Herd immunity underlies homologous recombination in stationary phase bacteria

Abstract

Homologous recombination is ubiquitous across evolution, generating variation through genetic reshuffling. Although bacteria lack meiotic recombination, homologous recombination has been known to extensively shape their genomes. However, the mechanisms and ecological conditions sustaining frequent recombination in bacteria remain unclear. Using an Escherichia coli–P1 phage model system, we reveal how frequent recombination emerges from CRISPR-based herd immunity to a generalized transducing phage (DNA vector). Herd immunity maintains phage-susceptible donor bacteria, recipient immune bacteria, and transducing phage, which promotes chromosomal gene flow that exceeds spontaneous mutation by up to two orders of magnitude. Notably, recombination begins as bacterial cultures enter stationary phase, when canonical mutation supply through population growth and DNA replication is strongly reduced, and recombination can be reinitiated by brief fresh nutrient pulses. Genetic analysis showed that stationary-phase recombination requires recA, recB, and recG, connecting this process to a defined homologous recombination pathway, and suggests a plausible transcriptional regulatory link to transition to nutrient limitation. Bacterial herd immunity therefore has an unexpected evolutionary consequence: it can promote homologous recombination by turning phages into de facto pollinators. This mechanism can help explain how high rates of homologous recombination may be sustained across bacterial populations and suggests that bacterial immune systems can shape molecular evolution not only by blocking infection but also by regulating genetic exchange.

Introduction

Homologous recombination (HR) is a pervasive evolutionary process that reshuffles standing genetic variation. By bringing alleles that arise in different individuals into the same genetic background, recombination can accelerate adaptation under clonal interference, reduce the accumulation of deleterious mutations, and increase evolutionary response in changing environments (Muller 1964; Charlesworth 1976; Otto and Lenormand 2002; Roze 2021). In population-genomic data, frequent HR can reduce linkage disequilibrium and generate near-random associations among loci (Cutter 2019). Although bacteria lack meiotic recombination, a majority of bacterial species show extensive signatures of homologous exchange. This frequent HR is evidenced by approximate linkage equilibrium in some bacterial populations (Maynard-Smith et al. 1993; Arnold et al. 2020; Birzu et al. 2025), rapid turnover of local phylogenies and decay of local correlations including linkage disequilibrium (Sakoparnig et al. 2021; Preska Steinberg et al. 2022; Wolff and Garud 2026), high ratios of recombination to mutation (r/m) (Torrance et al. 2024), and large recombined DNA blocks across bacterial genomes (Sakoparnig et al. 2021; Liu and Good 2024). Because bacterial HR is often biased toward exchange among close relatives, it can contribute to population cohesion and species integrity, while ecological divergence can restrict gene flow and reinforce genetic isolation (Shapiro and Polz 2015; Bobay and Ochman 2017; Arevalo et al. 2019; Conrad et al. 2024). Given that bacteria constitute most of the phylogenetic diversity represented in genome-based reconstructions of the tree of life (Hug et al. 2016), a central unresolved question of broad evolutionary relevance is therefore not whether HR shapes bacterial evolution but how bacterial populations can carry out such frequent homologous exchange.

Bacteria exchange DNA through several horizontal gene transfer (HGT) mechanisms, but none alone can readily explain the widespread genome-wide high HR rates inferred from natural populations. Natural transformation (uptake and incorporation of extracellular DNA) requires natural competence and opportunity and has been shown to exist for only a very limited subset of bacterial species (Johnston et al. 2014). Conjugation (unidirectional gene transfer) depends on direct cell-to-cell contact and maintenance of conjugative elements; is constrained by spatial structure, plasmid burden, and incompatibility; and only rarely mediates chromosomal HR (Thomas and Nielsen 2005). Generalized transduction (transfer of host DNA via phages) can transfer homologous chromosomal DNA, but it is an infrequent by-product of phage replication: most phage–host encounters result in lysis rather than productive transfer of host DNA, producing lysis-dominated dynamics of the bacterial host population and low transduction probabilities (Kleiner et al. 2020; Borodovich et al. 2022). Other proposed routes, including gene transfer agents and extracellular membrane vesicles, can mediate DNA transfer, but their quantitative contribution to genome-wide HR across natural bacterial populations remains less well established (Fulsundar et al. 2014; Lang et al. 2017). Thus, the ecological conditions and molecular mechanisms that can convert HGT into frequent, sustained bacterial HR remain mysterious.

Bacterial defense systems are likely to be central to this problem. These systems are traditionally viewed as barriers to HGT because they target phages, conjugative plasmids, and other mobile genetic elements (Rocha and Bikard 2022). However, comparative genomic analyses indicate that some defense systems, including restriction–modification systems and several additional defense families, are associated with elevated gene gain on evolutionary timescales, suggesting that defenses can sometimes facilitate rather than simply restrict genetic exchange (Oliveira et al. 2016; Kogay et al. 2024). Experimental work supports this possibility through what can be viewed as a protected-recipient mechanism. CRISPR-Cas immunity (a natural, adaptive immune system in bacteria) can enhance generalized transduction in Pectobacterium atrosepticum by protecting transductants (DNA recipients) from phage-induced lysis (Watson et al. 2018). Similarly, phage-inducible chromosomal islands in Staphylococcus aureus protect transductants and increase acquisition of chromosomal and plasmid-borne markers from purified phage lysates (Ibarra-Chávez et al. 2022). In other S. aureus systems, plasmid transduction occurs efficiently into recipients protected from lytic phage action by lysogenic superinfection immunity and restriction–modification systems (Mašlaňová et al. 2016). Under strong antibiotic selection, lysogenic immunity can also allow lysogens to acquire transduced antibiotic resistance markers from neighboring phage-susceptible bacteria in a population-level process called autotransduction (Haaber et al. 2016).

These studies show that protection from phage-induced lysis can increase transductant recovery. However, because they largely expose protected recipients to transducing particles under controlled assay conditions, they do not test whether immunity can maintain susceptible donors, immune recipients, and transducing phage together in a single evolving population. This distinction is critical: a one-off increase in transductant survival does not by itself explain persistent gene flow, high effective recombination rates, or near-random associations among loci. For bacterial HR to be evolutionarily consequential, polymorphism must be maintained within the system, because recombination can only reshuffle variation that remains present in the population.

Here we test whether this missing population-dynamic mechanism is herd immunity. Herd immunity occurs when immune individuals reduce pathogen transmission sufficiently to indirectly protect susceptible individuals (Fig. 1a). In bacterial populations, such dynamics can arise when a fraction of bacteria carries an immune system that interferes with phage replication. Immune bacteria reduce effective phage spread and thereby protect a susceptible subpopulation (Payne et al. 2018). This creates frequency-dependent selection: immune bacteria increase when phage pressure is high, but selection for immunity weakens near the herd-immunity threshold, allowing susceptible bacteria to persist (Fig. 1b). Importantly, very large bacterial population sizes (N ∼ 10⁹) weaken drift, making stochastic loss of susceptible bacteria unlikely. Then, if the phage is capable of generalized transduction, the protected susceptible bacteria can provide donor DNA, while immune bacteria act as recipients. Herd immunity could therefore maintain the ecological state required for sustained phage-mediated gene flow.

Figure 1.

Herd immunity and recombination dynamics in the experimental system. a) Illustration of principles of herd immunity. Without any immune individuals, each infected susceptible individual (cyan) infects (solid arrows) four more susceptible individuals. Thus, the pathogen spreads exponentially in the populations. With three-fourths of immune individuals (yellow) in the population, transmission of the pathogen is blocked (dashed lines), and some susceptible individuals are protected by herd immunity (cyan in black circles). b) Herd immunity-induced frequency-dependent selection on the immune strain (Immu). Starting from an initial frequency F₀, Immu can only reach the herd immunity threshold frequency H, maintaining the susceptible strain, Sus, in the population. c) Sus, which carries a nontargeting control plasmid encoding Cas9 (black arrow) with no spacers and a chromosomal cassette comprised of cfp and spectinomycin resistance genes (cfp-specR, cyan arrow), is being infected by a wild-type P1 phage (left), giving rise to phage progeny with a fraction of them carrying the inserted chromosomal cassette (right). d) Immu carrying a plasmid with Cas9 (black arrow), two P1 phage targeting spacers (purple squares), rfp gene for red fluorescence (red arrow), and a chromosomal cassette comprised of yfp and kanamycin resistance genes (yfp-kanR, yellow arrow) is being infected by the wild-type phage (left). Upon ejection of the wild-type phage DNA into Immu, the DNA is targeted and cleaved by the CRISPR-Cas9 system (right). e) When a stretch of the Sus donor DNA containing the cfp-specR cassette is injected into Immu (left), the transduced DNA is left intact and can recombine into the host cell chromosome. This gives rise to a strain expressing both chromosomally encoded cfp and yfp cassettes, as well as the plasmid encoded rfp (right).

We use CRISPR immunity as an experimentally tractable representative of bacterial immunity to test this population-dynamic mechanism. CRISPR-mediated herd immunity was shown experimentally (Payne et al. 2018), and CRISPR interference targets phage DNA directly, allowing immune bacteria to survive infectious phage particles while still receiving bacterial chromosomal DNA delivered by transducing particles. We therefore engineered E. coli populations containing CRISPR-immune recipients, phage-susceptible donors, and used the generalized transducing phage P1 to ask (i) whether CRISPR-mediated herd immunity can maintain gene flow, (ii) whether the resulting HR can exceed de novo mutation supply, and (iii) under which growth conditions recombination occurs.

Results

Experimental system for distinguishing HR from mutation

To measure HR and de novo mutation in the same population, we engineered two marked E. coli MG1655 strains. The phage-susceptible donor strain, hereafter Sus, carried a chromosomal cassette containing a cyan fluorescent protein gene and a spectinomycin resistance gene (cfp-specR) at the phage λ attachment site (Fig. 1c). The immune recipient strain, hereafter Immu, carried a chromosomal cassette containing a yellow fluorescent protein gene and a kanamycin resistance gene (yfp-kanR) at the phage P21 attachment site (Fig. 1d). These loci are separated by 389 kbp and therefore cannot be cotransduced by P1, which packages approximately 100 kb of DNA (Thomason et al. 2007).

Because endogenous CRISPR systems of E. coli MG1655 are inactive under standard laboratory conditions (Mitić et al. 2023), we provided Immu with engineered CRISPR immunity against P1. Immu carried a low-copy plasmid expressing Streptococcus pyogenes Cas9 and two spacers targeting the P1 genome. The same plasmid also carried a red fluorescent protein gene (rfp), allowing immune and susceptible bacteria to be distinguished during population-dynamic measurements. Sus carried the corresponding nontargeting control plasmid, expressing Cas9 but lacking P1-targeting spacers and rfp. Growth competition showed no significant difference between Sus and Immu under the experimental conditions (Figure S1).

This design allowed us to distinguish recombinants from spontaneous antibiotic resistance mutants after sampling. Recombinants acquired both chromosomal marker cassettes and therefore grew on kanamycin and spectinomycin while expressing both cfp and yfp. In contrast, spontaneous mutants survived double-antibiotic selection by acquiring resistance to the second antibiotic without acquiring the second fluorescent marker. By plating on selective plates, recombinant and mutant bacteria (measured in colony-forming units, CFUs) could therefore be scored separately after growth in nonselective culture conditions.

The system implements the proposed herd-immunity mechanism as follows: P1 replicates on Sus bacteria and generates progeny that include transducing particles carrying fragments of the Sus chromosome, including the cfp-specR locus (Fig. 1c). In Immu bacteria, infectious P1 DNA is targeted by CRISPR-Cas immunity, allowing the recipient cell to survive phage exposure (Fig. 1d). Transducing particles, however, deliver bacterial donor DNA, not phage DNA; this donor DNA is not targeted by the spacers and can recombine into the Immu chromosome (Fig. 1e). Thus, susceptible bacteria provide donor DNA and maintain the phage population, whereas immune bacteria survive infection and can acquire donor alleles by HR.

Herd immunity enables HR to exceed de novo mutation rate during entry into stationary phase

We first asked whether immune-mediated phage coexistence can enable HR at frequencies comparable to or exceeding de novo mutation supply. Overnight cultures of Immu and Sus were diluted 1:100 into fresh medium at a ratio close to the expected herd-immunity threshold, 95% Immu to 5% Sus, as estimated from our mathematical model (Figures S3, S4, Text). P1 was added at an initial multiplicity of infection (MOI) of 0.1, and cultures were sampled over 24 h as they grew from exponential phase into stationary phase. Spectinomycin and kanamycin were absent during growth, so recombinant and mutant genotypes were selected only after sampling.

Recombinant bacteria (CFUs) began to exceed mutant CFUs at 6 h postinoculation (hpi), when cultures entered stationary phase (Fig. 2a, b). From this point onward, recombinant CFUs continued to accumulate even though the total bacterial population no longer increased and later slightly declined (Fig. 2b). This caused a rapid increase in the frequency of recombinants in the population (Fig. 2c). By 24 hpi, recombinant CFUs were approximately 380-fold more abundant than spontaneous mutant CFUs (Fig. 2a). In contrast, mutant CFUs (Fig. 2a, right) remained comparatively stable, with only modest changes after entry into stationary phase. Thus, in the absence of antibiotic selection during growth, HR generated double-resistant genotypes much more efficiently than de novo mutation as cultures entered stationary phase.

Figure 2.

Recombination dynamics in initially exponentially growing cultures. a) Left: Concentration of bacteria (colony-forming units, CFUs), which survived kanamycin and spectinomycin selection on sampling plates due to recombination of chromosomally encoded cassettes (solid line, Rec.) and concentration of mutant CFUs with only one of the cassettes and a de novo mutation for resistance to the other antibiotic (dashed line, Mut.). Right: Zoom on mutant CFUs. Selection (kanamycin and spectinomycin) was not present in cocultures throughout incubation, only on sampling plates. b) Concentration of total CFUs (semi-log₁₀ scale) increases until 6 hpi, when the cultures reach stationary phase and their concentration starts slowly declining. c) Frequencies of recombinant (solid line) and mutant (dashed line) CFUs. d) Phage-susceptible bacteria remain in cultures throughout the experiment. e) Concentration of viable phages (expressed in plaque-forming units, PFUs) in experimental cultures peaks at 6 hpi, after which phage concentration converges to its equilibrium concentration around 10⁶ PFU/ml. In all panels the shaded areas around the curves show standard error of the mean.

This increase in recombination occurred in cultures where both susceptible donor bacteria and phage were maintained. Sus bacteria persisted throughout the experiment rather than being eliminated by P1 (Fig. 2d), and phage concentration (measured in plaque-forming units, PFUs) peaked during early growth before converging toward a stable concentration of approximately 10⁶ PFU/ml (Fig. 2e). This population structure is expected under herd immunity: immune bacteria reduce effective phage spread enough to protect a susceptible donor subpopulation, while susceptible bacteria maintain the transducing phage.

The coexistence of immune recipients, susceptible donors, and phage was required for frequent recombination. When Immu was replaced by a strain carrying the same yfp-kanR cassette but lacking P1-targeting spacers, recombination was almost absent (Figure S5). Likewise, when the initial Immu:Sus ratio was far from the range predicted to support herd immunity, susceptible bacteria and phage were not maintained, and HR was strongly reduced or lost (Figures S3 to S5). These controls show that recipient protection alone is not sufficient. Frequent HR required the population-dynamic condition in which immune recipients, susceptible donors, and transducing phage coexist.

Repeated nutrient pulses create recurrent windows of recombination

The first experiment showed that recombinant CFUs continued to accumulate after cultures entered stationary phase (Fig. 2a). However, P1 replication is inhibited in nongrowing E. coli (Łoś 2019), so transduction-mediated recombination should eventually decline once phage replication and growth of susceptible donor bacteria stop. Similar growth-phase dependence has been reported for plasmid conjugation in E. coli, where transfer occurs mainly during the transition from exponential growth to nongrowing phase and then plateaus (Headd and Bradford 2020). We therefore asked whether repeated and limited nutrient input can create recurrent opportunities for transduction-mediated HR.

To test this, we modified the culture conditions so that populations reached stationary phase earlier. Overnight cultures of Immu and Sus were mixed at 95% Immu to 5% Sus, inoculated with P1 at MOI = 0.1 and diluted only 1:2 into fresh nonselective medium. This allowed approximately one generation of growth before cultures entered stationary phase. Under these conditions, recombinant CFUs increased rapidly by 3 hpi and significantly exceeded mutant CFUs (Fig. 3a). Recombinant CFUs further increased by 6 hpi and then changed only modestly between 6 and 24 h postinoculation, while recombinant frequency remained approximately stable (Fig. 3b, c). Sus bacteria and phage also persisted throughout the experiment (Fig. 3d, e). Thus, transduction-mediated recombination occurred during entry into stationary phase but did not continue to increase indefinitely once cultures remained nutrient-limited.

Figure 3.

Recombination commences in early stationary phase. a) In cultures, which were initially diluted only 1:2, the concentration of recombinants (solid line, Rec.) starts increasing at 3 hpi and halts after 6 hpi. b) Total CFU concentration increases from initial 1:2 dilution and then fluctuates around stationary phase cell culture concentrations. c) Frequency of recombinants increases between 0 and 6 hpi and then fluctuates around a constant value. d) Phage-susceptible bacteria remain in cultures throughout the experiment. e) Phage concentration slowly converges to its equilibrium concentration around 10⁶ PFU/ml. In all panels the shaded areas around the curves show standard error of the mean.

We next asked whether recombination could be restarted by brief nutrient pulses. Cultures were again inoculated 1:2, and subsets were diluted 1:2 into fresh medium at 12 h and/or 24 h postinoculation. Samples were always taken immediately before dilution, so recombinant counts were measured in saturated cultures rather than during active regrowth. In undiluted controls, recombinant CFUs increased between 0 and 12 h and then did not increase significantly between 12 and 24 h. In contrast, cultures diluted at 12 h showed a significant increase in recombinant CFUs by 24 h. A second dilution at 24 h again produced a significant increase by 36 h, whereas cultures not diluted at 24 h did not show a significant increase over the same interval (Fig. 4). Recombinant frequencies showed the same qualitative pattern, showing that the increase in recombinant CFUs was not simply caused by changes in total population size (Figure S2).

Figure 4.

Recombination periodically recovers after short episodes of growth. The number of recombinant CFUs in cultures inoculated 1:2 increased significantly between 0 and 12 hpi (t = −9.01, P-value < 0.001), and then declined insignificantly until 24 hpi (empty circle, t = 1.26, P-value = 0.27). When cultures were diluted 1:2 at 12 hpi, the number of recombinants increased significantly at the sampling point 24 hpi (t = −3.28, P-value = 0.031) and, again, without dilution declined insignificantly at the sampling point 36 hpi (empty circle, t = 1.21, P-value = 0.29). The number of recombinant CFUs in cultures diluted 1:2 at 12 hpi as well as 24 hpi increased significantly when sampled at 36 hpi (t = −3.62, P-value = 0.022). Differences between mean recombinant CFU concentrations of diluted and nondiluted were significant both at 24 hpi (t = −4.69, P-value = 0.0094) and 36 hpi (t = −4.58, P-value = 0.01). Samples were always taken before dilution step ie from stationary phase saturated cultures. Error bars represent standard error of the mean.

From the 1:2 dilution experiment, we estimated that up to 689 detected recombination events of the reporter locus occurred per milliliter in a population of approximately 3.5 × 10⁹ CFU/ml. Because the 1:2 dilution allowed on average approximately one cell division, this corresponds to 1.97 × 10⁻⁷ detected recombination events at the reporter locus per generation. Assuming approximately uniform recovery of P1-transduced chromosomal fragments (Kleiner et al. 2020), extrapolation across the E. coli chromosome gives an estimated genome-wide recombination frequency of roughly one recombination event per 1,000 cells per dilution cycle (details of the calculation can be found in Text). This frequency is high enough to affect associations among loci over evolutionary timescales.

Transduction-mediated recombination requires recA, recB, and recG

To identify the molecular pathway responsible for transduction-mediated HR in stationary-phase recipients, we measured transduction into immune strains lacking core recombination proteins. We tested mutants affecting the two main HR initiation pathways: RecFOR, represented by ΔrecF, and RecBCD, represented by ΔrecB. We also tested two proteins involved in processing recombination intermediates, RuvA and RecG, using ΔruvA and ΔrecG. As a negative control, we included ΔrecA, because RecA is required for any homologous strand exchange (Bell and Kowalczykowski 2016).

As expected, no transductants were detected in ΔrecA (Fig. 5). Deletion of recF reduced transduction efficiency approximately two-fold relative to Immu, indicating that RecFOR contributes to the process but is not essential. In contrast, no transductants were detected in ΔrecB, showing that transduction-mediated HR under these conditions depends on RecBCD. Among the two tested pathways for processing recombination intermediates, ΔruvA did not significantly reduce transduction efficiency, whereas no transductants were detected in ΔrecG (Fig. 5).

Figure 5.

Transduction efficiency across deletion mutants. The wild-type Immu strain exhibits relative transduction efficiency = 1 (4147 recombinant CFU/ml per 5 × 10⁸ PFUs). ΔrecA shows no recombination as RecA is central to all recombination processes in the cell. ΔrecF (RecFOR pathway) shows a reduction in transduction efficiency to 51% of the wild-type Immu strain (t = 15.82, P-value < 0.001). In ΔrecB (RecBCD pathway) transduction efficiency drops to 0 (t = 75.6, P-value < 0.001). From two enzymes, which recruit RecBCD complex, ΔruvA and ΔrecG, the former shows no significant drop in transduction efficiency (t = 0.96, P-value = 0.39), whereas the latter shows transduction efficiency of 0. Error bars show standard error of the mean.

These results identify RecA, RecBCD, and RecG as essential components of transduction-mediated HR under stationary-phase conditions. The requirement for RecBCD is consistent with recombination initiated from linear chromosomal double-stranded DNA fragments delivered by transducing particles (Smith 2012). The complete loss of transductants in ΔrecG, together with the lack of a significant effect in ΔruvA, further shows that productive recombination in this assay depends specifically on RecG rather than RuvA-mediated processing of recombination intermediates.

Discussion

We identify a population-dynamic route by which generalized transduction can generate frequent HR in bacterial populations. In our E. coli–P1 system, CRISPR immunity did not simply protect recipient bacteria from phage-induced lysis, but by reducing effective phage spread, it maintained a susceptible donor subpopulation and allowed the transducing phage to persist. This coexistence is essential for establishing recurrent gene flow from susceptible donors into immune recipients. Recombination began as cultures entered stationary phase, exceeded spontaneous mutations in our assay by two orders of magnitude, and was restored by brief nutrient pulses. Thus, immune-mediated phage coexistence can create a temporal window in which genetic novelty is generated mainly by reshuffling standing variation rather than by new mutations.

The population-dynamic component is central. HR can influence molecular evolution only when polymorphism is maintained. If a single bacterial genotype fixes, or if phage-susceptible donors are eliminated, there is little homologous variation left to recombine. Herd immunity prevents this collapse by allowing immune recipients, susceptible donors, and phage to coexist. In this setting, recombination can repeatedly combine alleles from different genetic backgrounds and reduce associations among loci. This provides a mechanistic route by which bacteria could achieve a high effective contribution of recombination relative to mutations, as inferred from many bacterial genomes (Sakoparnig et al. 2021; Preska Steinberg et al. 2022; Liu and Good 2024; Torrance et al. 2024, 2025; Wolff and Garud 2026). Our experimental estimate, extrapolated across the chromosome, suggests that such gene flow can occur at frequencies high enough to affect linkage disequilibrium over evolutionary timescales.

Furthermore, we identified the recombination pathway required for incorporation of transduced chromosomal DNA. Recombination required RecA, RecBCD, and RecG, whereas RecFOR was only partially involved and RuvA was dispensable. The RecA and RecBCD dependence is consistent with recombination initiated from linear double-stranded DNA fragments delivered by transducing particles (Smith 2012). RecG is a helicase conserved across almost all bacterial species (Sharples et al. 1999; Rocha et al. 2005) that remodels branched recombination and replication intermediates, including D-loops and Holliday junctions, and can reverse stalled replication forks in vitro (Sharples et al. 1999; Azeroglu et al. 2016; Lloyd and Rudolph 2016). Its requirement in our assay is therefore consistent with processing recombination intermediates formed during the incorporation of transduced chromosomal DNA fragments.

Interestingly, the requirement for recG also suggests a plausible regulatory link between transduction-mediated HR and the physiological transition into stationary phase. recG is cotranscribed with spoT within the five-gene spo operon and is therefore coregulated with it at the transcriptional level (Lloyd and Sharples 1991; Edwards et al. 2011). SpoT controls cellular levels of the signaling nucleotides (p)ppGpp and is a key regulator of the stringent response—a broadly conserved bacterial response to nutrient limitation that slows growth, suppresses resource-intensive biosynthesis, conserves energy, and helps coordinate entry into stationary phase (Boutte and Crosson 2013; Urwin et al. 2024). This regulatory context suggests that RecG has a specific role in recombination during nutrient limitation and the transition to stationary phase, beyond its general function in DNA repair.

Intriguingly, the increase in recombination as nutrients become scarce parallels starvation-induced sexual reproduction in many unicellular eukaryotes. These reproduce clonally when nutrients are abundant but use independently evolved starvation-induced sexual reproduction in response to resource scarcity to reshuffle standing variation in ecologically unfavorable conditions (Giese 1935; Goodenough et al. 2007; Sieber et al. 2023).

We used CRISPR immunity because CRISPR-based herd immunity had already been demonstrated experimentally in E. coli (Payne et al. 2018), allowing a direct test of the population-dynamic hypothesis. Although the immune state was engineered to be plasmid-borne here, CRISPR-Cas systems on plasmids are widespread in bacteria and archaea (Pinilla-Redondo et al. 2022) making it a biologically plausible model. Furthermore, the mechanism described here is unlikely to be specific to CRISPR. Other defense systems that interfere with phage replication at the DNA level, including restriction–modification systems, BREX, and DISARM (Oliveira et al. 2016; Gordeeva et al. 2019; Bravo et al. 2022), could in principle generate similar coexistence regimes if they reduce effective phage transmission without eliminating susceptible hosts. Prophage-encoded superinfection immunity could have analogous effects. The common requirement is not CRISPR itself but partial population-level protection that preserves both susceptible donors and transducing phage.

This population-dynamic mechanism clearly separates our mechanism from recipient-protection assays. Previous work showed that CRISPR-Cas immunity, phage-inducible chromosomal islands, lysogenic immunity, and restriction–modification systems can increase recovery of transductants by protecting recipient bacteria from phage-induced lysis (Haaber et al. 2016; Mašlaňová et al. 2016; Watson et al. 2018; Ibarra-Chávez et al. 2022). These experiments established that protection can raise the probability that incoming DNA is recovered. However, they did not address whether protection can sustain HR within a growing and then nutrient-limited mixed population, nor did they compare the rates with de novo mutation supply. Our results show that recipient protection becomes evolutionarily consequential only when it produces a population structure that maintains donor bacteria and phage over time. In this context, herd immunity converts a one-off transfer advantage into a recurrent source of homologous gene flow and, in a sense, turns bacteriophages into de facto pollinators.

Several lines of evidence suggest that the ingredients required for herd-immunity-driven transduction-mediated HR occur in natural bacterial populations. Phages are abundant in microbial communities (Abedon 2008), and generalized transduction has been documented in diverse systems. For example, many induced Salmonella enterica prophages are capable of generalized transduction (Schicklmaier and Schmieger 1995). Virome and transductomics studies further show that measurable fractions of virions can carry host-derived DNA, although successful delivery and chromosomal establishment are expected to be much rarer (Kleiner et al. 2020; Borodovich et al. 2022). Comparative genomic analyses have inferred recombined DNA tracts that can fall within length scales compatible with phage-mediated transfer, although such patterns are not diagnostic of transduction on their own (Ha and Denver 2018; Sakoparnig et al. 2021). Natural populations also show extensive heterogeneity in CRISPR spacers, and bacterial defense repertoires vary widely across genomes and environments, creating the kind of differential susceptibility required for herd-immunity dynamics (Lopatina et al. 2019; Rocha and Bikard 2022; Meaden et al. 2025). The ecological relevance of the mechanism can therefore be tested by jointly measuring defense polymorphism, phage–host coexistence, packaging of host DNA, and transduction-mediated gene flow in natural populations.

Environmental conditions may favor immune-mediated HR in two complementary ways. First, bacterial doubling times in the wild are often much longer than in laboratory cultures (Gibson et al. 2018). Slower host growth is expected to reduce phage burst size, and lower burst size expands the parameter space in which herd immunity can maintain coexistence between immune bacteria, susceptible bacterial donors, and phage (Payne et al. 2018; Figures S3, S4). Thus, natural growth conditions could easily make the population-dynamic state required for transduction-mediated HR more accessible than it is under rapid laboratory growth conditions. Second, nutrient availability in natural microbial habitats often fluctuates (Zelenev et al. 2005; Wang et al. 2023). Our nutrient-pulse experiments show that such fluctuations can repeatedly move populations from growth into stationary phase, creating recurrent windows in which transduction-mediated HR can occur.

The evolutionary consequences of the population-dynamic process presented here depend on phage host range and ecological structure. Because phage-mediated transfer is constrained by host range, receptor compatibility, and defense susceptibility, it is expected to move DNA most efficiently among bacteria that remain ecologically and genetically compatible. This could allow frequent exchange within ecological populations while limiting the incorporation of highly divergent DNA. Such restricted gene flow may contribute to genetic cohesion within bacterial species or intraspecies units (Bobay and Ochman 2017; Conrad et al. 2024). Conversely, ecological divergence that changes phage susceptibility, receptor use, or defense composition could reduce phage-mediated gene flow and contribute to genetic isolation (Shapiro and Polz 2015; Arevalo et al. 2019).

Our results challenge the common view of bacterial immune systems as simple barriers to gene flow. While defense systems restrict mobile genetic elements and can limit HGT (Marraffini and Sontheimer 2008; Price et al. 2019; Wheatley and MacLean 2021; Rocha and Bikard 2022; Liu et al. 2025), their population-level effects can differ from their cell-level effects. By allowing immune and susceptible bacteria to coexist with phage, immunity can preserve the ecological conditions required for recurrent HR. This provides a plausible route by which defense system diversity may contribute to bacterial molecular evolution, not only by blocking infection but also by regulating when and between whom genetic exchange occurs.

Materials and methods

Bacterial strains construction

The Immu strain was a gift from Hande Acar Kirit, and its construction is described in Acar Kirit et al. (2020). In brief, it contains a chromosomal insertion of a cassette composed of a gene for yellow fluorescent protein (yfp, mVenus) under control of the constitutive phage lambda PR promoter, together with a gene for kanamycin resistance, at the phage P21 attB site of the parental MG1655 strain. This strain was transformed with the plasmid pCas9P1, which provided it with immunity to the P1 phage. pCas9P1 was derived from the pCas9 plasmid carrying a S. pyogenes truncated CRISPR type II system and chloramphenicol resistance. pCas9 was a gift from Luciano Marraffini (Addgene plasmid #42876). Two spacers targeting the P1 phage DNA were designed as oligonucleotides P1_DS_F and P1_DS_R (for all primer sequences, see Table S1), annealed together and inserted into pCas9 following a protocol described in Lutz and Bujard (1997).

Next, a gene for red fluorescent protein (rfp, mCherry) was cloned into the AxyI restriction site downstream of a chloramphenicol resistance gene. mCherry with a T1 transcription terminator was amplified by PCR from pZS*21mCherry using primers AxyI_mCherry_F and AxyI_T1_R, ligated into the pCas9P1 backbone and transformed into DH5alpha competent cells. It should be noted that our system lacks the genes for spacer acquisition because we aim to investigate the dynamics of already resident immune and susceptible bacteria in ratios close to the herd immunity threshold frequency.

The Sus strain was constructed from the parental strain MG1655 by insertion of a cassette containing a gene for cyan fluorescent protein (cfp, mCerulean) under the control of the constitutive promoter PN25 and a spectinomycin resistance gene. It was inserted into the phage λ attB site following a protocol described in Lutz and Bujard (1997). mCerulean was amplified by PCR from pTH2 (Addgene plasmid #84452) using a forward primer SphI_PN25_NdeI_CFP_F, which contained the PN25 promoter as an overhang, and a reverse primer HindIII_CFP_R. This strain was transformed with pCas9 containing no spacers, rendering the Sus strain chloramphenicol resistant but not immune to the P1 phage.

The P21 attB site containing the yfp-kanR cassette and the phage λ attB site containing the cfp-specR cassette are 389 kbp away from each other so that the two loci cannot be cotransduced by P1, which can only package up to 100 kb of DNA (Thomason et al. 2007).

All constructs were sequence verified and confirmed by phenotypic testing for fluorescence and antibiotic resistance.

Cultivation of bacteria and phages

Bacterial cultures were cultured in TPP round bottom 5 ml culture tubes (cat. #091106) in a New Brunswick Scientific Innova 43R Incubator Shaker at 37 °C and 250 rpm. LB broth (Sigma-Aldrich, cat. # L7275) was used as cultivation medium. LB broth with agar (Sigma-Aldrich cat. # L2897) was used to make LB agar plates.

All growth media and plates were supplemented with chloramphenicol 25 µg/ml. Liquid growth media were supplemented with 5 mM calcium chloride as it is required for adsorption of the phage P1 to bacterial cells. Plates used for selection of recombinant and de novo mutants (referred to as selection plates hereafter) were further supplemented with spectinomycin (35 µg/ml), kanamycin (35 µg/ml), and sodium citrate (10 mM) to suppress further phage adsorption on plates. Plates used for enumeration of CFUs did not contain selective antibiotics (ie contained only chloramphenicol 25 g/ml and sodium citrate 10 mM). Plates for total CFU counts were incubated overnight at 37 °C; selection plates were incubated for 48 h to make sure that even slowly growing mutant strains formed visible colonies.

Phage used in the study was the virulent variant of the P1 phage (P1vir, referred to as P1 in the study) and was obtained from the Guet lab strain and phage collection. P1 phage lysate for recombination experiments was produced on laboratory wild-type E. coli K12 MG1655 strain by addition of stock phage lysate (produced on the same strain) into exponentially growing culture of E. coli K12 MG1655 (optical density OD600 between 0.4 and 0.5) at multiplicity of infection MOI = 1. Standard double-layer soft agar assays were used to enumerate the number of PFUs in cultures. Resulting lysate had a concentration 2.8 ± 0.2 × 10¹⁰ PFU/ml. Plates used for plaque assays were supplemented with chloramphenicol (25 µg/ml) and calcium chloride (5 mM).

Competition assays

We inoculated Immu and Sus strains 1:1,000 by serial dilution into fresh LB and combined the cultures in a 1:1 ratio in 2 ml culture tubes. At the time of inoculation, we sampled from the mixed culture in quadruplicates and plated serially diluted samples on LB plates supplemented with 25 μg/ml chloramphenicol. Cultures were incubated under shaking conditions as described above for 20 h and sampled again.

Recombination experiments

Exponential phase experiments were conducted as follows: overnight cultures were inoculated 1:100 in fresh LB medium supplemented as described above. P1 phage was added to cultures at MOI = 0.1, and cultures were incubated as described above. At time intervals 3, 6, 12, 18, and 24 hpi, samples of 100 μl were spread onto selection plates. Simultaneously with sampling onto selection plates, samples were also serially diluted and plated onto LB agar plates without selective antibiotics (ie containing only 25 μg/ml chloramphenicol and 10 mM sodium citrate) to enumerate the total number of CFUs and performed double-layer soft agar assays to enumerate the number of PFUs in the cultures. All experiments were done in triplicate.

Plate imaging and colony enumeration

Plates were imaged using Zeiss AxioZoom.V16-Apotome2 fluorescent microscope equipped with 0.5× objective and filter cubes appropriate for CFP (Zeiss 47), YFP (AHF F46-003), and mRFP (Zeiss 63). The image of each plate was stitched from 12 individual images using the Zeiss Zen Blue software. Colonies on plates were counted using the OpenCFU 4.0.0 + dfsg-2 colony counting software (Geissmann 2013).

All colonies containing both, yfp-kanR and cfp-specR, were counted as recombinant CFUs no matter whether they also expressed rfp. About 1.5% of CFUs were not expressing rfp. We find two possible explanations for this observation. Because the phage cannot replicate on Immu, it cannot transduce the yfp-kanR cassette into Sus. Thus, the first possibility is that the nontargeting plasmid from Sus is transduced into Immu, and by HR the rfp gene is replaced by a sequence, which is missing it, resulting in a plasmid conferring immunity but not expressing rfp. Then the cfp-specR marker can be transduced into this strain generating a recombinant colony expressing both yfp and cfp but not rfp. Second possibility, the plasmid not targeting the phage is transduced from Sus into an immune recombinant strain. However, considering the frequency of Immu in the population (Figs. 2b, 3c), we find this explanation less likely than the first one.

All colonies containing only one of the two cassettes, ie either yfp-kanR or cfp-specR, were counted as mutant CFUs.

Knock-out strain experiments

Single gene knock-out strains from the Keio collection (Baba et al. 2006) with knock-outs of recA (JW2669), recB (JW2788), recF (JW3677), recG (JW3627), and ruvA (JW1850) were transformed with pCas9P1 plasmid. Knock-out strains as well as the control Immu strain were grown overnight (for 18 h) to their final OD, with the exception of ΔrecA, which grew substantially more slowly and reached its final OD after 42 h of incubation, so it was inoculated 24 h before all other strains. Samples from the cultures were serially diluted in SM buffer and plated on agar plates with 25 μg/ml chloramphenicol to enumerate the concentration of CFUs. A 150 μl of each strain was mixed with 50 μl of phage lysate grown on Sus strain (ie containing spectinomycin resistance and cfp cassette) with a concentration 1 × 10¹⁰ PFU/ml and incubated for 2 h in 37 °C in a shaking incubator. After incubation, 100 μl of each culture was plated on selection plates and incubated for 48 h. Plates were subsequently imaged, and colonies were counted as described above. For details on CFUs, OD measurements, and MOIs, see Table S2. All experiments were done in triplicate.

Statistical analyses and figures

Statistical significance of differences in data for Fig. 4 from different treatments and timepoints (diluted vs. nondiluted, 0 vs. 12, 12 vs. 24, 24 vs 36 hpi), for Fig. 5  (Immu vs. each deletion mutant), and for Figure S1 (Immu vs. Sus at time 0 and 20 hpi) were tested by the following statistical methods. First, each pair of data was tested for equal variances using Levene's test. Because the test did not reveal significant differences in variances in any of the sampled pairs, we used the two sample two-tailed Student's t-test to compare whether the means of the samples differed significantly. For each pair of samples, t-statistics and P-value are reported. Since all these experiments were performed in triplicates and the variances were equal, the number of degrees of freedom was equal to 4 for all statistics. All statistics were calculated using RStudio 2023.12.0 + 369.

Figures presenting data were generated using RStudio 2023.12.0  +  369 and packages ggplot2, readxl, dplyr, gridExtra, scales, patchwork, cowplot, and ggpubr. Graphics were drawn using Inkscape 1.3.2.

Author contributions

C.C.G, conceptualization, methodology, investigation, funding acquisition, supervision, and writing—review and editing; P.Pa., conceptualization, methodology, investigation, mathematical modeling, visualization, funding acquisition, project administration, supervision, writing—original draft, and writing—review and editing; J.P.B., funding acquisition and supervision; P.Pl., funding acquisition.

Supplementary materials

Supplementary material is available at Molecular Biology and Evolution online.

Funding

This work was supported by H2020 European Research Council EVOLHGT No. 648440 to JPB, European Commission Programme EXCELES project LX22NPO5103 “National Institute of Virology and Bacteriology” to PPl, and a bilateral Czech-Austrian mobility grant 8J3AT003 funded by MEYS CR and OeAD AT to PPa and CCG.

Data availability

The data underlying this article are available in the article and in its online material.