DNA methylation in demyelinated multiple sclerosis hippocampus

Multiple Sclerosis (MS) is an immune-mediated demyelinating disease of the human central nervous system (CNS). Memory impairments and hippocampal demyelination are common features in MS patients. Our previous data have shown that demyelination alters neuronal gene expression in the hippocampus. DNA methylation is a common epigenetic modifier of gene expression. In this study, we investigated whether DNA methylation is altered in MS hippocampus following demyelination. Our results show that mRNA levels of DNA methyltransferase were increased in demyelinated MS hippocampus, while de-methylation enzymes were decreased. Comparative methylation profiling identify hypo-methylation within upstream sequences of 6 genes and hyper-methylation of 10 genes in demyelinated MS hippocampus. Genes identified in the current study were also validated in an independent microarray dataset generated from MS hippocampus. Independent validation using RT-PCR revealed that DNA methylation inversely correlated with mRNA levels of the candidate genes. Queries across cell-specific databases revealed that a majority of the candidate genes are expressed by astrocytes and neurons in mouse and human CNS. Taken together, our results expands the list of genes previously identified in MS hippocampus and establish DNA methylation as a mechanism of altered gene expression in MS hippocampus.

Multiple sclerosis (MS) is an inflammatory, demyelinating, and neurodegenerative disease of the central nervous system (CNS) that affects more than two million people worldwide 1,2 . Among the spectrum of cognitive impairments, memory dysfunction is most common among MS patients 3,4 . Hippocampal demyelination is extensive in individuals with MS and modulates expression of neuronal genes involved in synaptic plasticity and memory function 5,6 .
Large-scale genome-wide association studies (GWAS) have identified MS susceptibility loci, including human leukocyte antigen loci and other immune-function related genes 7-10 ; however, their functional significance related to MS pathogenesis is still unknown. The relatively low concordance rate of single-nucleotide polymorphisms in monozygotic twins 11 , the presence of a strong gender bias, and the influence of migration on disease onset collectively suggest that the pathogenesis of MS likely results from a combination of both genetic and epigenetic factors 12,13 . Epigenetic modifications, including DNA methylation, histone modification, chromatin remodeling, and noncoding RNA regulation have been reported to regulate gene expression and to participate in the etiology of MS 14,15 .
DNA methylation occurs in special genomic regions called CpG islands, which contain greater than 50% cytosine and guanine nucleotides. It plays a major role in aberrant expression of genes that are important in several neurological diseases 16,17 as well as in memory formation and maintenance 18,19 . We previously compared and identified several genes and microRNAs in MS hippocampus that correlate with synaptic changes, memory dysfunction, and hippocampal demyelination 5,20 .
In this study, we investigated additional epigenetic mechanisms that alter gene expression in MS hippocampus. We found significant increases in mRNA levels of key DNA methyltransferase enzymes (DNMTs). Interestingly, the mRNA levels of the three DNA de-methylation enzymes (ten-eleven translocation methylcytosine dioxygenase 1-3; TET1-TET3), which catalyze hydroxy-methylation as well as the total level of hydroxy-methylated residues, were significantly decreased in MS demyelinated hippocampus. Several differentially methylated positions (DMPs) were also identified by comparing MS myelinated to demyelinated hippocampus. The methylation status of DMPs inversely correlated with mRNA levels of target genes that have been associated with neuronal survival and memory function. As methylation patterns in different cell types may contribute to overall methylation patterns 21 , human and mouse cell-specific databases indicated that a majority of the target genes were localized to neurons and astrocytes. Collectively, our results provide evidence that DNA methylation could play a major role in controlling gene expression in MS hippocampus.

Results
Demyelination correlates with changes in DNA methylation and de-methylation enzymes in MS hippocampus. In order to study DNA methylation, we measured levels of DNMT enzymes responsible for inserting and maintaining DNA methylation (DNMT1, DNMT3A, and DNMT3B) in myelinated and demyelinated MS hippocampi. Demyelination led to significant increases in mRNA levels of all three DNMTs in MS hippocampus (Fig. 1A). Immunohistochemical analysis showed that DNMT1 (Fig. 1B,C), DNMT3A (Fig. 1D,E) and DNMT3B (Fig. 1F,G) were primarily associated with hippocampal neurons in MS myelinated (Fig. 1B,D,F) and MS demyelinated (Fig. 1C,E,G) hippocampus.
In concert with DNMT enzymes inserting and maintaining methylation patterns, the TET family of de-methylation enzymes oxidizes methylcytosine (5 mC) to hydroxymethylcytosine (5 hmC) to remove methylation 22 . Three TET proteins (TET 1-3) have been identified in humans 23 , which all convert 5 mC to 5 hmC. Given the relationship between methylation and TET enzymes, we also quantified mRNA levels of TET1-3 in MS hippocampus ( Fig. 2A). The results showed a significant decrease in mRNA levels of all three TET genes in demyelinated hippocampi from individuals with MS. In order to test whether the lower mRNA levels of TET enzymes also correlated with levels of 5hmC residues, we measured the global levels of 5hmC within DNA isolated from MS myelinated and demyelinated hippocampi. The results showed that there was a significant decrease in the overall level of 5hmC within demyelinated hippocampus that correlated with the decreased mRNA levels of TET1-3 (Fig. 2B). We then conducted histochemical analysis to determine the cellular expression of 5hmC and we determined that neurons were the major cell type containing 5hmC in MS myelinated (Fig. 2C) and demyelinated hippocampus (Fig. 2D).

Loss of myelin correlates with DNA methylation changes within transcription start sites.
Imprinted gene expression is regulated by epigenetic mechanisms, particularly DNA methylation at DMPs. We did not detect any significant changes in overall 5mC levels in DNA from myelinated or demyelinated MS hippocampus (data not shown). We therefore used a more sensitive global array-based approach to identify DMPs by comparing myelinated (n = 8) and demyelinated (n = 7) MS hippocampus (Fig. 3A)     coverage in shore regions as well as 3091 probes at non-CpG sites 24 . Resultant methylation assays and global clustering showed that the methylation pattern could reliably differentiate between myelinated and demyelinated MS hippocampus (Fig. 3A). Demyelination resulted in identification of 144 DMPs (62 hypermethylated, 82 hypomethylated) within the assessed CpG sites (adjusted p < 0.05, delta β 20%; Fig. 3B). Localization of these DMPs in the context of genomic location was classified into different categories ( Fig. 3C): CpG islands, flanking CpG Islands (CGIs) (shores and shelves; 2-4 kb from CGIs), and open sea (not related to CpGs). We found 29 out of 62 (46.7%) DMPs detected within the CpG islands had increased methylation, whereas 26.8% (22 out of 82) of the DMPs within the CpG islands showed decreased methylation. We next asked whether these DMPs fall within the genomic location of well-characterized genes (UCSD genome browser hg119). The DMPs were mapped to either the '1 st exon' , 3′ and 5′ UTRs, or the 'body' (non-exonic) of the gene, or within 1500 bp of the transcription start site (TSS). The results showed that 92 out of 144 altered DMPs (65%) were mapped to 75 annotated genes (UCSD genome browser hg119). Several of the identified genes had multiple DMPs associated with their sequences (Table 2). To further validate our findings, we compared these identified genes with results obtained from an independent gene microarray database using MS hippocampus 5 . The results showed that a significant number of genes (43 out of 75) identified in the current study were also altered at the mRNA expression level in the previous study (gene symbols in bold letters, Tables 2 and 3).

Identification of cell types expressing transcripts containing DMPs in mouse CNS. After we
identified altered DMPs and their target genes in MS hippocampus, we compared the cellular expression of these target genes in mouse CNS using published mouse and human RNA-seq databases 25,26 . This comparison revealed that DMPs are associated with genes expressed by the four major cell types (astrocytes, neurons, oligodendrocytes, and microglia) in mouse and human CNS ( Table 2). Among identified hypomethylated targets (within UTR, gene body, and exon), 41% were specific to astrocytes, whereas 44% (13 out of 29) of the target genes were expressed by neurons in mouse CNS. The results also showed that 34% of the target genes were expressed by microglia/macrophage lineage cells in mouse (Table 2). Interestingly, hypomethylation identified 1 gene in mouse oligodendrocytes (F Box and WD repeat domain 8) and 4 in human oligodendrocytes (Small Nuclear Ribonucleoprotein 40 kDA; Bestrophin 3; Meis Homeobox 1; and Transmembrane Protein 165). None of the genes were expressed by both mouse and human oligodendrocytes. Among the hypermethylated transcripts, (within UTR, gene body, and exon), 26% of the genes were expressed by astrocytes and neurons (8 out of 30), while 20% were also expressed by microglia/macrophage cells in mouse brain ( Table 2). Four hypermethylated genes were oligodendrocyte-specific in mouse brain (McKusick-Kaufmna Syndrome; Inscuteable Homolog (Drosophila); Neurofascin; and Topoisomerase (DNA) I, Mitochondrial) and three (Chromosome 1 open reading frame 106; Neurofascin; and Shisa Family Member 2) were localized to human oligodendrocytes. Cellular analysis revealed that the hypermethylated, oligodendrocyte-specific Neurofascin 155, which is involved in maintenance of axoglial junctions 27 , is expressed by both mouse and human oligodendrocytes.
As the presence of DMPs within the promoter and TSS has the greatest possibility of affecting mRNA expression 14 , we identified the genes where a DMP was identified within 1500 bp of the TSS (Table 2). Our results showed that there were 14 DMPs (targeting 6 genes) that showed decreased methylation, while 10 DMPs (identified with 10 genes) showed increased methylation. Due to the presence of DMPs within 1500 bp of the TSS, we determined the cellular identification of the targets in both mouse and human RNA databases 25,26 . The comparative results identified several genes that are differentially expressed by mouse and human CNS cells ( Continued brain tissue is a heterogeneous mixture of several cell types, these results not only provide clues as to the cellular specificity of each of the target genes, but they also provide options for further genetic manipulations in mouse models. DNA methylation changes within transcription start sites of genes inversely correlate with mRNA levels of target genes. DNA methylation at the proximity of the TSS is generally considered to be a potent epigenetic modification that prohibits transcription factor (TF) recruitment, resulting in transcription suppression 28,29 . We therefore validated whether the presence of DMPs within 1500bp of the TSS affects mRNA levels in 4 hyper-methylated and 4 hypo-methylated target genes. The results showed (Fig. 4) that hypomethylation within the TSS1500 of AT-hook transcription factor (AKNA), Emopamil Binding Protein Like (EBPL), HECT domain, RCC1-like domain containing protein 6 (HERC6), and Secreted frizzled related protein 1 (SFRP1) led to significant increases in mRNA levels of these genes following demyelination. In contrast, hyper-methylation identified within the TSS1500 of Nescient helix-loop-helix 2 (NHLH2), Phospholipase C eta 1 (PLCH1), Transmembrane protein 132B (TMEM132B), and WD repeat domain 81 (WDR81) correlated with the decrease in mRNA levels of these genes following demyelination in MS hippocampus. Immunohistochemical analysis to confirm cellular localization, showed that SFRP1 (Fig. 4B,C), and PLCH1 (Fig. 4D,E) were primarily associated with hippocampal neurons in MS myelinated (Fig. 4B,D) and MS demyelinated (Fig. 4C,E) hippocampus.

Discussion
Epigenetic targets of gene regulation provide interesting targets for therapeutics because they are modifiable. The present study identifies DNA methylation as a correlate of demyelination and gene expression in MS hippocampus. Demyelination coincides with increases in mRNA levels of DNA methylation enzymes, with concomitant decreases in levels of DNA demethylation enzymes. Global methylation analysis identified 144 DMPs in MS hippocampus targeting several genes expressed by the major cell types in the human brain. As human tissue is not amenable to experimental manipulations, we also performed a comparative database search to determine the cellular specificity of the candidate genes in mouse brain. These data were further screened to identify the 25 DMPs localized within 1500 bp of TSS of 16 genes. Independent validation revealed that mRNA levels of the identified genes inversely correlated with DNA methylation status. Previous studies have shown that hippocampal demyelination is common in MS patients, which leads to loss of synaptic density and can influence expression of synaptic and neuronal genes as well as regulate the expression of neuronal miRNAs 5,20 . In this study, we identified candidate genes that are altered by methylation changes following demyelination in MS hippocampus and that may play a role in altering synaptic plasticity, memory performance, and neuronal survival in MS brains.   25,26 . To increase confidence in cellular identity, matching genes showing expression levels above the 50 th percentile were selected. Genes significantly altered in a comparison of MS hippocampus using previously published microarray-based analsis 5 are shown in bold.
The presence/loss of methylation within the TSS alters mRNA levels of the target genes. We identified 6 genes where demyelination led to decreased methylation. Among these, mRNA levels of AKNA were significantly increased following demyelination. AKNA is a major regulator of CD40 and CD40 ligand 30 and is reported in RNA seq databases to be expressed by microglia/macrophage cells (Table 3) in both mouse and human. Increases in microglial CD40 expression and interactions with its ligand CD40L have been shown to induce/stimulate the expression of tumor necrosis factor-alpha (TNF-α) and to initiate neuronal death 31 . Following demyelination, we detected a significant increase in methylation within the TSS of WDR81, a gene involved in neuronal survival 32 . Increased levels of AKNA in conjunction with decreased WDR81 following demyelination could therefore lead to increased levels of TNF-α and neuronal injury. In addition, we also identified hypomethylation and a significant increase in mRNA levels of SFRP1 following demyelination. SFRP1 is an inhibitor of the WNT signaling proteins, which regulate learning and memory as well as synaptic plasticity at central synapses 33,34 . Our comparative methylation analysis also identified hypermethylation (near the TSS) and decreased mRNA levels of NHLH2 and PLCH1 (Fig. 4). NHLH2 is a positive regulator of melanocortin receptors and modulates memory and learning 35,36 , while PLCH1 loss is generally associated with impaired working memory 37 . Differential methylation and concomitant increased levels of SFRP1 as well as decreased NHLH2 and PLCH1 could therefore lead to the decreased synaptic density and memory performance that have been previously reported following demyelination 5,20 .
The ongoing discovery of epigenetic factors underlying pathogenesis in MS is substantial 14 . These factors are relatively easily modifiable, in many cases utilizing readily available pharmacological agents targeting epigenetic modifiers. The study of DNA methylation in MS pathogenesis has been largely limited to comparisons of peripheral blood cells 14,15,38 . Using the same platform used in this study, 74 methylation sites (29 associated with either the MHC or the human leukocyte antigen (HLA-DRB1 region) were identified in a comparison between CD4+ T cells from patients with MS and healthy individuals 39 . In addition, significant differences in the methylation patterns associated with CD4 and CD8 T cells were detected in MS patients compared to healthy controls 40 . A genome-wide study of methylation in normal appearing white matter from MS patients identified 319  Table 3. Cellular Identity of hypo-and hyper-methylated CpGs near TSS in human and mouse. Genes where altered methylation was detected within 1500 bp were matched to corresponding cell types. Genes significantly altered in a comparison of MS hippocampus using previously published microarray-based analysis 5 are shown in bold. mRNA levels of genes measured using RT-PCR and presented in Figure 4 are italicized.
significantly hypermethylated and 220 significantly hypomethylated regions, which mapped to genes involved in oligodendrocyte survival and immune responses 41 . Our study is the first to provide insight into the landscape of methylation changes following demyelination in MS hippocampus. In addition, none of the genes identified in the current study was common to the candidates described in our previous collaborative study using MS normal appearing white matter 41 . This supports the need to develop datasets of gene changes from different gray matter regions in MS brains. Of great interest is the fact that the identified genes are expressed by different cell types in human and mouse brain. While similar DNA methylation studies in mouse are not currently possible due to the lack of a global mouse methylation profile chip, our species-specific cell identity provides the option to query, validate, and manipulate candidate genes for their possible role in memory function using animal models. Future tissue-, region-, and cell-specific analyses of epigenetic changes in MS brains need to be conducted in order to gain further insight into the pathogenesis of MS.

Materials and Methods
Human subjects and regulatory compliance. All brains were collected as part of the tissue procurement program approved by the Cleveland Clinic Institutional Review Board. Patient anonymity was strictly maintained and all tissue samples were handled in a coded fashion. All donors or their surrogates gave informed consent for their brains to be used for research studies. All experiments were carried out in accordance with the relevant Cleveland Clinic Institutional regulations and guidelines.
Tissue collection and characterization. Brains were removed according to a rapid autopsy protocol, sliced (1 cm thick), and then either fixed in 4% paraformaldehyde for morphological studies or rapidly frozen for biochemical analysis. Patient demographics are listed in Table 1. All hippocampi were characterized for demyelination by immunostaining using proteolipid protein (PLP) as described previously 5,20 . Briefly, frozen 30 µm sections were cut for characterization by immunohistochemistry and for assessment of demyelination. This was followed by collection of 3-4 subsequent sections for DNA isolation. No significant differences in disease duration (  environment using the ChAMP package, which integrates currently available 450 k analysis methods and also offers its own novel functionality 42 . After running basic quality control metrics, we performed a beta mixture quantile normalization method to adjust for bias introduced by the Infinium type 2 probe design 42 . All DNA methylation data files will be deposited at the GEO (https://www.ncbi.nlm.nih.gov/GEO) and can be accessed through the accession number GSE101658. As part of routine analysis of DNA methylation datasets in our study, the probes on the X chromosome were normalized for males and females separately and independently of autosomal probes because X chromosome inactivation causes significant gender differences in methylation patterns. In addition, to account for age-related DNA methylation changes, we employed the methods described by Horvath et al. 43 . Briefly, this multi-tissue predictor of age allows for the estimation of the DNA methylation age of most tissues and cell types. The predictor, which is freely available, was developed using 8,000 samples from 82 Illumina DNA methylation array datasets, encompassing 51 healthy tissues and cell types. Using these methods, we did not detect any differences in DNA methylation for predicted and actual age between MS samples. To determine the cellular identity of the methylation target, identified genes were queried against mouse and human cell-specific RNA sequencing databases 25,26 . To further validate the results and cellular specificity, we only selected genes that were above the 50 th percentile of expression level across all cell types.