Age-related macular degeneration (AMD) is a progressive, blinding disease affecting millions of elderly individuals worldwide1 2. Several genome-wide association studies (GWAS) have identified common variants associated with AMD in European-ancestry populations3 4 5 6, and recently, rare genetic variation at CFH, CFI, C3 and C9 were also shown to strongly associate with AMD in Europeans7 8 9 10. However, there are few such studies in Asians11. Importantly, Asians appear to have a distinct clinical presentation of the disease (for example, absence of drusen and minimal fibrous scarring in polypoidal choroidal vasculopathy, a variant of AMD accounting for 20–55% of Asian patients with exudative AMD) and different responses to treatment (for example, poorer response to inhibitors of vascular endothelial growth factor (VEGF) compared with patients of European ancestry)12 13. It remains unclear whether there are differences in underlying genetic characteristics of AMD between patients of Asian versus European ancestry. Concurrently, previous GWAS studies provide limited coverage of low-frequency coding variants, which may result in the loss of function and are often ethnic-specific. There is thus interest in genetic studies of AMD and other diseases beyond standard-content GWAS to discover potentially causative coding variants in different ethnic groups. To address these questions, the Genetics of AMD in Asians Consortium perform a genome-wide (GWAS) and exome-wide association study (EWAS) of advanced AMD solely on the exudative (neovascular) disease subtype in East Asians. Compared with standard-content GWAS arrays, the exome array has significantly increased marker density across the coding human exome, thus increasing power to detect disease associations located within the coding frame. EWAS of AMD have not been previously conducted in either Europeans or Asians. In this paper, we present data from eight independent AMD case–control collections enrolled across multiple sites in East Asia, totalling 6,345 exudative AMD cases and 15,980 controls. This is the largest sample, to our knowledge, of East Asians ever assembled for genetic studies of AMD. Results Association with previously identified AMD variants After genotype imputation, synchronization and stringent quality filters were performed, a total of 4,471,719 SNPs were assessed for the GWAS and 120,027 autosomal coding-frame SNPs for EWAS from 2,119 AMD cases and 5,691 controls (Table 1). Overall genomic inflation was very low (λ gc=1.031; Supplementary Fig. 1), suggesting minimal confounding of the disease association analysis by population stratification or other systematic study design biases. Data from the discovery stage analysis confirmed previously identified AMD variants in ARMS2-HTRA1 rs10490924 (P=1.20 × 10−103), CFH rs10737680 (P=7.54 × 10−38), CETP rs3764261 (P=1.66 × 10−12), ADAMTS9 rs6795735 (P=1.13 × 10−5), C2-CFB rs429608 (P=1.06 × 10−4), as well as CFI rs4698775 (P=7.5 × 10−4; Supplementary Table 1 and Supplementary Fig. 2). Our data also showed nominal evidence of replication in the same direction as the initial study for a further three previously reported variants (TGFBR1 rs334353, APOE rs4420638, and VEGFA rs943080; P 0.15, Supplementary Table 2). It is thus unsurprising that the European GWAS efforts have yet to detect this locus. VEGFA rs943080, a marker strongly associated with AMD in European-ancestry populations5, is in the vicinity of C6orf223, but its association with AMD is much weaker in this study of East Asians (Supplementary Tables 1 and 6), possibly reflecting the differences in therapeutic response to anti-VEGF treatment between Asians and Europeans13 31. The D47V mutation within SLC44A4 is not included in most of the routinely used genotyping arrays, but is now included as part of the exome array used in this study. The genomic region around the SLC44A4 locus is more complex, being located within the broad MHC region on Chromosome 6 between HLA class I and class II genes. As this region is very polymorphic, and allele frequency differences between cases and controls could be confounded by even minor population stratification, we thus reassessed the associations with AMD by adjusting for the first 10 principal components (PCs) of genetic ancestry. We did not observe any change in the association signals observed from our standard analysis, which adjusted for the first five PCs (analysis adjusting for the top 10 PCs; OR=1.39, P=2.33 × 10−7 in the discovery phase), consistent with previous observations in Asian studies with well-replicated associations within the MHC region32 33. SLC44A4 encodes for choline transporter protein-4, involved in sodium-dependent choline uptake by cholinergic neurons34. Defects in SLC44A4 have been linked to sialidosis, which presents with a spectrum of symptoms including eye abnormalities35. Prior studies on the basis of European patient collections have identified a total of four distinct AMD-associated loci on Chromosome 6 alone6. In this light, the burden of proof for the positive identification of C6orf223 and SLC44A4 (both are also located on Chromosome 6) is higher than usual due to the need for appropriate considerations of previously reported variants on the same chromosome. It is reassuring to note that both C6orf223 and SLC44A1 showed the strongest two signals of association with AMD outside of CFH, ARMS2-HTRA1 and CETP. Our exhaustive analyses using logistic regression adjusting for allele dosages at previously described SNP markers suggest that both C6orf223 and SLC44A1 are unrelated to those previously reported, and thus they likely represent Asian-specific genetic associations for AMD. Neither FGD6 nor the genes within its vicinity (Fig. 2c) have ever been previously implicated in any ocular disorders. FGD6 encodes for FYVE, RhoGEF and PH domain-containing protein 6, with its functions yet to be characterized. The Q257R mutation is less than half as common in Europeans (MAF=0.10) compared with East Asians (MAF=0.20–0.30), again possibly explaining the ability of our study to pick up this genetic effect. The identified genes were expressed in human retinal pigment epithelium (Supplementary Table 11). Of the three non-synonymous substitutions, the CETP D442G variant was predicted by both PolyPhen36 and SIFT37 to likely be causing damage to the protein structure/function, the FGD6 Q257R variant was predicted only by PolyPhen to be probably damaging but by SIFT to be tolerated and the SLC44A4 D47V variant was predicted by both tools to be benign or tolerated. Although the use of both prediction algorithms has been reported to be moderately sensitive, they suffer from lack of specificity38, and thus more evidence should be sought with regards to the FDG6 and SLC44A4 non-synonymous variants. Using HaploReg39, RegulomeDB40 and Encyclopaedia of DNA Elements (ENCODE)41 data, we identified variants within each of the four LD blocks in the 1000 Genomes Project (r 2>0.8 and 1% missingness should the SNP have a MAF 5%, the filtering criteria were set at >5% missingness. Statistical analysis For the discovery stage, all exudative AMD cases and controls appear well matched when visualized spatially on PC analysis for each sample collection on a per-country basis for Hong Kong, Japan and Singapore and according to self-reported ethnicity (ethnic Chinese for Hong Kong and Singapore, and ethnic Japanese for Japan; Supplementary Fig. 6), using previously reported criteria49, indicating that population stratification is unlikely to confound the association results. For both the discovery and replication stages, analysis of association with exudative AMD was carried out using 1-degree of freedom score-based tests using logistic regression. The tests model for a trend-per-copy effect of the minor allele on disease risk. For the discovery stage, we incorporated the top five PCs of genetic stratification into the logistic regression model to minimize the effect of residual population stratification50. We could not adjust for population stratification for the replication stage due to limited number of SNPs tested. Meta-analysis was conducted using inverse variance weights for each sample collection, which calculates an overall Z-statistic, corresponding P value and accompanying per-allele OR for each SNP analysed. Gene-based tests on mutational load was performed using the SKAT-O test16. The association between CETP D442G and serum HDL-c level was assessed using linear regression assuming an additive model of inheritance as previously described23 (due to serum HDL-c being distributed normally), with adjustment for age, gender and body mass index. Regional association and PC plots were analysed and plotted using the R statistical software package. Power calculations For the discovery stage (2,119 AMD cases and 5,691 controls), power calculations51 indicated that there is 80% power of detecting loci at P<1 × 10−4 (the threshold of association for bringing forward SNPs to the replication stage) at MAF as low as 10% with per-allele OR of 1.30. For rarer variants of higher penetrance, the discovery stage has 80% power of detecting loci at P<1 × 10−4 at MAF as low as 2% if the per-allele OR is at least 1.70. The entire sample (6,345 AMD cases and 15,980 controls) has 80% power to detect loci at P<5.0 × 10−8 at MAF as low as 2% if the per-allele OR is at least 1.55 or at MAF as low as 9% with per-allele OR of 1.25, in line with the effect sizes being reported in this study. Supplementary Table 15A shows the power calculations to detect SNPs at the threshold of P<1 × 10−4 in the discovery stage for bringing forward to the replication stage. Supplementary Table 15B shows the formal power calculations in the context of the combined discovery and replication stages. Author contributions C.-Y.C., K.Y., L.J.C., J.A., K.-H.P., C.P.P., N.Y., T.Y.W. and C.C.K. designed the study. C.-Y.C., C.M.G.C., M.M., P.D.C., I.Y.Y., A.L., R.M., A.H.K., S.Y.L., D.W., C.M.G.C., B.K.L., Y.S., H.N., Y.A.-K., N.G., A.T., K.M., S.Y., Y.S., H.I., T.I., S.H., T.Y.Y.L., H.C., S.T., X.D., F.W., P.Z., B.Z., J.S., J.-M.Y., W.P.K., R.M.v.D., Y.F., N.W., G.S.W.T., S.J.P., M.B., L.G., T.N., P.M., P.Z., S.-M.S., M.O., T.M., Y.K., S.J.W., H.C., H.-G.Y., J.Y.S., D.H.P., I.T.K., W.C., M.S., S.-J.L., H.W.K., J.E.L., C.K.H., T.H.L., S.-K.Y., T.A. and W.T.Y. gathered clinical data. J.P., S.D., I.N., Y.A.-K., F.M., P.O.S.T., F.L., X.Z., Y.S., B.G., R.D., Y.L., M.L.H., J.N.F., C.H.W., X.X., Jinlong Liang, J.M., X.J., Y.L., Jianjun Liu, K.S., E.N.V., J.X.B., Y.X.Z. and C.C.K. generated genetic data. C.-Y.C., K.Y., L.J.C., J.A., Lulin Huang, Lvzhen Huang, K.S.S., P.C., Jiemin Liao, P.G.O., Y.Y.T. and C.C.K. analysed the data. C.-Y.C., K.Y., L.J.C., J.A., Lulin Huang, Lvzhen Huang, E.S.T., X.X.L., Z.Y., K.H.P., C.P.P, N.Y., T.Y.W. and C.C.K. interpreted the data. C.-Y.C., K.Y., L.J.C., J.A., E.S.T., T.Y.W. and C.C.K. drafted the paper. All the authors contributed to revision of the paper. Additional information How to cite this article: Cheng, C.-Y. et al. New loci and coding variants confer risk for age-related macular degeneration in East Asians. Nat. Commun. 6:6063 doi: 10.1038/ncomms7063 (2015). Supplementary Material Supplementary Information Supplementary Figures 1-6, Supplementary Tables 1-15, Supplementary Methods and Supplementary References
