Genome-wide analysis captures the determinants of the antibiotic cross-resistance interaction network

Abstract

Understanding how evolution of antimicrobial resistance increases resistance to other drugs is a challenge of profound importance. By combining experimental evolution and genome sequencing of 63 laboratory-evolved lines, we charted a map of cross-resistance interactions between antibiotics in Escherichia coli, and explored the driving evolutionary principles. Here, we show that (1) convergent molecular evolution is prevalent across antibiotic treatments, (2) resistance conferring mutations simultaneously enhance sensitivity to many other drugs and (3) 27% of the accumulated mutations generate proteins with compromised activities, suggesting that antibiotic adaptation can partly be achieved without gain of novel function. By using knowledge on antibiotic properties, we examined the determinants of cross-resistance and identified chemogenomic profile similarity between antibiotics as the strongest predictor. In contrast, cross-resistance between two antibiotics is independent of whether they show synergistic effects in combination. These results have important implications on the development of novel antimicrobial strategies.

Figure 1. Cross-resistance interactions and their general properties.

(a) Network of cross-resistance interactions. Antibiotics are grouped according to their mode of action. An arrow from antibiotic A to antibiotic B indicates that adaptation to A decreased sensitivity to B in at least 50% of the evolved populations. (b) Distribution of the strength of cross-resistance interactions, as estimated by E-tests. (c) Cross-resistance interaction degrees of antibiotics. In-degree measures the number of antibiotic treatments which select for increased resistance against a given antibiotic while out-degree is defined as the number of antibiotics to which cross-resistance evolves when adapting to a given drug. The data are based on that of a. (d) The frequency of cross-resistance interactions between antibiotics is independent of whether they show physiological interactions (that is, synergy or antagonism), P=0.35, N=45, Kruskal–Wallis test. Aminoglycosides are excluded from the analysis as they show an especially large number of synergistic interactions and are strongly depleted in cross-resistance interactions with other antibiotics). Box plot presents the median and first and third quartiles, with whiskers showing either the maximum (minimum) value or 1.5 times the interquartile range of the data, whichever is smaller (larger).

Figure 2. Mutations identified in independently evolved lines.

Distribution of mutational events according to antibiotic (a), type (b) and size of DNA deletions (c). Laboratory-evolved mutator lines have accumulated exceptionally large numbers of mutations. The total number of putative loss-of-function mutations among point mutations, insertions and small deletions is 27% (b). (d) Observed mutations and known antibiotic resistance genes. Genes mutated in evolved lines are more likely to show significant sequence similarity to known antibiotic resistance genes than non-mutated ones (28 out of 143 versus 120 out of 4,358, P<10⁻¹⁴, Fisher’s exact test). Furthermore, genes showing sequence similarity to known resistance genes are enriched among genes mutated in multiple lines compared with those mutated in a single line (17 out of 47 versus 11 out of 96, P<0.005, Fisher's exact test). We identified genes showing significant sequence similarity to a set of genes curated in the Comprehensive Antibiotic Resistance Database using BLASTP search. In brief, we used the standalone NCBI BLASTP+ tool to identify E. coli genes that show highly significant similarity to any of the curated resistance or target genes (a conservative E-value cutoff of 10⁻³⁰ was applied).

Figure 3. Parallel evolution and cross-resistance.

(a) Mutational profiles of the 12 antibiotic selection regimes. Only those genes are shown that mutated in two or more of the 61 sequenced non-mutator laboratory-evolved lines. Mutations in promoters of multi-genic operons were associated with all genes encoded by the operon. The colour code indicates the number of cases when the same gene was independently mutated in different lines evolved under the same antibiotic pressure. (b) Heatmap of the average mutation profile similarity of two strains adapted to different (off-diagonal elements) and identical (diagonal elements) antibiotics. Mutation profile similarity between each pair of evolved lines was estimated by the Jaccard’s coefficient between their sets of mutated genes. Note that the map is symmetric. (c) Very-low average mutation profile similarities between strains adapted to different antibiotics are associated with low cross-resistance frequencies between antibiotic pairs. Mutation profile similarity was calculated as in b. Antibiotic pairs with mutation profile similarities <0.01 show significantly lower cross-resistance frequencies than the rest of the pairs (P<10⁻¹⁰, N=66, Wilcoxon rank-sum test), even when aminoglycosides are excluded (P<0.005, N=45). Dashed red curve indicates a smooth curve fitted by Loess regression (using the local polynomial regression fitting function of R).

Figure 4. Antibiotic properties and cross-resistance.

(a) Weak association between chemical structural similarity between antibiotic pairs and cross-resistance frequency (Spearman’s ρ=0.40, P<10⁻³, N=66), which disappears when aminoglycosides are excluded (ρ=0.21, P=0.18, N=45). Structural similarity between antibiotics was estimated by the Tanimoto similarity of their molecular fingerprints. (b) Correlation between chemogenomic profile similarity and overlap in the set of accumulated mutations during laboratory evolution (Spearman’s ρ=0.67, P<10⁻⁵, N=36). (c) Antibiotic pairs that frequently display cross-resistance interactions show relatively high overlap in their chemogenomic profiles (Spearman’s ρ=0.77, P<10⁻⁷, N=36). Dashed red curves on scatterplots A–C indicate smooth curves fitted by Loess regression. (d) Predicting antibiotic resistance phenotypes from genome sequences. Prediction performance for each antibiotic based on the set of accumulated mutations was measured by the area under the receiver operating characteristic (ROC) curve (AUC). This gives an overall measure of accuracy by taking into account both true positive and false positive rates across all possible cutoffs of the prediction score. Random prediction gives an AUC of 0.5. Variation in resistance among evolved strains can be predicted with 55–88% (76% average) accuracy, depending on the antibiotic studied. Special care was taken to avoid circularity in the predictions.