Mechanistic insights into cell-envelope damage to inform controllable ultrasonic inactivation of Escherichia coli: Integrating multi-field effects and machine learning.

Ultrasound is a promising non-thermal preservation technology, but limited mechanistic understanding of bacterial inactivation and viable but non-culturable (VBNC) formation-a state in which bacteria lose culturability but retain metabolic activity and resuscitation potential-constrains controllable process design. This study investigated Escherichia coli ATCC 25922 inactivation under multifield ultrasound (power density, temperature, pH, pressure), monitoring culturability, PMA-qPCR-estimated VBNC, membrane permeability, oxidative stress, lipid remodeling, and enzyme leakage. Machine learning (random forest, partial least squares regression, support vector regression) mapped processing parameters to survival and membrane damage endpoints, identifying key biochemical predictors of envelope integrity. VBNC formation peaked at moderate intensity, showing a non-monotonic trend; higher intensity and temperature favored direct lethality. Temperature and treatment time were the dominant factors driving culturability loss and VBNC formation; lipid peroxidation and alkaline phosphatase leakage were associated with outer- and inner-membrane permeability, respectively. Random forest models predicted culturable cells, VBNC estimates, and permeability markers with high internal accuracy (test R2 > 0.91), though external validation is needed. In commercial bayberry juice, three RF-screened thermosonication conditions reduced culturable E. coli below detection, but PMA-qPCR revealed residual VBNC, and no recovery occurred after 120 days at 4 °C. Overall, this study provides a mechanism-informed parameter- screening framework for tunable ultrasound processing and underscores the necessity of VBNC-aware detection and product-specific validation.