This is an uncorrected proof.
Figures
Abstract
Antibiotic resistance in bacteria frequently evolves due to mutations in drug-target or detoxifying genes. Pre-existing polymorphisms in these genes are likely to influence the evolvability of drug resistance. However, the role of polymorphisms in driving resistance evolution and the underlying molecular mechanisms are poorly understood. Here, we demonstrate that polymorphisms in a signalling pathway impact the evolvability of trimethoprim (TMP) resistance in Escherichia coli. When challenged with TMP, de-repression of the PhoQ-PhoP two-component signalling system in E. coli transcriptionally upregulates the drug target Dihydrofolate Reductase (DHFR), leading to drug resistance. We identified and characterised naturally occurring polymorphisms in PhoQ and DHFR that modulate intrinsic antibiotic susceptibility. These variants also altered the ability of E. coli to evolve de novo TMP resistance as a result of epistasis with adaptive mutations. Interestingly, a strain harbouring a less-evolvable DHFR variant acquired a novel mutation in PhoQ under TMP pressure that hyperactivated the signalling pathway, conferring high-level resistance but at a large fitness cost. This mutation was not observed in wild type but reached fixation rapidly in the background of the DHFR variant. Using RNA-sequencing we compare how natural variants and adaptive mutations in PhoQ affect the expression of PhoP-target genes and downstream regulatory pathways in E. coli. Finally, we uncouple the roles of resistance level by DHFR overproduction and fitness cost by activation of the RpoS regulon to explain why E. coli more frequently evolves to de-repress PhoQ than hyperactivate it under drug pressure. Our study, thus, demonstrates that pre-existing polymorphisms alter both, evolvability and mutation landscapes during antibiotic adaptation. This work establishes the PhoQ-PhoP-DHFR pathway as an experimental paradigm to understand the evolution of signalling pathways under environmental selection.
Author summary
Genetic variation is expected to enhance the evolvability of antibiotic resistance in bacteria. However, the impact of pre-existing genetic polymorphisms on evolvability and underlying mechanisms is poorly understood. In this study, we explore the impact of polymorphisms in the PhoQ-PhoP-DHFR pathway on the evolution of trimethoprim resistance in Escherichia coli. We show that polymorphisms in genes of this pathway are epistatic with adaptive mutations. As a result, pre-existing variants alter the outcomes of antibiotic challenge in terms of both resistance evolvability and the landscape of adaptive mutations. We then decipher how naturally occurring and drug-selected mutations in PhoQ affect the global transcriptome of E. coli and result in differing extents of pleiotropy. Finally, we uncouple the contribution of fitness cost and benefit of activating the PhoQ-PhoP signalling system during evolution and define drug concentration regimes in which different activity levels are likely to be favoured.
Citation: Yelpure C, Chinnaraj S, Topiwala K, Phadke J, Matange N (2026) Polymorphisms in a bacterial signalling pathway alter evolutionary routes of antibiotic resistance in Escherichia coli. PLoS Biol 24(8): e3003977. https://doi.org/10.1371/journal.pbio.3003977
Academic Editor: J. Arjan G. M. de Visser, Wageningen University, KINGDOM OF THE NETHERLANDS
Received: May 2, 2026; Accepted: August 12, 2026; Published: August 26, 2026
Copyright: © 2026 Yelpure et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The authors confirm that all data underlying the findings are fully available without restriction. Raw data from the study are provided in the Supplementary files S1, S2, S3, S4 and S5. Genome sequencing and RNA sequencing data have been uploaded to GenBank (PRJNA1398517 and PRJNA1399355).
Funding: This research was funded by the DBT/Wellcome Trust India Alliance (IA/I/20/2/505181) and ANRF-Advanced Research Grant, Govt. of India (ANRF/ARG/2025/000042/LS), both awarded to NM. CY is a recipient of a Senior Research Fellowship from the Department of Biotechnology, Government of India (DBT/2021-22/IISER-P/1541). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. URL: https://www.indiaalliance.org/ and https://anrfonline.in/ANRF/HomePage
Competing interests: The authors have declared that no competing interests exist.
Abbreviations: AMR, antimicrobial resistance; CFU, colony-forming units; DHFR, Dihydrofolate Reductase; FDR, false discovery rate; TMP, trimethoprim.
Introduction
Antimicrobial resistance (AMR) is the ability of bacteria to overcome growth inhibition by antibiotics. An increasing burden of AMR threatens to destabilise decades of medical advance and understanding its evolution may hold the key to slowing it down [1,2]. In addition to its biomedical relevance, antibiotic resistance is also an important model for understanding molecular, genetic, and evolutionary mechanisms of bacterial adaptation. Moreover, it has been used to study evolvability itself [3]. Evolvability was originally defined as the disposition of a biological system to generate and maintain variation. More recently, it is equated with adaptability, i.e., degree to which a biological system can undergo adaptive evolution [3–10]. The latter definition is particularly relevant for antibiotic resistance as predicting the potential for resistance evolution is key to mitigating its adverse effects.
Three processes dictate evolvability [3,8]. First, the generation of genetic variation, second, genotype-to-phenotype mapping of variants; and third, selection by the environment. In the case of adaptation to antibiotics, changes in evolvability have been traced to all three steps. Hypermutator strains that generate greater genetic variation emerge spontaneously during antibiotic challenge and have an enhanced capacity for adaptation to antibiotics [11–13]. Conversely, inhibiting transcription-coupled mutagenesis can slow down the evolution of drug resistance [14,15]. Likewise, epistasis among resistance-conferring mutations alters genotype-phenotype mapping and fitness, directly influencing resistance evolution [16,17]. Yet, empirical and experimental evidence linking these three processes and investigating how they affect one another is scarce. For instance, while it is intuitive that greater pre-existing genetic variation would tend to increase evolvability, how pre-existing polymorphisms alter the fitness landscape of adaptive mutations and, consequently, outcomes of antibiotic challenge is less well worked out.
A few studies have demonstrated the role of pre-existing sequence variants in antibiotic resistance. For instance, the MacLean group showed that mixed strain Pseudomonas infections had a higher propensity to develop antibiotic resistance [18]. Similarly, pre-existing chromosomal polymorphisms enhanced the evolvability of Escherichia coli towards colistin [19]. In some cases, molecular and genetic mechanisms have also been suggested. For example, epistasis between pre-existing variants and resistance-conferring mutations has been shown to enhance evolvability of drug resistance in Neisseria, Streptococcus and Staphylococcus through different molecular mechanisms like metabolism, drug efflux and cyclic nucleotide signalling [20–22]. Polymorphisms in bacterial two-component systems impact multiple phenotypes. For instance, polymorphisms in two-component systems affect traits such as quorum sensing (ComP/ComA in Bacillus [23]), pathogenesis (RocA/CsrR/CsrS in Streptococcus [24]) and pH sensing (ArsS in Helicobacter [25], EvgS in E. coli [26]). However, though the molecular basis of these phenotypes is understood, their impact over evolutionary timescales is unknown. In the present study, we bridge this gap by linking molecular mechanistic impacts of polymorphisms in a bacterial signalling pathway with their contribution to evolvability of drug resistance.
In earlier work [27], we discovered a gene regulatory pathway responsible for adaptation of Escherichia coli to Trimethoprim (TMP) (Fig 1A). TMP is a broad-spectrum antibiotic that inhibits Dihydrofolate Reductase (DHFR), encoded by the folA gene. The susceptibility of E. coli to TMP is contingent on the expression level of DHFR [28]. Upon TMP challenge, E. coli upregulates the expression of folA via cis-regulatory mutations in the promoter of folA or higher gene copy number [29]. However, one of the earliest loci implicated is mgrB, coding for a small membrane protein that inhibits the PhoQ-PhoP two-component signalling system. Mutations in mgrB that emerge spontaneously under TMP pressure either inactivate the gene or its promoter [27,30]. In the absence of MgrB, the PhoQ sensor kinase is de-repressed, resulting in greater phosphorylation of PhoP, its cognate response regulator. An increase in active PhoP enhances the expression of DHFR and improves fitness of E. coli in the presence of TMP (Fig 1A) [27,30].
A: Schematic representation of the PhoPQ two-component system and its target DHFR implicated in resistance to TMP in E. coli. Active PhoQ results in phosphorylation of the response regulator PhoP. PhoP positively regulates transcription of mgrB and folA (coding for the DHFR enzyme). MgrB inhibits the kinase activity of PhoQ, setting up a negative feedback loop. Evolution under TMP pressure results in mutational inactivation of mgrB leading to PhoQ de-repression. This increases DHFR transcription and confers trimethoprim resistance to E. coli. B: Top panel Distribution of naturally occurring polymorphisms in the PhoQ protein from E. coli. Position of each polymorphism in PhoQ (X-axis) and predicted stability (ΔΔG, Y-axis) are plotted. Values of ΔΔG between −1 and +1 kcal/mol were considered to be neutral. The frequency of occurrence of each variant is represented by a grey-to-red colour scale as shown. Domain boundaries are indicated above the plot. Bottom panel: Domain-wise distribution of polymorphisms in PhoQ and 4 different sensor kinases from E. coli. Enrichment of variants in the kinase domain is apparent in all cases. C, D: Distribution of naturally occurring polymorphisms in the PhoP (C) and DHFR (D) proteins from E. coli. As with PhoQ, position of each polymorphism (X-axis) and predicted stability (ΔΔG, Y-axis) are plotted. The frequency of occurrence of each variant is represented by a grey-to-red colour scale as shown. Domain boundaries are marked above the plot. The data underlying this Figure can be found in S1 File.
Though inactivation of MgrB results in low-level TMP resistance by itself, it synergizes with mutations in DHFR that arise subsequently to confer high-level drug resistance [27,30]. As a result, loss of MgrB acts as a gateway mutation that facilitates the evolution of TMP resistance in E. coli, such that preventing mutations in mgrB slows down the evolution of resistance [30]. The PhoQ-PhoP-DHFR pathway, therefore, provides an ideal system to investigate the impact of pre-existing genetic variation on the evolution of drug resistance. In this study, we combine bioinformatic data mining, allele reconstruction, structure-based mutagenesis, and experimental evolution to uncover genetic polymorphisms in PhoQ and DHFR that alter the intrinsic TMP susceptibility of E. coli, as well as the potential to evolve resistance. Using transcriptomics and co-culture experiments on natural variant and drug-selected mutants we ask how TMP concentration determines the optimal activity of PhoQ during adaptation. Finally, we uncouple the relative contributions of benefit due to increased DHFR and cost due to cross-activation of other transcriptional programs in determining PhoQ activity during evolution.
Results
Sequence variation within the PhoQ-PhoP-DHFR pathway among E. coli strains
To assess the extent of genetic polymorphism within the PhoQ-PhoP-DHFR pathway (Fig 1A), we compared their sequences across strains of E. coli. A total of 180, 204, and 150 protein sequences of PhoQ, PhoP, and DHFR, respectively, from strains catalogued in the NCTC3000 database [31] were aligned and polymorphisms were identified by comparing with E. coli K-12 MG1655 (S1 File). Since this database contains many historically archived strains, we expected most polymorphisms to have emerged in the absence of antibiotic pressure. Commensurate with its large size, PhoQ, the sensor histidine kinase, showed the greatest number of sequence variants at 28 polymorphic sites (Fig 1B). PhoP had an intermediate number of polymorphisms at 8 sites (Fig 1C) while DHFR was the most conserved with only 3 polymorphic sites (Fig 1D). MgrB showed 6 polymorphisms (S1 File). However, it also showed several frame-shift variants similar to those isolated from trimethoprim resistant E. coli [27,30].
For PhoQ, the largest number of polymorphisms were in its C-terminal histidine kinase domain (Fig 1B). Interestingly, an enrichment of polymorphisms within the kinase domain was also observed in other sensors such as BaeS, CpxA, EnvZ, and BasS, all of which are implicated in antibiotic resistance [32–36], indicating a general rather than PhoQ-specific trend (Fig 1B, S1 File). We next computationally predicted the ΔΔG values of folding of all variants compared to wild type (S1 File) [37]. Since functional changes are often associated with perturbation of structure and function, exemplified by studies on DHFR [38–40], we used these predictions as a proxy for activity. All sequence variants of PhoP, except one, were predicted to be neutral (i.e., predicted ΔΔG between −1 and +1 kcal/mol) (Fig 1C). Similarly, most PhoQ variants were also predicted to be near-neutral (Fig 1B). However, Val382Gly and Phe372Gly, both located in the kinase domain of PhoQ, were predicted as highly destabilising (ΔΔG of −3.3 and −3.1 kcal/mol, respectively). Two of the three variants of DHFR, i.e., Trp47Arg and Pro105Ala were also predicted to be destabilising (ΔΔG of −1.48 and −1.66 kcal/mol, respectively) (Fig 1D). Based on these analyses we surmised that proteins of the PhoQ-PhoP-DHFR pathway demonstrated different extents of sequence polymorphism and some variants, such as those in PhoQ and DHFR, may have consequences for TMP susceptibility of E. coli.
Val382Gly is a mildly activating polymorphism in PhoQ that facilitates the evolution of resistance
Given the large number of variants in PhoQ, we first asked whether naturally occurring sequence polymorphisms in this protein had phenotypic consequences for E. coli. We selected 8 polymorphisms based on their location in PhoQ and expressed them from a plasmid in E. coli ΔphoQ. PhoQ-Val382Gly enhanced colony formation on TMP-containing media, while the other 7 variants had near-neutral phenotypes (Fig 2A, S2 File). To rule out artefacts due to plasmid-based expression, we generated a scarless genomic knock-in of PhoQ-Val382Gly at its native chromosomal locus. Indeed, a small but significant increase in colony forming efficiency on TMP-supplemented agar (Fig 2B) and 1.6-fold higher IC50 (concentration required for 50% growth inhibition) in broth cultures (Fig 2C) was observed for the knock-in strain. Further, the promoter activity of PphoP and PfolA, both responsive to active PhoP [27,41], was higher by 40% and 30%, respectively, in the knock-in compared to wild type (Fig 2D). Taken together, these results identified Val382Gly to be a mildly activating, naturally occurring polymorphism in PhoQ.
A: Colony-forming efficiencies of 8 naturally occurring PhoQ variants at different trimethoprim concentrations. Individual sequence variants of PhoQ were introduced into a phoQ knockout strain on a multi-copy plasmid. Mean log10(CFU/mL) values from 3 independent experiments are plotted on a yellow-green colour gradient. Individual raw values are provided in S2 File. B: Colony forming efficiency of wild type (black) and the PhoQ-Val382Gly knock-in (magenta) strains across trimethoprim concentrations. Mean log10(CFU/mL) values ± SD from three replicates are plotted (*p-value<0.05, Welch’s t test). C: Trimethoprim IC50 values for wildtype and PhoQ-Val382Gly knock-in strains. Mean ± SD from 3 replicates is plotted as bars and individual values are shown (*p-values≤0.05, Welch’s t test). D: Promoter activity of PphoP and PfolA in the phoQ knockout (black) and PhoQ-Val382Gly knock-in strains (magenta). Promoter activity was measured as the ratio of GFP reporter fluorescence to OD at 600 nm and normalised to wild type (set to 1, marked by a dotted line). Mean ± SD of 3 replicates and individual data points are plotted (*p-values≤0.05, Welch’s t test). E: Crystal structure of the catalytic histidine kinase domain of PhoQ (PDB: 1ID0) is shown as a cartoon. Bound ATP-analog AMPPNP is shown as sticks and coloured by element (N: Blue; C: Magenta, O: Red, P: Orange). Residue Val382 in the ATP-binding helix, possible interacting residues Ile452 (3.6 Å) and Val448 (4.2 Å) and N-box residues (Asn385 and Asn389) are shown as sticks and coloured by element (C: Green; N: Blue; O:Red). F: Colony-forming efficiencies for different substitutions at Val382 in PhoQ. Mean log10(CFU/mL) values from 3 independent experiments are plotted on a yellow-green colour gradient. Individual raw values are provided in S2 File. G: PphoP promoter activity in E. coli wild type (black) and PhoQ-Val382Gly knock-in (magenta) strains over a range of Mg2+ concentrations. Promoter activity at each concentration of Mg2+ was normalised to activity in the absence of Mg2+ supplementation. Mean ± SD from 3 replicates is plotted. The data underlying this Figure can be found in S2 File.
Val382 is located in an alpha-helix proximate to the conserved ‘N box’ in the catalytic domain, which interacts with the α-phosphate of ATP (Fig 2E) [42]. To better understand why Val382Gly led to activation, we created other amino acid substitutions at this site that altered flexibility, polarity, size and charge. Of all the substitutions generated, only Pro and Gly, enhanced TMP resistance (Fig 2F, S2 File). On the other hand, substitution of Val382 with Arg or Trp reduced colony formation on TMP-supplemented media, while substitution with Asp resembled wild type (Fig 2F, S2 File). Since Pro and Gly are helix-breaking amino acids due to their distinct structural properties, we reasoned that loss of local helicity or enhanced flexibility of this region was likely to activate the kinase domain of PhoQ. Importantly, like the wild type receptor, PhoQ-Val382Gly continued to be inhibited by Mg2+ ions, which bind to the extracellular sensor domain and allosterically inhibit the catalytic domain [43] (Fig 2G). These results indicated that while Val382Gly resulted in higher basal activity due to a structural perturbation of the kinase domain, it did not alter signal transduction between the domains of PhoQ.
Strikingly, deleting mgrB synergised with PhoQ-Val382Gly to confer high level resistance to TMP (Fig 3A). Indeed, PhoQ-Val382Gly and ΔmgrB together resulted in ~17 fold higher IC50 of TMP compared to ~2.5 fold for ΔmgrB alone (Fig 3A and 3B), demonstrating strong positive epistasis (ε = 0.015, σε = 0.008) between these two alleles. The difference arose due to higher DHFR expression in the PhoQ-Val382Gly ΔmgrB strain, showing that PhoQ-Val382Gly led to more profound activation of the signalling pathway upon de-repression (Fig 3C). This result suggested that even though PhoQ-Val382Gly was itself only mildly activating, it could facilitate the evolution of high level resistance to TMP. In order to directly test this premise, we checked for the evolvability of the knock-in strain using a drug challenge assay. Here, 24 replicate populations each of wild type and the PhoQ-Val382Gly knock-in were subjected to 4 different TMP concentrations, i.e., 100 ng/mL (0.1xMICwt), 1 μg/mL (MICwt), 2 μg/mL (2x MICwt) and 4 μg/mL (4x MICwt), over a period of 5 days. At the end of each day, we passaged 1% of all populations into fresh growth media and assessed the number of surviving replicates of each genotype (Fig 3D). Evolvability was estimated as the number of replicates that evolved resistance, i.e., by estimating extent of adaptation. As expected, at MICwt and above the wild type showed a dramatic loss in viability, and though a few populations did evolve resistance, a majority were driven to extinction (Fig 3E). Interestingly, a greater number of replicates of the PhoQ-Val382Gly knock-in survived the drug challenge and evolved resistance under drug pressure (Cox proportional-hazards ratio at MICwt β1 = 3.27, 95% CI = 1.342 to 9.160", likelihood ratio = 6.948, p = 0.0084) (Fig 3E). Thus, the PhoQ-Val382Gly polymorphism, though only mildly activating, enhanced the evolvability of E. coli under TMP pressure.
A: Left panel. Trimethoprim dose response for wild type (black) and ΔmgrB (grey) strains of E. coli. Right panel. Trimethoprim dose response for PhoQ-Val382Gly knock-in strain (magenta) and its ΔmgrB derivative (grey). Growth at each concentration of TMP is normalised to growth in drug-free media. Mean ± SD from 3 replicates is plotted. Values of IC50 derived from curve-fitting (mean ± SD) are provided above the graph. B: Synergistic epistasis between PhoQ-Val382Gly and ΔmgrB alleles. Fold IC50 over wild type for indicated strains are plotted as bars and values are provided. The values of ε and σε indicating positive epistasis are shown above the bars. The expected fold change based on an additive interaction is indicated by a dotted line. C: DHFR expression in E. coli wild type, ΔmgrB, PhoQ-Val382Gly knock-in and its ΔmgrB derivative assessed using immunoblotting. FtsZ was used as a loading control. Representative data from 3 independent replicates are shown. D: Schematic for the antibiotic challenge assay used to assess evolvability of trimethoprim resistance for E. coli wildtype and the PhoQ-Val382Gly knock-in strains. E: Number of replicates of E. coli wildtype and the PhoQ-Val382Gly knock-in strains showing growth over 5 days of antibiotic challenge at different TMP pressures. The data underlying this Figure can be found in S2 File and S1 Raw Image.
A hypersensitive variant of DHFR with reduced evolvability reveals a novel activating mutation in PhoQ
We next turned our attention to the two polymorphisms in DHFR that were predicted to alter function, i.e., Trp47Arg and Pro105Ala. When expressed from a plasmid, both variants conferred a 30%–40% lower IC50 than wild-type DHFR (Fig 4A). Intrigued by the finding, we next sought to study the effect of combining these polymorphs with previously reported trimethoprim-resistant mutations in DHFR since intragenic epistasis has been reported for this protein [38,39,44]. Strikingly, combining Trp47Arg with at least 4 different resistance-conferring mutations led to substantially diminished TMP resistance (Fig 4A). A similar impact was observed with Pro105Ala, though the magnitude of this effect was milder (Fig 4A). Since Trp47Arg showed a more pronounced phenotype, we proceeded with this variant for further experiments. A trade-off between in vivo proteolytic stability and resistance has been reported for DHFR and is the molecular basis for negative epistasis among resistance conferring mutations in the protein [38,39]. To test if this mechanism also explained the impact of Trp47Arg, we analysed the expression level of plasmid-borne DHFR harbouring each of the 4 resistance conferring mutations, with and without the natural variant. Indeed, Trp47Arg exacerbated the proteolytic instability of resistant DHFR alleles in cells, leading to reduced expression levels, and explaining the observed negative epistasis (Fig 4B).
A: Trimethoprim IC50 values for E. coli strains heterologously expressing natural DHFR variants, resistant mutants and their combinations. Bars represent mean ± S.D. from at least 3 independent replicates. Individual values are also plotted (*p-values≤0.05, Welch’s t test). B: Expression level of heterologously expressed DHFR harbouring resistance-conferring mutation with and without Trp47Arg assessed by immunoblotting. Representative data from 3 independent replicates in shown. Ratio of expression level of each resistant mutant with and without Trp47Arg mutation is provided (fold R47/W47). C: Trimethoprim dose response for E. coli wild type (black) and the DHFR-Trp47Arg knock-in (blue) strains. Growth at each concentration of TMP is normalised to growth in drug-free media. Mean ± SD from 3 replicates is plotted. Values of IC50 derived from curve-fitting (mean ± SD) are provided above the graph. D: Left panel. Schematic for measurement of relative fitness of the DHFR-Trp47Arg knock-in strain. Right panel. Relative fitness of the DHFR-Trp47Arg knock-in strain in the absence or presence of TMP. Empty bars represent data from competitions between E. coli ΔlacZ and the DHFR-Trp47Arg knock-in, while solid bars represent competitions between E. coli wild type and the ΔlacZ derivative of the DHFR-Trp47Arg knock-in. Bars represent mean ± SD from 3 replicates. Individual data points are also plotted. No change in fitness (w = 1) is shown as a dotted line. Magnitude of cost upon addition of trimethoprim to growth media is also shown. E: Evolvability of E. coli wild type and the DHFR-Trp47Arg knock-in strain assessed using an antibiotic challenge assay. Number of replicates of each genotype showing growth over a period of 5 days at different TMP pressures are plotted. F: Prevalence of mutations in folA (gene duplication/single nucleotide changes), mgrB and dnaA/dnaB/diaA in 15 replicate evolving lineages of E. coli wild type subjected to 500 ng/mL of TMP after 12 passages. Number of replicate lineages in which mutations in each of the indicated genes were detected are shown as pie charts. G: Mutations at the folA, mgrB, dnaA and phoQ loci detected in the 4 surviving lineages of DHFR-Trp47Arg knock-in strain that had adapted to TMP. H: Immunoblot to assess the expression level of DHFR in the ancestral DHFR-Trp47Arg knock-in strain and 4 surviving lineages that were passaged in TMP. FtsZ was used as a loading control. Representative data from 3 replicates are shown. The data underlying this Figure can be found in S1 Raw Image and S2 and S3 Files.
Next, we generated a scarless knock-in of DHFR-Trp47Arg at the native folA locus. The knock-in strain showed a ~2-fold reduction in TMP IC50 (Fig 4C). It also showed a lower relative fitness than wild type in the presence of TMP (100 ng/mL), but not in drug-free media (Fig 4D), establishing it as hypersensitive to TMP. Given its negative epistasis with resistance-conferring mutations (Fig 4A), we investigated whether DHFR-Trp47Arg compromised evolvability of E. coli under TMP pressure. Indeed, the DHFR-Trp47Arg knock-in strain had a consistently lower evolvability at high drug pressure (Cox proportional-hazards ratio at MICwt β1 = 0.1504, 95% CI = 0.007961 to 0.8825, likelihood ratio = 4.482, p = 0.0343), and also led to a few replicates from 0.1xMICwt to go extinct (Fig 4E). Thus, DHFR-Trp47Arg induced hypersensitivity to TMP and also compromised the evolvability of E. coli under TMP pressure.
We asked whether the DHFR-Trp47Arg knock-in was at all capable of evolving TMP resistance and, if so, whether it adapted through similar mutations as the wild type. To facilitate adaptation, we performed a serial-transfer based laboratory evolution experiment using 15 replicate populations each of wild type and the DHFR-Trp47Arg knock-in. Both genotypes were exposed to 500 ng/mL of TMP (MICwt/2) for 12 growth cycles. Evolution experiments were performed in 3 sets of 5 replicates each, to minimise the chance of cross-contamination of replicates. As expected, at the end of 12 passages, all replicate populations of the wild type showed improved growth in TMP, though most replicates of the DHFR-Trp47Arg knock-in went extinct. However, 4 replicates survived and showed enhanced growth in TMP-supplemented media. Whole-genome sequencing revealed that TMP-adapted wild-type populations predominantly harboured inactivating mutations in mgrB, and point mutations or duplications in folA (Fig 4F, S3 File). Additionally, mutations in dnaA and its interacting proteins dnaB and diaA were also found repeatedly (Fig 4F, S3 File). Though some other loci also harboured mutations, at least two of the above genes (i.e. folA, mgrB and dnaA/dnaB/diaA) were mutated in all 15 replicates, implicating them as the major determinants of TMP resistance. This convergent pattern of mutations could not be attributed to artefacts such as cross-contamination, as even though the same loci were implicated across replicates, the precise mutations, as well as co-occurring mutations in other genes, were not identical (S3 File). Three out of 4 surviving populations of DHFR-Trp47Arg showed a similar mutational landscape, i.e., loss-of-function mutations in mgrB and amplification of the folA gene (Fig 4H, S3 File). Two of these three replicates also harboured mutations in dnaA (Fig 4H, S3 File). However, the fourth surviving population did not have mutations in mgrB, folA or dnaA. Instead, it harboured a single mutation, Ala20Pro, in the PhoQ sensor kinase (Fig 4H, S3 File). This mutation was absent from the list of naturally occurring PhoQ polymorphisms (S1 File). However, a single report identifying the same mutation from a clinical isolate of chlorhexidine/colistin resistant Klebsiella pneumoniae suggesting that it may be relevant under antibiotic pressure [45]. Interestingly, while mgrB mutations and folA amplifications co-occurred in the other three TMP-evolved populations, PhoQ-Ala20Pro was a stand-alone mutation (Fig 4H, S3 File). Nonetheless, it had elevated DHFR expression levels and showed comparable IC50 values to the other evolved populations (Fig 4G).
Substitution of Ala20 with Pro hyperactivates PhoQ
Since PhoQ-Ala20Pro was detected in the background of DHFR-Trp47Arg, we first asked whether the TMP resistant phenotype was due to a combined effect of the two alleles. We generated a plasmid-borne copy of PhoQ-Ala20Pro, introduced it into E. coli ΔphoQ (which had a wild-type folA locus) and found that it conferred high level of TMP resistance (Fig 5A, S2 File). We also generated a scarless genomic knock-in of PhoQ-Ala20Pro. This strain was resistant to TMP as well (Fig 5A and 5B, S2 File). Thus, PhoQ-Ala20Pro conferred TMP resistance independent of polymorphisms at the folA locus. The PhoQ-Ala20Pro knock-in showed substantially higher promoter activity of PfolA and PphoP compared to wild type (Fig 5C). Indeed, both promoters were stimulated to a greater extent than by MgrB deficiency (Fig 5C), demonstrating that PhoQ-Ala20Pro hyperactivated the two-component system beyond its de-repressed state. In line with this result, the IC50 of TMP for the PhoQ-Ala20Pro knock-in was higher than E. coli ΔmgrB (Fig 5B).
A: Colony-forming efficiencies of various mutations at Ala20 in PhoQ at different trimethoprim concentrations. Individual mutants of PhoQ were introduced into a phoQ knockout strain on a multi-copy plasmid. The last two columns represent colony forming efficiencies of the PhoQ-Ala20Pro knock-in (ΔphoQ::PhoQ-Ala20Pro) and ΔmgrB strains. Mean log10(CFU/mL) values from 3 independent experiments are plotted on a yellow-green colour gradient as indicated. Individual raw values are provided in S2 File. B: Trimethoprim IC50 for E. coli wild type, ΔmgrB and PhoQ-Ala20Pro knock-in strains. Mean ± SD from 3 replicates and individual data points are plotted (*p-value ≤0.05; Welch’s t test). C: Promoter activity for PphoP and PfolA in E. coli ΔphoQ, ΔmgrB and PhoQ-Ala20Pro knock-in strains, measured as GFP fluorescence/OD600 and normalised to wild type (set to 1). Mean ± SD from 3 replicates and individual data points are plotted (*p-value ≤0.05; Welch’s t test). D: Transmembrane region (helices H1 and H2) from a model of the PhoQ dimer generated using Alphafold. Helix H1 is shown in tan and H2 is shown as an olive cartoon. Ala20 is represented as sticks and coloured by element (C: Green). E: Promoter activity of PphoP in the presence of different concentrations of Mg2+ in E. coli wild type, ΔphoQ, ΔmgrB and PhoQ-Ala20Pro knock-in strains. Promoter activity at each concentration of Mg2+ was normalised to activity in the absence of Mg2+ supplementation. Mean ± SD from 3 replicates is plotted as bars. F: Promoter PphoP activity at 0 and 60 mM of Mg2+. GFP fluorescence was normalised to growth (OD at 600nm). Mean ± SD from at least 3 replicates for each strain is plotted as bars. Individual data points are also shown (*p-value ≤0.05; Welch’s t test). G: Epistasis between ΔmgrB and PhoQ-Ala20Pro alleles. Colony-forming efficiencies for PhoQ or PhoQ-Ala20Pro expressed from a plasmid (pPhoQWT/pPhoQA20P) in indicated genetic backgrounds, or the PhoQ-Ala20Pro knock-in strain (ΔphoQ::PhoQ-Ala20Pro) with and without mgrB at different TMP concentrations. Mean log10(CFU/mL) from 3 replicates is plotted on a yellow-green colour scale. Individual raw data are available in S2 File. H: Negative epistasis between ΔmgrB and PhoQ-Ala20Pro alleles measured using fold change in IC50 of trimethoprim (over wild type) for the genomic knock-in of PhoQ-Ala20Pro with and without mgrB. The values of ε and σε indicating negative epistasis are shown above the bars. Fold change in IC50 values based on an additive interaction is marked with a dotted line. I: DHFR expression level in the PhoQ-Ala20Pro knock-in strain and its ΔmgrB derivative assessed using immunoblotting. FtsZ was used as a loading control. Representative data from 3 replicates is shown. Fold change in expression upon deletion of mgrB is given below the immunoblot. J: Expression level of PhoQ protein determined using immunoblotting with an anti-HexaHis antibody. Wild type PhoQ or its mutants was expressed heterologously in the indicated genetic backgrounds. FtsZ was used as a loading control. Fold change for between mgrB-deficient and mgrB-expressing backgrounds is provided below the immunoblot. Representative data from 3 independent replicates is shown. The data underlying this Figure can be found in S2 File and S1 Raw Image.
Ala20 lies in the transmembrane helical region (TM1 helix) of PhoQ close to the cytosolic face of the inner membrane (Fig 5D) [46]. Since proline mutations introduce kinks in alpha-helices, we wondered if this might explain high activity of PhoQ-Ala20Pro. Like with Val382Gly, we generated a series of substitutions at Ala20 and tested their phenotype. Unexpectedly, PhoQ-Ala20Gly, which could also alter helicity, phenotypically resembled wild-type PhoQ (Fig 5A, S2 File). Substituting Ala20 with a negatively charged amino acid, i.e., Glu, resulted in TMP sensitivity compared to the wild type, while positively charged Lys phenocopied wild-type PhoQ (Fig 5A, S2 File). Interestingly, introduction of aromatic amino acids Trp or Phe, but not aliphatic amino acids Ile or Val, conferred TMP resistance, albeit to a lower extent than Pro (Fig 5A, S2 File). Thus, helix kinking or increased local flexibility were unlikely to be the reason for hyperactivity of PhoQ-Ala20Pro. Instead, activation of the receptor was mediated by an aromatic or heterocyclic ring at this position, possibly due to confined mobility of TM1 or altered interactions with the second PhoQ protomer.
Ala20Pro alters the impact of allosteric inhibitors of PhoQ
Since the TM1 helix is important for signal transduction [47], we asked how PhoQ-Ala20Pro responded to two known allosteric inhibitors, i.e., Mg2+ and MgrB. Adding magnesium to growth media at saturating concentrations suppressed PphoP promoter activity by ~35% in the wild type strain. For ΔmgrB and the PhoQ-Ala20Pro knock-in, a greater inhibition of ~50% was observed (Fig 5E). However, though PphoP activity was reduced to the level of the phoQ knockout in wild type and ΔmgrB, it remained ~2-fold higher in the PhoQ-Ala20Pro knock-in strain even at the highest magnesium concentration (Fig 5F). Since these experiments were performed in LB medium, which may have a basal concentration of Mg2+ ions, we also repeated these experiments in M9 (minimal medium) supplemented with 0.1, 1 and 10 mM Mg2+. Here too, increasing Mg2+ concentrations inhibited PphoP activity in a dose-dependent manner across all strains. At 0.1 mM Mg2+, loss of MgrB rather than activation by the Ala20Pro mutation led to higher activity at the lowest concentration tested, though at higher Mg2+ concentrations once again Ala20Pro showed greater activation (S1 Fig). Thus, the PhoQ-Ala20Pro receptor continued to be sensitive to magnesium; but this mutant could activate PhoP even at saturating concentrations of Mg2+. These data indicated that the Ala20Pro mutation altered the basal activity of PhoQ, possibly by inducing a conformational change that locked the kinase domains in an active conformation.
The second allosteric inhibitor of PhoQ, i.e., MgrB, binds in the transmembrane region. It has been suggested that displacement of MgrB by cationic peptides leads to receptor activation [48]. We wondered whether the Ala20Pro mutation also activated PhoQ by a similar mechanism, i.e., by preventing binding to MgrB. If true, deletion of mgrB from E. coli expressing PhoQ-Ala20Pro would have no impact on TMP resistance. On the other hand, if MgrB continued to inhibit PhoQ-Ala20Pro, loss of the repressor would further increase TMP resistance. Contrary to both expectations, MgrB-deficiency reduced colony formation on TMP-supplemented media for plasmid-borne as well as genomically encoded PhoQ-Ala20Pro (Fig 5G, S2 File). Fold change in TMP IC50 values confirmed negative epistasis (ε = −0.047, σε = 0.02) between PhoQ-Ala20Pro and ΔmgrB (Fig 5H). DHFR expression was also lower in the double mutant than in the PhoQ-Ala20Pro knock-in (Fig 5I).
Immunoblotting revealed that in the absence of MgrB, protein level of PhoQ-Ala20Pro was substantially lower than in its presence (Fig 5J). Wild type PhoQ protein expression was unaffected by MgrB. Interestingly, the level of the Ala20Trp mutant, which also resulted in TMP resistance, was unaffected by MgrB-deficiency (Fig 5J). In line with this observation, its resistant phenotype was also unaffected by loss of MgrB (S2 File). These results indicated that substitution of Ala20 to Pro led to destabilisation of the PhoQ protein and likely rendered it susceptible to proteolysis. MgrB stabilised the mutant protein, either by directly binding to it or through an indirect effect, leading to greater activation of PhoP and higher TMP resistance. Importantly, this also explained why PhoQ-Ala20Pro occurred as a stand-alone mutation rather than in combination with mgrB mutations in our evolution experiment.
Gene regulatory effects of PhoQ pathway mutants depend on its activity level
In E. coli, 55 protein coding genes are known to be direct targets of PhoP [49]. Interestingly, some of the targets of PhoP are themselves regulators of gene expression, and in a previous study we showed that loss-of-function mutations in MgrB affect the primary and secondary targets of this pathway [30]. We, therefore, used RNA-sequencing to ask how activation of the PhoQP pathway by a naturally occurring polymorphism and mutations that arose under antibiotic selection affected gene expression in E. coli. For these analyses, in addition to PhoQ-Val382Gly (mildly activated PhoQP) and PhoQ-Ala20Pro (hyperactivated PhoQP) knock-in strains, we also included wild type, ΔmgrB (de-repressed PhoQP) and the phoQ knockout for comparison.
The natural variant allele PhoQ-Val382Gly led to almost no change in the expression of the PhoP-regulon (Fig 6A, S4 File). However, de-repression due to loss of MgrB or hyperactivation due to PhoQ-Ala20Pro resulted in pronounced up-regulation of several PhoP-target genes (Fig 6A, S4 File). Notably, though PhoQ-Ala20Pro resulted in greater up-regulation of the many genes than loss of MgrB, this was not reflected in all PhoP-targets (Fig 6A, S4 File). For instance, levels of phoP transcript itself were comparable between the de-repressed and hyperactive states (8.5-fold and 7.5-fold over wild type, respectively). These data suggested the different levels of active PhoQP translated to different expression levels for only a sub-set of the PhoP regulon.
A: Heatmap showing the expression levels of individual genes of the PhoP regulon on a green-red scale as indicated. Genes are arranged according to their deregulation in the PhoQ-Ala20Pro knock-in strain B: A schematic showing two pathways cross-activated by the PhoQ/PhoP two component system. Activation of the RstAB-Asr pathway is adaptive in TMP, while activation of IraM-RpoS is maladaptive resulting in a fitness cost. C: Expression level changes for RstA-regulated genes in the indicated strains. Individual points indicate genes of the RstA regulon, while the mean expression of the regulon is shown as a line. Asr gene expression scales with PhoQ activity and is highlighted in blue. Statistical significance was tested using a Welch’s t test, *p-value ≤ 0.05. D: Relative fitness (w) of indicated PhoQ pathway mutants and their ΔrpoS derivates measured using competition with E. coli ΔlacZ in drug-free LB medium. No change in fitness would result in a value of 1 which is indicated by a line. Mean ± SD from at least 3 replicates is plotted as bars. Individual data points are also shown. Magnitudes of fitness costs of E. coli ΔmgrB and PhoQ-Ala20Pro knock-in strains are expressed as percentages and indicated above the respective bars. Statistical significance was tested using a Welch’s t test, *p-value ≤ 0.05. The data underlying this Figure can be found in S4 File.
Through its transcriptional targets, PhoP activates other regulatory pathways in E. coli (Fig 6D). One such pathway implicated in adaptation to TMP is the RstAB two-component system and its regulon, including the acid-resistance protein Asr (Fig 6B and 6C, S4 File) [30]. Asr expression was ~11-fold higher in the ΔmgrB strain, while it was up-regulated by ~51-fold in the PhoQ-Ala20Pro knock-in (Fig 6C). Interestingly, the most highly up-regulated PhoP-target gene in the PhoQ-Ala20Pro knock-in was iraM (Fig 6A and 6C, S4 File). Its expression was elevated ~1,000 fold in the PhoQ-Ala20Pro knock-in, while it was induced by ~12 fold in the ΔmgrB strain (Fig 6A, S4 File). We have shown earlier that IraM-dependent overproduction of RpoS-regulated genes imposes a fitness cost [30]. This fitness cost can be alleviated by deleting either iraM or rpoS. The massive overexpression of iraM in the PhoQ-Ala20Pro knock-in suggested that hyperactivation of PhoQ was even more costly than de-repression. In line with this result, the PhoQ-Ala20Pro knock-in had a lower relative fitness than the ΔmgrB strain in drug-free media, and deletion of rpoS completely eliminated this cost (Fig 6D). Importantly, PhoQ-Val382Gly led to only a mild induction of iraM (Fig 6A, S4 File) and no detectable cost to fitness (Fig 6D). Based on these results, we concluded that mutations altering PhoQ activity that emerged under TMP-selection led to deregulation of gene expression due to a combination of primary and secondary effects. On the other hand, a naturally occurring activating polymorphism only mildly perturbed the gene regulatory network of E. coli and was not detrimental for fitness in the absence of drug.
Fitness cost limits high activity mutations in PhoQ during adaptation to TMP
The above experiments demonstrated that for mutations altering PhoQ activity, TMP resistance-level scaled with fitness cost. This trend led us to ask how the optimal activity of the PhoQ-PhoP-DHFR pathway was set during evolution, i.e., what was the relationship between drug pressure and the most preferred activation state of PhoQ. First, we used pair-wise co-cultures to assess how TMP altered allele frequencies of mildly active (PhoQ-Val382Gly), derepressed (ΔmgrB) and hyperactive (PhoQ-Ala20Pro) mutants in competition with wild type. Competing strains were mixed at a 1:1 ratio (i.e., an initial frequency of 0.5) in drug-free or TMP-supplemented medium and changes in their frequency after a single growth cycle were assessed. Any deviation from a frequency of 0.5 indicated a preference for one of the two competing strains (Fig 7A).
A: Schematic for pairwise competition between PhoQ pathway mutants performed across different trimethoprim concentrations to determine the relationship between TMP selection pressure and optimal PhoQ activity B-H: Results of pairwise competition between indicated strains with difference PhoQ activity levels. For each competition a pair of lacZ-expressing and lacZ-deficient strains were mixed at an initial ratio of 1:1 (i.e., frequency of 0.5 for each of the two strains in the mixture). Frequency of each competitor after competition (left Y-axis) was calculated. Total CFU/mL of the mixed cultures is plotted on the right Y-axis as bars. Limit of detection (~106 CFU/mL) is shown as a dotted line. Mean ± SD from 3 replicates is shown. I: Competition between DHFR-Trp47Arg knock-in harbouring a ΔlacZ marker and TMP-adapted DHFR-Trp47Arg knock-in harbouring the PhoQ-Ala20Pro mutation (Line 7) at different TMP concentrations. Frequency of each competitor after competition (left Y-axis) was calculated. Total CFU/mL of the mixed culture is plotted on the right Y-axis as bars. Limit of detection (~106 CFU/mL) is shown as a dotted line Mean ± SD from 3 replicates is plotted. The data underlying this Figure can be found in S5 File.
In line with its mildly resistant phenotype and no detectable cost, PhoQ-Val382Gly knock-in and wild type remained equi-proportionate over much of the TMP concentration range. However, close to MICwt (~1 μg/mL) the PhoQ-Val382Gly knock-in was enriched, indicating that even though this polymorphism had a small impact on PhoQ activity, it provided sufficient benefit to be selected by TMP (Fig 7B). Yet, at TMP concentrations higher than MICwt, both, wild type and the knock-in were driven to extinction, demonstrating a narrow selection window (Fig 7B). On the other hand, ΔmgrB and the PhoQ-Ala20Pro knock-in strain increased in frequency over a much wider range of TMP concentrations, though neither was preferred in the absence of antibiotic commensurate with their fitness costs (Fig 7C and 7D). Key differences between de-repressed and hyperactivated PhoQ were apparent. E. coli ΔmgrB was selected over wild type starting from the lowest concentration of antibiotic tested, i.e., 50 ng/mL (MICwt/20), while the PhoQ-Ala20Pro knock-in required a higher drug concentration (100 ng/mL, MICwt/10) to offset its larger fitness cost (Fig 7C and 7D). Furthermore, though the ΔmgrB strain did not grow beyond MICwt, the PhoQ-Ala20Pro knock-in survived till the highest drug concentration tested.
Next, we performed pairwise competitions between ΔmgrB and the PhoQ-Ala20Pro knock-in. Interestingly, the PhoQ-Ala20Pro knock-in was enriched over ΔmgrB only at TMP concentrations approaching the MICwt, i.e., at 1 μg/mL or above (Fig 7E). On the other hand, ΔmgrB out-competed the hyperactive mutant at all sub-MICwt concentrations including in the absence of antibiotic (Fig 7E). This observation indicated that at sub-MICwt pressures fitness cost was the main factor determining the preferred adaptive strategy, while at higher pressures resistance level primarily determined the outcome. To confirm this idea, we performed similar competition assays between the PhoQ-Ala20Pro knock-in harbouring a deletion in rpoS and E. coli wild type or ΔmgrB. Indeed, the cost-compensated PhoQ-Ala20Pro knock-in out-competed both the wild type and ΔmgrB strains across the entire range of TMP pressures (Fig 7F and 7G). When both, de-repressed and hyperactive PhoQ were cost-compensated, once again, the PhoQ-Ala20Pro knock-in was selected over much of the TMP concentration range tested, verifying our model (Fig 7H).
The above results cemented the impact of fitness cost due to cross-activation of RpoS in limiting the evolution of high activity mutations in the PhoQ. The repeated isolation of mgrB mutations and the almost complete absence of mutations in PhoQ during several independent laboratory evolution experiments performed here and in the past [27,29,50] resonated strongly with this result. However, it raised the question of why a hyperactivating mutation in PhoQ was so readily isolated in the DHFR-Trp47Arg knock-in. A possible explanation was that since this strain was hypersensitive to begin with, the presence of a pre-existing deleterious mutation would alter the selection dynamics of PhoQ-Ala20Pro. To test this, we allowed the DHFR-Trp47Arg knock-in and the TMP-adapted population harbouring PhoQ-Ala20Pro to compete at different drug pressures. Interestingly, the adapted population was enriched over its ancestor even at 50 ng/mL of TMP and almost completed out-competed its ancestor by 100 ng/mL of TMP (Fig 7I). Thus, the presence of a deleterious natural variant allele at the DHFR locus facilitated the selection of a hyperactive mutant of PhoQ despite its high cost. These results demonstrated that polymorphisms in resistance genes not only altered the evolvability of bacteria under drug pressure, but also which mutations were likely to be selected by antibiotics.
Discussion
Genetic polymorphisms in bacterial two-component systems are associated with phenotypes such as stress tolerance and virulence. Despite mechanistic understanding of how polymorphisms alter bacterial physiology, their role in adaptation is not clear. In this study, we uncovered naturally occurring polymorphisms in the PhoQ-PhoP-DHFR pathway that modulate intrinsic susceptibility of E. coli to TMP, as well as evolvability under drug pressure. In the case of Val382Gly, an activating polymorphism in PhoQ, positive epistasis with an early adaptive mutation increased evolvability. On the other hand, the DHFR-Trp47Arg variant reduced evolvability by sensitising E. coli to TMP and negative epistasis with mutations in the drug target. Though opposing in their effects, these two results are linked by a common genetic mechanism, i.e., epistasis between pre-existing mutations and those that emerge during adaptation (Fig 8A). Indeed, epistasis is a strong force shaping adaptation and previous studies have highlighted its role in historical contingency during evolution [51], as well as propagation and amplification of small fitness effects over genetic/metabolic networks [52]. The present study revealed that epistatic interactions change evolutionary outcomes, both quantitatively and qualitatively. The quantitative impact was reflected in the frequency of replicate populations that evolved resistance during antibiotic challenge. The qualitative impact was seen as a change in the fitness landscape of adaptive mutations. Loss of mgrB, a first step mutation during the evolution of TMP resistance, is a low-level resistant mutant by itself [27]. But the pre-existence of PhoQ-Val382Gly substantially potentiated its resistance phenotype. Likewise, Trp47Arg in DHFR exacerbated the instability of resistance-conferring mutations, reducing their resistance level. On the other hand, though the high resistance phenotype of PhoQ-Ala20Pro was consistent regardless of the allele of DHFR, its selection as a spontaneous stand-alone mutation was facilitated by the Trp47Arg variant of DHFR.
A: Polymorphisms influence evolvability under drug pressure if they have epistatic interactions with adaptive mutations. Positive epistasis between pre-existing polymorphisms and adaptive mutations promotes the evolution of resistance, i.e., enhances evolvability of resistance. In contrast, negative epistasis diminishes evolvability. In addition, polymorphisms with negative epistasis can also drive adaptation to antibiotics through novel mutational paths. B: PhoQP signalling pathway across different mutants from the present study. The impact of PhoQ activity on trimethoprim resistance due to DHFR expression is correlated with fitness cost due to RpoS cross-activation. Genetic variation in the pathway is brought about by naturally occurring polymorphisms or drug-selected adaptive mutations. These genetic changes in the pathway activate signalling to different extents at the molecular level, thereby altering global transcriptome to different extents. Scaling of trimethoprim resistance level and fitness cost leads to changes in the concentration of antibiotic required to select for different levels of PhoQ/PhoP pathway activity. The presence of a hypersensitising allele of DHFR drove E. coli towards a high activity, high cost PhoQ mutant demonstrating how polymorphisms can change the mutational landscape of adaptation to antibiotics.
Differing evolvability due to genetic variation provides a likely explanation for why certain strains of pathogenic bacteria are more prone to evolve multidrug resistance than others. A few examples from published literature support this idea. For instance, Ortiz and colleagues used extensive genomics analyses to identify “pre-resistance mutations” in some lineages of Mycobacterium tuberculosis [53]. These polymorphisms, though not themselves primary determinants of resistance were associated with a higher propensity for multidrug resistance, and could serve as predictors of the risk of resistance evolution. Similarly, Jangir and colleagues showed that the presence of polymorphisms in cell wall biosynthetic genes of pathogenic E. coli strains enhanced evolvability of resistance to colistin, a last resort antibiotic that targets the lipopolysaccharide layer in gram-negative bacteria [19]. We propose that understanding how pre-existing polymorphisms interact with adaptive mutations is an important direction to pursue for predicting risk of resistance evolution. Genome-wide association studies (GWAS) on antibiotic resistant pathogens frequently identify a number of variants linked with resistance [54]. However, most experimental validation pipelines focus on highest effect variants that are usually themselves determinants of resistance, while the role of others remain uncertain. As an example, the study by Diaz Caballero and colleagues identified polymorphisms associated with beta-lactam resistance in Burkholderia multivorans [55]. In addition to mutations in AmpD, a gene that is known to be associated with beta-lactam resistance across bacteria, the authors identified a large number of variants at other loci that were statistically highly likely to be involved in resistance. Strikingly, phosphorelay mechanisms including two-component systems were enriched, though their precise role was unexplored [55]. We propose that validation pipelines must also take into account effects on evolvability of resistance, which in the case of regulatory proteins may be profoundly important as demonstrated by our work.
The PhoQ-PhoP two-component system, originally discovered in Mg2+ homeostasis and acid adaptation, has emerged as a major regulator of antibiotic resistance in Enterobacteria. In addition to modulating TMP resistance in E. coli, its role in colistin resistance in Klebsiella pneumonia is well studied [56]. Additionally, an association with carbapenem resistance in Enterobacter sp. has been reported as well [57]. The wide range of phenotypes are ascribed to several downstream effectors creating a cascade of gene regulatory changes. Some of these effectors are direct targets, while others are regulated by targets of PhoP [43]. A major downstream pathway that we analysed in the present work was the stress responsive RpoS regulon, which manifests in a fitness cost [30]. Importantly, since cost of PhoQ activation scaled with benefit (i.e., TMP resistance), it capped the evolution of PhoQ activity during adaptation and prevented the fixation of high activity mutations such as PhoQ-Ala20Pro during antibiotic challenge (Fig 8B). Indeed, pairwise competition experiments showed that unless drug pressure was high (i.e., > MICwt), moderate activation of PhoQ, i.e., by de-repression, was the “fitter” strategy. The relationship between cross-talk and evolution of signalling pathways has been an area of active investigation for more than a decade. For bacterial signalling, particularly two-component systems, cross-talk has been viewed as undesirable. Indeed, many studies have investigated the insulation of signalling pathways as an evolutionary mechanism to improve signalling fidelity and organismal fitness [58–60]. In contradiction, instances of cross-activating transcriptional programs triggered by two-component systems are known. For instance, in E. coli small ‘connector’ proteins synthesise the impact of different two-component systems on gene expression [61–63]. Similarly, more than one two-component signalling system activates the stress-responsive sigma factor, RpoS [64]. The results presented in the present study provide an explanation for this apparent dichotomy. Our work shows that when signalling pathways evolve towards higher activity levels, cost due to cross-activation is likely only relevant above a threshold of activity (Fig 8B). The PhoQ-Val382Gly mutation, for instance, activated PhoP to provide an advantage in TMP without detectable cost. Signalling pathways may, thus, have some ‘wiggle room’ for tuning their activity. Ultimately, the strength of selection for higher activity would determine whether a signalling pathway overcomes the cost of cross-activation during its evolution.
Curiously, the impact of different activating mutations in the PhoQ-PhoP pathway did not scale with active PhoP for all target genes. We noticed two distinct patterns in terms of the responsiveness of gene promoters to active PhoP levels. First class of promoters had low sensitivity, i.e., they were affected only at moderate and high PhoQ activity, but were unaffected by mild activation or inactivation of the signalling pathway. The phoPQ operon itself also fell into this category. Importantly, these promoters also showed saturation, i.e., their expression level was comparable between de-repressed and hyperactivated states of PhoQ (Fig 6A). The second type of promoter was highly sensitive to PhoQ activity. Indeed, the iraM promoter best exemplified this category as its expression level was altered upon loss and gain of PhoQ activity. Interestingly, despite being highly sensitive, expression of iraM did not saturate with PhoQ activity, i.e., de-repression and hyperactivation of PhoQ were clearly reflected in the levels of iraM (Fig 6A). When we analysed the promoters of iraM and phoP as representative of the two types, we noticed that both promoters had two PhoP binding sites each. For iraM, one of the sites had a greater similarity to the PhoP-binding consensus sequence, indicating high affinity, while the other was likely to be a low affinity site [64,65]. On the other hand, both PhoP sites in the phoP promoter were expected to be high affinity sites [66,67]. Further, both iraM promoter sites are predicted to be activating, while one of the PhoP-binding sites in the phoP promoter is expected to be repressive. Thus, differences in promoter architecture likely contributed to the differences in responsiveness to active PhoP. Though, these speculations require mechanistic dissection and experimental validation, their implications are significant, as differential impact of activation of a signalling pathway on its targets would mean that not all targets contribute equally towards the fitness effects of activating mutations. By extension, the evolution of a signalling pathway may be driven primarily by only a subset of its regulon and which subset drives evolution may in turn depend on the environment. We propose that the PhoQ-PhoP signalling pathway evolving under trimethoprim pressure may prove to be a good model to test these ideas.
In summary, the work presented here identifies the contribution of pre-existing polymorphisms in a pathway associated with TMP resistance in determining evolvability of resistance. We also uncover how benefit and cost shape the evolution of a signalling pathway under selection. These findings establish the PhoQ-PhoP-DHFR pathway as a paradigm for studying the evolution of a drug resistance-associated pathway at the molecular and organismal levels
Materials and methods
Bacterial strains and culture conditions
Luria Bertani broth (LB) or agar (LA) was used to culture E. coli strains. All strains were cultures at 37 °C and broth cultures were grown with shaking at 180 rpm. Growth media were supplemented with different concentrations of trimethoprim (TMP) or magnesium sulphate as needed. Chloramphenicol (Cmp, 30 μg/mL), ampicillin (Amp, 100 μg/mL), kanamycin (Kan, 30 μg/mL), gentamycin (Gent, 45 μg/mL) and streptomycin (Strp, 45 μg/mL) were added to the media for the selection of transformants and knockouts as needed. List of strains and plasmids used in this study are provided in Tables 1 and 2, respectively.
Identification of naturally occurring polymorphisms in PhoQ, PhoP, and DHFR
To identify polymorphisms in proteins of interest, genome sequences available through the NCTC3000 database were mined [31]. Sample accession number of sequenced strains from NCTC3000 were acquired from the European Nucleotide Archive and the corresponding genome sequence was obtained from the Genbank. Amino acid sequences for the proteins of interest from each of the genomes were identified by looking for the appropriate annotation. Multiple sequence alignment was performed for PhoQ, PhoP, DHFR, BaeS, BasS, CpxA, and EnvZ proteins using Clustal W in Mega-X [70]. Protein sequences of each of these genes from E. coli K-12 MG1655 (U00096.3) were used as reference. Any deviation from the reference was recorded as a naturally occurring polymorphism. For each polymorphism, the frequency of occurrence was calculated as the ratio of the number of sequences harbouring the sequence variant and total number of sequences analysed.
For PhoP, PhoQ, and other sensor kinases domain boundaries were identified using the Interpro database [71]. Protein stability (ΔΔG values) for mutant PhoQ, PhoP, and DHFR was predicted using I-Mutant2.0 [37]. Values of ΔΔG between −1 and 1 kcal/mol were considered to not affect the stability of the protein, while any negative value below −1 was classified as destabilising.
Site-directed mutagenesis of PhoQ and DHFR
PCR-based site-directed mutagenesis was used to introduce mutations into plasmid-borne phoQ and folA genes using a protocol described in Shenoy and Visweswariah, 2003 [72]. For the generation of PhoQ mutants, pCA24N-phoQ was used as a template, while for DHFR mutants, pPRO-folA was used as template. All mutagenic primers are listed in the S6 File. Amplified PCR product was digested overnight using DpnI enzyme to remove the wild-type template. Digested product was then transformed into E. coli DH10B competent cells and selected on antibiotic-containing LA plates. Mutant plasmids were isolated from transformants using Alkali Lysis method, screened using restriction enzymes and confirmed by Sanger sequencing (Barcode Biosciences, India). For double mutants of DHFR, appropriate single mutants were used as template for the mutagenesis PCR.
Measurement of trimethoprim resistance
Method 1: Colony formation assay.
Colony formation efficiency on trimethoprim-supplemented LA was performed as described in Balachandran and colleagues, 2025 [50]. Overnight-grown cultures of appropriate strains were serially diluted in LB using a 10-fold dilution series. Each dilution (10 μL) was spotted on LA plates containing different concentrations of trimethoprim. The trimethoprim concentrations for most assays ranged from 100 ng/mL to 6.4 µg/mL. Plates were incubated overnight at 37 °C, and the colony-forming units (CFU/mL) were determined. Log(CFU/mL) was used to compare growth across strains.
Method 2: Determination of Inhibitory concentration-50 (IC50).
The IC50 of trimethoprim from E. coli strains was determined using a Broth Microdilution Assay, as described in Vinchhi and Yelpure and colleagues, 2023 [30]. Briefly, glycerol stocks were revived in 3 mL of LB broth. Trimethoprim was serially diluted across the wells of a 96-well plate to obtain a 2-fold dilution series and 1% culture was inoculated. Final concentrations of trimethoprim ranged from 1 mg/mL to 0.47 ng/mL. The plates were incubated overnight at 37 °C, and optical density was measured at 600 nm using a microplate reader (Ensight, Perkin Elmer). Using GraphPad Prism software, the IC50 was determined by fitting the data to a nonlinear regression log-inhibitor versus response curve (GraphPad Prism, version 9.1.3) as described earlier [30,50]. The fold change in IC50 relative to wild type was calculated wherever necessary.
Generation of scarless genomic knock-in strains of PhoQ and DHFR
Genomic knock-in strains were generated for PhoQ-Val382Gly, DHFR-Trp47Arg, and PhoQ-Ala20Pro using Triparental Conjugation based on the method adapted from Low and colleagues, 2022 [73]. For PhoQ-Val382Gly and PhoQ-Ala20Pro, the phoQ gene, together with flanking regions of homology, was cloned in the pSEVA612S plasmid using a three-part Gibson assembly. Primers used for generating these clones are listed in S6 File. Part 1 was amplified using E. coli wildtype genomic DNA as template and PhoQ homology forward and the PhoQ mutagenic forward primers. For part 2, PhoQ homology reverse and PhoQ mutagenic reverse primers were used. The third part was the pSEVA612S plasmid linearised using the I-SceI primer and pSEVA612S-GFP plasmid was used as template. The PCR products were DpnI-digested overnight. The three parts were combined in a 3:1 insert:vector molar ratio using a Gibson Assembly Master Mix (New England Biolabs, U.S.A.) at 50 °C for 30 min. The mixture was then transformed into E. coli CC118 competent cells and plated on gentamycin-containing LA plates. GFP-negative colonies were screened using restriction digestion and positive clones were confirmed by Sanger sequencing (Genematrix LLP, Pune, India). To integrate the mutation into the E. coli genome, a triparental conjugation was performed. E. coli CC118 containing pSEVA-phoQ plasmid with appropriate mutation was used as the donor. The recipient strain E. coli ΔphoQ::KanR was transformed with the pACBSR plasmid, which contained a streptomycin-selectable marker gene and arabinose-inducible I-SceI enzyme. E. coli pRK2013 helper strain was used for conjugation of pSEVA612S-phoQ plasmids from the donor to the recipient. For conjugation, overnight culture (20 µL) of the donor strain was spotted on LA and dried completely. Next, overnight culture (20 µL) of the helper stain was spotted over the donor strain, dried completely and incubated for 2 hours at 37 °C. Subsequently, 40 μL of the recipient was overlaid, dried and incubated for 4 hours at 37 °C. The growth obtained was scraped off the agar plate using a sterile nichrome loop, resuspended in 100 µL of fresh LB medium, spread on LA containing gentamycin and streptomycin and incubated at 37 °C overnight to select for single crossover merodiploid colonies. Three randomly selected colonies were inoculated into 3 mL LB, and I-SceI enzyme and Lambda Red Recombinase present on the pACBSR plasmid were induced using 0.2% L-Arabinose. After 2 hours and 6 hours of induction at 37 °C a loopful of culture was streaked-out on LA plates and incubated overnight at 37 °C. Colonies from the LA plates were patched on kanamycin-containing plates, and those that had lost kanamycin resistance were selected. Scarless knock-ins were screened using genomic PCR and mutation-specific restriction digestion. The presence of the mutations was further confirmed using Sanger sequencing of PCR-amplified phoQ gene. Confirmed knock-ins were passaged on an LA plate for 5 days to cure the pACBSR plasmid from the cell and streptomycin-sensitive knock-ins were used for all experiments.
A similar procedure was followed for the generation of the DHFR-Trp47Arg knock-in. For the triparental conjugation, barring the recipient strain, all other steps were the same. The recipient used to generate the DHFR-Trp47Arg knock-in was E. coli K-12 MG1655 harbouring the pACBSR plasmid.
Measuring phoP and folA promoter activity
Promoter activity for phoP and folA was measured using a GFP-reporter plasmid. Promoter reporter constructs were acquired from the E. coli promoter library (Dharmacon; PphoP- PEC3876-98154831, PfolA-PEC3876-98155011). Strains harbouring the reporter plasmids were grown in 3 mL LB in the absence or presence of different concentrations of magnesium sulphate as needed. Overnight-grown cultures were diluted 1:1 in sterile LB, and 200 µL of the diluted cultures were used to measure GFP fluorescence (excitation-485 nm, emission-519 nm) and optical density (OD600). GFP fluorescence was normalised to OD600 to compare across strains. For testing the effect of magnesium, promoter activity in its presence was normalised to activity in the absence of magnesium.
Immunoblotting for DHFR and PhoQ
DHFR protein levels in different strains were estimated using immunoblotting with anti-DHFR polyclonal IgG as described earlier [27,38]. For hexa-Histidine tagged PhoQ, anti-6xHis monoclonal IgG (MA1135, ThermoFisher, U.S.A.) was used. For DHFR, 5 µg of total cellular lysate protein was electrophoresed on a 15% SDS-PAGE gel and then electroblotted onto a PVDF membrane. For PhoQ, lysates were electrophoresed on a 12% SDS-PAGE gel. Membranes were blocked using 5% BSA for 1 hour at room temperature and incubated in primary antibody (100 ng/mL of anti-DHFR polyclonal IgG; 1 µg/mL anti-6xHis monoclonal IgG) at 4 °C overnight. HRP-linked anti-rabbit IgG (1:20,000) for DHFR and HRP-linked anti-mouse IgG (1:4000) for PhoQ were used as secondary antibody. Chemiluminescence was used to detect protein bands. FtsZ protein levels were used as a loading control. Anti-FtsZ polyclonal antiserum was used at a concentration of 1:50,000. Band intensities were quantitated using ImageJ and compared as required.
Gene deletion in E. coli strains
All gene knockouts were generated using P1 transduction. P1 lysates were raised on appropriate knockout strain from the Keio Collection [74] carrying the Kanamycin-resistance marker and transduced into the appropriate host strain of E. coli. Kanamycin-resistant colonies were selected and knockouts were confirmed by PCR using gene specific primers. To generate an unmarked knockout, the kanamycin resistance marker was excised from the knockout strain using the pCP20 plasmid containing FLP recombinase, as described by Baba and colleagues, 2006 [74].
Determination of evolvability for PhoQ-Val382Gly and DHFR-Trp47Arg knock-in strains
The difference in spontaneous resistance acquisition frequency across strains was assessed using an antibiotic challenge assay. Twenty-four replicate populations of appropriate genotype were inoculated in LB containing 100 ng/ml (0.1xMICwt), 1,000 ng/mL(1xMICwt), 2,000 ng/mL (2xMICwt) and 4,000 ng/mL (4xMICwt) of trimethoprim in 96-well plates. The plates were incubated at 37 °C for 20–24 hours, which represented 1 growth cycle. At the end of a growth cycle, 1% of culture from each well was transferred to fresh LB containing trimethoprim. Antibiotic challenge was performed for 5 days, i.e., over 5 growth cycles. At the end of each growth cycle, optical density (OD600) was measured. OD600 > 0.1 was recorded as growth, while OD600 < 0.1 was recorded as no growth.
Strain competitions and relative fitness
Relative fitness (w) was calculated using direct competition between test and reference strains that were distinguished by the presence or absence of ΔlacZ::Cat marker. The marker was swapped between strains for all experiments to establish its neutrality. Reference and test strains were grown as monocultures overnight in LB. They were then diluted 1:100 and combined in a 1:1 ratio in 3 mL of LB with or without appropriate concentration of trimethoprim. The initial CFU/mL for test and reference strains was assessed by serially diluting the mixed culture and plating on LA containing IPTG (1 mM) and X-Gal (50 µg/mL). The strains were allowed to compete for 24 hours at 37 °C with shaking at 180 rpm, after which CFU/mL was measured again. Relative fitness was calculated using the formula w = ln(Tf/Ti)/ln(Rf/Ri), where Tf and Rf were CFU/mL at the final time point, and Ti and Ri were CFU/mL at the initial time point for test and reference strains, respectively.
For pairwise competitions across different trimethoprim concentrations, similar co-cultures were set up. The frequency of each strain in the competition was measured by dividing the CFU/mL of the test strain by total CFU/mL.
Adaptive laboratory evolution of trimethoprim resistance and genome sequencing
For laboratory evolution of trimethoprim resistance 15 replicates of DHFR-Trp47Arg knock-in and E. coli wild type were grown in the wells of a 96-well plate in LB supplemented with 500 ng/mL of trimethoprim for 12 growth cycles. Replicates were derived from independent cultures that originated from three different colonies of each genotype. Furthermore, evolution was performed in 3 sets, each consisting of 5 replicates, to prevent cross-contamination. Retrospective genome sequencing analyses was used to confirm that replicates within each set were indeed independent of each other, i.e., did not show identical mutations. Each growth cycle was for 24 hours after which 10% of the culture was passaged into fresh media. At the end of 12 passages, glycerol stocks were prepared for surviving lineages and stored at −80 °C.
For genome sequencing, frozen stocks of the surviving lineages were revived in 3 mL LB and grown till saturation. Genomic DNA was extracted using the phenol:choloroform:isoamyl alcohol and cleaned-up using spin columns. Paired-end, whole-genome Next Generation sequencing was performed on a MiSeq system (Illumina, USA) with read lengths of 150−200 bps. Library preparation and sequencing services were provided by Eurofins, India. Mutations were identified using Breseq pipeline [75]. The genome of E. coli K-12 MG1655 (Genbank:U00096.3) was used as reference. Raw sequencing reads are available in Genbank (PRJNA1398517).
Transcriptome analysis using RNA sequencing
For RNA-sequencing, 1% of the overnight-grown cultures of appropriate strains were inoculated into 3 mL LB broth and incubated at 37 °C for 3 hours with shaking at 180 rpm. Three independent biological replicates were grown in parallel for these analyses. Cultures were centrifuged and pellets were resuspended in 1 mL of RNAlater (Invitrogen, U.S.A). Total RNA was extracted using Trizol, and the quality of the extracted RNA was estimated using TapeStation (Agilent). Depletion of rRNA was performed using the Ribozero kit/ NEBNext rRNA depletion kit. Following library preparation, total RNA-sequencing was performed on Illumina Miseq platform with a 150 bp read length. Processed reads were aligned to the reference genome (NC_00096.3/NC_0000913.3) using STAR(v 2.7.10a) and Strand NGS v4.1. Abundance estimation was done for counting reads corresponding to genes, promoter regions, and genomic bins using featureCounts (1.34.0). RNA extraction, ribodepletion, library preparation, sequencing and preliminary analysis were performed by Eurofins Genomics, India and Strand Lifesciences, India. Gene names were inferred from the gene ID for E. coli using the Ecocyc database [76]. PhoP-regulated genes were identified RegulonDB [49]. Differential gene expression analysis was performed using DESeq2 and EdgeR. All strains were compared to the wild type to obtain the Log2(fold change). Statistical significance was tested using the Benjamini–Hochberg test. A false discovery rate (FDR) cut-off of 0.05 was used to identify statistically significant gene expression changes. Raw sequencing reads are available in Genbank (PRJNA1399355).
Statistical analyses
Testing for difference in evolvability across strains.
Cox proportional hazards model was used to assess differences in evolvability across strains in a drug-challenge assay. Analysis of the model and prediction of hazard ratio was performed using GraphPad Prism software. Hazard ratios were determined using growth as an event (designated as 1) and no growth was censored (designated as 0). In all cases wild type was considered as the reference for comparison. A positive hazard ratio suggested that the strain had higher evolvability than the wild type. The statistical significance of the model (based on p-values) was calculated using the Log-likelihood test. The hazard ratios and p-values from the model are reported in the text and figure legends.
Calculation of epistasis and statistical significance using the error propagation method
Pairwise epistasis (ε) between two loci was calculated using a multiplicative model as described by Trindade and colleagues [77], as follows
where, AB referred to wildtype, ab was the double mutant, and Ab/aB were the single mutants. The IC50 values of trimethoprim (representing fitness, W) were directly used to calculate pairwise epistasis (ε). If the double mutant (Wab) had a higher IC50 than the expected IC50 (WaB*WAb), the epistatic interaction was positive, and if expected IC50 exceeded the observed IC50, the interaction was referred to as negative epistasis [78].
We used error propagation method to test if the epistasis was significantly different from zero. Error σε was given by:
where, σ represents the standard deviation of the measured IC50 values.
Epistasis (ε) was significantly different from zero, if the calculated value of ε was greater than the error σε [77,79].
Supporting information
S1 Fig. PhoP-promoter activity in minimal media.
Activity of the PhoP promoter measured using a GFP reporter in wild type (MG1655), phoQ knockout (ΔphoQ), mgrB knockout (ΔmgrB) and PhoQ knock-ins (V382G and A20P). Cultures were grown overnight in M9 media supplemented with indicated concentrations of MgSO4. GFP fluorescence (arbitrary units) was normalised to growth (optical density at 600 nm). Mean ± SD are plotted as bars and individual data points are overlaid. The data underlying this Figure are tabulated below the graph.
https://doi.org/10.1371/journal.pbio.3003977.s001
(TIFF)
S1 File. Polymorphisms in the PhoQ-PhoP-DHFR-MgrB pathway identified from genome sequences of E. coli strains available in the NCTC database.
https://doi.org/10.1371/journal.pbio.3003977.s002
(XLSX)
S2 File. Raw numerical data used for plotting graphs in Figs 2–5.
https://doi.org/10.1371/journal.pbio.3003977.s003
(XLSX)
S3 File. Genomes sequencing of E. coli wild type or the DHFR-Trp47Arg knock-in strain evolved in trimethoprim (500 ng/mL).
https://doi.org/10.1371/journal.pbio.3003977.s004
(XLSX)
S4 File. RNA-sequencing of wild type or mutant E. coli strains.
For all mutants, wild type E. coli was treated as reference.
https://doi.org/10.1371/journal.pbio.3003977.s005
(XLSX)
S5 File. Raw numerical data from pairwise strain competition experiments used for plotting graphs in Fig 7.
https://doi.org/10.1371/journal.pbio.3003977.s006
(XLSX)
S6 File. List of oligonucleotides and PCR primers used in this study.
https://doi.org/10.1371/journal.pbio.3003977.s007
(XLSX)
S1 Raw Image. Uncropped immunoblot images for Fig 3C, 4B, 4H, 5I, and 5J.
https://doi.org/10.1371/journal.pbio.3003977.s008
(TIFF)
Acknowledgments
We thank Prof. Gad Frankel (Imperial College, London, U.K.) for sharing plasmids and strains for generating genomic knock-in strains. Ms. Avani Joshi and Mr. Ishaan Chaudhary are acknowledged for technical assistance.
References
- 1. GBD 2021 Antimicrobial Resistance Collaborators. Global burden of bacterial antimicrobial resistance 1990-2021: a systematic analysis with forecasts to 2050. Lancet. 2024;404(10459):1199–226. pmid:39299261
- 2. Merker M, Tueffers L, Vallier M, Groth EE, Sonnenkalb L, Unterweger D, et al. Evolutionary approaches to combat antibiotic resistance: opportunities and challenges for precision medicine. Front Immunol. 2020;11:1938. pmid:32983122
- 3. Pélabon C, Agudelo-Cantero GA, Araya Ajoy YG, Bolstad GH, Cheng C, Galis F, et al. Evolvability: progress and key questions. Bioscience. 2025;75(12):1042–57. pmid:41367904
- 4. Brookfield JF. Evolution: the evolvability enigma. Curr Biol. 2001;11(3):R106-8. pmid:11231170
- 5. Hansen TF, Pélabon C. Evolvability: a quantitative-genetics perspective. Annu Rev Ecol Evol Syst. 2021;52(1):153–75.
- 6. Houle D. Comparing evolvability and variability of quantitative traits. Genetics. 1992;130(1):195–204. pmid:1732160
- 7. Pigliucci M. Is evolvability evolvable? Nat Rev Genet. 2008;9(1):75–82. pmid:18059367
- 8. Riederer JM, Tiso S, van Eldijk TJB, Weissing FJ. Capturing the facets of evolvability in a mechanistic framework. Trends Ecol Evol. 2022;37(5):430–9. pmid:35153079
- 9. Szathmáry E, Smith JM. The major evolutionary transitions. Nature. 1995;374(6519):227–32. pmid:7885442
- 10. Wagner GP, Altenberg L. Perspective: complex adaptations and the evolution of evolvability. Evolution. 1996;50(3):967–76. pmid:28565291
- 11. Garcia-Clemente M, de la Rosa D, Máiz L, Girón R, Blanco M, Olveira C, et al. Impact of Pseudomonas aeruginosa infection on patients with chronic inflammatory airway diseases. J Clin Med. 2020;9(12):3800. pmid:33255354
- 12. Jolivet-Gougeon A, Kovacs B, Le Gall-David S, Le Bars H, Bousarghin L, Bonnaure-Mallet M, et al. Bacterial hypermutation: clinical implications. J Med Microbiol. 2011;60(Pt 5):563–73. pmid:21349992
- 13. Shibai A, Izutsu M, Kotani H, Furusawa C. Quantitative analysis of relationship between mutation rate and speed of adaptation under antibiotic exposure in Escherichia coli. PLoS Genet. 2025;21(3):e1011627. pmid:40153704
- 14. Deaconescu AM. Mfd - at the crossroads of bacterial DNA repair, transcriptional regulation and molecular evolvability. Transcription. 2021;12(4):156–70. pmid:34674614
- 15. Merrikh H, Kohli RM. Targeting evolution to inhibit antibiotic resistance. FEBS J. 2020;287(20):4341–53. pmid:32434280
- 16. Durão P, Balbontín R, Gordo I. Evolutionary mechanisms shaping the maintenance of antibiotic resistance. Trends Microbiol. 2018;26(8):677–91. pmid:29439838
- 17. Hughes D, Andersson DI. Evolutionary trajectories to antibiotic resistance. Annu Rev Microbiol. 2017;71:579–96. pmid:28697667
- 18. Diaz Caballero J, Wheatley RM, Kapel N, López-Causapé C, Van der Schalk T, Quinn A, et al. Mixed strain pathogen populations accelerate the evolution of antibiotic resistance in patients. Nat Commun. 2023;14(1):4083. pmid:37438338
- 19. Jangir PK, Yang Q, Shaw LP, Caballero JD, Ogunlana L, Wheatley R, et al. Pre-existing chromosomal polymorphisms in pathogenic E. coli potentiate the evolution of resistance to a last-resort antibiotic. Elife. 2022;11:e78834. pmid:35943060
- 20. Kobras CM, Monteith W, Somerville S, Delaney JM, Khan I, Brimble C, et al. Loss of Pde1 function acts as an evolutionary gateway to penicillin resistance in Streptococcus pneumoniae. Proc Natl Acad Sci U S A. 2023;120(41):e2308029120. pmid:37796984
- 21. Papkou A, Hedge J, Kapel N, Young B, MacLean RC. Efflux pump activity potentiates the evolution of antibiotic resistance across S. aureus isolates. Nat Commun. 2020;11(1):3970. pmid:32769975
- 22. Rubin DHF, Ma KC, Westervelt KA, Hullahalli K, Waldor MK, Grad YH. CanB is a metabolic mediator of antibiotic resistance in Neisseria gonorrhoeae. Nat Microbiol. 2023;8(1):28–39. pmid:36604513
- 23. Tortosa P, Logsdon L, Kraigher B, Itoh Y, Mandic-Mulec I, Dubnau D. Specificity and genetic polymorphism of the Bacillus competence quorum-sensing system. J Bacteriol. 2001;183(2):451–60. pmid:11133937
- 24. Bernard PE, Kachroo P, Eraso JM, Zhu L, Madry JE, Linson SE, et al. Polymorphisms in regulator of cov contribute to the molecular pathogenesis of serotype M28 Group A streptococcus. Am J Pathol. 2019;189(10):2002–18. pmid:31369755
- 25. Hallinger DR, Romero-Gallo J, Peek RM Jr, Forsyth MH. Polymorphisms of the acid sensing histidine kinase gene arsS in Helicobacter pylori populations from anatomically distinct gastric sites. Microb Pathog. 2012;53(5–6):227–33. pmid:22940419
- 26. Roggiani M, Yadavalli SS, Goulian M. Natural variation of a sensor kinase controlling a conserved stress response pathway in Escherichia coli. PLoS Genet. 2017;13(11):e1007101. pmid:29140975
- 27. Patel V, Matange N. Adaptation and compensation in a bacterial gene regulatory network evolving under antibiotic selection. Elife. 2021;10:e70931. pmid:34591012
- 28. Palmer AC, Kishony R. Opposing effects of target overexpression reveal drug mechanisms. Nat Commun. 2014;5:4296. pmid:24980690
- 29. Jena C, Chinnaraj S, Deolankar S, Matange N. Proteostasis modulates gene dosage evolution in antibiotic-resistant bacteria. Elife. 2025;13:RP99785. pmid:40073078
- 30. Vinchhi R, Yelpure C, Balachandran M, Matange N. Pervasive gene deregulation underlies adaptation and maladaptation in trimethoprim-resistant E. coli. mBio. 2023;14(6):e0211923. pmid:38032208
- 31. Dicks J, Fazal M-A, Oliver K, Grayson NE, Turnbull JD, Bane E, et al. NCTC3000: a century of bacterial strain collecting leads to a rich genomic data resource. Microb Genom. 2023;9(5):mgen000976. pmid:37194944
- 32. Baranova N, Nikaido H. The baeSR two-component regulatory system activates transcription of the yegMNOB (mdtABCD) transporter gene cluster in Escherichia coli and increases its resistance to novobiocin and deoxycholate. J Bacteriol. 2002;184(15):4168–76. pmid:12107134
- 33. Janssen AB, van Schaik W. Harder, better, faster, stronger: colistin resistance mechanisms in Escherichia coli. PLoS Genet. 2021;17(1):e1009262. pmid:33411745
- 34. Srinivasan VB, Vaidyanathan V, Mondal A, Rajamohan G. Role of the two component signal transduction system CpxAR in conferring cefepime and chloramphenicol resistance in Klebsiella pneumoniae NTUH-K2044. PLoS One. 2012;7(4):e33777. pmid:22496764
- 35. Tierney AR, Rather PN. Roles of two-component regulatory systems in antibiotic resistance. Future Microbiol. 2019;14(6):533–52. pmid:31066586
- 36. Yang K, Zhang T, Geng J, Long J, Yang H, Duan G, et al. Molecular mechanism of envZ regulating β-lactam antibiotic resistance and virulence in Escherichia coli. J Antimicrob Chemother. 2026;81(5):dkag104. pmid:41967056
- 37. Capriotti E, Fariselli P, Casadio R. I-Mutant2.0: predicting stability changes upon mutation from the protein sequence or structure. Nucleic Acids Res. 2005;33(Web Server issue):W306-10. pmid:15980478
- 38. Matange N. Highly contingent phenotypes of lon protease deficiency in Escherichia coli upon antibiotic challenge. J Bacteriol. 2020;202(3):e00561-19. pmid:31740490
- 39. Matange N, Bodkhe S, Patel M, Shah P. Trade-offs with stability modulate innate and mutationally acquired drug resistance in bacterial dihydrofolate reductase enzymes. Biochem J. 2018;475(12):2107–25. pmid:29871875
- 40. Rodrigues JV, Bershtein S, Li A, Lozovsky ER, Hartl DL, Shakhnovich EI. Biophysical principles predict fitness landscapes of drug resistance. Proc Natl Acad Sci U S A. 2016;113(11):E1470-8. pmid:26929328
- 41. Miyashiro T, Goulian M. Stimulus-dependent differential regulation in the Escherichia coli PhoQ PhoP system. Proc Natl Acad Sci U S A. 2007;104(41):16305–10. pmid:17909183
- 42. Marina A, Mott C, Auyzenberg A, Hendrickson WA, Waldburger CD. Structural and mutational analysis of the PhoQ histidine kinase catalytic domain. Insight into the reaction mechanism. J Biol Chem. 2001;276(44):41182–90. pmid:11493605
- 43. Groisman EA, Duprey A, Choi J. How the PhoP/PhoQ system controls virulence and Mg(2) homeostasis: lessons in signal transduction, pathogenesis, physiology, and evolution. Microbiol Mol Biol Rev. 2021;85(3):e0017620. pmid:34191587
- 44. Palmer AC, Toprak E, Baym M, Kim S, Veres A, Bershtein S, et al. Delayed commitment to evolutionary fate in antibiotic resistance fitness landscapes. Nat Commun. 2015;6:7385. pmid:26060115
- 45. Wand ME, Bock LJ, Bonney LC, Sutton JM. Mechanisms of increased resistance to chlorhexidine and cross-resistance to colistin following exposure of Klebsiella pneumoniae clinical isolates to chlorhexidine. Antimicrob Agents Chemother. 2017;61(1).
- 46. Lazaridi S, Yuan J, Lemmin T. Atomic insights into the signaling landscape of E. coli PhoQ histidine kinase from molecular dynamics simulations. Sci Rep. 2024;14(1):17659. pmid:39085378
- 47. Mensa B, Polizzi NF, Molnar KS, Natale AM, Lemmin T, DeGrado WF. Allosteric mechanism of signal transduction in the two-component system histidine kinase PhoQ. Elife. 2021;10:e73336. pmid:34904568
- 48. Jiang S, Steup LC, Kippnich C, Lazaridi S, Malengo G, Lemmin T, et al. The inhibitory mechanism of a small protein reveals its role in antimicrobial peptide sensing. Proc Natl Acad Sci U S A. 2023;120(41):e2309607120. pmid:37792514
- 49. Salgado H, Gama-Castro S, Lara P, Mejia-Almonte C, Alarcón-Carranza G, López-Almazo AG, et al. RegulonDB v12.0: a comprehensive resource of transcriptional regulation in E. coli K-12. Nucleic Acids Res. 2024;52(D1):D255–64. pmid:37971353
- 50. Balachandran M, Chatterjee R, Chaudhary I, Jena C, Matange N. Impeding pathways of intrinsic resistance in Escherichia coli confers antibiotic sensitization and resistance proofing. PLoS Biol. 2025;23(10):e3003443. pmid:41124183
- 51. Card KJ, LaBar T, Gomez JB, Lenski RE. Historical contingency in the evolution of antibiotic resistance after decades of relaxed selection. PLoS Biol. 2019;17(10):e3000397. pmid:31644535
- 52. Kryazhimskiy S. Emergence and propagation of epistasis in metabolic networks. Elife. 2021;10:e60200. pmid:33527897
- 53. Torres Ortiz A, Coronel J, Vidal JR, Bonilla C, Moore DAJ, Gilman RH, et al. Genomic signatures of pre-resistance in Mycobacterium tuberculosis. Nat Commun. 2021;12(1):7312. pmid:34911948
- 54. Mosquera-Rendón J, Moreno-Herrera CX, Robledo J, Hurtado-Páez U. Genome-Wide Association Studies (GWAS) approaches for the detection of genetic variants associated with antibiotic resistance: a systematic review. Microorganisms. 2023;11(12):2866. pmid:38138010
- 55. Diaz Caballero J, Clark ST, Wang PW, Donaldson SL, Coburn B, Tullis DE, et al. A genome-wide association analysis reveals a potential role for recombination in the evolution of antimicrobial resistance in Burkholderia multivorans. PLoS Pathog. 2018;14(12):e1007453. pmid:30532201
- 56. Huang J, Li C, Song J, Velkov T, Wang L, Zhu Y, et al. Regulating polymyxin resistance in Gram-negative bacteria: roles of two-component systems PhoPQ and PmrAB. Future Microbiol. 2020;15(6):445–59. pmid:32250173
- 57. Murtha AN, Kazi MI, Schargel RD, Cross T, Fihn C, Cattoir V, et al. High-level carbapenem tolerance requires antibiotic-induced outer membrane modifications. PLoS Pathog. 2022;18(2):e1010307. pmid:35130322
- 58. Capra EJ, Perchuk BS, Lubin EA, Ashenberg O, Skerker JM, Laub MT. Systematic dissection and trajectory-scanning mutagenesis of the molecular interface that ensures specificity of two-component signaling pathways. PLoS Genet. 2010;6(11):e1001220. pmid:21124821
- 59. Rowland MA, Deeds EJ. Crosstalk and the evolution of specificity in two-component signaling. Proc Natl Acad Sci U S A. 2014;111(15):5550–5. pmid:24706803
- 60. Townsend GE 2nd, Raghavan V, Zwir I, Groisman EA. Intramolecular arrangement of sensor and regulator overcomes relaxed specificity in hybrid two-component systems. Proc Natl Acad Sci U S A. 2013;110(2):E161-9. pmid:23256153
- 61. Kato A, Hayashi H, Nomura W, Emori H, Hagihara K, Utsumi R. A connecter-like factor, CacA, links RssB/RpoS and the CpxR/CpxA two-component system in Salmonella. BMC Microbiol. 2012;12:224. pmid:23031642
- 62. Kato A, Mitrophanov AY, Groisman EA. A connector of two-component regulatory systems promotes signal amplification and persistence of expression. Proc Natl Acad Sci U S A. 2007;104(29):12063–8. pmid:17615238
- 63. Mitrophanov AY, Groisman EA. Signal integration in bacterial two-component regulatory systems. Genes Dev. 2008;22(19):2601–11. pmid:18832064
- 64. Eguchi Y, Ishii E, Hata K, Utsumi R. Regulation of acid resistance by connectors of two-component signal transduction systems in Escherichia coli. J Bacteriol. 2011;193(5):1222–8. pmid:21193607
- 65. Bougdour A, Cunning C, Baptiste PJ, Elliott T, Gottesman S. Multiple pathways for regulation of sigmaS (RpoS) stability in Escherichia coli via the action of multiple anti-adaptors. Mol Microbiol. 2008;68(2):298–313. pmid:18383615
- 66. Kato A, Tanabe H, Utsumi R. Molecular characterization of the PhoP-PhoQ two-component system in Escherichia coli K-12: identification of extracellular Mg2+-responsive promoters. J Bacteriol. 1999;181(17):5516–20. pmid:10464230
- 67. Minagawa S, Ogasawara H, Kato A, Yamamoto K, Eguchi Y, Oshima T, et al. Identification and molecular characterization of the Mg2+ stimulon of Escherichia coli. J Bacteriol. 2003;185(13):3696–702. pmid:12813061
- 68. Kitagawa M, Ara T, Arifuzzaman M, Ioka-Nakamichi T, Inamoto E, Toyonaga H, et al. Complete set of ORF clones of Escherichia coli ASKA library (a complete set of E. coli K-12 ORF archive): unique resources for biological research. DNA Res. 2005;12(5):291–9. pmid:16769691
- 69. Zaslaver A, Bren A, Ronen M, Itzkovitz S, Kikoin I, Shavit S, et al. A comprehensive library of fluorescent transcriptional reporters for Escherichia coli. Nat Methods. 2006;3(8):623–8. pmid:16862137
- 70. Kumar S, Stecher G, Li M, Knyaz C, Tamura K. MEGA X: molecular evolutionary genetics analysis across computing platforms. Mol Biol Evol. 2018;35(6):1547–9. pmid:29722887
- 71. Blum M, Andreeva A, Florentino LC, Chuguransky SR, Grego T, Hobbs E, et al. InterPro: the protein sequence classification resource in 2025. Nucleic Acids Res. 2025;53(D1):D444–56. pmid:39565202
- 72. Shenoy AR, Visweswariah SS. Site-directed mutagenesis using a single mutagenic oligonucleotide and DpnI digestion of template DNA. Anal Biochem. 2003;319(2):335–6. pmid:12871732
- 73. Low WW, Wong JLC, Beltran LC, Seddon C, David S, Kwong H-S, et al. Mating pair stabilization mediates bacterial conjugation species specificity. Nat Microbiol. 2022;7(7):1016–27. pmid:35697796
- 74. Baba T, Ara T, Hasegawa M, Takai Y, Okumura Y, Baba M, et al. Construction of Escherichia coli K-12 in-frame, single-gene knockout mutants: the Keio collection. Mol Syst Biol. 2006;2:2006.0008. pmid:16738554
- 75. Deatherage DE, Barrick JE. Identification of mutations in laboratory-evolved microbes from next-generation sequencing data using breseq. Methods Mol Biol. 2014;1151:165–88. pmid:24838886
- 76. Keseler IM, Gama-Castro S, Mackie A, Billington R, Bonavides-Martinez C, Caspi R, et al. The EcoCyc database in 2021. Front Microbiol. 2021;12:711077. pmid:34394059
- 77. Trindade S, Sousa A, Xavier KB, Dionisio F, Ferreira MG, Gordo I. Positive epistasis drives the acquisition of multidrug resistance. PLoS Genet. 2009;5(7):e1000578. pmid:19629166
- 78. Kouyos RD, Silander OK, Bonhoeffer S. Epistasis between deleterious mutations and the evolution of recombination. Trends Ecol Evol. 2007;22(6):308–15. pmid:17337087
- 79. Borrell S, Teo Y, Giardina F, Streicher EM, Klopper M, Feldmann J, et al. Epistasis between antibiotic resistance mutations drives the evolution of extensively drug-resistant tuberculosis. Evol Med Public Health. 2013;2013(1):65–74. pmid:24481187