DNA methylation has a crucial role in various biological processes, such as development, differentiation, and gene expression ^[1][2][3][40,41,42]. Thus, comprehensive mapping becomes critical for addressing the functional role of this modification ^[4][43]. Various strategies have been developed to differentiate methylated and non-methylated C residues ^[5][44]. The initial lack of genome-wide approaches has restricted DNA methylation profiling to gene-specific evaluation using polymerase chain reaction (PCR) amplification of the target sequence. As DNA polymerases do not discriminate between C and 5mC, all potential differences in methylation are lost during classical PCR amplification ^[5][6][44,45]. To overcome this challenge, Frommer et al. developed a locus-specific method based on DNA treatment with sodium-bisulfite (SB) that leads to the conversion of all unmodified cytosine to uracil ^[6][45]. 5mC are resistant to deamination induced by SB, and are preserved during the PCR amplification by primers designed on the converted DNA. The resulting product is then analyzed by Sanger sequencing. Although very laborious, bisulfite sequencing PCR (BSP) is considered the gold standard to quantitatively study gene-locus specific DNA methylation ^[7][46].

The possibility to interrogate the entire genome using next generation sequencing technologies has expanded DNA methylation analyses ^[8][47]. The application of massive parallel sequencing to bisulfite-treated DNA has resulted in new genome-scale and -wide protocols, such as reduced representation bisulfite sequencing (RRBS) and whole genome bisulfite sequencing (WGBS), respectively. RRBS uses MspI, a methylation insensitive enzyme, to produce small fragments with CpG dinucleotides at both ends. Digested products are bisulfite converted and then sequenced ^[9][48]. Even though RRBS is more capable of covering a higher number of CpG loci within a given region than array-based techniques ^[8][10][47,49], the coverage across corresponding CpG-rich sequences might change among the samples tested, therefore introducing higher inter-sample variability and altering reproducibility ^[11][50]. As RRBS is biased for CpG-rich regions, such as CGIs, the coverage drops for CG shores, shelves, and open sea regions ^[11][50].

WGBS couples bisulfite-conversion of genomic DNA with high-throughput sequencing ^[12][51] and therefore provides a comprehensive and quantitative analysis of DNA methylome at single nucleotide resolution, without relying on restriction enzyme enrichment. Until few years ago, the high amount of DNA required as input, as well as the cost, limited its use. The recent advancement in sequencing platforms and sample preparation have now made WGBS more accessible and feasible, especially cost-wise, to allow projects with large sample sizes ^[13][52].

Likewise, non-bisulfite-based methods have also benefited from the introduction of high-throughput sequencing platforms. For example, immunoprecipitation of methylated DNA with an antibody recognizing 5mC residues coupled with sequencing (methylated DNA immunoprecipitation sequencing [MeDIP-Seq]) ^[8][47], is commonly used as an alternative approach to RRBS (Box 2).

MeDIP-Seq covers a higher number of regions than those normally screened by RRBS, but its efficiency is based on the specificity of the antibody, which can be biased toward hypermethylated sequences. Yet, neither RRBS nor MeDIP-Seq can determine the profile of virtually all CpG dinucleotides throughout the genome. Therefore, WGBS is considered the gold standard for full methylome analysis. More recently, the establishment of enzymatic methyl-Seq (EM-Seq), has added another free-bisulfite-based method to profile the entire methylome in instances when BS treatment is not suitable (i.e., fragmented DNA) or the DNA input is too low. EM-Seq relies on two sequential enzymatic reactions protecting 5mC and 5hmC from downstream deamination. The enzymatically converted DNA can then be processed for sequencing similar to WGBS ^[14][53].

Despite their sensitivity, all bulk sequencing methods are unable to dissect intra-cellular and intra-tumoral epigenetic heterogeneity within a specific cell population ^[15][54]. Single-cell technology has emerged as an invaluable tool to ascertain this heterogeneity ^[16][55]. WGBS and RRBS protocols have been optimized to carry out single-cell bisulfite sequencing (scBS) and single-cell RRBS (scRRBS), respectively ^[17][56]. While scBS is able to cover a higher number of CpG sites than scRRBS, the latter enables a better coverage of CGIs ^[18][57].

DNA methylation plays a critical role in the regulation of early development in humans and other mammals. In contrast to DNA sequences, DNA methylation is not inherited from gametes, as the parental DNA methylation pattern is erased at an early embryo stage ^[19][63]. During implantation, the DNA methylation profile is re-established, and the entire genome undergoes de novo methylation with the exclusion of CpG island-like regions, which elude this epigenetic modification due to the presence of RNA polymerase complexes that prevent access of de novo methylation machinery to the DNA. The resulting bimodal pattern is conserved throughout development and preserved for the whole lifespan of the organism unless unexpected alterations ^[20][21][64,65]. Yet, a group of genes, so-called “imprinted” genes, can escape this extensive reprogramming process occurring during embryonic development. Genomic imprinting is defined as the monoallelic, parental-specific expression of a gene in diploid cells, and as such, it is considered a form of gene regulation. Imprinted genes are epigenetically marked, i.e., “imprinted”, within differentially methylated regions (DMRs) in gametes, and such imprints are conserved after fertilization ^[22][66]. As a result, only one parental copy of the imprinted gene will be expressed, whereas the other copy will be silenced by DNA methylation ^[23][24][25][67,68,69]. The human insulin-like growth factor II (IGF-II)/H19 region is an example of paternally imprinted gene, in which methylation of the imprinting control sequence (ICR) regulates binding of the CTCF ^[26][27][70,71]. In mammals, the majority of imprinted genes affects growth of the embryo, placenta, and neonate. In that regard, paternally imprinted genes function as growth enhancers, whereas maternally imprinted genes function as growth repressors ^[28][72]. The loss of maternal DNA methyltransferases results in post-implantation lethality ^[28][29][16,72], suggesting the essential role of genomic imprinting in embryonic development and differentiation. Another category of imprinted genes include those involved in neurologic and behavioral regulation ^[30][73].

DNA methylation also partakes in protecting the structural integrity of the genome. It has been proposed that methylation of CpGs within parasitic DNA elements and retrotransposons, which account for 40% of the entire genome, operates as a genome defense system, in order to prevent the expression of these elements and preserve genomic stability ^[31][74].

Another puzzling and intriguing matter is whether the dynamic changes observed in response to biological and non-biological stimuli are paralleled by DNA methylation changes during the different phases of the cell cycle in physiological conditions. Substantial data with respect to this question are still lacking, and the few reported studies do not seem to agree with one another. Single-cell RNA sequencing results from murine embryonic stem cells (mESCs) point to a higher methylation rate in G1/S compared to other phases of the cycle, in line with previous observations in HeLa cells and human primary foreskin fibroblasts ^[32][33][75,76]. Similarly, Desjorbet et al. have demonstrated that the inheritance of DNA methylation marks efficiently occurs in late S phase, and cytosine methylation is completed in G2/M phase ^[34][77]. However, a different study analyzing DNA methylation levels in G0, G1, and G2 phases in low passage primary dermal fibroblasts did not confirm these cell cycle changes in DNA methylation patterns ^[35][78]. The specific cell type interrogated, along with not including the S phase in the analysis, could possibly account for these contrasting results, since hemi-methylated DNA sequences are challenging to evaluate with low coverage sequencing. Further analyses will need to delve into this question.