OBJECTIVE For determination of Salmonella concentration in meat various methods can be used depending on the expected level. When higher levels (102 to 103 bacteria or more per g) are anticipated, plate count techniques using selective agars, i.e. XLD, are appropriate whereas for low numbers (3 to 102 bacteria per g) a most probable number (MPN) method is recommended. Recently, a real-time PCR-based tool for determination of concentrations as low as 1.4 Salmonella per 20 cm (approx. 10 g) of cork-borer samples of pig carcasses has been developed (Krämer et al. 2011). However, compared to plate count techniques the MPN and real-time PCR methods are very labour intensive and might not be suitable when analysing many samples within a short timeframe. We suggest to use enrichment in the meat at relatively low temperatures (11 to 16°C) combined with predictive models as an alternative. Therefore, the objective of this study was to evaluate the possibility to carry out a Salmonella enrichment step in the meat itself and use the two species interaction model, presented by Møller et al. (2013), for quantifying levels of Salmonella Typhimurium DT104 in minced pork. METHODS A total number of 101 minced pork samples were inoculated artificially with various concentrations (from 10 to 106 bacteria per g) of stationary phase S. Typhimurium DT104. Counts of S. Typhimurium DT104 as well as natural microbiota in the samples were determined immediately after inoculation and again after an enrichment step in the minced meat for approx. 48 h at 11 – 16°C. A rearranged version of the expanded Jameson-effect species interaction model, suggested by Møller et al. (2013), was applied for prediction of the Salmonella concentration in the minced pork samples. Observed and predicted counts of S. Typhimurium DT104 (log10-units) prior to enrichment were compared visually and by the acceptable prediction zone (APZ) method, i.e. percentage of predictions being within ±0.5 log10-units of observed values. RESULTS A relatively good agreement between predicted and observed values was seen. However, only 56 % of the predictions were within the APZ. The model tended to over-predict counts from 3 log-units and above, whereas under-prediction to some extent was seen for counts below 3 log10-units. Over-prediction was most likely explained by uncertainty of the lag-time model for Salmonella, i.e. a short lag time would result in a lower initial count to get to the same count after enrichment as compared to a long lag time. In contrast, part of the under-prediction appeared to coincide with competitive growth of Citrobacter braakii and Hafnia alvei on the selective agar. Therefore, under-prediction more likely resulted from underestimation of the Salmonella count after enrichment. Whether competition between these species also took place in the meat during enrichment is not known. However, as the observed levels of the competitive species were below 5.5 log10-units it is questionable whether interaction with S. Typhimurium DT104 in the pork could have occurred. Omitting these samples and using the 56 observations below 3 log10-units improved the percentage of predictions within APZ to 63 %. CONCLUSIONS AND IMPACT OF THE STUDY A novel approach for determining Salmonella counts at low concentrations was proposed. Applying a simple plate count method, after cold enrichment (11-16°C for 2 d) in the pork sample itself in combination with predictive growth models, showed promising results. It indicates the potential of this approach as an alternative to meet the need for more sensitive methods, which are simple enough to be used in large-scale series of analysis.
8th International Conference on Predicive Modelling in Food - Icpmf8: Conference Proceedings, 2013, p. 23-24