OXA-48-Mediated Ceftazidime-Avibactam Resistance Is Associated with Evolutionary Trade-Offs.

The recent introduction of novel β-lactam/β-lactamase inhibitor combinations like ceftazidime-avibactam has increased our ability to treat infections caused by multidrug-resistant Gram-negative bacteria, including carbapenemase-producing Enterobacterales. However, the increasing number of cases of reported resistance to ceftazidime-avibactam is a concern. OXA-48 is a carbapenemase that has no significant effect on ceftazidime, but is inhibited by avibactam. Since isolates with OXA-48 frequently harbor extended-spectrum β-lactamases that are inhibited by avibactam, it is likely that ceftazidime-avibactam will be used to treat infections caused by OXA-48-producing Enterobacterales. Our data show that exposure to ceftazidime-avibactam can lead to changes in OXA-48, resulting in increased ability to hydrolyze ceftazidime and withstand the inhibitory effect of avibactam. Thus, resistance toward ceftazidime-avibactam among OXA-48-producing Enterobacterales should be monitored. Interestingly, the compromising effect of the amino acid substitutions in OXA-48 on other β-lactams and the effect of ceftazidime-avibactam exposure on the epidemic OXA-48 plasmid indicate that the evolution of ceftazidime-avibactam resistance comes with collateral effects.

To investigate the evolutionary implication of CAZ and CAZ-AVI treatment on OXA-48-producing isolates, we have studied the effects on both the bla OXA-48 gene and the epidemic plasmid associated with dissemination of OXA-48. Here, we report the occurrence of single (OXA-48:P68A) and double (OXA-48:P68A,Y211S) amino acid substitutions within OXA-48 as a response to CAZ and CAZ-AVI exposure, respectively. OXA-48:P68A demonstrated increased MICs toward CAZ, and OXA-48:P68A,Y211S showed increased MICs against both CAZ and CAZ-AVI. X-ray crystallography structures revealed that OXA-48:P68A leads to increased flexibility within the OXA-48 structure, likely contributing to elevated CAZ hydrolysis. Molecular modeling of OXA-48: P68A,Y211S showed an altered H-bond network due to Y211S. The alteration of this network is likely to affect the enzyme stability and confer higher CAZ resistance. In addition, Y211 in OXA-48 stabilizes AVI binding by aromatic stacking. In OXA-48: P68A,Y211S, the loss of this interaction might contribute to the reduced inhibitory activity of AVI. However, development of resistance toward CAZ and CAZ-AVI led to evolutionary trade-offs where (i) amino acid substitutions in OXA-48 compromised its carbapenemase and penicillinase activities and (ii) plasmid adaptation to CAZ and CAZ-AVI conferred a significant fitness cost as well as loss of stability.

Selection of mutants with decreased susceptibility toward CAZ and CAZ-AVI.
To investigate how the clinical use of CAZ and CAZ-AVI influences the evolution of OXA-48, as well as the plasmid carrying OXA-48, we initially subjected a clinical E. coli strain previously known to carry bla OXA-48 on a conjugative IncL plasmid (34,35) to PacBio sequencing. The genomic data revealed that the clinical strain harbored bla OXA-48 on a 65,499-bp IncL plasmid (p50579417_3_OXA-48; GenBank accession no. CP033880) with no other resistance genes (see Fig. S1 and reference 73). The plasmid was closely related to the epidemic IncL OXA-48 plasmid (40) and other OXA-48 plasmids (Fig. S1). We subsequently transferred the plasmid by conjugation into rifampin-resistant E. coli TOP10, isolated the plasmid, and transformed it into E. coli MG1655 (MP100 [Table 1]). Due to the lack of cephalosporinase activity of OXA-48, the introduction of p50579417_3_OXA-48 in E. coli MG1655 (MP101) did not result in any change in the MICs toward CAZ and CAZ-AVI (Table 2). MP101 was subsequently subjected to a two-step mutant selection regimen, using increasing concentrations of CAZ and CAZ-AVI up to 64ϫ the MIC of MP101. Analysis of selected mutants after CAZ (MP102) and CAZ-AVI (MP103) exposure showed that the MICs toward both CAZ and CAZ-AVI increased irrespectively of selection regimen (Table 2). With CAZ selection, the CAZ MIC increased 128-fold and the CAZ-AVI MIC increased 16-fold. Selection with CAZ-AVI resulted in a 128-fold increase in the MICs for both CAZ and CAZ-AVI (Table 2).   (Table 3). No change in the MIC toward CAZ-AVI was observed. For the P68A,Y211S (MP106), the MIC toward CAZ was increased by 32-fold and that toward CAZ-AVI was increased by 4-fold.
Changes in OXA-2 and KPC-2/3, leading to CAZ or CAZ-AVI resistance, respectively, have been shown to come along with functional constraints (20,26,29,31). Therefore, we performed MIC testing against a panel of ␤-lactams. Both the OXA-48:P68A and P68A,Y211S substitutions caused altered effects toward other ␤-lactams compared to the native OXA-48 ( Table 3). The effect against carbapenems was equal for both the single and double amino acid substitution, with a 4-to 16-fold decrease in the MIC against meropenem, imipenem, and ertapenem (Table 3). No change in MIC toward doripenem was observed. Moreover, expression of both mutants of OXA-48 resulted in reduced activity against piperacillin-tazobactam (32-fold MIC decrease) and temocillin (4-to 8-fold MIC decrease). For other ␤-lactams, including other cephalosporins such as cefepime and cefotaxime, the changes were within a 2-fold dilution step (nonsignificant).
Thermostability and enzyme kinetics of OXA-48:P68A and OXA-48: P68A,Y211S. OXA-48, OXA-48:P68A, and OXA-48:P68A,Y211S were expressed and purified (Ͼ95% purity). From an initial starting culture of 1 liter, yields of 23.9, 9.9, and 16.7 mg, respectively, were obtained. P68A and P68A,Y211S caused a reduction in the thermostability of OXA-48, with melting temperatures of 49.5 Ϯ 0.1°C and 46.7 Ϯ 0.2°C, respectively, compared to 53.5 Ϯ 0.1°C for the native OXA-48. We further determined the effect of the amino acid changes, with respect to both the hydrolytic activity against ␤-lactams as well as the inhibitory effect (50% inhibitory concentration [IC 50 ]) of AVI and tazobactam. Compared to native OXA-48, P68A and P68A,Y211S caused Ͼ10-fold and Ͼ20-fold increased catalytic efficiency (k cat /K m ) toward CAZ, respectively (Table 4). In addition, both substitutions conferred reduced hydrolytic activity toward penicillins (ampicillin and piperacillin) and carbapenems (imipenem and meropenem [ Table 4]). The reduced catalytic efficiency varied from 2-to ϳ600-fold compared to the native OXA-48. No change was observed for the cephalosporin cefepime. For OXA-48:P68A, we observed no change in the inhibitory activity of AVI (Table 5). However, the double substitution P68A,Y211S resulted in a Ͼ5-fold decrease in inhibitory activity of AVI. Moreover, both the P68A and P68A,Y211S substitutions led to an increased inhibitory activity of tazobactam with Ͼ20-fold and Ͼ3-fold reductions in the IC 50 compared to the native OXA-48 (Table 5).
P68A and P68A,Y211S increase flexibility of the active site. Two new crystal structures of OXA-48:P68A were obtained after soaking (see Table S1 in the supplemental material). Both structures displayed four molecules in the asymmetric unit arranged in two dimers (chains A to D). In the first OXA-48:P68A structure (resolved to 2.50 Å), we found CAZ bound to chain A/C ( CAZ binding forces R214 in the ␤5-␤6 loop to move out of the active site. As a result, R214 would then clash with parts of the ⍀-loop (D143 to S165) in the native OXA-48 ( Fig. 1B). Indeed, we found residues 149 to 161 (for both chains A and C) in the ⍀-loop of OXA-48:P68A to be disordered (Fig. 1B) and thus not visible in the electron density maps. Therefore, P68A seems to impose increased flexibility within OXA-48, and this enables both the ␤5-␤6 loop and ⍀-loop to occupy alternative conformations and a Errors are displayed as 95% confidence intervals based on a minimum of triplicates. b AMP, ampicillin; PIP, piperacillin; NIT, nitrocefin; CAZ, ceftazidime; FEP, cefepime; IMI, imipenem; MEM, meropenem.  consequently favors CAZ binding. Interestingly, the ␤5-␤6-loop region has been shown to be relevant for the carbapenemase versus ceftazidimase activity of OXA-48 (32,41). Furthermore, we found AVI covalently bound to S70 in all OXA-48:P68A-AVI chains. Superimposition of OXA-48:P68A-AVI with the structure of native OXA-48 in a complex with AVI (PDB no. 4S2K) (11) showed no difference in interaction with AVI (Fig. 1C). R214 and the ⍀-loop were in the same conformation as in the native OXA-48. Crystallization of OXA-48:P68A,Y211S was unsuccessful; therefore, we used molecular modeling and superimposed the obtained structure onto the structure of OXA-48:P68A-CAZ. In native OXA-48 (PDB no. 4S2P) and OXA-48:P68A, the Y211 side chain is part of a water-mediated H-bond network, including T234, P235, and M236, where the water molecule is centrally coordinated (Fig. 1D). In contrast, the formation of this H-bond network seems not feasible with serine at position 211. In principle, this could lead to a higher flexibility within the structure of OXA-48:P68A,Y211S and could be related to the ϳ4°C reduced thermostability, as well as to elevated CAZ hydrolysis, compared to OXA-48:P68A.
Effect of plasmid adaptations on resistance and host fitness. In order to investigate the effect of selection on the plasmid backbone, we isolated and transformed the plasmids after CAZ (p50579417_3_OXA-48-CAZ) and CAZ/AVI (p50579417_3_OXA-48-CAZ-AVI) selection into MP100. MIC determination of the resulting strains MP107 and MP108 mirrored the fold changes observed in the cloned expression vector system ( Table 6; see Fig. S2 in the supplemental material). This indicates that the P68A and P68A,Y211S in OXA-48 are the main contributors to the altered susceptibility profiles. To further investigate the role of the plasmid during resistance development, we measured the fitness costs of MP101, MP107, and MP108 in head-to-head competition assays. The native OXA-48 plasmid did not significantly reduce fitness (relative fitness [w] ϭ 1.01, P ϭ 0.509, df ϭ 6) in MP101 relative to the plasmid-free strain MP100. However, after CAZ-AVI adaptation, MP103 showed severely reduced fitness, by 18% (w ϭ 0.82, P ϭ 0.008, df ϭ 3) relative to the plasmid-free strain MP100 (Fig. 2).
In an attempt to isolate the fitness effects conferred by the plasmids after adaptation alone, we competed MP107 and MP108 with the native OXA-48 plasmid- carrying strain MP101. Both adapted plasmids significantly reduced fitness, by 14% (p50579417_3_OXA-48-CAZ; P ϭ 0.003, df ϭ 9) and 3% (p50579417_3_OXA-48-CAZ-AVI; P ϭ 0.026, df ϭ 6), respectively. Sequencing of the plasmids after selection revealed mutational differences on the plasmid backbone. Different point mutations in repA, 79G¡T (p50579417_3_OXA-48-CAZ) and 79G¡T (p50579417_3_OXA-48-CAZ-AVI) (42), as well as a 62-bp deletion in the DNA binding side of a xenobiotic response element (transcriptional regulator) in the p50579417_3_OXA-48-CAZ plasmid were observed. To investigate the effect of CAZ-AVI exposure on OXA-48 plasmid stability, we serially passaged MP101 harboring the native OXA-48 plasmid and the CAZ-AVIexposed strain MP103 for 350 generations under nonselective conditions. Consistent with the fitness data, the native OXA-48 plasmid was stably maintained throughout the experiment, whereas loss of p50579417_3_OXA-48-CAZ-AVI was observed after ϳ100 generations. The plasmid was completely lost from the population after ϳ250 generations (Fig. 3).

DISCUSSION
Treatment options for infections caused by carbapenemase-producing Gramnegative bacteria are limited, and the increased spread of these multidrug-resistant bacteria is considered a public health threat (1,2). Several novel ␤-lactamase inhibitors, including inhibitors of carbapenemases, have recently been introduced to resurrect the activity of existing ␤-lactams (43,44). This includes AVI, a novel diazabicyclooctane ␤-lactamase inhibitor able to inhibit class A and class D carbapenemases, such as KPC and OXA-48, respectively. Thus, the introduction of the combination CAZ-AVI offers a possible treatment option for class A and class D carbapenemase-producing Enterobacterales (16). Unfortunately, several reports are now describing the emergence of CAZ-AVI-resistant Enterobacterales, particularly KPC-producing K. pneumoniae (22,23,30). In contrast to carbapenemases such as KPC, the OXA-48 carbapenemase has an insignificant hydrolysis activity toward third-generation cephalosporins, including CAZ. However, the CAZ-AVI combination is a promising treatment option since OXA-48-  Table 1. producing isolates frequently also carry an ESBL (12,45,46). Thus, we investigated the effect of CAZ and CAZ-AVI on the evolution of OXA-48 and an OXA-48 plasmid closely related to the epidemic IncL plasmid frequently associated with bla OXA-48 (39,40).
Our results show that both CAZ and CAZ-AVI exposure resulted in mutations in bla OXA-48 that increased the ability of OXA-48 to hydrolyze CAZ. After CAZ-AVI exposure, we also identified a reduced inhibitory effect of AVI. Moreover, development of OXA-48 resistance toward CAZ/CAZ-AVI conferred collateral sensitivity where the carbapenem and penicillin activities of the enzyme were reduced. Several studies have reported OXA-48-like variants (OXA-163, OXA-247, and OXA-405) with increased activity toward extended-spectrum cephalosporins (36)(37)(38). These variants carry amino acid deletions/substitutions within and around the ␤5-␤6 loop also leading to reduced carbapenemase activity. OXA-247 carries the identified Y211S substitution in addition to a 4-amino-acid deletion (36). From these studies, it was unclear, however, whether this evolutionary trajectory could have been triggered by the exposure to extendedspectrum cephalosporins. We show that CAZ and CAZ-AVI have the potential to affect the evolution of OXA-48 and that the substitutions P68A and P68A,Y211S re-evolve the function of OXA-48 and specialize the enzyme toward ceftazidime hydrolysis. These results agree with other studies in which the adaptive changes in enzymes caused collateral effects, compromising or reversing the original function (20,29,31,47,48). While Y211 in OXA-48 is located at the ␤5-␤6 loop and is involved in AVI binding by the formation of an oxyanion hole (11,32), as well as in aromatic stacking with the side chain (ϳ4.3 Å), the role of P68 in the beginning of the ␣3 helix is poorly understood. In silico modeling of OXA variants identified loops flanking the ␣3 helix to play an important role for the carbapenemase or CAZ hydrolysis activity in those variants. In general, increased stability in these loop regions was also correlated with higher affinity to carbapenems, whereby more flexible loops revealed their catalytic proficiency toward CAZ (41).
The X-ray structure of OXA-48:P68A with four chains in the asymmetric unit revealed that CAZ binding requires higher structural flexibility. Generally, CAZ binding resulted in a displacement of the ␤5-␤6 loop carrying R214, as well as a disordered ⍀-loop (Fig. 1B). Docking experiments in OXA-48 have shown that CAZ hydrolysis is mechanistically unfeasible. This is mainly due to the high rigidity and the small active-site cavity confined by R214, as well as the length of the ␤5-␤6 loop (41,49). In contrast, a strongly shortened ␤5-␤6 loop and the absence of R214, as observed in OXA-163, expand the active-site cavity and allow OXA-163 to hydrolyze CAZ more efficiently (49). This supports the hypothesis that P68A increases the flexibility and changes the plasticity of the substrate binding site in OXA-48, allowing the hydrolysis of bulkier  Table 1). The native plasmid (p50579417_3_OXA-48) was stably maintained in MP101 over 350 generations without selection pressure. In the strain subjected to CAZ-AVI, MP103, the adapted plasmid (p50579417_3_OXA-48-CAZ-AVI) was purged out of the population within 350 generations. drugs such as CAZ. Decreased rigidity is also supported by a reduction in thermostability of ϳ4°C, compared to native OXA-48. Interestingly, OXA-163 (as well as OXA-247 and OXA-405) displayed collateral sensitivity toward carbapenems (36)(37)(38)50) but also cross-resistance toward some cephalosporins (e.g., cefotaxime). Molecular docking of OXA-163 suggests that cefotaxime is more embedded within the active site (49). However, cross-resistance toward other cephalosporins than CAZ was not observed for the OXA-48:P68A variant. Depending on their R2 side chain, cephalosporins display different binding behavior in metallo-␤-lactamases (51). While OXA-48:P68A allows CAZ binding by a displacement of the ␤5-␤6 and ⍀-loops, we hypothesize that for other cephalosporins, the active site of OXA-48:P68A might still be too narrow and rigid to achieve significant hydrolysis.
The double mutant carrying P68A,Y211S conferred a high level of resistance toward CAZ and elevated MIC levels against CAZ-AVI. Furthermore, P68A,Y211S caused a 20-fold increase in CAZ hydrolysis and a 5-fold reduction in the inhibitory activity of AVI, compared to native OXA-48. Therefore, our data suggest that OXA-48-mediated development of resistance toward CAZ-AVI is due to both increased enzymatic hydrolysis toward CAZ and reduced inhibitory activity of AVI. Molecular modeling of OXA-48: P68A,Y211S suggested the alteration of a water-mediated H-bond network between Y211 OH and T234 O and P235 O and M236 O (Fig. 1D). In OXA-48:P68A,Y211S, S211 is unlikely to be part of the same H-bond network due to its less-space-filling properties. The loss of H-bonds can increase enzyme flexibility; however, this is usually accompanied by decreased enzyme stability (52). Indeed, we found OXA-48:P68A,Y211S to be ϳ7°C and ϳ4°C less thermostable compared to native OXA-48 and OXA-48:P68A. We therefore believe that P68A,Y211S in OXA-48 further increases the flexibility of the active site, resulting in increased CAZ hydrolysis. In addition, S211 contributes to the reduced inhibitory activity of AVI by the loss of aromatic stacking, which stabilizes AVI in OXA-48 and OXA-48:P68A.
In addition, clinically relevant reversion of piperacillin-tazobactam resistance was shown for both the single and double OXA-48 mutants. Enzyme kinetics also revealed a double effect where the OXA-48 variants demonstrate reduced piperacillin hydrolysis activity and stronger inhibition by tazobactam. Similar effects were shown for CAZ-AVI-mediated mutations within KPC-2 and KPC-3 (29)(30)(31). Resensitization through exploitation of such evolutionary trade-offs could in principle provide the basis for alternative treatment strategies that potentiate the activities of earlier-generation ␤-lactams, and several strategies have recently been proposed (53)(54)(55)(56).
Since the dissemination of bla OXA-48 is partially linked to closely related IncL plasmids (39,40), we wanted to investigate the effect of CAZ and CAZ-AVI exposure on p50579417_3_OXA-48. Interestingly, the adapted plasmids showed a significant fitness cost compared to the native plasmid (Fig. 2), as well as reduced plasmid stability (p50579417_3_OXA-48-CAZ-AVI) in the absence of ␤-lactam selection (Fig. 3). Sequencing of the CAZ-and CAZ-AVI-exposed plasmids displayed mutations in a regulatory region (repA) known to be involved in plasmid copy number control (42,57). Upregulation of plasmid copy number has been shown previously to correlate with both decreased bacterial fitness and increased drug resistance (58). These data suggest that P68A and P68A,Y211S mutations, alone or in combination with putative copy number changes, might negatively impact the fitness of the adapted plasmids.
Mutations in bla KPC-2/3 conferring CAZ-AVI resistance have been observed in the clinical setting (19,22,28,30,59). However, to the best of our knowledge, OXA-48mediated reduced susceptibility to CAZ-AVI has not yet been reported in clinical settings (60). This might be due to the evolutionary trade-offs and reduced fitness reported here, which in principle may limit the occurrence and spread of resistance. Taken together, our data suggest that CAZ and CAZ-AVI conferred collateral sensitivity effects where the enzyme compromised its original carbapenemase and penicillinase activity. In principle, these evolutionary constraints and resensitizations can be exploited in order to improve future treatment protocols.
PacBio sequencing of E. coli 50579417. Genomic DNA of E. coli 50579417 was prepared from an overnight culture using the GenElute bacterial genomic DNA kit (Sigma-Aldrich, St. Louis, MO) according to the manufacturer's instructions. The DNA library was prepared using the Pacific Biosciences 20-kb library preparation protocol and size selection with a 9-kb cutoff using BluePippin (Sage Sciences, Beverly, MA). Sequencing was performed using the Pacific Biosciences RSII instrument using P6-C4 chemistry with a 360-min movie time and one single-molecule real-time sequencing (SMRT) cell. The sequences were assembled and polished at The Norwegian Sequencing Centre (https://www.sequencing .uio.no/) using HGAP v3 (Pacific Biosciences, SMRT Analysis Software v.2.3.0). Minimus2 from AMOS was used to circularize contigs, and RS_Resequencing.1 software (Pacific Biosciences, SMRT Analysis Software v.2.3.0) was used for correction of bases after circularization.
Selection of mutants with increased CAZ and CAZ-AVI MIC. Ten milliliters of an MP101 culture was grown in MH broth at 37°C overnight and then centrifuged for 10 min at 4,000 ϫ g, and the pellet was suspended in 1 ml MH broth. One hundred microliters was plated on MH agar containing increasing concentrations of either CAZ alone or in combination with AVI up to 2 mg/liter. For CAZ-AVI, we utilized the clinical ratio of 4:1, respectively. Colonies growing on the highest concentration were recovered and grown in a fresh overnight culture (MH broth) and subsequently plated on concentrations of CAZ or CAZ-AVI up to 32 mg/liter. Antibiotic susceptibility testing. For MIC determination, single colonies were suspended in 0.9% saline buffer to a 0.5 McFarland standard and further diluted 1:100 in MH broth. Fifty microliters of the bacterial suspension was loaded onto in-house-designed and premade Sensititre microtiter plates (TREK Diagnostic Systems/Thermo Fisher Scientific, East Grinstead, United Kingdom). The plates were incubated for 20 h at 37°C. Antibiotic susceptibility testing was performed in duplicates.
For a functional resistance profile, the native and mutated bla OXA-48 genes were cloned in the pCR-blunt II-TOPO vector (Invitrogen, CA) and expressed in E. coli TOP10 (Invitrogen, CA). The PCR product was obtained by Phusion High Fidelity PCR mastermix with High Fidelity buffer (New England Biolabs, MA) and preOXA-48 primers (61). Transformants were selected on LB agar plates containing 50 or 100 mg/liter ampicillin. Insertion size was verified by Sanger sequencing using M13 forward (5=-GTA AAACGACGGCCAG-3=) and reverse (5=-CAGGAAACAGCTATGAC-3=) primers.
Recombinant enzyme construction, expression, and purification. In order to construct OXA-48: P68A and OXA-48:P68A,Y211S, site-directed mutagenesis of bla OXA-48 in a pDEST17 vector was performed using the QuikChange II site-directed mutagenesis kit (Agilent Biosciences, Santa Clara, CA) (64,65). XL1-Blue competent cells were heat shock transformed with the constructed DNA. Point mutations were verified by Sanger sequencing using T7 primers, as described above. OXA-48 expression and purification were done as described previously (51,52). For the mutants, expression was performed in Rosetta 2(DE3)/pLysS. In general, cultures were grown to log phase in Terrific broth supplemented with ampicillin (100 mg/liter) at 37°C and 180 rpm. Enzyme expression was induced with 0.1 mM IPTG (isopropyl-␤-Dthiogalactopyranoside) and performed at 15°C and 180 rpm overnight. Harvested cells were sonicated, and recombinant proteins were purified as described previously (64).
Thermostability. Fluorescence-based protein thermostability was determined for OXA-48, OXA-48: P68A, and OXA-48:P68A,Y211S in an MJ minicycler (Bio-Rad) across a temperature gradient of 25 to 60°C (at a heating rate of 1°C per min). Thermostability was determined in 50 mM HEPES (VWR, PA) at pH 7.5 supplemented with 50 mM potassium sulfate (Honeywell, NC) using 0.2 mg/ml protein and 5ϫ SYPRO orange (Sigma-Aldrich, St. Louis, MO). The excitation and emission wavelengths of SYPRO orange are 470 and 570 nm, respectively. The melting temperatures were determined as the inflection point of the melting transition found from the first derivative. All experiments were performed in triplicates.
Crystallization, structure determination, and molecular modeling. OXA-48:P68A was crystallized by the sitting-drop method in 22% to 26% polyethylene glycol monomethyl ether 5000 (Sigma-Aldrich, St. Louis, MO) and 0.1 M BIS-Tris-propane buffer (Sigma-Aldrich, St. Louis, MO) at pH 6.5 to 7.5 at 4°C. Crystals were soaked for some seconds in CAZ (saturated) or AVI (saturated) in cryoprotector, which was in mother lipid and 10% ethylene glycol (Sigma-Aldrich, St. Louis, MO, USA), followed by being flash-cooled in liquid nitrogen. Diffraction data were collected on BL14.1 and BL14.2 BESSY II, Berlin, Germany, at 100 K at a wavelength of 0.9184 Å, and the diffraction images were indexed and integrated using XDS (66). AIMLESS was used for scaling (67). During scaling, the final data sets were carefully inspected (Table S1), where we aimed for high completeness: a CC 1/2 of Ͼ0.5 in the outer resolution shell and a mean above 1.0. Both structures were solved by molecular replacement with chain A of PDB no. 5QB4 (68) and the program Phenix 1.12 (69). Parts of the models were rebuilt using Coot (70). Plasmid sequencing. Plasmid DNA after CAZ and CAZ-AVI exposure was isolated as described above, and fragment libraries were constructed by using the Nextera kit (Illumina, Little Chesterford, United Kingdom) followed by 251-bp paired-end sequencing (MiSeq, Illumina). This was done according to the manufacturer's instructions. Paired-end sequence data were assembled using CLCbio's Genomics Workbench 8.0 (Qiagen, Aarhus, Denmark). Sequences were aligned against the native OXA-48 plasmid, and mutations were identified using DNASTAR (DNASTAR, Madison, WI).
Plasmid stability. MP101 and MP103 were evolved under nonselective conditions in MH broth for 350 generations. Initially, 990 l MH broth was inoculated with 10 l of an overnight culture and incubated for 12 h at 37°C and 700 rpm. Every 12 h, 10 l of the culture was transferred into 990 l MH broth. Plasmid stability was tested after 0, 50, 100, 125, 140, 155, 165, 250, and 350 generations. For each time point, an overnight culture was diluted 10 Ϫ4 to 10 Ϫ6 in 0.9% saline. One hundred microliters of each concentration was plated on MH agar and incubated overnight at 37°C. Subsequently, 100 single colonies were picked and streaked on MH agar containing 100 mg/liter ampicillin.
Data availability. Atom coordinates and structure factors for OXA-48:P68A and OXA-48:P68A-AVI have been deposited in the Protein Data Bank (PDB no. 6Q5F and 6Q5B, respectively). The plasmid sequence data (p50579417_3_OXA-48) are available in GenBank under accession no. CP033880. All other relevant data are available within this article, the supplemental material, or from the corresponding author upon request.

ACKNOWLEDGMENTS
We are very grateful to Jürgen Brem, Department of Chemistry, University of Oxford, United Kingdom, for providing the avibactam used in this study, as well as Linus Sandegren, Uppsala University, Sweden, for providing E. coli MG1655 (DA4201). Provision of beam time at BL14.1 and BL14.2 at Bessy II, Berlin, Germany, is highly valued. We thank Maria Chiara di Luca for sequencing E. coli 50579417, as well as Nicole Podnecky and Joao Pedro Alves Gama for their support in the analysis of the next-generation sequencing data. The PacBio sequencing was performed at the Norwegian Sequencing Centre.
Pål Jarle Johnsen was supported by Northern Norway Regional Health Authority, UiT-The Arctic University of Norway (project SFP1292- 16 The authors declare no competing financial interest.