Does the Association Between Eye Disease and Cognitive Function Vary by Genetic Risk of Cognitive Decline? An Analysis of Hospital Data With Replication in the Canadian Longitudinal Study on Aging

Abstract

Purpose

Age-related eye diseases are inconsistently associated with cognitive decline which could be due to effect modification. The purpose of this study was to investigate whether two genetic factors previously found to be associated with cognitive decline, the KIBRA (WWC1) and PDE7A/MTFR1 genes, modify the association between eye disease and cognitive function.

Methods

Data from a Montreal hospital-based cross-sectional study (n = 302) were used for the primary analysis. Candidate single-nucleotide polymorphisms (SNPs) rs17070145 (KIBRA gene) and rs10808746 (PDE7A/MTFR1 gene) were included in linear regression models to test for effect modification of the relationship between eye disease (glaucoma or age-related macular degeneration [AMD]) and cognitive function. Six oral cognitive tests were used. A replication analysis was done using the Quebec data from the Canadian Longitudinal Study on Aging (CLSA) (n = 4238). Effect modifications by expanded genomic regions around the candidate SNPs were tested.

Results

Three statistically significant interactions with two cognitive function measures —category verbal fluency and immediate story recall—were found in the Montreal study: glaucoma and AMD with rs17070145 and verbal fluency (P < 0.03) and glaucoma with rs10808746 with immediate story recall (P < 0.05). Similar interactions, although not with the same cognitive measure, were found in the CLSA: AMD with KIBRA and glaucoma with PDE7A/MTFR1.

Conclusions

Our results suggest that the KIBRA and PDE7A/MTFR1 genes may modify the association between eye disease and cognitive function. This knowledge may help to better understand the mechanism by which glaucoma and AMD are related to cognitive function.

Age-related macular degeneration (AMD) and glaucoma are leading causes of visual impairment in the world, affecting 196 million and 76 million people, respectively.^1,2 It has been suggested that eye diseases such as AMD and glaucoma and cognitive diseases such as Alzheimer's disease could share common biological mechanisms, including neuroinflammation and oxidative damage.³ Previous studies examining associations between eye disease and cognitive outcomes have had inconsistent results. For example, some studies have found associations between AMD and lower global cognitive function,^4–6 as well as worse memory and verbal fluency,⁴ whereas other studies have not found an association between AMD and adverse cognitive outcomes.^7–9 Similarly, some studies examining glaucoma and cognitive function have found an association,^7,9–11 but there are also other studies that have not.^12,13

Effect modification could explain some of the inconsistences in prior research. Perhaps eye disease and cognitive function are only associated in people at higher genetic risk of cognitive disease, and the frequency of those genetic risks varies by population. Two genetic factors have previously been found to be associated with cognitive function: single-nucleotide polymorphisms (SNPs) rs17070145, in the kidney and brain expressed protein (KIBRA) gene (also referred to as the WWC1 gene) on chromosome 5, and rs10808746 in the PDE7A/MTFR1 gene on chromosome 8.^14,15 KIBRA is associated with memory and is expressed primarily in brain regions related to memory, such as the prefrontal cortex and hippocampus.¹⁶ Findings from a 2012 meta-analysis suggested that the SNP rs17070145 in the KIBRA gene has a statistically significant association with episodic and working memory.¹⁴ Several studies have reported that T-allele carriers had better memory performance compared to non–T-allele carriers.^17,18 KIBRA may also be involved in retinal neuronal maintenance.¹⁹ The SNP rs10808746 is in the proximity of two adjacent genes on chromosome 8, PDE7A and MTFR1.¹⁵ The PDE7A gene is involved in the regulation of a secondary messenger cyclic adenosine monophosphate (cAMP) in the central nervous system, which is associated with cognitive impairment²⁰ and may be involved in the regulation of intraocular pressure and retinal ganglion cell survival.²¹ The G allele at rs10808746 was found to be associated with increased cognitive decline.¹⁵

The objective of this study was to investigate whether KIBRA and PDE7A/MTFR1 gene variants modified the association between eye disease and cognitive function in a hospital-based study. In this initial analysis, we focused specifically on the two candidate SNPs, rs17070145 and rs10808746, whereas our subsequent replication analysis using data from the Canadian Longitudinal Study on Aging (CLSA) examined the effects of gene variants within genomic regions surrounding the original candidate SNPs. Our hypothesis was that the associations between eye disease and cognitive function would vary by genetic risk of cognitive decline.

Methods

Montreal Hospital-Based Study

Study Design and Population

A cross-sectional study was conducted at Maisonneuve-Rosemont Hospital in Montreal, Canada.⁹ The overall goal of the study was to understand whether eye disease and cognition were related and why. To address this, patients with visually impairing eye disease were recruited. Patients in the ophthalmology clinics were recruited between 2016 and 2018 into one of three groups: (1) those having a diagnosis of late-stage AMD (geographic atrophy or neovascular disease) in both eyes with a better eye visual acuity worse than 20/40; (2) those having a diagnosis of primary glaucoma in both eyes with visual field mean deviation worse than or equal to –4 dB in the better eye (secondary glaucoma was excluded due to concerns that the primary cause of glaucoma could also affect cognitive function); and (3) those having normal vision and not being seen for suspected AMD or glaucoma with visual acuity better than 20/40 in both eyes and visual field mean deviation better than –4 dB in both eyes. A diagnosis of AMD entailed the use of fundus photographs, macular optical coherence tomography (OCT), and, as needed, fluorescein angiography. Patients underwent a dilated retinal examination by an ophthalmologist. A diagnosis of glaucoma was based on progression of a visual field, ocular imaging (Heidelberg retina tomography or OCT), or optic nerve head examinations (showing increasing excavation and rim thinning). People in the normal vision group were being seen for conditions such as early cataract, ocular hypertension, and posterior vitreous detachment.

Exclusion criteria included being under 65 years old, unable to respond for themselves, or scoring less than 10 on the Mini-Mental State Examination (MMSE) Blind version, which was used to assess global cognition. In the MMSE Blind version, eight items relying on vision are omitted.²² Individuals who had received eye surgery in the last 2 months were enrolled, but their data collection was delayed by 2 months to prevent their recovery from affecting their cognition.

There were 312 patients that met the final eligibility criteria and completed data collection. Only participants who self-reported their race/ethnicity as White or French-Canadian (97% of participants) were included in the analysis to avoid population stratification bias.²³ The final sample included 302 participants. Approval was received in 2015 from the Ethics Committee at Maisonneuve-Rosemont Hospital. The study was conducted according to the tenets of the Declaration of Helsinki. Written informed consent was obtained from each participant.

Data Collection

Eye disease information, including date of diagnosis, type of AMD or glaucoma, and previous treatments were obtained from participants’ medical charts. Binocular presenting visual acuity was measured using the Early Treatment of Diabetic Retinopathy Study visual acuity chart at 2 meters.^24,25 Visual acuity scores were converted to log of the minimum angle of resolution (logMAR). Visual field was measured in each eye using the Humphrey Frequency Doubling Technology perimeter full-threshold testing (Carl Zeiss Meditec, Jena, Germany).²⁶ Visual field data were determined as unreliable and were not used if false-positive results, false-negative results, or fixation losses were greater than 33% of trials. If a reliable visual field measure could not be obtained, the most recent visual field data were used from the medical chart as measured using the Humphrey Swedish interactive threshold algorithm standard 24-2 program. Data on age and sex were also collected from medical charts, and information on the highest level of education attained was self-reported. The presence of a physician's diagnosis of diabetes, heart disease, or stroke was also self-reported. Participants were asked if they smoked at least 100 cigarettes in their lifetime and whether they currently smoked.

Six cognitive tests were orally administered by trained interviewers: the verbal fluency test (letter and category), the digit span test (forward and backward), and the logical memory test (immediate and 30-minute delayed story recall). Approximately 30 minutes were needed to complete all of the tests, which were recorded and scored by trained personnel. The verbal fluency tests measured language and retrieval skills, the digit span tests evaluated verbal working memory, and the logical memory tests assessed verbal memory, encoding, and maintenance.²⁷ The letter verbal fluency test asked participants to name as many words as possible that begin with the letter P in 1 minute.^28,29 The category verbal fluency test asked participants to name as many animals as possible within 1 minute. The digit span forward test instructed participants to repeat back an increasingly longer list of numbers in forward order, and the backward test asked participants to repeat them back in backward order. Finally, the logical memory test required that participants recall as many details as possible of an 18-item short story both immediately after the story and 30 minutes later.²⁷ Saliva samples were collected with Oragene DNA kits (DNA Genotek, Ottawa, ON, Canada) and were genotyped using the Sequenom iPlex Gold (Sequenom, San Diego, CA, USA) technology.

CLSA Comprehensive Cohort

Study Design and Population

The CLSA Comprehensive Cohort is a national population-based prospective cohort study comprised of 30,097 participants ages 45 to 85 years at enrollment.³⁰ Baseline data from the Comprehensive Cohort, collected between 2012 and 2015, were used for our analysis. The Cohort includes randomly selected participants from seven provinces living within 25 to 50 km of 11 data collection sites located in Victoria, Vancouver, Surrey, Calgary, Winnipeg, Hamilton, Ottawa, Montreal, Sherbrooke, Halifax, and St. John's, Canada.³⁰ Participants were interviewed in the homes, and data were also collected at data collection sites.³⁰ People had to be community dwelling, not cognitively impaired, able to speak and understand French or English, and able to provide informed consent. The CLSA excluded non-citizens or permanent residents, those living on federal First Nations reserves or settlements, full-time members of the Canadian Armed Forces, and those in institutionalized care.³⁰ All participants provided written informed consent. Approval by the Research Ethics Board was obtained in July 2010 for all CLSA affiliated sites.

Data Collection

Glaucoma and AMD were measured by asking participants if a physician has ever diagnosed them with glaucoma or AMD. The oral tests used to measure cognitive function included the Rey Auditory Verbal Learning Test (RAVLT) for memory retention (immediate and 5-minute delayed recall),³¹ Controlled Oral Word Association Test (COWAT) for letter verbal fluency,³² Animal Naming Test (ANT) for category verbal fluency,³³ and Mental Alternation Test (MAT) for processing speed.^34,35 The RAVLT, ANT, and MAT were administered during in-home interviews, and the COWAT was administered at the data collection site.³⁴ For the RAVLT, participants were asked to learn and then recite 15 unrelated words. The ANT required participants to name as many animals as possible in 60 seconds, and the COWAT required participants to name as many words as possible beginning with a specified letter (e.g., F, A, S) in 1 minute. Two trials were done for the ANT, and three trials were done for the COWAT. Scores were summed for analyses. The MAT required participants to alternate between reciting the numbers 1 to 26 and the alphabet A to Z (i.e., 1-A, 2-B, 3-C, …).

Demographic measures such as age, sex, and education were reported using questionnaires during in-home interviews.³⁰ For education, we categorized participants as having low, medium, or high education. We defined low education as having no post-secondary education or having a certificate/diploma from non-university institutions, medium if participants had bachelor's degrees, and high if they had university degrees/certificates above a bachelor's degree. Diabetes status was measured by asking participants if they were ever diagnosed by a doctor with diabetes, borderline diabetes, or high blood sugar. If they said “yes,” then they were asked if they had type I diabetes, type II diabetes, or neither. Participants were asked if a doctor ever told them they had (1) heart disease, (2) a stroke or cerebrovascular accident, or (3) high blood pressure (hypertension). Participants were asked about whether they had smoked at least 100 cigarettes in their life and whether they currently smoked.

Non-fasting blood samples collected at data collection sites were used to produce genotype data for 26,622 participants, and the Affymetrix Axiom array was used to genotype 794,409 SNPs.³⁶ The Version 3 Release data followed most of the quality control checks used by the UK Biobank.^36,37 For example, marker-based quality control checks included genotype frequency consistency between batches and chromosomal sex, Hardy–Weinberg equilibrium (HWE), and discordance across control replicates. Sample-based quality control included checking for insertion/deletion, minor allele frequency (MAF), genotype missingness, relatedness, and heterozygosity. We removed participants with at least one third-degree or closer relative among the genotyped individuals, as well as outliers in heterozygosity or genotype missingness. The analyses were limited to participants of European descent (93% of the participants) to reduce the effect of population structure, as recommended by the CLSA.³⁶ In addition, to maximize ancestry homogeneity between our original and replication samples, we only included the CLSA participants from the province of Quebec in the replication as our original study participants were all from Montreal, Quebec. The final sample size included 4238 participants (Supplementary Fig. S1). The data also included imputed genotype data for approximately 308 million SNPs using the TOPMed reference panel at the University of Michigan Imputation Server.³⁶

Statistical Analysis

Quality control for genotype data from the Montreal Study and the CLSA was performed using PLINK 1.9³⁸ and 2.0.³⁹ SNPs with a MAF < 0.01, HWE < 10^–5, or missing genotype rate > 0.1 were excluded. The genotyped SNPs from the CLSA initially reported in reference to the GRCh37/hg19 genome build were converted to the GRCh38/hg38 build.⁴⁰

ANOVA and χ² tests were used to compare the characteristics of the three eye disease status groups. In our initial analysis, effect modification by the candidate SNPs of the association between each eye disease and cognitive function was tested using multiple linear regression models with the SNPs coded additively (i.e., 0, 1, or 2 minor alleles). The interaction term between eye disease status and the candidate SNP was used to assess effect modification. A separate model was run for each cognitive outcome, eye disease, and SNP combination. All cognitive outcomes were standardized to z-scores in the regression models. Every model was adjusted for the covariates age, sex, education, and diabetes. In a sensitivity analysis, we also adjusted for smoking, heart disease/stroke, hypertension and the APOE-e4 allele carrier status because this allele has been associated with both cognition (dementia) and AMD.^41,42 Genotypes for the two SNPs (rs429358 and rs7412) whose haplotypes define the APOE alleles were available in both studies. APOE alleles ε2, ε3, and ε4 correspond to the rs429358-rs7412 haplotypes T-T, T-C, and C-C, respectively. Unambiguous haplotypes can be obtained from the genotypes except for the double heterozygotes, which could be ε2/ε4 or ε1/ε3. The double heterozygotes were assigned to ε2/ε4 because the ε1 allele is extremely rare.

In our replication analysis in the CLSA, we selected a genomic region of ±10 kb around the candidate SNPs because the candidate SNPs may not be the causal variants but may be in linkage disequilibrium (LD) with causal variants in the regions. We extracted genotyped and imputed SNPs in the two candidate regions. For the KIBRA genomic region, this included 12 genotyped and 90 imputed SNPs; for PDE7A/MTFR1, there were two genotyped and 44 imputed SNPs. Imputation quality scores for all imputed SNPs were r² > 0.8. Genotyped data were unavailable for both candidate SNPs. Like the initial analysis, multiple linear regression models including an interaction term were used to test for effect modification by the KIBRA and PDE7A/MTFR1 gene variants. For every cognitive outcome, the model was applied to each SNP in the expanded genomic region. All cognitive outcomes were standardized to z-scores in the regression models. In addition to age, sex, education, and diabetes, the models were also adjusted for five principal components capturing ancestry. Additional adjustment for smoking, heart disease, stroke, hypertension, and APOE-e4 carrier status was performed in the CLSA replication as a sensitivity analysis. To control for testing multiple SNPs within each region, we pooled P values within the region to obtain an empirical P value accounting for LD, calculated with Fisher's method³⁷ using the R package poolr 1.1-1 (R Foundation for Statistical Computing, Vienna, Austria).⁴³ Effect modification plots were produced in the initial and replication analyses, and Manhattan-type plots displaying the significance in genomic regions were produced during the replication analysis. All analyses were performed using R 4.3.0⁴⁴ in RStudio.⁴⁵

Results

Descriptive Characteristics of Both Samples

The Montreal hospital-based study participant characteristics are presented in Table 1. Of the 302 participants, 41.1% (n = 124) had normal vision, 29.5% (n = 89) had AMD, and 29.5% (n = 89) had glaucoma. The three groups differed in age, education, cognitive test scores, and rs10808746 genotype (P < 0.05). Among the 4238 Quebec participants from the CLSA, 91.5% (n = 3879) had no AMD or glaucoma, 3.1% (n = 131) reported AMD, and 5.4% (n = 228) reported glaucoma (Table 2). Similar to the Montreal Study, the three groups differed in age and cognitive scores (P < 0.05). They also differed in their report of diabetes, heart disease, systemic hypertension, and smoking (P < 0.05). Genotype frequencies for the two candidate SNPs in the Montreal study and for the 148 SNPs included in the CLSA replication are shown in Table 1 and Supplementary Table S1, respectively.

Table 1.

Characteristics of the Montreal Hospital Study Participants by Eye Disease Status

	Normal Vision (n = 124)	AMD (n = 89)	Glaucoma (n = 89)	P
Age (y), mean ± SD	72.2 ± 5.7	83.2 ± 7.3	77.1 ± 7.6	<0.0001
Male sex, %	52.4	37.1	47.2	0.085
Education (y), mean ± SD	12.9 ± 4.0	10.3 ± 4.3	11.0 ± 4.0	<0.0001
Diabetes, %				0.599
Yes	27.4	21.3	24.7
No	72.6	78.7	75.3
Smoking status, %				0.887
Never	43.9	37.9	44.7
Former	49.6	54.0	49.4
Current	6.5	8.0	5.9
Heart disease/stroke, %				0.554
Yes	26.8	33.7	29.2
No	73.2	66.3	70.8
Systemic hypertension, %				0.068
Yes	41.9	58.0	46.6
No	58.1	42.0	53.4
APOE-ε4 carrier, %				0.195
Yes	23.4	13.5	20.2
No	76.6	86.5	79.8
Cognitive outcomes, mean ± SD
Letter verbal fluency test (1 trial)	14.0 ± 5.3	11.1 ± 5.6	12.2 ± 5.0	0.0004
Category verbal fluency test (1 trial)	17.1 ± 5.3	13.2 ± 4.9	14.3 ± 5.5	<0.0001
Digit span test, forward	11.7 ± 2.1	10.2 ± 2.5	10.3 ± 2.5	<0.0001
Digit span test, backward	6.6 ± 2.4	5.1 ± 1.9	5.6 ± 2.7	<0.0001
Immediate story recall	10.8 ± 4.1	7.9 ± 4.4	8.7 ± 3.7	<0.0001
Delayed story recall	8.5 ± 4.6	5.6 ± 4.2	6.5 ± 3.5	<0.0001
SNP genotypes, %
rs17070145 (chr5, KIBRA)				0.511
CC	42.7	42.7	38.2
CT	49.2	47.2	46.1
TT	8.1	10.1	15.7
rs10808746 (chr8, PDE7A/MTFR1)				0.0002
AA	22.6	15.7	41.6
AG	58.1	49.4	40.4
GG	19.4	34.8	18.0

P values were obtained from ANOVA or χ² tests.

Table 2.

Characteristics of the CLSA Study Participants by Eye Disease Status

	None (n = 3879)	AMD (n = 131)	Glaucoma (n = 228)	P
Age (y), mean ± SD	61.8 ± 9.9	71.2 ± 9.5	69.0 ± 9.1	<0.0001
Male sex, %	48.6	47.3	49.1	0.946
Education, %				0.318
Low	62.4	61.5	68.4
Medium	21.8	19.2	18.4
High	15.9	19.2	13.2
Diabetes, %				0.003
None	86.0	81.5	77.0
Type I	0.5	0.8	1.8
Type II	9.0	13.1	14.9
Other	4.5	4.6	6.3
Smoking status, %				0.027
Never	40.6	37.7	35.7
Former	47.4	57.7	52.0
Current	12.0	4.6	12.3
Heart disease, %				0.002
Yes	12.8	23.1	15.8
No	87.2	76.9	84.2
Stroke, %				0.436
Yes	1.6	3.1	1.8
No	98.4	96.9	98.2
Systemic hypertension, %				<0.0001
Yes	36.4	52.7	52.2
No	63.6	47.3	47.8
APOE-ε4 carrier, %				0.146
Yes	24.2	20.2	19.3
No	75.8	79.8	80.7
Cognitive outcomes, mean ± SD
RAVLT immediate recall	5.6 ± 1.8	5.1 ± 1.7	5.0 ± 1.8	<0.0001
RAVLT delayed recall	4.0 ± 2.0	3.2 ± 2.0	3.6 ± 2.0	<0.0001
Letter verbal fluency (3 trials)	36.1 ± 11.9	33.5 ± 10.7	33.1 ± 12.1	<0.0001
Category verbal fluency (2 trials)	39.0 ± 11.3	35.4 ± 12.3	36.3 ± 11.5	<0.0001
Mental alternation test	26.0 ± 9.5	23.2 ± 9.7	23.8 ± 9.1	<0.0001

P values were obtained from ANOVA or χ² tests.

Montreal Hospital-Based Study

Our initial analysis found three statistically significant interactions after adjusting for age, sex, education, and diabetes (Supplementary Table S2): AMD and glaucoma with rs17070145 for category verbal fluency (P = 0.021 and P = 0.004, respectively), and glaucoma with rs10808746 for immediate story recall (P = 0.042). For category verbal fluency, participants with AMD or glaucoma with a T allele scored worse than those with normal vision, whereas there was no difference between the eye disease groups for the CC genotype (Figs. 1A, 1B). For immediate story recall (Fig. 1C), participants with glaucoma and the AA genotype performed worse than those with normal vision, whereas participants with the GG genotype performed similarly to those with normal vision. Similar estimates were found in a sensitivity analysis adjusting for additional covariates (smoking, heart disease/stroke, hypertension and the APOE-e4 allele carrier status) (Supplementary Table S3). The two interactions with rs17070145 stayed significant (P < 0.05), whereas the P value for the interaction with rs10808746 was 0.0987, which is consistent with a loss of power after adjustment for non-significant covariates (P > 0.1081) and loss of sample size due to missing values for these covariates without meaningful changes in interaction effect of interest.

Figure 1.

Modification of the effect of eye disease status on cognitive outcomes by candidate SNPs. (A, B) Effect modification by rs17070145 (KIBRA). (C) Effect modification by rs10808746 (PDE7A/MTFR1). The SNP genotypes are on the x-axis, and the scores of the specific cognitive outcome are on the y-axis. Cognitive outcomes were standardized as z-scores. The colors of the lines correspond to one of the three eye disease status categories.

Replication in the CLSA

The statistical significance values of the eye disease status effect modification by variants in the KIBRA and PDE7A/MTFR1 genomic regions (adjusting for age, sex, education, diabetes, and the five principal components capturing ancestry) are presented in Figures 2 and 3, respectively. Further adjustment for smoking, heart disease, stroke, hypertension, and APOE-e4 did not change the results (data not shown). The lead SNPs (the most significant SNP within each region) for each cognitive outcome are presented in Supplementary Table S4. Based on the P values pooled across each of the two candidate genomic regions tested (controlling for multiple testing and reflecting the region significance), the interaction between glaucoma status and the PDE7A/MTFR1 genomic region was statistically significant for the RAVLT delayed recall cognitive outcome (pooled P = 0.010) (Fig. 3B). The lead SNP for the interaction between glaucoma status and the PDE7A/MTFR1 genomic region for RAVLT delayed recall, rs17395878, was also nominally significant (P = 0.006). In addition, interactions between AMD and the KIBRA genomic region were close to significant for letter verbal fluency (pooled P = 0.074; lead SNP rs6897615 P = 0.002) (Fig. 2C) and RAVLT delayed recall (pooled P = 0.120; lead SNP rs11134511 P = 0.017) (Fig. 2B). The effects of the lead SNPs rs11134511, rs6897615, and rs17395878 on the association between eye disease status and RAVLT or letter verbal fluency are shown in Figure 4.

Figure 2.

Significance of interactions between eye disease status and KIBRA gene variants in the 4238 CLSA participants from Quebec. The –log10 P values of the interaction term between eye disease (AMD or glaucoma) and every SNP within the KIBRA genomic region from the linear regression models for each cognitive outcome are displayed. Each plot corresponds to the following cognitive outcomes: (A) RAVLT immediate recall, (B) RAVLT delayed recall, (C) letter verbal fluency, (D) category verbal fluency, and (E) mental alternation test. The x-axis represents the genomic positions of the SNPs, and the points are clustered together by eye disease. The y-axis is the significance of the interaction between eye disease and the SNP on the cognitive outcome. The lead SNP is labeled, and the colors of the points represent the linkage disequilibrium (measured with r²) of each SNP with the lead SNP. The candidate SNP around which the region was selected is outlined with a yellow triangle. The imputed SNPs are represented by color-filled points; genotyped SNPs are blank. The green dashed lines are the pooled P values (reflecting the significance of the region). The black dashed line is the 0.05 P value threshold. The orange dashed line is the significance threshold after Bonferroni correction for the number of SNPs tested in the region.

Figure 3.

Significance of interactions between eye disease status and PDE7A/MTFR1 gene variants in the 4238 CLSA participants from Quebec. The –log10 P values of the interaction term between eye disease (AMD or glaucoma) and every SNP within the PDE7A/MTFR1 genomic region from the linear regression models for each cognitive outcome are displayed. Each plot corresponds to the following cognitive outcome: (A) RAVLT immediate recall, (B) RAVLT delayed recall, (C) letter verbal fluency, (D) category verbal fluency, and (E) mental alternation test. The x-axis represents the genomic positions of the SNPs, and the points are clustered together by eye disease. The y-axis is the significance of the interaction between eye disease and the SNP on the cognitive outcome. The lead SNP is labeled, and the colors of the points represent the linkage disequilibrium (measured with r²) of each SNP with the lead SNP. The candidate SNP around which the region was selected is outlined with a yellow triangle. The imputed SNPs are represented by color-filled points; genotyped SNPs are blank. The green dashed lines are the pooled P values (reflecting the significance of the region). The black dashed line is the 0.05 P value threshold. The orange dashed line is the significance threshold after Bonferroni correction for the number of SNPs tested in the region.

Figure 4.

Modification of the effect of eye disease status on cognitive outcomes by lead SNPs from the CLSA replication analysis. Plots are shown for effect modification by (A) rs11134511 (KIBRA), (B) rs6897615 (KIBRA), and (C) rs17395878 (PDE7A/MTFR1). The SNP genotypes are on the x-axis, and the scores of the specific cognitive outcome are on the y-axis. Cognitive outcomes were standardized as z-scores. The colors of the lines correspond to one of the three eye disease status categories.

Participants with AMD and the CC genotype for rs11134511 had a lower RAVLT delayed recall z-score than those without AMD or glaucoma, whereas there was no difference between the eye disease groups for those with an AA genotype (Fig. 4A). Participants with glaucoma and the AA genotype for rs6897615 had a lower letter verbal fluency z-score than those without AMD or glaucoma, but there was no difference between the eye disease groups for those with a CC genotype (Fig. 4B). Participants with glaucoma and a GG genotype for rs17395878 had higher RAVLT delayed recall scores than those without AMD or glaucoma, but there was no difference between the eye disease groups for those with the CC genotype (Fig. 4C).

Discussion

This study investigated the potential modification of the effect of age-related eye disease on cognitive function by gene variants within the KIBRA and PDE7A/MTFR1 genomic regions. The initial analysis using Montreal hospital-based study data examined two candidate SNPs, rs17070145 (KIBRA) and rs10808746 (PDE7A/MTFR1), followed by a replication analysis using the CLSA data examining expanded genomic regions of approximately 20 kb.

Results from our initial analysis identified statistically significant interactions between eye disease and SNPs rs17070145 and rs10808746 for certain cognitive outcomes. For category verbal fluency, the KIBRA SNP rs17070145 T-allele carriers with normal vision performed better than non–T-allele carriers, which is consistent with previous studies.^17,18 The biological mechanism through which KIBRA and glaucoma/AMD would interact is not known. Perhaps the advantage that the rs17070145 T-allele confers on memory is negated by the glaucoma or AMD disease processes or the vision loss that they cause. For immediate story recall, the PDE7A/MTFR1 SNP rs10808746 AA genotype was associated with better cognitive function in participants with normal vision, as seen with previous studies¹⁵; however, in those with glaucoma, cognitive scores were worse for those with an AA genotype but they were better in those with the GG genotype. This is consistent with a study in brain tumor patients that also found better cognitive function in those with the rs10808746 GG genotype.⁴⁶ The biological mechanism through which PDE7A/MTFR1 and glaucoma might interact is not known, although PDE7A/MTFR1 may regulate inflammation and oxidative injury.¹⁵

It is notoriously difficult to replicate gene–environment interactions due to the heterogeneity in the environmental and genetic characteristics, low statistical power, and inconsistent variable measurements.⁴⁷ Nonetheless, our large replication sample of similar ancestry as our discovery sample allowed us to identify a statistically significant interaction between glaucoma status and the PDE7A/MTFR1 genomic region after controlling for multiple testing with pooled P values. The interaction in the Montreal hospital-based study affected the outcome of immediate story recall, and the outcome affected in the CLSA was RAVLT delayed recall. Both cognitive outcomes deal with verbal memory retention. In addition, the CLSA replication of the interaction between AMD and KIBRA with category verbal fluency in the Montreal data was close to significant for the outcomes letter verbal fluency and RAVLT delayed recall. Although letter and category verbal fluency and RAVLT delayed recall capture different aspects of cognition (language and retrieval versus memory retention), the frontoparietal lobe of the brain is involved in both.^48,49

To our knowledge, our study is the first to examine the interaction between age-related eye diseases and candidate genetic factors other than APOE in their effect on cognitive outcomes. A previous study investigated interactions between baseline AMD severity status and APOE haplotypes in the development of cognitive impairment but did not find any significant interactions.⁵⁰ A strength of this work is the use of two independent samples of people with and without eye disease with similar quantitative cognitive measures. Limitations include the potential misclassification of AMD or glaucoma status in the CLSA data due to the use of self-report. However, a strong correlation (>0.9) between genetic associations using self-reported versus hospital recorded glaucoma was observed by DeBoever et al.⁵¹ Although the cognitive outcome measures in the Montreal Study and the CLSA were similar, they were not all exactly the same. The verbal fluency tests were exactly the same, but the immediate and delayed recall tests were different with different durations of delay (RAVLT vs. logical memory test), although both test memory and retrieval. Only participants of European descent from the province of Quebec were included in this study to maximize ancestry homogeneity and reduce the effect of population structure, thereby reducing the generalizability of our findings to populations not of European descent.

Our research suggests that KIBRA and PDE7A/MTFR1 gene variants are potential effect modifiers of the association between age-related eye disease and cognitive function. Further research is required to identify causal variants within the KIBRA and PDE7A/MTFR1 genomic regions. Improving our knowledge of all the genetic and environmental risk factors and their interaction on cognitive function will help us to understand how to better prevent and treat cognitive disease.