Machine Learning Establishes Single-Cell Calcium Dynamics as an Early Indicator of Antibiotic Response

Abstract

Changes in bacterial physiology necessarily precede cell death in response to antibiotics. Herein we investigate the early disruption of Ca²⁺ homeostasis as a marker for antibiotic response. Using a machine learning framework, we quantify the temporal information encoded in single-cell Ca²⁺ dynamics. We find Ca²⁺ dynamics distinguish kanamycin sensitive and resistant cells before changes in gross cell phenotypes such as cell growth or protein stability. The onset time (pharmacokinetics) and probability (pharmacodynamics) of these aberrant Ca²⁺ dynamics are dose and time-dependent, even at the resolution of single-cells. Of the compounds profiled, we find Ca²⁺ dynamics are also an indicator of Polymyxin B activity. In Polymyxin B treated cells, we find aberrant Ca²⁺ dynamics precedes the entry of propidium iodide marking membrane permeabilization. Additionally, we find modifying membrane voltage and external Ca²⁺ concentration alters the time between these aberrant dynamics and membrane breakdown suggesting a previously unappreciated role of Ca²⁺ in the membrane destabilization during Polymyxin B treatment. In conclusion, leveraging live, single-cell, Ca²⁺ imaging coupled with machine learning, we have demonstrated the discriminative capacity of Ca²⁺ dynamics in identifying antibiotic-resistant bacteria.

Keywords: E. coli; aminoglycosides; antibiotic resistance; calcium; machine learning; polymyxin B.