The Vibrio cholerae Seventh Pandemic Islands act in tandem to defend against a circulating phage

The current circulating pandemic El Tor biotype of Vibrio cholerae has persisted for over sixty years and is characterized by its acquisition of two unique genomic islands called the Vibrio Seventh Pandemic Islands 1 and 2 (VSP-I and VSP-II). However, the functions of most of the genes on VSP-I and VSP-II are unknown and the advantages realized by El Tor through these two islands are not clear. Recent studies have broadly implicated these two mobile genetic elements with phage defense. Still, protection against phage infection through these islands has not been observed directly in any V. cholerae El Tor biotype. Here we report the isolation of a circulating phage from a cholera patient stool sample and demonstrate that propagation of this phage in its native host is inhibited by elements in both VSP-I and VSP-II, providing direct evidence for the role of these genomic islands in phage defense. Moreover, we show that these defense systems are regulated by quorum sensing and active only at certain cell densities. Finally, we have isolated a naturally occurring phage variant that is resistant to the defense conferred by the VSP islands, illustrating the countermeasures used by phages to evade these defense mechanisms. Together, this work demonstrates a functional role for the VSPs in V. cholerae and highlights the key regulatory and mechanistic insights that can be gained by studying anti-phage systems in their native contexts.


Significance (Author summary)
The current pandemic strain of Vibrio cholerae carries two unique genomic islands. How these two islands confer evolutionary advantage to the pathogen is unknown. We show here the identification of a circulating phage that is sensitive to the defense systems present on these two islands and demonstrate how phage variants can evade these defenses. Our studies provide the first direct evidence showing the importance of these genomic islands in defending against phage in their native environments; and in doing so provide novel insight into the mechanisms of these highly conserved defense elements. a1111111111 a1111111111 a1111111111 a1111111111 a1111111111

Introduction
There have been seven recorded cholera pandemics caused by two V. cholerae biotypes: classical and El Tor. The current, longest lasting, (7 th ) cholera pandemic is caused by the El Tor biotype which has globally displaced the classical biotype in both endemic and clinical populations [1]. One of the defining genetic features of this 7 th pandemic strain is the presence of the two aptly named Vibrio Seventh Pandemic Islands (VSP-I and VSP-II [2]). In spite of their persistence in V. cholerae, the functions of most of the genes on VSP-I and VSP-II are unknown and the evolutionary advantages realized by the El Tor biotype through the acquisition of these two islands are not clear.
It is generally believed that concentrations of phage and V. cholerae inversely correlate in aquatic reservoirs, and this predator-prey relationship has been postulated to be one driving force in determining the severity and timing of cholera outbreaks [3,4]. In the last decade, three predominant lineages of virulent V. cholerae phages, called ICP1, ICP2, and ICP3, were isolated from Bangladeshi clinical samples [5]. While there is strong evidence for phage predation in cholera patients, only ICP1 was able to prey on V. cholerae in estuarine water. ICP2 and ICP3 are better adapted for predation in a nutrient rich environment [6]. A phage cocktail comprised of these three phages is efficient at killing V. cholerae both in vitro and in vivo [7]. While using phages to treat and prevent cholera is promising, the interaction between these V. cholerae phages and the bacterial host is often complex and a constant arms race exists between phages and their bacterial hosts [8]. Bacteria can acquire various phage resistance mechanisms through mutations and horizontal gene transfer. Conversely, due to a strong selective pressure, phages readily develop resistance to many bacterial phage defense systems [9]. Therefore, it is important to study phage-pathogen interactions to have a better understanding of transmission of disease, acquisition of new traits important for pathogenesis, and ultimately the evolutionary history of the cholera pandemics.
Recent studies have increasingly shown a connection between predicted phage defense genes and mobile genetic elements (MGEs) [9,10]. While there has been strong evidence that dedicated anti-phage islands such as the PLE [11] are present in V. cholerae, it is apparent that even previously well described MGEs frequently contain phage defense genes. For example, the V. cholerae integrative and conjugative element (ICE) SXT has been shown to carry an arsenal of anti-phage gene clusters [12] in addition to the antibiotic resistance genes that were initially characterized. Similarly, variants of the Vibrio Pathogenicity Island I (VPI-I) have been sequenced and found to contain anti-phage CRISPR systems [13].
Until recently, the only VSP-I genes with a described function are dncV and capV. Respectively these genes encode the cyclic-GMP-AMP(cGAMP) synthase [14] and the cGAMP sensing phospholipase [15]. When the native capV-dncV operon from V. cholerae is transplanted into a phage-sensitive E. coli strain, the recipient strain becomes resistant to certain phages [16]. Related cyclic-oligonucleotide synthase and effector pairs have been subsequently discovered and some were shown to have phage defense functions [17]. Collectively, these anti-phage systems are known as cyclic oligonucleotide-based antiphage signaling systems (CBASS) [18]. Moreover, it is now apparent that the first gene of the VSP-I island, vc0175, encodes a functional deoxycytidine deaminase named AvcD [19,20]. AvcD converts dCMP/dCTP to dUMP/ dUTP and alters the cellular nucleotide pools, which is predicted to negatively impact highly replicative elements including phages. Similar to CBASS, AvcD has been demonstrated to inhibit phage when expressed in E. coli [19]. So far, neither DncV/CapV nor AvcD has been shown to defend against any phage that infects V. cholerae where the system is natively found.
VSP-II is the larger of the two islands and like VSP-I it is only in recent years that its function is being examined. Study of this island is complicated by the fact that multiple circulating variants exist [21]. However, most standard V. cholerae laboratory strains appear to contain the complete repertoire of VSP-II genes. For years only the island's integrase and cognate recombination directionality factor were known [13]. It was recently discovered that a gene cluster in VSP-II (vc0513-0515) is a part of the zinc-dependent Zur regulon [22]. Specifically, vc0513 encodes a transcriptional activator VerA to activate expression of vc0512 (aerB) to alter chemotaxis and aggregation in an oxygen dependent manner.
During the course of the current study, a gene cluster on VSP-II was identified for defense against different MGEs [23]. It is suggested that the vc0492-490 operon (ddmABC) in VSP-II encodes a defense system to reduce plasmid stability and defend against phage. This novel finding greatly increases our understanding of why seventh pandemic El tor biotypes have maintained VSP islands in their genome and why plasmids have been remarkably unstable in these strains. However, while plasmid destabilization by VC0492-490 (DdmABC) was comprehensively shown in V. cholerae, phage defense was demonstrated via ectopic expression in E. coli and no Vibriophage has yet been identified that is susceptible to such a system [23].
Here we report the isolation of a circulating variant of the phage ICP3 from a cholera patient stool sample and demonstrate that propagation of this phage is inhibited by elements independently identified in both VSP-I and VSP-II, providing the first direct evidence on the roles of these genomic islands in phage defense in their native host. We further demonstrate that the defense element on VSP-II is transcriptionally controlled via quorum sensing, providing protection at specific cell densities. Moreover, by comparing the phages that are either resistant or sensitive to the VSP islands, we gain insight on the potential mechanism for evading these systems.

Isolation of VSP-I/VSP-II susceptible phage
Recent studies suggest phage defense systems are often found in mobile genetic elements present in bacterial genomes [24]. We therefore suspected that the Vibrio Seventh Pandemic Islands (i.e., VSP-I and VSP-II) ( Fig 1A) may also harbor anti-phage genes. However, so far, no V. cholerae phage has been reported to be sensitive to these two islands. Due to the prevalence of these two genomic islands in the circulating clinical isolates of V. cholerae [2,25,26], we hypothesize that novel phages that are sensitive to these islands exist and can be isolated. Therefore, we screened for novel phages from rice water stool (RWS) samples that had been collected from cholera patients in Bangladesh on permissive host, which is an O1 El Tor V. cholerae E7946 derivative lacking both VSP-I and VSP-II islands (ΔVSP) and deleted for prophages CTXF and K139 [27,28]. Frozen RWS samples that have been stored at -80˚C were thawed, bacteria and other debris were gently pelleted, and the supernatant was filter sterilized. The filtrate was then plated on soft agar plates containing the ΔVSP host (a general phage isolation and identification scheme is shown in (Fig 1B). Four RWS samples were examined, two of these samples contained no detectable plaques while the other two yielded myriad plaques on this particular V. cholerae host. These phage plaques were then picked for further examination. It has previously been shown that there are three dominant phages (ICP1, ICP2, and ICP3) circulating with V. cholerae in Bangladesh [5]. Using PCR primers specific to the gene encoding the unique DNA polymerase of each phage, we were able to categorize most of the newly isolated phage as closely related to ICP1, ICP2, or ICP3 (S1 Table).
The newly isolated phages were serially diluted and spotted on lawns of either the parental strain (WT) or ΔVSP (Fig 1B). Comparing the plaquing of these phages, we observed the majority showed no host preference. However, a number of phages from one stool sample appeared to form fewer and smaller plaques on WT than on ΔVSP (Figs 1B and 2A). In addition to displaying distinct plaque morphology on each host (Fig 2A), quantitatively, these differentially plaquing phages formed~2.5 times fewer plaques on WT than on ΔVSP (Fig 2B). While plaque morphology or efficiency of plating (EOP) can be influenced by many factors, our results suggest that the VSPs play an important role in altering the phage lifecycle. Based on whole genome sequencing, these novel phages all appeared to be variants of ICP3 (Accession number ON464735.2). We renamed one of these newly isolated VSP susceptible phage ICP3_2016_M1 (M1F) since it was isolated from a 2016 stool sample from Bangladesh. ICP3 is a roughly 38kb T7-like lytic phage with a characteristic podoviridae appearance [5]. Compared to previously sequenced ICP3 (HQ641340 [5]), M1F has homology with 99% coverage and 95.5% identity. It should be noted that this previously sequenced ICP3 was isolated using VSP-carrying El Tor strain [5]. Sensitivity to the VSPs is not observed in this previously isolated ICP3 (S1A Fig). Despite the similarities to previously sequenced ICP3, the substantial number of polymorphisms in the M1F genome make it difficult to determine any one specific factor responsible for the observed difference in VSP sensitivity.

Determination of VSP genes responsible for M1F targeting
In strains with either VSP-I or VSP-II completely deleted, plaque morphology and EOP of M1F were modestly changed and more closely resembled what was observed on WT (Fig 2A  and 2B). This implies that there is an element on each island that can act on M1F. Looking for homology to other anti-phage systems, we were unable to come up with strong candidates on VSP-II. However, VSP-I contains dncV, which encodes the enzyme for 3'-3'-cyclic-GMP-AMP (cGAMP) synthesis [14]. cGAMP activates the phospholipase encoded by the upstream gene capV leading to cell death [15]. This dncV-capV system had been expressed heterologously in E. coli to defend against phage infection [16] and therefore appeared a strong candidate for the VSP-I element for targeting M1F. However, we did not observe any differences in either  EOP was calculated by dividing the number of plaques from the strains ectopically expressing either avcD or ddmC by the number of plaques from the strains carrying the empty vector. Significance was determined by one-way ANOVA with Dunnett's plaque number or morphology when these genes were disrupted with and without VSP-II (S1B and S1C Fig).
To determine the specific VSP gene responsible for altering M1F infection, we took a systematic approach to determine the potential anti-phage element(s) on each island. A series of strains were constructed where a defined section of VSP was deleted in a host where the other island had been completely deleted. The regions selected for deletion are loosely based around the predicted operon structure and therefore vary in size. By plaquing M1F on these newly constructed "scanning deletion" strains and comparing the number of plaques to that on WT, we identified the regions on each island that appeared to target M1F. Specifically, we determined that the plaque number and morphology of M1F on the Δvc0175-176 ΔVSP-II and ΔVSP-I Δvc0490-493 strains were similar to those on the ΔVSP strain (S1C and S1D Fig). To further narrow down the gene on each island, we assayed M1F plaque formation on strains with deletions within these newly identified regions. Through this process, we determined that vc0175 (avcD) and vc0490 (ddmC) were the two genes on VSP-I and VSP-II, respectively, which cause differential M1F plaquing (Fig 2B and S1C-S1E Fig). Importantly, deletion of just these two genes together phenocopied the deletion of the entirety of both VSP-I (14kb) and VSP-II (27kb). Constitutive expression of either gene from a plasmid in the ΔavcD ΔddmC background reduces plaquing to approximately half of that of an empty vector in the same strain (Fig 2C), which is the exact difference we observed comparing the two single deletion mutants to the double deletion mutants ( Fig 2B). Moreover, constitutive expression of either avcD or ddmC in WT reduced plaquing as well, suggesting that the phage targets have not been saturated with these two phage defense proteins, and the production of these defense proteins is not subjected to a feedback inhibition mechanism ( Fig 2C). From these data we conclude that avcD and ddmC are the elements on VSP-I and VSP-II that inhibit the ability of M1F to plaque on V. cholerae. A recent report indicates ddmC (vc0490) forms an operon with vc0491(ddmB) and vc0492 (ddmC) [23], and the DdmABC function together as a complex, therefore, we assume disruption of any one of the genes in this operon is sufficient to abolish phage defense.

VC0175 (AvcD) and VC0490 (DdmC) inhibit phage replication
Having established the minimal VSP inhibitory elements, we began to characterize how these genes affect the phage life cycle in its V. cholerae host. Deletions of these genes in V. cholerae did not inhibit growth ( Fig 3A) which allowed us to examine the kinetics of M1F phage infection across different hosts. To do this we grew cells to mid-log phase and infected a multiplicity of infection (MOI) of 2, indicating there are an average of two phage for every bacterial cell. In contrast to the modest EOP differences observed when plaquing on a plate, the changes in lysis between strains was striking ( Fig 3A). The parental VSP + strain (WT) was nearly fully protected from lysis by M1F while the ΔavcD ΔddmC mutant was rapidly lysed by the phage. The protection mediated by these two systems were not dependent on the MOIs used (S2 Fig). We noticed the bacterial host recovered from phage-induced lysis and resumed growth several hours after infection ( Fig 3A and S2 Fig). When these recovered cells were re-infected with M1F, they remained sensitive to phage infection, suggesting these are not spontaneous resistant mutants. Together, these results directly demonstrate these genes provide a high level protection to a rapidly growing bacterial population against phage predation in liquid culture. multiple comparison test comparing the EOP of each strain to the control of the WT strain. Signifiers for significance were displayed based on p-values as follows: 0.1234(ns), 0.0332( � ), 0.0021( �� ), 0.0002( ��� ), <0.0001( ���� ).
https://doi.org/10.1371/journal.pgen.1010250.g002 cholerae ΔavcD ΔddmC strain containing the plasmid pMMB67eh or the same plasmid expressing avcD or ddmC was grown and infected with M1F at an MOI of 0.2. (C) Phage burst assays. Strains were grown to mid-log phase before being infected with phage at an MOI of 0.1. After a period of absorption cultures were further diluted 1/2500,1/25000, and 1/250000. Burst is calculated by dividing the total output phage by absorbed phage (T0-T10). Strains are indicated on the x-axis. P-values were determined by Unpaired T-tests with Welch's correction and significance shown as We were able to restore the protection by expressing these genes individually from a plasmid in the ΔavcD ΔddmC mutant. While expression of either gene from a plasmid reduced the overall lysis of a culture, avcD had a stronger effect in the conditions tested, nearly entirely ablating lysis (Fig 3B). Ectopic expression of ddmC did not completely prevent lysis but reduced the amount a culture lysed significantly versus an empty vector (Fig 3B). Although to a smaller extent, protection to phage-induced cell lysis was still observed at a higher MOI in the ΔavcD ΔddmC mutant when either gene was over-expressed (S3A Fig). At either MOI, incomplete complementation could be due to unbalanced gene expression between ddmC expressed ectopically from a low copy number plasmid and the chromosomally encoded ddmB and ddmA. Still, these results further indicate that that either gene is sufficient for protection against M1F. The lack of toxicity to the host (both WT and ΔVSP) when these genes are overexpressed also suggests that they are not inhibiting phage replication by simply hindering the V. cholerae's ability to grow (S3B and S3C Fig). We did however see a mild growth defect when the genes were expressed in E. coli, indicating that these genes cause minor growth inhibition in other bacteria (S3D Fig). While other groups had shown growth inhibition utilizing the same avcD expression construct [19], we suspect differences in the genetic background of the V. cholerae strains used in these studies (i.e., C6706 vs E7946) and the additional mutations we have introduced into our strains might account for the differing phenotypes.
The sufficiency of either gene to protect against productive M1F infection is also shown via a burst assay. While M1F produced approximately 50 phage per cell in VSP + (WT), ΔavcD, and ΔddmC strains ( Fig 3C) the burst size was doubled in strains with either both VSPs or both ddmC and avcD deleted, resulting in~100 phages released per cell (Fig 3C). This increase in burst could help partially explain the larger plaque size observed in these strains (Fig 2A) although it may not be the only contributing factor [29,30]. The burst sizes of M1F on these mutants that is nearly 2.5 times of WT which also matches the EOP data closely (Figs 2A and 3C).
Next, we tested if viral DNA replication of M1F was inhibited by avcD and ddmC in V. cholerae. To do this, we measured by qPCR the increase in viral DNA over the course of a replication cycle. As predicted, we observed more viral DNA in the ΔavcD ΔddmC mutant than the WT. This difference of~2x more DNA produced closely matched the overall burst differences ( Fig 3D). Together, these results suggest that AvcD and DdmC are most likely targeting phage DNA replication, reducing the overall number of phage genomes produced in each cell, and thereby reducing the total number of phage released from each cell.
AvcD has been shown to be a functional deoxycytidine deaminase capable of converting dCMP to dUMP, presumably changing the nucleotide pool in the cell to inhibit phage replication [19]. Another study showed that DdmC, working together with DdmB and DdmA, decreased plasmid stability in El Tor biotype by an unknown mechanism [23]. Similar to the previous study, we observed that DdmC appeared to be a SMC-Like protein based on Alphafold predictions [31]. These proteins are often required for proper chromosome segregation during replication and can play a role in plasmid maintenance.
Although AvcD and DdmC both target phage replication, because of the dissimilar predicted functions of these two systems, we reasoned that each system could have a unique target specificity. To evaluate if both systems similarly reduced plasmid uptake and stability, we � p<0.0332 and �� p<0.0021. (D) Fold change in M1F genome copy through infection. The indicated strain was grown to mid-log phase before being infected with M1F at an MOI of 0.02 and grown in LB at 37˚C on a roller. Immediately upon infection and at 20 minutes post infection a portion of the culture was removed and boiled. The boiled culture was then diluted 1:50 and used as template for qPCR with primers targeting the ICP3 DNA polymerase. Fold change was calculated by dividing the CT value at 20minutes by the value at 0 minutes for each culture.
https://doi.org/10.1371/journal.pgen.1010250.g003 measured the efficiency of strains to take up commonly used plasmids via conjugation from E. coli. Significantly more transconjugants were obtained from the ΔddmC recipient than those from WT or the ΔavcD mutant (Fig 4). This phenotypic divergence suggests that these two systems target MGEs with different specificity and it is probable that these systems use two distinct mechanisms to inhibit M1F DNA replication.

A SNP in ICP3_2016_M1 results in VSP sensitivity
Previously identified ICP3 did not show this susceptibility to the VSPs [5] (S1A and S1B Fig). Therefore, we sought to determine what causes M1F to be susceptible to the defense afforded

PLOS GENETICS
Phage defense systems in Vibrio cholerae by these two genes. As previously stated, M1F has 99.5% identity and 99% coverage when compared to a previously sequenced ICP3. While this was sufficient for us to categorize the isolated phage as ICP3, hundreds of differences remained between these two related phages. To isolate VSP resistant ICP3 with a more similar genetic background to M1F, we used the same RWS sample where M1F was isolated, and we re-isolated another ICP3 phage using the same scheme but for phages that did not show a preference for hosts with or without the VSPs. We successfully isolated a naturally occurring VSP resistant phage that we called ICP3_2016_M2 (M2F). Sequencing of this phage revealed that its genome was almost identical to M1F, with only 2 single nucleotide polymorphisms (SNPs) (Accession number ON464736.2). Both of these SNPs map to the gene that encodes the well conserved phage polymerase gp22 (S4 Fig). These VSP sensitivity determining mutations are both found in the exonuclease domain of the phage DNA polymerase but only one would result in an amino acid change (L94I). While this change is subtle, it occurs in a well conserved region proximal to the invariant non-catalytic aspartate that is essential for the ssDNA exonuclease function of the polymerase [32].
The minor genetic differences in M2F appeared to have a substantial impact on its ability to tolerate avcD and ddmC as there was no change in EOP when comparing plaques formed on the mutant strain to WT (Fig 5A). This is in direct contrast to the increase in EOP we observed on the same strains with M1F ( Fig 2B). When lysis assays were performed with each phage infecting at the same MOI (Fig 5B), WT resisted lysis by M1F but not with M2F (Figs 3A and 5B). It is also notable that while both M1F and M2F phages lysed the ΔVSP host strains to a similar degree, M2F appeared to initiate host cell lysis earlier than M1F (~20 mins for M1F vs~40 mins for M2F after phage addition), potentially suggesting differences in replication between the two phage. Our results suggest that M2F is not sensitive to inhibition conferred by avcD and ddmC; and we predict the resistance is likely due to a change in the structure and/or function of the phage DNA polymerase.

Protection by ddmABC is quorum sensing dependent
Phage infection relies on the active metabolism of the cell and therefore it is expected that many phages replicate more efficiently while strains are actively growing in mid-log phase. However, when optimizing infection conditions, we noticed a striking difference in lysis when phages were introduced at different cell densities of the V. cholerae host. During our standard infections with M1F at mid-log growth phase (OD~0.5), we observed the expected avcD ddmC dependent protection (Fig 6A). Expectedly, at higher cell density (OD~0.8), no lysis was detected at all in both WT and the ΔavcD ΔddmC mutant, which we assumed was due to the reduced metabolic state of the host at late log phase that was not conducive to phage infection ( Fig 6A). Intriguingly, at low cell density (OD~0.2), we observed lysis in all strains infected with M1F (Fig 6A) including the VSP + WT. This is in stark contrast to what we would have expected from strains containing the phage defense elements on VSP-I or VSP-II. These results suggest that the VSP phage defense systems are not active at low cell density and the phage protection phenotype at a higher cell density (OD~0.5) is a sign of more complex regulation. Furthermore, these results may help explain the discrepancy between complete protection in liquid cultures infected at a specific OD and infection on soft agar plates where developing plaques infect cells at varied growth states.
While avcD was previously shown to be important for quorum-sensing (QS) dependent multicellular aggregate formation in V. cholerae at high cell density [33], the link between ddmC (and the ddmABC operon) and QS is less clear. To test if ddm operon is also regulated by QS, we used a reporter plasmid containing the Photorhabdus luxCDABE luciferase operon under the control of the region immediately upstream of ddmA to measure gene expression in different QS mutants. Overall, relative light production increased with cell densities for all strains tested (Fig 6B). However, in the ΔluxO mutant that is genetically locked at QS high cell density state [34] [35], there was an increase in bioluminescence across different cell densities when compared to the WT (Fig 6B). In contrast, in the ΔcqsA ΔluxS strain missing the two known QS autoinducer synthases, and thus locked at low cell density QS state [34] [35], there was a decrease in bioluminescence across different cell densities when compared to the WT. Unlike the ΔcqsA ΔluxS strain, a hapR mutant which is also locked at low cell density QS state [34][35], did not show any change in luminescence when compared to the WT (Fig 6B). Together, these data strongly suggest that the ddmABC operon is controlled by QS [34][35][36] and is active only at higher cell densities. Specifically, the inactivity of this particular system at low cell density is due to the repression of gene expression by the Qrr sRNAs, AphA, and/or other unknown low cell density QS regulators [34][35][36] rather than the lack of a transcriptional activation by HapR, a mechanism that has been previously observed [37,38].
To test the QS-dependent phage protection, we infected different QS mutants with M1F at OD~0.5 where we would expect both avcD and ddmABC to be active and protective. Indeed, we observed complete protection in the WT strain as well as the ΔluxO mutant even at a high MOI, but not in the ΔcqsA ΔluxS strain. These results strongly suggest a functional connection between QS and protection from M1F infection.

Discussion
In this work, we have isolated a circulating novel variant of the Vibrio cholerae phage ICP3 (M1F) in a cholera patient stool sample that is susceptible to defense genes carried on the Vibrio Seventh Pandemic (VSP) Islands. We determined that both avcD on VSP-I and ddmC (or the ddmABC operon) on VSP-II are anti-phage elements that target M1F. We show that these systems inhibit DNA replication of the phage, reducing the overall number of phage released per cell. Moreover, these systems significantly protect V. cholerae cultures from lysis by M1F at both high and low MOIs. We have also isolated in the same stool sample a naturally occurring phage that is closely related to M1F that is VSP-resistant (called M2F), carrying changes in the proofreading domain of the phage DNA polymerase.
As previously stated, there are many SNPs throughout the phage genomes which differentiate the previously isolated ICP3 and the newly isolated ICP3_2016_M1 F or M2 F including the gene that encode GP22 (DNA polymerase). In contrast, there is exactly one predicted amino acid divergence in GP22 for the two new phage isolates (S4 Fig). Among the related T7 polymerase family, this residue is mostly a leucine and is next to the well conserved non-catalytic aspartate [32]. While M1 has a leucine in this position, M2F has an isoleucine. Looking through deposited sequences via the Basic Local Alignment Search Tool (BLAST), we were unable to identify any ICP3 or T7 sequences that had a similar leucine to isoleucine change at the equivalent position in the DNA polymerase. Therefore, M1F appears to be the ancestral isolate and it is very likely M2F is derived from M1F to allow better replication in circulating VSP-containing V. cholerae. However, the change in the M2F DNA polymerase may come with a fitness cost due to the high conservation of this region which suggests functional importance.
During the course of this study, three independent studies [19,20,23] identified the same gene clusters on VSP-I and VSP-II as potential anti-phage systems. Our work complements these findings but, importantly, shows function of these two islands in phage defense in the native host species and in a clinical setting. By identifying both sensitive and resistant V. cholerae phage isolates from the same patient stool sample, we have directly observed the arms race between the phage and its native bacterial host, illustrating how phage counteracts these defense systems within the native environment (i.e., within the human small intestine) and providing additional insights on the molecular mechanisms used by these systems for defending against phage attack.
AvcD was identified to be a deoxycytidine deaminase that functions to modify nucleotide pools. Interestingly, a small RNA called AvcI upstream of avcD controls the activity of AvcD post-translationally, resembling a toxin-antitoxin system [39]. Over-production of AvcD in the absence of AvcI leads to cell filamentation, a phenomenon similar to thymine-less death (TLD), even though AvcD only carries out deamination of both dCMP and dCTP. Expression of AvcD and its homologs from other species in E. coli leads to toxicity but also protection against some coli phages, suggesting modification of the nucleotide pools inside the bacterial host could lead to phage defense via direct inhibition of phage replication [19,20]. Our study uniquely shows that M1F is sensitive to alteration of the nucleotide pools mediated by AvcD in a system with no observed toxicity. This idea was further supported by our isolation of the VSP-insensitive M2F phage variant in which a polymorphism was identified in the exonuclease domain of the phage DNA polymerase. We predict that the M1F phage DNA polymerase is more sensitive to an imbalanced nucleotide pool, and this sensitivity could lead to blockage of replication or an "error catastrophe" that reduces phage fitness.
Deletion of ddmC or the whole ddmABC operon mimics the effect of removing VSP-II for defense against this phage. Recent study suggests that the three gene products function together to reduce plasmid stability in the pandemic V. cholerae El Tor biotype [23]. This system appears to be especially active towards small or medium sized plasmids [23]. Our data further support that ddmABC reduces conjugation of certain mid-sized plasmids (4.3kb and 8.8kb) from E. coli to V. cholerae. Expression of this operon in E. coli also confers resistance to some coliphages [23]. Our study further illustrates the importance of this operon in defending against a native circulating phage. The exact mechanism for phage protection conferred by this system remains unclear. Using bioinformatic analysis and structural prediction using Alphafold, we and others [23] predicted that DdmC (VC0490) contains a SMC domain common in certain chromosome partition proteins and therefore it is likely needed for recognizing and binding to specific features in the target DNA. DdmA (VC0492) is predicted to be an endonuclease, likely for processing and degrading target DNA. The small DdmB (VC0491) protein is predicted to be an ATPase which could function with nearby effectors to respond to foreign genetic materials [40]. It is therefore most likely that the nuclease activity of DdmA is being directed and regulated by DdmC and DdmB. Yet, the specific target and activator of this system remain elusive. VSP-resistant phage M2F, producing a slightly different DNA polymerase than that from the VSP-sensitive M1F, provides a clue that this defense system could either sense the polymerase directly, or more likely the phage DNA replication intermediates associated with the polymerase. Indeed, a phage defense system called Nhi, carrying both a nuclease and helicase domain, is recently identified to sense phage replication intermediates to direct DNA processing [41]. Regardless of the mechanism, although many versions of VSP-II circulate in the natural isolates of V. cholerae, nearly all VSP-II islands contain the ddmABC operon [21,42], suggesting there is significant pressure to maintain these core genes.
Moreover, recent study on ddmABC predicted the mechanism of defense to be abortive infection (ABI) [23], where a small number of infected bacterial cells sacrifice themselves to halt the propagation of the phage and protect the rest of the uninfected cells in the same population [43]. Data from our study suggests that ABI, at least in V. cholerae, may not be the sole acting mechanism for these systems. V. cholerae carrying functional VSPs appears to be fully protected from M1F at both low MOI of 0.02 (S2 Fig) and high MOI 2.0 (Fig 3A), suggesting that infected cells are still viable when phage are relatively abundant. Contrary to what we observed, a defense system using ABI would not be protective at high MOI and would result in characteristic cell lysis due to phage infection. Thus, our results imply that a specific molecule is being targeted/sensed by the defense system and this molecule is usually absent or present infrequently in the bacterial host. Since DdmABC act on a variety of plasmids as well as phages, the target likely is not a specific protein encoded by these foreign genetic elements. Again, based on the polymorphism existing between M1F and M2F, we speculate some replication intermediates are targeted or sensed by this system. Further analyses are required to identify the exact molecular mechanisms, but our work provides a phage-host combination that readily allows for such analysis.
While protection from phage through heterologous expression in foreign host such as E. coli is a very powerful approach, especially in large scale studies where demonstration of the phage defense activity is the key; these approaches may not inform the intricate functions of some antiphage systems. Our study uniquely illustrates that complex gene regulation exists for these phage defense genes in their native host environment and suggests conditions where phage defense systems are coordinated with other cellular processes. For instance, we show that these defense systems are under both growth phase and quorum sensing control: these systems are only expressed and become active at certain cell density, and therefore M1F phage cannot infect and lyse V. cholerae host carrying these defense systems at high cell density (e.g., late exponential phase/early stationary phase) unless quorum sensing (QS) is disrupted. Dependence of QS for phage protection in V. cholerae has been reported previously, however, such protection is mediated through a HapR-dependent production of hemagglutinin protease, and partly through downregulation of phage receptors [44]. In contrast, the QS-dependent protection we identified in this study is HapR independent. What is the possible driving force for connecting QS and phage protection? Since V. cholerae only develops natural competence and picks up foreign DNA at high cell density mediated by QS [45], QS mediated defense against foreign DNA could act as a checkpoint and prevent acquisition of unwanted MGEs by activating these defense systems. Alternately, upregulation of anti-phage systems during rapid (logarithmic) growth could be a mechanism to protect bacterial populations when they are the most susceptible to phage infection. Moreover, our isolation of both VSP-sensitive and -resistant phage directly from a cholera patient stool sample suggests that the human small intestine provides a unique environment to allow an active arms race between the bacterial host and the phage to occur. Specifically, during the early stage of intestinal colonization, V. cholerae has to be in a low cell density QS state to promote virulence gene expression [34,35]; however, such QS state prevents an optimal expression of the VSP phage defense genes which would favor phage infection.
While our ability to identify anti-phage systems has expanded rapidly, our data suggests that there are regulatory and mechanistic insights that can only be gained by studying these systems in their native hosts defending against endogenous phage. Outside of V. cholerae, there has also been increasing awareness of the arsenal of anti-phage genes contained on MGEs [46,47] and many components from these systems have been used as basic molecular biology tools for years [48,49]. As phages are so plentiful and diverse, undoubtedly there are more anti-phage mechanisms that are yet to be uncovered. Indeed, many (non-temperate phage) MGEs themselves are thought to be derived directly from phages, or derelict phage themselves [50,51]. By identifying new systems on these abundant but often ill-defined islands, we can help expand the "molecular toolbox" and pave the way for future innovation.

Bacterial strains and growth conditions
All V. cholerae strains used in this study were O1 El Tor. Bacterial strains utilized in this study are listed in S2 Table. WN6145 was the primary V. cholerae strain used in the study (provided by Dr. Andrew Camilli). This strain is derived from E7946, an El Tor Ogawa strain [52] to promote phage infection. Genetic alterations include elimination of phase variable site in the gene that encodes the common phage receptor the O1-antigen [53], deletion of the CTX prophage, and deletion of the kappa prophage. Bacteria were propagated in LB at 37˚C shaking with aeration. Strains containing P lac -bearing, RP4-derived pMMB67eh [54] were grown with 100μg/ mL Ampicillin, strains with pBBRlux::ddmA were grown with the addition of 2.5μg/mL Chloramphenicol. For growth curves, strains were grown in 200μL cultures in 96-well plates (Ther-moScientific Nunc 167008) at 37˚C with lid shaking. Bacterial growth and lysis curves of V. cholerae with M1F (Figs 2 & 6 and S2 & S3 Figs) or M2F (Fig 5) were performed with phage added at the indicated time and MOI at 37˚C shaking. For complementation experiments using pMMB67eh, LB broth and top agar was supplemented with 100μg/mL Ampicillin and 100μM IPTG (isopropyl-β-D-thiogalactopyranoside).

Generating V. cholerae mutant strains
Natural transformation was used to introduce antibiotic resistance markers into V. cholerae [55]. Unmarked deletions for the desired gene or gene cluster were made by splicing by overlap extension (SOE)-PCR and the resulting PCR products was transformed at the same time with PCR products where a cassette conferring kanamycin or spectinomycin resistance replacing lacZ. After selecting on the appropriate antibiotic colonies were screened for deletion of the unmarked locus [56]. Strains were made competent by growth on sterile chitin in instant ocean [55]. For marked constructs, approximately 500ng of PCR product was added to competent cells on chitin. When co-transforming, 200ng DNA containing the antibiotic resistance marker was used with 2μg unmarked PCR product. The oligonucleotide sequences used for PCR and sequencing reactions are provided in S3 Table. Plasmids were constructed via digestion with restriction endonucleases and subsequent ligation into vectors with T4 ligase. Insertions were initially identified by PCR and confirmed via sanger sequencing. In the ΔavcD deletion, the coding sequence of the gene between the start and stop codon were removed. Due to the AT rich start and proximity to ddmB, the ΔddmC deletion leaves the regions encoding the first 12 and the last 4 amino acid residues with an added stop codon added in place of the remaining coding region.

Isolation of Vibriophage from patient stool samples
Stool samples from cholera patients were previously collected as described [57]. Phage were isolated by scraping a small amount of -80˚C stool sample into a 1.5mL Eppendorf tube. These were thawed and gently spun to remove bacteria and other debris before the supernatant was passed through a 0.45μM filter (VWR 28145-485). Soft agar overlays were prepared by growing ΔVSP-I ΔVSP-II (WN7006) to OD~0.5 at 37˚C shaking, adding this culture to 0.5% Top Agar, and overlaying this mixture on LB plates. Serial dilutions of filtered stool samples were then applied on top of the overlay and incubated at 37˚C overnight. Single plaques were picked into 25μL TM buffer. Five μL of the plaque suspension was boiled, diluted 1:20 and used as template for ICP phage identification PCR using previously described primers [5]. Serial dilutions on these picked plaques were performed and plated on soft agar overlays with either ΔVSP-I ΔVSP-II (WN7006) or the isogenic VSP + strain (WN6145) V. cholerae to assess differences in plaquing between the two hosts. For those phages that were further studied, single plaques were picked from the ΔVSP-I ΔVSP-II (WN7006) plates of this second plating to subsequently have phage stocks prepared via PEG precipitation.

Phage infection
Phage stocks were generated on the permissive ΔVSP-I ΔVSP-II strain (WN7006) using polyethylene glycol (PEG) precipitation from liquid cultures using a protocol modified from [29,58]. V. cholerae was grown to OD 600~0 .3 in LB at 37˚C shaking, phage was added at an MOI of 0.2. Cultures were monitored until fully lysed (cleared). At this point DNAse, RNAse, and 0.002% (v/v) chloroform was added to eliminate bacteria and bacterial DNA/RNA. Bacterial debris were removed by adding NaCl to a concentration of 0.5M and centrifuging at 4˚C. The supernatant was then taken and 10% PEG 8000 was added. After overnight precipitation cultures were spun and the resultant pellet of phage was resuspended in TM buffer (10mM Tris-HCL pH 7.5, 100mM NaCl, 10mM MgCl 2 ). Phage stocks were further cleared by the addition of chloroform once more and centrifugation. The aqueous phase from this final centrifugation was then used as a working phage stock.
Titering of phage for was carried out by growing V. cholerae to mid-log, infecting cultures with diluted phage stocks, and allowing absorption to occur for 5-10 minutes before plating on 0.7% top agar. After a brief amount of time to solidify, plates were incubated at 37˚C overnight. Plaques were enumerated with the aid of magnification to ensure count of small plaques. EOP was calculated after titering equal number of phage on different V. cholerae strains at the same time ensuring that phage titer and incubation times remain the same. MOI was determined by calculating the colony forming units (CFU) per milliliter at a given OD 600 value and then adjusting the number of plaque forming units (PFU) to be added based on the desired MOI and expected CFU. Average burst size was determine by using one-step growth curves [59] performed with at least three biological replicates. Phage burst is reported as individual values with a bar representing the median ±SD (Standard Deviation) in Fig 2B.

Quantification of plaque size
After plaque plates had been incubated and counted plaque size was evaluated. To increase contrast of the plaques an overlay of 0.1% (2,3,5-Triphenyl-2H-tetrazolium chloride(TTC) (Thermofisher AAA1087009) was applied [60]. Plates were imaged on an Epson Perfection V800 Dual Lens scanner. Analysis of the images was performed in the program Fiji. Images were smoothed, the "Find Edges" function was used before automatic Thresholding (Li). The "analyze particles" function was then used to identify plaques selecting those larger than 4 pixels and with a circularity between 0.8 and 1.0.

Isolation of phage DNA and sequencing
Phage gDNA was isolated from lysate modified from [61]. Lysate was treated with DNase and RNase to remove bacterial DNA at 37˚C for 30 minutes before heat inactivation at 65˚C for 10 minutes. Phage capsids were disrupted by the subsequent addition of proteinase K and SDS. Phage DNA was then cleaned up using Zymo DNA Clean & Concentrator kit (D4029) and eluted in water. Illumina libraries were prepared for each phage using the standard Nextera XT DNA Library prep protocol. Sequencing was conducted at the Tufts University Core facility using an Illumina NextSeq 550 and Single-end 75nt reads. Genome sequence analysis was performed with CLC Genomic Workbench v20.0.4.

Real-time quantitative PCR
Reactions for qPCR experiments were carried out with PowerUp SYBR Green Master Mix (Thermofisher) on a CFX Connect Real-Time PCR Detection System (Bio-Rad) as previously described in [11]. Three independent samples were tested for each strain and each template was assayed in technical duplicate. Bacteria were grown to OD = 0.5 at 37˚C with aeration, at which point phage were added at an MOI = 0.1. Immediately upon infection and 20minutes post infection 20μL samples were taken, boiled, and diluted 1:50. This boiled lysate was used as template for qPCR using phage specific primers as previously described [5] with primers WNTP1342 (5'-ATTGTCGAGTGGGACAAAGG-3') and WNTP1343 (5'-ACCAACTC-GACGCATAGCTT-3').

Conjugation
Strains were grown overnight in LB at 37˚C shaking. V. cholerae strains used for these experiments contained ΔlacZ::spec (WT WLN7019, ΔavcD WLN7008, ΔddmC WLN7011, and ΔavcD ΔddmC WLN7020) for enhanced selection of transconjugants. Equal quantities of E. coli donor and V. cholerae recipient were mixed and spun gently before being resuspended in 80μL LB. This concentrated mixed was plated in 3x 20μL spots on LB with no drug supplements and incubated at 37˚C for 2.5 hours. Bacteria was scraped from the plates and resuspended in LB before being serially diluted and plated on LB polymyxin B (50U/mL) Spectinomycin (100μg/mL) and either Ampicillin (100μg/mL) for pMMB67eh or Kanamycin (100μg/mL). Colonies that arose after overnight incubation at 37˚C were enumerated.