Cleavage of Phosphorothioated DNA and Methylated DNA by the Type IV Restriction Endonuclease ScoMcrA

Many taxonomically diverse prokaryotes enzymatically modify their DNA by replacing a non-bridging oxygen with a sulfur atom at specific sequences. The biological implications of this DNA S-modification (phosphorothioation) were unknown. We observed that simultaneous expression of the dndA-E gene cluster from Streptomyces lividans 66, which is responsible for the DNA S-modification, and the putative Streptomyces coelicolor A(3)2 Type IV methyl-dependent restriction endonuclease ScoA3McrA (Sco4631) leads to cell death in the same host. A His-tagged derivative of ScoA3McrA cleaved S-modified DNA and also Dcm-methylated DNA in vitro near the respective modification sites. Double-strand cleavage occurred 16–28 nucleotides away from the phosphorothioate links. DNase I footprinting demonstrated binding of ScoA3McrA to the Dcm methylation site, but no clear binding could be detected at the S-modified site under cleavage conditions. This is the first report of in vitro endonuclease activity of a McrA homologue and also the first demonstration of an enzyme that specifically cleaves S-modified DNA.


Introduction
The sequence of the DNA bases contains the genetic information that is copied with great accuracy and inherited by successive generations. DNA may also carry so called epigenetic modifications that are added by host enzymes after replication. These modifications are lost or changed after transfer of the DNA to a new host by conjugation, transformation or transfection. Epigenetic modifications can also change in response to changing environmental conditions, and they are generally important for eukaryotic gene regulation including carcinogenesis [1].
In bacteria, epigenetic modifications have more specialized roles including the protection of self DNA against restriction endonucleases (REases) which cleave foreign, differently modified DNA. Potential invaders, such as bacteriophages, have developed DNA modifications and other measures to overcome these restriction barriers, and bacteria have evolved to restrict even the modified foreign DNA. One such strategy is for bacteria to restrict methylated DNA [2].
There are many methyl-specific REases. Some cleave DNA at specific methylated recognition sequences. Others make contact with their target DNA at a specific methylated recognition sequence, and then move along the DNA before cleaving at an undefined site. These Type IV methyl-specific REases are very diverse in their amino acid (aa) sequences, and they may consist of one or several peptides [3].
There are about 1303 putative Type IV REases in REBASE (http://rebase.neb.com), but only 3 have been biochemically characterized, whereas others are predicted based on bioinformatic analysis of DNA sequences. Quite unusually, some Type IV REases recognize methylated and also hydroxymethylated or glucosyl-hydroxymethylated DNA [4,5].
We have discovered a Type IV REase from a bacterium of the genus Streptomyces that cleaves methylated and phosphorothioated (S-modified) DNA that has a non-bridging oxygen of the phosphate group in the DNA backbone replaced by sulfur. The DNA S-modification is sequence specific, and a dnd (DNA degradation) gene cluster encoding four or five proteins is responsible for the S-modification [6,7].
Streptomycetes are filamentous soil bacteria that produce many chemically diverse antibiotics. The 8.7 Mb linear genome of Streptomyces coelicolor A3(2) has been fully sequenced and annotated [8]. S. coelicolor contains at least four methyl-specific restriction endonucleases that restrict (reduce or prevent) the introduction of methylated DNA e.g. from Dam + Dcm + Hsd + E. coli K-12 strains [9]. Therefore, DNA is generally passaged through a nonmethylating dam dcm hsd E. coli host before introduction into S. coelicolor [10,11].
Alternatively, the less restricting Streptomyces lividans 66 has been used as a recipient for methylated DNA from E. coli K. The genes of S. lividans and S. coelicolor are very similar and highly syntenous except for several genomic islands (GIs), some of which are mobile elements [12,13].
One of these methyl-specific endonucleases, ScoA3McrA (Sco4631) [4] is the focus of this report. As shown schematically in Figure 1, ScoA3McrA is located on the 17.1 kb mobile element Gi11, which is also known as the conjugative and integrative plasmid SLP1 [14].
The Type IV REase EcoKMcrA restricts DNA methylated at the sequences C5mCGG and has been shown to bind to this sequence, but in vitro DNA cleavage has not been observed [15][16][17]. Sequence comparison and phylogenetic analysis showed that the ScoA3 and EcoK McrA proteins have very little aa sequence similarities apart from the region surrounding the conserved HNH motif ( Figure S7B).
About one in 6000 phosphate groups of the backbone of S. lividans genomic DNA contain sulfur instead of a non-bridging oxygen at specific sequences [7,18]. The stereospecific Smodification (phosphorothioation) renders purified DNA susceptible to oxidative double-strand cleavage by Tris peracid which is generated at the anode during electrophoresis [19]. Five genes, dndA-E located on the apparently non-transmissible 93 kb SLG genomic island [13], are involved in DNA S-modification. Evidence for DNA S-modification has been discovered in several Streptomyces species, and in phylogenetically diverse prokaryotes including several economically important species [13,20].
The DNA S-modification of S. lividans does not seem to be accompanied by a cognate restriction endonuclease that could defend the strain against the invasion of bacteriophages that lack this modification, and S. lividans has traditionally been used as a permissive host for the isolation of many Streptomyces phages [11].
In this report we show that S-modified DNA (as well as methylated DNA) is restricted in vivo by the S. coelicolor methylspecific endonucleases ScoA3McrA. We also report site-specific in vitro cleavage of S-modified DNA by the purified enzyme. This is the first indication of a biological role of DNA S-modification, and also the first report of in vitro DNA cleavage by a McrA homologue.

Results
Failure to introduce the DNA sulfur modification gene cluster into S. coelicolor The dndA-E gene cluster of S. lividans is responsible for the DNA S-modification (phosphorothioation) of this strain [6]. We wanted to express this gene cluster in S. coelicolor which is closely related to S. lividans but lacks these particular genes. In order to generate a stable recombinant S. coelicolor strain, we cloned the dndA-E gene cluster into the wC31-derived integrative vector pSET152 which has an apramycin resistance marker for selection, and an origin of transfer for highly efficient conjugative transfer from E. coli ET12567/pUZ8002 (dam dcm hsd tra + ) to S. coelicolor M145. The plasmid vector without cloned DNA (pSET152) produced as expected c. 10 5 apramycin resistant S. coelicolor exconjugants per experiment (one Petri dish). Unexpectedly, pSET152::dndA-E (pHZ1904) produced only an average of nine apramycin resistant exconjugants per Petri dish ( Figure 2, Panel 1).
These apramycin resistant pSET152::dndA-E exconjugants were expected to express the dnd genes and generate S-modified S. coelicolor DNA that is visibly degraded during agarose gel electrophoresis (methods used in [21]). Only 49 out of 100 exconjugants showed the expected DNA degradation (Dnd + phenotype; Dnd stands for DNA degradation). The remaining 51 contained stable, S-free DNA (Dnd 2 phenotype, Figure S1A).
The dndA-E gene cluster of three Dnd 2 exconjugants was analysed and each contained a different mutation that was likely to abolish the S-modification activity ( Figure S1B).
One of these plasmids, pHZ1904* contained the entire dndA-E gene cluster with a single base insertion (frameshift mutation) in dndE which is required for DNA S-modification ( Figure S1B). pHZ1904* was excised in circular form from S. coelicolor and introduced into non-methylating E. coli ET12567/pUZ8002. In interspecific matings pHZ1904* produced about 3610 4 apramycin resistant S. coelicolor M145 exconjugants, almost as many as pSET152 without cloned DNA (Figure 2, Panel 1).

Author Summary
Bacteria frequently exchange genetic information among themselves. DNA from one species can be transferred efficiently to unrelated microbes. Bacteria have developed systems that restrict gene transfer. Many restriction systems recognize and destroy foreign DNA entering the cells, but there are also enzymes inducing suicide of cells that have been invaded by foreign genes that modify the host DNA. We describe a restriction endonuclease from an antibiotic-producing soil bacterium that cuts foreign methylated DNA and also foreign DNA containing sulfur. DNA sulfur modification occurs in diverse medically or industrially important microbes and has been shown to prevent cleavage of DNA. The most similar enzyme in the databases is the putative restriction endonuclease McrA from Escherichia coli which has not been observed to cleave DNA in a test tube. Our endonuclease showed no activity with magnesium, but it cleaved DNA in the presence of manganese ions. Therefore, we present two novelties: an unusual restriction endonuclease that cleaves sulfur-modified DNA and conditions that allow the study of the enzyme in a test tube.
We speculated that S-modification of S. coelicolor DNA by the intact dndA-E gene cluster was not tolerated because of a hypothetical S. coelicolor endonuclease that restricts S-modified DNA. The above 51 Dnd 2 exconjugants would therefore all contain a mutant, inactive dndA-E gene cluster, and the 49 Smodified (Dnd + ) exconjugants may have kept the dndA-E gene cluster intact but may have lost the hypothetical S. coelicolor sulfurspecific REase.
We therefore set out to cure the integrated pSET152::dndA-E from one of the 49 apramycin resistant Dnd + S. coelicolor exconjugants and expected that the resulting strain would be a mutant that allows the reintroduction of fresh pSET152::dndA-E at high frequency.
Unfortunately, we could not detect spontaneously apramycin sensitive derivatives of S. coelicolor pSET152::dndA-E. Instead, we cloned the dndA-E cluster into the highly unstable autonomously replicating Streptomyces plasmid pJTU412 [22] to generate pJTU1651. Introducing pJTU1651 into S. coelicolor M145 by conjugation from E. coli produced thiostrepton resistant Dnd + exconjugants. One of these was randomly selected and plated on non-selective medium to allow loss of pJTU1651. LG3, one of the cured (thiostrepton sensitive, Dnd 2 ) S. coelicolor gave c. 5610 4 apramycin resistant exconjugants both with pHZ1904 and with pHZ1904* (Figure 2, Panel 2). As expected, the pHZ1904 exconjugants were Dnd + .
These results proved that the process of introducing the dndA-E gene cluster into wild-type S. coelicolor selected for rare mutant derivatives that no longer restricted the establishment of this gene cluster.
Identification of the endonuclease gene that prevented the establishment of the dndA-E-expressing plasmid Several putative endonucleases genes had been identified in the S. coelicolor genome sequence [12]. The identification of the correct one was aided by the availability of the complete genome sequence of Streptomyces avermitilis which contains dndA-E homologues and produces S-modified DNA like S. lividans [23]. We were thus searching for a putative endonucleases gene of S. coelicolor that does not have a counterpart in S. avermitilis and found Sco4631 which is encoded by the genomic island Gi11 (also known as SLP1) and known to be absent from S. lividans [23]. Sco4631 was shown to restrict methylated DNA in vivo in Streptomyces [9] and is listed in REBASE as ScoA3McrA because it has limited similarity (37% identity including a HNH endonucleases motif in a 91 aa overlap) to EcoKMcrA from E. coli which restricts methylated and hydroxymethylated DNA in vivo, but has not been demonstrated to cleave DNA in vitro [4].
The following two experiments tested whether Sco4631 (ScoA3McrA) was indeed the gene that prevented the establishment of pHZ1904 in S. coelicolor: single crossover gene disruption using an internal fragment of sco4631 on a suicide vector plasmid (thiostrepton resistance) was used to produce S. coelicolor LG4 ( Figure S2). Both pHZ1904 (dndA-E + ) and pHZ1904* (dndE -) were introduced by conjugation from E. coli ET12567/pUZ8002 with equal high frequency into strain LG4 (c. 10 4 apramycin-resistant exconjugants per plate, Figure 2, Panel 3).
In addition, sco4631 including the upstream promoter was cloned into the pSAM2-derived integrating vector pPM927, resulting in pJTU1654, and introduced into HXY16, a S. lividans derivative that does not S-modify its DNA because it lacks the entire SLG genomic island including dndA-E. The resulting strain, LG5, gave only c. 100 apramycin resistant exconjugants with pHZ1904 but c. 10 4 exconjugants with pHZ1904* (Figure 2 These results proved that Sco4631/ScoA3McrA without the need for any other gene on Gi11 (SLP1) restricted the establishment of pHZ1904 which confers S-modification on its host.
Introduction of dndA-E into S. coelicolor resulted in the loss of the entire SLP1 sequence Since Sco4631 is encoded by Gi11, which is also known as the 17.1 kb conjugative plasmid SLP1, we wondered whether the LG4, S. coelicolor mutant from which only sco4631 was deleted; LG5, S. lividans derivative which lacks the dnd gene cluster and contains a cloned copy of the S. coelicolor gene sco4631; LG6 S. lividans derivative which lacks the dnd gene cluster. The graph shows that pHZ1904 which contains the dndA-E gene cluster (causes DNA S-modification; black bars), is specifically restricted by the wild type S. coelicolor strain. pHZ1904* (white bars), which differs from pHZ1904 by an inactivating frame shift point mutation in dndE, was not restricted. The cloning vector pSET152 without insert (grey bar) served as an additional, not restricted control. doi:10.1371/journal.pgen.1001253.g002 above 49 Dnd + pHZ1904 exconjugants might have lost the entire SLP1 sequence. PCR amplification of LG3 and ten similar exconjugants using outside flanking primers showed the same band. Sequencing of the PCR product of LG3 showed the entire Gi11 sequence was excised precisely, restoring the tRNA Tyr sequence into which SLP1 had originally inserted ( Figure S3A).
This was again consistent with the hypothesis that Sco4631 was responsible for restricting the establishment of pHZ1904 in S. coelicolor, and it demonstrated that SLP1 can be lost spontaneously from S. coelicolor ( Figure S3A). This has not been observed before because SLP1 is a highly efficient conjugative plasmid that would immediately reinfect cured strains in the absence of the dndA-E genes.
Expression of sco4631 in S. lividans resulted in the loss of the genomic island that contains the dndA-E gene cluster pJTU1654 (pPM927 derivative expressing sco4631, see above) was introduced by conjugation into wild type S. lividans 1326 and into S. lividans HXY16 (lacks SLG including dndA-E). A high frequency of 3610 4 thiostrepton exconjugants were obtained per plate with S. lividans HXY16, and 300-fold fewer exconjugants were obtained with wild type (Dnd + ) S. lividans 1326. The control pPM927 without cloned DNA transformed both strains with equal frequency ( Figure S4).
Ten randomly selected, independent exconjugants were examined using PCR amplification and sequencing of the amplified product. All of them had suffered a precise deletion of the entire 93 kb S. lividans SLG genomic island that contains the dndA-E gene cluster ( Figure S3B). These results again support the hypothesis that sco4631 and the dnd gene cluster are unable to coexist in the same host.
Cloning and expression of sco4631 and its mutant derivative in E. coli The coding sequence of sco4631 including the stop codon was cloned into the expression vector pET28a, generating pJTU1655 for producing amino-terminally His6-tagged Sco4631. A similar plasmid, pSco4631H508A, was constructed that contains an inactive mutant protein because the first histidine residue of the conserved motif (H 508 -N 521 -H 529 ) was changed to alanine. The two plasmids were introduced into E. coli BL21 (DE3)/pLysE and His 6 -Sco4631 and its mutant derivative were overexpressed for 5 h at 30uC. About 70% percent of the overexpressed protein was soluble in the supernatants and 30% was in the precipitates ( Figure  S5). The soluble proteins were purified using a Ni affinity column and stored at 220uC in Tris-Cl buffer pH 8.0 containing 50% glycerol.

His-tagged Sco4631 cleaves Dcm methylated DNA near the site of methylation
In the course of these experiments we noticed that the pET28a derivative containing the cloned sco4631 gene could not be transformed into Dam + Dcm + E. coli DH10B but the plasmid was stably maintained in the Dcm -E. coli BL21(DE3)/pLysE expression host and also in the non-methylating E. coli ET12567 (see Table S1).
This gave us the idea to test the in vitro DNA cleaving activity using as a substrate dcm methylated (Cm5CWGG) EcoRV prelinearized pOJ260 DNA isolated from E. coli DH10B.
No DNA cleavage was observed using standard restriction NEB buffers 1-4 without and with BSA or ATP. Endonuclease activity was, however, observed when Mn 2+ or Co 2+ was included in the reaction buffer ( Figure 3). Optimal cleavage with minimal unspecific star activity was achieved at 30uC for 5 min using 20 mM Tris-Cl pH 9.0, 50 mM NaCl, 1 mM Mn 2+ .
Under optimal conditions, about 25% of 100 ng pOJ260 was cleaved and produced at least five bands of differing intensity, indicative of a partial digest ( Figure 4A). The precise position of eight cleavage sites was determined by blunt-end ligating the digested pOJ260 DNA to a DNA fragment containing the bla gene conferring ampicillin resistance (see Materials and Methods). The cleavage/ligation sites were sequenced using PCR primers SeqF and SeqR reading away from the ends of the bla fragment. Each of eight cleavage sites was between 12 and 16 bp away from lividans 1326 DNA (100 ng) prepared by the Kirby mix procedure [11] was used as a substrate to react with 500 ng purified His 6 -Sco4631 protein at 30uC for 5 min in a total reaction volume of 20 ml. The metal ions added to the reaction buffer are indicated to the left, and the concentrations are above the lanes. The reactions were stopped by adding loading buffer containing 1% SDS, 50% glycerol and 0.05% bromophenol blue (Takara). After the reaction, the DNA samples were examined by 0.75% agarose gel electrophoresis. The gel was run at 5 V/ cm for 30 min and visualized by staining with ethidium bromide. The strong bands are high molecular weight DNA; the smear observed with Mn 2+ and Co 2+ concentrations $100 mM indicates DNA cleavage by Sco4631. doi:10.1371/journal.pgen.1001253.g003 a C5mCWGG Dcm methylation site. pOJ260 has ten Dcm methylation sites and six of them had nearby cut sites ( Figure 4C).
These results were consistent with our expectation that Sco4631 is a Type IV restriction endonuclease that cleaves near rather than precisely at its methylated DNA recognition site. (Note that EcoKMcrA does not restrict Dcm-methylated DNA.) The reaction buffer used in the above experiment resembles buffers that induce star activity (reduction of sequence specificity) in many Type II REases [24,25]. It might thus be that not all Dcm methylated sites are cleaved in vivo by Sco4631. The fact that cutting near one of the sites was observed six times independently suggested that the DNA recognition sequence of Sco4631 may extend beyond the sequence C5mCWGG, or that Dcm methylation of the plasmid was incomplete as was observed by H. ZHOU (in preparation). Also, the in vivo results of Gonzalez-Ceron et al. [9] showed that Sco4631 restricts DNA containing other DNA methylations.
His-tagged Sco4631 protein binds and cleaves in vitro a synthetic oligonucleotide containing a single Dcmmethylated site The above results were consistent with the in vivo observations suggesting that Sco4631 is a methyl-directed REase, but we could not be absolutely sure that E. coli DH10B had not added an unknown, additional modification to pOJ260 that was necessary for target selection by Sco4631. To test whether indeed the Dcm methylation alone was sufficient for cleavage by Sco4631, we synthesized two 164 bp double stranded DNA fragment that contained centrally the above preferred presumptive methylated site, and a 59 32 P end label either at the top or the bottom strand so The ethidium bromide-stained agarose gel shows EcoRV linearized pOJ260 DNA treated as indicated above the lanes. Incubation was for 5 min at 30uC in buffer containing 1 mM Mn 2+ . H, heat inactivated His 6 -Sco4631 used as a control; E (Enzyme) and E*, His 6 -Sco4631 and inactive His 6 -Sco4631(H508A), respectively, purified from E. coli BL21(DE3); K, kb ladder DNA standard; M (methylated), indicates Dcm methylated pOJ260 that was prepared from Dam + Dcm + host E. coli DH10B. U (unmethylated), indicates pOJ260 that was isolated from nonmethylating E. coli ET12567. The 0.9-2.6 kb bands generated by Sco4631 digestion of methylated DNA are relatively faint, indicating partial digestion. B. Prolonged incubation with His 6 -Sco4631 causes non-specific DNA degradation, also of unmethylated DNA. EcoRV linearized pOJ260 DNA (like in panel A) was incubated for 5-60 min as indicated above the gel. M, methylated DNA; U, unmethylated DNA. C. Sequences of the cleavage sites including the nearby Dcm methylation sites (C5mCWGG, shown in blue). Vertical arrows point to the cleaved bonds, most cuts were observed once except for the cut marked by a larger arrow that was observed in four independent samples. Note, there are four additional C5mCWGG sequences in pOJ260 (not shown) where no cleavage was observed. doi:10.1371/journal.pgen.1001253.g004 that nicking of both strands could be observed. The labelled DNA fragments were used to test for binding of Sco4631 using DNase I footprinting, and for detecting endonucleolytic activity. Figure 5A and 5B show a staggered cut in the position where four independent clones showed DNA cleavage ( Figure 4C, third sequence). Overexposed gels showed two additional, faint bands corresponding to single strand nicks on the bottom strand only on the right side of the Dcm methylation site as it is shown in Figure 5B and 5C. All these bands were absent in the controls using an unmethylated form of the oligonucleotide or heat treated Sco4631.
There may be a trivial explanation for the lack of faint bands, indicating secondary cleavage sites in the upper strand: most the DNA molecules that were cleaved at one of the secondary sites in the upper strand will also be cleaved at the primary site. Because of the 59 end-labeling, secondary cleavage cannot be detected.
DNase I footprinting revealed protection of 14 nucleotides centred around the Dcm methylation site C5mCTGG on the top strand (light blue nucleotides in Figure 5C), and weaker, asymmetric protection of the bottom strand containing the sequence C5mCAGG. These results confirm that Sco4631 binds specifically to the Dcm methylated DNA and covers additional bases outside the C5mCWGG sequences.
His-tagged Sco4631 protein cleaves in vivo S-modified DNA S. lividans DNA is S-modified (phosphorothioated) at specific positions. Only about one in 6000 DNA backbone phosphates contain sulfur in a non-bridging position. S-modification of S. lividans DNA seems to be incomplete, and certain sequences are preferentially S-modified. The Streptomyces multicopy plasmid pHZ209 contains such a preferentially modified site which was chosen for our study.
S-modified pHZ209 was isolated from Dnd + S. lividans 1326 and either cleaved oxidatively using Tris-peracid or using His-tagged Sco4631 protein and the same buffer that was used for the cleavage of Dcm methylated DNA. After digestion using EcoRV, which cuts pHZ209 once, three prominent bands and additional fainter bands were observed in both the Tris-peracid and the Sco4631-treated S-modified pHZ209 samples ( Figure 6A). The largest fragment represents 5.4 kb linearised pHZ209 generated from plasmids that were not cleaved by Tris-peracid or Sco4631. The other two smaller bands were of the sizes expected if pHZ209 was cleaved at or near the preferential S-modification site. Controls with pHZ209 that was isolated from a dnd 2 host did not produce these two smaller bands.
The precise Sco4631 cleavage sites were determined by blunt end cloning and sequencing. Tris-peracid cleaves the DNA backbone precisely at the site of the sulfur producing a 1 bp 59 overhanging staggered cut [19,21]. The Sco4631-induced cuts were found on both sides between 16 and 28 nt away from the Smodification (dotted line arrows in Figure 6B). This was consistent with the expectation that Sco4631, like other Type IV REases, bound to the S-modification but cleaved some distance away from the cleavage site [26][27][28]. This endonucleolytic cleavage of DNA is, of course, consistent with our initial observations that Sco4631 and dndA-E are unable to coexist in the same host. Sco4631 cleaves DNA at multiple possible sites on either side of the S-modification Oxidative cleavage of S-modified DNA breaks specifically the phosphorothioate bonds resulting in a 1 bp 59 staggered cut. The above sequence data ( Figure 6) suggested that Sco4631 cuts Smodified DNA rather imprecisely on either side near the sequence CGpsGCCG. The fact that cleavage occurred on both sides of the sequence suggests that Sco4631 may bind to both strands equally, i.e. the essential binding region may not extend beyond this palindromic sequence.
We could not exclude the possibility that the DNA propagated in S. lividans 1326 contained additional modifications that influenced the results. For this reason, we synthesized a 118 bp double-stranded oligonucleotide containing one phosphorothioate on each strand at the preferred sequence of pHZ209. Again, 59 32 P label was used separately on both the top and bottom strand so that cuts in either strand could be observed. Oligonucleotides without S-modified bases were used as controls.  Figure 5A) from the His 6 -Sco4631-digested S-modified DNA were cloned into pBluescript SK+, and 16 clones were end-sequenced from both sides using primers T3 and T7. Vertical arrows show where the cuts occurred, and the numbers indicate how many times each particular cut was observed. All 16 cuts occurred near the preferential Dnd-phosphorothioation modification sequence CGpsGCCG shown in blue. doi:10.1371/journal.pgen.1001253.g006 His-tagged Sco4631-specific cleavage was observed in both the top and the bottom strand, and on both sides of the S-modification (Figure 7). (Note that no cleavage occurred in the controls or at the site of S-modification which is marked by ** in Figure 7A and 7B.) Multiple cleavage sites were observed (solid arrows in Figure 7A and 7B), consistent with the data from the cloning and sequencing of individual cleavage sites (dotted arrows).
Surprisingly, no concentration dependent protection by Sco4631 of the S-modified DNA against DNase I was observed ( Figure S6), possibly because the S-modified DNA was cleaved too efficiently. This result may not apply precisely to the majority of Smodified sites because this preferred site is a 6 bp palindrome rather than the more usual 4 bp core palindrome. Also, the Smodification in the synthetic oligonucleotides is racemic while the DndA-E proteins add the sulfur stereospecifically in the R P configuration [18].
These results, however, confirmed that His 6 -tagged Sco4631 could cleave S-modified without interference from hypothetical DNA modifications that might have been present on naturally S-modified pHZ209.

Activity of Sco4631
We demonstrated that the S. coelicolor protein Sco4631 is a Type IV REase that cleaves Dcm-methylated and also S-modified (phosphorothioated) DNA in vitro near the respective modification sites. DNA cleavage required a special reaction buffer containing Mn 2+ or Co 2+ at pH 9.0. These conditions reduce the sequence specificity of some Type II REases [24,25]. Also the N-terminal His 6 tag that was added to facilitate protein purification might have reduced or even altered the sequence specificity of the protein that was used for the in vitro studies. The fact that all the observed DNA cleavage events occurred very specifically near the respective modification sites suggests that neither the chosen buffer nor the His 6 tag changed the enzyme specificity. Two other Type IV REases also have special buffer requirements for in vitro activity: GmrSD, encoded by an E. coli prophage, required Ca 2+ and UTP for DNA cleavage [5], McrBC required GTP and Mg 2+ for DNA binding [29]. Recently, Chan et al. reported that HNH nucleases generally have unusual buffer requirements [30]. Only MspJI from Mycobacterium, a remote homologue of E. coli Mrr, cleaved methylated DNA in a standard REase buffer [31]. (Please note that BseMII and BspLU11III have been reclassified as Type IIG REases [4,32].The in vivo Dcm-methylated or S-modified DNAs might have contained unknown additional modifications that could have influenced the in vitro target selection. We excluded this possibility by repeating the cleavage experiments using long (164 and 118 bp) double-stranded synthetic oligonucleotides that had the same sequence and DNA modifications as the preferred cleavage sites identified on in vivo methylated or S-modified DNA. The results confirmed the initial data, indicating that there had been no unknown in vivo modifications influencing our results. In addition, using the end-labelled modified oligonucleotides provided better information about the relative frequencies of DNA cleavage in different positions ( Figure 5, Figure 7).
We have thus shown that a few Dcm-methylated sites and one preferentially S-modified DNA sequence were cleaved by Sco4631. Gonzalez-Ceron et al. demonstrated, however, that also other sequence specific 5mC methylations lead to in vivo restriction in Streptomyces strains expressing Sco4631 [9]. We cannot exclude that Sco4631 can cleave DNA with methylated bases or Smodifications in many different possible target sequences. The observation that one of the six available Dcm sites in pOJ260 was cleaved four times while others were cleaved once or not at all could be due to statistical fluctuation, but it could also indicate that the uncut sites may have been undermethylated or that the DNA sequence around the methylation may influence the frequency of cleavage. It is also not possible to speculate about the importance of the surrounding DNA sequence for the enzymatic cleavage near S-modification sites because only a minority of the potential sites receive the sulfur, and some preferred sites are more strongly phosphorothioated than others [33]. Preferential S-modification alone, without differential cutting of sites depending on the surrounding DNA sequence, could have produced the major and minor bands in Figure 7A.
Gonzalez-Ceron et al. observed that a S. lividans strain expressing cloned sco4631 restricted pSET152 DNA methylated by Dam or M. TaqI 400-fold and 20-fold, respectively, while Dcm methylated pSET152 was restricted only four fold [9]. It therefore seemed surprising at first that we could introduce pET28a::-HisSco4631 into the Dam + Dcm 2 strain BL21(DE3), but not into the Dam + Dcm + strain DH10B. Tighter regulation of the lac promoter in the latter strain may have caused the difference rather than the methylation status.
The purified His 6 -Sco4631 was produced in E. coli that did not contain any other Streptomyces genes. This proved that Sco4631 acted without accessory proteins as are required for endonucleases activity by the McrBC family of Type IV REases [5,29].
Sco4631 cleaved the Dcm-modified DNA preferentially to the left of the modification site on the sequence shown in Figure 5, indicating that the sequence flanking the 5mC modification may influence where cleavage occurs. The S-modified DNA was cleaved more evenly on both sides of the S-modification (Figure 7), but there was no evidence for simultaneous cleavage on both sites.
It was surprising that the distance between the DNA modification and the cleavage sites was different for Dcm and Smodified DNA. Maybe the enzyme remains locked to the Dcm site producing a near cut 12-16 bp away, but moves away from the Smodified site before cleaving less precisely c. 16-28 bp away.
The above speculation would be consistent with the observation that DNase I protection footprints, which were made under conditions that allowed DNA cleavage, were detectable on both strands of the Dcm-methylated oligonucleotides, but they were absent from the S-modified oligonucleotides ( Figure S6). Note that the protection on each strand was asymmetric with respect to the methylated C.
The natural S. lividans DNA S-modification is chiral (P-S Rconfiguration, [18]) but the S-modification on the synthetic oligonucleotides was racemic and thus only one in four doublestranded oligonucleotides had the natural R-R conformation. It is unknown whether this may have affected protein binding or cleavage activity. The larger distance between the S-modification and the cleavage sites is, however, not an artefact of racemic Smodification because the larger distances were observed both with the naturally S-modified DNA and the synthetic S-modified oligonucleotide (see Figure 7B dotted arrows and solid arrows, respectively, for comparison). Sco4631 had been identified by Gonzalez-Ceron et al. as a methyl-specific REase that has some limited (37% in a region of 91 aa) amino acid sequence similarity to the 5mC and 5hmC-specific EcoKMcrA [35,36]. The two enzymes share a consensus HNH protein sequence motif that occurs in homing endonucleases and inteins, and in Group I and Group II introns [37,38]. In REBASE [4] Sco4631 is listed as ScoA3McrA. It should be noted, however, that EcoKMcrA does not restrict Dcm methylated DNA [16].

Biological functions of Sco4631
Because in vitro cleavage by Type IV REases has been elusive for many years, it was speculated that these enzymes may achieve restriction of modified DNA without DNA cleavage that could easily be reversed by re-ligation. Simple binding to the target DNA might prevent its replication or integration into the host genome [39]. By adding a third example of in vitro DNA cleavage by a Type IV REase we support the original idea that Type IV REases cleave DNA like all the other types of REases.
S. coelicolor restricts Dam, Dcm, Hsd and other methylated DNA very effectively using at least four different methyl-specific REases including Sco4631 [9,40]. However, S. coelicolor only weakly restricts S-modified DNA from S. lividans [41]. Unexpectedly, the integrating vector pHZ1904 was severely restricted by S. coelicolor even though it did not have any known DNA modification because it had been propagated in the non-methylating E. coli host ET12567 that also did not support expression of the dndA-E DNA S-modification genes. Mutant derivatives of pHZ1904 were, however not restricted, excluding the possibility of an unknown sequence-specific (as opposed to modification-specific) restriction system acting in S. coelicolor.
It seemed therefore likely that expression of the dndA-E gene cluster in S. coelicolor resulted in S-modification of the host DNA. Sco4631 would then cleave near the modified sites resulting in cell death.
S. coelicolor mutants that tolerated S-modification by DndA-E represented about 50% of the rare pHZ1904 exconjugants. We speculated that one of the putative S. coelicolor endonucleases without counterpart in S. lividans might cleave S-modified DNA. Such a nuclease was indeed found to reside on the genomic island Gi11 [12]. Gi11, also known as SLP1 is a 17.1 kb mobile element that can excise from the S. coelicolor genome and transfer by conjugation to S. lividans where it usually forms autonomously replicating plasmids [14]. These plasmids in S. lividans always lack a part of the original Gi11 sequence including sco4631 which would destroy the S-modified S. lividans DNA.
Occasionally, the entire SLP1 sequence integrates into the S. lividans genome [42]. It is not known whether these integrated elements feature a mutant sco4631 or whether dndA-E is inactivated or deleted from these strains.
It seems that Sco4631 does not effectively restrict the entry of Smodified DNA into S. coelicolor [43], but it very effectively prevents the establishment of mobile elements that contain gene clusters like dndA-E that cause S-modification of its DNA. Such suicidal processes are frequently used by plasmids containing stable kil (kill) and unstable kor (kill override) genes resulting in cell death as a result of plasmid loss [44]. Also, restriction-modification (R-M) systems which are often on mobile elements [45] are stabilized in a bacterial population because the R-activity persists longer than the M-activity after the genes have been lost from a cell [46]. Fukuda et al. proposed that an important function of Type IV REases in general is the sacrificial killing of cells that have newly acquired a DNA modification system [39].
It was expected that strains containing a dnd gene cluster would not contain a Sco4631 homologue. This was true for 40 out of 41 fully sequenced strains containing a dnd gene cluster. Pseudomonas fluorescens Pf0-1, however, contains both a complete dnd gene cluster and S-modified DNA, and it also contains two HNH proteins that are, however, very different from Sco4631 ( Figure  S7A, S7B).
It is not understood why bacteria need such defences against DNA modification systems. Dam methylation reduces the mutation rate by directing the mutHLS pathway to correct errors in the newly synthesized DNA strand, and DNA adenine methylation also has known regulatory roles in bacteria [47][48][49], but the functions of Dcm methylation and phosphorothioation remain unknown.
Importance of Sco4631 and DndA-E for the genomic islands Gi11 of S. coelicolor and SLG of S. lividans SLP1 in S. coelicolor and SLG in S. lividans are segregationally very stable genomic islands [13,50]. Spontaneous excision of each element in the absence of selective pressure could, however, be observed using sensitive PCR analysis ( Figure S3B) [13].
Our experiments demonstrate that both SLP1 and SLG can be cleanly excised from the respective genomes by natural processes.
The sequence-specific S-modification of DNA prevents DNA cleavage by some Type II REases whose specificity overlaps the Smodification site (Liang J, personal communication), and Salmonella enterica serovar Cerro 87 which has S-modified DNA, also contains a restriction system that cleaves foreign DNA that lacks the cognate S-modification [51]. The fact that S. coelicolor has in Sco4631 an effective defence against invasion by these genes indicates that DNA S-modification may have important biological functions that remain to be discovered.

Materials and Methods
Strains, plasmids and primers used are listed in Table S1 and  Table S2, respectively. E. coli and Streptomyces strains were grown in LB [52] and SFM medium [11], respectively, supplemented with 100 mg/ml ampicilin, 30 mg/ml apramycin, or 12 mg/ml thiostrepton as required. Transformation of E. coli DH10B with Streptomyces plasmids was performed using the machine EasyjecT Plus electroporator (EQUIBIO). Conjugative transfer of DNA from E. coli ET12567/pUZ8002 to Streptomyces was performed as described in [11], except that always approximately 10 10 recipient spores were used, and tenfold dilutions of donor cells (10 9 to 10 5 CFU) were used per 9 cm agar plate. The exconjugants were counted from appropriate dilutions after incubation at 30uC for 3 days on SFM plates. The number of E. coli and Streptomyces CFUs in each experiment was determined by making dilutions and plating LB agar and MS plates, respectively. The conjugation frequencies were expressed as the number of exconjugants per 10 9 CFU of donors. The calcium chloride transformation frequency of E. coli was determined using 10 ng of CCC plasmid DNA. Each experiment was repeated three times and the final frequency was the mean value. Investigation of Dnd phenotype (degradation of S-modified DNA by Tris-peracid) was carried out as described in [21].

Inactivation and heterologous expression of sco4631
An internal region of sco4631 was amplified by PCR using primers S31DF and S31DR, gel purified and inserted into the EcoRV site of pBluescript prior to transformation into pSET151 to give pJTU1653, which was then transferred from ET12567/ pUZ8002 into S. coelicolor M145 by two-parental mating [11]. A 2176 bp fragment containing intact sco4631 was amplified using KOD Plus (TOYOBO) and primers S31HEF (XbaI linker) and S31HER (XbaI linker). The resulting fragment was cut by XbaI and inserted into pBluescript, and then the XbaI insertion was purified from this intermediate and cloned into the XbaI site of pPM927 to give pJTU1654. pJTU1654 and pPM927 were transformed individually into ET12567/pUZ8002 for conjugation into S. lividans strains 1326 and HXY16.
Overexpression and purification of His-tagged Sco4631 and its mutant sco4631 was amplified by PCR using KOD Plus and primers S31OEF and S31OER, inserted into pBluescript, excised as an NdeI-EcoRI fragment and ligated between cognate sites of pET28a, generating pJTU1655. Site directed mutagenesis of pJTU1655 was carried out by using KOD-Plus-Mutagenesis Kit (TOYOBO) with primers H508A-F and H508A-R, resulting in pSco4631H508A. The expression constructs were then transformed into BL21(DE3)/pLysE respectively. 10 ml of the overnight culture of BL21(DE3)/pLysE pJTU1655 or BL21(DE3)/pLysE pSco4631H508A was inoculated into 1 L LB medium supplied with 50 mg/ml kanamycin and 34 mg/ml chloramphenicol. Then the culture was incubated at 37uC to OD 600 = 0.4, cooled to room temperature and 0.4 mM IPTG was added, followed by incubation for another 5 hours at 30uC. Then the cells were harvested and resuspended in 20 ml binding buffer (20 mM Tris-Cl, 20 mM imidazole and 150 mM NaCl pH 8.0) and lysed by sonication in an ice bath. After centrifugation (16000 g for 30 min at 4uC), the supernatant was applied to a HisTrap HP column (GE Healthcare) and purified using an Ä KTA FPLC (GE Healthcare) by eluting with imidazole linear gradient 20-500 mM. The product was desalted by a HiTrap Desalting column (GE healthcare) and stored in Tris-Cl buffer (50 mM, pH 8.0) with 50% glycerol at 220uC. Purified Sco4631 was visualized by Coomassie-stained 12% SDS-PAGE analysis and protein concentration determined using a Bradford Protein Assay Kit (Bio-Rad). The protein purity was determined by Quantity One (Bio-Rad) from the gel, and the purities of Sco4631 and Sco4631(H508A) are about 96.5% and 97.9%, respectively.

Optimization of in vitro conditions for cleavage activity of Sco4631
Total S. lividans 1326 DNA prepared by the Kirby mix procedure was used as a substrate for the Sco4631 cleavage activity assay. Purified Sco4631 protein (500 ng) and 100 ng DNA substrate were incubated at 30uC for 5 min in a total reaction volume of 20 ml. The reaction buffer contained 20 mM Tris-Cl (pH 8.0), 50 mM NaCl and each of seven divalent ions, Ca 2+ , Co 2+ , Cu 2+ , Mg 2+ , Mn 2+ , Ni 2+ or Zn 2+ at a concentration of 1 nM to 10 mM. The reactions were stopped by adding loading buffer (Takara) containing 1% SDS, 50% glycerol and 0.05% bromophenol blue. The DNA after the reaction was examined by 0.75% agarose gel electrophoresis. Mn 2+ or Co 2+ ions at concentrations between 100 mM and 10 mM gave maximal DNA cleavage. The optimal pH was determined in a similar manner using equivalent buffers of pH 5 to pH 10, supplemented with 1 mM Mn 2+ ; pH 9 was optimal for Sco4631-mediated DNA cleavage.
End sequencing of fragments generated by Sco4631 cleavage of Dcm-methylated or Dnd-phosphorothioated DNA The plasmids pOJ260 and pHZ209 were isolated from DH10B and S. lividans 1326, respectively, using the QIAGEN Plasmid Midi Kit (Qiagen). One microgram pOJ260 DNA was incubated with Sco4631 under the optimized conditions described above. Linearised pOJ260 was gel purified and blunted using Klenow Fragment (Fermentas) and dNTPs at 37uC for 10 min in a total volume of 20 ml (16reaction buffer, 0.05 mM dNTP mix, 5 Unit Klenow Fragment), followed by heating at 75uC for 10 min to stop the reaction. The blunted mix was ligated to a 1311 bp amplicon harboring a bla cassette and the ligation mix was transformed into E. coli DH10B. Plasmid DNA was then prepared from randomly selected ampicillin-resistant transformants for insert end-sequencing using the primers seqF and seqR. Similarly, pHZ209 was linearized using EcoRV and then incubated with Sco4631. The resulting DNA fragments were gel purified, blunted using Klenow fragment, and ligated into the EcoRV site of pBluescript for transformation into E. coli DH10B. Plasmid DNA was then prepared from randomly selected transformants for insert endsequencing using the primers T3 and T7.
Generation of 59 end-labeled methylated or phosphorothioated DNA substrates for cleavage and DNase I footprinting analyses A 164 bp PCR amplicon was generated using pOJ260 as a template, and the long primers MF and MR that contained a m5C residue on each strand within the C m5 C(A/T)GG Dcm-modification motif. An unmodified matching amplicon was also generated using primers UMF and UMR. Individual strands of the modified and unmodified amplicons were 59 end-labeled by pre-tagging the appropriate primer with c-32 P using the following protocol. 10 pmol of primer was incubated with [c-32 P]-ATP (Beijing Furui Co. Ltd) using 10U of T4 polynucleotide kinase (Promega) at 37uC for 30 min, followed by 90uC for 2 min to inactivate the T4 polynucleotide kinase. PCR reactions were performed in a total volume of 50 ml containing 2 ng template DNA, 10 pmol of each primer, 16 PCR buffer, 5% DMSO, and 5U Taq polymerase (Genescript). PCR products were purified using the QIAquick PCR Purification Kit (Qiangen). An identical strategy was used to synthesize and strand-specifically label the phosphorothioatemodified and unmodified pHZ209-derived 118 bp amplicons. The phosphorothioate modification was introduced between the tandem guanine residues in the Dnd-motif (GpsGCC) by prior chemical synthesis of the long PCR primers SF and SR. The matching unmodified amplicon was generated using the primers USF and USR.

Cleavage sites analysis and DNase I footprinting assay
The DNA substrates used in these analyses were as described above. For cleavage sites analysis, the reaction mixture containing 100 cps 32 P-labeled DNA substrate (10 nM), 20 mM Tris-Cl (pH 9.0), 50 mM NaCl, 1 mM MnCl 2 , 5% glycerol and 500 ng Sco4631 in a total volume of 20 ml was incubated at 30uC for 5 min. For the DNase I footprinting assay, 500 cps 32 P-labeled DNA substrate (50 nM) was added to 0-24 mg (18 mM) of Sco4631 under the same buffer conditions as above, and the reaction mix was then incubated on ice for 5 min prior to adding 2.5 ml DNase I buffer and 0.3 U of DNase I (Promega), and further incubation at 37uC for 1 min. The reaction was stopped by adding 100 ml stop solution (3 M ammonium acetate, 0.25 M EDTA, 1 mg/ml glycogen), and 50 ml phenol-chloroform. Samples were then denatured at 95uC for 2 min and loaded on an 8% polyacrylamide-urea gel. The DNA sequence ladder was generated using an fmol DNA Cycle Sequencing kit (Promega). After electrophoresis, the gels were dried and exposed to a Kodak X-ray film.

Accession number
The nucleotide sequence of pOJ260 has been deposited in GenBank under Accession number GU270843. Figure S1 Half of the S. coelicolor exconjugants contain mutated dnd gene clusters. A. Ethidium bromide-stained agarose gel showing total genomic DNA of S. coelicolor exconjugants run under conditions that favor the Tris-peracid-mediated cleavage of DNA phosphorothioate bonds. The exconjugants in lanes 1, 6, 7, 8, 10 and 11 had stable DNA and thus exhibited the Dnd 2 phenotype associated with DNA without S-modification. The DNA in lanes 2, 3, 4, 5 and 9 was visibly degraded (characteristic DNA smear) and thus exhibited the Dnd + phenotype associated with S-modified (phosphorothioated) DNA. B. Organisation of the dndA-E gene cluster and the structure of mutant gene clusters from three randomly selected Dnd 2 (without S) exconjugants. The bold C denotes the single-nucleotide insertion in pHZ1904*. Another isolate contained a Tn10 insertion (acquired in E. coli) in dndC, and the third isolate had suffered a deletion removing dndB-E. Found at: doi:10.1371/journal.pgen.1001253.s001 (0.71 MB TIF) Figure S2 Construction of the sco4631 mutant strain. An 1438 bp internal fragment of sco4631 was generated by PCR amplification using primers S31DF and S31DR (red arrows). This fragment was inserted into the suicide vector pSET151. Introduction of this construct into S. coelicolor M145 and thiostrepton selection resulted in single crossover integration into sco4631. The resulting strain LG4 lacked a functional copy of sco4631. tsr, thiostrepton resistance gene for selection in Streptomyces. Found at: doi:10.1371/journal.pgen.1001253.s002 (0.10 MB TIF) Figure S3 Confirmation that the genomic islands Gi11 ( = SLP1, encoding sco4631 near attR) and SLG (containing dndA-E) were precisely deleted from the S. coelicolor and S. lividans genomes, respectively. A. Gi11 resides in a tRNA Tyr sequence (yellow box) in S. coelicolor. The insertion regenerated tRNA Tyr at attL and created a 112 bp direct repeat at attR (black triangles). The primers UF and DR from outside the direct repeat were used to detect samples from which the Gi11 sequence had been deleted. The ethidium bromide-stained agarose gel shows the characteristic 1497bp PCR fragment obtained with S. lividans 1326 DNA (lane 2) and ten S. coelicolor strains that expressed the cloned dndA-E gene cluster cloned on pHZ1904 (lanes 3-12). Lane 1 shows that excision of Gi11 was not detectable in wild-type S. coelicolor M145. B. The genomic island SLG of S. lividans is also flanked by 15 bp direct repeats (green triangles). The primers LP1F and Rp1R from outside the direct repeats were used to detect excision of SLG. The ethidium bromide-stained agarose gel to the right shows that the characteristic 725 bp band was obtained using S. coelicolor M145 DNA (land 1) and DNA from ten S. lividans derivatives expressing the S. coelicolor gene sco4631 (lanes [3][4][5][6][7][8][9][10][11][12]. No such band was observed with DNA from wild-type S. lividans 1326 (lane 2). Found at: doi:10.1371/journal.pgen.1001253.s003 (0.61 MB TIF) Figure S4 Transfer of sco4631 from E. coli to S. lividans strains. HXY16 is the S. lividans derivative which lacks the dnd gene cluster; pPM927-sco4631 (pJTU1654) is pPM927 harboring sco4631 with its native promoter. E. coli strain used for conjugation is ET12567::pUZ8002.  Figure S7 Phylogenetic analysis of Sco4631. A. Unrooted phylogenetic tree showing the extreme diversity of Sco4631 and its closest homologues. The 560 aa sequence of Sco4631 was used as a template to search REbase and the non-redundant protein database using BlastP for Type IV REases. The 59 top scoring sequences (E,10 25 ) were aligned using ClustalW. The scale represents an evolutionary distance of 0.1. Red, ScoA3M-crA = Sco4631 is most similar to the 481 aa conserved hypothetical protein SSEG03462 of Streptomyces sviceus ATCC 29083 (389/ 466 = 83% identity) a producer of many oxidative antibiotic tailoring enzymes. The next closest homologue is the putative HNH endonuclease SkeORF2910P from the Actinomycete Sanguibacter keddieii DSM 10542 which was isolated from bovine blood. Purple, EcoKMcrA from the e14 prophage in E. coli K-12, and from three other E. coli strains containing the e14 prophage. EcoKMcrA has an evolutionary distance .0.8 from ScoA3McrA, and only 34/91 aa identity in the HNH region. Green, two putative HNH endonucleases from Pseudomonas fluorescens Pf0-1 which contains a complete dnd gene cluster and S-modified (phosphorothioated) DNA. This strain was thus not expected to contain a protein that cleaves S-modified DNA. The evolutionary distance of these proteins is about 1, and the identity is only about 40% in a 65 aa region containing the HNH conserved sequence motif. One of the proteins (McrA2P) may be inactive because it lacks two highly conserved amino acids (GE) in the centre of the HNH motif (Hx13Nx8H; see Figure McrA