Abstract
Background:
The Long Life Family Study (LLFS) is a family based, prospective study of healthy aging and familial longevity. The study includes two assessments of cognitive function that were administered approximately 8 years apart.
Objective:
To test whether APOE genotype is associated with change of cognitive function in older adults.
Methods:
We used Bayesian hierarchical models to test the association between APOE alleles and change of cognitive function. Six longitudinally collected neuropsychological test scores were modelled as a function of age at enrollment, follow-up time, gender, education, field center, birth cohort indicator (≤1935, or >1935), and the number of copies of ɛ2 or ɛ4 alleles.
Results:
Out of 4,587 eligible participants, 2,064 were male (45.0%), and age at enrollment ranged from 25 to 110 years, with mean of 70.85 years (SD: 15.75). We detected a significant cross-sectional effect of the APOE ɛ4 allele on Logical Memory. Participants carrying at least one copy of the ɛ4 allele had lower scores in both immediate (–0.31 points, 95% CI: –0.57, –0.05) and delayed (–0.37 points, 95% CI: –0.64, –0.10) recall comparing to non-ɛ4 allele carriers. We did not detect any significant longitudinal effect of the ɛ4 allele. There was no cross-sectional or longitudinal effect of the ɛ2 allele.
Conclusion:
The APOE ɛ4 allele was identified as a risk factor for poorer episodic memory in older adults, while the APOE ɛ2 allele was not significantly associated with any of the cognitive test scores.
INTRODUCTION
Cognitive decline, both normal and pathologic, is one of the most common complications of reaching older age. Preservation of good cognitive function or delaying the onset of cognitive decline is essential to maintaining quality of life in older adults, and it is important to identify risk factors of onset and rate of cognitive decline that can suggest therapeutic interventions. Several known factors including cardiovascular risk factors, alcohol use, smoking, and high systemic levels of inflammatory markers, as well as socioeconomic status contribute to the cognitive decline process [1, 2]. Cognitive decline patterns vary among older adults and are genetically regulated [3]. The apolipoprotein E (APOE) gene is one of the most important genes related to cognition. The gene has three alleles, namely ɛ2, ɛ3, and ɛ4, that result from the combination of the variations of two single nucleotide polymorphisms rs7412 and rs429358 [4]. Several studies have shown that carriers of the ɛ4 allele are at increased risk of dementia and Alzheimer’s disease, while the ɛ2 allele might have a protective effect against age-related neurodegenerative diseases [5, 6] and is associated with extreme human longevity [7, 8]. A relatively small number of studies have investigated the effect of both alleles on the rate of cognitive decline using longitudinally collected data [9–11]. The review by O’Donoghue [12] lists 40 studies of the association between APOE and cognition in longitudinal studies and only one study showed a protective effect of APOE ɛ2 on verbal episodic memory, while other studies showed a negative effect of APOE ɛ4 on various measures of cognition. Most of these studies were small (median sample size = 550), with largest sample size of 5,544 and length of follow-up ranging between 2 and 30 years (median = 5.6). Several factors may contribute to inconsistent findings, including sample size, short follow-up time, neuropsychological tests used, neurobiological mechanisms, and population ancestry.
The Long Life Family Study (LLFS) recruited over 5,000 individuals from longevous families. Participants underwent two in-person assessments, approximately 8 years apart, and attention, memory, and executive function were assessed through a battery of neuropsychological tests. The study included a relatively large sample of carriers of the APOE ɛ2 allele and provides a unique opportunity to assess whether APOE is associated with cross-sectional or longitudinal cognitive decline in this healthy aging cohort. In line with previous findings, we hypothesize that carriers of the APOE ɛ4 allele have increased risk for poorer cognitive function, while carriers of the ɛ2 allele are protected against cognitive decline.
METHODS
Study population
The LLFS is a multicenter longitudinal study for healthy aging and familial longevity that recruited 5,086 participants from three sites in the United States (Boston, New York, Pittsburgh) and one site in Denmark. The recruitment process and inclusion criteria for this study have been described [13] and were based on a metric of familial longevity that was calculated from the aggregated survival probabilities of family members [14]. The study recruited spouses of members of long-lived families as referents. The participants completed two in-person visits, where their physical and cognitive functions were assessed through questionnaires, performance measures, and neuropsychological tests. Approximately 4,700 participants provided blood samples for genotyping, and APOE alleles were determined from the SNPs rs7412 and rs429358 that were genotyped using real time PCR. APOE alleles were defined as ɛ2: rs7412 = T; rs429358 = T, ɛ3: rs7412 = C; rs429358 = T, ɛ4: rs7412 = C; rs429358 = C. All subjects provided informed consent and data are available via dbGaP (dbGaP Study Accession: phs000397.v1.p1).
Cognitive tests
Six neuropsychological tests were the main outcomes in our analyses. These tests include Verbal Fluency (category fluency for animals) to assess semantic memory and generativity; Digit Symbol Substitution Test (DSST) from the Wechsler Adult Intelligence Test (WAIS-R, 5) for processing speed; Digit Span forward and backward to measure working memory and attention; and Logical Memory (immediate and delayed recall) from the Wechsler Memory Scale –Revised (WMS-R, 4) to assess attention and episodic memory. The Mini-Mental State Examination (MMSE) was also administered but was not included in the analysis because of low variability.
Statistical analysis
Participants were divided into three genotype groups defined as: “APOE2 group”: carriers of the APOE genotypes ɛ2/ɛ2 or ɛ2/ɛ3; “APOE3 group”: carriers of the genotype ɛ3/ɛ3; “APOE4 group”: carriers of the genotypes ɛ3/ɛ4 or ɛ4/ɛ4.
We used APOE3 as reference group. We summarized participants’ characteristics using mean and standard deviation. We compared participants’ characteristics of the APOE2 and APOE4 groups to the APOE3 group using t-tests or χ2 tests. We analyzed the effect of ɛ2 and ɛ4 in two separate analyses using additive genetic models. To test for association between APOE alleles and each of the neuropsychological tests, we used Bayesian hierarchical modelling of the longitudinal values of each test score as a function of age at enrollment, follow-up time, gender, education, field center, birth cohort indicator (≤1935, or >1935), and the number of copies of ɛ2 or ɛ4 alleles. We started with a model with all main effects and interactions between the APOE variable, gender, education, and age at enrollment and follow-up time, and used a backward model selection algorithm to identify the significant interactions and main effects. All analyses were conducted in R3.5 and the Bayesian analysis was conducted using the rjags package. Full details of the model specification, the algorithm for model selection and an example of the full model in rjags are in the supplemental material. The LLFS data used in this analysis was frozen by June 2018.
RESULTS
Out of 5,086 LLFS participants, we excluded 22 participants with missing sociodemographic data, 387 participants with missing APOE genotype, and 90 participants with APOE genotype ɛ2/ɛ4. Tables 1 and 2 summarize demographic characteristics and cognitive test scores of the remaining 4,587 participants (1,785 in the older generation and 2,802 in the younger generation, Supplementary Tables 1 and 2 show similar information with breakdown of each APOE genotype). The APOE3 group (n = 3,038) was the most prevalent and was used as the referent group. Age at enrollment ranged from 71 to 110 years among the older generation (mean = 88.3 years; SD:7.71), and from 25 to 73 years among the younger generation (mean = 59.7 years; SD:7.10). In the older generation, the APOE2 group was older (89.4 years versus 88.3 years, p = 0.02), while the APOE4 group was younger (86.8 years versus 88.3 years, p < 0.004) than the APOE3 group at enrollment. At visit 2, there were no other significant differences in age, sex, education, percent deceased among genotype groups. In the younger generation, there were no significant differences in the demographic characteristics among the genotype groups. At enrollment, the APOE4 group had significantly lower DSST score (50.4 versus 51.7, p = 0.03) and lower Digit Span –Backward score (6.6 versus 6.9, p = 0.02) than the APOE3 group. At visit 2, the APOE4 group had lower DSST score than the APOE3 group (47.4 versus 48.9, p = 0.02). There were no significant differences in the cognitive test scores comparing the APOE2 group to the APOE3 in both generations at either visit.
Demographic characteristics and test scores of 1,785 older generation (born in or before 1935) LLFS participants
MMSE, Mini-Mental State Examination, DSST, Digit Span Substitution Test. aFollow up from Visit 1 through April 2018.
Demographic characteristics and test scores of 2,802 younger generation (born after 1935) LLFS participants
MMSE, Mini-Mental State Examination; DSST, Digit Span Substitution Test. aFollow up from visit 1 through April 2018.
Tables 3 and 4 and Supplementary Tables 3–6 show parameter estimates generated using the model selected with the credible interval algorithm. The overall conclusion is that there was no significant effect of the ɛ2 allele on either the baseline assessment or the rate of change over follow-up time on any of the neuropsychological tests, while the ɛ4 allele had a negative effect on the two logical memory tests at baseline but had no effect on their rate of decline. We describe below the details of the analysis of each test.
Parameter estimates of Animal Fluency, DSST, Digit Span tests by generation, without APOE genotype stratification
Parameter estimates in this table are generated using data from all subjects, and the generation specific effects were estimated using the interaction effects in Supplementary Tables 1 and 2. For example, the effect of age on the animal fluency score in the younger generation was estimated as the sum of the effects of age and 1935*age. Parameter estimates for ind1935 was only under the younger generation column because ind1935 was defined to have value 1 for younger generation and 0 for older generation.
Parameter estimates of Logical Memory tests: ɛ4 allele carriers versus non-ɛ4 allele carriers
Parameter estimates in this table are generated using data from all subjects, comparing ɛ4 carriers to non-ɛ4 carriers.
Animal fluency
Neither ɛ2 nor ɛ4 allele of APOE was associated with performance on animal fluency (Supp-lementary Tables 3 and 4). Age at enrollment, follow-up time, and some of the interactions with generation, sex, and education were significant (Supplementary Table 3), suggesting that the cross-sectional and longitudinal effects of age were different in the younger and older generations and were modified by sex and education. Table 3 describes the estimated age and follow-up effects by generation. Older age at enrollment was associated with a lower score (age effect=–0.17, 95% CI: –0.18, –0.15) and, for every year of follow-up, the score decreased by –0.06 points (95% CI: –0.09, –0.03) in the older generation while, in the younger generation, the effect of age at enrollment was not significant (age effect = 0.01, 95% CI: –0.03, 0.06). The analyses also predicted a significant increase in score for each year of follow-up time in the younger generation (0.18, 95% CI: 0.11, 0.25) that could be caused by a practice effect among the younger participants. Higher education had a positive effect on the score but slightly diminished with older age at enrollment (educ*age interaction effect = –0.01, 95% CI: –0.01, –0.005). The age effect was smaller in males (sex*age interaction effect = 0.04, 95% CI: 0.02, 0.05).
DSST
Only age at enrollment, gender and education were significantly associated with DSST score, while the effects of ɛ2 and ɛ4 alleles of APOE were not significant (Supplementary Tables 3 and 4). In the older generation, an older year of age at enrollment was associated with a decrease of 0.67 points (95% CI: –0.71, –0.64) on the DSST (Table 3). Follow-up time also had a negative effect on DSST (dage effect = –0.48, 95% CI: –0.54, –0.43). The negative effects of age at enrollment in the younger generation was smaller (–0.4, 95% CI: –0.48, –0.31), while the effect of follow-up time was not significant (–0.06, 95% CI: –0.20, 0.07). Higher education and female sex were associated with higher scores but the effect was reduced with older age at enrollment and longer follow up (educ*age interaction effect = –0.01, 95% CI: –0.01, –0.001; sex*age interaction effect = 0.12, 95% CI: 0.08, 0.15; sex*dage interaction 0.11, 96% CI 0.01,0.22).
Digit Span –Forward
Neither ɛ2 nor ɛ4 allele of APOE was associated with this test (Supplementary Tables 3 and 4). Age at enrollment, follow-up time, gender, and education were associated with the digit span forward score, in both the older and younger generations (Table 3). In the older generation, the score was expected to decrease by 0.03 points (95% CI: –0.03, –0.02) for every year of age at enrollment, and decrease by 0.12 points (95CI: –0.13, –0.11) for every year of follow-up time. In the younger generation, there was an estimated increase in forward span score as baseline age increased (0.02, 95% CI: 0.004, 0.04) and for each additional year in the follow-up time, the score decreased by –0.07 points (95% CI: –0.10, –0.05), thus suggesting a smaller rate of decline in the younger generation. Education was positively associated with the score (0.12 points, 95% CI: 0.10, 0.14) and the effect increased as follow-up time increased (educ*dage interaction effect 0.003, 95% CI: 0.0003, 0.01). Males tended to score higher by 0.13 points (95% CI: 0.04, 0.23) than females.
Digit span –backward
Similar to the Digit Span forward test, APOE was not associated with the backward span test score (Supplementary Tables 3 and 4). Older age at enrollment and longer follow-up time were negatively associated with the score only in the older generation (age effect = –0.03, 95% CI: –0.04, –0.02; dage effect = –0.04, 95% CI: –0.05, –0.03, Table 3), and had no significant effect in the younger generation. Higher education was positively correlated with the score but the effect decreased with older age at enrollment (educ*age interaction effect = –0.001, 95% CI: –0.002, –0.001). We did not detect any gender difference in this test.
Logical memory recall tests
As shown in Table 4, the ɛ4 allele had a negative effect on the logical memory tests (immediate recall ɛ4 allele effect = –0.31, 95% CI: –0.57, –0.05; delayed recall ɛ4 allele effect = –0.37, 95% CI: –0.64, –0.10) compared to carriers of ɛ3 or ɛ2. These effects were not modified by any of the other variables. In the older generation, older age at enrollment was associated with lower scores of both tests (immediate recall age effect = –0.10, 95% CI: –0.12, –0.09; delayed recall age effect = –0.12, 95% CI: –0.13, –0.10). However, consistent with a possible practice effect, follow-up time had positive effects on both tests (immediate recall dage effect = 0.06, 95% CI: 0.03, 0.08; delayed recall dage effect = 0.04, 95% CI: 0.01, 0.06).
The effects of age at baseline and follow-up time were different in the younger generation as indicated by the significant interactions of ind1935*age and ind1935*dage (Supplementary Table 6). In the younger generation, both older baseline age and follow-up time were associated with higher scores of both tests. In the immediate recall test, the analysis estimated an increase of 0.09 points (95% CI: 0.06, 0.12, Table 4) with every additional year increase in baseline age, and an increase of 0.20 points (95% CI: 0.14, 0.25) with every year of follow-up time. Similarly, in the delayed recall test, for every one-year increase in baseline age the score increased by 0.07 points (95% CI: 0.04, 0.11), and by 0.19 points (95% CI: 0.13, 0.25) for every year of follow-up time. The age affects were modified by sex and education. Male sex reduced the effect of age at enrollment (immediate recall sex*age interaction effect = 0.02, 95% CI: 0.002, 0.03; delayed recall sex*age interaction effect = 0.03, 95% CI: 0.02, 0.04, Table 4). The advantage of higher education diminished slightly as baseline age increased (immediate recall educ*age interaction effect = –0.003, 95% CI: –0.01, –0.001; delayed recall educ*age interaction effect = –0.003, 95% CI: –0.005, –0.001), and also diminished as follow-up time increased in immediate recall (educ*dage interaction effect = –0.01, 95% CI: –0.02, –0.002).
DISCUSSION
We conducted a comprehensive analysis of the effect of APOE alleles on age-related change in various different cognitive domains. The analyses confirm the negative effect of ɛ4 allele on episodic memory assessed by immediate and delayed recall on logical memory: compared to the ɛ2 and ɛ3 alleles, carriers of one or more ɛ4 alleles scored lower in both tests although they did not exhibit a faster rate of decline. We did not detect any significantly protective effect of the ɛ2 allele compared to the ɛ3 allele.
There is substantial literature linking the ɛ4 allele of APOE to poorer cognition at older age, faster rate of cognitive decline and higher risk for Alzheimer’s disease. The review by O’Donoghue and colleagues [12] identified 12 cross-sectional and 15 longitudinal studies that reported a significant negative association between ɛ4 and episodic memory. Our findings show poorer memory in carriers of ɛ4 compared to other genotypes, but did not detect a significantly faster rate of decline. A study of centenarians [15] also reported similar adverse effect of the ɛ4 allele on a memory cognitive test score using Beta regression modeling. Rawle et al. showed that ɛ4 homozygous carriers have a faster rate of cognitive decline compared to other genotype carriers [16] in a study of comparable sample size. LLFS is a study of healthy aging and longevity with approximately 50% fewer ɛ4 carriers compared to the study in Rawle et al. and the smaller number of ɛ4 carriers may have reduced the power of our study. Alternatively, the lack of a difference in the rate of decline in the older generation may be due to a survivor bias. A review paper by Smith et al. [17] had suggested that young ɛ4 carriers presented better mental performance compared to ɛ3/ɛ3 carriers. A study by Caselli et al. [18] aiming to address the transition from cognitive advantage to cognitive deficit in ɛ4 carriers showed that the longitudinal decline began before age 60 and had faster acceleration compared to ɛ3/ɛ3 carriers. In LLFS with an average age of 87 years, ɛ4 carriers may be survivors with increased resilience to the risk conferred by the ɛ4 allele and therefore are not showing the accelerated declines seen in other samples. In addition, Tao et al. [19] suggested that the adverse effect of the allele might be activated by chronic low-grade inflammation. We conducted an additional analysis of the two Logical Memory Recall tests adjusting for baseline C-Reactive Protein (CRP) level and the results (Supplementary Table 7) showed a reduction of the adverse effect of ɛ4 in the Logical Memory Immediate Recall test and no change of the ɛ4 effect on the Logical Memory Delayed Recall test.
Our findings on the effect of ɛ4 and cognition are consistent with other analyses conducted in the LLFS but expand the set of results to longitudinal assessments of cognitive function. For example, Kulminski and colleagues showed that the ɛ4 allele increases the lifetime risk of neurological disorders including dementia and Alzheimer’s disease by 98% in both LLFS men and women [20]. Barral et al. [21] defined exceptional cognitive performance using predominantly immediate and delayed memory, and showed that being in an exceptional cognitive performance family was significantly associated with being a non-carrier of the APOE ɛ4 allele.
The fact that Logical Memory was the only cognitive test affected by ɛ4 is consistent with early episodic memory changes in Alzheimer’s disease. Reduced verbal fluency has also been posited as a marker of early Alzheimer’s disease that affects semantic memory [22]. In our sample there was no relationship between animal fluency and ɛ4. It is possible that the difference between phonemic and semantic fluency, that is poorer semantic fluency relative to phonemic fluency, is more indicative of early Alzheimer’s disease [23] than semantic fluency in isolation.
Our analyses did not detect any significant protective effect of ɛ2 on cognition although the data set included 314 carriers of one or more ɛ2 alleles. The results regarding the effect of the ɛ2 allele on cognition have been mixed. The Religious Orders Study found that ɛ2 carriers had an annual increase in episodic memory score while the ɛ4 subgroup decreased more rapidly compared to ɛ3 carriers [24]. The study also found that ɛ4 carriers declined faster than ɛ3 in semantic memory and processing speed, but not in working memory. Four additional studies suggested the ɛ2 allele has a protective effect and is associated with reduced odds for developing cognitive impairment [11, 25–27]. In contrast, the ɛ2 allele was not significantly associated with cognitive decline in [5]. A study of 18,000 people by Marioni et al. [28] did not detect any relationship between the ɛ2 allele and learning and episodic memory (Logical Memory), processing speed (DSST), and a fluency test. A clustering analysis of the LLFS cohort based on pattern of cognitive change by Sebastiani et al. [29] detected a cluster of slowest changers of DSST that was enriched for ɛ2 carriers, and a cluster of fastest changers that was enriched for ɛ4, suggesting that the effect of ɛ2 may be modified by other factors.
Our study has some limitations. Only 55% of LLFS participants completed a second visit, though the dropout rates are comparable in each group (E2 = 46%, E3 = 45%, E4 = 40%). Supplementary Figure 2 shows a forest plot of the standardized mean differences of the neuropsychological tests among the three genotype groups at both visits and suggests that differences between participants who completed both visits and those that completed one visit should not affect the results. Secondly, having at most two time points restricts the analysis to a linear model rather than any nonlinear models. Lastly, our study sample is highly ethnically homogeneous that 99% of the study population are Caucasians. Therefore, we cannot infer if APOE alleles have different effects in different ethnic groups.
In conclusion, APOE ɛ4 allele was confirmed as a risk factor for episodic memory in older adults, while APOE ɛ2 allele was not significantly associated with any of the cognitive tests, and neither allele appear to modify the rate of cognitive decline.
Footnotes
ACKNOWLEDGMENTS
This work was supported by the National Institute on Aging (NIA cooperative agreements U19-AG023122, 5U01AG023744, 5U01AG023755, 5U01AG023749, 5U01AG023746, 5U01AG023712, R21AG056630, R01AG061844, and K01AG057798 to SA). The funders had no role in drafting this manuscript.
Thomas T. Perls, MD, MPH, Geriatrics Section, Department of Medicine, Boston University School of Medicine; Paola Sebastiani, PhD, Institute for Clinical Research and Health Policy Studies, Tufts Medical Center.