The endocannabinoid system, composed of endogenous lipids, their target receptors and metabolic enzymes, has been implicated in multiple biological functions in health and disease, both in the central nervous system and in peripheral organs. Despite the exponential growth of experimental evidence on the key role of endocannabinoid signalling in basic cellular processes, and on its potential exploitation for therapeutic interventions, much remains to be clarified about the respective regulatory mechanisms. Epigenetics refers to a set of post-translational modifications that regulate gene expression without causing variation in DNA sequence, endowed with a major impact on signal transduction pathways. The epigenetic machinery includes DNA methylation, histone modifications, nucleosome positioning and non-coding RNAs. Due to the reversibility of epigenetic changes, an emerging field of interest is the possibility of an ‘epigenetic therapy’ that could possibly be applied also to endocannabinoids. Here, we review current knowledge of epigenetic regulation of endocannabinoid system components under both physiological and pathological conditions, as well as the epigenetic changes induced by endocannabinoid signalling.
Endocannabinoids (eCBs) are endogenous lipids that activate the cannabinoid receptors, which are G-protein-coupled receptors also activated by Δ9-tetrahydrocannabinol (THC), the psychoactive component of marijuana [1, 2]. To date, two cannabinoid receptors have been cloned (CB1 and CB2) and two major eCBs have been characterized in mammals, N-arachidonoylethanolamine (anandamide, AEA)  and 2-arachidonoylglycerol (2-AG) [4, 5]. More recently other molecules, such as 2-arachidonoyl-glyceryl ether (noladin ether) , O-arachidonoyl-ethanolamine (virhodamine)  and N-arachidonoyl-dopamine , have been proposed as possible CB receptor agonists, yet their biological activity and pathological relevance await clarification.
CB1 and CB2 are the best-studied molecular targets of AEA and 2-AG, which bind to and activate them with different affinity. CB1 is the most abundant G-protein-coupled receptor found in the brain  and is responsible for mediating most of the neurobehavioral effects of THC [10, 11]. Consistent with the known eCB central effects, CB1 is abundant in brain areas involved in memory (e.g. hippocampus), motor coordination (e.g. basal ganglia, cerebellum) and emotional processes (e.g. prefrontal cortex) [12, 13]. CB1 is preferentially localized at the presynaptic level and eCBs synthesized by postsynaptic neurons can travel backwards and inhibit neurotransmitter release , i.e. the so-called retrograde signalling . CB1 is also present in peripheral tissues, including adipose tissue, liver and skeletal muscle .
CB2 is predominantly expressed in immune cells , where it appears to play a role in mediating the immunosuppressive effects of eCBs. However, recent findings suggest that CB2 is also present at low levels in some areas of the brain [18-20], where it is markedly activated upon insults , although the molecular details are rather complex and poorly understood.
Recently, AEA and 2-AG were reported by some authors , but contradicted by others , to activate GPR55, an orphan G-protein-coupled receptor, which shares low sequence homology (10–15%) with the other two CB receptors [24, 25].
AEA also interacts with several non-cannabinoid receptors , the best characterized of which is the transient receptor potential vanilloid 1 (TRPV1) channel, activated by AEA at an intracellular site . Other evidence has documented the interaction of eCBs with peroxisome-proliferator-activating receptors (PPARs) α and γ, although at high concentrations , with major implications for gene expression regulation .
eCB synthesis and degradation
AEA and 2-AG are prototypes of a much larger class of lipids, termed N-acyl-ethanolamines (NAEs) and monoacylglycerols (MAGs) respectively. Individual members of each group differ in length and degree of unsaturation of their acyl chains. Many potential biological activities have been attributed to NAEs and MAGs in vivo [30, 31]. However, most of them do not bind to CB receptors but seem to have an ‘entourage effect’, whereby they potentiate the activity of AEA or 2-AG at their receptors by inhibiting their degradation . Due to their lipophilic nature, eCBs cannot be stored in vesicles. Indeed, it is widely accepted that, unlike other mediators, eCBs are synthesized and released on demand through multiple biosynthetic pathways, which include N-acyl-phosphatidylethanolamine-hydrolysing phospholipase D for AEA  and sn-1-specific diacylglycerol lipases for 2-AG , as well as other biosynthetic enzymes that await further investigation .
The degradation of eCBs also occurs through multiple pathways, which include fatty acid amide hydrolase (FAAH) [36, 37] and monoacylglycerol lipase (MAGL) , the main hydrolytic enzymes for AEA and 2-AG respectively. Additionally, other enzymes showing an ‘amidase signature’, such as FAAH-2  and the N-acylethanolamine-hydrolysing acid amidase  which belongs to the choloylglycine hydrolase family, might hydrolyse AEA, thus releasing arachidonic acid and ethanolamine. Furthermore, both AEA and 2-AG, possibly under conditions in which the activity of MAGL or FAAH is suppressed, might become substrates for cyclooxygenase-2 and give rise to the corresponding hydroperoxy derivatives . These metabolites show different activity at CB1/2 or other ECS elements , or appear to act at new binding sites, giving pharmacological effects still obscure at molecular level [43, 44]. Overall, the physiological relevance of the oxidative pathways of eCBs still needs to be clarified . The degradation of eCBs by specific hydrolases follows their transport inside the cell . Because of their lipophilic nature, AEA and 2-AG can diffuse passively through lipid membranes. However, it appears that diffusion is accelerated by a rapid and selective carrier system. AEA appears to be taken up by several cell types at least in part via a facilitated transport mechanism, possibly mediated by a purported AEA membrane transporter, which can also carry 2-AG and is therefore referred to as endocannabinoid membrane transporter (EMT) . Very recently, a truncated FAAH, termed FAAH-1-like anandamide transporter, has been reported in neural cells, where it serves as an intracellular carrier that binds AEA but does not cleave it . Recent insights suggest that the metabolic control of the endogenous tone of AEA is complemented by intracellular trafficking via intracellular binding proteins, particularly fatty acid binding proteins , albumin and heat shock protein 70 , and storage in specific reservoirs (adiposomes) (for a detailed review see ref. ). Altogether biosynthetic, hydrolytic or oxidative enzymes exert a metabolic control of the endogenous tone, and hence the biological activity of eCBs, AEA, 2-AG and their congeners, together with their target receptors and metabolic enzymes, purported EMT and intracellular transporters, form the ECS (Fig. 1).
As far back as 1942, Waddington defined epigenetics as ‘the branch of biology which studies the causal interactions between genes and their products, which bring the phenotype into being’ . Epigenetics thus allows an understanding of the gap between genotype and phenotype through the study of reversible heritable changes in gene expression that occur independently of a change in DNA sequence. These heritable epigenetic modifications can be divided in DNA methylation, post-translational modifications of histones (acetylation, methylation, phosphorylation etc.), nucleosome positioning and small, non-coding RNAs (microRNA, endogenous small interfering RNA) (Fig. 2). The interaction between all these factors is responsible for the final regulation of gene expression. It is still under debate if all histone modifications as well as many non-coding RNAs should be considered epigenetic marks, because just a few of these modifications or RNAs are likely to be self-perpetuating and inherited . However, all of them can alter gene expression without involving DNA sequence alterations and thus are currently classified as epigenetic modifications.
The most widely studied epigenetic mechanism, the covalent modification of DNA, is catalysed by enzymes known as DNA methyltransferases (DNMTs) through the addition of a methyl (–CH3) group to cytosine residues at the 5-position of the pyrimidine ring [53, 54]. DNMT1 has been recognized as the maintenance DNMT regenerating the methyl-cytosine marks on the newly synthesized complementary DNA strand arising from DNA replication, whereas DNMT3a and DNMT3b induce de novo DNA methylation. There are also two other non-canonical enzymes: DNMT2 and DNMT3L. DNMT2 is not a true DNMT and its biological role remains unclear. DNMT3L is structurally closely related to DNMT3a and DNMT3b, but it appears to be inactive on its own since it needs to associate with both DNMT3a and DNMT3b; DNMT3L could be responsible for the recruitment of histone deacetylases to direct repression . DNA methylation typically occurs at cytosine residues of the CpG dinucleotides (CpG sites), usually rare in mammalian genomes (~ 1%) and partly clustered into so-called CpG islands . The latter are regions of the genome at least 200 bp long, with a GC percentage that is higher than 50% and with a ratio between observed/expected CpG number of at least 0.6. The observed number of CpG is the number of times a C is found followed immediately by a G. The expected number of CpG is calculated as the number of CpG dinucleotides you would expect in a specific window, based on the frequency of Cs and Gs in that window .
Sixty per cent of human genes have CpG islands in the promoter region or first exon, and DNA methylation in the promoter region is associated with suppression of gene transcription [53, 54]. The mechanism responsible for gene silencing starts with recruitment of methyl-DNA binding proteins (i.e. MeCP2) by methylated DNA. These in turn will recruit histone deacetylases (HDACs) that remove acetyl groups from histone proteins and chromatin-remodelling complexes. All these events cooperate in compacting chromatin structure and thus limiting the accessibility to transcriptional machinery. It should also be noted that DNA methylation does not occur exclusively at CpG islands. In fact, recently tissue-specific DNA methylation has been found at CpG island shores, outlying areas close to CpG islands (up to 2 kb distance), and strongly related to gene expression inactivation [58, 59].
Genomic DNA is packaged into a highly compact structure to form chromatin. The basic repeating units of eukaryotic chromatin are the nucleosomes, composed of short stretches of DNA (147 bp) wrapped around a core of histone proteins (H3, H4, H2A and H2B, each in two copies)  and joined together by linker DNA and linker histone H1. Regulation of chromatin structure and transcription is driven by post-translational modifications of the core histones, primarily in the N-terminal tails , that include acetylation, methylation, phosphorylation, ubiquitination, ADP addition and ribosylation . It is now well established that distinct histone modifications, occurring at selected residues and different times, on one or more tails, provide a ‘histone code’ that mediates distinct downstream events [63-66].
One of the most frequent epigenetic modifications is histone acetylation, associated with transcriptional activation  and occurring at different positions of histone H3 lysines (K9, K14, K18 and K23) and of histone H4 (K5, K8, K12 and K16) [61, 67]. Acetylation and deacetylation depend on the balance between histone acetyltransferases and HDAC activities. Unlike histone acetylation, histone methylation may be either activating or repressing gene expression, mainly depending on the sites of methylation  (Fig. 2). For example, K9 and K27 methylations of H3 are associated with transcriptional silencing, whereas methylations of K4, K36 and K79 of H3 have been linked to gene activation [68, 69]. Histone methylation, a relatively stable (and once thought to be irreversible) epigenetic mark, can interact with DNA methylation participating in chromatin remodelling [70, 71].
Nucleosome positioning also plays an important role in how chromatin structure regulates gene transcription. Depending on physical alterations of nucleosomes and how close they are to transcription start sites (TSSs), activators and transcription factors can access or not DNA, thus activating or inhibiting gene transcription. Nucleosome-free regions at the 5′ and 3′ UTRs of genes provide the sites of transcription machinery assembly . Gene activation occurs when there is loss of nucleosomes upstream of the TSS, and thus transcription factors can bind. In contrast, a gene repression occurs if the TSS is occupied by a nucleosome and thus the transcription machinery does not bind. There is a close relationship between nucleosome positioning and DNMTs, because the latter enzymes preferentially target nucleosome-bound DNA .
MicroRNAs (miRs) are RNAs 18–23 nucleotides long that function as post-transcriptional regulators. miR genes are transcribed into large primary miR (pri-miR), processed in the nucleus to a precursor (pre-miR) before being exported to the cytoplasm (mature miR), where they regulate mRNA translation by binding to complementary sequences inducing a silencing complex . Multiple mRNAs can be regulated by each miR . It has been estimated that miRs can target 30% of the human genes in a temporal and tissue-specific manner . miR transcriptional repression can occur via modification of histones due to the RNA-induced transcriptional silencing complex, which binds to miRs and induces post-translational modification of histone tails .
Epigenetic regulation of ECS components
Cannabinoid receptor genes interact with different transcription factors, many of which are implicated in DNA methylation and histone post-translational modifications . Among the different ECS components, so far most attention has been focused on the epigenetic regulation of CB1, in both preclinical and clinical contexts, probably because CB1 mRNA is deregulated in different pathological conditions and upon exposure to different drugs .
It has been observed that DNA hypermethylation of the CB1 gene (cnr1) promoter contributes to downregulation of CB1 transcription in colon cancer specimens . These findings further support the association of reduced CB1 expression with the progression of colorectal cancer , since it has already been suggested that CB1 provides intrinsic protection against colon inflammation .
Consistently, another study on the human epithelial colon cell line LS-174T shows that prostaglandin E2 downregulates CB1 gene expression by increasing DNA methylation of cnr1 promoter , thus promoting tumour growth.
A high (yet not significant) level of DNA methylation at cnr1 promoter has also been observed in rodents after maternal separation in the first generation germline , a finding that corroborates the already well-established role of CB1 in emotional regulation .
Another study showed that two epigenetic modulators, 5-aza-2′-deoxycytidine, an inhibitor of DNMTs that induces DNA hypomethylation, and trichostatin A, a histone deacetylase inhibitor that induces histone hyperacetylation, can regulate cnr1 and cnr2 expression in distinct different cells of the nervous and immune systems . The induction of CB1 and CB2 by these epigenetic modifiers was found in the same cells where the expression of the receptor protein was silenced. Indeed CB1 mRNA levels were increased in human Jurkat T cells but not in human neuroblastoma SH-SY5Y cells, and CB2 in SH-SY5Y cells but not in Jurkat T cells . Since CB1 was induced in Jurkat cells also in response to various cytokines (e.g. interleukin-4) [85, 86], it was suggested that epigenetic mechanisms and signalling pathways that can induce gene expression in T cells through cytokines are functionally connected .
In this context, it has also been hypothesized that the increase in the number of CB receptors by epigenetic mechanisms may be beneficial in order to potentiate the effects of CB1 ligands, e.g. as anti-inflammatory drugs in multiple sclerosis .
To date, there are few studies on the epigenetic regulation of ECS components in a clinical context. An upregulation of CB1 gene expression has been observed in the blood of patients with different eating disorders, but these alterations were not found to be associated with DNA methylation of the cnr1 promoter . In a very recent study, CB1 gene expression and methylation of its promoter were found to be altered in peripheral blood cells of subjects with THC dependence, where cnr1 promoter was found to be highly methylated, in keeping with a lower amount of CB1 mRNA . These findings led the authors to suggest CB1 epigenetic regulation in peripheral blood lymphocytes as an easily accessible biological marker for the study of THC action and dependence .
Moreover, the possible epigenetic regulation of ECS elements has recently been investigated in peripheral blood mononuclear cells from late onset (age > 65 years) Alzheimer's disease (LOAD) subjects. The faah gene clearly emerged as the only one to be altered in LOAD subjects due to a reduced DNA methylation at its promoter compared with healthy controls . Remarkably, the lowest levels of methylation were observed in patients with the most severe cognitive impairment . These findings suggest a possible role of FAAH as a peripheral biomarker and as a new potential therapeutic target for Alzheimer's disease. Additionally, faah gene was identified as the first direct target of 17β-oestradiol (E2) in Sertoli cells whereby the enhancement of faah promoter activity engages E2 receptor β (ERβ) and histone demethylase LSD1, and requires chromatin configuration competent for transcription . Overall, E2 induces epigenetic modifications at the faah proximal promoter compatible with transcriptional activation. After E2 treatment, a dramatic demethylation of the CpG island located in the proximal region of the faah promoter was observed, as well as demethylation of histone 3 lysine 9 (H3K9me3), an important mark associated with transcriptional activation. E2-stimulated transcription of faah gene is accompanied by a decline in H3K9me3 at the oestrogen-response element (ERE) 2/3 sites, suggesting a role for these modifications in the transcriptional activation by E2. Sertoli cells play a crucial role in supporting and regulating germ cell development in the testis, and sperm production is linked to the number of Sertoli cells . The FAAH substrate AEA induces apoptosis of Sertoli cells, and increasing its degradation reduces cell death . Thus, these findings could have a clear impact on spermatogenesis, suggesting E2 as a pro-survival hormone that promotes AEA removal by FAAH. Finally, it should also be mentioned that in a genome-wide integrative analysis of DNA methylation levels in glioblastoma patients, faah gene was one of the 13 found to display an inverse correlation between promoter methylation and expression level: faah promoter was hypermethylated and gene expression was downregulated . Current knowledge on epigenetic regulation of ECS components is summarized in Table 1.
Table 1. Overview of studies on the epigenetic regulation of ECS components. PGE2, prostaglandin E2; 5-Aza-dC, 5-aza-2′-deoxycytidine; TSA, trichostatin A; LOAD, late onset Alzheimer's disease; E2, β-oestradiol.
Human colon cancer
Increased DNA methylation at gene promoter associated with colorectal cancer progression
The possibility that eCBs could act as epigenetic factors was already proposed several years ago. These compounds can reduce gene expression of histones  while inducing alterations in the enzymes responsible for the addition and removal of functional groups (i.e. acetate, methyl, phosphate) to genome-associated proteins. The latter, by interacting with DNA, could be responsible for eCB-induced changes in transcriptional activity . Indeed, it has been reported that activation of CB receptors could induce alterations in the expression of key genes of various neurotransmitter systems .
Already a few years ago it was hypothesized that marijuana smoking might alter DNA methylation , on the basis of preliminary studies reporting the hypermethylation of DNA repair genes in marijuana-addicted individuals . More recently, experimental evidence has shown that maternal cannabis use alters the developmental regulation of mesolimbic dopamine D2 receptors in offspring through epigenetic mechanisms, specifically histone lysine methylation . It was also reported that adolescent THC exposure was associated with chromatin modifications, mediating in the nucleus accumbens of rats the upregulation of proenkephalin, the gene encoding the opioid neuropeptide enkephalin, through reduction of repressive histone 3 lysine 9 methylation . The reduction of dopamine D2 receptor mRNA after prenatal THC exposure was linked to an increase of histone 3 lysine 9 dimethylation (H3K9me2) and a decrease of histone 3 lysine 4 trimethylation (H3K4me3), modifications that have a well-documented antagonistic role in gene regulation during brain development . Based on these findings, it has been hypothesized that cannabis prenatal exposure, by inducing these and possibly other alterations of epigenetic mechanisms in the brain, could cause phenotypical abnormalities during development . A role for eCBs in the regulation of H3K9me3 has also been ascertained in glioma stem-like cells after differentiation. Using the selective CB1 agonist HU-210 and CB2 agonist JWH-133, an increased number of H3K9me3-labelled cells were observed that were CB1 and CB2 dependent, because both selective CB antagonists counteracted these effects .
In another report, an association between the increase of total histone 3 acetylation (generally corresponding to transcriptionally active chromatin) and CB1 downregulation in mutant huntingtin-inducible cells (HD43) was observed, a cellular model of Huntington's disease .
Moreover, gene-specific histone modifications have also been associated with eCB signalling. For instance, AEA protects neurons from inflammatory damage by inducing histone H3 phosphorylation of mitogen-activated protein kinase phosphatase-1 (MKP-1) gene in activated microglial cells, thus regulating MKP-1 expression and subsequent ERK-1/2 dephosphorylation . It has also been observed that THC and cannabinol can (a) activate the ERK-MAPK cascade, a crucial event for gene activation during mitosis induction, and (b) inhibit gap-junction intercellular communication, required to relieve growth suppression of cells . These events are both key epigenetic pathways and markers of tumour promotion in a rodent liver epithelial cell model . THC can also affect, in a dose-dependent manner, the expression of transcription factors, as well as transcriptional co-repressors such as histone deacetylase 3 (HDAC3) in the BeWo cell line, a model of trophoblast intercellular fusion and differentiation . The increase of HDAC3 expression following THC in these cells has led to the suggestion of a link between inhibition of cytotrophoblast cell-cycle progression and subsequent placental development .
Another recent study stated the possible role of eCBs in chromatin remodelling, in particular during spermiogenesis . Experiments have been carried out on male mice homozygous for a CB1-null mutation (CB1−/−). Using these animals, it has been observed that CB1 activation regulates chromatin remodelling of spermatids by either increasing transition protein 2 (Tnp2) levels or enhancing histone displacement, thus confirming that CB1 activity regulates Tnp2 expression as well as histone displacement/retention. Instead, in CB1 heterozygous mice (CB1+/−), Tnp2 levels were identical to those of CB1−/− whereas histone displacement/retention was not affected. Moreover, CB1 genetic loss induced a decrease in sperm chromatin quality, also linked to sperm DNA fragmentation. However, chromatin quality and DNA integrity of CB1+/− sperm were quite similar to those of wild-type animals, leading to the suggestion that histone displacement in CB1+/− could still protect DNA even if partially disrupted .
In another cellular model, human keratinocytes (HaCaT cells), the molecular mechanisms underlying the influence of eCBs, and in particular of AEA, on cell differentiation were evaluated by analysing the epigenetic regulation of key genes like keratins and transglutaminases. AEA affected the expression of keratin 1, keratin 10, involucrin and transglutaminase 5, and these changes appeared to be due to increased methylation of genomic DNA. In fact, treatment of HaCaT cells with 5-azacytidine, an inhibitor of DNA methylation, abolished the effect of AEA on keratin expression, and AEA itself was also able to inhibit gene transcription by inducing both specific and global DNA methylation . Moreover, it has been observed that AEA induced DNMT activity in differentiated keratinocytes in a CB1-dependent manner, and this effect was not the result of a direct interaction between AEA and DNMT but rather involved p38 MAPK signalling. Altogether, these results show a new activity of eCBs as transcriptional repressors via epigenetic mechanisms and, in the search of drugs able to reverse methylation abnormalities, they suggest a possible exploitation of eCB signalling in human diseases where DNA methylation is downregulated .
Finally eCBs can also affect DNA methylation in tissues known to be critical for replication and pathogenesis of human (HIV) and simian (SIV) immunodeficiency viruses . Indeed, among the 14 000 genes analysed, several of those relevant for viral entry, integration and production, as well as for inflammation, were found to be hypermethylated in the THC-treated SIV-infected macaques, compared with vehicle-treated SIV-infected macaques. Among these genes, more relevant changes occurred at CXCR4, nucleoporin 153 kDa (NUP153), RAN-binding protein 2 (RANBP2), human immunodeficiency virus type-I enhancer binding protein (HIVEP1), HIV-1 Rev binding protein (HRB), human immunodeficiency virus type-I enhancer binding protein 3 (HIVEP3), HIV TAT specific factor 1 (HTATSF1), mediator subunit complex 12 (MED12), CCAAT-enhancer binding protein delta (C/EBPD) and interleukin-1 receptor accessory protein (IL1RAP). All of them were found to be hypermethylated, with ~ 3- to ~ 24-fold increases over controls. The subsequent suppression of gene expression suggests a new role for THC that, through epigenetic mechanisms, might regulate key proteins required for viral kinetics in the brain . Table 2 summarizes current knowledge of the epigenetic effects of eCBs. Figure 3 shows both the epigenetic regulation of CB1 receptor and the epigenetic action of eCB signalling.
Table 2. Overview of studies examining epigenetic mechanisms triggered by ECS modulation
The key role of epigenetic mechanisms in different biological processes  and human diseases [113-116] has been clearly demonstrated over the last few years. Among the many genes and signalling pathways regulated by chromatin modification and/or DNA methylation , ECS elements are getting more and more attention. The ECS controls lipid signalling pathways by acting both in the central nervous system and in peripheral tissues, and thus a deeper understanding of its regulation might lead to the development of new clinical strategies for different human pathologies . Indeed, epigenetic alterations are reversible and specific interventions that target different epigenetic pathways might have a major impact on health problems, eventually providing a new avenue for innovative therapeutic approaches.
On a final note a few points remain to be addressed. Since CB1 in the central nervous system is preferentially localized presynaptically, it is of relevance to study how the signal will be translated to the nucleus. Moreover, it is important to understand how CB1 evokes epigenetic mechanisms, either by directly interacting with the epigenetic machinery or indirectly, for instance by regulating neurotransmitter release. Finally, the involvement of non CB1/CB2 receptors should be addressed.
A better understanding of the epigenetic regulation of eCB signalling as well as the eCB regulation of epigenetic mechanisms will be of great value for the possible design of more specific epigenetic drugs, also because those currently FDA approved, although promising, have both a non-specific target (non-chromatin proteins are also affected) and a genome-wide effect.
This investigation was partly supported by Fondazione TERCAS (grant 2009-2012 to MM). ADF was supported by a Fondazione TERCAS-Progetto Speciale Assegni di Ricerca 2011–2013 fellowship.