Genetic influences on susceptibility to rheumatoid arthritis in African-Americans

Abstract

Large meta-analyses of rheumatoid arthritis (RA) susceptibility in European (EUR) and East Asian (EAS) populations have identified >100 RA risk loci, but genome-wide studies of RA in African-Americans (AAs) are absent. To address this disparity, we performed an analysis of 916 AA RA patients and 1392 controls and aggregated our data with genotyping data from >100 000 EUR and Asian RA patients and controls. We identified two novel risk loci that appear to be specific to AAs: GPC5 and RBFOX1 (PAA < 5 × 10-9). Most RA risk loci are shared across different ethnicities, but among discordant loci, we observed strong enrichment of variants having large effect sizes. We found strong evidence of effect concordance for only 3 of the 21 largest effect index variants in EURs. We used the trans-ethnic fine-mapping algorithm PAINTOR3 to prioritize risk variants in >90 RA risk loci. Addition of AA data to those of EUR and EAS descent enabled identification of seven novel high-confidence candidate pathogenic variants (defined by posterior probability > 0.8). In summary, our trans-ethnic analyses are the first to include AAs, identified several new RA risk loci and point to candidate pathogenic variants that may underlie this common autoimmune disease. These findings may lead to better ways to diagnose or stratify treatment approaches in RA.

Figure 1

Consort diagram showing the phases and flow of the study. Phase 1 is a two-part GWAS of RA in AAs, which was jointly analyzed in a fixed-effects meta-analysis. Phase 2 is a TEMA, in which we combine the association summary statistics from our AA data (Phase 1) with those provided by Okada et al. (3) to conduct a random effects meta-analysis using METASOFT. In addition to computing association P-values, we generated M-values (see Materials and Methods). Phase 3 uses the TEMA data along with publicly available data to perform TEFM. We then prioritized candidate pathogenics variants by posterior probability and constructed credible sets for RA risk loci.

Figure 2

(A) Manhattan plot P-values from association testing of AAs with RA using calculated from the random effects meta-analysis conducted in Phase 2. Only P_AA values (association P-values generated using AA genotypes) are displayed in this plot. The x-axis indicates chromosome and position; the y-axis indicates –log₁₀(p). The horizontal red line is drawn at 5.0 × 10⁻⁹ and is the threshold for genomewide significance; the blue line is for suggestive evidence of association and is drawn at 1.0 × 10⁻⁶. Figure 2B–D is a locus zoom plot for each of the non-HLA associations detected in AAs. (B) chr8:114 980 000–115 150 000 (nearest gene: CSMD3). (C) chr13:92 900 000–93 050 000 in an intronic region of GPC5. Red coloration indicates the variants are more strongly linked to rs9516053, while blue coloration indicates variants more strongly linked to rs9589512. (D) chr16: 5 538 689–5 638 S689 in an intronic region of RBFOX1. (E) The independence of the associations of the PADI2 and PADI4 loci is shown. The y-axis is –log₁₀(combined P-value) from the TEMA of all three populations, and the x-axis is genomic position. Red coloration indicates the variants are more strongly linked to rs761426 (the index variant in PADI2) than to rs2301888 (the index variant in PADI4). For each of the two loci, darker shading indicates stronger LD. The blue line indicates genomic recombination rate; peaks of recombination may explain the close proximity of two independent associations.

Figure 3

(A–C) Plots of association P-values for RA in EURs, EASs and AAs (y-axis) and M-values (posterior probabilities that a genetic variant has an effect in a given population) (x-axis) in EURs (A) Asians, (B) and AAs (C). Larger values indicate increased likelihood the variant has an effect in the population plotted. Variants with M > 0.8 are colored blue and are considered to have evidence supporting effect replication in that population. Variants with M < 0.2 are colored red and are considered to have evidence against replication. Note that in (A) and (B) the HLA-DRB1 locus is shown at the top of the graph because it has P < 1 × 10⁻²⁵⁰ in EUR ancestry and < 1 × 10⁻¹³³ in Asian ancestry (see Supplementary Material, Table 2). (D and E) Scatter plots of effect size of RA risk loci in EURs versus M_AA (measure of evidence for association the same locus in AAs) (D) and EASs (E). EURs are used as the comparator because studies on this population have been larger than others to date. Loci are categorized as concordant (blue), discordant (red) or uncertain (gray). The color saturation is proportional to increasing MAF in EURs, while dot size is proportional to the MAF in the population analyzed. All ORs forced to be >1 by first taking the absolute value of beta then exponentiating. This is done in order to enable simple visual comparison of effect size.

Figure 4

 (A) Scatter plots of the association (−log₁₀P-value) of genetic variants in the CTLA4 locus (chr2:204 689 000–204 789 000) in AAs (left), EASs (middle) and EUR (right) to RA susceptibility. Below each scatter plot is a heatmap of LD between variants (r²) in each population. (B) Plot of the posterior probability of pathogenicity for each variant found within chr2:204 689 000–204 789 000. The two variants in red belong to the 90% credible set and are also shown in (C) The other variants are colored in blue. The region highlighted in the yellow panel represents an expanded view of the yellow vertical line in (C). (C) Gene diagram of the region containing the most likely candidate pathogenic variants. The top track is a partial gene diagram of CTLA4 showing Intron 3 and exon 4, the 3′UTR and intergenic region downstream. There is a dinucleotide repeat in the 3′UTR (marked (AT)₂₈) that is strongly linked with rs3087243. The (AT)₂₈ variant of this short tandem repeat is associated with decreased CTLA4 mRNA levels in autoreactive T-cell lines. The middle track shows the ENCODE TFBS. Variants rs3087243 and rs11571302 are the same as the two SNPs in the credible set shown in (B). The bottom track shows raw DNAse hypersensitivity signal in Tregs (blue), Th17 cells (brown), Th1 cells (yellow) and naïve T cells (purple).