Association of a History of Child Abuse With Impaired Myelination in the Anterior Cingulate Cortex: Convergent Epigenetic, Transcriptional, and Morphological Evidence

Child abuse is a major public health problem, affecting 5%−15% of all children in the Western world (1). Considerable clinical and epidemiological evidence shows that early-life adversity significantly increases the lifetime risk for stress-related psychiatric disorders, including depression and suicide. In particular, it has been suggested that this relationship may be mediated by various traits, including neuroticism (2), high levels of impulsivity and aggressive behaviors (3), and high anxiousness trajectories (4), which are typically observed in depressed individuals with a history of abuse. However, the neurobiological processes underlying this heightened vulnerability remain unclear.

Brain imaging studies have described brain network activity and structural abnormalities associated with child abuse, which may ultimately affect various aspects of cognitive and emotional processing (1). In particular, the anterior cingulate cortex, a critical region for the regulation of mood states, is among the brain areas most frequently associated with impairments in individuals with a history of child abuse. These include evidence of cortical thinning, alterations in network centrality and functional connectivity, and disruption of white matter integrity (reviewed in reference 1). While these studies bring invaluable insight into macrostructural adaptations associated with child abuse, underlying cellular and molecular alterations are largely unknown.

Recently, epigenetic regulation of gene expression has emerged as a mechanism of long-term genomic plasticity that has the potential to explain how early experiences may drive behavioral changes over the lifetime (5). In particular, accumulating data suggest that patterns of DNA methylation, a major epigenetic mark, are progressively established during brain development and may be disrupted by early-life adversity (6). We therefore hypothesized that child abuse, by affecting the epigenetic programming of key developmental processes, may lead to long-term structural and functional changes that potentiate the vulnerability to psychopathology.

In the present study, focusing on the anterior cingulate cortex, we used a cohort of subjects with a history of severe child abuse (assessed through psychological autopsies [7]). We first performed genome-wide screenings of DNA methylation and gene expression and found that child abuse was associated with a severe disruption of oligodendrocyte function. We further demonstrated that part of the epigenetic adaptations in abused subjects selectively occurred in oligodendrocyte-lineage cells, but not in neurons. Notably, these effects were absent in depressed suicide completers with no history of child abuse, strongly suggesting that they are not accounted for by psychopathology. Furthermore, we provide evidence showing that offspring of dams with low maternal care, a well-investigated rat model of early-life environmental variation, present myelin gene expression changes that correlate with the effects of child abuse in humans.

Finally, associated cellular and subcellular changes were assessed using stereology and coherent anti-Stokes Raman scattering (CARS) microscopy, representing the first high-throughput analysis of individual myelinated axons in human subjects. The results demonstrate that child abuse selectively affects small-diameter axons and their myelin sheaths.

Altogether, our studies reveal with unprecedented resolution the impact of child abuse on the anterior cingulate cortex epigenomic and histopathological architectures, and unveil oligodendrocytes and myelination as potential major substrates mediating the long-term consequences of child abuse.

Method

Brain Samples

Anterior cingulate cortex tissue was obtained from the Douglas-Bell Canada Brain Bank from 1) subjects who died suddenly without prolonged agonal state or medical illness, and with no psychiatric history (control group); 2) subjects who died by suicide in the context of a major depressive episode and had a history of severe child abuse (child abuse group); and 3) subjects who died by suicide in the context of a major depressive episode and had no history of child abuse (depressed group). The groups were matched for age, postmortem interval, and brain pH (8) (see the data supplement that accompanies the online edition of this article). Psychological autopsies were performed as described previously (9), with diagnoses assigned based on DSM-IV criteria. Characterization of early-life history was based on adapted Childhood Experiences of Care and Abuse interviews (10). We considered as severe adversity reports of nonrandom major physical and/or sexual abuse, up to age 15. Only cases with the maximum severity ratings of 1 and 2 were included. Histories of abuse were validated with available reports from medical charts, coroner files, or reports from youth protection services. Cingulate cortex tissue dissections and the rat model are described in the online data supplement.

Reduced Representation Bisulfite Sequencing (RRBS)

Libraries were prepared as described previously (see reference 11, the data supplement, and Table S1 in the data supplement).

Differential methylation.

To identify differentially methylated regions, we used a windows approach defined by clusters of CG sites. Windows were defined using Bumphunter (http://bioconductor.org/packages/release/bioc/html/bumphunter.html) as any CG within 50 bp of another CG, with no limit on the number of CGs in a given window but with a minimum of two. The CGs included in differential methylation analyses were determined as follows: only CGs with ≥5X coverage were included; 2) only CGs informed in >15 subjects in each group were included; 3) for each subject contributing to a given window, at least two-thirds of the CGs in that window were informed; and 4) the 0.1% of CGs showing the highest coverage were discarded as potential polymerase chain reaction (PCR) amplification.

Fluorescence-Activated Cell Sorting (FACS)

Nuclei preparation.

Frozen samples were homogenized, layered on a cushion of 0.8-M sucrose solution, and centrifuged at 2500 g (20 minutes at 4°C). Pellets were resuspended in blocking buffer containing primary antibodies against Sox10 (1:100; R&D Systems, Minneapolis) and NeuN (1:500; Alexa 700-conjugated; Novus, Littleton, Colo.). For the detection of Sox10+ nuclei, nuclei were incubated with a fluorophore-conjugated secondary antibody (Alexa Fluor 488, 1:500). Vybrant DyeCycle Violet stain was added to the nuclei solution.

Cell sorting.

Using a FACSAria Fusion (BD Biosciences, San Jose, Calif.), a gating was applied to filter singlets using physical parameters and violet fluorescence (405-nm laser, 525/50 filter). Nonoverlapping gates were adjusted to collect neuronal nuclei based on NeuN-Alexa 700 immunoreactivity (640-nm laser, 730/45 filter), and oligodendrocyte nuclei based on Sox10-Alexa 488 immunoreactivity (488-nm laser, 530/30 filter).

Targeted Bisulfite Sequencing on FACS-Sorted Nuclei

DNA extractions.

After overnight proteinase K digestion, samples were mixed with 770 μL of 20% PEG 8000 2.5-M NaCl and 200 μL of Agencourt AMPureXP Beads for every 200K nuclei. DNA extraction was subsequently conducted using a DynaMag-5 magnetic stand.

Library preparation.

We used our PCR-based library preparation procedure (12) (see the online data supplement and Tables S2–S4). Libraries were quantified using an Agilent TapeStation, pooled, diluted to 8 pM, and sequenced on an Illumina MiSeq.

Bioinformatic analyses.

After trimming universal primers, reads with a Phred quality score below 20 were discarded. Reads were aligned to the human genome hg19. CGs with a coverage lower than 10X were discarded.

RNA Sequencing

Only samples with RNA integrity numbers above 5 were used. Libraries were prepared at the Genome Quebec Innovation Center using the TrueSeq Stranded Total RNA Sample Preparation and Ribo-Zero Gold kits (Illumina) and quantified using PCR (KAPA Library Quantification). Three libraries were run per lane of an Illumina HiSEquation 2000 (100-bp paired-end), yielding ∼62 million reads/library (see Table S5 in the data supplement).

Differential expression analysis.

We used FASTX-Toolkit and Trimmomatic for adapter trimming, TopHat (version 2.1.0) for alignment, and HTSeq-count for gene-level quantification. Differential expression was analyzed using DESeq2, and validations were conducted using reverse transcription PCR and NanoString.

Gene set enrichment analysis.

As previously described (13), log₂ fold changes were obtained from the differential expression analysis. Genes were ranked based on fold changes. The GSEAPreranked tool and the C2 curated gene sets were used.

CARS Microscopy

Frozen blocks of white matter adjacent to the anterior cingulate cortex were fixed in 10% formalin, and 300-µm thick sections cut on a vibratome. A custom-built video-rate laser scanning microscope and two laser sources were used to probe CH₂ stretching bands in myelin (2,845/cm). To extract nerve fiber morphometric information, a segmentation algorithm (14) was applied. For each fiber, standard morphometric measurements were computed from the segmentation results using MATLAB (MathWorks, Natick, Mass.).

Statistical Analysis

Differential methylation analysis (RRBS) was performed using a general linear model with age and gender as covariates and applying the Benjamini-Hochberg correction. For differential expression analysis (RNA sequencing), DESeq2 was used to perform general linear models with age, gender, and RNA integrity number as covariates. NanoString data were analyzed similarly with general linear models using child abuse and psychopathology as fixed factors. For DNA methylation in FACS-sorted nuclei, cellular densities after immunohistochemistry, and CARS morphometry analyses, one-way and two-way analyses of variance were used, followed by Tukey’s honest significant difference post hoc tests. The significance threshold was set to 0.05.

Results

Epigenetic Effects of Child Abuse

Levels of DNA methylation were assessed with RRBS in the child abuse (N=27) and control (N=26) groups, covering on average 696,000 (SEM=40,000) CG dinucleotides per subject (see Figure S1 in the data supplement). Widespread differences were uncovered between groups (Figure 1A) (see also Figure S1E,F), and 115 genomic windows passed multiple-testing corrections (Benjamini-Hochberg, q<0.1). Among these, both hyper- and hypomethylation were detected in the child abuse group compared with the control group (see Figure S1F), suggesting that child abuse bidirectionally regulates epigenetic patterns in the cingulate cortex, as previously observed in the hippocampus (7). Strikingly, we noted that the three most significantly differentially methylated regions intersected with genes directly related to myelin and oligodendrocytes: LINGO3, of the LINGO family of proteins implicated in myelination (16); POU3F1, a transcription factor controlling myelination (17); and ITGB1, one of the integrins that mediate interactions between oligodendrocytes and axons (18).

Recent animal and human data suggest that DNA methylation patterns are cell type specific (6), and we therefore hypothesized that child abuse may selectively regulate oligodendrocytes through epigenetic processes. We conducted FACS sorting of nuclei from oligodendrocyte-lineage and neuronal cells (Figure 1B) (see also Figure S2 in the data supplement), using independent samples taken from adjacent tissue from the same subjects included in the RRBS study, and processed them through a targeted bisulfite sequencing analysis of DNA methylation in our top three RRBS differentially methylated regions. To control for the possible effects of depressive psychopathology and suicide, we included a third group of depressed individuals who died by suicide but had no history of child abuse (depressed group, N=25).

Interestingly, for both LINGO3 and POU3F1 (Figure 1C–H) (see also Figure S3 in the data supplement), decreased methylation in the child abuse group was confirmed and found to occur specifically in oligodendrocyte, but not neuronal, nuclei. These effects were absent in the depressed group. We found, however, no significant DNA methylation differences for ITGB1 in either Sox10+ or NeuN+ nuclei between groups.

Altogether, therefore, child abuse, but not suicide or depressive psychopathology, accounted for oligodendrocyte-lineage specific decreased DNA methylation in LINGO3 and POU3F1 genes, suggesting oligodendrocyte-specific epigenetic reprogramming as a result of child abuse.

Transcriptomic Effects of Child Abuse

We used RNA sequencing to further characterize genomic consequences of child abuse. Similar to the RRBS data, widespread differences were observed between the control and child abuse groups (see Figure S4E,F in the data supplement), with 3,734 differentially expressed genes (out of the 20,893 genes interrogated, i.e., 17.9% of p values <0.05 with no multiple testing correction). Compared with this genome-wide frequency, a list of 55 genes related to myelin and oligodendrocytes showed a strong enrichment for dysregulation (35/55, 63.6%) (see Figure S4F), which was confirmed by permutation testing (p<10⁻⁶) (see Figure S5 in the data supplement). Among these 55 genes, 32 genes were down-regulated and three up-regulated in the child abuse group (32/55 genes, Figure 2A–D). This enrichment for down-regulation was further confirmed by gene set enrichment analysis (enrichment score: −0.753; false discovery rate q<0.01) (Figure 2D). Down-regulations affected genes that are essential constituents of myelin (Figure 2A) and that critically control the synthesis of myelin lipids (Figure 2B) or the differentiation of oligodendrocytes (Figure 2C). Of note, this pattern of global down-regulation was absent in RNA sequencing data generated using tissue from the amygdala (see Figure S8 in the data supplement), suggesting that abuse-induced dysregulation of myelination may exhibit brain region specificity.

Focusing on the top differentially methylated genes identified by RRBS, we found that ITGB1 was strongly down-regulated in the child abuse group (Figure 2B), while the expression of LINGO3 (Figure 2B) and POU3F1 (Figure 2C) was not affected. Integrins are cell-cell interaction proteins, and numerous studies have shown that ITGB1, when expressed by oligodendrocytes, promotes myelination (18) by forming heterodimers with ITGA6 and ITGAV (20). Of note, the two latter genes were also down-regulated in the child abuse group (see Figure S7 in the data supplement), suggesting impaired communication between oligodendrocytes and nearby neurons or the extracellular matrix.

RNA sequencing results were validated using NanoString, and a strongly significant correlation was found between the two technologies (see Figure S4G). Notably, the global down-regulation of myelin genes was virtually absent in the depressed group (Figure 2E), suggesting that child abuse, but not depressive psychopathology, is specifically associated with impaired transcription of essential myelin genes.

To complement our findings, we used a well-established rodent model of maternal care (21) to measure how variation in early-life environmental experiences affects myelin in controlled experimental settings. We compared adult offspring of dams displaying high versus low levels of maternal behavior (high or low licking and grooming; see the data supplement) and quantified in the cingulate cortex the expression of the myelin-related genes that we observed differentially expressed in the human subjects as a function of child abuse. We observed a strong correlation (p<0.0001) between myelin gene expression changes in rats raised by low licking and grooming dams and expression changes in human subjects who had experienced child abuse (see Figure S9 in the data supplement). Although this model does not capture the whole spectrum of childhood abuse in humans, these results nevertheless strengthen the notion that myelin transcription is sensitive to the early-life environment.

Effects of Child Abuse on Oligodendrocyte Density

Genomic myelin alterations may reflect modifications in numbers of oligodendrocytes. Using stereology, we found that although the density of oligodendrocyte-lineage Sox10+ cells was similar between groups in the cingulate cortex gray matter (Figure 3B), compared with the control group it was significantly reduced in the cingulate cortex white matter of the child abuse group but not of the depressed group (Figure 3C).

In the adult brain, mature oligodendrocytes originate from oligodendrocyte progenitors. To investigate whether the effects of child abuse may stem from impaired proliferation in the oligodendrocyte lineage, we used the PDGFRα marker. We found that the density of PDGFRα+ cells was not different between groups (Figure 3C), suggesting that early-life adversity does not affect the proliferation of oligodendrocyte progenitors but reduces the pool of more mature myelinating oligodendrocytes in the cingulate cortex white matter.

Effects of Child Abuse on Myelin Ultrastructure

Molecular changes affecting myelin and associated with child abuse likely reflect the reduced number of oligodendrocytes observed in the cingulate cortex white matter of these individuals. To explore whether these changes may affect myelin integrity and abundance, we conducted a high-throughput morphometric analysis of myelin using CARS microscopy (Figure 4A–C) (25). We implemented this technique for its use in human brain tissue (see Figure S10A–C and Table S10 in the data supplement), and analyzed a total of 18,949 axonal fibers with similar cross-sectional orientations. As expected, axonal diameter was significantly correlated with g-ratio, the commonly used myelination index (Figure 4D) (see also Figure S12 in the data supplement).

The results showed a moderate but significant decrease in axonal diameter in the child abuse group (Figure 4E) compared with the two other groups. Furthermore, in agreement with the down-regulation of myelin genes described above, we measured a decrease in average myelin thickness that was specific to the child abuse group (Figure 4F). Finally, we found that the g-ratio was significantly higher in the child abuse group (Figure 4G). Altogether, the decrease in axon diameter associated with child abuse was accompanied by an even stronger decrease in myelin thickness, resulting in an increased g-ratio (Figure 4H). These results suggest that both axonal morphology and myelination are affected by early-life adversity.

Axon caliber and myelin sheath thickness are tightly regulated properties that reflect the length, conduction velocity, and firing rate of individual fibers (26), as well as their neuroanatomical localization (27). To gain deeper insight into the impact of child abuse on myelination, we categorized fibers by axonal diameter. Notably, the two smallest-caliber categories (0.5–1.25 μm), encompassing more than 50% of all fibers, showed reduced myelin thickness specifically in the child abuse group (Figure 4K). Accordingly, the g-ratio was increased in the child abuse group for these two categories (Figure 4L), indicating that early-life adversity preferentially affects small-diameter axons in the cingulate cortex. Finally, we note that child abuse was also associated with increased diameter in the largest axons (>1.75 μm, Figure 4J), which likely correspond to long-range subcortical connectivity of the anterior cingulate cortex (28). We speculate that the latter changes, which do not have an impact on the g-ratio of these fibers, may nevertheless alter functional coupling between the cingulate cortex and subcortical structures such as the amygdala and nucleus accumbens and contribute to altered emotional processing in abused subjects.

Discussion

MRI studies have suggested that early-life adversity may be associated with structural alterations in white matter (1). A recent report further provided evidence that interventions, such as placement in high-quality foster care, may partly counteract some of these effects, suggesting that white matter deficits could be reversed to alleviate lifelong consequences of early-life adversity (29). MRI technologies, however, offer limited specificity and spatial resolution: changes in diffusion tensor imaging fractional anisotropy, for example, may be related to changes in myelination, axonal number, axonal diameter, gliosis, or tissue edema (30). Most recent MRI strategies are able to more specifically interrogate myelin content (31), but they still only allow, at best, characterization of bundles of hundreds or thousands of axons. Here, we used state-of-the-art microscopy to precisely measure microstructural changes associated with early-life adversity. Our results provide direct evidence that axonal integrity and myelination of individual fibers may be disrupted in adults who have a history of abuse during childhood. Importantly, they suggest that this phenomenon predominantly affects small-diameter axons, possibly corresponding to cortico-cortical projections (32), while leaving other axonal populations unaffected. We propose that these selective myelin adaptations may provide a microstructural explanation for white matter defects previously detected by brain imaging studies of child abuse.

Beyond white matter myelin dysregulation, our results indicate that early-life adversity is also associated with profound epigenetic and transcriptomic alterations that affect oligodendrocytes located in the cingulate cortex gray matter. Using FACS, we conducted the first analysis of DNA methylation in human oligodendrocytes and showed that cell type–specific epigenetic reprogramming may account, at least in part, for child abuse effects on myelin. Accordingly, for POU3F1 and LINGO3, the results indicated that DNA methylation levels were specifically decreased in oligodendrocytes, but not neurons. Oligodendrocyte-specific differential methylation of these two genes had no detectable effect on their transcriptional activity at the whole tissue level (RNA sequencing data). Because of the dynamic relationship between DNA methylation and gene expression during brain development (33), it is possible that these two adaptations represent epigenetic “traces” of child abuse that no longer have an impact on gene expression in adulthood.

Ultimately, we found that individuals abused during childhood showed decreased expression of a large collection of myelin-related genes in the cingulate cortex, suggesting a strong impairment in oligodendrocyte function. Strikingly, this global down-regulation was completely absent in depressed suicide completers with no history of child abuse, and it correlated nicely with gene expression changes observed in an animal model of unfavorable maternal care. Altogether, these results document the crucial role of early-life experiences in setting appropriate myelin genes' activity in the cingulate cortex. They also suggest a pathophysiological model whereby global cingulate cortex myelin down-regulation following a history of child abuse is not necessary for the emergence of depression. Rather, it may represent a neurobiological vulnerability that results from child abuse and potentiates the risk of psychopathology throughout life in affected individuals, thereby contributing to higher impulsive, aggressive, or anxiety traits and more frequent substance use and depressive disorders (3). Finally, an important question raised by these findings is whether myelin changes also occur in resilient individuals, that is, those who remain psychiatrically healthy despite adverse experiences in early life.

A limited number of postmortem studies on depression have reported transcriptional dysregulation of myelin-related genes in the temporal (34) and prefrontal cortices (35, 36). Of note, these studies were conducted in cohorts that were not characterized for early-life experiences of care and abuse. In the present work, depressive symptoms and suicide were not associated with any changes in the expression of myelin genes in a third brain structure, the cingulate cortex. Therefore, our findings highlight the need for further research to disentangle potential brain region–specific effects of child abuse and depression on myelin.

Recent preclinical evidence clearly supports the essential contribution of the quality of early-life experiences to cortical myelination. Accordingly, postweaning social isolation in rodents leads to decreased myelination in the frontal cortex (37, 38) that does not recover after social reintegration, indicating that myelination is critically regulated in early life by the social environment. Our own results are concordant with these findings and suggest that in humans, early-life experiences also have an impact on myelination, at least in the cingulate cortex. Because myelination has been shown to adapt to neuronal activity during motor learning (39), we speculate that this phenomenon may be similarly modulated in the social brain by the quality of life experiences. By threatening the development of healthy attachment patterns, early-life adversity may lastingly disrupt oligodendrocyte differentiation, cellular density, and transcriptional programming within the cingulate cortex, in part through epigenetic processes. Where in the brain, when during development, and how at the molecular level these effects are necessary and sufficient to have an impact on affective regulation cannot, by definition, be determined through human postmortem studies but hopefully will be explored in animal models by conducting time-controlled, brain region–specific manipulations of myelination. Ultimately, we propose that persistent myelin alterations induced in the cingulate cortex by child abuse may contribute to behavioral dysregulation and the emergence of interpersonal difficulties (40), thereby potentiating the risk of depression and suicide.