Spatiotemporal microbial evolution on antibiotic landscapes

Main Text

The worldwide increase in antibiotic resistance has motivated numerous studies aimed at understanding the phenotypic and genotypic evolution of antibiotic resistance (1–7). These experiments have shed light on the tradeoffs constraining adaptive evolution in single and multi-drug environments (5, 6, 8, 9). However, most of our current knowledge about the evolution of resistance is based on laboratory setups with well-mixed environments (1–7, 10, 11).

In natural and clinical settings, bacteria migrate between spatially distinct regions of selection (5, 6, 8, 9, 12, 13). Theoretical models show that spatially structured pressures change the nature of selection: instead of competing with its neighbors for limited resources, an adapted individual needs only be the first capable of venturing and surviving in a new region (14, 15). A pioneering study, focusing on small population sizes, showed that structured microenvironments increase the rate of adaptation to antibiotics through highly reproducible genetic changes (16). It is unknown how evolution is shaped by the diversification potential and differences in adaptive constraints of large populations in spatial environments.

Here, we present a device for the evolution of bacteria that allows migration and adaptation in a large, spatially structured environment. The Microbial Evolution and Growth Arena (MEGA) plate consists of a rectangular acrylic dish (120 x 60 cm) in which successive regions of black-colored agar containing different concentrations of antibiotics are overlaid by soft agar allowing bacterial motility (Fig. 1A). Motile bacteria inoculated at one location on the plate deplete nutrients locally and then spread by chemotaxis to other regions (17). Only increasingly resistant mutants can spread into sections containing higher levels of antibiotic. The large size of the plate serves two purposes: it provides for a large population and mutational supply, and it maintains the antibiotic gradient despite diffusion (since drug diffusion time scales quadratically with distance while the bacterial front advances linearly, the large plate size prevents the antibiotic gradient from equilibrating over the duration of the experiment). Critically, once a mutant has exhausted the resources of a region of the plate, other mutants do not significantly chemotax to that region (since, without a nutrient gradient, they move diffusively). In this manner, mutational lineages can block each other physically, a phenomenon notably observed in biofilm formation (18). This partitioning of mutants into stable spatial domains also enables sampling of individual mutants for later analysis. Using periodic photography of the plate we constructed time-lapse movies of evolution (movie S1). Combining these with analysis of isolates, this system allows reconstruction of the phenotypic and genotypic evolutionary histories of evolving bacteria.

Challenging bacteria in spatial gradients of antibiotics leads to large increases in resistance through sequential adaptive steps across competing lineages (Fig. 1; movie S1). We first set up the MEGA-plate with symmetric four-step gradients of trimethoprim (TMP) or ciprofloxacin (CPR) proceeding inwards with tenfold increases in concentration per step (Fig. 1A, TMP: 0, 3, 30, 300, 3000x wild-type Minimum Inhibitory Concentration, MIC; CPR: 0, 20, 200, 2000, 20000x MIC) and inoculated the drug free regions with Escherichia coli. Bacteria swim and spread until they reach a concentration in which they can no longer grow (TMP: Fig. 1C and movies S1 and S2; CPR: movie S3). As resistant mutants arise in the population, their descendants migrate into the next step of drug concentration and fan out (Fig. 1C, 88h). Adjacent mutant lineages exclude each other and compete for limited space, resulting in some lineages entirely blocking off growth of others (Fig. 1C). When the winning lineages reach a further increased level of drug concentration at which they, too, are unable to grow, secondary mutations arise, and the process repeats. Ultimately, the bacteria reach and overspread the highest drug concentration, showing dramatic increases in drug resistance (phenotyping sampled mutants from the highest concentration region showed 10,000-fold increase in MIC for TMP, Fig. 1B; 100,000-fold increase for CPR, fig. S1). The adaption time (10 days in TMP; 12 days in CPR) is consistent with evolution in well mixed environments (4), yet is slower than reported adaptation rates in micro-spatial environments, likely a result of the additional time required to swim between concentration steps (9). It is possible that at different dimensions the MEGA-plate will yield different evolutionary dynamics; a wider front would increase the effective population size and thus the mutational supply, while a longer run between steps would increase selection among adjacent lineages.

To test the importance of the size of intermediate steps in the evolution of high-level resistance, we set up a variant of the MEGA-plate in which bacteria go from no drug to a high level directly or through one middle region of variable magnitude (Fig. 2. TMP: high step 3000; middle step 0, 3, 30, or 300 MIC. CPR: high step 2000; middle step 0, 2, 20, or 200 MIC). Bacteria were unable to adapt directly from zero to the highest concentration of either drug. Diffusive smoothing of these large steps enabled the appearance of partially resistant mutants, but their lineages did not advance (Fig. 2A, left). The addition of an intermediate concentration step enabled adaptation, although this was impeded when this middle step was too high (Fig. 2B). Even across the permissive intermediate steps, evolution often proceeded through multiple mutations taking advantage of the local gradients formed by diffusion (TMP: movie S4, CPR: movie S5). Thus, by progressing through colonization of regions with moderately challenging selective pressures, intermediate-resistance mutants can expand to sufficient numbers to facilitate the rise of high-resistance mutants. Analogous to evolutionary rescue in temporal selective gradients (19–21), a gradual spatial gradient allows adaptation to previously inhospitable environments. Unlike in a temporal gradient, though, a spatial gradient does not impose a minimal time for a mutant’s appearance and spread; at any time, a mutant appearing on the stalled front can expand and evolve further, provided it is sufficiently resistant to colonize the next step. Thus, concordant with theoretical predictions (22, 23), access to intermediate regions of moderate selection is critical for enabling a range of evolutionary paths to high-level resistance.

We next focused on the genotypic and phenotypic paths leading to high levels of resistance. We sequenced 21 isolates from the four-step TMP gradient experiment and 230 isolates from the multiple intermediate-step TMP experiment above. Samples separated into minimally and highly mutated, i.e. mutator phenotype, groups (>60 single nucleotide polymorphisms [SNPs] and indels for high; <12 for low; Fig. 3A). Similar separation was also seen when restricting the analysis to synonymous mutations to minimize differential effects of selection (fig. S2). All highly mutated isolates, but none of the others, had mutations in dnaQ (also called mutD), which was the only gene for which the presence of a mutation correlated perfectly with the mutator phenotype. This gene encodes DNA polymerase III which is critical to proofreading (22, 23). Isolates carrying mutated dnaQ alleles showed increased rates of mutations on rifampin disk diffusion assays (fig. S2)(24). These mutators appeared repeatedly in distinct locations on the plate and across experiments. Based on lineage reconstructions from the time-lapse video as well as genotypic relationships, the mutator phenotype emerged at least six times independently between the four-step and intermediate-step TMP experiments above (fig. S2; four different alleles of dnaQ were observed: V96E, I97N, I97S, I97T, where I97T appeared three independent times). While these mutator lineages accumulated mutations more rapidly, their rate of phenotypic adaption was similar to that of the less mutated isolates (reaching the highest level of resistance at roughly the same time, fig. S3 and movie S4). Indeed, the highly mutated lineages had a close to neutral ratio of non-synonymous to synonymous substitutions (Fig. 3B). In contrast, the less mutated isolates showed a high bias towards coding mutations, indicating that most of these mutations were likely adaptive (Fig. 3B).

Focusing on the non-mutator isolates, we identified a wide spectrum of putatively adaptive mutations for trimethoprim resistance. The most frequently mutated gene was the primary target of trimethoprim, folA (25), which encodes dihydrofolate reductase (DHFR) (Fig. 3C), with more mutations appearing as resistance increased. We also observed several genes that were repeatedly mutated yet are not involved in the folate biosynthesis pathway, and thus not primarily associated with trimethoprim resistance. These included stress response genes, such as those of the mar and sox operons, known to be important in general antibiotic and toxin resistance (26), as well as genes involved in transcription and translation, which have been shown to affect trimethoprim resistance (27). We also found that three genes not classically associated with trimethoprim resistance were repeatedly mutated, often with a probable loss of function (frameshift or nonsense): a phosphate transporter (pitA), shikimate kinase I (aroK), and a negative regulator of the PhoQP system (mgrB). Knockout of these genes in the ancestral strain confirmed their resistance phenotypes (fig. S4).

Mutations that increased resistance often came with a cost of reduced growth, which was subsequently restored by additional, compensatory, mutations (28–30). While some resistance-conferring mutations allowed colonization of high drug concentration regions without affecting growth, many lineages capable of growing in the high drug regions were deficient in yield, particularly during CPR resistance evolution (as measured by optical density, Fig. 4A–B, for CPR; fig. S3 for TMP). These yield deficient mutations were followed by compensatory mutations allowing growth to full density (Fig. 4A,B; movie S3; number of compensatory mutants observed in a single run: >50 for TMP and >500 for CPR). In the absence of a chemotaxis-inducing nutrient gradient, the compensatory mutants stayed localized behind the front – appearing in a characteristic pattern of localized spots spreading from single points (movie S3; Fig. 4A).

Focusing on evolution in CPR, we sampled and phenotyped compensatory mutants, finding that many had not only compensated for growth but had also increased in resistance, often beyond the resistance levels of the propagating front (Fig. 4C). Yet, as these mutants were engulfed by their parental lineage, they stayed constrained to the immediate vicinity in which they appeared and were unable to overtake the moving front. To test whether these compensatory mutants were capable of outcompeting the propagating front, in an additional evolution experiment we sampled the trapped compensatory mutants and moved them forward, re-inoculating them ahead of the still-moving front. These compensatory mutants were able to grow in a region where the front could not (Fig. 4D). Similarly, some trapped compensatory mutants were able to outcompete their parent when placed side-by-side on a fresh gradient plate (fig. S5). Hence, as compensatory mutations often occur behind the front, they are spatially restricted from contributing to the ultimate evolutionary course of the population. Indeed, in the rare cases when these compensatory mutations appeared at the front, and were not physically blocked, they accelerated the adaptive process (movie S3, 00:53; fig. S3). Thus, the fitness of the population is not driven by the fittest mutants, but rather by those that are both sufficiently fit and arise sufficiently close to the advancing front.

While the MEGA-plate is not intended to directly simulate natural or clinical settings, it captures unique aspects of evolution during range expansion. Evolution of high levels of resistance is enabled by intermediate regions of moderate selective pressure. Furthermore, as multiple lineages evolve in parallel, the propagating front can be led by lineages less fit than those trapped behind it.. It will be interesting to explore how adaptation rates and mutational diversity depend on other spatio-temporal parameters, including population density, mutation rate, and the relative expansion speed and spatial dimensions. Owing to the relaxed evolutionary constraints in range-expansion dynamics, the MEGA-plate is likely to reveal novel mutational pathways to high-level multi-antibiotic resistance. Further, the MEGA-plate can be adapted to a range of organisms and challenges beyond antibiotics. Importantly, differences in evolutionary dynamics between evolution under different selection pressures appear visually, simplifying both hypothesis generation and testing. Owing to this flexibility, the MEGA-plate is a platform for exploring the interplay of spatial constraints and evolutionary pressures. The MEGA-plate provides a physical analog of the otherwise abstract Muller plots of population genetics (31, 32) and of other elusive aspects of evolution, including diversification, compensatory mutations, and clonal interference. Its relative simplicity and ability to visually demonstrate evolution makes the MEGA-plate a useful tool for science education and outreach.

Supplementary Material