Analysis of Rare Variants in the C3 Gene in Patients with Age-Related Macular Degeneration

Age-related macular degeneration (AMD) is a progressive retinal disorder affecting over 33 million people worldwide. Genome-wide association studies (GWASs) for AMD identified common variants at 19 loci accounting for 15–65% of the heritability and it has been hypothesized that the missing heritability may be attributed to rare variants with large effect sizes. Common variants in the complement component 3 (C3) gene have been associated with AMD and recently a rare C3 variant (Lys155Gln) was identified which exerts a large effect on AMD susceptibility independent of the common variants. To explore whether additional rare variants in the C3 gene are associated with AMD, we sequenced all coding exons in 84 unrelated AMD cases. Subsequently, we genotyped all identified variants in 1474 AMD cases and 2258 controls. Additionally, because of the known genetic overlap between AMD and atypical hemolytic uremic syndrome (aHUS), we genotyped two recurrent aHUS-associated C3 mutations in the entire cohort. Overall, we identified three rare variants (Lys65Gln (P = 0.04), Arg735Trp (OR = 17.4, 95% CI = 2.2–136; P = 0.0003), and Ser1619Arg (OR = 5.2, 95% CI = 1.0–25; P = 0.05) at the C3 locus that are associated with AMD in our EUGENDA cohort. However, the Arg735Trp and Ser1619Arg variants were not found to be associated with AMD in the Rotterdam Study. The Lys65Gln variant was only identified in patients from Nijmegen, the Netherlands, and thus may represent a region-specific AMD risk variant.


Introduction
Age-related macular degeneration (AMD, MIM 603075) is a retinal disorder that causes progressive visual impairment in individuals aged over 50 years [1]. AMD primarily affects the macula, the central region of the retina, eventually leading to loss of central and sharp vision. It has been estimated that more than 33 million people suffer from vision loss due to AMD worldwide [2]. AMD is a multifactorial disease caused by a combination of genetic and environmental factors. GWASs identified common variants at 19 loci that influence disease susceptibility, accounting for 15-65% of the heritability [3,4].
It has been hypothesized that the remaining genetic fraction influencing the risk for development of AMD, the so-called missing heritability, may be explained by rare, highly penetrant variants [4]. Simulation studies suggested that common variants are insufficient to account for disease burden in densely affected AMD families and that rare penetrant variants would offer a likely explanation [5]. In addition, a meta-analysis of AMD GWASs suggested that each of the 19 loci may harbour several independent variants associated to AMD susceptibility [3]. The rare variant hypothesis is supported by the identification of rare missense mutations in the fibulin-5 (FBLN5) gene and the hemicentin-1 (HMCN1) gene in AMD patients [6,7]. In addition, rare, highly penetrant variants in the genes encoding complement factor H (CFH), complement factor I (CFI), complement component 3 (C3) and complement component 9 (C9) have recently been found to be associated with AMD [8][9][10][11][12].
Genetic studies have identified an important role for the complement cascade in the pathogenesis of AMD [13]. Interestingly, recent studies suggested a genetic overlap between AMD and atypical hemolytic uremic syndrome (aHUS), a life-threaten-ing renal disease caused by chronic, uncontrolled activation of the complement system. It has been reported that 4-10% of aHUS patients carry mutations in the C3 gene [14]. Moreover, diseasecausing mutations previously identified in aHUS patients, such as Arg1210Cys in CFH, Gly119Arg in CFI and Lys155Gln in C3 were found to confer a high risk of developing AMD [8][9][10][11][12]. However, the precise nature of this genetic overlap between two clinically distinct phenotypes remains unknown.
In this study, we explored the role of rare variants in the C3 gene in the pathogenesis of AMD. We performed a two-stage analysis to identify rare variants in the C3 gene. First, sequence analysis was carried out in a discovery set of 84 AMD cases from the EUGENDA cohort, and subsequently the frequencies of these variants were determined in replication sets from EUGENDA and from the Rotterdam Study consisting of 1474 AMD cases and 2258 controls. In addition, two recurrent aHUS-associated C3 mutations were genotyped in the entire AMD case-control cohort [15,16].

Results
To investigate the involvement of rare variants in the C3 gene in AMD, the exons and flanking introns of C3 were sequenced in a discovery cohort of 84 AMD cases (Table 1) from the Nijmegen area, the Netherlands. Sequencing identified three rare variants (MAF,1%; Lys155Gln/rs147859257, Arg735Trp/rs117793540 and Ser1619Arg/rs2230210) and two common variants (MAF$1%; Arg102Gly/rs2230199 and Pro314Leu/rs1047286) ( Table 2). None of the rare variants were found in 192 ethnicitymatched and age-matched controls. The Lys155Gln variant, which has recently been associated with AMD [10][11][12], was found in five cases, while variants Arg735Trp and Ser1619Arg were found in one case each. Bioinformatic algorithms SIFT and PolyPhen predicted the variants Arg102Gly, Lys155Gln, and Pro314Leu not to be damaging whereas Arg735Trp and Ser1619Arg were predicted to be damaging to the protein function (Table 2).
Next, the frequencies of the rare variants Arg735Trp and Ser1619Arg, as well as the common variants Arg102Gly and Pro314Leu were determined, in replication cohorts from EU-GENDA and from the Rotterdam Study, consisting of 1474 AMD cases and 2258 controls of European ancestry (Table 3). In addition, two recurrent aHUS mutations (Lys65Gln and Arg161Trp) were included in the analysis ( Table 3). The common variants Arg102Gly (OR = 1.2 [95% CI 1.1-1.4]; P = 0.001) and Pro314Leu (OR = 1.2 [95% CI 1.0-1.4]; P = 0.005) were signif-icantly associated with AMD. In the EUGENDA cohort, the rare variant Arg735Trp was found heterozygously in 8 and homozygously in one out of 718 AMD cases and heterozygously in one out of 1246 controls (OR = 17.4 [95% CI 2.2-136]; P = 0.0003). However, in the Rotterdam Study the rare variant Arg735Trp was found heterozygously in 1 out of 785 AMD cases and heterozygously in 2 out of 1048 controls, and was thus not associated with the disease. Rare variant Ser1619Arg was found heterozygously in six out of 718 AMD cases and heterozygously in two out of 1244 controls in the EUGENDA cohort (OR = 5.2 [95% CI .1.0-25]; P = 0.05). In the Rotterdam Study the rare variant Ser1619Arg was found heterozygously in 6 out of 835 AMD cases and heterozygously in 11 out of 1279 controls, and was thus not associated with AMD. The aHUS mutation Lys65Gln was found heterozygously in three out of 717 AMD cases and was not observed in 1246 controls (P = 0.05) in the EUGENDA cohort, but was not identified in the Rotterdam Study. Arg161Trp was found heterozygously in two out of 644 AMD cases and was not observed in 1142 controls (P = 0.13) in the EUGENDA cohort. In the Rotterdam Study the aHUS mutation Arg161Trp was observed heterozygously in one out of 320 AMD cases and heterozygously in one out of 483 controls (P = 1.0), and was not significantly associated with AMD in the EUGENDA cohort nor in the Rotterdam Study.
To determine whether the identified variants conferred disease risk independent of the two known common C3 variants (Arg102Gly/rs2230199 and Pro314Leu/rs1047286), a conditional logistic regression analysis was performed (Table 4). After conditioning on Arg102Gly/rs2230199, Arg735Trp remained associated with disease risk in the EUGENDA cohort (OR = 22.1, 95% CI = 2.8-173; P = 0.003). Similarly, after conditioning on Pro314Leu/rs1047286, Arg735Trp still showed association with disease risk in the EUGENDA cohort (OR = 22.0, 95% CI = 2.8-172; P = 0.003). In addition, Arg735Trp was significantly associated with disease risk in the EUGENDA cohort (OR = 22.1, 95% CI = 2.8-173; P = 0.003) after conditioning on both variants (Arg102Gly and Pro314Leu). Independent association with AMD could not be assessed for Lys65Gln and Ser1619Arg because too few data points were available to perform a reliable conditional analysis.

Discussion
In this study, three rare variants (Lys65Gln, Arg735Trp and Ser1619Arg) in the C3 gene were shown to be associated with AMD disease risk in our EUGENDA cohort. However, the Arg735Trp and Ser1619Arg variants were not found to be associated with AMD in the Rotterdam Study. The Arg735Trp and Ser1619Arg variants were also not associated with AMD in a recent study that analyzed the C3 gene in 2,493 AMD cases and controls [10]. The Lys65Gln variant was only identified in AMD patients from the Nijmegen area, while it was not found in the Rotterdam Study, nor in a cohort from Boston [10]. The Lys65Gln variant may therefore represent a region-specific AMD risk variant, being confined to the east of the Netherlands, which is confirmed by the occurrence of Lys65Gln in Dutch aHUS patients [16]. A fourth variant, Arg161Trp, was found in three AMD cases and in one control; therefore it cannot be ruled out that the variant is in fact associated with AMD but a larger sample size would be required to detect a significant association.
The complement system plays an important role in the pathogenesis of AMD [8][9][10][11][12]. The C3 gene encodes the complement component 3 protein, a central component of the complement cascade that plays a crucial role in clearance of pathogens and immune complexes. All three branches of the complement system, the classical pathway, the mannose-binding lectin pathway and the alternative pathway, converge at C3. At this stage, activation of C3 results in an amplification loop that ultimately forms a cytolytic membrane attack complex which targets pathogens. The system is tightly controlled by molecules like CFH and CFI. If the complement cascade is improperly regulated, host cells may also be subjected to complement attack, resulting in tissue damage and a spectrum of complementmediated diseases [17]. The rare variants reported in this study have previously been described for their association with aHUS, a chronic renal disorder caused by uncontrolled activation of the complement system, often leading to renal failure. Functional experiments have been performed in this context to determine the effect of the amino acid substitutions on C3 function. The Lys65Gln variant resides in the macroglobulin (MG1) domain of C3 and was shown to cause decreased binding to CFH [16]. Although not significantly associated in our study, the Arg161Trp variant was shown to cause hyperactive C3 convertase due to increased binding to factor B (CFB), and reduced binding to CFH and membrane cofactor protein (MCP; CD46), resulting in higher complement activity [15,16,18].
Recombinant protein studies have shown that the Arg735Trp variant performs normal in binding and cleavage assays [14]. However, since the Arg735Trp residue is located in the anaphylotoxin (C3a) domain of C3, it is not likely to show an effect in the performed assays that evaluated binding and activation of the C3b fragment. Recent studies in animal models suggest that C3a anaphylotoxin has specific functions in the retina and retinal pigment epithelium, but it remains unknown how an amino acid substitution in C3a may contribute to the development of AMD [19,20].
The Lys65Gln variant associated with AMD in this study has previously been associated with aHUS, further supporting the genetic overlap between AMD and aHUS. To rule out any renal pathology, the medical histories of AMD patients who carried this variant were evaluated, but no signs of aHUS or other renal pathologies were reported in these patients. Thus, although aHUS and AMD may overlap genetically, in this study no clinical overlap was shown, suggesting that compounding (genetic or environmental) factors contribute to a particular clinical phenotype. Such a notion is further supported by a study that showed that some aHUS patients carry multiple mutations in the complement factor genes [21]. In order to understand the shared associations, crossphenotype studies are warranted to unravel the mechanisms common and unique to aHUS and AMD. This will lead to more rational approaches to diagnosis and therapy by targeting these specific molecular targets.
The identification of rare penetrant AMD-associated variants may have relevance for diagnostic, predictive and therapeutic purposes, although the exact interpretation may remain a challenge. Recent studies have identified several highly penetrant rare variants (Lys155Gln in C3, Gly119Arg in CFI, Pro167Ser in C9 and Arg1210Cys in CFH), to be associated with AMD [8][9][10][11][12]. However, not all associations hold true among different populations. An example of this is the Arg1210Cys variant in CFH which was strongly associated in a North American AMD cohort [8] but not in an Icelandic AMD cohort [11] (Table 5). Genetic drift, founder effects or differences in genetic make-up that could compensate for a rare disruptive variant may underlie this phenomenon. In addition, the possibility that environmental effects could mask or enhance the penetrance of certain alleles between populations may also exist. To date, it remains unclear how these observations impact the predictive value of finding such variants in individuals. In contrast, variants that were proven to be functionally impaired and that are associated with AMD in several populations, such as the Lys155Gln variant in C3, have a much stronger predictive value.
In conclusion, we report a rare variant (Lys65Gln) at the C3 locus in patients with AMD, while an association with two other variants (Arg735Trp and Ser1619Arg) was not confirmed in other cohorts. This study further supports that rare variants contribute to the genetic variance of AMD, which may have implications for predictive testing and personalized medicine in AMD [10].

Cases and controls
Cases and controls included in this study were selected from the population-based Rotterdam Study, Rotterdam, The Netherlands and from the European Genetic Database (EUGENDA), a multicenter database comprising participants from the Cologne area, Germany and the Nijmegen area, the Netherlands. All participants of this study underwent extensive retinal imaging which has been described in detail elsewhere [22,23]. In short, AMD staging was performed by grading of stereo fundus photographs according to the standard protocol of the Rotterdam grading center and Cologne Image Reading Center (CIRCL). All subjects were classified on the basis of the eye with the more severe diagnosis. Cases were aged $50 years of age and AMD was classified by the presence of at least 15 intermediate (63-124 mm)

Sequencing
Sanger sequencing of the C3 (NM_000064) gene was performed in the discovery set, consisting of 84 AMD patients selected from the EUGENDA database. Primers were designed for all 41 coding exons and intron-exon boundaries by Primer3 software (Supplementary Material, Table S1). Polymerase chain reaction (PCR) was performed, and PCR amplicons were sequenced using an automated sequencer (BigDye Terminator, version 3, 3730 DNA analyzer; Applied Biosystems). Sequences were assembled and analysed using ContigExpress (Vector NTI Advance, Version 11.0, Life Technologies). Each newly identified variant was confirmed by a second independent PCR and bidirectional Sanger sequencing. The predicted effects of identified missense variants were examined using Polymorphism Phenotyping (PolyPhen) [24] and Sorting Intolerant from Tolerant (SIFT) [25].

Genotyping in the EUGENDA cohort
Variants (Lys65Gln, Arg102Gly, Arg161Trp, Pro314Leu, and Arg735Trp) were genotyped in the EUGENDA cohort using competitive allele-specific PCR assays (KASPar SNP Genotyping System, KBiosciences). KASPar genotyping was performed according to the manufacturer's protocol in a volume of 4 ml containing 10 ng of genomic DNA, 2.5 ml of 26reaction mix, and 0.069 ml of assay. Thermal cycling conditions included a preincubation step at 94uC for 15 min, 20 cycles of 94uC for 10 s, 57uC for 5 s, 72uC for 10 s, followed by 23 cycles of 94uC for 10 s, 57uC for 20 s, 72uC for 40 s. Plates were analyzed on a 7900 Fast Real-Time PCR system (Applied Biosystems). The Ser1619Arg variant was genotyped in the EUGENDA cohort by Amplification Refractory Mutation System (ARMS) [26] PCR (Supplementary Material, Table S2). PCR reactions were performed in a volume of 12.5 ml using 20 ng genomic DNA, 16 buffer, 2.5 mM MgCl 2 , 1 mM deoxyribonucleotide triphosphates, 0.2 mM of each primer, and 0.5 U Taq DNA polymerase (Invitrogen, Life technologies). Thermal cycle conditions included a pre-incubation step at 95uC for 5 min, 16 cycles of 95uC for 30 s, 69uC for 30 s, 72uC for 45 s, followed by 25 cycles of 95uC for 30 s, 67uC for 30 s, 72uC for 45 s. PCR amplicons were analysed by agarose gel electrophoresis. Each newly identified variant was confirmed by a second independent PCR and bidirectional Sanger sequencing.

Exome sequencing, exome chip analysis and genotyping in the Rotterdam Study
The occurrence of the C3 variants Lys65Gln, Arg161Trp, Arg735Trp, and Ser1619Arg in the Rotterdam Study (RS) was retrieved from exome chip and exome sequencing data. For exome sequencing purposes, genomic DNA of RS participants was Table 3. Genotyping of C3 variants in EUGENDA and Rotterdam samples.

Statistical analysis
A Fisher's exact test was performed to assess the association between each variant and AMD, and also to check for Hardy-Weinberg equilibrium. P-values were calculated two-sided, and values of ,0.05 were considered as statistically significant. Logistic  regression analysis was performed to check for genotype interactions, to estimate the odds ratios (OR) with 95% confidence intervals (CI) and to adjust for age. Conditional analysis was performed to identify secondary association signals at the C3 locus by accounting for two known AMD SNPs. Either Arg102Gly/ rs2230199 or Pro314Leu/rs1047286 was added independently and combined to the regression model as a covariate to test the effect of the variant of interest. Data were analyzed using SPSS software, version 18.0.

Supporting Information
Table S1 List of C3 gene sequencing primers. (DOC)