Promoter DNA Hypermethylation -- Implications for Alzheimer’s Disease

Abstract

Recent methylome-wide association studies (MWAS) in humans have solidified the concept that aberrant DNA methylation is associated with Alzheimer’s disease (AD). We summarize these findings to improve the understanding of mechanisms governing DNA methylation pertinent to transcriptional regulation, with an emphasis of AD-associated promoter DNA hypermethylation, which establishes an epigenetic barrier for transcriptional activation. By considering brain cell type specific expression profiles that have been published only for non-demented individuals, we detail functional activities of selected neuron, microglia, and astrocyte-enriched genes (AGAP2, DUSP6 and GPR37L1, respectively), which are DNA hypermethylated at promoters in AD. We highlight future directions in MWAS including experimental confirmation, functional relevance to AD, cell type-specific temporal characterization, and mechanism investigation.

Introduction of Alzheimer’s Disease

Alzheimer’s disease (AD) is a progressive neurodegenerative disease that impairs cognitive functions resulting in short-term memory loss and confusion, mood, psychological and behavioral alterations, appetite decline and inability to coordinate muscle movements. AD is incurable, and leads to secondary complications, ultimately causing death commonly due to pneumonia, pulmonary embolism, or cardiovascular disease. The pathology of AD is characterized by the presence of extracellular amyloid-β (Aβ) plaques, and neurofibrillary tangles in the soma and neuropil threads in neurites composed of hyperphosphorylated tau protein [1] The characteristic cerebral atrophy observed in AD is due to neuronal loss [2], and protein aggregates and cellular debris trigger activation of microglia and astrocytes in an innate immune response cascade in AD response [3] β-Amyloid accumulation in the brain is detectable decades before AD symptoms, whereas Tau deposits occur proximal to the manifestation [4-6] Further evidence suggests that extracellular oligomeric Aβ indirectly promotes tau accumulation by induction of de novo tau synthesis and redistribution of tau from the axon to the soma [1] Live imaging of brains has identified entorhinal cortex of the hippocampal formation and frontal lobes of the neocortex most susceptible to plaque formation [7], whereas the cerebellum remains refractory to plaque or fibrillary tangle build-up, even in extensively plaque deposited brains [8].

5–10% of AD cases are diagnosed under age 65 [9], and termed early onset AD (EOAD). Many EOAD patients show strong familial inheritance (familial AD) with dominant mutations in APP (amyloid β-protein precursor) [10], PSEN1 (presenilin 1, encoding PS1) [11,12], or PSEN2 (presenilin 2, encoding PS2) [13,14], which all converge on Aβ genesis pathway. APP is sequentially proteolyzed by β- and γ-secretase to produce Aβ peptides (36–43 amino acids) [15]. The PS1/2 homologs form the catalytic subunit of γ-secretase [15], but are differentially trafficked to late endosome/lysosome and plasma membrane via recycling and early endosomes, respectively [16] AD-associated APP and PSEN mutations have been extensively studied to date. For example, a reported EOAD APP mutation, E682K (glutamine-to- lysine substitution at codon 682) residing in an alternative beta-secretase 1 (BACE1) cleavage site, blocks alternative proteolysis and thus enforces canonical cleavage by BACE1, leading to amyloidosis [17]. Additionally, PSENs harbouring familial AD mutations preferentially yield the aggregation prone Aβ42 relative to the soluble form Aβ40 [16]. Herein, these coherently support the amyloid cascade model that suggests aberrant Aβ homeostasis (Aβ deposition and/or clearance) underlying AD pathogenesis [15].

Late onset AD (LOAD) is largely sporadic, and diagnosed over age 65, but otherwise clinically resembles EOAD. LOAD and sporadic EOAD patients frequently carry the E4 allele in apolipoprotein E (APOE) [18]. Interestingly, APOE in plasma and the central nervous system (CNS) represent independent pools of APOE, unable to cross the brain blood barrier. APOE in the CNS is predominantly expressed by astrocytes and also by microglia and neurons under pathological stress and distributes phospholipids and cholesterol for cellular repairing and membrane remodelling [19]. Many other risk modifying variants of low phenotypic penetrance have been detected in genetic studies, and are located in or close to genes implicated in cholesterol metabolism, immune response and endolysosomal pathways [9,15,20].

Notwithstanding significant genetic underpinning of AD, between a third and a half of AD cases are not attributable to heritable genetic predispositions [21]. This is exemplified by reports of monozygotic twin discordant for AD [22,23] and age of onset [24,25]. Divergent phenotypic manifestation without altering DNA sequence is associated with the regulation of gene expression (i.e. epigenetics) via mechanisms of DNA methylation, post-translational histone modifications, chromatin remodelling (i.e. targeted alteration of DNA-histone interactions in certain nucleosome), histone variants, and/or noncoding RNAs [26], and depending on the cellular context, these epigenetic modifications can be activating and/or repressive for gene expression. Both stable epigenetic modifications that persist over generations and transient chromatin modifications that are reversibly and dynamically deposited in response to cellular stimuli have been reported [27]. Epigenetic establishment, maintenance and changes are controlled by modulators (in particular, sequence-specific DNA binding proteins, namely transcription factors) that are writers, erasers, readers and/or effectors of epigenetic modifications. Evidence has shown site-specific de-methylation of DNA [28] in the entorhinal cortex of human post-mortem AD brains, which strongly argues for a role of DNA methylation in AD pathogenesis [8,26,29,30]. However, DNA methylation shows cell type-specific profiles pertinent to distinct cellular functions, and DNA methylation encodes disparate regulatory signals, depending on the genomic locations. In this review, we leveraged published single cell RNA-seq data of healthy brains to facilitate deconvolution of cell type specificity of genes marked with promoter hypermethylation in AD, based on the premise that expression of these genes is highly enriched in the respective cell types of healthy brains, and promoter hypermethylation engenders transcriptional repression. This strategy enables us to focus on gene targets at finer granularity, and to hypothesize the significance of their loss of function to AD development. Conversely, deconvoluting cell type-specificity of promoter hypomethylation in AD that are transcriptionally repressed in parenchymal cells of healthy controls poses challenges, due to the current shortage of single-cell transcriptomics data in AD. Thus, we limit the scope of our review to promoter hypermethylation in AD.

Mechanisms governing DNA methylation status

In mammals, DNA methylation primarily occurs at the 5-carbon of the cytosine base in the form of 5-methylcytosine (5mC), which frequently resides in CpG dinucleotides. Methylated CpGs (mCpGs) are hot spots for spontaneous deamination to thymine, and evolutionarily conserved mCpGs are comparatively deficient and interspersed sparsely in the genome [31]. Clusters of CpGs are enriched in CpG islands (CGIs) that are approximately 1000 base pairs in length [32] and often overlap with the sites of transcription initiation (promoters). Close to 70% of promoters are estimated to carry a CGI [33], and CGI-rich promoters are linked to genes of housekeeping, tissue-specific and developmental functions [34,35]. Promoter 5mC is expected to be associated with transcriptional repression by recruiting mCpG-binding domain (MBD) proteins that prohibit gene expression with corepressor complexes [36], or the methyl group hindering binding of transcription factors [37] in the classical view. The mammalian MBD protein family comprises of MeCP2, MBD1, MBD2 (the mCpG-binding subunit of the MeCP1 complex), MBD3 and MBD4, and other than MBD3, all bind to mCpG-rich DNA without sequence specificity [38]. Of note, MBD proteins interact with a large number of partners, including a distinctive set of corepressor complexes. To name a few, MeCP2 binds with mSin3a corepressor complex that also deacetylases histone, leading to dense chromatin compaction [39], and it has also been reported that mutations that attenuate binding of MeCP2 to m5C and the NCoR/SMRT corepressor complex are associated with Rett syndrome (a developmental neurological disorder) [40]. MBD1 interacts with SETDB1, a methyltransferase (KMT) of H3K9, responsible for chromosome condensation during cell cycle progression [41]. Moreover, as subunits of the MeCP1 complex, MBD2 couples with the NuRD (nucleosome remodeling and deacetylase) or Mi-2 corepressor in transcriptional silencing [42,43]. Gene body methylation correlates with transcriptional activity, whereby DNA methylation suppresses cryptic promoter usage or alternative splicing at highly transcribed genes, thereby protecting transcriptional fidelity [44,45]. Furthermore, DNA is hypermethylated at repetitive (transposable elements and their remnants) and non-repetitive intergenic regions where genome integrity along with silencing of transposons or other deleterious elements is ensured [46]. Thus, DNA methylation is associated with repression of viral elements, genomic imprinting (parent-of-origin specific gene expression), X chromosome inactivation (dosage compensation), regulating tissue-specific gene expression, and genome stability [46].

5mC is established de novo or maintained in the daughter strand of hemimethylated DNA by the DNMT (DNA methyltransferase) enzymes, DNMT3A/DNMT3B or DNMT1 in human [27]. 5mC demethylation can occur through passive dilution during iterative DNA replication or active conversion of methyl group catalyzed by enzyme families like the ten-eleven translocation (TET) enzymes [47]. Genome-wide loss of mammalian 5mC is restricted to two developmental waves, namely preimplantation development and primordial germ cell (PGC; the precursor of germ cells) development and migration, interrupted by a phase of concerted de novo DNA methylation during embryonic cellular differentiation [48]. Locus-specific 5mC demethylation is tightly regulated for cell lineage specification and cell identity maturation in the developing CNS and adult brain activities. During the burst of neurogenesis from E11.5 up to E14.5, the promoter of the Gfap (glial fibrillary acidic protein, an intermediate filament predominantly expressed in glial cells) gene is hypermethylated in neural progenitor cells (NPCs) to impede astrogliogenesis [49]. During this time window, abrogation of Dnmt1 gives rise to Gfap promoter hypomethylation in differentiating daughter cells of the neuronal lineage [50,51]. Subsequent promoter demethylation in NPCs promotes glial cell lineage development [49], a switch coinciding with lower expression of Dnmt3a [51]. Postnatal decline of Gfap transcription correlates with restoration of Gfap promoter hypermethylation [49], and de-repression of Gfap gene transcription via promoter hypomethylation in Dnmt3a knockout mice leads to anomalous astroglial differentiation [52]. The biphasic 5mC at the Gfap promoter abolishes Stat3 binding [53]. In addition to influencing CNS development, dysregulation of DNA methylation is causally linked to central and peripheral neurodegeneration in hereditary sensory and autonomic neuropathy type 1 (HSAN1) [54]. The HSAN1-associated DNMT1 missense mutations are autosomal-dominant, causing misfolding, precocious degradation and mislocalization [54]. Intriguingly, despite normal recruitment to replication foci in the S phase of the cell cycle, mutant DNMT1 fails to bind heterochromatin upon entry to G2 phase [54]. Diffused and locus-specific DNA demethylation in HSAN causes 8% loss of genome-wide 5mC, concurrent with the onset of dementia [54].

Overview of DNA methylome associations with AD

We reviewed methylome-wide association studies (MWAS) in population-based AD and control samples (Table 1), which were assayed on the Illumina Infinium HumanMethylation27 or HumanMethylation450 BeadArrays (henceforth 27K or 450K arrays). These two arrays interrogate 27,578 and 485,577 genome-wide CpGs, respectively [55]. The protocol for these arrays involve the bisulfite conversion of unmethylated cytosine to uracil, and thereby provides base pair resolution of CpG methylation status.

Table 1.

Summary of the methylome-wide association studies (MWAS) in AD using Illumina methylation arrays

Article	Tissue*	Sample type	Sample size		MWAS	Illumina platform
			AD	Control
Bakulski et al. 2012	FC	Bulk-cell	12	12	AD-control	Illumina array	27K
Sanchez-Mut et al. 2014	HIP	Bulk-cell	15	5	Braak-stage	Illumina array	27K
De Jager et al. 2014	DLPFC	Bulk-cell	708		Plaque burden	Illumina array	450K
Lunnon et al. 2014	EC, STG, PFC, CER, blood	Bulk-cell	122		Braak-stage	Illumina array	450K
Watson et al. 2016	STG	Bulk-cell	34	34	AD-control	Illumina array	450K
Smith et al 2018	STG	Bulk-cell	142		Braak-stage	Illumina array	450K
	PFC	Bulk-cell	144
Hernandez et al 2018	SFG	Neurons	14	18	AD-control	Illumina array	450K
Gasparoni et al 2018	OC	Neurons or glia	16	15	Braak-stage	Illumina array	450K
	FC	Bulk-cell	37	26
	TC	Bulk-cell	39	26
Mano et al 2017	ITG	Neurons	30	30	AD-control	Illumina array	450K

^*FC, frontal cortex; HIP, hippocampus; DLPFC, dorsolateral prefrontal cortex; EC, entorhinal cortex; STG, superior temporal gyrus; PFC, prefrontal cortex; CER, cerebellum; OC, occipital cortex; TC, temporal cortex; ITG, inferior temporal gyrus.

Bakulski et al [56] pioneered the early attempts of MWAS-based screening for differential DNA methylation changes in the frontal cortex (FC) of 12 LOAD patients and 12 cognitively normal controls. At a nominal P value cutoff of 0.05, they detected 948 differentially methylated CpG positions (DMPs) from 918 unique genes. Genes at the hypermethylated sites were enriched in transcription related molecular functions and biological processes while genes at the hypomethylated sites were associated with membrane transport and protein metabolism. Their top CpG site, with a 7.3% decrease in methylation in AD, was located upstream of TMEM59 (transmembrane protein 59), which encodes a protein responsible for post-translational glycosylation of APP and retention of APP in the Golgi apparatus[56]. TEME59 gene expression increased by more than 20% in LOAD cases compared to control samples. While there was no difference in the level of the full length TMEM59 protein between cases and controls, a shorter protein that bound the TMEM59 antibody was significantly higher in controls.

Sanchez-Mut et al [57] profiled DNA methylation in human hippocampus from 15 AD patients with varying Braak stages (I-II, III-IV, and V-VI) and 5 non-demented controls. By requiring a more than 25% methylation difference between Braak stages V-VI and controls, 4 CpG methylation sites from 3 genes were identified, including dual-specificity phosphatase 22 (DUSP22), claudin 15 (CLDN15), and quiescin Q6 sulfhydryl oxidase 1 (QSCN6). They verified that promoter hypermethylation of DUSP22 is associated with reduced RNA expression.

De Jager et al [58] examined DNA methylomic profiles associated with neuritic plaque burden in 708 dorsolateral prefrontal cortex (DLPFC) brain samples. At a Bonferroni correction P value threshold of 0.05, they identified 71 significant DMPs corresponding to 60 differentially methylated regions (DMRs), which comprise multiple consecutive DMPs, including two AD GWAS susceptibility loci, ABCA7and BIN1. To control for potential confounding effects resulting from the change in cell type composition in AD brains (loss of neurons) they included surrogate variables that captured the proportion of neurons in their analysis. 12 of these significant CpG sites were validated in data taken from a back-to-back publication of the Lunnon et al. study [59]. They further validated that these CpG were associated with altered RNA expression in the nearby genes ANK1, CDH23, DIP2A, RHBDF2, RPL13, RNF34, SERPINF1 and SERPINF2, in the temporal cortex region of an independent set of AD and control brains.

Lunnon et al [59] conducted a cross-tissue assessment of DNA methylomic changes associated with Braak staging in 4 brain regions, namely, entorhinal cortex (EC), superior temporal gyrus (STG), prefrontal cortex (PFC) and cerebellum (CER), in a cohort of 117 individuals by using the Illumina Infinium 450K arrays. Among the top 10 associated CpG loci identified in EC, two hypermethylated sites were located 91bp away from each other within the gene body of ankyrin 1 (ANK1). This hypermethylated region within ANK1 was also confirmed in the other two cortex brain regions, but not in CER. They found a significant association (P = 0.04) between the abundance of ANK1 isoform 5, 7 and 10 transcripts and AD-associated neuropathology. As methylation of ANK1 was also associated with amyloid plaques in De Jager et al [58], these results indicated a robust association of the ANK1 locus with AD-related neuropathology.

Watson et al [60] studied STG tissue from 34 LOAD patients and 34 matched controls. They identified 479 DMRs through a 1-kb sliding window analysis, about two-thirds of which were hypermethylated in AD. Genes overlapping with the DMRs were enriched for gene ontologies of neuron function and development, as well as cellular metabolism. Of the genes overlapped with the top 25 most significant DMRs, 8 had been previously reported in [58] and [59], including LOC100507547, PRDM16, PPT2, PPT2-EGFL8, PRRT1, C10orf105, CDH23 and RNF39.

Recently, Smith et al [61] analyzed DNA methylations associated with neuropathology of AD in PFC and STG samples from 147 individuals (87 middle- or late-stage AD patients and 60 controls). They identified 10 significant DMPs associated with Braak score, and 36 significant DMRs, including 6 hypermethylated DMRs within the HOXA gene cluster on chromosome 7, from upstream of the HOXA2 gene to the HOXA6 gene, most notably in the vicinity of HOXA3.

Since epigenetic modifications are cell type-specific, AD-associated methylation variations may be diluted in the aforementioned heterogeneous cell populations, in spite of imputation and controlling for cell type composition bias by the inclusion of in silico surrogate variables. To tackle these challenges, several recent studies performed cell type-specific AD-associated DNA methylation analyses.

Hernandez et al [62] applied laser-assisted microdissection to isolate pyramidal neurons from superior frontal gyrus (SFG) of 14 LOAD patients and 18 control samples. Adjusting for sex and age, 504 significant DMPs were reported, a majority (96.7%) of which were hypermethylated in AD brain derived pyramidal neurons. These loci were enriched for gene ontologies of T-tubule, axon, axonal growth cone and synapse, indicating epigenetic alterations of synapses and neurite arborization in LOAD. They also analyzed male-specific differential methylation analysis and identified 401 DMPs, nearly half of which overlapped with the results of all samples corrected for sex.

Two recent FACS (fluorescence-activated cell sorting) studies sorted neuronal and non-neuronal cells [63,64]. One study [63] isolated neuronal nuclei from inferior temporal gyrus (IFG) of postmortem brains in 30 AD patients and 30 age-matched normal controls. With multiple-test correction, no significant methylation deviation in AD was reported, probably due to small sample size. Nonetheless, 278 methylation sites attained mean methylation difference > 0.05 and P value < 0.05. 36 of the selected sites were clustered in 8 DMRs. 4 of the DMRs were located at the promoter CpG islands of BRCA1, ZNF714, AURKC, and LOC441666, while the other DMRs resided at intergenic regions or gene body of DUSP5P1, PCDHB7, SSR1, and ERICH1. The hypomethylation of BRCA1 promoter CpGs was correlated with increased RNA and protein expression of the gene in AD brains. A second study [64] conducted a bulk DNA methylation analysis in 52 control and 76 AD brains across frontal lobe (FL) and temporal cortex (TC), and reported no significant results. The authors isolated neuronal and glial nuclei from occipital cortex (OC) of 16 LOAD and 15 healthy brains [64]. Without specifying a genome-wide significance threshold, the top ranked DMPs were selected. Limited overlap was observed between neuronal and glia-specific Braak-associated DMPs, and 46 DMPs were common among the top 1000 DMPs of both cell types. Intriguingly, Braak-associated ANK1 hypermethylation was reported in glial cells but not in neurons.

Supplementary Table S1A-B shows the reported DMPs and DMRs from the above-stated MWAS. DMPs and DMRs are annotated for association with gene body or promoter regions. Promoter regions are defined as 1-kb flanking the transcription starting site (TSS) as in GENCODE gene annotation model GRCh37 release 19. For studies without significant DMPs or DMRs, we collected pertinent genes as shown in Supplementary Table S1C. At a P value threshold of 10⁻⁵, 94 promoter hyper-methylated genes and 24 hypo-methylated genes identified across the 9 studies were compiled in Supplementary Table S2. Promoter DNA hypermethylated genes were enriched for targets of transcription factors (AP4, NFAT and GATA_C) and regulators of cell proliferation, signalling transduction and stress response (Supplementary Table S3), while promoter DNA hypomethylated genes were enriched for LEF1 targets.

Since promoter CpG hypermethylation specifically correlates with transcriptional repression, by integrating published brain cell type-specific gene expression human data (Supplementary Table S2), which has only been published for cognitively normal patients [65,66], we aimed to assess the functions of genes in relevant cell types, which are likely to be repressed in AD, owing to promoter hypermethylation. In the following sections, we give a detailed account of the functional activities of selected neuron, microglia, and astrocyte-enriched genes (AGAP2, DUSP6 and GPR37L1, respectively) that are highly specifically expressed in the respective cell types, with MWAS evidence for promoter DNA hypermethylation in AD, suggesting transcriptional regulation modulated by DNA methylation in AD.

AD-associated promoter CpG hypermethylation at neuron-enriched AGAP2

AGAP2 (ArfGAP with GTPase domain, ankyrin repeat and PH domain 2) or PIKE (phosphoinositide 3-kinase enhancer) encoded by the gene AGAP2 in humans [67] is a Ras-like GTPase [68], present in both the activate guanosine triphosphate/GTP- or inactive guanosine diphosphate/GDP-bound state [69]. The exchange of GDP for GTP is stimulated by guanine nucleotide exchange factors (GEFs), and the intrinsic hydrolytic activity of GTP by GTPase is enhanced by GTPase activating proteins (GAPs) [69]. Phosphoinositides bind to the pleckstrin homology (PH) domain of PIKE [70,71], and regulate the subcellular localization and cellular activities of PIKE isoforms (PIKE-S, PIKE-L and PIKE-A) [72]. PIKE is upregulated in neurons, and almost silenced in glial cells. PIKE confers various neuroprotective effects, including anti-apoptosis, mediating long term potentiation (LTP, enhanced synaptic transmission upon frequent stimulation, insinuated in memory formation) and retrograde transport between endosome and trans Golgi network, as we will discuss.

PIKE-S is the first reported isoform, and is anchored to the nucleus for mediating phosphoinositide 3-kinase (PI3K)/protein kinase B (Akt) signalling [72]. In PC 12 cells, binding of nerve growth factor (NGF) to the tyrosine kinase receptor TrkA promotes PLC-γ1 relocalization from the cytoplasm to the nucleus, where it activates PIKE-S. PIKE-S interacts with and activates PI3K [72]. PI3K3 phosphorylates membrane-bound phospholipids phosphatidylinositol 4,5- bisphosphate (PIP2), generating phosphatidylinositol trisphosphate (PIP3), a secondary messenger and stimulator of downstream Akt [72]. Akt-catalyzed phosphorylation of target proteins including nucleoplasmin (a histone chaperone regulating chromosome decondensation in apoptosis [73]) confers resistance to apoptosis [72].

PIKE-L is densely localized in the postsynaptic density (PD) (a protein-rich layer coating the inner plasma membrane conspicuous under an electron microscope [74]) of excitatory neurons [75]. PIKE-L interacts with the scaffold protein Homer 1 (Homer protein homolog 1), mediating metabotropic glutamate receptor 1 (mG1uR1) signalling to PI3K activation, safeguarding neurons from excitotoxicity [75]. Additionally, PIKE-L inducibly binds to the GluA2 subunit of AMPAR (α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor, a class of ionotropic glutamate receptors), when PI3K-GluA2 association is engendered by NMDAR (N-methyl D-aspartate receptor, another class of ionotropic glutamate receptors) mediated Ca²⁺ influx [76]. The GluA2- PIKE-L axis formation further augments PI3K interaction with AMPAR, which is required for AMPAR insertion and retention on the postsynaptic membrane in LTP [76] In support of this model, PIKE-L attaches to GRIP1 (glutamate receptor-interacting protein 1) in the AMAPR- associated protein complex, and GRIP1 is a known interacting partner of GluA2 and kinesin (a microtubule motor protein) [77], critical for AMPAR trafficking to the plasma membrane [76]. PIKE knockout in mice disrupts NMDAR-dependent PI3K-AMAPR association, and subsequently attenuates AMPAR expression in LTP [76].

PIKE-A is widely detected in tumors (glioblastoma, astrocytomas [78] and others [79]), and lacks the N-terminal proline-rich domain, which is vital for interacting with SH3 domain containing proteins including PI3K [67]. Ablating PIKE-A weakens Akt activity, suggesting that PIKE-A may directly regulate Akt [79]. Additionally, PIKE-A co-localizes with AP-1 (a clathrin adaptor protein coating endosomes in the retrograde pathway [80,81]) and Rab4 (a small GTPase), which mark fast recycling endosomes [82]. Overexpression of PIKE-A shifts AP-1 from a perinuclear location (presumably trans Golgi network) to cytoplasmic punctate structures (presumably endosomes), and accelerates cargo recycling [82]. Ablation of PIKE-A impairs cargo sorting in the early endosome [83].

Crucially, brain derived neurotrophic factor (BDNF) stimulates PI3K/Akt signalling imparted through PIKE [84-86], and is endowed with neuroprotective effects for Aβ induced toxicity [85,86]. BDNF expression is reduced in the entorhinal cortex of AD patients [87], and PIKE is required for the neuroprotective effects of BDNF [88]. PIKE deficient mice exhibit compromised locomotion and spatial navigation [88]. Collectively, this supports the working model that aberrant repression of PIKE transcription through promoter DNA hypermethylation attenuates survival signalling activities in distressed neurons in AD.

AD-associated promoter CpG hypermethylation of microglia-enriched DUSP6

DUSP6 (dual specificity phosphatase 6) is a cytoplasmic enriched phosphatase, dephosphorylating threonine (abbreviated as T) and tyrosine (abbreviated as Y) residues in the T-E(abbreviation for glutamic acid)-Y triad [89] on the activation loop of ERK1/2 (extracellular signal-regulated kinases 1/2). ERK1/2 (critical for tau phosphorylation and Aβ processing in AD [90]) is one of the four major classes of mammalian mitogen-activated protein kinases (MAPKs) [91], which in addition include c-Jun amino (N)-terminal kinases 1-3 (JNK1-3, implicated in Aβ, protein misfolding or other cellular stress and cellular death signalling in AD [92]), p38 isoforms (α-δ, mediating chronic inflammation in AD [93]), and ERK5 (regulating nerve growth factor-effectuated neuroprotection [94]). MAPK is activated by MAPK kinase (MAPKK, MAP2K, MEK or MKK)-mediated phosphorylation, and phospho-activation of MAPKK is catalyzed by MAPK kinase kinase (MAPKKK, MAP3K, MEKK or MKKK) in response to signalling of receptor tyrosine kinases or G protein-coupled receptors (GPCRs) [91]. This 3-tier kinase signalling cassette is conserved in MAPK pathways [91]. MAPKs target copious sets of substrates, including transcription factors and MAPK-activated protein kinases (MAPKAPKs) among others [95]. Deactivation of MAPK is coordinated by type 1/2 serine/threonine phosphatases, protein tyrosine phosphatases and DUSPs [96]. Mitogen-activated protein kinase phosphatases (MKPs), the negative regulators of MAPKs that DUSP6 belongs to, are subfamily of DUSPs, which based on sequence homology and functional domains, are additionally classified into Cdc14 phosphatases (instructing mitotic exit [97]), PTENs (phosphatase and tensin homologs, tumor suppressors [98]), myotubularins (lipid phosphatase [99]), and atypical DUSPs [100]. MKPs are further categorized into inducible nuclear localized DUSP1, 2, 4 and 5, cytoplasmic enriched ERK1/2 specific DUSP6, 7 and 9, and JNK/p38 specific DUSP8, 10 and 16 [96].

Previously, transcriptional silencing of DUSP6 due to promoter hypermethylation was reported in pancreatic cancer [101]. Recently, DUSP6 was identified as a hub gene in the female-specific major depressive disorder co-expression network, andDUSP6 was downregulated in ventromedial PFC of female patients with major depressive disorder and female mice with distress [102].

DUSP6 is constitutively expressed in resting microglia, when phospho-ERK1/2 is low, and transiently downregulated with the culmination of ERK1/2 phosphorylation in LPS (lipopolysaccharide)-activated microglia, whilst total ERK1/2 abundance remains stable [103]. Expression of DUSP7 behaves similarly to DUSP6 in LPS-induced microglial activation, and inducible expression of DUSP1, 2, 4, and 5 oscillates in positive correlation with ERK1/2 phosphorylation [103]. Inhibition of ERK1/2 phosphorylation or an upstream ERK1/2 activator (MEK1) attenuated baseline, depressed or subsequently restored DUSP6 expression during the course of microglia activation and ensuing recovery [103]. Inhibition of DUSP6 augmented phosphorylated ERK1/2 (phospho-ERK1/2) in homeostatic microglia, but inconsistently, and microglia pretreated with a DUSP6 inhibitor were unable to elicit a full phospho-ERK1/2 response upon stimulation [103], suggesting the operation of a compensatory mechanism in dephosphorylation of ERK1/2 during the activation phase. We speculate that persistent AD-associated DNA hypermethylation at the DUSP6 promoter represses transcription, and in turn perturbs microglial homeostasis by dysregulation of phospho-ERK1/2. In corroborating with our speculation, DUSP6 was significantly downregulated in disease associated microglia of 5XFAD mice compared to homeostatic microglia (Fold-change from DAM to homeostatic microglia: - 1.397092; p: 10^-6.536861). Furthermore, Dusp6 is a PU.1 target gene [124], as reported in ChIP-seq analyses of PU.1 in BV2 cells (the rodent microglia cell line). PU.1 has been known as a master transcription factor inducing the myeloid lineages [104], and recruits TET2 and DNMT3B to hypomethylated and hypermethylated genes in monocyte-to-osteoclast differentiation [105]. Decreased PU.1 expression is associated with delayed onset of AD in humans [106]. Hence, the data collectively supports the plausibility that PU.1-dependent TET2 and DNMT3B recruitments may be perturbed in AD, leaded to the increased DNA methylation at the DUSP6 promoter.

AD-associated promoter CpG hypermethylation at astrocyte-enriched GPR37L1

The G-protein coupled receptor 37-like 1 (GPR37L1) and its closest homologue GPR37 are receptors for secreted prosaposin (neuro- and glial-protective [107]) and prosaptides (neuroactive peptide fragments of prosaposin) [108], albeit an early study characterized full-length GPR37L1 as constitutively active [109]. The amino terminus of GPR37L1, which is exposed to metalloprotease cleavage, is however indispensable for receptor activity [109]. Expression and secretion of the GPR37L1 ligand, prosaposin, is elevated by ischemia [110-112] and cellular stress [113]. Endocytosis of prosaposin is regulated by LRP1 (low density lipoprotein receptor-related protein-1),an APOE receptor expressed in neurons, glia and endothelial cells. LRP1 influences APP processing and therefore Aβ metabolism, and is implicated in AD susceptibility [114].

Gpr37l1 is expressed from postnatal day 8 in mice [115], with sustained expression throughout adulthood by astrocytes and some populations of oligodendrocyte precursor cells in the hippocampus, cerebral cortex, corpus callosum [115] and cerebellum [116]. Gpr37, however, is expressed in mature oligodendrocytes in hypothalamus, thalamus and corpus callosum [115], and Gpr37 is one of the substrates for Parkin, an E3 ubiquitin ligase implicated in pathogenesis of Parkinson’s disease [117]. Overexpression of Gpr37l1 was detected in glioma [118,119], and downregulation of Gpr37l1 was observed in medulloblastoma [120,121]. Complete deletion of Gpr37l1 exacerbates seizure susceptibility [122], and is associated with aberrant cerebellar development [123,124]. However, another study did not report precocious cerebellar development or seizures in Gpr37l1 null mice, and the loss of Gpr37l1 also did not impact on the number of astrocytes, oligodendrocytes, or excitatory/inhibitory synapses [115]. Elimination of Gpr37l1 had little effect on resting potentials of membranes in neurons or astrocytes [115]. GPR37L1 activation couples with Gɑs-mediated ERK1/2 signalling [108], and in Bergmann glia cells, interacts with patched 1, a receptor for sonic hedgehog (Shh) in modulating Shh pathway [124]. Glutamate uptake into astrocytes was suppressed by the prosaptide-Gpr37l1 ligand-receptor interaction [115]. In neurons, where Gpr37l1 is absent, prolonged response associated with repeated activation of the glutamate receptor NMDAR was attenuated with Gpr37l1 stimulation in murine brain slices [115]. The astrocytic Gpr37l1 crosstalk with neuronal NMDAR is not mediated by gliotransmitters released by astrocytes -- D-aspartate (a co-agonist for NMDAR) or tumour necrosis factor alpha (TNF-ɑ, a modulator of NMDAR activity) [115]. Nevertheless, astrocytic Gpr37l1 is upregulated in ischaemia, and it rescues hippocampal pyramidal neurons from cellular death [115]. This evidence suggests that astrocytic receptor Gpr37l1 regulates glutamate homeostasis, protecting neurons from excitotoxicity -- neuropathology documented in AD [125].

Perspective

The methylation arrays leveraged by current AD MWAS studies extensively cover promoter CGIs. For example, the 450K array probes 96 % of the CGIs, 92 % of the CGI shores (2 kb adjacent to CGIs) and 86 % of the CGI shelves (2-4 kb flanking CGIs) [75]. However, an unbiased genome-wide integration of distal or intergenic CpG sites at the comparable resolution relies on whole-genome bisulfite sequencing (for example, MethylC-seq in the bulk of cells [76]). Through studies of healthy brains, DNA methylome sequence data revealed tissue specificity [148] attributed to DNA methylation reprogramming during late-fetal and early postnatal periods [149] also dynamic in response to brain plasticity [150]. Comparable experimental approaches have not been applied to analyzing differential DNA methylation in AD.

It is also clear from this update that one of the outstanding knowledge gaps is locus-specific confirmation of promoter DNA hypermethylation at many of these candidate genes and at AD GWAS loci in AD cohorts, for example through methylation-specific polymerase chain reaction (PCR) [126]. Furthermore, differential methylation analyses in AD could notably benefit from knowledge of AD-associated DNA methylation alterations in relevant cell types or single cell methylation profiling [127]. Indeed, human brain cell types (for example, neurons and oligodendrocytes of the dorsolateral prefrontal cortex) show differential methylation profiles [128], and additionally, 21 subpopulations of prefrontal cortex neurons with distinct DNA methylation status have been reported [129]. Thus, single cell methylation profiling of AD brains could help tease out cell-specific methylation signatures.

The functional relevance of such promoter DNA hypermethylation to AD pathogenicity also remains unresolved, whether perturbations to promoter DNA methylation indeed alter AD phenotypic readouts, and at what time frame the switch of DNA methylation status occurs is pertinent to AD development. Do these changes represent an end-stage reflection of disease pathogenesis or more interestingly, an early response to cellular stress induced by Aβ? Some studies aimed to address these questions. One of the most comprehensively studied loci is the AD-specific BRCA1 promoter hypomethylation [63]. Promoter hypomethylation of BRCA1 was consistently observed in the 450K array and pyrosequencing data of neurons. Immunohistochemical staining showed BRCA1 localization coincided with advanced AD pathology [63]. Cytoplasmic BRCA1 is insoluble and co-aggregated with phosphorylated Tau in the neurons of the hippocampal CA1 region and entorhinal cortex, whereas in the occipital lobe or the cerebellum, BRCA1 is localized in the nucleus [63]. The study [63] showed that Aβ accumulation confers BRCA1 upregulation in vitro and in vivo, and induces DNA double-strand break, which is efficiently repaired by nuclear BRCA1. Additional insoluble Tau that is specific to advanced AD sequesters BRCA1 to the cytoplasm, compromising genome integrity [63]. Additionally, knockdown of BRCA1 in neurons with Aβ burden reduced short neurites and dendritic spines [63], consistent with AD brain pathology of humans and mice [130,131].

AD and a few other neurodegenerative disorders including Parkinson’s disease (PD), dementia with Lewy bodies (DLB) and frontotemporal dementia (FTD) are common in clinical manifestation as dementia (cognitive impairments and loss of independent functions), with shared pathological features of Aβ plaques, tauopathies (tau hyperphosphorylation or loss of function), and/or α-synuclein inclusions [132]. Common forms of dementia are also similar in global DNA methylation landscape [133]. At the gene level, the consensus data is sparse. The ANK1 gene is currently the well-studied example of locus-specific DNA hypermethylation in the entorhinal cortex consistent across AD, PD and Huntington’s disease, and only in brains of DLB or vascular dementia that also show AD pathology [134]. However, this DNA hypermethylation site does not reside within the gene promoter (hg19 chr8:41519302 to 41519420) [134]. Conversely, notable divergent promoter DNA methylation and expressions of the PIN1 gene have been reported for AD and FTD [135].

Furthermore, aging is a major risk factor for LOAD, and mechanisms associated with aging are also perturbed in AD, including inflammation, aberrant autophagy, mitochondrial dysfunction, vascular dysregulation, epigenetic alterations, and synaptic defeats [136]. A globally reduction in 5mC was reported in the hippocampus of AD patients [137], with locus-specific DNA methylation alterations. Studies suggest that the age-related DNA methylation dynamics correlate with chronological age [138], and are programmable [139], supported by the observation that the age-associated methylation states are consistent across samples and tissues [140-142]. Pervasive loss of DNA methylation was also observed in aging, and relates to the loss of constitutive (repeat-rich) heterochromatin integrity [143,144], and focal hypomethylation in aging occurs at CG-poor regions juxtaposing genes [144,145]. Conversely, age-associated DNA hypermethylation is manifested at CGIs [146]. Hence, it is highly speculative that DNA methylation changes due to aging may contribute to AD methylation landscape, and focal DNA methylation aberration specific to AD may play a critical role in AD development.

Our knowledge of DNA methylation changes in AD remains limited in part due to small sample sizes and in part due to the relative lack of single cell data. These considerations call for cost effective throughput approaches for population brain samples with cell type specificity, to pave the way for AD methylome compendium. Based on the data reviewed herein, the role of promoter DNA hypomethylation, intergenic and gene body DNA methylation aberrance and crosstalks between DNA methylation and other modes of epigenetic regulation in relation to AD pathogenesis warrants ongoing investigation and discussion.

Highlights.

• Hypermethylation dominates promoter DNA alterations in AD (94 of 118 genes).

• Promoter hypermethylated genes are enriched for targets of transcription factors (AP4, NFAT and GATA_C) and regulators of cell proliferation, signalling transduction and stress response.

• We integrated brain cell type-specific expression with promoter DNA hypermethylation, and discussed functions for neuron-, microglia- and astrocyte-specific candidate genes including AGAP2, DUSP6 and GPR37L1.