Cleide Møller, National Food Institute, Technical University of Denmark, Mørkhøj Bygade 19, DK-2860 Søborg, Denmark. E-mail: email@example.com
Aims: The aim of this study was to develop a model to predict cross-contamination of Salmonella during grinding of pork.
Methods and Results: Transfer rates of Salmonella were measured in three experiments, where between 10 and 20 kg meat was ground into 200-g portions. In each experiment, five pork slices of about 200 g per slice were inoculated with 8–9 log-units of Salmonella Typhimurium DT104 and used for building up the contamination in the grinder. Subsequently, Salmonella-free slices were ground and collected as samples of c. 200 g minced pork. Throughout the process, representative samples were quantitatively analysed for Salmonella. A model suggested by Nauta et al. (2005) predicting cross-contamination of Campylobacter in poultry processing and two modified versions of this model were tested.
Conclusions: The present study observed a tailing phenomenon of transfer of Salmonella during a small-scale grinding process. It was, therefore, hypothesized that transfer occurred from two environmental matrices inside the grinder and a model was developed. The developed model satisfactorily predicted the observed concentrations of Salmonella during its cross-contamination in the grinding of up to 110 pork slices.
Significance and Impact of the Study: The proposed model provides an important tool to examine the effect of cross-contamination in quantitative microbial risk assessments and might also be applied to various other food processes where cross-contamination is involved.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Salmonella has been linked to many foodborne illness cases across the world, and it is considered to be one of the main agents causing human gastroenteritis (Jimenéz et al. 2009). In Denmark, the locally produced pork was estimated as one of the most important sources of salmonellosis in 2009 (Anonymous, 2010). Contamination and cross-contamination with Salmonella start inside the pig slaughterhouse (De Busser et al. 2011) and originates from animals carrying Salmonella in their intestine. Along the slaughter line, several steps can be critical for Salmonella contamination: dehairing, polishing, removal of the intestines, removal of the pluck set and meat inspection procedures (Borch et al. 1996). During these steps, the carcass can be contaminated with faeces and bacteria can be spread all over the carcass and to subsequent carcasses. In the following handling and processing of pork through the cutting and retail steps, recontamination and cross-contamination with Salmonella may continue because infected and noninfected meat share the same cutting surfaces and processing equipment without cleaning and disinfection in between (EFSA, 2010). According to Hansen et al. (2010), this behaviour was most likely responsible for an increased prevalence of Salmonella in pork cuttings at retail in Denmark between 2002 and 2006, as the rise observed in 2006 at retail could not be ascribed to a rise in Salmonella carcass prevalence at slaughter.
The food-processing environment is an important but still poorly recognized and understood source of recontamination. Detailed proof and facts concerning this issue have only been published occasionally. Fortunately, the number of publications on investigations of processing environments is slowly increasing, demonstrating an increased awareness, but there are still only few data available that allow us to quantify the rate of transfer of pathogens from food to contact surfaces and vice versa during processing (De Boer and Hahné 1990; Reij and Den Aantrekker 2004). Studies that simulate and model the distribution of pathogens during processing operations are of major relevance to risk analysts to ascertain the importance of equipment sanitation, sources of potential product contamination and improved equipment design (Flores et al. 2006). Recently, a number of cross-contamination models have been published describing transfer of Listeria monocytogenes (Vorst et al. 2006; Aarnisalo et al. 2007; Keskinen et al. 2008; Sheen 2008), Escherichia coli O157:H7 (Pérez-Rodríguez et al. 2007; Sheen and Hwang 2010), and Staphylococcus aureus (Pérez-Rodríguez et al. 2007) during slicing of ready-to-eat products. Studies concerning transfer of Salmonella from chicken carcasses to cutting board (Jimenéz et al. 2009), from domestic washing-up sponge to kitchen surfaces and food (Mattick et al. 2003) and from raw chicken products during food preparation (De Boer and Hahné 1990), have also been performed. A few studies using Enterobacter aerogenes with attachment characteristics similar to Salmonella have also been conducted on chicken to investigate cross-contamination (Zhao et al. 1998; Chen et al. 2001). However, not much have been investigated regarding cross-contamination events in fresh-meat processing and according to Pérez-Rodríguez et al. (2008), pathogens may transfer to foods through many different types of events, such as recontamination and cross-contamination, which might be decisive in many of outbreaks. A work published by Berends et al. (1998), describing the ecology and epidemiology of Salmonella spp. in pork cutting lines of Dutch cutting plants and in butchers’ shops, has already mentioned the impossibility of accurate identification and quantification of all risks involved in contamination of pork with Salmonella, because of a lack of data. To our knowledge, models describing the transfer of Salmonella occurring during grinding of fresh meat have not yet been published. Therefore, this study investigated the transfer of Salmonella during grinding of pork. Different models, derived from the model developed by Nauta et al. (2005) to describe cross-contamination of Campylobacter in poultry processing, were applied to the obtained data, and a mathematical model was selected describing the grinding and pointing out how long the contamination was maintained during processing and in what concentration Salmonella was transferred. Finally, the model was challenged with different Salmonella concentrations inserted at several points during the grinding process.
Materials and methods
Pieces of vacuum-packaged fresh, deboned, deskinned pork leg, frequently used to produce cooked ham, containing 6–8% fat, measuring about 8 × 10 × 15 cm and weighing 1289 ± 109g were obtained from a central distributor. With the exception of one pack, which was reserved for inoculation, all the others were divided into five slices. From each of the slices that was selected for analysis, representative samples of 25 g were aseptically collected to be sure that the meat to be processed was Salmonella free. After removing of these 25 g, the investigated slices averaged 212 ± 50 g per slice. Appropriately diluted samples in Maximum Recovery Diluent (Oxoid Ltd – Thermo Fisher Scientific Inc., Greve, Denmark) were enumerated by spread plate (100 μl) on XLD – xylose lysine deoxycholate agar (Oxoid) and incubation at 37°C for 16–24 h.
As prior investigations had shown that Salmonella Typhimurium was the predominant serotype in retail pork cuttings in Denmark (Hansen et al. 2010), a strain of Salm. Typhimurium DT104 (77-20547-1 provided by Anders M. Hay Sørensen, National Food Institute, DTU, Denmark) isolated from pigs and carrying resistance to ampicillin, chloramphenicol, florfenicol, streptomycin and sulfa was chosen for the experiments and grown in 20-ml lysogeny broth (Oxoid) with shaking (200 rev min−1) overnight at 37°C. Subsequently, and prior to inoculation of pork, the culture was kept at 5°C for 24 h and then used directly (c. 109 CFU ml−1) in experiments conducted for building up and validation of the model, or diluted to c. 107 CFU ml−1 for validation challenge tests.
Inoculation of meat
To mimic the muscle structure that normally is infected in case of contamination with Salmonella, the whole piece of the obtained pork was surface inoculated with 10 drops of 100 μl of the Salmonella culture corresponding to 2 × 109 – 4 × 109 CFU per piece of meat, instead of inoculating the meat already sliced. The culture was spread on the whole surface of the side facing up using a Drigalski spatula (VWR International Ltd – Bie & Berntsen, Herlev, Denmark) and, subsequently, the meat was kept for 40 min to allow attachment of cells. The inoculated piece of pork was then divided into five slices, resulting in 108–109 CFU of Salmonella per slice. In total, the piece of pork was left uncovered on a table in the laboratory for 1 h at 21–22°C and 40% relative humidity before grinding took place.
A semi-industrial grinder (la Minerva® food service equipment, Italy – obtained from H.W. Larsen A/S, Copenhagen, Denmark) was used in a refrigerated room with temperature of about 5°C. Five slices of noninoculated meat were processed to create a matrix inside the grinder. Five inoculated samples were then ground and 40–90 slices of noninoculated pork meat were processed. Individual portions (213 ± 56 g) corresponding to each processed slice were collected in separate sterile Stomacher® bags (Seward, Worthing, UK). To ensure a homogenous distribution of Salmonella in the whole portion of meat, it was mixed two times for 1 min in a Stomacher® 400 circulator (Seward), intercalated with a step of manual blending, before sampling. Subsequently, 25 ± 0.2 g samples were diluted in 225 ml of brain heart infusion broth (Oxoid) and mixed in a Stomacher® 400 circulator for 2 min. Further appropriate dilutions were made in maximum recovery diluent (Oxoid) and drop-plated (3 × 10 μl) or spread-plated (1 × 100 μl or 3 × 333 μl) onto XLD agar with and without 100 mg l−1 ampicillin (Bristol-Myers Squibb, New York, NY, USA) at 37°C for 16–24 h. This procedure was repeated three times resulting in dataset 1, 2 and 3.
Model development and validation
Initially, results from the three Salmonella transfer studies were fitted to a model developed for predicting cross-contamination of Campylobacter in poultry processing (Nauta et al. 2005). This model is characterized by a four-parameter equation that assumes that the grinder can be described as one single environment, with constant transfer rates between the grinder and the meat. As a result of lack of fit, an extended model was studied: two new models were derived from the hypothesis, that the input of Salmonella is organized in two different matrices inside the grinder; one matrix where Salmonella reveals high transfer ability and a second where Salmonella demonstrates low transfer ability from the grinder to the meat. Mathematical details of the different models are described in the Results section. For estimation of model parameters, the residual sum of squares (RSS) was minimized using the solver function in MS Excel (Microsoft® Office Excel® 2007). The three models fitted to own experimental transfer results were compared by F-tests (Zwietering et al. 1990). Considering the RSS of the model, the number of observations and the number of model parameters, the root mean sum of squared errors (RMSE) (Ratkowsky 2004) and the bias-corrected version of Akaike information criterion (AICc) were calculated as measures for goodness of fit (Hurvich and Tsai 1989).
To validate the best-fit model, two challenge tests were performed. The first challenge test A (100 processed slices) was conducted by adding slices contaminated with 108–109 CFU of Salmonella per slice processed as 1st, 2nd and 3rd, 29th and 55th slices. A second challenge test B processed 110 portions where 1st, 2nd and 3rd slices were contaminated with 107 CFU of Salmonella per slice and the 19th and 35th slices had 108–109 CFU and 106–107 CFU of Salmonella, respectively. The inoculation procedure used in the validation trials was the same as adopted for model development. It is, however, important to mention that individual pieces of meat were used for each different concentration. As usual, the inoculated piece of pork was divided in five slices, then the selected slices were processed in the validation challenge test, and the remaining slices were discarded. In addition to the visual inspection of the data, bias and accuracy factors (Ross 1996) with log10 CFU per slice as the response variable were used to evaluate the performance of selected models.
Model predictions with low inoculation levels
Transfer of Salmonella in pork at the Danish retail level usually occurs at low concentrations (Hansen et al. 2010); therefore, the suggested model was used to predict transfer of Salmonella for low initial concentrations (103, 104 and 105 log10 CFU per slice), as at such low levels, the transfer of pathogens during processing is recognized as very challenging (Aarnisalo et al. 2007; Sheen 2008; Sheen and Hwang 2010).
Salmonella transfer model
Despite different number of processed slices in each of the three transfer trials, the profile of transfer of Salmonella during grinding of pork followed the same pattern. The cross-contamination of Salmonella occurred at two distinct rates through the process as shown in Fig. 1. First, the contamination builds up with ground meat from five contaminated slices, showing a (slightly increasing) plateau that balances the number of bacteria from each contaminated slide with the transfer from and to the grinder and inactivation in the grinder. At the first descending slope, fast transfer of Salmonella from the grinder to the minced meat occurred, whereas at the second slope, a ‘tail’ of low contaminated portions of ground pork represented a much slower transfer of Salmonella.
The model suggested by Nauta et al. (2005), predicting cross-contamination of Campylobacter in poultry processing, could efficiently describe the observed transfer of Salmonella during grinding of the first 20 slices but could not explain the ‘tail’ of low contaminated portions (results not shown). Therefore, it was hypothesized that the input of Salmonella is not organized in one environment as in the original model, but in two different matrices inside the grinder, as represented in Fig. 2. One matrix, where Salmonella is relatively loosely attached, is responsible for the fast transfer to the minced meat and from a second matrix, Salmonella’s transfer occurs at a slower rate. Based on this hypothesis, a modified version of Nauta et al. (2005) model was implemented. Modifications of the parameters and addition of an extra parameter to the model to describe the whole transfer were tested as shown in the following model equations:
This new model has seven parameters, a1, a2, b1, b2, c1, c2 and c3, which represent probabilities of transfer (a, b) or inactivation (c) per bacterial cell, as explained in Fig. 2. However, as it was possible to recover the inoculated number of Salmonella from pork slices (results not shown) and pork in general is an excellent substrate for survival and growth of Salmonella (Escartín et al. 2000), inactivation in the meat was assumed not to take place and c2, therefore, set to zero. Likewise, inactivation of Salmonella in environment 1 in the grinder was assumed unlikely to occur and c1 set to zero. In this case, the assumption was based on the fact that transfer from this environment would happen too fast for Salmonella to be inactivated and recovery of Salmonella from the ground meat portions was designed to minimize stress factors. As c1 and c2 both was set to zero, the model can be considered a five-parameter model (all with values between 0 and 1), which considers k slices of meat that are processed in a grinder to k portions of minced meat (i =1, 2,… k). The ith slice carries SiSalmonella (CFU per slice), and the resulting minced meat portion from slice i carries MiSalmonella (CFU per portion). The ‘contamination status’ of the grinder is gri. The probability of transfer per Salmonella cell from meat to grinder, environment 1 and to grinder, environment 2 is represented by a1 and a2, respectively. The backward transfer probabilities from the grinder (environments 1 and 2) to ground meat are given by b1 and b2. The survival in environment 2 of the grinder is represented by 1 − c3.
Fitting of models and goodness of fit
As shown in Table 1, three different models were fitted to the datasets obtained from the three transfer experiments. The first model, named 4p-1ge, was a four-parameter model that considered only one grinder environment and is identical to the Nauta et al. model from 2005. The second model, 4p-2ge, was a modified version of this model also with four parameters but taking into account two grinder environments (Eqn 1 with c1= c2= c3=0), and the third, 5p-2ge, was the new suggestion for a transfer model with five parameters as described previously (Eqn 1 with c1= c2=0). The suggested 5p-2ge model was evaluated as the best choice as it resulted in the lowest RMSE value and AICc score for all three datasets (Table 1). Low RMSE values, as those obtained for the 5p-2ge model, show that the observed and predicted transfer of Salmonella were very close (Valero et al. 2007) and the model with the lower AICc score, like the proposed 5p-2ge model, is more likely to be correct (Motulsky and Christopoulos 2003), and therefore, it is considered to have substantial support. This conclusion was also supported statistically by significant F-tests (P ≤0.033) when used to compare the suggested model (eqn 1) to the two other models for dataset 1, 2 and 3, respectively (Table 1). The difference between the three tested models was most pronounced for dataset 3, where the F-tests were highly significant (P < 0·001). Figure 3 is a visual example of fitting the three models to the dataset 2. It illustrates why the 5p-2ge was the superior model. The model 4p-2ge could not describe appropriately the observed build-up of Salmonella in the grinder while model 4p-1ge could not describe the observed data as it was not able to fit the ‘tailing’ phenomenon. Parameter estimates obtained from fitting the 5p-2ge model to each of the three datasets are shown in Table 2.
Table 1. Performance of the developed models when fitted to three different datasets
RMSE, root mean sum of squared error; AICc, Akaike information criterion.
*5p-2ge = suggested model with five parameters considering two grinder environments; 4p-2ge = modified version of Nauta et al. (2005) model considering two grinder environments; 4p-1ge = Nauta et al. (2005) model considering one grinder environment.
†For each dataset, different letters denote statistical difference according to the F-test.
Table 2. Parameter estimates from three datasets obtained from the transfer of Salmonella during grinding of pork, with initial level of 108–109 log10 CFU per each of the five contaminated slices
Source of parameters
Number of processed slices
1 − c3
Validation of transfer model
As opposed to the experiments performed when building the suggested model 5p-2ge, the five input slices carrying Salmonella were not only added in the beginning of the grinding process, but also at two later processing points in the two validation trials. As shown in Fig. 4, Salmonella-contaminated slices were added as 1st, 2nd, 3rd, 29th and 55th slices in trial A (Fig. 4a) and as 1st, 2nd, 3rd, 19th and 35th slices in trial B (Fig. 4b). In validation trial A, all five contaminated slices contained 108–109 CFU Salmonella, whereas in validation trial B, the Salmonella concentration was changed so that the first three slices contained 106–107 CFU, the fourth 108–109 CFU and the fifth 106–107 CFU. For validation trial A, comparisons of observed and predicted values resulted in bias factors of 0·95, 0·99 and 1·01 and accuracy factors of 1·07, 1·05 and 1·06, when using the parameter estimates from dataset 1, 2 and 3, respectively. Likewise for validation trial B, bias factors of 0·91, 0·93 and 1·01 and accuracy factors of 1·14, 1·12 and 1·07 were obtained applying parameters estimates from dataset 1, 2 and 3, respectively. These values of bias and accuracy factors indicate how good the model performed as the perfect agreement between predictions and observations will lead to a bias or accuracy factor of 1 (Ross 1996). Figure 4 shows that using parameter estimates obtained from fitting of dataset 3 to the 5p-2ge model predicted the observed ‘tailing’ phenomenon most accurately both in validation trial A and B. In large-scale studies as those related to cross–contamination, it is difficult to obtain a similarity in the size of the samples, what can be an explanation for the variation of the parameter estimates from different datasets. However, it was observed that the number of processed slices seems to have influence on the shape of the tailing phenomena and with larger experiment better parameter estimates were obtained, like those presented by dataset 3.
How the model can be used
The suggested model, 5p-2ge with parameter estimates obtained from dataset 3, was applied to simulate transfer of Salmonella at hypothetical lower levels of contamination. The cross-contamination was simulated with input concentrations of 105, 104 and 103Salmonella per slice as illustrated in Fig. 5. Looking at the profile of cross-contamination, it is possible to observe that in all examples, the tailing started after the 15th–16th processed slice but the contamination level of the tail corresponded to different levels depending on initial input of Salmonella, that is input of five slices with 104Salmonella per slice resulted in a tail of minced portions containing c. 1 CFU per 200 g, and an increase of ten times this value was noticed using input of 105 that results in tailing around 10 CFU per 200 g (Fig. 5). However, it is important to be aware that when using input of five slices with 105Salmonella per slice, at the 95th processed slice, levels of about 1 CFU of Salmonella could be present in a 2-kg processed piece of pork and then successively. The same elucidation can be applied to explain other hypothetical levels of Salmonella. Preliminary experiments, using input concentrations around 104–105Salmonella per slice, resulted in Salmonella counts under the quantification limit (<2 log10 CFU g−1) from portion 7 to portion 25. As a supplement to the direct counting methods, all these portions were incubated at 11–12°C for 2 days and then analysed again using the direct counting method. At this point, Salmonella not only could be detected in all portions but also appeared to be present in almost the same concentration (result not shown) confirming a tailing phenomenon.
During the grinding of pork, Salmonella present on a single piece of meat may be transferred to many portions of minced meat as a result of cross-contamination in the grinder. It is, therefore, important to be able to describe the cross-contamination by grinding, mathematically, to predict how many portions of ground meat may be contaminated with the pathogen from one single piece of meat. In this study, such mathematical model was developed and it was shown that it can be used for this purpose. During the development of the model, it was also taken into consideration that a tailing phenomenon of the transfer of Salmonella was observed during a small-scale grinding process. It was suggested that transfer occurred from two environmental matrices inside the grinder and the model was developed to match this hypothesis. Considering the low RMSE and AICc values supported by the F-tests results, the developed model satisfactorily fitted the observed behaviour of Salmonella during its cross-contamination in the grinding of up to 110 pork slices corresponding to 21 kg meat. During the past decade, a number of studies investigating the transfer of pathogens during slicing of ready-to-eat products (Vorst et al. 2006; Aarnisalo et al. 2007; Sheen 2008; Sheen and Hwang 2010) have been published. However, no other studies have yet modelled the transfer of Salmonella during a grinding process and only in few other studies, a similar distinct tailing phenomenon during cross-contamination was observed (Vorst et al. 2006; Aarnisalo et al. 2007; Sheen 2008; Sheen and Hwang 2010). In the study of Sheen and Hwang (2010), the cross-contamination of E. coli O157:H7 during slicing of ready-to-eat ham was modelled applying an empirical approach, and the selected model characterized the transfer as decreasing following exponential law. Comparison of the Sheen and Hwang (2010) model to the model developed in the present study revealed that the two models are mathematically similar, and it can be shown that the Sheen and Hwang (2010) model is basically the same as the model termed 4p-1ge (Nauta et al. 2005) considering one environment only. Despite this similarity, the proposed model should, however, be preferred as it includes the pieces of meat that are contaminated before grinding and it gives clear explanations of all the parameters involved providing an overview of the dynamics of a grinding process. As it is easier to understand, it also holds the potential to be universal, that is transferrable to cross-contamination dynamics for other food-processing steps. The fitted model obtained in this study is of course specific to the studied grinding process including the particular grinder applied. However, the structure of the model, and particularly its ability to predict the tailing phenomenon, seems relevant for different cross-contamination processes. Testing the model structure on data published in other transfer studies, where different food products, micro-organisms, concentration of pathogen and different routes of contamination (food product to slicer to food product or slicer to food product, using the same product or different products) were used, showed promising results. Applying the proposed 5p-2ge to literature data and obtaining R2 values close to one illustrates this claim. For example, when applying the data published by Vorst et al. (2006), simulating cross-contamination of L. monocytogenes during turkey slicing, R2=0·86, was found. When the data presented by Aarnisalo et al. (2007), regarding transfer of L. monocytogenes during slicing of gravad salmon, were used, R2 = 0·74 was obtained, and for the data published by Sheen and Hwang (2010) related to cross-contamination of E. coli O157:H7 during ham slicing, R2 was 0.78.
The good bias and accuracy factors, with values close to one, obtained when validating the suggested model, are an indication that the hypothesis of two matrices being responsible for the transfer could in fact be the explanation for the observed tailing. Nevertheless, it has yet to be elucidated what the two suggested environmental matrices consist of and how the transfer takes place at the physical level. The tailing phenomenon could also be interpreted as two subpopulations behaving differently in one environment, that is having different susceptibility to the environmental stress experienced during grinding and, thereby, different transfer abilities. More investigations are needed to determine the exact cause of the observed two-phase transfer of Salmonella during grinding. Furthermore, as Salmonella levels as high as 7–9 log10 CFU per slice were used for developing the model, it is not known whether tailing could be an artefact of high concentrations. Although levels of Salmonella lower than 3 log10 CFU per slice (Hansen et al. 2010) would have been more relevant to study, this was not performed because transfer modelling becomes very challenging at such low concentrations because of the random transfer pattern resulting in concentrations below the detection limit when using direct plate counting methods (Aarnisalo et al. 2007; Sheen and Hwang 2010). However, when applying Salmonella concentrations as low as 4–5 log10 CFU per slice and combining it with enrichment for 2 days at 11–12°C in the minced meat, the tailing phenomenon was confirmed. The complexity and difficulty of low-pathogen-level transfer faced in the present study were also described by Sheen and Hwang (2010), who stressed how relevant it is to obtain more information and knowledge to facilitate realistic risk predictions. It also suggests that new methods of enumeration should be developed to improve sensitivity and still continue to be simple enough for low concentration, large-scale studies to be realistic. Enrichment in the meat at relatively low temperatures from 11 to 16°C combined with accurate predictive models appears as an obvious solution.
Other factors such as varying fat contents of the meat, varying sizes of the meat pieces to be minced and growth of Salmonella inside the grinder could also influence the cross-contamination dynamics and perhaps be responsible for the observed tailing. Biological variation between the pieces of meat and Salmonella populations, as well as randomness, may also explain a large part of the variation between the parameter values as obtained from the three experimental datasets compared in this study (Table 2). In the present study, the effect of these factors was minimized by (i) use of lean meat with the same low fat content for all experiments, (ii) use of meat slices with approximately the same dimensions for all experiments and (iii) use of processing temperature and time conditions not supporting growth of Salmonella. Therefore, these factors could be excluded as main decisive factors of the observed tailing phenomenon.
Different models describing transfer behaviour of pathogens in food processing have already been published in the scientific literature with relatively good performance. However, most of them are empirical models and cannot explain the meaning behind the model parameters. Therefore, it is still important to develop models not only capable of describing the observed occurrence of cross-contamination taking into consideration the precise level of contaminant in the whole process, but also of giving a reasonable explanation to the phenomenon implicated in transfer of pathogens during food processing. This was achieved by the model proposed in the present study, and it is believed that the model structure is transferable to other cross-contamination scenarios. Furthermore, the proposed model presents an important tool to examine the effect of cross-contamination in case of low concentrations, for example in relation to quantitative microbial risk assessment investigations.
This research was financially supported by the Technical University of Denmark through the FoodDTU programme. The authors recognize the valuable and dedicated laboratory work of colleagues from the Division of Microbiology and Risk Assessment, National Food Institute, DTU.