Modeling and predicting growth and growth boundary of Bacillus cereus s.l. from phylogroups II, IV, V, and VI in starchy foods at or below 12°C.

Pathogenic Bacillus cereus s.l. can survive cooking of starchy foods and grow at chilled storage temperatures, highlighting foods with extended chilled shelf life as a risk factor. Some food administrations encourage use of predictive microbiology to support decisions of safe shelf lives. Therefore, the present study embarked on identifying a model from literature and/or expanding an existing model to enable accurate predictions of growth and no-growth responses of relevant B. cereus s.l. in starchy ready-to-eat and ready-to-cook foods when stored at temperatures at or below 12°C. The study focused on isolates belonging to psychrotolerant or mesophilic-psychrotolerant intermediary thermotypes in panC-groups II, IV, V, or VI and generated data for growth kinetics for various pH (4.8-7.8), a(w) (0.935-0.999) and storage temperatures (6.0-11.7°C) in 42 starchy foods (bulgur, couscous, pasta, potatoes, rice) and eight composite foods containing at least one starchy ingredient. Using 21 of the growth kinetics obtained for starchy foods, the five best performing of 10 available growth models were selected for improvement by product calibration and/or expansion with terms to consider the effect of interactions between temperature, pH and a(w). Of 410 updated models, nine showed promising performance and were evaluated using the remaining 21 growth kinetics obtained in starchy foods. Two models could be considered validated for these products with B(f) /A(f) -values of 0.87/1.21 and 1.01/1.32, respectively. Both models provided ≥75% correct predictions of the growth/no-growth responses and did not provide any fail-dangerous predictions. Further evaluation of these models for predictions of maximum specific growth rates (μ(max) , h(-1)) and growth/no-growth responses for a broader range of starchy foods used 33 challenge tests from the scientific literature and eight challenge tests from the present study, and remarkably showed that the performance of both models was poor for composite protein-rich starchy foods with B(f) -values ≤0.64 and A(f) -values ≥1.96, meaning these models should not be used for such products as μ(max) might be under-predicted creating unsafe situations. However, for other starchy foods, one of the validated models was found to be acceptable on the safe side with B(f) - and A(f) -values of 1.34 and 1.57, respectively.