Phages are viruses that infect bacteria and are important drivers of bacterial diversity and microbial community development7,8. Following host infection, temperate phages can undertake one of two lifestyles. They can enter the lytic cycle, whereby they exploit host resources, replicate, produce viral particles and kill their host6,9. Alternatively, phages can enter the lysogenic state and remain dormant (that is, as prophages) and be passed down to host progeny10. Phages can also cause chronic host infections in which they persist and extrude viral particles without killing the host11,12.

During lysogeny, model temperate phages, including phage lambda, produce a repressor (called cI) that binds and prevents expression from a promoter (termed PR) that controls the lysis genes9. Following DNA damage, activation of the host RecA protein leads to autoproteolysis and inactivation of the cI repressor13,14. Consequently, PR is de-repressed, which triggers phage replication, host-cell lysis and transmission of the phage to neighbouring bacteria. Other temperate phages harbour repressors that lack the peptidase domain responsible for autoproteolysis. Instead, a peptidase or antirepressor encoded elsewhere in the phage genome is activated by the host SOS response15,16. The understanding that all bacteria possess recA, coupled with the fact that phages are omnipresent, has led to the common view that the host SOS response is the universal prophage inducer. However, significant concentrations of potent DNA-damaging agents are rare in the environment and, increasingly, phages are being discovered that are not induced by DNA damage17. Together, these findings suggest that undiscovered induction triggers exist in nature.

Recent findings have revealed that quorum-sensing (QS) signals represent one SOS-independent induction trigger for phage lysis–lysogeny lifestyle transitions18,19,20. QS is a process of cell-to-cell communication that bacteria use to orchestrate collective behaviours. QS relies on the production, release and group-wide detection of and response to extracellular signalling molecules called autoinducers (AIs)21. Phages can harbour phage-to-phage QS-like communication systems, such as the arbitrium system in SPβ phages20. Alternatively, as in vibriophage VP882, they can monitor host bacterial QS-mediated communication pathways to tune the timing of the lysogeny-to-lysis switch to changes in host-cell density19. Phage VP882 is a linear plasmid-like prophage that encodes a homologue of the Vibrio QS receptor VqmA (called VqmAPhage), which is activated by a host-produced AI called DPO19,22. Following binding to DPO, VqmAPhage activates the transcription of a counter-oriented gene called qtip. Production of Qtip induces host-cell lysis. Our hypothesis is that through the surveillance of a bacterial-produced QS signal, the phage can integrate host-cell density information into its decision-making process. Moreover, by lysing its host at high cell density, the phage maximizes the probability of infecting other cells in the population.

Bacteria commonly harbour multiple prophages, a state called polylysogeny. How prophages residing in a polylysogenic host compete for host resources is not well understood. What is known is that following DNA damage, the number of virions released for lambdoid prophages is lower in polylysogens than in monolysogens23. This result suggests that in polylysogens, co-induced prophages compete for reproductive success. It is possible that in the context of polylysogeny, if a prophage possesses an alternative, non-DNA damage-dependent pathway to lysis, it could compete more effectively with co-residing prophages that cannot respond to the alternative induction cue. Here we sought to identify and characterize polylysogenic bacteria that harbour prophages that possess SOS-dependent and SOS-independent pathways to lysis. Our aim was to use them as models to explore within-host prophage competition.

The current work describes the discovery of SOS-independent, phage-encoded lysis-inducing modules that, despite bearing little resemblance to one another at the sequence level, share a common regulatory logic. Our characterization of several of these phages shows that they all use a transcription factor (TF) to activate the expression of a divergently transcribed gene encoding a small protein (smORF). The smORF induces the transition from lysogeny to lysis. The smORFs studied here lack homologues and predictable domains; however, they operate by inactivating the same respective target, the cI repressor, in the phage that encodes the smORF. The mechanisms by which the TFs regulate their partner smORF genes vary. In some cases, the TFs operate independently. In other cases, smORF expression requires a xenobiotic responsive element (XRE) family protein working in conjunction with a LuxR-type QS receptor or TF that requires a bacterial-produced AI ligand for activity.

The prophage-containing isolates on which we focus are polylysogenic. We show that the addition of a DNA-damaging agent leads to phage-mediated lysis of the bacteria and the release of a mixed population of phage particles. Single-cell analyses demonstrate that this outcome stems from a subset of host cells expressing lytic genes from only one of the phages, another subset expressing lytic genes from only the other phage and the final subset of cells expressing lytic genes from both phages. Unlike DNA damage, induction through the newly discovered regulatory modules results in gene expression from and near-exclusive production of the phage responsive to the specific input. Our results suggest that the activities of these SOS-independent pathways dictate the outcomes of prophage–prophage competition by expanding the range of stimuli to which specific prophages can respond.

Two mechanisms of phage-driven lysis

To advance studies of inter-prophage competition in polylysogenic bacteria, we conducted a search among sequenced phage genomes for genes encoding putative SOS-independent lysis–lysogeny modules located between repA and telN, which are hallmarks of all known linear plasmid-like phages. Our strategy was inspired by the arrangement of the vqmAPhage and qtip genes, which are encoded between repA and telN in vibriophage VP882 (ref. 19). We searched all prophage genomes at the National Center for Biotechnology Information (NCBI) database and six recently curated phage and phage-plasmid databases24,25,26,27,28,29. The genomes span diverse environmental, marine and human body sites, and we searched for convergently oriented repA and telN genes residing within 10 kb of each other. The search revealed 784 putative linear plasmid-like phage genomes, 274 of which contained distinct yet conserved repAtelN loci (Supplementary Table 1). In 271 out of the 274 genomes (99%), the gene upstream of repA encoded a cI-like DNA-binding protein adjacent to putative, divergently transcribed lytic, structural and regulatory genes. A panel of representative loci is shown in Fig. 1a. We constrained our search to phage genomes that harboured RecA-dependent, autoproteolytic cI repressors because autoproteolytic repressors exhibit a stereotypical response to SOS (that is, repressor cleavage), which is bioinformatically predictable and testable in vitro. Non-proteolytic cI repressors require additional phage-specific and/or host-specific factors for regulation. Therefore, we did not consider them for the current study. For phages that possess autoproteolytic cI repressors, we reasoned that any additional regulatory modules uncovered in their genomes would likely respond to SOS-independent inputs. Filtering phage genomes using these criteria led to 61 distinct loci. The majority of phages eliminated at this step (210 out of 271) encoded repAtelN loci that resembled those of the Escherichia coli linear plasmid-like phage called N15. This finding is likely a consequence of the overrepresentation of Enterobacteriaceae in the sequencing databases. N15 is known to be subject to antirepression30,31, and we determined that the repressor it encodes is non-proteolytic (Fig. 1b). The only member within the identified set of phages that harboured autoproteolytic repressors that had been previously investigated was phage VP882 (Fig. 1a,b), described above, and a singleton in our current cluster analysis (Supplementary Table 1). Many of the phages from our database searches come from metagenomics sequencing projects, are unobtainable and/or have no known host. Nonetheless, with the goal of probing the functions of these putative sensory input pathways, we focused on obtainable phage–host pairs.

Fig. 1: Variable gene content in an otherwise conserved locus of linear plasmid-like phages reveals TF–smORF modules that regulate lysis independently of SOS.
figure 1

a, Phylogenetic tree (left) and gene neighbourhoods (right) for 34 representative telN loci encoding convergently oriented telN and repA genes. Phage VP882 and Vibrio 1F-97 phage 72 are denoted with stars. The SOS-independent pathway components in phage VP882 (vqmAPhage and qtip) are in blue and orange, respectively. Genes encoding predicted autoproteolytic repressors (red) cluster together with respect to TelN phylogeny and are distinct from genes encoding non-cleavable repressors (navy). NCBI accession numbers are depicted to the right of each sequence. Scale bar indicates the number of amino acid substitutions per site. b, SDS–PAGE in-gel labelling of the non-proteolytic N15 phage repressor (HALO–cIN15) and the autoproteolytic phage VP882 repressor (HALO–cIVP882). The minus and plus symbols indicate the absence and presence, respectively, of 500 ng ml−1 ciprofloxacin used to induce the SOS response. M denotes the molecular weight marker (representative bands are labelled). c, Growth of Vibrio 1F-97 carrying aTc-inducible smORF72 in medium containing aTc (denoted smORF72), ciprofloxacin (denoted SOS) or water (denoted by the minus symbol). d, PR72lux expression in E.coli carrying an empty vector (designated V) or aTc-inducible smORF72 grown in medium containing aTc. The PR72lux plasmid carries cI72, which represses reporter expression. Relative light units (RLUs) were calculated by dividing bioluminescence by OD600. e, SDS–PAGE in-gel labelling of the Vibrio 1F-97 phage 72 repressor (cI72–HALO) produced by E.coli carrying aTc-inducible smORF72. The treatments –, SOS and smORF72 refer to water, ciprofloxacin and aTc, respectively. M as in b. f, PsmORF72lux expression from E.coli carrying an empty vector (V) or aTc-inducible tf72 grown in medium containing aTc. RLU as in d. g, Growth of Vibrio 1F-97 carrying aTc-inducible tf72 in medium lacking or containing aTc (white and blue, respectively). Data are represented as the mean ± s.d. with n = 3 biological replicates (c,d,f,g) or as a single representative image chosen from three replicates experiments (b,e).

Source Data

TF–smORF modules are SOS independent

The first isolate we investigated was Vibrio cyclitrophicus 1F-97 (henceforth called Vibrio 1F-97), which encodes a putative phage on genome contig 72 (hereafter called phage 72). Between the phage 72 repA and telN genes are genes encoding a putative TF (as judged by its predicted DNA-binding domain according to the NCBI conserved domain database) and a counter-oriented small, 171 nucleotide open reading frame (ORF) (hereafter called TF72 and smORF72, respectively). The sequences of these genes lacked nucleotide-level and amino acid-level identity to the vqmAPhage and qtip genes. However, their arrangement paralleled that of vqmAPhage and qtip in phage VP882. To test whether the DNA located between repA and telN on phage 72 controls the phage lysis–lysogeny transition, smORF72 was cloned under the control of an anhydrotetracycline (aTc)-inducible promoter on a plasmid (pTet–smORF72) and conjugated into Vibrio 1F-97. Addition of aTc to this recombinant strain led to a precipitous decline in the optical density at 600 nm (OD600), similar to when the potent SOS-activator ciprofloxacin was added. This result indicated that phage-driven lysis occurred (Fig. 1c). aTc administered to Vibrio 1F-97 harbouring an empty vector, or to Vibrio cholerae, Vibrio parahaemolyticus or E.coli (none of which contain phage 72) harbouring the pTet–smORF72 vector, did not affect growth. This result showed that the smORF72 protein alone does not induce lysis (Extended Data Fig. 1a,b). We eliminated the possibility that an additional predicted 165 nucleotide ORF encoded between repA and telN functions similarly to smORF72, as overexpression of the gene encoding this protein did not drive lysis (Fig. 1a and Extended Data Fig. 1c). Thus, host-cell lysis requires the presence of the phage and induction of the smORF72 gene. Supplementation with ciprofloxacin did not activate smORF72 expression in wild-type Vibrio 1F-97 (Extended Data Fig. 1d), which suggests that smORF72 production is independent of SOS. Notably, phage preparations obtained from Vibrio 1F-97 treated with aTc or with ciprofloxacin contained phage 72 particles, which indicates that phage 72 can be induced by both SOS-independent and SOS-dependent pathways (Extended Data Fig. 1e).

We next explored how induction of smORF72 promotes lysis. Important for this step is that our bioinformatics analysis revealed that 99% of the repAtelN loci (271 out of 274) contained genes upstream of repA encoding predicted phage-repressor proteins (called cI72 and its target promoter PR72, respectively, for phage 72). We verified the function of this repressor–promoter pair by fusing the PR72 promoter to lux on a plasmid (PR72lux). Recombinant E.coli carrying the construct made light, which decreased by 500-fold when the gene encoding cI72 was also introduced (Extended Data Fig. 1f). Consistent with this result, purified cI72 protein bound to the PR72 promoter in vitro (Extended Data Fig. 1g). Introduction of pTet–smORF72 into E.coli carrying the cI72–PR72lux plasmid restored high level light production, which indicated that production of smORF72 inactivates the cI72 repressor (Fig. 1d). Unlike DNA damage, smORF72 production did not lead to cI72 proteolysis (Fig. 1e). These results suggest that smORF72 is an antirepressor that inactivates its partner cI72 repressor protein through a nonproteolytic mechanism.

As noted above, a gene encoding a TF, tf72, lies adjacent to smORF72. TF72 is a logical candidate to be the regulator of smORF72 expression. Indeed, production of TF72 in E.coli harbouring the PsmORF72lux plasmid (in which the promoter of smORF72 is fused to lux) increased light output by 28-fold compared with the empty vector (Fig. 1f). Similarly, production of TF72 in Vibrio 1F-97 led to an increase in smORF72 transcription (Extended Data Fig. 1h), after which the culture lysed (Fig. 1g). These data suggest that TF72, through transcriptional activation of smORF72, drives phage-72-mediated lysis.

A central question is how tf72 expression is naturally regulated in phage 72 to induce the smORF72-mediated lysis cascade. In an effort to identify small-molecule inducers of phage 72, we introduced PsmORF72–lux into the Vibrio 1F-97 lysogen and we monitored both PsmORF72lux activity and growth (through OD600 measurements) following exposure to various compounds. We tested commercially available compound libraries (Biolog MicroArrays, around 2,000 conditions) and a curated library of antibiotics (a gift from the Seyedsayamdost Group, around 250 conditions). Among the antibiotic library compounds are known phage inducers, including DNA-damaging agents and regulators of reactive oxygen species. A reporter for E.coli phage lambda induction (PRlambdalux) was likewise assessed, which enabled us to determine whether inducers of PsmORF72lux and/or inhibitors of growth were specific for phage 72 or not. No test compound elicited more than a twofold increase in PsmORF72–lux expression or a decline in OD600 that was specific to phage 72 (Extended Data Fig. 2a). Thus, our screens did not produce potential inducers. Identifying induction stimuli for prophages that are non-SOS-inducible has proven difficult17. However, some phenothiazines, including common antipsychotic drugs, which have not previously been implicated in phage induction, drove low-level activation of both phage reporters (Extended Data Fig. 2b).

Vibrio 1F-97 harbours two prophages

Bioinformatics searches (BLASTp) of the phage-72-encoded TF–smORF module indicated that the closest homologue of the TF72 regulatory protein in NCBI (about 53% amino acid pairwise identity; Extended Data Fig. 3a) resided on a different contig within the same Vibrio 1F-97 genome, contig 63. Contig 63, like phage 72, harboured signatures of a linear plasmid-like phage (for example, genes analogous to telN and repA by synteny; Fig. 2a). Furthermore, DNA corresponding to contig 63 was present in phage preparations of ciprofloxacin-treated Vibrio 1F-97 cultures (Extended Data Fig. 3b). We now refer to this element as phage 63, and we classify Vibrio 1F-97 as polylysogenic for phage 63 and phage 72. The two phages shared little identity on a whole-genome basis (26.4% average nucleotide identity), and all analogous genes, with the exception of the two TFs, shared less than 30% amino acid identity (BLASTp). Furthermore, unlike phage 72, cloning and expression of the phage 63 DNA that intervened repA and telN did not induce lysis (Extended Data Fig. 3c,d). Rather, the gene encoding the homologous TF in phage 63 (tf63) was located between genes encoding predicted partition machinery (parAB) and an operon encoding predicted late genes (tail and assembly genes; Fig. 2a). tf63 resided adjacent to, but in the opposite orientation from, a gene encoding another hypothetical 183 nucleotide smORF (smORF63). smORF63 and smORF72 have only 9 identical residues in pairwise alignment (11% amino acid identity; Extended Data Fig. 3e). For phage 63, we constructed the analogous set of tools described above for phage 72 and determined that the par-associated tf63smORF63 pair encoded proteins that performed equivalent functions as the tf72smORF72 pair (Extended Data Fig. 4a–f). We were unable to identify a tf63smORF63-specific small-molecule inducer using the above compound screening strategy (Extended Data Fig. 4g). We conclude that the Vibrio 1F-97 polylysogen harbours two plasmid-like phages, each with SOS-independent pathways to lysis.

Fig. 2: Vibrio 1F-97 harbours two linear plasmid-like phages that control lysis through TF–smORF modules encoded in different genomic loci.
figure 2

a, Organization of genes on linear plasmid-like phages in Vibrio 1F-97. Genes are coloured by functional annotation as noted in the key. Circles below genomes denote positions of the tf (blue) and smORF (orange) genes located at repAtelN (phage 72) or by par genes (phage 63). All analogous genes, with the exception of those encoding the TFs, share less than 30% amino acid identity (BLASTp). b, PsmORFlux output from E.coli harbouring PsmORF72lux or PsmORF63lux and a second plasmid encoding an empty vector (V), aTc-inducible tf72, tf63 or the chimeric tf genes, designated tf72-5x and tf63-5x, each grown in medium containing aTc. The chimeric TF proteins (black bars) have five amino acids in their DNA-binding domains replaced from the other TF. Thus, TF72-5x possesses five amino acids from TF63 and TF63-5x possesses five amino acids from TF72. Extended Data Fig. 3a shows the exchanged residues and their locations. c, Light production from E.coli harbouring the cI-repressed reporter plasmids cI72–PR72lux (PR72lux) or cI63–PR63lux (PR63lux) and a second plasmid encoding an empty vector (V), aTc-inducible smORF72 or aTc-inducible smORF63, each grown in medium containing aTc. Data are presented as the mean ± s.d. with n = 3 biological replicates (b,c). RLU as in Fig. 1d (b,c).

Source Data

To examine the specificities of the two TF–smORF lysis pathways, we assessed whether either of the phage TFs could activate expression of the non-cognate smORF gene. Figure 2b shows that TF72 did not cause light to be produced from a PsmORF63lux reporter, and TF63 did not drive light production from the PsmORF72lux reporter. Notably, structural predictions of TF72 and TF63 using AlphaFold32 suggested that the two TFs most closely resemble each other (root mean square deviation (RMSD) = 1.3 Å) (Extended Data Fig. 5a). They also resemble restriction modification controller proteins (RMSD = 0.5–0.9 Å) and the helix-turn-helix TF ClgR from Mycobacterium smegmatis (RMSD = 1.0–2.5 Å) (Extended Data Fig. 5b). AlphaFold modelling pinpointed the residues in each phage TF that constitute the DNA-recognition helix α3 (Extended Data Fig. 5a). Exchange of five residues within the recognition helix and flanking carboxy-terminal loop reversed the TF72 and TF63 preferences for smORF72 and smORF63 promoter DNA (the chimeric proteins are called TF72-5x and TF63-5x, respectively; Fig. 2b and Extended Data Fig. 3a). Thus, promoter specificity is conferred by a minimal set of non-contiguous unique residues in each TF. Consistent with the strict specificities of the phage pathway components, neither smORF, when produced together with the non-cognate cI repressor (pTet–smORF63 with cI72–PR72lux; pTet–smORF72 with cI63–PR63lux), generated PR-driven light production (Fig. 2c). Thus, each phage-encoded TF–smORF module is specific at two levels: smORF expression and smORF-driven repressor inactivation.

QS controls TF–smORF modules

Our finding that phage genes encoding TF–smORF modules are not restricted to repAtelN genomic regions motivated us to expand our search. Preliminary BLAST searches revealed putative TF63 homologues in 117 predicted phage genomes, which were almost exclusively associated with the Vibrionaceae family (Supplementary Table 2). In this dataset, we identified TF63 homologues in four major genomic contexts: (1) in repAtelN loci; (2) within 5 kb of the partitioning gene parB; (3) in a heterogeneous genomic region containing small, variable ORFs of unknown function and; (iv), most rarely, within 10 kb of genes encoding predicted terminase and portal proteins. We used these findings to inform a second, guilt-by-association search for putative lysis-controlling modules that lack detectable sequence homology to TF63 beyond those encoded by Vibrionaceae revealed in the BLAST search.

The NCBI database was searched for sequences encoding homologues of ParB and the major capsid protein from phage 63. We chose to examine the parB-associated loci more deeply because this genomic context was one of the most common in our preliminary TF63 BLAST search and contained a conserved indicator gene (parB), which other common context groups lacked (Supplementary Table 2). The ParB search was anchored against a phage capsid protein to enrich for phage-associated parB loci instead of homologous sequences in plasmids or bacterial genomes. Capsid genes represent some of the most conserved viral sequences and were common among the genomes containing parB in our original set of 117 phages with TF63 homologues33. We filtered the output of this search for hits encoding an apparent TF within 10 kb of the predicted parB gene given the close association between TF63 and ParB in our earlier query. This search strategy revealed 56 distinct contigs (Supplementary Table 3), approximately 90% of which (50 out of 56) were putative linear plasmid-like phages based on detectable repAtelN loci. All 56 contigs were predicted to originate from phage genomes when analysed using VIBRANT, a hybrid machine-learning and protein similarity-based tool for phage identification34 (Supplementary Table 1).

In one par-associated node of interest containing five phages, each phage carried an operon containing genes encoding an XRE-like DNA-binding protein and a LuxR-type TF (Fig. 3a and Extended Data Fig. 6a). LuxR-type proteins contain amino-terminal acyl homoserine lactone (HSL) AI-binding domains and C-terminal helix-turn-helix DNA-binding domains35. The luxR genes for two of the identified phages (ARM81ld of Aeromonas sp. ARM81, and Apop of Aeromonas popoffii) were associated with host strains that we had previously identified in a bioinformatics search for phage-encoded LuxR proteins36. In that work, we showed that the phage LuxR proteins could bind the HSLs Aeromonads are known to produce. However, we were unable to deduce the functions of the phage LuxR proteins at the time owing to an inability to obtain an Aeromonas sp. ARM81 isolate with the phage and the genetic intractability of A.popoffii. To expand our understanding of the roles of these phage modules, we successfully obtained Aeromonas sp. ARM81 and we developed genetic tools to study A.popoffii.

Fig. 3: Activation of PsmORFApop requires XREApop, LuxRApop and the C4-HSL AI.
figure 3

a, Phylogenetic tree (left) of 28 representative ParB proteins. Gene neighbourhoods (right) for the 28 loci encoding parB (purple) and the major capsid protein (grey). ARM81ld, Apop and phage 63 are denoted with stars. Genes encoding predicted TFs that are not luxR or xre are in green. NCBI accession numbers are depicted to the right of each sequence. Scale bar as in Fig. 1a. b, PsmORFApoplux activity from a plasmid in E.coli carrying a second vector with the designated aTc-inducible gene (or genes). All media contained aTc. Supplementation with DMSO or C4-HSL is denoted by minus and plus symbols, respectively. c, PsmORFApoplux activity from a plasmid in E.coli carrying xreApop under its native promoter and a second vector with the designated aTc-inducible gene (or genes). Media contained the indicated combinations of water (–) or aTc (+), and DMSO (–) or C4-HSL (+). d, PsmORFApoplux activity from a plasmid in E.coli carrying luxRApop under its native promoter and a second vector with the designated aTc inducible gene (or genes). Treatments as in c. e, Proposed model for the regulation of the TF–smORF module in Apop. (1) XREApop activates the expression of the xreApopluxRApop operon, and (2) increased production of LuxRApop and XREApop occurs. (3) LuxRApop when bound to the C4-HSL AI ligand (pentagon), together with XREApop, activates expression of the counter-oriented smORFApop gene, and activates the expression of xreApopluxRApop. (4) smORFApop inhibits the cIApop repressor, (5) leading to host-cell lysis. The mechanism underlying the co-requirement for AI-bound LuxRApop and XREApop to activate PsmORFApop remains to be determined. Data are presented as the mean ± s.d. with n = 3 biological replicates (bd). RLU as in Fig. 1d (bd).

Source Data

First, we investigated the A.popoffii XRE–LuxR (XREApop–LuxRApop) pair. An unannotated 156 nucleotide smORF resided approximately 150 nucleotides upstream of the xreApopluxRApop operon near the par locus (Extended Data Fig. 6a). To determine whether it is regulated by XREApop–LuxRApop, a plasmid carrying aTc-inducible xreApopluxRApop was transformed into E.coli harbouring lux fused to the promoter of the candidate smORF (PsmORFApoplux). Light production from PsmORFApoplux commenced only when the E.coli reporter strain was supplied with C4-HSL (Extended Data Fig. 6b), an AI natively produced by Aeromonads37. Thus, XREApop and the LuxRApop–AI complex control smORF expression. Consistent with this finding, deletion of the smORF locus from Apop abolished XREApop–LuxRApop-AI-dependent lysis of A.popoffii but not ciprofloxacin-dependent lysis (Extended Data Fig. 6c). Analogous to what we show in Fig. 1d and Extended Data Figs. 1g and 4b,f, the Apop cI protein cIApop shifted PR-Apop DNA in vitro (Extended Data Fig. 6d), and smORFApop inactivated cIApop (Extended Data Fig. 6e). Lastly, unlike DNA damage, smORFApop did not lead to cIApop autoproteolysis (Extended Data Fig. 6f). The Aeromonas sp. ARM81 phage ARM81ld XRE–LuxR module functioned analogously to the A.popoffii XRE–LuxR module (Extended Data Fig. 7a–e). Thus, these XRE–LuxR-controlled modules likely act as orthogonal SOS-independent pathways to induction.

The Apop and ARM81ld phage TF–smORF modules are distinctive among the characterized set of TF–smORF modules given that two TFs (XRE and LuxR) are apparently involved in the activation of smORF expression. To define the individual and combined contributions of XREApop and LuxRApop, we generated expression vectors carrying only pTet–xreApop or only pTet–luxRApop and tested them in the PsmORFApop–lux assay. Neither XREApop nor LuxRApop alone drove reporter output irrespective of the presence of C4-HSL. This result indicated that in addition to AI, smORFApop expression depends on both TFs (Fig. 3b). To validate these findings, we engineered xreApop–PsmORFApoplux and luxRApop–PsmORFApoplux reporter plasmids, which carry the identical PsmORFApop–lux reporter sequence but include either full-length xreApop or full-length luxRApop controlled by the native promoter. We introduced pTet–xreApop or pTet–luxRApop into E.coli harbouring the reporters. Heterologous xreApop expression induced light production only when luxRApop was present on the reporter plasmid and AI was supplied (Fig. 3c,d). By contrast, heterologous luxRApop expression did not induce light production in either case, irrespective of the presence of AI (Fig. 3c,d). We interpret these results as follows: XREApop activates the expression of the xreApopluxRApop operon, driving both XREApop and LuxRApop production. Together, LuxRApop bound to C4-HSL and XREApop are required to activate smORFApop expression, which induces the phage lytic cascade and host-cell lysis (Fig. 3e). Confirming the proposed regulatory arrangement in vivo, quantitative PCR with reverse transcription (RT–qPCR) results revealed that overexpression of xreApop in A.popoffii led to the rapid activation of expression of xreApop and luxRApop, whereas overexpression of luxRApop did not (Extended Data Fig. 7f).

Induction dictates phage competition

Inter-prophage competition is predicted to arise in polylysogens when co-occurring prophages are induced by the same trigger (for example, DNA damage)23. Our above discovery of non-canonical phage lysis pathways established the possibility that the outcomes of such competitions could depend on the particular inducer that activates lysis. Vibrio 1F-97 cells lyse following phage induction by ciprofloxacin (DNA damage), production of smORF63 or production of smORF72 (Fig. 1c and Extended Data Fig. 4e). Nonetheless, the amounts of phage 63 and phage 72 virions released under each condition could reflect the consequences of inter-prophage competition. Whole-genome sequencing of phage particle preparations from Vibrio 1F-97 cultures induced by ciprofloxacin, smORF63 or smORF72 revealed that ciprofloxacin treatment drove the production of comparable quantities of DNA corresponding to both phage 72 and phage 63 (Fig. 4a). By contrast, cultures induced by smORF72 or smORF63 resulted in near-exclusive production of phage 72 particles and phage 63 particles, respectively (Fig. 4a). Thus, under non-SOS-inducing conditions, the phage produced depends on the smORF driving lysis.

Fig. 4: DNA damage drives discrete subsets of cells producing one, the other or both phages whereas TF–smORF modules exclusively drive production of only the phage encoding that module in polylysogenic hosts.
figure 4

a, Sequencing of viral particles prepared from Vibrio 1F-97 cultures carrying an empty vector, aTc-inducible smORF72 or aTc-inducible smORF63 grown in medium containing ciprofloxacin or aTc. Depth refers to the number of read counts. b, Top, representative smFISH images of phage 63 and phage 72 lytic genes in Vibrio 1F-97 cells following SOS activation or smORF induction as in a. Images are maximum z-projections of raw smFISH fluorescence for phage 72 (72; magenta) and phage 63 (63; cyan) genes with cells outlined in white. Bottom, phage 72 versus phage 63 smFISH intensity per Vibrio 1F-97 cell in the absence of phage induction (orange, n = 677 cells), with SOS activation (black, n = 516 cells), smORF72 induction (magenta, n = 375 cells) or smORF63 induction (cyan, n = 485 cells). Uninduced cells were used to determine boundaries (grey dotted lines; Methods) to delineate cells displaying no phage induction (bottom left quadrant), exclusively phage 72 induction (bottom right quadrant), exclusively phage 63 induction (top left quadrant), or both phage 63 and phage 72 induction (top right quadrant). Scale bar, 10 µm. c, Top, representative smFISH images of phage ARM81mr and phage ARM81ld early lytic gene expression in Aeromonas sp. cells following SOS activation or smORFARM81ld induction as in d. Bottom, images as in b for phage ARM81mr (mr; cyan) and phage ARM81ld (Id; magenta) genes. Phage ARM81mr versus ARM81ld smFISH intensity per Aeromonas sp. cell in the absence of phage induction (orange, n = 516 cells), with SOS activation (black, n = 463 cells) or smORFARM81ld induction (magenta, n = 477 cells). Scale bar and quadrants as in b (Methods). d, Detection of viral particles from Aeromonas sp. ARM81 carrying aTc-inducible smORFARM81ld grown in medium containing ciprofloxacin (black) or aTc (orange). Relative viral load is the amount of ARM81mr-specific or ARM81ld-specific DNA (cIARM81mr and cIARM81ld, respectively) in the induced samples relative to an uninduced sample measured by qPCR. e, Model for how TF–smORF modules influence inter-prophage competition. Left, a ubiquitous prophage inducer promotes competition because multiple prophages share host cell resources. Right, prophage-specific induction through a TF–smORF leads to exclusivity because only the induced phage garners host cell resources. Data are presented as the mean ± s.d. with n = 3 biological replicates (ac) or as the mean ± s.d. with n = 3 biological replicates and n = 4 technical replicates (d).

Source Data

To date, investigations of prophage induction in polylysogenic hosts have been limited to population-level measurements, as in Fig. 4a. Consequently, it is unclear whether in a single host bacterium, co-residing prophages can be simultaneously induced and acquire the resources needed for replication or, alternatively, whether only one of the co-residing prophages successfully replicates. Specifically, in the case of the Vibrio 1F-97 polylysogen, multiple mechanisms could produce the mixture of phage 72 and phage 63 particles that are produced following DNA damage. First, each host cell could produce both phage 72 and phage 63 particles. Second, each host cell could produce exclusively phage 72 particles or exclusively phage 63 particles. Third, a subset of host cells could produce exclusively phage 72 particles, another subset could produce exclusively phage 63 particles and a final subset could produce both phage 72 and phage 63 particles. To determine which possibility is correct, we used single-molecule RNA fluorescent in situ hybridization (smRNA-FISH) to mark and visualize phage induction in individual cells of the Vibrio 1F-97 polylysogen. Our strategy relied on the use of probes specific for phage 72 or for phage 63 genes that are expressed only during entry into the lytic cycle (Methods). Figure 4b shows that following DNA damage, a subset of cells (33%) exclusively harboured RNA from phage 72, a subset (9%) harboured RNA from phage 63 and a subset (28%) harboured RNA from both phages. Thus, DNA damage causes heterogeneity in phage induction at the single-cell level. By contrast, and as predicted on the basis of results presented in Fig. 4a, only phage-72-specific RNA was detected in cells induced by smORF72, and only phage-63-specific RNA was detected in cells induced by smORF63 (Fig. 4b). These findings indicate that activation of TF–smORF modules drive exclusivity in prophage induction at the single-cell level. Conversely, SOS activation leads to multiple populations of differentially induced cells, including those expressing the lytic genes of both phages and of either phage alone.

Previous work had determined that Aeromonas sp. ARM81 is polylysogenic and contains an integrated prophage, ARM81mr, in addition to the plasmid-like ARM81ld prophage38. We have no genomic evidence for an additional sensory pathway on the ARM81mr phage. We used the identical single-cell imaging smRNA-FISH analysis described above to discover the mechanism that gives rise to the sets of released phages from Aeromonas sp. ARM81 following induction with either ciprofloxacin or smORFARM81ld. The results mirrored those for Vibrio IF-97. DNA damage resulted in cells expressing lytic genes of exclusively phage ARM81mr (36%), exclusively phage ARM81ld (13%) or both phages (46%). Conversely, smORFARM81ld production drove the expression of only phage ARM81ld (Fig. 4c). Indeed, quantitation of the virions released following treatment with ciprofloxacin or following induction of xreARM81ldluxRARM81ld expression produced results consistent with the smRNA-FISH analysis. Specifically, administration of ciprofloxacin led to a 140-fold increase in ARM81mr phage particles and a 22-fold increase in ARM81ld particles relative to when no inducer was added (Fig. 4d). By contrast, induction of expression of xreARM81ldluxRARM81ld with C4-HSL led to a fivefold increase in ARM81ld particles, whereas ARM81mr particles were reduced by fivefold compared with their uninduced levels (amounting to a 700-fold reduction in ARM81mr particles compared with ciprofloxacin treatment; Fig. 4d). Together, our results from Vibrio 1F-97 and Aeromonas sp. ARM81 demonstrate that DNA damage results in the induction of and competition between co-residing prophages, whereas possession of an additional sensory pathway drives host-cell lysis exclusively by the phage that encodes the module (Fig. 4e). Thus, the particular induction cue dictates the distribution of phage particles produced by the polylysogenic host. Moreover, single and dual phage production is possible from an individual host cell.


Despite most bacteria being predicted to be polylysogenic, an understanding of the consequences of polylysogeny on the hosts and on their resident phages has been constrained by the limited models available and a lack of known prophage induction cues beyond the SOS response. Currently, competition among co-residing temperate prophages is generally considered a sprint, whereby in response to a single trigger (that is, the SOS response), differences in particular phage properties (for example, replication rates, packaging rates, burst size, among others), dictate how many particles of each phage are produced. Our current work in polylysogens harbouring phages with multiple pathways to lysis shows that differential phage induction can occur, and which phage or phages are produced varies depending on the induction cue. Although tuning into the host SOS response remains a universal mechanism by which prophages perceive host-cell stress, the additional sensory pathways we discovered here suggest that more specialized conditions exist that favour the induction of one phage over another.

In this work, we focused on discovering and characterizing SOS-independent lysis induction pathways on linear plasmid-like phages that encode autoproteolytic cI repressors. We uncovered a pattern in which previously unknown phage regulatory components are encoded adjacent to genes characteristic of linear plasmid-like phages (that is, near replication and partition machinery genes). To make headway, we constrained our initial work to this subset of phages, which constitute a specific subtype present in Gram-negative bacteria. We predict that SOS-independent pathways to lysis are widespread among phages but may require different search strategies to reveal them. Indeed, RecA-independent phage induction has been shown to occur in lambdoid phages in E.coli; however the molecular mechanisms that underlie these alternative pathways are unknown39.

Our single-cell imaging analyses further revealed that phage induction through the SOS cue drives a heterogenous outcome from polylysogens. That is, individual cells produce one type of phage, the other type of phage or a mixture of both phages. In the cases in which only one of the lytic genes of the phages is expressed in a host cell, we speculate that perhaps some stochastic process causes only one phage to enter the lytic cycle or, alternatively, one phage achieves a sufficient head start over the other phage and it monopolizes host resources, quenching production of the other phage. Our finding that a mixture of phages can be produced by a single host cell shows that multiple prophages can enter the lytic cycle and presumably compete for the host resources required to replicate and produce viral particles. Regarding Aeromonas, we note that although the smFISH analyses showed that the SOS cue drives similar levels of early lytic gene induction from both phages, phage ARM81mr ultimately outcompeted phage ARM81ld, producing tenfold more viral particles. Differences in rates of downstream processes (for example, replication and/or packaging) could drive this discrepancy. Our use of smRNA-FISH for imaging prophage induction in polylysogenic bacteria enabled this visualization of inter-prophage competition at the single-cell level. Going forward, this approach can be adapted to study other polylysogens and, for example, used for investigations of phage gene expression on rapid timescales during prophage induction or infection.

Regarding the molecular mechanisms that underlie inter-prophage competition, our results showed that induction of a TF–smORF module of a phage does not cross-activate the lytic programme of another co-resident prophage. In Vibrio 1F-97, this specificity was achieved at two levels: specific TF-mediated activation of smORF expression and specific smORF-driven repressor inactivation. AlphaFold modelling of the TFs revealed similarity in the global folds of the two proteins; however, the specificity for their respective target promoters is conferred by a few key residues within the DNA-recognition helix. In contrast to our findings, investigation of polylysogenic Salmonella strains revealed that some phage antirepressor proteins can inactivate cognate and non-cognate repressors, thereby enabling the synchronization of induction of multiple prophages40. This arrangement is thought to be vital for prophages with slow induction responses, allowing them to piggyback off of prophages that are more rapidly induced40. The finding that prophage induction by an additional cue does not trigger induction of co-residing prophages could be a strategy that proves especially successful when host resources are limiting because it ensures exclusive reproduction and dissemination of only the induced phage.

The two phage-encoded LuxR proteins investigated here required partner XRE proteins to activate the expression of their counter-oriented smORF genes, a requirement that does not exist for most bacterial LuxR QS receptors. One recent example, however, demonstrates that the activity of the Pseudomonas aeruginosa LuxR receptor, called RhlR, is modulated by direct interaction with the PqsE protein41. Whether the phage LuxR and XRE proteins interact directly, and if so, if the mechanism parallels that for RhlR–PqsE, remains to be determined. Finally, we found that xre and luxR do not always co-occur in phage genomes. A tBLASTn search using XREApop as the query revealed three contigs, likely from linear plasmid-like phages, in different Shigella sonnei genomes (Extended Data Fig. 6a). Unlike in the Aeromonas phages, these putative Shigella phages lack luxR genes. Shigella are not known to produce HSL AIs, thus a LuxR component might not provide benefits to the phage in the Shigella host. It is possible that particular sets of regulatory components encoded on different plasmid-like phages have been shaped through evolution, presumably by relevant host-sensory cues.

The genes encoding the phage TFs TF72, TF63, XREApop–LuxRApop and XREARM81ld–LuxRARM81ld all required synthetic induction to drive lysis, a situation that mirrors our findings for VqmAPhage in phage VP882 (ref. 19). Although not investigated here, it is possible that these regulators are produced in infected cells and they function to activate lysis before any additional newly infecting phages can establish lysogeny. Determining the roles of these modules in all stages of the phage life cycle is a focus of our ongoing work. Thus far, we have not succeeded in identifying the natural inducers of these TF–smORF cascades. However, we note that DNA-damaging compounds, the universally used phage inducers, do not generally occur in nature at the concentrations used in experiments. Thus, the identities of the natural cues that drive lysis even in intensively studied (that is, SOS-dependent) prophages remain unknown. Presumably, the signals that induce the additional pathways to lysis revealed here, as well as others that are identified going forward, could be particular to each host–phage partnership and the niche in which they reside. We propose that although discovered long after the SOS cue, some or all of these TF–smORF pathways may be key determinants of phage lifestyle transitions and the arbiters of inter-prophage competition in real-world settings.


Sequence retrieval, RepA–TelN and ParB-associated loci

To identify examples of phage genomes with convergently oriented telN and repA genes, we examined sequences from the following databases: NCBI nt, IMG/VR v.3 (specifically the file IMGVR_all_nucleotides.fna)28, Cenote Human Virome Database v.1.1 (CHVD_clustered_mash99_v1.fna)24, Global Ocean Virome 2 database (GOV2_viral_populations_larger_than_5Kb_or_circular.fna)42, the Gut Phage Database (GPD_sequences.fna)25, a curated set of linear plasmid-phages (retrieved through NCBI accession numbers in table S4 of ref. 26) and the Metagenomic Gut Virus Catalog (mgv_contigs.fna)29. In February 2022, most databases were searched using both tBLASTn and profile-HMM-based search strategies (the exception being NCBI nt, which was too large to annotate anew using profile HMMs). A manual examination of established lysis-control loci19 revealed that the associated telN gene always encoded a protein that matched closely to Pfam profile PF16684, whereas the following set of seven families encompassed the diversity of observed repA sequences: PF13362, PF02399, PF08707, PF10661, PF02502 and PF13604. Thus, we extracted these seven families from Pfam (v.34)43 into a custom HMM database. We then used the gene finder MetaGeneMark9 to predict ORFs from the aforementioned nucleotide files using default parameters. The resulting proteins were used in a profile HMM search with HMMER3 (ref. 44) against the repA/telN custom database, and nucleotide sequences were extracted that contained both telN and repA genes only if they were convergently oriented and within 10 kb of one another. To complement this search strategy, we also used the predicted TelN and RepA proteins from vibriophage VP882 in a tBLASTn search against these nucleotide databases (NCBI accession numbers YP_001039865 and YP_001039868, respectively). We retained only the hits with e-values better than 0.001, which also covered >50% of the query sequence. For consistency, we re-annotated all retrieved sequences using a common method, as follows. We used the gene-finder MetaGeneMark45 to predict ORFs using default parameters. We next used their amino acid sequences in a profile HMM search with HMMER3 (ref. 44) against TIGRFAM46 and Pfam43 profile HMM databases. The highest scoring profile was used to annotate each ORF. As above, we further refined our database by considering only contigs with convergently oriented telN and repA genes within 10 kb of one another. The HMM-based and tBLASTn-based search strategies produced highly redundant (but not identical) sequence sets, as the same databases were used and many identical sequences are listed across multiple databases. Thus, we dereplicated our combined sequence files using cd-hit-est with the following parameters: ‘-c 1.0 -aS 1.0 -g 1 -d 0’. We manually examined the DNA sequence located between each telN and repA gene from this dereplicated set, extracting the intervening nucleotides to a second locus-specific dataset. We dereplicated these sequences as described above to produce the final set of 274 loci referenced in the text and detailed in Supplementary Table 1. The NCBI nucleotide accession that encodes phage 63 (NZ_AIDA02000063) is dated 20 March 2022, after our initial February search. The sequence for phage 72 appears more than once in the NCBI nucleotide database (NZ_AIDA02000072, dated 20 March 2022, and KP795532, dated 18 November 2019). Presumably, these similar entries result from separate sequencing analyses of identical (or similar) bacteria. For these reasons, our February BLAST search only initially revealed phage 72, whereas phage 63 was subsequently discovered.

To investigate phage genomes for potential TF63-associated marker genes, we first performed a preliminary BLASTp search against the NCBI nr/nt database using TF63 as a query (accession number WP_016786069). We filtered the top 5,000 hits from this search for those with over 35% amino acid identity across 70% of the query sequence and retrieved the corresponding nucleotide file in NCBI, retaining the DNA ±50 kb from the gene boundaries. This analysis produced 744 sequences that were then filtered for those of predicted phage origin using the phage prediction tool VIBRANT34 (v.1.2) with default parameters. VIBRANT indicated that 200 sequences in this set were putatively phage-derived. Some of these putative phage TF63 sequences aligned poorly to our original query, so a second tBlastn search was performed using the original TF63 protein. From the 200 putative phage genomes, only those with tBlastn hits that had an e-value better then e−10 were retained. Finally, we removed one short sequence (<5 kb) and dereplicated the remaining phage sequences at 95% nucleotide identity with cd-hit-est using the following parameters: ‘-c 0.95 -aS 0.95 -g 1 -d 0’. This strategy produced 117 sequences that were then uniformly annotated as described above and manually categorized into the genomic context groups described in Supplementary Table 2. The phage prediction tool VIBRANT was also used to assign a likely phage origin to the sequences listed in Supplementary Table 1, as described above.

To identify examples of putative lysis-control loci associated with the parB gene, we again performed a BLASTp search against the NCBI nr/nt database, this time using the ParB protein from phage 63 as a query (accession number WP_016786072). From the top 10,000 proteins revealed by this search, we retrieved the corresponding nucleotide file in NCBI and examined the locus surrounding parB by downloading DNA ±50 kb from the gene boundaries (this strategy retrieved the full contig sequence for many of the associated nucleotide files). To identify parB genes associated with the phage structural locus present in phage 63, we used tBLASTn to query these sequences with the major capsid protein from this phage (accession number WP_016786053), which is one of the most conserved phage protein folds47. We filtered for hits with over 25% amino acid identity across 70% of the query sequence, revealing 121 putative parB loci present on 118 distinct sequences. These 121 loci were dereplicated as described for telN and repA and filtered for the presence of a full-length parB gene and a predicted TF within 10 kb of parB. This analysis produced 56 sequences, which are referenced in Supplementary Table 3.

Phylogenetic analysis of TelN and ParB

From a manual examination of the retrieved RepA-encoding and TelN-encoding sequences, we observed that all TelN proteins shared the same profile HMM as their top-scoring annotation (PF16684.8). This pattern stood in contrast to predicted RepA-encoding sequences, which hit best to one of six different protein families. The consistent TelN annotation indicated to us that this protein may be conserved despite extensive sequence divergence among the phage genomes considered, and thus, it provided a good phylogenetic bellwether for this locus. Indeed, the set of non-redundant, full-length TelN proteins from these datasets aligned well using MUSCLE with default parameters48. A preliminary phylogenetic tree was generated from these sequences using the Geneious Tree Builder per the UPGMA clustering method and with a Jukes–Cantor distance model. From this tree, a subset of 34 sequences was selected to best represent the full TelN phylogenetic tree and the genetic diversity encoded by its associated gene neighbourhoods. Figure 1a depicts a phylogenetic tree generated from these 34 representative sequences, which was produced using PhyML with the LG substitution model and bootstrapped 100 times. A similar procedure was used to produce the ParB phylogeny depicted in Fig. 3a. In this case, a preliminary phylogenetic tree was produced from the set of 56 ParB proteins in the full dataset using the Geneious Tree Builder per the UPGMA clustering method and with a Jukes–Cantor distance model. From this tree, a subset of 28 sequences was selected to best represent the full ParB phylogenetic tree and the genetic diversity encoded by its associated gene neighbourhoods. This final ParB tree was produced using PhyML with parameters identical to those used for TelN.

Prediction of autoproteolytic cI proteins

All 271 predicted cI proteins were used to perform a batch search of the NCBI Conserved Domain (CD) database. The output of this search included conserved domains and other protein features for each cI protein queried. Among these features were predicted catalytic sites that corresponded to known catalytic residues in canonical autoproteolytic cI proteins14. Thus, we used this feature table to define the 61 predicted autoproteolytic cI proteins and labelled the remaining 210 proteins as putatively non-autoproteolytic. The clustering analysis presented in Supplementary Table 1 (summary statistics tab) was performed using the DNA sequences between repA and telN genes and grouped sequences that shared 80% nucleotide identity over 95% of the length of the shorter sequence in the same cluster. We used the program ‘cd-hit-est’ with the following parameters: -c 0.8 -aS 0.95 -g 1 -d 0.

Phylogenetic analysis of linear plasmid-like phages encoding par-associated xre modules in Shigella and Aeromonas

NCBI BLASTp of the DNA sequence encoding XREApop was used as a representative of the three Aeromonas phages identified in the par-associated analysis (detailed above) retrieved three predicted linear plasmid-like phages in Shigella. For each phage genome, genes were called using Prodigal (v.2.6.3)49, and gene diagrams were constructed. Genes were annotated using Prokka (v.1.11)50, and annotations were supplemented with NCBI BLASTp searches by hand51. The phage genome tree (Extended Data Fig. 3a) was constructed using VICTOR52, producing an average support of 78%. The numbers above branches are pseudo-bootstrap support values from 100 replications.

Bacterial strains and growth conditions

E.coli and Aeromonads were grown with aeration in Luria–Bertani (LB-Miller, BD-Difco) broth. Vibrio strains were grown in LB medium with 3% NaCl. All strains were grown at 30 °C. Strains used in the study are listed in Supplementary Table 4. Unless otherwise noted, antibiotics, were used at the following concentration: 100 μg ml−1 ampicillin (Sigma), 50 μg ml−1 kanamycin (GoldBio) and 5 μg ml−1 chloramphenicol (Sigma). Inducers were used as follows: E.coli: 200 μM isopropyl β-d-1-thiogalactopyranoside (IPTG, GoldBio) and 50 ng ml−1 aTc (Clontech); Vibrios: 500 ng ml−1 ciprofloxacin (Sigma) and 50 ng ml−1 aTc; Aeromonads: 1 μg ml−1 ciprofloxacin and 5 ng ml−1 aTc. C4-HSL was supplied at a final concentration of 10 μM except for the experiment shown in Fig. 3b, in which it was administered at specified doses, as indicated in the figure.

Cloning techniques

All primers and dsDNA (gene blocks) used for plasmid construction, qPCR and electrophoretic mobility shift assays (EMSAs) listed in Supplementary Table 5 were obtained from Integrated DNA Technologies or Twist Bioscience. Gibson assembly, intramolecular reclosure and traditional cloning methods were used for all cloning. PCR with iProof was used to generate insert and backbone DNA. Gibson assembly relied on the HiFi DNA assembly mix (NEB). The Apop-based and ARM81ld-based PRlux reporter constructs in E.coli (JSS-3346k and JSS-3348k, respectively) required the addition of a second copy of the cI repressor under its native promoter in the plasmid for function (Supplementary Table 5). All enzymes used in cloning were obtained from NEB. Plasmids used in this study are listed in Supplementary Table 6. Transfer of plasmids into Vibrio 1F-97, A.popoffii, and Aeromonas sp. ARM81 was carried out by conjugation followed by selective plating on TCBS agar supplemented with kanamycin, LB plates supplemented with ampicillin and kanamycin, and LB plates supplemented with kanamycin and chloramphenicol, respectively.

Growth, lysis and reporter assays

Overnight cultures were back-diluted 1:100 with fresh medium with appropriate antibiotics before being dispensed (200 μl) into 96-well plates (Corning Costar 3904). Cells were grown in the plates for 90 min before ciprofloxacin, aTc or C4-HSL was added as specified. Wells that did not receive treatment received an equal volume of water or DMSO. A BioTek Synergy Neo2 Multi-Mode reader was used to measure OD600 and bioluminescence. RLUs were calculated by dividing the bioluminescence readings by the OD600 at that time.


Overnight cultures were back-diluted 1:100 and grown for 90 min before administration of aTc or ciprofloxacin at the indicated concentrations. In Extended Data Figs. 1d and 7f, cells were collected at T = 0 and 15 min, and in Extended Data Fig. 1h, at T = 0 and 90 min following induction. Collected cells were treated with RNAProtect Bacteria Reagent (Qiagen) according to the supplier’s protocol. Total RNA was isolated from cultures using an RNeasy Mini kit (Qiagen). RNA samples were treated with DNase using a TURBO DNA-free kit (Thermo). cDNA was prepared as described using SuperScriptIII Reverse Transcriptase (Thermo). SYBR Green mix (Quanta) and an Applied Biosystems QuantStudio 6 Flex Real-Time PCR detection system (Thermo) were used for real-time PCR. Each cDNA sample was amplified in technical quadruplicate and data were analysed by a comparative CT method, in which the indicated target gene was normalized to an internal bacterial control gene (rpoB).

qPCR and viral preparation

Viral preparations consisted of non-chromosomal DNA (RQ1, RNase-Free DNase, Promega) prepared from 1 ml of cells of the indicated strains. Overnight cultures were back-diluted 1:100 and grown for 90 min before being divided into 3 equal volumes and exposed to treatments as specified. Cultures were grown for an additional 5 h before collection of cell-free culture fluids (Corning SpinX). qPCR reactions were performed as described above for RT–qPCR assays. Next, 1 µl of purified non-chromosomal DNA was used for each qPCR reaction. Data were analysed by normalizing the CT values of samples treated with ciprofloxacin or aTc to the CT values of samples treated with water using a primer set to the indicated phage. Viral preparations for whole-genome sequencing (SeqCenter) were prepared exactly as described for qPCR, except by column purification (Phage DNA Isolation Kit, Norgen Biotek).

Small-molecule screening for PsmORF 72lux and PsmORF 63lux inducers in Vibrio 1F-97

Overnight cultures of Vibrio 1F-97 strains harbouring a plasmid encoding either PsmORF72lux or PsmORF63lux were back-diluted 1:100 and grown for 90 min before being dispensed into 23 × 96-well Biolog plates (PM3b-PM25) and 3 × 96 well plates of curated compounds from the Seyedsayamdost Group (Princeton). Plates were incubated at 30 °C overnight before measurement of OD600 and bioluminescence. Biolog conditions or library compounds that inhibited growth were retested in a Vibrio 1F-97 strain harbouring a reporter for phage lambda induction (cI–PRlux)53. The lambda reporter is SOS responsive but not responsive to a TF–smORF module. The goal was to determine whether any condition or compound was a general phage inducer.


Custom Stellaris RNA FISH probes were designed against sets of early lytic genes in each phage under study using the Stellaris RNA FISH Probe Designer v.4.2 (LGC, Biosearch Technologies; Supplementary Table 7). Probes were labelled with Quasar 670 (phage 63 and phage ARM81ld) or CAL Fluor Red 590 (phage 72 and phage ARM81mr) (Biosearch Technologies). smFISH was performed as previously described54, with minor modifications. In brief, cells were fixed with 1 ml cold 1× PBS and 3.7% formaldehyde and mixed at room temperature for 30 min. Samples were subjected to centrifugation at 400g for 8 min and the clarified supernatant discarded. Cells in the pellets were washed twice with 1× PBS and subjected to centrifugation at 600g for 3.5 min following each wash. Cells in the pellets were resuspended in 300 µl water and permeabilized by the addition of 700 µl 100% ethanol with mixing at room temperature for 1 h. Samples were subjected to centrifugation at 600g for 7 min and the supernatant discarded. Cells in the pellets were resuspended in 1 ml Stellaris RNA FISH wash buffer A (Biosearch Technologies) containing 10% formamide, mixed for 5 min at room temperature, and the cells were collected by centrifugation at 600g for 7 min. Cells in the pellets were resuspended in 50 µl Stellaris RNA FISH hybridization buffer containing 10% formamide and 4 µl of each probe stock (12.5 µM) and incubated overnight at 37 °C. A 10 µl aliquot of the hybridization reaction was added to 200 µl wash buffer A followed by centrifugation at 600g for 3.5 min. Samples were resuspended in 200 µl wash buffer A, incubated for 30 min at 37 °C and subjected to centrifugation at 600g for 3.5 min. This step was repeated. Cells were stained with 50 µg ml–1 DAPI in wash buffer A for 20 min at 37 °C, washed with 200 µl Stellaris RNA FISH wash buffer B and resuspended in 10 µl 1× PBS. Next, 1 µl aliquots of these cell samples were dispensed into No. 1.5 glass coverslip bottomed 24-well plates (MatTek), along with 30 µl VectaShield mounting medium and the samples were covered with agarose pads. Imaging was performed with a Nikon Eclipse Ti2 inverted microscope equipped with a Yokogawa CSU-W1 SoRa confocal scanning unit. Samples were imaged with a CFI Apochromat TIRF ×60 oil objective lens (Nikon, 1.49 numerical aperture) with excitation wavelengths of 405, 561 and 640 nm and with 0.4 µm (for Aeromonas phages) or 1 µm (for Vibrio phages) z-steps. Images were captured through a ×2.8 SoRa magnifier. Images were processed using Nikon NIS-Elements software.

Cells in the images were segmented in Fiji software from maximum z-projections of the DAPI channel. Groups of cells that could not be accurately resolved were excluded from downstream analysis and coordinates of remaining cells were exported. smFISH data were analysed using custom Python scripts. In brief, images were convolved with a Gaussian function to remove noise. Spots were next detected as local maxima with intensities greater than a threshold set based on a negative control image (negative controls for each probe set are as follows: phage 72, smORF63 induced; phage 63, smORF72 induced; phage ARM81mr, smORFARM81ld induced; phage ARM81ld, no induction). Each spot was fitted with a 3D Gaussian function to determine the integrated, background-subtracted spot intensity. In instances with multiple spots residing in close proximity, a 3D multi-Gaussian fit was performed. Spots were assigned to cells and the summed intensity from all spots in a cell were reported. Cells with total intensities ≤1 were assigned a pseudovalue of 1. Supplementary Table 8 provides summary data for Vibrio 1F-97 phage 72 and phage 63 and Supplementary Table 9 provides summary data for Aeromonas sp. phage ARM81ld and phage ARM81mr.

In vitro HALO-cI repressor cleavage and in-gel HALO detection

Assessment of cleavage of HALO-cI proteins in response to DNA damage or smORF induction was carried out in E.coli according to a previously described method19, with minor modifications. In brief, overnight cultures of E.coli T7Express lysY/Iq carrying the indicated HALO fusion plasmid and cognate, aTc-inducible smORF vector were diluted 1:200 in medium and grown for 2.5 h with shaking. IPTG (200 µM) was added to the cultures before they were divided into 3 equal volumes followed by administration of the relevant treatment as specified. The treated cultures were incubated without shaking for an additional 2.5 h. Cells were collected by centrifugation (16,100g for 1 min), resuspended in BugBuster containing 1 µM HALO-Alexa660 (excitation/emission of 663/690 nm). The cleared supernatant, collected after centrifugation of the lysate (16,100g for 10 min), was loaded onto a 4–20% SDS–PAGE stain-free gel. Gels were imaged using an ImageQuant LAS 4000 imager under the Cy5 setting for HALO-Alexa660. Exposure times never exceeded 30 s.

Protein production and purification

Plasmids harbouring genes encoding HIS–HALO–cIApop and HIS–HALO–cIARM81ld were introduced into E.coli BLR21(DE3) (Millipore Sigma) and a plasmid carrying HIS–HALO–cI63 was introduced into E.coli BL21(DE3) (Invitrogen). For protein production, the strains were grown overnight at 15 °C with 1 mM IPTG. In all cases, cells were pelleted at 3,000g for 10 min followed by resuspension in buffer A (150 mM NaCl, 20 mM Tris pH 7.5, 1 mM TCEP). Complete EDTA-free protease inhibitor cocktail tablets (Millipore Sigma) and 10 units of DNase I (Thermo) were added to each resuspended pellet and the cells were lysed by sonication. The insoluble fractions were separated from the soluble material by centrifugation of the lysates at 26,000g for 40 min. The soluble fractions were collected and applied to Ni-NTA Superflow resin (Qiagen). The resin was washed with 5 column volumes (CVs) of buffer A. Proteins were eluted in 2.5 CVs of buffer A and a gradient of 100–300 mM imidazole. The HIS–HALO tags were cleaved from the cIApop, cIARM81ld and cI63 proteins by treatment with 1 mg HIS-TEV Plus protease overnight at 4 °C. The samples were re-applied to Ni-NTA Superflow resin and the proteins were captured at around 85–95% purity by washing the resin with 1 CV of buffer A.

cI72–HALO–HIS protein was produced in E.coli BLR21(DE3) by growth at 37 °C for 3 h with 1 mM IPTG. cI72–HALO–HIS protein was purified from cell pellets as described above for the other cI proteins; however, to circumvent insolubility and aggregation, the HALO–HIS tag was not cleaved. Following initial purification on Ni-NTA Superflow resin, the eluate was concentrated and loaded onto a Superdex-200 size exclusion column (GE Healthcare) in Buffer A. cI72–HALO–HIS at about 85% purity was collected.


DNA probes PR63, PR72, PR-Apop and PR-ARM81ld were generated using plasmids JSS-3131, JSS-3129, JSS-3346k and JSS-3348k, respectively, and the primers listed in Supplementary Table 5. Each EMSA reaction contained 20 ng of DNA probe (8–14 nM depending on the probe). The concentration of cIApop and cIARM81ld protein used in each reaction ranged from 200 nM to 800 nM. Higher concentrations of the cI63 and cI72–HALO–HIS proteins (800 nM to 3,200 nM) were required owing to their limited solubility. The cI proteins with their respective probes were combined in binding buffer (50 mM NaCl, 20 mM Tris pH 7.5, 1 mM TCEP) and incubated at room temperature for 15 min. The samples were subjected to electrophoresis on a Novex 6% DNA retardation gel (Thermo) in 1× TBE at 100 V for 45 min. Double-stranded DNA was stained with SYBR Green I nucleic acid gel stain (Thermo) for 20 min. After washing with 1× TBE, gels were imaged using an ImageQuant LAS 4000 imager under the SYBR Green setting.

AlphaFold and DALI structural prediction of phage TF72 and TF63

AlphaFold2 (ref. 32) and AlphaFold-Multimer55 were used to predict the structures of TF72 and TF63 from the protein sequences. The predicted structures were uploaded to the DALI server56 for a heuristic PDB search. Structural predictions and homologues predicted by the DALI server were aligned and visualized using PyMOL57.

Quantitation and statistical analyses

The following software were used to collect and analyse data generated in this study: GraphPad Prism 9 for analyses of growth and reporter-based experiments; Gen5 v.3.11 for collection of growth and reporter-based data; MetaGeneMark v.3.26, HMMER3 v.3.3.2, VIBRANT v.1.2, MUSCLE v.3.8.31, Geneious Prime v.2022.2.2, SnapGene v.6, Prodigal 2.6.3, PhyML v.3.3.20180621, CD-HIT v.4.8.1 and BLAST 2.13.0+ for analyses of publicly available data and primer design; QuantStudio for qPCR data collection; Nikon NIS-Elements for acquisition of FISH micrographs; and FIJI v.2.9.0/1.53r and Python v.3.7.6 for image analyses. AlphaFold 2.1.1, PyMOL Open GL 2.1 and DALI (accessed 26 November 2022) were used for protein structural predictions and analyses. Data are presented as the mean ± s.d. with n = 3 biological replicates starting from separate bacterial colonies measured on the same day. The number of technical and independent biological replicates for each experiment are indicated in the figure legends.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.