Phage are integral components of microbial ecosystems that can direct ecosystem functioning and dynamics with consequences for human health, biotechnology, and elemental cycling1–5. The enormous influence of phage stems from them being the most abundant biological entity on Earth while also being effective predators6–8. They have highly selective host ranges, which can cause specific changes to microbial abundances, diversity, and interactions that can modify ecosystem functioning and stability9–11. They also impose strong selection pressures on their hosts that can drive ecosystem dynamics over ecological and evolutionary timescales12–14.
The effectiveness of phage predation depends on whether the host is in a surface-associated state (e.g., biofilms, colonies, aggregates, etc.)15,16. This is typical as the majority of microbial life is associated with surfaces17-19. Surface association can increase phage predation by increasing local phage concentrations and the duration of physical contacts with host cells20. However, surface association can also repress phage predation by causing changes to host physiology and local environmental conditions21,22. This includes forming an extracellular matrix that can slow phage transport to host cells22,23, creating regions of low metabolic activity that are less susceptible to phage24–26, and inducing the secretion of molecules that inhibit phage24,27. Surface association also enables the process of microbial spatial self-organization, whereby different microbial populations arrange themselves across space as a consequence of their traits, local environmental conditions, and interactions with neighboring cells28-30. This can result in spatial patterns that physically protect or expose host cells to phage15,21,22,31. Phage predation can also feedback on spatial self-organization and, in turn, modify local environmental conditions and interactions32,33. Thus, there is a complex interplay between phage predation and spatial self-organization that can determine the dynamics and functioning of surface-associated microbial ecosystems.
Here, we hypothesize that phage predation can drive the spread of plasmid-encoded antibiotic resistance by modifying microbial spatial self-organization. More precisely, we hypothesize that phage predation increases the spatial intermixing of different microbial populations during surface-associated growth, consequently increasing the number of cell-cell contacts and promoting conjugation-mediated plasmid transfer between them. Our hypothesis is based on fundamental principles of surface-associated microbial growth. Only those cells located at the biomass periphery typically have access to resources replenished from the environment, and only those cells therefore grow and contribute to new biomass (Fig. 1a; referred to as the active layer)28,34,35. Because their population sizes are small, they are subject to stochastic fluctuations that cause different microbial populations to spatially segregate along the biomass periphery (Fig. 1a), which is referred to as spatial demixing28,35-37. Briefly, the interfaces between microbial populations stochastically meander during growth, which can cause neighboring interfaces to coalesce (Fig. 1a). This reduces the number of cell-cell contacts between different microbial populations and the probability that plasmid transfer will occur between them38,39. However, phage have more ready access to, and are therefore more likely to predate on, cells located at the biomass periphery, which are the cells undergoing the most rapid spatial demixing (Fig. 1a). We therefore expect phage predation to slow spatial demixing, preserve more cell-cell contacts between different microbial populations, and promote plasmid transfer between them. Stated alternatively, we expect phage predation to hinder any one microbial population from dominating the biomass periphery and consequently increase spatial intermixing and plasmid transfer (Fig. 1a).
To test our hypothesis, we performed surface-associated growth experiments with pairs of competing strains of the bacterium Escherichia coli in the presence or absence of phage. The strains can engage in the conjugation-mediated transfer of plasmid R388, which is self-transmissible and encodes for cyan fluorescent protein (CFP) and resistance to chloramphenicol (Fig. 1b)40. We refer to one strain as the R388 donor and the other as the potential recipient (Fig. 1b). We mix the strains together, grow them across nutrient-rich surfaces, infect them with the T6 lytic phage41, and track the extent of R388 transfer using confocal laser-scanning microscopy (CLSM) (Fig. 1c). We complement our experiments with individual-based computational simulations to identify the mechanisms by which phage predation reshapes microbial spatial organization and increases plasmid transfer during surface-associated growth.
Phage predation increases R388 spread during surface-associated microbial growth
We first quantified the effect of phage predation on the spread of R388 as the R388 donor and potential recipient grow together across nutrient-amended agar surfaces. We find that phage predation increases the spread of R388 even in the absence of positive selection for R388 in the form of added chloramphenicol (Fig. 2a, b). This is supported by three lines of evidence. First, the number of spatially discrete transconjugant regions (i.e., sectors composed of cells expressing CFP and RFP and appearing magenta) is larger when phage are present (two-sample two-sided Welch test; P = 5.2 x 10-8, n = 5) (Fig. 2c). Second, transconjugants comprise a greater proportion of the total biomass when phage are present (two-sample two-sided Welch test; P = 6.0 x 10-8, n = 5) (Fig. 2d). Third, the total number of transconjugants is larger when phage are present (two-sample two-sided Welch test; P = 7.6 x 10-6, n = 5) (Fig. 2e). This is true even though phage predation reduces the total biomass size (two-sample two-sided Welch test; P = 1.8 x 10-6, n = 5) (Fig. 2f). Overall, nearly all the potential recipients receive R388 when phage are present (Fig. 2a), despite the fact that transconjugants grow slower than R388-free counterparts (R388 reduces the growth rate by approximately 5%)42. In contrast, only those potential recipients lying at the interfaces with the R388 donor typically receive R388 when phage are absent (Fig. 2b). Thus, when phage are present, the production of transconjugants exceeds their removal via phage predation and out-competition by R388-free counterparts. These outcomes remain valid when reducing the initial relative abundance of the R388 donor by up to 8-fold (Extended Data Fig. 1), demonstrating that only a few R388 donors are needed for the positive effect of phage predation on R388 transfer to manifest. They also remain valid in oxic conditions (Extended Data Fig. 2a, b) and when reducing the nutrient concentration by 90% (Extended Data Fig. 2c, d).
Natural transformation and transduction are not important R388 transfer mechanisms
We next tested whether conjugation-independent mechanisms of horizontal gene transfer (i.e., natural transformation and transduction) can explain our results, as E. coli can acquire plasmid-encoded genes via conjugation-independent mechanisms43. To test this, we prepared a phage-induced lysate of the R388 donor and applied it to the potential recipient during surface-associated growth. We concurrently applied heat-inactivated phage to test for natural transformation or viable phage to test for transduction. For both treatments, CFP is undetectable throughout the entire biomass (Extended Data Fig. 3a), which is expected if natural transformation and transduction are negligible. We then suspended the biomass that received the lysate and streaked it onto chloramphenicol-amended agar plates. We do not observe any growth regardless of whether we apply heat-inactivated or viable phage, which is again expected if natural transformation and transduction are negligible. Finally, we repeated the experiments with DNase I to degrade free DNA. We find that DNAse I has no effect on the number of transconjugants (two-sample two-sided Welch test; P = 0.38, n = 5) (Extended Data Fig. 3b), which is expected if natural transformation is negligible. Taken together, our data establish that phage predation does not promote the transformation or transduction of R388 or its associated genes, and that conjugation-independent mechanisms therefore cannot explain our results.
Phage predation increases R388 transfer by slowing spatial demixing
Our hypothesis posits that phage predation slows the spatial demixing of different microbial populations along the biomass periphery (Fig. 1a), which in turn decreases the number of cell-cell contacts and the extent of R388 transfer between them. To test this, we quantified the effect of phage on the magnitude of spatial intermixing between the R388 donor and potential recipient. We find that spatial intermixing is significantly higher when phage are present (two-sample two-sided Welch test; P = 1.9 x 10-7, n = 5) (Fig. 2g), providing evidence that phage predation does indeed slow spatial demixing. This remains valid if R388 is absent from the experiments (Extended Data Fig. 4), demonstrating that it is an R388-independent effect.
To provide further evidence that phage predation slows spatial demixing, we performed an experiment where we inoculate the phage at a distance of 1 mm from where we inoculate the mixture of the R388 donor and potential recipient on the nutrient-amended agar surface. The phage must therefore diffuse across the surface to predate on their host. We expect that the side of the biomass facing towards the phage inoculation area is exposed to more phage, and thus has higher spatial intermixing and more extensive R388 transfer, than the side facing away. Indeed, we observe higher spatial intermixing on the side facing towards the phage inoculation area (two-sample two-sided Welch test; P = 0.019, n = 3) (Fig. 3a, c). We also observe significantly more transconjugant regions on the side facing towards the phage inoculation area (two-sample two-sided Welch test; P = 7.3 x 10-5, n = 3) (Fig. 3a, d), which is expected if the extent of spatial intermixing determines the extent of R388 transfer.
Finally, we performed a third experiment where we reduce the dosage of phage applied to the biomass of the R388 donor and potential recipient (Extended Data Fig. 5). We expect that at a sufficiently low phage dosage, phage predation will be patchy across the biomass. This should generate correlated variance in the magnitude of spatial intermixing and transconjugant proliferation along the biomass periphery, where local regions with higher phage predation have higher spatial intermixing and more transconjugants. This is indeed what we observe (Extended Data Fig. 5).
Phage predation slows spatial demixing by reshaping spatial organization
When analyzing the surface-associated growth experiments, it is evident that the interfaces between the R388 donor and potential recipient are straighter when phage are present (two-sample two-sided Welch test; P = 0.005, n = 5) (Fig. 2h). This is particularly evident when we inoculate the phage at a distal location, where the interfaces become significantly straighter immediately after contact with phage (Fig. 3b) and are straighter on the side of the biomass facing towards the phage inoculation area (two-sample two-sided Welch test; P = 0.0008, n = 5) (Fig. 3e). Straighter interfaces are less likely to coalesce with each other during surface-associated growth28, thus slowing the spatial demixing of different microbial populations and maintaining more cell-cell contacts and promoting plasmid transfer between them.
Why does phage predation cause straighter interfaces to form? We hypothesize that phage predation shifts the location of fastest growth from the biomass periphery to the interior. In the absence of phage, cells at the biomass periphery grow fastest due to their preferential access to resources replenished from the environment (Fig. 4a). In the presence of phage, cells at the biomass periphery are preferentially predated on, thus shifting the location of fastest growth to the interior (Fig. 4a). Importantly, the biomass periphery has lower cell packing densities and lower rotational ordering of individual cells (Fig. 4b), and we therefore expect peripheral growth to result in more meandering interfaces. In contrast, the interior has higher cell packing densities and higher rotational ordering of individual cells (Fig. 4b), and we therefore expect interior growth to result in straighter interfaces.
To test this, we built an individual-based computational model that allows us to quantify how phage predation affects spatial self-organization. We formulated our model such that phage predate on cells located at the biomass periphery and observe the same effect as in our experiments. When phage are absent, cells at the biomass periphery grow fastest. These cells have low rotational ordering and form meandering interfaces that frequently coalesce with each other, resulting in rapid spatial demixing (Fig. 4c and Supplementary Movie 1). In contrast, when phage are present, cells at the biomass periphery are continuously predated on, which causes cells in the interior to grow fastest (Fig. 4d and Supplementary Movie 1). These cells have high rotational ordering, which results in straighter interfaces (two-sample two-sided Welch test; P = 2.3 x 10-4, n = 5) (Fig. 4f) and higher spatial intermixing (two-sample two-sided Welch test; P = 2.2 x 10-16, n = 5) (Fig. 4g). Indeed, we found that cells have high rotational ordering during phage predation in our experiments (Extended Data Fig. 6).
To further test this mechanism, we performed additional simulations without phage but where we prevent growth in the two layers of cells at the biomass periphery; we therefore directly set the location of fastest growth to the interior (Fig. 4e). This results in interfaces with straightness comparable to those observed with phage (mean interface straightness for interior growth = 0.993, SD = 0.0009; mean interface straightness with phage = 0.995, SD = 0.0003) (Fig. 4f and Supplementary Movie 1). The magnitude of spatial intermixing is also comparable to that observed with phage (mean spatial intermixing for interior growth = -0.47, SD = 0.04; mean spatial intermixing with phage = -0.42, SD = 0.03) (Fig. 4g). Taken together, our simulations demonstrate that the key ingredient for the formation of straighter interfaces is to shift the location of fastest growth from the biomass periphery to the interior where cells have higher packing densities and higher rotational ordering, which reduces interface coalescence and slows spatial demixing.
Phage predation promotes extensive plasmid proliferation in the absence of selection
Because conjugation-mediated plasmid transfer requires direct contact between a plasmid donor and a potential recipient cell, we expect transconjugants to emerge along the interfaces between those microbial populations. To test this, we integrated a plasmid transfer module into our individual-based computational model that sets a defined probability of plasmid transfer when a plasmid donor and a potential recipient cell come into physical contact. When phage are absent, transconjugants exclusively localize along the interfaces between plasmid donor and potential recipient regions and do not substantially proliferate (Fig. 5a and Supplementary Movie 2), which is consistent with our experimental results (Fig. 2b). When phage are present, transconjugants are not confined to the interfaces but instead proliferate and eventually displace potential recipients (Fig. 5b and Supplementary Movie 3), which is again consistent with our experimental results (Fig. 2a).
Why do transconjugants proliferate when phage are present even though the plasmid reduces the growth rate? We hypothesize that the plasmid transfer probability and the large number of cell-cell contacts created by phage predation are sufficiently high to counteract the effects of out-competition by faster growing plasmid-free counterparts. To test this, we varied the plasmid transfer probability in our model between 0.0001 and 0.001. When phage are absent, the number of transconjugants consistently increases as the plasmid transfer probability increases (Fig. 5c, d and Extended Data Fig. 7b). When phage are present, there are always significantly more transconjugants regardless of the plasmid transfer probability (two-sample two-sided Welch tests; P < 2.2 x 10-16, n = 5) (Fig. 5c, d and Extended Data Fig. 7). Moreover, once the plasmid transfer probability exceeds 0.0003, nearly all the potential recipients receive the plasmid and the frequency of transconjugants approaches 0.5 within the entire biomass (Fig. 5c, d and Extended Data Fig. 7). Thus, when phage are present, a relatively low plasmid transfer probability can result in significant proliferation of transconjugants even though they grow slower than plasmid-free counterparts.