Mapping DNA methylation signatures to identify epigenetic variation across subcortical regions of the human PTSD brain

Abstract

Background

Post-traumatic stress disorder is a mental disorder that may occur in the aftermath of severe psychological trauma. Epigenetic changes in the brain may play a critical role in understanding the neurobiology of PTSD by linking environmental traumatic stress exposure to lasting alterations in gene expression that shape neuronal function.

Methods

We examined 1,065,750 DNA methylation (DNAm) sites from 171 donors including neurotypicals, PTSD, and major depressive disorder cases across six regions implicated in the fear circuitry of the brain. We performed RNA-seq to examine changes in gene expression and link these changes to changes in DNAm at nearby sites in a case control manner. We created a single cell-type atlas of DNAm using a single nucleus RNA-seq reference panel to map epigenetic changes to specific cell types. Finally, we leveraged a human PTSD ketamine trial to associate blood DNAm biomarkers of ketamine efficacy with specific changes in DNAm in brain.

Results

We found significant differential methylation for PTSD near 195 genes and to further resolve the changes we observe, we constructed a cell type-specific DNA methylation atlas defined for changes to the PTSD methylome across 6 cell types. To identify potential therapeutic intersections for PTSD, we found significant methylation levels in the MAD1L1, ELFN1, and WNT5A genes in PTSD patients who responded to ketamine. Finally, to better understand the unique biology of PTSD, we analyzed matching methylation data for a cohort of MDD donors with no known history of trauma or PTSD.

Conclusions

Our results implicate DNAm as an epigenetic mechanism underlying the molecular changes associated with the subcortical fear circuitry of the PTSD brain.

INTRODUCTION

Post-traumatic stress disorder is a debilitating psychiatric disorder with an approximately 7% lifetime prevalence in the general population(1–3). PTSD typically emerges following extremely stressful life events, such as direct threats to one’s life. Particularly when chronic, it is often comorbid with other psychiatric diagnoses including major depression(4) and substance use disorder(5). Because the diagnostic criteria for PTSD specify that it is a lasting clinical condition arising from discrete environmental exposures (trauma), epigenetic mechanisms are very likely to contribute substantially to its pathophysiology.

Methylation of DNA (at both cytosine nucleotides (CpGs) and neuronal non-cytosine nucleotides) is a critical, epigenetic regulator of genome architecture, gene expression, and cell function(6). These processes are important for mammalian brain development(7), aging(8), disease(9–11), and response to external stimuli such as stress(12). Epigenetic changes evoked by stress are thus encoded into the genome and can serve as a link between the genetic architecture and the response (i.e., gene expression). In this way, epigenetic changes are a path through which traumatic stress and other environmental exposures influence PTSD vulnerability and resilience and converge mechanistically with underlying risk for PTSD(13). Elucidating this interplay is thus fundamental to advancing our understanding of the etiology and pathophysiology of PTSD. Previous studies of the PTSD methylome have predominantly examined peripheral tissues such as blood(14–17). These findings have largely centered on changes in inflammatory/immune response and glucocorticoid signaling. However, because epigenetic changes are tissue-, region-, and cell-specific, peripheral studies are limited in their ability to inform the understanding of brain epigenetic regulation of gene expression.

In this study, we focused on subregions of the amygdala and the hippocampus because these are among the most well-understood components of the brain’s fear circuitry with known functional engagement in PTSD(18). The amygdala is comprised of several nuclei with discrete functions including the basolateral nucleus (BLA), the central nucleus (CeA) and the medial nucleus (MeA). Neuroimaging and animal studies have found that the amygdala modulates the fear response(19–21) and recent studies have demonstrated that individuals with PTSD exhibit greater amygdala activation relative to comparison subjects(22,23). Further, functional imaging studies have examined connectivity between the prefrontal cortex (PFC) and the amygdala and observed impaired inhibition in PTSD subjects(24). The hippocampus is primarily involved in storing memory. Notably, gross hippocampal volume is decreased in PTSD patients compared to traumatized controls who did not develop PTSD(25). We included three subregions of the hippocampus in our analysis: the dentate gyrus (DG), CA subfields (CA), and the subiculum (Sub).

We generated DNAm data from six postmortem primary brain regions from 171 individual donors using the targeted next-generation sequencing of bisulfite-converted DNA (targeted methyl-seq). After rigorous quality control, this provided genomic coverage across 1,065,750 CpG sites representing 22,544 genes, in a much larger sample size (117 cases versus 54 controls) than previous studies, across two unique diagnostic groups (PTSD and major depressive disorder) compared to neurotypical controls in brain regions (subregions of the amygdala and hippocampus) not extensively examined in previous large DNA methylation cohorts(26). The goal of this effort was to comprehensively measure epigenetic responses (i.e., DNA methylation differences at both CpG and non-CpG sites) across discrete brain regions with known involvement in PTSD pathophysiology. To further resolve the changes we observe, we take advantage of high dimensional single cell RNA-sequencing data to construct a cell type-specific atlas of PTSD methylation changes across all major CNS cell types. We have also included a non-PTSD psychiatric comparison group of individuals with MDD due to the high comorbidity between the two disorders(27). We find differential DNAm signatures across all six regions including many non-overlapping, sex-specific differences. Many DNAm signals are present near genes regulating GABAergic transmission such as ELFN1 which has previously been implicated in PTSD interneuron dysfunction(28,29) and glucocorticoid signaling including CRHR1. We also find that DNAm changes aggregate at genes previously implicated in genetic risk for PTSD and its associated clinical phenotypes including KANSL1, MAD1L1, and CRHR1(30). Notably, we also identified significant changes in methylation of the ELFN1, MAD1L1, and WNT5A genes in response to the rapid-acting antidepressant ketamine in a cohort of PTSD patients. Taken together, this work represents a unique and powerful resource for exploring DNA methylation changes in human subcortical regions important for psychological stress pathology and provides critical neurobiological targets for future development of therapeutics.

RESULTS

Widespread DNAm differences in subcortical brain regions

We obtained DNAm data from 171 human postmortem individuals balanced for sex, including donors diagnosed with PTSD, major depressive disorder (MDD), and neurotypical controls (Supplementary Table 1) using targeted sequencing of bisulfite converted DNA (Targeted Methyl-Seq). We mapped the DNAm landscape of six postmortem brain regions: three amygdala nuclei: basolateral (BLA), central (CeA), and medial (MeA); and three hippocampal subregions: Cornu Ammonis (CA) subfields, dentate gyrus (DG), and subiculum (Sub) using a custom bioinformatic analysis pipeline (Figure 1A). Additionally, our cohort was well balanced for sex (PTSD: 31 females, 30 males; MDD: 24 females, 33 males; and healthy controls: 21 females, 32 males). The targeted methyl-sequencing initially included capture for 5M CpGs across the human genome. After extensive quality control (Methods) and removal of CpGs on the sex chromosomes, we were left with 1,065,750 autosomal probes for analysis (Supplementary Table 2).

Figure 1. DNA isolated from different brain regions display widespread differences in CpG methylation.

(A) Bioinformatics analysis pipeline. (B) Individual scores from principal components analysis of all samples included in the study. PC2 maximally separate subjects by sex. (C) PC1 maximally separate subjects by brain regions. (D) heatmap of median methylation levels of the 1,200 CpG sites (top 200 most significant CpG sites for each region that distinguish each of the six brain regions from the five others); only samples with data from every brain region were included. Hierarchical clustering of these 1,200 sites reveals subregion similarities within each primary region. (E) Heatmap of z-score transformed odds ratios for genic, CpG and chromatin features showing the log odds ratio of Bonferroni-corrected significant CpG sites within the feature versus excluded. Significant subcortical CpGs are more enriched in bivalent promoter, weak and active enhancer regions (P < 0.0001) and less enriched in TSS, ZNF genes, and gaps (P < 0.0001).

Principal component (PC) analysis across DNAm levels revealed that the top PCs were associated with sex (PC2: 1.88% variance) (Figure 1B). Consistent with previous DNAm studies(11) of human brain, we found that neuronal composition accounted for one of the top components of variation, explaining 3.98% (PC1) and 1.78% (PC 3) DNAm variance (Supplementary Figure 1A). Processing and sequencing batch effects have been reported to significantly affect DNAm levels. However, we did not find a major effect on variance attributable to batch (Supplementary Figure 1B). These processing issues are likely ameliorated in our pipeline as bisulfite conversion occurred in the same plates as the downstream sequencing (Methods).

DNAm differences between PTSD subjects and neurotypical controls

We compared differences in CpG methylation levels for each region between our control and PTSD cohorts controlling for sex, age at death, ancestry, brain bank, PMI, neuronal proportion and smoking status (Supplementary Table 1). We used dispersion shrinkage for sequencing (DSS)(31) which was initially developed for the analysis of bisulfite sequencing data of individual CpGs(32). In the combined sex analyses, we identified 6 FDR significant DMCs (P_B-H < 0.05). We found one hypermethylated DMC in the CA for the SNAR-G2 gene (test statistic = 5.6). We found five hypomethylated DMCs for microRNA miR54812 (test statistic = −5.7) and NDUFAF1(test statistic = −5.6), HORMAD2-AS1(test statistic = −5.5), and NXN (test statistic = −5.9) genes in the DG, and SLC22A1 (test statistic = −5.4) in the Sub (Supplementary Figure 2A and Supplementary Table 4). We performed baseline comparisons of DNAm between control males and females. Similar to previous reports, we identified 1932 FDR-significant DMCs (Supplementary Table 4). We found the CeA had the most differences with 441 DMCs and the CA had the fewest with 216. In females, we found six significant DMCs in the BLA, 10 in the CeA, 16 in the MeA, 19 in the CA, 86 in the DG, and 13 in the Sub (Supplementary Figure 2B shows the top ten FDR significant sites; the full list is available in Supplementary Table 4). We also found non-overlapping male-specific gene-associated DMCs in the BLA (3 genes), the CeA (8 genes), the MeA (15 genes), the CA (11 genes), the DG (15 genes), and in the Sub (16 genes) (Supplementary Figure 2C and Supplementary Table 4). The most significant genes with DMCs for all regions and by sex are shown in Table 1.

Table 1.

The top 5 FDR significant DMCs in the univariate analysis for combined-sex, and two sex-specific analyses.

Analysis	Chromosome	Position (bp)	Region	Closest Gene (Location)	Test Statistic	P	Adj. P
Combined	19	799710	DG	NXN	−5.8737	4.26E-09	0.0043
	6	9533078	DG	AC097493.4	−5.7705	7.90E-09	0.0043
	17	41403096	DG	NDUFAF1	−5.6408	1.69E-08	0.0061
	21	49029942	CA	NTF6G	5.5576	2.73E-08	0.0296
	24	30080197	DG	HORMAD2	−5.5301	3.20E-08	0.0086
Female	9	4221321	CNA	SDK1	−6.3581	2.04E-10	0.0002
	4	237157075	DG	AC107079.1	−6.3181	2.65E-10	0.0002
	13	89787631	DG	AP004833.2	−6.2778	3.43E-10	0.0002
	3	3297053	Sub	PRDM16	−6.2539	4.00E-10	0.0004
	4	99593922	CA	AFF3	−6.1477	7.86E-10	0.0009
Male	15	94968558	DG	AL139381.1	6.3734	1.85E-10	0.0002
	4	131443806	CA	NOC2LP2	6.3396	2.30E-10	0.0002
	4	117859040	MNA	HTR5BP	6.2142	5.16E-10	0.0006
	16	104317737	Sub	AL512357.2	−6.211	5.27E-10	0.0006
	9	107952197	DG	LAMB1	−6.1054	1.03E-09	0.0006

DNAm changes at PTSD risk loci

We found that 11,470 CpG sites (of the 42,582 CpG sites (26.9%), P < 0.05) were significantly, differentially methylated within the LD regions spanning 137 PTSD genome-wide significant loci. We found significant differences in DNA methylation ratios for MAD1L1 (P < 0.0001, test statistic = −4.2), CAMKV (P = 0.0061, test statistic = −2.7), KCNIP4 (P = 0.0012, test statistic = 3.2), KANSL1 (P = 0.0023, test statistic = −3.0), CRHR1 (P = 0.0004, test statistic = −3.5), and TCF4 (P = 0.0004, test statistic = −3.5) and suggestive changes in HSD17B11 (P = 0.0114, test statistic = −2.5) and SRPK2 (P = 0.0015, test statistic = −3.2) (Figure 3A).We specifically examined methylation changes at nine genes identified by the MVP PTSD GWAS(30) and plotted their differential methylation for each subregion (Figure 3B). We found significant changes for KCNIP4, HSD17B11, MAD1L1, SRPK2, KANSL1, CRHR1, and TCF4 for at least one brain subregion. Because we identified a large portion of DMCs near PTSD risk genes (26.9%), we reasoned it was likely that enrichment of DMCs maybe occurring near or within other genes implicated in PTSD pathophysiology. Therefore, we explored the possibility of enriched methylation differences within PTSD genetically regulated genes identified by TWAS(28). We found significant enrichment of DNAm changes in the interneuron gene ELFN1 (Figure 3A) (O.R. = 12.25; P < 0.0001)(28).

Figure 3. Examples of DNAm changes for 10 GWAS-positive loci for PTSD.

(A) Examples of group-specific DNA methylation ratios for ten PTSD risk genes. MAD1L1, CAMKV, KCNIP4, SPRK2, KANSL1, CRHR1, and TCF4 are significant risk genes (identified by GWAS) for PTSD. ELFN1 is a significant TWAS hit for PTSD, and SLC32A1 is a previously identified transcriptomic key driver in PTSD. For each boxplot, y-axis dots show methylation levels at a specific CpG site. P-value corresponds to differential methylation. For box plots, center line is the median, limits are the IQR, and whiskers are 1.5X the IQR. (B) Methylation changes at nine genes identified by the largest PTSD GWAS using data from MVP were examined. Significant changes were found for KCNIP4, HSD17B11, MAD1L1, SRPK2, KANSL1, CRHR1, and TCF4 for at least one brain subregion. Significant enrichment of DNAm changes were also found in the interneuron gene ELFN1, which was a PTSD genetically regulated genes identified by TWAS.

PTSD DNAm cell type-specific atlas

Our epigenomic analysis of PTSD regions implicated several diverse biological processes including GABAergic and neuroimmune signaling, suggesting possible involvement of different cell types in PTSD etiology. To better resolve the individual contributions of brain cell types in the PTSD, we generated a DNAm-atlas from our bulk/homogenate DNA methylomes. We used estimated cell-type fractions for all major CNS cell types including astrocytes (Astro), excitatory neurons (EXC), inhibitory neurons (INH), microglia (MG), oligodendrocytes (OLG) and oligodendrocyte precursor cells (OPC). We applied CellDMC(33), a pipeline designed to detect cell type-specific differential DNAm (DMCTs) signals using single cell RNA-sequencing data reference. We detect 102 FDR significant (P_B-H < 0.05) cell type-specific differential DNAm CpGs (DMCTs) across all six brain regions and cell types. Most FDR significant PTSD-associated DMCTs occurred in OLGs of the MeA, with lower but significant numbers across all cell types, most notably in OPCs and MG (Figure 4A). We identified FDR significant cell type DMCTs near the PTSD GWAS risk genes MAD1L1 (P_B-H = 0.017) and ESR1 (P_B-H = 0.0017) in OLGs (Figure 4B). We also noted strong enrichment of cell type specific DMCTs in upstream regions and promoters across all cell types (Figure 4C, blue). In addition, we found significantly more hypomethylated DMCTs in the CpG features of INH and MG and significantly more hypermethylated DMCTs in OLGs and OPCs (Supplementary Figure 9C, D, green). We observed significant enrichment (P < 0.05) of PTSD GWAS signals near hypomethylated DMCTs in OLGs, and of hypermethylated DMCTs in INH neurons and OPCs (Supplementary Figure 9C, D, salmon). Overall, cell type specific methylation analysis suggests a mostly inhibitory neuron and oligodendrocyte origin of risk for PTSD.

Figure 4. Brain Cell-type specific DNAm atlas in PTSD.

(A) Bar plots display the number of cell type-specific PTSD-associated DMCTs, calculated using CellDMC. Hypermethylated DMCs are on the left and hypomethylated DMCs are on the right (B). An example region of cell type specific DNA methylation in oligodendrocytes of the MeA at PTSD GWAS risk loci MAD1L1 (top) and ESR1 (bottom). Yellow highlight indicates position of significant methylation change. (C) Heatmap of cell type specific DMC enrichments in genic features (blue), CpG features (green), and PTSD GWAS risk variants (salmon) across brain regions and cell types. *P value<0.05, ** P value <0.005, ***P value <0.0005.

Blood DNAm signatures are associated with ketamine response treatment in PTSD patients

In order to link our findings with potential clinical significance, we explored the methylation patterns in peripheral blood of PTSD patients who responded to the rapid acting antidepressant ketamine(34). Intravenous ketamine was administrated twice a week to 97 veterans and active-duty service members with PTSD for four weeks. PTSD severity was assessed by self-reported PTSD check list for DSM-V (PCL-5) prior to each ketamine infusion and 24h post first and last infusions. Responder status was determined by the slope of trajectory on the PCL-5 score. DNA methylation was measured at 24-hr post 1^st and 8^th ketamine infusions and 2–3 days post 4^th ketamine infusion. We found significant ketamine-induced methylation variation in the several PTSD associated genes identified in our study (Figure 5). We observed significant differential methylation to ketamine treatment in MAD1L1 (e.g., at five sites at chr7:1866310 (P = 0.005), chr7: 1952716 (P = 0.001), chr7: chr7:1952724 (P = 0.001), chr7:2116690 (P = 0.015), chr7:2116692 (P = 0.003)), ELFN1 (e.g. chr7:1714000 (P = 0.001), chr7:1714057 (P = 0.009), chr7:1708567 (P = 0.002)), and in WNT5A (e.g. chr3:55504935 (P = 0.003)) associated with ketamine response. All significant ketamine sites were within 1.5kb of one or more significant PTSD CpGs in postmortem brain (P < 0.05, Supplementary Data Table 14) except for the nearest CpG to WNT5A (15kb). These single gene analyses suggest broader clinical implications for methylation changes in PTSD patients.

Figure 5. Examples of ketamine altered DNA methylation ratios for three PTSD genes.

Significant hypomethylation at five sites in the promoter of MAD1L1 in patients who responded to ketamine, three in the ELFN1 promoter and one in the WNT5A promoter. For each boxplot, y-axis dots show methylation levels at a specific CpG site. P value corresponds to differential methylation. For box plots, center line is the median, limits are the IQR, and whiskers are 1.5X the IQR.

Distinguishing DNAm associations of PTSD and MDD diagnoses

PTSD is highly co-morbid with major depressive disorder (MDD) with more than 50% of newly diagnosed cases of PTSD also experiencing depressive symptoms(27). Therefore, we included analyses of an MDD cohort balanced for age, sex, PMI, and with a similar history of drug use to disentangle the unique and divergent biological processes of both disorders. We analyzed this MDD cohort in the same manner as our PTSD cohort (Figure 1A) and corrected for the same covariates (age at death, sex, ancestry, cell type proportion, brain bank, PMI score, and smoking status) to make relevant biological comparisons between the two diagnostic groups (Supplementary Figure 10, Supplementary Data Table 15).

We found that between disorders, DMCs were most similar between the same regions than in any other regional or multi-regional comparison and unique for PTSD and MDD (Figure 6B). We next directly compared the genes associated with these DMCs to better partition differences across diagnoses (Figure 6C) and identify overlapping and convergent biological processes. Genes whose DNAm level was inversely correlated with gene expression and were more highly expressed in PTSD compared with MDD (P < 0.05) were associated with neuronal processes and the synapse (Figure 6C), particularly in the Sub. Notably, we found changes in the synaptic gene CAMK2A (LFC = 0.3, P = 0.0193) and the glutamate receptor GRIK4 (LFC = 0.23, P = 0.0266). Genes that were more highly enriched in MDD compared to PTSD were associated with neuroinflammatory processes and leukocyte activation (Figure 6D). These findings suggest decreased involvement of neuroimmune processes and microglia activity in PTSD versus depression consistent with previous functional neuroimaging studies of reduced TSPO availability in patients with PTSD versus increases in those with major depression(35).

Figure 6. Common and divergent epigenetic signatures between PTSD and MDD.

(A) Bar plot shows number of significant CpG sites in the PTSD case-control analysis (pink) and MDD case-control analysis (gray). (B) UpSet plot shows the shared DMCs across brain regions and diagnosis groups. (C, D) Enrichment analysis for genes with DNAm levels inverse to gene expression stratified by increases in PTSD (C) and increases in MDD (D). One asterisk (*) indicates adjusted p-value < 0.1, and two asterisks sign (**) indicates adjusted p-value < 0.01 (E) Locus zoom-in plots on chromosome 7 for PTSD combined-sex analysis and (F) MDD combined-sex analysis. The significant DMCs were plotted by subregion (red-BLA, orange-CeA, yellow-MeA, purple-CA, blue-DG, and green-Sub. Nominal significant (P<0.05) DMCs are plotted in blue and non-significant DMCs are colored in gray.

We found several significant DMCs in the MAD1L1 locus, a major risk gene for both PTSD and MDD (Figure 6E and 6F). MAD1L1’s nearest neighbor gene is ELFN1 (approximately 68 kbp away) and both genes are in high linkage disequilibrium. Previous studies have identified significant changes in ELFN1 gene expression in the dorsolateral prefrontal cortex of the PTSD brain and ELFN1 was a significant TWAS hit in the same study(28). Interestingly, we found multiple DMCs in the intronic and exonic regions of ELFN1 that were specific to PTSD amygdala subregions and a single less significant DMC in the MDD Sub. These findings reveal unique regulation of ELFN1 in the amygdala of individuals with PTSD and suggest alterations in GABAergic signaling that differ from MDD.

DISCUSSION

Here we present the most comprehensive DNA methylation study of the postmortem PTSD brain. Because each region of the brain has a unique cellular composition and regulatory landscape that contributes to neurobiological function, studying several brain regions simultaneously is essential to translating knowledge of epigenetic modifications into insight of molecular risk processes. We were able to cast a wide net to catch PTSD-related genomic alterations by examining the epigenomes of six brain subregions concurrently: three subnuclei of the amygdala and three subregions of the hippocampus and found and analyzed a sizable number of sites that differed in methylation depending on the region. It is important to note that gene expression patterns are not the same across amygdala and hippocampal brain areas, even if differential methylation is shared across them. To validate the functional impact of the DNAm changes we observed, we performed genome-wide transcriptomic analysis on donors and regions used in our methylation cohort. Importantly, we found that greater than 50% of case-control DEGs aligned with significant DMC hyper/hypo- methylation patterns, suggesting the epigenomic changes we observe have mechanistic roles underlying dysregulated gene expression in PTSD brain. It is important to note that DNAm is not the only epigenetic modification capable of altering the transcriptome. Histone modifications and the three-dimensional nature of DNA (chromatin looping) can also alter gene expression(36) and likely play additional or complementary roles in control of gene expression with DNAm. PTSD changes to these systems may explain why DNAm alone does not account for all changes to the transcriptome. Future studies examining the role of additional epigenetic changes are thus warranted.

To better understand the biological processes underlying the differential methylation signals we identified, we performed gene network analysis using Ingenuity Pathway Analysis (IPA) from Qiagen(37). IPA curates a large database of empirically derived biological interactions providing a powerful tool for identifying how genes (in this case genes that differentially methylated in PTSD) are coordinated in regulating biological process and identifying hub genes with significantly large numbers of molecular connections to other genes. Our gene network analysis identified several hub genes including ESR1 (Estrogen Receptor 1). ESR1 has previously been implicated female PTSD biology(38) and is involved in transcriptional regulation of many processes including neuroinflammation a role consistent with observed cell type specific changes in ESR1 methylation in microglia (Figure 5, Supplementary Data Table 6). We observed moderate enrichment for DMCs with risk loci for PTSD (26.9%) suggesting genetic regulation of CpG methylation in diseased tissue. In combination with cell type-specific differential DNAm calling (Figure 5), we were able to detect a possible inhibitory neuron and oligodendrocyte origin for PTSD. We found enrichment of PTSD-risk GWAS loci among PTSD hypermethylated INH and hypomethylated OLG DMCTs suggesting DNAm may play an important role in mediating genetic risk for PTSD.

We observed significant DMCs and DMHs within the MAD1L1 gene that is a significant risk gene for PTSD(30), MDD(39), Schizophrenia and Bipolar Disorder(40) pointing to a convergent role across neuropsychiatric disorders. However, our results also point to significant changes in methylation patterns in MAD1L1’s neighboring gene ELFN1 that appear to be PTSD-specific (Figures 6A&E and 6F). We have previously reported ELFN1 gene expression changes in the frontal cortex of PTSD and MDD patients(28). To date, no evidence has emerged implicating ELFN1 involvement in these disorders stemming from subcortical gene expression dysregulation. The lack of significantly methylated CpGs in ELFN1 in the MDD amygdala strongly suggests a unique role in PTSD subjects. Future studies of GABAergic synaptic transmission disruption in PTSD amygdala are thus warranted. In addition, a focus on cell-type specific studies of changes in both CpG methylation and CpH methylation will provide the necessary resolution to determine the specific effects these changes are having in specific neuronal subtypes.

Several studies have suggested that DNA methylation can be altered by antidepressants and promote symptom remission(41). We therefore reasoned that epigenetic modification of PTSD genes with DMCs could be clinically relevant. Therefore, we explored DNA methylation in the peripheral blood of PTSD patients treated with the rapid acting antidepressant ketamine (both responders and non-responders), with the aim of understanding the potential contribution antidepressants might play in the epigenome of PTSD. PTSD patient responders to ketamine had significant decreases in methylation in three PTSD hub genes (ELFN1, MAD1L1, and WNT5A) compared to those who did not respond to ketamine. Ketamine treatment response is based on DNA methylation changes across three time points, long after the drug was fully metabolized, suggesting longer term epigenetic changes in PTSD patients. Previous studies have also identified hypomethylation of the MAD1L1 gene associated with PTSD among post-combat male military cohorts(42). Together, this evidence suggests that DNA methylation of these genes might serve as a biomarker for PTSD diagnosis and treatment. Further investigation into the relationship between ketamine, DNA methylation, and PTSD is needed to identify genome-wide epigenetic changes associated with rapid acting antidepressant administration.

In summary, we present results from the largest DNA methylation study of human brain subcortical regions to date. These results support the role of DNAm as a critical epigenetic modulator of gene expression in PTSD etiology. We detect cell type-specific changes to the PTSD methylome by generating a unique and powerful DNAm atlas. Peripheral changes in DNAm after ketamine administration occurred near CpGs altered in the PTSD brain at genetically linked and neurobiologically relevant genes. PTSD risk genes were major hub genes in methylation gene networks suggesting a genetic component to downstream molecular functions. Finally, we identified convergent and divergent epigenetic regulatory mechanisms between PTSD and MDD highlighting common and divergent pathways. More generally, we illustrate the value of integrating epigenetic data across multiple brain regions to study complex psychiatric disorders.

MATERIALS and METHODS

Targeted Methylation Enrichment

Targeted next-generation sequencing of bisulfite converted DNA (targeted Methyl-seq) was achieved using the Roche KAPA HyperPrep Kits in combination with the SeqCap Epi CpGiant Probes. A panel of more than 2.1 million long oligonucleotide DNA probes (total capture size of 80.5 Mb) were used to interrogate >5.5 million methylation sites per sample at single-nucleotide resolution. The provides coverage for 99% of Refseq genes (hg19) (including miRNAs), promoter regions, and 96% of CpG Islands and CpG island shores.

For each sample, DNA concentration was measured using a ThermoFisher Scientific Qubit Fluorometer and 250ng was subsequently fragmented using a Covaris E210 Focused-ultrasonicator to generate ~300bp DNA fragments. 5% unmethylated PhiX DNA was spiked in to normalize nucleotide imbalance and better calibrate Illumina base calling. All samples beginning with DNA extraction through sequencing were completely randomized by donor, brain region and sex. Fragmented DNA samples were then converted into Illumina libraries using the KAPA HyperPrep Kit. In brief, DNA fragments were end repaired, A-tailed and ligated to barcoded Illumina adapters. Each library was then bisulfite converted using the EZ DNA Methylation-Lightning Kit by Zymo Research and amplified by polymerase chain reaction (PCR).

The bisulfite converted libraries were then hybridized with the SeqCap Epi CpGiant Probes in order to enrich the libraries for the > 5.5 million methylation sites across the genome. Bisulfite conversion occurred in the same plates as sequencing to minimize potential batch effects caused by additional handling. Enriched libraires were subsequently amplified by PCR before being pooled for sequencing on an Illumina NovaSeq 6000 Sequencing System. DNA libraries were run on an Agilent Bioanalyzer after shearing and size selection for quality control purposes. Three pools of 332 samples each were prepared. Each pool was sequenced on a NovaSeq 6000 S4 flow cell at 150bp paired-end, generating about 9Gb of data per sample. This corresponds to 88 – 97X coverage across the samples (average of 92X).

Bioinformatics

Differentially Methylated CpG Sites

To identify potentially differentially methylated CpG sites (DMCs) between control and PTSD and control and MDD individuals, a beta binomial linear regression model with an arc-sine link function was performed on each CpG site individually utilizing the DSS(31) Bioconductor package in R. This was the main analysis of the study and all additional analyses are considered exploratory. Age, sex, post-mortem interval (PMI), smoking status, ancestry, and tissue source were included as co-factors in the model. We included estimated cell type proportion and inferred ancestry information from genotype data in the model. The models were run separately for each brain region and also for males and females jointly and individually (sex was excluded as a co-factor in the sex-specific models). In all of our univariate analysis for combined sex and sex-specific comparisons, Benjamini-Hochberg procedure (FDR) correction is applied per brain region using threshold of 0.05. We also reported the Wald test statistic which represents both direction and magnitude of change in methylation. A positive test statistic represents hypermethylation and negative represents a change to hypomethylated state.

Differentially Methylated CpH sites

To identify differentially methylated CpH sites between control and PTSD, we employed a beta binomial linear regression model with an arc-sine link function on each CpH site individually utilizing the DSS(31) Bioconductor package in R. Age, sex, post-mortem interval (PMI), smoking status, ancestry, tissue source, and estimated cell type proportion were included as co-factors in the model. The models were run separately for each brain region. Benjamini-Hochberg procedure (FDR) was applied per brain region and per type of CpH sites for the correction of multiple hypothesis testing. Ultimately, the results were combined to gain insights on differentially methylated sites at the CpH level. Benjamini-Hochberg correction is applied using threshold of 0.05.

Supplementary Material

Supplement Description (2 files):

Supplement Methods, Results, Figures S1–S10, Table S1–S15 Legends

Tables S1–S15

Figure 2. Network analysis.

Gene network analysis was performed using IPA for DMCs in each differential methylation pattern. The most significant gene networks (network score > 30) from the six single-regional analyses were obtained and merged by integrating the hub genes and their nearest neighbors in IPA. Pie-circles were used to indicate the differential methylation regions for genes that have more than two neighboring genes. The amygdala regions (BLA: red, CeA: orange, and MeA: yellow) are in warm colors and the three hippocampus regions (CA: purple, DG: blue, and Sub: green) are in cold colors. Genes SMARCA4, ESR1, TCF4, WNT3A, and WNT5A were hub genes with the greatest number of nearest neighbors.

KEY RESOURCES TABLE

Resource Type	Specific Reagent or Resource	Source or Reference	Identifiers	Additional Information
Add additional rows as needed for each resource type	Include species and sex when applicable.	Include name of manufacturer, company, repository, individual, or research lab. Include PMID or DOI for references; use “this paper” if new.	Include catalog numbers, stock numbers, database IDs or accession numbers, and/or RRIDs. RRIDs are highly encouraged; search for RRIDs at https://scicrunch.org/resources.	Include any additional information or notes if necessary.
Biological Sample	human postmortem tissue: 61 PTSD: 31 females, 30 males; 57 MDD: 24 females, 33 males; and 53 healthy controls: 21 females, 32 males	National PTSD Brain Bank
Commercial Assay Or Kit	Agilent SureSelectX Methyl-seq library	Agilent	G9651A
Commercial Assay Or Kit	RNeasy Mini Kit	Qiagen	74104
Commercial Assay Or Kit	SMARTer® Stranded RNA-seq Kit	Takara	634411
Commercial Assay Or Kit	Roche KAPA HyperPrep Kits	Roche	KK2801
Sequence-Based Reagent	EZ DNA Methylation-Lightning Kit	Zymo	D5030-E
Software; Algorithm	R packages for anlysis	https://github.com/mjgirgenti/PTSDSubcorticalMethylation

Code Availability

All code used in this study is freely available online and can be found at https://github.com/mjgirgenti/PTSDSubcorticalMethylation

Data Availability

DNAm and RNA-seq count data count data generated and/or analyzed during the current study are available at https://www.girgentilab.org/datasets. Raw data will be made available upon reasonable request from the Department of Veterans Affairs National Center for PTSD.