A Prospective Study of 2 Major Age-Related Macular Degeneration Susceptibility Alleles and Interactions With Modifiable Risk Factors FREE

Age-related macular degeneration (AMD) is a potentially blinding disease with lifestyle and genetic risk factors.¹ A common single nucleotide polymorphism in the gene for complement factor H (CFH) (rs1061170; sequence, T1277C; protein, Y402H) is strongly associated with AMD,^2- 12 and its consistency and plausible biological reasoning suggest that the relationship is causal. Linkage analyses also identified a second major gene for AMD on chromosome 10q, with the strongest evidence for a coding single nucleotide polymorphism in the predicted gene LOC387715.^7,10 Although the evidence of association of these variants with AMD is strong, the magnitude of prior estimates must be considered in light of the study design and the selected nature of cases and controls. In most instances, the cases were prevalent and had features of advanced AMD,^2- 10 and the controls in some studies were restricted to individuals who had no family history of AMD,^4,6 both of which instances might bias risk estimates upward.¹³

In addition, the high prevalence of CFH Y402H and LOC387715 A69S in the white population suggests the possibility of important modifying factors. Elucidation of these modifying factors may increase understanding of disease pathogenesis and suggest lifestyle changes that may prevent AMD or delay the disease onset in carriers of predisposing genetic variants.

We therefore investigated the associations between CFH Y402H and LOC387715 A69S and AMD in a prospective, nested case-control study within the Nurses' Health Study and the Health Professionals Follow-up Study. We aimed to further clarify the magnitude of these relationships and determine whether modifiable risk factors act synergistically with CFH or LOC387715 to alter the risk of AMD.

The Nurses' Health Study is an ongoing cohort study of 121 700 primarily white US female registered nurses who were 30 to 55 years old in 1976. The Health Professionals Follow-up Study includes 51 529 mostly white US male health care professionals who were 40 to 75 years old in 1986. All participants completed a mailed questionnaire every 2 years, on which information was obtained on lifestyle factors, including height and weight (from which we calculated the body mass index as weight in kilograms divided by the square of height in meters), cigarette smoking history,¹⁴ regular aspirin use, and diet (assessed via validated semiquantitative food frequency questionnaires).^15- 16 Risk factor information was collected in a similar manner in both cohorts except for regular aspirin use, which was assessed on the basis of reported use of aspirin for 15 or more days per month among women and 2 times per week or more among men. This difference in definition was necessitated by the different questions used in the 2 cohorts. We used exposure information at baseline, which was defined as the time of blood collection. We obtained blood samples from 32 826 Nurses' Health Study participants (1989-1990) and from 18 162 Health Professionals Follow-up Study participants (1993-1995); these form the study base. The study protocol was approved by the institutional review board of Brigham and Women's Hospital and the Harvard School of Public Health's Human Subjects Committee.

Procedures for our 2-stage documentation of incident AMD have been previously described and validated.^14,16 On the biennial study questionnaires, we asked participants about the diagnosis of AMD and requested permission to review their medical records. We sent a letter to the participant's eye physician to obtain information on the date of diagnosis, best-corrected visual acuity at the time of diagnosis and at the most recent examination, and the presence of chorioretinal lesions (eg, drusen, retinal pigment epithelial changes [including atrophy, hypertrophy, and detachment], geographic atrophy, subretinal neovascular membrane, and disciform scar). Only those participants in whom we confirmed the presence of 1 or more typical AMD lesions with a visual acuity of 20/30 or worse attributable to these lesions, who were first diagnosed as having AMD after the date of receipt of the baseline blood specimen, and who were 50 years or older were selected as cases for the present study. The visual acuity criterion is included to reduce the possibility of surveillance bias, which is more likely for early AMD, and to limit the analysis to those cases of greatest clinical and public health importance. We classified cases as neovascular AMD if there was a retinal pigment epithelial detachment, subretinal neovascular membrane, or disciform scar not due to other causes (eg, histoplasmosis or choroidal rupture). Dry AMD included those cases with the documented presence of drusen and/or retinal pigment epithelial changes but with no signs of neovascular AMD. We classified participants based on the most severely affected eye.

To balance our use of these valuable resources with maximization of study power and our ability to perform subgroup analyses, we aimed to select 2 control subjects for each case of dry AMD and 3 controls for each case of neovascular AMD. We selected controls at random from study participants in the same cohort as the case. The controls were still at risk of AMD at the time the case was diagnosed as having AMD, they were within 1 year of the same age, and they reported having undergone an eye examination in the past 2 years.

We extracted DNA from the buffy coat fraction of the centrifuged blood specimens using a commercially available blood kit (QIAmp; Qiagen Inc, Valencia, Calif). All samples were genotyped using the same sequence detection system (ABI Prism 7900HT; Applied Biosystems, Foster City, Calif) in 384-well format. The 5′ nuclease assay (TaqMan; Applied Biosystems) was used to distinguish the 2 alleles of each polymorphism. Polymerase chain reaction amplification was performed on 5 to 20 ng of DNA using 1× TaqMan universal polymerase chain reaction master mix (No Amp-erase UNG; Applied Biosystems), 900nM forward and reverse primers, 200nM FAM-labeled probe (Applied Biosystems), and 200nM VIC-labeled probe (Applied Biosystems) in a 5-μL reaction volume. Amplification conditions on a dual-plate thermal cycle (AB 9700; Applied Biosystems) were as follows: 1 cycle at 95°C for 10 minutes followed by 50 cycles at 92°C for 15 seconds and 58°C for 1 minute. The TaqMan primers and probe sequences CFH Y402H are the same as those used and reported previously by Hageman et al.⁶ Primers and probe sequences for LOC387715 A69S (ABI Assays-By-Design; Applied Biosystems) are available from the authors on request. All laboratory personnel were masked to case-control status, and split-masked quality control samples were inserted to validate genotyping procedures with 100% concordance for genotypes.

We performed χ² analysis to test for departures from Hardy-Weinberg equilibrium. We compared genotype and allele frequencies between cases and controls using χ² tests and used conditional logistic regression to estimate the incidence rate ratios (IRRs) and 95% confidence intervals (CIs) for each genotype adjusted for other risk factors. We initially obtained separate estimates of the IRR in each cohort and tested for heterogeneity. Because there was no evidence of heterogeneity between the cohorts (P=.80 for CFH Y402H and P=.59 for LOC387715 A69S), we provide herein only pooled data from the 2 cohorts.

We initially modeled the allelic effects using a multiplicative (ie, log-additive) coding scheme such that we created a single variable for each locus coded 0 for homozygous wild-type, 1 for heterozygous, and 2 for homozygous variants. We then fit an unconstrained (ie, codominant) model using separate indicator variables for heterozygous and homozygous variants. We considered models for each locus separately, as well as 2-locus models for the main effects and 2-locus models that included terms for interaction (the least constrained models). To arrive at the best-fitting model, we compared these alternative models using the Akaike Information Criterion.¹⁷ As a rule of thumb, an Akaike information criterion difference of less than 2 suggests substantial evidence in favor of the more parsimonious model.

We extended the preferred models to control for potential confounding by other risk factors, by fitting multivariate models adjusted for current cigarette smoking (yes vs no), obesity (body mass index ≥30 vs <30), regular aspirin use (yes vs no), alcohol intake (continuous), consumption of fruits (continuous), and the intake ratio of ω-6 to ω-3 fatty acids in the diet (continuous). We used a model-based method based on the observed data to calculate the attributable fraction in the population as a measure of the proportion of AMD cases to which each polymorphism contributes^18- 19 (Handan Wand, PhD, Donna Spiegelman, PhD, and Stephen Walter, PhD; unpublished data; 2000).

We fit additional models to simultaneously estimate the stratum-specific IRR for the joint effects of each variant and a prespecified group of modifiable risk factors. We were primarily interested in interactions with cigarette smoking and obesity because these factors can stimulate complement activation and/or other inflammation-related pathways. In addition, we explored possible modification by regular aspirin use, dietary intake of fruits, the intake ratio of ω-6 to ω-3 essential fatty acids, and alcohol consumption. To evaluate possible departures of the IRR among cases jointly exposed to each genetic variant and modifiable risk factor from the product of the IRR for gene only and that of the IRR for the risk factor only, we formed interaction terms by multiplying the ordinal variable representing the number of risk alleles by a dichotomous variable representing each risk factor. We dichotomized continuous risk factors for reasons of power and ease of interpretation into the following categories: intake of fruits (the lowest fifth vs the upper four fifths of the distribution among the controls), the intake ratio of ω-6 to ω-3 fatty acids (the upper two fifths vs the lower three fifths of the distribution among the controls), and alcohol consumption (the upper fifth vs the lower four fifths of the distribution among the controls).

Finally, we conducted a meta-analysis to summarize the association between CFH Y402H and LOC387715 A69S and AMD risk among the published studies. We tested for heterogeneity among the reported risks using a χ² statistic and calculated the summary estimates from the random-effects model of DerSimonian and Laird,²⁰ which incorporates within-study and between- study variability. All P values are 2 sided.

The study group consisted of 457 cases, including 293 with incident dry AMD matched with 586 controls and 164 with incident neovascular AMD matched with 485 controls. The mean ± SD age at AMD diagnosis was 68.7 ± 5.0 years. Dry AMD cases were primary early AMD characterized by drusen (17.7%), pigment epithelial changes (33.2%), or both (49.1%), but also included 2 cases of central geographic atrophy. We found significant differences in the expected direction between the AMD cases and controls in the baseline distributions of obesity (P=.04), cigarette smoking (P<.001), dietary intake of fruits (P=.03), and consumption of alcohol (P=.01). Regular aspirin use (P=.27), dietary intake of lutein and zeaxanthin (P=.71), and the intake ratio of ω-6 to ω-3 fatty acids (P=.22) were not different between the cases and controls (Table 1).

Genotype data were successfully obtained for 437 AMD cases (95.6%) and for 1015 controls (94.8%) for CFH, and 445 AMD cases (97.4%) and 1041 controls (97.1%) for LOC387715. For both CFH and LOC387715, the genotype distribution among the controls was in Weinberg equilibrium (P=.80 and P=.36, respectively). In unadjusted analyses (Table 2), we observed a prevalence of 47.6% for CFH Y402H heterozygosity and 29.7% for CFH Y402H homozygosity among AMD cases compared with 45.5% and 12.9%, respectively, among controls (P<.001). There was also a highly significant association between LOC387715 A69S genotype and risk of AMD (P<.001), with 15.3% of AMD cases and 3.9% of controls being homozygous for this variant.

In conditional logistic regression models not adjusted for other risk factors, there were highly significant associations between the CFH genotype and risk of AMD, regardless of the genetic model assumed (data not shown). Based on the comparison of the Akaike information criterion differences, the best-fitting model included log-additive effects of CFH Y402H and LOC387715 A69S and their interaction. However, we could not exclude on statistical grounds the more parsimonious model that demonstrated the log-additive effects of the 2 genes but not their interaction (Akaike information criterion difference, 1.74).

In further analyses undertaken by modeling the independent log-additive effects of both genetic variants, estimates were virtually unchanged after additionally controlling for other AMD risk factors. Specifically, compared with individuals who were homozygous for the low-risk allele, the risk of developing AMD increased by a factor of 1.98 (95% CI, 1.64-2.40) among those individuals heterozygous for CFH Y402H, and by a factor of 3.92 (95% CI, 2.69-5.76) among those homozygous for CFH Y402H. Corresponding estimates for subgroups of dry and neovascular AMD cases are presented in Table 3. Similarly, the multivariate-adjusted IRR (95% CI) for developing AMD was 2.38 (1.92-2.96) for subjects heterozygous for LOC387715 A69S and 5.66 (3.69-8.76) among homozygous subjects compared with those with no LOC387715 A69S risk alleles. The IRR for neovascular AMD among subjects homozygous for LOC387715 A69S was 8.53 (4.33-16.97) compared with those with no A69S alleles (Table 3).

Given the evidence of supermultiplicative joint effects of the 2 genetic variants observed in our initial model comparison, we simultaneously estimated the stratum-specific IRR of AMD for each possible combination of CFH Y402H and LOC387715 A69S genotypes (Table 4). Estimates from this model demonstrated a 50.48-fold increased risk of AMD among subjects who were homozygous for the risk variant at both loci compared with subjects with no risk alleles at either locus, but the 95% CI was wide (10.8-236.6). Calculation of the population-attributable fractions showed an attributable fraction (95% CI) of 44.6% (39.0%-50.3%) for CFH Y402H, 34.3% (30.1%-38.8%) for LOC387715 A69S, and 62.8% (57.9%-67.5%) for both genes together.

Testing of multiplicative interaction terms demonstrated no statistically significant departures from multiplicative joint effects between the IRR for CFH Y402H and cigarette smoking (P=.72 for interaction) or obesity (P=.56 for interaction) or between the IRR for LOC387715 A69S and these factors (Table 5). However, because the lack of such statistical interaction on the multiplicative scale does not necessarily imply the absence of synergy in biological effects or translate well into public health messages,^21- 22 we also calculated the IRR for each genotype–risk factor combination to identify subjects at particularly high risk of AMD (Table 5). Although statistically consistent with multiplicative effects, these estimates indicate that, compared with nonobese subjects with no CFH Y402H risk alleles, there is a 4-fold increased risk among participants homozygous for CFH Y402H who are not obese, whereas the risk is 12-fold among homozygous participants who are also obese. Corresponding IRRs for nonsmoking vs smoking homozygous carriers of CFH Y402H are 4.23 and 8.69, respectively (Table 5). Similarly, for LOC387715 A69S, compared with nonsmoking subjects with no risk alleles, the estimated IRRs are 6.33 for homozygous A69S carriers who do not smoke and 22.47 among current smokers. We observed no statistically significant departures from multiplicative interaction between CFH Y402H and regular aspirin use (P=.88 for interaction), fruit intake (P=.44 for interaction), ω-6 to ω-3 fatty acid intake ratio (P=.81 for interaction), or alcohol consumption (P=.34 for interaction) or between LOC387715 A69S and these factors (P=.10, P=.77, P=.42, and P=.51 for interaction, for regular aspirin use, fruit intake, ω-6 to ω-3 fatty acid intake ratio, and alcohol consumption, respectively). For these factors, comparison of the stratum-specific IRR did not suggest large differences in risk conferred by either genotype among exposed vs unexposed participants, with the possible exception of LOC387715 and regular aspirin use. Compared with participants with no A69S risk alleles who used aspirin regularly, the risk of AMD was increased 7-fold among subjects homozygous for A69S who did not use aspirin regularly, but just more than 2-fold among regular aspirin users who were homozygous for A69S.

To provide perspective on the consistency of the evidence for AMD susceptibility conferred by these 2 genes, we performed a meta-analysis to summarize available data from studies published as of the time of our analysis. Tests for heterogeneity were not significant for estimates of CFH Y402H heterozygosity (P=.29) or homozygosity (P=.22). Summary estimates for CFH Y402H demonstrate an increased risk of 2.27 (95% CI, 2.03-2.53) among heterozygotes and 5.33 (95% CI, 4.57-6.22) among homozygotes (Figure 1). Similarly, tests for heterogeneity were not significant for LOC387715 A69S (heterozygotes, P=.06; homozygotes, P=.51), and pooled estimates were 2.18 (95% CI, 1.67-2.85) among heterozygotes and 7.14 (95% CI, 5.54-9.20) among homozygotes (Figure 2).

Age-related macular degeneration is a result of the combined effects of genetic and lifestyle risk factors and is the most common cause of blindness in the United States.^1,26- 27 Recently, 2 major genetic susceptibility alleles for AMD were discovered in prevalent case-control studies,^2- 10 and the present study of incident AMD confirms that most of the AMD cases in these 2 large cohorts (62.8%) involve one or both of these variants, and our results provide additional evidence regarding the complexity of interactions between these susceptibility alleles and modifiable risk factors.

Data from our controls suggest that the CFH Y402H allele is present in 58.4% of the US white population. The role of CFH in innate immunity implies an ability to influence a variety of biological systems and conditions in addition to its effect on risk of AMD.²⁸ The LOC387715 A69S allele is also common, with 33.5% of controls having at least 1 risk allele. It remains unclear whether A69S is the causal variant involved in AMD, and little is known of the function of LOC387715. Nonetheless, it is clear that the expression of these genes is not specific to the eye and that pleiotropic effects are likely. Such effects could alter the clinical course or survival of individuals who carry the risk alleles and may introduce selection bias when prevalent rather than incident cases are studied.²² Using our validated methods, we therefore undertook a large, prospective, nested case-control study and replicated findings from other groups.^2- 10 Our statistical summary of the available evidence confirms highly significant 2- and 5-fold increases in the incidence of AMD among those with 1 and 2 copies, respectively, of CFH Y402H. In addition, we also confirm in our large prospective cohorts that LOC387715 A69S (or a closely linked locus) is a second major susceptibility allele for AMD, which combined evidence suggests confers a 2-fold higher risk among heterozygotes and a 7-fold higher risk among homozygotes.

We also examined whether the effect of these variants is influenced by cigarette smoking or obesity, as well as other modifiable risk factors. We found no statistical evidence of departures from multiplicative interaction between these variants and modifiable risk factors. However, there is a strong argument that one must separate the concept of joint biological effects from the issue of statistical testing of multiplicative interaction terms.^21- 22 Epidemiologists have consequently proposed surrogate measures in case-control studies to test for departures from the additivity of risks, which relate more closely to public health messages,^21- 22 but there is debate regarding whether current approaches are valid and robust.²⁹ We therefore provided estimates of the joint effects of these 2 susceptibility genes, as well as the joint effects of each variant with modifiable risk factors, which are consistent with a multiplicative relationship of public health importance. For example, subjects homozygous for CFH Y402H have a 12-fold increased risk if they are also obese and nearly a 9-fold increased risk if they also smoke cigarettes, compared with persons who are exposed to neither the genetic variant nor the modifiable risk factor. In addition, cigarette smoking appears to at least multiply the risk conferred by LOC387715 A69S, showing a 22-fold higher risk among subjects homozygous for A69S who also smoked cigarettes. These data are supported by the recent findings of stronger effects of CFH Y402H³⁰ and LOC387715 A69S among smokers.²⁵ Despite no significant overall effect of aspirin on the risk of AMD, these data suggest that regular aspirin use may reduce the risk conferred by homozygosity for LOC387715 A69S. This possibility may be worth investigating in larger studies.

Strengths of the present study include its prospective design, large sample size, and prospective information on lifestyle exposures. One potential limitation relates to our method for ascertainment of the AMD cases. Owing to prohibitive costs and logistical difficulties, we cannot perform standardized clinical assessments of retinal status among participants of these large and geographically dispersed cohorts. Nonetheless, we have demonstrated that our case ascertainment method has high specificity,^16,31 which ensures minimal bias in a prospective study.²² Furthermore, that some participants who were selected as controls may at a later point develop AMD is not a limitation but rather a predictable characteristic of the risk set sampling method that does not introduce bias.²² The consistency of our findings linking CFH Y402H and LOC387715 A69S with AMD, as well as prior work from our laboratory^{15- 16,31- 33} on modifiable risk factors for AMD, provides further reassurance of the validity of these results, compared with studies in which photographic evidence of case and control status was used.^2- 10,25 Nevertheless, the generalizability of the findings is uncertain owing to the selected nature of these cohorts of health care professionals. The main concern relates to the imprecision of stratum-specific interaction effect estimates due to the sparseness of data. Some of these exposures, such as cigarette smoking, are less common in our cohorts of health care professionals than in the general population or in previous case-control studies,^7,25 making it more difficult to detect interactions. Further study of interactions in prospective studies with larger samples would be helpful in addressing this issue.

There is growing evidence that exposure to stressors such as UV radiation, oxidation, and infection results in the development of a chronic subclinical inflammatory response propagated by the complement cascade.³⁴ The intensity of this response may depend in part on genetic predisposition, for example, that conferred by common genetic variants such as CFH Y402H. The function of LOC287715 is unknown, but our data suggesting synergy with CFH and a stronger impact of the A69S variant among cigarette smokers support the hypothesis that this or a closely linked gene may also play a role in closely related pathways. The high-risk CFH Y402H variant is predicted to have functional relevance through its ability to decrease the effectiveness of CFH-mediated complement inhibition. Some risk factors for AMD, such as cigarette smoking and obesity, are also known to promote inflammatory activity.³⁵ These observations taken in the context of the high prevalence of these polymorphisms suggest that quantifying their interrelationships with modifiable risk factors for AMD is of substantial public health importance. Data from the present study suggest that, in situations in which underlying levels of inflammatory activity are likely to be elevated, the impact of carrying AMD-associated risk alleles is magnified.

The high prevalence of the CFH Y402H and LOC397715 A69S and their strong association with AMD raises the question of the utility of population-based genetic testing. The existence of interactions with modifiable lifestyle factors may provide further impetus for screening individuals who are at potentially greater risk, for example, cigarette smokers or the obese. Knowledge of the substantial risk of AMD among individuals homozygous for either or both of these major AMD-associated variants might help motivate these individuals to stop smoking, lose weight, modify other risk factors, and have regular eye examinations. On the other hand, the high prevalence of these variants relative to the proportion of individuals who develop visually significant AMD implies a relatively low positive predictive value for genetic testing. Ongoing research suggests that joint consideration of other loci within these or other genes may improve the predictive value of genetic testing in the future.^6,36 Screening is thought to be most appropriate in those instances in which early detection would result in a more favorable outcome.³⁷ Although options for early intervention in AMD are currently few, further development of new treatments that show promise in slowing the progression of visual loss from AMD^38- 40 may provide an additional rationale for genetic testing as more is learned about the interrelationships of the multiple genetic and environmental determinants of AMD.

Studies such as this one are useful to broaden the clinical perspective on the influence of the combined effects conferred by common genetic polymorphisms and modifiable risk factors. The CFH Y402H and LOC387715 A69S genetic variants appear to contribute to most of the AMD cases in the population and to act synergistically with modifiable risk factors. These data reinforce the complexity of the interactions among the genetic and modifiable risk factors in the pathogenesis of AMD. In addition, they support the need for renewed attention to risk modification strategies to reduce the public health impact of this leading cause of blindness and visual impairment.

Correspondence: Debra A. Schaumberg, ScD, OD, MPH, Division of Preventive Medicine, Brigham and Women’s Hospital, 900 Commonwealth Ave E, Third Floor, Boston MA 02215 (dschaumberg@rics.bwh.harvard.edu).

Submitted for Publication: July 20, 2006; final revision received September 14, 2006; accepted September 19, 2006.

Financial Disclosure: None reported.

Funding/Support: This study was supported in part by grants EY013834, EY009611, CA87969, CA49449, and HL35464 from the National Institutes of Health. The TaqMan primers and probe sequences for CFH Y402H were obtained from Gregory Hageman, PhD, University of Iowa, Iowa City.