Antibiotics Drive Expansion of Rare Pathogens in a Chronic Infection Microbiome Model

Abstract

Chronic (long-lasting) infections are globally a major and rising cause of morbidity and mortality. Unlike typical acute infections, chronic infections are ecologically diverse, characterized by the presence of a polymicrobial mix of opportunistic pathogens and human-associated commensals. To address the challenge of chronic infection microbiomes, we focus on a particularly well-characterized disease, cystic fibrosis (CF), where polymicrobial lung infections persist for decades despite frequent exposure to antibiotics. Epidemiological analyses point to conflicting results on the benefits of antibiotic treatment yet are confounded by the dependency of antibiotic exposures on prior pathogen presence, limiting their ability to draw causal inferences on the relationships between antibiotic exposure and pathogen dynamics. To address this limitation, we develop a synthetic infection microbiome model representing CF metacommunity diversity and benchmark on clinical data. We show that in the absence of antibiotics, replicate microbiome structures in a synthetic sputum medium are highly repeatable and dominated by oral commensals. In contrast, challenge with physiologically relevant antibiotic doses leads to substantial community perturbation characterized by multiple alternate pathogen-dominant states and enrichment of drug-resistant species. These results provide evidence that antibiotics can drive the expansion (via competitive release) of previously rare opportunistic pathogens and offer a path toward microbiome-informed conditional treatment strategies. IMPORTANCE We develop and clinically benchmark an experimental model of the cystic fibrosis (CF) lung infection microbiome to investigate the impacts of antibiotic exposures on chronic, polymicrobial infections. We show that a single experimental model defined by metacommunity data can partially recapitulate the diversity of individual microbiome states observed across a population of people with CF. In the absence of antibiotics, we see highly repeatable community structures, dominated by oral microbes. Under clinically relevant antibiotic exposures, we see diverse and frequently pathogen-dominated communities, and a nonevolutionary enrichment of antimicrobial resistance on the community scale, mediated by competitive release. The results highlight the potential importance of nonevolutionary (community-ecological) processes in driving the growing global crisis of increasing antibiotic resistance.

Keywords: antibiotic resistance; chronic infection; cystic fibrosis; experimental microbiome; infection microbiome.

FIG 1

Schematic outline of the CF meta-community approach. All experiments are derived from a 10-species menu that captures the majority of CF microbiome diversity across a cohort of 77 people with CF (57) and is consistent with microbiome content across the CF literature (15–19, 21–24). The 10-species meta-community is exposed to 10 treatments (in 5× replication) and propagated for 5 serial passages. The experimental design results in 250 individual synthetic microbiome observations.

FIG 2

The 5-fold replicated synthetic CF microbiomes converge toward a single stable state in the absence of antibiotic perturbations. Five replicate synthetic microbiomes were grown anaerobically in artificial sputum medium. The community composition was estimated by 16S rRNA gene amplicon sequencing at time 0 and at every 2-day passage (x axes) into fresh medium (10% transfer of 2 mL culture volume). The colored bars represent the relative abundance of each species in the community (left y axis), while the black line represents the total bacterial abundance per mL (right y axis, log scale). Each panel represents a separate replicate experiment. Strain information is provided in Table 1 (our default P. aeruginosa strain is mucoid PDO300).

FIG 3

Varying the pathogen composition has minimal impact on community composition. Each panel represents the average of 5 replicates in the absence of antibiotics; the mucoid PA with SA panel is the average from Fig. 1. Figure details are the same as those described for Fig. 1. Data on individual replicates per treatment are presented in Fig. S2. Mucoid and nonmucoid PA, P. aeruginosa strains PAO1 and PDO300, respectively. SA, S. aureus.

FIG 4

Antibiotic treatments produce large community fluctuations and alternative community states. Columns represent distinct antibiotic treatments (the first “no antibiotics” control column is reproduced from Fig. 1), and rows represent 5 replicates. The left axes measure community composition (bar charts); the right axes measure total bacterial abundance per mL (black lines). Experimental procedures, sampling, and analysis were performed as described for Fig. 1. Fresh antibiotics were resupplemented at each passage. Total abundance data by species are presented for each treatment and time point in Fig. S3.

FIG 5

Absolute pathogen densities are variable and often increased under antibiotic exposures. Each dot corresponds to the fold change difference of an individual replicate of species-specific final time point absolute density under defined antibiotic treatments, compared to the mean value of the no-antibiotic control (data redrawn from Fig. 3). Mer, meropenem; cip, ciprofloxacin; tob, tobramycin. Asterisks denote significantly higher final densities in the presence of antibiotic compared to antibiotic-free controls (competitive release; one-tailed Wilcoxon test; *, P < 0.05; **, P < 0.01).

FIG 6

Antibiotic resistance testing does not consistently predict species presence/absence in a community context. For each species-drug combination, we assessed predicted survival (MIC in rich medium [Table S2] > experimental concentration) and observed survival (relative abundance of at least 1% averaged across all five replicates at the final time point [Fig. 4]). True-positive cases (predicted and observed present) are coded in gray; true negatives (predicted and observed absent) are in white. False positives (predicted present, observed absent; evidence for competition) are in red, and false negatives (predicted absent, observed present; evidence for facilitation) are in blue. Species order was determined through clustering via stringdist (118). In Discussion we address the caveat that single-species MIC measures are taken under distinct growth conditions.

FIG 7

S. aureus growth in meropenem is facilitated by coculture with B. cenocepacia. Experiments were conducted in rich medium (Tryptone Soya Yeast Extract [TSYE] broth) in room air, in the presence or absence of 10 μg/mL meropenem and for each species either grown alone (monococulture) or together (coculture) in a 96-well plate with hourly shaking. At 0 and 48 h, cells were serially diluted and plated at concentrations of 10⁻² to 10⁻⁷ onto either mannitol salt agar (for S. aureus) or LB agar with 500 mg/L gentamicin (for B. cenocepacia).

FIG 8

Drug-resistant pathogens are consistently enriched as a functional class across all drug treatments. Fold change differences for the sum of drug-resistant pathogens (B. cenocepacia, A. xylosoxidans, and S aureus) compared to no-antibiotic control; details as in Fig. 4. Asterisks mark significant competitive release; one-tailed Wilcoxon test; *, P < 0.05; **, P < 0.01.

FIG 9

Antibiotics drive pathogen enrichment in experimental microbiomes, producing community structures that overlap clinical sputum communities. PCA visualization of experimental microbiome data (colored triangles and squares, summarizing data in Fig. 4) plus clinical microbiome data across a cohort of 77 people with CF (gray/black circles, black/severe signifies low lung function [57]). (A) Squares illustrate experimental initial conditions. (B and C) Triangles are final compositions after 5 serial passages (10 days), in the absence (B) or presence (C) of antibiotics. Colors denote experimental condition (see key). Each experimental treatment is replicated 5-fold, producing highly repeatable dynamics in the absence of antibiotics (blue triangles, B) and variable pathogen enriched outcomes following antibiotic treatment (C). Antibiotics were supplemented at each passage at clinically relevant concentrations (meropenem, 15 μg/mL; tobramycin, 5 μg/mL; ciprofloxacin, 2.5 μg/mL). Each point is a single microbiome sample (species resolution for clinical samples via the DADA2 plugin in QIIME2 [57, 119]). Ordination is the PCA of centered log-ratio transformed relative abundances.

FIG 10

Most endpoint experimental taxa fall within the range of clinically observed relative frequencies. The relative abundances of taxa in synthetic microbiome inocula and endpoints (30 samples) compared to 77 clinical cohort observations (57). The box represents the interquartile range (from 25% to 75% of samples), with the horizonal line at the median. Outliers are represented as dots (two-tailed Welch’s t test versus clinical data with Bonferroni multiple testing correction; *, corrected P < 0.001).