Background

Obsessive-compulsive disorder (OCD) is a debilitating neurodevelopmental condition that affects up to 3% of the worldwide population according to the World Health Organization [1]. OCD is characterized by intrusive thoughts and repetitive behaviors in a large time-consuming manner [2, 3].

Cortical areas [anterior cingulate gyrus (ACC), dorsolateral prefrontal cortex (dlPFC) and orbitofrontal cortex (OFC)] maintain the main projections to the ventral striatum areas [nucleus accumbens (NAC), caudate nucleus (CN) and putamen (PT)] [2, 4], both involved with OCD symptoms, paradigms [5], and treatment response [6]. The cortico-striato-thalamo-cortical circuitry (CSTC) is altered in the brain of OCD individuals, which includes three relevant loopings of indirect pathways with the respective cortical connection: the affective (ACC, NAC and thalamus), dorsal cognitive (dlPFC, CN and thalamus) and ventral cognitive (OFC, PT and thalamus) circuits, which are related to the affective and reward processing, the working memory and executive function, and the motor and inhibitory response, respectively [2].

Neuroimaging MRI studies using whole-brain voxel-based morphometry (VMB) revealed that changes in anatomical structures from both affective and cognitive (executive) circuits are consistently described in OCD cases and were related to variation in symptom severity [7]. Diffusion-weighted magnetic resonance imaging was associated with gene expression alterations confirming the tripartite model of striatum organization and connection model [8]. We explored this model by using the differentially expressed (DEGs) and coexpressed genes modules in CN, NAC and PT brain tissues from OCD cases and controls, revealing the involvement of cell communication, cell response, synaptic transmission and plasticity for all striatum areas [9].

Different studies demonstrated that OCD etiology is a multifactorial condition with both polygenic and environmental risk factors [3, 10]. The impact of environmental factors may reflect changes in DNA methylation (DNAm), an epigenetic modification that consists in the addition of a methyl group (CH3) to carbon at the fifth position of cytosine (C) [11]. DNA methylation intermediates the interaction between genetic and environmental factors involved with psychiatric disorders [12]. Specifically for OCD, DNAm has been investigated in peripheral tissues, such as blood [13,14,15] including mononuclear cells [16] and saliva [17]. Differentially methylated CpG (cytosine-phosphate-guanine) sites between OCD patients and controls were only partially able to group patients (67%) in an unsupervised clustering analysis. These CpG sites were located in genes enriched for actin cytoskeleton, cell adhesion molecules (CAMs), actin binding, transcription regulator activity, and other cellular pathways [13]. By evaluating methylation levels from selected CpG sites, no changes were observed in 14 genes previously associated with OCD [14]. However, OXTR, the oxytocin receptor gene, presented higher methylation in the OCD patients and correlated with severity and oxytocin was associated with the regulation of complex socio-cognitive processes [15]. DNAm levels of a CpG site located in the first intron from SLC6A4 were higher in the saliva of pediatric and adult OCD patients compared to controls but no alteration was observed for SLC6A4 expression [17]. The opposite way, higher methylation of two CpG sites located at BDNF exon 1 correlated with higher expression in OCD patients [16].

These data, derived from surrogate tissues, point to the necessity of exploring DNAm in the brain areas associated with OCD to verify its possible contribution for the disease. Furthermore, the methylation clock [18] in brain tissues from patients with OCD should be disclosed as it is a complement of tissue senescence. The DNA methylation clock is associated with physiological ageing but also is associated with changes according to stress exposition along life [19] and age acceleration was associated with other psychiatric disorders [20, 21].

Considering our small sample size to explore DNA methylation comparing cases and controls, to avoid false positive results, we performed gene network analysis of DNAm integrated with transcriptomic data in the brain areas associated with OCD.

Results

Characterization of OCD cases and controls

Individuals from both groups (OCD cases and controls) presented similar socio-demographic characteristics and were matched by sex and age. All individuals were older than 50 years and did not have a history of clinical dementia at the time of death (Table 1).

Table 1 Demographic characteristics of obsessive-compulsive disorder (OCD) cases and controls
Full size table

DNA methylation data comparing OCD cases and controls

For all brain areas, comparison between OCD cases and controls did not point to differentially methylated CpG sites (DMSs) after multiple correction tests (adjP≤0.05). Additional file 2: Table S1 presents CpG sites with a p-value < 0.0005 for each brain area. The EpiMod algorithm, which is based only in methylation data and protein-protein-interaction (PPI) networks, from Functional Epigenetic Modules (FEM) analysis, infers differential methylation hotspots called modules. It showed enriched gene modules for all brain areas between OCD cases and controls: five modules for ACC, nine for OFC, five for NAC, eight for CN and three for PT (Additional file 2: Table S2). ACC and PT modules were predominantly hypomethylated, NAC and CN were mostly hypermethylated and OFC had both hypermethylated and hypomethylated modules (Additional file 1: Fig. S1).

Considering differentially methylated regions (DMRs), four out of five brain areas presented mostly a hypomethylation pattern: 70 DMRs for ACC (22 hypermethylated and 48 hypomethylated), 356 for OFC (140 hypermethylated and 216 hypomethylated), 75 for CN (26 hypermethylated and 49 hypomethylated), 106 for PT (11 hypermethylated and 95 hypomethylated). Only NAC had more hypermethylated than hypomethylated DMRs (n = 174, 138 hypermethylated and 36 hypomethylated) (Additional file 2: Table S3). Some genes located in DMRs were also identified by EpiMod-FEM analyses, such as HLA-DQB1 and HLA-DQA1. DMRs located at ADARB2 and UGT2B15 were hypermethylated in OFC DMSs, but do not survive after the multiple test correction (p-value: 1.33E-05/ Δβ: 0.186; p-value: 3.60E-04/ Δβ: 0.208, respectively).

DNA methylation and gene expression integration data

As we have a small sample size, methylation and transcriptomic data integration from the same brain areas from OCD cases and controls resulted in a more robust approach. The list of differentially expressed genes (DEGs) comparing OCD cases and controls from NAC, CN and PT was retrieved from a previously published study by our group [9]. To search DEGs for ACC and OFC, RNASeq data were preprocessed using the same parameters as described in the methods section. First, data were integrated by using genes from DMRs associated with genes, FEM enriched modules from methylation and DEGs from the RNASeq data (p < 0.01) (Additional file 2: Table S4) to construct a PPI network for each brain area. To achieve a more accurate analysis regarding connected transcriptomic and methylation data, we selected only edges that connected nodes from different lists, e.g. one node from the DEG list and another from the DMR or FEM lists. All nodes were classified by network centrality measures (Additional file 2: Table S5) and by using a 95 percentile as threshold for each topological measure ranked list, we observed that genes from DEGs, DMRs and FEM are represented. For OFC PPI, 40 (23%), 51 (29%), 86 (48%); from NAC PPI, 44 (29%), 41 (27%), 69 (45%); from CN PPI, 186 (66%), 51 (18%) and 46 (16%); and for PT PPI, 44 (53%), 11 (13%) and 28 (34%) (Fig. 1). Interestingly, HLA-DQA1, reported as differentially methylated in CN DMR and FEM analyses, had a high classification in the rank according to its degree and closeness measures in the network. Searching for genes previously associated with OCD in the networks, all brain areas have specific genes as well as genes represented in more than one brain area (Table 2).

Fig. 1
figure 1

PPI networks from STRING [86] for the five brain areas using DEG, genes with DMRs and genes from FEM modules. Network plots were created using the igraph (v.1.2.4.2) library

Full size image
Table 2 Genes (nodes) belonging to the final networks that were already associated with OCD
Full size table

For each network, we searched for functional module enrichment analysis as described in the methods section, considering non-redundant ontologies/pathways with a minimum of 5 genes. PPI networks from all brain areas were enriched for G-protein signaling pathway, immune response, apoptosis and synapse biological processes. Also, all areas but CN were enriched for different behaviors, including feeding, learning and memory. ACC, OFC, NAC and CN were also enriched for axon, dendrite, purine process, response to stress, GTPase and MAPK activity. Regarding exclusive processes, ACC PPI network was enriched for cAMP signaling, OFC for inflammatory response and interferon-gamma signaling pathway, NAC for acetylcholine receptor activity and cholinergic synaptic transmission, and PT for transcriptional regulation. CN PPI network had the higher number of enriched processes, such as myelination, glial cell development, peripheral nervous system development, regulation of I-kappaB kinase/NF-kappaB signaling, PI3K signaling, Ras and Rho proteins signal transduction, type I interferon and JAK-STAT signaling pathways, JUN kinase activity, MHC class II receptor activity and regulation of transcription in response to stress.

Considering the REACTOME, GPCR signaling was enriched in all areas. NAC, CN and PT were also enriched for MAP kinase activity. Some areas have more specific pathways such as neurotransmitter receptors and postsynaptic signal transmission in ACC, interferon gamma signaling and MHC class II antigen in OFC, Rho GTPase cycle, interferon signaling, neddylation and pyruvate metabolism in NAC, interleukin-17 signaling, class I MHC processing, neddylation and cellular senescence in PT. CN was enriched for interferons gamma, alpha/beta signaling, interleukins 3, 4, 5, 10 and 13 signaling, Rho GTPases signaling, MHC class I and II antigen presentation, transcriptional regulation by RUNX family genes, NMDA and GABAB receptors (Additional file 2: Table S6).

DNA methylation and gene expression integration data from DMRs not located at genes

To present a comprehensive assessment of DMRs, including those not located at genes, they were submitted to ENCODE [22] to be annotated to TFs. We identified seven TFs for ACC, 15 for OFC, 20 for NAC, 21 and 20 for CN and PT, respectively. Genes targeted by the TFs were searched within DEGs (Additional file 2: Table S7). In relation to TFs targeted DEGs, we identified eight for ACC, 19 for OFC, 13 for NAC, 68 for CN and 22 PT. TFs and their targeted DEGs as well as its activation/repression relation were used as connections to construct a network (Fig. 2).

Fig. 2
figure 2

Regulatory Networks for TFs from DMRs and targeted DEGs for the five brain areas. Network plots were created using the igraph (v.1.2.4.2) library

Full size image

Some TFs but not their targets were shared between areas and some TFs-DEGs pairs were shared between areas, although sometimes the target could be up or downregulated for the different areas. Mostly, they were involved with cellular processes such as cell growth, cell survival, cell proliferation and/or differentiation, cell death, immune and inflammatory response and apoptosis (Table 3). Different DMRs from CN and PT are localized in a binding site of the TF REST, with its target CACNA1H being down-regulated in both areas. CACNA1H was also targeted by EGR1, another TF that had a binding site in DMRs from both areas. In other cases, different DMRs in two brain areas, i.e. CN and OFC, were located in the binding site of the same TF, including SP1, that target CD44 which was downregulated in OFC but upregulated in CN. There were also exclusive TFs for all brain areas (Additional file 2: Table S7).

Table 3 Transcription factors (TFs) and targeted differentially expressed genes (DEGs) shared between the five brain areas
Full size table

Targeted DEGs were submitted to WebGestalt [23] revealing enrichment for immune response processes for all areas (Fig. 3). Regarding specific enrichments, we highlight interferon gamma signaling for ACC, ERK cascade for OFC, acetylcholine process for NAC, interleukin signaling for CN and regulation of DNA-templated transcription in response to stress for PT.

Fig. 3
figure 3

Enrichment results for Regulatory Networks for TFs from DMRs and targeted DEGs for the five brain areas

Full size image

DNA methylation age

We investigated the methylation age variations according to Horvath’s method [18]. In agreement with the chronological age, OFC, NAC, CN and PT from OCD had the DNAm age older than the respective areas of the control group. Only ACC presented DNAm age slightly younger than the chronological age for the OCD group. Although both AA difference and AA residuals presented a higher aging trend for the OCD group for almost all areas, the comparisons between OCD and control groups were not significant (Fig. 4).

Fig. 4
figure 4

Comparison of chronological and DNAm age for each brain area from OCD and control groups. A: Chronological age; B: DNAm age; C: AA Difference; D: AA Residuals

Full size image

Discussion

We explored DNA methylation and transcriptome data in post-mortem brain tissue associated with obsessive-compulsive disorder, searching for differences in the molecular mechanisms comparing OCD patients and controls. Regarding transcriptome analysis, there are some studies using blood and brain tissues, although for methylome analysis previous evidence comes from only peripheral assessment as blood [13, 14]. Due to the small sample size in our study, we searched DNA methylation differences using three approaches based on the assumptions that underlie DNA methylation studies: DMS, DMRs and gene modules as well as an integrative analysis of methylome and transcriptome data (Additional file 1: Fig. S2).

We did not find DMSs with a corrected p-value for any brain area, not an unexpected result considering our sample size. CpG sites with p-value < 0.0005 and methylation differences ≥20% between OCD and controls, ADARB2, UGT2B15, HLA-DQB1 and HLA-DQA1 were recurrently identified by the different DNA methylation approaches suggesting that these genes have a role in OCD. A DNA methylation study in saliva of OCD patients reported that differentially CpG sites were only evident when more symptomatic cases were used [24].

Gene networks from methylation and gene expression integrated data pointed to the involvement of the G-protein signaling pathway and small GTPase signal transduction. Components from the GPCR pathway are expressed at different levels in all physiological systems, including the nervous and immune systems [25]. GPCRs signaling regulate, among others, the actin–cytoskeleton dynamic by activating small GTPases [26]. Several genes from the actin binding processes that are modulated by epigenetic regulation [27] were associated with OCD in peripheral blood of patients [13]. Actin binding maintains and modulates dendritic spines, growth cone and axon guidance [28, 29]. Axons and dendrites contain a specialized transcriptome capable of producing synaptic proteins independently of the cell soma [30]. In our study, we observed enriched processes for neuronal and glial structure, synapse compounds or signal transduction. Usually, synaptic inputs reach neurons via dendrites, in a postsynaptic position. This information is processed by cellular machinery and the output goes to the axon, arriving at the presynaptic area [31]. In all brain areas, we identified enriched processes for presynaptic and postsynaptic alteration, and related to parts of these both mechanisms suggesting again that the quality of synapses, gap junctions and consequently receptors and information transmission could be altered as a consequence of the disruption of DNAm in the individuals with OCD.

A member of the GPCRs family, the GABAB receptor, was identified in CN. GABAB receptors produce slow and prolonged inhibitory signals via G proteins, interact with several neurotransmitter receptors and regulate receptor activity. These receptors are broadly expressed in the nervous system and both GABAB pre and postsynaptic reduce neurotransmitter release and hyperpolarizing neurons, respectively. Altered GABAB receptor function has been related with neurological and psychiatric disorders [32, 33]. Also, GABAB has a functional crosstalk with N-methyl-D-aspartate (NMDA), an ionotropic glutamate receptor enriched for CN. NMDA has a role in activity-dependent changes in synaptic plasticity [32]. Glutamate receptors are involved with physiological and pathological processes in the central nervous system as well as the efficiency of synaptic transmission [34, 35] and are associated with OCD [36], with direct and indirect pathways operating on CSTC circuitry [37]. Moreover, individuals with OCD have higher GABA levels as well as GABA/glutamate ratio compared to health controls [38].

Enriched processes identified in NAC not found in the other brain areas were related to acetylcholine receptor activity and cholinergic synaptic transmission. An animal model showed that cholinergic interneurons’ activity in NAC were associated with adaptive cue-motivated behavior, present in many psychiatric conditions [39]. Acetylcholine seems to not interfere with signal attenuation in the cholinergic system of compulsive behavior in a rat model of OCD [40]. Nevertheless, in pediatric obsessive-compulsive disorder, specifically in children with PANDAS (Pediatric autoimmune neuropsychiatric disorders associated with streptococcal infections), cholinergic interneurons activity in the striatum may contribute to pathophysiology in children with rapid-onset OCD symptoms [41].

Additionally, we explored the potential involvement of transcription factors that corroborated the previous findings, such as the involvement of the immune system, replicated in all enrichment analysis for all brain areas. Immunological and neuroinflammatory alterations have been associated with psychiatric disorders, such as OCD. The role of the immune system in the pathophysiology of OCD is mediated by different processes, i.e. interleukin signaling, interferon gamma and MHC receptor activity [42, 43], also replicated in our analysis. Regarding the involvement of cytokines in OCD, they affect the central nervous system by altering neurotransmitter systems. We observed enriched pathways for the anti-inflammatory cytokines IL-17 for PT and IL-4, IL-5, IL-10, IL-13 for CN, reinforcing the role of cytokines in the pathogenesis of OCD [44]. The proinflammatory cytokine interferon gamma (IFN-γ) was also altered in CN and OFC. MHC receptor activity involves genes that control polymorphic proteins from the immune system [55], including HLA-DPB1 that was associated with OCD in a GWAS study comparing OCD cases, healthy controls and combined parents-child trios [45]. Here, HLA-DPB1 appears as down-regulated in ACC and OFC.

Some cognitive functions, such as learning and memory, and behaviors, such as feeding and locomotory, were enriched in our analysis for ACC, OFC, NAC and PT. Also, for OFC, synaptic assembling included BDNF among the genes, whose role was related to brain volume by neuroimaging as well as cognition outcomes, including social behavioral changes in OFC-amygdala circuit [46, 47]. There is an increasing amount of evidence supporting the association between BDNF and OCD phenotype and neurobiology [16, 48, 49]. Although it is still unclear how OCD symptoms/dimensions might be related to the immune alterations, some inflammatory processes were associated with the psychopathology of OCD by compromising cognitive functions.

We also identified an enrichment for the MAPK cascade, which is related to innate and adaptive immune response [50]. This pathway is involved in the regulation of conservative mechanisms in eukaryotic cells, including apoptosis and cell differentiation [51], and can coordinate and respond differently to stimuli including hormones and receptors [52]. ERK-MAPK pathway mediates several mechanisms involved in the pathogenesis of OCD. Specifically, the TrkB/ERK-MAPK pathway overactivation contributes to restoring normal behavior in an animal model of OCD. Moreover, up-regulation of the Ras/ERK-MAPK pathway was involved in the development of OCD-related disorders [53]. ERK and RAS cascade reactions signal to MAPKs, activating signal transduction, transcription factors and regulating gene expression. ERK cascade, which was enriched in our integration analysis for OFC, CN and PT, is activated by different stimuli, such as G protein-coupled receptors. RAS, the most prominent member of the small GTPase family, is activated after receptor phosphorylation, propagating signals that result in the activation of sequential kinases, including ERK [33, 54]. Along with RAS, other small GTPase presented enrichment in our results in CN, the Rho protein signal transduction was enriched in NAC. RHO regulates cytoskeleton reorganization, cell cycle progression and MAP kinase signal transduction. Proteins from the Rho GTPase family are expressed in the central nervous system and involved in cytoskeletal plasticity, especially in the actin cytoskeleton, modulating cell and axon migration [33]. Other pathways involved with Ras/MAPK, such as PI3K and cAMP signaling pathways. PI3K interacts with RAS as one of its main effector pathways and cooperates with MAPK pathways inducing DNA synthesis [54]. The regulation of PI3K activity through insulin-related signaling was described as a key player in OCD etiology, affecting dendritic spine and synapse formation [55]. cAMP signaling, also enriched in ACC, OFC, CN and PT, regulates the activation of MAPKs and also plays the opposite role, blocking the MAPK pathway through the binding of Raf-1 to Ras [54]. cAMP levels and activity was altered in patients with OCD compared to controls [56]. cAMP also plays a structural role in dendritic spines and in the structural enlargement of spines. Additionally, postsynaptic cAMP mechanism enhances structural potentiation of spines playing a key role in the regulation of structural synaptic plasticity [29].

Finally, the immune response has a close relation with stress response [57], biological processes that were enriched in all brain areas except for PT. Stress conditions play an important role in disrupting neurotransmission, synaptic plasticity, and cognitive functions such as memory and learning [57, 58]. An increasing amount of evidence of accelerated epigenetic age in some patients with neurodegenerative, psychiatric and cardiovascular diseases has been previously described [18, 19, 59]. In agreement to the chronological age, we observed older DNAm age for OCD group, as well as a higher aging trend regarding AA difference and AA residuals, which also agrees with the enrichment finding of apoptotic processes for all areas, aging enriched for OFC, CN and PT, and cellular senescence enriched for PT. Although not significant, it was noteworthy that in disagreement with other areas, ACC was the only region that showed negative age acceleration (AA) residual and presented a younger DNAm age for OCD group. A positron emission tomography study reported inflammation at OCD’s neurocircuitry in several brain areas, including OFC, CN, PT, thalamus and ventral striatum, but not ACC [60]. In ACC, reduced glutathione levels were found with aging and brain atrophy in mice with EAAC1 mutation. EAAC1 is a transporter of cysteine, which is a precursor for neuronal glutathione synthesis [61]. In addition, glutamatergic system gene variants were associated with lower concentrations of glutamate in the CSTC circuitry, particularly ACC, of patients with OCD [62, 63].

It is important to note that as a case-control study, our findings could be a result of the disease rather than a cause. As psychiatric disorders present high comorbid and genetic correlations [64], possibly our findings are not exclusive for OCD, although we identified genes previously associated with OCD and used strict inclusion and exclusion criteria for patients selection, seeking to ensure high quality and reduce biases as best as possible [65]. Also, considering limitations associated with small sample size and post-mortem collection of clinical information, our results must be viewed as a preliminary contribution regarding brain methylation patterns in OCD.

Conclusions

Our findings confirm the involvement of previously associated genes and biological processes in OCD as well as report differences related to specific brain areas. These findings point to a role of cellular communication, inflammatory processes and behavior mediated by DNA methylation in OCD brain tissues. The main findings were related to the immune system, reaffirming the current literature findings about its involvement with OCD. We conclude that changes in DNA methylation are involved with OCD and further studies are needed to characterize alterations in different paths in each brain area.

Methods

Participants and OCD clinical assessment

Brain samples were retrieved from the Sao Paulo Autopsy Service from the University of São Paulo as part of the psychiatric disorders collection of the Biobank for Aging Studies. Three psychiatrists performed a screening interview encompassing clinical, functional, cognitive and psychiatric parameters (detailed in [66]) to a family member or close caregiver who had at least weekly contact with the deceased. For individuals with a presumably diagnosis of OCD, a second clinical evaluation with the same informant was done for the confirmation of the diagnosis. The complete evaluation comprised the Structured Clinical Interview for DSM IV Axis I disorder (SCID) [67], the Yale- Brown Obsessive-compulsive scale (Y-BOCS) [68] and a short version of DY-BOCS [69]. The questionnaire was modified to ask the questions about the deceased case. As standard procedure in the second evaluation, informants showed the medication (or medication boxes) that were used. After the second assessment, a best estimated diagnosis procedure was done by two psychiatrists [9, 65]. Briefly, from 109 individuals, 72 were assigned to the psychiatric group, being 22 diagnosed with OCD by clinical parameters and 50 diagnosed with other psychiatric disorders (i.e. bipolar disorder, major depression, Tourette syndrome, schizophrenia and others). Most individuals were unmedicated to psychiatric medications. Thirty-seven individuals did not fulfill psychiatric diagnosis criteria. Of the 22 OCD cases, eight had the best estimated diagnosis and were included in this study (Additional file 1: Table S8). Eight controls without any psychiatry diagnosis, matched by age, sex and brain’s hemisphere were selected from the same Biobank. All individuals were 50 years of age or older, without a history of clinical dementia and no other clinical comorbidity which could result in hypoxia or brain autolysis (as high postmortem hour, hospitalization with mechanical breathing, chronic obstructive pulmonary disease, kidney failure and previous cerebrovascular accident). These individuals were considered as healthy controls in the study. Brain tissues from both OCD cases and healthy controls were collected from cortico-striatal areas ACC, OFC, NAC, CN and PT (Additional file 1: Fig. S3). More detailed information concerning sample collection and the complete evaluation can be found elsewhere [9, 65, 70].

DNA methylation data

DNAm profiling and data quality control

Brain samples were dissected to isolate the cortico-striatal areas and preserved in cryotubes at − 80 °C. DNA was extracted using QIAsymphony DNA Kit (Qiagen, Hilden, Mettmann, Germany) on QIAsymphony platform, according to the manufacturer’s instructions. DNA was bisulfite-converted using the EZ DNA Methylation kit (Zymo Research, Irvine, California, USA), according to manufacturer’s instructions and hybridized in the Illumina Infinium HumanMethylation450 BeadChip array (Illumina Inc., San Diego, California, USA). Raw data were extracted by the iScan SQ scanner (Illumina) using GenomeStudio software (v.2011.1), with the methylation module v.1.9.0 (Illumina), into IDAT files, which were imported to R statistical environment using minfi package [71]. Quality control steps removed 16,217 probes associated with SNPs (Single nucleotide polymorphism) (that contained SNPs at the CpG interrogation site or at the single nucleotide extension), 10,007 probes with unreliable measurements (p > 0.05), 26,357 probes located in specific contexts (non-CpG sites) and 10,871 probes located in the sexual chromosomes. All samples passed quality control parameters resulting in 422,060 probes to be analyzed. Background was corrected using the noob method [72] and cell composition was estimated using FlowSorted.DLPFC.450 k function [73] as implemented in the minfi package [71]. Cell composition was evaluated for each brain region considering the groups to be compared. We used the compareGroups package [74] that pointed to OFC as presenting differences in cell composition. Thus, cell composition was considered as a variable to be corrected only when OCD cases and health controls were compared for the OFC region. ChAMP package was used to identify and correct batch effects and biological variables (sex) [75, 76]. Distribution of Infinium I and II probes fluorescence measurements was normalized by functional normalization (FunNorm) method [77].

Identification of DNAm changes related to OCD and methylation clock

For each brain region, we applied the linear model function from limma [78] to M-values (loggit of B-values) to identify differentially methylated CpG sites (DMSs) [79]. As parameters, we considered adjusted p-value (adjP) ≤0.05 after multiple testing corrections using the Benjamini and Hochberg method. Age was included as a variable for all brain areas and cellular composition was included only for OFC. As multiple CpG sites may map to the same gene, and the effect of DNAm can be dependent on its location in relation to the gene (promoter, body), we used Functional Epigenetic Modules (FEM) analysis. The FEM package [80], as implemented in ChAMP, identifies subnetworks (protein interaction modules - PPI) where a significant number of members exhibit differential DNAm in relation to the phenotype of interest. To identify differentially methylated regions (DMRs), DMRcate was applied to the methylation values [81], considered significant those with FDR (False Discovery Rate) < 0.05 and methylation differences (delta beta, Δβ) ≥10%. DNAm age was calculated with the beta values using Horvath’s method [18]. The algorithm also calculated de age acceleration (AA), which can be used to determine how fast tissues are aging, i.e., whether the DNAm age of a given tissue is consistently higher (or lower) than expected [18, 82].

Transcriptome data

The list of differentially expressed genes (DEG) comparing OCD cases and controls from NAC, CN and PT was retrieved from a previously published study by our group [9]. For ACC and OFC, RNASeq data were preprocessed using the same parameters [9]. ACC dataset was composed of 8 controls and 8 OCD samples and OFC for 5 and 6 controls and OCD samples, respectively. Any ribosomal RNA that bypassed the depletion process from the read counts of both the ACC and OFC regions was removed. To filter genes that were lowly expressed, we converted the read counts to counts per million (CPM) values using the edgeR library (v.3.28.1) [83], choosing only transcripts that had at least 0.3 CPM in 50% for a group (controls or OCD). For both datasets, we used the DESeq2 library (v.1.26) [84] to normalize the raw counts with the rlog function and the surrogate variable analysis (SVA) library (v.3.34) [85] to estimate any hidden covariates, using sex, age, laboratory batch and OCD diagnostic as our variables of interest. The DESeq2 were also used to perform the differential expression analysis, using as covariates in the linear model the same variables of interest with the addition of surrogate variables estimated by the SVA for each brain region (3 surrogate variables for the ACC and 2 surrogate variables for the OFC).

Bioinformatic integrative analysis

Integrative analysis considering DEGs and methylation differences (CpG sites mapped to genes)

For the five brain areas, we merged the three datasets (DMR and FEM genes list from methylation analysis and DEGs from the RNASeq analysis) and use the STRING Database [86] to construct a PPI network for each brain area, where, we used all active interaction sources with our list of genes, selecting only the edges with a high confidence score (≥0.7). From this PPI network, we selected edges that connected nodes from 2 different sources (DMR, DEG or FEM lists), to remove nodes that were exclusively connected with its own DMR, DEG or FEM dataset and to guarantee that for each PPI network, we had the major connected component integrating both transcriptomic and methylation data. To calculate network centralities metrics, we used the NetworkX library (v.1.9.1), choosing the metrics: Degree - evaluate the number of connections from each node; Neighbor degree - the number of the degrees from each node’s immediate neighbor; Closeness - also a measure of the influence of a node in the network, but based on its node’s positioning; Eigenvector - measure the global influence from a node [87]; KATZ centrality - a variation from the Eigenvector centrality used on social networks, capable to measure both the local and the global influence of a node; Betweenness - measures the number of paths between a pair nodes that cross a given node. A gene high betweenness might be capable to connect isolated elements from a network; Clustering Coefficient - measure the number of triangles in the network, in other words, the number of immediate neighbors that are also connected between each other [88]. The selected nodes were re-submitted to STRING [86] (using the same parameters) and PPI enrichment analysis was applied to evaluate if the brain region networks have more interactions, considering both GO and REACTOME, non-redundant ontologies/pathways with a minimum of 5 genes.

Integrative analysis considering combination between transcription factors annotated from DMRs (not mapped to genes) and DEGs

After identifying the DMRs, they were annotated for transcription factors (TFs) using ENCODE (Encyclopedia of DNA Elements) [22], wgEncodeRegTfbsClusteredV3 track available from Genome Browser’s Data Integrator [89]. To match the TFs associated with the DMRs to the DEGs previously identified for our dataset [9], we used the TRRUST v2 database [90]. Additionally, annotated TFs and its targets were used to construct regulatory networks, indicating the type of connection (activation, repression or unknown interaction). Targeted DEGs were submitted to WebGestalt [23] for GO and REACTOME enrichment analysis, considering only non-redundant ontologies with a minimum of 5 genes related and FDR < 0.05.