Evaluating the predictive power of combined gene expression dynamics from single cells on antibiotic survival

Heteroresistance and persistence are examples of mechanisms that can allow otherwise drug-susceptible bacteria to survive and resume growth after antibiotic exposure. These temporary forms of antibiotic tolerance can be caused by the upregulation of stress response genes or a decrease in cell growth rate. However, it is not clear how the expression of multiple genes contributes to tolerance phenotypes. Using fluorescent reporters for stress-related genes, we conducted real-time measurements of expression prior to, during, and after antibiotic exposure. We first identified relationships between growth rate and reporter levels based on auto- and cross-correlation analysis, revealing consistent patterns where changes in growth rate were anticorrelated with fluorescence following a delay. We then used pairs of stress gene reporters and time-lapse fluorescence microscopy to measure the growth rate and reporter levels in cells that survived or died following antibiotic exposure. Using these data, we asked whether combined information about reporter expression and growth rate could improve our ability to predict whether a cell would survive or die following antibiotic exposure. We developed a Bayesian inference model to predict how the combination of dual reporter expression levels and growth rate impacts ciprofloxacin survival in Escherichia coli. We found clear evidence of the impact of growth rate and gadX promoter activity on survival. Unexpectedly, our results also revealed examples where additional information from multiple genes decreased prediction accuracy, highlighting an important and underappreciated effect that can occur when integrating data from multiple simultaneous measurements.IMPORTANCETransient increases in bacterial antibiotic tolerance can result in treatment failure despite an infection initially presenting as susceptible, presenting a significant challenge in antibiotic therapy. This phenomenon can also provide a window of opportunity for bacteria to acquire permanent genetic resistance mutations. Although understanding the underlying mechanisms of these antibiotic tolerance phenotypes is crucial for developing effective approaches to treatment, current approaches for studying these transient phenotypes have limitations. Here, we use fluorescent reporters to monitor the expression of genes involved in stress response over time, aiming to link expression with antibiotic survival outcomes. Our results reveal a counterintuitive finding: monitoring multiple gene reporters does not necessarily improve our ability to predict antibiotic survival outcomes compared to single gene reporters. This result emphasizes the need for a deeper mechanistic understanding of the relationship between stress response gene expression and antibiotic tolerance.