Strong selective environments determine evolutionary outcome in time-dependent fitness seascapes

Abstract

The impact of fitness landscape features on evolutionary outcomes has attracted considerable interest in recent decades. However, evolution often occurs under time-dependent selection in so-called fitness seascapes where the landscape is under flux. Fitness seascapes are an inherent feature of natural environments, where the landscape changes owing both to the intrinsic fitness consequences of previous adaptations and extrinsic changes in selected traits caused by new environments. The complexity of such seascapes may curb the predictability of evolution. However, empirical efforts to test this question using a comprehensive set of regimes are lacking. Here, we employed an in vitro microbial model system to investigate differences in evolutionary outcomes between time-invariant and time-dependent environments, including all possible temporal permutations, with three subinhibitory antimicrobials and a viral parasite (phage) as selective agents. Expectedly, time-invariant environments caused stronger directional selection for resistances compared to time-dependent environments. Intriguingly, however, multidrug resistance outcomes in both cases were largely driven by two strong selective agents (rifampicin and phage) out of four agents in total. These agents either caused cross-resistance or obscured the phenotypic effect of other resistance mutations, modulating the evolutionary outcome overall in time-invariant environments and as a function of exposure epoch in time-dependent environments. This suggests that identifying strong selective agents and their pleiotropic effects is critical for predicting evolution in fitness seascapes, with ramifications for evolutionarily informed strategies to mitigate drug resistance evolution.

Keywords: Antibiotic resistance; Escherichia coli; experimental evolution; fitness seascape; fluctuating selection; microbial evolution; phage resistance; pleiotropy; sub‐MIC; time‐dependent selection.

Figure 1

Overview of experimental design to systematically study evolution under fitness seascapes. The main experiment on the left was a 48‐day serial passage experiment where initially isogenic E. coli was subjected to a control environment without antimicrobials; time‐invariant environments with single antimicrobials or all three antimicrobials combined; and time‐dependent environments encompassing all permutations of four 12‐day epochs, including the three antimicrobials and one antimicrobial‐free environment. The antimicrobials were used at subinhibitory concentrations (0.5 × MIC). This translates into a fitness cost seen as an intrinsic growth rate (r) of the ancestral bacteria relative to the antibiotic‐free environment of 0.88, 0.46, and 0.36 for nalidixic acid, rifampicin, and spectinomycin, respectively. The full experiment was repeated with and without initial introduction of phage representing an alternative type of selection pressure. Each unique treatment combination was replicated 16 times, amounting to a total of 928 independent populations, which were cross‐phenotyped over time against low‐level resistance to each of the antimicrobials. In addition, a single dominant clone was isolated from all surviving end‐point populations (N = 900) and phenotyped for growth (optical density [OD] at 600 nm after 24‐hour culture) at several concentrations of each antimicrobial. To investigate the underlying molecular evolution, a subset of 235 clones representing different regimes and divergent low‐level resistance outcomes were also subjected to whole‐genome sequencing.

Figure 2

Contingency of phenotypic outcome and mutation landscape on selective regime. Panels A and B show resistance dynamics over time for nalidixic acid and rifampicin, respectively (mean ± bootstrapped 95% confidence intervals; 32 replicates per mean data point; data in presence and absence of phage pooled). Gray rectangles denote epoch boundaries for time‐dependent protocols. Panels C, D, E, and F show low‐level resistance outcomes to nalidixic acid, rifampicin, spectinomycin, and phage T4, respectively, for clones isolated from each population at the experimental end‐point (logistic regression expected value ± 95% confidence intervals; 16 replicates per mean data point). Resistance has been quantified as a binary variable and indicates the ability to grow at levels exceeding the minimum inhibitory concentration of the ancestral bacterial strain. (G) Mutational landscape. The heat map on the left shows the proportion of sequenced clones containing a nonsynonymous (or infrequently synonymous) mutation in a gene recurrently hit in the dataset. The genes have been ordered by total number of hits. The bar plot on the right shows the median mutation count for clones in each history (lack of bar indicates median of 0 mutations). The y‐axis labels indicate the antimicrobial therapy protocol with the following encoding for each of the four 12‐day experimental epochs: X = antimicrobial‐free environment; N = nalidixic acid; R = rifampicin; S = spectinomycin; A = all three antimicrobial compounds combined.

Figure 3

Pleiotropy and fitness effect of resistance. (A) Low‐level nalidixic acid and spectinomycin resistance after 48 days in time‐invariant rifampicin environment (mean ± bootstrapped 95% confidence intervals). (B) Influence of phage resistance on whether mutations in the genes nadR, rpoB, or spoT produce a low‐level spectinomycin resistant phenotype (mean ± bootstrapped 95% confidence intervals). The data are for clones from phage‐exposed environments (for which phage resistance phenotype was determined). (C) Fitness effect of phage resistance (mean ± bootstrapped 95% confidence intervals). Fitness has been quantified as optical density (OD) at 600 nm wavelength after 24‐hour culture in liquid medium. The value has here been related to the mean growth of the clones from the control treatment (absence of antimicrobials and phage). The data for all the figures is for clones isolated from populations at the experimental end point (N _total = 900), with subset treatments or phenotypes included in a particular analysis indicated in the figure or legend.

Figure 4

Antimicrobials driving resistance evolution in time‐dependent regimes. (A) Rifampicin resistance over time as a function of rifampicin exposure epoch (both in presence and absence of phage that had no effect on selective antimicrobial as it did for nalidixic acid). (B) Nalidixic acid resistance over time in the absence of phage as a function of rifampicin exposure epoch. (C) Nalidixic acid resistance over time in the presence of phage as a function of nalidixic acid exposure epoch. All the data are shown as mean resistance ± bootstrapped 95% confidence intervals, and are based on N = 928 populations. The shaded area indicates the relevant (antimicrobial color code) exposure epoch.

Figure 5

Predictability of past drug exposure and future resistance evolution in time‐dependent regime. (A) Degree of predictive power obtained for antimicrobial and phage exposure past based on phenotypic data from populations and clones at the end‐point of 48‐day serial passage experiment. The white dashed line indicates the level of predictability of the estimated factor by chance. (B) Mutual information between exposure history (rows) and end‐point clone antimicrobial resistance phenotype states (columns). The exposure histories are described at various levels of detail ranging from coarse (phage status and epoch of exposure to an antimicrobial separately) to full information (phage status and antimicrobial epoch order considered simultaneously). For example, the phage presence label of the histories carries substantial information on spectinomycin resistance status of the end points clones.