Fight not flight: Parasites drive the bacterial evolution of resistance, not escape

In the face of ubiquitous threats from parasites, hosts can evolve strategies to resist infection or to altogether avoid parasitism, for instance by avoiding behavior that could expose them to parasites or by dispersing away from local parasite threats. At the microbial scale, bacteria frequently encounter viral parasites, bacteriophages. While bacteria are known to utilize a number of strategies to resist infection by phages, and can have the capacity to avoid moving towards phage-infected cells, it is unknown whether bacteria can evolve dispersal to escape from phages. In order to answer this question, we combined experimental evolution and mathematical modeling. Experimental evolution of the bacterium Pseudomonas fluorescens in environments with differing spatial distributions of the phage Phi2 revealed that the host bacteria evolved resistance depending on parasite distribution, but did not evolve dispersal to escape parasite infection. Simulations using parameterized mathematical models of bacterial growth and swimming motility showed that this is a general finding: while increased dispersal is adaptive in the absence of parasites, in the presence of parasites that fitness benefit disappears and resistance becomes adaptive, regardless of the spatial distribution of parasites. Together, these experiments suggest that parasites should rarely, if ever, drive the evolution of bacterial escape via dispersal.

Methods
Bacteria & Phage Strains and Culturing

We used the widely-studied model system of P. fluorescens SBW25 and its strictly-lytic phage Phi2 (Brockhurst et al. 2007), which putatively binds the bacterial lipopolysaccharide (LPS) (Scanlan and Buckling 2012). We used variations of King’s B (KB) medium (the standard SBW25 bacterial media, see experimental evolution section below): 10 g/L LP0037 Oxoid Bacteriological Peptone, 15 g/L glycerol, 1.5 g/L potassium phosphate, and 0.6 g/L magnesium sulfate. Amplified phage stocks were created by infecting an actively growing SBW25 liquid culture, incubating while shaking for 24 hours at 28°C, then adding 25 µL chloroform per mL culture and storing at 4°C.

Measuring Phage & Bacterial Density

Phage and bacterial concentrations were quantified using standard plating and optical density measurements. Phages were quantified in plaque-forming-units per mL (PFU/mL) using a plaque assay: phage suspensions and 100 µL of an overnight bacterial culture in the appropriate KB concentration were added to 4 mL of 0.5% agar KB media at 45°C then poured over 1.5% agar KB plates. After overnight incubation at 28°C plaque forming units were counted. Bacteria were quantified by plating serial dilutions onto 1.5% agar KB plates and counting the number of colony-forming-units (CFU). In liquid culture, bacterial density was quantified by measuring the optical density at 600 nanometers (OD600) with a spectrophotometer, then translating that to CFU/mL using a previously-established standard curve.

Experimental Evolution

Experimental evolution was conducted in plates containing media with 3 g/L agar. Approximately 22 hours before inoculation or passage, 25 mL of media was poured into each plate with an internal diameter of 90 mm. For the global parasite treatment, media was cooled to 50°C and Phi2 was added before pouring to reach a final phage concentration of 106 PFU/mL. For the control and local parasite treatments, no phage was added before pouring.

Initial inoculations for timepoint 0 were taken from an exponentially growing culture of ancestral P. fluorescens that had been shaken at 250 rpm. For the control and global treatments, 5 µL of a bacterial suspension with 5 x 107 CFU/mL was spotted onto the center of the plate. For the local treatment, phages were added to the bacteria to achieve 5 µL of a mixture containing 5 x 107 CFU/mL bacteria and 5 x 106 PFU/mL phages, then spotted onto the center of the plate. These concentrations were chosen so that the initial inoculum droplet of bacteria would be exposed to approximately the same number of phages between the local and global treatments.

After bacteria are inoculated in the center of the plate, nutrient consumption creates a spatial gradient in chemoattractants. This gradient induces bacterial dispersal from the center of the plate towards the periphery, primarily via flagella-driven swimming. Thus, population-level dispersal is the result of both bacterial growth and motility (Fraebel et al. 2017). In these conditions, phages like Phi2 only passively diffuse (Sampedro et al. 2015).

After initial inoculation, populations were then propagated to a new set of plates daily for 14 transfers, while re-creating the same phage spatial distribution each time. After ~24 hours incubation, a 300 µL pipette tip set to 5 µL was stabbed into the agar immediately past the visible cell front and drawn up slowly, to ensure no bubbles entered the tip. This 5 µL sample was then spotted onto the center of a new plate. For the global treatment, plates were poured in the same manner as initial inoculations, incorporating phages throughout the media. For the local treatment, 2.5 µL of a phage stock with 107 PFU/mL was added onto the 5 µL droplet on the new plate, yielding the same area-density as the initial inoculations. For the control treatment, no phages were added. After 14 transfers, evolved isolates were taken by spreading diluted suspensions of the 5 µL sample onto 1.5% agar KB plates and incubating for 48 hours. By sampling from the cell front, our experimental design inherently selects for increased dispersal in all treatments. We chose this sampling method because we expected it would maximize the evolutionary benefits of increased dispersal, reducing the chance that parasite selection for escape would be obscured by sampling that selected against increased dispersal.

This experiment, with five replicate populations in each of the three parasite distribution treatments (Fig 1), was carried out twice under different media and incubation conditions. In the first experiment, bacteria and phages were incubated at 30°C; in the second experiment, bacteria and phages were incubated at 29°C. Previous work in other labs found that Phi2 does not grow at 30°C but can grow at 29°C (Padfield et al. 2019). Similarly, we found that Phi2 formed plaques on ancestral bacteria at 29°C but not at 30°C. Accordingly, we label the experiment at 29°C as “more favorable” for parasite growth, and label the experiment at 30°C as “less favorable” for parasite growth. The number of generations was estimated by dividing the total duration of the experiment by the average ancestral growth rate (Fig S1) from each experiment.

Nutrient conditions also differed between the two experiments. Bacteria growing in media with 3 g/L agar can disperse via swarming or swimming motility (Kohler et al. 2000; Harshey 2003). For experimental tractability, nutrient concentrations in the media were altered to prevent the formation of swarms. At each temperature, we tested KB formulations with 3 g/L agar, either 25%, 50%, 75%, or 100% concentrations of glycerol and peptone (the carbon and nitrogen sources), and 50% concentrations of potassium phosphate and magnesium sulfate. For each experiment, we then selected the highest-nutrient formulation that did not induce bacterial swarming. For the first experiment, in the conditions relatively less favorable for phage growth, the chosen media was “50%/50% KB”; for the second experiment, relatively more favorable for phage growth, the chosen media was “25%/50% KB”. Note that, because the two experiments were not conducted concurrently, and because of the media and temperature differences between the two experiments, comparisons should only be made within each experiment.

Experimental evolution populations were blocked such that each block contained one population in each of the control, local, and global treatments. In some cases, plates showed no growth after 24 hours of incubation. In these cases, all three populations in the block were re-inoculated from the previous transfer, which had been stored at 4°C. This occurred seven times (Table S1), predominantly in the first two transfers, when phage killing was most frequent. Additionally, one global and one local population, both in the experiment with less favorable conditions for phage growth, became contaminated during the experiment and were excluded from analysis.

Quantifying Bacterial Dispersal/Growth in Soft Agar

Bacterial growth during experimental evolution and dispersal (sometimes called ‘migration rate’ in the literature) of evolved bacterial isolates were measured by photographing plates at 300 dpi. For experimental evolution, plates were photographed immediately before passaging. For evolved isolates, 5 µL of a bacterial suspension at 5x107 CFU/mL was spotted onto the center of a 25mL 0.3% agar plate of the appropriate media, then incubated for approximately 24 hours before photographing. To reduce batch effects, each block was poured from the same bottle of media the day before inoculation, although the drying time was not precisely standardized. After photographing, the area of the circular bacterial growth in these images was measured manually using Fiji (Schindelin et al. 2012). This area was normalized for the exact time of incubation under the assumption that radius increases linearly with time (Croze et al. 2011). For evolved isolates, the batch-corrected dispersal rate is reported.

Quantifying Bacterial Resistance to Phage

Bacterial resistance to phage was quantified by tittering a stock of the phage on lawns of the ancestral bacteria and evolved isolates, then reporting the batch-corrected number of plaques observed on each isolate.

Statistical Analysis

All analyses and visualization were carried out in R 4.2.2 (R Core Team 2022) and RStudio 2022.12.0+353 using packages reshape (Wickham 2007), tidyr (Wickham et al. 2023b), dplyr (Wickham et al. 2023a), ggplot2 (Wickham 2016), ggh4x (Brand 2022), lme4 (Bates et al. 2015), pbkrtest (Halekoh and Højsgaard 2014), gcplyr (Blazanin 2023), cowplot (Wilke 2020), ggrepel (Slowikowski 2022), ggtext (Wilke and Wiernik 2022), magrittr (Bache and Wickham 2022), and purr (Wickham and Henry 2023), along with the Okabe and Ito colorblind-friendly palette (Okabe and Ito 2008). Any other statistical functions, like prcomp, were built-in to R. See https://github.com/mikeblazanin/trav-phage for all analysis code. A permanent record of all data and code is available from Dryad.

Simulations

To simulate bacterial dispersal with different parasite spatial distributions, we modified the widely-used Patlak-Keller-Segel model of bacterial chemotaxis and growth (Patlak 1953; Keller and Segel 1970) to include phage populations (Eqs. 3 – 6). These models have been previously validated to recreate in vitro bacterial population behavior with great precision (Cremer et al. 2019; Li et al. 2020; Ping et al. 2020; Mattingly and Emonet 2022).

In this model we track the population density of hosts (N), parasites (P), resources (R), and attractants (A) through both space and time, using parameters as defined in Table 1. Simulations were initiated with parasites excluded (Control), gaussian distributed centered on the origin (peak height =  parasites/ , sd = ) (Local), or uniformly distributed (2.5 parasites/ ) (Global). Across all treatments, hosts were initiated gaussian distributed centered on the origin (peak height 1010 cells/ , sd = 20 ), resources were initiated uniformly at 50 mM, and attractants were initiated uniformly at 2 mM.

To assess the fitness landscape between resistance (i) and dispersal (χ), growth rate (cR), attractant consumption (cA), or yield (Y), the host population was split into two equally-sized sub-populations that shared the same initial distribution. One sub-population (the resident) had the parameter values listed in Table 1, while the other (the invader) had one or two parameter values which differed from the resident. See Table S3 and the Supplemental Materials for how parameter values were determined. Following this setup, the simulation was run for 20 simulated hours. At the end of the simulation, the fitness of the invader was calculated as log10(invader frequency/resident frequency). All models were implemented in Matlab (Mattingly and Emonet 2022) and all code is available at https://github.com/jeremymoore558/ks_phage. Visualization code is available at https://github.com/mikeblazanin/trav-phage. A permanent record of all data and code is available from Dryad.