Identification of PGF as a New Gene for Neovascular Age-Related Macular Degeneration in a Chinese Population

Abstract

Purpose: To determine the associations of the VEGFA, VEGFB, and placental growth factor (PGF) genes with neovascular age-related macular degeneration (nAMD) and polypoidal choroidal vasculopathy (PCV).

Methods: Seven single-nucleotide polymorphisms (SNPs) in VEGFA, three SNPs in VEGFB, and five SNPs in PGF were genotyped in 1722 unrelated Chinese participants, including a Hong Kong cohort of 214 nAMD patients, 236 PCV patients, and 365 controls, and an independent Shantou cohort of 189 nAMD patients, 187 PCV patients, and 531 controls, using TaqMan genotyping assays.

Results: Placental growth factor SNPs rs2268615 (G allele, P = 0.0047; odds ratio [OR] = 1.54, 95% confidence interval [CI], 1.14–2.08) and rs2268614 (G allele, P = 0.015; OR = 1.46, 95% CI, 1.07–1.97) were associated with nAMD. A significant omnibus haplotype association with nAMD was detected for a two-SNP window containing rs2268615 and rs2268614, with a haplotype G-G conferring a 1.54-fold increased risk (P = 0.0042) in the Hong Kong cohort and a 1.42-fold risk (P = 0.012) in the Shantou cohort. Pooling of the Hong Kong and Shantou data enhanced the association of nAMD with rs2268615 (P = 0.0022; OR = 1.38, 95% CI, 1.12–1.69; I² = 0%), rs2268614 (P = 0.0067; OR = 1.33, 95% CI, 1.08–1.63; I² = 0%), and the G-G haplotype (P = 0.0013; OR = 1.46, 95% CI, 1.16–1.84; I² = 0%). In contrast, the PGF SNPs and haplotype were not associated with PCV. Our results also revealed no association of SNPs in VEGFA and VEGFB with nAMD or PCV.

Conclusion.: Placental growth factor is a susceptibility gene for nAMD in a Chinese population, providing new evidence to support a biological role of PGF in choroidal neovascularization.

Age-related macular degeneration (AMD) is a leading cause of irreversible central visual impairment among elderly individuals in developed countries. Age-related macular degeneration is characterized by degenerative features at the macula affecting the photoreceptors and retinal pigment epithelium (RPE), often presenting with macular drusen in the early stage. Advanced AMD can be classified into geographic atrophy (dry AMD) and neovascular AMD (nAMD; or wet AMD). Prevalence of advanced AMD has been estimated to be 0.59% in Caucasian populations and 0.56% in Asian populations aged 40 to 79 years.¹ Neovascular AMD accounts for the majority of severe visual loss in Asian AMD patients. Retinal pigment epithelium detachment, subretinal fluid, hemorrhage, and fibrotic scar are common features in nAMD.² 

Polypoidal choroidal vasculopathy (PCV) is another serious macular disease usually presented with subretinal polyp-like lesions, RPE detachment, retinal edema, and subretinal hemorrhage. In PCV, the inner choroidal vascular networks terminated in polypoidal lesions, which are different from the generalized choroidal neovascularization (CNV) in nAMD and best detected by indocyanine green angiography (ICGA).³ Polypoidal choroidal vasculopathy is reportedly more prevalent among Asians and African-Americans compared with Caucasians.⁴ It might account for approximately 20% to 50% of nAMD in Asian populations while less than 10% in Caucasians.⁵ In Chinese, the prevalence of PCV was approximately 0.5% in people aged 40 years and older and over 1.3% in those aged 65 years and older.⁶ It remains debatable whether PCV is a subtype of nAMD or a distinct clinical entity. Generally, PCV was considered a variant of nAMD due to the similarities in phenotypic features; however, they are dramatically different in natural history, prevalence, and treatment response. These discrepancies lead to believe that nAMD and PCV are two separate entities.⁷ 

Both AMD and PCV are multifactorial in etiology, involving multiple environmental and genetic risk factors. A recent genome-wide association study (GWAS) involving more than 77,000 study subjects showed that polymorphisms in 19 genetic loci, which included complement factor H (CFH), age-related maculopathy susceptibility 2 (ARMS2), and VEGFA, accounted for 15% to 65% of the total genetic contribution to AMD.⁸ No GWAS of large scale has been conducted in PCV, but there are reported associations of PCV with AMD-associated genes such as CFH, ARMS2-HTRA1, complement component 2 (C2), and cholesteryl ester transfer protein (CETP).^9,10 In a recent systematic review and meta-analysis, the majority of genes that were associated with PCV were also associated with AMD. One clear exception was the ARMS2-HTRA1 polymorphisms, which were significantly different between nAMD and PCV,⁷ suggesting that the genetic components of AMD and PCV are partially different. Therefore, genetic studies involving both AMD and PCV will help to identify gene variants of similar or differential effects between the two. 

Differences between nAMD and PCV also present in their treatment responses to anti-VEGF therapies. Most patients with nAMD respond well to anti-VEGF monotherapy, while PCV patients usually require combined anti-VEGF and photodynamic therapy.⁵ In 2012, aflibercept, a fusion protein that blocks VEGFA, VEGFB, and placental growth factor (PGF), was approved for treating macular degeneration.¹¹ Intravitreal injection of aflibercept achieved equivalent effect in improving the best-corrected visual acuity (BCVA) and preventing BCVA loss in nAMD while required fewer injections when compared with ranibizumab.¹² Aflibercept is also effective for treatment-resistant nAMD.¹³ 

The PGF, a target of anti-VEGF therapy, plays an important role in angiogenesis.¹⁴ We therefore hypothesize that PGF could be a susceptibility gene for nAMD and/or PCV. In this study, we determine the associations of the PGF, VEGFA, and VEGFB genes with nAMD and PCV separately, using haplotype-tagging single-nucleotide polymorphisms (SNPs) and reported functional SNPs. Our results revealed PGF as a novel putative gene for nAMD. 

Materials and Methods

Study Participants in the Hong Kong Cohort

This study involved a total of 1722 unrelated Chinese participants recruited separately from Hong Kong and Shantou, two different cities in southern China. The Hong Kong cohort included 815 participants: 214 patients with nAMD, 236 with PCV, and 365 healthy controls (Supplementary Table S1), recruited from the eye clinics of the Prince of Wales Hospital and the Hong Kong Eye Hospital, Hong Kong. All patients received complete ophthalmic investigations, including BCVA measurement, applanation tonometry, slit-lamp biomicroscopy, fundus photographs, fluorescein fundus angiography (FFA), and ICGA. All AMD patients recruited in this study had nAMD in at least one eye. Polypoidal choroidal vasculopathy was diagnosed upon a choroidal origin of polypoidal lesions as shown by ICGA.^10,15–17 Diagnosis of nAMD and PCV were distinguished by FFA and ICGA. Subjects with any eye having nAMD and PCV lesions concurrently or other causes of CNV were excluded. Control subjects were recruited from people who attended the clinics for an unrelated eye condition and recruited according to the following criteria: (1) age 60 years and older, (2) no identifiable signs of AMD or macular degeneration of any cause, and (3) no any other major eye diseases, except for mild senile cataracts and/or mild refractive errors. 

The study protocols were approved by institutional Ethics Committees in the Chinese University of Hong Kong and the Joint Shantou International Eye Centre. Written informed consents were obtained from all subjects after explaining the nature of the study. All study procedures were performed in accordance with the tenets of the Declaration of Helsinki. 

Selection of SNPs and Genotyping

Haplotype-tagging SNPs were selected from the PGF gene in the HapMap CHB population, using the HapMap Genome Browser release #27 dataset (in the public domain, http://hapmap.ncbi.nlm.nih.gov/). The tagger-pairwise method was used, with an r² cut-off of 0.8 and a minor allele frequency (MAF) cut-off of 0.05. Three SNPs were generated, namely rs2359192, rs2268615, and rs2268616. These three SNPs captured all alleles across the PGF locus with a MAF greater than 0.05 and a mean r² of 0.974. In addition, two PGF SNPs, rs11850328 and rs2268614, which had been correlated with the serum level of PGF,¹⁸ were also selected. Thus, a total of five PGF SNPs were included. The haplotype-tagging SNPs in the VEGFB gene were selected using the same criteria and two SNPs were picked with a mean r² of 1.0: rs4930152 and rs11603042. However, rs11603042 failed the assay design so that it was replaced by SNP rs594942 in high linkage disequilibrium (LD; r² = 0.925). In addition, a common coding variant (rs12366035, p.Asp136Asp) in VEGFB was selected. 

In this study, we also assessed the association of the VEGFA gene with nAMD and PCV. We selected seven candidate SNPs based on previous studies. SNP rs943080 had a significant association with AMD in European populations⁸ and a marginal association with AMD in Asians.¹⁹ In a recent meta-analysis two VEGFA SNPs, rs1413711 and rs833061, were linked to AMD susceptibility.²⁰ SNP rs699947 was a strong determinant of the anatomical outcome after photodynamic therapy in AMD.²¹ SNP rs833070 was associated with AMD risk and retinal thickness in patients receiving anti-VEGF therapy.²² SNP rs833069 was associated with the development and progression of AMD.²³ Another SNP rs3025039 was associated with a decreased serum level of VEGF.²⁴ 

Genomic DNA from the whole blood was extracted using a commercial kit (Qiagen QIAamp DNA Blood Mini kit; Qiagen, Hiden, Germany) according to the manufacturer's protocol. All of the 15 candidate SNPs, including five in PGF, three in VEGFB, and seven in VEGFA, were genotyped in all of the Hong Kong participants, using TaqMan SNP assays (Applied Biosystems, Foster City, CA, USA) with the Roche LightCycler 480 Real-Time PCR System (Roche, Basel, Switzerland) according to the manufacturer's instructions. 

Replication Study

The two PGF SNPs (rs2268615 and rs2268614) that showed significant association with nAMD and the VEGFB SNP rs594942 that showed association with PCV in the Hong Kong cohort were genotyped in an independent cohort of 907 unrelated participants recruited from the Joint Shantou International Eye Centre (JSIEC), Shantou, China, including 189 patients with nAMD, 187 with PCV, and 531 control subjects. The recruiting criteria were the same with that adopted for the Hong Kong sample. 

Statistical Analysis

All SNPs were assessed for Hardy-Weinberg Equilibrium (HWE) using the exact test in PLINK (v1.07; in the public domain, http://pngu.mgh.harvard.edu/purcell/plink/).²⁵ Five SNPs, including four in VEGFA (rs699947, rs833061, rs833070, and rs1413711) and one in PGF (rs2268616), were excluded for further analysis because of deviation from HWE in the control group (P < 0.05). Allelic and genotypic distributions of each SNP among different study groups were compared by χ² test or Fisher's exact test in PLINK. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated with the nonrisk allele as reference. Logistic regression analysis was used to adjust the association for sex. Bonferroni correction was adopted to adjust the P values in multiple testing. A P value of less than 0.005 (P = 0.05/10, where 10 was the number of SNPs included in data analysis) was considered statistically significant.^26–28 

Pairwise LD and haplotype associations were assessed using Haploview²⁹ and PLINK. The variable-sized sliding-window method was used to capture the optimum markers in haplotype association.³⁰ Omnibus test for global haplotype association of each window was conducted using PLINK. The window that gave a smallest P value in the omnibus tests was considered the optimum window.^31,32 A haplotype-specific association test was then conducted for the haplotypes within that window. The permutation test was used to correct the P values for haplotype associations (number of iterations = 10,000). 

To evaluate the effects of the PGF, VEGFA, and VEGFB SNPs in the context of other AMD genes and the genetic interaction among them, we obtained the genotypes of two major AMD and PCV-associated SNPs, CFH rs800292 and HTRA1 rs11200638, from our previous studies.^10,15,33 The genetic effects of two significant PGF SNPs, rs2268615 and rs2268614, were estimated in the context of both CFH rs800292 and HTRA1 rs11200638 in an additive model by using logistic regression analysis in PLINK. The CFH SNP rs1061170 was not involved in the logistic regression analysis since it is rare and not associated with nAMD in Chinese.^34,35 Furthermore, pairwise interaction analysis was conducted to detect epistatic effects among the SNPs in VEGFA, VEGFB, PGF, CFH, and HTRA1 using the epistasis option in PLINK. 

To combine the data of the two PGF SNPs from the Hong Kong and Shantou cohorts, we performed the Mantel-Haenszel χ² to obtain the pooled-OR and 95% CI using a fixed-effect model based on heterogeneity test results (I² ≤ 50%).^36,37 The test was performed using Review Manager (RevMan, version 5.2; The Cochrane Collaboration, Copenhagen, Denmark). 

Results

Characteristics of Study Subjects

Supplementary Table S1 showed the characteristics of the study participants. There were more males in patient groups than in controls. Therefore, sex was adjusted by logistic regression in the association analysis. Because we purposely recruited subjects older than 60 years as controls, the age was not adjusted in the association analysis. 

Individual SNP Association

The call rate of each SNP was greater than 99.7%. The G allele of the PGF SNP rs2268615 was overrepresented in nAMD, conferring a significant risk effect (P = 0.0047; OR = 1.54, 95% CI, 1.14–2.08; Table 1). Another SNP, PGF rs2268614, showed a similar trend of an association with nAMD (G allele, P = 0.015, OR = 1.46, 95% CI, 1.07–1.97). Logistic regression analysis revealed that these two SNPs, rs2268615 (P = 0.0093; OR = 1.50, 95% CI, 1.11–2.03) and rs2268614 (P = 0.019; OR = 1.46, 95% CI, 1.06–1.99), remained at similar levels of significance after adjusting for sex. These two PGF SNPs were not significantly associated with PCV, although they showed the same trend of effect as that in nAMD (Table 1). In addition, a borderline association was detected between VEGFB rs594942 and PCV (P = 0.038, OR = 1.31, 95% CI, 1.02–1.69; Table 1), but this association could not withstand multiple correction. The other two SNPs in PGF, two in VEGFB and three in VEGFA were not associated with nAMD or PCV. The five SNPs that were excluded due to deviation from HWE (i.e., VEGFA rs699947, rs833061, rs833070, and rs1413711, and PGF rs2268616) were not associated nAMD or PCV (data not shown). 

Table 1

 

Allelic Association of SNPs in the VEGF-A, VEGF-B, and PGF Genes With nAMD and PCV

 

To determine whether the effects of the PGF SNPs were independent of the two major AMD-associated SNPs, CFH rs800292 and HTRA1 rs11200638, we assessed the associations in logistic regression models. PGF rs2268615 and rs2268614, respectively, remained associated with nAMD after conditioning on CFH rs800292, HTRA1 rs11200638 and sex, indicating independent effects of the PGF SNPs (Supplementary Table S2). In epistatic analysis, no significant SNP*SNP interaction was identified in either nAMD or PCV (data not shown). 

Haplotype-Based Association Analysis

We performed a sliding-window haplotype association analysis of the PGF gene, with the window sizes ranging from two to four. In nAMD, the most significant omnibus association was identified from a two-SNP window defined by rs2268615 and rs2268614 (P_omnibus = 0.012 at 2 degrees of freedom). Three haplotypes were detected in this window, of which two showed a significant association (Table 2). A haplotype G-G, presented in 82.3% of nAMD and 75.2% of controls, conferred an increased risk (P = 0.0042, permutation P = 0.02; OR = 1.54, 95% CI, 1.02–2.33), while the haplotype T-A, detected in 15.8% of nAMD and 23.0% of controls, was protective (P = 0.0029, permutation P = 0.014; OR = 0.63, 95% CI, 0.41–0.97; Table 2). In contrast, no significant omnibus association was identified for PCV (data not shown). 

Table 2

 

Haplotype-Based Association of the PGF Gene in nAMD

 

Replication Study and Pooled Analysis

In the Shantou Chinese samples, the two PGF SNPs, rs2268615 (P = 0.12, OR = 1.25, 95% CI, 0.94–1.65; Table 1) and rs2268614 (P = 0.15, OR = 1.23, 95% CI, 0.93–1.63; Table 1), did not show a statistically significant association with nAMD, but the ORs were toward the same trends as that in the Hong Kong sample. A significant omnibus association was detected for the haplotypes defined by rs2268615 and rs2268614 (P_omnibus = 1.42 × 10⁻⁴ at 3 degrees of freedom). The haplotype G-G, defined by the risk alleles, was associated with nAMD (P = 0.012, permutation P = 0.11; OR = 1.42, 95% CI, 1.08–1.88; Table 2). In contrast, no single SNP (including VEGFB rs594942) or haplotype was associated with PCV (Table 1). 

By pooling the data of the two PGF SNPs and the haplotypes in the Hong Kong and Shantou samples, SNP rs2268615 (P = 0.0022; OR = 1.38, 95% CI, 1.12–1.69; I² = 0%) and the haplotype G-G (P = 0.0013; OR = 1.46, 95% CI, 1.16–1.84; I² = 0%) showed an enhanced, significant association with nAMD, while rs2268614 (P = 0.0067; OR = 1.33, 95% CI, 1.08–1.63; I² = 0%) showed a similar trend towards an association (Fig.). In contrast, the two PGF SNPs and the haplotype G-G were not associated with PCV (Supplementary Fig. S1). 

Discussion

In this study, we have, for the first time, identified a significant association between a haplotype-tagging SNP rs2268615 in the PGF gene and nAMD. Another PGF SNP rs2286614 also showed a trend toward an association with nAMD. Logistic regression suggested that the effects of rs2268615 and rs2268614 were independent of CFH rs800292 and HTRA1 rs11200638. In addition, a significant omnibus haplotype association with nAMD was detected for a two-SNP window containing rs2268615 and rs2268614 in both the Hong Kong and Shantou cohorts. Finally, pooling the data of the Hong Kong and Shantou cohorts revealed enhanced associations of nAMD with the two PGF SNPs and the haplotype G-G defined by their risk alleles, with low intercohort heterogeneity. In contrast, no SNP or haplotype showed a significant association with PCV. Thus, our data indicates PGF as a susceptibility gene for nAMD, but not PCV. 

The PGF gene, spanning a 13.94-kb region on chromosome 14, encodes the PGF, which is a member of the VEGF protein family and shares homology with VEGFA. Placental growth factor is involved in pathological angiogenesis via two receptors, Fms-related tyrosine kinase 1 and Neuropilin 1.³⁸ Placental growth factor directly stimulates the growth and migration of endothelial cells, and synergistically amplifies the action of VEGFA.³⁹ In the retina, PGF is proangiogenic on retinal endothelial cells.⁴⁰ It is also expressed in the intact choroid and could be upregulated during the course of laser-induced CNV in mice.¹⁴ Furthermore, CNV could be prevented in mice by knocking out or knocking down PGF, or blocking the PGF receptor with a neutralizing antibody.^14,41 

In a previous study, a SNP rs1042886, located in the 3′-untranslated region of PGF, was associated with pre-eclampsia (OR for risk allele A: 1.5, P = 0.010).⁴² In our present study, the PGF SNP rs2359192, also located in the 3′-untranslated region, was not associated with nAMD or PCV. Instead, we identified two PGF SNPs, rs2268615 (G; OR, 1.54) in intron 2 and rs2268614 (G; OR, 1.46) in intron 3, conferred increased risks to nAMD. To our knowledge, this is the first time that these two PGF SNPs are associated with a human disease. Notably, the AA genotype of PGF rs2268614 had been correlated with an elevated level of plasma PGF in a population-based sample of Caucasian origin.¹⁸ The rs2268614 is located at the binding sequence of a transcription GA-binding factor.¹⁸ Therefore, this SNP is likely to implicate in the regulation of PGF expression. In a recent study, a significant upregulation, instead of downregulation, of systemic PGF was detected in AMD patients treated with aflibercept.⁴³ This finding echoes to our finding that the PGF SNP rs2268614, which is reportedly correlated with a higher PGF plasma level, is protective for nAMD. Notably, in mice, knocking out or knocking down of PGF was found to prevent CNV formation.¹⁴ Therefore, the roles of systemic PGF levels should be different between human nAMD and animal model of CNV. 

We found no significant association of VEGFA and VEGFB with nAMD and PCV. Churchill et al.⁴⁴ first reported a SNP rs1413711 in VEGFA to associate with AMD in a Caucasian cohort. Later, some studies had found VEGFA as a risk factor for AMD,^23,45,46 whereas some others did not.^47–49 In a GWAS, the VEGFA SNP rs943080 was significantly associated with AMD in European populations (P = 9 × 10⁻¹⁶, OR = 0.87 for the minor allele C).⁸ In contrast, a borderline association of rs943080-C with AMD was reported in Asians (P = 0.041, OR = 0.91).¹⁹ In our present study, rs943080-C was not significantly associated with nAMD (P = 0.42, OR = 1.13) or PCV (P = 0.43, OR = 1.12) in the Hong Kong Chinese cohort (Table 1). We also studied other reported VEGFA SNPs of functional impacts. The VEGFA SNP rs833069 showed the same trend of effect (OR = 1.23 and 1.15 in nAMD and PCV, respectively) as that in a previous study,²³ but the associations were not statistically significant. This is likely due to the relatively small sample sizes in this study. Nevertheless, our data suggested that VEGFA is not a major genetic factor for nAMD and PCV in our study population. 

There are several limitations in this study. First, the haplotype defined by rs2268615 and rs2268614 showed a borderline association with nAMD in the Shantou cohort. Also, the individual SNPs did not show a statistically significant association with nAMD. However, the ORs were toward the same trends as that in the Hong Kong sample. Hong Kong and Shantou, two different cities in southeast China about 400-km apart, share some genetic and social similarities. Approximately 1/6 of the Hong Kong residents are from the Chaoshan district, where Shantou is the biggest city (in the public domain, http://www.cafiu.org.cn/english/NewsInfo.asp?NewsId=1578). In our previous studies, we found similarities and disparities in the genetic associations of eye diseases between the Hong Kong and Shantou study populations. A common variant rs4236601 in the CAV1/CAV2 locus was associated with POAG in both the Hong Kong (OR = 5.01) and Shantou (OR = 5.47) cohorts.⁵⁰ In contrast, ABCC5 rs1401999, a SNP associated with PACG, showed different trends of effect (OR = 1.16 for Hong Kong and OR = 0.85 for Shantou).⁵¹ Therefore, despite the Shantou cohort provided a positive replication in the present study, studies in other cohorts are warranted to verify the role of PGF in nAMD. Second, smoking status for a portion of our study subjects was not available; therefore smoking was not adjusted in the analysis. 

In conclusion, we have identified PGF SNPs rs2268615 and rs2268614 as new susceptibility gene markers for nAMD in Chinese, indicating a biological role of the PGF gene in AMD mechanism. Further replication studies are warranted to extend our findings to other ethnic populations and verify the role of PGF in nAMD.1552 

Figure

 

The forest plots of pooling the samples from Hong Kong and Shantou cohorts compared (A) rs2268615(G), (B) rs2268614(G), and (C) G-G defined by rs2268615 and rs2268614 between nAMD and control. Squares indicate the study-specific OR. The size of the box is proportional to the percent weight that each study contributed in the pooled OR. Horizontal lines indicate 95% CI. A diamond indicates the summary OR with its corresponding 95% CI.

Acknowledgments

The authors thank all the participants in this study. 

Supported by the National Natural Science Foundation of China (81500764, LJC; China), a Direct Grant of the Chinese University of Hong Kong (4054119, CPP; Hong Kong), and the Endowment Fund for Lim Por-Yen Eye Genetics Research Centre, Hong Kong. 

Disclosure: L.J. Chen, None; L. Ma, None; W.K. Chu, None; T.Y.Y. Lai, None; H. Chen, None; M.E. Brelén, None; S.S. Rong, None; A.L. Young, None; P.O.S. Tam, None; M. Zhang, None; C.P. Pang, None