The impact of DNA methylation in Alphaproteobacteria

Alphaproteobacteria include bacteria with very different modes of life, from free‐living to host‐associated and pathogenic bacteria. Their genomes vary in size and organization from single circular chromosomes to multipartite genomes and are often methylated by one or more adenine or cytosine methyltransferases (MTases). These include MTases that are part of restriction/modification systems and so‐called orphan MTases. The development of novel technologies accelerated the analysis of methylomes and revealed the existence of epigenetic patterns in several Alphaproteobacteria. This review describes the known functions of DNA methylation in Alphaproteobacteria and also discusses its potential drawbacks through the accidental deamination of methylated cytosines. Particular emphasis is given to the strong connection between the cell cycle‐regulated orphan MTase CcrM and the complex network that controls gene expression and cell cycle progression in Alphaproteobacteria.


Introduction
Methylated bases can be found on the genomes of organisms from all kingdoms of life. The role of these epigenetic marks has been mostly studied in eukaryotes, where they are involved in a variety of processes such as cellular differentiation, embryogenesis, genomic imprinting and cancer development (Egger et al., 2004;Schubeler, 2015;Smith and Meissner, 2013). Methylated cytosines and adenines are however also frequently found in bacteria and can represent more than 2% of the bases in a bacterial genome (Marinus and Lobner-Olesen, 2014;Sanchez-Romero et al., 2015). Methylated bases (m5C, m4C and m6A) are often used by bacteria to discriminate their own DNA from exogenous DNA entering the cell during horizontal gene transfers (HGT). In this case, motifs regularly spread on self-DNA are methylated by a DNA methyltransferase (MTase) to protect them from cleavage by a restriction endonuclease (RE) partner. The MTase and the RE then belong to a so-called restriction-modification (R/M) system (Vasu and Nagaraja, 2013;Loenen et al., 2014). RE cleave non-methylated motifs detected on exogenous DNA, thereby acting as 'immigration controllers' limiting the efficiency of HGT.
In addition to these R/M-associated MTases, bacteria often encode orphan MTases that are no longer associated with an RE partner. The most recent estimate comparing 230 bacterial and archaeal genomes is that ~half of these bacteria encode minimum one active orphan MTase (Blow et al., 2016). Although these have lost their capacity to control HGT, there is more and more evidence that several of these are very important, if not essential, for cell cycle progression, virulence or genome maintenance (Collier, 2009;Marinus and Casadesus, 2009;Sanchez-Romero et al., 2015;Adhikari and Curtis, 2016). The two best known examples are the conserved Dam MTase found in many Gammaproteobacteria and the CcrM MTase found in most Alphaproteobacteria (Adhikari and Curtis, 2016). These two MTases methylate adenines found in specific motifs, creating m6A epigenetic marks that can affect the activity of specific DNA binding proteins such as transcription factors, endonucleases or initiators of DNA replication. Dam-dependent methylation is, for example, essential for the replication of one of the chromosomes of Vibrio cholerae (Marinus and Lobner-Olesen, 2014) and required for the phase-variable expression of several virulence factors in clonal populations of Salmonella enterica (Garcia-Pastor et al., 2018). Interestingly, many bacterial pathogens also express other phase-variable orphan MTases that control randomly switching regulons, called phasevarions, encoding proteins involved in pathogenesis, host-adaptation and antibiotics resistance (Atack et al., 2018).
This review focuses on the functions of DNA methylation in Alphaproteobacteria. This class of bacteria displays very diverse life cycles as it includes free-living, host-associated and intracellular bacteria that sometimes act as plant, animal or human pathogens.

Methylated DNA motifs and DNA methyltransferases in Alphaproteobacteria
Until recently, the detection of methylated bases involved the use of complex methods such as the use of methylation-sensitive restriction enzymes and Southern blotting to detect the methylation state of specific DNA motifs on bacterial genomes. Although time-consuming, these methods proved efficient to discover the function of the best characterized orphan and cell cycle-regulated MTase CcrM. Initially discovered in Caulobacter crescentus, CcrM homologs are however found in nearly every Alphaproteobacteria, except Rickettsiales and Magnetococcales (Wright et al., 1997;Brilli et al., 2010;. C. crescentus is a free-living bacterium commonly used to study the regulation of the bacterial cell cycle because its cell cycle can be synchronized (Collier, 2016). It replicates asymmetrically into two distinct cell types: a stalked cell that immediately replicates its chromosome (S-phase) while it elongates into a pre-divisional cell, and a swarmer cell that cannot replicate its chromosome (G1-phase) before it differentiates into a stalked cell (Fig. 1A). CcrM methylates adenines in 5′-GANTC-3′ motifs in late pre-divisional cells, when the replication of the unique circular chromosome of C. crescentus has just ended (Zweiger et al., 1994). Considering that DNA replication is semi-conservative, GANTC motifs stay in a hemi-methylated state (only the old DNA strand is methylated) for some time after their replication and until cells reach the pre-divisional stage. The duration of this period depends on the location of the motif on the chromosome. Indeed, motifs located close to the origin of replication stay hemi-methylated during most of the S-phase of the cell cycle, while motifs located near the chromosomal terminus are methylated immediately after their replication (Fig. 1A). This model was recently confirmed at the genome scale thanks to the use of Single Molecule Real Time (SMRT) sequencing, which allows the detection of methylated motifs on bacterial genomes (Flusberg et al., 2010;Davis et al., 2013). This technique revealed that most GANTC motifs are efficiently methylated by CcrM on the C. crescentus genome, highlighting the efficiency of this enzyme (Kozdon et al., 2013;. It is unclear whether CcrM is a processive enzyme that remains attached to DNA and methylates multiple GANTC motifs before dissociating from the DNA (Berdis et al., 1998;Shier et al., 2001;Woodcock et al., 2017) or a distributive enzyme that dissociates after each methylation reaction (Albu et al., 2012;Maier et al., 2015).
Analysis of the C. crescentus methylome also revealed the existence of four other functional DNA MTases: two are adenine MTases predicted to be part of R/M systems, while the two others are supposedly orphan cytosine MTases (Kozdon et al., 2013). Interestingly, motifs methylated by these two orphan MTases appear as only partially methylated, suggesting the possible existence of methylation patterns or phasevarions controlled by these MTases, although their function remains elusive. So far, the methylomes of ~20 Alphaproteobacteria have been characterized using SMRT sequencing (Kozdon et al., 2013;Blow et al., 2016) and bioinformatics searches indicate that genes encoding putative DNA MTases are frequently found in alphaproteobacterial genomes (Roberts et al., 2015;Oliveira et al., 2016). The function of some of these MTases has now been characterized in more details. These include six DNA MTases that are part of R/M systems in the marine bacterium Agrobacterium gelatinovorum (Suzuki et al., 1996b), in the methylotrophic bacterium Paracoccus aminophilus (Dziewit et al., 2011), in the photosynthetic bacterium Rhodobacter sphaeroides (Scavetta et al., 2000;, in the plant symbiont Rhizobium leguminosarum (Rochepeau et al., 1997), and in the bioethanol and acetic acid producing Zymomonas mobilis (Kerr et al., 2011) and Acetobacter pasteurianus (Suzuki et al., 1996a) bacteria.

Regulation of CcrM during the cell cycles of Alphaproteobacteria
As mentioned earlier, the activity of CcrM is cell cycle-regulated in C. crescentus, so that it happens only once DNA replication is finished in late pre-divisional cells. Interestingly, if CcrM is over-produced throughout the whole cell cycle, cells from C. crescentus and from the plant symbiont Sinorhizobium meliloti display cell division defects and over-initiate DNA replication Wright et al., 1997). CcrM over-production also leads to nodulation defects with Mesorhibozium loti (Ichida et al., 2009) and to intracellular replication defects in the animal and human pathogen Brucella abortus (Robertson et al., 2000). All together, these observations demonstrate that the temporal control of CcrM activity is critical in many Alphaproteobacteria. The C. crescentus CcrM protein is only detectable in pre-divisional cells due to tight transcriptional and post-transcriptional control mechanisms (Fig. 1). A common feature found in C. crescentus (Reisenauer et al., 1999), B. abortus (Bellefontaine et al., 2002;Francis et al., 2017), S. meliloti (Pini et al., 2015) and Mesorhizobium huakuii (Peng et al., 2014) is that the conserved CtrA regulator is responsible for the timed activation of ccrM transcription (Fig. 1B). This response regulator is the most abundant and active in swarmer and pre-divisional C. crescentus cells (Fig. 1A), where it controls the expression of hundreds of genes (Laub et al., 2002;Collier, 2016). In addition, the SciP regulator prevents the activation of ccrM by CtrA in swarmer cells ( Fig. 1) (Gora et al., 2010;Tan et al., 2010;Collier, 2016). Noteworthy, there is also evidence suggesting that CcrM represses its own expression through the methylation of the ccrM promoter region (Fig. 1B) Stephens et al., 1995). Once synthesized, CcrM is a particularly unstable protein degraded by the Lon protease ( Fig. 1)  , enabling variations of CcrM levels as ccrM transcription fluctuates during the C. crescentus cell cycle.

The variable essentiality of DNA MTases in Alphaproteobacteria
CcrM was the first DNA MTase identified as being potentially essential for the survival of at least certain Alphaproteobacteria, including C. crescentus (C. Stephens et al., 1996), B. abortus (Robertson et al., 2000), S. meliloti (Wright et al., 1997), R. sphaeroides (Burger et al., 2017), the plant pathogen Agrobacterium tumefaciens (Curtis and Brun, 2014) and the phototrophic environmental bacterium Rhodopseudomonas palustris (Pechter et al., 2015) cultivated in standard laboratory conditions. This finding was however challenged more recently concerning C. crescentus. Indeed, a ccrM deletion mutant could be constructed when C. crescentus was cultivated in slow growing conditions instead of fast growing conditions (Gonzalez and Collier, 2013). The resulting ∆ccrM cells were significantly elongated, but their viability was comparable to wild-type cells. Since then, CcrM was also found to be dispensable in a few other Alphaproteobacteria, including the environmental bacterium Brevundimonas subvibrioides (Curtis and Brun, 2014) and, most likely, the pollutant-degrading bacterium Sphingomonas wittichii (Roggo et al., 2013).
Besides ccrM, the gene encoding another putative orphan DNA methyltransferase from R. palustris has been recently shown to be un-disruptable during a transposon mutagenesis followed by deep sequencing (Tn-Seq) experiment. Interestingly, this gene shows similarities with the Dcm cytosine MTase that is mostly found in Gammaproteobacteria where it is usually dispensible (Pechter et al., 2015). This finding suggests that cytosine methylation can sometimes play essential roles in Alphaproteobacteria. This is however not always the case, since cytosine MTases found in C. crescentus and B. subvibrioides were shown to be dispensable using similar Tn-Seq experiments (Christen et al., 2011;Curtis and Brun, 2014).

Impact of CcrM-mediated methylation on the C. crescentus transcriptome
Methylated bases can serve as epigenetic signals influencing gene expression in prokaryotes and eukaryotes. Then, an attractive hypothesis is that many orphan DNA MTases found in bacteria, which have lost their use as 'immigration controllers' may instead be used to modulate gene expression. C. crescentus CcrM is one of these examples. Indeed, an early study demonstrated that the activity of the ctrAP1 promoter controlling the expression of the CtrA master regulator, is influenced by its methylation state (Fig. 2) (Reisenauer and Shapiro, 2002). Considering that GANTC motifs are under-represented in the genomes of bacteria that have a ccrM homolog but over-represented in intergenic regions in these genomes, CcrM-mediated DNA methylation could potentially have a much broader impact on gene expression in Alphaproteobacteria . Proof of this came from a study comparing the transcriptomes of wild-type, ∆ccrM and ccrM-overexpressing C. crescentus cells . Strikingly, the expression of ~10% of the C. crescentus genes appeared as modulated in response to the absence of ccrM, with a significant enrichment for genes that have minimum one GANTC motif in their promoter region and for genes whose expression is cell cycle-regulated or essential. Examples of genes fitting each of these criteria notably included genes encoding proteins involved in DNA replication (ligA, gyrA or nrdA), chromosome decatenation (nstA) and cell division (ftsZ, mipZ or ftsN), uncovering a strong connection between CcrM-mediated methylation and cell cycle progression (Fig. 2). A complementary study demonstrated that the activity of the ftsZ promoter is very low when its GANTC motif remains un-methylated, providing insufficient levels of FtsZ for the division of C. crescentus cells cultivated in fast-growing conditions (Gonzalez and Collier, 2013). Supporting this model, an experimental evolution approach showed that ∆ccrM cells must accumulate mutations that promote ftsZ transcription to survive under fast-growing conditions (Gonzalez and Collier, 2015).
Whether CcrM-mediated DNA methylation is required for the expression of cell division genes in other Alphaproteobacteria remains untested. It however appears that GANTC motifs are not systematically found in ftsZ promoter regions from other Alphaproteobacteria, even in bacteria where ccrM is essential for viability such as Rhizobiales (Gonzalez and Collier, 2013). This observation indicates that CcrM-mediated DNA methylation affects the transcription of other genes or other processes required for cell cycle progression in certain Alphaproteobacteria.

The coordinated action of CcrM and GcrA regulates gene expression as a function of the cell cycle
Analysis of the CcrM regulon in C. crescentus revealed that it contained a strong over-representation of GcrAregulated genes . This finding suggested that the GcrA global cell cycle regulator could act as an epigenetic regulator, whose activity may be affected by the methylation state of GANTC motifs in its target promoters (Fig. 2). In addition, the phenotypes of ∆gcrA and ∆ccrM mutants displayed striking similarities, being incapable of dividing under fast-growing conditions Collier, 2013, 2015;Murray et al., 2013). GcrA is a cell cycle-regulated transcriptional regulator that controls the expression of hundreds of genes when  . To simplify, a blue lollipop indicates that the promoter region carries one or more GANTC motifs. Many of these promoters are also activated by GcrA (Holtzendorff et al., 2004;Fioravanti et al., 2013;Haakonsen et al., 2015) or by the (p)ppGpp alarmone (Boutte and Crosson, 2011;Sanselicio & Viollier, 2015). Dashed red arrows indicate that binding of GcrA to the promoter region was detected in vivo during chromatinimmunoprecipitation assays described in (Fioravanti et al., 2013) or (Haakonsen et al., 2015). Red arrows indicate that the binding of purified GcrA onto the promoter was confirmed in vitro (Fioravanti et al., 2013). PTS Ntr indicates the nitrogen-related phosphoenolpyruvate phosphotransferase system sensitive to glutamine levels (Ronneau et al., 2016). Note that this diagram does not include all the genes regulated by CcrM, GcrA or (

PTS Ntr
SpoT nrdA it is present in S-phase cells (Holtzendorff et al., 2004;Fioravanti et al., 2013;Haakonsen et al., 2015) and it is often co-conserved with CcrM in Alphaproteobacteria (Brilli et al., 2010;Murray et al., 2013). Its mechanism of action remained unclear for over a decade after its discovery, in large part due to its unique domains that did not resemble canonical DNA binding domains. Still, several recent studies shed some light on how GcrA modulates gene transcription in C. crescentus. GcrA appears to be a dual function protein that can interact with the Sigma 73 housekeeping factor (RpoD) loaded onto the RNA polymerase and that can bind to methylated promoter regions through an atypical but dedicated DNA binding domain (Fioravanti et al., 2013;Haakonsen et al., 2015;Wu et al., 2018). A crystal structure of this domain revealed the existence of two protein pockets that are probably responsible for the detection of m6A modifications (Wu et al., 2018). Consistent with its role as an epigenetic transcriptional regulator, purified GcrA appears to have more affinity for certain promoter regions when they are in a methylated, rather than in a non-methylated state (Fig. 2) (Fioravanti et al., 2013). Interestingly, the affinity of GcrA for hemi-methylated motifs on a few tested promoter regions (including the mipZ promoter) is dependent on the DNA strand that is methylated, suggesting that GcrA might regulate certain genes asymmetrically in early pre-divisional cells. Chromatin-immunoprecipitation followed by deep sequencing (ChIP-Seq) experiments confirmed that GcrA binds preferentially, but not exclusively, to Sigma 73 -dependent promoters that harbor GANTC motifs in vivo. Also, GcrA has lower affinity for chromosomal regions in cells that lack the CcrM MTase, confirming the influence of GANTC methylation on GcrA activity (Fioravanti et al., 2013;Haakonsen et al., 2015). Overall, the current model, supported by the analysis of crystal structures of GcrA domains, is that GcrA preferentially activates Sigma 73 -dependent promoters that harbor fully-methylated GANTC motifs through a direct interaction with the domain of Sigma 73 that recognizes the -10 element of core promoters (Haakonsen et al., 2015;Wu et al., 2018). GcrA binding sites can be located in core promoter regions, but also at a significant distance, including downstream of transcriptional start sites (TSS) (Wu et al., 2018).
Considering that GcrA and Sigma 73 are bound to the ftsZ promoter in vivo (Fioravanti et al., 2013;Haakonsen et al., 2015) and that the ftsZ promoter is activated in response to its methylation by CcrM (Gonzalez and Collier, 2013), it is likely that GcrA is the epigenetic regulator that activates ftsZ transcription in stalked and pre-divisional C. crescentus cells, although this was not directly tested in vitro. If this epigenetic mechanism controlling cell division is conserved in some other Alphaproteobacteria, it may explain why gcrA homologs sometimes appear to be required for the fitness of certain Alphaproteobacteria such as R. palustris (Pechter et al., 2015).
Interestingly, a connection between GcrA/CcrMmediated regulation and (p)ppGpp mediated regulation has unexpectedly been uncovered in C. crescentus (Fig. 2) (Sanselicio and Viollier, 2015). (p)ppGpp is the alarmone that is produced in response to glucose and nitrogen starvation (Ronneau et al., 2016;Hallez et al., 2017;). It controls key cellular processes and regulates the expression of many genes in C. crescentus through mechanisms that are not yet fully understood (Boutte and Crosson, 2011;. Among these, several are co-regulated by the GcrA epigenetic regulator, such as ctrA and mopJ (encoding a pleiotropic regulator). Supporting this connection, it was shown that gcrA can be disrupted in fast-growing C. crescentus cells over-expressing (p)ppGpp (Haakonsen et al., 2015). This finding suggests that epigenetic control by GcrA/CcrM can be modulated in response to environmental conditions in Alphaproteobacteria, although the precise mechanism is still unknown.
Although the connection between GcrA and CcrM now appears as obvious, there are still many C. crescentus genes whose expression is different in ∆ccrM cells compared to wild-type cells, but that are not regulated by GcrA or that are not under the control of the housekeeping Sigma 73 factor (Narayanan et al., 2015;Zhou et al., 2015;Adhikari and Curtis, 2016;Collier, 2016). There are also other Alphaproteobacteria that display an essential CcrM MTase but that lack a gcrA homolog, as it is the case for certain strains of A. tumefaciens (Curtis and Brun, 2014). These observations open up new research directions to better understand how epigenetic signals can affect gene expression and cellular processes in Alphaproteobacteria.

Hypomethylated DNA motifs in Alphaproteobacteria and epigenetic patterns
In Gammaproteobacteria, Dam-mediated regulation often depends on the protection of bi-stable promoters by specific transcription factors so that GATC motifs recognized by Dam are found non-methylated in a subset of cells within clonal populations. OxyR, Lrp and Fur, for example, have been shown to create such 'epigenetic patterns' mediating phase variations in Escherichia coli and Salmonella enterica (Wion and Casadesus, 2006). Thus, the presence of hypo-methylated promoter regions can be an indication that such promoters may be under epigenetic control mechanisms.
A first indication that epigenetic patterns may exist in Alphaproteobacteria came from a study characterizing the M. loti genome using restriction landmark genome scanning (RLGS) to detect GANTC motifs that are hypo-methylated on this genome (Ichida et al., 2007). This study revealed that ~170 GANTC motifs were protected from methylation by CcrM in free-living cells, while much fewer were detectable in bacteroids within nodules. The fact that CcrM-dependent epigenetic patterns change during the establishment of symbiosis suggests that epigenetics may contribute to the control of plant-microbe interactions. Using the same method, CcrM-dependent epigenetic patterns were also identified on the genomes of two other plant interacting bacteria: Bradyrhozobium japonicum and A. tumefaciens (Ichida et al., 2007).
The more recent use of the SMRT sequencing technology also revealed the existence of hypo-methylated GANTC motifs on genomes of Alphaproteobacteria that encode a CcrM MTase. The estimation was that ~30-50 GANTC motifs are frequently hypo-methylated on the C. crescentus and Celeribacter marinus genomes (Kozdon et al., 2013;Yang et al., 2016). A more quantitative method based on the use of a methylation sensitive restriction enzyme (HinfI) followed by Illumina-based deep-sequencing (REC-Seq) also confirmed the existence of hypo-methylated sites on the C. crescentus and the S. meliloti genomes (Ardissone et al., 2016). In C. crescentus, these epigenetic patterns do not change as a function of the cell cycle (Kozdon et al., 2013), but do respond to environmental changes such as phosphate availability (Ardissone et al., 2016) or switches from rich to minimal media (Gonzalez and Collier, unpublished observation). In ccrM over-expressing cells, very few of these GANTC motifs remain hypo-methylated Ardissone et al., 2016). This observation indicates that CcrM may compete with DNA binding proteins that occlude GANTC motifs from their methylation by CcrM in wild-type cells. A recent study has shown that the conserved MucR1 and MucR2 proteins, which regulate a transcriptional switch during the S-to-G1 phase transition in C. crescentus (Fumeaux et al., 2014), prevent the efficient methylation of about half of the hypo-methylated GANTC motifs of the C. crescentus genome, most likely through a competition with CcrM ( Fig. 3) (Ardissone et al., 2016). Similar observations were made for the MucR regulator of S. meliloti. There is however no direct evidence that certain promoters regulated by MucR1/2 display bi-stable activities or are subject to phase variation as it is the case for some hypo-methylated promoters in Gammaproteobacteria (Casadesus and Low, 2013;Sanchez-Romero et al., 2015). Still, other hypo-methylated promoters may have such properties. Hypo-methylated GANTC motifs may also play a role in cell division, DNA replication or nodulation control in Alphaproteobacteria since ccrM overproduction most often leads to defects in these processes (Wright et al., 1997;Robertson et al., 2000;Kahng and Shapiro, 2001;Ichida et al., 2009). It is nevertheless worth noting that hypo-methylated GANTC motifs are generally not conserved among different Alphaproteobacteria (Blow et al., 2016), suggesting that epigenetics controls different genes/processes in different bacteria or that the hypo-methylation of many of these motifs has no impact on gene expression.

Influence of DNA methylation on genome maintenance and evolution
Adenine methylation is sometimes involved in DNA mismatch repair (MMR) or in controlling the initiation of DNA replication in Gammaproteobacteria that encode a Dam methylase (Adhikari and Curtis, 2016). Such effects have been rarely tested directly in Alphaproteobacteria. It was however shown that a C. crescentus ∆ccrM strain has a spontaneous mutation rate that is similar to a wildtype strain, showing that CcrM-dependent methylation is not required for MMR in this Alphaproteobacterium . Similarly, this mutant does not display obvious defects in DNA replication control. Then, there is, so far, no experimental indication that DNA methylation is involved in DNA replication control or in MMR in Alphaproteobacteria.
Although R/M systems usually reduce HGT (Fig. 4), R/M systems can sometimes be shut-down in response Alphaproteobacteria. MucR proteins participate in a S-to-G1 phase transcriptional switch controlling cell cycle progression (Fumeaux et al., 2014). Most GANTC motifs are either hemi-(HM) or fullymethylated (FM) in C. crescentus (Kozdon et al., 2013;. Binding of MucR to promoter regions that carry GANTC motifs can block their methylation by CcrM (Ardissone et al., 2016), generating hypo-methylated GANTC motifs that can become nonmethylated (NM) after two rounds of replication, creating MucRdependent epigenetic patterns. Such patterns have been directly detected in C. crescentus and S. meliloti (Ardissone et al., 2016 to unfavorable environmental conditions in bacteria. This may promote DNA uptake and integration by recombination in these conditions, influencing HGT for adaptation (Jeltsch, 2003). In addition, it has been known for a long time that cytosine methylation by orphan or R/M MTases can have an impact on genome evolution. Indeed, methylated cytosines, although sometimes useful for the regulation of gene expression in bacteria, are mutation hot-spots due to their accidental deamination into thymines (Jeltsch, 2003;Marinus and Lobner-Olesen, 2014). TG mismatches must be repaired before the next round of DNA replication, otherwise they turn into undetectable C-to-T mutations. Bacteria can use up to three systems to detect and repair such DNA mismatches (Fig. 4): the highly conserved MMR process (Lenhart et al., 2016), the bacterial very small patch (VSP) repair process (Marinus and Lobner-Olesen, 2014), and DNA glycosylases that can remove damaged bases by base excision repair (BER) (Jacobs and Schar, 2012). Interestingly, such DNA glycosylases might act as enzymes to de-methylate some bases in bacterial genomes in order to modulate epigenetic patterns (Fig. 4), as it is the case in eukaryotes (Schuermann et al., 2016).

Perspectives
There are still many outstanding questions awaiting answers, notably to understand how m6A marks can influence gene expression, whether methylated cytosines can modulate gene expression in Alphaproteobacteria, whether there are phase variation mechanisms mediated by epigenetic switches in Alphaproteobacteria, what are the roles of poorly conserved orphan MTases and whether there are de-methylating processes in bacteria to modulate or re-set epigenetic patterns. With the advent of novel technologies to analyze epigenomes and their expression (Beaulaurier et al., 2015;Blow et al., 2016), we anticipate that major discoveries are likely to be made in the near future.