History of Antibiotic Adaptation Influences Microbial Evolutionary Dynamics During Subsequent Treatment

Antibiotic resistance is a growing healthcare concern whereby bacterial infections are increasingly difficult to eradicate due to their ability to survive antibiotic treatments [1]. There have been reported cases of resistance for nearly every antibiotic we have available [2]. Coupled with the fact that the antibiotic discovery pipeline has slowed over the past few decades [3], there is a dire need to find better treatment strategies using existing antibiotics that can slow or even reverse the development of resistance.


Antibiotic resistance is a growing healthcare concern whereby bacterial infections are increasingly difficult to eradicate due to their ability to survive antibiotic treatments [1]. There have been reported cases of resistance for nearly every antibiotic we have available [2]. Coupled with the fact that the antibiotic discovery pipeline has slowed over the past few decades [3], there is a dire need to find better treatment strategies using existing antibiotics that can slow or even reverse the development of resistance.

Adaptive laboratory evolution is a technique that can be used to study and test evolutionary principles in a highly controlled laboratory setting [4]. Microorganisms with short generation times such as bacteria are especially amenable to adaptive laboratory evolution and can be adapted to an environment through repeated cycles of growth in a specific media environment, dilution of the culture, and subsequent passaging into fresh media [5]. Multiple replicates for each condition can be evolved in parallel to investigate the reproducibility of evolutionary dynamics [6]. The evolutionary trajectories of the bacteria can be measured as they adapt to different nutrients and stressors over time [7]. Whole-genome resequencing on the evolved strains can subsequently be used to determine the mutations that occurred that may be associated with the observed phenotypes [8,9].

Adaptive laboratory evolution can be used to study the development of antibiotic resistance in bacterial pathogens [10]. Resistance to antibiotics is an evolutionary response of bacteria to withstand and survive the effects of the stressor. Deliberately evolving bacteria to withstand antibiotics through experimental evolution can yield insights into the evolutionary dynamics and trajectories of this adaptive process [11,12]. These evolution experiments can provide a longer-term perspective, which can yield information for the design of novel treatment strategies that can reduce the rate of resistance evolution or potentially even reverse the effects of resistance [13–15].

Recent studies have explored how adaptation to an antibiotic can cause bacteria to concurrently become more susceptible or more resistant to other drugs, an effect termed collateral sensitivity or collateral resistance [14,16,17]. Collateral sensitivities between drugs have been used to design drug cycling strategies and to explain the decreased rate of adaptation to certain antibiotics [12,14,18–23]. Drug deployment strategies that exploit such collateral sensitivities between pairs of antibiotics to minimize resistance evolution have been tested in vitro. A recent study determined the collateral sensitivity drug interaction network in Escherichia coli and demonstrated how an alternating sequential treatment of 2 reciprocal, collaterally sensitive antibiotics can slow down the rate of resistance evolution [14]. In this drug cycling strategy, the development of resistance to 1 drug concurrently increased the sensitivity to the second drug, and this allowed wild-type cells to outcompete the resistant cells when exposed to the second drug. In a different study, evolution experiments of alternating sequential therapies of pairs of antibiotics were performed on Staphylococcus aureus, and the study showed that the alternating treatments slowed the rate of resistance development compared to single-drug treatments [12]. Consistent with the Ecoli study, this study found that collateral sensitivities could explain the evolutionary constraints in the cases in which alternating treatment resulted in decreased resistance development compared to the single-drug treatment.

Most of the prior studies that test the use of alternating antibiotic therapies to reduce the rate of resistance development employ an adaptive laboratory evolution scheme in which the drugs are switched at daily or subdaily intervals with the purpose of testing if rapidly changing antibiotic environments can diminish the rate of drug-resistance adaptation [12,19,23,24]. In this study, we expand on these prior works, but we are not focused on studying the evolutionary dynamics of bacteria adapted to rapidly changing drug environments. Rather, we explore the evolutionary dynamics of sustained, longer treatments of drugs and how the development of high levels of resistance to 1 drug influences the subsequent dynamics of sustained adaptation to a second drug. In clinical settings, when antibiotic cycling strategies are employed, they are used typically at the level of the hospital ward, and the cycling of antibiotics are often done at monthly intervals [25,26]. The rationale here is that if resistance to 1 drug arises after frequent use in a ward, switching to an antibiotic of a different class may allow resistance rates to the withdrawn drug to stabilize or even fall, enabling the first drug to be efficiently reintroduced again at a later time [27]. This practice of cycling drugs of different classes over the course of monthly intervals is done empirically, and it remains unclear how these regimens constrain the evolutionary dynamics of antibiotic resistance development. Here, we explore the evolutionary trajectories of bacteria as they evolve high levels of resistance to 1 antibiotic and the subsequent trajectories as the selection pressure from the first drug is withdrawn and replaced with the sustained pressure of a different drug. It remains unexplored how prior adaptation to 1 drug environment affects the evolutionary dynamics of a bacterial population during subsequent adaptation to a second drug in terms of the amount of resistance it can potentially develop and the resistance profile of the first drug. Collateral sensitivities and collateral resistances between 2 drugs have been studied in the context of adaptation to single drugs [21], i.e., as bacterial populations evolve and become resistant to 1 drug, do the cultures concurrently become more resistant or sensitive to other drugs? In this study, we focus not on if bacteria become concurrently more resistant or sensitive to other drugs, but rather, if adaptation to 1 drug constrains or potentiates the evolutionary dynamics to sustained adaptation to a second drug. How does the history of adaptation to 1 drug influence the subsequent adaptation to a second drug? If there are such historical dependencies, can we use this knowledge to design sequential therapies that slow down the evolution of resistance to the drugs used? What happens to the previously developed resistance once the drug pressure is taken away or switched to a different drug? Do compensatory adaptations sustain the high resistance, or do the bacteria revert and become susceptible again [28]? The answers to these questions are important for understanding how bacteria adapt to different antibiotic environments. Bacterial pathogens have complex evolutionary histories and elucidation of any historical dependencies of antibiotic resistance evolution would allow for rational forecasting of future resistance development and would aid in the design of strategies for mitigating antibiotic resistance.


Adaptive evolution of Paeruginosa to sequences of antibiotics

To test how different antibiotic-resistance backgrounds affect the subsequent adaptation dynamics when evolved to a new antibiotic, we used a laboratory evolution approach to evolve Paeruginosa to all 2-drug sequences of the 3 clinically relevant drugs piperacillin (PIP), tobramycin (TOB), and ciprofloxacin (CIP). In each of the experimental sequences, Paeruginosa was subjected to 20 days of adaptation to each drug by serially passaging parallel replicate cultures to increasing concentrations of the drugs followed subsequently by 20 more days of adaptation to each of the 3 drugs or to lysogeny broth (LB) media without a drug (Fig 1A). Additional parallel replicates were adapted to LB media without a drug for 40 days as a control. For each drug treatment, changes in the resistance to the other 2 drugs were concurrently measured (Fig 1B). Minimum inhibitory concentration (MIC) gradients in microtiter plates were used to simultaneously measure the drug resistance level and to propagate the bacteria daily. To adapt the bacteria to a drug, a sample is taken from the population from the well of the highest drug concentration that allowed for growth (i.e., MIC/2) and then used to inoculate a new MIC gradient. This serial dilution cycle is done daily. More explicitly, 20 μl of culture is sampled from the well of the highest concentration that allowed for growth, then diluted in 5 ml of fresh LB media, and then this diluted culture is used to inoculate a new MIC gradient. This dilution protocol results in a daily dilution factor of the bacterial population of 1/500 (Materials and methods, Fig 1B). S1 Data provides the estimated number of generations per day for the evolved lineages based on the daily measurements of the OD600. For each lineage, the OD600 values are fairly consistent from day to day, and so with a dilution factor of 1/500, the cultures undergo approximately 9 generations of growth per daily dilution cycle (S1 Data).

Fig 1. Adaptive evolution of Paeruginosa to 3 antibiotics.

(A) Ancestral Paeruginosa PA14 was evolved daily for 20 days to piperacillin (PIP), tobramycin (TOB), ciprofloxacin (CIP), and lysogeny broth (LB) media. In the following 20 days, the 1-drug–resistant lineages were passaged further to the first drug, as well as subpassaged to the other 2 drugs, and to LB media. (B) Bacteria were taken from the highest concentration that allowed growth (defined as OD600 > 0.1), diluted in fresh LB, and inoculated into fresh minimum inhibitory concentration (MIC) gradients, corresponding to a daily dilution factor of 1/500. After overnight incubation, the process is then repeated.

We observed differences in final resistance levels to the different drugs depending on the history of past treatments (or lack of treatments), an effect we call drug-order–specific effects of adaptation. Our results show that a history of past drug adaptation can affect the rate at which resistance can potentially arise when subsequently adapted to a new antibiotic. Furthermore, in some cases, adaptation to a second drug or to LB can partially or fully restore sensitivity to the first drug. These observations suggest that in order to limit the rate of development of antibiotic resistance, it is important to consider which drugs a bacterial population may have been exposed to in the past when choosing which drugs to subsequently deploy.

The 3 drugs tested have different mechanisms of action and are clinically used to treat Paeruginosa infections [29]. Piperacillin is a beta-lactam that inhibits cell wall synthesis [30]; tobramycin is an aminoglycoside that binds to the prokaryote ribosome and inhibits protein synthesis [31]; and ciprofloxacin is a fluoroquinolone that binds DNA gyrase and inhibits DNA synthesis [32]. We chose to study these 3 antibiotics because of their common use in the clinical setting to treat Paeruginosa infections [29], their diverse mechanisms of action, and their well-studied resistance mechanisms [33]. Adaptive evolution for 20 days to these drugs individually resulted in 1-drug–resistant mutants denoted as PIPR, TOBR, and CIPR. The Day 20 PIPR, TOBR, and CIPR populations had averages of 32-, 64-, and 64 times higher MICs to piperacillin, tobramycin, and ciprofloxacin, respectively, compared to their initial levels. To determine if the population MICs that were measured during the course of the adaptive laboratory evolution experiments were representative of individual colony MICs, we retrospectively measured the MICs of cultures grown from multiple revived colonies from the saved frozen stocks (S2 Data). Overall, 73% of the retrospectively measured MICs were within one 2-fold dilution step of the originally measured population MICs, which suggests that the reported MIC values for each of the evolved lineages are well representative of the bacterial populations (S1 Fig).

By following how the resistance to each of the 3 drugs changes for each of the drug sequences (Fig 2; S2 and S3 Figs and S3 Data), we observed 3 types of drug-order–specific effects in the MIC profiles (Fig 3). Note that for now, we focus on summarizing the different drug-order–specific effects (as seen by the changes in drug MICs), and later, we discuss several hypotheses for the underlying mechanisms of the drug-order–specific effects based on analysis of the genomic mutations of the adapted lineages. In the first type of drug-order–specific effects, adaptation to a second drug or to LB restores the susceptibility to the first drug (Fig 3A). In these experiments, we were first interested to see if the increases in MICs of the 1-drug–resistant lineages (Day 20 PIPR, TOBR, and CIPR) were permanent or transient. By evolving them to LB and hence removing the selection pressure of the drug for 20 days, we observed that the high MICPIP was maintained in Day 40 PIPRLB (Fig 3A [top], p = 0.80; Fig 2A), while MICTOB declines (leading to partial resensitization) in Day 40 TOBRLB (Fig 3A [middle], p < 0.0001; Fig 2E), and MICCIP declines (although not significantly) in Day 40 CIPRLB (Fig 3A[bottom], p = 0.18; Fig 2I). Thus, for these 3 treatments, removal of the antibiotic pressure can maintain the high resistance or lead to resensitization in a drug-specific manner. Similar trends were seen in a recent adaptive evolution study whereby Paeruginosa was evolved to tobramycin, ciprofloxacin, piperacillin/tazobactam, meropenem, and ceftazidime, followed by subsequent adaptation in the absence of the drug (growth medium only) to determine the effects of removing the drug selection pressure [34]. Similar to the patterns seen in our study, they observed that the tobramycin-resistant cultures partially resensitized, the ciprofloxacin-resistant cultures had a modest resensitization, and the 3 beta-lactam–evolved cultures maintained high levels of resistance.


Fig 2. Minimum inhibitory concentration (MIC) time courses of adaptive evolution. See image here: https://doi.org/10.1371/journal.pbio.2001586.g002

Plots show the MICs of the treatments to the 3 drugs and lysogeny broth (LB) over time. The top (A, B, C), middle (D, E, F), and bottom (G, H, I) rows show the MICs to piperacillin (PIP), tobramycin (TOB), and ciprofloxacin (CIP), respectively. The first, second, and third columns show the MICs of the PIPR, TOBR, and CIPR lineages, respectively. The dotted black lines mark the Day 20 MICs of the 3 drugs (i.e., MICPIP of Day 20 PIPR in (A), MICTOB of Day 20 TOBR in (E), and MICCIP of Day 20 CIPR in (I)). Error bars show SEM of 4 replicates per treatment. See S3 Data for the raw data.


Fig 3. Summary of the drug-order–specific effects. See image here: https://doi.org/10.1371/journal.pbio.2001586.g003

(A) The Day 20 PIPR, TOBR, and CIPR lineages were partially or fully resensitized to piperacillin (PIP), tobramycin (TOB), and ciprofloxacin (CIP), respectively, during subsequent adaptation to the other 2 drugs and/or lysogeny broth (LB). For example, in the top plot, a one-way ANOVA of the data shown yielded p < 0.01, suggesting that the mean MICPIP values of the 5 PIPR lineages being compared are different. A post-hoc Tukey’s honest significant difference (HSD) test then shows that the MICPIP of Day 40 PIPRCIPR is significantly less than that of Day 20 PIPR (p < 0.05), indicating that resensitization to piperacillin occurred after subsequent ciprofloxacin adaptation but not after tobramycin (p = 0.90) or LB (p = 0.80) adaptation. Similar types of comparisons are done for the TOBR and CIPR lineages in the middle and bottom plots, and we observed that not every type of subsequent adaptation led to resensitization. The table above the plots summarizes which subsequent adaptations (columns) led to the resensitization of the 3 drugs in their respective lineages (rows). (B) The MICTOB of Day 40 PIPRTOBRwas less than that of Day 20 TOBR (p < 0.05, Tukey’s HSD test), while the MICTOB of Day 1 TOBR and Day 20 PIPR were comparable. This suggests that the rate of tobramycin adaptation when there is a prior history of piperacillin adaptation is less than the rate of tobramycin adaptation when starting from the ancestor. (C) When bacteria are adapted to 2 drugs, the order of adaptation to those 2 drugs can lead to differences in the endpoint minimum inhibitory concentrations (MICs). For example, in the first plot, adaptation to ciprofloxacin followed by piperacillin led to a higher final MICPIP than the reverse order (MICPIP of Day 40 CIPRPIPR versus Day 40 PIPRCIPRp < 0.05) when they had initially comparable MIC values (MICPIP of Day 1 PIPR versus Day 1 CIPR). The second and third plots show how the final levels of MICPIP and MICCIP are different when tobramycin adaptation is followed by ciprofloxacin adaptation or with the reverse order. For all 3 panels, the asterisks denote p < 0.05 (Tukey’s HSD test), not significant (n.s.) denotes p > 0.05, and the color of the asterisks denotes which drug MIC is being compared. In the plots, for each lineage being shown, the black bar denotes the mean of the 4 individual replicate values (gray dots). See S3 Data for the raw data and S2 Textfor the calculations of the statistical tests.

Next, we were interested to see if evolving the 1-drug–resistant lineages to the other 2 drugs would show the same patterns seen as when evolved to LB. Interestingly, we saw unique outcomes for each of the 3 lineages. When Day 20 PIPR was evolved to tobramycin, the MICPIP of Day 40 PIPRTOBR remained high (p = 0.90), similar to how the MICPIP of Day 40 PIPRLB remained high (Fig 3A [top]). This result suggests that subsequent tobramycin adaptation has no role in altering the high piperacillin resistance. This specific order of drug treatments can then result in multidrug-resistant Paeruginosa cultures that are resistant to both piperacillin and tobramycin. On the other hand, when Day 20 PIPR was evolved to ciprofloxacin, the resulting cultures became resensitized to piperacillin (Fig 3A [top], p < 0.05), and the MICPIP declined to levels comparable to those of the initially susceptible cultures (MICPIP of Day 1 PIPR versus Day 40 PIPRCIPRp = 0.80), indicative of a full resensitization. Since resensitization did not occur after subsequent adaptation to tobramycin or LB, we suspect that the subsequent ciprofloxacin adaptation had an active role in the resensitization to piperacillin in such a way that tobramycin and LB did not. These results show that if a piperacillin-resistant culture (that is also sensitive to tobramycin and ciprofloxacin) is evolved to tobramycin, multidrug resistance can occur. However, if it is evolved to ciprofloxacin, despite the fact that ciprofloxacin resistance increases, the culture becomes susceptible to piperacillin again, making piperacillin a potentially rational choice for further treatment.

When Day 20 TOBR was evolved to ciprofloxacin, partial resensitization occurred (MICTOB of Day 20 TOBR versus Day 40 TOBRCIPRp < 10−5), and the MICTOB of Day 40 TOBR-CIPR fell to a comparable level as that of Day 40 TOBRLB (p = 0.98) (Fig 3A [middle]). This result suggests that the resensitization seen during the subsequent ciprofloxacin adaptation is not caused by the selection pressure of ciprofloxacin, but rather by the absence of the selection pressure of tobramycin. On the other hand, evolving Day 20 TOBR to piperacillin also led to a partial resensitization (MICTOB of Day 20 TOBR versus Day 40 TOBRPIPRp < 0.05) but not as much as it did when Day 20 TOBR was evolved to ciprofloxacin (MICTOB of Day 40 TOBRPIPRversus Day 40 TOBRCIPRp < 0.01) and LB (MICTOB of Day 40 TOBRPIPR versus Day 40 TOBRLB, p < 0.05). Because of this difference, we suspect that the maintenance of the comparably high tobramycin resistance is a consequence of the piperacillin selection pressure, since we observed that adaptation in the absence of the drug pressure in LB led to substantially greater resensitization. This case highlights how the removal of all drug pressures may lead to the resensitization of the culture more than with the treatment of the culture to a new drug. In conjunction with the result that subsequent tobramycin adaptation of Day 20 PIPR still maintained a high MICPIP, this case then also shows how regardless of the order, sequential adaptation to piperacillin and tobramycin leads to multidrug resistance of the 2 drugs.

Lastly, when Day 20 CIPR was evolved to piperacillin and tobramycin, both treatments lead to a partial resensitization to ciprofloxacin (Fig 3A [bottom]). During subsequent tobramycin adaptation, the decrease in the MICCIP from Day 20 CIPR to Day 40 CIPRTOBR (p < 0.01) was marginally more than the decrease in the MICCIP from Day 20 CIPR to Day 40 CIPRPIPR (p < 0.05) during subsequent piperacillin adaptation. As mentioned above, subsequent adaptation of Day 20 CIPR to LB led to a decrease in MICCIP that was not statistically significant; however, we argue that the decrease is comparable to that seen when adapted to piperacillin and tobramycin as the final MICCIP of Day 40 CIPRLB was not significantly different than that of Day 40 CIPRPIPR (p = 0.93) and that of Day 40 CIPRTOBR (p = 0.53). Hence, in this case, evolution of a ciprofloxacin-resistant culture to either a different drug or to a no-drug condition led to a partial resensitization of ciprofloxacin. Interestingly, we also observed that the resensitization that occurred during subsequent piperacillin adaptation happened more quickly than the resensitization that occurred during subsequent tobramycin and LB adaptation (Fig 2I). After 5 days of subsequent piperacillin adaptation (Day 25 CIPRPIPR), the MICCIP was significantly different than that of Day 20 CIPR (p < 0.001), while this was not the case after 5 days of subsequent tobramycin (p = 1.00) or LB (p = 0.57) adaptation. These cases in which partial or full resensitization to the first drug occurs after adaptation to a second drug or LB highlight opportunities in which resistance to 1 drug can potentially be reversed by treating with a second drug or by removing the drug pressure completely.

In the second type of drug-order–specific effects, prior adaptation to a first drug reduces the rate of subsequent adaptation to a second drug (such that the endpoint level of resistance to that second drug is lower compared to the amount of resistance developed when the Day 0 Ancestor is directly evolved to that second drug). We observed that evolution first to piperacillin reduces the rate of subsequent evolution to tobramycin (Fig 2D and 2E). That is, the MICTOB of Day 40 PIPRTOBR was less than that of Day 20 TOBR (Fig 3B, p < 0.05). This observation suggests that, in some cases, different bacterial populations may evolve resistance to a given antibiotic at different rates depending on the history of prior adaptations that the populations have experienced. Having knowledge of prior adaptations may then potentially be used to slow down the development of resistance to a drug if that drug is selected rationally. Interestingly, we observed no cases in which prior drug adaptation led to an enhancement in the rate of adaptation to a second drug.

The last type of drug-order–specific effects is when the final MIC of a drug is different after adaptation to a 2-drug sequence compared to after adaptation to the opposite order of the 2 drugs (Fig 3C). This third type of drug-order–specific effect exists as a consequence of a combination of the first type of effect (resensitization of the 1-drug–resistant lineages during subsequent adaptations to other drugs) and specific cases of collateral sensitivities during the adaptation of certain lineages. First, the MICPIP was higher when piperacillin was used after ciprofloxacin (Day 40 CIPRPIPR) compared to when piperacillin was used before ciprofloxacin (Day 40 PIPRCIPR) (Fig 3C [top], p < 0.05). In this case, adaptation to piperacillin first led to high levels of piperacillin resistance, and subsequent adaptation to ciprofloxacin led to the resensitization to piperacillin as discussed before (Fig 2A). On the other hand, even though adaptation to ciprofloxacin first led to a collateral sensitivity to piperacillin (S4A Fig [right], p < 0.01), subsequent adaptation to piperacillin resulted in a final MICPIP comparable to that of Day 20 PIPR (Fig 2C).

Next, we observed that during the adaptation to tobramycin followed by ciprofloxacin and vice versa, the final MIC values of piperacillin and ciprofloxacin were different depending on the order of adaptation to the 2 drugs (Fig 3C [bottom and middle]). With regards to the difference seen in the final MICCIP (Fig 3C [bottom], p < 0.05), the partial resensitization to ciprofloxacin starting from Day 20 CIPR during subsequent tobramycin adaptation (Fig 2I) resulted in the MICCIP to be less than adaptation to tobramycin first, followed by ciprofloxacin (Fig 2H). Finally, it was interesting that even though piperacillin was not the direct selection pressure, there was a difference in the final MICPIP level whether ciprofloxacin adaptation occurred after tobramycin adaptation or vice versa (Fig 3C [middle], p < 0.01). In this case, initial adaptation to tobramycin first did not affect the MICPIP (Fig 2B), but subsequent adaptation to ciprofloxacin resulted in a collateral sensitivity to piperacillin (S4C Fig, p < 0.01). On the other hand, as previously mentioned, adaptation to ciprofloxacin first initially resulted in the collateral sensitivity to piperacillin (S4A Fig [right], p < 0.01); however, the MICPIP returned to baseline values during subsequent adaptation to tobramycin (Fig 2C). Thus, regardless if ciprofloxacin adaptation occurred before or after tobramycin adaptation, ciprofloxacin adaptation led to piperacillin collateral sensitivity. However, in order to take advantage of this collateral sensitivity, ciprofloxacin adaptation should be used after tobramycin adaptation, rather than vice versa. In a contrasting example, we also found it interesting that while ciprofloxacin adaptation also led to collateral sensitivity of tobramycin, subsequent piperacillin adaptation did not cause the MICTOBto return to baseline levels (Fig 2F) in the manner in which subsequent tobramycin adaptation returned the MICPIP to baseline values (Fig 2C). Altogether, these cases highlight how treating an infection with a sequence of 2 drugs can result in different resistance profiles depending on the order used.

All the cases of collateral sensitivity that were observed occurred during ciprofloxacin treatment whereby ciprofloxacin adaptation resulted in a lower MIC of piperacillin or tobramycin compared to baseline levels (S4 Fig). First, adaptation to ciprofloxacin starting from the Day 0 Ancestor resulted in collateral sensitivity to both piperacillin (Fig 2C; S4A Fig [right], p < 0.01) and tobramycin (Fig 2F; S4A Fig [left], p < 0.0001). Next, adaptation to ciprofloxacin starting from both the 1-drug–evolved lineages Day 20 PIPR (Fig 2D) and Day 20 TOBR (Fig 2B) resulted in collateral sensitivity to tobramycin (S4B Fig, p < 0.01) and piperacillin (S4C Fig, p < 0.01), respectively. These results suggest that regardless of historical background, ciprofloxacin adaptation results in collateral sensitivity to the other 2 drugs. While we observed that collateral sensitivity of other drugs occurs only during ciprofloxacin adaptation, a recent study in which Paeruginosa ATCC 27853 was evolved to different antibiotics reported that evolution to tobramycin resulted in collateral sensitivity to piperacillin-tazobactam and ciprofloxacin, whereas we did not observe this effect [34]. Also, this study did not observe that adaptation to ciprofloxacin resulted in collateral sensitivity to piperacillin and tobramycin, as we reported here. We suspect that these inconsistences may be due to strain-specific differences in the different Paeruginosa strains used (strain PA14 was used in this study).

We were interested in measuring the fitnesses of the evolved lineages to see if the adaptations to the different treatments altered their growth dynamics. We measured the growth curves (OD600) of the 68 evolved replicate lineages as well as the Day 0 Ancestor in quadruplicate grown in LB for 24 hours (S5 Fig). The exponential growth rates were subsequently calculated from the growth curves (S6A Fig, S1 Text and S4 Data). While we observed many different growth rates across the different lineages, we did not observe any correlation between the growth rate and the change in MIC between the Day 20 1-drug–evolved lineages and the subsequent Day 40 lineages (i.e., altered growth rates could not explain the cases in which subsequent adaptation led to the maintenance of high resistance or resensitization to the first drug) (S6B Fig and S1 Text).

Genomic mutations of adapted lineages

We hypothesized that genomic mutations acquired during adaptive evolution contributed to the drug-order–specific effects observed in the MIC profiles. We sequenced genomes of the Day 0 Ancestor, Day 20 PIPR, TOBR, CIPR, and LB Control lineages and the Day 40 1-drug–and 2-drug–evolved lineages, as well as the LB Control lineages. Genome sequencing of the Day 20 and Day 40 mutants revealed a total of 201 unique mutations across the 56 samples consisting of 77 SNPs, 31 insertions, and 93 deletions (Fig 4; S7 Fig, S1 and S2 Tables). The 77 SNPs were found within 49 genes. Two SNPs were synonymous, and 6 were intergenic. To test how representative the sequencing results were of the mutant populations, we used PCR and Sanger sequencing to test for the presence of specific mutations in multiple colonies of different lineages after reviving the lineages from the saved frozen samples. We used the primers from S3 Table to test for the presence of 1 mutation from 1 replicate of each lineage, with 4 colonies of each lineage. Overall, while there may be limited heterogeneity in the populations with respect to a few of the mutations, the large majority of the mutations were homogeneous in the populations and fixed within the lineages (S1 Text).


Fig 4. Genomic mutations of the evolved lineages. See image here: https://doi.org/10.1371/journal.pbio.2001586.g004

Mutations for the Day 20 and Day 40 mutants are plotted according to position on the chromosome. Each lineage is labeled and has 4 tracks for the 4 replicates per treatment. The inner set of tracks are the Day 20 1-drug–evolved lineages, the middle set of tracks are the Day 40 1-drug–evolved lineages, and the outer set of tracks are the Day 40 2-drug-evolved lineages. The color of the track denotes the treatment during the first 20 days. The color of the plotted mutation denotes during which treatment the mutation occurred. For example, a blue dot on a yellow track denotes a mutation in a Day 40 CIPRTOBR replicate lineage that occurred during tobramycin adaptation (i.e., between Day 21 and Day 40). For the Day 40 1-drug–evolved lineages, circles denote mutations that occurred during the first set of 20 days, and triangles denote mutations that occurred during the second set of 20 days. Large rectangles denote large genomic deletions. Numbers in parentheses next to gene names indicate the number of unique mutations that occurred in that gene.

While some genes were mutated during evolution to all drugs, other mutations were drug-specific and were related to the drugs’ primary mechanisms of action as would be expected (S4 Table). Genes encoding transcriptional regulators for multidrug efflux pumps were commonly mutated during evolution to all 3 drugs (mexCmexRmexSnalCnalDnfxBparS) [35]. Ribosomal proteins (rplJrplLrpsLrplF) [36] and NADH dehydrogenase subunits (nuoBnuoGnuoLand nuoM) [37,38] were frequently mutated during tobramycin evolution. The most commonly mutated gene was fusA1, which encodes elongation factor G and was mutated in 11 different replicate lineages adapted to tobramycin. fusA1 has been observed to be mutated in clinical isolates of Paeruginosa [39–41], as well as in adaptive evolution studies to aminoglycosides in Paeruginosa [34] and Ecoli [12,16,18]. Mutations in fusA1 may also contribute to altered intracellular (p)ppGpp levels, which may modulate virulence in Paeruginosa [41]. Mutations in gyrA and gyrB were observed during ciprofloxacin evolution, but none were observed in parC and parE (the other genes of the quinolone resistance-determining region [29]). Lastly, genes encoding peptidoglycan synthesis enzymes (dacCmpl) and beta-lactamase regulators (ampR) were mutated during piperacillin treatment. Many of these genes have also been observed to be mutated during human host adaptation of Paeruginosa [42], highlighting the importance of several of these clinical resistance determinants (S1 Text).

We next analyzed the genomic mutations to see how the historical context affects which mutations occur during adaptation to a drug. For example, how do the mutations that occur during adaptation to piperacillin only (Day 20 PIPR and Day 40 PIPR) compare to the mutations that occur during piperacillin adaptation when there is a prior history of adaptation first to tobramycin (Day 40 TOBRPIPR) or ciprofloxacin (Day 40 CIPRPIPR)? To this end, we first categorized the genes in which mutations occurred into 23 broad categories based on the available literature and on the PseudoCAP functional classifications from the Pseudomonas Genome Database [43] (Table 1). Next, for each lineage, we tallied the number of times a gene in a functional category was mutated across the 4 biological replicates for each of the lineages (Fig 5). For a complete list of genes in each functional classification and descriptions of the genes, see S2 Table.

Table 1. Functional classifications of the mutated genes.


Fig 5. The frequency of mutated genes during piperacillin (PIP), tobramycin (TOB), and ciprofloxacin (CIP) adaptation depending on the historical background. See image here: https://doi.org/10.1371/journal.pbio.2001586.g005

The number of unique mutations observed in a gene in a functional class (rows) is shown based on the intensity of the color across all 4 biological replicates for each of the lineages (columns). The lineages are grouped according to the final (or only) drug that the lineage was adapted to in order to compare how historical context affects how often genes in the functional classes are mutated. For example, the first 4 columns (with red shading) correspond to the frequency of genes mutated in the lineages that were adapted to piperacillin only (Day 20 PIPR and Day 40 PIPR) and piperacillin after prior adaptation to a first drug (Day 40 TOBRPIPR and Day 40 CIPRPIPR). Note that the data in the Day 40 PIPR column correspond to additional mutations that occurred (between Day 21 and 40) and do not double count the ones from Day 20 PIPR column. As an example of how different genes are mutated during piperacillin adaptation under different historical contexts, the cells outlined by the dashed-black box show that regardless of historical context, all lineages that underwent piperacillin adaptation had mutations in genes related to the MexAB-OprM efflux pump. However, only the lineage that had prior ciprofloxacin adaptation (Day 40 CIPRPIPR) had mutations in genes related to the MexCD-OprJ and MexEF-OprN efflux pumps. Lastly, none of the piperacillin-adapted lineages had mutations in genes involved in the MexXY-OprM efflux pump. The cells outlined by the dashed-purple boxes show that while subsequent adaptation of Day 20 PIPR to ciprofloxacin (Day 40 PIPRCIPR) resulted in several mutations in genes involved in efflux pumps, subsequent adaptation to tobramycin (Day 40 PIPRTOBR) resulted in no mutations in genes involved in efflux pumps. The corresponding mutations that occurred are explicitly listed at the bottom. See the main text for more details of how this difference may play a role in the resensitization to piperacillin during subsequent ciprofloxacin adaptation of Day 20 PIPR. See S1 Table for the complete list of mutations and S2 Table for descriptions of the mutated genes.

We observed several general trends in the genes mutated during adaptation to the 3 drugs, depending on their historical context. In the lineages adapted to piperacillin, we saw history-dependent trends in the mutated genes that were related to multidrug efflux pumps (Fig 5, dashed-black box). While all the piperacillin-adapted lineages had mutations in genes related to the MexAB-OprM efflux pump (which is the primary efflux pump of piperacillin [44]) such as nalD and mexR (whose products repress the expression of mexAB-oprM [45]), the Day 40 CIPRPIPR lineage had additional mutations in the structural subunit genes of the other efflux pumps MexCD-OprJ (mexC) and MexEF-OprN (mexF). Lastly, no mutations in genes related to the MexXY-OprM pump were observed in any of the piperacillin-adapted lineages. With regard to adaptation to piperacillin only, most of the mutations that occurred in genes related to MexAB-OprM occurred within the first 20 days, with only a few additional mutations occurring between Day 21 and 40. Regardless of historical context, metabolic and cell wall genes tended to be frequently mutated in piperacillin-adapted lineages, whereas metabolic and cell wall genes did not seem to be consistently mutated across the tobramycin-adapted and ciprofloxacin-adapted lineages. This result is perhaps due to the fact that the primary target of piperacillin is cell wall (peptidoglycan) synthesis, which is largely a metabolic process. Interestingly, we also observed that the lineages adapted only to piperacillin (Day 20 PIPR) sustained large chromosomal deletions that were not seen in the lineages in which there was prior tobramycin or ciprofloxacin adaptation (Day 40 TOBRPIPR and Day 40 CIPRPIPR). We discuss and explore the potential implications of these large deletions below.

The tobramycin-adapted lineages consistently had mutations occur in ribosomal subunit genes and other ribosomal machinery genes, regardless of historical context. In the lineages adapted only to tobramycin, mutations in genes related to the ribosome, membrane, energy, and NADH dehydrogenase tended to occur by Day 20, followed by mutations in efflux pump–related genes by Day 40. The mutations in genes related to membrane, NADH dehydrogenase, and energy likely reflect the unique requirement of the proton-motive force for the uptake of aminoglycoside antibiotics [46], and the mutations occurring during tobramycin adaptation may contribute to the resistance by reducing the proton-motive force [16]. While we observed mutations in the NADH dehydrogenase genes in the lineages adapted only to tobramycin, we saw no such mutations in the lineages in which prior piperacillin or ciprofloxacin adaptation occurred (Day 40 PIPRTOBRand Day 40 CIPRTOBR). Also, while efflux pump–related genes were mutated in the Day 40 TOBR and Day 40 CIPRTOBR lineages, no such mutations were seen in the Day 40 PIPRTOBRlineages in which prior adaptation to piperacillin occurred (Fig 5, dashed-purple boxes).

The mutations in the ciprofloxacin-adapted lineages were fairly consistently distributed regardless of historical context. For all ciprofloxacin-adapted lineages, mutations were seen in genes related to DNA/RNA synthesis as expected, as well as in genes related to membrane, flagella, efflux pumps, metabolism, and transcriptional regulators. Mutations related to the MexAB-OprM, MexCD-OprJ, and MexEF-OprN efflux pumps (mostly in genes encoding negative regulators of the pumps) are seen in the ciprofloxacin-adapted lineages, reflecting the ability of these different pumps to extrude ciprofloxacin; however, no mutations were seen in genes related to MexXY-OprM, even though this pump is also known to contribute to fluoroquinolone resistance [44]. Further experiments in measuring the gene expression of the different efflux pumps may help elucidate the roles that these pumps play in contributing to the different drug-order–specific effects.

Next, we sought to determine if the patterns in mutated genes could explain the mechanisms of some of the drug-order–specific effects that were observed in the MIC time courses. We first discuss the cases of resensitization or maintenance of high resistance in which the 1-drug–evolved lineages were subsequently adapted to the other 2 drugs or to LB (Fig 3A). While subsequent adaptation of Day 20 PIPR to LB and tobramycin maintained high piperacillin resistance, subsequent adaptation to ciprofloxacin led to full resensitization to piperacillin (Fig 3A [top]). We hypothesize that these differences stem from the different efflux pump-related genes that were mutated in these lineages (Fig 5, dashed-purple boxes). Evolution of the Day 0 Ancestor to piperacillin resulted in 2 different SNPs in nalD, and 1 SNP in mexR across the 4 biological replicates of Day 20 PIPR, likely leading to the overexpression of the MexAB-OprM efflux pump [45]. We suspect that MICPIP remained high during subsequent adaptation to LB and tobramycin due to continued overexpression of MexAB-OprM.

However, when Day 20 PIPR was adapted to ciprofloxacin, several mutations occurred in genes related to other efflux pumps, including 1 in mexA, 2 in nfxB, and 2 in mexS (Fig 5, dashed-purple boxes). In particular, mexS encodes a negative regulator of the expression of MexEF-OprN, and mutations in this gene likely lead to the overexpression of the efflux pump [47]. Interestingly, expression of MexEF-OprN has been observed to correlate inversely with the expression of MexAB-OprM [47,48]. Hence, we suspect that the resensitization to piperacillin when Day 20 PIPR was subsequently adapted to ciprofloxacin may be have been due to a concurrent decrease in MexAB-OprM expression (leading to reduced piperacillin efflux) as MexEF-OprN expression increased. That is, it is possible that the mutations that occurred during ciprofloxacin adaptation that led to the overexpression of MexEF-OprN negated the effects of the mutations that occurred during prior piperacillin adaptation that led to overexpression of MexAB-OprM. Furthermore, we observed no mutations in efflux pump–related genes in Day 40 PIPRTOBR (Fig 5, dashed-purple boxes), which supports the notion that because no mutations occurred, which would have negatively correlated with the expression of MexAB-OprM, expression of this efflux pump was maintained throughout the subsequent adaptation to tobramycin, and hence the MICPIP remained high.

We observed that subsequent adaptation of Day 20 TOBR to LB and ciprofloxacin resulted in a partial resensitization to tobramycin, and that while subsequent adaptation to piperacillin also led to a significantly lower MICTOB, it was not as low as that of Day 40 TOBRLB and TOBRCIPR(Fig 3A [middle]). In this case, the partial resensitization during subsequent adaptation to LB may be attributable to adaptive resistance of aminoglycosides in Paeruginosa. Adaptive resistance is a phenomenon in which resistance to a drug is transiently induced in the presence of the drug, and resistance recedes upon the removal of the drug [49]. In contrast to acquired resistance, which is mediated through genetic mutations, adaptive resistance is explained by phenotypic alterations that allow for temporary increases in resistance. Paeruginosa is known to exhibit adaptive resistance to aminoglycosides [50,51], and it is primarily mediated through up-regulation of MexXY-OprM during drug exposure and subsequent down-regulation after the removal of the drug [52]. We suspect that the partial resensitization during subsequent ciprofloxacin adaptation is also a consequence of adaptive resistance once the tobramycin selection pressure is removed. We further speculate that during the initial adaptation to tobramycin, the increase in tobramycin resistance was a combination of adaptive resistance and acquired resistance from accumulation of the mutations as seen in Day 20 TOBR. Thus, the resensitization during subsequent LB and ciprofloxacin adaptation was not a full resensitization but rather a partial one, perhaps reflecting the remaining contribution of the acquired resistance. Lastly, with regards to Day 40 TOBRPIPR, it is unclear how subsequent piperacillin adaptation seemingly resulted in the maintenance of high MICTOB compared to that of Day 40 TOBRLB and TOBRCIPR. We hypothesize that the subsequent piperacillin adaptation somehow counteracted the resensitization effects of adaptive resistance, even when the tobramycin selection pressure was removed.

The mechanism of ciprofloxacin resensitization when Day 20 CIPR was subsequently adapted to LB, piperacillin, and tobramycin remains unclear (Fig 3A [bottom]). While reversion of aminoglycoside sensitivity has been the most characterized case of adaptive resistance in Paeruginosa, other studies have suggested that adaptive resistance may be prevalent in other classes of antibiotic classes as well, and that it may be mediated by epigenetic processes such as methylation and stochastic gene expression [53], particularly affecting the expression of efflux pumps [54]. It could be possible that adaptive resistance partially explains the resensitization to ciprofloxacin. We also note that qualitatively, there was much more variability in the MIC time courses between the individual replicates of the CIPR lineages, as seen by the larger error bars in Fig 2I, compared to that of the PIPR (Fig 2A) and TOBR (Fig 2E) lineages. Taken together, further investigation of the partial ciprofloxacin resensitization is needed.

While we observed clear cases of collateral sensitivity develop to piperacillin and tobramycin during the course of ciprofloxacin adaptation (S4 Fig), other adaptive evolution studies of Paeruginosa evolved to ciprofloxacin showed mixed results. In one study, the adaptation of Paeruginosa ATCC 27853 to ciprofloxacin showed no change in the MIC of 3 different beta-lactams (including piperacillin-tazobactam), nor of tobramycin [34]. In another study, while no statistical significances were assigned, adaptation of Paeruginosa PAO1 to ciprofloxacin appeared to result in slight collateral sensitivities to piperacillin-tazobactam and tobramycin in some of their replicates. Nevertheless, in our study, we hypothesize that the collateral sensitivity to piperacillin and tobramycin during ciprofloxacin adaptation is attributable to the mutations seen in nfxB (which encodes a transcriptional repressor that regulates MexCD-OprJ [55]) in the Day 20 CIPR lineages. Three of the Day 20 CIPR replicates had deletions in nfxB(15, 13, and 16 base pairs), likely resulting in the inactivation of NfxB and concomitant up-regulation of MexCD-OprJ and increased ciprofloxacin resistance [56]. In fact, nfxB mutants have been reported to be hypersusceptible to certain beta-lactams and aminoglycosides [57,58].

Lastly, with regards to the decreased rate of tobramycin adaptation given a history of prior piperacillin adaptation (Fig 3B), we attribute this effect to the large chromosomal deletions that were sustained in 3 of the 4 Day 20 PIPR replicates. The consequences of these deletions are discussed in more detail in the subsequent sections of the manuscript. In summary, based on the genomic mutations, we have presented our interpretations of potential mechanisms that contribute to the drug-order–specific effects. These include how historical context can influence the frequency of mutations in certain genes, the varying contributions of adaptive and acquired resistance to total resistance, and specific cases of inverse correlation of the expression of different efflux pumps. While mutations are likely not the sole determinants of the differences [34,59], many of the observed genomic mutations can partially explain the drug-order–specific effects.

Drug history dependence of pyomelanin hyperproduction

One striking mutation we observed was that 3 of the 4 replicates of Day 20 PIPR (Day 20 PIPR-1, -2, and -3) had large, approximately 400 kbp deletions (corresponding to approximately 6% of the genome) in a conserved region of the chromosome (Fig 4 [large red rectangles]; S5 Data), suggestive of selective genome reduction [60–63], and have been associated with directed repeats [64] and inverted repeats [65] at the boundaries of the deletions. These large deletions were also fixed in the corresponding Day 40 PIPRTOBR, Day 40 PIPRCIPR, and Day 40 PIPRLB lineages. Interestingly, the 3 PIPR lineages with these large deletions hyperproduced the brown pigment pyomelanin during piperacillin evolution, and this visually observable phenotype also persisted when evolved to tobramycin (PIPRTOBR), ciprofloxacin (PIPRCIPR), and LB (PIPRLB). The loss of hmgA as part of the large chromosomal deletions correlates exactly with the pyomelanin phenotype of these lineages. Indeed, hmgA mutants of Paeruginosa hyperproduce pyomelanin [66]. This observation shows that evolving to piperacillin results in a high probability of sustaining large deletions spanning hmgA, which results in the pyomelanogenic phenotype. However, when we evolved the Day 20 TOBR and CIPR lineages to piperacillin to yield the Day 40 TOBRPIPR and Day 40 CIPRPIPR lineages (4 replicates each), none of them became pyomelanogenic, suggesting that prior history of tobramycin or ciprofloxacin adaptation leads to a lower propensity of becoming pyomelanogenic when subsequently evolved to piperacillin. Interestingly, 1 of the Day 20 TOBR replicates became pyomelanogenic when subsequently evolved to ciprofloxacin (Day 40 TOBRCIPR-2). Hence in this study, pyomelanin hyperproduction is a consequence of piperacillin and ciprofloxacin evolution, yet the likelihood to evolve this visually striking and observable phenotype depends on the history of prior drug adaptation.

While the 3 PIPR lineages that produced pyomelanin were not significantly more resistant to piperacillin than the nonpyomelanogenic PIPR lineage, pyomelanin-producing strains have been observed clinically [60] and have been shown to be more persistent in chronic lung infection models [66]. We tested the reproducibility of this example of a phenotypic dependence on the history of drug adaptation with a higher throughput approach. Starting with clonal populations of Day 0 Ancestor, Day 20 TOBR, and Day 20 CIPR, we seeded 92 replicate populations of each lineage into microplates, and we used a 96-pin replicating tool to serially propagate these populations and evolve them to increasing concentrations of piperacillin daily. The lineages that started from Day 0 Ancestor had the highest propensity to become pyomelanogenic (Fig 6A) compared to lineages starting from Day 20 TOBR (Fig 6B) or Day 20 CIPR (Fig 6C). Still, certain lineages starting from Day 20 TOBR and Day 20 CIPR did also produce pyomelanin, albeit with less propensity than starting from Day 0 Ancestor (Fig 6D; S8–S10 Figs).


Fig 6. Wild-type Paeruginosa has a higher propensity to become pyomelanogenic when evolved to piperacillin compared to tobramycin-resistant (TOBR) and ciprofloxacin-resistant (CIPR) lineages. See image here: https://doi.org/10.1371/journal.pbio.2001586.g006

We tested how common it was for piperacillin adaptation to lead to pyomelanin hyperproduction under different historical backgrounds. Ninety-two replicates of (A) Day 0 Ancestor, (B) Day 20 TOBR-3, and (C) Day 20 CIPR-1 were passaged daily to low, increasing concentrations of piperacillin for 10 days. Photographs of Day 7 of passaging show how the Ancestor had a higher propensity of evolving the pyomelanin phenotype during piperacillin treatment compared to evolution of Day 20 TOBR-3 and Day 20 CIPR-1. The complete set of photographs for all lineages tested is shown in S8–S10 Figs. (D) The number of visibly brown wells was tracked daily over the course of the 10 days of piperacillin evolution. Overall, Day 0 Ancestor had the highest propensity to become pyomelanogenic during piperacillin evolution, followed by Day 20 CIPR-3 and Day 20 TOBR-2. Interestingly, the number of brown wells for these lineages did not increase monotonically over time, suggesting heterogeneity in these populations, and that nonpyomelanogenic subpopulations outcompeted the pyomelanogenic ones in the wells that transiently turned brown.

Drug-order–specific effects in clinical isolates

To explore the relevance of our laboratory evolution results clinically, we tested for the drug-order–specific MIC evolutionary dynamics in clinical isolates. We first tested the evolutionary dynamics of clinical isolates that were resistant to piperacillin but susceptible to tobramycin and ciprofloxacin. We evolved 3 piperacillin-resistant clinical isolates of Paeruginosa to piperacillin, tobramycin, and ciprofloxacin for 10 days and tracked how the piperacillin resistance changed in these lineages. If the results from the adaptive evolution experiment applied to these piperacillin-resistant clinical isolates, then we would expect that evolving to tobramycin would not affect the high piperacillin resistance, but evolving to ciprofloxacin would restore susceptibility to piperacillin. As mentioned previously, evolving Day 20 PIPR to LB did not result in a reduction of MICPIP, which suggests that the resensitization to piperacillin when Day 20 PIPR was evolved to ciprofloxacin was a consequence of the switch to the ciprofloxacin drug pressure. Of the 3 isolates we tested, the evolutionary dynamics of 2 of these isolates matched these expectations (Fig 7; S11 Fig and S3 Data). After normalizing to Day 1 MIC values, the MICPIP after 10 days of ciprofloxacin adaptation was significantly less than the MICPIP after 10 days of LB adaptation in isolate #2 (Fig 7B, p < 0.05) and in isolate #3 (Fig 7C, p < 0.001), indicating resensitization to piperacillin during ciprofloxacin adaptation. This observation suggests that the MIC evolutionary dynamics we observed are not limited to laboratory strains of Paeruginosa and may be observed in diverse strains of Paeruginosa, including those originating from human patients. Note that these 3 clinical isolates were isolated from different patients, and their phylogenetic relatedness between each other and to the laboratory PA14 strain used in our study is untested. In isolate #1, there was no significant difference in the normalized MICPIP values after 10 days of adaptation to tobramycin, ciprofloxacin, and LB (Fig 7A, p = 0.237, one-way ANOVA). Interestingly, this isolate evolved higher levels of piperacillin and ciprofloxacin resistance than the other 2 isolates (S11 Fig and S3 Data), which suggests the possibility that adaptation to ciprofloxacin in these higher piperacillin-resistant cultures could still result in a restoration of piperacillin susceptibility.


Fig 7. Clinical isolates with high piperacillin resistance become resensitized to piperacillin following adaptation to ciprofloxacin. See image here: https://doi.org/10.1371/journal.pbio.2001586.g007

To see if we could recapitulate the adaptation dynamics of the minimum inhibitory concentration of piperacillin (MICPIP) when Day 20 PIPR was evolved to tobramycin and ciprofloxacin, we evolved 3 piperacillin-resistant clinical isolates of Paeruginosa to piperacillin, tobramycin, ciprofloxacin, and lysogeny broth (LB). (A) While the first isolate did not show restoration of piperacillin sensitivity during ciprofloxacin evolution as anticipated, (B and C) the other 2 isolates recapitulated this effect. In clinical isolates #2 and #3, the relative changes in the MICPIP when the isolates were evolved to ciprofloxacin were significantly different from the relative changes when evolved to LB at Day 10 (p < 0.05 and p < 0.001, respectively). For each of the 3 isolates, a one-way ANOVA was first performed on the Day 10 MICPIP values of the lineages evolved to LB, tobramycin, and ciprofloxacin. Error bars show SEM of 3 replicates per treatment. See S2 Text for the calculations of the statistical tests and S11 Fig for the original, prenormalized data.

In the next set of evolution experiments, we investigated the role that the large chromosomal deletions play in a drug-order–specific effect. We had observed that compared to the Day 20 PIPR replicate that did not have a large deletion, the 3 Day 20 PIPR replicates with the large deletions, when subsequently evolved to tobramycin, developed less tobramycin resistance (S3 Data and S12 Fig). This observation suggests that the large deletions were involved in reducing the subsequent rate of tobramycin resistance evolution given a prior history of piperacillin adaptation. A recent study isolated 4 pairs of clinical isolates of Paeruginosa, in which each pair consisted of a pyomelanogenic (PM) isolate and a “parental wild-type (WT)” nonpyomelanogenic isolate [64]. In each of the 4 pairs, the only genomic difference between the pyomelanogenic (denoted APM, BPM, CPM, and DPM) and its corresponding parental wild-type isolate (denoted AWT, BWT, CWT, and DWT) was the presence of large chromosomal deletions that overlap with parts of the deletions seen in Day 20 PIPR-1, -2, and -3 (Fig 8E; S5 Data). Indeed, all of the large deletions encompass hmgA, whose loss accounts for the pyomelanin phenotype. We used these 4 pairs of clinical isolates to test the hypothesis that the large deletions play a role in lowering the rate of tobramycin resistance evolution. We evolved the 4 pairs of isolates to tobramycin using the same daily serial passaging technique used throughout this study and tracked the MICs of tobramycin, piperacillin, and ciprofloxacin over the course of 15 days (Fig 8; S3 Data and S13 Fig). At the end of the 15 days, we saw that APM, BPM, and CPM had lower relative increases in MICTOB, compared to AWT (p < 0.01), BWT(p < 0.05), and CWT (p < 0.05), respectively (Fig 8A–8C). These data provide support for the idea that the large chromosomal deletions do indeed play a role in reducing the rate of tobramycin adaptation, and potentially even in limiting the maximum level of tobramycin resistance that can be developed comparatively. In the case of the fourth pair, we saw that DWTand DPM had comparable increases in MICTOB over the course of the tobramycin adaptation (Fig 8D, p = 1.00). It can be speculated that some combination of the presence or loss of specific genes in DPM led to this evolutionary trajectory that is different from the other 3 pyomelanogenic isolates. We would also like to point out that within each pair, the “WT” and “PM” isolates vary in initial Day 1 MICTOB. The BPM and BWT pair was the most disparate pair, as BPM had a much lower MICTOB than BWT (S13 Fig).


Fig 8. Clinical isolates with large chromosomal deletions have lower rates of tobramycin (TOB) resistance evolution. See image here: https://doi.org/10.1371/journal.pbio.2001586.g008

To see if large chromosomal deletions played a role in reducing the rate of tobramycin resistance evolution, 4 pairs of clinical isolates were evolved to tobramycin. Each pair consisted of a pyomelanogenic (PM) isolate with a large deletion, and its corresponding nonpyomelanogenic parental isolate that does not have a large deletion (“wild-type” [WT]) [64]. As anticipated, we observed that (A) APM, (B) BPM, and (C) CPM had lower relative increases in the minimum inhibitory concentration of tobramycin (MICTOB) compared to AWT, BWT, and CWT, respectively. However, (D) DWT and DPM had comparable relative increases in MICTOB. Asterisks denote p < 0.05 of a two-sample ttest after the raw MIC values were normalized by subtracting the average Day 1 MICTOBfor each evolved lineage. See S2 Text for the calculations of the statistical tests. Error bars show SEM of 3 replicates per treatment (except BPM-2, which had 2 replicates). (E) The large deletions of the 4 “PM” isolates are located in the same region as the deletions of Day 20 PIPR-1, -2, and -3, and all of the deletions encompass hmgA, whose loss causes the hyperproduction of pyomelanin. See S13 Fig for the original, prenormalized data. Interestingly, a recent study also observed large genomic deletions spanning hmgA when Paeruginosa PAO1 was evolved to meropenem, which is another beta-lactam antibiotic [65]. These mutants were also pyomelanogenic. The large deletions in both our study as well as this recent study also span mexX and mexY, which encode portions of the efflux pump that is a significant determinant of aminoglycoside resistance [67]. The loss of these genes in the 3 PIPRreplicates may partially explain why subsequent tobramycin adaptation is limited compared to the replicate that did not sustain the large deletion.


This study presents evidence of how the evolutionary history of bacterial adaptation to antibiotics can complicate strategies for treating infections and for limiting the further development of multidrug resistance. Exposing bacteria to fluctuating environments has been shown to be a potentially good strategy for slowing down the development of resistance [12,19,68]. More broadly, mechanisms of memory and history dependence in bacterial systems are being uncovered to better understand the dynamics of bacterial survival and adaptation in changing environments [69–71]. For example, a recent study showed that the survival of Caulobacter crescentus in response to a high concentration of sodium chloride stress depended on the duration and timing of an earlier treatment of a moderate concentration of sodium chloride and that this effect was linked to delays in cell division, which led to cell-cycle synchronization [72]. Another study described what they call response memory, which is when a gene regulatory network continues to persist after the removal of its external inducer. The study showed that in Ecolilac induction transiently continued when the environment was switched from lactose to glucose, which may be beneficial when the environment fluctuates over short timescales [73]. The results of those studies as well as the results from this study challenge the notion that bacteria respond solely to their present environment. Bacteria can encounter different stressors over time such as osmotic, oxidative, and acidic stress, and other studies have looked at how adaptation to 1 stressor protects the bacteria against other stressors [7,74]. Another example of bacteria adapting to changing environments is how Paeruginosa, which can be found in the natural environment in the soil and water, can readily adapt to a human host under the right conditions and consequently become pathogenic [75].

There are several factors involved in the emergence of antibiotic resistance that are clinically important that are not considered in this study. We have not taken into account any pathogen/host interactions, such as the role of the immune system. We also do not take into consideration the pharmacokinetics of the drug and the time-dependent fluctuation of drug concentration as experienced by the bacteria in a human-host environment. Furthermore, the dosages of clinical regimens are typically much higher than the wild-type MIC, and the evolutionary dynamics of the bacteria under these conditions may be different from those seen in our study, in which the drug pressure is slowly increased over time. We neglect to consider the role of horizontal gene transfer, which is a common mechanism of antibiotic resistance transfer, and focus rather on the role of de novo mutations acquired during adaptation. Because of the nature of the serial passaging method, we may be selecting for fast growers that may not necessarily have mutations that confer the most amount of resistance in terms of the MIC. We used a strong selection pressure in this study by propagating from the highest concentration of drug that showed growth, but it has been shown that weak antibiotic selection pressure can greatly affect the adaptive landscape [76,77]. Lastly, these bacteria were evolved to 1 antibiotic at a time, and we do not know how different mutant lineages would adapt if competed against each other. It would be interesting in the future to conduct competition experiments to measure the fitness of the different lineages with respect to each other.

While adaptive evolution of clinical isolates suggests that the drug-order–specific effects are clinically relevant, actual clinical studies must be performed to test the true clinical applicability of these effects. A major challenge that still needs to be addressed is how to translate the results of in vitro adaptive evolution experiments to effective therapies that can be used in a clinical setting [78]. For example, while we observed that piperacillin adaptation often led to the large chromosomal deletions and concomitant pyomelanin hyperproduction, the clinical isolates that had the large deletions (Fig 8) were not necessarily resistant to piperacillin. On the other hand, the other set of clinical isolates, which did have resistance to piperacillin, did not have the large deletions (Fig 7). Disparities between the phenotypic and genotypic adaptations such as this will need to be studied further in terms of strain-specific differences, actual history of antibiotic exposure, and other factors that are beyond the scope of this study.

Despite these caveats, there are several key factors of this study that provide confidence in the claims made. We saw consistency in the parallel replicates for the treatment lineages. An interesting exception is Day 40 PIPRTOBR-4, which had a higher final tobramycin resistance compared to Day 40 PIPRTOBR-1, -2 and -3, which we believe is attributed to the large genomic deletions seen in the first 3 replicates but not in the fourth replicate. We observed parallel evolution in which several genes were mutated independently across multiple lineages, and overall, there were less than 15 mutations per 20 days of evolution, which suggests positive selection. Furthermore, many of the mutated genes are also observed in clinical isolates of Paeruginosa, further giving credence to the clinical relevance of these mutations.

As mentioned previously, studies that have looked at alternating treatments of antibiotics have primarily looked at the effects of rapid switching, typically at daily or subdaily intervals. One of such recent studies evaluated how Ecoli responded to 136 different sequential treatments of subinhibitory concentrations of doxycycline and erythromycin, with each treatment consisting of 8 “seasons” of 12-hour-long adaptation periods to 1 of the drugs [19]. Using final optical density as an endpoint metric, the study found that 5 of the sequential treatments could clear the bacteria at the end of the eighth season. Interestingly, one of those 5 successful treatments consisted of 4 seasons of erythromycin, followed subsequently by 4 seasons of doxycycline. On the other hand, the treatment consisting of 4 seasons of doxycycline followed by 4 seasons of erythromycin did not manage to clear the bacteria at the end of 8 seasons. While the experimental setup is much different compared to that of this present study in terms of organism, antibiotics used, duration of treatment, and endpoint metric, these 2 treatments (4 seasons of erythromycin then 4 seasons of doxycycline and vice versa) are quite analogous to the types of treatments tested in our present study. The fact that these authors found a difference in the outcomes of this pair of opposite sequential treatments may suggest that drug-order–specific effects similar to those presented in our study may play a role in the evolutionary dynamics of their experiments.

Cycling between 2 drugs that exhibit collateral sensitivity to one another has been proposed and tested as a strategy to slow down the rate of resistance development [14]. Studies that have systematically tested for collateral sensitivities across a variety of antibiotics in Ecolihave consistently discovered that when Ecoli is adapted to drugs of the aminoglycoside class, it develops collateral sensitivity to several other drugs of different classes including beta-lactams, DNA synthesis inhibitors, and protein synthesis inhibitors [14,16,77]. In our present study, we tested 1 aminoglycoside (tobramycin), and we did not observe any collateral sensitivity arise to piperacillin or ciprofloxacin during adaptation to tobramycin. Instead, we saw collateral sensitivity to piperacillin and tobramycin arise as Paeruginosa was adapted to ciprofloxacin, which is a DNA synthesis inhibitor. While we only tested 1 drug in each of 3 drug classes, the dissimilarity of collateral sensitivity profiles between those studies and this present study may highlight how collateral sensitivity profiles may be organism-specific and drug-specific. Further supporting this idea, these prior studies also showed that while adaptation to drugs of the aminoglycoside class as a whole tended to lead to collateral sensitivity to other drug classes, not every aminoglycoside drug that was tested induced the same collateral sensitivity profiles.

While we did observe cases of collateral sensitivity, the main focus of our study was not to look at how resistance profiles to other drugs concurrently change during adaptation to 1 drug, but rather to see how adaptation to 1 drug influences the future evolutionary dynamics as the resistant population adapts to a new drug. Additionally, we wanted to see how adaptation to the second drug affected the resistance profile of the drug that the bacteria originally developed resistance to. Our sustained drug adaptation scheme can be thought of as being more akin to month-long antibiotic cycling at the level of the hospital ward or the environments that bacteria in persistent chronic infections are exposed to. The history-dependent evolutionary dynamics presented in this study highlight the complexity of bacterial adaptation to multidrug therapies, serve as a framework for forecasting evolutionary trajectories based on genetic and phenotypic signatures of past adaptation, and ultimately help elucidate our fundamental understandings of the evolutionary forces that drive resistance adaptation.

Asymmetrical evolutionary responses in changing environments have been studied in terms of collateral sensitivity/resistance [14,16], temperature [79], other abiotic stresses [7], and in cancer treatments [80]. Here, we present the concept of drug-history–specific effects in multidrug resistance adaptation, whereby the history of adaptation to 1 antibiotic environment can influence the evolutionary dynamics during subsequent adaptation to another antibiotic environment. These history-specific effects have direct clinical implications on optimizing antibiotic treatment strategies to slow and prevent the emergence of dangerous multidrug-resistant bacterial pathogens.

Materials and methods

Ethics statement

The set of Paeruginosa clinical isolates collected from the University of Virginia Health System (presented in Fig 7) were deidentified, did not require Institutional Review Board approval for their use, and were anonymized. The Hocquet Paeruginosa clinical isolates (which were originally collected by the authors of the Hocquet study [64]; presented in Fig 8) also did not require Institutional Review Board approval for their use and were anonymized.

Experimental study design

We evolved, in parallel, 4 independent replicates for each lineage in the primary adaptive evolution experiment and 3 independent replicates for each of the clinical isolates to balance the statistical power of the conclusions with the technical feasibilities of the daily serial propagations. In the primary adaptive evolution experiment, we concluded the 1-drug evolution at the end of 20 days because the resistance levels of the evolved lineages to their respective drugs were saturated or close to saturated at that point. The clinical isolates from Fig 7 and from Fig 8 were evolved for 10 and 15 days, respectively, because the similarities and differences of the drug-specific effects to those of the primary adaptive evolution experiment were readily apparent at that point.

Media, growth conditions, and antibiotics

MIC plates were made daily using the broth microdilution method with the standard 2-fold dilution series [81]. LB was used as the growth medium for all experiments (1% tryptone, 0.5% yeast extract, 1% NaCl). Antibiotics tested include piperacillin sodium, tobramycin, and ciprofloxacin HCl (Sigma). Aliquots of 1 mg/ml and 10 mg/ml antibiotic stocks were made by diluting the antibiotic powders in LB and were stored at −20°C. New frozen drug aliquots were used on a daily basis.

Adaptive laboratory evolution

A frozen stock of Paeruginosa PA14 was streaked on an LB agar plate, and a single colony was inoculated into 4 ml of LB, which was then grown overnight at 37°C, shaking at 125 RPM. This antibiotic-susceptible culture, denoted as the Day 0 Ancestor, was diluted to an OD600 of 0.001 (approximately 106 CFU/ml) and then inoculated into 3 identical MIC plates consisting of concentration gradients of piperacillin and tobramycin. A sample of the ancestor was saved in 25% glycerol and stored at −80°C. The 3 MIC plates were used to serially propagate cultures evolved to LB media, piperacillin, and tobramycin, with 4 biological replicates per condition. Wells for growth control (media + culture) and sterility control (media) were included in each MIC plate. For adaptation to LB media, bacteria were sampled from the growth control well. MIC plates were placed in a plastic container (to prevent evaporation) and incubated at 37°C with shaking at 125 RPM (Thermo Scientific MaxQ 4000). MIC plates were incubated daily for approximately 23 hours.

At the end of incubation, growth in the MIC plates was determined using a plate reader (Tecan Infinite M200 Pro). Growth was defined as OD600 > 0.1 after background subtraction. We recorded the MIC of each lineage for each drug, which was defined as the lowest antibiotic concentration tested that did not show growth (S1 and S3 Data). To propagate, cultures were passaged from the highest concentration that showed growth (i.e., MIC/2) from the corresponding MIC drug gradient. For adaptation to LB, cultures were passaged from the growth control well that contained only LB without any drug. For each culture to be passaged, the culture was first diluted by a factor of 1/250 in fresh LB (e.g., 20 μl of the culture was diluted in 5 ml of LB), which was then inoculated in fresh piperacillin and tobramycin drug gradients in the new day’s MIC plate. Wells of the MIC plate thus contained 100 μl of double the final concentration of the antibiotic and 100 μl of the diluted culture. Hence, the cultures were diluted by a total factor of 1/500 daily. Daily samples were saved in 25% glycerol and stored at −80°C. For Day 21, the piperacillin and tobramycin evolved cultures were subcultured in additional MIC plates such that they could subsequently be evolved to tobramycin and piperacillin, respectively.

A similar protocol was used to establish the ciprofloxacin-evolved lineages (CIPR). Starting with a clonal population of the Day 0 Ancestor, 4 replicates were established and propagated daily under ciprofloxacin treatment for 20 days. CIPR was then subpassaged to piperacillin and tobramycin to establish the CIPRPIPR and CIPRTOBR lineages in addition to continued ciprofloxacin evolution.

To establish the PIPRCIPR and TOBRCIPR lineages, bacteria from the frozen stocks of Day 20 PIPR and TOBR were revived on LB agar plates, and clonal populations were evolved to ciprofloxacin to establish these lineages. Similarly, to establish the PIPRLB, TOBRLB, and CIPRLB lineages, bacteria from the frozen stocks of Day 20 PIPR, TOBR, and CIPR were revived on LB agar plates, and clonal populations were evolved to LB.

Lastly, the MIC to ciprofloxacin was retrospectively measured for the Control, PIPR, TOBR, PIPRTOBR, and TOBRPIPR lineages. Frozen stocks were revived and plated on LB agar plates. The notation for the day numbering is such that Day X PIPR means X days exposure to piperacillin. For consistency, stocks were revived from Days 0 (Ancestor), 5, 10, 15, 19, 20, 25, 30, 35, and 39 for Control, PIPR, and TOBR. One day of exposure to ciprofloxacin would yield Days 1, 6, 11, 16, 20, 21, 26, 31, 36, and 40 MICs to ciprofloxacin. For PIPRTOBR and TOBRPIPR, stocks were similarly revived from Days 20, 25, 30, 35, and 39 and exposed to ciprofloxacin to measure Days 21, 26, 31, 36, and 40 MICs to ciprofloxacin. S3 Data shows the MICs to piperacillin, tobramycin, and ciprofloxacin, respectively, for all the lineages. Note that not all drug MICs were measured on a daily basis for all lineages.

During analysis of the mutations, we deduced that there were some cross-contaminations between replicates in a few lineages. Namely, we saw sets of mutations that were identical in 2 replicates. We believed that the most likely explanation was that the following 7 lines were cross-contaminated sometime between Day 21 and Day 40: CIPRPIPR-3, CIPRPIPR-4, TOBR-1 CIPRTOBR-1, CIPRTOBR-2, CIPRTOBR-4, and CIPR-3, where the number denotes the replicate. To redo these lineages, the corresponding Day 20 replicate frozen stocks were revived on LB agar plates. Then clonal populations were used to redo the propagation as described before. For example, CIPR-3 was evolved to piperacillin for 20 days to redo CIPRPIPR-3. We performed Sanger sequencing of replicate-specific mutations (S3 Table) on the Day 40 mutants to confirm successful propagation of the cultures.

Whole-genome sequencing

Frozen samples of Day 0 Ancestor, Day 20 Control, PIPR, TOBR, CIPR, Day 40 Control, PIPR, TOBR, CIPR, PIPRTOBR, PIPRCIPR, TOBRPIPR, TOBRCIPR, CIPRPIPR, and CIPRTOBR were streaked on LB agar plates and incubated at 37°C. Agar plates were submitted to Genewiz Incorporation for sequencing service. A single colony from each plate was chosen for DNA extraction, library preparation, multiplexing, and sequencing using 101-bp paired-end reads with the Illumina HiSeq 2500 platform. Reads were aligned to the reference PaeruginosaPA14 genome (NC_008463.1) with coverage ranging from 113X to 759X. This large range is due to the fact that we submitted samples for sequencing in 3 batches and had different numbers of samples for each batch but had relatively the same number of reads per batch. Nevertheless, the coverage was more than sufficient to identify the SNPs, insertions, and deletions in the genomes. The sequencing reads for Day 0 Ancestor and the 56 drug-evolved lineages are available via the NCBI SRA database (www.ncbi.nlm.nih.gov/sra), accession number SRP100674, BioProject number PRJNA376615.

Reads were aligned and mutations were called using the breseq pipeline [82] using default settings. All reported mutations were visually inspected by viewing the read alignments in IGV and the breseq output files, and mutations with less than 80% frequencies were not counted. The full list of mutations is presented in S1 and S2 Tables. The circos software package [83] was used to plot the mutations by genomic position for Fig 4 and the positions of the large chromosomal deletions in Fig 8.

We confirmed some of the mutations using Sanger sequencing. For each of the Day 20 PIPR, TOBR, and CIPR replicates, we chose 1 mutation each to confirm (S3 Table). We also used these to confirm that replicates were not contaminated before submitting them for whole-genome sequencing. These mutations were also confirmed in each of the Day 40 lineages that were derived from the Day 20 PIPR, TOBR, and CIPR replicates.

Reproducing drug-history dependence in the pyomelanin phenotype during piperacillin evolution

Clonal populations of Day 0 Ancestor, Day 20 TOBR-1, -2, -3 and -4, and Day 20 CIPR-1, -2, -3, and -4 were grown in LB starting from the frozen samples. These cultures were diluted in LB to OD600 of 0.001. On Day 1, in 96-well plates, 100 μl of the diluted cultures were inoculated with 100 μl of 4 μg/ml piperacillin (to yield a final concentration of 2 μg/ml piperacillin). Ninety-two wells were used to establish independent replicate populations exposed to piperacillin. Cultures were incubated at 37°C with shaking at 125 RPM. On Day 2, replicate populations were passaged using a 96-pin replicator tool (V&P Scientific, VP246, 100–150 μl per pin) into 200 μl of 4 μg/ml piperacillin. This protocol was continued until Day 10 with a final concentration of 20 μg/ml piperacillin. For each plate, 2 wells were used as sterility controls (only LB), and 2 wells were used as growth controls (LB with bacteria, without drug). Photographs were taken daily (S8–S10 Figs), and the number of visibly brown wells was recorded.

Testing for drug-order–specific evolutionary dynamics in clinical isolates

Three clinical isolates of Paeruginosa with high piperacillin resistance and low tobramycin and ciprofloxacin resistance were obtained from the University of Virginia Health System and were evolved to the 3 drugs in the same manner as the main adaptive evolution experiment starting from frozen samples. These isolates were first confirmed to actually be Paeruginosa with PCR by using primers that specifically amplify the 16S rRNA region of Paeruginosa [84]. Three replicates of each isolate were evolved to each of the 3 drugs for 10 days, and their MICs to the 3 drugs were measured as before. In separate subsequent experiments, the 3 clinical isolates were evolved to LB with 3 replicates each. The MICPIP was measured for 10 days (S3 Data). This measurement was done by inoculating into piperacillin concentration gradients to measure the MICPIP but sampling and passaging from the “growth control” well (LB with bacteria, without drug) to adapt to LB.

The 4 pairs of clinical isolates of Paeruginosa from the Hocquet study [64] were evolved to tobramycin for 15 days with 3 parallel replicates each, with the exception of BPM, which had 2 replicates due to cross-contamination in the third replicate. The MICs for piperacillin and ciprofloxacin were also measured every 5 days (S3 Data). At the end of the 15 days of evolution, primers amplifying part of the hmgA gene were used to check for the presence of the gene in the “WT” isolates and the absence of the gene in the “PM” isolates (S3 Table).

Statistical significance of drug-order–specific effects in MIC profiles

All statistical comparisons of MIC values were performed on the log2 transformed values. Unless noted otherwise, one-way ANOVAs were performed on the MICs of the relevant lineages. If the p-value from the ANOVA was less than 0.05, a post-hoc Tukey’s honest significant difference (HSD) multiple comparisons test was then performed to determine which pairs of treatments were significantly different from each other. The Tukey’s HSD tests report 95% confidence intervals for the true mean difference for each pairwise comparison. If the confidence interval does not contain 0, then the 2 groups being compared have significantly different means at the p = 0.05 level. To also assess the comparisons using nonparametric statistic tests, Kruskwal-Wallis tests followed by post-hoc Dunn’s multiple comparisons tests were also performed. All of the Kruskal-Wallis tests yielded comparable results to the one-way ANOVA at the alpha = 0.05 significance level, and the conclusions are the same for the key comparisons that drive the results highlighted in the manuscript. For a complete set of calculations, see S2 Text.

For the comparisons presented in Fig 3, treatments being compared consist of those listed on the x-axis of each graph in the figure. For the comparisons presented in Fig 7, the raw MIC values for each lineage were first normalized by subtracting the average Day 1 MIC of each of their respective lineages. For each of the 3 clinical isolates, a one-way ANOVA and a Kruskal-Wallis test were performed on the Day 10 MICPIP values of the lineages evolved to LB, tobramycin, and ciprofloxacin (piperacillin-adapted lineages were excluded in the comparisons). The Tukey’s HSD test and Dunn’s test were then performed to see if the Day 10 MICPIP values of the lineages evolved to tobramycin and ciprofloxacin were significantly different from the lineages evolved to LB. For the comparisons presented in Fig 8, the raw MIC values for each lineage were first normalized by subtracting the average Day 1 MIC of each of their respective lineages. A two-sample t test and a Wilcoxon rank sum test were performed for the Day 15 MICTOB values of the “WT” and “PM” lineages evolved to tobramycin in each of the 4 pairs of isolates. Calculations were done in MATLAB R2016b, using the functions “anova1” for one-way ANOVA, “multcompare” for Tukey’s HSD test, “ttest2” for two-sample t test, and “ranksum” for the Wilcoxon rank sum test. The Kruskal-Wallis test was done with the “kruskal.test” command in R, and the Dunn’s test was done with the “posthoc.kruskal.dunn.test” command with the PMCMR R package [85].


  1. Andersson DI, Hughes D. Antibiotic resistance and its cost: is it possible to reverse resistance? Nat Rev Microbiol. Nature Publishing Group; 2010;8: 260–271. pmid:20208551
  2. Ling LL, Schneider T, Peoples AJ, Spoering AL, Engels I, Conlon BP, et al. A new antibiotic kills pathogens without detectable resistance. Nature. 2015;517: 455–459. pmid:25561178
  3. Cooper MA, Shlaes D. Fix the antibiotics pipeline. Nature. 2011;472: 32. pmid:21475175
  4. Conrad TM, Lewis NE, Palsson BØ. Microbial laboratory evolution in the era of genome-scale science. Mol Syst Biol. 2011;7: 509. pmid:21734648
  5. Adams JP, Rosenzweig F. Experimental Microbial Evolution: History and Conceptual underpinnings. Genomics. Elsevier Inc.; 2014;104: 393–398. pmid:25315137
  6. Elena SF, Lenski RE. Evolution experiments with microorganisms: the dynamics and genetic bases of adaptation. Nat Rev Genet. 2003;4: 457–69. pmid:12776215
  7. Dragosits M, Mozhayskiy V, Quinones-Soto S, Park J, Tagkopoulos I. Evolutionary potential, cross-stress behavior and the genetic basis of acquired stress resistance in Escherichia coli. Mol Syst Biol. Nature Publishing Group; 2013;9: 643. pmid:23385483
  8. Long A, Liti G, Luptak A, Tenaillon O. Elucidating the molecular architecture of adaptation via evolve and resequence experiments. Nat Rev Genet. Nature Research; 2015;16: 567–582. pmid:26347030
  9. Lang GI, Desai MM. The spectrum of adaptive mutations in experimental evolution. Genomics. Elsevier B.V.; 2014; pmid:25269377
  10. Palmer AC, Kishony R. Understanding, predicting and manipulating the genotypic evolution of antibiotic resistance. Nat Rev Genet. Nature Publishing Group; 2013;14: 243–248. pmid:23419278
  11. Toprak E, Veres A, Michel J-B, Chait R, Hartl DL, Kishony R. Evolutionary paths to antibiotic resistance under dynamically sustained drug selection. Nat Genet. Nature Publishing Group; 2012;44: 101–5. pmid:22179135
  12. Kim S, Lieberman TD, Kishony R. Alternating antibiotic treatments constrain evolutionary paths to multidrug resistance. Proc Natl Acad Sci. 2014;111: 14494–14499. pmid:25246554
  13. Jansen G, Barbosa C, Schulenburg H. Experimental evolution as an efficient tool to dissect adaptive paths to antibiotic resistance. Drug Resist Updat. Elsevier Ltd; 2013;16: 96–107. pmid:24594007
  14. Imamovic L, Sommer MOA. Use of Collateral Sensitivity Networks to Design Drug Cycling Protocols That Avoid Resistance Development. Sci Transl Med. 2013;5: 204ra132–204ra132. pmid:24068739
  15. Baym M, Stone LK, Kishony R. Multidrug evolutionary strategies to reverse antibiotic resistance. Science (80-). 2016;351: aad3292–aad3292. pmid:26722002
  16. Lázár V, Pal Singh G, Spohn R, Nagy I, Horváth B, Hrtyan M, et al. Bacterial evolution of antibiotic hypersensitivity. Mol Syst Biol. 2013;9: 700. pmid:24169403
  17. Szybalski W, Bryson V. Genetic studies on microbial cross resistance to toxic agents. I. Cross resistance of Escherichia coli to fifteen antibiotics. J Bacteriol. 1952;64: 489–499. pmid:12999676
  18. Munck C, Gumpert HK, Wallin AI, Wang HH, Sommer MOA. Prediction of resistance development against drug combinations by collateral responses to component drugs. Sci Transl Med. 2014;6: 262ra156. pmid:25391482
  19. Fuentes-Hernandez A, Plucain J, Gori F, Pena-Miller R, Reding C, Jansen G, et al. Using a Sequential Regimen to Eliminate Bacteria at Sublethal Antibiotic Dosages. PLoS Biol. 2015;13:4: e1002104. pmid:25853342
  20. Rodriguez de Evgrafov M, Gumpert H, Munck C, Thomsen TT, Sommer MOA. Collateral Resistance and Sensitivity Modulate Evolution of High-Level Resistance to Drug Combination Treatment in Staphylococcus aureus. Mol Biol Evol. 2015;32: 1175–1185. pmid:25618457
  21. Pál C, Papp B, Lázár V. Collateral sensitivity of antibiotic-resistant microbes. Trends Microbiol. 2015;23: 401–407. pmid:25818802
  22. Lázár V, Nagy I, Spohn R, Csörgő B, Györkei Á, Nyerges Á, et al. Genome-wide analysis captures the determinants of the antibiotic cross-resistance interaction network. Nat Commun. 2014;5: 4352. pmid:25000950
  23. Roemhild R, Barbosa C, Beardmore RE, Jansen G, Schulenburg H. Temporal variation in antibiotic environments slows down resistance evolution in pathogenic Pseudomonas aeruginosa. Evol Appl. 2015;8: 945–955. pmid:26640520
  24. Perron GG, Gonzalez A, Buckling A. Source–sink dynamics shape the evolution of antibiotic resistance and its pleiotropic fitness cost. Proc R Soc B. 2007;274: 2351–2356. pmid:17650474
  25. Sarraf-Yazdi S, Sharpe M, Bennett KM, Dotson TL, Anderson DJ, Vaslef SN. A 9-year retrospective review of antibiotic cycling in a surgical intensive care unit. J Surg Res. Elsevier Ltd; 2012;176: e73–e78. pmid:22445457
  26. Abel zur Wiesch P, Kouyos R, Abel S, Viechtbauer W, Bonhoeffer S. Cycling Empirical Antibiotic Therapy in Hospitals: Meta-Analysis and Models. Wilke CO, editor. PLoS Pathog. 2014;10:6: e1004225. pmid:24968123
  27. Brown EM, Nathwani D. Antibiotic cycling or rotation: A systematic review of the evidence of efficacy. J Antimicrob Chemother. 2005;55: 6–9. pmid:15531594
  28. Gifford DR, Maclean RC. Evolutionary Reversals of Antibiotic Resistance In Experimental Populations Of Pseudomonas aeruginosa. Evolution (N Y). 2013;67: 2973–2981. pmid:24094347
  29. Lister PD, Wolter DJ, Hanson ND. Antibacterial-Resistant Pseudomonas aeruginosa: Clinical Impact and Complex Regulation of Chromosomally Encoded Resistance Mechanisms. Clin Microbiol Rev. 2009;22: 582–610. pmid:19822890
  30. Lodise TP, Lomaestro B, Drusano GL. Piperacillin-tazobactam for Pseudomonas aeruginosa infection: Clinical Implications of an Extended-Infusion Dosing Strategy. Clin Infect Dis. 2007;44: 357–363. pmid:17205441
  31. Shawar RM, MacLeod DL, Garber RL, Burns JL, Stapp JR, Clausen CR, et al. Activities of Tobramycin and Six Other Antibiotics against Pseudomonas aeruginosa Isolates from Patients with Cystic Fibrosis. Antimicrob Agents Chemother. 1999;43: 2877–2880. pmid:10582875
  32. Chaudhry NA, Flynn HW, Murray TG, Tabandeh H, Mello MO, Miller D. Emerging Ciprofloxacin-Resistant Pseudomonas aeruginosa. Am J Ophthalmol. 1999;128: 509–510. pmid:10577596
  33. Chatterjee M, Anju CP, Biswas L, Anil Kumar V, Gopi Mohan C, Biswas R. Antibiotic resistance in Pseudomonas aeruginosa and alternative therapeutic options. Int J Med Microbiol. Elsevier GmbH.; 2016;306: 48–58.
  34. Feng Y, Jonker MJ, Moustakas I, Brul S, ter Kuile BH. Dynamics of Mutations during Development of Resistance by Pseudomonas aeruginosa against Five A ntibiotics. Antimicrob Agents Chemother. 2016;60: 4229–4236. pmid:27139485
  35. Terzi HA, Kulah C, Ciftci IH. The effects of active efflux pumps on antibiotic resistance in Pseudomonas aeruginosa. World J Microbiol Biotechnol. 2014;30: 2681–2687. pmid:24964907
  36. Hosokawa K, Park NH, Inaoka T, Itoh Y, Ochi K. Streptomycin-resistant (rpsL) or rifampicin-resistant (rpoB) mutation in Pseudomonas putida KH146-2 confers enhanced tolerance to organic chemicals. Environ Microbiol. 2002;4: 703–712. pmid:12460278
  37. Suzuki S, Horinouchi T, Furusawa C. Prediction of antibiotic resistance by gene expression profiles. Nat Commun. Nature Publishing Group; 2014;5: 5792. pmid:25517437
  38. Schurek KN, Marr AK, Taylor PK, Wiegand I, Semenec L, Khaira BK, et al. Novel Genetic Determinants of Low-Level Aminoglycoside Resistance in Pseudomonas aeruginosa. Antimicrob Agents Chemother. 2008;52: 4213–9. pmid:18824604
  39. Feliziani S, Marvig RL, Luján AM, Moyano AJ, Di Rienzo JA, Krogh Johansen H, et al. Coexistence and Within-Host Evolution of Diversified Lineages of Hypermutable Pseudomonas aeruginosa in Long-term Cystic Fibrosis Infections. PLoS Genet. 2014;10:10: e1004651. pmid:25330091
  40. Markussen T, Marvig RL, Gómez-Lozano M, Aanæs K, Burleigh AE, Høiby N, et al. Environmental Heterogeneity Drives Within-Host Diversification and Evolution of Pseudomonas aeruginosa. MBio. 2014;5: 1–10.
  41. Chung JCS, Becq J, Fraser L, Schulz-Trieglaff O, Bond NJ, Foweraker J, et al. Genomic Variation among Contemporary Pseudomonas aeruginosa Isolates from Chronically Infected Cystic Fibrosis Patients. J Bacteriol. 2012;194: 4857–4866. pmid:22753054
  42. Marvig RL, Sommer LM, Molin S, Johansen HK. Convergent evolution and adaptation of Pseudomonas aeruginosa within patients with cystic fibrosis. Nat Genet. 2015;47: 57–64. pmid:25401299
  43. Winsor GL, Lam DKW, Fleming L, Lo R, Whiteside MD, Yu NY, et al. Pseudomonas Genome Database: improved comparative analysis and population genomics capability for Pseudomonas genomes. Nucleic Acids Res. 2011;39: D596–600. pmid:20929876
  44. Aeschlimann JR. The Role of Multidrug Efflux Pumps in the Antibiotic Resistance of Pseudomonas aeruginosa and Other Gram-Negative Bacteria. Pharmacotherapy. 2003;23: 916–924.
  45. Morita Y, Cao L, Gould VC, Avison MB, Poole K. nalD encodes a second repressor of the mexAB-oprM multidrug efflux operon of Pseudomonas aeruginosa. J Bacteriol. 2006;188: 8649–8654. pmid:17028276
  46. Karlowsky JA, Saunders MH, Harding GAJ, Hoban DJ, Zhanel GG. In Vitro Characterization of Aminoglycoside Adaptive Resistance in Pseudomonas aeruginosa. Antimicrob Agents Chemother. 1996;40: 1387–1393. pmid:8726006
  47. Sobel ML, Neshat S, Poole K. Mutations in PA2491 (mexS) Promote MexT-Dependent mexEF-oprN Expression and Multidrug Resistance in a Clinical Strain of Pseudomonas aeruginosa. J Bacteriol. 2005;187: 1246–1253. pmid:15687188
  48. Maseda H, Sawada I, Saito K, Uchiyama H, Nakae T. Enhancement of the mexAB—oprM Efflux Pump Expression by a Quorum-Sensing Autoinducer and Its Cancellation by a Regulator, MexT, of the mexEF-oprN Efflux Pump Operon in Pseudomonas aeruginosa. Antimicrob Agents Chemother. 2004;48: 1320–1328. pmid:15047536
  49. Fernández L, Breidenstein EBM, Hancock REW. Creeping baselines and adaptive resistance to antibiotics. Drug Resist Updat. 2011;14: 1–21. pmid:21288762
  50. Skiada A, Markogiannakis A, Plachouras D, Daikos GL. Adaptive resistance to cationic compounds in Pseudomonas aeruginosa. Int J Antimicrob Agents. Elsevier B.V.; 2011;37: 187–193. pmid:21295448
  51. El’Garch F, Jeannot K, Hocquet D, Llanes-Barakat C, Plésiat P. Cumulative Effects of Several Nonenzymatic Mechanisms on the Resistance of Pseudomonas aeruginosa to aminoglycosides. Antimicrob Agents Chemother. 2007;51: 1016–21. pmid:17194835
  52. Hocquet D, Vogne C, Garch F El, Gotoh N, Lee A, Lomovskaya O, et al. MexXY-OprM Efflux Pump Is Necessary for Adaptive Resistance of Pseudomonas aeruginosa to Aminoglycosides. Antimicrob Agents Chemother. 2003;47: 1371–1375. pmid:12654672
  53. Adam M, Murali B, Glenn NO, Potter SS. Epigenetic inheritance based evolution of antibiotic resistance in bacteria. BMC Evol Biol. 2008;8. pmid:18282299
  54. Motta SS, Cluzel P, Aldana M. Adaptive Resistance in Bacteria Requires Epigenetic Inheritance, Genetic Noise, and Cost of Efflux Pumps. PLoS One. 2015;10:3: e0118464. pmid:25781931
  55. Purssell A, Poole K. Functional characterization of the NfxB repressor of the mexCD-oprJ multidrug efflux operon of Pseudomonas aeruginosa. Microbiology. 2013;159: 2058–2073. pmid:23924707
  56. Poole K, Gotoh N, Tsujimoto H, Zhao Q, Wada A, Yamasaki T, et al. Overexpression of the mexC-mexD-oprJ efflux operon in nfxB-type multidrug-resistant strains of Pseudomonas aeruginosa. Mol Microbiol. 1996;21: 713–724. pmid:8878035
  57. Masuda N, Sakagawa E, Ohya S, Gotoh N, Nishino T. Hypersusceptibility of the Pseudomonas aeruginosa nfxB Mutant to beta-Lactams Due to Reduced Expression of the AmpC beta-Lactamase. Antimicrob Agents Chemother. 2001;45: 1284–6. pmid:11257048
  58. Masuda N, Gotoh N, Ohya S, Nishino T. Quantitative Correlation between Susceptibility and OprJ Production in NfxB Mutants of Pseudomonas aeruginosa. Antimicrob Agents Chemother. 1996;40: 909–913. pmid:8849250
  59. Plucain J, Suau A, Cruveiller S, Médigue C, Schneider D, Le Gac M. Contrasting effects of historical contingency on phenotypic and genomic trajectories during a two-step evolution experiment with bacteria. BMC Evol Biol.; 2016;16: 1–11.
  60. Ernst RK, D’Argenio DA, Ichikawa JK, Bangera MG, Selgrade S, Burns JL, et al. Genome mosaicism is conserved but not unique in Pseudomonas aeruginosa isolates from the airways of young children with cystic fibrosis. Environ Microbiol. 2003;5: 1341–1349. pmid:14641578
  61. Moran NA. Microbial Minimalism: Genome Reduction in Bacterial Pathogens. Cell. 2002;108: 583–586.
  62. Rau MH, Marvig RL, Ehrlich GD, Molin S, Jelsbak L. Deletion and acquisition of genomic content during early stage adaptation of Pseudomonas aeruginosa to a human host environment. Environ Microbiol. 2012;14: 2200–2211. pmid:22672046
  63. Lee MC, Marx CJ. Repeated, Selection-Driven Genome Reduction of Accessory Genes in Experimental Populations. PLoS Genet. 2012;8:5: e1002651. pmid:22589730
  64. Hocquet D, Petitjean M, Rohmer L, Valot B, Kulasekara HD, Bedel E, et al. Pyomelanin-producing Pseudomonas aeruginosa selected during chronic infections have a large chromosomal deletion which confers resistance to pyocins. Environ Microbiol. 2016;18: 3482–3493. pmid:27119970
  65. Cabot G, Zamorano L, Moyà B, Juan C, Navas A, Blázquez J, et al. Evolution of Pseudomonas aeruginosa Antimicrobial Resistance and Fitness under Low and High Mutation Rates. Antimicrob Agents Chemother. 2016;60: 1767–1778. pmid:26729493
  66. Rodríguez-Rojas A, Mena A, Martín S, Borrell N, Oliver A, Blázquez J. Inactivation of the hmgA gene of Pseudomonas aeruginosa leads to pyomelanin hyperproduction, stress resistance and increased persistence in chronic lung infection. Microbiology. 2009;155: 1050–1057. pmid:19332807
  67. Morita Y, Tomida J, Kawamura Y. MexXY multidrug efflux system of Pseudomonas aeruginosa. Front Microbiol. 2012;3: 1–13.
  68. Nichol D, Jeavons P, Fletcher AG, Bonomo RA, Maini PK, Paul JL, et al. Steering Evolution with Sequential Therapy to Prevent the Emergence of Bacterial Antibiotic Resistance. PLoS Comput Biol. 2015;11:9: e1004493. pmid:26360300
  69. Casadesús J, D’Ari R. Memory in bacteria and phage. BioEssays. 2002;24: 512–518. pmid:12111734
  70. Lambert G, Kussell E. Quantifying Selective Pressures Driving Bacterial Evolution Using Lineage Analysis. Phys Rev X. 2015;11016: 1–10. pmid:26213639
  71. Wolf DM, Fontaine-Bodin L, Bischofs I, Price G, Keasling J, Arkin AP. Memory in Microbes: Quantifying History-Dependent Behavior in a Bacterium. PLoS One. 2008;3:2: e1700. pmid:18324309
  72. Mathis R, Ackermann M. Response of single bacterial cells to stress gives rise to complex history-dependence at the population level. Proc Natl Acad Sci. 2016;113: 4224–4229. pmid:26960998
  73. Lambert G, Kussel E. Memory and Fitness Optimization of Bacteria under Fluctuating Environments. PLoS Genet. 2014;10:9: e1004556. pmid:25255314
  74. Lagator M, Colegrave N, Neve P. Selection history and epistatic interactions impact dynamics of adaptation to novel environmental stresses. Proc R Soc London B Biol Sci. 2014;281: 20141679. pmid:25232137
  75. Lyczak JB, Cannon CL, Pier GB. Establishment of Pseudomonas aeruginosa infection: Lessons from a versatile opportunist. Microbes Infect. 2000;2: 1051–1060.
  76. Mira PM, Meza JC, Nandipati A, Barlow M. Adaptive Landscapes of Resistance Genes Change as Antibiotic Concentrations Change. Mol Biol Evol. 2015;32: 2707–2715.
  77. Oz T, Guvenek A, Yildiz S, Karaboga E, Tamer YT, Mumcuyan N, et al. Strength of Selection Pressure Is an Important Parameter Contributing to the Complexity of Antibiotic Resistance Evolution. Mol Biol Evol. 2014;31: 2387–2401.
  78. Kupferschmidt K. Resistance fighters. Science (80-). 2016;352: 758–61. pmid:27174968
  79. Mongold JA, Bennett AF, Lenski RE. Evolutionary adaptation to temperature. IV. Adaptation of Escherichia coli at a niche boundary. Evolution (N Y). 1996;50: 493.
  80. Lee MJ, Ye AS, Gardino AK, Heijink AM, Sorger PK, MacBeath G, et al. Sequential Application of Anticancer Drugs Enhances Cell Death by Rewiring Apoptotic Signaling Networks. Cell. Elsevier Inc.; 2012;149: 780–794. pmid:22579283
  81. Andrews JM. Determination of minimum inhibitory concentrations. J Antimicrob Chemother. 2001;48 Suppl 1: 5–16.
  82. Deatherage DE, Traverse CC, Wolf LN, Barrick JE. Detecting rare structural variation in evolving microbial populations from new sequence junctions using breseq. Front Genet. 2015;5: 1–16. pmid:25653667
  83. Krzywinski M, Schein J, Birol I, Connors J, Gascoyne R, Horsman D, et al. Circos: an Information Aesthetic for Comparative Genomics. Genome Res. 2009;19: 1639–1645. pmid:19541911
  84. Spilker T, Coenye T, Vandamme P, LiPuma JJ. PCR-Based Assay for Differentiation of Pseudomonas aeruginosa from Other Pseudomonas Species Recovered from Cystic Fibrosis Patients. J Clin Microbiol. 2004;42: 2074–2079. pmid:15131172
  85. Pohlert T. The Pairwise Multiple Comparison of Mean Ranks Package (PMCMR) [Internet]. 2014. http://cran.r-project.org/package=PMCMR

Leave a Reply

History of Antibiotic Adaptation Influences Microbial Evolutionary Dynamics During Subsequent Treatment

Antibiotic resistance is a growing healthcare concern whereby bacterial infections are increasingly difficult to eradicate due to their ability to survive antibiotic treatments [1]. There have been reported cases of resistance for nearly every antibiotic we have available [2]. Coupled with the fact that the antibiotic discovery pipeline has slowed over the past few decades [3], there is a dire need to find better treatment strategies using existing antibiotics that can slow or even reverse the development of resistance.

Scroll to top