DNA methylation in APOE: The relationship with Alzheimer's and with cardiovascular health

Abstract Introduction Genetic variation in the apolipoprotein E (APOE) gene is associated with Alzheimer's disease (AD) and risk factors for cardiovascular disease (CVD). DNA methylationat APOE has been associated with altered cognition and AD. It is unclear if epigenetic marks could be used for predicting future disease. Methods We assessed blood‐based DNA methylation at 13 CpGs in the APOE gene in 5828 participants from the Generation Scotland (GS) cohort. Using linear mixed models regression, we examined the relationships among APOE methylation, cognition, cholesterol, the family history of AD and the risk for CVD. Results DNA methylation at two CpGs was associated with the ratio of total cholesterol and HDL cholesterol, but not with cognition, family history of AD, or the risk of CVD. Discussion APOE methylation is associated with the levels of blood cholesterol, but there is no evidence for the utility of APOE methylation as a biomarker for predicting AD or CVD.

CVD-related death in carriers of the 4-allele, 1 this has not always been observed. 10

APOE methylation in AD and CVD
Epigenetic modifications have been associated with many human disorders. 11 DNA methylation (DNAm) is most commonly observed as the addition of a methyl group to the 5-position of cytosine in the context of a cytosine-guanine dinucleotide (CpG 12 ). Many studies have found a link between AD and DNAm. 12,13 Candidate gene studies for APOE have shown that DNAm in this gene is associated with dementia [14][15][16][17] with neuritic amyloid plaque burden, 18 and with cognitive ability. 19 Associations have been suggested between epigenetic mechanisms and risk factors for cardiovascular disease. [20][21][22][23] A recent systematic review identified 34 candidate gene studies on DNAm in CVD. 4 However, few studies have explored modifications specifically in APOE, and there have been some conflicting results: while Karlsson et al. 15 found no evidence for differences in APOE DNAm in blood between patients with CVD and healthy controls, Ji et al. 24 reported APOE hypermethylation in the blood of patients with coronary heart disease compared to controls. No AD-or CVD-related associations were reported in the EWAS catalogue 25 ; for cg14123992, cg04406254, cg05501958, and cg21879725, associations with other traits were reported (Table S1 in supporting information).
To help clarify this role of APOE, we use data from Generation Scotland (GS), a large population-based cohort, the size of which is several times greater than samples in previous studies, providing a robust analysis of APOE DNAm and cognitive and vascular health. We included both AD and CVD in our analysis for the following reasons: First, APOE is both highly relevant to the genetics of AD, as well as of importance to cardiovascular health due to its function of clearing lipoproteins and cholesterol from the blood. 7,8 Second, the state of current knowledge is relatively limited for APOE DNAm in AD and for APOE DNAm in CVD.
Third, CVD and cardiovascular risk factors strongly influence the risk of AD and are a relevant research goal in the context of AD.
We characterize blood-based DNAm in the promoter region, 2nd and 3rd exons and introns, and 4th exon of the APOE gene, and explore associations among APOE DNAm and a variety of markers of cognitive function, AD, vascular health, and CVD. Due to the importance of developing new clinical biomarkers for health outcomes before the onset of disease (the neuropathological hallmarks of dementia start to appear decades prior to onset), we restrict our sample to ages between 30 and 65 years.

DNA Methylation and APOE measurements
DNAm in peripheral blood samples was profiled in two analysis sets (Set 1: n = 5087 in 2017 and Set 2: n = 4450 in 2019), using the Illumina HumanMethylationEPIC BeadChip as described previously. 29 Briefly, low quality samples, probes with low detection rates, and participants for which the predicted sex did not match the recorded sex were excluded (see Methods S1 in supporting information).
We removed related participants from Set 1 using a genetically determined relationship cut-off of >0.05 (GCTA GREML 30 ) to reduce the potential influence of shared genetics on the findings. The participants in Set 2 were unrelated to each other and to those in Set 1. We restricted the analysis to CpGs located on chromosome 19 between 45,409,039 and 45,412,650 bp, which corresponded to the region encompassing the APOE gene (UCSC GRCh37/hg19 genome build).
To restrict the age of our sample, we removed participants younger than 30 years and older than 65 years. After combining the two sets, our final sample consisted of a total of 13 CpGs in 5828 participants ( Figure S1 in supporting information).

Characterization of APOE methylation
We examined the DNAm status of 13 CpGs in the APOE gene (Figure 1A). Twelve CpGs exhibited similar DNAm levels to those previously described (Table S2 15

Approximations for the risk of AD and of CVD
When completing the questionnaires, participants had reported on the presence of "AD" and "heart disease" for themselves and for their family members. Only three participants reported having AD, and 203 participants (3.5%) reported having CVD. We therefore created approximate measures of risk that were based on reports by participants about these illnesses for close relatives (parents, siblings, and grandparents).
For both AD and CVD, each relative was assigned either 0 or 1 (0: absence of the disorder, 1: presence of the disorder) based on the report of the participant. To derive each risk measure, we calculated the weighted sums of family records for each participant as follows: The choice of weights used for a given family member was based on kinship/relatedness between that relative and the participant. The participants that self-reported as having AD or CVD were excluded from the models that included the risk of AD or the risk of CVD as predictor variables, respectively. Due to a low number of relatives with AD and a consequential skew in the distribution of AD risk, AD risk was transformed into a categorical variable AD class (low risk: no close relatives with AD, n = 4854; high risk: at least one close relative with AD, n = 974). Using this approach, we were able to estimate the risk for each disorder for every participant in the study.

Covariates
For alcohol consumption, we did not include participants that have quit drinking (n = 373) or whose answers relating to recent alcohol consumption according to their own assessment did not correspond to their usual drinking pattern (n = 1557). Information on smoking was processed as described previously 31 ; we did not include participants

Statistical analysis
For all numerical variables, outliers-defined as scores beyond four interquartile ranges from the median-were excluded. For the analyses exploring effects on DNAm as an outcome, linear mixed models were used; for analyses exploring effects on AD class as an outcome, logisticregression models were used.
Each model was run twice: First as a basic adjusted model that included the variable(s) as defined in the hypothesis as predictors, and cell counts, processing batch, and analysis set as covariates. Next, a fully adjusted model was run that included additional covariates (sex, smoking, alcohol consumption, education, deprivation index, and-in some models-physical activity, heart rate, body mass index [BMI], APOE genotype; see Methods S3 in supporting information).
The predictor-specific covariates used in the fully adjusted model All statistical analyses were performed in R.

Sample characteristics
Among the 5828 participants, 3399 (58.3%) were female and 2429 (41.7%) were male. The age range was 30 to 65 years and the mean age was 52.7 years (Table 1). APOE genotype frequencies were comparable to those described previously for the British population (Table 1). 10,33 DNAm measures resembled previous reports 19

and most
CpGs appeared normally distributed upon visual inspection ( Figure   S2 in supporting information). Correlations between DNAm levels in blood (this study) and brain tissue (publicly available datasets) for the 13 CpGs in this study ranged from -0.30 to 0.51 for the whole brain  (Table S2).

Associations between CpGs
The correlations in DNAm for all pairwise combinations between the 13 CpGs ranged from -0.49 to 0.78, with a mean absolute   Table S4 in supporting information).

Association between DNA methylation and genotype
After correcting for multiple testing, there was evidence for an association between APOE carrier status and DNAm at five CpGs in the fully adjusted models: cg14123992 ( 2 = 0.003, P = 3.9 × 10 -4 ), cg04406254

Age-dependent drift in DNA methylation
In the basic-adjusted model, 7/13 CpGs showed an association with age after correction for multiple testing (R 2 range: 4.0 × 10 -5 -0.03;

APOE methylation and cognitive function
We observed no association between general cognitive ability and DNAm in the fully adjusted regression models after correction for multiple testing (Table S8 in supporting information). There were associations between the individual cognitive tests and DNAm levels in the basic but not fully adjusted regression models (Table S9 in supporting   information).

APOE methylation and the risk of AD
To determine whether APOE carrier status was associated with AD class (0: absence of the disorder, 1: presence of the disorder), a logistic regression was run, with AD class as the outcome variable and . We observed no associations between DNAm levels and AD risk in the basic or fully adjusted models after correcting for multiple testing (Table S10 in supporting information).

APOE methylation and the risk of CVD
We observed no association between APOE methylation and the risk measure of CVD (Table S11 in

Correlations between CpGs and age-drift in DNA methylation
In this study, we used DNAm and phenotypic data from a large cohort, GS, to explore DNAm in the APOE gene and its association with risk factors for AD, CVD, and blood cholesterol.
We observed correlations among CpGs, which had been reported before. 19 Compared with Liu et al., 19 the correlations observed in our study were stronger and more of them were negative. In contrast to the findings of Karlsson et al., 15 the DNAm levels of five CpGs correlated with APOE carrier status for the different APOE alleles. This difference may be due to the increased power of our study over that of Karlsson et al. 15 We observe associations between most CpGs and imputed proportions of white blood cells. It has been suggested that relative numbers of cell subtypes might be caused by the same regulatory perturbations that give rise to certain phenotypes. 38 We did not explore potential associations between AD or cardiovascular risk factors, and cell proportions in this study, but it represents an interesting possibility for future research.
We replicated a finding by Ma et al. 22

Relationships among APOE methylation, AD, and cognition
Neither cognition nor family history of AD were associated with APOE methylation. This might be due to a lack of appreciable changes in APOE methylation before the onset of symptoms of AD. Most previous studies that associated differential APOE methylation with AD were conducted on tissue from patients that had either been screened positive for cognitive dysfunction, 15  and Liu et al. 19 used blood, while Shao et al. 16 and Wang et al. 17 used both. In fact, the latter were not able to replicate their findings from brain tissue in blood. Blood represents an attractive medium for identifying biomarkers for disease. Indeed, it has been reported that patients with AD and healthy controls can be distinguished based on gene-expression patterns in blood. 40 However, the APOE gene is differently expressed between brain tissue and blood 16,17 and APOE CpGs exhibit relatively modest correlations between blood and brain DNAm. [34][35][36] Thus, blood-based DNAm may exhibit AD-associated changes that are distinct from DNAm changes in the brain or they might appear later in the course of the disorder. Finally, while some studies have reported AD-associated changes in APOE methylation as described above, little research has been done on the topic and few-if any-replication studies have been performed to validate the effects.
Moreover, some prominent studies that investigated associations between DNAm and AD across the entire genome did not report APOE to be altered in the disorder. 13,41

Relationship between APOE methylation and blood cholesterol
We did not replicate the finding 24 of an association between APOE methylation and CVD; our results are in line with reports by Karlsson et al. 15 and Sharma et al. 42 However, we did find a negative association between APOE methylation and the ratio of the total to HDL cholesterol at cg08955609 and cg18768621; a finding that-to our knowledge-had not been reported before. However, due to the relatively small effect size, replication in other large cohorts or metaanalyses is required to confirm these findings. Moreover, our mea-

Limitations and future directions
The results of the present study offer additional insight into the epi- study is based on DNAm data from the blood, which does not necessarily correspond to DNAm patterns in the brain. This might preclude the identification of potential epigenetic changes prior to disease onset and allows only limited insight into underlying biological processes. However, due to the multitude of peripheral biological processes associated with AD, blood may be a legitimate tissue for DNAm studies. Finally, this study adopts a candidate gene approach; while APOE is a biologically plausible candidate for implications in AD and CVD, genome-wide approaches may inform on the relative importance of APOE.
In conclusion, we showed that CpGs in the APOE gene exhibit correlations within and between distinct regions of the gene and that DNAm levels at some CpGs of the APOE gene correlate with the APOE genotype. Furthermore, we found an association between DNAm level at cg8955609 and cg18768621, and blood cholesterol. We did not find differences in the levels of APOE methylation between individuals at low risk and individuals at high risk of developing either AD or CVD.

Andrew M. McIntosh has received research support from Eli Lilly and
The Sackler Trust. He has also received speaker fees from Illumina and Janssen.