Assessment of Escherichia coli bioreporters for early detection of fungal spoilage in postharvest grape berries

Table grapes are popular for their taste and nutritional value, but they are vulnerable to fungal spoilage that can cause economic losses and food safety concerns. The early detection of the causal agents is essential to implement effective prevention and control measures. Escherichia coli bioreporters are cost-effective, sensitive, and noninvasive tools promising for this purpose. However, their effectiveness for discriminating multiple fungal spoilers in grape berries has not been assessed. In this study, two fungal spoilers, Aspergillus niger and Botrytis cinerea, were isolated from rotting grape berries and identified using phylogenetic analysis. E. coli bioreporters were then noninvasively exposed to inoculated grape berries by three pathogens A. niger, A. westerdijkiae and B. cinerea. SoxSp, fabAp and pspAp bioreporters were among the most inducible ones by the infection-related volatile organic compounds. The response pattern was classified using six machine-learning predictors in combination with various preprocessing techniques. The top-performing model was a random forest (RF) model that utilized the highly-correlated variables (RHCV) preprocessing technique, which achieved 100% accuracy in predicting infection stages of A. niger in grape berries. The most robust predicator to discriminate the infection stages of A. westerdijkiae and B. cinerea was generated using spatialSign preprocessing in combination with support vector machine with radial kernel (svmRadial), both achieving prediction accuracy of 92%. Lastly, RHCV preprocessing coupled with svmRadial classifier was the best method to discriminate between the three fungal spoilers on grape berries, with a prediction accuracy of 80%. The study showed that E. coli bioreporters are highly accurate in predicting early fungal spoilage in grape berries but have moderate prediction specificity in discriminating multiple fungal spoilers.