CFH and ARMS2 Polymorphisms Interact with Zinc Supplements in Cognitive Impairment in the Women’s Health Initiative Hormone Trial

Source of study subjects

Study data from the WHI HT [13] was provided through an investigator agreement with the NCBI database of Genotypes And Phenotypes (dbGAP) [15]. The WHI HT was sponsored by the National Heart, Lung, and Blood Institute (ClinicalTrials.gov Identifier: NCT00685009) and was a randomized, double blind, community-based, placebo-controlled clinical trial of postmenopausal women recruited from June 1995 to July 2002 (estrogen plus progestin versus placebo) or to February 2004 (estrogen-alone versus placebo) to investigate the effects of postmenopausal hormone replacement therapy (HRT) on human health. Subjects were between 50 and 79 years of age and did not have medical conditions that predicted a survival of less than three years or had characteristics predicted to diminish study adherence [13]. In total, 27,347 subjects were recruited by WHI HT investigators. We restricted our analysis to women for whom genotyping data and serial mental status examinations were available, leaving 7,700 women for the analysis of the interaction of zinc supplements and genetics on cognitive decline. This sample is believed to be representative of the study population as a whole.

Assessment of cognitive status

Cognitive function was determined by trained surveyors using the 3MS [16]. The 3MS score assesses attention, concentration, orientation, memory, language, constructional praxis, abstract thinking, and list generation, generating a score from 0 to 100. While cognitive function is a continuous variable, its measurement can fluctuate transiently. Scores <79 have been validated previously as consistent with cognitive impairment and were used to dichotomize mental status for the purpose of this analysis similar to a previous study of AMD patients [2, 17].

Supplement use history

During the WHI HT period of follow up the reported daily-ingested quantities of over 25 individual dietary supplements, including elemental zinc, were recorded. This information was provided as part of an investigator agreement by the National Institutes of Health, dbGAP [15].

Ethics and oversight

The WHI HT received ongoing oversight by a Data and Safety Monitoring Committee and the US National Institutes of Health and the Office for Human Research Protection [13]. Each study center (forty) provided independent IRB approval and elicited individual participant consent [13]. Ethics review of our analysis of genotype interaction with zinc supplements in cognitive impairment was provided by the University of Toronto, Ontario, Canada.

Genotype assignment and interpretation

Genotyping for study subjects was performed by WHI HT investigators using a combination of array-based, PCR and next-generation sequencing techniques (dbGAP Study accession: phs000200.v10.p3). Genotype information was available for 7,700 subjects at the ARMS2 AMD risk polymorphism (rs10490924), for the CFH AMD risk-associated SNPs (rs3766405 and rs412852) [7], and for 2 SNPs (rs7412 and rs429358) which determine APOE alleles [18].

CFH AMD risk-haplotype definition was described previously [7]. Individuals with homozygous CC at both CFH SNPs were considered high risk, those with heterozygous CT genotypes at each locus or those homozygous CC at rs3766405 and heterozygous CT at rs412852 were considered AMD intermediate risk, and all other genotype combinations at these 2 CHF loci were considered AMD low risk.

For the ARMS2 locus, patients with homozygous TT genotypes at rs10490924 were considered to have a high-risk AMD genotype (2 risk alleles), GT combinations were considered intermediate AMD risk (1 risk allele) and individuals with homozygous GG were low risk AMD (0 risk allele). For the purpose of analysis, ARMS2 intermediate and high risk were combined because of the relative rarity of homozygous risk alleles and considered collectively as the “high risk” group [7]. APOE alleles were determined though genotyping at rs7412 and rs429358 with the APOE allele defined as a haplotype with cytosine at each of these loci [19].

In our main model of CI risk, we considered allelic dosage modeling of the number of CFH risk haplotypes and binary ARMS2 status (combining 1 and 2 risk ARMS2 allele groups because of rarity of the risk allele in the population) resulting in 6 genetic groups which we refer to as C_XA_Y, where C refers to CFH where X is 0, 1, or 2 risk alleles and A represents ARMS2 where Y is 0 or 1/2 risk alleles.

Statistical analysis

7,700 women had both 3MS scores and genetic data available, 7,486 of which self-reported ‘White’ or ‘Hispanic/Latino’ as their ethnicity. This group was selected for analyses because AMD genetics have been studied primarily in these ethnic groups. Twenty-two women (0.29%) had missing Education status which was imputed using the majority group (“post-High School”). The status of APOE mutation could not be ascertained for 2 women (0.03%) and these were excluded from analysis. Out of 25,805 CI assessments available for this set of 7,486 women, 15 (0.06%) had missing 3MS score (including one subject with a singular observation) and these data points were dropped from analysis. After these exclusions, 7,483 subjects with 25,786 data points were available for analysis.

To examine the effect of zinc supplementation on CI we used a repeated-measures logistic regression model with CI as an outcome. In the main-effects model, we considered the risk of CI as functions of age (with cubic splines), HRT (dichotomous), education (5 levels), zinc supplementation (dichotomized to ≤7.5 mg /day (“un-supplemented”) or >7.5 mg/day (“supplemented”), CFH risk alleles (using allelic coding), and ARMS2 risk alleles (dominant coding) (Table 2). The gene×treatment model added the effect of interaction between Zinc and ARMS2 and CFH polymorphisms.

Zinc supplementation status

Zinc intake (mg/day) was determined from micronutrient supplements survey data, and was collected without knowledge of individual CFH and ARMS2 genotypes (WHI form 45). No information was available on the temporal relationship between zinc and food intake. Dietary zinc intake was not considered in our analysis because of the confounding effects of recall bias. Micronutrient surveys were completed at approximately 3-year intervals during the study, though not always at the time of the 3MS score determination. To associate zinc intake with 3MS scores we used a heuristic algorithm, which prioritized the temporal proximity of zinc and 3MS score information, considering the determinations to be coupled if they were sampled within one year of each other. For more temporally separated zinc and 3MS scores, the algorithm computed a weighted average of 2 zinc intake determinations flanking the 3MS score assessment, giving twice the weight to the zinc intake reported prior to 3MS score. This averaging heuristic affected 119 (0.46%) of the observations.

Half of 25,786 recorded zinc intake values (longitudinal observations) were zero (12,951 or 50.2%). RDA-level supplementation (7.5 mg–15 mg/d) was reported in 39.1% of surveys. Supplementation levels >15 mg were rare and reported in only 8.2% of surveys. Zinc intake associated with each 3MS score was used to consider subjects as “supplemented” or “un-supplemented” (at a given assessment) for analysis, using a 7.5 mg/day threshold. Those taking at least 7.5 mg/day were considered “supplemented” while those reporting less than 7.5 mg of daily zinc consumption were considered “un-supplemented”. We performed initial analysis of potential confounding of zinc intake by other covariates using ANOVA models of per-subject mean zinc intake and an F-test.

Age and education

Age and education correlate strongly with the development of cognitive impairment as measured using 3MS scores [20]. Education levels were reported in 11 categories of increasing schooling years, which we re-categorized into 5 levels. These were included in subsequent multivariate models in an unrestricted fashion allowing for education level adjustment of 3MS scores. For each 3MS score we determined the age of the subject at time of cognitive assessment, using “Age at Screening” and “Days since Screening” information provided and rounded to the nearest year. The age effect was modeled using penalized natural cubic splines with 3 degrees of freedom using Generalized Additive Modeling approach [21]. Briefly, the age range is divided into a small number of panels, and within each panel age is allowed to take a cubic shape, with a restriction that overall shape of age effect across the whole range is continuous and smooth. The degree of smoothness is further controlled by penalizing the parameters underlying the individual cubic functions, forcing them to be smaller than they would be if no penalization was invoked. The continuity and smoothness restrictions, coupled with the amount of penalization determines the equivalent degrees of freedom utilized to model age, which we set here to be three. This allows the effect of age on risk of CI to be modeled as a smooth, non-linear curve, which is a more flexible and realistic assumption compared to usual linear model parameterization and increases the chance that any observed effect of zinc and genetics is properly controlled for potential age confounding.

To account for reported association between HRT and increased risk of CI, we utilized a binary HRT status variable (pooling estrogen-only and progesterone + estrogen arms together) in our model [20].

In our repeated-measures logistic regression model, CI (3MS score <79) was used as the outcome, age (using cubic spline expansion), 5-level education status, number of APOE risk alleles, and HRT status were considered as potential confounders. The number of CFH risk alleles, the number of ARMS2 risk alleles, and zinc intake (“un-supplemented” or “supplemented”) were the primary variables of interest. A time-to-event analysis, such as with Cox Proportional Hazards model, was considered but rejected due to long assessments intervals, the infrequency of assessments, and occasional reversion of mental status scores. The time-to-event analysis and repeated measures logistic regression allows us to perform classical and well-accepted statistical inference on assessing genetic effect modification of zinc supplementation on CI and make the results directly comparable to similar studies in AMD which also used repeated measures logistic regression [6] and Cox regression [11]. Our modeling was done using the “gam” function in R within an “mgcv” package [22] which allows for both repeated measures and non-linear age effects, while providing a conventional Generalized Linear Models hypothesis testing framework. Less conventional models such as Latent Growth Curve models focusing more on analysis of changes in cognitive function may offer an alternative approach [23, 24].

Regression analysis

On average, across zinc supplementation groups, an increased number of CFH risk alleles was associated with a decreased risk of CI (p = 0.03). Compared to subjects with no CFH risk alleles, the OR for developing CI was 0.73 (p = 0.05) for those with one risk allele, and 0.57 (p = 0.02) for those with two risk alleles, showing an allelic dosage effect. On average, across genetic groups, zinc supplementation was associated with reduced risk for CI (OR 0.70, p = 0.023). The number of ARMS2 risk alleles had no significant impact on the risk of CI (Table 2).

We analyzed the genetic modification of zinc supplementation on the risk of developing CI by considering a model with interaction effects between zinc supplementation and genotype groups defined by CFH and ARMS2 risk allele number. We previously reported that genotype groups comprised of CFH and ARMS2 variants correlate with the treatment effect of zinc on progression from intermediate AMD to the neovascular form of advanced AMD. Individuals in some genotype groups experience an increased risk for progression to choroidal neovascular disease if using the AREDS dietary supplement compared to placebo [11].

We identified a statistical interaction between CFH and ARMS2 risk allele genotype groups and zinc supplementation (p = 0.028) in a model of CI. We then examined the association of zinc on the development of CI in 6 predefined genotype groups. We found that the overall beneficial zinc effect appears to be confined to those individuals who have at least one risk CFH risk haplotype and no ARMS2 risk alleles. In this genotype group, a significant reduction in CI risk was observed (OR = 0.456 for C₁A₀, OR = 0.206 for C₂A₀, p = 0.002). The C_XA_Y nomenclature is explained above in “Genotype Assignment and Interpretation (Table 3). For other genetic subgroups, zinc supplementation was not associated with either reduced or increased risk of CI.

The effects of known risk factors for CI and confounders were significant with the direction of effect as expected, with the exception of HRT, which was not significant (p = 0.087) (Table 4 and Fig. 1). The effects of education and age were both significant (p < 0.001 with 4 and 3 degrees of freedom, respectively) and showed a beneficial effect of higher education, and a detrimental effect of increasing age. The number of APOE4 alleles was also associated with increased risk of CI (OR = 2.80, p < 0.001).

OPEN IN VIEWER

Fig.1

Effect of Age on risk of CI using a natural cubic spline model with 3 degrees of freedom, adjusted for HRT, education, zinc use, and genetic effects.