Roles of DNA adenine methylation in host^pathogen interactions: mismatch repair, transcriptional regulation, and more

The DNA adenine methyltransferase (Dam methylase) of Gammaproteobacteria and the cell cycle-regulated methyltransferase (CcrM) methylase of Alphaproteobacteria catalyze an identical reaction (methylation of adenosine moieties using S -adenosyl-methionine as a methyl donor) at similar DNA targets (GATC and GANTC, respectively). Dam and CcrM are of independent evolutionary origin. Each may have evolved from an ancestral restriction-modiﬁcation system that lost its restriction component, leaving an ‘orphan’ methylase devoted solely to epigenetic genome modiﬁcation. The formation of 6-methyladenine reduces the thermodynamic stability of DNA and changes DNA curvature. As a consequence, the methylation state of speciﬁc adenosine moieties can affect DNA–protein interactions. Well-known examples include binding of the replication initiation complex to the methylated oriC , recognition of hemimethylated GATCs in newly replicated DNA by the MutHLS mismatch repair complex, and discrimination of methylation states in promoters and regulatory DNA motifs by RNA polymerase and transcription factors. In recent years, Dam and CcrM have been shown to play roles in host–pathogen interactions. These roles are diverse and have only partially been understood. Especially intriguing is the evidence that Dam methylation regulates virulence genes in Escherichia coli , Salmonella , and Yersinia at the posttranscriptional level.


Introduction
Postreplicative DNA methylation superimposes on the primary DNA sequence secondary information that has significance for DNA transactions such as transcription, transposition, initiation of chromosome replication, and prevention of mutations by DNA repair (reviewed in Løbner-Olesen et al., 2005;Casadesus & Low, 2006;Wion & Casadesus, 2006;Low & Casadesus, 2008). The most common postreplicative base methylations are N 6 -methyladenine (6-meA) and 5-methylcytosine, which are found in both prokaryotes and eukaryotes, and N 4 -methylcytosine, which is restricted to bacteria. The chromosomes of the model Gammaproteobacteria Escherichia coli and Salmonella enterica serovar Typhimurium (S. enterica hereafter) contain about 20 000 6-meA residues that are the products of two distinct methyltransferases. Most adenine methylations occur in the sequence GATC catalyzed by the enzyme DNA adenine methyltransferase (Dam). In E. coli, about 600 6-meA residues are due to the action of M.EcoK, part of a classical type I restriction/modification system. A third adenine methyltransferase, YhdJ, is not produced under normal laboratory conditions, and its role in cellular metabolism remains unknown (Broadbent et al., 2007). The adenine methyltransferases discussed in this article are not part of a restriction/modification system and are often referred to as orphan or solitary methyltransferases. An exception, however, is the modification subunit of a type III DNA restriction system that regulates gene expression in Haemophilus influenzae (Fox et al., 2007).
Another model organism that has been used to study the physiological roles of DNA adenine methylation is Caulobacter crescentus, a member of the alpha branch of Proteobacteria. In C. crescentus, the cell cycle-regulated methyltransferase (CcrM) recognizes and methylates the sequence GANTC, and has a role in cell cycle-regulated events (Marczynski & Shapiro, 2002). CcrM methylation in C. crescentus and a few other members of the Alphaproteobacteria are included in this review.
The Dam enzyme is encoded by the dam gene, and much of our knowledge about the cellular functions of Dam came from studying dam mutants in E. coli and S. enterica. The properties of these mutants showed that the most-affected DNA transactions are Dam-directed mismatch repair, initiation of chromosome replication, and regulation of gene expression. As discussed below, altered gene expression patterns in dam mutants impair host-pathogen interactions. In addition to the known examples of transcriptional regulation, there are hints that Dam methylation may also influence gene expression by posttranscriptional mechanisms. We list the instances of posttranscriptional regulation and speculative models to explain them. This article is a companion to other recently published reviews (Løbner-Olesen et al., 2005;Casadesus & Low, 2006;Wion & Casadesus, 2006;Heusipp et al., 2007;Low & Casadesus, 2008).

Dam methylation The Dam methyltransferase
The E. coli Dam enzyme, which is a monomer in solution, catalyzes the transfer of the methyl group from S-adenosyl-L-methionine (SAM) to the N 6 position of the adenine residue in GATC sequences, using base flipping to position the base in the enzyme's catalytic site. The natural substrate for the enzyme is hemimethylated DNA, where one strand is methylated and the other is not. This is the configuration of DNA immediately behind the replication fork. Doublestranded DNA is a better methyl acceptor than denatured DNA, and there is little difference in the rate of methylation between unmethylated and hemimethylated DNA (Herman & Modrich, 1982). The enzyme appears to have two SAMbinding sites; one is the catalytic site and the other increases specific binding to DNA, probably through an allosteric transformation (Bergerat et al., 1991). Dam is thought to bind the template and to slide processively along the DNA, methylating about 55 GATC sites per binding event (Urig et al., 2002). The atomic structure of Dam complexed with DNA has been solved to 1.89 Å resolution in the presence of S-adenosyl-homocysteine (a product of the Dam reaction) (Horton et al., 2006). The structure shows both nonspecific backbone contacts and specific contacts with the GATC bases. Importantly, the aromatic ring of Y119 intercalates into the DNA between GA and TC, thereby flipping the adenine into the enzyme's active site. The unpaired T residue can adopt an intrahelical or an extrahelical position. Four other important contacts are made: K9 to G, L122 and P134 to C, and R124 to T. These and flanking phosphate contacts by conserved residues (R95, N126, N132, and R137) position Dam on the DNA duplex.
There are about 130 molecules of Dam per E. coli cell, and this level is optimal to allow a period of time between synthesis of the extending nucleotide chains and methylation of the GATC sequences within them . The actual time between synthesis and methylation can be rapid for plasmid molecules (2-4 s) (Stancheva et al., 1999) or about 1 min for chromosomal DNA in slow-growing cells with a doubling time of about 100 min (Campbell & Kleckner, 1988). Increases or decreases in the number of Dam molecules can profoundly alter the physiological properties of the cell. Adenine methylation reduces the thermodynamical stability of DNA and alters DNA curvature, thereby affecting DNA-protein interactions at certain GATC-containing DNA motifs (Wion & Casadesus, 2006). Steric hindrance of protein binding by the methyl group is also conceivable (Wion & Casadesus, 2006).
The cellular level of Dam is regulated mainly by transcription. The dam gene transcripts arise from five distinct promoters. The major dam promoter (P 2 ) is located 3 kb upstream of the gene , and is regulated by the growth rate: the faster the growth rate, the greater the level of transcript. This makes sense for Dam as fast-growing cells, which contain multiple replication origins, are expected to require a greater concentration of the enzyme than slow-growing ones, which contain few replication origins. Dam is a substrate for the Lon protease, and there might be regulation of the enzyme level by this mechanism (Calmann & Marinus, 2003).
Dam competes with two other proteins, MutH and SeqA, for hemimethylated GATC substrate sites. These two proteins act before Dam to participate in the removal of replication errors (MutH) and to form the compacted and properly supercoiled chromosome structure for the nucleoid (SeqA). Increasing the cellular level of Dam causes a decrease in the amount of hemimethylated DNA, and prevents these two proteins from carrying out their functions, leading to an increased mutation rate and a change in supercoiling of the chromosome, respectively (Herman & Modrich, 1981;Marinus et al., 1984;Løbner-Olesen et al., 2003).
Although Dam methylase is a highly processive enzyme, it may become less processive at GATC sites flanked by specific DNA sequences (Peterson & Reich, 2006). Reduced processivity may allow competition between Dam and specific DNA-binding proteins, thus permitting the formation of nonmethylated GATCs. For instance, the E. coli chromosome contains about 36 specific, unmethylated dam sites (Ringquist & Smith, 1992;Wang & Church, 1992;Hale et al., 1994;Tavazoie & Church, 1998). The number of unmethylated sites in the chromosome varies depending on the growth phase and the growth rate, suggesting that the proteins that bind to them could be involved in gene expression or in the maintenance of chromosome structure. The unmethylated dam sites appear to be mostly (Ringquist & Smith, 1992) or completely (Palmer & Marinus, 1994) modified in strains overproducing Dam, suggesting that the enzyme competes with other DNA-binding proteins at these specific sites. Evidence for competition between Dam and other DNA-binding proteins at several unmethylated sites has been obtained, as discussed in more detail below (see Regulation of gene expression). Alternatively, or in addition, some GATC sites in DNA structures [e.g. non-B-form DNA such as H-DNA (Parniewski et al., 1990)] are relatively resistant to methylation at the normal cellular level of the enzyme. Palindromic structures containing GATCs are also relatively resistant to Dam methylation (Allers & Leach, 1995) (Fig. 1).
In addition to the unmethylated GATC sites discussed above, persistent hemimethylated sequences have been detected in the chromosome (Ogden et al., 1988;Campbell & Kleckner, 1990). These are distinct from the transiently hemimethylated GATC sites that occur immediately behind the replication fork due to the time lag between DNA replication and Dam methylation. The persistent hemimethylated sites are discussed in more detail below (see Initiation of chromosome replication).

Dam-directed mismatch repair
Errors that arise from DNA replication need to be removed from the newly synthesized DNA strand, and not from the parental strand. In E. coli and S. enterica, this discrimination is achieved by virtue of the hemimethylated state of DNA behind the replication fork: the newly synthesized DNA is not methylated, but the parental strand is methylated (Pukkila et al., 1983). The base mismatch formed by a replication error (e.g. G-T) is recognized and bound by the MutS protein, which recruits the MutL protein (a molecular matchmaker) to form a ternary complex with MutH. The latent endonuclease activity of MutH is unmasked by the complex, and the enzyme cleaves the unmethylated strand 5 0 to the G at a nearby GATC site. MutH is then displaced from the complex by the UvrD helicase. UvrD unwinds DNA and the exposed single strand is degraded by exonucleases until the mismatch is removed. The resultant gap is filled in by the DNA polymerase III holoenzyme, and the nick is sealed by DNA ligase. Finally, the hemimethylated GATC is symmetrically methylated by Dam (reviewed in Iyer et al., 2006). Because MutH is active on hemimethylated, but not on fully methylated DNA, mismatch repair action is confined to the hemimethylated region behind the replication fork.
Among the evidence supporting the above model is that both the lack of Dam methylation and the overproduction of Dam lead to the same result: an increase in the spontaneous mutation frequency (Marinus & Morris, 1974;Herman & Modrich, 1981). Overproduction of Dam leads to premature methylation of new DNA, thereby preventing MutH action if a mismatch is present. In turn, lack of Dam results in the loss of strand discrimination, leading to the use of the parental strand as a template for mismatch repair with a 1 2 probability. Single-and double-strand breaks have been detected in the chromosome of dam mutants as a consequence of mismatch repair (Marinus & Morris, 1974;Wang & Smith, 1986). Homologous recombination is required to repair the double-strand breaks, and this explains why mutations inactivating homologous recombination are synthetically lethal in a dam mutant background (Marinus, 2000).

Initiation of chromosome replication
As mentioned above, persistent hemimethylated sites have been detected at the origin of chromosome replication, oriC, and the region surrounding it (Campbell & Kleckner, 1990). This region includes the dnaA gene, which is located 43 kb from oriC. DnaA initiates chromosome replication by binding to oriC and facilitating duplex opening to load DnaB helicase and DNA polymerase III holoenzyme. The persistence of the hemimethylated state is due to the high density of GATC sequences in oriC (11 in 245 bp) and in the promoter region of dnaA (8 in 219 bp), providing multiple binding sites for the SeqA protein. The SeqAinduced hemimethylated state in this region of the chromosome lasts for about one-third of the cell cycle (sequestration), but the mechanism by which it is relieved is not known. The purpose of sequestration is to prevent reinitiation from oriC from occurring more than once per cell cycle. For initiation to occur most efficiently, oriC and the dnaA promoter region must be fully methylated. This also contributes to ensuring that initiation occurs only once per cell cycle (Braun et al., 1985;Yamaki et al., 1988). In S. enterica, SeqA may play replication-related roles similar to those described in E. coli (Prieto et al., 2007). In Vibrio cholerae, both Dam methylation and SeqA are essential (Julio et al., 2001;Saint-Dic et al., 2008), and SeqA overproduction causes DNA replication arrest (Saint-Dic et al., 2008). In fast-growing E. coli or S. enterica cells, the time required for chromosome replication exceeds the doubling time. Under such conditions, E. coli and S. enterica cells contain multiple copies of oriC due to initiations that occurred two or three generations ago. These origins fire simultaneously during the cell cycle, leading to synchronous initiation, which is thought to be due to the immediate release of DnaA from an origin after initiation (reviewed in Nielsen & Løbner-Olesen, 2008). This release will temporarily increase the DnaA/oriC ratio in wild-type cells for the remaining fully methylated origins. After initiation, other mechanisms ensure that DnaA is not in the proper conformation for initiation. Among these mechanisms is a reduction in the transcription of the dnaA gene. Sequestration by SeqA after initiation keeps the dnaA promoter region in a hemimethylated state, which reduces transcription initiation because the dnaA promoter GATC sequences need to be fully methylated for maximal expression (Braun et al., 1985).
In E. coli dam cells, there is no sequestration by SeqA; consequently, DnaA can immediately rebind origins after the first initiation event, and initiate a second time when the concentration of the active form of DnaA is high enough. Transcription from the dnaA gene continues throughout the cell cycle although at a reduced level. Dam methylation, therefore, is not essential for replication initiation; rather, the cell uses methylation to discriminate between old and new origins.

Regulation of gene expression
Because the state of GATC sites (methylated, unmethylated, and hemimethylated) can affect specific binding of DpnI (cuts methylated DNA), DpnII (cuts unmethylated DNA), Dam, SeqA, and MutH, it is not surprising that the presence of this tetranucleotide in the promoter or the regulatory sequences can affect gene expression by regulating binding of RNA polymerase or transcriptional regulators (Fig. 2, Table 1). The promoter region of the dnaA gene discussed above, for example, is maximally active in the fully methylated state, consistent with its biological role. In contrast, there is evidence that specific protein binding yields about 36 unmethylated GATCs in the E. coli chromosome (reviewed in detail by Casadesus & Low, 2006). Nine such GATCs are in the cyclic AMP-binding protein (CAP)binding sites preceding the mtlA, cdd, flhD, gcd, ycdZ, yffE, ppiA, and proP operons (Wang & Church, 1992), suggesting that gene expression might be modulated by Dam methylation through differential CAP binding. Other genes with GATCs that overlap with protein-binding sites are hrsA, kdgT (Fnr), pspA, yjdG (IHF), fep (Fur), carA (CarP, IHF), agn43 (flu) (OxyR), ppiA (Lrp, CAP), and yhiP (Lrp) (Hale et al., 1994;Tavazoie & Church, 1998). Data supporting specific binding of a regulatory protein either in vivo or in vitro are only available for a fraction of the genes listed. Additional unmethylated GATC sites were found in the noncoding regions of rspA, ydjL, yahM, bhsA, yjdD, yhiP, yiaK, yidX, and yihU/V genes (Hale et al., 1994;Tavazoie & Church, 1998), although their significance is not known.
Studies on the pap operon have provided the most detailed evidence that unmethylated GATCs are involved in transcriptional control (reviewed in Casadesus & Low, 2006). Pyelonephritis-associated pilus (Pap) expression is regulated by a phase variation mechanism in which individual cells either express pili (phase-on) or not (phase-off). When Pap pilus gene expression is in the phase-off state, GATC1028 is fully methylated and GATC1130 is unmethylated. Conversely, in the phase-on state, the methylation state at these two sites is reversed. In a strain overproducing Dam, the transition from phase-off to phase-on is prevented, whereas in a dam mutant, the opposite transition does not occur. The mechanism of phase variation involves competition between Dam and the transcriptional activators Lrp and PapI. Lrp is required for methylation protection of GATC1130, and both Lrp and PapI are required for protection of GATC1028 (Casadesus & Low, 2006). Other pilus systems also appear to be under Dam control, but they have not been analyzed as deeply as pap (Casadesus & Low, 2006). Formation of Dam methylation patterns also regulates the agn43 gene of E. coli, which encodes a nonfimbrial adhesin (Henderson & Owen, 1999;Waldron et al., 2002;Wallecha et al., 2002). In addition to unmethylated sites, there is also evidence that hemimethylated GATCs can control gene expression. Transposition of Tn10 is regulated by the methylation state of two specific GATC sites in IS10 right (Roberts et al., 1985). Overproduction of Dam decreases transposition, whereas it is increased in a dam mutant. One of the GATC sites overlaps the À 10 region of the transposase (tnp) promoter, while the other is near the inner end of IS10 in the target area for transposase action. In DNA that is not being replicated, these sites are methylated and inert for transposition. Upon replication, these sites become hemimethylated, but only one of the hemimethylated species is activated for transposition. In a wild-type strain, the transposase promoter is only active in the IS10 species that presents methylation of the transposase-coding strand and unmethylation of the noncoding strand. Coupling of trans-posase synthesis and activity to hemimethylation implies that transposition is repressed for most of the cell cycle, and can only be induced when the element is replicated. The asymmetry imposed at the replication fork means that only one of the two copies of the element can transpose. Hence, one copy can remain in place while the other finds an alternative location. Coupling transposition to replication may help to prevent the potentially deleterious effects of excessive transposition (Roberts et al., 1985). Other transposons such as Tn5 and Tn903 and the insertion element IS3 also use Dam methylation to control transposition (Curcio & Derbyshire, 2003).
Another case of transcriptional activation by strandspecific hemimethylation has been described in the traJ gene of the S. enterica virulence plasmid, albeit with the difference that the regulatory GATC is not located in the promoter itself, but in an upstream binding site for the transcriptional activator Lrp (Camacho & Casadesus, 2002). Another difference is that the active configuration is opposite to that of IS10: methylation of the traJ noncoding strand permits Lrp binding and subsequent traJ transcription, but methylation of the coding strand does not (Camacho & Casadesus, 2005).
Several E. coli promoters have GATC sites in their À 10 or À 35 regions. These include promoter regions for the sulA, trpS, trpR, tyrR, and glnS genes, and expression of these genes is increased in dam mutants (reviewed by Plumbridge, 1987;Barras & Marinus, 1989;Marinus, 1996). It is not known whether expression of these genes is increased in a hemimethylated configuration, but even if it were, the physiological role for coupling their transcription to replication is not obvious.
In the finP gene of the Salmonella virulence plasmid, which encodes a small regulatory RNA, Dam methylation prevents repression by the nucleoid protein H-NS . In contrast, the overlapping traJ gene is also repressed by H-NS, but in a Dam-independent manner. Protection from H-NS repression is still observed when a GATC that overlaps the À 10 module of the finP promoter is eliminated by site-directed mutagenesis . This observation suggests that the effect of Dam methylation on finP transcription is not local, but global, perhaps reflecting, among several possibilities, the existence of structural differences between dam 1 and dam nucleoids.
Global gene expression analysis comparing wild-type and dam mutants using microarrays has been performed in E. coli and S. enterica (Oshima et al., 2002;Løbner-Olesen et al., 2003;Robbins-Manke et al., 2005;Balbontin et al., 2006). The results are difficult to compare, given the differences in the strain backgrounds, media, arrays, and other experimental conditions, as well as in the goals of the experiments. However, upregulation of SOS gene expression in the dam background was detected in each case, and decreased motility in two out of three studies. Dam methylation also occurs in many bacteriophages that infect enterobacteria. The regulation of phage genes by Dam and the role of Dam methylation in P1 development have been reviewed elsewhere (Marinus, 1996;Wion & Casadesus, 2006).

CcrM methylation Caulobacter
Caulobacter crescentus, a member of the Alphaproteobacteria, has defined morphological stages in its life cycle. The DNA methyltransferase in this organism is CcrM, which methylates adenine moieties in the sequence GANTC (Marczynski & Shapiro, 2002). Although the DNA methylation target of CcrM is similar to that of Dam, CcrM belongs to a different family of methyltransferases. In contrast to Dam, CcrM is more active on hemimethylated than unmethylated DNA. However, like Dam, CcrM is highly processive. Unlike Dam in E. coli or S. enterica, CcrM is an essential function in Caulobacter, and is not present at all stages of the life cycle. Both Dam and CcrM are substrates for the Lon protease (Reisenauer et al., 1999).
The life cycle of Caulobacter involves differentiation into two cell types: stalked cells and swarmers. Chromosome replication occurs only in stalked cells, and involves the sequential action of three key unstable regulators: DnaA, GcrA, and CtrA (Collier et al., 2007). The genes for these regulators are located sequentially on the chromosome, with dnaA closer to the origin of replication (Cori) and ctrA at the most distal location. The action of these regulators, acting as a transcriptional cascade, is determined by the state of methylation of chromosomal DNA. DnaA initiates chromosome replication at the fully methylated Cori in a manner similar to that described in E. coli. Because CcrM is not present at this stage, replication produces two hemimethylated daughter DNA molecules during fork progression. As in E. coli, expression of the dnaA gene, which lies near Cori, is attenuated on hemimethylated DNA, thereby reducing the possibility of premature initiation. DnaA also activates transcription of the gcrA gene, whose product controls transcription of replication genes encoding DNA polymerase III holoenzyme, DNA helicase, and primase. GcrA in turn activates transcription of the ctrA gene, which contains two GANTC sequences in the upstream regulatory region of the promoter and one close to the À 35 hexamer (Collier et al., 2006). Again, this promoter is active only when hemimethylated; expression of the gene is, therefore, coordinated with the cell cycle (Reisenauer & Shapiro, 2002). CtrA binds Cori to prevent premature initiation. In addition, CtrA activates transcription of the ftsZ and ccrM genes, and represses transcription of gcrA. FtsZ is a key cell division protein and its CtrA-controlled production can couple chromosome replication and cell division. Transcription of the ccrM gene occurs only in the hemimethylated state, and is activated by CtrA binding to the upstream regulatory region of ccrM (Collier et al., 2007). This arrangement ensures that the concentration of CcrM increases toward the end of the replication cycle. The ccrM promoter also contains two GANTC sequences, presumably ensuring autoregulation of the gene (Reisenauer et al., 1999). The production of CcrM is followed by methylation of the daughter chromosomes, which silences the ctrA and ccrM genes and activates transcription of dnaA. CcrM also prepares Cori for replication initiation by fully methylating it (Table 2).
After cell division, the DNA of both cell types is fully methylated (Marczynski & Shapiro, 2002). In the swarmer cell, CtrA remains bound to Cori and the CcrM protein is degraded, preventing further methylation and thereby ensuring that the origin is hemimethylated and inert for further initiation. In the stalked cell, however, CtrA is destroyed by proteolysis, allowing initiation to proceed on the fully methylated Cori (Marczynski & Shapiro, 2002).

Other bacteria
At least 20 other members of the alpha subdivision of Proteobacteria contain CcrM homologs (Reisenauer et al., 1999). In Agrobacterium tumefaciens, Sinorhizobium (Rhizobium) meliloti, and Brucella abortus, the ccrM gene is indeed essential for viability (Wright et al., 1997;Robertson et al., 2000;Kahng & Shapiro, 2001), and the ccrM genes from S. meliloti and C. crescentus are functionally interchangeable (Wright et al., 1997). When overproduced in any of these organisms, CcrM causes defects in cell division, cell morphology, and initiation of DNA replication. All the above data suggest that the physiological functions of CcrM in Caulobacter might be conserved in these other species.

Roles of DNA adenine methylation in host--pathogen interactions
Evidence for a relationship between Dam methylation and bacterial virulence was first provided by the regulation of adhesin-encoding genes such as the pap operon of E. coli and others (Casadesus & Low, 2006). However, the role of Dam methylation in the infection of model animals was first    (Fig. 3). A simple genetic approach was to compare the lethal dose 50% (LD 50 ) of a dam mutant with that of the wild type. Additional details about the infection process were provided by examination of animal organs and in vitro studies using cell cultures. In bacterial species where DNA adenine methylation is essential, an alternative strategy was to examine the effects of Dam and CcrM methylase overproduction. Although not known to occur in nature, Dam methylase overproduction provides a useful laboratory tool, both to overcome viability problems and to detect cases in which undermethylation is a critical factor for gene expression. However, Dam overproduction in E. coli leads to a seqA phenotype (Løbner-Olesen et al., 2003). If this is also the case in other organisms, certain phenotypic effects observed on Dam overproduction might be the result of SeqA deficiency.

Salmonella enterica
Dam methylation is an essential factor in Salmonella virulence, and its absence causes severe attenuation in the mouse model: the LD 50 of a dam mutant is 10 000-fold higher than that of the wild type by the oral route, and 1000-fold higher intraperitoneally (Garcia-del Portillo et al., 1999;Heithoff et al., 1999). Lack of SeqA causes a more modest decrease in Salmonella virulence, and attenuation is only observed by the oral route (Prieto et al., 2007). In the last few years, a combination of genetic screens, transcriptomic and proteomic analyses, cell culture studies, and organ analysis upon mouse infection have provided insights into the causes underlying the extreme attenuation of Salmonella dam mutants. To date, the following virulence-related defects have been described: (1) Salmonella dam cells show a reduced capacity to interact with the intestinal epithelium, due to impaired epithelial cell invasion (Garcia-del Portillo et al., 1999). The main cause of this defect seems to be inefficient activation of genes in S. enterica pathogenicity island I (SPI-1) (Balbontin et al., 2006). Inefficient SPI-1 expression in dam mutants reflects the existence of lowered levels of the main SPI-1 transcriptional activator, HilD. However, Dam-dependent regulation of hilD appears to be posttranscriptional, and therefore indirect (J. López-Garrido & J. Casadesus, unpublished data).
(2) Reduced motility, another relevant defect of Salmonella dam mutants, may also contribute to inefficient invasion. Transcriptome analysis has shown that dam mutants have multiple alterations in the expression of flagellar genes in a pattern too complex to be readily deciphered or even modeled (Balbontin et al., 2006).
(3) Lack of Dam methylation causes envelope instability, with the release of outer membrane vesicles and leakage of proteins (Pucciarelli et al., 2002). Vesicle release has been tentatively associated with impaired binding of Tol and PAL envelope proteins to peptidoglycan. Protein leakage may also be a side effect of envelope fragility. One factor contributing to envelope instability in dam mutants may be reduced transcription of the lppB gene, which encodes Braun lipoprotein (Balbontin et al., 2006).
(4) The std fimbrial operon, which is tightly repressed under laboratory conditions, undergoes derepression in dam mutants, and the StdA fimbrial protein becomes one of the most abundant proteins detected in cell extracts (Alonso et al., 2005). Ectopic production of Std fimbriae contributes to virulence attenuation in S. enterica dam mutants, as indicated by the observation that an stdA dam strain outcompetes a dam strain during mouse infection (Jakomin et al., 2008). It is possible that massive expression of Std fimbriae in dam mutants may interfere with signal exchange between the host and the pathogen, and may additionally overalert the host immune system. Dam   methylation probably controls binding of transcriptional regulators to the std upstream activating sequence, which contains a cluster of three GATC sites. Genetic screens for std regulators have identified the GATC-binding protein SeqA and RosE, a homolog of the arginine repressor ArgR, as repressors of std expression Jakomin et al., 2008). In turn, the poorly known HdfR protein, a LysR relative, is an activator of std expression in dam and seqA mutants. Interestingly, derepression of the std operon by dam and seqA mutations occurs in only a fraction of the bacterial culture, suggesting the occurrence of either bistable expression or phase variation (Jakomin et al., 2008).
(5) Salmonella dam mutants are extremely sensitive to bile salts, a defect that may compromise their survival in the hepatobiliary tract Pucciarelli et al., 2002). The main extracellular niche for Salmonella in persistent infections and during chronic carriage is the gall bladder, which contains high concentrations of bile. Because of their envelope defects, dam mutants are more sensitive to the detergent activity of bile salts. In addition, lack of DNA strand discrimination for mismatch repair makes dam mutants more sensitive to the DNA-damaging activity of bile salts (Prieto et al., 2004). The relevance of bile-induced DNA damage during animal infection is illustrated by the ample repertoire of Salmonella DNA repair functions required to cope with bile-induced DNA lesions: besides Dam-directed mismatch repair, bile resistance also requires base excision repair, SOS translesion synthesis, and RecBmediated recombinational repair (Prieto et al., 2006). Although bile salts are weak mutagens, long exposure to high concentrations of bile (e.g. in persistent and chronic infections) might increase genetic polymorphism in Salmonella populations. This view is consistent with the high frequency of chromosome rearrangements known to occur in Salmonella typhi (Echeita & Usera, 1998) and other hostadapted serovars (Liu & Sanderson, 1995;Liu & Sanderson, 1998) (Table 3).

Enterohemorrhagic E. coli OH157:O7
A critical step during colonization and pathogenesis by enterohemorrhagic E. coli is the formation of 'pedestals' that result from the accumulation of actin filaments beneath adherent bacteria, elevating them above the surrounding cell surfaces (Hayward et al., 2006). Wild-type E. coli O157:H7 show relatively poor pedestal formation on cultured mammalian cell lines, while deletion of the dam gene results in a dramatic increase in both adherence and actin pedestal formation (Campellone et al., 2007). Increases in adherence and pedestal formation in vitro correlate with elevated protein levels of intimin, Tir, and another secreted protein, EspF U .
Dam methylation plays an additional role in enterohemorrhagic E. coli by controlling the production of a virulence factor, Shiga toxin 2 (Stx2) (Murphy et al., 2008). This toxin is encoded by a lambdoid prophage that has a relatively low threshold for induction. During infection, prophage induction may occur in a fraction of the bacterial population, thereby permitting Stx2 release.

Haemophilus influenzae
Certain strains of H. influenzae, a causative agent of respiratory tract infections, require Dam methylation for efficient invasion of both endothelial and epithelial cell lines (Watson et al., 2004). In other strains, however, dam mutants are fully invasive. The cause of these strain-specific differences is not known (Watson et al., 2004).
In addition to Dam, H. influenzae possesses a DNA methyltransferase (Mod) that is part of a type III restriction-modification system and undergoes phase-variation expression due to DNA repeat instability. Mod has been shown to regulate a number of H. influenzae genes, some positively and others negatively (Srikhanta et al., 2005). Phase variation of Mod expression may cause random switching of genes such as dnaK, potentially involved in cell adhesion, hbpA, which encodes a heme transport protein, and several genes for surface proteins of unknown function. Phase variation of Mod may thus provide a mechanism for the generation of diversity in H. influenzae populations by controlling a phase-variable regulon or 'phasevarion' (Srikhanta et al., 2005;Fox et al., 2007). Systems of this kind might exist in other bacterial pathogens as Helicobacter pylori and Neisseria meningitidis.

Pasteurella multocida
In the bovine respiratory pathogen P. multocida, overproduction of Dam methylase causes attenuation in the mouse model, suggesting that Dam methylation may control the expression of virulence genes (Chen et al., 2003). It is not known whether the Pasteurella Dam methylase, which is closely related to that of H. influenzae, is essential or dispensable.

Actinobacillus actinomycetemcomitans
Synthesis and secretion of leukotoxin, a potential virulence factor related to RTX pore-forming hemolysins, is exacerbated in dam mutants of the periodontal disease agent A. actinomycetemcomitans. Furthermore, A. actinomycetemcomitans dam mutants show reduced invasion of epithelial cells (Wu et al., 2006).

Klebsiella pneumoniae
In K. pneumoniae, an opportunistic pathogen causing respiratory and urinary tract infections, lack of Dam methylation causes partial attenuation upon intranasal or intraperitoneal inoculation of mice (Mehling et al., 2007). The mild attenuation of Klebsiella dam mutants has not been hitherto correlated with altered expression of known virulence genes.

Campylobacter jejuni
Knockout of a putative DNA methyltransferase gene (cj1461) in the intestinal pathogen C. jejuni causes reduced motility, aberrant flagellar appearance, and hyperadherence to epithelial cells, accompanied by reduced invasion (Kim et al., 2008). Some of these traits are reminiscent of virulence-associated defects described previously in Salmonella dam mutants. However, the putative Cj1461 protein shows little homology with Dam methylase, and neither its DNA methylation activity nor its DNA target have been determined so far.

Yersinia enterocolitica
Dam methylation is essential in certain strains of Yersinia, and dispensable in others (Julio et al., 2001;Robinson et al., 2005). In Yersinia strains in which Dam methylation is essential, Dam methylase overproduction does not impair growth (Julio et al., 2001). However, Dam-overproducing strains are attenuated in the mouse model, and their avirulent phenotype is pleiotropic (Julio et al., 2001(Julio et al., , 2002. A relevant defect of Y. enterocolitica Dam overproducers is enhanced invasion capacity, probably associated with transcriptional alterations in invasin genes inv and ail, and with changes in the composition of lipopolysaccharide O-antigen . The latter phenotype may involve posttranscriptional control. Furthermore, Y. enterocolitica Dam-overproducing strains show impaired secretion of Yop effector proteins, which become insensitive to Ca 21mediated control (Julio et al., 2002). The latter defect is associated with enhanced degradation of LcrG, which in turn reflects increased transcription of the gene encoding ClpP protease (Fälker et al., 2005). An additional trait that may contribute to attenuation in Dam-overproducing strains is enhanced bacterial motility . Some such defects have also been described in viable dam mutants of the related species Yersinia pseudotuberculosis and Yersinia pestis (Robinson et al., 2005;Taylor et al., 2005).

Vibrio cholerae
Vibrio cholerae mutants lacking Dam methylase are not viable. However, as described above for Y. enterocolitica, Dam overproduction does not impair bacterial growth. Vibrio cholerae overproducers of Dam methylase are attenuated in the suckling mouse model, but the causes of attenuation remain to be established (Julio et al., 2001).

Aeromonas hydrophila
Aeromonas, a promiscuous pathogen of humans and animals, requires Dam methylase for viability. However, as in similar cases reported above, investigators have been able to examine the involvement of Dam methylation in pathogenesis by constructing Dam methylase-overproducing strains (Erova et al., 2006a, b). Attenuation was observed upon intraperitoneal infection of mice with a Dam overproducer, and several virulence defects were identified in vitro: (1) reduced cytotoxicity associated with type III secretion; (2) reduced motility; and (3) enhanced cytotoxic and hemolytic activities associated with the Act enterotoxin, which is secreted by a type II secretion system. All virulence-related alterations associated with Dam methylase overproduction disappeared when critical amino acids within its DNA methylation motif were eliminated, thereby confirming that DNA adenine methylation is involved in A. hydrophila pathogenesis (Erova et al., 2006a, b).

Brucella abortus
Overproduction of CcrM methylase decreases the proliferation of B. abortus inside murine macrophages, suggesting that CcrM methylation may play a role in intracellular replication, which is a hallmark of Brucella infections (Robertson et al., 2000). Because CcrM is essential in this species, inhibitors of the enzyme have been sought and considered as potential antimicrobials (Benkovic et al., 2005).

Posttranscriptional regulation by DNA adenine methylation: facts
Very short patch (VSP) repair in E. coli A system that altered the frequency of recombinants by correcting T-G mismatches with repair tracts shorter than 20 bp was discovered during a study of homologous recombination in bacteriophage lambda, and was termed VSP repair (Lieb, 1983). A hotspot for C to T transitions in the lambda cI gene turned out to be located in a DNA cytosine methyltransferase (Dcm) recognition sequence, CCAGG, altering it to CTAGG (Coulondre et al., 1978). Deamination of 5-methylcytosine (5-meC) yields thymine, and thus creates a T-G mismatch. Mutations due to 5-meC deamination occur in the stationary phase, but not in exponentially growing bacteria, and the role of VSP repair is to prevent the resulting mutagenic event by restoring C-G pairs before DNA replication (Lieb & Bhagwat, 1996;Lieb & Rehmat, 1997). T-G mismatches are recognized and cleaved 5 0 to the T by the Vsr endonuclease (Hennecke et al., 1991); conventional base excision repair involving DNA polymerase I and DNA ligase then follows (Lieb & Bhagwat, 1996). There is also a requirement for the MutS and MutL proteins of Damdirected mismatch repair, but their role is uncertain (Bhagwat & Lieb, 2002). The level of Vsr in the wild type is low in logarithmic-phase cells, and high in stationary-phase cells, as expected from the biological rationale for VSP (Macintyre et al., 1999). The E. coli vsr gene is in a transcriptional unit with the dcm gene. The 3 0 end of the dcm gene is overlapped by the first six codons of the vsr gene, which is in a 11 register relative to dcm (Dar & Bhagwat, 1993). Such an overlap is uncommon in E. coli, and in this case may serve to couple the expression of these genes. Both dcm and vsr appear to be transcribed into a single mRNA, and translation of vsr appears to be dependent on translation of the upstream dcm-coding sequence (Dar & Bhagwat, 1993). However, Western analysis showed that the Vsr level varies with the growth rate, while the level of Dcm does not change during the exponential and stationary phases of growth. The mechanism by which this is achieved is not known. The location of the promoter and its mode of regulation are also unknown.
Surprisingly, VSP repair is reduced in E. coli dam mutants as measured by an increase in the mutation frequency of CCAGG to CTAGG (Bell & Cupples, 2001). Western blotting indicated that, unlike the wild type, there was no increase in the Vsr level upon entry into the stationary phase in dam cultures. However, the level of Dcm remained unaltered. Because the vsr and dcm genes are cotranscribed, it was concluded that regulation of Vsr in a dam mutant is probably achieved by a posttranslational mechanism (Bell & Cupples, 2001). Because the vsr mRNA levels in dam 1 and dam strains were not determined, an effect on mRNA stability cannot be excluded.
Pedestal formation in enterohemorrhagic E. coli O157:H7 As described in the host-pathogen interactions section above, dam mutants of enterohemorrhagic E. coli show increased adherence and pedestal formation in vitro, which is correlated with elevated protein levels of three effector proteins: intimin, Tir, and EspF U . However, the increased levels of effectors did not result from an increase in mRNA levels as measured by microarrays, Northerns, and reverse transcriptase (RT)-PCR, suggesting a posttranscriptional mechanism of regulation (Campellone et al., 2007). To further investigate the basis of this observation, an E. coli O157:H7 hfq mutant was constructed, and pedestal formation was as robust as

Lipopolysaccharide composition in Y. enterocolitica
Dam overproduction in Y. enterocolitica causes numerous metabolic alterations, including a change in the composition of lipopolysaccharide O-antigen, which contains increased amounts of lipid A core without O-antigen subunits Heusipp et al., 2007). The O-antigen gene cluster consists of two transcriptional units, but the transcript levels in the Dam overproducer, as measured by RT-PCR, of representative genes in each cluster (ddhA, gne, and rosA) were unchanged relative to the wild type. Thus, the modulation of lipopolysaccharide structure seems to involve an unknown posttranscriptional mechanism .

Transcription of SPI-1
As described in the host-pathogen interactions section above, pathogenicity island SPI-1 is essential for virulence of S. enterica. Transcriptomic analyses of S. enterica model strain SL1344 and dam derivatives showed that transcription of invasion genes in pathogenicity island SPI-1 was decreased in the absence of Dam methylation (Balbontin et al., 2006). In agreement, using a lac transcriptional fusion to one of the SPI-1 genes (sipC), b-galactosidase activity was found to increase when a culture of the wild type entered the stationary phase. In contrast, dam mutants showed a low level of expression both in the exponential phase and in the stationary phase (Balbontin et al., 2006). Transcriptional data were recapitulated at the protein level: more SipC was found in the wild type than in the dam mutants. A similar effect of Dam methylation was found for other representative SPI-1 genes, suggesting that the whole island might be under Dam control.
Transcriptional control of pathogenicity island SPI-1 is complex and multilayered (reviewed by Jones, 2005). Transcription of the hilA gene in SPI-1 appears to be central for the expression of the other genes on the island. Expression of hilA is activated by RtsA, HilC, and HilD, the latter being more important. HilD is a member of the AraC/XylS family of transcriptional activators. Expression of hilD is in turn modulated by the products of the csrA and csrB genes, which have opposite effects on transcript levels. CsrA destabilizes specific mRNA molecules by interactions at the ribosomebinding site. The csrB gene encodes an untranslated RNA that binds CsrA to prevent mRNA degradation. While deletion of the csrB gene has only a mild effect on hilA transcription and none on invasion, deletion of csrA or overexpression of CsrA both reduce hilA transcription about 10-fold, and invasion of epithelial cells about 100-fold. CsrA, therefore, appears to have both positive and negative regulatory effects on SPI-1 gene expression.
Posttranscriptional regulation by DNA adenine methylation: hypotheses Proteolysis Posttranscriptional regulation can occur at the level of the message and/or at the protein level. There are many examples of 'unstable' regulatory proteins whose concentration is determined by the balance between synthesis and degradation by specific proteases. For example, one model for E. coli VSP repair could be that proteolysis is more active on Vsr during logarithmic growth than in the stationary phase. This would imply that a specific protease activity is decreased in cells approaching the stationary phase of growth. In dam mutants, this decrease would not occur, leading to continued proteolysis. Similar arguments can be made for the other systems described above. Whatever the mechanism of posttranscriptional regulation turns out to be, the key proteins will be subject to proteolysis, even though this may not be the primary regulatory mechanism. For further discussion, however, we will assume that proteolysis is not the primary mechanism of posttranscriptional control.

mRNA translation
For some of the examples of posttranscriptional regulation listed above (e.g. VSP repair in E. coli), mRNA levels have not been determined, leaving open the possibility that mRNA stability is altered in dam mutants. In the other examples above, the steady-state mRNA levels have been measured and are unchanged. An appealing model for such situations is to invoke translational regulation through the involvement of a small RNA molecule, either an antisense RNA or a small noncoding RNA (sRNA). An example of Dam-mediated regulation was described in the synthesis of FinP, an antisense RNA acting on the traJ transcript (Torreblanca et al., 1999;. Transcription of finP is decreased in dam mutants . In addition to this example, microarray analysis indicates that many noncoding regions of the E. coli chromosome show either increased or decreased transcript levels in a dam mutant vs. a wild type (Campellone et al., 2007). For each example of posttranslational regulation discussed below, we offer a hypothesis invoking small RNA molecules.
The vsr transcript also includes, and is preceded by, the dcm-coding sequence. While the Dcm protein is present at the same steady-state level in both the exponential and the stationary phases, the Vsr protein is induced in the stationary phase (Bell & Cupples, 2001). An sRNA could be induced upon entry into the stationary phase, allowing translation of the vsr gene, but not affecting the upstream dcm sequence. The sRNA could, for instance, unmask a readthrough region of the transcript or stabilize the 3 0 end. This hypothesis fits with the observation that many sRNAs are induced as cells enter the stationary phase (Majdalani et al., 2005). In dam mutants, transcription of the particular sRNA might be downregulated, thereby preventing vsr translation.
The EHEC tir transcript also encodes the gene (eae) encoding intimin, but not that for EspF U . Because there are examples of an sRNA molecule binding to two separate messengers, coregulation of translation of espF U and tir is possible. In this case, an sRNA might bind constitutively to the mRNAs to prevent translation (negative regulation). Alternatively, some structural feature of the mRNA might prevent translation, and the sRNA might modify it to allow translation (positive regulation). By a signal currently unknown, transcription of the sRNA gene might be altered when the organism finds itself at the right place in the alimentary tract, allowing translation of the messages. In a dam mutant, synthesis of the sRNA would be altered, allowing constitutive translation of the effector messages.
A similar type of model can be proposed for lipopolysaccharide composition changes in Yersinia. Overexpression of Dam is known to alter transcription profiles such that genes that are not normally expressed are activated, and these could include loci for small regulatory RNAs and/or antisense RNAs. Such changes in sRNA or antisense RNA levels might be due to Dam preventing SeqA from accessing its substrate sites. Hence, it would be interesting to analyze lipopolysaccharide composition in a Yersinia seqA mutant.
For pathogenicity island SPI-1, a role for sRNA or antisense RNA could also be invoked for one of the many regulators known to activate transcription of the island. As mentioned above, one of these regulators is a small RNA, csrB. Increased csrB transcription in a dam mutant could affect the level of CsrA, which in turn activates SPI-1 gene expression through HilD. However, preliminary evidence suggests that neither csrA nor csrB are under Dam methylation control (J. López-Garrido & J. Casadesus, unpublished data).
Most sRNA molecules require Hfq for binding to their cognate mRNAs. If the above models require sRNAs, then the effects of dam methylation should be mirrored in an hfq mutant. A positive correlation would support the model while a negative one would exclude Hfq-dependent sRNAs, but not antisense RNAs.
Practical uses of DNA adenine methylation: vaccines and DNA adenine methylase inhibitors The strong attenuation of dam mutants, combined with their capacity to persist at low levels in animal organs (causing an almost asymptomatic infection), makes Salmonella dam strains appropriate for use as live vaccines. In fact, dam mutants of S. enterica have been shown to elicit immune responses in chickens and calves with paramount efficiency (Dueger et al., 2001(Dueger et al., , 2003Heithoff et al., 2001). An oral, live dam vaccine has also been described in H. influenzae (Watson et al., 2004). Interestingly, a viable dam mutant of Y. pseudotuberculosis was found to protect mice against infection by the wild type, and also to cross-protect against plague (Robinson et al., 2005). An alternative and efficient strategy for the design of live vaccines against Y. pseudotuberculosis is Dam overproduction (Julio et al., 2002).
A negative trait that may hamper the use of dam mutants as live vaccines for humans is hypermutation, caused by the lack of DNA strand discrimination for mismatch repair. In Salmonella, lack of Dam methylation gives rise to a 10-15fold increase in spontaneous mutation rates (Torreblanca & Casadesus, 1996). Dam overproduction does not provide a solution to overcome this problem: early studies in E. coli dam mutants showed that Dam overproduction increases mutation rates well above Dam absence (Marinus et al., 1984). Similar observations have been made in Salmonella: expression of Dam methylase from a multicopy plasmid increases the spontaneous mutation rate over 400-fold (Torreblanca & Casadesus, 1996). Increased mutation rates might, however, allow the use of dam vaccines in livestock animals, if animal health regulations permit.
Attenuation of dam strains in a variety of human pathogens has also raised the possibility of using Dam or CcrM methylase inhibitors as antibacterial drugs (Benkovic et al., 2005;Mashhoon et al., 2006). In Alphaproteobacteria, such drugs would be bactericidal. In Salmonella and other pathogens in which Dam methylation is not essential, Dam inhibitors could be expected to attenuate virulence by transforming wild-type bacteria into phenocopies of dam mutants. Because Dam methylation is a dispensable function in enteric bacteria, inhibitors specifically targeted at Dam methylase should be harmless for the normal intestinal flora. A drug of this kind should also be harmless for the host, because adenine methylation is rare, if not absent, in mammalian cells (Ratel et al., 2006).
Whatever the fate of Dam-based vaccines and Dam inhibitors in Salmonella and other pathogens, neither strategy can be envisaged as universally valid. Enterohemorrhagic E. coli dam mutants show increased production of both actin pedestals and Stx (Campellone et al., 2007;Murphy et al., 2008). They are, therefore, potentially more virulent, and useless as live vaccines. In turn, administration of a Dam-inhibiting drug to infected animals might increase enterohemorrhagic E. coli virulence. An additional potential problem is that a Dam inhibitor might increase the spontaneous mutation rate in the intestinal flora.