Interaction networks, ecological stability, and collective antibiotic tolerance in polymicrobial infections

Abstract

Polymicrobial infections constitute small ecosystems that accommodate several bacterial species. Commonly, these bacteria are investigated in isolation. However, it is unknown to what extent the isolates interact and whether their interactions alter bacterial growth and ecosystem resilience in the presence and absence of antibiotics. We quantified the complete ecological interaction network for 72 bacterial isolates collected from 23 individuals diagnosed with polymicrobial urinary tract infections and found that most interactions cluster based on evolutionary relatedness. Statistical network analysis revealed that competitive and cooperative reciprocal interactions are enriched in the global network, while cooperative interactions are depleted in the individual host community networks. A population dynamics model parameterized by our measurements suggests that interactions restrict community stability, explaining the observed species diversity of these communities. We further show that the clinical isolates frequently protect each other from clinically relevant antibiotics. Together, these results highlight that ecological interactions are crucial for the growth and survival of bacteria in polymicrobial infection communities and affect their assembly and resilience.

Keywords: antibiotics; infection; microbiology; systems biology.

Fig. 1.

Quantifying ecological interactions between bacteria isolated from polymicrobial UTIs. (A) Schematic illustrating a bladder environment infected with a community consisting of different bacterial species. (B) Preparation of conditioned medium for interaction experiments. Bacteria were grown in AUM for 48 h, followed by centrifugation, filtration, and replenishment with fresh nutrients. (C) Interaction measures from growth in conditioned medium. The growth in the conditioned medium is compared with the growth in the reference medium (black trace) (SI Appendix) for each of the 72 isolates. For positive interactions (red trace), the growth yield in conditioned medium (N_c) exceeds the growth in the reference medium (N_u), ε = log(N_c/N_u) > 0. For negative interactions (blue trace), the growth in the conditioned medium (N_c) is exceeded by the growth in the reference medium (N_u), ε = log(N_c/N_u) < 0. (D) Examples of growth curves in triplicate of Enterococcus faecalis (Upper) and Citrobacter koseri (Lower) in medium conditioned by C. koseri (colored traces) and reference medium (black traces). The growth of E. faecalis is positively affected by the conditioned medium (red curve above black curve), whereas C. koseri is negatively affected by medium conditioned by another isolate from the same species (blue curve below black curve).

Fig. 2.

The global UTI interaction network reflects species phylogeny and is enriched for competition but depleted of exploitation. (A) Pairwise interaction matrix depicting the interactions in yield (maximum reached OD₆₀₀) of 72 UTI isolates in conditioned medium prepared from these same isolates; the interaction measure is ε = log(N_c/N_u) (SI Appendix). The acceptors (columns) are grown in the conditioned medium of the donors (rows). Interactions cluster according to phylogeny, as can be seen from the 16S rDNA phylogenetic tree on the left. The isolates are symmetrically ordered on the horizontal and vertical axes. Ecoli, E. coli; Ent, Enterococcus spp. (E. faecalis and E. faecium); KECS, Klebsiella spp. (K. pneumoniae, K. oxytoca), Enterobacter cloaca, Citrobacter koseri, Serratia liquefaciens, and Pantoea sp4; Pm, P. mirabilis; Ps, Pseudomonas spp. (P. aeruginosa, P. fluorescens); St, Staphylococcus spp. (S. aureus, S. haemolyticus, S. capitis); one isolate belonging to Morganella morganii is located between Pseudomonas and P. mirabilis. The lower left to upper right diagonal depicts the self-interactions. (B) Statistical analysis of phylogenetically clustered interactions. Donors are ordered on the vertical axis and acceptors on the horizontal axis. Red and blue correspond to respective positive and negative interactions, respectively; dark and light gray backgrounds depict significant over- or underrepresentation of the most prominent interactions compared with randomized matrices. A white background depicts nonsignificant over- or underrepresentation (SI Appendix). Statistics on Morganella morganii are omitted due to small sample size (n = 1). (C) Statistical analysis of all two-node motifs in the measured interaction network. Reciprocal interactions with the same sign (−/− and +/+) and commensalism (+/0) are overrepresented, whereas amensalism (−/0) and exploitation (+/−) are depleted. Z scores were calculated based on the comparison with an ensemble of random networks preserving the degree of distribution (SI Appendix).

Fig. 3.

Polymicrobial UTI communities are often ecologically stable, and the number of taxa per community is close to the limit of stability. (A) Interactions within complete communities compared with interactions in an ensemble of randomized communities (SI Appendix). Competitive interactions (−/−) are slightly enriched, while exploitation (+/−) is slightly depleted in communities. No cooperative interactions (+/+) occur in the communities (SI Appendix). (B) Intracommunity microbial interactions in four stable communities. Each depicted complete community consists of all taxa that were isolated from one host. Red arrows indicate positive interactions; blue arrows indicate negative interactions. Eco, E. coli; other abbreviations are as in Fig. 2. (C) Examples of population dynamics for one of the stable communities with one fixed point in the absence of voiding. Lines of the same color show independent simulations starting from different initial conditions (SI Appendix); the initial population size and inoculation order do not affect the final population size. (D) Fraction of stable randomly assembled communities as a function of the number of taxa in a community. The stability of the communities drops with increasing number of taxa. The observed number of four to five taxa is close to the typical number of taxa of randomly assembled stable communities. The black asterisk indicates the number of taxa and fraction of observed coexisting communities. (E) Steady-state population size for different UTI species as a function of the effective dilution rate for a stable community (C and SI Appendix). The effective dilution rate (mimicking repeated voiding of the bladder) markedly affects the stability of the community. The gray band shows the voiding patterns ranging from a healthy bladder to a bladder with an increased postvoid residual urine volume (SI Appendix, Fig. S9).

Fig. 4.

Bacteria in polymicrobial UTIs frequently protect each other from antibiotics. (A) Histogram of the change in tolerance of 72 isolates in 14 conditioned media to the trimethoprim-sulfamethoxazole combination (SI Appendix, Fig. S10A). (B) As in A, but for nitrofurantoin. (C) Interaction matrix depicting the effect on the tolerance to the trimethoprim-sulfamethoxazole combination of 14 conditioned media on 72 isolates in conditioned medium (SI Appendix). Abbreviations are as in Fig. 2.