Genome-wide association study of age-related macular degeneration identifies associated variants in the TNXB–FKBPL–NOTCH4 region of chromosome 6p21.3

INTRODUCTION

Age-related macular degeneration (AMD) is a leading cause of visual loss in Western populations (1,2), reducing the quality of life of tens of millions of older people worldwide. It affects the macular region of the retina, which has a high density of photoreceptors for detailed central vision. Early in the disease, deposits called drusen form along Bruch's membrane, which separates the retinal pigment epithelium (RPE) from the underlying choroid (3). The later stages of the disease are characterized by focal atrophy of the RPE and overlying photoreceptors (geographic atrophy, GA) and/or growth of new blood vessels from the choroid through Bruch's membrane into the RPE (choroidal neovascularization, CNV) (3). Both of these processes can result in the loss of central vision.

Susceptibility to AMD is influenced by age, environmental and genetic factors (3). Smoking is the most important environmental risk factor (4). Striking progress has been made in understanding the genetics of AMD (5,6). Common sequence variants in the complement pathway genes CFH (7–10), C2 (complement component 2)–CFB (complement factor B) (11,12) and C3 (complement component 3) (13,14) are established risk factors and there is another risk locus in the vicinity of CFI (complement factor I) (15,16). This and other evidence points to the activation of the alternative complement pathway as an important component of the pathogenesis of AMD (17,18). Variants at the ARMS2 (age-related maculopathy susceptibility 2)–HTRA1 (HtrA serine peptidase 1) locus are strongly associated with AMD (19,20), but the mechanism is uncertain (21). More recently reported risk loci at LIPC (hepatic lipase) (22–24), CETP (cholesteryl ester transfer protein) (22–24), TIMP3 (TIMP metallopeptidase inhibitor 3)–SYN3 (synapsin III) (22–24) and VEGFA (vascular endothelial growth factor A) (24) implicate lipid metabolism, matrix homeostasis and control of angiogenesis as additional factors in the pathogenesis of AMD. The known genetic risk variants do not account for all the heritability of AMD. We have therefore carried out a genome-wide association study (GWAS) in the UK population to identify additional susceptibility loci.

RESULTS

We conducted a GWAS in the UK population obtaining genotypes for 893 cases and 2199 controls that passed quality control metrics (Materials and Methods, Table 1, Supplementary Material, Table S1). Subjects were typed with either the Illumina 300k array (150 cases) or the Illumina 550k array (743 cases and 2199 controls). Quality control criteria (Materials and Methods) were not met for 24 125 single-nucleotide polymorphisms (SNPs) in the 300k array and 68 531 SNPs in the 550k array. A small number of additional SNPs were excluded after visual inspection of their cluster plots so that subsequent analyses were based on 286 135 and 488 867 genotyped SNPs from the 300k and 550k arrays, respectively. We imputed variants using the CEU HapMap II reference panel and combined results from the two arrays following imputation for a total of 2 272 849 SNPs (Materials and Methods). Unless otherwise specified, the results discussed here did not show evidence of heterogeneity between the two platforms. The genomic inflation factor (λ) was 1.014 using all the genome-wide results and 1.007 after excluding the test statistics observed at the previously reported CFH, C2–CFB and ARMS2–HTRA1 loci; the quantile–quantile plots showed no widespread departure from the expected distribution of P-values (Supplementary Material, Figs S1 and S2). Therefore, no further correction for inflation was applied to the test statistics. Replication studies were conducted on a sample of up to 1411 advanced AMD cases and 1431 controls (Supplementary Material, Table S2).

DISCUSSION

We have carried out the first GWAS of AMD in the UK population. As expected, we found association with the well-established susceptibility loci ARMS2–HTRA1 (19,20), CFH (7–10), CFB–C2 (11,12), C3 (13,14), CFI (15,16) and APOE (28). There was also evidence of association for more recently reported risk loci at VEGFA (24) and LIPC (22–24), but not CETP or TIMP3–SYN3 (22–24). Our discovery experiment had limited power to detect these weaker associations because of the sample size and the use of younger unexamined controls (29).

At the CFI locus, the strongest association was with an SNP in the neighbouring CCDC109B gene as reported by others (23) rather than with the originally reported SNP rs10033900 closer to CFI (16). This is unsurprising as we have previously reported lack of association with rs10033900 in two independent UK case–control samples (15) and it was a subset of cases from these studies that was used in our GWAS. The other reported association at this locus with rs2285714 in PLA2G12A (phospholipase A2, group XIIA) (22) showed no evidence of association in our discovery analysis. Further studies are needed to clarify the origin of the association signal in this region.

The association at the CFH locus on chromosome 1q32 was extensive, as found in other GWAS (22–24). It spanned a megabase of DNA encompassing many genes. Conditional analysis of the region identified two independent association signals, one characterized by SNP rs10801555 (an almost perfect proxy for rs10611760, Y402H) and the other by rs1329428 (a proxy for rs1410996). These two independent association signals have been reported previously (12,22,25).

On chromosome 10q26, there was a strong association spanning the ARMS2 and HTRA1 genes as found by others (22–24), but high LD prevented further localization of the association signal.

On chromosome 6p21.3, there was a strong association with rs541862, a perfect proxy for a nonsynonymous change R32Q (rs641153) in CFB which has a well-established association with AMD, the minor allele being protective (11,12). This was one of several strongly associated SNPs spanning C2, CFB, RDBP (RD RNA-binding protein) and SKIV2L in a region of high LD (Fig. 1A) which makes the dissection of the statistical association difficult. Support for rs641153 as the causal variant comes from functional studies showing that the 32Q variant of CFB has reduced activity compared with 32R because of weaker binding to C3b (30). Another associated SNP from this region, rs438999 (R151Q) in the SKIV2L, has also been suggested as a causal variant (31), and a protective effect of rs429608 in the same gene has been reported (27). Our data showed evidence of association with both rs438999 and rs429608, but conditional analysis did not provide support for an association at SKIV2L independent of the signal from rs541862 in CFB (Supplementary Material, Table S6).

On chromosome 6p21.3, we also found a strong association reaching genome-wide significance for SNP rs2071277 in the NOTCH4 gene some 250 kb from CFB. Conditional analysis confirmed that this was independent of the C2–CFB signal represented by rs541862 and also provided support for a second independent association with rs12153855 in the intervening TNXB gene. Only four haplotypes defined by these three SNPs were observed (Table 5), including the protective haplotype CTT characterized by a C allele at rs541862 which is the well-established protective association reported for C2–CFB. The other three haplotypes had a T allele at rs541862. The commonest of these was TTT, which we designated as the reference haplotype. The next most common haplotype TTC was associated with an increased risk for AMD, giving an OR of 1.14 (95% CI 1.05–1.25); the fourth haplotype TCC was associated with higher risk, the OR being 1.47 (95% CI 1.29–1.67).

Both rs12153855 in TNXB and rs2071277 in NOTCH4 are intronic and unlikely to be causal variants. SNP rs12153855 has a perfect proxy rs9391734 in the 5′ untranslated region of the neighbouring gene FKBPL, which might have functional significance. So we have identified novel associated variants but do not have evidence to pinpoint the causal variants responsible for these associations, as is the case with other susceptibility loci for AMD and many other complex diseases. All three genes could play a role in the pathogenesis of AMD as we review in what follows. However, further research will be needed to determine whether or not this is the case. Other disease associations have been reported for this region of chromosome 6 and notably an association with dementia and variants near NOTCH4 and AGER (advanced glycosylation end product-specific receptor) (32).

TNXB encodes the extracellular matrix protein tenascin-X, which is present in the macular Bruch's membrane/choroid complex (33). Various functions have been attributed to tenascin-X, which could have relevance to AMD. It has been shown to modulate collagen deposition (34). There is also evidence that it modulates the mechanical properties of collagen networks (35) and is involved in the maturation and/or maintenance of collagen and elastin networks (36); both collagen and elastin are present in Bruch's membrane. In addition, forms of tenascin-X have been shown to bind to VEGFA and VEGFB (vascular endothelial growth factor B) and enhance their pro-angiogenic activity (37).

FKBPL encodes the FKBP-like protein which belongs to the family of FK506-binding proteins and has a role in intracellular signalling (38). Of particular interest in the context of AMD, FKBPL also appears to have a separate role in controlling angiogenesis as a secreted protein that inhibits endothelial cell migration, tubule formation and angiogenesis through a CD44-mediated pathway (39).

NOTCH4 encodes one of the four cell surface receptors of the Notch signalling pathway which has a key role in cell differentiation decisions across many cell types and at different stages of cell lineage progression (40). Notch signalling is important in embryonic development, the regulation of tissue homeostasis and the maintenance of stem cells. Notch signalling is particularly important in vascular development, not only in embryogenesis but also in pathological angiogenesis, prompting research on Notch pathway inhibitors as potential agents to disrupt tumour-related angiogenesis for cancer therapy (41).

Notch signalling plays an important role in the developing retinal vasculature by regulating the specification of endothelial cells into stalk and tip cells (42). In studies of mice, NOTCH4 is strongly expressed in retinal arterial endothelial cells (43). Moreover, Notch signalling has been shown to influence pathological angiogenesis in a study of laser-induced CNV in rats (44). Following laser photocoagulation, inhibition of Notch signalling by intravitreal or subcutaneous administration of the gamma secretase inhibitor DAPT resulted in a substantial increase in the CNV volume compared with controls; conversely, stimulation of Notch signalling by intravitreal administration of Jagged1 peptide, a Notch ligand, substantially reduced the CNV volume (44). In the same report, evidence was presented that Notch signalling contributes to angiogenic homeostasis by providing a counterbalance to proangiogenic pathways such as that mediated by VEGF (44). This led the authors to propose the Notch pathway as a potential therapeutic target for the treatment of CNV, which is the severest form of AMD (44). This might complement the use of anti-VEGF antibodies, which is the mainstay of current treatment. Our finding that SNPs at the NOTCH4 locus are associated with AMD provides support for the involvement of Notch signalling in AMD and points specifically to a role for the NOTCH4 receptor. However, if this were mediated through an alteration in angiogenic homeostasis, we might expect the association to be specifically with CNV. Most of the association signal does indeed come from CNV because this is the condition affecting most of the cases in our study, but the subgroup analysis suggests that there is also an association in the minority of cases with GA. This raises the possibility that other functions of the Notch pathway may be more important, such as its role in the differentiation of immune cells, including macrophages, which, as recent evidence suggests, may play a key part in the pathogenesis of AMD (45).

In summary, our UK GWAS has shown the expected association between AMD and known risk loci at ARMS2–HTRA1, CFH, C2–CFB, C3, CFI–CCDC109B and APOE and with more recently reported risk loci at LIPC and near VEGFA. In addition, we have found novel associations with variants in TNXB–FKBPL and the neighbouring gene NOTCH4 on chromosome 6p21.3 which are independent of the associated SNPs at the nearby C2–CFB locus. TNXB, FKBPL and NOTCH4 are all plausible AMD susceptibility genes, but further research will be needed to identify the causal variants and determine whether any of these genes are involved in the pathogenesis of AMD.

MATERIALS AND METHODS

SUPPLEMENTARY MATERIAL

Supplementary Material is available at HMG online.

FUNDING

This work was supported by the Medical Research Council, UK (grant G0000067 to J.R.W.Y., A.T.M., D.G.C., A.C.B.; separate grants to A.R.W., A.C.B., A.F.W.); the Macular Disease Society (grants to J.R.W.Y., A.T.M., A.J.L. and S.E.); the Guide Dogs for the Blind Association (OR2006-02d to A.T.M., J.R.W.Y., A.R.W., D.G.C., C.B.); Fight for Sight (Mercer Fund to A.R.W.); the Wellcome Trust (grants 061860, 091388, 061858 and 091157 to D.G.C.); the Juvenile Diabetes Research Foundation (grants 4-2000-948, 9-2005-25 and 9-2011-253 to D.G.C.); the Giles' Family Memorial Funds (H.-T.L.); the Macula Vision Research Foundation (A.F.W.); the Chief Scientist Office, Scotland (CZB/4/79 to B.D., C.H., A.M.A., A.F.W.); The Brian Mercer Charitable Trust (A.J.L.); The British Council for the Prevention of Blindness (A.J.L.); The TFC Frost Charitable Trust (A.J.L.); The Gift of Sight (A.J.L.); Centre National de Génotypage (T.L.) and Inserm (Centre d'Investigation Clinique des Quinze-Vingts, J.-A.S.); the Department of Health's NIHR Biomedical Research Centre for Ophthalmology at Moorfields Eye Hospital and UCL Institute of Ophthalmology (A.T.M., A.R.W., J.R.W.Y., C.B.); and the Manchester NIHR Biomedical Research Centre (P.N.B.). The views expressed in the publication are those of the authors and not necessarily those of the Department of Health. We acknowledge the use of genotype data from the British 1958 Birth Cohort DNA collection, funded by the Medical Research Council (G0000934) and the Wellcome Trust (068545/Z/02). H.C. and M.D. have been supported by Cancer Research UK (C3 48/A3758 and A8896, C48/A6361); the Medical Research Council UK (G0000657-53203); the Scottish Executive Chief Scientist's Office (K/OPR/2/2/D333, CZB/4/449); and a Centre Grant from CORE as part of the Digestive Cancer Campaign. The Lothian Birth Cohort collection was supported by the Biotechnology and Biological Sciences Research Council, a Royal Society–Wolfson Research Merit Award to I.J.D. and two awards from the Chief Scientist Office of the Scottish Government's Health Directorates (ETM/55 and CZB/4/505).

Supplementary Material