The phage growth limitation system in Streptomyces coelicolor A(3)2 is a toxin/antitoxin system, comprising enzymes with DNA methyltransferase, protein kinase and ATPase activity

The phage growth limitation system of Streptomyces coelicolor A3(2) is an unusual bacteriophage defence mechanism. Progeny ϕ C31 phage from an initial infection are thought to be modi ﬁ ed such that subsequent infections are attenuated in a Pgl þ host but normal in a Pgl (cid:2) strain. Earlier work identi ﬁ ed four genes required for phage resistance by Pgl. Here we demonstrate that Pgl is an elaborate and novel phage restriction system that, in part, comprises a toxin/antitoxin system where PglX, a DNA methyltransferase is toxic in the absence of a functional PglZ. In addition, the ATPase activity of PglY and a protein kinase activity in PglW are shown to be essential for phage resistance by Pgl. We conclude that on infection of a Pgl þ cell by bacteriophage ϕ C31, PglW transduces a signal, probably via phosphorylation, to other Pgl proteins resulting in the activation of the DNA methyltransferase, PglX and this leads to phage restriction.


Introduction
Bacteria have evolved a plethora of diverse mechanisms to evade killing by bacteriophages. The mechanisms can act at any stage of the phage life cycle from preventing phage adsorption right through to inhibition of cell lysis and release of progeny phage (Hoskisson and Smith, 2007). The phage growth limitation (Pgl) system of Streptomyces coelicolor confers resistance against the temperate bacteriophage ϕC31 and its homoimmune relatives (Chinenova et al., 1982;Laity et al., 1993). This system is called 'growth limitation' as phage infecting a Pgl þ strain for the first time undergoes a normal single burst to produce progeny phage but the progeny is attenuated for growth in a second round of infection. The progeny is however able to form normal plaques on a Pgl À host (Fig. 1). The mechanistic explanation of the Pgl phenotype proposed by Chinenova et al. (1982) is that during the initial round of infection the progeny phage are modified and then restricted in the second round of infection. This system differs fundamentally from typical R-M systems, where, if the phage DNA becomes modified, there can be an escape from restriction and rapid spread of infection through sensitive bacteria. In Pgl, however, even if modification fails during the first burst, it is likely that it will occur in subsequent cycles thereby severely limiting spread of infection ( Fig. 1; Sumby and Smith, 2002). Furthermore Pgl may confer an added advantage to a clonal population as being Pgl þ amplifies phage that might infect and kill phage-sensitive competitors.
Previous work has identified four genes, located in two operons (pglWX and pglYZ) 6 kbp apart, required for Pgl in S. coelicolor (Bedford et al., 1995;Laity et al., 1993;Sumby and Smith, 2002). Bioinformatic searches on the products of the pgl genes revealed several protein motifs (Fig. 2). The product of PglX is predicted to bind AdoMet and has an N 6 -adenine DNA methyltransferase motif, PglW is a putative serine/threonine protein kinase that contains a typical Hanks-like protein kinase domain, PglY possesses putative Walker A and Walker B motifs for the binding and hydrolysis of ATP/GTP and PglZ is predicted to have a conserved alkaline phosphatase-like fold (Pfam: PF08665). While the predicted function of PglX as a DNA methyltransferase fits well with the proposed mechanism by Chinenova et al. (1982) the Pgl system appears to involve additional activities that are novel to R-M systems, not least the putative kinase activity of PglW. Transcriptional analysis of pgl genes indicates that both operons are transcribed, even in the absence of phage infection (Bedford et al., 1995;Sumby and Smith, 2002). While transcriptional up-regulation of these genes is plausible during infection, we cannot rule out a role for the control of protein function through phosphorylation given that PglW is a predicted protein kinase. Indeed it has been shown that two isoforms of PglZ were detected with different isoelectric points in 2D-PAGE gels, suggesting that this protein could be post-translationally modified (Hesketh et al., 2002).
Another feature of the Pgl system is that it is subject to high frequency phase variation in which a Pgl þ strain gives rise to a Pgl À strain with a frequency of 10 À 2 to 10 À 3 and Pgl À to Pgl þ with a frequency of 10 À 3 to 10 À 4 (Chinenova et al., 1982;Laity et al., 1993). The phase variation has been attributed to a variation in length of a G tract in pglX (Sumby and Smith, 2003). Switching the phage resistance on and off may help to ease the strong selection within the phage population to mount a counter defence (Bayliss et al., 2006).
Here we set out to test the major bioinformatic predictions of Pgl protein functions. In the process we identified a toxin/antitoxin system comprising a toxic PglX protein, shown to be a DNA methyltransferase, and an antitoxin, PglZ. We also demonstrate that the protein kinase activity of PglW, and the Walker A motif of PglY are required for a functional Pgl system. We present a model as to how the Pgl proteins might confer phage resistance in S. coelicolor through an elaborate and novel R-M-like system.

Results
We set out to test whether mutations in the predicted functional motifs of the Pgl proteins were required for the phage defence phenotype (Fig. 2). The strategy used was to generate knock-out or null mutations in each pgl gene and then complement these null mutants with either wild type (wt) or mutant alleles introduced ectopically into the chromosome using the ϕC31 int/attP system (Bierman et al., 1992). Where possible we obtained further proof of Pgl protein function by heterologous expression of Pgl protein in Escherichia coli and biochemical assay of the partially purified extracts.

PglX is a DNA methyltransferase
Bioinformatic analysis of the PglX sequence suggested the presence of an N 6 -adenine methyltransferase motif (NPPY) at 378-381 aa (Roth et al., 1998;Scavetta et al., 2000;Fig. 2). A ΔpglX null mutant, SPHX, was constructed using REDIRECT technology that involves recombineering of the kanamycin-resistant cosmid SCIF2, replacing the pglX coding sequence (SCO6627) with an apramycin resistance gene. The apramycin resistance gene is flanked by loxP sites and also contains an origin of transfer (oriT)to enable conjugation of the cosmid into the Pgl þ S coelicolor strain, M145 (Gust et al., 2003;Redenbach et al., 1996). Double crossovers that retain the apramycin marker but had lost the kanamycin resistance marker were then infected with Cre-phage in a transient infection to remove the apramycin resistance marker (Khodakaramian et al., 2006). SPHX was sensitive to ϕC31 and could be complemented by the introduction of pPS8003 encoding a His-tagged version of PglX ( Fig. 3A; Sumby and Smith, 2002).
To test whether the putative methyltransferase domain of PglX was necessary for Pgl function, the tyrosine residue (Y381) of the conserved NPPY motif was targeted by site directed mutagenesis (Sumby and Smith, 2002). This tyrosine residue in other methyltransferases is structurally essential for catalysis where it is required for flipping out the target base from the DNA double helix prior to methylation (Roth et al., 1998). The resulting plasmid, pPH1002 (encoding PglX Y381A -His 6 ) was conjugated into SPHX (ΔpglX)totest in vivo activity of the mutated pglX allele. PglX Y381A did not restore phage resistance to SPHX suggesting that the putative methyltransferase motif is essential for the Pgl system (Fig. 3A).
To assay DNA methyltransferase activity of PglX, attempts were made to express a C-terminally His-tagged PglX fusion in E. coli and to enrich extracts by affinity chromatography. The 136 kDa PglX-His 6 from E. coli was barely detectable using a 6 Â His antibody in Western blots of the enriched proteins (data not shown). Nevertheless, low levels of methyltransferase activity using ϕC31 DNA as a substrate were observed ( Fig. 3B-D). In a time course of methyltransferase activity, incorporation of 3 Hmethyl groups into TCA precipitable material increase over the period of 60 min (Fig. 3C) and the level of incorporation was dependent on the amount of protein added (data not shown). An extract of PglX Y381A -His 6 was prepared in an identical procedure to the expression and enrichment of PglX-His 6 , but methyltransferase activity was almost undetectable (Fig. 3B), ruling out the possibility that the observed activity was due to endogenous E. coli enzymes. Further controls confirmed that incorporation of label into TCA precipitable material was dependent on DNA addition and could be competed by addition of unlabelled AdoMet (Fig. 3D).  Chinenova et al. (1982). The white-filled, phage-like particles represent unmodified phage, and the black-filled, phage-like particles represent modified phage. Pgl þ (black) and Pgl À (grey) host mycelia are represented as curved lines. These data strongly suggest that PglX can methylate DNA in an in vitro assay.

A pglZ null mutation is lethal in strains with a wild type pglX gene
In silico predictions on the function of PglZ are limited to a region (527-704 aa) annotated as a 'PglZ domain' (Pfam: PF08665), which falls within a family of proteins called the 'alkaline phosphatase clan' (Finn et al., 2010;Fig. 2). We attempted to create a knockout mutant of pglZ in M145. REDIRECT technology was used to replace the pglZ ORF in the cosmid SC4G2, with the apramycin marker generating a cosmid SC4G2:ΔpglZ::apra. When this cosmid was introduced into M145 by conjugation, an extremely low frequency of double recombinants (kanamycin sensitive, apramycin resistant) was obtained (Table 2). Three putative M145::ΔpglZ colonies were propagated and were found to be phage sensitive as expected. However, introduction of pglZ-His 6 encoded by the integrating plasmid, pPH1001, was unable to complement the phage sensitive phenotype of any of these recombinants (Fig. 4A). To demonstrate that pPH1001 was able to complement a pglZ À defective allele, the plasmid was introduced into J1934, a Pgl À strain constructed by Bedford et al. (1995) by insertional inactivation resulting in the deletion of the 3 0 end of pglZ encoding the Cterminal 130 amino acids (referred to as pglZ 1-834 ). The plasmid pPH1001 complemented the pglZ 1-834 allele J1934 to give phage resistance (Fig. 4A). These data suggest that the M145::ΔpglZ strains made by the REDIRECT approach had acquired a secondary mutation, possibly in one of the other pgl genes.
The most likely site for a second site mutation is the G-tract present within pglX that had been shown previously to inactivate Fig. 3. PglX methyltransferase activity is required for the Pgl phenotype. (A) SPHX (ΔpglX) was challenged with 3 Â 10 2 pfu/ml of ϕC31cΔ25, grown on J1929 (pglY À ) Bedford et al. (1995). Wild type Streptomyces coelicolor M145 (pglX þ ), SPHX pPS8003 (ΔpglX/pglX þ ) and SPHX pPH1002 (ΔpglX/pglX Y381A ) were challenged with 1 Â 10 5 pfu ml À 1 ϕC31cΔ25. (B) Methyltransferase activity of heterologously expressed PglX-His 6 (squares) and PglX Y381A -His 6 (circles) using 3 H-AdoMet as the methyl donor to ϕC31cΔ25 DNA isolated from J1929 (pglY À ) Bedford et al. (1995). Data are the mean of two replicate experiments. (C) Time dependent methyltransferase activity of PglX-His 6 on ϕC31cΔ25 DNA isolated from J1929 (pglY À ). Data are the mean of three replicate experiments and error bars represent the SD of the data. (D) Dependence of methyltransferase activity of PglX-His 6 on protein, ϕC31cΔ25 DNA and competition by addition of 10 nM unlabelled AdoMet to indicate specificity of activity (data are the mean of two replicate experiments).
Pgl (Sumby and Smith, 2003). Sequencing through the G-tract indicated that in two of the three M145::ΔpglZ strains the number of G residues had contracted (from 8 to 7 G nucleotides; Fig. 4B (i) and (ii)), while in the third strain the G-tract was as for the Pgl þ wild type M145 (8 G nucleotides; Fig. 4B (i) and (ii)), however this strain was also not complemented by pPH1001 indicating that the secondary mutation must be elsewhere in the genes required for Pgl. The provision (via integration of pPH1001) of a second copy of pglZ in the Pgl þ wild type strain M145 enabled the disruption of the native pglZ at a frequency of recombination that is typical when an inessential gene is targeted, such as pglW by the cosmid SCIF2:ΔpglW::apra (Table 2; Gust et al., 2003;Khodakaramian et al., 2006).
To confirm the essentiality of an intact pglZ,wefirst performed alignments of related PglZ proteins from the sequence databases to identify conserved residues that we could target for mutagenesis. Alignment of PglZ homologues indicated the presence of conserved residues D535 and D694. These residues fall within the predicted alkaline phosphatase fold (525-792 aa; Fig. 2; Quevillon et al., 2005). D535 and D694 in PglZ were both changed to alanine by site directed mutagenesis in the plasmid pPS5045 and the mutant alleles were subcloned into the integrating vector pIJ6902 (Huang et al., 2005) to generate pPH1007 (encoding PglZ D535A -His 6 ) and pPH1008 (encoding PglZ D694A -His 6 ). After conjugation of pPH1007 and pPH1008 into J1934, containing the pglZ 1-834 allele, the exconjugants failed to complement the Pgl À phenotype indicating that both pglZ D535A and pglZ D694A alleles were defective (Fig. 4C). Plasmid, pPH1008 (encoding PglZ D694A -His 6 ), was then conjugated into M145 and exconjugants were used as recipients for the SC4G2:ΔpglZ::apra cosmid, but the formation of the pglZ knockout strains was again prevented (Table 2). These experiments confirmed that the pglZ D694A allele was unable to confer antitoxin activity in the presence of an otherwise intact Pgl system.
These observations indicate that PglZ may be interacting with PglX to inhibit a toxic activity. To test whether mutations in other pgl genes would permit the disruption of pglZ, the cosmid, SC4G2: ΔpglZ::apra, was conjugated into strains SPHX (ΔpglX), SPHW (ΔpglW) and SLMY (ΔpglY)(Tables 1 and 2). Deletions of pglZ were obtained at low frequency, similar to that observed for the deletion of pglZ in a wild type background, except in SPHX (ΔpglX), where deletion of pglZ was obtained at the normal frequency ( Table 2). The G tracts in one SPHW::ΔpglZ and one SLMY::ΔpglZ strains were sequenced and they had suffered a contraction and an expansion, Fig. 4. Residues in the conserved PglZ domain are required for the Pgl phenotype. (A) Strains (indicated) were challenged with 1 Â 10 4 to 1 Â 10 5 pfu ml À 1 of ϕC31cΔ25 grown on J1929 (pglY À ). (B) Alignment of G-tract in pglX in M145, ΔpglZ null mutants indicating the expansion or contraction of the G-tract in these strains. DNA sequences starting at sequence coordinate 7353977 from (i) the reference parent strain M145 (Genbank accession: NC_003888), (ii) three independent ΔpglZ exconjugants from M145, (iii) a ΔpglZ exconjugant from SPHW, (iv) a ΔpglZ exconjugant from SLMY. (C) Mutagenesis of conserved residues in PglZ fail to complement the pglZ mutant J1934. Strains (indicated) were challenged with 1 Â 10 4 to 1 Â 10 5 pfu ml À 1 of ϕC31cΔ25 grown on J1929 (pglY À ). respectively ( Fig. 4B (iii) and (iv)). These data are indicative of a specific suppression of a toxic activity of PglX by PglZ. We hypothesise that the pglZ 1-834 allele in J1934 might retain the protective activity required against an intact pglX.

PglW has kinase activity
If PglZ prevents the toxic activity of PglX, then the presence of a signalling mechanism that regulates Pgl activity would enable tight control of the system and prevent unwanted toxicity. Bioinformatic analysis of the PglW sequence reveals the presence of two putative protein kinase domains; a tyrosine kinase domain at 195-490 aa (Prosite; PS00109) and a putative serine/threonine protein kinase (STPK, Hanks-type) domain (Prosite; PS00108) at 530-816 aa (Fig. 2). However only the putative STPK domain has a predicted ATP binding site that includes the central core of the catalytic loop and the invariant lysine. This residue (K677; Sumby and Smith, 2002)w a s targeted by asymmetric PCR mutagenesis. It has previously been shown that the equivalent residue to K677 is absolutely conserved in the ATP binding domain of such proteins, and it is believed to be essential for autophosphorylation and the phosphotransfer reaction (Hanks and Hunter, 1995;Hanks et al., 1988;Young et al., 2003).
The resulting plasmid pPH1012 (encoding PglW K677A -His 6 )was conjugated into SLMW (ΔpglW) to test whether the pglW K677A -His 6 allele could complement for the deletion of pglW and restore phage resistance. The PglW K677A -His 6 mutation resulted in a phage-sensitive phenotype, indicating that the putative Hankslike protein kinase domain is both functional and required for a Pgl þ phenotype (Fig. 5A). The gene encoding PglW-His 6 was cloned into the E. coli expression vector pT7-7 to generate pPS5012 and introduced into E. coli BL21 DE3(pLysS). No expression of PglW-His 6 was detected. Examination of the codon usage at the start of the pglW ORF showed the presence of several codons that are rarely used in E. coli. The first 19 codons were therefore optimised for expression in E. coli generating the expression plasmid pPS5025. The frequency of transformation of E. coli BL21 DE3 (pLysS) by pPS5025 was very low and expression trials of the few colonies that were obtained indicated the presence of insoluble and truncated proteins (data not shown). It seems likely that PglW-His 6 is toxic in E. coli BL21 and only plasmids that have suffered mutations can be established, explaining the poor transformation frequencies. At this stage we do not understand the basis for the toxicity of PglW although PglW has a putative, but as yet uncharacterised, Nterminal nuclease-related domain (NERD) motif (Fig. 2).
The plasmid pPS5025 was used as a template in an in vitro expression system to generate sufficient full length PglW-His 6 to test for autokinase activity (Molle et al., 2003). Incubation of His 6 -pglY in pT7-7 This work Table 2 Conjugation efficiency in various strains upon introduction of the PglZ deletion cosmid.

PglY has ATPase activity
Bioinformatic analysis of the PglY sequence suggested the presence of a Walker A (ATPase) motif (GSFGSGKS) at 75-82 aa (Bedford et al., 1995). It has previously been shown that these motifs are involved in nucleotide binding, and is found in many protein families (Karata et al., 1999;Ramakrishnan et al., 2002). We decided to test whether the ATPase motif is required for resistance to ϕC31 as a means to determine the role PglY might have in the Pgl phenotype.
The essential consecutive lysine and serine residues (K81, S82) were targeted by site directed mutagenesis in the plasmid pPS8008, which encodes PglY (Sumby and Smith, 2002). The resulting plasmid pPH1003 (encoding His 6 -PglY K81A/S82A )w a s conjugated into SLMY (ΔpglY). The K81A/S82A double substitution in PglY resulted in a phage-sensitive phenotype suggesting that ATP binding and/or hydrolysis is essential for conferring resistance to bacteriophage infection via the Pgl system (Fig. 6A).
N-terminally His-tagged PglY was expressed in E. coli and purified by nickel affinity chromatography. A single band was observed at $1 6 0k D a ;t h i sb a n dw a se x c i s e da n ds u b j e c t e dt op e p t i d em a s s fingerprinting by MALDI-TOF mass spectrometry and positively identified as PglY (top hit against the NCBInr protein database). The same approach was subsequently used for purifying His 6 -PglY K81A/S82A . The His-tagged wild-type and mutant PglY proteins were tested for their ability to bind and hydrolyse ATP. His 6 -PglY was found to hydrolyse ATP with a substrate affinity (K M )f o rA T Po f 0.5 mM, the K M of the mutant protein His 6 -PglY K81A/S82A was 200fold higher (100 mM; Fig. 6B). These data indicate that PglY is a functional ATPase. PglY nucleotidase activity was specificforATP ,and was found not to hydrolyse the other nucleotides tested (AMP, ADP, GMP, GDP, and GTP; data not shown). The non-metabolisable ATP analogue, ADP-NP was tested for its ability to inhibit the activity of PglY in vitro. The addition of equivalent molar amounts of ADP-NP and ATP, and subsequent 2-fold dilutions of ADP-NP in each assay resulted in a severely inhibited rate of hydrolysis of ATP, indicative of ADP-NP binding and inhibition of activity.
Taken together these data indicate an essential role for the ATPase activity of PglY in the Pgl system.

Discussion
Previous work describes the phenotype of Pgl in which an infection of S. coelicolor Pgl þ by phage ϕC31 undergoes a single burst but subsequent infections by progeny phage of S. coelicolor Pgl þ are attenuated (Bedford et al., 1995;Laity et al., 1993). A logical mechanistic explanation of these observations is that the phage is modified in the first infection but the second infection is restricted. Modified phage is able to proceed through a normal infection cycle in a Pgl À strain. In this work we demonstrated that PglX is indeed a DNA methyltransferase, as predicted by the bioinformatics searches. A strong genetic interaction between pglX and pglZ implies that PglX is toxic and that toxicity is suppressed in strains that are pglZ þ or contain the truncated pglZ 1-834 allele in J1934. This interaction resembles a toxin/antitoxin system. Many phage resistance mechanisms rely on some type of toxin/antitoxin pair to enable phage restriction and host immunity (Makarova et al., 2013). Examples include the restriction-modification systems where host protection from the endonuclease is usually conferred by DNA modification. However S. coelicolor is unusual as it is known to contain methyl-specific endonucleases (Gonzalez-Ceron et al., 2009;MacNeil, 1988). It is therefore feasible that it is the DNA methyltransferase activity of PglX that is toxic in S. coelicolor and the antitoxin, PglZ, inhibits this. Although we have shown that PglX has methyltransferase activity in vitro we were not able to demonstrate the presence of methylated DNA from progeny phage or from M145 (Pgl þ ) undergoing an infection using antibodies against DNA containing N 6 -methyladenine. Recently a phage defence system from Bacillus cereus called bacteriophage exclusion or BREX, has been described, that is mediated in part by a homologue of PglX (Goldfarb et al., 2014). In the BREX system the target for the methyltransferase encoded by the B. cereus PglX homologue was elucidated by PacBio sequencing. The target, TAGGAG (the underlined A is methylated), is modified in uninfected cells, but infecting phage DNA was not modified despite containing multiple target sites. In the BREX system phage replication is prevented in the first infection, apparently through cessation of phage DNA replication. Thus although BREX and the Pgl systems are mediated by a core of orthologous proteins (PglX, PglY and PglZ in S. coelicolor and PglX, BrxC and PglZ in B. cereus)it seems there are significant differences in their mechanisms of resistance. We hypothesise that Pgl is adapted to cause phage resistance in the context of a host in which there is no detectable DNA methylation (using antibodies to DNA containing either C 5methylcytosine or N 6 -methyladenine; data not shown) and in which there appears to be general methyl-specific restriction (Gonzalez-Ceron et al., 2009;MacNeil, 1988). In this context host DNA might be unmethylated during growth of uninfected host cells and the Pgl system would be inherently toxic unless the methyltransferase activity is highly regulated. We propose therefore that any DNA methylation occurs only during phage infection. The ATPase activity of PglY was also shown to be required for phage resistance, possibly implying the need for a motor to drive a processive activity, similar to Type I R-M systems.

A model for the mechanism of Pgl
The model proposed by Chinenova et al. (1982) suggested that Pgl would differ from all known methyl-specific restriction systems owing to the proposed marking of phage or phage DNA and flagging it for restriction in later rounds of infection. We propose a model as to how the activities of the Pgl proteins could mediate the Pgl phenotype in vivo (Fig. 7).
The model resembles a Type I R-Ms y s t e mi nw h i c ht h e modification and restriction activities are governed by the modification status of the DNA (Dryden et al., 2001;Murray, 2002). We propose that the Pgl proteins switch between three activity states: resting, modifying and restricting. In uninfected cells we propose that the Pgl proteins are in a 'resting complex' in which PglZ suppresses the toxic activities of PglX (Fig. 7). Evidence that the Pgl proteins are transcribed was obtained in previous work and PglZ was detected in   7. Proposed two step model for the Pgl mechanism by modification and then restriction. In uninfected cells Pgl proteins assemble into a 'resting complex' in which the toxic activity of PglX is inhibited by an interaction between PglX and PglZ. Following infection the resting complex changes (presumably triggered by a signal that is specific for phage ϕC31 and its relatives) to become either a DNA modifying complex or a restricting complex, dependent on whether the incoming phage is unmodified or modified, respectively. The modifying complex is likely to have N 6 -adenine methylation activity through the activity of PglX (blue). We propose that the restricting complex is activated by infection by modified phage. Candidate Pgl proteins that could mediate restriction are indicated in red. PglW P and PglZ P are putatively phosphorylated isoforms due to the kinase activity of PglW that transduce a signal to the cells of phage infection (more details are provided in the text). the S. coelicolor proteome (Bedford et al., 1995;Hesketh et al., 2002;Sumby and Smith, 2002). In the model infection by phage coming from a Pgl À host causes a change in the activities of Pgl proteins to m o d i f yp r o g e n yp h a g e ,m o s tl i k e l yb yN 6 -adenine methylation through the activity of PglX (Fig. 7). The putative N 6 -adenine methyltransferase activity could modify all the DNA in the infected cell or just targets in ϕC31 and its relatives. The trigger for switching between resting and modifying activities of the Pgl proteins might also be specifict oi n f e c t i o nb yϕC31 and its relatives. To avoid restriction of the modified phage DNA in the first infection cycle, we propose that Pgl proteins in the modifying state cannot switch directly to a restricting state or be reversed back to the resting state. We also propose that host methyl-specific restriction enzymes either do not recognise the modification in progeny phage or that the modification occurs late in the phage replication cycle. Thus in agreement with the observations of Chinenova et al. (1982),modified phage progeny are released in a normal burst. Infection of Pgl þ s t r a i n sw i t ht h i sm o d i fied phage triggers the activation of the restricting activity of Pgl (Fig. 7). The mechanism of restriction is not clear but could be mediated by the PglW NERD domain, the host cell methyl-specific restriction endonucleases or an unidentified motif in PglX that is responsible for conferring toxicity. If the latter is the case, PglX could resemble some Type IIS restriction/modification systems that are contained within a single polypeptide (e.g. BpmI and Eco57I; Pingoud et al., 2005). We propose that PglW and PglZ sense the phage infection and/or presence of modified or unmodified phage DNA and control the activity of PglX and other Pgl proteins. As PglW has kinase activity and as PglZ have been identified in two isoforms in a proteomics experiment, we propose that control of the Pgl activity state is dependent on phosphorylation ( Fig. 5; Bentley et al., 2002;Hesketh et al., 2002;Lomovskaya et al., 1980;Pingoud et al., 2005).
It is known that bacteriophage are an important force for driving bacterial evolution (Weinbauer and Rassoulzadegan, 2004), and the evolution of several types of phage resistance mechanisms, important for avoiding infection and lysis (R-M systems, Abi systems, CRISPR-Cas) suggests evading the deleterious effects of phage is highly selective (Makarova et al., 2013). Data presented here indicate that Pgl is a complex R-M-like system that demonstrates yet further biological novelty in the cellular mechanisms of defence against bacteriophages.

Bacterial strains, plasmids, growth conditions and conjugal transfer from E. coli to Streptomyces
The S. coelicolor strains used in this study are summarised in Table 1. All strains were cultivated on mannitol and soya flour (MS) agar (Hobbs et al., 1989). Plaque assays were performed as in Kieser et al. (2000) on Difco nutrient agar. Bacteriophage ϕC31 cΔ25 was used throughout this work as described in Kieser et al. (2000). Plasmids were conjugated into Streptomyces from the E. coli strain ET12567 (dam À dcm À hsdS) containing pUZ8002 to provide the transfer functions (MacNeil, 1988). Plasmids used are summarised in Table 2. Plasmids for expression of the His-tagged Pgl proteins in E. coli were made as follows starting with the PglX-His 6 expression plasmid, pPS5032: A 5772 bp BamHI fragment encoding PglX from pPS1001 (Sumby and Smith, 2002)w a s inserted into BamHI cut pARO191 (Parke, 1990) to generate pPS2002. An NdeI site was introduced at the pglX start codon by replacing a 482 bp HindIII-FseI fragment with a 408 bp PCR fragment generate using primers MMUTF and MMUTR and cut with HindIII and FseI to generate pPS3009. DNA encoding the His 6 tag was added to the 3 0 end of pglX by replacing a 46 bp AatII fragment with a PCR fragment cut with AatII generated from primers IF2 new and MENDSQ to form pPS5028. The ORF encoding PglX-His 6 was then cut out from pPS5028 with the NdeI and BamHI sites and the fragment was inserted into NdeI-BamHI cut pT7-7 to form pPS5032. The PglZ-His 6 expression plasmid, pPS5045 was made as follows: A 5269 bp SstI-BsiWI fragment from pPS5001 (pGem7 containing a 8166 bp NruI fragment encoding pglYZ from cosmid 4G2 inserted into the SmaI site; Sumby and Smith, 2002) was replaced with a 218 bp SstI-BsiWI fragment generated by PCR with primers ZNT and ZHisR to introduce an NdeI site at the start of pglZ, creating pPS5042. A HindIII site was then introduced just before the stop codon in pglZ by replacing a 282 bp SphI fragment with a PCR fragment from primers ZCHisF and ZCHisR, generating plasmid pPS5043. The NsiI-HindIII fragment was then inserted into pPSCHis fusing the 3 0 end of pglZ to DNA encoding an in frame His 6 -tag, forming pPS5045. (pPSCHis was constructed from pGEM7 by insertion of a 51 bp DNA fragment encoding a His 6 -tag flanked by HindIII and EcoRI sites). The PglW-His 6 expression plasmid, pPS5012, was constructed as follows: The 1 kbp XbaI-NotI fragment from pPS1001 (Sumby and Smith, 2002) was replaced with a PCR fragment generated using primers KtipF and KtipR resulting in an NdeI overlapping the start of the pglW ORF, generating pPS3005. A DNA fragment encoding the His 6 -tag was then added at the 3 0 end by replacing the BamHI-EcoRI fragment in pPS3005 with a PCR fragment made using primers KiHis/IRT3 and generating pPS5007. The NdeI-EcoRI fragment from pPS5007 was then subcloned into pT7-7 to generate pPS5012. pPS5025 containing the optimised codons at the start of the pglW ORF was created by replacing the NdeI-NotI fragment with a PCR product, generated using primers KCODON and KiseqR, also cut with NdeI-NotI. The His 6 -PglY expression plasmid was generated as follows: pPS5070, encoding an Nterminal His-tag fused to a truncated PglY, was constructed by inserting a PCR fragment, generated using primers YHisF and YHisR and then digested with AatII and SphI, into pGEM7. The entire pglY ORF frame was then reconstructed in pPS5071 by inserting the SphI-PvuII fragment from pPS5060 (Sumby and Smith, 2002) into SphI/SmaI-cut pPS5070. The His 6 -pglY gene was then inserted into pT7-7 using the NdeI-BamHI restriction sites. The pIJ6902-derived, pglZ-containing integrating vector, pPH1001, was constructed by subcloning the NdeI/EcoRI fragment from pPS5045. Primers for PCR reactions are listed in Table S1.

Disruption of Pgl genes and marker removal in Streptomyces
The pglX and pglZ null mutants were created according to the protocol of Gust et al. (2003) using the pIJ774 apramycin resistance cassette, flanked with loxP sites. In the double mutants (Table 1) the marker was removed as described (Khodakaramian et al., 2006) prior to disruption of the second target gene. Oligonucleotides for creating the disruptions are listed in Table S1.

Site-directed mutagenesis
Point mutants in the Pgl proteins were created using sitedirected mutagenesis, performed using the QuikChange XL sitedirected mutagenesis kit (Stratagene): primer sequences are detailed in Table S1. The exception to this protocol was in pglW, where the parental vector (pPS8002) was too large for Quickchange. The PglW K677A allele was produced by asymmetric PCR of a 900 bp region using the primers in Table S2.Afirst round of PCR created a product containing the mutant allele, with a flanking natural StuI site (pglW K677A F and pglW Asym FlankR), and a second PCR created the mutant allele with a flanking SrfI site (pglW K677A R and pglW Asym FlankF). The products were then mixed and a final round of PCR using the outer PCR primers (pglW Asym FlankF and pglW Asym FlankR) resulted in a final product of 900 bp representing a region of pglW with the desired mutation. This fragment was cloned into pGEM-T-Easy, and the mutation confirmed by sequencing. The resulting plasmid, pPH1004, was cut with SrfI and StuI, giving a 460 bp product, which was used to replace the natural fragment in pPS8002, creating the vector pPH1012 (pglW K677A).

Overproduction of Pgl proteins
Pgl proteins were overproduced as His 6 -tagged fusions. The plasmids were introduced in to E. coli Rosetta (DE3) (Novagen) and protein expression induced by the addition of 0.1 M IPTG (isopropyl-β-D-thiogalactopyranoside) in exponentially growing cells (optical density 0.5 at 600 nm). The resulting His-tagged proteins were purified by nickel affinity chromatography, and their identity confirmed by MALDI-TOF mass spectrometry.
The exception to this was PglW, which could not be produced in any DE3 lysogens. The protein was expressed from the plasmid pPS5025 using the EcoPro T7 coupled in vitro transcriptiontranslation system (Merck Biosciences), with the mass of the expressed protein confirmed by incorporation of 35 S-methionine in to the translation product according to the manufacturers instructions.

ATPase assays
The ability of wild-type and mutant PglY proteins to hydrolyse nucleotides was assayed at 30 1C in 40 mM HEPES Á HCl (pH 8.0), 10 mM MgCl 2 , 10 mM dithiothreitol, and 0.1 mg ml À 1 bovine serum albumin. Nucleotides and protein was added at the indicated concentrations. Reactions (10 ml) were pre-incubated at 30 1C for 10 min with PglY prior to the addition of the nucleotide. Reactions were stopped by the addition of 2.5 ml of stop buffer (100 mM Tris Á HCl (pH 7.5), 5% SDS, 200 mM EDTA, and 10 mg ml À 1 proteinase K) and incubation at 37 1C for 20 min. The hydrolysis of ATP was detected by measurement of the release of inorganic phosphate using acidic ammonium molybdate and malachite green according to the method of McGlynn et al. (2000).