Antibiotic combination therapy is used clinically to broaden antibiotic coverage whilst mitigating the evolution of resistance, as such, it is the recommended first line of treatment in time-sensitive infections such as sepsis. Of particular interest is the use of combinatorial treatment that exploit the phenomena of collateral sensitivity, whereby decreased susceptibility towards one drug is associated with increased susceptibility towards a second drug. Of even greater clinical significance is reciprocal collateral sensitivity where this interaction is bidirectional, and this is therapeutically useful as bacteria become trapped in a cycle that ensures that they always remain susceptible to at least one of the drugs used in treatment, even if they develop resistance to the other. However, whilst collateral sensitivity is frequently measured within in vitro laboratory experiments, it is an understudied concept and there is scant evidence within clinical data, partially due to the lack of curated hospital MIC datasets. Now, a small number of studies exploring clinical MIC data do exist, for example a recent study of 448K antibiotic susceptibility test results from hospitals in Pittsburgh concluded that reciprocal collateral sensitivity is rare, whilst concurrent resistance (i.e. dual resistance and dual sensitivity) is common. We plan to analyse the ATLAS dataset to assess whether this scarcity of reciprocal collateral sensitivity is mirrored on a larger, global dataset, and where collateral interactions are identified, we hope to evaluate their clinical relevance. We will quantify these drug pair dependencies as mutual information scores, as well as utilising the unique feature of ATLAS, raw MIC data, to explore the directionality of MIC associations between drug pairs. We know that collateral sensitivity can be heterogenous amongst strains of the same species, even amongst replicates of the same bacterial isolate, and so we will also quantify cases of reciprocal collateral sensitivity within subpopulations that are conditioned on resistance towards a single antibiotic. Additionally, it is possible that antibiotic triplicates are required for collateral sensitivity rather than just pairs, and we will address this question using a Markov random field. Finally, we aim to identify patterns of dual sensitivity and dual resistance, both on a species level and when species identify is unknown (i.e. when treatment is empirical) to identify antibiotic pairs that are optimal for the avoidance of cross-resistance.
Center for Global Research Data