Predictive Microbial Modeling of Enterococcus faecium NRRL B-2354 Inactivation during Baking of a Multicomponent Low-Moisture Food

Abstract

ABSTRACT The use of baking ovens as a microbial kill step should be validated based on results of thermal inactivation models. Although traditional isothermal models may not be appropriate for these dynamic processes, they are being used by the food industry. Previous research indicates that the impact of additional process conditions, such as humidity, should be considered when validating thermal processes for the control of microbial hazards in low-moisture foods. In this study, the predictive performance of traditional and modified thermal inactivation kinetic models accounting for process humidity were assessed for predicting inactivation of Enterococcus faecium NRRL B-2354 in a multi-ingredient composite food during baking. Ingredients (milk powder, protein powder, peanut butter, and whole wheat flour) were individually inoculated to achieve ∼6 log CFU/g, equilibrated to a water activity of 0.25, and then mixed to form a cookie dough. An isothermal inactivation study was conducted for the dough to obtain traditional D- and z-values (n = 63). In a separate experiment, cookies were baked under four dynamic heating conditions: 135°C, high humidity; 135°C, low humidity; 150°C, high humidity; and 150°C, low humidity. Process humidity measurements; time-temperature profiles for the product core, surface, and bulk air; and microbial survivor ratios were collected for the four conditions at six residence times (n = 144). The traditional isothermal model had a high root mean square error (RMSE) of 856.51 log CFU/g, significantly overpredicting bacterial inactivation during the process. The modified model accounting for the dynamic time-temperature profile and process humidity data was a better predictor with an RMSE of 0.55 log CFU/g. These results indicate the importance of accounting for additional process parameters in baking inactivation models and that model performance can be improved by utilizing model parameters obtained directly from industrial-scale experimental data. HIGHLIGHTS