Introduction

The exposure to chronic and severe negative life experiences during early childhood is associated with the development of a host of physical and mental health problems later in life1. Adverse childhood experiences (ACEs) include physical, sexual and verbal abuse, physical and emotional neglect, witnessing violence at home, a family member suffering from addictions, mental health issues or incarcerated, and losing a parent to separation, divorce or other reason2. Children who have experienced four or more ACEs are more likely to develop long lasting health issues such as diabetes, heart disease, overweight or obesity, cancer, respiratory disease, mental health conditions, alcohol and drug abuse, interpersonal and self-directed violence and sexual risk taking3.

There is growing evidence suggesting that epigenetic modulation is one of the molecular mechanisms through which stressors interact with the genome. Epigenetic information regulates gene expression and, although relatively stable, the epigenetic landscape is highly sensitive to environmental exposures4,5. DNA methylation is one of the most widely studied epigenetic modifications in which a methyl group is added to a cytosine residue, most commonly in the context of cytosine-guanine dinucleotides (CpG). Children exposed to severe adversity show DNA methylation changes in genes involved in the vulnerability to stress, neurotransmission, inflammatory responses and behavior6,7,8,9,10. Negative childhood exposures can trigger DNA methylation changes in genes that modulate anxiety and related phenotypes, such as the oxytocin receptor, glucocorticoid receptor, serotonin transporter gene, brain-derived neurotrophic factor and glutamate receptor11,12,13,14,15. Early-life maternal and paternal stressors are predictive of DNA methylation changes detected in adolescents16 and both ACEs and DNA methylation changes at the glucocorticoid receptor gene have been associated with increased risk of psychopathologies during adolescence17. Moreover, adverse experiences have been associated to an accelerated biological aging18. The deviation between the DNA methylation age and the chronological age is a measure of the epigenetic aging rate19,20. In children, the Pediatric-Buccal-Epigenetic (PedBE) clock is a tool to measure the biological age, providing an understanding of the environmental exposures that might influence child health and disease21. Recent findings show that psychologically adverse or violent home environments can accelerate epigenetic aging in youth22. Similarly, neighborhood violence or elevated parental depressive symptoms have been associated with both emotional distress and accelerated epigenetic aging in children18,23,24,25. Importantly, an accelerated rate of epigenetic aging predicts the risk of many chronic conditions such as obesity, cancer, Alzheimer's disease, cardiovascular disease, and all-cause mortality risk26.

Recent research shows that positive childhood experiences predict positive outcomes in long-term health and can also neutralize the negative impact of ACEs on adult health27. In this context, interventions to increase awareness and understanding of childhood adversities and to promote family connection have been proposed as strategies to influence health and well-being later in life28,29,30. In addition, multimodal programs that combine several approaches such as cognitive behavioral therapy, exercise, yoga, music, art, EMDR (Eye Movement Desensitization and Reprocessing) therapy, individual counselling and interactions with animals have been proposed to improve wellbeing and mental health in child victims of multiple ACEs31,32. Notably, in rodents, an enriched environmental model, which includes cognitive, somatosensorial, motor and visual stimulation, reduces the negative psychological and behavioral consequences of early adversity by modulating trauma-related epigenetic marks and improving neurogenesis and synaptic plasticity33,34,35.

We recently described the protocol and mental health impact of a 1-week multimodal intervention group (n = 44 girls) program for adolescents (aged 13–16 years) reporting 4 or more ACEs36. After completing the program, the intervention group showed significant reduction in trauma-related outcomes (− 73% in the Short PTSD Rating Interview (SPRINT) scale; − 26% in the Child PTSD Symptom Scale (CPSS)) and a 57% improvement in attention/awareness-related outcomes Mindful Attention Awareness Scale-Adolescents (MAAS-A). This program addresses trauma through evidence-based therapeutic approaches, in an enriched environment that provides social, somatosensory and cognitive stimulation. Based on the literature discussed above, we hypothesize that these conditions may trigger DNA methylation changes in genes involved in the pathophysiology of multiple ACEs, such as vulnerability to stress, neurotransmission, inflammatory responses, behavior and cell aging. We hypothesize that some of the DNA methylation changes may correlate with the mental health improvements that we have previously reported in the same sample36, providing insights for future mechanistic research. To start testing this hypothesis, we profiled genome-wide DNA methylation levels in saliva samples from control and intervention group participants, at baseline (T1) and post-intervention (T2), in order to detect potential physiologically relevant DNA methylation changes.

Results

Intensive multimodal 1-week group program causes genome-wide alterations in DNA methylation

To identify the impact of the intervention on DNA methylation levels at each CpG on the Human MethylationEPIC array (N =  > 850,000 sites), we used an ANCOVA model adjusting for DNA methylation level at baseline (T1), BMI, age, ACEs score and cell type proportions (see details in “Methods”). This approach revealed that 49 DML exhibited a p value < 0.001 and a change in DNA methylation level greater than 10% (Table 1), while 195 DML showed a p value < 0.001 and a change in DNA methylation level greater than 5% (Supplementary Table S1).

Table 1 Intervention-sensitive differentially methylated loci (DML) with p-value lower than 0.001 and a DNA mean difference (T2–T1) of 10% or more (n = 49).
Full size table

Out of the 49 DML, 87% showed an increase in DNA methylation level from baseline to post-treatment and 37 DML reside in known genes. These 49 DML were distributed across all human chromosomes except the Y chromosome (Fig. 1a) and were most often found within gene bodies (57%), followed by 5′ untranslated regions (27%), and gene promoter regions of genes (up to 1500 basepairs upstream of the gene transcription start site) (16%) (Fig. 1b). Most of the DML were in open sea regions (more than 4 kb from a CpG island) (64%) and 12% were located within CpG islands (Fig. 1c). Considering the probe locations included on the array, the genomic region and location enrichments of the DML were not significant (p value > 0.05).

Figure 1
figure 1

(a) Manhattan plot of intervention-sensitive differentially methylated loci (DML). The X-axis represents the chromosomal position and the Y-axis represents the significance on a − log 10 scale. The red and dashed line indicates the threshold for the cut-off p value < 0.001 and DNA methylation mean difference (T2−T1) of 10% or more; (b) Percent distribution to standard genomic features of DML with available information (n = 37). 5′UTR = 5′ untranslated region’ 3′UTR = 3′ untranslated region; TSS = transcription start site; TSS200 = 0–200 bp upstream of TSS; 44 TSS1500 = 200–1500 bp upstream of TSS to standard genomic features; (c) Percent distribution of intervention-sensitive DML (n = 49) to island relative positions. Shores are considered regions more than 4 kb from CpG islands, shelves are regions 2–4 kb from CpG islands, and other/open sea regions are isolated CpG sites in the genome that do not have a specific designation.

Full size image

Functional roles of intervention-sensitive DML

Using a meta-database restricted to the 49 DMLs to identify molecular interactions for network biology (ConsensusPathDB-human tool), we conducted a pathway analysis and found a significant enrichment of functional interactions associated with the nervous, endocrine, immune systems, and processes involved in cancer, diabetes and cardiovascular disease (top 20 pathways with FDR q-value < 0.03, Table 2; all pathways with FDR q-value ≤ 0.05, Supplementary Table S2). These findings support links to neurophysiological processes affected by childhood adversity3.

Table 2 Top 20 functional interactions of the 49 meditation-sensitive DML (p value < 0.001 and mean difference (T2−T1) > 10%) using the ConsensusPath tool.
Full size table

Sequence motif enrichments to identify transcription factors binding sites among the 49 intervention-sensitive DMLs revealed 21 significantly enriched motifs (E-value < 0.05, Table 3). The top 5 sequence motifs corresponded to binding sites for ETV4, ZN341, ETV2, SP1, and BC11A transcription factors, which are involved in cell differentiation, regulation of immune homeostasis, blood cell differentiation, immune responses, cancer, cardiovascular disease, diabetes and brain development, respectively, among other biological processes (UniProt database).

Table 3 Transcription factor motif enrichment analysis of intervention-sensitive DML.
Full size table

Impact of multimodal intervention on epigenetic age acceleration

Pearson's correlation analysis revealed no association between baseline Intrinsic Epigenetic Age Acceleration (IEAA) and ACE total score (n = 44; p value = 0.43: r = − 0.13). The analyses of the three categories of adversity assessed by the standard ACE questionnaire (i.e. abuse, neglect and household challenges), revealed a weak but significant positive correlation between IEAA and exposure to abuse (emotional, physical and sexual) (p value = 0.03: r = 0.33) while neglect (emotional and physical) and household challenges (separation from biological parents, witnessing domestic violence, household substance abuse, mental illness in household and having incarcerated family members) were not associated with epigenetic accelerated aging (neglect: p value = 0.07: r = 0.27; household challenges: p value = 0.13: r = − 0.23). No significant difference was found in DNA methylation age or Intrinsic Epigenetic Age Acceleration (IEAA) between groups, calculated at T1 and T2 (Fig. 2; Supplementary Table S3a). The intervention did not have any significant impact on the participants’ IEAA according to the ANCOVA model (coefficient = − 0.661, SE = 0.874, p value = 0.454) (Supplementary Table S3b).

Figure 2
figure 2

(a) Positive correlation between baseline Intrinsic Epigenetic Age Acceleration (IEAA) and exposure to abuse (p value = 0.03: r = 0.33). IEAA positive values indicate that biological age is higher than chronological age, whereas negative values indicate that biological age is lower than chronological age. Abuse score was calculated as the sum of individual scores for emotional, physical and sexual abuse on the 10-item ACE scale. (b) IEAA adjusted by cell type proportions in control and intervention groups before and after the program. No effect of the intervention on IEAA was detected (Δ IEAA (T2−T1) control vs intervention group, p value = 0.23; Supplementary Table S2).

Full size image

Correlation between psychological and DNA methylation outcomes.

Since we previously reported a significant improvement in attention/awareness-related outcomes and a reduction in trauma-related outcomes following the 1-week intervention group program36, we next sought to identify DNA methylation changes related to psychological outcomes by comparing differences in DNA methylation levels and changes in the scores for Attention Awareness Scale-Adolescents (MAAS-A), trauma (the Short PTSD Rating Interview (SPRINT)), and the Child PTSD Symptom Scale (CPSS)) at baseline (T1) and post-intervention (T2). This approach revealed significant correlations of DNA methylation levels at 274 CpGs with MAAS-A scores (p value < 1 × 10–3, r > 0.5, Supplementary Table S4). However, none of these CpGs corresponded to the intervention-sensitive DML described above and they did not show significant functional enrichment (Supplementary Table S5). Improved SPRINT and CPSS scores significantly correlated with DNA methylation levels at 160 and 202 CpGs, respectively (p value < 1 × 10–3, r > 0.5, Supplementary Tables S6 and S7). Two of these genes corresponded to the intervention-sensitive DMLs described above: SIRT5 gene (Sirtuin 5; p value: 0.0001, r = − 0.59) and TRAPPC2L gene (Trafficking Protein Particle Complex Subunit 2L; p value: 0.00002, r = − 0.55; Supplementary Table S1. The DNA methylation levels at 35 CpGs correlated with both CPSS and SPRINT scores and Fisher test confirmed that the CpG overlap between scales was significant (p value < 1 × 10–5). This observation is consistent with the fact that both SPRINT and CPSS scales measure PTSD-related outcomes and that the results from both scales were highly correlated in our previous report (r = 0.833, p value < 1 × 10–3)36. Annotation of these 35 CpGs to genes revealed the known functions of the encoded proteins (Table 4) and an enrichment analysis detected functional interactions involved in metabolic, cardiovascular, immune and neural signaling (q-value < 0.04, Supplementary Table S8).

Table 4 Function (Uniprot database) of the genes associated to the 35 CpGs found to correlate with both CPSS and SPRINT scales.
Full size table

Discussion

Here we describe a genome-wide DNA methylation analysis from saliva samples, as an extension of our previous study that showed the mental health benefits of an intensive multimodal 1-week group program involving mindfulness training, artistic expression and EMDR in adolescent girls with a history of 4 or more ACEs (full details on the program protocol and psychological outcomes are described in Roque Lopez et al.36).

Forty-nine DML were sensitive to the intervention with a methylation change greater than 10% (p value < 0.001). Fifty-four percent of these DML were located in the body of genes, of which 76% showed increases in DNA methylation levels post-intervention, which is generally associated with active transcription in proliferative tissues37.

Although DNA methylation analysis from saliva samples might be not representative of other tissue type programming, some reports have shown correlations between DNA methylation levels in brain, blood and saliva38,39,40,41. A biological pathway-enrichment analysis of the 49 intervention-sensitive DML-associated genes suggests the modulation of several functional processes associated with diseases linked to early childhood adversity, including several biological processes involved in neural signaling and substance abuse disorders (e.g., glutamate receptor, beta agonist/beta blocker, cholinergic, glutamatergic, serotoninergic and dopaminergic synapses and opioid, oxytocin and endocannabinoid signaling, long-term depression and potentiation). These findings are consistent with other reports showing that ACEs can trigger DNA methylation changes in genes that modulate mental health and behavior, such as serotonin transporter and glucocorticoid receptor genes11,12, brain-derived neurotrophic factor14 and glutamate receptor15, oxytocin receptor12,13. DML-associated genes also were enriched in processes involved in neural signaling and substance abuse disorders (e.g., glutamate receptor, beta agonist/beta blocker, cholinergic, glutamatergic, serotoninergic and dopaminergic synapses and opioid, oxytocin and endocannabinoid signaling, long-term depression and potentiation). In addition, these DML-associated genes were significantly enriched in processes involved in cardiovascular health (e.g., endothelins, vascular smooth muscle contraction, thromboxane A2 receptor and calcium signaling, beta-agonist/beta-blocker pathways), diabetes (e.g., insulin secretion, leptin signaling, pancreatic secretion, AGE-RAGE signaling) and cancer (e.g., choline metabolism, WNT, ErbB and EGF-EGF receptor signaling, cancer-related microRNAs, NOTCH signaling), which are non-communicable diseases more likely to appear in 18 year old adults or older with a history of at least 4 ACEs than in those with none3. Inflammation also has been reported in stress-related disorders42,43 and the enrichment analysis suggests that the intervention may regulate inflammation through the modulation of IL8- and chemokine G-coupled receptor CXCR1- and CXCR2-mediated signaling. Furthermore, stress-related DNA methylation changes were associated with the enrichment in several hormone networks (e.g., follicle stimulating hormone signaling, thyroid hormone synthesis and signaling, androgen receptor signaling, aldosterone synthesis and secretion), which are regulated by hypothalamus-pituitary endocrine axes known to be sensitive to stress and childhood adversity44,45,46,47. Consistent with these findings, the top 5 significantly enriched DNA sequence motifs corresponding to transcription factors binding sites are involved in the regulation of similar processes. ETV4 and ETV2 are transcription factors of the ETS family that have been largely involved in carcinogenesis48 and cardiovascular disease49. Specificity protein 1 (SP1) is associated with different types of cancer, neurological and cardiovascular disease50,51 and ZNF341 is involved in immune-mediated disorders and infection susceptibility by regulating IL-6 signaling52. BCL11A is involved in β-hemoglobinopathies, cancer and type II diabetes53, neurogenesis54 and midbrain dopaminergic neurons55.

In our study we found no evidence of association between IEAA and ACE total score, probably because 90% of the participants had a history of 4 or more ACEs. However, our analyses of the three categories of adversity (i.e. abuse, neglect and household challenges), revealed a weak but significant correlation between IEAA and exposure to abuse (emotional, physical and sexual) but not to the other ACE categories. These findings are consistent with data from a prospective study with 974 children showing that girls from age 0–14 years exposed to abuse (i.e., emotional or physical), but not to other individual types of ACEs, presented DNA methylation age acceleration56. On the other hand, no effect in the epigenetic aging trajectory was detected in response to the intervention, probably due to its short duration. Future prospective studies including follow-up care and evaluation will be required to explore a putative association of the intervention with changes in the epigenetic aging trajectory in subjects with a history of multiple ACEs.

The DNA methylation changes post-intervention correlated with the CPSS, SPRINT and MAAS-A measured psychological outcomes at 202, 160, and 274 CpGs, respectively. However, only two of these DML, annotated to the SIRT5 and TRAPPC2L genes, showed a change in DNA methylation level greater than 5% (p value < 0.001). SIRT5 (change in DNA methylation = 13%) was associated with CPSS scores and TRAPPC2L (change in DNA methylation = 7%) was associated with SPRINT scores. SIRT5 is a member of the sirtuin family of proteins located predominantly in the mitochondrial matrix, and it protects cells from oxidative stress57,58. The effect of traumatic stress on oxidative components and redox-state homeostasis has been documented59. These data suggest that the epigenetic modulation of antioxidant-related pathways may be relevant to the psychological benefits of the intervention. SPRINT scores negatively correlated with the DNA methylation levels at the body of TRAPPC2L gene, which is involved in intracellular vesicle-mediated transport events60 and is functionally associated with neurodevelopmental delay/intellectual disabilities in individuals homozygous for a missense variant61.

Taken together, our data support the contribution of epigenetic mechanisms in mediating the effects of the 1-week intervention group program for adolescents exposed to 4 or more ACEs. Future studies are required to examine the functional implications of these changes (i.e., expression levels and activity of candidate genes). The potential relationships of these findings with physiological outcomes may help identify molecular targets aimed to prevent the onset of health disorders and improve the long-term health trajectory in individuals with 4 or more ACEs. Although this level of exposure to adversity increases the risk of adult onset of chronic health problems, behavioral risk, and mortality3, ACE screening is not yet integrated into primary care. One of the arguments is the scarce evidence on therapeutic strategies for children or adolescents with a history of multiple victimization62. However, the early screening of ACEs is seen by several authors as a promising way to promote child well-being through policy, health education and evidence-based programs for families, children and adolescents63,64. Results presented in our previous study36, data presented here and recent evidence from other studies31,32,65,66 are starting to provide the scientific background to encourage further discussions on future avenues for prevention and treatment of ACEs. Although this study describes a promising short intervention for adolescents with multiple ACEs, the participants may still need group or individual follow-up support in order to enhance and strengthen the benefits from this program. Future prospective studies to assess the stability of the epigenetic changes resulting from the intervention and their potential long-term influence on health are warranted.

Methods

Participants

We recruited forty-four adolescent girls, aged 13–16 years, from the foster care system of the Colombian Institute of Family Well-Being (ICBF). All participants were partially or totally separated from their biological families due to inadequate parental care, including abuse and neglect. Exclusion criteria were cognitive impairment, self-harming behavior within the last 6 months, suicidal behavior or ideation or current substance dependence. The flowchart of participants invited, screened, enrolled and completing the study as well as the participants’ demographic information that we could collect have been fully described in our previous report36. Participants and their legal representatives provided a written informed consent. We randomized the participants into two groups using a random-number generator. All subjects (intervention and control group; n = 44) underwent parallel and identical assessments at baseline and post-intervention. Participants were informed to which group they had been assigned after the baseline assessment (T1) and subjects assigned to the control group were immediately invited to attend the same program at the end of the study. All the assessments were carried out at the youth care centers from which the participants were recruited.

This research was performed in accordance with relevant guidelines/regulations and in accordance with the Declaration of Helsinki. Informed consent was obtained from all participants and their legal guardians.

Intervention

The intervention was performed during a school holiday week (June 20–27th, 2019) and it was conducted at a nature retreat facility in Santander, Colombia.

The intervention program included an early morning routine starting with an awakening with soft music and a hot beverage in the garden, followed by a 30 min yoga session67 and a guided loving kindness and compassion meditation to cultivate positive affective states68. After a healthy breakfast, participants attended a mindfulness practice for adolescents69. The program included several sessions per day of artistic expression through art and craft, dramatic play, dance, and music. On days 5 and 6, participants attended two EMDR group protocol sessions/day. During that same week, the control group was engaged in holiday activities proposed by the ICBF. While the intervention and the control group, in their respective locations, engaged in some similar activities (e.g., dance, acting, physical exercise, games, movies), the control group activities did not include approaches to specifically treat traumatic experiences or to promote attentional and emotional regulation. For full details of the intervention program and the control group activities, see Roque López et al.36.

DNA Isolation and Methylation microarray

Before and after the 1-week intervention, saliva samples (1 ml) from all the participants (n = 22/group) were collected using Oragene saliva collection kits and DNA was isolated according to the manufacturer’s protocol. DNA concentration was determined using a Qubit fluorometer (Life Technologies) and normalized to 20 ng/μl for the methylation microarray. Bisulfite conversion was performed with the EZ methylation Gold-kit (cat# D5005, Zymo Research) and the Illumina Infinium MethylationEPIC Beadchip Array was used to quantitatively interrogate at single-nucleotide resolution over 850,000 CpG sites across the genome (Biotech Center, University of Wisconsin-Madison).

Pre-processing of human MethylationEPIC data

Raw intensity data files were imported into R environment. R package minfi was used to assess sample quality, calculate the detection p value of each tested probe, and normalize data70. Two samples were discarded as their mean detection p value exceeded 0.05. Probes were normal-exponential out-of-band (noob) normalized with dye correction, followed by quantile normalization. No samples showed incorrect sex prediction based on methylation levels. Probes were filtered if at most one sample exhibited a detection p value > 0.01, contained a SNP, reported methylation at a SNP, measured methylation at a CH dinucleotide site, had at most one sample with a detection p value > 0.01 or were known cross-reactive probes71,72. These filtration criteria resulted in 688,000 probes used for further analysis.

Beta values were obtained through minfi and were further converted to M-values for differential analysis.

Identification of differentially methylated loci (DML)

Linear regression for each tested CpG using an ANCOVA model was employed using R package limma73. The treatment effect (difference between the intervention and control group) on DNA methylation level, was estimated using an analysis of covariance (ANCOVA) of the outcome (T2) with the baseline (T1) as covariate. BMI, age, ACE score and cell type proportions (surrogate variables) were also included as covariates. In this model, the mean posttest difference between the groups is used to estimate the outcome (DNAm T2 ~ Group (control/int) + age + bmi + ACE + DNAm T1 + surrogate variables)74. Surrogate variables were assessed by R package sva75, which identified a total of 3 variables. For quality control purposes cell type proportions were also calculated using R package RefFreeEWAS. The correlation of p values between these two approaches was 0.86, indicating the accuracy of both measures. P values corrected and uncorrected by FDR were obtained. To assess systematic bias of the linear regression model, the genomic inflation factor was calculated for the obtained p values, yielding a genomic inflation factor of ~ 1, suggesting no bias in these methods. In our study, the relatively small sample size, together with some characteristics inherent to the array (measure of continuous variables in large cell numbers, non-variability of many sites on the array, correlation between neighboring probes on the array) likely resulted in the absence of FDR adjusted DML76,77. However, an FDR adjustment assumes independence in the comparisons, and DNA methylation levels across the genome are not independent. Thus, several studies have taken an approach that requires a larger effect size (i.e., > 10%) with a more liberal p value cut-off78,79,80,81. Therefore, to detect intervention-sensitive DML, we established as cut-off a p value ≤ 0.001 combined with an average difference in methylation between T1 and T2 greater than 10%.

Functional analysis

Gene ontological enrichment of biological processes were identified using the ConsensusPathDB-human database as implemented in the Functional Enrichment module of the EASIER R package82. This database integrates interaction networks in Homo sapiens including metabolic, biochemical and gene regulatory signaling and drug-target interactions. FDR-corrected p values < 0.05 were considered significant.

The DNA sequences flanking the DML of (+ /− 250 nucleotides) were used to find enriched motifs using the AME suite package (MEME Suite online platform)83. An E-value cut-off of 0.05 was established to identify significantly enriched motifs, as recommended by MEME developers83.

Estimation of the impact of the multimodal intervention on epigenetic age acceleration

We explored the associations between Intrinsic Epigenetic Age Acceleration (IEAA) and ACE scores using the basal DNA methylation data from both groups and the ACE scores that we previously described in the same sample36. Child epigenetic age based on the Pediatric-Buccal-Epigenetics’ (PedBE) clock21 was calculated using the methylclock R package84. The package provides the following parameters: (i) DNA methylation predicted age (biological age) in years, (ii) age acceleration, difference between DNAm and chronological age in years; (iii) Intrinsic Epigenetic Age Acceleration (IEAA), obtained after regressing chronological age and cell type proportions on biological age. Pearson's correlation analysis was used to explore associations between basal Intrinsic Epigenetic Age Acceleration (IEAA) adjusted by cell type proportions, ACE total score and the number of ACEs from each one of the three categories of adversity (i.e. abuse: emotional, physical and sexual; neglect: emotional and physical; household dysfunction: separation from biological parents, witnessing domestic violence, household substance abuse, mental illness in household and having incarcerated family members), assessed by the 10-item ACE questionnaire derived from the Kaiser Permanente ACEs Study85 (full details on frequency and patterns of ACEs in this sample are described in Roque Lopez et al.36). We used an ANCOVA model (see “Methods”) to assess the potential impact of the intervention on IEAA. This model included group (intervention or control) as the independent variable, IEAA at T2 from both groups as the dependent variable, and it was adjusted by basal IEAA (T1), BMI, and ACE score, considering p values < 0.05 as significant.

Correlation between psychological phenotypic measures and DNA methylation

PTSD and awareness and attention-related outcomes of this intervention in this same sample were assessed by SPRINT, CPSS and MAAS-A scales and are fully reported in our previous report36. Here we conducted correlations between changes in the above-mentioned scales and changes in DNA methylation (T2−T1) of each CpG.

Linear regression for each tested CpG using an ANCOVA model was employed using R package limma73. Separate models for each psychological scale were constructed, controlling for age, BMI and ACES score. Surrogate variables were assessed by the R package sva75. To assess systematic bias of the linear regression model, the genomic inflation factor was calculated for the obtained p values, yielding a genomic inflation factor of ~ 1, suggesting no bias. Pearson’s correlation coefficients (r) were calculated for continuous variables of interest with beta-values. Correlations with an uncorrected p value < 1 × 10–3, and a correlation coefficient r > 0.5 were considered significant for the current study.