In vertebrates, CpGs dinucleotides are underrepresented (one fifth of the expected frequency) and in 60–90 % of cases, these dinucleotides are methylated on the carbon 5 of the cytosine residue (m5C) [10, 11]. Cytosine is the main chemically modified base in higher eukaryotes [12].

Cytosine methylation protects DNA from enzymatic digestion by restriction enzymes and initial global digestion pattern analysis combined with the recent mapping of methylated CpGs across the murine and human whole genomes revealed that CpG poor regions, including exons, introns, intergenic regions, and repeated sequences, are highly methylated (from 80 to 90 % are CpG methylated). However, a very few CpGs-rich regions appeared sensitive to the action of these enzymes suggesting that some regions of the genome contain unmethylated CpGs [13]. These small regions of 0.5–2 kb, called CpG islands represent approximately 1 % of the genome and are characterized by a G + C content above 50 % and a CpG/GpC ratio close to 1 [14, 15]. CpG islands are frequently found upstream of constitutively expressed genes [16] or in 40 % of tissue-specific genes [14].

In a given genome, the methylation patterns are precisely established with a lack of methylation of CpGs islands and a more general methylation of other dispersed CpG sequences. A few exceptions of hypermethylated CpG islands have been described in the case of parental imprinting and dosage compensation of one of the two X chromosome [17, 18]. Unlike the DNA sequences, methylation profile is not inherited from the gametes. Almost all methylation is erased in the early embryo and a new pattern is established at the time of implantation. Furthermore, this uneven bimodal distribution of methylated CGs depends on the type of tissue, with brain or oocytes harboring a very low level of methylation [19–22]. Methylation of promoter regions is usually associated with gene silencing but beside regulation of transcription, splicing is also influenced by DNA methylation [23] possibly through the binding of proteins with a high affinity for methylated DNA, which would regulate the binding of splicing factors [24]. The intragenic DNA methylation promotes transcriptional elongation by inhibiting transcription from alternative promoters contained within genes [25, 26] and methylation regulates alternative promoter usage, short and long noncoding RNAs production, alternative RNA processing, as well as enhancer activity. Furthermore, a very high proportion of methylated CpG is found at transposable elements and contributes to transcriptional repression putting DNA methylation in the center of the multiple pathways involved in genome regulation.

Eukaryotic DNA is tightly packed around histone octamers to form nucleosomes. Nucleosomes must be packed to achieve the 10,000–20,000-fold compaction necessary to fit a genome into a nucleus of a few microns. Beside DNA methylation, compaction of the chromatin fiber also depends on changes at the level of nucleosome. A few examples of factors or modifications involved in nucleosome compaction but also in transcription, replication, and repair are developed below.

In interphase nuclei, it was first supposed that chromosomes would entangle after decondensation. However, at the end of the eighteenth century, Rabl and Boveri demonstrated that plant chromosomes conserve polarity in interphase nuclei and hypothesized that chromosomes occupy discrete territories within the nucleus. Moreover, initial studies of the β-globin loci in humans and mice have shown that active promoters, enhancers, and distant regulatory elements located on linear DNA regions, at a distance of tens of kilobases are co-localized in the interphase nucleus forming a central regulatory structure associated with RNA polymerase II called “chromatin hub” [80, 81]. A clear demonstration for the existence of chromosome territories (CTs) came approximately three decades ago with the work of Cremer and collaborators who observed that UV laser irradiation of discrete regions of Chinese hamster cell nuclei damages only a small subset of mitotic chromosomes [82]. Furthermore, distinct territories do not overlap and homologous chromosomes generally occupy non-adjacent territories [83].

Based on these observations and on publications highlighting the importance of nuclear positioning, we now know that the organization of the genome within the eukaryotic nucleus is not random and dynamic and that the spatial distribution of DNA sequences controls nuclear processes.

Gene-poor CTs are usually found at the nuclear periphery whereas gene-rich chromosomes localize internally [84, 85]. This spatial organization of CTs based on gene-density is conserved in primates [86], whereas the idea that chromosome size influences localization remains conflicting and partly unresolved [83, 85, 87]. Consequently, the plasticity of the chromatin fiber is limited by interactions and higher order-structures, which may be tailored by transcriptional activity. Thus, differentiation processes, requiring the sequential expression of specific genes, may participate in the shaping of chromosomal domains depending on cell-type specific transcriptional competence.

The nuclear periphery is generally associated with heterochromatin and silent chromatin modifications while the repositioning of genes away from the nuclear periphery is usually associated with relocalization of loci away from the nuclear envelope. For instance, upon differentiation of mouse ES cells into neural progenitors, the Mash1 locus is displaced away from the nuclear envelope concomitantly with Mash1 gene activation, while expression of some neighboring genes is not affected [88]. This displacement correlates with histone modifications principally at the Mash1 gene. Interestingly, ES cells mutated for histone and DNA methyltransferases (HMTs and DNMTs, respectively) do not exhibit changes in the association of the locus to the nuclear periphery and it was suggested that histone acetylation rather than methylation could be functionally linked to the control of chromatin association to the nuclear rim indicating a link between epigenetic changes and subnuclear distribution of functional domains.

A large proportion of the genome is transcribed into noncoding RNAs, which participate in the regulation of gene expression. Only the example of long noncoding RNAs is developed here. Long noncoding RNAs (or LncRNA) are heterogeneous group of noncoding transcripts with a size of 200 nucleotides or more [89]. These RNAs transcribed by RNA polymerase II are the most common noncoding RNAs in mammals and are predominantly found in the nucleus. LncRNAs are transcribed from intergenic regions, in antisense, overlapping, intronic and bidirectional orientations relative to protein-coding genes or from regulatory regions such as UTR, promoters or enhancers. Their expression is tightly regulated and, as observed for protein-coding genes, regions encoding LncRNAs are enriched in H3K4 trimethylation at promoters and H3K36 is associated with LncRNAs elongation [90]. The LncRNA have a 5-methyl-guanosine 5′ cap and are often spliced and polyadenylated but devoid of evident open reading frame. Their intrinsic nucleic acid nature confers to LncRNAs the capacity to bind proteins and to mediate base pairing interactions as described for other noncoding RNA such as microRNAs, small nucleolar RNAs, or small nuclear ribonucleoprotein particles. Expression of LncRNAs seems to be tightly regulated with a high tissue- and differentiation-dependent specificity.

Most of the nuclear LncRNA have been characterized for their role as epigenetic modulator at specific loci [89, 91–94] and thanks to their flexible and modular scaffold, they form higher order structures that facilitates interaction with proteins (reviewed in [95]). By recruiting DNMT3, the polycomb repressive complex (PRC2) [89, 96] or the H3K9 methyltransferase [97, 98], they contribute to the formation of heterochromatin and gene repression. The first and best-characterized examples of LncRNAs are those involved in X inactivation and parental imprinting (for review, [95, 99, 100]).

Most of the autosomal genes are expressed from both alleles. However, a certain number of genes are expressed from only one allele per genome, depending on their parental origin. This phenomenon named “parental imprinting” involves the deposition of different epigenetic marks involved in the designation of the parental origin and the maintenance of this parent-of-origin designation during development and differentiation. Approximately 100 imprinted genes have been identified in mammals so far (http://​har.​mrc.​ac.​uk/​research/​genomic_​imprinting; http://​igc.​otago.​ac.​nz/​home.​html).

DNA methylation plays a key role in genomic imprinting. Most of the imprinted genes are organized in loci spanning 1–3 Mb in length containing biallelically expressed genes interspersed with paternally expressed genes and maternally expressed genes and long noncoding RNAs. Imprinted loci are regulated by CpG-rich cis-acting regulatory elements named Imprinting Control Regions (ICR) of a few kilobases in length and regulated by differential DNA methylation (Differentially Methylated Region, DMR) depending on the parental origin. The ICR methylation pattern is established in the gametes by DNMT3A and DNMT3L [31, 37, 142] and maintained during early embryogenesis by DNMT1 [143, 144]. Marks inherited from the parent are erased in the primordial germ cells of the offspring and then restored according to the sex of the embryo [144]. Beside imprinting loci, a few isolated imprinted genes have also been described and are usually regulated by differential methylation and posttranslational modification of histones.

Most of the imprinted genes are implicated in prenatal development, in particular in the brain where imprinted genes are involved in modulating metabolic axes, behavior, learning, and maternal care.

A number of imprinting disorders have been described including transient neonatal diabetes [145], maternal and paternal uniparental disomy of chromosome 14 [146], growth disorders such as Beckwitt–Wiedeman and Silver–Russell syndromes [147–149] linked to the 11p15 locus or pseudoparathyroidism 1b linked to the 20q locus.

Prader–Willi syndrome (PWS) and Angelman syndrome (AS) are clinically distinct complex disorders linked to chromosome 15q11-q13. They are both characterized by neurological, developmental, and behavioral phenotypes. Both disorders can result from microdeletion, uniparental disomy, or defect in the 15q11-q13 imprinting center. The abnormality is carried by the paternal chromosome for PWS or the maternal allele for AS. PWS is characterized by neonatal hypotonia, childhood obesity, hypogonadism, cognitive impairment, and behavioral manifestation including obsessive-compulsive symptoms and hypothalamic insufficiency. This syndrome is linked to de novo paternal 15q11-q13 deletion in 70 % of cases, maternal disomy in 29 %, and methylation defect in 1 %. AS is characterized by mental retardation, speech impairment, ataxia, seizure, and microcephaly. In 70 % of cases, the disease is caused by deletion of the 15q11.2-q13 locus, paternal disomy in 7 % of cases, methylation defect (3 %), or epimutation in rare cases.

Interestingly, the 15q11.2-13.3 contains numerous genes with critical neurological functions such as CHRNA7 (cholinergic receptor nicotinic alpha 7), UBE3A and GABRB3 and deletions and duplications of the locus are observed in 1–3 % of patients with autism suggesting a key role for this locus in the neuronal development [150]. The main cause of AS is the loss of function in the brain of the UBE3A gene which codes for the E6-AP protein implicated in the transfer of small ubiquitin molecules to certain target proteins, to enable their degradation and might regulate synapse function through protein degradation pathways. The pathogenic mechanisms that underlie autism and AS due to the excess or lack of UBE3A are still unknown. These phenotypes could be caused by the misregulation of UBE3A targets influencing spine density, synapse, and neuronal function.

Beside protein-coding genes, several noncoding RNAs are also present in the PWS/AS region including the neuron specific SNRPN long noncoding RNA and a cluster of small nucleolar RNA transcripts (sno-RNAs) encompassing the UBE3A–AS region. In addition, the imprinting control region is bound by MeCP2, which is mutated in the Rett mental retardation-syndrome (OMIM 312750; [151, 152]). By binding to the PWS-ICR, MeCP2 has a role in loop formation acting as a neuronal chromatin structural organizer during neuronal maturation. Among the genes regulated by formation of these higher-order structures at the 15q11.2-13.3 locus, MeCP2 might positively regulate the expression of the CHRNA7 gene which is significantly reduced in Rett syndrome and frontal brain in autistic patients [140, 153, 154].

Overall, these results suggest overlapping epigenetic pathways between locus-specific abnormalities and autism.

Repeat expansion diseases are caused by an increase in the number of trinucleotide repeats and share some common features. These short repeats are polymorphic in the general population. Beyond a certain number of repeats, they become unstable and tend to increase at the next generation (dynamic mutation). The disease appears when the number of triplets exceeds a certain critical threshold. Trinucleotide repeats are either found in coding region resulting in the production of a protein with altered function (i.e., Huntington disease), or in noncoding regions resulting in an altered transcription and loss of protein synthesis (Fragile-X syndrome, Friedreich’s ataxia, myotonic dystrophy) [155], and pathophysiology of these diseases implicate epigenetic changes and heterochromatin formation [156].

Fragile X syndrome (FXS) (MIM # 300624) is the most common form of inherited mental retardation with a prevalence of 1/2,500 (Orphanet Database), characterized by a delayed language, autistic-like behavior and sometimes dysmorphic signs.

FXS is caused by expansion of a CGG trinucleotide in the 5′ UTR of the FMR1 gene located on Xq27.3. In the normal population, the number of repeats varies from 6 to 54. Between 55 and 200 CGG, the repeats become meiotically unstable (premutation) especially in the maternal germ line while the full mutation corresponds to more than 200 repeats and is accompanied by a hypermethylation of the repeats and neighboring sequences [157, 158]. FMR1 hypermethylation could act directly by preventing the binding of transcription factors. However, features of inactive chromatin have also been observed on mutated alleles like lysine hypoacetylation of histone H3 and H4 and H3K9 methylation [159, 160] suggesting that expansion of methyl-sensitive repeats contributes to the phenotype. These changes in the chromatin structure lead to the silencing of the FMR1 gene and absence of the FMRP protein, a regulator of translation in brain neurons [161, 162]. FMRP binds approximately 4 % of the mRNA produced in the brain and may be involved in the nuclear export and targeting of certain transcripts to the ribosome [163]. In the synapse, FMRP plays a role in regulating translation of proteins involved in certain pathways including the mGluR-LTD receptor (metabotropic glutamate receptor long term depression-dependent) [164]. Its loss of function would explain the cognitive deficits found in the fragile X syndrome.

Another interesting disease linked to triplet expansion is the Friedreich’s ataxia (FRDA) (MIM # 2293000), an autosomal recessive neurodegenerative disease that begins in childhood and is characterized by difficulties to coordinate movements, dysarthria, loss of reflexes, pes cavus, scoliosis, cardiomyopathy, and diabetes mellitus. The causative gene, FXN, encodes the Frataxin, a protein involved in the assembly and transport of iron-sulfur proteins of the mitochondrial respiratory chain. FRDA is caused by the expansion of a GAA trinucleotide in the first intron of FXN. Repeats range from 6 to 34 in the general population and are over 66 in patients. Expansion is associated with a decrease in FXN transcription level and different mechanisms could be implicated in gene silencing [165].

First, it has been shown that GAA repeats adopt an unusual conformation able to inhibit transcription in vitro [166]. Then, a role for heterochromatin formation has been proposed based on the observation of patient’s samples [167, 168]. Moreover, in FRDA patients specific CpG sites are hypermethylated in the FXN intron 1 compared to control and hypoacetylation of histones H3 and H4, hypermethylation of H3K9 are thought to modulate FXN promoter activity [167, 168].

Moreover, in favor of a role for chromatin remodeling and heterochromatinization in the pathophysiology of these two examples of triplet expansion diseases, reactivation of FMR1 and FXN transcription are induced by DNA hypomethylating (5-aza-2′-deoxycytidine) or histone hyperacetylating agents [169, 170].

Hereditary sensory neuropathy type 1 (HSAN1) is an autosomal dominant neurodegenerative pathology that clinically manifests as loss of sensation, dementia, and hearing loss (MIM # 614116). Affected individuals have early mortality and often require total care because of dementia, loss of ambulation from predominant sensory ataxia, and hearing loss [171, 172]. Genome-wide linkage analysis of several kindred with HSAN1 symptoms led to the identification of mutations in the gene encoding the DNA methyltransferase 1 (DNMT1) on 19p13.2 as the cause of the pathology. The DNMT1 protein contains a C-terminal catalytic domain and an N-terminal regulatory region involved in protein–protein interactions and required for the activation of the catalytic activity carried by the C-terminal region [173]. The identified mutation occurs in the targeting-sequence domain that regulates DNMT1 binding to chromatin during the late S phase and is responsible for the persistent association during the G2 and M phases. During replication DNMT1 is associated with the replication fork and remethylate the neosynthesized DNA strand. Heterochromatin is replicated at the end of the S-phase and persistent association of DNMT1 after the S phase could be required for the proper remethylation of this heavily methylated part of the DNA. The mutations in HSAN1 impair DNMT1 targeting to heterochromatin in G2 but the direct link between DNMT1 mutation and the HSAN1 phenotype requires further investigation together with the phenotypical consequences and tissue specificity.

Described in 1966, the Rett syndrome (OMIM # 312750) is a rare disease (prevalence 4/100,000), dominant and X-linked predominantly affecting women. It is characterized by a normal psychomotor development until the age of 6–18 months and the emergence of learning disabilities, autistic behavior, stereotyped hand movements and encephalopathy due to a progressive neurological dysfunction [174]. Demonstrated for the first time in 1999, the causal Rett syndrome mutations are located in the MECP2 gene on Xq28 [175]. MECP2 is ubiquitously expressed, but its highest expression level was found in the brain and fluctuates with neuronal maturation [176, 177]. Most of the pathogenic mutations affect the Methyl Binding (MBD) and Transcription Repression Domains (TRD), and numerous reports have clearly demonstrated that loss or altered expression of the MeCP2 methyl-binding protein leads to the neurological defects observed in the Rett syndrome. Nevertheless, the mechanisms underlying the Rett phenotype are not completely understood. MeCP2 protein might be involved in the epigenetic regulation of a number of genes during neuronal development and maturation such as BDNF but also imprinted genes such as DLX5 and 6 (Distal-Less Homolog) or UBE3A [178–180]. MeCP2 is involved in the global organization of chromatin and long distance looping, in the repression of endogenous retrovirus transcription [178, 181] but also as a modulator of alternative splicing with the RNA binding protein YB1 (Y-Box Binding Protein 1) [182]. In addition, MECP2 binds non-methylated DNA and may act as a transcriptional activator through its interaction with CRB1 at the promoter of target genes [182]. More recently, In vitro and in vivo approaches demonstrated that MeCP2 requires methylated CpGs flanked by a run of at least four A/T for efficient DNA binding conferring to MeCP2 its target specificity and explaining why only a few genes are dysregulated in Mecp2-deficient mice.

Cases with duplication of MECP2 are characterized by severe motor dysfunction, mental retardation, and premature death, indicating that MECP2 dosage is finely regulated in neuronal tissues.

The brachydactyly and mental retardation syndrome is linked to mutations in the gene encoding the HDAC4 type IIA histone deacetylase (OMIM # 600430). This pathology linked to deletion of the 2q37 locus, is a rare disorder characterized by a wide clinical spectrum involving developmental delay, skeletal abnormalities, such as brachydactyly type E, facial dysmorphia, and inconsistently, heart defect. This syndrome affecting 1/10,000 individuals (Orphanet) is also characterized by mental retardation and autism. The HDAC4 protein deleted in this pathology acts as a corepressor of transcription and is composed of a N-terminal domain that interacts with transcription factors and a C-terminal nuclear export signal allowing the phosphorylation-dependant transfer of the protein from the cytoplasm to the nucleus [183]. In particular, HDAC4 is a corepressor of transcription factors such MEF2C and RUNX2 and plays a key role in the regulation of genes involved in the osteogenic, chondrogenic, myogenic, and neurogenic differentiation pathways [184–187]. In the central nervous system, HDAC4 protects neurons from apoptosis [113] but might also regulate a number of key genes implicated in the neuronal differentiation pathway suggesting that mutation in the HDAC4 gene modifies expression of a number of genes involved in brain function and neuronal survival.

Kleefstra syndrome (KS; MIM 610253) characterized by severe intellectual disability, hypotonia, microcephaly, epilepsy, and development defects [188, 189], is caused by microdeletions in the distal long arm of chromosome 9q and haploinsufficiency or intragenic loss of function mutations in the gene encoding the EuHMTASE 1/GLP involved in the deposition of Lysine 9 residues on histone H3 to euchromatin and gene repression of individual genes [190–192]. This was established by identification of three patients with features of the syndrome and carrying either mutations or balanced translocation in EHMT1 in which deletion size does not correlate with disease severity since patients with mutations in EHMT1 are as severely affected as those with 9qter submicroscopic deletions encompassing other genes. The role of EHMT1 in the brain and in the etiology of KS is not fully understood but recent data obtained in Drosophila Melanogaster suggest a role for this protein in learning and memory [193]. Ehmt1 ^+/− mice also exhibit learning and memory deficits associated with a significant reduction in dendritic arborization and number of mature spines in hippocampal CA1 pyramidal neurons [194]. The genes targeted by this histone methyltransferase are not known but EHMT1 haploinsufficiency might be involved in the learning deficits and synaptic dysfunction observed in patients with Kleestra syndrome.

Type 2 Weaver syndrome (OMIM # 614421) is a rare condition characterized by an abnormal over-growth, tall stature, advanced bone age, a marked macrocephaly, hypertelorism, facial dysmorphism, mental retardation, and learning disabilities [195]. This pathology, mostly sporadic, is caused in some patients by heterozygous mutations in the EZH2 gene on chromosome 7q36.1 [196]. EZH2, Enhancer of Zeste 2 encodes a member of the Polycomb Repressive Complex 2 (PRC2), which catalyzes the trimethylation of lysine 27 of histone H3 (H3K27). As described above, the EZH2 enzyme together with the other members of the PRC2 complex, SUZ12 and EED are involved in the control of cell differentiation, cell lineage determination, stem cell maintenance, X-chromosome inactivation, or genomic imprinting. EZH2 mutations found in the Weaver syndrome cause a loss of the catalytic activity. Furthermore, somatic EZH2 mutations are also encountered in different types of cancers including hematological malignancies [197, 198] suggesting that the congenital mutations found in the Weaver syndrome might also contribute to the increased incidence of cancers, in particular acute lymphoblastic leukemia, observed in this syndrome.

Moreover, in some patients affected with the Weaver syndrome, mutations in the NSD1 gene (nuclear receptor SET domain containing protein–1 gene; OMIM # 117550), which encodes a H3K36 and H4K20 histone methyltransferase have also been observed [199–201]. The NSD1 protein contains a Su(Var)3-9, enhancer of zeste, trithorax (SET) domain responsible for its histone methyltransferase activity, a multiple plant homeodomain (PHD) that associates with chromatin and a proline-tryptophan-tryptophan-proline (PWWD) involved in protein–protein interactions. NSD1 methylates H3K36 and H4K20 in vitro but also non-histone proteins conferring to NSD1 a versatile role in both activation and repression of transcription [202], but its specific function in the brain remains partly understood.

Autism spectrum disorders (ASDs) are a range of complex neurodevelopmental disorders with significant overlap with a number of diseases. ASDs are characterized by repetitive and stereotyped behavior, limited social development and impaired language skills and higher cognitive functions. Cortical organization and circuitry requires coordination of a number of developmental processes such as specification of neuronal identity, neuronal migration, and wiring of neuronal circuits regulated by a number of transcription factors and epigenetic changes occurring at different developmental steps. Because ASD is an early onset neurodevelopmental disorder, alterations during neocortex development are likely involved in the pathology. Furthermore, evidence from recent genetic studies suggests that synaptic dysfunction underlies ASD. In most cases, the etiology is poorly understood and likely involves polygenic factors but also environmental influence and epigenetic dysregulations as exemplified by discordance in monozygotic twins [203]. The involvement of epigenetics is reviewed and discussed here.

As discussed earlier, the MeCP2 protein plays a key role in CNS development. A correlation between reduced MeCP2 expression and autism has been described. Decreased MeCP2 levels have been found in the postmortem frontal cortex samples from autistic patients and in the fusiform gyrus involved in face processing. In the different samples analyzed, decreased MeCP2 expression is associated with hypermethylation of its promoter region suggesting that aberrant DNA methylation causes MeCP2 silencing [140]. However, considering the multiple roles of this protein, its targets and its role either as a silencer or activator of transcription is not known. Nevertheless, it is worth-noting that MeCP2 is linked to the Rett syndrome, classified as ASD.

Beside MeCP2, Methylation changes in ASD have been observed in a certain number of other genes. Their respective roles in disease onset or penetrance are not fully understood and the mechanisms leading to this localized hypermethylation are unknown. In ASD postmortem brain tissues, hypermethylation correlated with decreased expression has been described in the promoter region of the Oxytocin receptor involved in social memory and ability to recognize other individuals [204, 205], the Bcl2 gene [206], the RORA gene (retinoic acid-related orphan receptor alpha) [207], β Catenin [208, 209], or Protein kinase C beta gene (PRKCB1) [210]. Moreover, in a recent study based on 19 autism cases and 21 unrelated controls, methylation level of 485,000 CpG island has been determined using the Infinum HumanMethylation450 BeadChip. Four differentially methylated regions (DMR) have been identified [211]. One of these DMR is located in the 3′ UTR of the PRRT1 gene and overlaps with DNAseI hypersensitive sites. Little is known on the role and regulation of PRRT1 in human but it might be involved in hippocampus function and mutations of other genes of the PRRT family have been observed in a number of neurological disorders. The second DMR was observed in cerebellar tissue, in the promoter of the SDHAP3 gene, associated with a noncoding RNA and a small coding RNA of unknown function. The third DMR identified is located in the promoter region of TSPAN32 and C11orf21. TSPAN32 is involved in cellular immunity and acts as a structural and cell signaling scaffold protein resembling other tetraspanins identified as implicated in schizophrenia and bipolar disorders. Interestingly, this DMR is located within an imprinted region of the genome. The fourth DMR is located upstream of the ZFP57 and overlaps with the 5′ end of an alternatively spliced EST. Interestingly, ZFP57 is instrumental in targeting DNA methyltransferase to imprinted regions during development. As described above, a number of other imprinted loci have been associated with autism suggesting a key role for parent-of-origin methylation in ASD.

Another regulatory pathway possibly affected by epigenetic changes in ASD is the Neurexin-Neuroglin pathway. A large number of mutations have been described in genes such as neuroligins, neurexins, and SHANK that play a role in the formation and the maintenance of synapses. SHANKs are scaffolding proteins of the postsynaptic density of glutamatergic synapses. Mutations or CNVs in the SHANK3 gene have been associated with the Phelan–McDermid syndrome microdeletion syndrome (PMS) in which autistic features are prominent, but also in ASD [212, 213] and schizophrenia associated to intellectual disability (ID) and poor language [214]. The 5′ promoter region of the SHANK3 gene contains five CpG islands and one specific CpG island (number 2) modulates SHANK3 expression in a tissue-dependent manner [215]. The isoform-specific expression of SHANK3 is altered in ASD brain tissues with SHANK3 hypermethylation and treatment with the 5-aza-cytidine DNMT inhibitor modifies the level of SHANK3 methylation together with the isoform-specific expression in cultured cells [216]. Interestingly, the SHANK3 gene is well conserved from rodents to human and the neonatal expression of certain transcripts decrease the methylation of CpG island 2 [217] suggesting a methylation-dependent SHANK3 regulation during development. The cause of increased methylation in specific SHANK3 CpG islands in ASD brain tissues is not known but these observations support an association between hypermethylation, expression changes, and susceptibility to ASD.