An epigenetic candidate–gene association study of parental styles in suicide attempters with substance use disorders

Abstract

Suicide attempts (SA) are prevalent in substance use disorders (SUD). Epigenetic mechanisms may play a pivotal role in the molecular mechanisms of environmental effects eliciting suicidal behaviour in this population. Hypothalamic–pituitary–adrenal axis (HPA), oxytocin and neurotrophin pathways have been consistently involved in SA, yet , their interplay with childhood adversity remains unclear, particularly in SUD. In 24 outpatients with SUDs, we examined the relation between three parental dysfunctional styles and history of SA with methylation of 32 genes from these pathways, eventually analysing 823 methylation sites. Extensive phenotypic characterization was obtained using a semi‐structured interview. Parental style was patient‐reported using the Measure of Parental Style (MOPS) questionnaire, analysed with and without imputation of missing items. Linear regressions were performed to adjust for possible confounders, followed by multiple testing correction. We describe both differentially methylated probes (DMPs) and regions (DMRs) for each set of analyses (with and without imputation of MOPS items). Without imputation, five DMRs in OXTR, CRH and NTF3 significantly interacted with MOPS father abuse to increase the risk for lifetime SA, thus covering the three pathways. After imputation of missing MOPS items, two other DMPs from FKBP5 and SOCS3 significantly interacted with each of the three father styles to increase the risk for SA. Although our findings must be interpreted with caution due to small sample size, they suggest implications of stress reactivity genes in the suicidal risk of SUD patients and highlight the significance of father dysfunction as a potential marker of childhood adversity in SUD patients.

Investigating methylation levels in stress related pathways, we found significant interactions in paternal dysfunction and a risk of suicidal attempt in patient with multiple addictions under stabilized methadone treatment.

1. INTRODUCTION

According to the World Health Organization, more than 700 000 people die from suicide every year ¹ —corresponding to more than 1% of the total death toll worldwide. Suicidal behaviour, defined as all acts perpetrated by someone to kill him/herself, is a prerequisite to suicide. Therefore, the study of suicide attempts (SAs) can be considered as a valid proxy to reduce the morbidity and the mortality associated with suicidal behaviour in general. ²

Suicidal behaviour has multifactorial and complex aetiology, encompassing clinical (mostly psychiatric), environmental (demographic, sociological, economic) and biological factors. Childhood adversity has been identified as a major early environmental factor for suicidal behaviour in adulthood. ³ However, not all individuals exposed to childhood adversity will develop suicidal behaviours in adulthood, suggesting that there are large inter‐individual differences in the vulnerability to suicidal behaviour. Furthermore, compared to childhood trauma, other forms of childhood adversity, such as poor parental bonding or dysfunctional parenting, have received much less attention. However, clinical reports showed significant relationship between poor parental bonding and suicide. ⁴ , ⁵

Epigenetic mechanisms, such as DNA methylation, may be involved in the complex interaction between environmental and biological factors that is thought to elicit suicidal behaviour. DNA methylation is implicated in the regulation of gene expression. Differences in DNA methylation have been reported in patients with suicidal behaviour that had been exposed to childhood adversity in several biological pathways and genes involved in the response to stress, ⁶ , ⁷ including the hypothalamus–pituitary–adrenal (HPA) axis ⁸ and the neurotrophin pathway, ⁹ , ¹⁰ compared to unexposed individuals. One methylation site in the SKA2 gene (spindle and kinetochore associated complex subunit 2, a chaperon protein for the glucocorticoid receptor into the nucleus) was significantly more methylated in neuronal and glial nuclei of individuals with suicidal behaviour compared to controls. ¹¹

To date, studies investigating the interaction of DNA methylation and childhood adversity and in suicidal behaviour mainly focused on childhood maltreatment. Much less attention has been paid to parental style/caregiving, ¹² although clinical reports also showed significant relationship between poor parental bonding and suicidal behaviour. Additionally, there is consistent evidence from animal models and clinical studies that parental bonding is associated with epigenetic changes in the HPA axis ¹³ and in the oxytocin pathway (oxytocin receptor gene, OXTR). ¹⁴ In humans, childhood maternal care has also been associated with differential DNA methylation of the genes coding for brain‐derived neurotrophic factor (BDNF) and OXTR in peripheral blood cells in adult men and women and DNA methylation in low versus high maternal care group in the BDNF and OXTR genes. ¹⁵

Finally, most studies investigating the role of childhood adversity in suicidal behaviour have focused on psychiatric disorders (e.g. mood disorders, PTSD ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ ) other than substance use disorders (SUDs). Yet, history of SA is reported by 30%–40% of outpatients with SUDs, ²⁰ , ²¹ and SUDs remain a strong risk factor for suicidal behaviour. ²² Furthermore, in this population, childhood adversity is common and has previously been associated with SA, ²³ whether childhood trauma or parental bonding was considered. For instance, our group reported a significant positive correlation between maternal abuse and a higher number of lifetime SAs in women with opioid use disorders (OUD). ²⁴

We aimed to extend the findings regarding suicidal behaviour and parental style in SUD populations using epigenetics. More precisely, we evaluated the differences in DNA methylation of several genes in the HPA axis, oxytocin and BDNF pathways, between patients having attempted suicide or not. In addition, we investigate the mediating effects of dysfunctional parental styles in this relationship. Our hypothesis is that differences in methylation between outpatients with SUD for suicide attempters compared to those, who did not attempt and who are non‐attempters would be greater in case of parental dysfunction.

2. MATERIALS AND METHODS

2.1. Sample

We included 24 French self‐declared European descent subjects (16 men and eight women) from a multicentric study ²⁵ aimed at characterizing the pharmacokinetics and pharmacogenomics of methadone treatment. Participants were treatment‐seeking outpatients with lifetime OUD (defined as DSM‐IV opioid dependence ²⁶ ) in full remission ≥12 months under steady methadone regimen for at least 3 months.

Inclusion criteria were as follows: age >18 years, affiliation to a social security system. Exclusion criteria were any specially protected person (legally protected adults, compulsory admission), presenting a contraindication to blood sampling and inability to master French language. Patients with alcohol, cocaine or benzodiazepine dependence in the last 12 months, contraindication to oral midazolam, pregnant or breastfeeding women were also excluded—due to the pharmacokinetics purpose of the original study.

Data were obtained during a single interview conducted by licensed psychologists or medical doctors. Socio‐demographic criteria were collected according to a standard procedure. Clinical data collected included number of cigarettes per day, history of lifetime SUD (DSM‐IV dependence was considered for the study) for cocaine, opioids, cannabis and benzodiazepines (according to section E of the Structured Clinical Interview for DSM‐IV, SCID‐IV ²⁷ ) and the existence of current medication for mood disorders (antidepressants or mood stabilizing agents) or psychotic disorders. These medication data were used not only as confounders for methylation levels but also as a proxy for the presence of current/recent psychiatric disorders.

History of SA was characterized using the ‘suicide’ section of the Diagnostic Interview for Genetic Studies, v 4.0 (DIGS 4.0 ²⁸ ) through a binary response: ‘yes/no’ to the question ‘have you ever tried to kill yourself?’. Of note, this was the screening question for extended assessment of the worst lifetime attempt.

2.2. MOPS questionnaire: parental styles

The Measure of Parental Style (MOPS) is a self‐administered questionnaire ²⁹ derived from the Parental Bonding Instrument (PBI) questionnaire. ³⁰ It is a retrospective assessment of parenting styles in three categories: ‘overcontrol’, ‘abusive’ and ‘indifferent’. Situations/parental attitudes are assessed using the following wording: ‘During your first 16 years how “true” are the following statements about your MOTHER's/FATHER's behavior towards you’ applied to 15 Likert‐type items for each parent (mother and father) ranging from 0 not at all true to 3 extremely true, for example, ‘Verbally abusive of me’ or ‘Over controlling of me’. Although they partially overlap with widely used measures of childhood adversity such as trauma assessments, MOPS scores likely addresses complementary items and focus more on the retrospective perception of parental behaviour rather than on obvious traumatogenic attitudes. There is no cut‐off score, the total score for each category showing the degree of different types of parenting styles experienced by the individual. Higher scores at the MOPS indicate more dysfunctional parental styles. There were four (17%) individuals with missing values for the MOPS. Thus, we provide two sets of analyses: one with raw MOPS data (N = 20) and one after imputation of the missing MOPS scores (N = 24). For the latter, we used fully conditional specification implemented by the MICE algorithm in the ‘mice package’ for R studio (Version 1.4.1103). ³¹ This procedure imputed the four MOPS scores with an item‐wise manner, which we deemed relevant as an add‐on to the raw analysis aimed at increasing gene discovery by maximizing sample size.

Healthy controls were not included to avoid confounding by epigenetic factors of SUD psychopathology and psychiatric medication. ³²

All participants gave written informed consent after they had received a complete description of the study. The study was approved by the Ethics Committee of Paris (CPP Ile‐de‐France VI) and was registered to the clinicaltrials.gov website (NCT00894452). The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and the 1975 Declaration of Helsinki, as revised in 2008.

2.3. DNA methylation quantification

Genomic DNA was extracted from the collected peripheral blood cells, with the Wizard Genomic DNA Purification Kit (Promega, Charbonnières‐les‐Bains, France) according to the manufacturer's instructions. Sodium bisulfite conversion of 500 ng of genomic DNA was performed with an EZ DNA Methylation Kit (Zymo Research, CA, USA) according to the manufacturer's instructions.

DNA methylation profiling was performed with Human‐Methylation450k BeadChips (Illumina, CA, USA), which analyse the methylation status of 485 577 cytosine‐guanine dinucleotides (CpG) sites covering 99% of all RefSeq genes, with an average of 17 probes per gene, 41% of which were in proximal promoters. Whole‐genome amplification, labelling, hybridization and scanning were performed according to the manufacturer's instruction at IntegraGen® (Evry, France). All arrays were imaged using an Illumina BeadArray Reader.

2.4. Gene selection

We applied a custom selection of genes from the HPA, oxytocin and BDNF pathways identified through the KEGG pathways database (https://www.genome.jp/kegg/pathway.html). We limited most of the selection to extracellular ligands/receptors as to avoid gathering too many unspecific genes from intracellular effectors. As most previous studies of candidate genes in the literature tend to focus on a more restricted choice of genes (generally one or two), we substantially extended this selection to provide a better overview of these pathways. We relied on redundancy and strength of association in different articles. We referred to reviews involving in the epigenetics of suicidal behaviour and childhood maltreatment ³³ and specific articles in the different pathways:

1. for the BDNF pathway: KEGG neurotrophin signalling pathway (map 04722), genes encoding extracellular proteins, only main ligand neurotrophins (BDNF, NTF3, NTF4, CNTF, NTF) and their receptors (NTRK2, NTRK3, CNTFR), one of the major genes regulating BDNF (SORT1) and SKA2. ³⁴ , ³⁵ The same selection was also applied in a previous study from our group ³⁶ focusing on suicidal behaviour;

2. for the HPA axis: we have selected from a review of the genetic HPA regulation: various signalling (CRH, CRHR1, CRHR2, AVP and POMC), regulatory (CRHBP, FKBP5, FKBP4, PROP1, NR3C1 and NR3C2) and inflammatory (SOCS3, STAT3, LIF and ANXA1) targets associated with the HPA axis ³⁷ ;

3. for the oxytocin pathway: we have selected from a review of the oxytocin pathway gene network ³⁸ in the human brain: OXT, OXTR, CD38, TRPM2, EGFR, ³⁹ NPR1, NPR2 and NPR3. ⁴⁰

We used hg19 coordinates for each gene (https://www.ensembl.org/index.html). CpG islands are mainly located in promoter regions which are typically located directly upstream or at the 5′ end of the transcription start site and 100–1000 bp long. ⁴¹ In order to include the most of CpG islands we chose to add 1 K base pairs before/after each gene based on the hg19 human genome version. The 32 genes harboured 823 probes in our DNA methylation array after quality control (QC)—described below.

2.5. Pre‐processing and QC

The raw intensity files (IDAT) generated by the BeadChips array were imported into the R programming environment using RStudio (version 1.4.1103). ⁴²

DNA methylation status is represented by the beta (β) value, which can range from 0 (no methylation) to 1 (complete methylation). The β value is calculated by means of the following equation: β = M/(M + U + α) where M = methylated intensity, U = unmethylated intensity and α is a small offset added to avoid dividing by small values. This is the default equation of the getBeta() function in the ad hoc minfi R package, where the default value for α is 100. Each CpG site reports an average β value obtained from the average of Ms and Us across approximately 30 bead replicates. ⁴³

We evaluated the similarities and differences between the samples by visual inspec using multi‐dimensional scaling and density plots. The methylated and unmethylated signal intensities were evaluated using the function plotQC() from minfi package. ⁴⁴ We also evaluated performance across samples checking the sample‐dependent and sample independent controls for bisulfite conversion I and II, extension, hybridization, non‐polymorphic, specificity I and II and target removal. We excluded cross‐hybridization probes. ⁴⁵

Samples and CpG sites that failed QC check were removed: At the genome level, 3656 sites were removed as beadcount <3 in 5% of samples, and 10 168 sites having 1% of samples with a detection p‐value greater than 0.01 were removed.

Normalization of the DNA methylation data was performed using the pre‐process Funnorm() function in the minfi package. Singular value decomposition was checked using ChAMP package in R, ⁴⁶ in order to confirm that no underlying technical variability remained after QC. More precisely, we checked the values obtained before and after normalization and before and after moving out sex chromosome.

After performing QC, we selected only the probes that target CpG site located within the candidate genes. The final number of samples analysed and probes investigated was, respectively, 24 and 823.

We estimated cell counts with the function estimateCellCounts2() from the FlowSorted.blood.EPIC package in R. ⁴⁷ The cell types that were estimated in our sample were as follows: CD8+, CD4+, NK, B cells, monocytes and neutrophils.

2.6. Statistical analysis

The data are described by mean +/− standard deviation or frequencies %. Bivariate analyses were conducted to identify the variables associated with SA (Chi² or Mann–Whitney U, α = 0.05, two‐tailed tests) that should be entered as covariates/cofactors in the regression models. To find differentially methylated positions (DMPs), we applied two linear regression models where DNA methylation (beta value) was the outcome: first against ‘history of SA’ alone and then against the interaction ‘history of SA’ X ‘parental style scores’. Six distinct parental styles from the MOPS questionnaire were considered. Each was analysed separately. The models were (1) methylation ~ SA + age + sex + estimated cell proportions + number of cigarettes per day + antipsychotics + mood medications and (2) methylation ~ SA × MOPS + age + sex + estimated cell proportion + number of cigarettes per day + antipsychotics + mood medications. To account for multiple testing, a false discovery rate (FDR) procedure using the Benjamin–Hochberg (BH) method was applied, and values below 0.05 after correction were considered significant.

We also investigated the differentially methylated regions (DMRs) using the comb‐p tool ⁴⁸ (python version 3.6). We choose a 200 kb window based on the recommended protocol by Yehuda et al, ⁴⁹ considering a minimum of two probes per DMR.

3. RESULTS

3.1. Socio‐demographic and parental style data

Among the total sample of 24 patients, mean age was 43 years (+/−9), and 67% (n = 16) of them were men. Nine patients (38%) reported a history of SA. Patients' socio‐demographic characteristics and clinical features are described in Table 1. There was no significant difference in the clinical and sociodemographic data between the sample with non‐imputed and the sample including imputed MOPS scores (Table S1).

TABLE 1.

Socio‐demographic characteristics and clinical features for patients.

Characteristics	Mean +/− SD Or N (%)	p‐value (SA vs. non SA)	N
Age	42.9 (+/−9.3)	0.53	24
Sex		0.099	24
Men	16 (66.7%)
Women	8 (33.3%)
History of suicide attempt	9 (37.5%)		24
Number of cigarettes per day	16.2 (+/−12.1)	0.098	24
Cocaine dependence lifetime	16 (66.7%)	1	24
Alcohol dependence lifetime	11 (45.8%)	0.675	24
Cannabis dependence lifetime	17 (70.8%)	0.356	24
Benzodiazepine dependence lifetime	9 (37.5%)	0.033	24
MOPS mother
Abuse mother	5.08 (+/−3.21)	0.952	22
Indifferent mother	2.67 (+/−4.55)	1	24
Overcontrol mother	5.30 (+/−3.11)	0.904	22
MOPS father
Abuse father	4.54 (+/−5.21)	0.276	20
Indifferent father	5.45 (+/−6)	0.926	20
Overcontrol father	5.12 (+/−3.19)	0.132	20
Mood treatment	8 (33.3%)	0.021	24
Neuroleptic treatment	7 (29.2%)	0.031	24
Current opioid use	4 (16.7%)	0.615	24
Current cannabis	17 (70.8%)	0.061	24
Current alcohol use	11 (45.8%)	0.206	24
Current benzodiazepine use	18 (75%)	0.635	24
12‐month cocaine use	11 (45.8%)	0.423	24

3.2. Investigation of the association between ‘history of SA’ and DNA methylation in candidate genes

We did not find any significant association between SA and methylation levels at the FDR < 0.05 (lowest FDR p‐value = 0.581 with both samples excluding/including imputed MOPS scores. Table 2 shows associations with p‐values <1).

TABLE 2.

Five strongest DMPs associations with ‘history of suicide attempt’ for 24 patients with substance use disorder.

Probe	Estimate β regression	Std. error	t value	p‐value	FDR value	Chr
cg17238830	0.056	0.012	4.5	7.230 × 10−4	0.581	5
cg20930320	0.017	0.016	1.06	0.310	0.998	1
cg17285605	0.003	0.008	0.407	0.691	0.997	1
cg17513787	−0.003	0.006	−0.456	0.657	0.997	1
cg09131339	−0.011	0.011	−0.978	0.348	0.997	1

3.3. Investigation of the association between the interaction ‘history of SA × parental styles’ and DNA methylation in candidate genes after correction for multiple testing and adjustment for confounders

In the sample with raw MOPS scores (Table 3A), one significant DMP (NTF3 cg26401796) remained associated with the interaction terms between history of SA and father indifference (FDR p = 0.036).

TABLE 3A.

Significant DMPs associated with the interaction ‘history of suicide attempt’ × ‘parental style MOPS’ in outpatients with substance use disorder (raw MOPS scores, N = 20).

Probe	Estimate β regression	Std. error	t value	p‐value	FDR value	Chr	Position	Gene annotation
Indifferent father
cg26401796	0.008	6 × 10−4	13.18	4.48 × 10−5	0.036	12	38 719 359	NTF3 (Intron–enhancer)

In the sample with including imputed MOPS scores (Table 3B), there were two significant DMPs (cg16012111 and cg27637521, mapped to FKBP5 and SOCS3, respectively) remained associated with the interaction terms between history of SA and all father dysfunctional styles (−0.13 < β < −0.074, maximum FDR p = 0.04).

TABLE 3B.

Significant DMPs associated with the interaction ‘history of suicide attempt’ × ‘parental style MOPS’ and for 24 patients with substance use disorder (with the including imputed MOPS scores, N = 24).

Probe	Estimate β regression	Std. error	t value	p‐value	FDR value	Chr	Position	Gene annotation
Abuse father
cg16012111	−0.130	0.015	−6.85	4.46 × 10−5	0.018	6	35 656 758	FKBP5 (Island)
cg27637521	−0.086	0.012	−7.02	3.62 × 10−5	0.018	17	76 355 202	SOCS3
Indifferent father
cg16012111	−0.099	0.015	−6.52	6.71 × 10−5	0.03	6	35 656 758	FKBP5 (Island)
cg27637521	−0.083	0.012	−6.41	7.69 × 10−5	0.03	17	76 355 202	SOCS3
Overcontrol father
cg16012111	−0.088	0.013	−6.73	5.14 × 10−5	0.04	6	35 656 758	FKBP5 (Island)
cg27637521	−0.074	0.012	−6.08	1.18 × 10−4	0.04	17	76 355 202	SOCS3

No significant associations were found between interaction terms involving SA and mother parenting styles and DNA methylation in candidate genes, whether MOPS scores were imputed or left as is. Interaction plots (Figures 1, 2, 3) were used to inspect the relationships between SA history, methylation levels and father dysfunction levels.

FIGURE 1.

Differences in methylation (β value) in cg26401796 (NTF3) in patient who attempt suicide or not with history of indifferent father.

FIGURE 2.

Differences in methylation (β value) in cg16012111 (FKBP5) in patient who attempt suicide (SA 2) or not (SA 1) with history of different father styles. SA, suicide attempt (1: no, 2: yes); PEREABUS, abuse father (A); PEREINDIFF, indifferent father (B); PEREOC, overcontrol father (C).

FIGURE 3.

Differences in methylation (β value) in cg27637521 (SOCS3) in patient who attempt suicide (SA 2) or not (SA 1) with history of different father style. SA, suicide attempt (1: no, 2: yes); PEREABUS, abuse father (A); PEREINDIFF, indifferent father (B); PEREOC, overcontrol father (C).

In the sample with raw MOPS scores, Figure 1 shows the weak interaction between NTF3 cg26401796 and father indifference as a function of lifetime SA, where lifetime suicide attempters show inverted relationship between the methylation levels for the highest father indifference.

In the sample including imputed MOPS scores, for FKBP5 cg16012111, Figure 2 shows that only ‘overcontrol father’ involved actual interaction, while such interaction was confirmed for all father dysfunctional styles regarding SOCS3 cg27637521 in Figure 3. The methylation level of FKBP5 cg16012111 increases as father overcontrol also increases in both SA groups but that this increase is more pronounced in patients who had not versus patients who had attempted suicide. The interaction was opposite for SOCS3 cg27637521: The methylation level increases with father dysfunction in patients with a history of SA, while it slightly decreases in those without.

3.4. Identification of DMRs

The DMR analysis in the sample with raw MOPS scores extended the results from the DMP analysis. Thus, at the significance level Šidák p < 0.05 (Table 4A), we found DMRs in three other genes (NTF3, OXTR and CRH) for father abuse, two of which (OXTR and CRH) were also associated with SA x father indifference.

TABLE 4A.

Differentially methylated regions (DMRs) for different parental styles significant at the 0.05 level. Z Šidák p‐values represent p‐values after multiple testing correction (raw MOPS scores, N = 20).

Probe	Chr	Start	End	N probes	p‐values	Z Šidák p	Gene annotation
Indifferent father
cg03987506	3	8 810 549	8 810 593	2	1.98 × 10−4	0.004	OXTR
cg00078085
cg17305181	8	67 090 581	67 090 799	5	3.99 × 10−5	1 × 10−4	CRH
cg18640030
cg08215831
cg19035496
cg23409074
Abuse father
cg03987506	3	8 810 549	8 810 593	2	1.70 × 10−4	0.003	OXTR
cg00078085
cg17305181	8	67 090 581	67 090 799	5	4.16 × 10−5	1.53 × 10−4	CRH
cg18640030
cg08215831
cg19035496
cg23409074
cg20462512	12	5 543 155	5 543 270	2	3.09 × 10−4	0.002	NTF3
cg27423216

There were fewer DMRs at the significance level Šidák p < 0.05 (Table 4B) in the sample including imputed MOPS scores, where six regions corresponding to the DMP analysis were evidenced.

TABLE 4B.

Differentially methylated regions (DMRs) for different parental styles significant at the 0.05 level. Z Šidák p‐values represent p‐values after multiple testing correction (sample including imputed MOPS scores, N = 24).

Probe	Chr	Start	End	N probes	p‐values	Z Šidák p	Gene annotation
Abuse father
cg10508317	17	76 355 146	76 355 203	2	1.55 × 10−5	2.19 × 10−4	SOCS3
cg27637521
Indifferent father
cg10508317	17	76 355 146	76 355 203	2	2.14 × 10−5	3.01 × 10−4	SOCS3
cg27637521
Overcontrol father
cg10508317	17	76 355 146	76 355 203	2	1.14 × 10−6	1.60 × 10−5	SOCS3
cg27637521

4. DISCUSSION

We performed a candidate epigenetic analysis in 24 patients with SUDs, examining methylation levels as a function of the interaction between the risk for suicide attempt and three dysfunctional parental styles (abuse, indifferent, overcontrol measured for fathers and mothers, specifically). Using a conservative analysis pipeline, we showed associations between different parental styles, SAs and hyper‐ or hypomethylation of genes in the oxytocin (OXTR), HPA (CRH and—after imputation—FKBP5 and SOCS3) and BDNF (NTF3) pathways after correction for multiple testing and adjustment for key confounders (i.e. age, biological sex, medication use, tobacco smoking level and cell count). To the best of our knowledge, there was no previous publication reporting epigenetic changes in stress‐related pathways as a function of past suicidal behaviour and childhood adversity in any SUD population. Our findings support a plausible role of stress‐related pathway gene methylation in the risk for suicide attempt in patients with OUD, which represent an overlooked population in suicide research overall. We applied conservative bioinformatics on a thoroughly‐selected clinical sample—yet, the sample remained representative of outpatients with severe SUDs. Methylation data were overall of high quality, with no biological sample excluded. We focused on a particularly original measure of childhood adversity, parental styles measured by the MOPS questionnaire, which may be a useful complement to measures of childhood trauma. We used multivariate imputation of MOPS data to maximize sample size, presented here as an additional analysis.

Both our findings regarding the oxytocin and the HPA pathways are consistent with the pathophysiology of the stress response and with previous literature. Our finding involving OXTR showed a weak effect size, which was yet larger than for the other findings with non‐imputed MOPS scores. Lower serum oxytocin concentrations have been reported in recent suicide attempters compared to healthy controls. ⁵⁷ In an epigenetic study, OXTR methylation was a predictor of changes in SUD symptomatology over time, but this effect was independent from childhood trauma. ⁵⁸ The A allele of rs53576 OXTR was associated with suicide attempt history in a sample with current depressive episode. ⁵⁹ Regarding CRH, there have been extensive epigenetic studies of methylation alterations of the HPA axis, particularly in the case of history of adverse childhood events. ⁶⁰ A previous study found two promoter sites within CRH (cg23409074 and cg1903496) hypomethylated in high‐severity compared to a low‐severity SA population. ⁶¹ Associations have also been reported between CRH‐binding protein gene SNP rs1500 and reduced improvement in cocaine abuse among patients after 1 year of methadone treatment. Symptoms in SUD patients may thus be mediated in part by genetic variants in HPA axis. ⁶² Regarding both OXTR and CRH, it remains unclear whether possible alterations in gene expression are due to external stressors or reflects the severity of a given disorder. Finally, as regards NTF3, we could only identify one study where NTF3 levels in the hippocampus were lower for suicide victims compared with controls. ⁵⁵ It is noticeable that our group previously reported associations between severe suicidal behaviour and an SNP from another gene in the neurotrophin pathway (BDNF rs10835210) in SUD—highlighting its potential role in suicidal behaviour in this population. ³⁶

Although they were obtained after multivariate imputation of MOPS score that may have been missing due to the absence of the corresponding parent, our findings of altered methylation in FKBP5 and SOCS3 merit a brief discussion. In a genetic study performed in cocaine‐dependent individuals—thus close to ours in terms of clinical setting, a genetic polymorphism of FKBP5 had also been associated with SA in interaction with childhood trauma. ²³ In an epigenetic study, the polymorphism rs1360780 was associated with hypomethylation of FKBP5 transcription start site after history of childhood trauma, increasing glucocorticoid resistance. ⁵⁰ These findings—and ours—are in line with a body of literature showing interactions between FKBP5 genotypes, methylation and early trauma in the risk of several neuropsychiatric diseases. ⁵⁰ , ⁵¹ , ⁵² , ⁵³ Likewise, SOCS3 may represent a promising biomarker of suicidal behaviour ³³ , ³⁴ , ⁵⁴ given its key role in immune physiology and disorders and in inflammation.

The results of our candidate gene study can be deemed exploratory, since methylation changes in stress‐related pathways are most likely to occur altogether in a given individual/population, especially in case of severe childhood adversity. Future studies should focus on the interaction between these different pathways and be more integrated. For instance, and in relationship with our study, authors reported interactions between the activity of the endogenous opioid system and both those of the oxytocin and HPA axis pathways. ⁶³

We did not find any significant association with maternal styles, which was rather unexpected. Most previous studies focused on maternal behaviour and their offspring and little on paternal behaviour. This could reflect a lack of statistical power to detect parental dysfunction in women, who were under‐represented in our sample (N = 8 vs. 16). MOPS scores are sensitive to the respondent's gender. ²⁹ , ³⁰ There was no association between SA and MOPS splitting by gender (Table S2), and sex was included as a confounder in all methylation analyses. This could also reflect the influence of father parenting in mammals, which may have been overlooked in humans. ⁶³ However, the definition of ‘father styles’ in standardized scales such as the MOPS must be examined with caution. Definitions likely vary from one culture to another and even from one family to another. With that regard, the total father MOPS scores for each sub‐category were high compared to the original study conducted in patients with depressive disorders. ²⁹ Moreover, the parental medical history was not available in this sample, which may have represented a potential confounder for MOPS scores.

As already stated in the Methods, we performed a within‐case study, which may lack the reliability provided by control group results. However, this design was deemed relevant given the very strong differences expected between outpatients from addiction care centres in the Paris area and healthy subjects. Ill controls, e.g., patients with mood disorders, may help in interpreting our findings. We analysed a small sample selected for stable response to methadone, which may have limited our ability to (i) detect significant associations for mother styles since there were only eight women and since MOPS scores are sensitive to the respondent's gender ²⁴ (calling for the inclusion of more women in methylation studies in OUD) and (ii) analyse specific suicidal phenotypes beyond ‘lifetime SA’. ⁵⁶ We did not ascertain psychiatric diagnoses by focusing on suicidal behaviour as a transdiagnostic ‘disorder’. With that regard, however, the data about current psychotropic medication was a useful proxy for psychiatric disorders. The benzodiazepine dependence lifetime variable has been not included as we could not retrieve any evidence of altered methylation levels as a function of benzodiazepine use. This issue should clearly be addressed by future studies, since it is likely that chronic benzodiazepine exposure induces methylation changes, at least for the genes they target.

The methylation data were derived from blood cells DNA, which limits inference of the effects of environmental variables on the brain. However, our findings are located in CpG islands, which have been found to be strongly conserved across brain areas and blood in control subjects. ⁶⁵ Methylation levels were generated in 2015, and there are now arrays with wider coverage. This may have limited the number of CpG sites we could have found associated with SA. We also note that there is evidence of high correlation between the 450 k array and the Illumina EPIC array (850 k sites), which would allow our sample to be meta‐analysed with potential analyses on new samples. Finally, we did not measure gene expression levels and could thus not disentangle enhancing from repressing effects of hypermethylation in the current study.

5. CONCLUSIONS

We have identified differences in DNA methylation in three key stress‐related genetic pathways (HPA axis, oxytocin and BDNF pathway) in patients under stable remission from OUD using methadone, who previously attempted suicide, in interaction with paternal dysfunction (abuse, indifferent, overcontrol). These results suggest that father style should deserve our attention, since the few available studies focused on the maternal role. We could not conclude as to this role of gene methylation in the pathophysiology of SA in OUD is specific to early adversity estimated by the MOPS or is more global as an integrated model of SA liability involving other epigenetics and genomic factors, the severity of SUD itself and more proximal stressors—which are numerous in OUD. More global and integrative analyses of multi‐omics data ⁶⁶ have already been done in complex human diseases and could also help with that regard. Our findings open avenues for identifying markers of the vulnerability to parental dysfunction in eliciting suicidal behaviour. However, it remains unclear at this point whether this vulnerability relates to the co‐occurrence between psychiatric and substance use disorders in general or are as a direct driver of suicide.

AUTHOR CONTRIBUTIONS

Clara Chrétienneau designed the study, performed statistical analyses and wrote the first draft of the manuscript and the present final version. Conceptualization: Florence Vorspan, Cynthia Marie‐Claire and Vanessa Bloch. Methodology: Cynthia Marie‐Claire, Leticia M. Spindola and Stéphanie Le Hellard. Investigation: Florence Vorspan, Vanessa Bloch and Clara Chrétienneau. Writing—original draft preparation: Clara Chrétienneau and Romain Icick. Writing—review and editing: Clara Chrétienneau, Romain Icick, Florence Vorspan, Leticia M. Spindola and Stéphanie Le Hellard. Supervision: Jean‐Louis Laplanche, Frank Bellivier and Florence Vorspan. Project administration: Florence Vorspan and Stéphane Mouly. Funding acquisition: Florence Vorspan and Vanessa Bloch. All authors critically reviewed the content of the manuscript and approved the final version for publication.

CONFLICT OF INTEREST STATEMENT

Florence Vorspan had congress fees paid by pharmaceutical companies (CAMURUS AB, RECORDATI, ACCORD Pharmaceutical). Romain Icick performed a paid talk for PIERRE FABRE, with the payment transferred in full to a non‐profit association (‘Les amis de l'Espace Murger’). The other authors have no conflict of interest to declare.

ETHICS STATEMENT

The authors state that they have obtained appropriate institutional review board approval, in accordance with the principles outlined in the Declaration of Helsinki for all human experimental investigations, including obtention of written informed consent from all the participants involved.

Supporting information

Table S1. Comparison between the whole sample and the sample with missing MOPS data. N (%) or medians (IQR) are shown. ^aperformed on the sample with MOPS data for both mother and father.

Table S2. A) Comparison of MOPS scores by SA status in women. Medians (IQR) are shown.

Chrétienneau C, Spindola LM, Vorspan F, et al. An epigenetic candidate–gene association study of parental styles in suicide attempters with substance use disorders. Addiction Biology. 2024;29(4):e13392. doi: 10.1111/adb.13392

DATA AVAILABILITY STATEMENT

The data presented in this study are available on request from the corresponding author.