Strain-specific Growth Parameters are Important to Accurately Model Bacterial Growth on Baby Spinach in Simulation Models.

Digital tools to predict produce shelf life have the potential to reduce food waste and improve consumer satisfaction. To address this need, we (i) performed an observational study on the microbial quality of baby spinach, (ii) completed growth experiments of bacteria that are representative of the baby spinach microbiota, and (iii) developed an initial simulation model of bacterial growth on baby spinach. Our observational data showed that the predominant genera found on baby spinach were Pseudomonas, Pantoea and Exiguobacterium. Rifampicin resistant mutants (rifR mutants) of representative bacterial subtypes were subsequently generated to obtain strain-specific growth parameters on baby spinach. These experiments showed that: (i) it is difficult to select rifR mutants that do not have fitness costs affecting growth (9 of 15 rifR mutants showed substantial differences in growth, compared to their corresponding wild type strain) and (ii) based on estimates from primary growth models, the mean (geometric) maximum population of rifR mutants on baby spinach (7.6 log10 CFU/g, at 6˚C) appears lower than that of the spinach microbiota (9.6 log10 CFU/g, at 6˚C), even if rifR mutants did not have substantial growth-related fitness costs. Thus, a simulation model, parameterized with the data obtained here as well as literature data on home refrigeration temperatures, underestimated bacterial growth on baby spinach. The root mean square error of the simulation's output, compared against data from the observational study, was 1.11 log10 CFU/g. Sensitivity analysis was used to identify key parameters (e.g., strain maximum population) that impact the simulation model's output, allowing for prioritization of future data collection to improve the simulation model. Overall, this study provides a roadmap for the development of models to predict bacterial growth on leafy vegetables with strain-specific parameters, and suggests that additional data is required to improve these models.