Association analyses identify six new psoriasis susceptibility loci in the Chinese population

Psoriasis is a chronic inflammatory hyperproliferative skin disease affecting up to 3% of the population, with considerable ethnic variation. Family-based analyses and population-based epidemiological studies have confirmed the genetic basis of psoriasis¹ and suggested a genetic influence upon its age at onset². In addition to earlier linkage analyses, genome-wide and candidate gene association studies have implicated many genomic regions in the pathogenesis of psoriasis^3–6, not all of which have been confirmed. However, these genetic signals do not fully account for the observed variation in genetic liability to psoriasis, suggesting that additional genetic factors remain to be discovered. Furthermore, differences in the results of similarly-powered GWAS of psoriasis in Chinese⁵ and European⁴ populations suggest the existence of allelic and/or locus heterogeneity between different populations.

To explore additional susceptibility factors for psoriasis, we performed a multistage association study based on our first-stage GWAS (including our initial GWAS samples reported in ref. 5 ⁵ and an additional 95 healthy controls, see Online Methods), followed by replication in 6 independent cohorts consisting of Chinese Han (Replications 1 and 2), Chinese Uygur (Replication 3), as well as European population from Germany (Replication 4) and the United States (Replications 5 and 6). We further genotyped SNPs from previously reported loci in European population in the samples of Replication 2 to investigate the possibility of genetic heterogeneity of disease susceptibility between different populations. The experimental design is summarized in Figure 1.

In Replication 1, 61 SNPs from 51 regions were selected in the GWAS dataset (see Online Methods), and genotyped in an independent sample of 4,610 cases and 5,373 controls from Chinese Han population (Supplementary Table 1). Combined analysis (GWAS and Replication 1) showed that 4 SNPs met the genome-wide significant threshold (P<5.0×10⁻⁸): rs3762999 at 5q33.1 (TNIP1/ANXA6), P_Combined=1.1×10⁻¹², OR=1.24; rs999556 at 5q33.1 (TNIP1/ANXA6), P_Combined=4.3×10⁻¹³, OR=1.24; rs2431697 at 5q33.3 (PTTG1), P_Combined=1.1×10⁻⁸, OR=1.20; rs3751385 at 13q12.11 (GJB2), P_Combined=1.7×10⁻¹⁰, OR=0.85.

After applying standard quality-control procedures as described elsewhere⁵, 20 SNPs with P<0.05 from Replication 1 were genotyped in another independent cohort of 2,024 cases and 5,495 controls of Chinese Han (Replication 2, Supplementary Table 1). In the joint association analysis using samples from GWAS, Replication 1 and Replication 2, 9 SNPs annotated in 7 loci reached genome-wide significance: rs151823 at 5q15 (ERAP1), P_Combined=9.3×10⁻⁹; rs999556/rs3762999 at 5q33.1 (TNIP1/ANXA6), P_Combined=3.8×10⁻²¹ and P_Combined=4.6×10⁻¹⁸; rs2431697 at 5q33.3 (PTTG1), P_Combined=1.1×10⁻⁸; rs10088247/rs7007032 at 8p23.2 (CSMD1), P_Combined=4.5×10⁻⁹ and P_Combined=3.8×10⁻⁸; rs3751385 at 13q12.11 (GJB2), P_Combined=8.6×10⁻⁸ (P_{GWAS+Replication1}=1.7×10⁻¹⁰); rs514315 at 18q22.1 (SERPINB8), P_Combined=5.9×10⁻⁹; and rs9304742 at 19q13.41 (ZNF816A), P_Combined=2.1×10⁻⁹ (Table 1). All of these 9 SNPs showed consistent association evidence independently in each replication panel and in the combined analysis of Chinese Han. The association at rs3751385, however, showed significant heterogeneity (I² =50%) between the GWAS/replication 1 and 2 samples (Supplementary Table 2). Consequently, the analysis of the combined GWAS+R1+R2 samples (under a random effect model) yielded less significant association (P_Combined=8.57×10⁻⁸) than the analysis of the GWAS+R1 samples (P_Combined=1.7×10⁻¹⁰). The association at rs3751385 in the GWAS+R1 samples satisfied the genome-wide criteria for significance, although the association in the GWAS+R1+R2 samples was barely below the genome-wide significance. In addition, SNP rs2431697 reached genome-wide significance (P=1.11×10⁻⁸) in the combined results of GWAS and Replication 1, although its genotyping failed in Replication 2. Association results of the other 52 SNPs are summarized in Supplementary Table 3.

Of the 9 SNPs, two (rs999556 and rs3762999) in strong linkage disequilibrium (LD) with each other (r²=0.97) were located in the same region (TNIP1/ANXA6) as the SNP rs17728338, which is associated with psoriasis in European population⁴. Multivariate logistic regression analysis showed that the association of rs999556 and rs3762999 was dependent upon association at rs17728338 in our GWAS dataset, providing evidence that all three SNPs detect the same disease association in the Chinese Han population. To address this question in European population, we performed multivariate logistic regression analysis using two European GWAS samples that are, in part, subsets of replication 4 and 5. The results showed that of these three SNPs, only SNP rs17728338 was strongly associated with psoriasis in the European population (P=1.7×10⁻⁸), SNP rs999556 was not associated (P=0.96), SNP rs3762999 was also not associated (P=0.062), and the association with SNPs rs999556 and rs3762999 was dependent upon the association with rs17728338. Thus although LD in this region is apparently more extensive in Chinese than in European population, SNP rs17728338 is the best known tag for the TNIP1/ANXA6 susceptibility locus for psoriasis in both populations.

To further assess the impact of the 6 new loci on genetic risk for psoriasis in diverse populations, we analyzed genotyping data from four additional cohorts: Chinese Uygur (Replication 3: 539 cases, 824 controls), Germany (Replication 4: 823 cases, 1,840 controls) and the U.S. (Replication 5: 2,470 cases, 2,348 controls and Replication 6: 254 nuclear families described elsewhere^7–9). Interestingly, the C allele of rs9304742 in the ZNF816A region showed a consistent protective effect among Chinese Han (P=2.1×10⁻⁹, OR=0.88), Chinese Uygur (P=1.7×10⁻³, OR=0.77), and Germans (P=7.9×10⁻³, OR=0.84); suggestive evidence for a protective effect of the A allele of rs11084211 (which is in strong LD with the C allele of rs9304742) was also seen in the U.S. cohort from Replication 5 (P=1.6×10⁻², OR=0.90) (Table 2). In addition, rs151823 at 5q15 (ERAP1) showed consistent association (P=2.9×10⁻⁵, OR=0.69) in the Chinese Uygur sample, and rs3751385 at 13q12.11 (GJB2) showed consistent association (P=3.6×10⁻³, OR=0.79) in the German sample. However, in the combined analysis of the Germany and US case-control samples (see Online Methods), none of these newly-associated SNPs conferred significant risk for psoriasis, although it was not possible to test association with ZNF816A (Table 2). Indeed, all the six loci showed significant heterogeneity between the Chinese and Caucasian samples (I²>0.40, Supplementary Table 4).

Because the PTTG1 SNP rs2431697 is located fairly close to IL12B (within ~1.1 Mb), we also carried out a multivariate logistic regression analysis conditioning on the genetic effect of both rs7709212 and rs3213094 within the IL12B locus in our initial GWAS, and found that rs2431697 still showed significant association in the Chinese Han population (P=2.7×10⁻³). This suggests that effects of both loci are independent. In contrast, no association was seen for PTTG1 in the European studies (Table 2), whereas IL12B is strongly associated⁴.

In an attempt to uncover susceptibility genes underlying each of the association signals, we investigated the patterns of recombination and LD around the associated SNPs and identified the gene(s) located within each region of LD (Supplementary Fig. 1). Of these 9 SNPs, 7 were within 6 genes or LD blocks at 5q15 (ERAP1), 5q33.3 (PTTG1), 8p23.2 (CSMD1), 13q12.11 (GJB2), 18q22.1 (SERPINB8), 19q13.41 (ZNF816A), where a single gene was found within the LD block harboring the association. The remaining two SNPs located at 5q33.1, where two genes (TNIP1 and ANXA6) were identified in the LD region. TNIP1 has been considered as a susceptibility gene for psoriasis in European populations due to its important role in working downstream of TNF-α to negatively regulate NF-κB⁴ and the fact that its encoded protein binds to the product of another psoriasis gene (TNFAIP3). However, ANXA6 might also be implicated in the pathogenesis of psoriasis via early and late stages of epidermal growth factor receptor (EGFR) downregulation¹⁰.

To further explore the potential biological implications of the other six newly discovered susceptibility genes (ERAP1, PTTG1, CSMD1, GJB2, SERPINB8, ZNF816A), we performed a literature search. ERAP1 (Endoplasmic reticulum aminopeptidase 1) is an interferon-γ-induced aminopeptidase with several proposed biological functions in the endoplasmic reticulum, including the trimming of peptide antigens to optimal length for binding to MHC class I molecules¹¹. Furthermore, ERAP1 is also associated with ankylosing spondylitis, another major Class I-associated autoimmune disease, in both Caucasian and Chinese populations^12,13. Given that HLA-Cw6 is much more strongly associated with Type I than with late-onset (Type II) psoriasis in both Chinese and European population¹⁴, it is of interest that ERAP1 was preferentially associated with Type I psoriasis in the combined Chinese Han datasets (rs151823, P=6.5×10⁻³, OR=0.88, Supplementary Table 5).

PTTG1 (Pituitary tumor transforming gene) encodes a multifunctional protein with roles in the control of mitosis, cell transformation, DNA repair, and gene regulation¹⁵, with in vitro angiogenic activity¹⁶, at least in part through the regulation of VEGF¹⁷, which in turn has been reported to be overexpressed psoriasis plaques¹⁸. Notably, the TNFAIP3, TNIP1, and PTTG1 regions are now genetically implicated in both psoriasis and systemic lupus erythematosus (SLE)^4,19,20.

CSMD1 (CUB and Sushi multiple domains 1) is a tumor suppressor gene expressed in areas of regenerative growth such as skin and epithelial cells²¹. Higher expression levels of CSMD1 correlated with tumor differentiation, suggesting that it could contribute to the pathogenesis of psoriasis by influencing keratinocyte differentiation²².

Connexins form gap junctions, which play an important role in regulating homeostasis and differentiation in many tissues²³. Encoded by GJB2 (Gap junction protein beta 2), connexin 26 is highly up-regulated in psoriasis lesions²⁴ and a transgenic mouse overexpressing GJB2 in suprabasal keratinocytes exhibits several features of psoriasis²⁵.

SERPINB8 (Serpin peptidase inhibitor clade B member 8) is one of a family of serine protease inhibitors and is up-regulated in psoriasis lesions²⁶. It has been suggested that this family of protease inhibitors contributes to the pathophysiology of psoriasis and other common inflammatory skin diseases^27,28.

ZNF816A (Zinc finger protein 816A) encodes a zinc-finger protein. Zinc finger domains function as specific modules for protein recognition, and have multiple regulatory functions, such as the recognition of RNA and proteins²⁹. Furthermore, ZNF816A belongs to the same functional class of proteins as ZNF313 (Zinc finger protein 313), which was recently identified as a novel psoriasis susceptibility gene³⁰. Like ZNF313, ZNF816A appears to be associated with Type I (early onset) psoriasis in the combined Chinese Han datasets (rs9304742, P=1.5×10⁻³, OR=0.85, Supplementary Table 5).

Along with genome-wide exploration, we further investigated 15 SNPs from 8 loci reported to be associated with psoriasis in European populations⁴ in Replication 2 (see Online Methods). In addition to TNIP1 rs17728338 highly associated with psoriasis in the combined Chinese Han data (GWAS and Replication 2: 3,163 cases and 6,722 controls), only three genes, IL23R, IL13 and TNFAIP3, showed suggestive evidence for association (P_Combined=4.2×10⁻⁵, OR=1.23 for IL23R; P_Combined=1.0×10⁻³, OR=0.90 for IL13 and P_Combined=3.7×10⁻³, OR=1.16 for TNFAIP3). No nominal associations were observed for the remaining 8 SNPs within IL1RN, TSC1, IL23A/STAT2 and SMARCA4 (Supplementary Table 6). In most cases, the minor alleles and their allele frequencies (Supplementary Table 6) were similar between the European and Chinese Han populations, suggesting genetic heterogeneity of psoriasis within the two ethnic groups. However, it is possible that the power of the sample was not sufficient to detect risk SNPs with a low minor allele frequency in the Chinese Han population, as is likely to be the case for rs2066808 and rs12983316 (MAF<0.05, Supplementary Table 6). It is also possible that differences in patterns of linkage disequilibrium account for the ethnic diversity, since some of the associated variants identified in European population were not sufficiently tagged by any SNP screened in our GWAS.

This study was designed to maximize statistical power in a cost-effective manner by adopting a multistage analysis strategy in several large Chinese and European population samples, resulting in the identification of six novel susceptibility loci. Our findings increase the number of genetic risk factors of psoriasis, some of which have also been implicated in other autoimmune diseases. Our results also highlight new and plausible biological pathways in psoriasis, suggest additional genetic factors that may contribute to its age at onset, and provide insight into genetic heterogeneity of psoriasis across different ethnic populations. Further studies using large samples drawn from diverse ethnic populations will be required to provide a more comprehensive understanding of the global genetics of psoriasis.