Introduction

Age-related macular degeneration (AMD) is the leading cause worldwide of severe visual impairment among people older than 55 years of age [1]. Its prevalence is likely to increase with the exponential growth in the aging population [1, 2]. Advanced AMD can be divided into two subtypes based on the nature and extent of the patient’s retinal damage: dry AMD (also called geographic atrophy, GA) and wet AMD (also called neovascular AMD). Wet AMD is the main type of advanced AMD in Asia, while GA rarely occurs in the Han Chinese population [2]. Wet AMD causes vision loss due to abnormal blood vessel growth in the choriocapillaris and retina; it leads to more vision loss than the dry form of the disease [3]. Wet AMD can be further classified into choroidal neovascularization (CNV) and polypoidal choroidal vasculopathy (PCV). The latter is markedly more common in Asia than in Europe [2]. Wet AMD is a complex disease caused by a combination of genetic and environmental factors [4]. In 2005 and 2006, complement factor H (CFH) [5,6,7] and ARMS2/HTRA1 [8, 9] were identified as two significant AMD risk loci by genome wide association analysis (GWAS). Thus far, at least 59 loci have been associated with AMD (Table S1).

The association of coding and untranslated region (UTR) variants with wet AMD is often ethnicity-specific. Some of the coding variants, such as rs9332739 (E318D) in C2 [10], rs2230199 (R102G) and rs1047286 (P314L) in C3 [11], and rs4151667 (L9H) and rs641153 (R32Q) in CFB [12], have only been validated in Caucasians, whereas rs2303790 (D442G) in CETP [13] has only been validated in East Asians. Most of the reported SNPs associated with AMD are not in the coding and UTR gene regions.

The coding and UTR variants in the reported AMD loci have not yet been systematically investigated in Han Chinese subjects. In our previous study of whole exome sequencing (WES) for wet AMD, we only reported on the SNP/loci that surpassed the genome-wide significance association (P < 4.24 × 10−7) in the discovery phase [14]. Based on our previous WES work for the wet AMD study [14], we systematically investigated the contribution of coding and UTR variants in the 72 genes located within the 47 reported wet AMD genetic susceptibility loci in a Han Chinese population.

Methods

Ethics statement

The study protocol was approved by Sichuan Provincial People’s Hospital’s Ethics Committee, Sun Yat-sen University’s Ethics Committee, Chinese University of Hong Kong’s Ethics Committee, the Joint Shantou International Eye Center of Shantou University and Chinese University of Hong Kong’s Ethics Committee. Written informed consent was obtained from all participants involved in this study. All the experiments were performed in accordance with the relevant guidelines and regulations, including any relevant details.

Study sample descriptions

All study subjects were Han Chinese from South China. The subjects were unrelated and were recruited from the Department of Ophthalmology of the Sichuan Provincial People’s Hospital, Zhongshan Ophthalmic Center, Chinese University of Hong Kong, and the Joint Shantou International Eye Center, as previously described [14]. The patients received complete ophthalmic investigations, including fluorescein angiography and indocyanine green angiography (ICGA). PCV was diagnosed based on the onchoroidal origin of the polypoidal lesions detected with ICGA. CNV and PCV were distinguished using the fluorescein angiography and ICGA findings. The control subjects were ≥50 years old and also received ophthalmic examinations. Those with macular degeneration from any cause were excluded from the study.

During the first stage of the study, we analyzed the genomic sequences from 349 patients with wet AMD (Table 1). There were 194 patients with PCV, and 155 patients with CNV. During the replication stages, we analyzed the genomic sequences from 1341 patients (764 with PCV and 577 with CNV). This study included a higher number of male patients (60–70%), which is compatible with previous studies that have shown that men are more susceptible than women to wet AMD.

Table 1 Sample collections of the cohorts for wet AMD (PCV and CNV) cases and controls
Full size table

WES

WES was performed in the discovery stage as previously described [14].

Inclusion of genes from the previously reported AMD association studies

Through a literature search, we identified 47 reported loci associated with AMD that involved 72 genes (Table S1). Using our previously published WES results, we analyzed the coding and UTR variants of these 72 genes using the cutoff value of P < 10 × 10−3 for wet AMD.

Genotyping

The Sequenom mass-array system and SNaPshot genotyping were used for the replication studies.

Statistical analysis

Single-variant association analysis for SNPs was performed with case-control association analysis and Fisher’s exact test model using PLINK version 1.9. The parameters included minor-allele frequency (MAF) cutoffs >0.01, genotypic success rate >90%, individual call rate >90% and Hardye Weinberg equilibrium >1 × 10−4. The combined meta-analysis power was calculated across the WES discovery and the replication collections using PLINK version 1.9.

Power calculation

The power calculations [15] of the collected data indicated that there was a 99% probability of detecting loci at P < 1 × 10−3 with a MAF of 30% if the per-allele OR was 1.30 and the disease prevalence was 5% in the multiplicative genetic model for the entire sample (1690 wet AMD cases and 2033 controls). For the discovery stage (349 wet AMD cases and 1250 controls), the power calculations indicated that there was 92% probability of detecting loci at P < 1 × 10−3 with a MAF of 30% if the per-allele OR was 1.30 and the disease prevalence was 5% in the multiplicative genetic model.

Results

In the coding and UTR regions, a previously reported SNP (rs10490924, p.A69S in ARMS2) at the ARMS2/HTRA1 loci showed the strongest association with wet AMD, with no significant differences between PCV and CNV (P = 2.37 × 10−18, OR = 2.11 for wet AMD; P = 2.12 × 10−9, OR = 1.93 for PCV; P = 2.05 × 10−9, OR = 2.42 for CNV). The following four previously reported SNPs in the CFH gene had strong associations with wet AMD, and again there were no significant differences between PCV and CNV: rs3753396 (p.Q672Q) (P = 2.60 × 10−10, OR = 1.35 for wet AMD; P = 7.35 × 10−8, OR = 1.80 for PCV; and P = 2.22 × 10−5, OR = 1.90 for CNV), rs1065489 (p.E936D) (P = 4.21 × 10−10, OR = 2.53 for wet AMD; P = 1.21 × 10−7, OR = 1.84 for PCV; and P = 2.71 × 10−5, OR = 2.01 for CNV), rs2274700 (p.A473A)(P = 8.21 × 10−10, OR = 1.81 for wet AMD; P = 1.33 × 10−6, OR = 1.75 for PCV; and P = 4.31 × 10−6, OR = 1.89 for CNV), and rs800292 (p.V62I) (P = 1.83 × 10−7, OR = 1.74 for wet AMD; P = 2.32 × 10−5, OR = 2.05 for PCV; and P = 3.11 × 10−5, OR = 1.80 for CNV).

In addition to these known SNPs, another 25 SNPs showed an association with wet AMD (P < 10 × 10−3) in the coding and UTR regions; these 25 SNPs have not been previously reported in Chinese and/or Asian populations (Table 2). Using the Sequenom mass-array method, we searched for these SNPs in another Chinese cohort of 885 wet AMD patients and 572 controls (Table 1) whose members were recruited from the Guangdong Province. Table 3 presents the results of this replication. In this stage, rs2736911 (c.C112T, p.R38X) in ARMS2 showed the strongest association with wet AMD (P = 9.95 × 10−4, OR = 0.69). Interestingly, CNV (P = 2.69 × 10−5, OR = 0.54), not PCV (P = 0.12, OR = 0.82), was the main contributor to this association (Table 3). SNPs in FILIP1L and BBX also showed nominal associations.

Table 2 Allelic association of the selected genes in the discovery stage (Exome Sequencing stage) (P < 10–3 for wet AMD)
Full size table
Table 3 Allelic association of the selected genes in the replication stage
Full size table

When we combined the findings from both the discovery and replication stages (Table 4), rs2736911, located in ARMS2, was still strongly associated with wet AMD (P = 2.34 × 10−6, OR = 0.66), and there was a stronger association with the CNV subtype (P = 8.51 × 10−6, OR = 0.66) than the PCV subtype (P = 0.011, OR = 0.77). Four other novel SNPs in three loci, including rs189132250 in BBX (3q12.1), rs6078 in LIPC (15q21.3), rs144351944 in FILIP1L (3q12.1) and rs1800978 in ABCA1 (9q31.1) showed a weak association with wet AMD (Table 4).

Table 4 Combined association of the selected genes in the discovery and replication stages by meta-analysis
Full size table

For the four newly-discovered SNPs in these three loci (rs189132250, rs6078, rs144351944, and rs1800978), we performed a second replication in another independent Chinese cohort from Shantou in the Guangdong Province (456 wet AMD patients, including 241 PCV cases and 215 CNV cases, and 211 controls). The results are shown in Table 6. The rs189132250 variant in BBX and the rs1800978 variant in ABCA1 showed weak associations with wet AMD (Table 5).

Table 5 Allelic association of the selected genes in the second replication stage
Full size table

Combining the results of the discovery, first replication and second replication stages (Table 6) revealed that rs189132250 in BBX (c.A1329G, p.A443A) was associated with wet AMD (P = 8.76 × 10−5, OR = 6.0) and that it had a stronger association with CNV (P = 8.56 × 10 −7, OR = 9.8) than with PCV (P = 0.02, OR = 3.8). The rs144351944 variant in the UTR5 of FILIP1L was associated with wet AMD, with PCV, not CNV, being the main contributor to the association (P = 8.44 × 10−4, OR = 1.34 for wet AMD; P = 4.91 × 10−5, OR = 1.52 for PCV; and P = 0.45, OR = 1.26 for CNV). Similarly, the rs1800978 variant in the UTR5 of ABCA1 was also associated with wet AMD, with PCV, not CNV, being the main contributor to the association (P = 4.14 × 10−4, OR = 1.26 for wet AMD; P = 1.85 × 10−4, OR = 1.33 for PCV; and P = 0.015, OR = 1.23 for CNV; Table 6). The association of rs6078 variant in LIPC with wetAMD is P = 2.27 × 10−4, OR = 1.23. However, the random-effects meta-analysis p-value P(R) for wetAMD is only 0.403 and the heterogeneity I2 is 94.73 that indicate the inconsistent results between these three cohorts. This result (Table 6) suggested that rs6078 variant in LIPC is not associated with wetAMD, CNV or PCV. Supplementary Figure 1 shows the forest plots of the four SNPs from each of the independently analyzed cohorts. According to the ORs, the rs189132250 variant in BBX carried a high risk for wet AMD. We also performed a rare variant burden assay for these four genes (with a MAF < 0.01 cut-off for the SNPs) using the SKAT R package; however, no significant signals were found in these genes (Table S2).

Table 6 Combined association of the selected four SNPs in the discovery and replication stages by meta analysis
Full size table

We then evaluated the associations between wet AMD and the four SNPs in the three previously reported loci. Using the SNaPshot genotyping method, this analysis identified three new potential SNPs: rs13095226 (3q12.1) [16], rs1883025 (9q31.1) [10], rs493258 [17], and rs10468017 (15q21.3) [10]. No significant associations were identified for these SNPs (Table S3). However, among these four SNPs, the rs1883025 variant in ABCA1 had an LD (r²) = 0.7 with the rs1800978 variant (Table S4) and showed a consistent trend of association with PCV (P = 0.08, OR = 1.24), as did rs1800978 (P = 3.0 × 10−4, OR = 1.33).

Discussion

The currently available data indicate that genetic influences are responsible for up to 70% of a patient’s AMD risk [18]. Consequently, research should focus on the functional effects of AMD-associated gene variants, particularly the effects of coding and UTR variants, in AMD progression. Our wet AMD studies in China have demonstrated the association of wet AMD with SNPs in the coding and UTR regions of the GWAS-reported loci. This study validated the wet AMD associations of the five previously reported SNPs (rs10490924 in the ARMS2/HTRA1 region [14, 19] and rs3753396 [14, 20], rs1065489 [14, 21], rs2274700 [14, 22], and rs800292 [14, 23] in CFH). The rs77466370 variant in the FGD6 gene was associated with wet AMD due to the PCV contribution (P = 7.75 × 10−6, OR = 1.55), as previously reported [14]. Another SNP, rs2303790 in CETP, had a P value larger than 1 × 10−3 in the discovery cohort; we confirmed that this SNP was associated with wet AMD in a previous study of a Han Chinese population [13].

In this study, the rs2736911 variant, which encodes a stopgap substitution in ARMS2 (p.R38X) and is independent of rs10490924, showed an association with wet AMD. The rs2736911 variant was first reported in a Polish population in 2012 [24]. To date, the association of rs2736911 with wet AMD has only been experimentally validated in Korea [25]. In 2015, Zhan Ye et al. indicated that rs2736911 was weakly associated with wet AMD (P = 0.0011, OR = 0.69) using a PheWAS analytical approach [26]. However, a meta-analysis of 23 published AMD studies did not find a strong association between rs2736911 and AMD (P = 0.122, OR = 0.77) [27]. In the Chinese cohort of the current study, we confirmed that the rs2736911 variant was associated with wet AMD (P = 2.90 × 10−6, OR = 0.66). By comparing the PCV and CNV subtypes, we found that rs2736911 was more effective in protecting against the CNV subtype (P = 7.57 × 10−6, OR = 0.66) than the PCV subtype (P = 0.012, OR = 0.77).

The rare variant rs189132250 in BBX (c.A1329G, p.A443A) is located in 3q12.1 near COL8A1. In the discovery stage of this study, no SNPs in the coding or UTR regions of COL8A1 showed an association beyond 10−3. The rs189132250 variant showed an association with wet AMD (P = 8.75 × 10−5) and had a stronger association with the CNV subtype (P = 8.56 × 10−7) than with the PCV subtype (P = 0.183). BBX is a transcription factor that is necessary for cell cycle progression from the G1 phase to the S phase. In a European ancestry GWAS study, Neil Caporaso et al. reported that this gene region was associated with cigarette smoking [28], an environmental risk factor for AMD [29]. The rs189132250 variant in BBX increases the risk of wet AMD, implying an interaction between this gene and the smoking environmental factor.

FILIP1L is also located in 3q12.1 near COL8A1. The rs144351944 variant in the UTR5 of FILIP1L was nominally associated with wet AMD (P = 8.44 × 10−4 for wet AMD; P = 4.94 × 10−5 for PCV; and P = 0.45 for CNV). FILIP1L has been predicted to regulate the antiangiogenic activity of endothelial cells; when FILIP1L was overexpressed in endothelial cells, it inhibited cell proliferation and migration and increased cell apoptosis [30]. FILIP1L was previously reported to be strongly expressed in tumor stroma and colon cancer vasculature (at a protein level) [31]. Previous reports have shown that FILIP1L is expressed in ovarian epithelial cells and downregulated in cancer [32, 33]. FILIP1L may be involved in wet AMD progression through the VEGF signaling pathway [31].

Except for rs1800978, which is in the reported LD of ABCA1 [10, 34], the three other newly-identified SNPs in this study were not in the same LD as the SNPs reported in the previous studies. This indicates that our WES analysis detected novel sequences that had been missed by GWAS. We could not evaluate the predicted function of these significant SNPs with respect to pathogenesis because the gene expression regulatory SNPs were not captured by WES. The association of these SNPs with wet AMD could be related to the LD of the other SNPs in the gene expression regulatory region.

In summary, we carried out a comprehensive analysis of coding and UTR variants in Chinese cohorts of wet AMD. In addition to the previously reported four coding variants (rs800292, rs2274700, rs3753396, and rs1065489) in CFH as well as the rs10490924 and rs2736911 variants in the ARMS2/HTRA1 region, we identified two novel LDs (located in rs189132250 in BBX and rs144351944 in FILIP1L regions) that were associated with wet AMD. We also validated the wet AMD association of rs1800978 in ABCA1, which is in the same LD as the reported SNPs [10, 34]. These findings provide new information regarding the coding and UTR regions of the known AMD GWAS loci in Chinese cohorts.