Genetics and Age-Related Eye Disease Study Formulation Interaction in Neovascular Age-Related Macular Degeneration

Abstract

Purpose:

This work aims to determine whether previously defined genotype risk groups interact with Age-Related Eye Disease Study formulation (AREDS-F) use in progression to neovascular age-related macular degeneration (nvAMD).

Methods:

We conducted a case-only study of 265 nvAMD patients. Patients were anonymously genotyped at the complement factor H and age-related maculopathy susceptibility 2 loci and segregated into genotype groups (GTGs) defined by specific combinations of risk alleles. Physicians, who were blind to patients’ GTGs, obtained patients’ AREDS-F use history. The facility performing genetic analysis was blind to the AREDS-F use history. We used logistic analysis to estimate the interaction coefficient between AREDS-F use and GTG 2 vs GTG 3 in a general-population model.

Results:

The odds ratio of numbers of patients reporting prior AREDS-F use to nonuse for GTG 2 vs GTG 3 was 4.18 (P = .001). Logistic regression, correcting for nongenetic risk factors, gave an estimate of the β for interaction of AREDS-F with genotype of 1.57 (P = .001). This estimates a corrected odds ratio associated with the effect of interaction of 4.81 between those in GTG 2 compared with those in GTG 3.

Conclusions:

Our data indicate an interaction between GTGs and AREDS-F use that is consistent in size and direction with previously published reports, which had found that using AREDS-F supplements significantly increases the risk of nvAMD for some users and significantly protects other users.

Introduction

Age-related macular degeneration (AMD) is the leading cause of visual disability in the industrialized world and the third leading cause globally. ¹ AMD preferentially damages the macula, the central region of the retina. ² The Age-Related Eye Disease Study (AREDS) classified AMD as early, intermediate, or advanced based on macular phenotype and visual acuity. ³ Advanced AMD takes 2 principal forms: neovascular AMD (nvAMD), which refers to pathologic angiogenesis and its sequelae, and central geographic atrophy (GA), which refers to localized atrophy of central macular tissue and associated structures. ² The interaction of age, environment, and genetics influences the risk of developing AMD. Genetic factors and retinal phenotype are predictors of advanced disease. ⁴ Polymorphisms in the complement factor H (CFH) and age-related maculopathy susceptibility 2 (ARMS2) genes have the greatest impact on the progression to nvAMD or GA. ^{5 -7}

The AREDS has found that an AREDS formulation (AREDS-F)—composed of a specific combination of high-dose antioxidants (β-carotene, vitamin C, and vitamin E) and high-dose zinc—reduces the risk of progression from intermediate to advanced AMD by 25%. ³ Although advanced AMD was defined in the AREDS as either nvAMD or central GA, the demonstrated reduction in progression to advanced AMD was because of decreased progression to nvAMD and not to central GA, even after long-term evaluation. ^3,8

Multiple publications have evaluated the influence of genetic profiles on the response to the AREDS-F. ^{9 -12} All these analyses were performed on data from individuals described in AREDS report No. 8, ³ the only placebo-controlled, long-term study of the effect of nutritional therapy on AMD progression. Awh and colleagues ⁹ have concluded that genotype groups (GTGs) defined by combinations of variants in the CFH and ARMS2 genes can identify individuals who benefit substantially from using the AREDS-F and others who experience increased risk of progression from AREDS-F exposure. Vavvas et al ¹³ analyzed a larger subset of the AREDS data and have found a difference in response to the AREDS-F as a function of GTG when using nvAMD as an end point.

Not all investigators have accepted the clinical utility of genetic testing of patients using the AREDS-F. ^{14 -17} Chew and colleagues ¹⁴ and Assel et al ¹⁵ concluded that all patients with intermediate AMD benefit from the AREDS-F, irrespective of GTG. Their analyses were likely weakened by several factors, including their use both of central GA and nvAMD as clinical end points, as shown in a published reanalysis of the Assel et al data. ¹⁸

The expense of longitudinal studies has limited the understanding of the relationship between GTGs and AREDS-F use on AMD outcomes. Only 1 prospectively collected case-control data set (the AREDS ³ ) is available. All participants in a follow-up study, the AREDS2, received an antioxidant and zinc supplement. There was no placebo control group. This makes the AREDS2 data unsuitable to validate a pharmacogenetics interaction.

One alternate approach to assess whether there is an interaction between 2 independent variables is a case-only design, the validity of which has been established. ^19,20 Unlike a case-control study, in which exposures are compared between individuals with a condition and those without, a case-only study evaluates only individuals with the condition. ²⁰ A statistical interaction between 2 variables, in this case GTG and AREDS-F use, can be estimated if there is no significant interdependence between the variables.

We used a case-only design to compare the ratio of AREDS-F users vs nonusers between those within GTG 2 vs GTG 3. We measured the interaction between GTG and AREDS-F use in progression to nvAMD using a logistic regression statistical model.

Methods

Participants

Patients who were at least age 65 years at the time of nvAMD onset were recruited from community-based retina practices in Ohio, Pennsylvania, and California. We excluded any patient with prior AMD genetic testing. We recorded age, sex, smoking status, and body mass index (BMI) as calculated by height and weight.

All participants had a recent change in their vision or clinical findings that indicated recent-onset nvAMD. We excluded patients with fibrosis at the time of nvAMD diagnosis. Fellowship-trained retina specialists diagnosed nvAMD based on the funduscopic examination and ocular coherence tomography findings. We excluded patients with a history of vitrectomy, macular laser treatment, or choroidal neovascular membranes for causes other than AMD.

An independent data coordinating center (JRL Research and Consulting) collated deidentified genotype results and AREDS-F use histories. None of the investigators were aware of either the AREDS-F use history or GTG.

AREDS-F Use

We defined AREDS-F use as taking at least 7 doses per week of any commercial AREDS-F product containing the AREDS-prescribed dose of antioxidants and 40 mg zinc for at least 5 years up to the time of the onset of nvAMD in either eye. ^3,21 A decision to include individuals taking half the AREDS-recommended 80-mg daily dose of zinc was made to reflect a common pattern of community use of the AREDS-F. The AREDS2 demonstrated that daily use of 25-mg zinc was biologically equivalent to 80 mg, suggesting that 40 mg of zinc per day was also equivalent. ²¹

We defined non–AREDS-F use as the consumption of the AREDS-F at any dose for an aggregate of less than 30 days prior to the onset of nvAMD. To avoid diluting any potential associations, we were careful to exclude patients who declared occasional AREDS-F use or who provided an unreliable use history. For example, excluded patients might have been unsure when they commenced AREDS-F use or which vitamin they used. We expected recall of AREDS-F use to be unrelated to GTG. Use of a standard daily multivitamin was not an exclusion criterion, but we excluded individuals using a zinc supplement for nonocular indications. Study participants did not have AMD genetic testing prior to the onset of wet AMD, because genetic testing could influence the decision whether to take the AREDS-F. We collected data on known AMD confounders—age, sex, smoking status, and BMI.

Genotyping and GTG Assignment

A College of American Pathology and Clinical Laboratory Improvement Amendments–certified laboratory (Arctic Medical Laboratories) performed CFH and ARMS2 genotyping for AMD on deidentified, coded samples. The service provider was blinded to the AREDS use data. A mass array matrix-assisted laser desorption/ionization-time of flight mass spectrometer (Agena Biosciences) was used to determine genotype at CFH single nucleotide polymorphisms rs3766405, rs412852, and the ARMS2 indel (382_815 del443ins54).

To be consistent with the literature, ¹³ we defined the CFH high-risk haplotype as homozygous cytosine at rs3766405 and rs412852. We defined the risk ARMS2 genotype as the 3′ insertion/deletion polymorphism (indel), associated previously with ARMS2 RNA instability. ²²

Awh et al described the scientific rationale for the GTGs. ⁹ Given the relative scarcity of homozygous CFH low-risk alleles and ARMS2 homozygous high-risk alleles, individuals homozygous for these alleles were grouped with individuals heterozygous for the corresponding risk alleles. Participants with 0 or 1 CFH and no ARMS2 risk alleles (C01A0) comprised GTG 1. GTG 2 individuals had 2 CFH and no ARMS2 risk alleles (C2A0). GTG 3 individuals had 0 or 1 CFH and 1 or 2 ARMS2 risk alleles (C01A12). GTG 4 participants had 2 CFH and 1 or 2 ARMS2 risk alleles (C2A12) (Table 1.)

Table 1.

Features of the Genotype Groups.

Genotype group	CFH risk rs3766405/rs412852	ARMS2 372_815del443ins54
GTG 1	Low/intermediate risk: all except CC/CC (C01)	Low risk: biallelic ancestral (A0)
GTG 2	High risk: CC/CC (C2)	Low risk: biallelic ancestral (A0)
GTG 3	Low/intermediate risk: all except CC/CC (C01)	High risk: 1 or 2 indels present (A12)
GTG 4	High risk: CC/CC (C2)	High risk: 1 or 2 indels present (A12)

Abbreviations: ARMS2, age-related maculopathy susceptibility 2; CC/CC, homozygous cytosine at both CFH loci; CFH, complement factor H; GTG, genotype group.

Sample Size Calculations

Previous studies of AREDS data ^9,12,13 found that GTG 3 (low CFH risk, high ARMS2 risk) individuals derive strongest risk reduction of nvAMD from AREDS-F use, whereas GTG 2 individuals (high CFH risk, low ARMS2 risk) experience an increased risk of nvAMD if treated with the AREDS-F. These studies found that the AREDS-F did not affect the risk of progression to nvAMD for patients in GTG 1 or GTG 4.

We performed sample size calculations based on previously reported risk ratios, ^12,13 seeking 80% power and α = .05 to detect a difference between GTG 2 and GTG 3 participants in the interaction between GTGs and AREDS-F use in progression to nvAMD. We used established methods to calculate sample size for logistic regression to adjust the estimated odds ratios for known confounders (age, sex, smoking status, and BMI). ²³ Given the reported frequency of GTG 2 and GTG 3 among nvAMD cases, ¹³ we estimated the target sample size across all GTGs to be approximately 220 cases with the exact number depending on the observed ratio of AREDS-F users to AREDS-F nonusers.

Statistical Analyses

If the effect of AREDS-F were the same for all GTGs, the odds of AREDS-F use in cases should not differ significantly between GTG 2 and GTG 3. A significant difference in the odds of AREDS-F use among patients in different GTGs would be evidence of an interaction between GTGs and AREDS-F use in progression to nvAMD.

We conducted all analyses using R statistical software (https://cran.r-project.org). All statistical code used in this study is available on request.

We determined differences in patient demographic parameters using the t test for continuous variables (BMI and age) and the χ² test for categorical variables (sex and smoking status). We computed P values and (Wald-type) 95% CIs, as appropriate, to assist in establishing the significance of results.

Results

Study Data

Approximately 700 patients were approached for study entry from May 2018 to January 2019. About one-third of these patients could not provide a reliable history of AREDS-F use. The most common reasons for exclusion based on AREDS-F history were uncertainty about the duration of AREDS-F use or uncertainty about the brand of AREDS-F supplement used. Among patients meeting study inclusion criteria, we enrolled 46 patients with a reliable history of AREDS-F use and 219 patients with a reliable history of AREDS-F nonuse. Approximately 95% of qualified patients invited to participate in the study chose to enroll. We unintentionally exceeded our recruitment target. Table 2 summarizes demographic and clinical aggregate information of patients taking the AREDS-F and those not taking the AREDS-F.

Table 2.

Characterization of the Study Cohort.

Parameter	AREDS-F not taken (n = 219)	AREDS-F taken (n = 46)	P
Age, y	78.02	80.06	.11
Female sex, %	58.4	65.2	.49
Smokers %, no/quit/yes	40.3/47.3/12.4	54.3/39.1/6.5	.17
White ethnicity, %	100	100
BMI mean, kg/m2	28.18	28.82	.14

Abbreviations: AREDS-F, Age-Related Eye Disease Study formulation; BMI, body mass index.

After AREDS-F use determination, we ascertained genotypes, which had the distribution shown in Table 3.

Table 3.

Distribution of Genotype Among Age-Related Eye Disease Study Formulation Users and Nonusers.^a

GTG	AREDS-F users	Non-AREDS-F users	Odds ratio users:nonusers
1 (n = 37, 13.9%)	5	32	0.16
2 (n = 47, 17.7%)	15	32	0.47
3 (n = 119, 45.1%)	12	107	0.11
4 (n = 62, 23.3%)	14	48	0.29

Abbreviations: AREDS-F, Age-Related Eye Disease Study formulation; GTG, genotype group.

^aThe odds ratio of GTG 2 vs GTG 3 for AREDS-F use to nonuse was 4.18 (P = .001).

Logistic Regression

To estimate the interaction of GTG and AREDS-F use in progression to nvAMD, we performed logistic regression. ²⁰ In our primary analysis, we compared GTG 2 (n = 47) cases with those with GTG 3 (n = 119). The principal measure of the predictive strength of a dependent variable in a regression study is the β-coefficient, which is here defined as the change in the logarithm of the odds of the target outcome (AREDS-F use) in GTG 2 compared with the baseline group (ie, log ratio of odds in GTG 2 to comparator GTG 3).

The β-coefficient for GTG 2 compared with GTG 3 was 1.57, specifying an odds ratio of nvAMD associated with AREDS-F use of 4.81 (P = .001) (Table 4). This indicates an interaction between GTGs 2 and 3 and AREDS-F use in progression to nvAMD.

Table 4.

Logistic Regression Coefficients and Odd Ratios for Genotype Group 2 Compared With Genotype Group 3 With Age-Related Eye Disease Study Formulation Use With Respect to Outcome Variable, Adjusted for Sex, Age, Smoking, and Body Mass Index.

	β-coefficient (95% CI)	Exp β (95% CI)	P
GTG 2 vs GTG 3	1.57 (0.68-2.47)	4.81 (1.95-11.74)	.001

Abbreviations: Exp, exponentiation; GTG, genotype group.

When analyzed without adjusting for confounders, the odds ratio of AREDS-F use in GTG 2 compared with GTG 3 individuals was 4.18 (P = .001).

Conclusions

AMD is associated with genes whose products interact with components of the AREDS-F. ^{24 -27} A clinically significant interaction of genetic variants with the AREDS-F on the development of advanced AMD would permit a personalized approach to potentially life-long supraphysiologic doses of antioxidant vitamins and zinc. Prior to our study, all analyses of the interaction of CFH and/or ARMS2 genetic risk and the AREDS-F were conducted using data from AREDS patients. Klein et al performed the earliest analysis and reported a lack of efficacy of the AREDS-F in preventing all forms of advanced AMD in individuals having 2 high-risk CFH risk alleles. ¹¹ Awh and colleagues found a similar but more powerful correlation using a classification system based on combinations of CFH and ARMS2 risk allele numbers. ^9,10

Other reports disputed a clinically significant interaction between GTGs and AREDS-F use on the progression to advanced AMD. ^{14 -16} They noted that GTGs were defined post hoc, leading to overfitting of data to a model and elimination of statistical significance after Bonferroni multiple testing correction. ¹⁵ Assel et al, ¹⁵ using an independent AREDS-derived dataset, did not find a statistically significant interaction between AREDS-F and CFH risk alleles on progression to advanced AMD. This analysis was insensitive to the distinction between nvAMD and central GA as 2 forms of advanced disease. By including progression due to central GA in their smaller independent data set, Assel et al ¹⁵ diluted the data and made it less likely to identify increased risk of nvAMD in patients with CFH risk alleles who used AREDS-F. AREDS report No. 8 showed that AREDS-F use reduced progression to advanced AMD, ³ but this benefit was due to reduced progression to nvAMD, not reduced progression due to central GA. Seddon et al ^{4 -6} first reported an interaction between genetic groups and AREDS-F use in progression to nvAMD. They found no significant interactions between GTGs and AREDS-F use for progression to central GA.

To test the hypothesis that the GTGs defined by Awh et al ^9,10 interacted with AREDS-F use in progression to nvAMD, Vavvas et al identified a novel validation data set of 299 GTG 2 or GTG 3 individuals treated with AREDS-F or placebo. ¹³ Their data set likely had significant overlap with the independent data set reported 2 months later by Assel et al. ¹⁵ Vavvas and colleagues found that the hazard ratio of progression to nvAMD for GTG 2 patients who used AREDS-F compared with placebo was 4.9 (P = .021), and the hazard ratio of GTG 3 individuals who used AREDS-F compared with placebo was 0.36 (P = .003). AREDS-F conferred substantial risk of nvAMD for GTG 2 participants and substantial protection from nvAMD for GTG 3 participants. This study used previously defined GTGs and a previously determined end point—nvAMD. Therefore, there was no question about whether the investigators overfitted the data, and a multiple testing correction was not applicable.

Nonetheless, these results prompted a few ophthalmologists to recommend genetic testing prior to prescribing AREDS-F supplements. We decided to study the issue in patients with nvAMD and a confirmed AREDS-F use history who were drawn from community retina practices. If there were no relationships between genotypes and AREDS-F use on progression to nvAMD, the frequency of past AREDS-F use should not vary among GTGs. If the conclusions of Vavvas et al ¹³ were correct, we should expect patients with nvAMD with GTG 2 (high CFH and low ARMS2 risk) to have a greater proportion of AREDS-F use compared with those with GTG 3 (low CFH and high ARMS2 risk).

We identified an interaction between GTGs 2 and 3 and the use of the AREDS-F in patients with nvAMD. The coefficient of association between the AREDS-F and GTG 2 (with GTG 3 as a baseline group) was 1.58 (P = .001), which is similar to the one observed in the AREDS data subset analyzed by Vavvas and colleagues (1.37, P = .04). ¹³ Our observed interaction coefficient implies 4.81-fold greater odds for AREDS-F use in GTG 2 individuals compared with GTG 3. Using case-only data, we cannot demonstrate that AREDS-F users in GTG 2 have increased risk of nvAMD compared with placebo, or that AREDS-F users in GTG 3 have decreased risk of nvAMD compared with placebo. However, our data are consistent with these possibilities because we found interactions between GTGs 2 and 3 with AREDS-F use in progression to nvAMD that accord with the findings of Vavvas et al. ¹³

Possible Confounding Factors

Owing to study inclusion criteria, AREDS-F users must have had AMD without neovascularization (intermediate AMD) for a minimum of 5 years, during which they took supplements. It is possible that patients in different GTGs progressed from intermediate AMD to nvAMD at different rates, making observed GTG ratios a function of elapsed time rather than different responses to AREDS-F treatment. To address this possibility, we examined the AREDS data, which included well-documented, accurate time-to-progression information. ³ In placebo-treated AREDS patients, the length of time to progress from enrollment in the study as category 3 or category 4 (the categories for which the AREDS found AREDS-F use beneficial) to new nvAMD was unrelated to GTG (Table 5, raw data on file). These data indicate that it was very unlikely that this possibility confounded our data.

Table 5.

Progression Time to Neovascular Age-Related Macular Degeneration in Placebo-Treated Age-Related Eye Disease Study Participants With Intermediate Age-Related Macular Degeneration as a Function of Genotype Group.^a

GTG	Mean time to nvAMD for category 3 and 4 individuals, y
1	5.89
2	5.22
3	5.13
4	5.13

Abbreviations: GTG, genotype group; nvAMD, neovascular age-related macular degeneration.

^aThere was no statistically significant difference in the time to progression to nvAMD when comparing GTG 2 with GTG 3 (P = .22).

We used a case-only study design, which can detect an interaction under the assumption that the variables are causally independent, that is, AREDS-F use was not influenced directly or indirectly by participants’ GTG. Genetics might influence AREDS-F use if genetic variation frequency and access to health care were found to vary by ethnic grouping. Our patients were all white, making this possibility very remote.

The independence of AREDS-F use and GTG would also be violated if GTGs influenced clinical disease characteristics, leading to different rates of AREDS-F use. To address this, we studied whether there were differences in phenotypes between GTG 2 and GTG 3 at the time of participant enrollment in the AREDS. If GTG 2 or GTG 3 had different clinical features, we would expect a difference in their ratio in the 4 different AREDS phenotypic categories at study enrollment. ³ There were no significant differences between GTG 2 and GTG 3 in the four AREDS categories (Table 6).

Table 6.

Distribution of Age-Related Eye Disease Study Categories Among Age-Related Eye Disease Study Participants at Enrollment as a Function of Genotype Groups.^a

GTG	Category 1	Category 2	Category 3	Category 4
1	291	223	373	74
2	62	51	145	67
3	171	172	403	186
4	28	37	240	138
	Ratio GTG 2:GTG 3 Category 1	Category 2	Category 3	Category 4
	.363	.297	.360	.360

Abbreviations: GTG, genotype group.

^aχ² for GTG 2 vs GTG 3 comparing category 2 with 3 was P = .30, and χ² for GTG 2 vs GTG 3 comparing category 1 + 2 (for whom Age-Related Eye Disease Study report No. 8 did not show a benefit from AREDS-F) vs category 3 + 4 (who did benefit from AREDS-F) was P = .51.

AREDS2 report No. 14 appears to dispute this finding. ²⁸ Individuals with a CHF risk allele were more likely to have more extensive drusen, and those with an ARMS2 risk allele were more likely to have evidence of nvAMD. AREDS2 report No. 14 identified phenotypic features that developed at any time during the AREDS2, and all participants in this study received a supplement with zinc. If AREDS-F use predisposes to nvAMD in patients with CFH risk alleles, it would not be surprising if AREDS-F use also predisposed to pre-nvAMD features, such as drusen. If this were the case, it could account for the predilection of individuals with the CFH risk allele to have more extensive drusen.

Although it is not possible to prove complete causal independence between AREDS-F use and genotype, we have investigated and rejected what we regard as the most plausible confounders. If genotypes did influence the decision to take AREDS-F to a degree, the interaction was very unlikely to have been of sufficient size to account for the magnitude of effect we observed.

It is estimated that 7 million Americans have intermediate AMD, putting them at risk of nvAMD. ²⁹ Approximately 15% of patients with AMD have GTG 2. ⁹ Many ophthalmologists and optometrists recommend the AREDS-F for all patients with intermediate AMD. Our study data indicate an interaction between GTGs and use of AREDS-F, which is consistent with previously published conclusions that patients with high CFH risk and low ARMS2 risk have an increased risk of nvAMD because of AREDS-F use. Although further research regarding this pharmacogenomic interaction would be valuable, we think that the existing evidence supports genetic testing of patients at risk for nvAMD before commencing AREDS-F treatment.