Novel continuous experimental evolution methodology uncovers rapid resistance development and cross-resistance

TL;DR

A new method for studying antibiotic resistance shows how bacteria quickly develop resistance and cross-resistance to multiple drugs.

Contribution

A novel continuous experimental evolution system was developed to study rapid resistance and cross-resistance.

Findings

Resistant mutants emerged in all three Enterobacter cloacae complex strains within four days of cefepime exposure.

One mutant showed cross-resistance to cefiderocol and ceftazidime-avibactam due to a mutation in the blaMIR-11 gene.

Abstract

Antimicrobial resistance poses a significant global health threat. Experimental evolution studies are crucial in understanding resistance mechanisms and thereby informing strategies to preserve antibiotic efficacy. We developed a novel continuous experimental evolution system enabling uninterrupted medium exchange with a rising antibiotic gradient, using standard laboratory equipment. We applied this system to three Enterobacter cloacae complex strains isolated from urinary tract infections in Germany between 1990 and 1992, which therefore had no prior exposure to cefepime, a fourth-generation cephalosporin approved in Germany in 2004. After four days of exposure to a cefepime gradient, resistant mutants emerged in all three strains. Notably, one mutant exhibited cross-resistance to the novel antibiotics cefiderocol and ceftazidime-avibactam, due to a single missense mutation in the β-lactamase gene blaMIR-11. Our study demonstrates the effectiveness of this novel approach for investigating antimicrobial resistance development and cross-resistance mechanisms, as well as identifying and characterizing a mutation attributed to major cross-resistance.

Figures

Schematic overview of the continuous experimental evolution system.A medium reservoir containing medium with appropriate antibiotic concentration (room temperature) supplies an interim reservoir (37 °C) and thereby constantly ramps the antibiotic concentration in this reservoir. The constantly stirred medium from the interim reservoir is then in turn pumped into the multiple culture flasks, and old medium is removed from the culture flasks to the waste with the same flow rate. Bacteria are cultivated in the culture flasks under constant agitation and at 37 °C. Created in BioRender.com.

Visualization of the minimum inhibitory concentrations fold change for all replicates included in the continuous experimental evolution and schematic overview of the mutation in *bla*_MIR-11_ of the cefiderocol-resistant PBIO4880 evolvant.A Fold change of minimum inhibitory concentrations of evolvant strains compared to respective parent strains based on values shown in Table [2](#Tab2). The resistance profile of the evolvant strains is shown by differing shapes. Fold changes are plotted as their minimum value (i.e., ≥16 is plotted as 16). PBIO4886 replicate 1, and all three replicates of PBIO

Full text

Introduction

Antimicrobial resistance (AMR) is the cause of one of the most critical global health crises of the 21^st^ century. In 2019, bacterial AMR was estimated to be directly responsible for more than 1 million deaths, threatening to undermine the foundations of modern medicine if new drugs and effective mitigation strategies are not rapidly developed and implemented^1^. This challenge is particularly acute with multidrug-resistant (MDR) Gram-negative bacteria, which can possess sophisticated mechanisms to evade nearly all available antimicrobial agents. Shown for example in the case of the recently described cefiderocol (FDC) resistance of a convergent Klebsiella pneumoniae strain isolated as part of a clinical outbreak^2^.

A pathogen of increasing concern in the context of AMR is the Enterobacter cloacae complex (ECC), a leading cause of nosocomial infections, including life-threatening bloodstream infections and pneumonia^3,4^. ECC strains are notorious for rapidly acquiring resistance, often through overexpression or mutation of their chromosomal AmpC-type β-lactamases and the acquisition of resistance determinants on mobile genetic elements^5^, resulting in MDR phenotypes. To treat these recalcitrant infections, clinicians increasingly rely on last-resort antibiotics such as FDC (a siderophore cephalosporin) or the combination drug ceftazidime-avibactam (CAZ-AVI; a β-lactam-β-lactamase inhibitor combination). However, the emergence of resistance and, critically, cross-resistance to these key agents is now a major clinical threat, demanding urgent investigation into the evolutionary pathways involved.

Despite the urgent need to predict resistance development, current experimental evolution (EE) studies often use discontinuous, batch-culture methods^6^. These methods impose abrupt and episodic selection pressures that do not replicate the gradual, continuous antibiotic concentration gradients encountered by pathogens in vivo during patient therapy or sub-therapeutic dosing^7,8^. This methodological limitation prevents accurate assessment of the true speed of resistance evolution and obscures the fine-grained genetic trajectories that promote cross-resistance to multiple critical antibiotics simultaneously^9^.

To address this knowledge gap, we developed a novel continuous EE system using standard laboratory equipment to maintain uninterrupted medium exchange with a precise, incrementally rising antibiotic gradient. This platform mimics the prolonged, low-concentration selection pressure characteristic of clinical treatment failure, providing a more biologically relevant model for AMR research compared to classical transfer approaches.

In this study, we applied this continuous flow-through system with an increasing cefepime (CEF) gradient to three clinical ECC strains, which were isolated prior to the widespread use of CEF (between 1990 and 1992). This unique opportunity allowed us to study how resistance evolves during a bacterium’s first exposure to this antibiotic. We comprehensively investigated the genomic and phenotypic mechanisms driving the observed resistance and investigated cross-resistance to the last-resort drugs FDC and CAZ-AVI. This proof-of-principle study demonstrates an affordable, reproducible and simple methodology for future high-throughput AMR studies.

Results

Development of a continuous experimental evolution setup

The development of this novel continuous EE methodology focused on three main points compared to previously established EE methods using antibiotics as selection pressure: (i) circumventing population bottlenecks, (ii) improving accessibility and resource efficiency, and (iii) minimizing manual intervention during the experiment.

First, a significant disadvantage of discontinuous EE approaches, such as serial transfer, is the introduction of population bottlenecks and the resulting undesirable effects, such as genetic drift^10–12^. To eliminate these issues, we designed a continuous EE system that sustains uninterrupted bacterial growth. Our system maintains the bacterial population in a constant state of exponential growth by employing the simple principle of a constant influx of fresh medium and efflux of old medium. This maintains a steady culture volume and eliminates the need for serial passaging. The continuous flow was precisely achieved using standard laboratory peristaltic pumps.

Second, a limiting factor in the widespread adoption of continuous EE systems is the need for specialized equipment. Popular sophisticated methods such as the morbidostat or even bioreactor-based systems, for example, represent the gold standard for continuous EE studies but require non-standard and expensive components that are not typically available in most laboratories^9^. To ensure accessibility, we based our system solely on standard laboratory equipment. Additionally, to improve resource efficiency, we adapted the system to allow parallel cultivation of multiple replicates. The system consists of two peristaltic pumps, a medium reservoir, a waste reservoir, several culture flasks, and a multi-neck bottle serving as an interim reservoir. The culture flasks and interim reservoir were kept in an incubator at 37 °C with constant agitation using magnetic stirrer plates. Fresh medium containing a fixed antibiotic concentration, kept at room temperature, is continuously added to the interim reservoir using the first pump at a steady flow rate (V̇_1_), resulting in a steadily increasing antibiotic concentration over time. The rising antibiotic concentration gradient was chosen independently of the used strains and their specific doubling time, to represent the clinical context as it would be expected during treatment. The second pump distributes this medium to all culture flasks and removes old medium to the waste reservoir with an identical flow rate (V̇_2_), thus keeping the culture volume constant. With a V̇_1_ of 0.45 mL/min, the antibiotic concentration in the culture flasks approximately matches that of the medium reservoir after two days (Fig. 1). The medium reservoir can be replaced every 2 days, depending on the desired final concentration and the chosen rate of antibiotic increase.Fig. 1. Schematic overview of the continuous experimental evolution system.A medium reservoir containing medium with appropriate antibiotic concentration (room temperature) supplies an interim reservoir (37 °C) and thereby constantly ramps the antibiotic concentration in this reservoir. The constantly stirred medium from the interim reservoir is then in turn pumped into the multiple culture flasks, and old medium is removed from the culture flasks to the waste with the same flow rate. Bacteria are cultivated in the culture flasks under constant agitation and at 37 °C. Created in BioRender.com.

Finally, the design significantly reduces the experimental burden. The only manual inputs required during the entire EE run are the initial inoculation, periodic sampling of the cultures, and replacement of the medium reservoir (antibiotic refill). This minimal intervention, combined with the ability to evolve multiple replicates in parallel, establishes this method as a versatile, high-throughput, continuous EE platform.

As a proof-of-principle application, we used our novel continuous EE method to investigate the development of CEF resistance in three clinical strains belonging to the ECC. The three extended-spectrum β-lactamase-producing strains were originally isolated from urine samples at the University Hospital Schleswig-Holstein in Kiel (Germany) between 1990 and 1992 during routine diagnostics for urinary tract infections and have been cryopreserved since then. The isolation dates predate the medical approval of CEF (USA: 1996; EU: 2004), suggesting that the strains had no prior clinical exposure to this specific antibiotic.

Genomic analysis identified the three strains as E. roggenkampii sequence type (ST)165 (PBIO4880), E. hormaechei ST1131 (PBIO4886), and E. hormaechei ST108 (PBIO4914). Subsequent analysis using AMRFinderPlus revealed that all strains carried one class C serine β-lactamase, which typically confers resistance to β-lactam antibiotics such as penicillins and third-generation cephalosporins. PBIO4880 additionally carried the glutathione-transferase fosA, conferring fosfomycin resistance. We further determined the plasmid content: PBIO4886 was plasmid-free, while the other two strains (PBIO4880 and PBIO4914) each carried three plasmids (Table 1). Phenotypic antimicrobial susceptibility testing using the VITEK-2 COMPACT platform confirmed an almost identical resistance pattern against penicillins and third-generation cephalosporins. Crucially for the EE study, all three strains were confirmed as susceptible to CEF and according to EUCAST guidelines^13^, with minimum inhibitory concentrations (MICs) of up to 2 µg/mL (Supplementary Table S1).Table 1. Overview of the main characteristics of the tested strainsPBIO4880PBIO4886PBIO4914SpeciesE. roggenkampiiE. hormaecheiE. hormaecheiSTST165ST1131ST108OriginurineurineurineYear of sampling199019911992Resistance genesblaMIR-11, fosA**blaACT-16bla_ACT-55_No. of plasmids303All three strains belong to the Enterobacter cloacae complex and were isolated from urinary tract infections in a German tertiary care hospital between 1990 and 1992. Resistance genes were predicted using AMRFinderPlus.

Continuous experimental evolution revealed resistance mechanisms to cefepime and cross-resistance to cefiderocol

For this proof-of-principle study, three replicates of each of the three ECC strains were cultured in parallel. Inoculation cultures served as the parent strains for subsequent comparative analyses. To achieve a final CEF concentration of 16 µg/mL, which is equivalent to twice the EUCAST breakpoint, the system used two separate medium reservoirs containing 8 µg/mL and 16 µg/mL CEF in broth. Using the established flow rate (V̇_1_) of 0.45 mL/min and replacing the medium reservoir after two days, the final target concentration was reached after four days. All replicates remained culturable after sampling the evolved strains (evolvants) for phenotypic and genotypic investigation.

All evolvant strains, except for two replicates of PBIO4880, exhibited a CEF-resistant phenotype. The MIC of these resistant strains increased 8- to ≥64-fold compared to their respective parent strains. Although the two PBIO4880 evolvant strains (DF4880_2R and DF4880_3R) remained CEF-susceptible after the EE, their MIC increased 4- to 8-fold, reaching 2 µg/mL. A major contributor to the complexity of AMR is cross-resistance, which refers to the emergence of resistance to antibiotics due to exposure to different antibiotics with a similar mode of action^10,14^. This effect has been previously described for the novel cephalosporin antibiotic FDC. This antibiotic bypasses traditional resistance mechanisms using a Trojan horse strategy combined with enhanced structural stability. A single base pair duplication, occurring without exposure of the strains to FDC, conferred resistance to this sophisticated antibiotic. The mutation was located in a gene encoding a siderophore receptor (cirA) and led to a frameshift and early stop codon^2^. Therefore, the FDC MIC of all CEF-resistant strains was determined to investigate potential cross-resistance effects. Most evolvant strains were FDC-susceptible and had a 2- to 4-fold increase in their MIC. The CEF-susceptible evolvant strains DF4880_2R and DF4880_3R were also FDC-susceptible and showed no increase in their MIC compared to the respective parent strains. The CEF-resistant replicate of PBIO4880 (DF4880_1R) was FDC-resistant, with a ≥ 64-fold increase in MIC compared to the parent strain (Table 2). Plotting the fold-change of observed CEF and FDC MICs for all evolvant strains revealed a distribution into three distinct clusters (Fig. 2A). Cluster I comprised CEF-susceptible strains (DF4880_2R and DF4880_3R) exhibiting a 4-to 8-fold increase in CEF MIC but no change in FDC MIC. Cluster II contained all replicates of PBIO4886 and PBIO4914, which were CEF-resistant with 8-to ≥16-fold elevated MICs, as well as a moderate increase in FDC MIC of 2-to 4-fold. Cluster III included CEF- and FDC-resistant strain DF4880_1R, which exhibited an extensive MIC elevation of ≥64-fold for both antibiotics.Fig. 2. Visualization of the minimum inhibitory concentrations fold change for all replicates included in the continuous experimental evolution and schematic overview of the mutation in blaMIR-11 of the cefiderocol-resistant PBIO4880 evolvant.A Fold change of minimum inhibitory concentrations of evolvant strains compared to respective parent strains based on values shown in Table 2. The resistance profile of the evolvant strains is shown by differing shapes. Fold changes are plotted as their minimum value (i.e., ≥16 is plotted as 16). PBIO4886 replicate 1, and all three replicates of PBIO4914 overlap at one position. B Visualization of the single nucleotide substitution (g.934 G > C) in blaMIR-11 of the evolvant strain (DF4880_1R) compared to the respective parent (DF4880_1S) and the resulting amino acid substitution (p.A312P). C Cartoon representation of the AlphaFold3-predicted structure of the mutated MIR-11 protein (light blue), with the changed amino acid colored in yellow, superimposed onto the structure of AmpC BER from Escherichia coli (E. coli) (7CIN), with the α/β domain, helical domain, Ω loop, and R2 loop colored gray, pink, green, and red, respectively^25,26^. D Excerpt from B showing the β-lactamase domain interface with the Ω loop and the R2 loop in detail to highlight the position of the changed amino acid (yellow) in the protein structure.Table 2. Overview of phenotypic resistances of parent and evolvant strains, as well as mutations identified in evolvant strains compared to their respective parent strain, for all replicates included in the continuous experimental evolutionStrainReplicateParentEvolvantMutationCEF [µg/mL]FDC [µg/mL]CEF [µg/mL]FDC [µg/mL]PBIO488010.25 (S)2 (S)≥16 (R)≥128 (R)blaMIR-11 (g.934 G > C; p.A312P)20.25 (S)2 (S)2 (S)2 (S)n.a.30.5 (S)2 (S)2 (S)2 (S)n.a.PBIO488611 (S)0.25 (S)8 (R)0.5 (S)rpoA (g.79 A > C; p.T27P)21 (S)0.25 (S)≥16 (R)1 (S)ampD (g.244 G > A; p.G82D)31 (S)0.25 (S)≥16 (R)0.5 (S)dnaK (g.25 C > T; p.L9L), ampD (g.283 G > A; p.W95^a^)PBIO491412 (S)1 (S)≥16 (R)2 (S)blaACT-55 (g.952 G > A; p.V318M), malT (g.631 A > T; p.Q211L)22 (S)1 (S)≥16 (R)2 (S)-32 (S)1 (S)≥16 (R)2 (S)blaACT-55 (g.919_927dupAGCGACAGT; p.S307_S309dup), F0F1 ATP synthase subunit alpha (g.376 G > T; p.V126F)^a^Minimum inhibitory concentrations for cefepime (CEF) were determined by broth microdilution and for cefiderocol (FDC) using the ComASP® Cefiderocol 0.008-128 μg/mL Kit (Liofilchem, Roseto degli Abruzzi, Italy) and interpreted according to EUCAST v15 guidelines^20^. Evolvant strains were compared to their respective parent strains using breseq^23^.n.a. not applicable (not sequenced due to CEF susceptibility), R resistant, S susceptible.^a^The mutation of the F0F1 ATP synthase subunit alpha gene identified in DF4914_3R compared to the respective parent strain DF4914_3S is the identical genotype as found in the wildtype PBIO4914.

Subsequent analysis focused on the evolvant strains from Clusters II and III, as these demonstrated more pronounced phenotypic shifts, including a CEF-resistant phenotype as well as a change in FDC MIC compared to the parent strains (Fig. 2A). To elucidate the underlying resistance mechanism, these selected evolvant strains, their parent strains and the wild-type strains were characterized via whole-genome sequencing. Hybrid assemblies of the wild-type strains served as references for the parent strains, and the evolvant strains were compared to their respective parent strain. Two PBIO4914 evolvant strains (DF4914_1R and DF4914_3R) had mutations in the class C serine β-lactamase blaACT-55. DF4914_1R had a missense mutation leading to an amino acid exchange (p.V318M), which corresponds to V298M according to the “structural alignment-based numbering of class C β-lactamases” (SANC) scheme^15^. For the third replicate of PBIO4914 (DF4914_3R), a duplication of nine nucleotides was identified, leading to three additional amino acids (p.S307_S309dup) at position 289 based on the SANC scheme. Both mutations are located in the R2 loop structure of the β-lactamase domain interface. Additionally, DF4914_1R also had a missense mutation in malT, which encodes a transcriptional activator of the maltose regulon. For the parent strain DF4914_3S, a mutation in the F0F1 ATP synthase subunit alpha gene compared to the wild-type PBIO4914 was identified, which was not present in the evolvant strain DF4914_3R. No mutation was identified in the second replicate of PBIO4914 (DF4914_2R) compared to the parent strain, within the detection limits of our WGS pipeline, and the mechanism underlying resistance in this lineage remains unresolved. The second and third replicates of PBIO4886, DF4886_2R and DF4886_3R, respectively, had mutations in ampD, which encodes the negative regulator of AmpC β-lactamases. In DF4886_2R, a missense mutation was identified, and in DF4886_3R, a nonsense mutation and an additional missense mutation in the dnaK gene, encoding a heat shock protein 70 family chaperone, were identified. Replicate one of PBIO4886 (DF4886_1R) carried a missense mutation in the gene encoding the RNA polymerase alpha subunit (rpoA). Finally, the FDC-resistant replicate of PBIO4880 (DF4880_1R) exhibited a single missense mutation in the class C serine β-lactamase gene blaMIR-11 (Table 2).

Missense mutation in blaMIR-11 confers cross-resistance to cefiderocol and ceftazidime-avibactam

The identified single missense mutation of the FDC- and CEF-resistant evolvant strain DF4880_1R resulted in an amino acid substitution in the β-lactamase MIR-11 (p.A312P). This substitution is located at position 292 according to the SANC scheme (Fig. 2B) and, based on annotations, is part of the R2 loop structure at the domain interface^15^. Additionally, the AlphaFold3-predicted structure of the mutated MIR-11 was compared with the crystal structure of class C serine β-lactamase AmpC BER from E. coli^16,17^ (Fig. 2C). This β-lactamase shares 73.63% amino acid identity at 93% coverage with MIR-11 and the protein structures have a root mean square deviation of 1.336 Å or 1.356 Å for the wild-type MIR-11, according to an analysis using PyMOL. The comparison also shows the location of the substituted amino acid (shown in yellow) within the R2 loop structure (Fig. 2D). The R2 loop is part of the active site of serine β-lactamases, along with the Ω loop at the opposite boundary and the nucleophilic Ser64 in between. The structure of these components is crucial for accommodating the R1 and R2 side chains of β-lactams^17–21^. Ser64 is responsible for the nucleophilic attack on the lactam ring, which is subsequently broken in the first step of the hydrolysis of β-lactam antibiotics^17,22^. Mutations located in the R2 and Ω loops of various β-lactamases have previously been implicated in conferring resistance to cephalosporins, such as CAZ and CEF^23–25^. It has also been shown that the insertion of two alanines in the R2 loop region of AmpC BER, compared to the AmpC EC2, resulted in a wider active site and improved catalytic efficiency^17^.

To validate the correlation between the missense mutation in the blaMIR-11 gene and the observed phenotypic resistance to CEF and FDC, the wild-type and mutated genes were heterologously expressed in E. coli DH5α. Both variants, including their promoter and terminator sequences, were cloned into a vector to facilitate expression at near-native levels. Comparison between both revealed a 16-fold and 8-fold increase in MIC for CEF and FDC, respectively, for the strain expressing the mutant gene compared to the strain expressing the wild-type gene. In addition, DF4880_1R also presented phenotypic resistance to CAZ-AVI, which is indicated for treatment of complicated urinary tract infections caused by MDR ECC, with a > 64-fold increase in MIC compared to the parent strain (Table 3 and Supplementary Table S1). The E. coli DH5α strain expressing the mutated blaMIR-11 gene showed a 16-fold increase in MIC compared to the strain expressing the wild-type gene. The novel bicyclic boronate β-lactamase inhibitor taniborbactam has been shown to be a promising candidate in combination with CEF for lowering MIC in serine and metallo-β-lactamase-producing isolates^26–28^. Here, we observed a 64-fold decrease in the MIC for the strain expressing the mutated blaMIR-11, indicating that the divergence in the domain interface of the enzyme did not affect taniborbactam susceptibility. According to EUCAST guidelines, the E. coli DH5α strain expressing the mutated blaMIR-11 gene only presented phenotypic resistance to CEF, but the substantial increase in MIC for CEF, FDC, and CAZ-AVI highlights and validates the influence of the single nucleotide substitution on resistance development in the investigated E. roggenkampii strain (Table 3). Further investigation of the fitness of the FDC-resistant mutant DF4880_1R compared to the parent strain DF4880_1S revealed no differences, as determined by growth kinetics (Fig. S1A&B). Additionally, a comparative analysis of the β-lactamase activity using the chromogenic cephalosporin nitrocefin revealed no significant difference between parent and mutant strain as well as between the E. coli DH5α strains expressing the mutant and the wild-type blaMIR-11 gene (Fig. S1C).Table 3. Overview of minimum inhibitory concentrations (MIC) of cefepime (CEF), cefiderocol (FDC), ceftazidime-avibactam (CAZ-AVI), and cefepime-taniborbactam (FTB)CEF [µg/mL]FDC [µg/mL]CAZ-AVI [µg/mL]FTB [µg/mL]PBIO48800.250.50.5^a^n.a.DF4880_1S0.2520.25^a^n.a.DF4880_1R≥16≥128≥16^a^n.a.E. coli DH5α0.1250.0640.125≤0.06E. coli DH5α^pCR2.1^10.0640.1250.125E. coli DH5α^pCR2.1_blaMIR11-WT^0.50.0640.25≤0.06E. coli DH5α^pCR2.1_blaMIR11-MT^80.540.125Shown are the parent and evolvant strains of PBIO4880 obtained during the experimental evolution, as well as the wild-type PBIO4880 strain. Additionally, the MICs of E. coli DH5α transformed with a plasmid carrying either the wild-type or mutated blaMIR-11 gene, or the empty vector, were compared with those of the wild-type strain.n.a., not applicable.^a^MICs were determined by broth microdilution, using ComASP® Cefiderocol 0.008–128 μg/mL Kit (Liofilchem, Roseto degli Abruzzi, Italy) for FDC, or by the VITEK® 2 COMPACT (bioMérieux) platform.

Discussion

Our study presents a novel, high-throughput system for continuous EE that overcomes the main limitations of conventional methods. While the morbidostat or bioreactor-based approaches remain the gold standard for continuous EE studies, our design utilizes standard laboratory equipment and simple peristaltic pumps to provide a versatile and accessible alternative. This setup ensures constant medium exchange and a continuously increasing antibiotic gradient, prioritizing ease of implementation and throughput. This simple yet straightforward setup greatly reduces hands-on time and makes continuous EE accessible to many laboratories, streamlining resistance evolution research.

By eliminating the need for optical density-dependent adjustments or discontinuous drug gradients, our method maintains a steady, constant selection pressure that favors the emergence of resistant mutants while still allowing population survival^9,10^. Although this linear increase in antibiotic concentration differs from the fluctuating multi-dose in vivo pharmacokinetics, it shares the critical feature of being independent of bacterial population dynamics. While this also removes the requirement for optical density sensors and therefore simplifies the hardware requirements, it introduces a potential limitation due to the lack of growth-dependent feedback control like the morbidostat employs. In addition, by avoiding the population bottlenecks inherent in most discontinuous EE approaches, our system more closely models the sustained selection environment found in vivo when considering the core interaction between bacteria and antibiotics.

In just four days of exposure to an increasing CEF gradient, all replicates of three ECC strains remained culturable. Over 77% (7/9) of the total replicates cultivated simultaneously exhibited phenotypic CEF resistance, with an increase in MIC of 8- to ≥64-fold. For two replicates, we could not confirm phenotypic resistance according to EUCAST guidelines. This is likely due to the sampling process, which imposes an artificial bottleneck and may, by chance, select susceptible clones that survived the antibiotic challenge in the overall population. This is based on their own increase in MIC of 4- to 8-fold, reaching 2 µg/mL, just below the EUCAST breakpoint of >4 µg/mL, and on population dynamics, such as the protection provided by other resistant clones in the population^10,29^ (Fig. 2A). Overall, these findings highlight the effectiveness of our method and the adaptability of the ECC strains. Our system provides the benefits of continuous selection pressure while remaining high-throughput and low-complexity. This is accomplished by enabling parallel cultivation of multiple replicates using only standard laboratory equipment and minimal manual intervention. Future research will determine whether the observed efficiency is consistent across diverse bacterial species and drug combinations.

Four of the seven resistant evolvant strains carried mutations directly in or related to serine class C β-lactamases. Two replicates of PBIO4914 had a mutation in the blaACT-55 gene, resulting in divergence in the R2 loop structure of the β-lactamase. Previous studies have shown that mutations in the R2 loop structure of serine class C β-lactamases can increase catalytic efficiency against cephalosporins such as CEF, leading to resistant phenotypes^25^. For the β-lactamase AmpC BER, the insertion of two amino acids in the R2 loop region could be traced to a wider active site and higher catalytic efficiency (kcat/Km) for cephalosporins and imipenem^17^. In contrast, two replicates of PBIO4886, carrying the β-lactamase gene blaACT-16, had mutations in the ampD gene. AmpD regulates AmpC β-lactamase expression, and mutations have been shown to cause overexpression of β-lactamases, resulting in higher MICs and resistance^30–33^. The other CEF-resistant replicate of PBIO4886 had a missense mutation in rpoA, a gene previously implicated only in ampicillin resistance in E. coli and enhanced susceptibility in Pseudomonas aeruginosa due to reduction of the MexEF-OprN efflux pump activity^34,35^. The mutation in rpoA may influence global transcription patterns in a way that indirectly affects either the β-lactamase expression itself or other resistance mechanisms. For most of the evolvant strains, we identified genetic differences compared to their respective parent strains that are consistent with the observed resistance phenotypes, which allowed us to propose plausible mechanistic scenarios. However, we acknowledge that these scenarios remain simplified and do not capture all potential layers of regulation and interaction. The CEF-resistant evolvant of PBIO4914, for which no clear genetic driver was detected (DF49142R), highlights this limitation particularly clearly. Even under these controlled laboratory evolution conditions, not all resistance mechanisms can be fully resolved by WGS-based approaches.

The CEF-resistant replicate of PBIO4880 (DF4880_1R) also showed phenotypic resistance to FDC and CAZ-AVI, which we traced to a single missense mutation in the serine class C β-lactamase gene blaMIR-11. The resulting amino acid substitution, from alanine to proline at position 292 according to the SANC scheme, is located in the R2 loop structure^15^ (Fig. 2B, C). As previously mentioned, these types of mutations have been implicated in resistance to various cephalosporins. Shields et al. reported a clinically evolved E. hormaechei strain that was phenotypically resistant to CAZ-AVI and CEF, among others, and also showed reduced susceptibility to FDC after the patient was treated with CEF. The resistance and cross-resistance effects were attributed to a deletion of two amino acids in the R2 loop of an AmpC β-lactamase (p.A292_L293del)^36^. During a clinical trial of FDC, a resistant E. cloacae strain emerged after FDC treatment in a patient with nosocomial pneumonia. The resistant isolate had a MIC of 8 µg/mL, representing an 8-fold increase, and carried a missense mutation in the β-lactamase ACT-17 (p.A313P), which is very similar in position and involved amino acids to the one we observed (p.A312P / p.A292P according to the SANC scheme), resulting in a ≥ 64-fold increase in MIC^37,38^. In both cases, expression of the mutated gene in E. coli, compared to the wild-type gene, led to an increase in MIC ( > 32-fold for p.A292_L293del in AmpC of E. hormaechei and 2-fold for p.A313P in ACT-17 of E. cloacae), comparable to the effect we observed in E. coli (8-fold increase) (Table 3). A comparative analysis using Clustal Omega and all blaMIR genes available in the NCBI GenBank database (accessed on 06.08.2025) revealed that blaMIR-36 also carries a missense mutation, resulting in an alanine to threonine substitution at position 292^39,40^. This highlights the clinical relevance of our findings and that this type of commonly occurring mutation can even result in resistance against the novel antibiotic FDC. Similar to previous results for K. pneumoniae, where mutations conferring resistance to FDC also restored susceptibility to carbapenems, the mutation in DF4880_1R also presented an evolutionary trade-off^41^. The evolvant strain was phenotypically susceptible to the carbapenem ertapenem with a 4-fold decrease in MIC compared to the parent strain, which was phenotypically resistant (Supplementary Table S1). Despite the significant increase in CEF MIC attributable to the mutated blaMIR-11 gene when expressed in E. coli, the combination of CEF and the β-lactamase inhibitor taniborbactam led to a 64-fold reduction in MIC compared to CEF alone. Evidently, the amino acid exchange did not decrease taniborbactam efficacy. To our knowledge, only one case has been described in which mutations in a class C β-lactamase conferred taniborbactam non-susceptibility, as a clinical E. coli strain resistant to CAZ-AVI, showed no difference in MIC between CEF and cefepime-taniborbactam (FTB; 4 µg/mL). The strain carries an AmpC β-lactamase CMY-185 with four amino acid substitutions, one of which is located in the Ω loop and another one in helix 11^42,43^. The exact influence of the amino acid exchange from alanine to proline, leading to the FDC-resistant phenotype in DF4880_1R, remains to be elucidated. Previously, proline residues have been implicated in enhancing secondary structure stability of β-lactamases as well as increasing protein expression levels^44–46^. We were not able to discern any differences in fitness between the parent strain DF4880_1S and the respective evolvant strain DF4880_1R, showing that the single amino acid exchange solely affects the phenotypical resistance (Fig. S1). Despite not being a significant difference, DF4880_1R showed a tendency for higher nitrocefin hydrolysis activity compared to the respective wild-type; however, this trend could not be confirmed for the E. coli strain expressing the mutant gene compared to the one carrying the wild-type variant (Fig. S1C). It has been previously shown that AmpC mutations leading to resistance against extended-spectrum cephalosporins can, in turn, reduce their catalytic efficiencies against narrow-spectrum cephalosporins. AmpC BER from E. coli had lower rates of hydrolysis but a higher affinity for cephalosporins compared to AmpC EC2^47^. Overall, this highlights the significance of the identified mutant strain, exhibiting resistance to extended-spectrum cephalosporins like the novel and sophisticated FDC with evolutionary trade-offs that are limited to the overall resistance pattern and do not affect virulence or fitness. Furthermore, this is a case of genuine cross-resistance, which developed in a naïve strain that had definitely no prior evolutionary exposure to cefiderocol, due to its isolation date.

Due to the nature of this proof-of-concept study, only one colony per replicate was investigated. This represents a bottleneck because of the random nature of sampling and excludes factors such as population heterogeneity. In future studies using this methodology, it is therefore prudent to expand downstream analysis to the population scale, employing methods such as single-cell sorting and sequencing. A further limitation of our study is the absence of transcriptomic and proteomic data. We did not perform any RNA-based or proteome-wide analyses for any of the evolvant strains. Consequently, our interpretation of resistance mechanisms is restricted to stable genetic changes detectable by our WGS pipeline in combination with phenotypic susceptibility testing. Additionally, the fixed dilution rate and drug gradient used in this EE system represent a trade-off. The advantages of this simplification, which enables accessibility of the method and mirrors aspects of clinical antibiotic therapy, come at the expense of feedback control. This could potentially lead to the extinction of the bacterial population or growth beyond the exponential phase if the strains are not compatible with the chosen parameters in terms of doubling time or frequency of resistance mutations. Furthermore, it is currently unclear whether this system would be able to select for multi-step mutations, which are required for phenotypic resistance to some antibiotics, such as quinolones.

In summary, our novel continuous EE methodology enabled rapid, high-throughput detection of CEF resistance and significant cross-resistance to FDC and CAZ-AVI in historical EEC strains, traced to a single nucleotide substitution in a class C β-lactamase. The system retains the key advantages of continuous evolution methodologies while overcoming the limitations of low throughput and specialized equipment. Due to the unique advantages and trade-offs of this system, this system functions as a complementary methodology to the sophisticated morbidostat method and expands the toolkit available for EE studies. These results highlight the low genetic barrier for cross-resistance to emerge, a major contributor to the global threat of AMR. Future research will use this methodology to investigate diverse bacterial species and drug combinations at a larger scale and to further elucidate adaptation mechanisms at the population and transcriptional levels.

Methods

Bacterial strains and growth conditions

The list of bacterial strains used in this study can be found in Supplementary Table S2. Generally, strains were stored at −80 °C in LB containing 20% (v/v) glycerol, and 5 mL LB was inoculated with single colonies from LB agar plates and incubated at 37 °C and 180 rpm overnight. For the heterologous expression experiments, cultures were supplemented with 50 mg/L kanamycin. For growth kinetics, overnight cultures were used to inoculate fresh LB medium to an optical density at 600 nm of 0.01 in a 96-well plate in triplicate. Measurements were performed every 15 min for 22 h at 37 °C, with 30 s of double orbital shaking before each measurement. The resulting growth curves were used to determine the area under the curve of the optical density at 600 nm.

Continuous EE

Continuous EE was performed in 100 mL flasks containing 30 mL cation-adjusted Mueller-Hinton-Bouillon 2 (MH2) (Sigma Aldrich, USA-90922) medium, employing magnetic stirring bars to facilitate agitation at 80 rpm using the MULTISTIRRER magnetic stirrer F203A0178 (Velp Scientifica™, Italy). The culture flasks and the multi-neck bottle (interim reservoir) containing fresh medium without added antibiotics, which was also continuously stirred at 300 rpm using the Hei-PLATE Mix ‘n’ Heat Core+ (135 mm) (Heidolph Scientific Products GmbH, Germany), were kept at 37 °C for the duration of the experiment. The culture flasks were inoculated with overnight cultures to an optical density (600 nm) of 0.1. To facilitate a continuous gradient of antibiotic concentration in the culture flask, medium containing 8 µg/mL CEF from a 2 L reservoir, which was UV-protected and kept at room temperature, was transferred into the interim reservoir using an IPC 4 ISM930 peristaltic pump (Ismatec Laboratoriumstechnik, Germany) at V̇_1_ of 0.45 mL/min over the course of two days. This interim reservoir, containing medium with a seamlessly increasing concentration of CEF, supplied all culture flasks using a second peristaltic pump, IPC-N 24 ISM939 (Ismatec Laboratoriumstechnik, Germany), at V̇_2_, which is equivalent to V̇_1_ divided by the number of culture flasks. With the same flow rate as the influx of fresh medium (V̇_2_), old medium was continuously removed into the waste, using the same pump. Validation of the flow rate was performed using a fixed concentration of a bromophenol blue solution as the reservoir, and the concentration in the culture flasks was measured at different time points (n = 3). To minimize tubing length and reduce the risk of dead volume or backflow, the waste reservoir was placed inside the incubator (37 °C) adjacent to the culture flasks. Although this configuration was used to maintain system integrity and temperature stability within the lines, the waste reservoir can also be kept at room temperature outside the incubator, depending on setup requirements. The maximum number of culture flasks that can be employed in parallel mainly depends on the capacity of the employed pumps regarding the number of channels, as well as their maximum flow rate. After 2 days, the reservoir was exchanged for medium with a CEF concentration of 16 µg/mL, leading to a final concentration in the culture flasks of four times the EUCAST breakpoint of CEF (16 µg/mL) after another 2 days^20^. Samples were collected directly after inoculation (parent) and after reaching the final antibiotic concentration (evolvant), using a 3-way stopcock situated between each culture flask and the waste. After plating 100 µL of the 1000 µL sample onto LB agar plates containing 4 µg/mL CEF and incubation at 37 °C overnight, single colonies were cryo-conserved in LB containing 20% (v/v) glycerol and 4 µg/mL CEF (Fig. 1.)

Cloning

The complete genes of the wild-type blaMIR-11 from DF4880_1S and the mutated variant from DF4880_1R, including native promoter and terminator sequences (identified via bPROM and ARNold^48^), were amplified via PCR using the primers MIR11_For and MIR11_Rev (Supplementary Table S3). After purification, fragments were cloned into the pCR™2.1 vector according to the TA Cloning™ Kit (Thermo Fisher, Waltham, USA) instructions and transformed into E. coli DH5α. Following selection via 50 mg/L kanamycin, and verification by PCR using the M13 primers, the insertion into the plasmids was validated by Sanger sequencing (Supplementary Table S3).

Antimicrobial susceptibility testing

The VITEK® 2 COMPACT (bioMérieux, Marcy-l'Étoile, France) platform was used to facilitate the phenotypic antimicrobial susceptibility testing. Additionally, to determine the minimal inhibitory concentrations (MIC) of CEF, FTB and CAZ-AVI, a broth microdilution assay was performed according to ISO 20776-1 in triplicate. The MIC of FDC was assessed in triplicate using the ComASP® Cefiderocol 0.008-128 μg/mL Kit (Liofilchem, Roseto degli Abruzzi, Italy) according to the manufacturer’s instructions. Additionally, the MIC of colistin was determined using Colistin Ezy MIC™ Strips (CL) 0.016–256 mg/mL (HiMedia Laboratories, Mumbai, India) according to the manufacturer’s instructions in triplicate. MIC values were evaluated according to EUCAST version 14.0 guidelines^13^.

DNA isolation and whole-genome sequencing (WGS)

Total genomic DNA of wild-type isolates, parent and evolvant strains was isolated using the MasterPure DNA Purification Kit for Blood Version II (LGC Biosearch Technologies, Hoddesdon, United Kingdom). Isolated DNA was sequenced by SeqCenter (Pittsburgh, MA, USA) using Illumina sequencing by synthesis (SBS) technology. The sequencing libraries were prepared using the tagmentation-based and PCR-based Illumina DNA Prep kit and custom IDT 10 bp unique dual indices (UDI) with a target insert size of 280 bp. No additional DNA fragmentation or size selection steps were performed, and sequencing was performed on an Illumina NovaSeq X Plus sequencer in one or more multiplexed shared-flow-cell runs, producing 2 × 151 bp paired-end reads. Demultiplexing, initial quality control and initial adapter trimming was performed with bcl-convert v. 4.2.4 (https://support-docs.illumina.com/SW/BCL_Convert/Content/SW/FrontPages/BCL_Convert.htm).

The wild-type isolates (PBIO4880, PBIO4886 and PBIO4914) were additionally sequenced using Nanopore sequencing. Sample libraries were prepared using the PCR-free Oxford Nanopore Technologies (ONT) Ligation Sequencing Kit (SQK-NBD114.24) with the NEBNext® Companion Module (E7180L) to the manufacturer’s specifications. No additional DNA fragmentation or size selection was performed. Nanopore sequencing was performed on a MinION Mk1B (or compatible GridION) sequencer using R10.4.1 flow cells in one or more multiplexed shared-flow-cell runs. Run design utilized the 400 bps sequencing mode with a minimum read length of 200 bp. Adaptive sampling was not enabled. Guppy v. 6.4.6 was used for super-accurate basecalling (SUP), demultiplexing, and adapter removal (dna_r10.4.1_e8.2_400bps_modbases_5mc_cg_sup.cfg). No quality trimming has been performed.

Genome assembly, annotation, and genomic analysis

Raw paired-end sequencing reads were contaminant-filtered (e.g. PhiX), again adapter-trimmed and quality-trimmed using BBDuK from BBMap v. 39.03 (https://sourceforge.net/projects/bbmap/). Quality control of raw and trimmed reads was done with FastQC v. 0.12.1 (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/). Short read-only assemblies were created using shovill v. 1.1.0 (https://github.com/tseemann/shovill) in combination with SPAdes v. 3.15.5^49^ at a maximum sequence coverage of 100x without using the integrated correction step. A manual polishing step was performed by mapping all trimmed paired-end reads to the draft assembly using BWA-MEM2 v. 2.2.1^50^ and polishing the assembly with Polypolish v. 0.5.0^51^. Hybrid assemblies with a short read-first approach (using trimmed reads as obtained above) were created using Unicycler v. 0.5.0^52^ with SPAdes v. 3.15.4^49^ and Racon v. 1.5.0^53^. Completeness/contamination of draft and completed genome assemblies was assessed with CheckM v. 1.2.2^54^. Closed genomes of wild-type isolates were annotated using Bakta v. 1.8.2 with full database version 5.0^55^.

Species assignment was performed with logic integrated in Kleborate v. 2.3.2^56^, and Multilocus Sequence Typing (MLST) was done with mlst v. 2.23.0 (https://github.com/tseemann/mlst) using the scheme for the EEC^57^ hosted at the PubMLST website. Short nucleotide variant (SNV) analysis was conducted using breseq v. 0.38.2^58^. For this, short reads of parent strains were first mapped to the closed genome of the corresponding wild-type isolate, and SNVs were applied to these references using the gdtools APPLY command. The resulting closed parent genomes served as a reference for the mapping of short reads from evolvant strains.

Chromogenic β-lactam assay

The chromogenic cephalosporin nitrocefin was used to compare the activity of the wild-type and mutated beta-lactamase MIR-11^59^. Overnight cultures were used to inoculate 5 mL LB with a 1:100 dilution and incubated at 37 °C and 180 rpm for 3 h. After washing with PBS, the cultures were adjusted to an optical density at 600 nm of 1 in PBS. In a 96-well plate, 100 µL 0.5 mM nitrocefin was added to 100 µL bacterial suspension and incubated at 37 °C for 20 min. Absorption at 390, 486, and 600 nm was measured after 30 s of double orbital shaking.

Supplementary information

Echelmeyer et al._revised_Supplemental_Material_clean_version