Collateral sensitivity identified in surveillance data shows limited reproducibility in clinical isolates

Although collateral sensitivity has emerged as a potential strategy for combating antimicrobial resistance, Sebastian T Tandar and colleagues had identified collateral sensitivity in only 3·0% of species–antibiotic pairs through the analysis of multicentre antimicrobial surveillance datasets comprising over 5 million minimum inhibitory concentration measurements.1 Specifically, collateral sensitivity pairs of several clinically important multidrug-resistant pathogens, including Acinetobacter baumannii and Pseudomonas aeruginosa, were associated with a higher colistin minimum inhibitory concentration. These findings highlighted the rarity of collateral sensitivity in clinical surveillance data but suggested relevance for guiding novel antimicrobial strategies.
To assess the translational applicability of these collateral sensitivity pairs, we experimentally determined the minimum inhibitory concentrations in clinical isolates of A baumannii2 and P aeruginosa3 using standard broth microdilution assays. Among the isolates tested, only one A baumannii strain (52082) showed a collateral sensitivity pattern, characterised by an elevated colistin minimum inhibitory concentration and increased susceptibility to levofloxacin (appendix p 1). However, no reproducible collateral sensitivity was observed for antibiotic pairs tested across the remaining isolates, and high colistin minimum inhibitory concentrations were not associated with increased susceptibility to other antibiotics (appendix p 1).
Several factors could explain why our experimental observations did not fully corroborate surveillance-derived associations. Analysing large and heterogeneous surveillance datasets might generate statistical correlations that are not conserved at the level of individual strains. Collateral sensitivity is highly dependent on context and is evolutionarily dynamic, owing to variations in genetic background, selection history, and temporal effects.4 Each clinical isolate might differ substantially in terms of evolutionary trajectories, with previous exposures and compensatory mutations that obscure collateral sensitivity effects. Consequently, collateral sensitivity relationships might be transient, rendering them challenging to reproduce outside controlled conditions. In our previous experimental evolution study of Enterococcus faecalis, we found that collateral sensitivity was difficult to replicate and led to different outcomes, depending on the method used.5 Finally, minimum inhibitory concentration-based surveillance lacks mechanistic validation, limiting the causal interpretation of inferred collateral sensitivity interactions.
Hence, our findings suggest that although collateral sensitivity can be observed in individual clinical isolates, its occurrence is inconsistent and not generalisable. We highlight the need for integrating experimental validation of clinical isolates with large-scale surveillance analyses before translating collateral sensitivity-based strategies into clinical practice. Surveillance data provide valuable hypotheses, but multiple factors, including strain-level heterogeneity and evolutionary dynamics, are likely to constrain the clinical utility of collateral sensitivity-guided therapies.