct for the stratification.

In contrast with the current study, after recent non-European migrants were excluded from the recent WTCCC study, in which the 7q SNP association was originally identified(9), there was little evidence for substructure as measured by extreme differences in SNP allele frequency in individuals from different geographic areas within the UK. Indeed, the uncorrected λ^GC in the British RA study was 1.03, compared with 1.43 in the NARAC GWAS, indicating the WTCCC study had a much more genetically homogeneous population. These observations led the investigators to conclude that it was unnecessary to correct the associations for stratification in their study. Although the allele frequency of this SNP did not exhibit extreme geographic variation (p<10⁻⁶), it would be interesting to determine whether a stratification correction, such as the principle components based correction used here, would nonetheless influence this particular association. Interestingly, the WTCCC identified SNP, rs11761231, has recently failed to replicate in a British RA replication collection (7).

The current study raises the issue of population stratification effects in case-control studies, particularly in complex populations. Because sample sizes used are often very large and the allele frequency differences detected are modest, even minor differences in the racial or ethnic makeup of cases and controls have the potential to create false positive associations reflective of these differences rather than disease-associated genetic differences between cases with disease and healthy controls. This observation helps explain the inherent difficulty in replicating candidate gene association studies performed in complex populations. It is not always easy to implement stratification correction given the requirement for WGA data or within Europe ancestry informative marker (19, 20)genotypes for each individual. However, controlling for stratification should not only avoid false positive associations, but should also increase the power to detect true associations.