Evolution can be propelled by natural selection, it can wander with the chance effects of mutation and genetic drift, and it can be constrained by history, whereby past events limit or even potentiate the future (Travisano et al., 1995; Keller and Taylor, 2008; Meyer et al., 2012; Kryazhimskiy et al., 2014; Rebolleda-Gomez and Travisano, 2019). The relative roles of these forces has been debated, with the constraints of history the most contentious (Blount et al., 2018). A wealth of recent research has shown that evolution can be surprisingly repeatable when selection is strong even among distantly related lineages or in different environments (Lieberman et al., 2011; Lassig et al., 2017; Turner et al., 2018), but disparate outcomes become more likely as the footprint of history (i.e. differences in genetic background caused by chance and selection in different environments) increases (Blount et al., 2018; Benton et al., 2021; Mahrt et al., 2021) (For definitions of the forces and their role in the evolution of antibiotic resistance, see Box 1). In the absence of chance and history, selection will cause the most fit genotype to fix in the particular environment, and provided this variant is available, evolution will be perfectly predictable (Bailey et al., 2015; Lassig et al., 2017). However, historical and stochastic processes inevitably produce some degree of contingency, making evolution less predictable, reflecting the importance of evolutionary history (Blount et al., 2008; Meyer et al., 2012; Bajić et al., 2018; Blount et al., 2018; Card et al., 2019; Galardini et al., 2019). The evolution of a new trait, whether by horizontally acquired genes or de novo mutation, is a stochastic process that depends on available genetic variation capable of producing a new trait (Khan et al., 2011; Salverda et al., 2011). As any other evolved trait, antimicrobial resistance (AMR) is subjected to these three evolutionary forces (Box 1).

Antibiotics can impose strong selection pressure on microbial populations, leading to highly repeatable evolutionary outcomes (Vogwill et al., 2014; Lukačišinová et al., 2020; Scribner et al., 2020), with the level of parallelism predicted to depend on the strength of antibiotic pressure (Wistrand-Yuen et al., 2018). However, evolutionary history can also alter the distribution of fitness effects of AMR mutations, their mechanisms of action, or their degree of conferred resistance (Eyre-Walker and Keightley, 2007; Hall and MacLean, 2011; Yen and Papin, 2017; Barbosa et al., 2019). The phenotypic effect of any given mutation acquired is contingent on prior events and will determine the potential of further adaptations to a given environment (Travisano et al., 1995). For example, the effects of a given mutation can vary in different genetic backgrounds (epistasis) or in different environments (pleiotropy) and those mutations can constrain further adaptations (Trindade et al., 2009; Hall and MacLean, 2011; Yen and Papin, 2017; Gifford et al., 2018; Santos-Lopez et al., 2019). Additionally, chance differences in the mutations acquired, their order of occurrence, or compensatory mutations that decrease resistance costs can affect the eventual level of resistance and its evolutionary success in the population (Salverda et al., 2011; Wistrand-Yuen et al., 2018).

The study of mutational pathways to AMR has become accessible by applying population-wide whole-genome sequencing (WGS) to experimentally evolved populations (for a review, see Baquero, 2021). Growth in antibiotics will select for resistant phenotypes whose genotypes can be determined by WGS, and their frequencies and trajectories indicate relative genotype fitness. When large populations, 1 × 10⁷ CFU/mL or higher, of bacteria are propagated, the probability that every base pair is mutated at least once approaches 99% after ~80 generations (Lynch et al., 2016; Santos-Lopez et al., 2019). Yet chance still remains important because most mutations are initially rare and subject to genetic drift until they reach a critical frequency of establishment, when selection dominates their fate (Heffernan and Wahl, 2002; Good et al., 2017; Cooper, 2018). Furthermore, many mutations arise concurrently and those with higher fitness tend to exclude other alleles, known as clonal interference. Thus, the success of new mutations will be determined by their survival of drift, the chance that they co-occur with other fit mutants, and by their relative fitness, which is shaped by selection and history (Nguyen Ba et al., 2019).

The contributions of history, chance, and selection to evolution can be measured using an elegant experimental design (depicted in Figure 1A, Box 1, and described in detail in the Methods) introduced by Travisano et al., 1995, in which replicate populations are propagated from multiple ancestral strains with different evolutionary histories. This experimental design has been used to quantify effects of these forces and to predict evolution in prokaryotes, eukaryotes and even digital organisms (Travisano et al., 1995; Flores-Moya et al., 2008; Keller and Taylor, 2008; Meyer et al., 2012; Kryazhimskiy et al., 2014; Matos et al., 2015; Rebolleda-Gomez and Travisano, 2019; Bundy et al., 2021), but has not been applied to calculate effects of these forces in the evolution of AMR, one of the most critical threats in modern medicine. Here we use this framework to measure the relative roles of history, chance, and selection in the evolution of AMR phenotypes and genotypes in the ESKAPE (Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumannii, Pseudomonas aeruginosa, and Enterobacter spp.) pathogen Acinetobacter baumannii, a leading agent of multidrug-resistant infections worldwide and named as an urgent threat by the CDC (CDC, 2019). Quantifying contributions of these evolutionary forces is essential if we are ever to predict the evolution of drug resistance of pathogens, including HIV and malaria, and of various cancers (Hughes and Andersson, 2015; Verlinden et al., 2016; MacLean and San Millan, 2019; Pokhriyal et al., 2019; Gerstung et al., 2020).

Figure 1 with 1 supplement see all

Download asset Open asset

Figure 1—source data 1

Concentrations of CAZ and IMI (mg/L) added to the broth at different intervals of the evolution experiments.

https://cdn.elifesciences.org/articles/70676/elife-70676-fig1-data1-v2.docx

Download elife-70676-fig1-data1-v2.docx

Figure 1—source data 2

Minimum inhibitory concentration (MIC) values for all ancestors and evolved clones by treatment.

https://cdn.elifesciences.org/articles/70676/elife-70676-fig1-data2-v2.csv

Download elife-70676-fig1-data2-v2.csv

Previously (Santos-Lopez et al., 2019), we propagated a single clone of A. baumannii (strain 17978-mff) for 12 days or 80 generations in increasing concentrations of the fluoroquinolone antibiotic ciprofloxacin (CIP). In that experiment, which established the history for the present study and is analogous to prior exposure in a clinical setting, three replicate populations each were propagated in biofilm conditions or planktonic conditions (hereafter B1–B3 and P1–P3 respectively, Figure 1B). These environments selected for different genetic pathways to CIP resistance and replicate populations also diverged by chance, which produced the genetic and phenotypic histories of the ancestral strains in the current study (Figure 1C,D, Figure 1—source data 1). Key historical differences include reduced ceftazidime (CAZ) resistance in B populations but increased CAZ resistance in P populations (Figure 1C Santos-Lopez et al., 2019).

In the current study, the ‘selection’ phase (Figure 1B) involved experimental evolution in increasing concentrations of the cephalosporin CAZ or the carbapenem imipenem (IMI) for 12 days via serial dilution of planktonic cultures. CAZ or IMI concentrations were doubled every three days (ca. 20 generations), starting with 0.5× minimum inhibitory concentration (MIC; Figure 1—source data 1) for each clone and finishing with 4× MIC, where maximum killing has been observed with β-lactams antibiotics (Nightingale, 1980). Each population was therefore exposed to the same selective pressure during evolutionary rescue. In this study design (Figure 1A, Supplementary Text), the extent of increased resistance represents selection, effects of chance are the phenotypic variation among triplicate populations propagated from the same ancestor, and differences between populations derived from different ancestors quantifies effects of history (Figure 1B).

While the scale of this experiment could seem small, it is well suited for studying the evolution of resistance as 160 generations correspond to ca. 100 days, 15 days, or 170 days of growth in patients of Escherichia coli, P. aeruginosa, or Salmonella enterica, respectively (Gibson et al., 2018). In addition, the genetic contributions of chance, history, and selection were determined by sequencing whole populations to a mean site coverage of 358 (S.D. ± 106) bases at the end of the experiment.

Contributions of evolutionary forces under antibiotic treatment

Antibiotic treatments are designed to achieve sufficient concentration in vivo to clear the infection and prevent the development of new resistant mutants. However, for several reasons including poor drug pharmacokinetics, poor drug distribution, or poor patient compliance, antibiotic concentrations are often below the MIC in body compartments (Andersson and Hughes, 2014). It is expected that as drug concentrations increase, the strength of selection relative to other forces also increases. We therefore analyzed resistance phenotypes of the whole population after 3 days of evolution under subinhibitory drug concentration and after 12 days of evolution in increasing drug levels that concluded at four times the MIC. We analyzed population-wide resistance instead of measures of single isolates because heterogeneity can determine the success or failure of an antibiotic treatment in clinical scenarios (Sánchez-Romero and Casadesús, 2014; Dewachter et al., 2019).

We estimated the role of each force as described by Travisano et al., 1995. Briefly, we estimated the effect of history as the square root of the variance among all propagated populations; the effect of chance as the square root of the variance between the replicates propagated from the same ancestor, and the effect of selection was calculated as the difference in grand mean of the propagated replicates and their ancestors (see Materials and methods for details of this calculations). We estimated effects of these forces during propagation in two antibiotics, CAZ and IMI, and present results of each treatment sequentially. First, after 3 days of growth in subinhibitory concentrations of CAZ, history explained the largest variation in resistance phenotypes (61.7% of variation, p<0.05), with 30.7% for selection and only 7.6% chance (Figure 2A,E, Materials and methods). As expected, CAZ resistance increased overall, but some individual populations did not differ significantly from their ancestor (populations P2, P3, Figure 2A). However, by day 12, following propagation in 4× MIC CAZ, the amount of variation explained by selection increased to 47.8% and effects of history dropped to 31.4% (Figure 2B,E), indicating that strong selective pressures can diminish or erase the effects of history.

Figure 2 with 1 supplement see all

Download asset Open asset

Figure 2—source data 1

Estimated statistics for history, chance, and selection forces.

By means of a nested linear mixed model, the estimated coefficients representing the forces are shown, in addition to the confidence intervals at a α = 0.05 significance level generated by bootstrapping and the Bayes factors computed by a Bayesian analysis. BF₁₀ is the ratio between the probabilities of the alternative and null model and therefore, it measures the degree of evidence of including the force. BF₁₀ < 1 null evidence, 3 > BF₁₀ > 1 weak evidence, 20 > BF₁₀ > 3 positive evidence, 150 > BF₁₀ > 20 strong evidence, BF₁₀ > 150 very strong evidence. This is also source data for Figure 3.

https://cdn.elifesciences.org/articles/70676/elife-70676-fig2-data1-v2.xlsx

Download elife-70676-fig2-data1-v2.xlsx

Previous studies have shown that other evolved traits such as fitness itself show declining adaptability: less fit populations adapt faster and to a greater extent than more fit populations when propagated under the same environmental conditions (Wiser et al., 2013; Kryazhimskiy et al., 2014), which would lead to reduced variance in fitness traits among populations. This homogeneity indeed emerged as prolonged CAZ selection overcame historical variation. Populations with lower initial MICs, which by necessity were exposed to lower concentrations of CAZ, increased their resistance level more than populations with higher MICs (Figure 2—figure supplement 1), implying weak selection for further resistance in populations exceeding the MIC threshold and hence declining rates of resistance gains. This finding also suggests that the level of evolved resistance converges and may be predictable (Meyer et al., 2012; Kryazhimskiy et al., 2014), but effects of genetic background remain (Figure 2). Strong antibiotic selection has the potential to overcome but do not entirely eliminate historical differences in resistance.

Evolutionary trade-offs arise from past antibiotic selection

Evolutionary trade-offs occur when changes in a given gene or trait increase fitness in one environment but reduce fitness in another. For example, a history of adaptation to one antibiotic could alter resistance and subsequent evolution in the presence of a subsequent antibiotic. The phenomena of cross-resistance and collateral sensitivity are specific examples of pleiotropy, where the mechanism of resistance to the initial drug either directly increases or decreases resistance to other drugs, respectively (Pal et al., 2015). Additionally, the resistance mechanism could interact with other genes or alleles in the genome, a form of epistasis, and also promote or impede resistance evolution. We hypothesized that resistance mechanisms arising during selection in CAZ would alter resistance to other antibiotics both by genotype-independent (pleiotropy) and genotype-dependent (epistasis) mechanisms. Recall that during the history phase of the experiment (Santos-Lopez et al., 2019), populations propagated in increasing concentrations of CIP became from 4- to 200-fold more resistant to CIP (Figure 1C, Santos-Lopez et al., 2019). Some of these strains also became more resistant to CAZ (populations P1–P3), while others became more susceptible (populations B1 and B3, for more details, see Santos-Lopez et al., 2019), and given that these populations originated from the same ancestor, this variation in collateral resistance phenotypes is best explained by pleiotropy. In the current study, after 12 days evolving in the presence of CAZ, the grand mean of CIP resistance levels did not change, so history was the dominant force shaping the MIC to CIP (Figure 3A). However, if we analyze the P and the B populations independently, B populations became significantly more sensitive to CIP but the P populations did not (Figure 3A), showing that the emergence of collateral sensitivity may depend on prior selection in different environments. These results also indicate that CAZ resistance mechanisms interact with CIP resistance in potentially useful ways.

Figure 3

Download asset Open asset

We also tested if evolving in the presence of CAZ-altered resistance to the carbapenem antibiotic IMI (Figure 3B). As CAZ and IMI are both β-lactam antibiotics and mutations in efflux pumps can alter resistance to both (Lee et al., 2017), we predicted selection in CAZ would also increase IMI resistance and further, that the contributions of each evolutionary force to IMI resistance would follow that measured for CAZ (Figure 2B). As expected, all 12 populations evolved in CAZ became more resistant to IMI (two-tailed nested t-test p<0.0001, t = 7.507, df = 34), and selection was the most important force (p<0.05), explaining almost 44.3% of the variation, while history contributed 23.0% and chance 32.2% (Figure 3E).

Phenotypic divergence despite genetic parallelism

When multiple lineages evolve independently in the same environment, phenotypic convergence is usually observed, but the genotypes may be more variable (Meyer et al., 2012; Bedhomme et al., 2013; Kryazhimskiy et al., 2014). In our experiment, large populations were exposed to strong antibiotic pressure, so we predicted convergence at the genetic level owing to few solutions that improve both fitness and resistance (Lenski, 2017; Cooper, 2018). We conducted whole-population genomic sequencing of all populations at the end of the experiment to identify all contending mutations above a detection threshold of 5% and analyzed the genetic contributions of history, chance, and selection using Manhattan distance estimators as a metric for the genotypic distance between populations (Figure 4). We calculated the genotypic role of chance as the mean distance between evolved populations sharing the same ancestor; history as the mean distance between evolved populations with different ancestors, after subtracting the effect of chance; and selection as the mean distance between ancestral and evolved populations, after subtracting the effects of chance and history. Using these metrics, we infer that evolution in CAZ at the genotypic level was shaped more by selection than history, but the opposite was seen in IMI, and effects of chance were similar in both experiments (Figure 4).

Figure 4 with 1 supplement see all

Download asset Open asset

Figure 4—source data 1

Putative driver mutations and resistance levels of the replicate populations after 12 days evolving in presence of CAZ or IMI.

The average resistance levels (mg/L) and SEM are shown in the table. Replicates highlighted acquired the same mutation.

https://cdn.elifesciences.org/articles/70676/elife-70676-fig4-data1-v2.docx

Download elife-70676-fig4-data1-v2.docx

Figure 4—source data 2

Complete list of mutated genes from the sequenced populations and clones.

https://cdn.elifesciences.org/articles/70676/elife-70676-fig4-data2-v2.xlsx

Download elife-70676-fig4-data2-v2.xlsx

Clinical CAZ-resistant A. baumannii isolates commonly acquire mutations that increase the activity of Acinetobacter drug efflux (ade) pumps (Lee et al., 2017). In the history phase of CIP selection, biofilm lines (Figure 1B) selected mutations in adeL, the regulator of the adeFGH pump, which produce collateral sensitivity to CAZ and other β-lactams (Figure 1D). In contrast, P lines became cross-resistant to CAZ by adeN mutations that regulate the adeIJK complex or pgpB mutations that are also regulated by adeN (Figure 1D; Santos-Lopez et al., 2019). Subsequently, evolution in increasing concentrations of CAZ selected at least one mutation in adeJ in 16/18 populations (Figure 4A); this gene encodes the permease subunit of AdeIJK that is a known cause of CAZ resistance (Lee et al., 2017). The two exception populations instead acquired mutations in adeN, in ACX60_RS2390, a gene of unknown function, and in ftsI, the target of CAZ. Evolution in IMI also selected mutations in the ftsI gene in all populations (Figure 4B); this gene encodes penicillin binding-protein 2, one of the most common causes of de novo resistance to IMI in clinical isolates (Lee et al., 2017). Therefore, evolution in β-lactam antibiotics generated convergent evolution regardless of the genetic background (Vogwill et al., 2014; Scribner et al., 2020).

Yet despite these genetic similarities, replicate populations reached different resistance levels (Figure 2B,D). As the resistant phenotype was measured in mixed populations with diverse genetic backgrounds, it is possible that even though a resistance allele is fixed, different genotypes within each population could explain the phenotypic differences. Evidence of this heterogeneity might be seen when comparing the five replicate IMI populations that acquired the same mutation in ftsI (A579V) but differ in resistance levels by up to fourfold (Figure 4 and Figure 1—source data 1). Another potential explanation for different phenotypes associated with mutations in the same gene is that different mutations may produce different resistance levels. Evidence for this possible explanation is seen when comparing replicate populations derived from ancestor P1, where different SNPs in adeJ (Figure 4, Figure 1—source data 2) produce varied resistance (Figure 2), perhaps by altering the function of this permease in different ways. Follow-up experiments with reconstructed variants in isogenic backgrounds are needed to test this hypothesis. To summarize, both varied pleiotropy of different mutations in the same drug targets and interactions between mutations in different drug targets may constrain AMR evolution.

Stephen Jay Gould famously argued that replaying the tape of life is impossible because historical contingencies are ubiquitous (Gould, 1990). The evolution and spread of AMR provide a test of this hypothesis because countless evolution experiments are initiated each day with each new prescription to combat infections caused by bacteria with different histories. Previous studies suggest that the predictability of antibiotic resistance – or the fidelity of the replay – depends on the pathogen, the antibiotic treatment, and the growth environment (Vogwill et al., 2014; Gifford et al., 2018; Wistrand-Yuen et al., 2018; Card et al., 2019; Santos-Lopez et al., 2019; Scribner et al., 2020). Here, we have quantified contributions of history, chance, and selection to AMR evolution, using six different ancestors replicated in each of two different antibiotic treatments. In the end, selection is unsurprisingly the predominant force in the evolution of AMR and produced convergent evolution even at the nucleotide level in some instances. Yet history and chance play clear and important roles in the emergence of new resistance phenotypes (Figures 3B,D,5, Vogwill et al., 2014), the extent of evolved resistance (Figures 2 and 3), the generation of collateral sensitivity networks, (Pal et al., 2015), and the predictability of the final resistance phenotype (Figures 1 and 4, Gifford et al., 2018; Scribner et al., 2020). If we consider that the established history of these experimental populations is shallow – the result of only 80 prior generations of growth in a different antibiotic that selected between one and three mutations – it is remarkable how deeply these genotypes were imprinted, resulting in divergent evolutionary trajectories under stringent selection in new drugs. Our data also suggest that, as in Drosophila (Teotonio and Rose, 2000), viruses (Bedhomme et al., 2013) and yeast (Rebolleda-Gomez and Travisano, 2019), history and chance may determine the reversibility of acquired traits (Figure 5).

Figure 5

Download asset Open asset

This probability of reversion is potentially clinically important because exploitable collateral sensitivity networks can arise, such as the tradeoff between CIP resistance and β-lactam resistance identified here (Pal et al., 2015). Finally, our data reveals that evolution of AMR follows a clear diminishing return pattern, where antibiotic pressure selects for mutations with progressively smaller phenotypic effects as the population is treated with higher antibiotic concentrations (Figure 2—figure supplement 1). This result mirrors findings in the original Travisano et al. paper (Travisano et al., 1995), where populations that were pre-adapted to compete well in maltose did not adapt further, but populations with major deficiencies in maltose evolved to become just as fit. This result may be instructive for AMR management: on the one hand, more resistant populations at the outset did not increase this phenotype further, but on the other hand, more susceptible lines rapidly compensated for this deficit.

Our experiment was performed in planktonic cultures and was limited to a sensitive strain of A. baumannii treated with a single fluoroquinolone followed by one of two β-lactam drugs. These were deliberate experimental design choices that allowed careful assessment of the evolutionary forces at play in a rapidly evolving population but may be considered limitations for some broader applications. Despite these limitations, our finding that history and chance are ancillary forces compared to the strength of selection imposed by antibiotics is universal and is well supported by the literature. For instance, exposure to fluroquinolones in Gram-positive or Gram-negative bacteria commonly selects for mutations in gyrA (Seward and Towner, 1998; Weigel et al., 1998; Hooper and Jacoby, 2015). However, we also observed that history and chance can play important roles in resistance evolution in certain specific environments. For example, the reversions in adeL are probably lifestyle dependent and would not be expected to occur if we replay the experiment in the biofilm lifestyle instead of planktonic.

Finally, our experiment focuses solely on de novo mutations and does not allow the opportunity for horizontal gene transfer from other species or strains, which is the principal mechanism of the emergence of AMRs in most clinical settings (MacLean and San Millan, 2019). However, genetic background also affects the fitness of transmissible elements (Alonso-del Valle et al., 2021) and epidemiological data indicate that evolutionary history constrains the persistence of resistance mediated by plasmids (Dunn et al., 2019; León-Sampedro et al., 2021). The framework defined here illustrates the potential to identify genetic and environmental conditions where selection is the most dominant evolutionary force and it predictably produces antagonism between resistance traits. With ever greater knowledge of the present state, we gain hope for guiding the future to exploit the past.

All data generated or analyzed in this study are included in the manuscript, supporting files, or at https://github.com/sirmicrobe/chance_history_selection, where raw experimental values and statistical analysis code is shared. All sequences were deposited into NCBI under the BioProject number PRJNA485123 and accession numbers can be found in Supplementary file 1.

1. 1. Santos-Lopez A
   2. Marshall CW

   (2020) NCBI Sequence Read Archive

   ID PRJNA485123. Predicting the emergence of antibiotic resistance.

   https://www.ncbi.nlm.nih.gov/sra?linkname=bioproject_sra_all&from_uid=485123

1. 1. Alonso-del Valle A
   2. León-Sampedro R
   3. Rodríguez-Beltrán J
   4. DelaFuente J
   5. Hernández-García M
   6. Ruiz-Garbajosa P
   7. Cantón R
   8. Peña-Miller R
   9. San Millán A

   (2021) Variability of plasmid fitness effects contributes to plasmid persistence in bacterial communities

   Nature Communications 12:2653.

   https://doi.org/10.1038/s41467-021-22849-y

   • PubMed
   • Google Scholar
2. 1. Andersson DI
   2. Hughes D

   (2014) Microbiological effects of sublethal levels of antibiotics

   Nat Rev Microbiol 12:465–478.

   https://doi.org/10.1038/nrmicro3270

   • PubMed
   • Google Scholar
3. 1. Bailey SF
   2. Rodrigue N
   3. Kassen R

   (2015) The effect of selection environment on the probability of parallel evolution

   Molecular Biology and Evolution 32:1436–1448.

   https://doi.org/10.1093/molbev/msv033

   • PubMed
   • Google Scholar
4. 1. Bajić D
   2. Vila JCC
   3. Blount ZD
   4. Sánchez A

   (2018) On the deformability of an empirical fitness landscape by microbial evolution

   PNAS 115:11286–11291.

   https://doi.org/10.1073/pnas.1808485115

   • PubMed
   • Google Scholar
5. 1. Baquero F

   (2021) Evolutionary pathways and trajectories in antibiotic resistance

   Clinical Microbiology Reviews 10:e00050-19.

   https://doi.org/10.1128/CMR.00050-19

   • Google Scholar
6. 1. Barbosa C
   2. Römhild R
   3. Rosenstiel P
   4. Schulenburg H

   (2019) Evolutionary stability of collateral sensitivity to antibiotics in the model pathogen Pseudomonas aeruginosa

   eLife 8:e51481.

   https://doi.org/10.7554/eLife.51481

   • Google Scholar
7. 1. Barrick JE
   2. Colburn G
   3. Deatherage DE
   4. Traverse CC
   5. Strand MD
   6. Borges JJ
   7. Knoester DB
   8. Reba A
   9. Meyer AG

   (2014) Identifying structural variation in haploid microbial genomes from short-read resequencing data using breseq

   BMC Genomics 15:1039.

   https://doi.org/10.1186/1471-2164-15-1039

   • PubMed
   • Google Scholar
8. 1. Baym M
   2. Kryazhimskiy S
   3. Lieberman TD
   4. Chung H
   5. Desai MM
   6. Kishony R

   (2015) Inexpensive multiplexed library preparation for megabase-sized genomes

   PLOS ONE 10:e0128036.

   https://doi.org/10.1371/journal.pone.0128036

   • PubMed
   • Google Scholar
9. 1. Bedhomme S
   2. Lafforgue G
   3. Elena SF

   (2013) Genotypic but not phenotypic historical contingency revealed by viral experimental evolution

   BMC Evolutionary Biology 13:46.

   https://doi.org/10.1186/1471-2148-13-46

   • PubMed
   • Google Scholar
10. 1. Benton ML
    2. Abraham A
    3. LaBella AL
    4. Abbot P
    5. Rokas A
    6. Capra JA

    (2021) The influence of evolutionary history on human health and disease

    Nature Reviews. Genetics 22:269–283.

    https://doi.org/10.1038/s41576-020-00305-9

    • PubMed
    • Google Scholar
11. 1. Blount ZD
    2. Borland CZ
    3. Lenski RE

    (2008) Historical contingency and the evolution of a key innovation in an experimental population of Escherichia coli

    PNAS 105:7899–7906.

    https://doi.org/10.1073/pnas.0803151105

    • Google Scholar
12. 1. Blount ZD
    2. Lenski RE
    3. Losos JB

    (2018) Contingency and determinism in evolution: Replaying life’s tape

    Science 362:aam5979.

    https://doi.org/10.1126/science.aam5979

    • PubMed
    • Google Scholar
13. 1. Bolger AM
    2. Lohse M
    3. Usadel B

    (2014) Trimmomatic: a flexible trimmer for Illumina sequence data

    Bioinformatics 30:2114–2120.

    https://doi.org/10.1093/bioinformatics/btu170

    • PubMed
    • Google Scholar
14. Preprint

    1. Bundy JN
    2. Ofria C
    3. Lenski RE

    (2021) How the Footprint of History Shapes the Evolution of Digital Organisms

    bioRxiv.

    https://doi.org/10.1101/2021.04.29.442046

    • Google Scholar
15. 1. Card KJ
    2. LaBar T
    3. Gomez JB
    4. Lenski RE

    (2019) Historical contingency in the evolution of antibiotic resistance after decades of relaxed selection

    PLOS Biology 17:e3000397.

    https://doi.org/10.1371/journal.pbio.3000397

    • PubMed
    • Google Scholar
16. Report

    1. CDC

    (2019) The biggest antibiotic-resistant threats in the U.S

    Centers for Disease Control and Prevention.

    https://www.cdc.gov/drugresistance/biggest-threats.html

    • Google Scholar
17. 1. Cooper VS

    (2018) Experimental evolution as a high-throughput screen for genetic adaptations

    MSphere 3:e00121-18.

    https://doi.org/10.1128/mSphere.00121-18

    • PubMed
    • Google Scholar
18. 1. Dewachter L
    2. Fauvart M
    3. Michiels J

    (2019) Bacterial Heterogeneity and Antibiotic Survival: Understanding and Combatting Persistence and Heteroresistance

    Molecular Cell 76:255–267.

    https://doi.org/10.1016/j.molcel.2019.09.028

    • PubMed
    • Google Scholar
19. 1. Dunai A
    2. Spohn R
    3. Farkas Z
    4. Lázár V
    5. Á G
    6. Apjok G
    7. Boross G
    8. Szappanos B
    9. Grézal G
    10. Faragó A

    (2019) Rapid decline of bacterial drug-resistance in an antibiotic-free environment through phenotypic reversion.Landry CR, Wittkopp PJ, editors

    eLife 8:e47088.

    https://doi.org/10.7554/eLife.47088

    • PubMed
    • Google Scholar
20. 1. Dunn SJ
    2. Connor C
    3. McNally A

    (2019) The evolution and transmission of multi-drug resistant Escherichia coli and Klebsiella pneumoniae: the complexity of clones and plasmids

    Current Opinion in Microbiology 51:51–56.

    https://doi.org/10.1016/j.mib.2019.06.004

    • PubMed
    • Google Scholar
21. 1. Durão P
    2. Balbontín R
    3. Gordo I

    (2018) Evolutionary mechanisms shaping the maintenance of antibiotic resistance

    Trends in Microbiology 26:677–691.

    https://doi.org/10.1016/j.tim.2018.01.005

    • PubMed
    • Google Scholar
22. 1. Eyre-Walker A
    2. Keightley PD

    (2007) The distribution of fitness effects of new mutations

    Nat Rev Genet 8:610–618.

    https://doi.org/10.1038/nrg2146

    • PubMed
    • Google Scholar
23. 1. Flores-Moya A
    2. Costas E
    3. Lopez-Rodas V

    (2008) Roles of adaptation, chance and history in the evolution of the dinoflagellate Prorocentrum triestinum

    Die Naturwissenschaften 95:697–703.

    https://doi.org/10.1007/s00114-008-0372-1

    • PubMed
    • Google Scholar
24. 1. Galardini M
    2. Busby BP
    3. Vieitez C
    4. Dunham AS
    5. Typas A
    6. Beltrao P

    (2019) The impact of the genetic background on gene deletion phenotypes in Saccharomyces cerevisiae

    Molecular Systems Biology 15:e8831.

    https://doi.org/10.15252/msb.20198831

    • PubMed
    • Google Scholar
25. 1. Gerstung M
    2. Jolly C
    3. Leshchiner I
    4. Dentro SC
    5. Gonzalez S
    6. Rosebrock D
    7. Mitchell TJ
    8. Rubanova Y
    9. Anur P
    10. Yu K

    (2020) The evolutionary history of 2,658 cancers

    Nature 578:122–128.

    https://doi.org/10.1038/s41586-019-1907-7

    • PubMed
    • Google Scholar
26. 1. Gibson B
    2. Wilson DJ
    3. Feil E
    4. Eyre-Walker A

    (2018) The distribution of bacterial doubling times in the wild

    Proc Biol Sci 285:789.

    https://doi.org/10.1098/rspb.2018.0789

    • PubMed
    • Google Scholar
27. 1. Gifford DR
    2. Krasovec R
    3. Aston E
    4. Belavkin RV
    5. Channon A
    6. Knight CG

    (2018) Environmental pleiotropy and demographic history direct adaptation under antibiotic selection

    Heredity 121:438–448.

    https://doi.org/10.1038/s41437-018-0137-3

    • PubMed
    • Google Scholar
28. 1. Good BH
    2. McDonald MJ
    3. Barrick JE
    4. Lenski RE
    5. Desai MM

    (2017) The dynamics of molecular evolution over 60,000 generations

    Nature 551:45–50.

    https://doi.org/10.1038/nature24287

    • PubMed
    • Google Scholar
29. Book

    1. Gould SJ

    (1990)

    Wonderful Life: The Burgess Shale and the Nature of History

    W. W. Norton & Company.

    • Google Scholar
30. 1. Hall AR
    2. MacLean RC

    (2011) Epistasis buffers the fitness effects of rifampicin- resistance mutations in Pseudomonas aeruginosa

    Evolution 65:2370–2379.

    https://doi.org/10.1111/j.1558-5646.2011.01302.x

    • PubMed
    • Google Scholar
31. 1. Heffernan JM
    2. Wahl LM

    (2002) The effects of genetic drift in experimental evolution

    Theoretical Population Biology 62:349–356.

    https://doi.org/10.1016/s0040-5809(02)00002-3

    • PubMed
    • Google Scholar
32. 1. Hooper DC
    2. Jacoby GA

    (2015) Mechanisms of drug resistance: quinolone resistance

    Ann N Y Acad Sci 1354:12–31.

    https://doi.org/10.1111/nyas.12830

    • PubMed
    • Google Scholar
33. 1. Hughes D
    2. Andersson DI

    (2015) Evolutionary consequences of drug resistance: shared principles across diverse targets and organisms

    Nat Rev Genet 16:459–471.

    https://doi.org/10.1038/nrg3922

    • PubMed
    • Google Scholar
34. 1. Jones RM
    2. Nicas M
    3. Hubbard A
    4. Sylvester MD
    5. Reingold A

    (2016) The Infectious Dose of Francisella Tularensis (Tularemia

    Applied Biosafety 10:227–239.

    https://doi.org/10.1177/153567600501000405

    • Google Scholar
35. 1. Keller SR
    2. Taylor DR

    (2008) History, chance and adaptation during biological invasion: separating stochastic phenotypic evolution from response to selection

    Ecol Lett 11:852–866.

    https://doi.org/10.1111/j.1461-0248.2008.01188.x

    • PubMed
    • Google Scholar
36. 1. Khan AI
    2. Dinh DM
    3. Schneider D
    4. Lenski RE
    5. Cooper TF

    (2011) Negative epistasis between beneficial mutations in an evolving bacterial population

    Science 332:1193–1196.

    https://doi.org/10.1126/science.1203801

    • PubMed
    • Google Scholar
37. 1. Kryazhimskiy S
    2. Rice DP
    3. Jerison ER
    4. Desai MM

    (2014) Global epistasis makes adaptation predictable despite sequence-level stochasticity

    Science 344:1519–1522.

    https://doi.org/10.1126/science.1250939

    • PubMed
    • Google Scholar
38. 1. Kugelberg E
    2. Lofmark S
    3. Wretlind B
    4. Andersson DI

    (2005) Reduction of the fitness burden of quinolone resistance in Pseudomonas aeruginosa

    J Antimicrob Chemother 55:22–30.

    https://doi.org/10.1093/jac/dkh505

    • PubMed
    • Google Scholar
39. 1. Lassig M
    2. Mustonen V
    3. Walczak AM

    (2017) Predicting evolution

    Nature Ecology & Evolution 1:77.

    https://doi.org/10.1038/s41559-017-0077

    • Google Scholar
40. 1. Lee CR
    2. Lee JH
    3. Park M
    4. Park KS
    5. Bae IK
    6. Kim YB
    7. Cha CJ
    8. Jeong BC
    9. Lee SH

    (2017) Biology of Acinetobacter baumannii: Pathogenesis, Antibiotic Resistance Mechanisms, and Prospective Treatment Options

    Frontiers in Cellular and Infection Microbiology 7:55.

    https://doi.org/10.3389/fcimb.2017.00055

    • PubMed
    • Google Scholar
41. 1. Lenski RE

    (2017) Convergence and Divergence in a Long-Term Experiment with Bacteria

    Am Nat 190:S57–S68.

    https://doi.org/10.1086/691209

    • PubMed
    • Google Scholar
42. 1. León-Sampedro R
    2. DelaFuente J
    3. Díaz-Agero C
    4. Crellen T
    5. Musicha P
    6. Rodríguez-Beltrán J
    7. de la Vega C
    8. Hernández-García M
    9. R-GNOSIS WP5 Study Group
    10. López-Fresneña N
    11. Ruiz-Garbajosa P
    12. Cantón R
    13. Cooper BS
    14. San Millán Á

    (2021) Pervasive transmission of a carbapenem resistance plasmid in the gut microbiota of hospitalized patients

    Nature Microbiology 6:606–616.

    https://doi.org/10.1038/s41564-021-00879-y

    • PubMed
    • Google Scholar
43. 1. Lieberman TD
    2. Michel JB
    3. Aingaran M
    4. Potter-Bynoe G
    5. Roux D
    6. Davis MR
    7. Skurnik D
    8. Leiby N
    9. LiPuma JJ
    10. Goldberg JB

    (2011) Parallel bacterial evolution within multiple patients identifies candidate pathogenicity genes

    Nat Genet 43:1275–1280.

    https://doi.org/10.1038/ng.997

    • PubMed
    • Google Scholar
44. 1. Lukačišinová M
    2. Fernando B
    3. Bollenbach T

    (2020) Highly parallel lab evolution reveals that epistasis can curb the evolution of antibiotic resistance

    Nature Communications 11:3105.

    https://doi.org/10.1038/s41467-020-16932-z

    • PubMed
    • Google Scholar
45. 1. Lynch M
    2. Ackerman MS
    3. Gout JF
    4. Long H
    5. Sung W
    6. Thomas WK
    7. Foster PL

    (2016) Genetic drift, selection and the evolution of the mutation rate

    Nature Reviews Genetics 17:704–714.

    https://doi.org/10.1038/nrg.2016.104

    • PubMed
    • Google Scholar
46. 1. MacLean RC
    2. San Millan A

    (2019) The evolution of antibiotic resistance

    Science 365:1082–1083.

    https://doi.org/10.1126/science.aax3879

    • PubMed
    • Google Scholar
47. 1. Mahrt N
    2. Tietze A
    3. Künzel S
    4. Franzenburg S
    5. Barbosa C
    6. Jansen G
    7. Schulenburg H

    (2021) Bottleneck size and selection level reproducibly impact evolution of antibiotic resistance

    Nature Ecology & Evolution 5:1233–1242.

    https://doi.org/10.1038/s41559-021-01511-2

    • Google Scholar
48. 1. Matos M
    2. Simões P
    3. Santos MA
    4. Seabra SG
    5. Faria GS
    6. Vala F
    7. Santos J
    8. Fragata I

    (2015) History, chance and selection during phenotypic and genomic experimental evolution: replaying the tape of life at different levels

    Front. Genet 6:71.

    https://doi.org/10.3389/fgene.2015.00071

    • PubMed
    • Google Scholar
49. 1. Meyer JR
    2. Dobias DT
    3. Weitz JS
    4. Barrick JE
    5. Quick RT
    6. Lenski RE

    (2012) Repeatability and contingency in the evolution of a key innovation in phage lambda

    Science 335:428–432.

    https://doi.org/10.1126/science.1214449

    • PubMed
    • Google Scholar
50. 1. Nguyen Ba AN
    2. Cvijović I
    3. Rojas Echenique JI
    4. Lawrence KR
    5. Rego-Costa A
    6. Liu X
    7. Levy SF
    8. Desai MM

    (2019) High-resolution lineage tracking reveals travelling wave of adaptation in laboratory yeast

    Nature 575:494–499.

    https://doi.org/10.1038/s41586-019-1749-3

    • PubMed
    • Google Scholar
51. 1. Nightingale C

    (1980)

    Pharmacokinetics of the oral cephalosporins in adults

    The Journal of International Medical Research 8:2–8.

    • PubMed
    • Google Scholar
52. 1. Pal C
    2. Papp B
    3. Lazar V

    (2015) Collateral sensitivity of antibiotic-resistant microbes

    Trends in Microbiology 23:401–407.

    https://doi.org/10.1016/j.tim.2015.02.009

    • PubMed
    • Google Scholar
53. 1. Palaci M
    2. Dietze R
    3. Hadad DJ
    4. Ribeiro FKC
    5. Peres RL
    6. Vinhas SA
    7. Maciel ELN
    8. do Valle Dettoni V
    9. Horter L
    10. Boom WH
    11. Johnson JL
    12. Eisenach KD

    (2007) Cavitary disease and quantitative sputum bacillary load in cases of pulmonary tuberculosis

    Journal of Clinical Microbiology 45:4064–4066.

    https://doi.org/10.1128/JCM.01780-07

    • PubMed
    • Google Scholar
54. 1. Paterson IK
    2. Hoyle A
    3. Ochoa G
    4. Baker-Austin C
    5. Taylor NGH

    (2016) Optimising antibiotic usage to treat bacterial infections

    Scientific Reports 6:37853.

    https://doi.org/10.1038/srep37853

    • PubMed
    • Google Scholar
55. Preprint

    1. Pennings PS
    2. Ogbunugafor CB
    3. Hershberg R

    (2021) Reversion Is Most Likely under High Mutation Supply, When Compensatory Mutations Don’t Fully Restore Fitness Costs

    bioRxiv.

    https://doi.org/10.1101/2020.12.28.424568

    • Google Scholar
56. 1. Pokhriyal R
    2. Hariprasad R
    3. Kumar L
    4. Hariprasad G

    (2019) Chemotherapy resistance in advanced ovarian cancer patients

    Biomarkers in Cancer 11:1179299X19860815.

    https://doi.org/10.1177/1179299X19860815

    • PubMed
    • Google Scholar
57. 1. Rebolleda-Gomez M
    2. Travisano M

    (2019) Adaptation, chance, and history in experimental evolution reversals to unicellularity

    Evolution 73:73–83.

    https://doi.org/10.1111/evo.13654

    • PubMed
    • Google Scholar
58. 1. Salverda ML
    2. Dellus E
    3. Gorter FA
    4. Debets AJ
    5. van der Oost J
    6. Hoekstra RF
    7. Tawfik DS
    8. de Visser JA

    (2011) Initial mutations direct alternative pathways of protein evolution

    PLOS Genetics 7:e1001321.

    https://doi.org/10.1371/journal.pgen.1001321

    • PubMed
    • Google Scholar
59. 1. Sánchez-Romero MA
    2. Casadesús J

    (2014) Contribution of phenotypic heterogeneity to adaptive antibiotic resistance

    PNAS 111:355–360.

    https://doi.org/10.1073/pnas.1316084111

    • PubMed
    • Google Scholar
60. 1. Santos-Lopez A
    2. Marshall CW
    3. Scribner MR
    4. Snyder DJ
    5. Cooper VS

    (2019) Evolutionary pathways to antibiotic resistance are dependent upon environmental structure and bacterial lifestyle.Kirkegaard K, editor

    eLife 8:e47612.

    https://doi.org/10.7554/eLife.47612

    • PubMed
    • Google Scholar
61. 1. Scribner MR
    2. Santos-Lopez A
    3. Marshall CW
    4. Deitrick C
    5. Cooper VS

    (2020) Parallel evolution of tobramycin resistance across species and environments

    MBio 11:e00932-20.

    https://doi.org/10.1128/mBio.00932-20

    • PubMed
    • Google Scholar
62. 1. Seward RJ
    2. Towner KJ

    (1998) Molecular epidemiology of quinolone resistance in Acinetobacter spp

    Clinical Microbiology and Infection 4:248–254.

    https://doi.org/10.1111/j.1469-0691.1998.tb00052.x

    • PubMed
    • Google Scholar
63. 1. Teotonio H
    2. Rose MR

    (2000) Variation in the reversibility of evolution

    Nature 408:463–466.

    https://doi.org/10.1038/35044070

    • PubMed
    • Google Scholar
64. 1. Travisano M
    2. Mongold JA
    3. Bennett AF
    4. Lenski RE

    (1995) Experimental tests of the roles of adaptation, chance, and history in evolution

    Science 267:87–90.

    https://doi.org/10.1126/science.7809610

    • PubMed
    • Google Scholar
65. 1. Trindade S
    2. Sousa A
    3. Xavier KB
    4. Dionisio F
    5. Ferreira MG
    6. Gordo I

    (2009) Positive epistasis drives the acquisition of multidrug resistance

    PLOS Genetics 5:e1000578.

    https://doi.org/10.1371/journal.pgen.1000578

    • PubMed
    • Google Scholar
66. 1. Turner CB
    2. Marshall CW
    3. Cooper VS

    (2018) Parallel genetic adaptation across environments differing in mode of growth or resource availability

    Evolution Letters 2:355–367.

    https://doi.org/10.1002/evl3.75

    • PubMed
    • Google Scholar
67. 1. Verlinden BK
    2. Louw A
    3. Birkholtz LM

    (2016) Resisting resistance: is there a solution for malaria?

    Expert Opin Drug Discov 11:395–406.

    https://doi.org/10.1517/17460441.2016.1154037

    • PubMed
    • Google Scholar
68. 1. Vogwill T
    2. Kojadinovic M
    3. Furio V
    4. MacLean RC

    (2014) Testing the role of genetic background in parallel evolution using the comparative experimental evolution of antibiotic resistance

    Molecular Biology and Evolution 31:3314–3323.

    https://doi.org/10.1093/molbev/msu262

    • PubMed
    • Google Scholar
69. 1. Wagenmakers EJ

    (2007) A practical solution to the pervasive problems ofp values

    Psychonomic Bulletin & Review 14:779–804.

    https://doi.org/10.3758/BF03194105

    • Google Scholar
70. 1. Weigel LM
    2. Steward CD
    3. Tenover FC

    (1998) gyrA mutations associated with fluoroquinolone resistance in eight species of Enterobacteriaceae

    Antimicrob Agents Chemother 42:2661–2667.

    https://doi.org/10.1128/AAC.42.10.2661

    • PubMed
    • Google Scholar
71. 1. Wiser MJ
    2. Ribeck N
    3. Lenski RE

    (2013) Long-Term Dynamics of Adaptation in Asexual Populations

    Science 342:1364–1367.

    https://doi.org/10.1126/science.1243357

    • PubMed
    • Google Scholar
72. 1. Wistrand-Yuen E
    2. Knopp M
    3. Hjort K
    4. Koskiniemi S
    5. Berg OG
    6. Andersson DI

    (2018) Evolution of high-level resistance during low-level antibiotic exposure

    Nat Commun 9:1599.

    https://doi.org/10.1038/s41467-018-04059-1

    • PubMed
    • Google Scholar
73. 1. Yen P
    2. Papin JA

    (2017) History of antibiotic adaptation influences microbial evolutionary dynamics during subsequent treatment

    PLOS Biol 15:e2001586.

    https://doi.org/10.1371/journal.pbio.2001586

    • PubMed
    • Google Scholar

Acceptance summary:

This work uses an elegant and well-designed experimental evolution strategy to investigate the roles of history, chance, and selection on the evolution of antibiotic resistance in the clinical pathogen Acinetobacter baumannii. The authors show that while history impacts the evolution of antimicrobial resistance, this effect decreases with increasing drug selection strength and indicates that natural selection is a dominant driver. The work presents clear, unambiguous data on the importance of antibiotic exposure on the evolution of resistance.

Decision letter after peer review:

Thank you for submitting your article "The roles of history, chance, and natural selection in the evolution of antibiotic resistance" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Dominique Soldati-Favre as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Alan McNally (Reviewer #2).

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

While the work addresses a very relevant and interesting question, the reviewers raised some concerns that need to be addressed in order to improve the manuscript. Additional experimental details should be included, and clarification is needed regarding some of the concepts used. More importantly, the limitations of the approach need to be specifically addressed and some of the claims and interpretations should be toned down, particularly given the absence of a drug-free control.

1) A main concern is in the interpretation of populations having the same genetics but different phenotypes (eg lines 300-310) – this conclusion is wrong given the experimental design. The authors provide several explanations but leave out a high likelihood alternative which is that different experiments consist of mixed populations with diverse genetic makeups. Alternatively, if the authors meant instead to focus on specific mutations (ie same mutations/localized mutations) regardless of the background, then this conclusion is trivial – it is known that pleiotropy exists.

2) I agree with the author's definitions of change due to history vs change due to selection given the evolution they designed – and I found this quite innovative and overall strong. However, the definition and calculations of evolution due to chance here are shaky given the three replicates only. There are plenty of extremely rigorous estimates of genetic drift during bacterial evolution (see all of Lenski's work) – three populations is quite small to make that claim and run the statistics. Including chance as a possible contributor didn't add much to the story, and it raised more questions than answers given the extremely diverse genetics that were observed, and that data of the mixed populations below 75% was not included.

3) Experiments are flawed without a no drug control. Was the reversion of CIP resistance specific to the new drug selection, or is it simply the lack of CIP pressure? This is important as the authors make claims about drug-specific evolutionary tradeoffs and collateral sensitivity. The methods in line 428: "We froze 1 mL of the control populations on days…" but this is the only place a control population is mentioned.

4) Much of the data was difficult to follow, such as the fold changes for MICs – for example, following the discussion of CIP resistance, maintenance, and reversion was nearly impossible with the reference to figure 4 fold changes only. Also going back and forth between IMI and CAZ, which is discussed in parallel near the beginning, but then broken down later on, is another example.

5) Key experimental details are missing. What volume were the populations propagated in? Population experiments were not well described – how were replicate MICs initiated? How much volume was used for sequencing? So on and so forth. These are particularly important to interpret the results.

6) Why was a 75% threshold used to determine alleles if the populations were sequenced to a level of 5%?

7) Despite being described in great detail in the methods, it doesn't come across to non-modeler such as myself exactly how selection, history, and chance were quantified in the phenotypic experiments shown in figure 2. I am a bit clearer as to how these were quantified for the genomic data. However, I think a broad readership may benefit from a layman's description of how these were differentiated. It is vital to the science and data shown and I would certainly have appreciated a clarification, especially for the phenotypic data in figure 2.

8) In line 307. In the absence of bacterial genetic experiments to confirm that these historical infections are actually driving the 4 fold differences in phenotype, I think this inference needs to be toned down.

9) Line 343: Is there a benefit to be had of fitness experiments in antibiotic-free medium to confirm the supposition made here?

10) In some cases, I found the framing of the results in terms of history, chance, and selection to be a bit overly general, which sometimes obscured the specific results being reported. The paper could be improved by using more specific language-perhaps restricted in scope- in describing and interpreting the results, both because 1) it's not obvious to me that the results would apply generally to antibiotic resistance beyond the very nice, but potentially system-specific, results presented here; and 2) the terms themselves (history, chance, selection) could conceivably have different meanings in different contexts (more on this below).

11) The study design has been used in numerous previous studies; it is well established, elegant, and has given rise to many new insights. However, as I understand it, there are some inherent assumptions of the approach that should be briefly discussed. Most notably, does the approach inherently assume that the effects of history, chance, and selection are additive (or perhaps linear) in some sense (in terms of phenotypic variance measurements or the Manhattan-based genotype metric)? While this simplifying assumption seems critical to the power of the approach, it is not clear to me that this assumption holds in general. When I try to think of this in terms of, say, a simple population dynamics model, the terms history, chance and selection are themselves somewhat nebulous, and it's not clear to me that they could be unambiguously and uniquely defined even in simplified theoretical models (or more directly, that the variance-based phenotype measures correspond to well-defined features or parameters of population dynamics models). I say this not to criticize the approach-again, its power lies in the simplicity of the design and the intuitive value of separating these three evolutionary features and attempting to quantify their contributions. But I think the article could be strengthened by briefly discussing the underlying assumptions-ideally by pointing to previous work (if it exists) that establishes that the features are additive and measurable in the sense required by the experimental design. If not, I think it would be worth discussing that limitation briefly, as I worry the inherently nonlinear nature of these very complicated, evolving systems could lead us to misinterpret the results. Given the general success of similar approaches in past work, I suspect the authors have thought through these issues in detail; discussing those points might open the paper to a broader audience not intimately familiar with all the previous studies.

12) One example related (but not identical to) the point above: in the current experiment, the role of history is defined in terms of previous selection conditions (drug and growth phase). But the new evolution experiment itself has multiple time points, and even qualitatively distinct epochs (sub- and super-MIC drug). So one might argue that history is playing a role continuously throughout the experiment-history not merely of the previous selection in fluoroquinolones, but also history of the previous time points / epochs of the new evolution (β-lactams). My point is that it is important to clearly define the terms at all stages and discuss, at least briefly, the limitations of the definitions that are chosen.

13) While history, chance, and selection are quantified at both the genetic and phenotype level, it's not clear to me that these numbers can be directly compared to one another (though it's tempting to do so!). Could the authors briefly comment on the connection between these measures-that is, when (and to what extent) one would expect correlations between them (e.g. high levels of historical influence at the genetic level leads to high levels of historical influence at the phenotype (MIC) level….assuming the definitions used here).

14) Do the authors have any thoughts on how the results might be affected by the fact that the new evolution experiments take place in planktonic (rather than biofilm) conditions? How might the results differ if they had been performed in biofilm, and what could you learn from the fully symmetric experiment (P/ B initial strains evolved in both P and B new selection conditions). The authors may wish to discuss this avenue for future work.

Essential revisions:

While the work addresses a very relevant and interesting question, the reviewers raised some concerns that need to be addressed in order to improve the manuscript. Additional experimental details should be included, and clarification is needed regarding some of the concepts used. More importantly, the limitations of the approach need to be specifically addressed and some of the claims and interpretations should be toned down, particularly given the absence of a drug-free control.

We thank the reviewers and editors for their careful consideration of our manuscript. We agree with all reviewer comments and edited the paper accordingly. We have revised the manuscript for general clarity, and we believe that the revised manuscript has been improved through the changes proposed by the reviewers. We have included the missing experimental details, we have softened some conclusions, and we have included two sections about our assumptions and limitations of the study. We also clarified that we propagated drug-free controls, which was not clear in the previous manuscript. We thank the editor and reviewers for improving this manuscript.

1) A main concern is in the interpretation of populations having the same genetics but different phenotypes (eg lines 300-310) – this conclusion is wrong given the experimental design. The authors provide several explanations but leave out a high likelihood alternative which is that different experiments consist of mixed populations with diverse genetic makeups. Alternatively, if the authors meant instead to focus on specific mutations (ie same mutations/localized mutations) regardless of the background, then this conclusion is trivial – it is known that pleiotropy exists

We agree with the reviewer (see also comment #8). We have modified the text acknowledging that we are measuring the phenotype of mixed populations that could be genetically heterogeneous, and we have also toned down the conclusion of this section. (Lines 324-332)

Text explicitly addressing this comment:

"As the resistant phenotype was measured in mixed populations with diverse genetic backgrounds, it is possible that even though a resistance allele is fixed, different genotypes within each population could explain the phenotypic differences" and regarding the effects of specific mutations: "Follow-up experiments with reconstructed variants in isogenic backgrounds are needed to confirm this hypothesis."

2) I agree with the author's definitions of change due to history vs change due to selection given the evolution they designed – and I found this quite innovative and overall strong. However, the definition and calculations of evolution due to chance here are shaky given the three replicates only. There are plenty of extremely rigorous estimates of genetic drift during bacterial evolution (see all of Lenski's work) – three populations is quite small to make that claim and run the statistics. Including chance as a possible contributor didn't add much to the story, and it raised more questions than answers given the extremely diverse genetics that were observed, and that data of the mixed populations below 75% was not included.

We appreciate these important points and are glad to provide more nuance to the argument that the role of chance can be determined with our experimental design.

We agree that three populations seems like a small number of replicates to be able to study the effect of chance. However, we are not analyzing the effect of chance in just three replicates, but rather in 3 replicates of 6 ancestors in two independent experiments. In fact, both the genotype and the phenotype results show that most of the replicates started from a single ancestor followed a unique evolutionary trajectory, highlighting the importance of chance in our experiment. Chance is one of the principal forces determining variance in evolution. In the absence of chance, evolution from a single genotype will lead to the fixation of the most fit genotype in every experiment. We only detect 3 events of parallel evolution at the nucleotide level in replicates coming from the same ancestor (B3 and P2 in CAZ, and P1 in IMI) but each of those populations acquired unique mutations during the experiment. The fact that we don’t detect the same evolutionary trajectories in the replicates from a given ancestor -- a result that would be expected in the absence of chance -- shows that the experimental setup is able to detect the effects of chance in the evolution of AMR.

Once we acknowledge that the role of chance can be detected in our experiment and knowing that the number of replicates is limited, we decided to assess the statistical evidence of each force by means of a Bayesian approach, which works particularly well when sample sizes are small and exhibits some advantages over frequentist methods (Wagenmakers, 2007). Specifically, for a given force we computed the Bayes factor that results from comparing the regression model excluding that force against the full model that includes all three forces. This approach enables assessment of how powerful the included force is with respect to its absence according to the observed data. In the case of chance, the support for its inclusion was very strong in most scenarios. On the other hand, an alternative, traditional analysis using bootstrap showed that the contribution of chance was statistically different to the contribution of history in most of the cases analyzed (Figures 2E and 3E).

Finally, we would like to clarify that we have included in our analysis the data of all mutations detected above 5% of frequency. Please also see point #6.

3) Experiments are flawed without a no drug control. Was the reversion of CIP resistance specific to the new drug selection, or is it simply the lack of CIP pressure? This is important as the authors make claims about drug-specific evolutionary tradeoffs and collateral sensitivity. The methods in line 428: "We froze 1 mL of the control populations on days…" but this is the only place a control population is mentioned.

We understand the concerns of the reviewer and we agree with them. With our experimental setup, we can conclude only that the reversions were produced in planktonic conditions, in the presence of CAZ or IMI and in the absence of CIP. We have clarified this in the text (Line 370).

We understand that the reversions were produced when the populations were exposed to CAZ or IMI, and whether they also happen in the absence of the antibiotic is beyond the scope of the paper. See Dunai et al., eLife 2019, a brilliant example where genotypic reversions in antibiotic-free environments were analyzed.

We did not find it necessary to propagate all six clones in triplicates in the absence of CIP to analyze the role of the evolutionary forces during evolution in a new antibiotic. However, we valued the need for a drug-free control of the overall experiment. Therefore, we propagated control populations of the original ancestor in the absence of antibiotic for 12 days. Two replicates were sequenced and the two intergenic mutations observed in the control populations were subtracted from the propagated lines in the presence of CAZ or IMI. We apologize for this confusion and we have clarified it (Lines 484, 501).

4) Much of the data was difficult to follow, such as the fold changes for MICs – for example, following the discussion of CIP resistance, maintenance, and reversion was nearly impossible with the reference to figure 4 fold changes only. Also going back and forth between IMI and CAZ, which is discussed in parallel near the beginning, but then broken down later on, is another example.

The figure of the fold changes is just a summary of figures 2 and 3 in a different representation. We apologize for the confusion, and to improve the clarity of the manuscript, we have referenced the text and figure 4 (Lines 322, 365, 367 and 372). In addition, to facilitate the reading of the paper, we have outlined which sections describe results from the CAZ evolution and which section describes IMI results (Lines 173-176). We believe that describing both experiments together was more challenging to follow and decided to split them in two parts.

5) Key experimental details are missing. What volume were the populations propagated in? Population experiments were not well described – how were replicate MICs initiated? How much volume was used for sequencing? So on and so forth. These are particularly important to interpret the results.

Thank you. We have carefully reviewed the methods and clarified all missing details (Lines 472-475, 491-494, 503-509, 516-519, 547-548).

6) Why was a 75% threshold used to determine alleles if the populations were sequenced to a level of 5%?

We believe that this confusion comes from the legend of the Figure 4: “Only genes in which mutations reached 75% or greater frequency or that became mutated in more than one population are shown here”. We only used a 75% threshold to represent the mutations in Figure 4. We decided to show just mutations with >75% of frequency for clarity in the figure. All mutations detected in the experiment can be found in Table S3. Importantly, for all the analyses, including the estimations of the forces at the genotypic level, all observed mutations were used (Lines 264-265). We have clarified this in the methods section (Lines 547-548).

7) Despite being described in great detail in the methods, it doesn't come across to non-modeler such as myself exactly how selection, history, and chance were quantified in the phenotypic experiments shown in figure 2. I am a bit clearer as to how these were quantified for the genomic data. However, I think a broad readership may benefit from a layman's description of how these were differentiated. It is vital to the science and data shown and I would certainly have appreciated a clarification, especially for the phenotypic data in figure 2.

We added a brief description of how the forces were differentiated in the main text just before the initial results and in the methods (Lines 168-176).

8) In line 307. In the absence of bacterial genetic experiments to confirm that these historical infections are actually driving the 4 fold differences in phenotype, I think this inference needs to be toned down.

We agree that we cannot definitively determine whether previous mutations (a historical effect) or new mutations acquired during in different genes (chance + selection) are driving the 4-fold changes. Therefore, we have toned down the conclusion (Lines 325 – 336). We have acknowledged this comment with the line:

" Follow-up experiments with reconstructed variants in isogenic backgrounds are needed to confirm this hypothesis"

9) Line 343: Is there a benefit to be had of fitness experiments in antibiotic-free medium to confirm the supposition made here?

This is a really interesting point. We do not believe that fitness experiments in antibiotic-free medium confirm the supposition made in the line 343. Evolution experiments can be viewed as optimum competition experiments. Arguably, these designs are the best experiments to test the fitness of a given genotype, as the genotype of interest is competing not only against one strain like in pairwise competitions, but with all other contending mutations that arise each generation. The high level of parallelism found is the best indicator that the reversions are adaptive in those propagated populations.

We do not know if the reversions produce any benefit in the absence of antibiotics, and whether the reversions are beneficial in the absence of CAZ or IMI is trivial for this experimental setup, as those genotypes were selected in the presence of CAZ or IMI. However, we can confirm that those reversions were adaptive in planktonic conditions, in the absence of CIP, and in the presence of CAZ/IMI.

We have clarified that the reversed mutations initially imposed a fitness cost in presence of CAZ or IMI and in the absence of CIP (Line 388).

10) In some cases, I found the framing of the results in terms of history, chance, and selection to be a bit overly general, which sometimes obscured the specific results being reported. The paper could be improved by using more specific language-perhaps restricted in scope- in describing and interpreting the results, both because 1) it's not obvious to me that the results would apply generally to antibiotic resistance beyond the very nice, but potentially system-specific, results presented here; and 2) the terms themselves (history, chance, selection) could conceivably have different meanings in different contexts (more on this below).

Thanks, we better acknowledge these limitations. We have included a paragraph discussing the generality of our results in the discussion (Lines 437-450), and we have included a paragraph analyzing the assumptions done to analyze the forces (Lines 570-591). Also, see responses to #11, #12 and #13.

11) The study design has been used in numerous previous studies; it is well established, elegant, and has given rise to many new insights. However, as I understand it, there are some inherent assumptions of the approach that should be briefly discussed. Most notably, does the approach inherently assume that the effects of history, chance, and selection are additive (or perhaps linear) in some sense (in terms of phenotypic variance measurements or the Manhattan-based genotype metric)? While this simplifying assumption seems critical to the power of the approach, it is not clear to me that this assumption holds in general. When I try to think of this in terms of, say, a simple population dynamics model, the terms history, chance and selection are themselves somewhat nebulous, and it's not clear to me that they could be unambiguously and uniquely defined even in simplified theoretical models (or more directly, that the variance-based phenotype measures correspond to well-defined features or parameters of population dynamics models). I say this not to criticize the approach-again, its power lies in the simplicity of the design and the intuitive value of separating these three evolutionary features and attempting to quantify their contributions. But I think the article could be strengthened by briefly discussing the underlying assumptions-ideally by pointing to previous work (if it exists) that establishes that the features are additive and measurable in the sense required by the experimental design. If not, I think it would be worth discussing that limitation briefly, as I worry the inherently nonlinear nature of these very complicated, evolving systems could lead us to misinterpret the results. Given the general success of similar approaches in past work, I suspect the authors have thought through these issues in detail; discussing those points might open the paper to a broader audience not intimately familiar with all the previous studies.

Thank you for this thoughtful comment. To clarify, the history, chance, and adaptation study design itself does not make an assumption about the linearity or additivity of effects. The only linear assumption adopted in this study was through the use of a nested linear mixed model to estimate the forces in the phenotypic case. We opted to use such a model for its simplicity. Using more complex models that could account for non-linear effects, while being promising direction, was beyond the scope of the study.

Briefly, in the case of phenotypic variation, variation between populations is simply partitioned between three possible pools: (i) differences between ancestral and evolved populations (selection), (ii) differences between evolved populations with different ancestors (history) and (iii) differences between evolved populations with the same ancestor (chance). For genotypic variation, the same logic applies, using differences in the frequencies of mutations attributed to the same three pools. Where non-additive effects become important to consider is in interpreting the differences between the phenotypic and genotypic metrics, which we discuss further below (see #13). We added discussion of these assumptions in Lines 591-613.

12) One example related (but not identical to) the point above: in the current experiment, the role of history is defined in terms of previous selection conditions (drug and growth phase). But the new evolution experiment itself has multiple time points, and even qualitatively distinct epochs (sub- and super-MIC drug). So one might argue that history is playing a role continuously throughout the experiment-history not merely of the previous selection in fluoroquinolones, but also history of the previous time points / epochs of the new evolution (β-lactams). My point is that it is important to clearly define the terms at all stages and discuss, at least briefly, the limitations of the definitions that are chosen.

We agree that history is playing a continuous role during the experiment. According to Travisano, (Travisano et al., 1995) “the set of potential adaptations is severely limited to inherited constitution, so that at every moment the course of evolution is contingent on prior (historical) events”. We now acknowledge this more clearly that history plays a continuous role in lines 71-74 and 606-613.

For example, the mutations acquired at day 1 on the experiment arose by chance, were selected for in presence of CAZ or IMI, with a crucial importance of the previous evolutionary history, i.e. evolution in presence of CIP. If we jump, for example, to day 3, the previous mutations acquired on day 1 are right now historical mutations. We can apply this logic to any combination possible: mutations on day 2 are historical mutations if you analyze their evolutionary consequences at day 4, 5, 6… It just depends on the question that you want to answer. What we are asking here is how the previous exposure to an antibiotic X in lifestyles A and B, influences the further adaptation to an antibiotic Y. Therefore, we arbitrarily chose that everything happened before the exposure to the second antibiotic was just evolutionary history.

Analyzing the literature, most of the experiments following the Travisano design chose arbitrarily what they consider evolutionary history and from when they consider selection + chance (this even varies somewhat in the original article). Then, the roles of the forces are analyzed at the end of the experiment. However, there is an elegant paper from Rebolleda-Gomez and Travisano, where they analyze the contribution of the three forces periodically during their whole experimental evolution and not just at the end of the experiment (Rebolleda-Gomez and Travisano, Evolution 2019).

13) While history, chance, and selection are quantified at both the genetic and phenotype level, it's not clear to me that these numbers can be directly compared to one another (though it's tempting to do so!). Could the authors briefly comment on the connection between these measures-that is, when (and to what extent) one would expect correlations between them (e.g. high levels of historical influence at the genetic level leads to high levels of historical influence at the phenotype (MIC) level….assuming the definitions used here).

This is an interesting question. The contributions of history, chance, and selection at the genotypic and phenotypic levels would be same if every mutation that arose had an equal effect on the phenotype (or at least that the frequency of each mutation in the population was proportional to its phenotype) and phenotypic effects were additive, with no epistasis. The greater the deviation from those assumptions, the greater the differences will be between the genotypic and phenotypic roles of history, chance and selection. However, the differences between genotypic and phenotypic contributions may suggest additional information about the evolutionary system. In our results, the differences between the genotypic and phenotypic analyses were modest, suggesting relatively small deviations from these assumptions. The largest distinction was that in imipemen-selected populations, history contributed about 45% of genetic variation (Figure 4), but only about 33% of variation in imipenem resistance. This indicates that the mutations contributing to history, those in which parallel changes were observed in populations derived from the same ancestor, had a smaller effect on imipenem resistance than mutations associated with other categories. Interestingly, the “historical mutations” in these populations occurred most often in populations with ancestors previously evolved under biofilm conditions, implying constraining effects of this prior environment. It may be that some of these mutations were adaptations to the return to a planktonic environment, rather than to antibiotic selection. We further explain the connection between the phenotypic and the genotypic metrics in the methods (Lines 570-585).

14) Do the authors have any thoughts on how the results might be affected by the fact that the new evolution experiments take place in planktonic (rather than biofilm) conditions? How might the results differ if they had been performed in biofilm, and what could you learn from the fully symmetric experiment (P/ B initial strains evolved in both P and B new selection conditions). The authors may wish to discuss this avenue for future work.

Thanks for the comment. This is an interesting question, and we have asked this question ourselves too. We decided to run the experiment planktonically for one reason: after the CIP exposition, planktonic populations showed cross resistance to CAZ via mutations in adeJ and biofilm populations showed collateral sensitivity to CAZ via mutations in adeL. We believed that mutations in adeL had the potential to constrain evolution in CAZ while mutations in adeJ would be selected for. In fact, we detect mutations in adeB, which is regulated by adeJ.

We believe that the general take home message of the paper would be the same: selection is unsurprisingly the predominant force in the evolution of AMR and produced convergent evolution even at the nucleotide level in some instances. Yet history and chance play clear and important roles in the emergence of new resistance phenotypes, and possibly more so in biofilm. However, we argue that the reversions probably are lifestyle dependent, and therefore the likelihood of detecting the same reversions is low if we replayed the experiment in biofilm.

We have included this discussion in the manuscript (Lines 437-450).

Author details

1. Alfonso Santos-Lopez

   Department of Microbiology and Molecular Genetics, School of Medicine, University of Pittsburgh, Pittsburgh, United States

   Present address

   Servicio de Microbiología. Hospital Universitario Ramón y Cajal and Instituto Ramón y Cajal de Investigación Sanitaria, Madrid, Spain

   Contribution

   Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Validation, Visualization, Writing – original draft, Writing – review and editing

   Contributed equally with

   Christopher W Marshall

   For correspondence

   alfonsosantos2@hotmail.com

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-9163-9947

2. Christopher W Marshall

   Department of Microbiology and Molecular Genetics, School of Medicine, University of Pittsburgh, Pittsburgh, United States

   Present address

   Department of Biological Sciences, Marquette University, Milwaukee, United States

   Contribution

   Formal analysis, Methodology, Software, Visualization, Writing – original draft, Writing – review and editing

   Contributed equally with

   Alfonso Santos-Lopez

   For correspondence

   christopher.marshall@marquette.edu

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-6669-3231

3. Allison L Haas

   Department of Microbiology and Molecular Genetics, School of Medicine, University of Pittsburgh, Pittsburgh, United States

   Contribution

   Formal analysis, Investigation, Methodology

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-2154-4328

4. Caroline Turner

   Department of Microbiology and Molecular Genetics, School of Medicine, University of Pittsburgh, Pittsburgh, United States

   Present address

   Department of Biology, Loyola University Chicago, Chicago, United States

   Contribution

   Formal analysis, Methodology

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-1347-518X

5. Javier Rasero

   Department of Psychology, Carnegie Mellon University, Pittsburgh, United States

   Contribution

   Formal analysis, Methodology, Software

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-1661-2856

6. Vaughn S Cooper

   1. Department of Microbiology and Molecular Genetics, School of Medicine, University of Pittsburgh, Pittsburgh, United States
   2. Center for Evolutionary Biology and Medicine, University of Pittsburgh, Pittsburgh, United States

   Contribution

   Conceptualization, Funding acquisition, Project administration, Supervision, Writing – original draft, Writing – review and editing

   For correspondence

   vaughn.cooper@pitt.edu

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-7726-0765

• 5,231

  Page views

• 647

  Downloads

• 13

  Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.