Background

Alzheimer’s disease (AD) is the most common cause of dementia. The development of AD starts with a preclinical stage where pathophysiological changes in brain amyloid-β are present, [1,2,3,4] while cognition is still intact. [5, 6] Once amyloid-β pathology is present it can take several years to decades to develop dementia. [7] AD drug development programs are shifting their focus to individuals in the earliest disease stage, because this is when individuals could benefit most from disease-modifying therapy. For this, it is important to be able to identify cognitively unimpaired individuals at risk of future cognitive decline in the ageing population.

Genetic risk factors might help to identify individuals with increased risk of develo** AD. With heritability estimates of 58–79%, late-onset AD is a highly heritable disease. [8, 9] A well-known major genetic risk factor for AD is the apolipoprotein E (APOE) ɛ4 allele. APOE ɛ4 heterozygotes are at three-fold greater risk of AD and ɛ4 homozygotes at eight- to fifteen-fold compared to APOE ɛ4 non-carriers. [10] Additionally, it has been shown that APOE ɛ4 carriers have a higher prevalence of amyloid-β pathology at younger ages relative to non-carriers [5] and that APOE ɛ4 predicts cognitive decline in healthy controls. [11, 12] However, up to 40% of AD cases do not carry an APOE ɛ4 allele and the age at onset of AD dementia in APOE ɛ4 carriers varies widely, indicating that other genes or environmental risk factors are involved in causing the disease. Genome wide association studies (GWAS) in AD showed that, in addition to APOE ɛ4, over 80 genetic variants were significantly associated with AD. [13] By combining these multiple genetic variants, a so-called ‘polygenic risk score’ (PRS) can be calculated for each individual. Typically, higher PRS scores reflect increased risk for AD. [14] In cognitively unimpaired individuals, a high AD-PRS has been shown to be associated with amyloid-β pathology [15,16,17,18] and cognitive decline. [17,18,19,20,21] However, some studies did not find an association between AD-PRS and amyloid-β pathology [16] or cognitive decline. [22,23,24] Discrepant findings could be related to variability in the proportion of amyloid-β positive individuals in cognitively unimpaired groups, which ranges from 24 to 33%[5]. Other explanations for discordant results could be related to the number of genetic variants included in the AD-PRS, variability of cognitive tasks or composite scores used, or the follow-up duration. Furthermore, some studies included the APOE genotype in the AD-PRS, while other studies did not [16, 25,26,27]. Thus, it remains unclear to what extent the independent aspects of amyloid-β, APOE genotype and AD-PRS are related to cognitive decline in cognitively unimpaired individuals.

The aim of the present study was to investigate how APOE genotype, AD-PRS (APOE region excluded) and amyloid-β pathology relate to decline in cognition over time in cognitively unimpaired older individuals, and whether such effects are observed for specific cognitive domains (i.e., memory, attention, executive function and language).

Methods

Participants

Participants were part of the EMIF-AD PreclinAD study, which is a study investigating risk factors for amyloid-β pathology and cognitive decline in cognitively unimpaired older adults. [28] The PreclinAD study recruited participants from two sites: the Manchester and Newcastle Age and Cognitive Performance Research Cohort (ACPRC) in Manchester [29] and the Netherlands Twin Register (NTR) [30] in Amsterdam. Baseline inclusion criteria for the EMIF-AD PreclinAD study were age ≥ 60 years, Telephone Interview for Cognitive Status modified score of ≥ 23, [31] delayed recall score > − 1.5 SD of demographics-adjusted normative data on the Consortium to Establish a Registry for Alzheimer’s Disease 10-word list, [32] global Clinical Dementia Rating (CDR) score of 0, with a score of 0 on the memory sub domain [33] and a 15-item Geriatric Depression Scale score of < 11. [34] Baseline exclusion criteria were any neurological, systemic or psychiatric disorder that could cause cognitive impairment. Included participants had cognitive data available, an AD-PRS, an amyloid-β positron emission tomography (PET)-scan and/or APOE genotype data at their baseline visit (n = 276, of which 97 were complete twin pairs). All participants underwent an extensive baseline assessment including Mini-Mental State Examination (MMSE), neuropsychological assessment, acquiring years of education, and blood sampling. At two-year follow-up, MMSE, CDR and neuropsychological assessment were repeated, which were also used to monitor disease progression. According to protocol, if cognitive impairment was suspected by the study physician a neurologist was consulted and, if necessary or possible, a diagnostic work-up in the Alzheimer Center Amsterdam was performed. Each participant gave written informed consent, in accordance with the Medical Ethics Review Committee of the VU University Medical Center, the National Research Ethics Service Committee North West—Greater Manchester South and the Helsinki Declaration of 1975.

Neuropsychological assessment

In Manchester, neuropsychological testing was performed during the research facility visit at both baseline and follow-up. In Amsterdam, neuropsychological testing at baseline was performed at home, and during follow-up at the VU University Medical Center or at home when participants were unable to come to the hospital (n = 19, 10%). We used the following tests to assess memory performance: the total immediate recall and delayed recall of the Rey Auditory Verbal Learning Test (RAVLT), [35, 36] the twenty minute recall of the Rey Complex Figure Test (RCFT), [37, 38] the total errors of the Cambridge Neuropsychological Test Automated Battery (CANTAB) Paired Associate Learning (PAL) test, [39, 40] and the total score of the Face Name Associated Memory Examination (FNAME) names and occupations delayed recall. [41, 42] To assess attention, we used the Trail Making Test (TMT) part A, [43] the forward condition of the Digit Span (total span score), [44] the simple accuracy score of the CANTAB Reaction Time (RTI) test and the mean response latency of the CANTAB Rapid Visual information Processing (RVP) test. [40] To test executive function, we used the TMT part B [43] (corrected for TMT part A), the backward condition of the Digit Span (total span score), [44] letter fluency (in English testing the letters F A S and in Dutch the letters D A T [45]) and the between errors score of the CANTAB Spatial-working Memory (SWM) test. [40] For language function, we used category fluency (animal fluency) one minute and the graded naming test (GNT). [46,47,48,49] FNAME data were missing in 22 participants (8%) at baseline and in 20 participants (9%) at follow-up. CANTAB RVP data were missing in 30 participants (11%) at baseline and in 31 participants (14%) at follow-up, due to lack of time or fatigue effects. For all other tests, 0-3% of test scores were missing.

Genoty** and polygenic risk scoring

All genetic variants in our cohort were determined by applying standard genoty** and imputation methods and we applied established quality control methods. [50, 51] The genotype imputation method has been previously described. [52] In short, all individuals were genotyped using Illumina Global Screening array (GSA) with shared custom content (Illumina, Incl). We used high-quality genoty** in all individuals (individual call rate > 98%, variant call rate > 95%). All individuals’ reported sex matched with their genetic sex. Variants that departed from Hardy–Weinberg equilibrium were excluded at p < 1 × 10− 6. Genotypes were prepared for imputation using bcftools (v1.9) [53] for removing ambiguous (single nucleotide polymorphisms) SNPs, and flip** and swap** alleles to align to GRCh37/hg19. This was followed by haplotype phasing using SHAPEIT2 [54] and imputation of unobserved genotypes using Minimac3 [55] using a precompiled Haplotype Reference Consortium (HRC) reference panel. [56] We calculated a weighted individual AD-PRS based on the 83 genetic variants that showed genome-wide significant (GWS, p < 5e− 8) evidence of association with AD. [13]APOE haplotypes were not included in the AD-PRS. The selected variants were directly genotyped (median genoty** rate = 1) or imputed with high quality (median imputation score R² = 0.98, minimum R² = 0.5). The AD-PRS was generated by multiplying the genotype dosage of each risk allele for each variant by its respective weight and then summing across all variants. [57] Weights used can be found in (Table S1). [13] For analyses we used an ordinal APOE genotype variable (0 = APOE ɛ2ɛ2:ɛ2ɛ3; 1 = APOE ɛ3ɛ3; 2 = APOE ɛ2ɛ4:ɛ3ɛ4; 3 = APOE ɛ4ɛ4) and a normalized AD-PRS (mean = 0, standard deviation = 1). From all participants APOE data was missing for one participant (0.4%) and AD-PRS data was missing for five participants (1.8%). This study is independent of the discovery of the 83 genetic variants associated with AD, because participants were not included in Bellenguez et al. (2022).

Amyloid-β PET-scan

At baseline, all participants were scanned from 90 to 110 min (4 × 5 min frames) after intravenous injection of 185 MBq (± 10%) [18F]flutemetamol. [28, 58] In Manchester, PET-scans were performed on a high-resolution research tomography brain scanner (HRRT; Siemens/CTI, Knoxville, TN, USA) at the Wolfson Molecular Imaging Centre of the University of Manchester. A 7-min transmission scan using a 137Cs point source was acquired for subsequent attenuation and scatter correction of the PET data. [59] In Amsterdam, PET-scans were performed on a Philips Ingenuity TF PET-MRI (Philips Healthcare, Cleveland, USA) at the Amsterdam UMC, location VU University Medical Center. Full acquisition details have been described previously. [28, 60] Immediately prior to each part of the PET-scan, a T1-weighted gradient echo pulse MRI scan was acquired for attenuation correction of the PET data. Three GE healthcare (GEHC) certified readers (SFC in Manchester; BNMvB and LEC in Amsterdam) visually rated standardized uptake value images as amyloid-β negative (predominantly white matter uptake) or positive (binding in one or more cortical brain region or striatum unilaterally), blinded to clinical and demographic data and according to GEHC guidelines. [61] Majority visual read was applied for both Manchester and Amsterdam obtained PET-scans.

Statistical analysis

We imputed missing cognitive test scores (0–14% of participants per test) by single imputation, using predictive mean matching as method [62] with follow-up time, genetic relatedness, baseline age, sex, education, amyloid-β status, and MMSE score as predictors. After imputation, baseline and two-year follow-up individual cognitive test scores were standardized using their baseline mean and standard deviation across the total sample (n = 285) (Fig. 1). CANTAB PAL, RVP and SWM scores and the TMT part A and B scores were inverted by multiplying the z score with − 1, so that for all tests lower scores reflect worse performance. Next, we averaged z-scores of the domain specific tests into four cognitive composite scores (i.e., memory, attention, executive function, and language).

Fig. 1
figure 1

Flowchart of participant selection. None of the participants developed AD dementia or MCI between baseline and follow-up. Abbreviations: EMIF-AD = European Information Framework for Alzheimer’s Disease; amyloid-β = amyloid-beta; APOE = apolipoprotein E; PRS = polygenic risk score

Full size image

First, we compared demographic variables between amyloid-β normal and amyloid-β abnormal groups using generalized estimating equations (GEE), including a random effect to account for family-relatedness in the Amsterdam cohort. We studied associations of APOE genotype (continuous dose: 0 = APOE ɛ2ɛ2:ɛ2ɛ3; 1 = APOE ɛ3ɛ3; 2 = APOE ɛ2ɛ4:ɛ3ɛ4; 3 = APOE ɛ4ɛ4) and AD-PRS (z-score) with amyloid-β status using separate GEE models (amyloid-β ~ APOE; amyloid-β ~ AD-PRS, respectively), including age, sex and center as covariates and APOE genotype and population substructure (principal components (PC) 1–3) depending on AD-PRS or APOE genotype as predictor. Next, we studied associations of APOE genotype (continuous dose), AD-PRS (z-score) and baseline amyloid-β status with changes in memory functioning over time using linear mixed effects models (LMMs). Models included subject-specific random intercepts and fixed slopes, a random effect for family, thereby correcting for clustering in the data, and were adjusted for age, sex, years of education and center. APOE genotype or population substructure (PC1-3) were added depending on using APOE genotype or AD-PRS as predictor. First, we tested with separate models the effects of APOE genotype, AD-PRS and amyloid-β status on memory decline with memory composite score as outcome (Memory ~ APOE×time; Memory ~ AD-PRS×time; Memory ~ amyloid-β×time, respectively). Next, we tested in separate models main and interaction effects of APOE genotype and AD-PRS on memory decline (Memory ~ APOE×AD-PRS×time), of APOE genotype and amyloid-β status on memory decline (Memory ~ APOE×amyloid-β×time) and AD-PRS and amyloid-β status (Memory ~ AD-PRS×amyloid-β×time). Finally, we tested the 4-way interaction of APOE×AD-PRS×amyloid-β×time on memory. Interaction terms were removed when not significant (p > 0.05). All models were repeated using cognitive composite scores of attention, executive function and language as outcome.

All analyses were corrected for multiple testing using the false discovery rate (FDR). FDR-corrected p values < 0.05 were considered significant. [63] Statistical analyses were performed in RStudio version 3.6.1 “Action of the Toes” (http://www.r-project.org/), using the following packages: mice_3.13.0, gee_4.13-20, lme4_1.1–26, lmerTest_3.1-2, emmeans_1.4.8, ggplot2_3.3.5.

Results

Demographics

We included 276 cognitively unimpaired older participants (Fig. 1) (mean age 74.7 ± 9.7, 63% female, 11.5 ± 2.7 years of education, mean follow-up 2.0 ± 0.4 years). Fifty-two participants (19%) had a positive visual read of their PET-scan. These participants were older, and had a lower composite memory, executive function and language score compared to participants with an amyloid-β negative PET-scan (Table 1).

Table 1 Baseline and follow-up sample characteristics
Full size table

Associations of APOE genotype and AD-PRS with amyloid-β status

Using separate models, we found that both higher APOE dose scores and higher AD-PRS were associated with a higher probability of amyloid-β positivity (odds ratio [OR] 3.99 [95% confidence interval (CI) 2.29 to 6.84], per allele; OR 1.43 [95% CI 1.02 to 2.00], respectively) (Fig. 2).

Fig. 2
figure 2

Associations between APOE genotype, AD-PRS and amyloid-β status. Values given are odds ratio [95% confidence interval] as estimated by Generalized estimating equations (predictor: APOE genotype or normalized AD-PRS; outcome: amyloid-β status (reference = amyloid-β negative status), including random effect for family-relatedness (comparable with logistic regressions, but corrected for clustering in the data), adjusted for age, sex, center, and population substructure (PC1-3). The odds ratio for the AD-PRS reflects the odds of having an amyloid-β positive status per one standard deviation increase in the AD-PRS. Amyloid-β status was based on visual read [18F]flutemetamol PET standardized uptake value images. Abbreviations: AD = Alzheimer’s Disease; APOE = apolipoprotein E; CI = confidence interval; OR = odds ratio; PC = principal components; PRS = polygenic risk score

Full size image

Associations of APOE genotype, AD-PRS and amyloid-β with cognitive functioning

At baseline, no significant associations were observed between APOE genotype, AD-PRS and amyloid-β status with memory performance. Across the total group, memory scores declined over two years (β(SE) = -0.03(0.01), p = 0.04). Higher APOE dose was associated with steeper decline in memory functioning (β(SE) = -0.05(0.02), p = 0.008; Table 2; Fig. 3 A). Individuals with a high AD-PRS showed steeper decline in memory functioning over time than individuals with low AD-PRS (β(SE) = -0.04(0.01), p = 0.006; Table 2; Fig. 3B). Individuals with abnormal amyloid-β showed steeper decline in memory functioning over time than individuals with normal amyloid-β (β(SE) = -0.08(0.04), p = 0.02; Table 2; Fig. 3 C). We then examined the interactions of APOE genotype and AD-PRS on the rate of memory decline, which was not significant. Repeating analyses including APOE genotype and AD-PRS as main effects in the same model, we found that both higher APOE genotype dosage and a high AD-PRS were associated with steeper decline in memory over time (Table 2). When examining interactions of APOE genotype dosage and amyloid-β status on the rate of memory decline, we found a significant interaction effect (p-interaction = 0.02) (Fig. 4). Repeating analyses after stratifying for amyloid-β status, we found that a higher APOE dose was associated with steeper memory decline in individuals with abnormal amyloid-β (β(SE) = -0.13(0.06), p = 0.03), and no association was found in individuals with normal amyloid-β (β(SE) = -0.02(0.02), p = 0.46; Table 2). When examining interactions of AD-PRS and amyloid-β status on the rate of memory decline, no significant effects were found. Repeating analyses including AD-PRS and amyloid-β status as main effects in the same model, we found that both abnormal amyloid-β and high AD-PRS were related to steeper decline in memory over time (Table 2). The model including the 4-way interaction term (APOE×AD-PRS×amyloid-β×time) was not significant. After removing ɛ2ɛ4 carriers from the sample, the results and interpretation did not change (Table S2).

Table 2 Baseline and annual change effects of APOE genotype, PRS and amyloid-β status in memory composite scores
Full size table
Fig. 3
figure 3

Effects of APOE genotype, AD-PRS and amyloid-β status on changes in memory composite scores. Model-based estimation of longitudinal changes in memory composite score, based on APOE genotype (A), polygenic risk score (PRS) for Alzheimer’s Disease (B) and amyloid-β status (C). Neuropsychological tests were z-transformed and averaged for the composite memory score. APOE genotype dosage variable used as factor for plots: ɛ2 heterozygous/homozygous (APOE ɛ2ɛ2:ɛ2ɛ3), ɛ3ɛ3 (APOE ɛ3ɛ3), ɛ4 heterozygous (APOE ɛ2ɛ4:ɛ3ɛ4) and ɛ4 homozygous (APOE ɛ4ɛ4). AD-PRS depicted in tertiles. Amyloid-β status was based on visual read [18F]flutemetamol PET standardized uptake value images. Fixed-effect covariates were baseline age, sex, education, center and population substructure. For statistics see Table 2. AD = Alzheimer’s disease; APOE = apolipoprotein E; PET = positron emission tomography; PRS = polygenic risk score

Full size image
Fig. 4
figure 4

Interaction effects between APOE genotype and amyloid-β status with memory functioning over time. Amyloid-β status was based on visual read [18F]flutemetamol PET standardized uptake value images. APOE dosage variable used as factor for plots: ɛ2 heterozygous/homozygous (APOE ɛ2ɛ2:ɛ2ɛ3, n = 34), ɛ3ɛ3 (APOE ɛ3ɛ3, n = 157), ɛ4 heterozygous (APOE ɛ2ɛ4:ɛ3ɛ4, n = 77) and ɛ4 homozygous (APOE ɛ4ɛ4, n = 7). Neuropsychological tests were z-transformed and averaged for the composite memory score. Interaction effect was statistically significant, see Table 2. Abbreviations: APOE = apolipoprotein E

Full size image

Examining composite attention, executive function and language scores, we found across the whole group all scores improved over two years (attention: β(SE) = 0.05(0.02), p = 0.02; executive function: β(SE) = 0.06(0.02), p = 0.004; language: β(SE) = 0.11(0.02), p < 0.001). However, individuals with higher APOE dose showed steeper decline in language functioning (β(SE) = -0.07(0.03), p = 0.04; Table S3, Figure S1G). When examining interactions of APOE genotype and AD-PRS as main effects in the same model, we found that higher APOE genotype dosage was associated with steeper decline in language over time, while AD-PRS was not (Table S3). We then examined interactions of APOE genotype dosage and amyloid-β status on the rate of language decline, and we found a significant interaction effect (p-interaction = 0.003) (Figure S2). Repeating analyses after stratifying for amyloid-β status, we found that higher APOE dose was associated with steeper language decline in individuals with abnormal amyloid-β (β(SE) = -0.22(0.07), p = 0.002), and no association was found in individuals with normal amyloid-β (β(SE) = -0.02(0.04), p = 0.62; Table S3). No significant associations were found for APOE genotype, AD-PRS and amyloid-β status with attention and executive function performance over time.

Discussion

In a cognitively unimpaired older population, we found that both higher APOE genotype dosage and higher AD-PRS were related to amyloid-β abnormality. Furthermore, we observed in separate models that AD-PRS, APOE genotype dosage and abnormal amyloid-β were associated with steeper decline in memory functioning over time in separate models. The association of higher APOE genotype dosage with steeper memory decline was dependent on amyloid-β abnormality, whereas AD-PRS explained variance in memory decline over time independently of APOE genotype dosage and amyloid-β status. In addition, APOE genotype dosage was associated with steeper decline in language functioning over time, and this association was dependent on amyloid-β abnormality. These results suggest that amyloid-β status, APOE genotype and AD-PRS all explain parts of cognitive decline in older individuals with initially normal cognition.

Our findings that both higher APOE genotype dosage and higher AD-PRS were associated with amyloid-β pathology replicates previous findings in cognitively unimpaired individuals. [64, 65] This suggests that besides APOE ɛ4 other genes may contribute to amyloid-β abnormality. Pathway-enrichment analyses [13] of the SNPs that were included in the AD-PRS showed most significant genes to be related to amyloid-β and tau, but also to lipid metabolism and immunity, highlighting the role of microglia and its potential interaction with amyloid precursor protein (APP) metabolism in AD development. [66]

At baseline, none of the variables examined were associated with memory performance as measured by a composite consisting of five memory tests. We previously did find an association between abnormal amyloid-β and lower baseline RCFT 3-minute recall scores in the Amsterdam cohort of the EMIF-AD preclinAD study. [67] Combining the two EMIF-AD cohorts in the sample of this study, using a composite memory score that combines multiple tests instead of separate tests and assessing associations with LMMs instead of GEEs may explain the different findings. When examining memory performance over time, we observed that an abnormal amyloid-β status was related to steeper decline in memory functioning over time compared to a normal amyloid-β status, which is in line with previous preclinical AD studies. [68,69,70] We also found that both a high AD-PRS and high APOE genotype dosage was related to decline in memory functioning over time. This observation is largely consistent with earlier studies in cognitively unimpaired individuals. [18, 19, 22, 24, 71] We further extend the literature by showing that APOE genotype dosage and high AD-PRS explained variance in memory decline independently. Further research is needed to understand whether specific risk genes contribute to the rate of cognitive decline in preclinical AD. While the effect of high APOE genotype dosage was specific for individuals with abnormal amyloid-β (which is in line with a previous study [72]), the high AD-PRS was related to steeper memory decline independently of amyloid-β status. Possibly, some SNPs included in the AD-PRS reflect risk for cognitive decline, through pathways which might be unrelated to amyloid-β pathology. Previous studies using pathway analysis methods on AD GWAS results showed that genetic variants linked to AD risk were associated with the immune system, [73] endocytosis, cholesterol metabolism, amyloid-β clearance and tau metabolism. [74, 75] Future research should investigate if pathway-specific PRSs contribute to distinct aspects of cognitive decline in AD. Additionally, APOE genotype dosage was associated with decline in language performance over time, dependent on amyloid-β abnormality. We previously showed in cognitively unimpaired individuals that the very first cognitive changes in the early stages of AD are to be found in memory and language domains. [76] As all cognitive scores improved over two years, reflecting learning effects, and we found no relation between APOE genotype, AD-PRS and amyloid-β status and decline on attention and executive function, it is possible with a longer follow-up period these individuals may start to show decline in these domains as well. Furthermore, our study evaluated genetic risk factors for AD development while previous population studies indicated that environmental factors also influence dementia risk. [77] We previously showed environmental factors to contribute to onset of amyloid-β aggregation, but not particularly to cognitive decline. [76] For future research, it will be of interest to investigate the contribution of risk genes and their possible interaction with environmental risk factors to AD development.

A strength of the present study is the relatively large sample of cognitively unimpaired older individuals with a substantial age range (60–94), AD biomarkers, genetic risk data, and repeated neuropsychological assessment available. Possible limitations are that by using a dichotomous variable for amyloid-β burden, more sensitive information could have been lost, because binary visual reads disregard the potential significance of quantifying early, pathological, amyloid-β accumulation. [78] In relation to APOE genotypes, we placed both APOE ɛ2ɛ4 and ɛ3ɛ4 carriers into the same category of the APOE genotype variable even though previous studies have shown that ɛ3ɛ4 carriers may have worse outcomes than ɛ2ɛ4 carriers. [79] However, removing ɛ2ɛ4 carriers from our sample did not change the results nor interpretation (Table S2). Since it has been shown that predictive performance of AD-PRS based on European samples is lower in non-European ancestry samples, [80] our results might not be directly generalizable due to the fact our study was solely based on a European sample. Although AD-PRS was associated with steeper memory decline in our study, effect sizes were small, which may limit its potential clinical use. Additionally, none of the participants progressed to mild cognitive impairment or dementia yet. In this study, 50 participants did not complete follow-up assessment and their current cognitive status is unknown. Comparing their baseline characteristics to remaining individuals in the study showed that they were older and were more often female and APOE ɛ4 carriers compared to the 226 individuals who were assessed at follow-up (see Table S4). The follow-up duration of an average of two years may have been too short to observe decline or clinical progression in initially cognitively unimpaired individuals, who take on average 5–6 years to show decline or clinical progression. [81] For this reason we aim to continue following our cohort.

Conclusion

Our findings provide further support among cognitively unimpaired older individuals that, in addition to abnormal amyloid-β and APOE genotype, the AD-PRS is also associated with memory decline. Future research should further investigate which specific genetic variants contribute to memory decline and through which mechanism, because this will be important for develo** treatments that may prevent decline in cognition.