|Home | About | Journals | Submit | Contact Us | Français|
Evidence supports a role for epigenetic mechanisms in the pathogenesis of late-onset Alzheimer’s disease (LOAD), but little has been done on a genome-wide scale to identify potential sites involved in disease. This study investigates human postmortem frontal cortex genome-wide DNA methylation profiles between 12 LOAD and 12 cognitively normal age- and gender-matched subjects. Quantitative DNA methylation is determined at 27,578 CpG sites spanning 14,475 genes via the Illumina Infinium HumanMethylation27 BeadArray. Data are analyzed using parallel linear models adjusting for age and gender with empirical Bayes standard error methods. Gene-specific technical and functional validation is performed on an additional 13 matched pair samples, encompassing a wider age range. Analysis reveals 948 CpG sites representing 918 unique genes as potentially associated with LOAD disease status pending confirmation in additional study populations. Across these 948 sites the subtle mean methylation difference between cases and controls is 2.9%. The CpG site with a minimum false discovery rate located in the promoter of the gene Transmembrane Protein 59 (TMEM59) is 7.3% hypomethylated in cases. Methylation at this site is functionally associated with tissue RNA and protein levels of the TMEM59 gene product. The TMEM59 gene identified from our discovery approach was recently implicated in amyloid-β protein precursor post-translational processing, supporting a role for epigenetic change in LOAD pathology. This study demonstrates widespread, modest discordant DNA methylation in LOAD-diseased tissue independent from DNA methylation changes with age. Identification of epigenetic biomarkers of LOAD risk may allow for the development of novel diagnostic and therapeutic targets.
Worldwide changes in demography are leading to a rapid increase in the numbers of older adults at risk for dementia. Accordingly, the global prevalence of dementia is expected to quadruple from an estimated 35.6 million cases in 2010 to 115.4 million cases in 2050 . The global financial burden of dementia in 2010 was $604 billion (US dollars) including direct medical bills, formal social care, and informal care provided by unpaid caregivers .
Alzheimer’s disease (AD), a progressive, fatal neurodegenerative disease, is the most prevalent form of dementia. Less than 2% of AD cases represent early-onset AD (EOAD)  defined by disease onset prior to age 60 and genetic mutations in amyloid-β protein precursor (AβPP), presenilin-1 (PSEN-1), or presenilin-2 (PSEN-2) genes [3, 4]. Mutations in these genes dysregulate the AβPP pathway and directly lead to amyloid-β (Aβ) plaque accumulation, a major pathological hallmark of AD.
The remaining vast majority of cases are sporadic, termed Late-Onset Alzheimer’s Disease (LOAD) because they manifest symptoms after age 60. Approximately 60% of LOAD cases carry at least one apolipoprotein ε4 allele (APOE-ε4), while the global population prevalence of the allele is only approximately 22% [5, 6]. Pooled data on LOAD from recent, collaborative, large genome-wide association studies (GWAS) reported the population attributable risk for APOE variants was between 19% and 35% . GWAS also identified additional LOAD risk alleles (CLU,PICALM, BIN1, CR1, ABCA7, MS4A, EPHA1, CD33, CD2AP) that contribute added risk in population subsets [8–11]. These risk factor genotypes are neither necessary nor sufficient for LOAD development. Twin studies revealed incomplete concordance [12, 13] and variable age at onset  even among monozygotic pairs, highlighting the complex etiology of LOAD. These observations underscore the likely importance of non-genetic factors in LOAD etiology and spurred recent investigations of the epigenetics of AD.
Epigenetics is the study of heritable changes in gene expression that occur without changes to the underlying DNA sequence. Methylation  and hydroxymethylation  at the 5′ site on cytosines in cytosine-guanine (CpG) dinucleotides are important epigenetic modifications associated with gene expression in the human brain. Specific marks distinguish brain regions [17, 18] and epigenetic differences in human brain tissues have been associated with such neurological diseases as schizophrenia and bipolar disorder . Epigenetics is also a mechanism by which environmental exposures can translate to human disease [20, 21].
In AD cases lacking highly penetrant genetic susceptibility, the etiology of amyloid dysregulation is not well understood. Altered epigenetic regulation of tau and amyloid processing genes has been observed across multiple brain regions and is a potential mechanism for disease [22–24]. Human postmortem case-control studies observed global hypomethylation in the entorhinal cortex of AD subjects  and in the temporal neocortex neuronal nuclei of an AD monozygotic twin relative to their cognitively normal twin . Evidence for epigenetic involvement in AD pathogenesis spans human studies in various tissues, animal models, and cell culture, and was recently reviewed [27–29].
Significant transcriptome-wide gene expression differences have been observed between brain tissues of LOAD cases and controls [30, 31]. However, previous AD research on DNA methylation as a regulator of gene expression evaluated DNA methylation at 5′ promoter regions of a few candidate genes selected based on a priori hypotheses about AD molecular mechanisms. The current research provides a semi-unbiased, quantitative, genome-wide discovery of locations of DNA epigenetic differences in human frontal cortex brain tissue between LOAD cases and controls, which allows for identification of novel disease-associated genes. The gene identified in this study that best distinguished cases and controls was technically validated using an additional sensitive and quantitative method of DNA detection. This mark was also validated using a second population of samples. The functional significance of this DNA methylation mark was further confirmed by gene expression and protein quantification assays.
The NIA funded Michigan Alzheimer’s Disease Center (MADC) (P50AG008671; PI: Sid Gilman) maintains a well-clinically characterized cohort of AD and cognitively normal control subjects, many of which agreed to participate in autopsy and donated to the MADC Brain Bank. Upon autopsy, each left hemisphere was fixed in 10% neutral formalin for neuropathological diagnosis. The right hemisphere was sectioned coronally, flash frozen, and archived in MADC freezers at −80°C. Frozen tissue blocks 0.5 cm3 (50–90 mg) in size were dissected at −20°C from the mid-frontal gyrus of the frontal lobe and provided for this study. MADC frozen tissues were previously used in high quality expression  and proteomic studies .
Twelve age- and gender-matched pairs of LOAD cases (clinical diagnosis and Braak score ≥4) and controls (clinically confirmed and Braak score ≤2) were used for the genome-wide discovery phase of the project and for gene-specific technical validation. An additional thirteen matched pairs were included in the population validation phase, which included gene-specific DNA methylation, gene expression, and protein quantification studies. The demographic characteristics of all 50 brains included in this study are described in Table 1. Postmortem intervals (PMI) in hours for AD cases used in the Discovery Set were as follows: 3, 4, 7, 7, 7.75, 8, 8.75, 9, 11, 12, 14, 24. PMI in hours for controls used in the Discovery Set were as follows: 6, 6, 13.5, 14, 17, 18, 18, 18, 19.3, 20.5, 21.25, 24.5. Gray matter for DNA methylation, expression, and protein analysis was excised from the tissue sample and used in this study and vascular lesions were avoided.
DNA was extracted from all 25 matched pair samples using the Promega Maxwell Tissue DNA Kit (Madison, WI) according to manufacturer’s instructions. APOE genotyping was assayed using the Applied Biosystems TaqMan method (Foster City, CA) according to manufacturer’s instructions using the ABI 7900 HT machine .
DNA was bisulfite-treated using the Zymo EZ DNA Methylation Kit (Orange, CA) with a modified thermal cycling protocol (98°C for 10 min, 64°C for 17 h). Genome-wide DNA methylation was assessed with the Infinium HumanMethylation27 BeadArray (Illumina) performed at the University of Michigan DNA Sequencing Core facility in accordance with manufacturer’s instructions and previously published . Six cases and six control samples were randomly applied to each of two 12-sample arrays to avoid biasing case-control differences by batch effect. BeadArrays were imaged using the Illumina BeadArray Reader. Image processing and intensity data extraction are standard components of the BeadScan software that is associated with the BeadArray Reader. The Illumina BeadStudio Software generated percent methylation estimates (beta values) for each probe set based on Cy3 and Cy5 fluorescence intensities. Data was background normalized and exported for further processing.
All statistical analysis was performed with the R Statistical Software (version 2.10.1). CpG sites that failed on 10% of samples were not included in subsequent analyses. Linear models adjusting for age and gender were fit across all CpG sites using the limma package (version 3.2.3). As is standard in microarray analyses, empirical Bayesian variance methods were incorporated to site-specific moderated t-tests . The linear model used for each individual CpG site was as follows:
Top hits by case status were identified by p < 0.05 and false discovery rates were calculated. Additional analyses were performed on the top hit. We compared the basic linear model above to a model containing PMI using an F-test. We also compared the R2 goodness of fit of two simple linear regression models containing only either PMI or Case Status as predictors.
Samples were hierarchically clustered by the single linkage method across the top 26 hits by case status. Positional gene set enrichment analysis was performed using Gene Set Enrichment Analysis (GSEA) to determine statistical over-representation of disease specific epigenetic marks within chromosomal cytogenic bands containing at least 15 genes . Enrichment in promoter and 3′ UTR regulatory motifs of disease associated genes was determined by GSEA . Biological processes and molecular functions associated with LOAD gene lists were established using Gene Ontology .
Site-specific methylation technical (of the original 12 Discovery Set matched pairs) and population (of an additional 13 matched pairs with an expanded age range) validation of the top CpG hit was determined by bisulfite-pyrosequencing on the Qiagen Pyromark MD instrument (Valencia, CA). Using Pyromark Assay Design Software, a custom pyrosequencing assay was designed to include the two CpG sites present on the Illumina array (Target IDs cg01182697 and cg20793071). This amplicon is located in the promoter of the gene that most statistically distinguished LOAD cases and controls on the Illumina array. Primers were complementary to bisulfite treated DNA in regions without CpG nucleotides (Table 2). The region of interest was amplified by bisulfite-PCR with the following thermal cycling protocol: 15 min activation at 95°C; 50 cycles of 30 s denaturation at 94°C, 30 s annealing at 58°C, 1 min extension at 72°C; followed by a final extension for 10 min at 72°C. Serial dilutions of 100% methylated and unmethylated controls were used to test for any bias in amplification for each assay. Internal bisulfite conversion quality controls were incorporated at original sequence non-CpG cytosines by including C nucleotides in the dispensation order, which should be fully converted to Ts following bisulfite treatment.
Incorporation of either a T (for an unmethylated cytosine) or C (for a methylated cytosine) at each CpG provides a quantitative measure for consecutive CpG sites throughout the region sequenced. The level of methylation for each CpG within the target region of analysis was quantified using the Pyro Q-CpG Software. Primers and pyrosequencing assay file information including nucleotide dispensation orders and sequences to analyze are in Table 2.
Functional relevance of top methylation marks distinguishing AD cases and controls was assessed via SYBR green Real Time PCR gene expression assays. RNA was extracted from all 25 matched pair samples using an adjacent portion of the same gray matter sample used for DNA. RNA was extracted using the Qiagen RNeasy Lipid Tissue Kit (Valencia, CA), following homogenization with the Qiagen Tissue Lyser instrument. Assays were designed using Genscript software (Piscataway, NJ). cDNA was generated with the Bio Rad iScript cDNA Synthesis Kit (Hercules, CA) and the primers are listed in Table 3. Quantitative PCR assays were run with the iQ SYBR Green Supermix (Bio Rad) on the CFX96 C1000 Thermal Cycler (Bio Rad). CFX Manager software (Bio Rad) was used to determine the threshold cycle (CT) and perform inter-plate normalizations. CT values relative to β-actin levels were used to compute a fold change between matched pairs. A subset of samples was analyzed on the Affymetrix GeneChip Human Genome U133 Plus 2.0 Array in the University of Michigan Sequencing Core using standard protocols. We determined that β-actin is a suitable control in these samples as gene expression of β-actin did not differ by case control status at any of the 6 probe sets included in the Affymetrix Array (p: 0.13, 0.18, 0.23, 0.43, 0.48, 0.77). Gene expression data were evaluated for normality using R Statistical Software. To determine if higher methylation values were associated with decreased expression of target genes, Spearman correlation coefficients were calculated between CpG methylation (as measured by pyrosequencing) and the expression level data.
Adjacent portions of the same 25 matched-pair, gray matter tissue used for DNA were homogenized and protein extracted in Thermo Scientific RIPA buffer (Burlington, Ontario). Protein concentration was quantified using the Thermo Scientific Pierce BSA assay (Burlington, Ontario). Protein (25 μg) was loaded on 10% SDS-polyacrylamide gels for Western blot analysis. Transmembrane protein 59 antibody was purchased from Novus Biologicals (Littleton, CO) and Transmembrane protein 59 control protein was purchased from OriGene (Rockville, MD). The blots were imaged on the VersaDoc 5000MP instrument (Bio Rad) with Quantity One densitometry software (Bio Rad) under a consistent, constrained area. The levels of Transmembrane protein 59 were standardized to the corresponding tubulin band.
DNA was extracted from the frontal cortex of 25 age-and gender-matched LOAD case and control human postmortem pairs (Table 1). Sixteen pairs were male and nine pairs were female. LOAD cases and controls did not differ by age (p = 0.68). The mean PMI was 12.8 h, ranging from 3–28 h. Controls had significantly longer PMI than LOAD cases (p = 0.0004). The mean number of years in storage was 11.8, ranging from 2–21 years. Cases and controls did not statistically differ in the number of years in storage (p = 0.07).
In the Discovery Set, a randomly selected subset of 12 matched pairs ranging in age from 69–95 (mean age 79.8) were analyzed for genome-wide DNA methylation using the Illumina Infinium HumanMethylation27 BeadArray. The BeadArray represented 27,578 CpG sites corresponding to 14,475 unique genes. The average number of CpG sites per gene was 1.9, and 92.0% of CpG sites were within 1000 bp of a transcription start site. CpG sites within CpG islands were overrepresented on the array, as 72.5% of sites were within CpG islands. The average number of CpG sites per sample with Illumina detection p > 0.05 (considered failing) was 71.1 (range 11–360 sites). CpG sites that failed on more than 2 samples (>10% samples) were not included for further analyses (n = 171 sites).
The global distribution of 5′-cytosine modifications at all CpG sites measured by the array was bimodal, and the distribution of methylation levels was very similar between cognitively normal controls (Fig. 1A) and AD cases (Fig. 1B). There was a large cluster of sites that had less than 10% methylation (15,735 in controls and 15,619 in AD cases) based on the mean of each group of 12 samples. A second cluster of sites was modified between 75% and 100% (5,226 sites in controls and 5,162 sites in AD cases).
Of the 27,578 sites on the array, 25,380 sites were located in promoter regions as defined within 1,000 bp of a transcription start site. Only 2,198 sites were outside of known promoters. Promoter CpG sites had a median of 5.1% methylation (IQR = 2.1–5.2) while non-promoter sites had a median of 59.7% methylation (IQR = 11.0–84.0). This observation did not depend on AD case status. On the array, 20,006 sites were located within CpG islands and 7,572 sites were outside of CpG islands. CpG island sites had a median methylation of 3.2% (IQR = 1.7–9.6) and non-CpG island sites had 74.6% median methylation (IQR = 45.5–85.4). This did not depend on AD case status. Our discovery set of 24 samples ranged in age from 69–95 years (Table 1), and age was an important predictor of methylation. There were 2,416 sites and 2,227 unique genes associated with age (based on p = 0.05). Of these sites, 1,294 were hypermethylated with increasing age and 1,122 were hypomethylated with age. The top 25 CpG sites associated with age among controls are listed in Table 4.
Following age and gender adjusted linear fit models with empirical Bayseian standard error adjustments, 948 CpG sites representing 918 unique genes were significantly associated with AD case status (based on p = 0.05). Among these 948 sites, the maximum mean methylation difference between AD cases and controls was 19.5% at a CpG site 249 base pairs from the predicted TSS of C21orf56 on chromosome 21 (cases 34.8% methylated versus controls 15.9% methylated). The mean observed disease specific methylation difference across the 958 sites was 2.9% (IQR = 0.88–4.2).
The top 26 autosomal CpG sites distinguishing cases and controls (as defined by FDR) are depicted in a heatmap (Fig. 2). Samples clustered on case status with the exception of two control samples. One of those controls was the oldest control subject in the study at 94 years of age. The top 25 CpG sites that were significantly different by case status are outlined in Table 5.
Gene ontology analysis of the top 958 disease specific sites revealed hypermethylation in several molecular functions and biological processes associated with transcription. The top 10 molecular functions enriched for hypermethylation with AD cases were RNA polymerase II transcription factor activity (Fig. 3A), RNA binding, GTPase regulator activity, cytokine binding, DNA binding, lyase activity, ATPase activity, transcription factor activity, and nucleoside triphosphatase activity. Similarly, the top 10 biological processes associated with hypermethylation were nucleic acid metabolic process, DNA replication, regulation of nucleic acid metabolism, regulation of DNA dependent transcription, regulation of RNA metabolic process, regulation of cell cycle, DNA dependent transcription, positive regulation of RNA metabolic process, DNA metabolic process, RNA biosynthetic process, and nervous system development.
Hypomethylation was enriched at functions and processes related to membrane transport and protein metabolism. The top 10 molecular functions associated with hypomethylation in AD cases are electron carrier activity, cation transmembrane transporter activity, metal ion transmembrane transporter activity, enzyme binding, rhodopsin-like receptor activity, cation channel activity, integrin binding, phosphoric ester hydrolase activity, G-protein coupled receptor activity, and peptidase activity. The top 10 biological processes associated with hypomethylation in LOAD cases were carboxylic acid metabolic processes (Fig. 3B), organic acid metabolic process, biosynthetic process, cation transport, nitrogen compound metabolic process, amine metabolic process, negative regulation of developmental process, programmed cell death, apoptosis, and anti-apoptosis.
Several promoter and 3′ UTR regulatory binding motifs were enriched in the disease associated gene list. Hypermethylation in LOAD cases was observed in genes containing binding site motifs for transcription factors POU3F2 (p < 0.001) and HOXA4 (p = 0.004), and microRNAs MIR-9 (p = 0.002), MIR-518C (p < 0.001), MIR-1 (p = 0.025), and MIR-326 (p = 0.019). Genes containing MIR-140 (p = 0.04) and NFE2 (p = 0.019) motifs were hypomethylated in LOAD cases.
Positional gene set analysis of the 958 disease associated CpG sites identified enrichment of hypomethylation at the chromosomal location19q13 (Normalized Enrichment Score (NES) = 1.24), in cyto-genetic band region where the APOE gene is located . In addition, hypermethylation in LOAD cases was seen at 19q13 (NES = −0.72), a candidate location for genetic linkage with LOAD . We visually identified broad regions of altered methylation in AD brains compared to control brains (Fig. 4). These include the p-arm of chromosome 14 and distal q-arm of chromosome 3 (hypomethylated in AD brains compared to control brains) and the p-arms of chromosomes 10 and 17 (hypermethylated in AD brains compared to control brains). Chromosome 15 had the highest density of observed disease-specific methylation differences.
We checked the list of disease specific hits identified with the BeadArray for genes known to be associated with AD. There were two CpG sites on the array corresponding to each of the following genes: ABCA7, APOE, ABPP, BACE1, BDNF, BIN1, CD2AP, CD33, CLU, CR1, EPHA1, MAPT, MS4A6A, PICALM PSEN2, and TOMM40. There was one site representing PSEN1 on the array and this site was associated with AD (Cases mean methylation = 1.6%, Controls mean methylation = 2.6%; p = 0.034; cg11490446). Gene expression of the probe set for PSEN1 at exon 2 differed by case status based on the results of the Affymetrix gene expression array (probe 207782 s at, p = 0.0076, fdr = 0.35). The Spearman correlation coefficient linking expression of this gene expression probe set and methylation measured by the Illumina BeadArray is −0.61 (p value = 0.0014). AD cases were less methylated and had higher expression of this probe. There was no difference gene expression in the other five probe sets for PSEN1. One of the two sites corresponding to EPHA1 was associated with hypermethylation with age (p = 0.029; cg02376703). One of the two sites associated with PSEN2 was associated with hypomethylation with age (p = 0.030; cg25514304). The remaining CpG sites within LOAD candidate genes were not associated with differential methylation by case status or age.
We also assessed known human imprinted genes for their association with AD, focusing on CpG sites located within Differentially Methylated Regions (DMRs) . The Illumina array contains 10 sites in the DMR for DIRAS3, 8 sites each for PLAGL1 and GNAS, 7 sites for ZIM2, 4 sites for PEG10, 3 sites for PEG3, 2 sites for MEST, and 1 site each for the genes GRB10, KCNQ1, and SNURF. One of the sites in the DMR for DIRAS3 was more highly methylated in AD cases (43.4%) than controls (38.5%) (p = 0.024; cg21808053). One of the sites in the DMR for GNAS was hypomethylated with age among controls (p = 0.012; cg21625881) and the site for KCNQ1 was hypermethylated with age (p = 0.023; cg27119222).
After adjusting for age and gender, the highest ranking site (FDR = 0.36, p = 0.000013) associated with LOAD was a CpG site upstream of Transmembrane Protein 59 (TMEM59). TMEM59 is responsible for post-translational glycosylation of AβPP and leads to retention of AβPP in the Golgi apparatus . AD cases had 7.3% lower methylation at TMEM59 than controls, and this difference was more profound in older subjects (Fig. 5A). Methylation of TMEM59 was significantly associated with age in cases relative to controls (p = 0.013). In a second TMEM59 model, we added PMI as a predictor and compared the goodness of fit of the two nested models using an F-test. PMI did not improve the model fit and PMI is not a statistically significant predictor of methylation at the TMEM59 site. In a simple linear regression model with Case Status as the only predictor of TMEM59 methylation, the model fit R2 was 0.597. The model fit R2 with PMI as the only predictor was 0.132.
The methylation findings were technically validated by pyrosequencing a single assay containing both CpG sites from the Illumina array that were associated with TMEM59. The CpG density 1000 bp flanking the top hit on either side is 1.5%. Pyrosequencing the original 24 discovery samples confirmed the difference between cases and controls at TMEM59 was 2.7% (Fig. 5B and D). Methylation was again associated with age in cases relative to controls in the technical validation (p = 0.0084). Pyrosequencing of an additional 26 matched pair samples across an expanded younger age range (61–94) confirmed the age-associated reduction in methylation (p = 0.0022), and the association with AD case status was not statistically significant at an alpha of 0.05 (p = 0.088) (Fig. 5C).
We determined expression levels of TMEM59 at four points along the 8 exon transcript (Fig. 6A) (including the beginning, end, and two middle sites) to functionally validate the DNA methylation results with the TMEM59 gene. At the four locations along the transcript that were assayed by real time PCR, controls had lower RNA expression levels than cases (Fig. 6B). For the first assay on the transcript, AD cases had 24.9% higher expression (fold change 1.33; p = 0.0013). For two assays in the middle of the transcript, AD cases had 20.5% (fold change 1.26; p = 0.0071) and 28.3% (fold change 1.40; p = 0.056) higher expression. The assay at the end of the transcript showed AD cases to have 21.5% (fold change 1.27; p = 0.0036) higher expression than controls. DNA methylation and RNA expression were negatively correlated at TMEM59 (Spearman correlation coefficient = −0.274, p = 0.0083).
To further investigate the functional implications of the observed DNA methylation and gene expression differences of TMEM59, we measured the protein levels by Western blot in the full set of 25 case brains and 25 control brains. No differences were observed for the full length 36 kDa protein (p = 0.68) (Fig. 6 C), but AD cases had reduced levels of a shorter protein that was also bound by the antibody specific for the C-terminus of TMEM59 (p = 0.040) (Fig. 6C, E). The quantity of shorter protein was associated with age (Fig. 6D).
We performed a genome-wide, semi-unbiased quantitative comparison of frontal cortex DNA methylation from age- and gender-matched LOAD cases and controls. The CpG site most strongly associated with case status was located in the promoter region of TMEM59, a gene recently implicated in AD pathogenesis . This gene is involved in the post-translational modification of AβPP and causes retention of AβPP in the Golgi apparatus . The magnitude of methylation difference at this site between cases and controls was very modest (7.3% difference using the Illumina array), but the direction of association was confirmed using an alternate method of DNA methylation detection (2.7% difference using pyrosequencing). In an expanded population including a higher number of younger cases and controls, an interaction between age and case status was detected. Thus, age modifies the association between disease status and methylation. There was not a primary association between case status and methylation when the younger population was included. In the original sample of LOAD cases and controls, TMEM59 DNA methylation levels corresponded to functional changes in TMEM59 gene expression. LOAD cases had lower methylation and higher expression of TMEM59 than control samples. No differences in the level of the full length TMEM59 protein were observed between cases and controls; however a smaller protein that bound the TMEM59 antibody was significantly higher in controls. This TMEM59 protein size pattern is consistent with the TMEM59 control protein lysate. The shorter protein may represent a proteolytic product of the full length protein. The shorter protein is approximately 17 kDa, which could also be consistent with translation of an alternative RNA transcript beginning at exon 5 of the TMEM59 gene. The observed differences in protein expression levels are consistent with epigenetic regulation. Further molecular research is needed to better understand the gene expression and protein regulation and potential role of DNA methylation at this site.
Well-studied genes known to be involved in AD pathogenesis or identified through GWAS for genetic association with LOAD were largely not associated with disease-specific DNA methylation differences in this study. A notable exception was PSEN1, which was modestly hypomethylated in LOAD cases. Consistent with previous human postmortem tissue studies, PSEN1 showed very low levels of methylation in our population . Here, LOAD cases had reduced DNA methylation that was associated with increased PSEN1 gene expression, suggesting the DNA methylation change may be functional at this site. Studies in mice and neuroblastoma cell lines demonstrate that PSEN1 gene expression is regulated by DNA methylation at specific promoter CpG sites and this regulation depends on B vitamin availability . The correlation between DNA methylation and gene expression in our study support the cell line and mouse model findings. The Illumina HumanMethylation27 BeadArray platform used in this study also allowed for discovery of novel gene associations with AD. For example, methylation change was observed within the DMR for the imprinted gene, DIRAS3. Genomic imprinting in AD is a potential mechanism to explain epidemiological parent-of-origin inheritance observations [45, 46].
Greater than 900 genes (6% of genes featured on the array) were differentially methylated by case status after adjusting for age and gender. Overall, the disease related methylation effect size (2.9%) was relatively modest, and the global methylation distributions of AD cases and controls were similar. Together these findings suggest that DNA methylation may play a role in LOAD and the individual effects at each CpG site may be subtle. The magnitude and absolute number of DNA methylation changes observed in this study are consistent with previous reports in the literature performed on candidate gene subsets. In a case-control study of prefrontal cortex DNA methylation of twelve genes, only two genes were associated with AD status and the differences in methylation were less than 10% .
Gene set enrichment analysis revealed key patterns in the identified set of disease associated CpG sites. First, gene ontology analysis showed hypermethylation of genes involved in transcription and DNA replication, while membrane transporters were hypomethylated. Second, hypermethylation was enriched at genes containing POU3F2 binding motifs. POU3F2 is a transcription factor critical in central nervous system development that regulates Nestin gene expression, a protein important for radial axon growth . Third, positional analysis showed hypomethylation with case status at 19q13 and hypermethylation at 19p13, cytogenic band regions genetically linked with AD [5, 40].
LOAD cases and control sample groups were similar with respect length of time in storage, but LOAD cases had shorter PMI then control samples (p = 0.0004). During PMI, samples may be exposed to damaging lower pH conditions and higher temperatures where enzymes may be active. PMI was not a significant confounder at the TMEM59 site, however, PMI may be a factor at other specific locations throughout the DNA methylome.
This study measured genome-wide DNA methylation differences between LOAD case and control subjects aged 69 to 95. Across this relatively short age range, DNA methylation was associated with age at over 2,400 CpG sites, representing more than 8% of the sites on the BeadArray. Both hyper- and hypomethylation was observed. Previous studies have observed global hypomethylation with age in the brain , but gene-specific studies of aging and methylation have noted varying patterns . These results further support age as an important covariate to consider in statistical models of DNA methylation in late life.
Age is a major factor in epigenetic change in the brain , potentially confounding or modifying disease specific associations. In a study of cerebral cortex DNA from gestation to 104 years of age, eight of fifty loci showed increases in methylation through late life and two sites presented changes suggestive of an acceleration of age-related change in a subset of samples with LOAD . Additional evidence supports increased age-dependent epigenetic drift with disease. In LOAD prefrontal cortex samples representing a 30 year age range, an age-specific epigenetic drift was more prominent among cases compared to controls. The average methylation in promoters of MTHFR and APOE increased by 6.8%, while control samples decreased with age by 10.6% .
Cultured cells can potentially have very different epigenetic profiles than primary cells as an artifact of growth in culture , and thus use of primary human frontal cortex tissue is a strength of this study. DNA methylation is brain region specific and greater differences have been seen between the cerebellum and cortex regions than by gender, age, PMI, race, or cause of death . This study consistently used frontal cortex tissue because of its role in advanced AD. As with many studies of epigenomic, transcriptomic, or proteomic changes in the human brain, the tissue samples represented populations of mixed cell types, an important limitation, which may have resulted in attenuated effects. Epigenomes are cell type specific [15, 51], and brain cell types have different roles in AD . The AD brain has an active changing cell population including neuronal loss and glial activation  that may in part be responsible for the observed results. DNA methylation data, however, was not enriched in inflammatory mediators, which would have supported changes due to gliosis. This study considered brain region specific methylation and as epigenomic platforms require lower input DNA amounts, future research may be able to also consider cell type specific changes.
Results from large DNA methylome and transcriptome maps of the human brain suggest that intragenic CpG sites rather than promoter CpG islands may better correlate with transcription . The genome-wide sites included on the Illumina Infinium HumanMethylation27 BeadArray are more likely to be located within promoter region CpG islands. Important methylation events located elsewhere throughout the genome would be missed using this platform and may be included in future research utilizing alternative technologies.
These results must be interpreted with caution because this study had a small sample size relative to the large number of CpG site comparisons and the magnitude of observed methylation differences between LOAD cases and controls was moderate. The results from the array were technically validated at the top CpG site, but it is not clear whether this observation will be consistent across populations. While a small study, we identified modest DNA methylation differences as a potential event in LOAD.
We would like to thank Cheryl Teboe from the Data Core at the Michigan Alzheimer’s Disease Center (MADC) for providing the clinical data on the samples used in this study. Lisa Bain from the MADC Neuropathology Core provided the tissue samples used in this study. Susan Dagenais at the University of Michigan DNA Sequencing Core (PI: Robert Lyons) ran the Illumina BeadArray for this study. This study was supported by an anonymous gift to the MADC, the University of Michigan NIEHS Center (P30ES017885), and the Michigan CTSA (NIH grant number UL1RR024986). This analysis was supported by the NIA T32 Interdisciplinary Studies in Public Health and Aging (T32AG027708).
Authors’ disclosures available online (http://www.j-alz.com/disclosures/view.php?id=1109).