Evolutionary Stabilization of Cooperative Toxin Production through a Bacterium-Plasmid-Phage Interplay

Single-cell analysis of prophage-mediated S. Tm lysis under colicin-inducing conditions.We previously showed that the gene encoding ColIb (cib) in S. Tm is unimodally expressed in the population upon induction (18). This observation was somewhat surprising, as in contrast, group A colicins are expressed bimodally, fitting a division of labor strategy (10, 11). Since ColIb is released by prophage-mediated lysis, we hypothesized that this mechanism may introduce stochasticity and consequently heterogeneous population behavior. To address this, we first quantified expression of ColIb (cib) and prophages at the single-cell level using a set of fluorescent gene reporters (see Tables S1 and S2 in the supplemental material). S. Tm SL1344 harbors four prophages: the lambdoid prophages Gifsy-1, Gifsy-2, and ST64B as well as the P2-like prophage SopEΦ (14). In the wild-type S. Tm (S. Tm^WT), prophages are induced by the SOS response upon addition of the DNA-damaging drug mitomycin C. A gfp reporter plasmid (p^{Pcib gfp}), reporting ColIb production at the single-cell level (18) was transformed into wild-type S. Tm SL1344 (S. Tm^WT) and prophage-cured (S. Tm^ΔPh) cells (Tables S1 and S2). Bacteria were grown in liquid culture over the course of 4 h under ColIb/prophage-inducing (iron-chelatin diethylenetriaminepentaacetic acid [DTPA] and mitomycin C, see Materials and Methods) and noninducing conditions, and gfp expression was analyzed by fluorescence-activated cell sorting (FACS). Prophage-deficient S. Tm^ΔPh p^{Pcib gfp} exhibited substantially higher green fluorescent protein (GFP) mean fluorescence intensities (MFI) compared to S. Tm^WT under all investigated conditions, particularly in the presence of mitomycin C (Fig. 1B). A size-dependent effect regarding the differences in the MFI values between S. Tm^WT p^{Pcib gfp} and S. Tm^ΔPh p^{Pcib gfp} was ruled out by plotting the cell size (forward scatter [FSC]) against the MFI (see Fig. S1 in the supplemental material). This observation confirmed our previous data (14) indicating a scenario where ColIb (or GFP in this case) is not released in the absence of prophages and therefore accumulates inside the bacteria.

Next, we investigated the gfp expression dynamics of individual S. Tm^WT p^{Pcib gfp} and prophage-cured S. Tm^ΔPh p^{Pcib gfp} cells under ColIb/prophage-inducing conditions using time-lapse microscopy in a CellASIC ONIX microfluidic platform. S. Tm microcolonies were grown in microfluidics chambers under ColIb-inducing conditions. Notably, lysis of individual S. Tm^WT cells was observed already within 2 h after mitomycin C exposure, while a part of the cib-expressing population did not lyse until the end of the measurement (Fig. 2A; see also Movie S1 in the supplemental material). In contrast, no lysis was observed in case of prophage-cured S. Tm^ΔPh, even >3 h after mitomycin C exposure (Fig. 2B; Movie S2).

• Open in new tab
• Download powerpoint

FIG 2

A fraction of the ColIb-producing S. Tm population simultaneously expresses ST64B prophage lysis genes. (A and B) Live-cell microscopy using the Onix microfluidic system (Millipore). Microcolonies of S. Tm^WT p^{Pcib gfp} (A) and prophage-deficient S. Tm^Ph p^{Pcib gfp} (B) were cultured in microfluidic plates for 3 h in LB only. Subsequently, medium was exchanged with LB supplemented with 100 μM DTPA for 20 min to induce cib expression. Next, LB supplemented with 100 μM DTPA and 0.5 μg/ml MitC was applied to induce the bacterial SOS response and full expression of cib and induce the prophage(s). After 20 min, cells were again exposed to LB supplemented with 100 μM DTPA for the remaining time of the experiment. Microscopic images obtained by live-cell microscopy (GFP-channel only) of S. Tm^WT p^{Pcib gfp} (A) and prophage-deficient S. Tm^Ph p^{Pcib gfp} (B) microcolonies. White arrows point to bacteria that are about to lyse. (C) S. Tm^WT p^{Pcib gfp} (blue) and S. Tm^{lysST::T7 pol} p^{PT7 sfgfp} (orange) were grown in a 96-well plate in either LB with respective supplements (0.5 μg/ml MitC; 100 μM DTPA) for 4 h. Fluorescence signal intensities of GFP (cib) and sfGFP (ST64B prophage lysis genes) were determined by flow cytometric analysis, respectively. Background fluorescence signals of S. Tm^WT (black), grown under the same conditions, are shown as control. (D) Quantification of percent GFP (cib) “ON” cells from three replicate FACS experiments as described above for panel C.

Prophages introduce phenotypic heterogeneity in S. Tm lysis.On the basis of the results of the time-lapse microscopy analysis, we reasoned that phenotypic heterogeneity in the ColIb system is introduced at the level of prophage-mediated cell lysis and therefore colicin release, rather than at the level of cib expression. S. Tm prophage ST64BΦ is from S. Tm ST64BΦ is a lambdoid phage often found in S. Tm virulent strains and of major importance in ColIb release (14, 19, 20). A double reporter strain (S. Tm^{lysST::T7 pol} p^{PT7 rfp Pcib gfp}) was constructed to simultaneously monitor ColIb production and induction of ST64BΦ lysis genes on the single-cell level (Tables S1 and S2). This strain relies on a signal amplification system based on the T7 polymerase gene T7 pol, integrated in the ST64B prophage lysis genes, and a plasmid-encoded red fluorescent reporter gene tagRFP-T under the control of the T7 promoter in addition to the ColIb gfp reporter. Therefore, red fluorescence indicates activation of ST64BΦ lysis genes and GFP indicates the induction of ColIb expression in the same bacterial cell. This reporter unequivocally shows that only a small fraction of ColIb-producing (GFP-positive [GFP⁺]) cells also expressed the lysis genes (tag RFP-T ⁺; Fig. S2 and Movies S3 and S4). In order to quantify the relative frequency of ST64B prophage lysis gene induction in S. Tm via FACS analysis, we generated another reporter strain, S. Tm^{lysST::T7 pol} p^PT7sfgfp (Tables S1 and S2). Again, signal amplification is based on the T7 polymerase gene T7 pol, integrated in the ST64B prophage lysis genes, and a T7 promoter-superfolder gfp (sfgfp) fusion on a plasmid. Therefore, ST64BΦ lysis gene induction is now monitored via sfgfp expression. Using this reporter strain and S. Tm^WT p^{Pcib gfp} (Fig. 1B and Fig. S1), it was confirmed that under ColIb/prophage-inducing conditions (DTPA, mitomycin C), ST64BΦ lysis genes are expressed only in a subset of the S. Tm population (20.13%), while cib is expressed unimodally (Fig. 2C and D) in about 80% of the population, as shown previously (18). In summary, using different reporter strains, it was shown that S. Tm expresses cib unimodally, while prophage lysis genes are expressed bimodally.

ColIb production does not affect bacterial growth rate.Although the ColIb system does not mediate costly cell lysis itself, ColIb production may have a negative impact on S. Tm fitness by imposing a metabolic burden. Moving to study the effect of ColIb production on S. Tm growth, we extended our strain tool set. For this, we used S. Tm strain ATCC 14028 (S. Tm^{WT 028}) as the basis because this strain originally does not carry a ColIb plasmid. From there, we generated a derivative carrying ST64BΦ only (S. Tm^{WT 028 ΔGif 1,2,3}) and a prophage-deficient variant (S. Tm^{WT 028 ΔPh}). All strains were transformed with pColIb (p2; Table 1) to explore the effect of ColIb on S. Tm fitness in different environments. We also cured the strains from P2 to address whether previous carriage of P2 had any effect on the fitness of the strains, e.g., by compensatory mutations. We grew cultures under control and various ColIb-inducing conditions and recorded optical density (optical density at 600 nm [OD₆₀₀]) as a proxy for growth hourly. During 4 h of growth, the most prominent differences between WT and phage-deficient S. Tm strains appeared under SOS-inducing conditions (Fig. S3). Here, S. Tm^{WT 028 ΔGif 1,2,3} carrying ST64BΦ only grew to slightly lower OD₆₀₀ in the first 2 h of the experiment, followed by a sudden decline in OD₆₀₀ after 2 h under SOS-inducing conditions. The prophage-deficient variant S. Tm^{WT 028 ΔPh} showed positive growth rates over the entire 4 h. Notably, comparing the growth rates of strains with and without pColIb, we find no difference in any of the investigated conditions (Fig. S3). This suggests that carriage of pColIb and toxin expression does not measurably affect S. Tm growth rates.

Ecological model of the tripartite interaction of S. Tm, the colicin plasmid, and a temperate phage in a group B colicin system.In the experiments described above, we demonstrated that prophage induction and subsequent ColIb release by cell lysis take place only in a subpopulation of ColIb-producing S. Tm. Even though ColIb increases the competitiveness of an S. Tm population (21), from an ecological perspective, colicin release by cell lysis is extremely costly. This presents a dilemma, as individuals then have an incentive to evade the cost of death or resulting lower growth rates due to colicin production, while the populations’ invasion of the gut relies on cooperative release of ColIb. Consequently, maintaining stability of intrastrain cooperation is a challenge for colicin-producing bacterial populations. As prophages contribute to ColIb release in our case, we wondered under which conditions phage-mediated colicin release is an evolutionary stable strategy in S. Tm.

In order to study the dynamic interdependencies in this tripartite system of Salmonella, colicin, and a prophage during infection of the mammalian gut, we devised a model based on ordinary differential equations (ODEs) (Fig. 3; see also Text S1 in the supplemental material for details of the model). The model takes the following players into account: S. Tm (s), a plasmid encoding the colicin Ib (cib) and colicin Ib immunity (imm) genes, a temperate S. Tm phage, and a competing species (e), as, e.g., commensal E. coli (Fig. 3; Table S3). The mammalian gut is modeled as a chemostat with a periodic inflow of S. Tm (s₀) to depict subsequent infections. The considered time scale of the interaction dynamics is ecological. Therefore, only S. Tm infection dynamics are addressed, while emerging S. Tm mutants are not considered. Even though S. Tm (s) may cause infections of different mammalian hosts in nature, pathogen host-to-host transition was not explicitly considered in the scale of this model.

• Open in new tab
• Download powerpoint

FIG 3

Transition graph for the ecological model of S. Tm, group B colicin plasmid and phage based on ordinary differential equations. The S. Tm population was structured into six compartments, according to the presence of prophage (right column), colicin plasmid (middle row) and induction of colicin expression and lysis (red cells, lower row). The S. Tm population is denoted with s, the superscript indicates the presence (+) or absence (−) of the prophage, and the subscript indicates the presence (+) or absence (−) of the plasmid. The additional c in the subscript indicates that the bacterium is producing colicin. s−−, S. Tm without colicin plasmid and without prophage; s−+, S. Tm without colicin plasmid and with prophage; s+−, S. Tm with colicin plasmid and without prophage, not producing colicin but immune to colicin; s++, S. Tm with colicin plasmid and with prophage, not producing colicin but immune to colicin; s+c−, S. Tm with colicin plasmid and without prophage, producing and immune to colicin, no colicin release; s+c+, S. Tm with colicin plasmid and with prophage producing and immune to colicin, colicin release. Furthermore, free phages (Φ), free colicin (C), and a competitor population (e) are shown. See Text S1 in the supplemental material for further details of the model.

ColIb can increase fitness of the S. Tm population in particular during inflammation-induced Enterobacteriaceae blooms, where S. Tm competes against close relatives (21). During blooms, the S. Tm population, as well as competing species, strongly expand due to increased availability of oxygen and altered nutritional environment (22). This effect was modeled as temporarily increased inflow Ds₀ (t) of all bacterial species. Furthermore, inflammation provides environmental cues (e.g., reactive oxygen species, iron limitation), which trigger ColIb production and prophage-mediated cell lysis and induce expression of the ColIb receptor (cirA) on competing E. coli (21, 23). Thus, we assumed that ColIb production pays off only when S. Tm grows in inflammation-induced blooms. To model inflammation-induced blooms, we further introduced periodic rounds of SOS signal as inducer of colicin expression and phage lysis (fluctuating environment).

Colicin plasmids and temperate phages act as “infectious” agents, which spread mainly by conjugation and the release of free phages Φ upon S. Tm host lysis, respectively (Fig. 3). S. Tm harboring the colicin plasmid (s+−) (Fig. 3, s row) are immune to the colicin, by expression of the cognate immunity protein. However, they do not produce colicin, unless there is an environmental trigger, as induction by SOS stress (Fig. 3, lower row) (s+−_c). The temperate phage is integrated as prophage in the S. Tm genome (s−+). Prophage activation is induced by the bacterial SOS response as well. Therefore, in plasmid and prophage carrying S. Tm (s++), SOS triggers lead to colicin production (s++_c), followed by lysis-mediated release of both colicin and free infectious phages (lysis rate δ). Upon infection of new hosts, the phage lyses new host cells or integrates into the bacterial genome with the probability ρ. Using parameter values determined from a combination of experimental data, literature research, and estimations (Table S3), the resulting ODEs describe the dynamics of the tripartite system of Salmonella, colicin, and a prophage and can now be used to study trait evolution.

Adaptive dynamics analysis shows that phage-mediated lysis in the ColIb system is an evolutionary stable strategy.Evolutionary stability of costly traits is often investigated by in silico competition experiments between players using prototypical, fixed opposing strategies. Thus, evolutionary stability of colicin release via phage-mediated lysis can be studied by investigating the competition of a wild-type strain with a mutant strain that does not lyse at all (17). However, in temperate phages, changes in the lysis rate occur by a stepwise process rather than by a drastic shift. For example, mutations in CI repressor operator binding sites can lead to subtle changes in lysis/lysogeny decisions (24). Evolutionary invasion analysis (adaptive dynamics) uses a set of mathematical modeling techniques to analyze the long-term evolution of a trait and thereby takes the stepwise adaptation of a strategy into account (25, 26). This type of model starts with an arbitrary wild-type (resident) strategy. From time to time, rare mutants (invaders, e.g., with an altered lysis rate) appear and attempt to invade the resident population and thereby might replace the wild type, leading to a change of the residents’ trait. Eventually, a stable state can be reached where no further mutant invasion is possible, which is referred to as the evolutionary stable strategy (ESS) of a trait. Instead of implementing a few prototypical strategies only, adaptive dynamics introduces a whole range of strategies and thereby reveals the direction in which evolutionary forces successively change a specific trait (convergent stability) (25, 26).

Using adaptive dynamics, we analyzed evolutionary stability of lysis from the phages’ perspective by invading mutant phages that have an altered lysis rate compared to the resident phage. Two different scenarios of phage-mediated lysis were considered. (i) Under stress conditions, prophage-mediated lysis of the lysogenic S. Tm population is induced by the SOS response at a given rate. (ii) Upon a new infection by phages, the majority of the S. Tm population lyses, and a certain fraction becomes lysogenic. Therefore, the dynamics of phage-mediated lysis are determined by two parameters, the lysis rate δ of lysogenic S. Tm and the probability ρ to become lysogenic. To investigate evolution of phage-mediated lysis in our ecological model, mutant phages with altered lysis rates δ_m and ρ_m were introduced at very small numbers, appearing as free infectious phages or as prophages in S. Tm. Using the resulting altered set of model equations, the Lyapunov exponent (rate of separation) was determined numerically for every adaptation step. If the Lyapunov exponent was positive, the mutant is able to spread; if it was negative, the mutant dies out. The resulting dynamics of the trait (given by the parameters δ and ρ) are presented as pairwise invasibility plots (PIP) (Fig. 4).

• Open in new tab
• Download powerpoint

FIG 4

Evolutionary stability of the tripartite system of S. Tm, ColIb, and phage. Evolutionary stability of the tripartite system of S. Tm, ColIb, and phage was modeled when stress signals reliably (A and B) or nonreliably (C and D) indicate Enterobacteriaceae blooms. Inflammation-induced Enterobacteriaceae blooms are characterized by high abundance of S. Tm and potential competitors (e.g., E. coli). Environmental cues in blooms stimulate both induction of ColIb expression and prophages and represent a hot spot for colicin-dependent competition and phage transmission. (A) For the first case, it was assumed that the SOS signal (white bar) signals the occurrence of blooms (shaded bar) reliably, i.e., that ColIb and the phage are induced only in periods of Enterobacteriaceae blooms, when potential competitors are present at high abundance and colicin production really pays. (B) Pairwise invasibility plot (PIP) of joint evolution of lysis rate (δ_stress) of lysogenic S. Tm and probability ρ to become lysogenic upon new infection by phages. The arrows within the gray areas indicate how the parameters are changed by evolution, the colored paths represent example paths of mutant populations. The PIP indicates that in case of a reliable signal, evolution drives δstress to arbitrarily large values. (C) In the second case, it was assumed that stress signals Enterobacteriaceae blooms nonreliably, i.e., that ColIb and the phage are also induced in periods in between blooms in the absence of competitors. (D and E) Resulting PIPs of joint evolution of lysis rate (δ_stress) and probability to become lysogenic upon new infection of phages (ρ). The arrows within the gray areas indicate how the parameters are changed by evolution, the colored paths represent example paths of mutant populations, and the solid dot indicates the evolutionarily stable strategy (ESS). Specifically, panel D shows the full tripartite model, and panel E shows a model for only S. Tm and phage without colicin production and no competition. See Text S1 for parameters and details.

Using this approach, we investigated to what extent the reliability of an SOS signal as an indicator of Enterobacteriaceae blooms (representing an environmental condition, where colicin release will help to win against competitors) would influence the evolutionary fate of lysis. First, it was assumed that the SOS signal reliably indicates Enterobacteriaceae blooms, induced by inflammation (Fig. 4A). The resulting PIP indicates that in this scenario, evolution drives the lysis rate δ_stress to arbitrarily large values (Fig. 4B). If the signal was reliable, a higher lysis rate permits a more rapid release of phages during a bloom. As a result, this would allow a mutant with a higher lysis rate to spread and outcompete the wild type.

In the mammalian gut, other stimuli (drugs, metabolites) besides an inflammatory response could trigger the bacterial SOS response. Therefore, the SOS signal may not reliably indicate an Enterobacteriaceae bloom. For this reason, we implemented a second scenario, with an unreliable SOS signal (Fig. 4C). Here, an ESS exists for δ_stress > 0 and ρ > 0 (Fig. 4D). The resulting finite lysis rate reflects that stress is correlated with but not reliably signals inflammation. In this case, it is of advantage when a subpopulation of phage-infected S. Tm remains lysogenic, buffering the population from extinction. Thus, we conclude that the resulting ESS is a phage-mediated bet-hedging strategy caused by the unreliability of the SOS signal.

Colicin release by Salmonella lysis is driven by the selfish nature of temperate phages.We hypothesized that S. Tm temperate phage-dependent lysis is not optimized to release the colicin but to maximize the spread of the phages. To investigate whether colicin release is required to convey evolutionary stability in the phage lysis system, we implemented an alternative scenario without colicin release and without competitor. Notably, a positive lysis rate (δ_stress > 0) was still found to reach an ESS (Fig. 4E). Thus, the release of colicin by S. Tm lysis is not required to convey evolutionary stability of the temperate phage. Obviously, for temperate phages, bacterial death and the concomitant release of phage particles presents a private good. Our results indicate that the selfish behavior of temperate phages is the driving force for the ESS of Salmonella host lysis. The lysis-deficient ColIb system employs the evolutionary stable phage system to become evolutionary stable itself. This raises several interesting questions pertaining to the evolutionary relationship between the two systems, which remain open at the current stage of investigation.

Distribution of colicins and phage lysis gene families on the genus Salmonella.To gain insight into evolution of ColIb release by temperate phages, we next investigated the distribution of phage lysis genes and group B colicins in Salmonella strains more globally, using a genome-based approach. Salmonella enterica is the type species of the genus Salmonella. It comprises six subspecies and more than 2,500 serovars (27). At the time this analysis was performed, 9,499 genome sequences of S. enterica strains were available in the NCBI reference sequence (RefSeq) database, which we analyzed for the presence of colicin and lysis gene families (Table S4). To this end, we generated profile hidden Markov models (HMMs) for group A and group B colicins and phage lysis gene families and screened S. enterica chromosomal and plasmid sequences for matches. The rationale behind screening for phage lysis genes instead of all prophage-related sequences was to primarily capture functional prophages with intact lysis function. To account for the high sequence similarity of different colicin gene families, we filtered putative colicin matches to the best hit per genome. In case of phage lysis gene families, we allowed multiple hits.

Prophages are widely distributed in the genus Salmonella (28). We also found that 9,196 of 9,499 (96.8%) S. enterica genomes have at least one lysis gene family in their chromosomal or plasmid sequence (in total 42,968 matches). Thus, only 303 genomes lack lysis genes at all. In total, 1,213 S. enterica genomes (12.8%) show matches to colicin HMM profiles (Fig. 5). Interestingly, genomes that are lysis positive (n = 9,196) seem to harbor colicin genes (group A and B) more often (12.9%, n = 1,183 with colicin match) than lysis-negative genomes (9.9.% of lysis-negative genomes, n = 30 with colicin match), but this difference is not significant (one-sided Fisher’s test, P = 0.072). Based on profile HMM matches, group B colicins are more abundant in S. enterica genomes (70.1%, n = 850) compared to group A colicins (29.9%, n = 363). We asked whether lysis-positive genomes (n = 9,196) harbor group B colicins more often than lysis-negative genomes (n = 303). In the subset of lysis-positive genomes, 9.0% (n = 830) harbor a colicin from group B, while this is only the case for 6.6% (n = 20) of lysis-negative genomes. Group A colicins are present in 3.8% (n = 353) of lysis-positive genomes and 3.3% (n = 10) of lysis-negative genomes. We conclude that B colicins are not more prevalent in Salmonella enterica genomes that do contain lysis genes (one-sided Fisher’s test, P = 0.084).

• Open in new tab
• Download powerpoint

FIG 5

Distribution of colicin and lysis gene families in 9,499 S. enterica subsp. enterica genomes. The majority of S. enterica genomes contain phage lysis gene families, while colicin is present only in a subset of S. enterica genomes, quantified based on profile HMM matches. Mapping of colicin profile HMMs to S. enterica genomes are filtered based on the best hit, due to the high similarity of some colicin gene families. The 10 most prevalent serovars are annotated.

Further, we stratified the Salmonella enterica genomes by serovar to check for serovar-specific differences in colicin and lysis gene abundances (Fig. 6). The frequency of group B colicins was highest in serovar Heidelberg, Typhimurium, and Derby (in this order). Notably, ColIb was the most prevalent group B colicin. ColIb is highly similar to ColIa at the protein level and cannot be distinguished based on an HMM. Thus, colicins annotated as ColIb in the analysis also comprise ColIa genes. Group A colicins were highly enriched in serovar Weltevreden (colicin E9), Heidelberg, and Derby (E1), but with limited occurrence. Interestingly, no colicins of any type were found in serovar Typhi. This could be accounted for by the fact that S. Typhi causes systemic rather than gastrointestinal infection, where it does not have to compete against the microbiota. Besides, serovar Typhi also did not harbor phage lysins of the spanin type. This suggests that serovar Typhi has a different spectrum of temperate phages. Overall, these data point at serovar-specific distribution and ecology of group A and B colicins in S. enterica.

• Open in new tab
• Download powerpoint

FIG 6

Prevalence of colicin and lysis gene families in S. enterica subsp. enterica serovars. Mapping of colicin profile HMMs to S. enterica genomes are filtered based on the best hit, due to the high similarity of some colicin gene families. Lysis-associated gene families are not filtered by the best hit and are stratified by the serovar. Prevalence of group A and group B colicins and phage lysins are shown for the 10 most common serovars, namely, serovars Typhi (n = 2142), Typhimurium (n = 1088), Enteritidis (n = 807), Newport (n = 285), Heidelberg (n = 267), Derby (n = 172), Dublin (n = 171), Weltevreden (n = 167), Agona (n = 145), and Infantis (n = 144).

Finally, we focused our analysis on serovar Typhimurium, our experimental model system. Notably, only a small set of the 1,088 genomes harbor group A colicins (0.74%, n = 8), while colicin group B is found in 11.8% (n = 128) of genomes. In serovar Typhimurium, lysis-associated gene families seem to be omnipresent (1,087 of all screened serovar Typhimurium genomes). This hindered us from analyzing whether group B colicins are also more frequently associated with prophages than group A. However, it suggests that due to high prevalence of prophages in serovar Typhimurium genomes, this serovar may generally be able to release and benefit from group B colicins.