Escape mutations circumvent a tradeoff between resistance to a beta-lactam and a beta-lactamase inhibitor

Beta-lactamase inhibitors are increasingly used to counteract microbial resistance to beta-lactam antibiotics mediated by beta-lactamase enzymes. These inhibitors compete with the beta-lactam drug for the same binding site of the beta-lactamase, thereby generating an inherent evolutionary tradeoff: enzyme mutations that increase its activity against the beta-lactam drug also increase its susceptibility towards the inhibitor. It is unclear how common and accessible are mutants that escape this adaptive tradeoff. Here, systematically constructing and phenotyping a deep mutant library of the ampC beta-lactamase gene of Escherichia coli, we identified escape mutations, which even in the presence of the enzyme inhibitor allow growth at beta-lactam concentrations far exceeding the native inhibitory levels of the wildtype strain. Importantly, while such escape mutations appear for combinations of avibactam with some beta-lactam drugs, for other drugs escape phenotypes are completely restricted. Amplicon sequencing of the selected mutant pool identified these escape mutations and showed that they are rare and drug specific. For the combination of avibactam with aztreonam, an escape phenotype was conferred via multiple substitutions in a single conserved amino acid (Tyr 150). In contrast, a different set of mutations showed an escape phenotype for cefepime, and no escape mutants appeared for piperacillin. The differential adaptive potential of ampC to combinations of avibactam and different beta-lactam drugs can help guide drug treatments that are more resilient to evolution of resistance.


Main text
Resistance to beta-lactam antibiotics, mediated primarily through beta-lactamases, is growing as a major threat to public health. Beta-lactams are commonly used for the treatment of a range of bacterial pathogens, constituting the most widely prescribed antibiotic class 1,2 . These antibiotics are characterized by a core of a beta-lactam ring and are classified by their moieties into four major families: penicillins, cephalosporins, carbapenems, and monobactams 3 . The irreversible binding of these drugs to the peptidoglycan cross-linking enzymes (Penicillin Binding Proteins, PBPs) leads to cell death and lysis. Resistance to beta-lactams has become widespread in recent years, mainly through drug degradation by beta-lactamases , most notably by the serine beta-lactamases of classes A and C 4,5 . Hydrolyzing the core beta-lactam ring, these enzymes decrease the effective drug concentration, thereby conferring increased resistance [6][7][8] . The effectiveness of these enzymes is often further improved either by mutations that increase their expression or by intragenic structural mutations that enhance their efficacy and specificity [9][10][11][12][13][14] . To overcome beta-lactamase mediated antibiotic resistance, beta-lactam antibiotic treatments are often supplemented by a beta-lactamase inhibitor [15][16][17] . These inhibitors compete with the drug for the same binding site, yet being resilient to degradation they block enzyme activity thereby restoring antibiotic efficacy. While class A beta-lactamase inhibitors, such as clavulanate or tazobactam , are commonly used, efficient inhibitors of the widespread class C beta-lactamases have only recently been introduced.
Class C beta-lactamases confer resistance to broad-spectrum cephalosporins, penicillins, and monobactams 18,19 . Like other beta-lactamases, they are highly transferable and widespread among gram-negative bacteria, especially in clinical isolates 20,21 . Class C beta-lactamases catalyze the hydrolysis of the beta-lactam ring by a conserved catalytic Serine (Ser 64 in Escherichia coli ampC ) and an activating conserved Tyrosine (Tyr 150) 19,[22][23][24] . A major obstacle to effective treatment of pathogens that carry class C beta-lactamases is the insensitivity of the enzyme to classical beta-lactamase inhibitors; the different structure of the pocket of class C enzymes compared to class A prevents the classical inhibitors from binding to its pocket 25,26 .
Avibactam is a novel non-beta-lactam molecule that inhibits the activity of class C as well as of other beta-lactamases [27][28][29] . It is resilient to class C beta-lactamase hydrolysis and inhibits their activity by binding the catalytic Serine covalently yet reversibly 28 . Avibactam can substantially decrease the resistance level of class C beta-lactamase-carrying bacteria to beta-lactam drugs 30 . Indeed, avibactam was recently approved for clinical use in combination with the cephalosporin ceftazidime and a combination of avibactam with the monobactam aztreonam is currently in clinical trials 31,32 . It is therefore important to understand the potential for evolutionary adaptation of the enzyme to combinations of avibactam with different beta-lactams.
Resistance to beta-lactamase inhibitors is typically associated with increased susceptibility to beta-lactam drugs, representing inherent constraints on adaptive mutations. While mutations that increase beta-lactamase expression can increase resistance to the drug alone and in combination with the inhibitor, resistance to the drug-inhibitor treatment via structural intragenic mutations is inherently constrained: due to the structural similarity between the drug and the inhibitor, mutations that increase drug degradation also tend to increase the affinity of the enzyme to the inhibitor 7,33-37 . Of course, this functional tradeoff between resistance to the drug and the inhibitor does not completely preclude mutations that increase resistance to the combination [38][39][40][41] . Indeed, even without escaping this tradeoff, a mutation providing strong resistance to one compound, even at the cost of mild sensitivity to the other, can provide an overall increased resistance to the combination 34,40,41 . Less is known though about mutations that escape this tradeoff, allowing the mutant to survive, even in the presence of the inhibitor, at drug concentrations that kill the wildtype in the absence of the inhibitor (native inhibitory concentration). In particular, it is unclear how prevalent are such escape mutations, how specific they are to different beta-lactams and whether they might be accessible via a single amino acid substitution.
Here, focusing on E. coli AmpC enzyme as a model for class C beta-lactamases, we constructed a systematic single amino acid substitution mutant library and identified resistance and escape mutations to avibactam paired with different beta-lactam drugs. Measuring growth of the mutants across gradients of different drugs, with and without avibactam, we found that escape mutations are accessible but only for some drugs and not others. These escape mutations allow bacteria to grow even at the presence of avibactam at the high drug concentrations required to kill the wildtype without avibactam. Sequencing the selected mutant library, we identified these escape mutations and found that they are rare and drug-specific.
To systematically study the tradeoffs between mutations that confer resistance to beta-lactam and beta-lactamase inhibitors, we constructed a library of mutants inside the beta-lactamase  To characterize the level of resistance that mutations in ampC can confer, we pooled all the MAGE mutants and selected the pooled library on gradients of five different antibiotics, representing all major beta-lactam families: the penicillins piperacillin (PIP) and ampicillin (AMP); the monobactam aztreonam (ATM); the cephalosporin cefepime (FEP); and the carbapenem meropenem (MER). Each drug was applied with or without the beta-lactamase inhibitor avibactam (AVI). Culture density was monitored over time for the mutant library as well as the parental strain (WT, expressing unmutated ampC ). As drug concentration increased, the cultures took longer to reach detectable OD, yet grew at similar rates (Fig. 2a). To characterize these time delays we measured t th , the time in which culture density crossed a set threshold, OD th . Next, for each drug concentration, we represented the measured delay by an estimated initial density of resistant mutants , where g is bacterial growth rate ( Fig. 2a; The initial density of resistant mutants OD 0 Res calculated based on t th , and plotted for the WT (black) and the MAGE library (grey) on gradients of four beta-lactam antibiotics individually (filled symbols) and when supplemented without AVI (no-fill symbols).
The drug concentration beyond which the η value of drops below a density OD 0 Res corresponding to a single initial mutant is determined (triangles on the bottom axis). The shift in the concentration required to inhibit the mutant library in the presence of AVI, compared to that required for the WT in the absence of AVI, is indicated by a purple arrow. A left-pointing arrow indicates that at the presence of AVI, no single mutant could grow at the drug-only MIC, while a right-pointing arrow is an indication for escape mutants.
Wildtype AmpC mediated resistance varied substantially for different drugs, as well as its potential for resistance improving mutations. Supplementing cultures with avibactam lowered the resistance of the wildtype to piperacillin, aztreonam, cefepime and ampicillin (Fig. 2b), but not to meropenem (supplementary fig. 4) indicating that active AmpC confers differential resistance to these drugs. Considering the phenotypes of the wildtype ampC and the MAGE mutant library without avibactam, we found that while no intragenic mutations in ampC led to increased ampicillin resistance, resistance to the other three drugs vastly improved (Fig. 2b, triangles). We next asked whether these mutations that increase resistance to piperacillin, aztreonam and cefepime in the absence of avibactam can possibly also escape the drug-inhibitor tradeoff.
Contrasting the drug concentration required to inhibit the mutant library in the presence of avibactam with the native inhibitory concentration, we find that escape mutations are accessible for some drugs and are not accessible in others. While no escape mutants are observed for piperacillin ( , left-pointing arrow, Fig. 2b ), we fitted the measured abundance data of each of the mutants across all drug concentrations with a dose-response model and determined the drug concentration that inhibits the mutant growth by 50% (Fig. 3a, Methods; Supplementary fig. 6).
Analyzing the inhibitory concentrations of each of the mutants, we found that resistance to different drugs is mediated through only partially overlapping sets of mutations. Determining the IC50 of all mutants ( ), we identified, for each of the drugs, residues for which multiple IC50 AB M ut beneficial substitutions exist (Fig. 3a, R148 for FEP; Y150 for ATM; E272 and N289 for PIP; S287 for both PIP and FEP). The diversity of beneficial substitutions for each of these residues indicates that these positions were maladapted (Fig. 3a). These residues were usually essential for AmpC activity on other drugs, possibly explaining their maladaptiveness (Fig. 3b). We further identified many mutations conferring high levels of resistance to both piperacillin and cefepime, but no cross high-resistance between aztreonam and the other drugs was found (Fig. 3b).
However, a single mutation, E272I, had a small effect that extended over the three drugs (Fig.   3b). To link the mutations that affect drug resistance to the enzyme structure and its active site, we computationally docked each of our compounds to a free AmpC enzyme (AutoDock Vina 45 ; Fig. 3c; Methods). We found that while substitutions of residues in loop H10 (Fig. 3c green, residues 287-296) to Glycine or Proline tend to confer resistance to both piperacillin and cefepime, they tend to have a negative effect on resistance to aztreonam (Fig. 3a). Such mutations are expected to destabilize the helix structure into a loop 46 and hence can make more space for piperacillin and cefepime that carry a large ring adjacent to the beta-lactam ring.
Similarly, we found multiple substitutions of the conserved Y150 residue sensitizing towards cefepime and piperacillin, yet conferring resistance to aztreonam. These resistance-conferring mutations were next examined for their ability to escape the drug-inhibitor tradeoff. acid is marked with an X and mutations which were not included in our design are marked in gray. Mutations which did not confer resistance to any drug were not enriched and could not be identified in our data are marked with a dot. The resistance level of the latter to all drugs is expected to be similar to the WT or smaller. b. The inhibitory concentration of each mutant is plotted for the different drugs to assess cross-resistance. WT inhibitory concentration is marked with a dashed line and a gray background represents our resolution as determined by drug dilution factor between measurements. c. The predicted conformation of the three antibiotics within the AmpC pocket. The antibiotics were docked using AutoDock Vina to the 3D crystal structure of AmpC beta-lactamase (PDB 1KVL) with the S64G replacement to emulate the WT physico-chemical environment. The enzyme is shown in ribbon, and regions that contain residues that affect resistance are colored in the same colors as in 3a-b. Residues for which multiple substitutions confer antibiotic resistance are highlighted and shown in stick representation. Antibiotics are colored coded by atoms (grey for C, red for O, blue for N, yellow for S and while for H).
Pairing our measurements of mutant inhibitory concentrations for drug-only and drug-inhibitor combinations, we characterized the drug-inhibitor resistance tradeoffs and identified escape mutations. For each mutant, we compared the changes in resistance to the drugs with and without avibactam relative to the wildtype ( Fig. 4; Supplementary fig. 7). As expected, mutants with increased resistance to the beta-lactam drugs were often more sensitive to avibactam and mutants with increased resistance to avibactam became more sensitive to the beta-lactam drugs (Fig. 4a-d). Mutants with increased resistance to both the drug and the drug-avibactam combination, also appeared, but commonly these mutants too were not able to grow at the native inhibitory concentration required to inhibit the wild-type in absence of avibactam ( Fig.  4b-d; orange shade with blue horizontal lines in Fig. 4a). Yet, importantly, some rare mutations did confer resistance to the drug beyond the native inhibitory concentration of the wildtype even 4a). In agreement with our phenotypic measurements (Fig. 2b), these escape mutants were only identified for aztreonam and cefepime and not for piperacillin. Moreover, even for cefepime and aztreonam, such mutations are rare and drug-specific. While substitutions of the conserved Y150 to non-aromatic positively charged amino acids (R, K), or small amino acids (G and A), as well as to S can escape the tradeoff between resistance to aztreonam and avibactam, they are sensitizing to cefepime and do not confer resistance to the cefepime-avibactam combination ( Fig. 4c-d; each such substitution appears in all accessible codons in all drug concentrations). Conversely, mutations that alter N346 to the large volume amino acids F, Y, W as well as the S237H and R148P substitutions can escape the tradeoff between resistance to avibactam and cefepime but are sensitizing to aztreonam and hence do not escape the avibactam-aztreonam combination (Fig. 4c-d). To validate and better characterize the escape phenotype, we isolated two of these mutants, Y150A and N346W, and challenged them on a 2-D gradient of avibactam and either aztreonam or cefepime respectively. Identifying the IC50 isobole, the line in concentration space where bacterial growth is inhibited by 50% (Fig. 4e-f; Methods), we found that these escape mutations indeed increase the resistance to the inhibitor-drug combination even above the high levels needed to kill the wildtype without avibactam.  Supplementary fig. 8). While weak escape phenotype appeared for ceftazidime, all other tested beta-lactams prevented substantial resistance to the drug-inhibitor combinations (Fig. 4g). The mutants that escaped the ceftazidime-avibactam tradeoff (R148N and N346P; Methods) were similar to mutations that escaped the tradeoff between avibactam and cefepime, which is also a cephalosporin (R148P and N346F/Y/W), suggesting fine tuning of resistance to the specific beta-lactam class.

Discussion
Systematically screening ampC beta-lactamase mutants for mutations that escape the tradeoff between resistance to beta-lactam drugs and beta-lactamase inhibitor, our measurements show that such escape mutations appear only for some drugs and are rare and drug-specific. In these evolutionary susceptible drug-inhibitor combinations, even a single amino-acid change in the beta-lactamase enzyme AmpC can confer resistance levels exceeding the native concentration that the wild-type can sustain without the inhibitor. Such escape mutations can potentially jeopardize treatment efficacy of bacterial pathogens by drug-inhibitor combinations. In contrast, no single amino acid mutants appear to escape the piperacillin-avibactam tradeoff. While most escape and resistance-conferring mutations were identified in a single avibactam concentration, the two escape mutants that were tested on a range of avibactam, N346W and Y150A, showed that these escape mutants enjoy an ability to grow in the native inhibitory concentration of the beta-lactam drug at a wide range of avibactam concentrations. While further investigations of the pharmacokinetic properties of the drugs and the inhibitor are required, our identification of drug-inhibitor combinations permissive and non-permissive to escape mutations can serve as basis for evolutionary-informed choice of combination drug treatments which minimize the potential for resistance.

Methods
Identifying the ampC catalytic pocket. Residues delineating the catalytic pocket of AmpC were selected by their distance from the ceftazidime drug bound to AmpC enzyme in the 3D crystal structure (PDB ID: 1IEL). We defined two layers according to their distance to the drug, residues within 5Å from the bound drug form the first layer as many of them serve as hotspot for substrate binding or catalytic activity, while more distant residues within a distance from 5Å to 7Å from the drug form the second layer, which are generally linked to the maintenance of the pocket 3D structure.   with the default scoring function. In the AutoDock Vina configuration files, the parameter num_modes was set to 20 and exhaustiveness to 48 to improve the searching space and accuracy. We identified the enzyme pocket based on the location of cephalothin on PDB ID 1KVL. We chose all the rotatable bonds in ligands to be flexible during the docking procedure, and we kept all the protein residues rigid. We assigned the Gasteiger atomic partial charges to the protein using the AutoDockTools package 48 . Antibiotics with their closed beta-lactam ring configuration were obtained from ZINC15 database 49  Data availability. The data sets generated and/or analysed during the current study are available from the corresponding author upon reasonable request.