A large-scale replication study identifies TNIP1, PRDM1, JAZF1, UHRF1BP1 and IL10 as risk loci for systemic lupus erythematosus

Abstract

Genome-wide association studies have recently identified at least 15 susceptibility loci for systemic lupus erythematosus (SLE). To confirm additional risk loci, we selected SNPs from 2,466 regions that showed nominal evidence of association to SLE (P < 0.05) in a genome-wide study and genotyped them in an independent sample of 1,963 cases and 4,329 controls. This replication effort identified five new SLE susceptibility loci (P < 5 × 10⁻⁸): TNIP1 (odds ratio (OR) = 1.27), PRDM1 (OR = 1.20), JAZF1 (OR = 1.20), UHRF1BP1 (OR = 1.17) and IL10 (OR = 1.19). We identified 21 additional candidate loci with P ≤ 1 × 10⁻⁵. A candidate screen of alleles previously associated with other autoimmune diseases suggested five loci (P < 1 × 10⁻³) that may contribute to SLE: IFIH1, CFB, CLEC16A, IL12B and SH2B3. These results expand the number of confirmed and candidate SLE susceptibility loci and implicate several key immunologic pathways in SLE pathogenesis.

Systemic lupus erythematosus (SLE) is a chronic inflammatory autoimmune disease characterized by the presence of antibodies to nuclear self-antigens. Many of the lupus autoantibodies recognize nucleic acids and nucleic acid binding proteins, which in turn activate Toll-like receptors, leading to the production of type I interferon¹. Despite considerable clinical heterogeneity, SLE ranks among the most heritable of common autoimmune diseases, with a sibling risk ratio of ~30 (ref. ²). Recent genome-wide association (GWA) and candidate gene studies have identified at least 15 common SLE risk alleles that achieve genome-wide significance (P < 5 × 10⁻⁸). These include genes encoding proteins important for adaptive immunity and the production of autoantibodies (HLA class II alleles, BLK, PTPN22 and BANK1) and proteins with roles in innate immunity and interferon signaling (ITGAM, TNFAIP3, STAT4 and IRF5)^3–10. To identify additional risk loci, we performed a targeted replication study of SNPs from 2,466 loci that showed a nominal P value of <0.05 in a recent GWA⁷ scan of 1,310 individuals with lupus (cases) and 7,859 controls. We also genotyped SNPs from 23 previously reported SLE risk loci, 42 SNPs implicated in other autoimmune diseases and over 7,000 ancestry-informative markers (Fig. 1).

Figure 1.

Overview of experimental design. Variants were selected from loci with P < 0.05 in a genome-wide scan of 1,310 cases and 7,859 controls, previously reported SLE risk loci, confirmed loci from other autoimmune diseases and over 7,000 ancestry-informative markers, and these variants were incorporated into an Illumina custom SNP array. The array was genotyped in independent cases and controls from the United States and Sweden. 823 of the Swedish controls were genotyped using the Illumina 310K SNP array. Variants were analyzed as described in Online Methods.

We designed a custom SNP array (Illumina Infinium II) consisting of over 12,000 variants and genotyped two independent SLE case and control populations from the United States (1,129 SLE cases and 2,991 controls) and Sweden (834 SLE cases and 1,338 controls). Included among the US controls were 2,215 Alzheimer’s disease case-control samples, which were judged to be acceptable as controls because the genetic factors underlying SLE and Alzheimer’s disease are expected to be independent. We next applied data quality filters to remove poorly performing samples and SNPs, population outliers, duplicate samples and related individuals (see Online Methods). Following these quality control measures, we examined a final set of 10,848 SNPs (Fig. 1). Association statistics for 3,735 SNPs were calculated and corrected for population stratification using 7,113 ancestry-informative markers (see Online Methods).

We first examined 25 SNPs (from 23 loci) that were previously reported to be associated with SLE (Table 1 and Supplementary Table 1). We found further evidence of association for 21 of the variants (P < 0.05), including 9 loci that reached genome-wide significance (P < 5 × 10⁻⁸) in the current combined dataset. Among the loci with genome-wide significant results were HLA-DRB1 (HLA*DR3 or DRB1*0301), IRF5, TNFAIP3, BLK, STAT4, ITGAM, PTPN22, PHRF1 (also called KIAA1542) and TNFSF4 (also called OX40L). The analysis also provided additional evidence of association for variants at nine loci for which a single previous study reported genome-wide levels of significance: HLA-DRB1 (HLA*DR2 or DRB1*1501), TNFAIP3 (rs6920220), BANK1, ATG5, PTTG1, PXK, FCGR2A, UBE2L3 and IRAK1-MECP2.

Table 1.

Replication results of previously reported SLE risk loci

SNP	Chr	Critical region	P				Gene of interest	Risk allele	Risk allele frequency	OR (95% CI)
			GWAS	US	Sweden	Combined
Variants with P < 5 × 10−8 in the current dataset
rs3135394a	6p21.32	32.027–32.874	7.8 × 10−22	1.8 × 10−26	8.3 × 10−21	2.0 × 10−60	HLA-DRB1b	G	0.10	1.98 (1.84–2.14)
rs7574865a	2q32.2	191.609–191.681	3.0 × 10−19	6.4 × 10−16	2.7 × 10−12	1.4 × 10−41	STAT4	T	0.23	1.57 (1.49–1.69)
rs2070197a	7q32.1	128.276–128.476	n.a.	1.4 × 10−16	4.1 × 10−9	5.8 × 10−24	IRF5	C	0.11	1.88 (1.78–1.95)
rs11860650a	16p11.2	31.195–31.277	5.3 × 10−11	1.8 × 10−5	9.2 × 10−8	1.9 × 10−20	ITGAM	T	0.13	1.43 (1.32–1.54)
rs2736340	8p23.1	11.331–11.488	5.5 × 10−8	4.6 × 10−9	0.0028	7.9 × 10−17	BLK	T	0.25	1.35 (1.27–1.43)
rs5029937a	6q23.3	138.174–138.284	1.0 × 10−4	2.4 × 10−7	3.1 × 10−5	5.3 × 10−13	TNFAIP3	T	0.03	1.71 (1.51–1.95)
rs2476601	1p13.2	113.963–114.251	3.3 × 10−5	4.5 × 10−5	1.5 × 10−5	3.4 × 10−12	PTPN22	A	0.10	1.35 (1.24–1.47)
rs4963128	11p15.5	0.485–0.664	0.0021	1.5 × 10−5	8.7 × 10−4	4.9 × 10−9	PHRF1	C	0.67	1.20 (1.13–1.27)
rs2205960	1q25.1	171.454–171.523	9.5 × 10−6	0.030	6.7 × 10−4	6.3 × 10−9	TNFSF4	T	0.23	1.22 (1.15–1.30)
Variants with a previous report of P < 5 × 10−8
rs9271366a	6p21.32	32.446–32.695	0.0079	7.4 × 10−4	8.3 × 10−5	1.4 × 10−7	HLA-DRB1c	G	0.16	1.26 (1.18–1.36)
rs6920220a	6q23.3	138.000–138.048	9.9 × 10−4	5.2 × 10−4	0.049	4.0 × 10−7	TNFAIP3	A	0.21	1.17 (1.10–1.25)
rs2269368	Xq28	152.743–152.943	2.5 × 10−5	n.a.	0.0049	7.5 × 10−7	IRAK1-MECP2	T	0.14	1.11 (1.01–1.22)
rs2431099	5q33.3	159.813–159.821	1.5 × 10−5	0.16	0.047	1.6 × 10−6	PTTG1	G	0.52	1.15 (1.09–1.22)
rs5754217	22q11.2	20.240–20.315	0.0060	8.4 × 10−4	0.018	2.3 × 10−6	UBE2L3	T	0.19	1.20 (1.13–1.27)
rs2245214a	6q21	106.749–106.876	0.032	4.3 × 10−6	0.35	1.2 × 10−5	ATG5	G	0.37	1.15 (1.09–1.21)
rs10516487	4q24	102.930–103.134	0.097	0.091	0.0015	8.3 × 10−4	BANK1	G	0.70	1.11 (1.04–1.18)
rs2176082a	3p14.3	58.214–58.443	0.010	0.012	0.0031	1.2 × 10−5	PXK	A	0.28	1.17 (1.10–1.25)
rs1801274	1q23.3	159.724–159.746	4.1 × 10−4	n.a.	n.a.	4.1 × 10−4	FCGR2A	G	0.50	1.16 (1.09–1.20)
Variants with a previous report of P > 5 × 10−8
rs280519a	19p13.2	10.387–10.430	7.1 × 10−4	n.a.	0.036	7.4 × 10−5	TYK2	A	0.48	1.13 (1.06–1.21)
rs10156091	7p21.3	8.134–8.154	0.095	0.0031	8.7 × 10−4	6.5 × 10−4	ICA1	T	0.10	1.16 (1.06–1.27)
rs2022013	1q25.3	181.538–181.670	0.26	2.05 × 10−5	2.8 × 10−4	0.0015	NMNAT2	T	0.60	1.09 (1.03–1.16)
rs7829816	8q12.1	56.985–57.025	0.49	0.76	0.19	0.17	LYN	A	0.79	1.05 (0.96–1.17)
rs2071725	22q13.2	41.908–41.970	0.63	0.34	0.29	0.30	SCUBE1	G	0.86	1.09 (0.98–1.20)
rs5744168a	1q41	n.a.	n.a.	1.00	0.40	0.67	TLR5	G	0.94	1.02 (0.94–1.12)
rs509749	1q23.3	158.993–159.067	0.64	0.94	0.93	0.76	LY9	G	0.96	1.01 (0.91–1.12)

Critical region here is defined as the minimal region containing variants with r² > 0.4 in the HapMap CEU population and is reported in HG18 coordinates (Mb). P values calculated from indicated case-control population (GWAS: 1,310 cases and 7,859 controls; US: 1,129 cases and 2,991 controls; Sweden: 834 cases and 1,338 controls; combined: 3,273 cases and 12,188 controls) and combined P values were calculated as described in Online Methods. Risk allele is reported relative to + reference strand. Risk allele frequency is the frequency in control chromosomes. OR is the combined odds ratio as described in Online Methods.

^aIndicates markers that were imputed, as described in Online Methods, in the GWAS samples and directly genotyped in the replication samples.

^brs3135394 has an r² = 0.87 to the HLA*DR3 (DRB1*0301) allele.

^crs9271366 has an r² = 0.97 to the HLA*DR2 (DRB1*1501) allele. See Supplementary Table 1 for expanded summary statistics. n.a, not available due to failure to pass quality control measures (TYK2, FCGR2A and IRAK1-MECP2), or the specific variant was not present in the genome-wide array (TLR5 and IRF5); however, rs2070197 (IRF5 region) is in strong LD with rs10488631, which had a P = 2 × 10⁻¹¹ in the genome scan.

An earlier candidate gene study⁹ identified MECP2 as a potential risk locus for SLE; however, in the current dataset, SNPs near IRAK1, a gene critical for Toll-like receptor 7 and 9 signaling and located within the identified region of linkage disequilibrium (LD) surrounding MECP2, showed the strongest evidence of association. Similar findings were recently reported¹¹, and further work will be required to determine the causal allele in the IRAK1-MECP2 locus. We found additional evidence of association for three loci (TYK2, ICA1 and NMNAT2) that had previously shown significant but not genome wide–level evidence for association^6,10. For four previously implicated variants (LYN, SCUBE1, TLR5 and LY9), no evidence of association was observed in the combined dataset.

To identify previously unknown SLE risk loci, we examined 3,188 SNPs from 2,446 distinct loci that showed evidence of association to SLE in our genome-wide dataset⁷, which comprised 502,033 SNPs genotyped in 1,310 SLE cases and an expanded set of 7,859 controls. Using this dataset, we imputed over 2.1 million variants using Phase II HapMap CEU samples as a reference (see Online Methods) and generated a rank-ordered list of association statistics. Variants with P < 0.05 were selected for possible inclusion on the custom replication array. For efficient genotyping, we identified groups of correlated variants (r² > 0.2) and then carried out selection of at least two SNPs from each group where all SNPs had P < 0.001. For the remaining groups, the SNP with the lowest P value in the group was included. In the replication samples, we calculated the association statistics (see Online Methods) and observed a significant enrichment of the replication results relative to the expected null distribution (Fig. 2). Excluding previously reported SLE risk alleles, there were 134 loci with P < 0.05 (64 expected; P = 2 × 10⁻¹⁵) and 12 loci with P < 0.001 (1 expected; P = 1 × 10⁻⁹), suggesting the presence of true positive associations.

Figure 2.

Newly discovered genome-wide significant associations in SLE. (a–e) Association results from the GWA scan are plotted on the y axis versus genomic position on the indicated chromosome on the x axis within a 500-kb region surrounding the loci defined by (a) TNIP1, (b) PRDM1, (c) JAZF1, (d) UHRF1BP1 and (e) IL10. The meta-analysis P value for the most strongly associated marker is indicated by a red square. P values from the genome scan are shaded to indicate LD to the genome-wide associated variant: red, r² > 0.8; yellow, r² > 0.5; green, r² > 0.2; gray, r² < 0.2. Along the bottom are the recombination rates from the CEU HapMap (light blue line) and the known human genes (blue). A previously reported and independent SLE risk locus at the nearby ATG5 gene is indicated (b; rs2245214). (f) Histogram of P values of 1,256 independent SNPs (r² < 0.1 to any other SNP in the array) in the 1,963 case and 4,329 control replication samples. Under a null distribution, the expected density of results is indicated by the dashed line. A significant enrichment of results with P < 0.05 was observed.

The replication study identified five new SLE risk loci with a combined P value that exceeded the genome-wide threshold for significance (P < 5 × 10⁻⁸): TNIP1, PRDM1, JAZF1, UHRF1BP1 and IL10 (Table 2 and Supplementary Table 2). These loci are discussed in more detail below.

Table 2.

Newly discovered SLE risk loci in the combined dataset

SNP	Chr.	Critical region	P				Gene of interest	Risk allele	Risk allele frequency	OR (95% CI)
			GWAS	US	Sweden	Combined
Genome-wide significant loci
rs7708392a	5	150.419–150.441	4.5 × 10−7	7.7 × 10−4	1.2 × 10−5	3.8 × 10−13	TNIP1	C	0.24	1.27 (1.10–1.35)
rs6568431	6	106.675–106.705	6.1 × 10−6	0.0016	0.0050	7.1 × 10−10	PRDM1	A	0.38	1.20 (1.14–1.27)
rs849142a	7	28.108–28.223	4.5 × 10−7	0.10	5.4 × 10−4	1.5 × 10−9	JAZF1	T	0.49	1.19 (1.13–1.26)
rs11755393a	6	34.658–35.090	0.0014	3.7 × 10−4	5.1 × 10−4	2.2 × 10−8	UHRF1BP1	G	0.35	1.17 (1.10–1.24)
rs3024505	1	205.007–205.016	2.6 × 10−6	0.062	1.8 × 10−4	4.0 × 10−8	IL10	A	0.16	1.19 (1.11–1.28)
Loci with combined P value < 1 × 10−6
rs10911363a	1	181.672–181.816	2.0 × 10−4	1.5 × 10−5	0.52	9.5 × 10−8	NCF2	T	0.27	1.19 (1.12–1.26)
rs12444486a	16	84.548–84.576	3.5 × 10−5	0.021	0.026	1.9 × 10−7	IRF8	T	0.50	1.16 (1.10–1.23)
rs11013210a	10	23.181–23.337	1.6 × 10−5	0.013	0.12	2.0 × 10−7	ARMC3	T	0.21	1.18 (1.11–1.26)
rs1874791a	1	67.563–67.687	3.1 × 10−5	0.012	0.11	3.4 × 10−7	IL12RB2	A	0.18	1.18 (1.10–1.26)
rs9782955	1	233.893–234.107	6.4 × 10−6	0.057	0.12	4.6 × 10−7	LYST	C	0.74	1.18 (1.11–1.26)
rs7683537a	4	185.805–185.914	1.6 × 10−4	0.11	0.0013	7.6 × 10−7	MLF1IP	T	0.82	1.23 (1.14–1.33)
rs428073	12	117.706–117.315	1.7 × 10−5	0.22	0.0079	7.7 × 10−7	TAOK3	T	0.69	1.18 (1.11–1.26)
rs497273a	12	119.610–119.891	5.0 × 10−5	0.068	0.021	8.2 × 10−7	SPPL3	G	0.65	1.14 (1.08–1.21)
Loci with combined P value < 1 × 10−5
rs1861525	7	25.097–25.183	8.5 × 10−5	0.16	0.0027	1.9 × 10−6	CYCS	G	0.05	1.27 (1.12–1.45)
rs921916	7	50.193–50.205	4.8 × 10−4	0.027	0.014	2.0 × 10−6	IKZF1	C	0.18	1.15 (1.07–1.23)
rs7333671	13	73.177–73.198	2.2 × 10−4	0.14	0.0027	2.2 × 10−6	KLF12	G	0.08	1.22 (1.11–1.34)
rs12992463	2	22.312–22.464	2.1 × 10−5	0.23	0.023	2.6 × 10−6	–	A	0.50	1.12 (1.06–1.19)
rs12620999	2	237.616–237.770	1.6 × 10−5	0.040	0.45	3.1 × 10−6	COPS8	C	0.19	1.13 (1.06–1.21)
rs503425a	11	118.079–118.198	0.0012	3.3 × 10−4	0.43	3.3 × 10−6	DDX6	C	0.20	1.16 (1.08–1.24)
rs10742326a	11	34.733–34.809	1.4 × 10−4	0.017	0.21	3.6 × 10−6	APIP	G	0.59	1.14 (1.08–1.21)
rs4766921a	12	117.835–117.883	4.6 × 10−5	n.a.	0.036	4.6 × 10−6	KIAA1853	G	0.67	1.18 (1.09–1.27)
rs11951576a	5	6.741–6.866	2.5 × 10−5	0.42	0.014	4.6 × 10−6	POLS-SRD5A	C	0.69	1.14 (1.08–1.22)
rs6438700	3	123.355–123.454	7.4 × 10−5	0.23	0.020	5.5 × 10−6	CD86	C	0.82	1.18 (1.09–1.27)
rs6486730a	12	127.830–127.840	8.2 × 10−5	0.16	0.049	6.9 × 10−6	SLC15A4	G	0.41	1.13 (1.07–1.19)
rs4748857a	10	23.529–23.654	2.2 × 10−4	0.68	1.3 × 10−4	6.9 × 10−6	C10orf67	C	0.73	1.16 (1.09–1.24)
rs3914167a	5	39.426–39.454	1.8 × 10−4	0.24	0.0081	7.6 × 10−6	DAB2-C9	G	0.27	1.15 (1.09–1.23)

Samples, critical region, P values, risk alleles and ORs are as defined in the Table 1 legend.

^aIndicates markers that were imputed, as described in Online Methods, from the GWAS samples and directly genotyped in the replication samples. See Supplementary Table 2 for expanded summary statistics.

A variant (rs7708392) on 5q33.1 that resides within an intron of TNIP1 (encoding TNF-α-induced protein 3 (TNFAIP3)-interacting protein 1) was significantly associated with SLE in all three cohorts and had a combined P = 3.8 × 10⁻¹³ (Fig. 2). Variants near TNIP1 were recently found to contribute to risk of psoriasis¹²; however, the SLE and psoriasis risk variants are separated by 21 kb and appear to have distinct genetic signals (r² = 0.001). TNIP1 and TNFAIP3 are interacting proteins¹³, but the precise role of TNIP1 in regulating TNFAIP3 is unknown. The association of multiple distinct variants near TNFAIP3 with SLE^4,14, rheumatoid arthritis¹⁵, psoriasis¹² and type 1 diabetes¹⁶ suggests that this pathway has an important role in regulating autoimmunity.

A second confirmed risk variant (rs6568431, P = 7.12 × 10⁻¹⁰) was identified in an intergenic region between PRDM1 (PR domain containing 1, with ZNF domain, also known as BLIMP1) and ATG5 (APG5 autophagy 5-like). The signal at rs6568431 appears to be distinct from the previously reported⁶ SLE risk allele within ATG5 (rs2245214, Table 1), as rs6568431 has an r² < 0.1 with rs2245214, and rs2245214 remains significantly associated with SLE (P < 1 × 10⁻⁵) after conditional logistic regression incorporating rs6568431 (Fig. 2).

The promoter region of JAZF1 (juxtaposed with another zinc finger gene 1) is a third newly confirmed SLE locus (rs849142, P = 1.54 × 10⁻⁹). Of note, this same variant was previously associated with risk of type 2 diabetes¹⁷ and with height variation¹⁸. A separate prostate cancer risk allele near JAZF1 (rs10486567)¹⁹ showed no evidence for association in the current study.

A fourth newly identified SLE risk locus is defined by a nonsynonymous allele (R454Q) of UHRF1BP1 (ICBP90 binding protein 1; rs11755393, P = 2.22 × 10⁻⁸). This allele encodes a nonconservative amino-acid change in a putative binding partner of UHRF1, a transcription and methylation factor linked to multiple pathways²⁰. The UHRF1BP1 risk allele is in a region of extended LD that encompasses multiple genes, including SNRPC (small nuclear ribonucleoprotein polypeptide C), which is part of a RNA processing complex often targeted by SLE autoantibodies.

The fifth newly identified SLE locus is IL10 (interleukin-10; rs3024505, P = 3.95 × 10⁻⁸; Fig. 2). IL10 is an important immunoregulatory cytokine that functions to downregulate immune responses²¹, and variation in IL10 has inconsistently been reported to be associated with SLE²². The variant associated with SLE is identical to a SNP recently identified as contributing to risk of ulcerative colitis²³ and type 1 diabetes²⁴, suggesting the possibility of shared pathophysiology in the IL10 pathway across these disorders.

Using a significance threshold of P < 1 × 10⁻⁵ in the combined replication sample, we identified 21 additional SLE candidate risk loci (Table 2 and Supplementary Table 2). Less than one locus (0.01 loci, specifically) with P < 1 × 10⁻⁵ was expected under a null distribution for the meta-analysis (P = 8 × 10⁻⁷⁷), suggesting that several of these loci are likely to be true positives for association to SLE. Notable candidate genes in this list include: (i) IRF8 (interferon regulatory factor 8), which was implicated in a previous GWA study (GWAS)⁴ and whose family members IRF5 and IRF7 are within confirmed SLE risk loci; (ii) TAOK3 (TAO kinase 3), a kinase expressed in lymphocytes, and the disease-associated variant is a missense allele (rs428073, N47S); (iii) LYST (lysosomal trafficking regulator), mutations of which cause Chediak-Higashi syndrome in humans, a complex disorder characterized by a lymphoproliferative disorder; and (iv) IL12RB2 (interleukin 12 receptor, beta 2), a locus which includes IL23R and SERPBP1 but appears distinct from the IL23R variants reported in inflammatory bowel disease, psoriasis and ankylosing spondylitis²⁵.

A noteworthy feature of recent GWAS is the large number of overlapping loci found to be shared between different complex diseases²⁶. We tested 42 variants from 35 loci that were previously reported as autoimmune-disease risk alleles for association with SLE (Table 3 and Supplementary Table 3). No single locus had an unadjusted P value < 5 × 10⁻⁸; however, we found an enrichment of previously identified disease-associated alleles. From the 35 loci tested (42 total variants), there were 5 alleles with unadjusted P < 0.0004 (less than 1 expected by chance, P = 4.4 × 10⁻¹²) and with P < 0.05 after a Bonferroni correction for the 35 pre-specified loci. For each of the five variants, the SLE-associated allele matched a previously reported allele and had the same direction of effect (Table 3). We observed a highly significant association to SLE of a missense allele of IFIH1 (rs1990760, P = 3.3 × 10⁻⁷) that has previously been associated with type 1 diabetes and Graves’ disease^27,28. We also observed an association of SLE with a missense allele (R32Q) of CFB (complement factor B, rs641153) that resides in the HLA class III region and is a validated risk allele for age-related macular degeneration²⁹. This missense allele in CFB is not in significant LD with other HLA region variants associated with SLE (DR2/DR3) and remained significant (P < 0.05) after conditional logistic regression analyses that incorporated DR2 and DR3. The HLA is a complex genetic region, but it is noteworthy that the rs641153 SNP has a protective effect nearly identical to that of the reported age-related macular degeneration (AMD) risk allele²⁹. Further validation of the five candidate disease alleles is required.

Table 3.

Candidate autoimmune loci with evidence of association to SLE

SNP	Gene	Chr	P				Combined corrected	Risk allele	Risk allele frequency	OR	Phenotype	References
			GWAS	US	Sweden	Combined
rs1990760	IFIH1	2	3.2 × 10−5	0.015	0.0039	3.34 × 10−7	1.12 × 10−5	T	0.60	1.17	T1D, Graves’	27,28
rs641153a	CFB	6	0.0079	n.a.	0.0011	1.4 × 10−4	0.0049	G	0.91	1.30	AMD	29
rs12708716a	CLEC16A	16	0.15	1.3 × 10−4	0.062	1.6 × 10−4	0.0056	A	0.64	1.16	T1D, Addison’s, MS	32–34
rs6887695a	IL12B	5	0.014	0.04	0.03	1.7 × 10−4	0.0060	G	0.68	1.13	Psoriasis, IBD	12,35
rs17696736	SH2B3	12	0.0036	0.12	0.19	4.0 × 10−4	0.014	T	0.50	1.08	T1D, Celiac, SLE	33,36,37

All alleles in the table either were identical to the reported variants or have r² > 0.8 to the reported variant and were the same risk allele with the same direction of effect. Samples, individual and combined P values, risk allele frequency and OR are as described in Table 1 legend. Combined-corrected P value is the Bonferroni-corrected P value for the 35 previously reported risk loci. Other autoimmunity associations: T1D, type 1 diabetes; AMD, age-related macular degeneration; MS, multiple sclerosis; IBD, inflammatory bowel disease. See Supplementary Table 3 for expanded summary statistics and a complete list of variants tested.

^aIndicates markers that were imputed, as described in Online Methods, from the GWAS samples and directly genotyped in the replication samples.

Using 26 SLE risk alleles (21 previously reported loci in Table 1 plus the 5 newly identified SLE loci), we performed several additional analyses. First, we performed pairwise interaction analysis with the previously confirmed loci, and, consistent with previous literature from SLE⁶ and other complex diseases³⁰, we observed no evidence for non-additive interactions. Using conditional logistic regression analyses, we found no evidence for multiple independent alleles contributing to risk at any of the individual risk loci. We next estimated the percent of variance explained by each of the confirmed SLE risk alleles using previously described methods³⁰. HLA-DRB1 (HLA*DR3), IRF5 and STAT4 were each estimated to account for over 1% of the genetic variance, whereas the remaining loci each accounted for less than 1% of the variance. Together, the 26 SLE risk loci explain an estimated 8% of the total genetic susceptibility to SLE.

Targeted replication of GWAS results is an efficient study design to confirm additional risk loci³¹. However, there are few available data as to the probability of replicating results that fall short of accepted P value criteria for genome-wide significance. In the current study, all variants with P < 0.05 from the original GWAS were included for replication. The lower a locus’ P value is in the GWAS, the higher is the probability of that locus reaching candidate or confirmed status in the replication meta-analysis (Fig. 3). Of note, no candidate or confirmed loci were obtained in the current study from the group of variants with a GWAS P value between 0.05 and 0.01, despite accounting for ~50% of all variants tested in the replication. These results may be useful in helping guide future targeted study designs, although clearly the size of the original GWAS population, the replication sample size, the disease architecture and the effect size of the candidate disease-associated variants need to be carefully considered in planning replication efforts.

Figure 3.

Percentage of newly discovered variants reaching candidate (P < 1 × 10⁻⁵) and confirmed (P < 5 × 10⁻⁸) status in the meta-analysis stratified by the P value in the original GWAS.

These data provide further evidence that common variation in genes that function in the adaptive and innate arms of the immune system are important in establishing SLE risk. Although each of the identified alleles accounts for only a fraction of the overall genetic risk, these and other ongoing studies are providing new insight into the pathogenesis of lupus and are suggesting new targets and pathways for drug discovery and development.

METHODS

Methods and any associated references are available in the online version of the paper at http://www.nature.com/naturegenetics/.