A vitalistic approach for non-thermal inactivation of pathogens in traditional Greek salads

Transcription

1 Food Microbiology,,, ^ Available online at on doi:./yfmic. ORIGINAL ARTICLE A vitalistic approach for non-thermal inactivation of pathogens in traditional Greek salads Panagiotis N. Skandamis ;,KennethW.Davies, PeterJ. McClure, Konstantinos Koutsoumanis and ChryssoulaTassou A vitalistic model was developed for the e ects of low ph, chill temperatures and oregano essential oil on Salmonella enterica subsp. enterica serotype Enteritidis PT (S. Enteritidis PT )and Escherichia coli O:H NCTC in taramasalad and aubergine salad, respectively. The model was based on the concept of the existence of heterogeneous response to stress within a genetically identical bacterial population that is exposed to stress at lethal levels. The imposed stress conditions were mathematically expressed as an argument of a cumulative normal distribution function, to predict the proportion of survivors at a given time of exposure. For each food, the modelling procedure consisted of estimating parameters using nonlinear least squares in one step, resulting in a single equation. The graphical illustration of the tted model and data con rmed a good t. Moreover, model comparisons with literature data and Pathogen Modelling Program were carried out to underline the importance and speci city of developing predictive models for survival if pathogens in situ. # Elsevier Science Ltd. All rights reserved. Introduction Mathematical modelling of bacterial kinetics can be a useful tool to quantitatively express the e ect of intrinsic and extrinsic factors on spoilage and foodborne pathogens in foods. To date, much work has been focused on growth kinetics (Ross and McMeekin ) as well as on kinetics of microbial heat inactivation (Cole et al., Anderson et al., Blackburn et al., Humpheson et al. ). However, there is less quantitative data available for the combined e ect of heat with other factors, such as NaCl and/or ph (Dodds, Blackburn et al. *Corresponding author. Fax: gjn@aua.gr, Membre et al., Breand et al., Gaillard et al., Chhabra et al., Juneja et al. a,b). There is a similar scarcity of data for non-thermal inactivation at sub-optimal conditions in model foods and real foods (Buchanan et al.,, Linton et al., Little et al., Koutsoumanis et al., Skandamis and Nychas ). In studies of heat inactivation, microbial death rate has usually been determined with the traditional log-linear model (log number of survivors vs time) based on the rst-order kinetics. Since this model does not describe well the sigmoidal or semi-sigmoidal survival curves with shoulder region and/or tailing region, typical sigmoidal models (e.g. Gompertz, logistic, and Baranyi) or the logistic-based model of Kamau et al. () have been applied successfully Received: May Agricultural University of Athens, Department of Food Science and Technology, Laboratory of Microbiology & Biotechnology of Foods, Iera Odos, Athens, Greece Unilever Research Colworth, SEAC- Microbiology Section, Sharnbrook, Bedfordshire MK LQ, UK Institute of Technology of Agricultural Products, National Agricultural Research Foundation, S.Venizelou, Lycovrisi Athens, Greece -//+ $./ r Elsevier Science Ltd. All rights reserved.

2 P.N.Skandamisetal. (Koutsoumanis et al., Skandamis and Nychas,, Xiong et al., Whiting et al. ). It should be noted that both traditional log-linear and sigmoidal models, as well as those included in the cited references, account for descriptions of bacterial death at one condition with time only variations in time. Consequently, there is lack of inactivation models that predict bacterial survival vs time and express the inactivation kinetics in response to a given range of stress factors. Kilsby et al. () published a novel probabilistic model for thermal inactivation of Clostridium botulinum and Salmonella Bedford at di erent ph and A w.the basis of this model di ers from that of previous vitalistic models (Cole et al., Anderson et al., Blackburn et al. ) in the assumption that within a nominally genetically identical bacterial population, the individual cells are not equally resistant to an externally imposed stress. This variation has been used as the basis for modelling. Such an approach has proven advantageous in the sense that (a) it has provided an acceptable t of the types of survival curves listed above and (b) allowed for one-step modelling (single equation) of kinetics in response to environmental conditions. Single equation models have the advantage over modelling in two stages, because the latter have errors involved at both primary and secondary stages. The present work was based on the data published by Koutsoumanis et al. () and Skandamis and Nychas () and the objective was to describe the survival of two pathogens, Escherichia coli O:H and S. Enteritidis PT, in two di erent emulsi ed foods (i.e. traditional Greek appetizers, taramasalad and aubergine salad) as a function of refrigerating temperature, ph and oregano essential oil, with an appropriate probability distribution of the surviving bacteria during exposure to di erent conditions. Materials and Methods Bacterial strains Salmonella enterica supsp. enterica serotype enteritides PT E. coli O:H NCTC and (S. Enteritidis PT ) were the micro-organisms examined (Koutsoumanis et al., Skandamis and Nychas ). The E. coli O:H strain used was non-toxigenic strain and was provided by I. Ogden (Applied Food Microbiology Group, Department of Medical Microbiology of the University of Aberdeen, Scotland), while the S. Enteritidis PT was provided by Professor R.G. Board (University of Bath, UK). Experimental design In taramasalad survival of S. Enteritidis PT was studied at a range of temperatures (, C, C, C), ph (?,?,?) and oregano essential oil concentrations (%,?%,?%,?%,?% v/w). Similarly, to investigate survival of E.coli O:H in aubergine salad, four temperatures, (C, C, C, C) three ph (?,?,?) and four oregano essential oil concentrations (%,?%,?%,?% v/w) were selected. An inoculum of cfu g was used for both micro-organisms at all experimental conditions. Foods and additives The preparation procedure for both foods as well as the full experimental procedure are described by Koutsoumanis et al. () and Skandamis and Nychas (). Statistical model The modelling approach used was that of Kilsby et al. () who described death kinetics as a distribution of inactivation times. A model was tted of the form Log N t ¼ Log½ð Stress functionþš; ðþ N where N t is the number of microbes at time t, and is the normal probability integral, de- ned by: ðxþ ¼ Z x pffiffiffiffiffi e x = dx: ðþ Stress function is a mathematical expression that describes how the experimental condi-

3 A vitalistic approach for non-thermal inactivation tions a ect the inactivation times of the bacterial population. In particular, the stress function was de ned by an empirical algebraic expression with additive and multiplicative terms of environmental variables (T, phand % v/w oregano essential oil concentration) in the rst or second order. Since the same environmental variables (T, ph, % essential oil concentration) are tested on two di erent micro-organisms and two di erent food ecosysterms, separate stress functions should be developed for each case. The procedure for tting the model has two steps described by Kilsby et al. (), with a separate N term given to each data series and was performed with SAS Software. Evaluation of model performance To investigate whether the models could describe the experimental data su ciently, as well as to evaluate the applicability of the models in similar conditions of ph, temperature, and even essential oils in di erent foods, model predictions were tested against the current experimental data. Food validation (with independent data) were also carried out. The former was performed on the basis of comparison of predicted and observed times for log reduction (including survival and death phase) of S. enteritidis PT and E. coli O:H in taramasalad and aubergine salad, respectively. The performance criteria employed were graphical, together with the bias and accuracy factors as modi ed by Baranyi et al. (). The latter indices were also calculated for the models of Koutsoumanis et al. () and Skandamis and Nychas (), which were developed with the same data-sets. In addition, comparison of current models was made with the predictions (time for log reduction) of USDA Pathogen Modelling Program v?(buchanan ) for every combination of ph and temperature without oregano essential oil. Food validation was carried out with literature data, and with experimental data produced by challenge studies with E. coli O:H inoculated in commercial Greek appetizers, such as aubergine salad (ph?), taramasalad (ph?) and tzatziki (ph?) provided by Olympus Foods, all with or without the addition of oregano essential oil (?%, % and % v/w) at CandC. In particular, commercial salad packs were aseptically opened and after inoculation appropriate volumes of oregano essential oil were added to obtain the desired % v/w concentration. Thorough mixing was performed and all packs were placed in the experimental storage temperature until sampling. Survivor curves of E. coli O:H with respect to time were then generated to validate the model graphically. As far as literature validation was concerned, the values of ph and temperature that were reported in the articles were introduced in the current predictive models were used to obtain times for log or log reduction and these values were compared with the published data. The choice of response variable (time for log or log reduction) was dependent on the available information in each article. Results The results for both data-sets (i.e. taramasalad and aubergine salad) are presented in Figs and. The distribution-based model does not have a horizontal lower asymptote. The reason that Figs and can be seen to have a lower at portion is because the software always produces a surface, and when tted values would be below the axis, the surface is shown as horizontal. The upper plateau on these graphs is a result of two factors:. Results obtained at time zero, when plotted on a log (time) scale would appear at minus in nity. In order to have a nite axis these have been arbitrarily positioned at.this plateau would appear longer if minus in nity had been approximated by a larger negative number, and so its length is quite arbitrary. The upper plateau does not always appear level at extremely short times. A larger negative value to represent Log (zero) would rectify this, but has not been adopted because it would compress the more important part of the plot.. When a mild stress was used, there may be some delay before microbial numbers drop appreciably on a log scale, and produced a

4 P.N.Skandamisetal. Figure. Graphical illustration of normal distribution model for inactivation model of S. Enteritidis PT in taramasalad at various ph, temperatures and oregano essential oil concentrations. Open symbols are below surface and solid symbols are above surface. Surface represents the model.

5 Figure. Graphical illustration of normal distribution model for inactivation model of E. coli O:H NCTC in aubergine salad at various ph, temperatures and oregano essential oil concentrations. Open symbols are below surface and solid symbols are above surface. Surface represents the model. A vitalistic approach for non-thermal inactivation

6 P.N.Skandamisetal. short horizontal relationship of microbial numbers with time or log (time). Taramasalad The survival of S. Enteritidis PT in taramasalad implied a combined e ect of ph and storage temperature with oregano essential oil (Fig. ) that was evident in results obtained from samples stored at high temperatures and low ph. of E. coli O:H (Fig. ). In fact, the e ects of low ph and/or high concentrations of oregano essential oil can be summarized in two main conclusions: (a) increase in the death rate and (b) reduction in the survival period of E. coli O:H (Fig. ). It has not been possible to show a statistical relationship between temperature and pathogen response. These observations are re ected in the stress function of the developed model, following the tting procedure described above for taramasalad: Log N t N ¼ Log LogðtimeÞþ bph þ ceo þ dph fpheo gph þ heo ipheo : ðþ In such cases, there was a higher antimicrobial e ect in samples supplemented with the essential oil compared with the unsupplemented ones (Fig. ). It should also be noted that the inactivation of S. Enteritidis PT was higher at lower ph values. The log-reduction of S. Enteritidis PT vs time was modelled using a normal distribution, replacing the stress function of Eqn () with a function of temperature, ph and oregano essential oil concentration (EO). A nonlinear least-squares regression, resulted in the following equation: The model suggests that temperature is not a signi cant term within the experimental region, and these data can be modelled without taking its e ect into account. The parameter estimates are shown below, and the tted model is illustrated in Fig. together with observed data: ¼ ; b ¼ ; c ¼ ; d ¼ ; f ¼ ; ¼ ; g ¼ ; h ¼ ; i ¼ : Log N t ¼ Log Log ðtimeþþþaeo bphþct þ dt fteoþgpht N hpheo ieo jeo : ðþ The model ts well with the parameters listed below, and has an RMSE of?. ¼ ; a ¼ ; b ¼ ; c ¼ ; d ¼ E ; f ¼ ; g ¼ ; ¼ ; h ¼ ; i ¼ ; j ¼ ; where time is given in days, N t the bacterial population at time t and N the initial inoculum size. The graphical illustration of the model is given in Fig.. Aubergine salad ph and concentration of oregano essential oil strongly in uenced both the survival and death This model was a poor t (RMSE of?). The residuals were examined for clues as to why this should be, but there did not appear to be any obvious pattern there, indicating a certain tendency. Evaluation of model performance Table shows how the accuracy of the current models in predicting times for log reduction (Figs and ) is not quite as good as the models previously published (Koutsoumanis et al., Skandamis and Nychas ). It shows, however, that the current models are less biased. Figures (a) and (a) show how the current model gives predicted values that are in the same region as the observed values. For

7 A vitalistic approach for non-thermal inactivation Table. Bias and accuracy factors of previous models and vitalistic models of the present study. The response variable is time for log reduction (t D ) Model Food ^ micro-organism n a Temperature range Koutsoumanis et al. () Taramasalad inoculated with S. Enteritidis PT ph range Oregano essential oil Bias t D Accuracy t D b ^C?^??^% (v/w)?? Vitalistic model ^C?^??^% (v/w)?? Skandamis and ^C?^? ^?% (v/w)?? Nychas () Aubergine-salad inoculated with E. coli O:H NCTC Vitalistic model ^C?^? ^?% (v/w)?? a Number of curves. b Cases (out of in total) that contained insu cient data-points for calculation of t D values were excluded. Predicted time for log reduction (A) Predicted time for log reduction (B) Observed time for log reduction Observed time for log reduction Predicted time for log reduction (A) Predicted time for log reduction (B) Observed time for log reduction Observed time for log reduction Figure. Comparison of observed time for log reduction of S. Enteritidis PT in taramasalad and predicted time for log reduction by: (a) the vitalistic model, and (b) the model of Koutsoumanis et al. (). Figure. Comparison of observed time for log reduction of E. coli O:H NCTC in aubergine salad and predicted time for log reduction by (a) the vitalistic model, and (b) the model of Skandamis and Nychas (). the previous model however (Figs (b) and (b)), where observed values are low, the predicted values tend to be too high, and where observed values are high, the predictions are consistently low. Moreover, all models seemed to provide on average slightly overestimated

8 P. N. Skandamis et al. Table. Comparison of observed times for log reduction (t D ) and those predicted from USDA Pathogen Modelling Program (PMP) v?and the vitalistic models Strain Temperature (C) ph Oregano essential oil (v/w) Observed t D (days) a PMP predicted t D (days) Vitalistic predicted t D (days) S. Enteritidis?? PT?????????????????????? E. coli???? :H??????????????????????????? a In cases that the reduction of E. coli O:H in aubergine salad and of S. Enteritidis PT in taramasalad was smaller than log cfu g until the end of storage period, t D was indicated as and, respectively. These numbers correspond to the total experimental duration of challenge tests that were used to develop the relevant models. predictions ( fail-safe ), except for the model for Salmonella in taramasalad of Koutsoumanis et al. (). In contrast to the models of the present study, USDA Pathogen Modelling Program v?(pmp), predicted signi cant lower inactivation times ( fail-dangerous ) for Salmonella and signi cant higher inactivation times ( fail-safe ) for E. coli O:H (Table ). The PMP predictions of t D for E. coli O:H were generally longer than the vitalistic ones. For validation in food, a total of comparisons for E. coli O:H and comparisons for S. Enteritidis PT of times for log or log reduction in various foods are presented (Tables and ). The highest deviations between predictions and published data were evident for Salmonella, whereas predictions of model of E. coli O:H approximated the majority of published data in di erent foods, even in ph conditions outside the range of model development (Tables and ). One hundred per cent of Salmonella comparisons and % of those for E. coli O:H showed faster reduction times than those predicted by the models ( fail-safe ). Furthermore, among the three commercial Greek appetizers, better simulation of E. coli O:H survival was evident in aubergine salad and tzatziki (Figs and ). In contrast, the model failed to predict survival of pathogen in taramasalad apart from cases where no oregano essential oil was added (Fig. ). Discussion Bacterial inactivation can be in uenced by many factors utilized in traditional preservation systems (e.g. ph, A w, heat), emerging (ultra high pressure, electric pulse elds), or alternative natural preservatives (e.g. essential oils, bacteriocins). Many of these factors are

9 Table. Comparison of predicted and observed values for time for log and log reduction (t D, t D )ofs. Enteritidis PT, with observed data from literature Food substrate Strain(s) Temp. (C) Gilt-head seabream mixed with olive oil, oregano and lemon juice S. Enteritidis PT ph Observed t D (days) Predicted t D (days) Observed t D (days) Predicted t D (days) References? Tassou et al. () Diced tomatoes S. Baildon? Weissinger et al. () Home-made mayonnaise acidi ed with S. Enteritidis? Lock and Board () acetic acid and preincubated at C PT Home-made mayonnaise acidi ed with? Lock and Board () acetic acidfno preincubation Chicken salad made with real mayonnaise in the Mixture of? Erickson et al. () presence of indigenous ora (lactic acid bacteria) ATCC strains b Macaroni salad made with reduced calorie mayonnaise in the presence of indigenous ora (lactic acid bacteria)? Erickson et al. () Tzatziki (cucumber and yogurt salad ) S. Enteritidis? Tassou et al. () PT Tzatziki? Tassou et al., () Taramasalad ( sh roe salad) S. Enteritidis? Tassou et al. () PT Taramasalad ( sh roe salad)? Tassou et al. () a The time to log or log reduction was chosen for comparison depending on the available data in the literature. b,,,,,,,,,,,. A vitalistic approach for non-thermal inactivation

10 Table. Comparison of predicted and observed (literature) values of time for log and log reduction (t D, t D ) a of E. coli NCTC Food-substrate Temp. (C) ph Observed t D (days) Predicted t D (days) Observed t D (days) Predicted t D (days) References Strain(s) Ground roasted beef slurry using di erent acidulants (acetate, citrate, lactate) Buttermilk (di erent brands with di erent fat content) Sour cream (di erent brands from di erent types of milk) Yogurt (di erent brands with di erent fat content) Yogurt (traditional and bi do yogurt)? ^ ^ Abdul-Raouf et al. () Mixture c of C, P, B, A-CL, -?^? Dineen et al. ()??^? ^ Dineen et al. () SEA c ^ Dineen et al., ()? ^? Massa et al. () Ecoli(E-D) O:H VT Mayonnaise (reduced-calorie) d??? Hathcox et al. () Mixture b of C, E, F,, Blue cheese sauce (based on mayonnaise)? Weagant et al. () Mixture b of SEA c -, SEA-, SEA- Seafood sauce (based on mayonnaise)? Weagant et al. () Apple cider? d Ryu and Beuchat () EO (venison jerky isolate) Orange juice? d? Ryu and Beuchat () Apple cider d? d??semancheck and Golden () (human isolate) Salami (di erent initial A w (? and?))? ^? ^? Clavero and Beuchat () Mixture b of B, CA-, E, P, Dry sausage? Glass et al. () Mixture b of, CL,, P, B Fermented sausage? Du y et al. () Meat isolates a Thetimetolog or log reduction was chosen for comparison depending in the available data in the literature. b Mixture including human, beef and other meat isolates. c Shiga-like toxin producer. d Prediction outside temperature or ph range of model. P.N.Skandamisetal.

11 A vitalistic approach for non-thermal inactivation C % EO C.% EO (A) (B) Log CFU/g Log CFU/g C % EO (C) Log CFU/g C % EO (D) Log CFU/g Log CFU/g (E) C % EO C % EO Log CFU/g (F) Log CFU/g C % EO (G) Figure. Survival of E. coli O:H NCTC in commercial aubergine salad (ph?) stored at C (a^d) and C (e^g) without oregano essential oil (a,e), with?% v/w, (b) % v/w, (c,f) and % v/w (d,g) oregano essential oil; EO: essential oil, solid lines represent the simulation curve predicted by the model, dotted lines indicate the plating threshold and square symbols are the observed data points. utilized in manufacturing processes and product formulation, which also take advantage of interactive e ects.with respect to the traditional salads of the present study, the combination of chill temperatures, low ph and oregano essential oil were also found to signi cantly

12 P. N. Skandamis et al. Log CFU/g (A) (B) o C % EO o C.% EO Log CFU/g Log CFU/g (C) Log CFU/g (E) o C % EO o C % EO (D) Log CFU/g o C % EO o C % EO (F) Log CFU/g (G) Log CFU/g o C % EO Figure. Survival of E. coli O:H NCTC in commercial taramasalad salad (ph?) stored at C (a^d) and C (e^g) without oregano essential oil (a,e), with?% v/w, (b) % v/w, (c,f) and % v/w (d,g) oregano essential oil; EO: essential oil, solid lines represent the simulation curve predicted by the model, dotted lines indicate the plating threshold and square symbols are the observed data points. reduce the numbers of E. coli O:H and S. Enteritidis PT, in aubergine salad and taramasalad, respectively (Figs and ). In a previous work published by Koutsoumanis et al. () and Skandamis and Nychas (), a two-stage modelling procedure was used. This involved tting survival curves with the Baranyi model (as primary model), then expressing the survival kinetics as a quadratic function (secondary model). The choice of an appropriate mathematical model, as well as the identi cation and quanti cation of the con-

13 A vitalistic approach for non-thermal inactivation Log CFU/g o C % EO (A) Log CFU/g o C % EO (B) Log CFU/g (C) Log CFU/g (E) o C % EO o C % EO o C % EO Log CFU/g (D) Log CFU/g o C % EO (F) Figure. Survival of E. coli O:H NCTC in commercial tzatziki salad (ph?) stored at C(a, c, e) and C (b, d, f) without oregano essential oil (a,d), with % v/w, (b,e) and % v/w (c,f) oregano essential oil; EO: essential oil, solid lines represent the simulation curve predicted by the model, dotted lines indicate the plating threshold and square symbols are the observed data points. trolling factors that a ect micro-organisms in foods are two di cult tasks in predictive modelling (Baird-Parker and Kilsby ). The latter task is traditionally carried out through a reductionalist concept according to Ross and McMeekin (), which correlates microbial responses to speci c ecological determinants, such as ph, A w, temperature, etc.these factors are easy to adjust in liquid media. However, simplifying modelling by using broths with xed conditions may not lead to good predictions of microbial response in foods, especially solid or structured foods (Pin and Baranyi ). This is of great importance for the development of models for pathogen inactivation, since potential hazards should not be ignored and/or underestimated. In this regard, the models of the present study were both developed in foods. Moreover, the probabilistic approach of the present study (Kilsby et al. ) generally seemed to overcome the aforementioned de ciencies and provide an alternative, easier, and more exible modelling methodology for both foods, in a one-stage process. Models tted in two stages have a source of error associated with each stage, and seldom,

14 P.N.Skandamisetal. if ever, are the combined sources of variation compared with deviations of individual points from the model. For this reason, it is not easy to grasp how well any two-stage model agrees with the data from which it is derived. In contrast, single equation approaches enable data and the tted model to be presented on the same graphs in a more representative manner than that of other typical models. If one or more data series show a trend that di ers from the rest, it can be detected easily. The modeller can then consider improvements that could take the new trend into consideration, or alternatively highlight the need to investigate data from individual series more carefully. If there are su cient data series to describe the model well, it is possible to include some short series that do not show all the features of the longer ones, because they will not signi cantly a ect the estimation of parameters for the full model. Another advantage of a single equation model is the ability to check numerically how well it ts over the whole data space. From this comparison, a simple judgement can be made on the adequacy of the model in describing the observed microbial response. Individual components of such a model can describe relevant features such as slope or position of any shoulder/tail. It is worth mentioning that there is limited information on modelling non-thermal inactivation of bacterial counts as a single function of environmental variables. To asses a predictive model, especially one that relates to survival of pathogens, it is essential to determine its performance using both data with which it was developed, and independent data in the range of model validity. This will render the predictive model a useful tool for industry to assess the risks or bene ts of a new formulation or technology. There are models validated with literature data to determine whether they provide realistic estimates of microbial behaviour in foods, and identify possible factors (strain to strain variability, food formulation, presence of additional controlling factors, such as competitive ora and preservatives) that result in poor agreement of a model with independent data. The models of the present study were initially compared graphically and mathematically (using bias and accuracy factors) with two previous models developed for the same data-sets (Figs and ; Table ). Such comparison con rmed the internal validity of the models and showed that the suggested approach can be as e ective as the classical methodology (primary and secondary empirical models) used by the majority of modellers. It is important to have models that can predict microbial behaviour in real foods. Current models were therefore compared to those developed for liquid media, for the purpose of validation. Indeed, comparison of current models with PMP v?software, which is based on data from liquid media con rmed that the current models perform better (Table ). PMP v? includes general models that provide predictions for thermal inactivation of Cl. botulinum, and survival of E. coli O:H, Salmonella spp., Listeria monocytogenes and Staphylococcus aureus, for ph ranging from? to? adjusted with HCl and/or lactic acid (^%) (Buchanan ). It is acknowledged that models based on data from liquid media will not generally be totally accurate in their predictions for speci c food ecosystems. Moreover, it is strictly recommended that predictions of Software in foods to which ph is adjusted with organic acids are reliable only if the acidulant in food is lactic acid. If other acidulants are used (e.g. citric acid, acetic acid) then zero value should be selected for lactic acid input in Software Tables and predictions should be based upon the ph alone. It is notable, however, that under these circumstances, predictions of PMP v?software for E. coli O:H were signi cantly closer to the observed data than those for Salmonella (Table ). Speci cally, predictions of PMP were very conservative (high t D )in the fail-safe region for E.coli, whereas very low t D in the fail-dangerous region were given for Salmonella. In the next stage, literature and challenge tests with commercial Greek salads served as two sources of independent data for model validation. Comparison of models with literature data led to similar conclusions to those derived from comparison of current models with PMP v?. For instance, in Table, it seems that PMP does not predict well the response seen in taramasalad, and there are some large deviations between the model predictions for Salmonella inactivation and the published data

15 A vitalistic approach for non-thermal inactivation (Table ). However, the E. coli O:H model in aubergine salad indicated wider applicability in a variety of foods, including foods formulated and manufactured in a variety of ways (Table ). Since the responses of E. coli O:H in aubergine salad at the conditions of the present study seem to largely approach the general patterns of survival of E. coli in numerous foods with low ph, it is expected that the relevant model would be applicable too. Therefore, the poor agreement of Salmonella model with literature (Table ), raises the question of which are the factors that a ect Salmonella di erently compared to other foods of similar ph. Taramasalad is a homogeneous and nutritious product, which contains signi cant amounts of oily compounds (oil and sh fat), extensively dispersed throughout the whole food matrix. This should be considered in relation to assumption that foods with high fat content are likely to enhance survival of pathogens at low ph, either by containing fatty regions within which bacterial cells may be protected from exposure to antimicrobial agents, or by quenching of hydrophobic substances, such as essential oils or phenolics, within the fat content (Rico-Munoz and Davidson, ). Evidence for reduction of inhibitory potential of essential oil in such oily food as taramasalad was generated by model validation with commercial Greek appetizers supplemented with oregano essential oil (Figs ^). Indeed, challenge tests with E. coli O:H in commercial taramasalad, indicated that as high concentrations of oregano essential oil as % at ph? (of taramasalad) and C were incapable of reducing the pathogen more than log cfu g within days (Fig. ). Such behaviour could not be predicted by the model, since the experimental data that were used for its development in aubergine salad suggested that even?% oregano essential oil at the above ph and temperature can cause log decline in E. coli O:H counts within only days. In contrast, the vitalistic model seemed to predict better the survival of E.coli in commercial aubergine salad and tzatziki, which contain quite smaller levels of fat compared to taramasalad (Figs and ). The results of the present study could be summarized by the following points: () good performance of both models on the original data-sets, () PMP predictions did not approximate the survival model of S. Enteritidis PT and E.coli in aubergine salad and taramasalad, () vitalistic model S. Enteritidis PT predicted slower inactivation than data from other foods, () vitalistic model of E.coli O:H could predict survival of pathogen in commercial Greek salads supplemented or not with essential oil as natural preservative, and () PMP approximated the survival of E. coli in aubergine salad. Considering the wide range of other foods, the importance of oregano essential oil as natural preservative and the advantages of the present modelling approach, these results show the need for further investigation on the applicability of such approach in numerous foods with their safety being dependent on di erent controlling variables, including natural antimicrobials as alternative preservatives. Moreover, the present study attempted to address the need to consider the speci city and natural variation of food ecosystems and expanded data for pathogen survival that could be included in the application of tertiary models (McDonald and Sun ). Acknowledgements This research is partly funded by EU (DGXII), Project FAIR-ct-. References Abdul-Raouf, U. M., Beuchat, L. R. and Ammar, M. S. () Survival and growth of Escherichia coli O:H in ground, roasted beef as a ected by ph, acidulants and temperature. Appl. Environ. Microbiol., ^. Anderson, W. A., McClure, P. J., Baird-Parker, A. C. and Cole, M. B. () The application of a log-logistic model to describe thermal inactivation of Clostridium botulinum B at temperatures below?c. J. Appl. Microbiol., ^. Baird-Parker, A. C. and Kilsby, D. C. () Principles of predictive food microbiology. J. Appl. Bacteriol. S.S. S^S. Baranyi, J., Pin, C. and Ross,T. () Validating and comparing predictive models. Int. J. Food Microbiol., ^.

16 P. N. Skandamis et al. Blackburn, C. de W., Curtis, L. M., Humpheson, L., Billon, C. and McClure, P. J. () Development of thermal inactivation models for Salmonella enteritidis and Escherichia coli O:H with temperature, ph and NaCl as controlling factors. Int. J. Food Microbiol., ^. Breand, S., Fardel, G., Flandrois, J. P., Rosso, L. and Tomassone, R. () Model of the in uence of time and mild temperature on Listeria monocytogenes nonlinear survival curves. Int. J. Food Microbiol., ^. Buchanan, M. S., Golden, M. H.,Whiting, R. C., Philips, J. G. and Smith, J. L. () Non-thermal inactivation models for Listeria monocytogenes. J. Food Sci., ^. Buchanan, R. L. () Developing and distributing user-friendly application software. J. Ind. Microbiol., ^. Chhabra, A.T., Carter,W. H., Linton, R. H. and Cousin, M. A. () A predictive model to determine the e ects of ph, milk fat, and temperature on thermal inactivation of Listeria monocytogenes. J. Food Protect., ^. Clavero, M. R. and Beuchat, L. R. (). Survival of Escherichia coli O:H in broth and processed salami as in uenced by ph, water activity and temperature and suitability of media for its recovery. Appl.Environ.Microbiol., ^. Cole, M. B., Davies, K. W., Munro, G., Holyoak C. D. and Kilsby, D. C. (). A vitalistic model to describe the thermal inactivation of Listeria monocytogenes. J. Ind. Microbiol., ^. Dineen, S. S., Takeuchi, K., Soudah, J. E. and Boor, K. J. (). Persistence of Escherichia coli O:H in dairy fermentation systems. J. Food Prot., ^. Dodds, K. L. () An introduction to predictive microbiology and the development and use of probability models with Clostridium botulinum. J. Ind. Microbiol., ^. Du y, L. L., Grau, F. H. and Vanderlinde, P. B. () Acid resistance of enterohaemorrhagic and generic Escherichia coli associated with foodborne disease and meat. Int. J. Food Microbiol., ^. Erickson, J. P., Mickenna, D. N.,Woodru, M. A. and Bloom, J. S. () Fate of Salmonella spp. Listeria monocytogenes, and indigenous spoilage microorganisms in home-style salads prepared with commercial real mayonnaise or reduced calorie mayonnaise dressings. J. Food Prot., ^. Gaillard, S., Leguerinel, I. and Mafart, P., () Model for combined e ects of temperature, ph and water activity on thermal inactivation of Bacillus cereus spores. J. Food Sci., ^. Glass, K. A., Loe elholz, J. M., Ford, J. P. and Doyle, M. P. () Fate of Escherichia coli O:H as affected by ph or sodium chloride and in fermented, dry sausage. Appl. Environ. Microbiol., ^. Hathcox, A. K., Beuchat, L. R. and Doyle, M. P. () Death of Enterohemorrhagic Escherichia coli O:H in real mayonnaise and reduced-calorie mayonnaise dressing as in uenced by initial population and storage temperature. Appl. Environ. Microbiol., ^. Humpheson, L., Adams, M. R., Anderson, W. A. and Cole, M. B. () Biphasic thermal inactivation kinetics in Salmonella enteritidis PT. Appl. Environ. Microbiol., ^. Juneja,V. K. and. Eblen, B. S.(a) Predictive thermal inactivation model for Listeria monocytogenes with temperature, ph, NaCl, and sodium pyrophosphate as controlling factors. J. Food Prot., ^. Juneja, V. K., Marmer, B. S. and Eblen B. S. (b) Predictive model for the combined e ect of temperature, ph, sodium chloride, and sodium pyrophosphate on the heat resistance of Escherichia coli O : H. J. Food Safety, ^. Kamau, D. N., Doores, S. and Pruitt, K. M. () Enhanced thermal destruction of Listeria monocytogenes and Staphylococcus aureus by the lactoperoxidase system. Appl. Environ. Microbiol., ^. Kilsby, D. C., Davies, K. W., McClure, P. J., Adair, C. and Anderson, W. A. () Bacterial thermal death kinetics based on probability distributions: the heat destruction of two important food pathogens. J. Food Prot., ^. Koutsoumanis, K., Lambropoulou, K. and Nychas, G-J. E. (). A predictive model for the non-thermal inactivation of Salmonella enteritidis in a food model system supplemented with a natural antimicrobial. Int. J. Food Microbiol., ^. Linton, R. H., Carter,W. H., Pierson, M. D., Hackney, C. R. and Eifert, J. D. () Use of a modi ed Gompertz equation to predict the e ects of temperature, ph, and NaCl on the inactivation of Listeria monocytogenes Scott A heated in infant formula. J. Food Prot., ^. Little, C. L., Adams, M. R., Anderson,W. A. and Cole, M. B. () Application of a log^logistic model to describe the survival of Yersinia enterocolitica at sub-optimal ph and temperature. Int. J. Food Microbiol., ^. Lock,J.L.andBoard,R.G.()ThefateofSalmonella enteritidis PT in home-made mayonnaise prepared from arti cially inoculated eggs. Food Microbiol., ^. Massa, S., Altleri, C., Quaranta, V. and De Pace, R. (). Survival of Escherichia coli O:H in yoghurt during preparation and storage at C. Lt. Appl. Microbiol., ^. McClure, P. J., Beaumont, A. L., Sutherland, J. P. and Roberts, T. A. () Predictive modelling of growth of Listeria monocytogenes. The e ects on growth of NaCl, ph, storage temperature and NaNO. Int. J. Food Microbiol., ^. McDonald, K. and Sun, D-W. () Predictive food microbiology for the meat industry: a review. Int. J. Food Microbiol., ^.

17 A vitalistic approach for non-thermal inactivation Membre, J. M., Majchrzak, V. and Jolly, I () Effects of temperature, ph, glucose, and citric acid on the inactivation of Salmonella typhimurium in reduced calorie mayonnaise. J. Food Prot., ^. Pin, C. and Baranyi, J. () Predictive models as means to quantify the interactions of spoilage organisms. Int. J. Food Microbiol., ^. Rico-Munoz, E. and Davidson, P. M. () E ect of corn oil and casein on the antimicrobial activity of phenolic antioxidants. J. Food Sci., ^. Ross,T. and McMeekin,T. A. () Predictive microbiology. Int. J. Food Microbiol., ^. Ryu, J-H. and Beuchat, L. R. () In uence of acid tolerance responses on survival, growth, and thermal cross-protection of Escherichia coli O:H in acidi ed media and fruit juices. Int. J. Food Microbiol., ^. Semanchek, J. and Golden, D. A. (). Survival of Escherichia coli O:H during fermentation of apple cider. J. Food Prot., ^. Skandamis, P. and Nychas, G.-J. E. () Development and evaluation of a model predicting the survival of Escherichia coli O:H NCTC in home-made eggplant salad under various temperatures, ph and oregano essential oil concentrations. Appl.Environ.Microbiol., ^. Tassou, C. C., Drosinos, E. H. and Nychas, G. J. E. () E ects of essential oil from mint (Mentha piperita) on Salmonella enteritidis and Listeria monocytogenes in model food systems at Cand C. J. Appl. Bacteriol., ^. Tassou, C. C., Drosinos, E. H. and Nychas, G. J. E. () Inhibition of resident microbial ora and pathogen inocula on cold fresh llets in olive oil, oregano, and lemon juice under modi ed atmosphere or air. J. Food Prot., ^. Weagant, S. D., Bryant, J. L. and Jinneman, K. G. () An improved rapid technique for isolation of Escherichia coli O:H from foods. J. Food Prot., ^. Weissinger, W. R., Chantarapanont, W. and Beuchat, L.R. () Survival and growth of Salmonella baildon in shredded lettuce and diced tomatoes, and e ectiveness of chlorinated water as a sanitiser. Int. J. Food Microbiol., ^. Whiting, R. C., Sackitey, S., Calderone, S., Morely, K. and Phillips, J. G. (). Model for the survival of Staphylococcus aureus in nongrowth environments. Int. J. Food Microbiol., ^. Xiong, R., Xie, G., Edmodson, A. E. and Sheard, M. A. () A mathematical model for bacterial inactivation. Int. J. Food Microbiol., ^.