P1 Ref Endonuclease: A Molecular Mechanism for Phage-Enhanced Antibiotic Lethality

Ref is an HNH superfamily endonuclease that only cleaves DNA to which RecA protein is bound. The enigmatic physiological function of this unusual enzyme is defined here. Lysogenization by bacteriophage P1 renders E. coli more sensitive to the DNA-damaging antibiotic ciprofloxacin, an example of a phenomenon termed phage-antibiotic synergy (PAS). The complementary effect of phage P1 is uniquely traced to the P1-encoded gene ref. Ref is a P1 function that amplifies the lytic cycle under conditions when the bacterial SOS response is induced due to DNA damage. The effect of Ref is multifaceted. DNA binding by Ref interferes with normal DNA metabolism, and the nuclease activity of Ref enhances genome degradation. Ref also inhibits cell division independently of the SOS response. Ref gene expression is toxic to E. coli in the absence of other P1 functions, both alone and in combination with antibiotics. The RecA proteins of human pathogens Neisseria gonorrhoeae and Staphylococcus aureus serve as cofactors for Ref-mediated DNA cleavage. Ref is especially toxic during the bacterial SOS response and the limited growth of stationary phase cultures, targeting aspects of bacterial physiology that are closely associated with the development of bacterial pathogen persistence.


Author Summary
The bacteriophage P1-encoded Ref protein is a novel class of endonuclease, cleaving only DNA to which the bacterial RecA protein is bound. The physiological role of such an enzyme has been entirely enigmatic. In this article, we define Ref as a P1 function that amplifies the lytic cycle under conditions when the bacterial SOS response is induced due to DNA damage. DNA-damaging antibiotics such as ciprofloxacin are much more lethal in bacteriophage P1 lysogens, and Ref is uniquely responsible for the complementation. In a phenomenon related to reported examples of phage-antibiotic synergy, Ref is highly toxic when expressed in bacteria in the absence of other P1 functions. Ref cleaves DNA in partnership with the RecA proteins from at least two prominent pathogens. The

Introduction
As multidrug resistance in important bacterial pathogens becomes an ever more serious health crisis [1], a need for new therapeutic strategies is evident [2][3][4][5]. An older approach, abandoned in the west in the 1940s but generating renewed interest, is phage therapy, as recently reviewed [6]. Phage therapy is slowly advancing as an approach to improve food safety, enhance water quality, provide an alternative to antibiotics in food-producing animals, facilitate environmental biocontrol of multidrug resistant pathogens on surfaces in hospitals, and treat wounds in humans [7]. Currently, phage therapy is approved in the United States only for control of pathogens in food sources and plants [8,9].
Phage therapy may prove useful, not just as an alternative to antibiotics, but also as a complement. Phage-antibiotic synergy or PAS-has been noted in multiple reports involving a variety of bacterial species [10][11][12][13][14][15]. Although the degrees of complementarity vary, the phenomenon does not typically reflect synergy in the genetic sense; i.e., where the effects of the phage and antibiotic are reliably greater than additive. The established definition of PAS is a phenomenon whereby sub-lethal concentrations of certain antibiotics can substantially stimulate the host bacteria's production of virulent phage [15]. The overall result is a complementation of antibiotic action by phage in reducing bacterial survival.
Whereas PAS, or in general a combination of antibiotic and phage therapy, represents a promising alternative therapeutic avenue for multidrug resistant pathogen infections, progress is constrained in part by a lack of understanding of the molecular mechanism(s) underlying the phage contribution. In this report, we explore the molecular basis of PAS using a classical phage system, the bacteriophage P1 of Escherichia coli. In this case, we show that the origin of an observed PAS phenomenon can be traced to the activity of a single P1-encoded gene, called ref for recombination enhancement function [16][17][18].
First described in 1951 [19], bacteriophage P1 remains a mainstay of bacterial molecular biology. Unlike many other temperate phage, the P1 genome (93.6 kbp;~117 genes) is not integrated into the host chromosome, but instead is maintained as a low copy number, autonomous plasmid [20]. As a lysogen, the C1 repressor prevents expression of P1 lytic genes [21]. When the host cell is placed under stress, the phage P1 lytic cycle is induced [22][23][24]. As defined in the current study, part of that cycle involves the P1 RecA-dependent nuclease Ref.
The Ref protein is an HNH-superfamily endonuclease (named for Histidine-Asparagine-Histidine, the three catalytic residues), with the novel property that it will only cleave DNA to which bacterial RecA protein is bound [25]. The Ref protein has a globular C-terminal domain that includes the nuclease active site [25]. The unstructured and highly charged N-terminal domain (76 amino acid residues) is required for single-strand and double-strand DNA binding [26]. Ref cleaves single-stranded DNA bound by RecA. It will also cleave both strands of the target duplex in DNA displacement loops (D-loops) that are created by RecA [25][26][27]. Both forms of RecA-bound DNA are found in bacterial cells, but become particularly numerous when the cell's DNA is damaged. The nuclease active site resides in the C-terminal domain and consists of three histidine residues that coordinate a Zn 2+ ion [25]. Changing one histidine (His 153) to alanine eliminates almost all of the nuclease activity [25,27]. Based on HNH nuclease precedent [28], we have recently constructed a mutant converting two Zn 2+ coordination residues (His 153 and His 134) to alanine to eliminate Zn 2+ binding and completely inactivate nuclease activity.
In spite of the recent advances in understanding Ref biochemistry in vitro, the in vivo role of the Ref endonuclease has not been addressed. Expression of Ref in E. coli enhances some classes of recombination events in a RecA-and RecBCD-dependent manner [16,17,29]. Deletion of the ref gene in bacteriophage P1 had no apparent effect on lysogenic or lytic cycles in early reports [17,18]. The RecA-dependent endonuclease function of Ref was unexpected [25], and the activity remains both unprecedented and perplexing. The current study began as an attempt to understand why an enzyme like Ref would evolve and be maintained by a bacteriophage.
In addition to its Ref-related activity, the bacterial RecA recombinase plays multiple roles in the maintenance of bacterial genome stability [30][31][32][33][34][35]. RecA forms nucleoprotein filaments on single-stranded DNA, and functions directly in all recombination processes via its DNA pairing and strand exchange activities [36][37][38][39][40][41]. RecA also plays a pivotal role in induction of the SOS response, by cleaving the bacterial LexA repressor in response to DNA damage. This results in the expression of a large number of genes that are directly or indirectly involved in DNA repair [42][43][44]. An example of an indirect effect on DNA repair can be seen in the SOSinducible gene sulA, the product of which interacts with FtsZ to prevent formation of the Zring and delay cell division until the genome is faithfully repaired [45,46]. Individual cells in a bacterial population that induce the SOS response or are otherwise in a slow growth state sometimes survive antibiotic treatment [47]. The SOS response has thus drawn attention as a potential therapeutic target in addressing multidrug resistance and its genesis [48][49][50][51][52][53][54].
There are no bacterial proteins more central to SOS induction than the RecA protein. This in turn provides a rationale for the evolution of a bacteriophage function that interacts with RecA. To enhance its lytic cycle under conditions of cellular stress, the bacteriophage P1 targets the SOS response with Ref: binding DNA, cleaving genomic DNA to which the RecA protein is bound, and causing cell filamentation. As we demonstrate here, these activities substantially enhance the lethality of antimicrobial agents that damage DNA and induce SOS. Ref is most toxic to bacteria in the precise situations where antibiotics are least toxic: during SOS and stationary phase; making it a potential candidate for combination phage/antibiotic therapeutic regimens.

Results
There were two goals to this study. We wished to determine the physiological function of the P1-encoded Ref endonuclease, and also wished to explore the effects of this enzyme when expressed in E. coli in the absence of other phage factors. We first explore an example of enhancement of antibiotic activity by bacteriophage P1. We then define the function of the ref In Escherichia coli, bacteriophage P1 lysogens enhance the toxicity of ciprofloxacin P1 lysogens having or lacking a functional ref gene, as well as non-lysogens were grown to log phase then treated with 8 ng/mL ciprofloxacin (Fig 1A). Immediately preceding treatment, and every two hours after for six hours, cells were removed from the culture, diluted, and plated. After an overnight incubation cells were counted to determine the number of viable E. coli cells Phage-antibiotic synergy with bacteriophage P1 and ciprofloxacin. A. Survival of P1 lysogens after treatment with 8 ng/mL ciprofloxacin. Log phase cultures of E. coli MG1655, EAR2 (P1ref+) or EAW195 (P1Δref), were adjusted to 5x10 5 CFU/mL and were treated with water (open markers) or 8 ng/mL ciprofloxacin (filled markers) with six hours outgrowth. The average CFU/mL and standard deviation of three biological replicates is reported (error bars are present, but smaller than markers in most cases). * = p-value<0.0001, when compared to MG1655 treated with 8 ng/mL ciprofloxacin. B. Optical density of P1 lysogens treated with ciprofloxacin. Log phase cultures of above strains were treated with water (open markers) or 4 ng/mL ciprofloxacin (filled markers) with outgrowth at 30°C. The average OD 595 and standard deviation of five biological and two technical replicates for each condition is reported. C. Optical density of P1 lysogens induced with a temperature shift. Log phase cultures of strains in (B) were shifted from 30°C to 42°C at each post-infection time point. Non-lysogens, ref + and Δref lysogens all grew similarly in the absence of ciprofloxacin (open markers). We next compared P1 lysogens and non-lysogens that were treated with ciprofloxacin, a DNA gyrase and topoisomerase IV inhibitor that creates double strand breaks in DNA by trapping covalent protein-DNA complexes during normal enzyme activity [55]. At 2 and 6 hours post-cipro treatment, there were significantly (p-value<0.0001) less WT P1 lysogens (ref + ) that survived the ciprofloxacin treatment than nonlysogens. The killing enhancement observed in the P1 lysogens was entirely dependent on the presence of an intact ref gene. When the ref gene was deleted from the lysogen, survival after ciprofloxacin treatment returned to non-lysogen levels.
E. coli P1 lysogens' response to ciprofloxacin treatment was also tracked by observing the optical density of the cultures over time in the presence and absence of a relatively low concentration of ciprofloxacin (4 ng/mL, 1/4-1/2x MIC) ( Fig 1B). Ref + P1 lysogens reached stationary phase at a lower cell density than Δref lysogens in the absence of ciprofloxacin treatment. When treated with ciprofloxacin, a portion of the P1 ref + lysogens appeared to lyse upon reaching stationary phase, while Δref lysogens entered stationary phase normally. Treating E. coli cultures with 8 ng/mL ciprofloxacin for more than four hours in the presence of the ref gene led to complex filamentation and lysis phenotypes (as detailed later). By treating with 4 ng/mL ciprofloxacin, we were able to moderately reduce the effects of the antibiotic and view a less severe phenotype more suited to basic observation by optical density changes.
Entry into the lytic cycle was enhanced in ref + lysogens only when the lytic cycle was induced using a DNA-damaging antibiotic. In the presence of a temperature shift, which cleaved the thermo-labile C1 repressor of P1Cm C1.100, the lysis profiles of both ref + and Δref phage were identical ( Fig 1C). The ref gene is a lytic function of bacteriophage P1 Next, we confirmed that ref gene expression occurs during the P1 lytic cycle. E. coli P1 lysogens (EAR2) harboring the thermo-labile phage C1 lytic repressor variant (C1.100) were grown to log phase at 30°C and then shifted to 42°C, inducing expression of all lytic cycle genes (Fig 2A  and 2B). Levels of ref transcript were measured using rt-qPCR every 10 minutes following the temperature shift until cultures were clear, indicating completion of lysis between 40 and 60 minutes after lytic induction (Fig 2A). Expression of ref peaked at 30 minutes following lytic induction, with a 73-fold increase in expression, when normalized to levels measured prior to lytic induction. Changes in culture density during this period were controlled for by using exactly 1 μg of total RNA from each isolate as a template for reverse transcription. To put the fold-induction of the ref gene during lytic induction into context, the maximum fold-induction of four other P1 genes under the control of the C1 repressor and four P1 genes not under the control of the C1 repressor were determined using rt-qPCR on the same total RNA samples ( Fig 2B). The maximum fold-induction of ref expression was higher than that observed for all of the genes not controlled by C1 repressor. It was also higher than the levels observed for ban, which is controlled by the C1 repressor. The expression of ref was lower and peaked slightly later after lytic induction than the C1 repressor-controlled genes ssb, dmt, and kilA. The induction of ref is thus within the range of induction seen for other C1 repressor-controlled genes during the lytic cycle.
to induce temperature-sensitive phage lysis and optical density was tracked as in (B). The average and standard deviation of three biological and three technical replicates is reported.

Ref expression is repressed by E. coli H-NS
Given the apparent stress that ref expression places on the E. coli host, we thought it possible that the ref gene might be repressed not only by the phage C1 repressor, but also by at least one E. coli host factor. The 5 0 regulatory region of ref contains a binding site for E. coli H-NS with sequence 5'-CGATAAA-3', a strong match to the H-NS consensus binding site 5'-tCGA-TAAATT-3' [57]. Purified H-NS bound the 5 0 regulatory region of the ref gene, as demonstrated with an electrophoretic mobility shift assay (EMSA) using a radiolabeled dsDNA fragment with the sequence of the 5 0 regulatory region of ref ( Fig 3A). The affinity is similar to a documented high-affinity binding site of H-NS (the downstream regulatory element (DRE) of bglG [58]) and is stronger than the affinity for randomly selected GC-rich DNA, suggesting H-NS binds specifically to the 5 0 regulatory region of ref.
H-NS is found in E. coli and other close relatives, such as Salmonella typhimurium [59]. It binds AT-rich DNA, which often includes virulence factors such as several of the pathogenicity islands in Salmonella [60,61]. The E. coli genome has a GC content of 50.8%, while the P1 genome is somewhat more AT-rich, with a GC content of 47.3% [20]. H-NS binding nucleates at high-affinity binding sites [57,62], leading to transcriptional silencing by interacting with RNA polymerase at promoter sequences (reviewed in [63]). In this way, E. coli can prevent the expression of horizontally acquired genes that may be detrimental to its survival, as tested here with ref.
In Fig 3B, binding of the 5 0 regulatory region of ref in P1 lysogens is demonstrated using digital droplet PCR (ddPCR) and chromatin immunoprecipitation (ChIP). Lysis of EAR2 was induced with a temperature shift. Cells were cross-linked using formaldehyde just before the temperature shift and 30 minutes after the temperature shift. Using an anti-H-NS antibody, the ChIP assay pulled down E. coli and P1 DNA to which H-NS was bound. The presence and amount of specific genes in the IP sample were quantified using ddPCR. The actively transcribed gene atpH was used as a negative control for background levels of H-NS binding, whereas bglB is a positive control known to bind high levels of H-NS [64]. Signals for ref were normalized to input DNA at the 30 min time points to account for increases in the amount of P1 phage DNA compared to E. coli DNA during P1 lytic replication. The ref gene is bound by H-NS more strongly than atpH P1 lysogens, but the binding is significantly reduced 30 minutes after lysis is induced. switch, C1 repressor controlled genes = red. Fold-induction data was obtained and calculated as in A. Values within 20% of the maximum fold-induction for each gene in each biological replicate were averaged to obtain the maximum fold induction time and value for each gene. C. Lysogens formed upon infection of MG1655 with P1Cm C1.100. MG1655 culture was supplemented with CaCl 2 and incubated with P1Cm C1.100 phage, followed by the addition of sodium citrate to chelate calcium ions and prevent superinfection. After recovery in LB, cells were serially diluted and plated on LB and LB/chloramphenicol (shown here) to select for phage lysogens. D. Colonies from plates in C were counted and the lysogenization efficiency (lysogens/viable cells) average and standard deviation from three biological replicate infections are reported. * = p-value<0.0001.

Expression of the N-terminal DNA binding domain of Ref triggers the E. coli SOS response
We next turned to an exploration of the physiological effects of ref gene expression. The endonuclease function of Ref [25] led us to hypothesize that ref expression might induce the bacterial SOS response. We used an SOS reporter plasmid encoding green fluorescent protein (GFP) under the early SOS-inducible recN promoter and transformed it into strains MG1655 (nonlysogen), EAR2 (P1ref + lysogen) and EAW195 (P1Δref lysogen). Thus, the intensity and duration of the SOS response can be monitored by measuring cellular fluorescence at the emission maximum for GFP [65]. Once reaching log phase, cells were treated with ciprofloxacin at 8 ng/ ml (1/2 the reported MIC for E. coli MG1655; see Methods). Growth was monitored by measuring optical density of the cultures at 600 nm and fluorescence was measured at 509 nm. Data was normalized as detailed in "Materials and Methods." To account for reduced growth rates in some strains, fluorescence was divided by optical density. Ref + lysogens show a higher level of expression from the early SOS recN promoter, when compared to a Δref lysogen or a non-lysogen ( Fig 4A). Normalized fluorescence/OD values cannot be directly compared between panels A and B because panel A utilized a high-copy reporter plasmid, while panel B utilized a low-copy plasmid to allow for a high copy Ref expression plasmid. These results indicated that Ref expression results in SOS induction. However, it was the highly charged and unstructured N-terminal DNA binding domain of Ref rather than the globular nuclease domain that was both necessary and sufficient to cause the SOS signal from the recN promoter.

Ref expression produces SOS-independent cell filamentation
We also imaged cells studied in Fig 4B to determine if the cell filamentation that generally accompanies the SOS response was occurring as a result of Ref expression (Fig 4C). Cells were grown to 1x10 8 CFU/mL and treated with 1% arabinose to express protein. As expected, the EV control and RefΔN76 expression did not produce filamentation, as they were not undergoing SOS. Cells expressing WT Ref protein and exhibiting high levels of SOS filamented extensively and were on average~3x as long as EV control cells (quantified in Fig 4D). DNA was present throughout the length of the filaments, indicating that DNA replication was continuing even though cell divison was not. However, the N-terminal domain alone (RefΔC110) was insufficient to produce the same degree of cell filamentation, even though it was the construct most effective in the induction of SOS. Ref nucexpression did not generate cell filamentation,  [67,68]. LexA3 is a non-cleavable variant of the LexA repressor of the SOS regulon [69,70]. An absence of RecA or the presence of LexA3 should block SOS induction. The absence of SulA would normally suppress filamentation during SOS.
Plasmids encoding the EV control, WT Ref and RefΔC110 expression were introduced into ΔrecA, sulA -, and lexA3 sulAbackgrounds. Log-phase cells (1x10 8 CFU/mL) were treated with arabinose to induce protein expression, allowed to continue growing for four hours, stained with DAPI, and imaged using brightfield and fluorescent illumination (Fig 5A). absence of phage P1. E. coli strains EAR86 (EV), EAR87 (pRef) EAR88 (pRefΔC110), EAR 120 (pRefΔN76), EAR121 (pRef nuc-) and EAR123 (pRbsR) harboring protein expression plasmids and an SOS reporter plasmid were grown to log-phase at 30°C. Cultures were treated with 1% arabinose to induce protein expression (time = 0). Fluorescence and optical density were measured as in panel A. Normalized fluorescence average and standard deviation of at least two biological and two technical replicates is reported for each condition. C. Microscopy of E. coli cells expressing Ref variants. Log phase cultures of EAR61 (EV), EAR62 (pRef), EAR73 (pRefΔC110), EAR98 (pRefΔN76), and EAR105 (pRef nuc-) at 1x10 8 CFU/mL were treated with 1% arabinose. Cells were outgrown for 4 hours and incubated with DAPI before imaging at 600x magnification using brightfield and fluorescence channels. Scale bar = 10 μm, representative images shown. D. Quantification of cell length data from (C) was obtained using the MicrobeTracker plugin for MatLab [66]. Each counted cell is represented by a single data point with the average and standard deviation for the data shown. At least 100 cells were measured for each condition, except WT Ref in which only 39 cells could be found. **** = p-value <0.0001.  Additionally, the E. coli protein RbsR was under the same promoter and conditions (black circles) as a control, and had no effect on cell growth or viability, indicating that cell death was not a general effect of high level protein expression. In a ΔrecA background, cells expressing Ref showed minimal growth or death, diverging from EV control cells between 6 and 8 hours after Ref induction and (unlike the EV control) never approached stationary phase.
The effect of expressing Ref in combination with various antibiotics was then examined ( Fig  6B-6D). Cultures of WT E. coli were grown to log phase as above and treated with 1% arabinose and various antibiotics (-drug data in Fig 6B is also shown in Fig 6A). Ciprofloxacin ( Fig  6B) and mitomycin C (Fig 6C) both function as antibiotics by creating DNA damage [71,72], and both are capable of inducing the SOS response [53,73]. Ref expression enhanced the antibiotic effectiveness of both of these antibiotics, decreasing survival by approximately 2-logs (red squares) compared to the EV controls (blue circles). Trimethoprim functions by inhibiting bacterial dihydrofolate reductase, a key enzyme in producing thymidine for DNA replication [74]. Although trimethoprim can somewhat induce the SOS response [75], it does so indirectly, as it does not produce DSBs de novo. E. coli sensitivity to trimethoprim was only moderately increased (less than 1 log) by the expression of Ref (Fig 6D).
Ref reduces the capacity of E. coli to recover from ciprofloxacin treatment. Strains from Fig  6B were allowed to continue growing overnight after treatment with arabinose with or without ciprofloxacin. Images of the cultures are shown in Fig 6E, with corresponding viability counts standard deviation for the data shown. An average of 90 cells (range: 39-188) were counted for each condition. ** = p-value <0.0001 when compared to EV in same background, * = p-value <0.05 when compared to same vector in WT background. C. Cell survival after expression of Ref. WT E. coli strains EAR61 (EV), EAR62 (pRef), EAR73 (pRefΔC110), ΔrecA strains EAR64 (EV), EAR65 (pRef), EAR74 (pRefΔC110) were treated as in (A). Cells were plated for viability and the average and standard deviation of at least three biological replicates for each condition are reported (error bars small and not visible in some cases). Significant p-values are noted in S1 Table. doi:10.1371/journal.pgen.1005797.g005 , and ΔrecA strains EAR64 (EV) and EAR65 (pRef) were grown to log-phase, adjusted to 5x10 5 CFU/mL, treated with 1% arabinose, and outgrown for one day at 30°C. Cells were plated and the average and standard deviation of CFU/mL for at least three biological replicates for each condition are reported. Significant pvalues can be found in S1 Table. B. WT E. coli strains EAR61 (EV) and EAR62 (pRef) were grown as described in (A), but with the addition of water or 8 ng/ in Fig 6F and brightfield and DAPI-stained fluorescent microscopy images in Fig 6G. Cells harboring the EV control in the absence of ciprofloxacin continued to grow and maintain normal viability and cell morphology (Figs 6E-6G). In the presence of ciprofloxacin, these same cells recovered viability overnight, although a large number of dead cells are seen in the microscopy images (indicated by cells where DAPI fluorescence saturated the camera, as dead cells are more permeable to DAPI.). Cells expressing Ref in the absence of ciprofloxacin did not recover viability overnight (approx. 4 log reduction in viability, compared to EV control, Fig 6F), are filamentous, and large numbers of dead cells with high DAPI permeability are seen (Fig 6G).  [26]. Expression of these 76 amino acids on their own (Ref ΔC110) results in induction of the SOS response (Fig 4) and decreased growth rates (Fig 6). However, cells recover from Ref ΔC110 expression alone if given sufficient time. The C-terminal domain (RefΔN76) has little effect on its own. Expression of the complete Ref protein confers considerable toxicity in E. coli, especially when expressed in the presence of moderate doses of ciprofloxacin. Expression of Ref results in substantial cell filamentation even in genetic backgrounds that block SOS induction ( Fig 5). As the filamentation occurs even in the absence of RecA, we hypothesize that Ref may inhibit cell division via a mechanism unrelated to its nuclease function.
We propose a model of Ref action during the bacteriophage P1 lytic cycle, in which Ref converts the E. coli SOS response to a bacterial liability. The strategy is employed when DNA damage occurs and the host cell (and thus the lysogenic phage) is in jeopardy (Fig 8). When E. coli lysogens are subjected to DNA damage, RecA filaments form on exposed single-stranded DNA. This triggers the induction of the bacterial SOS response, which in turn will block cell division and induce the phage lytic cycle [76][77][78][79]  independently of the SOS response through a yet-undiscovered mechanism. How each of these functions could benefit the phage is explored in more detail below.
Ref expression occurred during the lytic cycle (Fig 2A), as expected due to the presence of a C1 repressor binding site occluding one of the ref promoters [20]. The ref induction levels during the lytic cycle fell within a range seen from other genes with C1 repressor binding sites ( Fig  2B). Conversely, the presence of ref inhibited the lysogenic cycle of the phage (Fig 2C and 2D). These results, combined with previous research showing that ref activity from its native promoter was noticeably higher in cells with defective or absent C1 repressors [16,17] There are several possible ways the phage can benefit from ensuring that the host cell remains in SOS. First, a binding site for the SOS repressor LexA occurs in the bacteriophage P1 genome, overlapping the promoter for P1 coi [20]. During SOS, the Coi protein is expressed and binds to P1 C1 repressor, preventing it from repressing lytic cycle genes [24]. Lytic replication is the only way for a phage to survive if the host is unable to repair DNA damage via the SOS response, making this coupling of SOS and lytic replication crucial for phage survival. Second, the main subunit of E. coli DNA polymerase III (dnaE) is essential for the lytic replication of bacteriophage P1 [80]. SOS promotes the use of the Y-family error-prone polymerases Pol IV and Pol V for host replication at sites of damage, freeing up Pol III for phage replication [81][82][83].
The DNA binding activity of Ref is centered in the unstructured N-terminal domain [26]. Ref binding of DNA appears to play a role in all phenotypes we observe with Ref expression, as RefΔN76 does not induce SOS, cause cell filamentation, or lead to cell death (Figs 4B, 4C and  6A). DNA binding proteins can stall replication forks, creating gaps and breaks in the DNA [84]. Most replisome stalling is due to interactions with DNA-protein complexes [85]. Replication restart pathways are needed, and if the fork is not reloaded DNA damage will persist [86].
Actively transcribing RNA polymerase complexes must also contend with DNA-bound proteins and generally pause when encountering other DNA-protein complexes [87]. Finally, Ref-DNA complexes present a barrier to RecA-mediated DNA repair, a function required for bacterial survival in the presence of DNA-damaging agents. DNA binding activity could explain why Ref is still toxic in the absence of recA, even though RecA is required to create a substrate for Ref endonuclease activity [25].
The phage genome would not be immune from the DNA binding effects of Ref, but the presence of many copies of the phage genome during the lytic cycle would mitigate any effects. A single instance of Ref binding to the E. coli genome that results in unrepaired DNA damage will lead to death of the host. To contrast, a single instance of Ref binding to the phage genome that results in unrepaired DNA damage will reduce the burst size by a single phage.
Ref cleaves RecA-bound DNA in vitro [25][26][27]. It is likely that Ref's RecA-dependent DNA cleavage activities elevated DNA damage in P1-infected cells in the original studies on Ref [16][17][18]29]. Ref only affected lysis of P1-infected cells in the presence of DNA damage, and had no effect when lysis was induced using the temperature sensitive C1 repressor (Fig 1B and 1C).  [88,89]. While one might expect a nuclease like Ref to inhibit cell division by cleaving the DNA being segregated to daughter cells, the filamentation observed here was independent of nuclease activity (Fig 4C  and 4D) and the SOS response (Fig 5). We are currently investigating potential interactions between Ref and the divisome in more detail. At this time, it is unclear exactly what benefit delaying cell division confers to the phage. Others have suggested that phage inhibition of cell division will increase host adaptations to stress [90] or prevent compartmentalization of the host that would interfere with phage replication or packaging [88]. Either explanation could fit our data.
Ciprofloxacin treatment of E. coli lysogens may be an example of phage-antibiotic synergy (PAS) [10], with our data supporting a model not yet seen in other cases of PAS. Most antibiotics are effective only if cells are in active metabolism, and bacteria can persist through antibiotic treatment if they enter stationary phase [47]. Ref expression is clearly most toxic as cells approach stationary phase (Fig 6A). Thus, the toxicity of Ref complements the toxicity of most antibiotics. Ref nuclease activity and DNA binding contribute to this activity. Since the identity of Ref as an HNH endonuclease family member was not originally evident from its gene sequence [25], it is tempting to speculate that other families of phage-encoded endonucleases with a function similar to Ref may exist.
There is potential for more general use of Ref in therapeutic applications of phage. The use of toxic phage proteins of various classes as alternatives or complements to antibiotics has been examined previously, but Ref is not a member of any of these classes [91]. Ref will utilize RecA proteins from at least two prominent pathogens (Neisseria gonnorhoeae and Staphylococcus aureus) as cleavage partners. Since no other P1 functions are required for its synergistic effect with antibiotics, the ref gene is portable. In principle, ref could be cloned into any of a wide range of phages that are being tested for use in agriculture and medicine. We propose that any antibiotic that induces the bacterial SOS response may in principle be rendered more effective in the presence of a complementing phage expressing the ref gene as a cloned adjuvant.

Reagents/media/growth conditions
For liquid growth, cells were grown in Luria-Bertani (LB) broth at 37°C with 250 rpm shaking, unless otherwise noted. P1 lysogens were always grown at 30°C to prevent the induction of the thermo-inducible CI repressor, which would cause phage to enter the lytic cycle. For growth on solid medium, cells were plated on LB plus 1.5% (wt/vol) agar. M9 minimal media was prepared using a standard recipe [92]. Chloramphenicol (MP Biomedicals) was dissolved in 95% ethanol to make a 25 mg/mL stock solution and was used at 25 μg/mL for all selections. Ampicillin (Sigma) was dissolved in water to make a 100 mg/mL stock and was used at 100 μg/mL for all selections. Ciprofloxacin was dissolved in water at 5 mg/mL. Mitomycin C was dissolved in water at 0.5 mg/mL and stored protected form light. Trimethoprim was dissolved in DMSO at 5 mg/mL. All antibiotics were serially diluted in water to obtain working concentrations. Larabinose was dissolved in water at 20% (w/v) and used at a working concentration of 1% (w/ v) to induce protein expression from pBAD-myc-HisA-derived plasmids.

Preparation of phage from lysogens
Phage stocks (listed in Table 1) were prepared from E. coli lysogens using a modified plate lysis method. Overnight cultures of lysogen (100 μL) were mixed with 3 mL melted and cooled top agarose (LB+0.7% agarose), and 15 mM CaCl 2 and poured on top of a room-temperature LB plate. The plate was incubated right-side up at 42°C overnight to allow phage to lyse the cells. The next day, 5 mL SM buffer (5.8 g NaCl, 2 g MgSO 4 7 H 2 O, 50 mL 1 M Tris-Cl pH 7.5, 5 mL 2% gelatin in 1 L H 2 O) was added to each plate and rocked at 4°C for 2 hours. The SM buffer (with phage) was transferred to a sterile glass test tube and a few drops of chloroform were added to kill any remaining E. coli. The tube was spun down to pellet cell debris, and the supernatant was removed and stored in a fresh, sterile glass test tube at 4°C. Phages were titered using standard methods [92] to ensure similar PFU/mL (within 1 log) were obtained.

Plasmid construction
Plasmid pEAW903 is Super-Glo GFP under the control of the E.coli recN promoter. The 200 bp upstream of the E. coli recN gene were inserted into pQBI63 (Qbiogene), replacing the T7 promoter. PCR was used to generate the 200bp upstream of recN gene with E. coli MG1655 DNA as template, and primers consisting of a BglII site followed by MG1655 bases 2749617-  2749637, and a NheI site followed by a NdeI site (bases 5326-5315 of pQBI63) and bases 2749813-2749793 of MG1655. The PCR product was digested with BglII and NheI and ligated into pQBI63 digested with the same enzymes. Plasmid pEAW915 was functionally similar, but contained the recN promoter fused to Super-Glo GFP in the pACYC184 background, allowing for plasmid selection with chloramphenicol [94]. Plasmid pEAW427 (RbsR in pBAD) was constructed by using E. coli MG1655 DNA as the template in a PCR with one primer consisting of a NcoI site followed by bases 4-23 of the rbsR gene, and the other primer consisting of an EcoRI site followed by bases 1243-1226 of the rbsR gene. The PCR product was digested with NcoI and EcoRI and was inserted into pBAD-Myc-His A digested with the same enzymes.

Strain construction
All strains are derivatives of E. coli K-12 and are listed in Table 1. EAR2 was constructed by lysogenizing MG1655 with P1Cm C1.100 produced from strain BR4666. Briefly: 100 μL of P1Cm C1.100 produced from BR4666 was incubated with 200 μL MG1655 overnight culture and 2 μL 1 M CaCl 2 (5 mM final concentration, necessary for absorption of phage) for 30 minutes at 30°C (to prevent induction of temperature-sensitive phage) without shaking. Then, 1 mL LB and 100 μL 1 M sodium citrate were added to prevent superinfections of the bacteria and the culture was allowed to recover at 30°C for 30 minutes with agitation. Cells were pelleted, resuspended in 100 μL LB, and plated on a LB plate with chloramphenicol for selection of lysogens. Plates were grown at 30°C overnight.  [95]. pEAW507, a plasmid containing a mutant FRT-KanR-wt FRT cassette, was used as a template in a PCR. The primers consisted of the 51 bases before the promoter of the ref gene +21 bases before one FRT, and the 51 bases after the stop of ref +21 bases after the other FRT. The PCR product was electroporated into EAR2/pKD46, and a kanamycin resistant colony was selected. DNA from this colony, designated EAW195, was used as a template in a PCR to confirm the presence of the FRT-KanR-FRT replacing the ref gene on the lysogen. EAW20 is MG1655ΔrecA and was constructed using a variation of the method in [95]. A plasmid containing the 200 bases upstream of the start of the recA gene, followed by a FRT-KanR-FRT cassette, and the region starting 6 bases downstream of the stop of the recA gene through the recX gene, was constructed. This plasmid, designated pEAW413, was used as a template in a PCR with an upstream primer consisting of bases 84-64 before the start of the recA gene, and a downstream primer consisting of the 10 bases after the stop of the recX gene, and the last 8 bases of the recX gene. The PCR product was electroporated into MG1655 containing the plasmid pKD46. A kanamycin resistant colony was used as template in a PCR, and the product was sequenced to confirm the presence of the FRT-KanR-FRT cassette replacing the recA gene.
Strains EAR15, 16, and 17 were constructed using standard methods by transforming plasmid pEAW903 into chemically competent MG1655, EAR2, and EAW195, respectively and selecting for ampicillin-resistant colonies.
Strains EAR86, EAR87, EAR88, EAR120, EAR121 and EAR123 were constructed using standard methods by co-transforming chemically competent MG1655 with the plasmids indicated in Table 1 and selecting for colonies both ampicillin and chloramphenicol resistantance.
Survival curves with drug or arabinose added at 5x10 5 CFU/mL Drug concentrations used were all at the MIC as determined experimentally using standard methods [97] with E. coli MG1655: ciprofloxacin used at 8 ng/mL, mitomycin C used at 5 μg/ mL, trimethoprim used at 0.5 μg/mL. Independent colonies were grown overnight at 30°C. Cultures were diluted 1:100 in fresh LB and grown at 30°C until OD 600 was at least 0.35. Cultures were adjusted to OD 600 = 0.35 with LB (approximately 1x10 8 CFU/mL), then diluted 1:200 in LB, to give a density of approximately 5x10 5 CFU/mL. Each culture was split in two, and drug was added to a final concentration as indicated to one of the aliquots. For experiments testing the effect of Ref expression from a plasmid, 1% L-arabinose was also added to all cultures (including empty vector controls). Immediately after addition of arabinose and/or drug, samples were removed and serially diluted 1:10, plated on LB plates supplemented with appropriate selection drugs, and grown overnight at 30°C. The rest of the culture was incubated at 30°C overnight, with samples withdrawn periodically, diluted, and plated as before. Reported data are the average and standard deviation of at least three biological replicates.

Cell lysis by phage (induced with ciprofloxacin)
Independent colonies of strains EAR2 and EAW195 were picked from freshly streaked plates and grown overnight at 30°C. Cultures were diluted 1:100 in fresh LB and grown at 30°C until OD 600 was near 0.4. Cultures were adjusted to OD 600 = 0.3 with LB and 200 μL of each culture was pipetted into two individual wells of a clear 96 well plate, and 10 μL ciprofloxacin was added to a final concentration of 4 ng/mL. A prewarmed plate reader was used to incubate at 30°C and shake while the OD at 595 nm was read every 10 minutes. Reported data is the average and standard deviation of five biological and two technical replicates.

Cell lysis by phage (induced with a temperature shift)
Strains EAR2 and EAW195 were diluted 1:100 in fresh LB and grown at 37°C until OD 600 was no more than 0.4. Then, 100 μL of each culture was pipetted into three separate wells on a clear 96 well plate and 100 μL of prewarmed LB (50°C) was added to each well and the plate was immediately placed in a prewarmed (42°C) plate reader. The plate was shaken and incubated at 42°C and the OD at 595 nm was read every two minutes. Reported data is the average and standard deviation of three biological and three technical replicates.

rt-qPCR
Three independent overnights of EAR2 were grown at 30°C. Overnight cultures were diluted 1:100 in fresh LB plus 5 mM calcium chloride. Cultures were grown on an orbital shaker at 180 rpm at 30°C until the OD at 600 nm was approximately 0.4. An equal volume of fresh LB prewarmed to 50°C was added to each culture and the temperature of the incubator was shifted to 42°C to induce the lytic cycle of the phage. Immediately, a 1 mL sample was taken from each culture and mixed with 2 mL RNA protect Bacteria Reagent (Qiagen) and processed according to manufacturer literature. Additional samples were removed from the main cultures every 10 minutes following the temperature shift until the cultures lysed (became clear), then one additional sample was taken. Samples were treated as above. Cells were lysed and total RNA was isolated using Protocol 7 from the RNeasy Mini Kit (Qiagen). Total RNA was treated with TURBO DNase (Life Technologies) for one hour at 37°C. DNase was removed from the samples using the "RNA Cleanup" protocol from the RNeasy Mini Kit. cDNA was made from 1 μg total RNA using the Superscript III First-Strand Synthesis System (Life Technologies) and 1 μL cDNA was used as a template in qPCR with the 2x SYBR Select Master Mix (Life Technologies), according to manufacturer literature. Primer pairs used are indicated in Table 2. CysG was used as a constitutively expressed reference gene. Each cDNA sample was amplified in triplicate, and the resulting C T values for each technical replicate were averaged. Data were analyzed using the ΔΔC T method as described in [56], where ΔC T was determined by normalizing target gene expression to cysG expression and ΔΔC T was determined by normalizing the normalized target gene expression at time = t to normalized target gene expression at time = 0. Reported are the average fold-changes from time = 0 of three biological replicates for ref. For all other genes, all timepoints with a fold-induction within 20% of the maximum fold-induction for each biological replicate were averaged and the standard deviation is reported.

Lysogenization efficiency
Overnight cultures of E. coli MG1655 were diluted 1:100 in fresh LB and grown at 37°C until OD 600 was at least 0.4 (but no more than 0.6). Cultures (500 μL) were washed twice with CM buffer (5 mM CaCl 2 and 10 mM MgSO 4 ) and resuspended in 500 μL CM buffer. Phage prepared above (10 μL) were diluted with 90 μL LB. One tube of each phage was prepared for each biological replicate. MG1655 in CM buffer (200 μL) was added to each phage tube and incubated for 30 minutes at 30°C without agitation. Then, 1 mL LB and 100 μL 1 M sodium citrate were added and cultures were allowed to recover at 30°C for an additional 1.5 hours. Samples were washed with 200 μL M9 buffer twice and resuspended in 200 μL M9 buffer. Cell suspensions were serially diluted 1:10 and spotted on an LB plate to measure viability and an LB plate supplemented with 25 μg/mL chloramphenicol to measure P1 lysogens. Plates were incubated overnight at 30°C and colonies were counted the next morning. Lysogenization efficiency was calculated as CFU/mL chloramphenicol-resistant colonies divided by CFU/mL viable colonies and average and standard deviation of at least three biological replicates is reported.

Electrophoretic mobility shift assay
The 5' regulatory region of the ref gene was PCR amplified from strain EAR2 using primer pair ref5'reg. The high affinity region of bglB was amplified from pMK110 [99] using primer pair bglB. A GC rich region likely to not bind DNA was amplified from pBR322 using primer pair GC. All primer sequences are found in Table 2. DNA was either gel purified or spermidine precipitated using standard methods. 200 pM DNA was 5'-end labeled with [γ-32P]ATP using PNK (New England Biolabs). EMSA was performed as described in [99] and run on a 4% polyacrylamide gel.

ChIP and digital droplet PCR
Cells for immunoprecipitation were grown to mid-log and lysis was induced as in rt-qPCR experiments, except that the starting culture volume was 200 mL. Immediately after the addition of 200 mL prewarmed media, a 100 mL sample was withdrawn and 10 mM sodium phosphate and 1% formaldehyde were added and the sample was shaken for 5 min at 37°C to crosslink proteins DNA. The reaction was quenched with 100 mM ice cold glycine for 30 minutes on ice. Another 100 mL aliquot of cells was removed and processed 30 minutes after the temperature shift. Cells were harvested, washed with PBS, and frozen at -80°C. H-NS and bound DNA was immunoprecipitated with an H-NS antibody and processed as described in [100]. DNA present at different time points was quantified with the QX100 digital droplet PCR (Bio-Rad). PCR mixes contained 1x EvaGreen supermix, 1:1000 dilution of IP DNA, and 250 nM of each primer to amplify different regions of the E. coli and P1 phage genomes. Input DNA was run alongside as a control for total amount of DNA present. Three primer sets (

SOS response assays
Assays were performed as in [94]. Briefly, overnight cultures were diluted 1:100 in fresh LB and 200 μL was added to the wells of a black-walled, clear-bottom 96 well plate. The plate was incubated at 30°C and shaken orbitally on a plate reader. After 3 hours of growth, lysogens (Fig 4A) had 20 μL ciprofloxacin (diluted in water) added to a final concentration of 8 ng/mL. Protein expression strains (Fig 4B) had 20 μL arabinose added to a final concentration of 1% (instead of ciprofloxacin) to induce protein expression. Wells that did not receive ciprofloxacin or arabinose had 20 μL of water added. The plate was incubated at 30°C shaken orbitally. Every 10 minutes, the plate was briefly shaken linearly and the optical density (OD) at 600 nm and fluorescence at 509 nm (excitation at 474 nm). All values were blank-corrected by subtracting out the average OD or fluorescence value of the blank wells at each timepoint. This resulted in some negative fluorescence values, particularly for cultures that were not undergoing SOS. All fluorescence values were converted to positive numbers by adding a constant to every reading. Fluorescence is reported as the change in fluorescence from time 0. Error bars represent the standard deviation of the data.
Survival curves with arabinose added at 1x10 8

CFU/mL
Cultures were grown to OD 600 = 0.35 as in "arabinose survival curve at 5x10 5 CFU/mL" experiments. Each culture (approx. 1x10 8 CFU/mL) was split in two, and L-arabinose was added to a final concentration of 1% to one of the aliquots. Immediately, 100 μL of each culture was removed and serially diluted 1:10, plated on LB plates with appropriate selection antibiotics, and grown overnight at 30°C. The rest of the culture was incubated at 30°C for four hours, with 100 μL samples withdrawn periodically. These samples were diluted and plated as before.
Reported data are the average and standard deviation of at least three biological replicates.

Microscopy of cells expressing Ref variants
Cultures were grown at treated as in "arabinose survival curve at 1x10 8 CFU/mL" and were allowed to grow at 30°C for four hours (or overnight) after arabinose treatment. A 5 μL aliquot of each culture was removed and incubated with an equal volume of 10 μM DAPI for 5 minutes at room temperature. Then, 1-2 μL of the mixture was placed on a 1.5% agar pad, allowed to soak in, and flipped on to a glass microscope slide. All pictures were taken within a half hour of mounting.
All the images were acquired on a commercial Olympus IX83 inverted microscope equipped with an Olympus UPLSAPO60XS, 1.42 NA silicone oil immersion objective and Hamamatsu C11440 charge-coupled device (CCD) camera operated by a personal computer (PC) running MetaMorph for Olympus software. Brightfield images were obtained using an X-Cite 120 halide arc lamp (Lumen Dynamics Group, Inc.) and Olympus IX3-LWUCDA motorized long working distance (LWD) condenser. The fluorescence excitation for DAPI was performed using the Olympus standard components, and the emission for DAPI was obtain using a 5060C-OFF-Zero (-ZERO pixel set mounted in cube, Semrock) filter set. Contrast of images was adjusted for publication using ImageJ or Adobe Photoshop. Cell length measurements were obtained using the MicrobeTracker plugin for Matlab [66] and were converted to μm using the conversion factor 0.1083 μm/pixel.