Evolution of beta-lactam resistance causes fitness reductions and several cases of collateral sensitivities in the human pathogen Haemophilus influenzae

TL;DR

This study shows how Haemophilus influenzae evolves resistance to beta-lactam antibiotics, often leading to reduced fitness and unexpected sensitivities to other drugs.

Contribution

The study reveals that resistance evolution in H. influenzae leads to unpredictable phenotypes and collateral sensitivities.

Findings

Resistance to beta-lactam antibiotics was driven by mutations in ftsI and OmpP2 genes.

Over 50% of resistant clones became more susceptible to aminoglycosides, clarithromycin, and colistin.

Some mutations caused cross-resistance to meropenem and fluoroquinolones.

Abstract

The evolution of antimicrobial-resistant pathogens poses a global health threat, and it is unclear if evolutionary trajectories to resistance lead to predictable phenotypes. We analyzed antimicrobial resistance (AMR) evolution in the human pathogen Haemophilus influenzae in controlled evolution experiments with increasing concentrations of either ampicillin, cefotaxime, or ceftriaxone. We isolated 315 clones from different time points of six independent experiments and characterized changes in genome sequences, bacterial fitness, and minimum inhibitory concentrations (MICs) to 14 antibiotics. Resistance evolution under ampicillin and cefotaxime was mainly driven by mutations in the ftsI gene, encoding the penicillin-binding protein 3. However, ceftriaxone exposure repeatedly selected for amino acid substitutions in the outer membrane protein P2 (OmpP2). Some OmpP2 mutants reproducibly showed phenotypic heterogeneity not only for ceftriaxone but also for fluoroquinolones, rifampicin, and tetracycline. Bacterial fitness assessments revealed trade-offs between resistance-associated mutations and growth, though no systematic correlation of MIC increase and growth deficit was detected. Over 50% of the selected clones became more susceptible to aminoglycosides, clarithromycin, and colistin, while some mutation patterns resulted in cross-resistance to meropenem and fluoroquinolones. Overall, beta-lactam antibiotics reproducibly selected mutants with increased MICs, but evolutionary pathways and resulting phenotypes remain unpredictable. These findings highlight the complexity of resistance evolution and suggest that future AMR treatment strategies may need to consider strain-specific and collateral effects more closely.

Figures

Increasing number of mutations and changes in beta-lactam minimum inhibitory concentrations (MICs) in six evolution experiments. Number of mutations (black) and minimum inhibitory concentration fold increase compared to the parental strain Rd KW20 for ampicillin (Amp, red), cefotaxime (Ctx, green), and ceftriaxone (Cro, blue) in selected clones over 20 passages (two or five clones per passage). Bacterial population (A) without antibiotic exposure and exposed to increasing concentrations of (B) ampicillin, (C) cefotaxime, (D) ceftriaxone replicate 1, (E) ceftriaxone replicate 2, and (F) ceftriaxone replicate 3. Shaded areas represent the 95% confidence interval.

Maximum likelihood phylogeny of H. influenzae Rd KW20 clones evolved in the presence of ampicillin (A, 53 clones), cefotaxime (B, 56 clones), and ceftriaxone (C, 57 clones). Amino acid substitutions in the penicillin-binding protein 3 (PBP3) and presence of mutations in specified genes as well as minimum inhibitory concentration (MIC) changes over time are color coded. Strains that exceeded the clinical breakpoint for ampicillin/cefotaxime/ceftriaxone according to EUCAST breakpoints are marked with a white star at the MIC heatmap. The specified genes are those in which at least 20 clones from the whole set of selected clones exhibited mutations but did not mutate under antibiotic-free conditions.

Crystal structure of transpeptidase domain (residues 254–610) of the penicillin-binding protein 3 (PBP3) of H. influenzae (26) with superimposed ampicillin (A, colored in cyan) and cefotaxime (B, magenta) molecules. In vitro selected amino acid substitutions in PBP3 Thr332lle, Thr443Ala, Ala530Ser, Asp551Tyr, and Ala561Glu are in red. Known positions for amino acid substitutions defining established PBP3 resistance groups are in green.

Bacterial fitness of selected clones derived from evolving populations exposed to ampicillin (Amp), cefotaxime (Ctx), and ceftriaxone (Cro). Bacterial fitness measured as relative growth difference between the ancestral strain (H. influenzae Rd KW20) and the selected clone. A relative fitness of 1 indicates no fitness cost, whereas a value greater or less than 1 indicates increased or decreased fitness, respectively. The black bars represent standard deviations. *Padj < 0.05, Padj < 0.01, *Padj < 0.001. The heatmap indicates MIC values of Amp, Ctx, and Cro (in mg/L) of the respective strain (already presented in Fig. 2). Stars indicate clinical resistance above the respective EUCAST breakpoints. Hatched panels indicate phenotypic heterogeneity. Clones that were derived from the same evolution experiment are grouped together on the y-axis.

Collateral effects of increased MICs against individual beta-lactam antibiotics. Heat map of log2-transformed relative increase in MIC of ampicillin, cefotaxime, ceftriaxone (underlying MIC values already presented in Fig. 1), meropenem (beta-lactams), amikacin, kanamycin, gentamicin (aminoglycosides), levofloxacin, moxifloxacin (fluoroquinolones), clarithromycin (macrolide), vancomycin (glycopeptide), rifampicin (ansamycin), colistin (polymyxin), and tetracycline (tetracycline) in mutant clones relative to the MIC of the parental strain H. influenzae Rd KW20. Selected clones were evolved in the multi-step evolution experiment. Stars indicate clinical resistance above the respective EUCAST breakpoints (not available for aminoglycosides, clarithromycin, vancomycin, and colistin). Hatched panels indicate phenotypic heterogeneity. Clones that were derived from the same evolution experiment are grouped together on the y-axis.

Equations

Full text

INTRODUCTION

Antimicrobial resistance (AMR) is a major public health challenge globally (1). The World Health Organization has identified several priority pathogens to inform research and development as well as public health interventions to combat the AMR crisis (2). One of these priority pathogens is ampicillin-resistant Haemophilus influenzae strains. H. influenzae is a common gram-negative opportunistic pathogen residing in the human upper respiratory tract (3, 4). The bacterium can cause a range of infections, from mild illnesses like sinusitis and otitis media to severe invasive diseases such as meningitis and bacteremia, particularly in vulnerable populations (5). Despite the successful implementation of a vaccine against H. influenzae serotype b strains, non-typeable strains (NTHi), i.e., strains lacking a polysaccharide capsule, have become prevalent (6) and evolved resistance to aminopenicillins such as ampicillin and amoxicillin in many world regions (7–9).

While amoxicillin-clavulanate and ampicillin-sulbactam are used worldwide to overcome beta-lactamase-mediated resistance (predominantly TEM-1 and ROB-1), third-generation cephalosporins such as cefotaxime and ceftriaxone also effectively bypass these enzymes (10, 11) and are increasingly favored in clinical practice (12). This is due to their favorable tolerability and their ability to achieve high serum and cerebrospinal fluid concentrations, making them especially suitable for the treatment of severe infections like sepsis and meningitis caused by H. influenzae (13). However, resistance to beta-lactams can also evolve in beta-lactamase-negative strains via alterations in the penicillin-binding protein 3 (PBP3), encoded by the ftsI gene (14). Mutations in ftsI, particularly in the transpeptidase domain of PBP3, reduce the binding affinity for different beta-lactam antibiotics (15).

Current understanding is limited regarding whether resistance evolution in H. influenzae follows identical genetic and phenotypic trajectories across different beta-lactam classes and whether such evolutionary pathways are reproducible. In particular, it is unclear if evolution consistently leads to cross-resistances against multiple beta-lactam antibiotics or if variability in resistance mechanisms leads to diverse outcomes. Here, laboratory-based evolution experiments provide a powerful approach to address these questions (16). By simulating selective pressures in vitro, these experiments allow for tracking genetic changes and phenotypic adaptations as resistance evolves. In addition, this experimental approach facilitates an assessment of evolved collateral effects, such as cross-resistance or collateral sensitivity (i.e., variants with evolved resistance against certain antibiotics show an increase in susceptibility to other antibiotics), which may inform about epistatic interactions, and possibly new treatment regimens against resistant strains (17).

Therefore, the aim of the current study was to enhance our understanding of the predictability of resistance evolution and the resulting collateral effects. In particular, we investigated the evolutionary pathways leading to resistance against either ampicillin, cefotaxime, or ceftriaxone in H. influenzae using a multi-step in vitro evolution approach. By subjecting the reference strain H. influenzae Rd KW20 to increasing concentrations of these antibiotics, we tested whether evolution results in identical or divergent genetic trajectories and phenotypes. We employ short- and long-read sequencing technologies to map mutations across the genome and assess their impact on minimum inhibitory concentrations (MICs) for different antibiotics. Furthermore, we evaluated whether the evolution of resistance is reproducible by examining the consistency of mutation patterns and collateral effects on other antibiotics.

MATERIALS AND METHODS

Bacteria and culture methods

The beta-lactamase-negative reference strain H. influenzae Rd KW20 and evolved mutant strains were routinely grown from −80°C freezer stocks on chocolate agar (Mueller–Hinton broth with 5 g/L yeast extract, 17 g/L agar, 5% defibrinated horse blood, and 20 mg/L beta-nicotinamide adenine dinucleotide [beta-NAD]) at 37°C in a humid atmosphere containing 5% CO_2_. Colonies grown on chocolate agar served as inoculum for liquid cultures in Mueller–Hinton broth supplemented with 5 g/L yeast extract, 20 mg/L beta-NAD, and 15 mg/L hemin, hereafter referred to as Haemophilus test medium (HTM), grown at 37°C with shaking at 100 or 200 rpm. Liquid cultures were routinely plated on HTM agar (formulation of HTM with 17 g/L agar added).

Antibiotic susceptibility testing

To determine the antibiotic starting concentrations for the multi-step evolution experiment, MICs of ampicillin, cefotaxime, and ceftriaxone were tested for the reference strain using broth microdilution methodology according to European Committee on Antimicrobial Susceptibility Testing (EUCAST) protocol, but using HTM. Ampicillin sodium salt, cefotaxime sodium salt, and ceftriaxone disodium salt hemi(heptahydrate) were purchased from Sigma-Aldrich.

Gradient diffusion strips (Liofilchem MIC Test Strips) on Mueller–Hinton fastidious agar (BD) were routinely used for all isolated single clones evolved in the evolution experiment to determine the MICs of ampicillin, cefotaxime, and ceftriaxone. To check for collateral effects, 17 clones (3–4 per evolved population, selected based on differing mutational patterns and/or distinct beta-lactam MICs) were additionally tested for clarithromycin, kanamycin, colistin, vancomycin, levofloxacin, rifampicin, amikacin, tetracycline, gentamicin, and meropenem susceptibility using Liofilchem MIC Test Strips. Susceptibility interpretation was performed according to EUCAST breakpoints (v.13.0 2023), which are 1 mg/L for ampicillin, 0.125 mg/L for cefotaxime, and 0.125 mg/L for ceftriaxone.

Multi-step evolution experiment

Populations of the susceptible reference strain H. influenzae Rd KW20 were evolved under antibiotic pressure (ampicillin, cefotaxime, and ceftriaxone) in HTM, starting with an antibiotic concentration of 0.5× MIC of the reference strain (determined using broth microdilution methodology). One population was evolved in the absence of any antibiotic pressure (control). Evolution in the presence of ceftriaxone was performed three times to investigate reproducibility of resistance evolution.

A volume of 20 µL of single-colony resuspension (a single colony of H. influenzae Rd KW20 grown on chocolate agar resuspended in 100 µL of HTM) was used to inoculate 2 mL cultures containing either 0.125 mg/L of ampicillin, 0.008 mg/L of cefotaxime, 0.002 mg/L of ceftriaxone (0.002 mg/L), or no antibiotic (first passage) (Fig. S1).

The evolving 2 mL cultures were grown in 12-well microtiter plates at 37°C, 100 rpm and 5% CO_2_ for 4 days per cycle, with daily 100-fold dilution into 2 mL of fresh HTM. At each passage, one culture was grown at the same antibiotic concentration as in the previous passage (back-up culture), while three additional cultures were inoculated containing twice the antibiotic concentration of the previous passage. Whenever at least one of the cultures with higher antibiotic concentration showed visible growth after 24 h, this culture was used to inoculate the next passage. This approach allowed for a stepwise increase in antibiotic concentration throughout the experiment without extinction. The highest antibiotic concentration that supported visible growth after 20 passages was recorded as the final concentration: 1 mg/L for ampicillin, 0.125 mg/L for cefotaxime, and 0.064 mg/L for ceftriaxone. At the end of each 4-day cycle, 20 µL of culture was spotted onto chocolate agar and incubated for 3 days at 37°C and 5% CO_2_. After incubation, the biomass of the spotted cultures was resuspended in 1 mL of HTM, and 20 µL of this resuspension was used to initiate the next cycle. This process was repeated for a total of five cycles, adding up to 20 passages in liquid medium. In the final cycle, cultures were not spotted onto chocolate agar but directly processed. Glycerol stocks (HTM with 20% glycerol) of each passage were stored at −80°C.

Single clones were isolated from each passage by dilution streaking of the respective glycerol stocks on chocolate agar. In general, two clones were isolated per passage. However, whenever the antibiotic concentration was increased compared to the previous passage, five clones were isolated instead. Isolated clones were grown in 3 mL of HTM to determine MICs of ampicillin, cefotaxime, and ceftriaxone using gradient diffusion strips (Liofilchem MIC Test Strips) and to isolate DNA using the DNeasy UltraClean Microbial Kit (Qiagen) for DNA sequencing.

Whole genome sequencing

Short-read DNA libraries were prepared from extracted genomic DNA with a modified Illumina Nextera XT library kit protocol as described previously (18). Libraries were sequenced with 150 bp paired-end reads on an Illumina NextSeq 2000 (Illumina, San Diego, CA, USA). Short-read data (fastq files) are available in the European Nucleotide Archive (ENA) under BioProject PRJEB86018. Accession numbers are available in Table S1. Mutations, i.e., single-nucleotide polymorphisms (SNPs), short insertions, and deletions, were called with a reference mapping pipeline (19). Briefly, short reads were mapped to the reference genome H. influenzae Rd KW20 (American Type Culture Collection 51907) (BioSample SAMN39831887) using the Burrows–Wheeler Aligner (BWA) (20), and allele calling was done with the Genome Analysis Toolkit (21). Mutations were called with the following thresholds: four reads mapped in both forward and reverse orientations, a minimum of eight reads with a PHRED score of ≥20 and ≥75% allele frequency. Indels in the gene hdsM (i.e., repetitive region) were excluded. For subsequent phylogenetic analysis, SNPs were concatenated when at least 95% of all isolates fulfilled the abovementioned thresholds for read coverage, PHRED score, and frequency at individual genome positions.

Long-read sequencing was performed with the PacBio Sequel II system (Pacific Biosciences, California, USA) for selected clones at different time points from each evolution experiment to identify structural genomic changes. DNA libraries were prepared with the SMRTbell Express Template Prep Kit (v.3.0). PacBio raw data files (.bam) are available at the ENA under BioProject PRJEB86018. De novo genome assembly was performed using the PacBio SMRTlink software (v.11.0) and its “Microbial Genome Analysis” application, with the genome length set to 2 Mb, a HiFi read minimum quality of 20, and a downsample coverage of 100. For the resulting assemblies, the same origin was set using the first 50 bp of dnaA. This sequence was identified using the Seqkit toolkit (v.2.9.0), and the origin was adjusted accordingly. Subsequently, each assembly was aligned to the reference genome using progressive Mauve (v.2.4.0), and the detected variations were extracted from the alignment. Indels of at least four bases in length were further examined through visual inspection of the assembly and reference mapping in these regions to characterize the mutation more precisely (tandem repeat variation, inversion, deletion, and insertion).

Long-read sequencing detected deletions and inversions in ompP2, which were not detected using Illumina sequencing. To check for the presence of structural changes affecting ompP2, the BWA alignment files of all 315 clones were checked manually for variations in ompP2 using Geneious Prime (v.2024.0.5) (https://www.geneious.com/).

Phylogenetic and Bayesian analysis

FastTree (v.2.1.9) (22) was used to calculate phylogenetic trees for each evolution experiment with a general time reversible (GTR) substitution model and Gamma20 likelihood optimization to account for evolutionary rate heterogeneity among sites and 500 resamples. FigTree (v.1.4.4) (https://github.com/rambaut/figtree/releases) was used to root phylogenetic trees at midpoint, and iTol (v.6.8.1) (https://itol.embl.de/) was used for visualization. A phylogenetic molecular clock signal was first assessed with TempEst (v.1.5) (23) by the correlation of sampling day and root-to-tip distance. We further performed a permutation-based linear regression analysis using the lmPerm (v.2.1.0) R package to test the significance of the regression model without assuming normality by computing P values through randomized permutations of the data. Mutation rates were calculated with BEAST (v.2.5) (24) for all evolution experiments, except one replicate experiment exposed to ceftriaxone with a poor correlation of SNPs and sampling time, and a non-converging Markov chain Monte Carlo sampling. We employed a coalescent exponential population model, log-normal clock rate, a strict molecular clock, tip dating (day of the experiment), a GTR substitution model, 20 million iterations with 10% burn-in, and sampling every 2,000 traces/trees. Inspection of BEAST log files with Tracer (v.1.7.2) showed an adequate mixing of the Markov chains, and all parameters were observed with an effective sample size of >200.

Competitive growth assay

Pairwise competitive growth assays were performed to estimate the in vitro relative fitness of mutants evolved in the multi-step evolution experiment compared to the parental strain (H. influenzae Rd KW20). Investigated clones were selected to represent a diverse range of mutational patterns. Briefly, five single colonies grown on chocolate agar served as inoculum for 3 mL pre-cultures that were grown until mid-exponential phase (Fig. S2). Pre-cultures were diluted to an OD_600_ of around 0.025. Equal volumes (1.5 mL) of diluted mutant and parental strain cultures were added in 15 mL falcon tubes. Co-cultures were grown in triplicate for 4 h at 37°C with shaking at 200 rpm. Initial and final colony-forming units (CFU) of the mutant and parental strain were determined by standard plate counting on HTM agar (total CFU) and on HTM agar containing 0.016 g/mL cefotaxime (mutant CFU), respectively. Even though co-cultures were inoculated to a 1:1 ratio of the OD_600_ of the two competing strains, initial CFU counts sometimes differed between the two strains. Initial CFU ratios of the parental strain to that of the mutant strain of 1.0:0.5 to 1.0:2.0 were accepted, as the formula used below accounts for unequal starting CFUs. Relative fitness was assessed as the ratio of the Malthusian parameter of the mutant to that of the parental (reference) strain. The following formula considering population sizes of the mutant and the parental strain was used for the calculation of fitness of each strain in relation to the parental strain (25):

$$[eqn]$$

The formula calculates the fitness (fitness_m_) of the mutant strain relative to the reference strain (Rd KW20). Here, A represents the population size, with subscripts m for the mutant strain and r for the reference strain. The time points 0 and 4 h indicate the population size at inoculation and after 4 h, respectively. A fitness of 1 indicates no fitness cost, whereas a ratio greater than or less than 1 indicates increased or decreased fitness compared to the parental strain, respectively.

Crystal structures

The crystal structure of the transpeptidase domain (amino acids 254–610) of PBP3 of H. influenzae was obtained from the RCSB Protein Data Bank (entry ID 6HZO deposited by Bellini et al. [26]). Detected PBP3 substitutions were visualized in this crystal structure using PyMOL (v.3.0) (https://www.pymol.org/). Ampicillin and cefotaxime were superimposed using crystal structures of PBP3 of Mycobacterium tuberculosis (entry ID 6KGW) and Staphylococcus aureus (entry ID 3VSL). The structure of full PBP3 (entry ID AF-P45059-F1-v4) was obtained from the AlphaFold Protein Structure Database.

AlphaFold Server powered by AlphaFold 3 (27) was used to predict structures of wild-type and mutated outer membrane protein P2 (OmpP2). Predicted structures were visualized using PyMOL (v.3.0).

Statistical analyses

The association between specific mutations and a reduction in susceptibility was analyzed using Mann–Whitney U tests. For this purpose, the ampicillin, cefotaxime, and ceftriaxone MIC values of two sets of evolved clones were compared. The clones in the two sets differed genetically only by the mutation or mutation combination under investigation (based on short-read sequencing approach and the manual check of ompP2 mutations in the BWA alignments), aside from detected tandem repeat variations in the hsdM gene. For each clone set, a minimum of three genetically identical clones was required. To minimize the risk of false-positive results due to multiple testing, P values were adjusted using the Benjamini–Hochberg correction. Given that all clones were derived from the same ancestral strain, we acknowledge the possibility of non-independence among observations and interpreted statistical outcomes with caution.

A one-sample t-test [considering the Malthusian parameters ln(Am, 4h/Am, 0h) and ln(Ar, 4h/Ar, 0h)] with Benjamini–Hochberg correction for multiple testing was used to assess the significance of fitness gain/loss of mutants compared to the reference strain. Pairwise t-tests with Benjamini–Hochberg correction were conducted to identify significant differences between fitness values of clones isolated from the same evolving population.

A permutation test was conducted to evaluate the association between average fitness and antimicrobial susceptibility (MIC) to ampicillin, cefotaxime, and ceftriaxone. For each antibiotic, the observed correlation between relative fitness value and MIC was calculated. The null distribution was generated by randomly shuffling the MIC values within each group and recalculating the correlation for 10,000 permutations. The P value was determined by comparing the observed correlation to the null distribution, assessing the statistical significance of the observed association.

RESULTS

Evolution of beta-lactam resistance is mainly driven by ftsI and ompP2 mutations

We exposed the reference strain H. influenzae Rd KW20 in a multi-step evolution experiment to increasing concentrations of the three beta-lactam antibiotics ampicillin, cefotaxime, and ceftriaxone. Bacterial populations evolved over 20 transfers in individual experiments, which are equivalent to approximately 200 generations. Overall, we investigated the genomes and MICs of 315 clones, isolated across time from six independent evolution experiments, including a drug-free control and three replicate experiments with ceftriaxone. We identified between 11 and 25 mutations (including SNPs and indels) in individual experiments over 20 days of exponential growth, resulting in a linear increase in the number of mutations across time. SNP-based phylogenies and root-to-tip linear regressions (R^2^ = 0.31–0.86) indicated moderate to strong support of a constant molecular clock in all evolution experiments, except for one replicate exposed to ceftriaxone (R^2^ = 0.11) (Table S3). Based on a Bayesian coalescent model, assuming a strict molecular clock and exponential population growth, the mutation rates ranged from 1.0 × 10^−7^ to 2.1 × 10^−7^ substitutions per site per day.

The MICs against the three beta-lactam antibiotics of individual clones did not change significantly in the drug-free control experiment (Fig. S3). In the drug-free control, measured MIC differences were within the range of technical variation, i.e., one doubling dilution compared to the MICs of the parental strain Rd KW20, which were 0.19 mg/L for ampicillin, 0.008 mg/L for cefotaxime, and 0.003 mg/L for ceftriaxone. We also did not detect any mutations in the main drug target ftsI in the evolved control population (Table S1).

In contrast, exposure to ampicillin, cefotaxime, and ceftriaxone resulted in a pronounced increase of beta-lactam MICs (Fig. 1). Specifically, exposure to ceftriaxone led to strong MIC increases; in one clone, the MIC increased up to 64-fold against ceftriaxone relative to the ancestor strain. Exposure to ampicillin resulted in more moderate MIC increases against all three beta-lactam antibiotics. Furthermore, populations exposed to third-generation cephalosporins showed a more pronounced MIC increase against ceftriaxone and cefotaxime as compared to ampicillin.

Increasing number of mutations and changes in beta-lactam minimum inhibitory concentrations (MICs) in six evolution experiments. Number of mutations (black) and minimum inhibitory concentration fold increase compared to the parental strain Rd KW20 for ampicillin (Amp, red), cefotaxime (Ctx, green), and ceftriaxone (Cro, blue) in selected clones over 20 passages (two or five clones per passage). Bacterial population (A) without antibiotic exposure and exposed to increasing concentrations of (B) ampicillin, (C) cefotaxime, (D) ceftriaxone replicate 1, (E) ceftriaxone replicate 2, and (F) ceftriaxone replicate 3. Shaded areas represent the 95% confidence interval.

Clinical resistance against cefotaxime (EUCAST breakpoint: >0.125 mg/L) was observed for 4 out of 55 (7.3%) clones evolved under cefotaxime exposure and for 28 out of 169 (16.6%) clones evolved in three evolution experiments under ceftriaxone exposure (Fig. 2; Fig. S3). Exposure to ceftriaxone further selected 1 out of 169 (0.6%) ceftriaxone-resistant (EUCAST breakpoint: >0.125 mg/L) clone and 10 out of 169 (5.9%) ampicillin-resistant (EUCAST breakpoint: >1 mg/L) clones. Cross-resistance against ampicillin and cefotaxime was observed for five clones evolved under ceftriaxone exposure (Fig. S3). Overall, 20 days of exposure to ampicillin did not select for clinical resistant clones against any of the three tested drugs.

Maximum likelihood phylogeny of H. influenzae Rd KW20 clones evolved in the presence of ampicillin (A, 53 clones), cefotaxime (B, 56 clones), and ceftriaxone (C, 57 clones). Amino acid substitutions in the penicillin-binding protein 3 (PBP3) and presence of mutations in specified genes as well as minimum inhibitory concentration (MIC) changes over time are color coded. Strains that exceeded the clinical breakpoint for ampicillin/cefotaxime/ceftriaxone according to EUCAST breakpoints are marked with a white star at the MIC heatmap. The specified genes are those in which at least 20 clones from the whole set of selected clones exhibited mutations but did not mutate under antibiotic-free conditions.

As expected, increasing MICs against the beta-lactam antibiotics were associated with mutations in ftsI (Fig. 2; Fig. S3). Out of 136 evolved clones that exhibited at least one ftsI mutation, 18 were resistant according to EUCAST breakpoints for cefotaxime, but none were resistant to ampicillin or ceftriaxone. Specifically, we identified combinations of the following five amino acid (amino acid) substitutions in individual clones in the transpeptidase domain: p.Thr332Ile (c.968C>T), p.Thr443Ala (c.1327A>G), p.Ala530Ser (c.1591G>T), p.Asp551Tyr (c.1652G>T), and p.Ala561Glu (c.1684C>A). Exposure to both ampicillin and cefotaxime selected for p.Ala530Ser. Under cefotaxime conditions, p.Ala530Ser, p.Thr443Ala, and p.Thr332Ile evolved in a stepwise manner. The substitutions p.Asp551Tyr and p.Ala561Glu were detected independently in separate clones, each arising contemporaneously (detected in the third passage) in one population exposed to ceftriaxone. The detected ftsI mutations were significantly associated with an MIC increase in beta-lactams (Table S4). In a crystal structure model of PBP3, all five ftsI mutations were either in proximity to other well-known resistance-associated amino acid positions (Ile449, Arg517, and Asn526) or close to the binding site of the antibiotics (Fig. 3).

Crystal structure of transpeptidase domain (residues 254–610) of the penicillin-binding protein 3 (PBP3) of H. influenzae (26) with superimposed ampicillin (A, colored in cyan) and cefotaxime (B, magenta) molecules. In vitro selected amino acid substitutions in PBP3 Thr332lle, Thr443Ala, Ala530Ser, Asp551Tyr, and Ala561Glu are in red. Known positions for amino acid substitutions defining established PBP3 resistance groups are in green.

Furthermore, ftsI mutations differentially affected the susceptibility against ampicillin, cefotaxime, and ceftriaxone. While we observed a significant MIC increase for ampicillin (twofold increase, Padj = 0.008) and ceftriaxone (threefold increase, Padj = 0.004) associated with amino acid substitution p.Ala530Ser, the secondary amino acid substitution p.Thr443Ala did not significantly affect ampicillin susceptibility but was associated with a twofold increase in the MIC for cefotaxime (Padj = 0.019) (Table S4). Out of the five detected PBP3 substitutions, p.Asp551Tyr showed the highest MIC fold increase toward all three beta-lactams with a maximum eightfold increase of the MIC of ceftriaxone (Padj = 2E-06).

Surprisingly, clones isolated from populations exposed to ceftriaxone (replicates 1 and 3) did not harbor any mutations in ftsI. However, we repeatedly selected different mutations in the ompP2, i.e., p.Thr126fs (c.376_377insT), p.Val353fs (g.902541_902708del), as well as a structural rearrangement comprising a genomic inversion starting at codon 353, resulting in p.Gly354fs (g.902541_902708inv). The latter two mutations were not detected with the short-read reference mapping approach but were identified based on the long-read assemblies. However, both mutations could also be detected in clones for which no long-read sequencing data were available through a manual check of the BWA alignments. In addition, various tandem repeat variations in different genes could be detected using the long-read sequencing data (Table S2). Mutation p.Thr126fs introduces a stop codon at position 129 in ompP2, resulting in a truncated protein of only 128 amino acids instead of 359 amino acids (Fig. S4). The other two mutations starting at codon 353 (i.e., g.902541_902708inv and g.902541_902708del) affect the C-terminus of OmpP2, specifically the last beta-strand of the beta-barrel (Fig. S5). The mutation g.902541_902708inv evolved independently in two replicate experiments. All three ompP2 mutations were significantly (Padj < 4E-05) associated with a MIC increase against all three beta-lactam antibiotics, with a more pronounced effect for ceftriaxone susceptibility (i.e., 11.5-fold MIC increase, as compared to 2.0- and 3.9-fold increases for ampicillin and cefotaxime, respectively). Overall, 104 out of 315 evolved clones carried ompP2 mutations, of which 10 were resistant to ampicillin, 14 to cefotaxime, and 1 to ceftriaxone according to EUCAST breakpoints. However, all resistant clones with ompP2 mutations harbored additional mutations, for example, in the pnp gene. Clones with ompP2 mutations only exhibited highly increased MICs, but these were below the respective EUCAST clinical breakpoints (Table S4).

Interestingly, strains harboring the inversion mutation (g.902541_902708inv) resulting in p.Gly354fs were phenotypically heterogeneous concerning beta-lactam susceptibility. Respective strains showed two ellipses for all three beta-lactams on the agar plates with gradient diffusion strips (Fig. S6). The lower MIC corresponded to the MIC observed for the parental strain ± two doubling dilutions.

Besides mutations in ftsI and ompP2, we detected 13 different mutations in pnp (encoding polyribonucleotide nucleotidyltransferase) in clones with and without pre-existing ftsI or ompP2 mutations isolated from populations exposed to cefotaxime, and in two replicated experiments under ceftriaxone. Of those 13 mutations, seven induced a premature stop codon (p.Glu18*, p.Lys47*, p.Glu80*, p.Gln159*, p.Glu220*, p.Glu471*, and p.Met601*); three resulted in a frameshift (p.Val423fs, p.Gly444fs, and p.Glu580fs); and three were non-synonymous SNPs (p.Ala440Val, p.Gln505Lys, and p.Gly626Cys). p.Glu220* and p.Glu580fs were associated with a significant (Padj <0.034) MIC increase for all three beta-lactams in clones with p.Thr126fs in OmpP2 (Table S4). The effect of the other detected mutations in pnp could not be determined due to an insufficient number of clones carrying the specific mutation pattern. Four different mutations in rpoB (encoding DNA-directed RNA polymerase subunit beta), three of them in close proximity (p.Leu149Phe, p.Arg151His, and p.Gly154Asp), were detected in clones isolated from populations exposed to ampicillin and ceftriaxone (replicate 2). Furthermore, clones derived from ampicillin, cefotaxime, and ceftriaxone (replicate 2) exposed populations evolved four different mutations in udk (encoding uridine kinase) and spoT (encoding guanosine-3′,5′-bis(diphosphate) 3′-pyrophosphohydrolase). Two mutations in prs (encoding ribose-phosphate pyrophosphokinase) were detected in ampicillin-exposed clones as well as two mutations in Rd_08000 (encoding a putative TonB-dependent receptor) in ceftriaxone-exposed clones (replicates 1 and 3). However, only minor effects were observed, with a maximum median MIC fold increase of 3 for mutations in genes other than ftsI and ompP2.

Evolved resistant mutants pay a fitness cost under drug-free conditions

To investigate possible fitness costs of specific mutants and identify putative compensatory mutations that minimize the fitness costs imposed by resistance-associated mutations, a competitive growth assay was performed. Most of the evolved clones exhibited a lower fitness compared to the parental strain (Fig. 4; Table S1), strongly indicating an evolutionary trade-off between resistance and competitive fitness in the absence of antibiotics. However, three strains (E_0.5Amp_07a, E_1Amp_17a, and E_0.032Ctx_07a) did not show a fitness/growth defect. One of these clones (E_0.5Amp_07a) harbored putative beta-lactam susceptibility-decreasing mutations in ftsI and rpoB (28–30) accompanied by a mutation in udk. Another clone (E_1Amp_17a) showed the same mutation pattern but with additional mutations in rpoA, plsB, and prs.

*Bacterial fitness of selected clones derived from evolving populations exposed to ampicillin (Amp), cefotaxime (Ctx), and ceftriaxone (Cro). Bacterial fitness measured as relative growth difference between the ancestral strain (H. influenzae Rd KW20) and the selected clone. A relative fitness of 1 indicates no fitness cost, whereas a value greater or less than 1 indicates increased or decreased fitness, respectively. The black bars represent standard deviations. *Padj < 0.05, **Padj < 0.01, **Padj < 0.001. The heatmap indicates MIC values of Amp, Ctx, and Cro (in mg/L) of the respective strain (already presented in Fig. 2). Stars indicate clinical resistance above the respective EUCAST breakpoints. Hatched panels indicate phenotypic heterogeneity. Clones that were derived from the same evolution experiment are grouped together on the y-axis.

Pairwise t-tests were conducted to compare all clones from a given evolution experiment. Particularly interesting are comparisons between clones that differ by only a single mutation, as these allow for the evaluation of fitness effects of individual mutations within a specific genetic background. Comparison of fitness assay results of E_1Amp_17a and E_1Amp_13a revealed that the rpoA mutation restored the wild-type fitness level in the respective mutational background (Table S5). Clone E_0.032Ctx_07a exhibits putative beta-lactam susceptibility-decreasing mutations in ftsI and pnp (31, 32) accompanied by mutated plsB. In this case, comparison with fitness assay results of E_0.016Ctx_05b revealed that evolution of the plsB mutation resulted in wild-type fitness in the respective mutational background. Moreover, fitness decrease could directly be addressed to and quantified for specific mutations in ftsI (p.Asp551Tyr), ompP2 (p.Thr126fs and p.Val353fs), and pnp (p.Glu580fs and p.Glu220*). Correlating MIC increases and relative fitness of evolved clones revealed a negative correlation between MIC and fitness for cefotaxime (coefficient = −0.45, P = 0.019), but no correlation between MICs and fitness for ampicillin and ceftriaxone (Table S6).

Evolved beta-lactam resistance is commonly associated with collateral sensitivity toward aminoglycosides, clarithromycin, vancomycin, and colistin

To assess the expression of collateral effects, we characterized resistance to 14 antibiotics from different classes for 17 clones isolated from the evolution experiments. We identified a consistent pattern of decreased susceptibility for beta-lactam antibiotics (Fig. 5). This also includes increased MICs of the carbapenem meropenem. In this case, two clones (E_1Amp_10b and E_1Amp_19b) with increased MICs of ampicillin, cefotaxime, and ceftriaxone expressed cross-resistance to meropenem at above EUCAST breakpoint levels. Surprisingly, two clones evolved under ceftriaxone exposure exhibited increased susceptibility (i.e., collateral sensitivity) to meropenem. These two clones additionally showed collateral sensitivity toward all non-beta-lactam antibiotics, excluding tetracycline. In general, collateral sensitivity is most often expressed toward the three considered aminoglycoside drugs (amikacin, kanamycin, and gentamicin), toward the macrolide clarithromycin, as well as the polymyxin drug colistin, possibly indicating a robust emergence of collateral sensitivity upon beta-lactam resistance evolution in H. influenzae.

Collateral effects of increased MICs against individual beta-lactam antibiotics. Heat map of log2-transformed relative increase in MIC of ampicillin, cefotaxime, ceftriaxone (underlying MIC values already presented in Fig. 1), meropenem (beta-lactams), amikacin, kanamycin, gentamicin (aminoglycosides), levofloxacin, moxifloxacin (fluoroquinolones), clarithromycin (macrolide), vancomycin (glycopeptide), rifampicin (ansamycin), colistin (polymyxin), and tetracycline (tetracycline) in mutant clones relative to the MIC of the parental strain H. influenzae Rd KW20. Selected clones were evolved in the multi-step evolution experiment. Stars indicate clinical resistance above the respective EUCAST breakpoints (not available for aminoglycosides, clarithromycin, vancomycin, and colistin). Hatched panels indicate phenotypic heterogeneity. Clones that were derived from the same evolution experiment are grouped together on the y-axis.

Moreover, the two identified phenotypically heterogeneous OmpP2 mutants consistently expressed phenotypical heterogeneity (i.e., two ellipses with a gradient diffusion test) toward a variety of antibiotics, including next to the beta-lactams also fluoroquinolones, rifampicin, and tetracycline.

DISCUSSION

This study characterizes the in vitro evolution of resistance against three clinically relevant beta-lactam antimicrobials in the opportunistic human pathogen H. influenzae. Our results revealed an approximately linear increase in the number of mutations across time, while susceptibility against the antibiotics decreased more steeply during early passages and tended to plateau in the later passages, with dynamics varying across the different beta-lactams and replicates. These increases in resistance are mainly mediated by mutations in previously characterized antimicrobial-resistant genes and, in two populations, are additionally associated with the emergence of phenotypic heterogeneity. Generally, evolved resistance against one antibiotic caused significant fitness reductions in drug-free environments and led to altered phenotypic responses to other drug classes, including numerous cases of collateral sensitivity, especially toward aminoglycosides and additional drug classes. Understanding evolutionary pathways and predicting their impact on future treatment regimens will be crucial to reduce bacterial adaptation and selection of antimicrobial resistances.

Respiratory infections with H. influenzae are usually treated with an aminopenicillin in combination with a beta-lactamase inhibitor (to overcome beta-lactamase-mediated resistance). In case of treatment failure or for severe invasive infections such as sepsis or meningitis, the use of third-generation cephalosporins such as cefotaxime and ceftriaxone is endorsed. However, it is still unclear whether cephalosporin resistance can also develop through alternative evolutionary pathways beyond the well-known PBP3 group III mutations. Furthermore, while resistance to ceftriaxone and cefotaxime is currently rare in clinical isolates (33), understanding potential evolutionary pathways to resistance and collateral effects is crucial for anticipating future threats and optimizing treatment strategies.

To address this, we performed in vitro evolution experiments by exposing H. influenzae Rd KW20 to increasing concentrations of ampicillin, cefotaxime, and ceftriaxone and monitored genomic changes. Except for p.Thr322Ile, all PBP3 substitutions that were identified in these experiments were previously reported (in combination with other amino acid substitutions) for patient-derived isolates and exhibited decreased susceptibility to cephalosporins and/or ampicillin (p.Thr443Ala [34–37], p.Ala530Ser [34, 35, 37–40], p.Asp551Tyr [34], and p.Ala561Glu [39]). Notably, none of the PBP3 ampicillin resistance group-defining mutations (p.Met377Ile, p.Ser385Thr, p.Leu389Phe, p.Ile449Veal, p.Ala502Val/Thr, p.Arg517His, and p.Asn526Lys) evolved in the evolution experiments, although p.Arg517His and p.Asn526Lys are two well-known driver mutations for ampicillin and cefotaxime resistance (41, 42). However, PBP3 p.Thr322Ile and p.Thr443Ala, which evolved under cefotaxime exposure, are in close spatial proximity to Asn526 and Arg517, respectively. Interestingly, the resulting structural changes in the beta-lactam binding site of PBP3 seem to mainly affect resistance to cephalosporins rather than ampicillin. In in vitro evolution experiments with NTHi performed by Jakubu et al. (43), one strain with wild-type ftsI evolved p.Thr443Ala in the presence of the cephalosporin cefuroxime. In accordance with our results, the respective population showed only a MIC increase for the cephalosporin but not ampicillin. Hence, cephalosporins can select for amino acid substitutions that do not affect ampicillin susceptibility.

However, we also observed convergent evolution of p.Ala530Ser under both ampicillin and cefotaxime selection pressure. Respective clones showed significantly increased MICs, especially of ampicillin and ceftriaxone, demonstrating that amino acid substitution in PBP3 can show collateral effects for beta-lactam antibiotics. Further studies that elucidate to what extent specific ftsI mutations and mutation patterns affect third-generation cephalosporin susceptibility as well as information on treatment outcome are essential to evaluate this and improve clinical decision-making.

Interestingly, we also observed clones with highly increased beta-lactam MICs that did not harbor any ftsI mutations. Ceftriaxone exposure led to the evolution of three different ompP2 mutations, resulting in a profound effect on MICs of all three beta-lactam antibiotics. OmpP2 is one of the major outer membrane proteins in H. influenzae with a mass of approximately 36–42 kDa. The protein functions as a porin with an exclusion limit of about 1.4 kDa (44, 45). Studies analyzing the ompP2 gene of H. influenzae reported a high conservation of the 16 beta-strands building up the beta-barrel but increased variation and accumulation of SNPs in the eight surface-exposed loop regions (46–49), which can result in antigenic variation (50, 51). By expression of OmpP2 and the AcrAB (multidrug efflux pump) encoding genes of H. influenzae in an acrB and ompF knockout strain of Escherichia coli, Zwarma et al. provided evidence that sensitivity to beta-lactams in H. influenzae actually is affected by OmpP2. While the efflux pump AcrAB-TolC actively transports antibiotics from the cytoplasm and periplasm out of the cell, OmpP2 counterbalances this efflux by leaking the drugs back into the cell (52).

Our findings confirm that diverse mutations in ompP2 can significantly contribute to highly increased beta-lactam MICs in H. influenzae (49, 53, 54). These ompP2 mutations seem to be particularly selected under ceftriaxone exposure. The ompP2 frameshift mutation p.Thr126fs detected in this study resulted in a stop codon at amino acid position 129 (instead of position 361) and therefore probably generated a non-functional protein. The other two mutations resulting in p.Val353fs (deletion) and p.Gly354fs (inversion) affected only the last beta-strand of the porin, though AlphaFold predicted the mutated sequence to not alter the beta-barrel structure. Therefore, decreased influx of beta-lactam antibiotics might have resulted from an altered electrostatic profile within the porin channel due to residue properties of the changed amino acids (55–57). Most studies on beta-lactam resistance in H. influenzae focus on PBP3, which makes evaluation of the clinical relevance of ompP2 mutations and their contribution to treatment failure difficult.

Additional genes, with possibly more moderate effects on ampicillin, cefotaxime, and/or ceftriaxone susceptibility, are rpoB, pnp, and spoT. Mutations in these three genes have already been associated with antibiotic resistance in other studies (58) or were selected in comparable evolution experiments with H. influenzae (43, 54), E. coli (59), and Staphylococcus aureus (60). RNA polymerase subunit β is essential for transcription, and even though specific mutations in rpoB are known to cause rifampicin resistance, the gene was also recently associated with beta-lactam susceptibility in Neisseria gonorrhoeae (28), S. aureus (28), and Bacillus subtilis (30). The highly conserved PNPase, which influences mRNA and small RNA stability, is described to affect resistance and tolerance to ciprofloxacin and aminoglycosides in Pseudomonas aeruginosa, respectively (31, 32). The pyrophosphohydrolase encoded by spoT synthesizes and hydrolyzes alarmone ppGpp, which induces the expression of stress resistance components and is involved in resistance and tolerance against different antibiotics, including beta-lactams, for example, in P. aeruginosa and Helicobacter pylori (61–63). In our study, the effect of mutations in rpoB, pnp, and spoT genes was limited to a twofold MIC increase. Therefore, one of the detected mutations alone does not result in beta-lactam resistance. However, these novel associations with decreased beta-lactam susceptibility in H. influenzae highlight the complex interplay of various genetic factors in resistance mechanisms.

Drug resistance mutations often entail a reduced fitness in the absence of antibiotic pressure (64, 65) as the antibiotic target is often involved in important cellular processes (66). Indeed, almost all of the clones examined exhibited a significant fitness deficit, with significant variability across mutants. Notably, mutations associated with a higher increase in MIC do not necessarily entail a greater fitness deficit. This is exemplified by the amino acid substitution p.Asp551Tyr in PBP3, which resulted in only a moderate fitness decrease of approximately 9% (within the wild-type genomic background) but was associated with an eightfold increase of the MIC of ceftriaxone. Fitness levels can influence the evolutionary success of a mutant. It is assumed that mutants with little or no fitness deficit tend to be more successful (67). This is supported by studies on Mycobacterium tuberculosis and Staphylococcus aureus, which discovered mutations with the lowest fitness costs to be the ones most frequently identified in resistant clinical isolates (68–70).

In addition, the frameshift mutation in ompP2 was associated with a relatively moderate fitness deficit of approximately 12%. At the same time, this mutation provided a significant advantage in the presence of the antibiotic, with a 23.5-fold increase in the MIC of ceftriaxone. In the evolution experiment, the ompP2 frameshift mutations outcompeted the inversion mutant over time, as both mutations were initially detected in the second passage, but just the frameshift mutation persisted until passage 20. However, applied bottlenecks, such as the 1:100 dilution used for passaging in this study, inherently impose a random selection, as only a small, randomly chosen subset of the growing population can proliferate in the next passage, thus increasing the effect of genetic drift. Genetic drift influences the manifestation of specific mutations, as certain variants may be lost or disproportionately represented purely by chance (71–73). A specificity of OmpP2 for NAD as well as a highly reduced uptake of NAD in ompP2 knockout strains has been described for H. influenzae (74). This might explain the reduced fitness observed for clones with mutated ompP2, as H. influenzae is dependent on extracellular NAD (75) due to the lack of enzymes necessary for de novo biosynthesis (76).

In cases of entailed fitness costs, fitness can be restored by reverting the resistance-causing mutation, which is just possible in the absence of antibiotic pressure (77), and by the acquisition of compensatory mutations (64). We identified that the p.Thr443Ala substitution in PBP3 mitigates the fitness deficit caused by p.Ala530Ser in PBP3 (and p.Val423fs in pnp), while this amino acid substitution is also associated with a slight increase in the MIC of third-generation cephalosporins. Such mutations likely present an increased risk of being selected in the presence and the absence of the antibiotic.

Resistance-conferring mutations can additionally affect susceptibility to other antibiotics. Simultaneously acquired resistance to another antibiotic is referred to as cross-resistance, while collateral sensitivity describes an evolutionary trade-off where antibiotic resistance to one antibiotic confers increased susceptibility to another antibiotic. Robust collateral sensitivity phenotypes could help to develop new evolutionary-based treatment strategies and therefore may reduce resistance emergence (78).

We observed collaterally increased MICs of fluoroquinolones in ompP2 mutated clones with decreased beta-lactam susceptibility. This influence of OmpP2 on fluoroquinolone susceptibility in H. influenzae is supported by evolution experiments with clinical isolates of H. influenzae in the presence of moxifloxacin that selected for ompP2 mutations and MICs exceeding CLSI breakpoints for moxifloxacin (54). In our experiments, MICs did not exceed EUCAST clinical breakpoints for moxifloxacin (>0.125 mg/L) and levofloxacin (>0.06 mg/L). However, the success of treatment with fluoroquinolones might be reduced in these cases. For the macrolide clarithromycin, we observed differential effects in the ompP2 mutants. These observations may be clinically relevant, as H. influenzae could be treated with co-trimoxazole, macrolides, fluoroquinolones, or tetracycline in cases of known resistance or intolerance to beta-lactam antibiotics (79).

Overall, our study identified the emergence of collateral sensitivity toward the three tested aminoglycosides against the macrolide clarithromycin, the glycopeptide vancomycin, and also the polymyxin drug colistin. The reduced MIC to vancomycin in one evolution experiment was particularly surprising, as this antibiotic is only active against gram-positive bacteria. This effect might be linked to the detected mutation in the mltC gene, as increased susceptibility to vancomycin under acidic pH has been observed in a MltC/MltE deletion mutant of Escherichia coli (80). Overall, these patterns of evolved collateral sensitivity were observed for more than 50% of the tested clones. Even though they did not emerge in all cases, collateral sensitivities seem to occur frequently toward specific antibiotics and may thus be a promising focus for optimizing antibiotic treatment designs.

The therapeutic potential of such evolutionary trade-offs clearly deserves further study, especially using combinatorial and sequential treatments with a beta-lactam and an aminoglycoside or clarithromycin. Unfortunately, due to the ototoxicity and nephrotoxicity, which are common side effects of aminoglycosides, a potential combinatorial or alternating treatment would not be appropriate for simple respiratory tract infections caused by H. influenzae (81), but rather of value for severe infections like meningitis or sepsis. In fact, beta-lactam–aminoglycoside combinations are used in cases like neonatal meningitis or severe sepsis of unknown etiology, where H. influenzae is only rarely the causative pathogen. Nevertheless, in such cases, according to our findings, this therapy could be effective.

Limitations

H. influenzae Rd KW20, a well-characterized reference strain, was used as the starting point for all evolution experiments. This represents a limitation, as evolutionary trajectories and phenotypic effects of mutations may vary by genetic background (82). While our findings offer a valuable foundation, the generalizability of resistance-associated fitness and collateral effects should be confirmed in diverse clinical isolates. Future studies incorporating the detection of resistance mutations identified here will be essential to assess their broader clinical relevance across diverse genetic backgrounds.

Conclusion

Overall, this study revealed the fast evolution of beta-lactam resistance in the opportunistic human pathogen H. influenzae. Evolved resistance is mediated by an accumulation of mutations in known target genes and in two populations associated with the emergence of heterogeneity in susceptibility. Almost all resistant clones suffered from reduced competitive fitness in the absence of antibiotics and often expressed collateral sensitivity toward other drug classes, especially aminoglycosides. These findings indicate robust evolutionary trade-offs that are caused by the resistance-conferring mutations and may represent a promising focus for optimizing antibiotic treatment designs. Similarly, a further analysis of novel resistance mutations and of epistatic effects may additionally assist in improving the development of new antimicrobial treatment strategies.