Can epigenetics shine a light on the biological pathways underlying major mental disorders?

A significant proportion of the global burden of disease can be attributed to mental illness. Despite important advances in identifying risk factors for mental health conditions, the biological processing underlying causal pathways to disease onset remain poorly understood. This represents a limitation to implement effective prevention and the development of novel pharmacological treatments. Epigenetic mechanisms have emerged as mediators of environmental and genetic risk factors which might play a role in disease onset, including childhood adversity (CA) and cannabis use (CU). Particularly, human research exploring DNA methylation has provided new and promising insights into the role of biological pathways implicated in the aetio-pathogenesis of psychiatric conditions, including: monoaminergic (Serotonin and Dopamine), GABAergic, glutamatergic, neurogenesis, inflammatory and immune response and oxidative stress. While these epigenetic changes have been often studied as disease-specific, similarly to the investigation of environmental risk factors, they are often transdiagnostic. Therefore, we aim to review the existing literature on DNA methylation from human studies of psychiatric diseases (i) to identify epigenetic modifications mapping onto biological pathways either transdiagnostically or specifically related to psychiatric diseases such as Eating Disorders, Post-traumatic Stress Disorder, Bipolar and Psychotic Disorder, Depression, Autism Spectrum Disorder and Anxiety Disorder, and (ii) to investigate a convergence between some of these epigenetic modifications and the exposure to known risk factors for psychiatric disorders such as CA and CU, as well as to other epigenetic confounders in psychiatry research.

In the past decade, growing evidence has shown a link between epigenetic processes, and a range of mental health disorders (Binder, 2017). Epigenetic modifications refer to functional changes in DNA structural packaging or associated proteins without structural alteration of the DNA sequence itself (Jaenisch & Bird, 2003). This biological mechanism has important implications on how genes are expressed and how the chromatin is packaged, thus modifying subsequent protein translation within regionally specific parts of the central nervous system (Binder, 2017). The most studied epigenetic process in humans is DNA methylation (DNAm) ( Table 1 for definitions of key terms). Indeed, recent parallel evidence suggests that differential DNAm profiles are associated with exposure to childhood adversity (CA) as well as cannabis use (CU) (Kandaswamy et al., 2020;Markunas et al., 2020;Nöthling, Malan-Müller, Abrahams, Hemmings, & Seedat, 2020). This suggests that epigenetic factors may account for some of the non-explained variance in genetics studies and possibly mediate Table 1. A glossary of key epigenetic terms and biological function of genes involved in pathways discussed in this review

Gene names and key terms Biological function and definition
Epigenetic terms DNA-methylation DNA-meth is the covalent addition of a methyl group to the 5 th carbon of a Cytosine (C) base, resulting in a 5-methylcytosine (5-mC) base. Epigenetic is the major process by which the environment can alter gene expression Candidate gene approach Explores methylation on certain genes of interests based on a priory hypothesis. It often examines whether DNAm changes in different CG sites within specific genes are related to a particular phenotype.

Epigenome-Wide Association Studies
Examines the association of DNAm changes (otherwise called methylome-wide association studies (MWAS)) across the entire genome for a particular phenotype, using a hypothesis-free paradigm. EWAS have been performed with increasingly powerful techniques and have moved from pioneer CpG-island microarrays studies that interrogated around 12.000 sites across the DNA (Mill et al., 2008) to more advanced techniques such as Infinium MethylationEPIC BeadChip, that covers more than 850 000 CpG methylation sites (Yong et al., 2016).

Histone acetylation studies
Histone acetylation is a dynamic epigenetic modification that functions in the regulation of DNA-templated reactions, such as transcription. This lysine modification is reversibly controlled by histone (lysine) acetyltransferases and deacetylases.

Methylome-wide association studies (MWAS)
Test a genome-wide set of methylation sites for association with an outcome of interest.

SLC6A4
Regulated serotonergic signalling via transporting 5-HT from synaptic spaces into presynaptic neurons. SLCA2 is involved in the recapture of the Encodes one of several forms of glutamic acid decarboxylase, an enzyme which is responsible for catalysing the production of gamma-aminobutyric acid from L-glutamic acid PVALB Encodes for Parvalbumin protein, essential for neural synchronisation in some neurons in the CNS GRIN1 (2,2B 3B, D1) The protein encoded by this gene is a critical subunit of N-methyl-D-aspartate receptors, members of the glutamate receptor channel superfamily. It plays an important role in the plasticity of synapses GRIA 2, 3 Encodes for the Glutamate Ionotropic Receptor AMPA Type Subunit 2 and 3 Glutamate receptors, which are the predominant excitatory neurotransmitter receptors in the mammalian brain MARLIN-1 (synonym of JAKMIP1) codes for a protein that may play a role in the microtubule-dependent transport of the GABA-B receptor KCNJ6 Encodes a member of the G protein-coupled inwardly-rectifying potassium channel family of inward rectifier potassium channels. This type of potassium channel allows a greater flow of potassium into the cell than out of it and thus regulates circuit activities in neural cells. Expressed in GABAergic synapses HELT Protein Coding gene involved in DNA-binding transcription factor activity and protein dimerisation activity. It is a transcriptional repressor gene which is known to function as a selector gene that determines GABAergic over glutamatergic fate in the mesencephalon

GRIK2
Codes for the Glutamate Ionotropic Receptor Kainate Type Subunit 2. Glutamate receptors are the predominant excitatory neurotransmitter receptors in the mammalian brain

SLC6A12
Transports betaine and GABA. May have a role in the regulation of GABAergic transmission in the brain through the reuptake of GABA into presynaptic terminals, as well as in osmotic regulation.
GABBR1, 2 Encodes a receptor for GABA that functions as a heterodimer with GABA(B) receptor 1 and 2. Defects in this gene may underlie brain disorders such as schizophrenia and epilepsy.
(Continued ) 1646 Luis Alameda et al. the interactions between genotype and known environmental risk factors in influencing the onset of complex diseases (Relton & Smith, 2010). Initially, epigenetic research in psychiatry used a candidate gene approach, and progressively, research moved to Epigenome Wide Association studies (EWAS) ( Table 1). While both designs have their advantages and limitations, the breadth of coverage of EWAS offers a more informative insight on biological pathways. This is based on the rational that chromatin conformation and transcriptional regulation are influenced by a set of methylated or unmethylated cytosines across a region, rather than specific CpG sites in isolation (Mill et al., 2008) Different biological pathways have been implicated in the aetio-pathogenesis across multiple mental disorders. Some of these are pathways related to neurotransmission such as serotonin (Provenzi, Giorda, Beri, & Montirosso, 2016), dopamine or GABA/glutamatergic processes (McCutcheon, Krystal, & Howes, 2020); while others pathways involve inflammation (Cullen et al., 2019), oxidative stress (Steullet et al., 2016), synaptic plasticity and neurogenesis (Claudino, Gonçalves, Schuch, Martins,

Gene names and key terms Biological function and definition
Encodes perforin a pore-forming cytolytic protein found in the granules of cytotoxic T lymphocytes (CTLs) and natural killer cells (NK cells)

CD7
Encodes a transmembrane protein which is a member of the immunoglobulin superfamily found on thymocytes and mature T cells that plays an essential role in T-cell interactions and also in T-cell/B-cell interaction during early lymphoid development MPG Encodes N-methylpurine DNA glycosylase a specific type of DNA glycosylase involved in the recognition of a variety of base lesions, including alkylated and deaminated purines, and initiating their repair via the base excision repair pathway MPOG A member of the XPO subfamily of peroxidase enzyme most abundantly expressed in neutrophil granulocytes MARC2 a The protein encoded by this gene is an enzyme found in the outer mitochondrial membrane that reduces N-hydroxylated substrates CEMIP a Cell migration-inducing and hyaluronal-binding protein, known as KIAA1199, has been shown to bind hyaluronic acid and catalyse its depolymerisation its depolymerisation independently of CD44 and hyaluronidases Oxidative stress

GGT6
Encored for a gamma-glutamyltransferase, that plays a key role in glutathione homoeostasis by providing substrates for its synthesis GSTM5 (Glutathione S-Transferase Mu 5), important for glutathione homoeostasis

NR3C1
Encodes the human glucocorticoid receptor protein, which is the receptor to which cortisol and other glucocorticoids bind miR124 A microRNA that targets NR3C1 FKBP5 (2, 1B) Encodes the FK506 binding protein, a member of the immunophilin protein family which may play a role in immunoregulation and basic cellular processes involving protein folding and trafficking

SKA2
Encodes for a component of the spindle and kinetochore-associated protein complex, which is a protein complex involved in regulating chromosomal segregation. SKA2 is important in facilitating GR nuclear transactivation.

Cannabinoid system
CNR1 and CNR2 a Encodes the cannabinoid receptor gene

DNMTS
This gene encodes an enzyme that transfers methyl groups to cytosine nucleotides of genomic DNA OXTR Encodes oxytocin, which is a neuropeptide hormone produced by the hypothalamus and released into systemic circulation by the posterior pituitary Tobacco signature genes AHRR a The protein encoded by this gene participates in the aryl hydrocarbon receptor (AhR) signalling cascade, which mediates dioxin toxicity, and is involved in the regulation of cell growth and differentiation F2RL3 a Encodes a member of the protease-activated receptor subfamily, part of the G-protein coupled receptor 1 family of proteins. This receptor plays a role in blood coagulation, inflammation and response to pain GFI1 Encodes a nuclear zinc finger protein that functions as a transcriptional repressor. This protein plays a role in diverse developmental contexts, including haematopoiesis and oncogenesis

MYO1G
Is a plasma membrane-associated class I myosin that is abundant in T and B lymphocytes and mast cells.. This myosin is required during immune response for detection of rare antigen-presenting cells by regulating T-cell migration a genes related to cannabis use.
& Rocha, 2020), or the stress response system (Hypothalamic Pituitary adrenal Axis -HPA) (Wesarg, Van Den Akker, Oei, Hoeve, & Wiers, 2020). It is important to take into account that some of these processes participate in disease pathogenesis in a parallel manner, such as via the redox system and through the glutamatergic/GABAergic imbalance (Hardingham & Do, 2016); or the immune system and the stress response (Pariante, 2017). Although these processes are often explored within discrete categorical clinical conditions, they often overlap transdiagnostically. For instance, alterations in serotonin pathways are linked to both depression and psychosis phenotypes (Selvaraj, Arnone, Cappai, & Howes, 2014).
In this review, we set to appraise firstly, the evidence of DNAm modifications both from candidate genes and EWAS studies, associated either specifically or transdiagnostically with psychiatric conditions, and secondly, if these DNAm modifications map onto biological pathways. Thirdly, we will explore if the existing findings from studies on DNAm changes associated with CA and CU, two of the environmental exposures most consistently associated with psychiatric disorders (Lindert et al., 2014;Mandelli, Petrelli, & Serretti, 2015;Marconi, Di Forti, Lewis, Murray, & Vassos, 2016;Sideli, Quigley, La Cascia, & Murray, 2020a;Varese et al., 2012), point at the same biological pathways therefore contributing to the understanding of how these environmental exposures increase transdiagnostic and specific psychiatric liability. Details on methodological considerations can be found in Online Supplementary Material (SM).

Evidence of epigenetic processes in major transdiagnostic pathways
In this section we will review the evidence, predominantly from case-control studies pointing at an association between DNAm changes in the Serotoninergic, Dopaminergic pathways, Excitatory inhibitory balance (including the Glutamatergic and GABAergic dysfunction), Synaptic plasticity and Neurogenesis; the Immune system, Inflammation and Oxidative stress and the major mental conditions (focusing on Eating Disorders (ED): anorexia nervosa (AN) and bulimia nervosa (BN), Autism Spectrum Disorder (ASD), BD and Psychotic Disorder, Depression, Post Traumatic Stress Disorder (PTSD) and Anxiety Disorders). Summary of findings is illustrated in Fig. 1; findings on HPA-axis and its association to environmental risk factors are presented in Section 'The epigenetic signature of childhood adversity and cannabis use' and Fig. 2 and Online Supplementary Table S1 (SM) summarises the characteristics of the articles mentioned in that section. Table 2 summarised the key elements of studies finding evidence of a link between DNAm on genes involved in each biological pathway across all disorders. Screening
Evidence on ED so far has not found an association with SLC6A4 DNAm and AN (Boehm et al., 2019;Pjetri et al., 2013;Steiger et al., 2019).
Another well-explored gene of interest in the serotoninergic pathway is MAO-A (Shih & Thompson, 1999) which is involved in monoamine degradation and it has established linked with depression (Meyer et al., 2006). While studies in depression have found inconsistent DNAm changes (Domschke et al., 2015;Melas & Forsell, 2015;Melas et al., 2013); in candidate gene studies in anxiety disorders, the evidence points at a pattern of hypomethylation (Ziegler et al., 2016) as shown in acrophobia (Schiele et al., 2018) and obsessive compulsive disorder (OCD) (Domschke et al., 2012). Moreover, increased MAO-A DNAm has been suggested as a potential useful marker of better response to psychotherapy in anxiety disorders (Schiele et al., 2020;Ziegler et al., 2016).
Overall, we find a transdiagnostic link between DNAm changes in genes involved in the serotoninergic pathway, with limited evidence in ED (findings on PTSD discussed in Section 'The epigenetic signature of childhood adversity and cannabis use' and described in Table 2).

The dopaminergic pathway
It is widely accepted that dopaminergic dysregulation stands as one of the most supported hypotheses for the pathogenesis of SCZ and psychosis as a whole (Jauhar et al., 2018;McCutcheon et al., 2020). Studies examining DNAm in the blood of patients with SCZ as compared with controls have showed both higher and lower DNAm levels in different DA receptor's genes; with decreased DNAm in DRD3  and DRD4 ; and in other dopamine receptors (Kordi-Tamandani, Sahranavard, & Torkamanzehi, 2013b;Yoshino et al., 2016).
In ED, findings of DNAm changes affecting dopaminergic genes DAT and DRD2 are mixed (Frieling et al., 2010;Pjetri et al., 2013). It has been suggested that DNAm variation in the dopamine pathway in ED may be related to comorbid Borderline Personality Disorder (Borderline PD)  and exposure to CA (Section 'The epigenetic signature of childhood adversity and cannabis use' and Online Supplementary Table S1 (SM)).
None of the EWAS studies conducted in ASD has found evidence supporting an association with DNAm changes involved in the Dopaminergic pathway.
Overall, recent findings support a link between DNAm changes in genes involved in the dopaminergic pathway related to neurodevelopmental disorders such as SCZ, with limited evidence suggesting a link with other conditions.

Glutamatergic/GABAergic Pathway and excitatory/inhibitory balance
Alterations in glutamatergic and GABAergic pathways, which can lead to either excitatory/inhibitory imbalance, have been reported to play a role in the etiopathogenesis of psychotic disorders (McCutcheon et al., 2020) and ASD (Marotta et al., 2020). Furthermore, N-Methyl-D-aspartic acid or N-Methyl-D-aspartate (NMDAR) hypofunction as well as a decrease in the parvalbumin-expressing fast-spiking interneurons (PVI), both processes being essential for the excitatory/inhibitory balance, have been widely shown to be involved in psychotic disorders (Thuné, Recasens, & Uhlhaas, 2016). Note: This figure summarises the evidence presented in this review, highlighting the idea that some biological pathways linking environmental risk factors with mental health disorders via epigenetic changed in the form of DNAm are transdiagnostics (e.g immune system/inflammation) while others seem to be more specific (e.g dopaminergic system). (1) The environmental risk factors row and epigenetic modifications row suggest links between childhood adversity (CA), and Cannabis use (CU) and DNAm changes mapping to biological pathways which are also functionally related (Serotoninergic, Dopaminergic pathways, Glutamatergic & GABAergic pathway, Neurogenesis, Immune system & Inflammation and Oxidative stress). (2)The epigenetic modifications row and mental health disorders row illustrate the evidence, from case-control studies, of an association between DNAm changes in these pathways and the major mental health conditions (Eating Disorders (anorexia nervosa and bulimia nervosa) Post-traumatic stress disorder, Anxiety Disorders, Psychotic Disorder, Bipolar disorders, Depression and Autism Spectrum Disorders). (3) The arrows connecting the three rows show the potential mediating role of DNAm changes linking CA and CU and risk to develop mental health conditions. The thickness of the lines shows the robustness of the evidence reported in the literature review. The items "genotype: and "other risk factors" are added to highlight the influence of genetic factors and environmental confounders in DNAm studies. The dotted line connecting eating disorders with the pathways indicate that literature was limited and mixed not allowing to draw clear links with the pathways.
In terms of EWAS Mill and colleagues (Mill et al., 2008) performed the first EWAS in post-mortem brains of SCZ and BD subjects compared to controls, and found DNAm changes associated with SCZ and BD at loci involved in glutamatergic (GRIA 2, GRIND3B) and GABAergic (MARLIN-1, KCNJ6, HELT) neurotransmission, supporting previous candidate genes results. Findings related to GRIA family genes have been replicated in latter EWAS studies (Aberg et al., 2012;Numata et al., 2014), and other EWAS studies have confirmed DNAm changes in genes involving GABAergic neurotransmission (SLC6A12 and GABBR1) (Hannon et al., 2021).
In ASD, an EWAS study on histone acetylation in participants with the disorder compared to controls found an enrichment of hyperacetylated sites in genes involved in GABA receptor activity (Ramaswami et al., 2020), although this has not been previously found on DNAm (Wong et al., 2018).
Lastly, a Depression EWAS (Nagy et al., 2015) of post-mortem brains of suicide victims and controls found 115 differentially methylated regions (DMRs), which included regions related to GRIK2.
Overall, there is evidence linking DNAm changes on genes involved in the glutamatergic pathway mainly with psychosis, with some evidence suggesting a link with ASD and MDD
RELN is a good studies gene that has been linked to SCZ (Costa et al., 2002). An aberrant DNAm status in RELN has been found in SCZ and BD patient as compared with controls (Fikri et   In the gene and DNAm columns, CA + (with an arrow pointing up) reflects the presence of a positive association between the DNAm in probes located in those genes and CA; CA-(with an arrow pointing down) reflects a negative association. The disorder column shows in which mental health condition that association has been found. Lastly, the clinical outcomes column shows the presence of evidence linking DNAm, with a particular clinical phenotype; CA + indicated that the association between DNAm and the clinical outcome was related to CA.    Interestingly, peripheral blood hypomethylation in the RELN promoter was associated with poor cognitive functioning (Alfimova, Kondratiev, Golov, & Golimbet, 2018).
In ASD, an EWAS study in post-mortem brain found dysregulation in the pathway of phosphatidylinositol 3-kinase (PI3K) activity (Wong et al., 2019), an enzyme that is involved in cellular growth, proliferation and differentiation, and which has been previously been associated with SCZ (Law et al., 2012).
As a whole, transdiagnostic evidence suggests an involvement of DNAm changes in neurogenesis an neural plasticity.

Immune system and inflammation
Abundant evidence supports the role of neuroinflammation and altered immune processes in the aetiopathogeneses of various mental conditions (Mazza, Lucchi, Rossetti, & Clerici, 2020;Pariante, 2017).
In depression, a discordant monozygotic twin study based on peripheral blood, found 39 DMRs associated to a lifetime history of MDD, which were significantly enriched in biological pathways associated to cytokine secretion (Zhu et al., 2020). Another EWAS on post-mortem brain of people with late-MDD (Hüls et al., 2020), found altered DNAm in the YOD1 locus, which is dysregulated in depression (Howren, Lamkin, & Suls, 2009) and its implicated in the regulation of inflammatory processes (Schimmack et al., 2017).
Two EWAS studies in PTSD found differences in DNAm across genes part of biological pathways involved with inflammation and immune response (Kuan et al., 2017;Uddin et al., 2010).
Furthermore, an EWAS study from patients suffering from Panic Disorder found enrichment in genes involved in the regulation of lymphocytes (Shimada-Sugimoto et al., 2017).
We can conclude that there is transdiagnostic, rather than specific, a link between DNAm changes in the immune system and inflammation and neural plasticity, although evidence is more robust in SCZ.

Oxidative stress
There is converging evidence pointing at a role of redox dysregulation as a possible mechanism involved in the aetiopathogenesis of both ASD (Bjørklund et al., 2020) and psychosis (Perkins, Jeffries, & Do, 2020). Oxidative stress has been shown to play a role in epigenetic modifications, enhancing inflammatory gene transcription (Rahman, Marwick, & Kirkham, 2004). Oxidation of 5mC to the 5-hydroxymethylcytosine (5hmC) is considered a key step in the reversibility of DNA methylation, which can have important therapeutic implications. Moreover, glutathione,  (Hing et al., 2018), Schizophrenia (Di Carlo et al., 2019), and eating disorders (Thaler and Steiger, 2017), therefore studies mentioned here are just examples of the literature in this particular domain. When various genes are reported in the same pathway and the same study, but no specific information on clinical/biological outcome or specific direction if the association is provided, these genes have been put in the same row (e.g Asberg et al., and Liu et al.,). When an arrow is next to the author's name it reflects the direction of the DNAm of the particular gene in in relation to the condition ↑ : increased ↓: decreased DNAm. When in column 1 there is no arrow is because information could not be obtained or was not clear, and the presence of that gene indicates the association of DNAm in that gene with the respective condition (differently methylated). When a three step sequence separated by an arrow is presented, this refers to mediation analyses (e.g peng et al.,: CA → ↑SLC6A4 → depressive symptoms: DNAm of SLC6A mediates the effect of CA on depressive symptoms). CA: childhood adversity; CU: cannabis use. Definition of each gene is presented in Table 1. DLPFC: Dorsolateral prefrontal cortex. ASD: autism spectrum disorder; SCZ: schizophrenia. PTSD: post-traumatic stress disorder; Borderline PD: Borderline personality disorder; MDD: major depression disorder; BD: bipolar disorder.

1656
Luis Alameda et al.
the major antioxidant in the brain, is involved in the methionine cycle, and depletion of glutathione can be detrimental for the DNAm process (García-Giménez, Roma-Mateo, Perez-Machado, Peiro-Chova, & Pallardó, 2017). Although evidence examining this pathway in the context of epigenetics is scarce, some EWAS have shown interesting results: one study examined prospectively the association of EWAS methylation changes with the transition to psychosis (Kebir et al., 2017), and found an enrichment of pathways related to oxidative stress regulation in those transitioning. Furthermore, an EWAS study in Borderline PD found differences in methylation in GCT6, which is important in glutathione metabolism (Arranz et al., 2021).
The epigenetic signature of childhood adversity and cannabis use The characteristic of the studies described in this section can be found in Online Supplementary Table S1 (SM), and in Table 2.

Hypothalamus pituitary-adrenal axis pathway
While multiple studies have explored the link between epigenetic modification involved in the HPA-axis, and psychiatric disorders, recent evidence is beginning to indicate that some of these epigenetic modifications might follow exposure to CA. The latter is a robustly replicated risk factors for many psychiatric disorders (Online Supplementary Table S1 (SM) summarises the main findings on studies examining the link between DNAm and genes involved in the HPA-axis, and key findings are summarised in Fig. 2). As a whole, as illustrated in Fig. 2, at the level of NR3C1 there is consistent evidence on a positive correlation between CA and DNAm in Borderline PD and MDD and some clinical outcomes Martin-Blanco et al., 2014;Perroud et al., 2011;Radtke et al., 2015), and a negative correlation with anxiety and PTSD (Labonte, Azoulay, Yerko, Turecki, & Brunet, 2014;Schechter et al., 2015;Wang et al., 2017;Yehuda et al., 2015). Lower DNAm in FKBP5 is associated with CA in psychosis and PTSD (Klengel et al., 2013;Misiak et al., 2020); while in depression, 3 studies found no such link (Bustamante et al., 2018;Farrell et al., 2018;Klinger-König et al., 2019), as opposed to another study . As for NR3C1, findings on FKBP5 DNAm are different across disorder, suggesting a divergent transdiagnostic mechanism involving in HPA related genes (see Fig. 2). The SKA2 interacts with adversity scores in predicting lifetime suicide attempt (Kaminsky et al., 2015), and mediated the association between reduced cortical thickness and PTSD (Sadeh et al., 2016a) and suicide phenotypes (Sadeh et al., 2016b).

Childhood adversity
With regards to the serotoninergic pathway, while hypomethylation in SLC6A is associated with resilience to PTSD (Koenen et al., 2011), hypermethylation of SLC6A has been linked to exposure to CA and associated with the worst clinical presentation in MDD (Kang et al., 2013). Moreover, hypermethylation in the 5-HT3A-R gene appeared to mediate the link between exposure to adversity and higher severity of disease parameters in a mixed sample of BD and Borderline PD (Perroud et al., 2016).
Moreover, hypomethylation of MAOA, a gene important for the degradation of serotonin and DA (Xu, Jiang, Gu, Wang, & Yuan, 2020), appears to partially mediate the known association between CA and depressive symptoms, alongside with other stress-related genes such as BDNF and NR3C1 and SLC64 (Peng et al., 2018). Moreover, MAOA DNAm was negatively correlated to life events in a sample of Panic Disorder (Domschke et al., 2012).
In relation to DA, one study in patients with bulimia spectrum disorders found no differences in DRD2 DNAm when comparing those with exposure and non-exposure to trauma .
At the level of the Glutamatergic pathway, one study found that exposure to CA was associated with decreased DNAm in GAD in a sample of Panic Disorder (Domschke et al., 2013). Lastly, a candidate gene study (Engdahl, Alavian-Ghavanini, Forsell, Lavebratt, & Rüegg, 2021) and an EWAS (Weder et al., 2014) linked CA to increased methylation levels in GRIN2B/ GRIND1 genes, suggesting evidence that this change may lead to the onset of depressive symptoms.
As a whole, DNAm changes in some of the genes that have been previously linked to major psychiatric conditions (Section 'Evidence of epigenetic processes in major transdiagnostic pathways', Table 2), such as SLC6A, 5HT3A-R, A (MAOA), BDNF, GAD and the GRIND family, (related to the serotoninergic, and glutamatergic pathways respectively) are also associated to CA. This suggests that some of the DNAm changes attributed to these disorders may be partially related to the consequence of CA exposure, as illustrated in Fig. 1.
Cannabis use CU and in particular heavy use (Marconi et al., 2016) has been consistently associated with increased risk for PD, but to a lesser degree for other psychiatric conditions (Sideli, Trotta, Spinazzola, La Cascia, & Di Forti, 2020b). In recent years, candidate genes studies from peripheral blood have reported changes in DNAm associated with heavy CU in genes involved in dopamine transmission, such as DRD2 (Gerra et al., 2018), DAT1 (Grzywacz et al., 2020) and COMT (Van der Knaap et al., 2014) and in the CB1 and CB2 receptors genes part of the endocannabinoid system (Rotter et al., 2012;Tao et al., 2020). The latter playing an important role in brain development and synaptic transmission.
A recent study investigated the effect of heavy CU with and without tobacco on EWAS (Osborne et al., 2020). The analyses in the sample with both cannabis and tobacco use identified differentially methylated sites in 2 genes, AHRR and F2RL, previously reported to be affected by tobacco exposure. Within the sample of cannabis users without tobacco, while none of the differentially methylated loci reached EWAS significance, an exploratory analysis showed enrichment for genes involved in the signalling pathway, including glutamatergic transmission, brain function and mood disorders. Moreover, these exploratory analyses show two differentially methylated sites significantly associated with both only CU and cannabis with tobacco, which are within the MARC2 gene. The latter previously linked to adverse effects to antipsychotics in schizophrenia (Åberg et al., 2010) and within the CUX1 gene which is involved in neuronal development (Platzer et al., 2018).
Furthermore, recent whole blood and cell-specific Methylome-wide association studies (MWAS) from a sample of adolescents with CU disorder pointed at many methylation sites Psychological Medicine relevant to brain function and to neurodevelopment (Clark et al., 2021). These included CpGs located in the CLMN gene and the SENP7 gene, expressed in the brain and playing a role in brain developmental and synaptic function and organisation (Juarez-Vicente, Luna-Pelaez, & Garcia-Dominguez, 2016;Marzinke & Clagett-Dame, 2012). Interestingly, the pathway analyses based on the cell type-specific significant DNAm changes associated with CU implicated pathways such as the Slit-Robo signalling (granulocytes) under the regulatory control of the endocannabinoid system during brain cortical development (Alpár et al., 2014), the ErbB signalling pathway (T-cell) and pathways involved in DNA repair (B-cell) (Clark et al., 2021).

Inflammation, oxidative stress, synaptic plasticity and neurogenesis
Childhood adversity A number of EWAS studies conducted in clinical samples have reported an association between exposure to CA and DNAm changes across genes involved in inflammation. For instance, a study (Prados et al., 2015) found a positive correlation between the IL17RA DNAm and CA in a Borderline PD and MDD sample. Other evidence suggests a negative correlation between DNAm in genes enriched for immune pathways (such as TLR1 and TLR3) and CA in PTSD subjects (Uddin et al., 2010); while the TNFRSF13C gene was differently methylated between Borderline PD participants with and without CA (Arranz et al., 2021) (See Online Supplementary Table S1 (SM) -EWAS section).
Candidate genes studies have linked CA with DNAm changes in BDNF Thaler et al., 2014;Weder et al., 2014), consistently with EWAS data reporting DNAm changes affecting genes involved in neurogenesis (Prados et al., 2015;Uddin et al., 2010). For instance, three EWAS studies in BD (Comes et al., 2019), Borderline PD (Arranz et al., 2021) and MDD (Lutz et al., 2017) have consistently shown changes in DNAm in genes from the POU family that are associated with CA (POU6F2, POU5F1 and POU3F1 respectively), which are genes involved in myelinisation and neurogenesis (Online Supplementary Table S1 (SM) -EWAS section).
A recent EWAS study found differences in DNAm of the GGT6 gene that were associated with exposure to CA in a sample of Borderline PD patients; GGT6 is key for glutathione homoeostasis (Arranz et al., 2021), it is also the main antioxidant and redox regulator that has previously been associated with SCZ aetiopathogenesis (Steullet et al., 2016). Further evidence is summarised in Online Supplementary Table S2 (SM).
As a whole, candidate gene and EWAS studies suggest a link between CA and genes involved in the inflammatory and neurogenesis pathways, with some preliminary evidence suggesting a link between CA and DNAm and oxidative stress genes (Fig. 1).

Cannabis use
The largest to date case-control study to examine the effect of lifetime CU on DNAm reported an epigenome-wide-significantly differentially methylated CpG site within the CEMIP gene (Markunas et al., 2020). The CEMIP gene, involved in hyaluronic catabolism, which has been shown to have an important role in inflammation, immune processes as well as associated with BD and SCZ previously (Petrey & de la Motte, 2014).
The confounding effect of smoking is clearly evidenced in an EWAS study on peripheral blood of SCZ patients (Hannon et al., 2016). A similar study examining the impact of CA on the epigenome in a general population found that tobacco consumption was an important confounding when examining the signature of CA (Marzi et al., 2018). Whether some of these epigenetic changes associated with tobacco exposure could also mediate the already reported link between tobacco use and increased risk of psychosis, it is an important question yet to be determined (Gurillo et al., 2015), and tobacco smoking should be accounted for in the future epigenetic studies in psychiatry.

Alcohol use and abuse
There is some initial evidence to suggest that alcohol use is associated with DNAm changes (Liu et al., 2016;Wang, Xu, Zhao, Gelernter, & Zhang, 2016;Weng, Wu, Lee, Hsu, & Cheng, 2015). Enrichment analyses examined DNAm in alcohol users have revealed enrichment in pathways related to neural degeneration (Weng et al., 2015), and in genes important for neurogenesis (NPDC1), inflammation (HERC5) and in GABA receptors (a receptor delta and B receptor subunit 1); all of which are pathways previously associated with different mental disorders, as shown in Fig. 1. However, studies rarely account for such covariates, which is currently a limitation of current literature.

Psychiatric medication
The extent of the data reporting the DNAm changes associated with psychiatric medication would require a separate review. Indeed, there is consistent evidence that pharmacological agents can trigger DNAm in similar or opposite directions than those attributed to the disease. For example, Lithium, Carbamazepine and Quetiapine, often prescribed for the treatment of BD, are associated with decrease methylation of SLC6A4 (Asai et al., 2013;Sugawara et al., 2015), in contrast with the hypermethylation reported in BD in that gene (Table 2). Similarly, studies who have investigated the effect of antipsychotic medication, have shown, on the one hand, that Haloperidol affects DNAm in leucocytes of SCZ patients (Melas et al., 2012), while on the other hand, a recent EWAS study showed that Clozapine exposure leads to DNAm differences in patients with treatment-resistant SCZ ) as compared to controls. Thus, it is key to consider the possibility that some of the changes in DNA pathways may be led by agents rather than the disease itself, highlighting the need to account for medication in future studies and to consider epigenetics as a potential mediating mechanism of action of the beneficial effects of medication in the brain.

Summary and outstanding questions
As illustrated in Fig. 1, many of the epigenetic dysregulations we report are transdiagnostic, such as those affecting the

1658
Luis Alameda et al.
serotoninergic, inflammatory and neurogenesis pathways, while others such as the Glutamatergic/GABAergic pathway are shared between a couple of disorders (e.g. SCZ and ASD), or disorder specific such as the dopaminergic pathway in PDs. These are pathways that have been classically implicated in the aetiopathogenesis of psychiatric phenotypes; additional emerging pathways such as oxidative stress remain to be further explored. Moreover, CA, is transdiagnostically associated to psychiatry morbidity, and seems to play a role in the DNAm dysregulation of many of these pathways. Furthermore, the preliminary DNAm changes so far reported associated with CU affect pathways previously link to psychosis, suggesting potential mediating venues to be tested in clinical populations (Fig. 1).
In addition, CA is associated with DNAm changes both in the general population (Kandaswamy et al., 2020) as well as in clinical samples with a psychiatric diagnosis (Online Supplementary Table S1 (SM)). This might suggest that the DNAm changes associated to CA exposure predate disease onset and could represent a marker of acquired psychiatric liability. However, evidence formally testing mediating pathways EWAS level between CA and the main clinical conditions is non-existent in humans. Candidate gene studies tend to find the inconsistent direction of the association between CA across disorders, or findings are inconsistent within disorders as shown in Fig. 2 and Table 2 and Online Supplementary Table S1 (SM). One explanation could be that there are other causative partners that are being missed in the equation, that may explain such inconsistency, such as the role of genotype, gene expression or a more thorough assessment of specific adversities in the context of protective factors and its link with more carefully selected clinical phenotypes.
The existing findings from epigenetics research need to be appraised in the context of well-known technical limitations epigenetics, such as the blood-brain inconsistencies, tissue-type specificity (Bakulski, Halladay, Hu, Mill, & Fallin, 2016;Nikolova & Hariri, 2015) and the candidate gene v. EWAS issue (see Palma-Gudiel, Córdova-Palomera, Leza, & Fañanás, 2015). Moreover, evidence suggests that variation in DNAm depends not only on the environment, but also on genetic factors (Bell & Spector, 2012). Although some studies presented in this review have found evidence that some genotypic variation in some risk alleles can influence DNAm (Klengel et al., 2013;Melas et al., 2013;Perroud et al., 2016), EWAS addressing the joint effect of genotype and environment are still in its infancy (Min et al., 2021). Addressing this issue will prove methodologically challenging, but methods quantifying the genetic influences on DNAm, such as the methylation quantitative trait loci (mQTL) should be used in relation to the presence of environmental insults. Moreover, studies included in this work are often small (Online Supplementary Table S1 (SM)) and thus underpowered, except some exceptions , which presents the need to create collaborative efforts allowing meta-analysis of comparable epigenetic data. Furthermore, evidence of the environmental exposure impact through epigenetic modification in psychiatric diseases or phenotypes is still limited, with studies focusing mainly on exposure to CA and only preliminary results of the effect of cannabis. Given the replicated but differential impact of multiple environmental risk factors in major psychiatric disorders , future studies exploring epigenetic variation as a mediator between genetic vulnerability and various environmental factors (not only CA) should be addressed, using novel methods specifically developed for mediation using EWAS data (Liu et al., 2021). Another important factor is the phenotypic characterisation for environmental exposure. For instance, most of the studies in this work have used broad measures of adversity, using a composite cumulative score, rather than differentiating between neglect of abuse. The same can be said for the measures of CU which little reflects the level of exposure none to affect psychiatric liability. Moreover, the outcomes are often considered as SCZ, or MDD or even major psychoses (combining SCZ & BD), which are extremely heterogeneous entities, involving microphenotypes (Maj et al., 2021), and which accordingly may have very different biological underpinnings. Evidence is showing that there are some levels of specificity between adversity subtypes and symptoms domains, for example, abuse is more related to positive symptoms while neglect is not  and that CU is associated with paranoia (Freeman et al., 2005). Thus, using a composite measure of CA and broadly defined conditions when trying to understand specific mediating epigenetic pathways may consider such specific links between environment and psychopathology first. Accordingly, this work suggests that some biological pathways are operating transdiagnostically, and therefore a phenotypic characterisation based on clinical dimensions may be more biologically informative than diagnostic categorisations. Furthermore, the timing of environmental exposure should be addressed, given evidence that a disruption in epigenetic programming occurs across different time windows throughout the life span (Massicotte, Whitelaw, & Angers, 2011). In this line, the lack of information on the timing of trauma and of CU initiation could explain some of the inconsistencies mentioned in our review (Fig. 2). For example, we reviewed studies showing increased methylation of the serotonin transporter in depressed individuals exposed to trauma (usually when adversity occurs before adulthood), which contrasts with the lower methylation in the same gene in PTSD, when exposure tends to be later in life.

Conclusions
Future Research should include the influence of gender and how it can modulate the links between DNAm and mental disorders, or how it can affect the influence of CA on DNAm. More effort should go towards designing studies that integrate genetic data with the often-neglected effect of environmental exposures (e.g. recreational drugs and psychotropic medication). Specifically, collaborative efforts between geneticists, epigeneticists and epidemiologists will lead to increased understanding of how the DNAm changes mapping to specific pathways, might mediate the biological link between environmental exposures and increased liability to specific or transdiagnostic psychiatric morbidity.
Supplementary material. The supplementary material for this article can be found at https://doi.org/10.1017/S0033291721005559.