Epigenetic dysregulation of enhancers in neurons is associated with Alzheimer's disease pathology and cognitive symptoms

Abstract

Epigenetic control of enhancers alters neuronal functions and may be involved in Alzheimer's disease (AD). Here, we identify enhancers in neurons contributing to AD by comprehensive fine-mapping of DNA methylation at enhancers, genome-wide. We examine 1.2 million CpG and CpH sites in enhancers in prefrontal cortex neurons of individuals with no/mild, moderate, and severe AD pathology (n = 101). We identify 1224 differentially methylated enhancer regions; most of which are hypomethylated at CpH sites in AD neurons. CpH methylation losses occur in normal aging neurons, but are accelerated in AD. Integration of epigenetic and transcriptomic data demonstrates a pro-apoptotic reactivation of the cell cycle in post-mitotic AD neurons. Furthermore, AD neurons have a large cluster of significantly hypomethylated enhancers in the DSCAML1 gene that targets BACE1. Hypomethylation of these enhancers in AD is associated with an upregulation of BACE1 transcripts and an increase in amyloid plaques, neurofibrillary tangles, and cognitive decline.

Fig. 1

Fine-mapping of DNA methylation changes at enhancers, genome-wide, in AD neurons. In prefrontal cortex neurons of individuals with no/mild, moderate, and severe AD neurofibrillary tangle pathology (Braak stage 1–6; n = 101 individuals), DNA methylation was profiled at enhancer and promoter regions across the genome using bisulfite padlock probe sequencing. a Manhattan plot showing differentially methylated regions associated with the severity of neurofibrillary tangle pathology, after controlling for sex, age, postmortem interval, and neuronal subtype proportion. −log₁₀(P) refers to the significance of differentially methylated cytosine enrichment at enhancers, with the sign corresponding to the direction of methylation change (hyper or hypomethylated). Threshold for genome-wide significance (dashed line) is q < 0.05. Enhancers in DSCAML1 highlighted in red. b Image of ETS motif, the most strongly overrepresented TF motif within the differentially methylated enhancers in AD (p < 10⁻¹⁸, binomial distribution model in oPOSSUM). Motif sequence logo is provided by MotifMap. c Contribution of CpG and CpH sites to DNA methylation differences at enhancers in AD. Motif analysis depicts a predominant contribution of CpH sites, especially CpA. d DNA methylation changes at enhancer regions in the DSCAML1 gene in neurons that correspond to the severity of AD pathology. Dashed red line is genome-wide significance threshold (q < 0.05). Red dots denote enhancers that were also differentially methylated before the arrival of neurofibrillary tangles in the prefrontal cortex (Braak 1–4). Tracks for neuronal (NeuN + ) H3K27ac and H3K4me3, from PsychENCODE (pink, n = 9 individuals) are shown. e Averaged % DNA methylation across Braak stage for the most significantly disrupted enhancer region in DSCAML1 (chr11: 117,504,514–117,506,898) in AD neurons. F = female; M = male. *q < 10⁻¹⁴, robust linear regression model followed by hypergeometric test. The boxplot center line is the median, the lower and upper limits are the first and third quartiles (25th and 75th percentiles), and the whiskers are 1.5× the interquartile range

Fig. 2

Chromatin conformation analysis in cortical neurons identifying the gene targets of the differentially methylated enhancers in AD. Chromatin interactions, genome-wide, were determined using Hi-C data from human prefrontal cortex. a Circos plot showing significant interactions between differentially methylated enhancers in AD neurons and gene promoters (TSS ± 2 kb). AD enhancers acted in cis for all their target gene promoters (average 587 kb ± 37 kb enhancer–target promoter distance; n = 1224 enhancers). b, c Pathways affected by genes with differentially methylated enhancers in AD, as determined by MetaCore. Top ten GO processes and disease pathways show that in our unbiased, genome-wide analysis, the enhancers dysregulated in AD neurons affected genes involved in neurogenesis and amyloid neuropathies (q < 0.05, hypergeometric distribution; red dashed line). d Chromatin interactions of the enhancers in DSCAML1 that were differentially methylated in AD neurons (all interactions ± 400 kb of DSCAML1 shown). Enhancers in DSCAML1 targeted the promoters of BACE1 (shown in red) as well as other genes

Fig. 3

Omics analysis discovered high-confidence networks involved in AD. Using the FOREST OmicsIntegrator, we merged our DNA methylation sequencing dataset profiling enhancers in AD neurons and transcriptomic sequencing data generated from the no/mild, moderate, and severe AD prefrontal cortex (Braak stage 1–6). Molecular pathways affected in AD neurons, as determined by network analysis are shown, and highlight two major networks involved in pathological progression in AD neurons. Pathway analysis by MetaCore of these two networks show their primary involvement in cell cycle reentry and amyloid neuropathies, respectively (q < 0.001, hypergeometric distribution). Hub genes: UBC and CUL3; subnetwork genes connected to CUL3: APP, CTNNB1, PIK3R1, ESR1, HSP90AB1, and PPP2CA are shown in black

Fig. 4

DNA methylation at DSCAML1 intron 3 is linked to BACE1 gene expression and to the pathology and cognitive symptoms of AD. a Correlation of DNA methylation at a DSCAML1 CpG site (cg07533617) with BACE1 mRNA expression in the AD prefrontal cortex. DNA methylation status at cg07533617 (situated in DSCAML1 intron 3 at an enhancer interacting with the BACE1 gene promoter) was inversely correlated with BACE1 mRNA expression during the early stages of AD (Braak stage ≤ 4, n = 101 AD patients; p < 0.005, robust linear regression) but not in late stage (Braak stage ≥ 5, n = 76 AD patients; p = 0.21, robust linear regression). Analysis performed using the ROSMAP dataset, which contained both transcriptomic data generated by RNA-seq and genome-wide DNA methylation data generated by 450K Illumina DNA methylation arrays from the same AD patients. Correlation between residuals of DNA methylation (beta values of probe cg07533617, x-axis) and BACE1 mRNA levels (FPKM, y-axis) shown. b Significance of correlation between DNA methylation at cg07533617 in DSCAML1 intron 3 and AD pathology/cognitive symptoms; determined using ROSMAP DNA methylation and pathological/clinical data (n = 465 AD and controls; 251 AD and 214 controls). DNA hypomethylation at cg07533617 in DSCAML1 intron 3 was significantly correlated with increased amyloid pathology and neurofibrillary tangle density, as well as a decline in episodic memory, perceptual speed, and global cognitive function (*p < 0.05, linear mixed model with annual cognitive measures as the longitudinal outcomes and DNA methylation as the predictor). a, b All analyses of ROSMAP data adjusted for sex, age, postmortem interval, years of education, neuronal cell proportion, as well as in a RIN

Fig. 5

Genetic variation within DSCAML1 intron 3 enhancers affects BACE1 mRNA levels in the prefrontal cortex. ROSMAP data containing AD patients and controls (n = 278 individuals) that had both genotype data (using genome-wide SNP arrays) and transcriptome analysis (by RNA-seq). The extended genomic area around BACE1 (±500 kb) is shown. Haplotypes (n = 53) determined by Haploview. Analysis of haplotype association with BACE1 expression that controlled for age, sex, postmortem interval, Braak stage, years of education, RIN, and neuronal cell proportion was performed. Pink track shows the significance of the haplotype association with BACE1 expression (dashed red line is q < 0.05). BACE1 expression in the prefrontal cortex was influenced by two haplotypes in DSCAML1 (q < 10⁻⁴, robust linear regression model). These two haplotypes overlapped four enhancers determined to be epigenetically dysregulated in AD neurons, supporting their involvement in BACE1 regulation. Three haplotypes near the BACE1 TSS were also associated with BACE1 expression. Haplotype information (including SNP ID) and their population frequencies are shown for the five haplotypes significantly associated with BACE1 expression. D’ values, measure of linkage disequilibrium, are shown in the boxes, with red boxes indicative of a strong linkage disequilibrium between SNPs

Fig. 6

Accelerated CpH methylation changes with aging in AD neurons. Age-dependent changes in CpH methylation in neurons from individuals with no/mild, moderate, and severe AD pathology (Braak stage 1–2, 3–4, 5–6, and n = 38 individuals, 32, and 31 individuals, respectively). Age analysis was performed on CpH sites (n = 3661 CpHs) in enhancers relevant to AD (n = 848 enhancers showing both epigenetic and associated gene transcript differences in AD, as identified by Omics integration analysis). a Aging changes in % CpH methylation at enhancers in neurons. CpH methylation decreased with age in individuals with no/mild AD pathology (p < 0.005, robust linear regression model), but not in moderate or severe AD. Analysis of % CpH methylation changes with age adjusted for sex, postmortem interval, and neuronal proportion. b Box plot showing the difference between the epigenetic age and chronological age in neurons of moderate and severe AD cases. The epigenetic aging calculator used CpH sites to determine the epigenetic ages of the moderate and severe AD groups. CpH methylation aging was accelerated in severe AD neurons (p < 0.05, paired t-test). The boxplot center line is the median, the lower and upper limits are the first and third quartiles (25th and 75th percentiles), and the whiskers are 1.5× the interquartile range. c CpH methylation age relative to chronological age in neurons of severe AD cases (Braak stage 5–6). Each dot represents a severe AD case, and the line demarks where chronological age matches epigenetic age. Purple dots are samples in which epigenetic age is greater than chronological age. Severe AD cases typically show an epigenetic age of 80 years or older regardless of chronological age

Fig. 7

Proposed model displaying the contribution of enhancer dysregulation to the development and progression of AD. In aging, neurons gradually lose CpH methylation at enhancers, especially at enhancers regulating genes involved in the early cell cycle. Aging and other AD risk factors that trigger improper attempts of neurons to renter the cell cycle promotes the formation of neurofibrillary tangles in the AD brain and is neurotoxic. In AD neurons there is also a hypomethylation of enhancers in DSCAML1 that occurs early in disease. Enhancers in DSCAML1 activate BACE1 expression, which leads to the increased production of the Aβ peptides. Aβ peptides promotes an accelerated loss of DNA methylation at enhancers in AD neurons, thereby facilitating aberrant cell cycle reactivation. Aβ peptides also forms plaques, which fuels the spread of neurofibrillary tangle pathology across the brain. Consequently, epigenetic changes at enhancers in AD neurons contributes to the pathology, neurodegeneration, and cognitive symptoms of AD