no growth models for Zygosaccharomyces rouxii associated with acidic, sweet intermediate moisture food products

International Journal of Food Microbiology 192 (2015) 51–57

Contents lists available at ScienceDirect

International Journal of Food Microbiology journal homepage: www.elsevier.com/locate/ijfoodmicro

Growth/no growth models for Zygosaccharomyces rouxii associated with acidic, sweet intermediate moisture food products C.L. Marvig, R.M. Kristiansen, D.S. Nielsen ⁎ Section of Food Microbiology, Department of Food Science, University of Copenhagen, Denmark

a r t i c l e

i n f o

Article history: Received 12 March 2014 Received in revised form 18 June 2014 Accepted 20 September 2014 Available online 28 September 2014 Keywords: Low water activity Low pH Osmophilic yeast spoilage G/NG model incorporating time

a b s t r a c t The most notorious spoilage organism of sweet intermediate moisture foods (IMFs) is Zygosaccharomyces rouxii, which can grow at low water activity, low pH and in the presence of organic acids. Together with an increased consumer demand for preservative free and healthier food products with less sugar and fat and a traditionally long self-life of sweet IMFs, the presence of Z. rouxii in the raw materials for IMFs has made assessment of the microbiological stability a significant hurdle in product development. Therefore, knowledge on growth/no growth boundaries of Z. rouxii in sweet IMFs is important to ensure microbiological stability and aid product development. Several models have been developed for fat based, sweet IMFs. However, fruit/sugar based IMFs, such as fruit based chocolate fillings and jams, have lower pH and aw than what is accounted for in previously developed models. In the present study growth/no growth models for acidified sweet IMFs were developed with the variables aw (0.65–0.80), pH (2.5–4.0), ethanol (0–14.5% (w/w) in water phase) and time (0–90 days). Two different strains of Z. rouxii previously found to show pronounced resistance to the investigated variables were included in model development, to account for strain differences. For both strains data sets with and without the presence of sorbic acid (250 ppm on product basis) were built. Incorporation of time as an exploratory variable in the models gave the possibility to predict the growth/no growth boundaries at each time between 0 and 90 days without decreasing the predictive power of the models. The influence of ethanol and aw on the growth/no growth boundary of Z. rouxii was most pronounced in the first 30 days and 60 days of incubation, respectively. The effect of pH was almost negligible in the range of 2.5–4.0. The presence of low levels of sorbic acid (250 ppm) eliminated growth of both strains at all conditions tested. The two strains tested have previously been shown to have similar tolerance towards the single stress factors included in the study, but when the stress factors were combined the two strains showed difference in their ability to grow illustrating the importance of including more strains when developing growth/no growth models. The developed models can be useful tools for development of new acidic sweet IMFs. © 2014 Elsevier B.V. All rights reserved.

1. Introduction Microbiological spoilage of food products has large economic consequences for the food industry and chemical preservatives have therefore for decades been used to prevent spoilage. However, now the food industry is facing increasing consumer demands for preservative free and healthier food products with less sugar and fat. Sweet intermediate moisture foods (IMFs), including jams, bakery products, and chocolate fillings, have generally been considered as microbiologically stable due to their low water activity, caused by high amounts of sugar and other soluble compounds, and low amounts of water (between 20 and 50% (w/w)). In addition, chemical preservatives, ⁎ Corresponding author at: Department of Food Science, Section for Food Microbiology, University of Copenhagen, Rolighedsvej 26, 1958 Frederiksberg C, Denmark. Tel.: +45 35333287. E-mail address: [email protected] (D.S. Nielsen).

http://dx.doi.org/10.1016/j.ijfoodmicro.2014.09.021 0168-1605/© 2014 Elsevier B.V. All rights reserved.

such as sorbic acid and benzoic acid, have traditionally been used to prevent spoilage of IMF products. Driven by consumer demands, IMF producers are increasingly developing sweet IMF products with less fat, sugar and preservatives, which have made the products much more exposed to spoilage by growth of osmophilic yeast and xerophilic moulds leading to visible growth, production of gas and off-flavours (Fleet, 1992; Mossel and Sand, 1968; Smith et al., 2004). Zygosaccharomyces rouxii is the most common spoilage organism in sweet IMFs (Fleet, 1992; Jermini et al., 1987; Martorell et al., 2005; Marvig et al., 2014). It is able to grow at relatively low aw (N0.65) and tolerates low pH values. Besides, Z. rouxii is very tolerant towards the preservatives sorbic acid and benzoic acid (Membré et al., 1999; Praphailong and Fleet, 1997; Restaino et al., 1983; Rojo et al., 2014; Tokuoka, 1993; Vermeulen et al., 2012). Z. rouxii is often present in the production environment, which makes it difficult to avoid contamination of the final products (Marvig et al., 2014). Therefore, the product recipes need to be designed with the aim of avoiding growth of spoilage

52

C.L. Marvig et al. / International Journal of Food Microbiology 192 (2015) 51–57

organisms. For this purpose predictive growth/no growth (G/NG) models have previously been developed for growth of Z. rouxii in sweet IMFs including the factors aw (0.76–0.88), pH (5.0–6.2), ethanol (0–14.5% (w/w) in water phase), sorbic acid (1500 ppm), acetic acid (0–3.5% (w/w) in water phase) and temperature (8–25 °C). It was here demonstrated that at 22–25 °C, the microbial stability of sweet IMFs cannot be guaranteed by simply lowering pH and aw to 5.0 and 0.76, respectively (Vermeulen et al., 2012, 2014). Several IMFs, such as fruit based chocolate fillings and jams, have lower aw and pH-values than included in the models of Vermeulen et al. (2012, 2014). Based on this, the present study developed additional G/NG models covering acidic sweet IMFs incorporating the factors aw, pH, ethanol and Time. To illustrate strain variability in model development, two datasets were generated; one for Z. rouxii MUCL 44388, which was also included in the G/NG models of Vermeulen et al. (2012, 2014) and one for Z. rouxii M-5-12-L, isolated from chocolate filling (Marvig et al., 2014). Both strains have shown high tolerance towards low aw, low pH, high ethanol concentration and presence of sorbic acid making them suitable model organisms (Marvig et al., 2014). The two datasets were generated with and without the presence of low concentrations of sorbic acid (250 ppm). 2. Material and methods 2.1. Preparation of growth media Sabouraud broth (SAB) (Oxoid, Basingstoke, UK) was used as basic medium for the growth of Z. rouxii strains. This medium was modified to mimic chocolate fillings as described previously (Vermeulen et al., 2012). In short, high amounts of sugar (50% (w/w), glucose (G-8270, Sigma-Aldrich, Steinheim, Germany) and fructose (F-0127, SigmaAldrich) in a 1:1 ratio) were added (modified SAB). The media were varying in (i) aw (0.65–0.80, four equidistant levels) by adding glycerol (Sigma-Aldrich), (ii) pH (2.5–4.0, four equidistant levels) by adding HCl (UN 1789, Merck, Darmstadt, Germany) and (iii) ethanol concentration (0–16 % (w/w) in water phase, five equidistant levels) (Merck). All these media were made with and without sorbic acid (250 ppm on total medium basis) added as potassium sorbate (Sigma-Aldrich). To establish the water activity, calibration curves were used describing aw as a function of the added concentrations of glycerol to the modified SAB (Vermeulen et al., 2012). The exact aw value was determined by a Labmaster aw equipment (Novasina, Stork Intermes NV, Berchem, Belgium). The pH was measured with a pH electrode (MeterLab, Lyon, France). 2.2. Inoculum preparation and inoculation For model development two strains were used: (i) Z. rouxii MUCL 44388, isolated from honey (Vermeulen et al., 2012) and (ii) Z. rouxii M-5-12-L, isolated from chocolate filling (Marvig et al., 2014). The strains were taken with an inoculation needle from stock cultures stored at −80 °C, inoculated in 5 ml Sabouraud medium supplemented with 15% (w/v) glucose-fructose (1:1 ratio) (supplemented SAB) and incubated at 25 °C for 48 h. The purity of the cultures was checked on plate count agar (PCA, Oxoid) supplemented with 15% (w/v) glucose-fructose (1:1 ratio) (supplemented PCA). To standardize the inoculums a second subculture was prepared by transferring 100 μl of the 48 h culture to 10 ml supplemented SAB and incubated at 25 °C for 48 h. To determine growth/no growth, optical density measurement data were used using the same procedure as described in Vermeulen et al. (2012, 2014). Briefly, 180 μl of the specific media was added to 96well microplates (TC-plates, Greiner Bio-one). To avoid changes in the specific media due to addition of the inoculum, the pre-culture was diluted to the appropriate inoculation level using the specific medium. Next, the inoculum culture (20 μl) was added to 96 well microtiterplates

to obtain a final inoculation density of approximately 4.5 log CFU/ml. The inoculum density was determined by plating on supplemented PCA. Finally, the microtiterplates were first covered with a Breath-easy film (Fiers NV, Kuurne, Belgium) and afterwards, closed with a lid. The lid was sealed with silicone (Bostik MultiFog 2640) and parafilm (Sigma-Aldrich, USA). The microtiter plates were stored in closed jars, where the relative humidity was kept constant by the presence of a glycerol-solution with the same aw as the media in the plate. 2.3. Data generation for model development All combinations of aw (aw = 0.65, 0.70, 0.75 and 0.80), pH (pH = 2.5, 3.0, 3.5 and 4.0) and ethanol (concentrations of 0%, 4.5%, 9% and 14.5% (w/w) in water phase) were tested in a full factorial design in 16 replicates at 25 °C for 90 days. Separate data sets were generated for the models with Z. rouxii MUCL 44388 and Z. rouxii M-5-12-L with and without sorbic acid. The optical density (OD) of the media was measured at 600 nm using a microplate reader (Varioskan flash, ThermoScientific, Denmark). Before each measurement the plates were agitated for 1 min. The wells were measured for 90 days two times a week (on average), leading to at least 20 data points for each growth curve. Growth was defined when OD was consistently above the average plus three times the standard deviation of the blank. Data were automatically processed using a custom Excel® macro application (Vermeulen et al., 2012). OD–growth curves of all wells were visually checked to prevent false positive results (e.g. air bubbles). At the end of the incubation period, strain purity was checked by streaking out on supplemented PCA agar. Wells that did not show any turbidity were directly plated on the same agar to assess whether inactivation occurred. 2.4. Development of growth/no growth models The G/NG data were used to develop G/NG models for Z. rouxii MUCL 44388 and Z. rouxii M-5-12-L with pH, aw and ethanol concentration as explanatory variables, after 30, 60 days and 90 days of incubation (3-dimensional models). Additional G/NG models were established in which time was incorporated as an extra explanatory variable giving a 4-dimensional model. No models were developed for the datasets with sorbic acid since growth was not observed in any of the conditions tested. In all cases a logistic regression model was used to describe the data. The equation of this model consists of a polynomial (right-hand   p (left-hand side) with p the probability side) and logit (p) = ln 1‐p of growth (Eqs. (1) and (2)). 2

logit ðpÞ ¼ b0 þ b1  aw þ b2  pH þ b3  Eth þ b4  aw 2 2 þb5  pH þ b6  Eth þ b7  aw  pH þ b8  aw ð1Þ  Eth þ b9  pH  Eth

logit ðpÞ ¼ b0 þ b1  aw þ b2  pH þ b3  Eth þ b4  Time 2 2 2 2 þb5  aw þ b6  pH þ b7  Eth þ b8  Time þb9  aw  pH þ b10  aw  Eth þ b11  aw  Time þ b12  pH  Eth þ b13  pH  Time þ b14  Eth  Time

ð2Þ In these equations Eth (% (w/w)) is the ethanol concentration in the water phase. aw, pH and time are, respectively the water activity, pH and time in days. The water phase was defined as the mass of water added to the medium (Dang et al., 2011). bi (i = 0, …, 14) are the parameters to be estimated. The models were fitted in SPSS 21.0 (SPSS, Inc., Chicago IL., USA) using linear logistic regression, according to the procedure described in Vermeulen et al. (2007). Briefly, this implies that the main effects of aw, pH, ethanol and time were forced to stay in the final model equation independent on their p-value. The quadratic and

C.L. Marvig et al. / International Journal of Food Microbiology 192 (2015) 51–57

interaction terms were selected by the forward stepwise procedure, based on the significance of the likelihood-ratio criterion (p = 0.001). Model building stopped when no more variables met entry or removal criteria. The predicted G/NG interfaces were plotted in Matlab®R2012b (Mathworks, Inc., Natick, MA, USA). Goodness-of-fit statistics considered were: (i) − 2 log L with L the likelihood in its optimum, (ii) Akaike's Information Criterion (AIC = −2 (maximum log likelihood − number of variables in model)) and (iii) Hosmer–Lemeshow statistic. The predictive power was measured by c (the concordance index), which is equal to the area under the ROC-curve (Receiver Operating Curve) (Agresti, 2002). 3. Results and discussion 3.1. Effect of time For each of the two Z. rouxii strains three 3-dimensional models (a w, pH and ethanol (% w/w in water-phase)) were generated for the datasets without the presence of sorbic acid (Eq. (1)). In addition, 4-dimensional models incorporating time was fitted for the complete datasets without the presence of sorbic acid (Eq. (2)). The parameter estimates with standard deviations, and the goodness-of-fit statistics are given in Tables 1 and 2 for the two strains. The goodness-of-fit statistics and predictive power indicate a relatively high c-value and % predicted correct for all of the models. Also the Hosmer–Lemeshow statistic, − 2lnL and AIC values indicate good fit of the models. The Z. rouxii M-5-12-L model for 90 days shows higher Hosmer–Lemeshow values than the other models indicating a worse fit of the data. However, this statistic parameter is strongly influenced by one single poor prediction and should be interpreted with caution (Gysemans et al., 2007). The 3D-models for 30 and 60 days for Z. rouxii MUCL 44388 had standard deviations larger than estimates. Since the goodness-of-fit statistics are good, it is probably caused by an uneven distribution of growth and no growth data. It makes the estimates non-significant and therefore only the 3D-model for 90 days is included for the Z. rouxii MUCL 44388 strain. Incorporation of time as an extra explanatory variable includes more growth data and solved this problem. The G/NG boundaries for the 4D-models, with time incorporated as

53

an extra explanatory variable, follow the boundaries created in the 3D-models (Fig. 1). Only the curve for the 3D-models of 30 days for the Z. rouxii M-5-12-L strain differs from the 4D-models (Fig. 1A). It was, therefore, concluded that incorporation of time as an extra explanatory variable increased the reliability of the models and the 4D-models were used. For acidic sweet IMFs time is an important factor, since the shelf-life of the products in general is very long. It is therefore of great interest for the confectionary industry to have time as an explanatory variable in the models. The latter has previously been tried by Vermeulen et al. (2012), who experienced a loss of predictive power when incorporating time as an explanatory variable. However, here time was incorporated as a fifth factor and the number of factors might have influenced the predictive power more than the factor time itself. Incubation time up to 60 days had an influence on the G/NG boundaries, whereas the change of the G/NG boundaries from 60 to 90 days was limited. During the first 60 days of incubation, the G/NG boundary moved towards more stringent aw conditions, whereas only limited change in the ethanol concentration at which the organisms started to grow was observed after 30 days of incubation (Fig. 2). In the G/NG model made by Vermeulen et al. (2012) (a w : 0.76–0.88, ethanol: 0–14.5%, pH: 5.0–6.2, acetic acid: 0–3.5%) it was concluded that the major effect of the environmental stress conditions could be observed after 30–40 days of incubation, which was confirmed by the results of (Vermeulen et al., 2014). However, this was only valid for products stored at room temperatures. When the storage temperature was lowered below 12 °C the G/NG boundaries changed during the first 60 days of incubation (Vermeulen et al., 2014). It illustrates the importance of following the G/NG experiments for long enough time to ensure observation of the full probability of growth with the time being dependent on the stringency and type of environmental stress factors. 3.2. Effect of pH, ethanol and water activity The two 4D-models developed for the two strains of Z. rouxii both showed that the effect of pH was almost negligible between pH 2.5 and 4.0 when no ethanol was present. In the presence of ethanol a decrease in growth probability was observed for Z. rouxii M-5-12-L at

Table 1 Parameter estimates and the goodness-of-fit statistics for the growth/no growth models for Z. rouxii MUCL 44388. Variables

aw Eth pH Time aw2 Eth2 pH2 Time2 aw • Eth aw • pH aw • time Eth • pH Eth • time pH • time Constant

3D model—30 days

3D model—60 days

3D model—90 days

4D model

Estimate

St. dev.

Estimate

St. dev.

Estimate

St. dev.

Estimate

St. dev.

N.S. N.S. N.S. – N.S. N.S. N.S. – N.S. N.S. – N.S – – N.S

N.S. N.S. N.S. – N.S. N.S. N.S. – N.S. N.S. – N.S – – N.S.

152.804 N.S. 20.665 – N.S. N.S. −2.803 – N.S. N.S. – N.S. – – −143.372

26.971 N.S. 4.991 – N.S. N.S. 0.747 –– N.S. N.S. – N.S. – – 21.197

559.793 52.488 0.955 – N.S. N.S. N.S. – −81.352 N.S. N.S. – – −381.577

56.231 5.407 0.318 – N.S. N.S. N.S. – 8.309 N.S. – N.S. – – 38.355

1255.080 7.735 1.780 0.751 −651.650 −1.573 N.S. −0.004 N.S. N.S. N.S. N.S. −0.059 N.S. −596.692

254.915 0.636 0.235 0.060 167.931 0.127 N.S. 0.000 N.S. N.S. N.S. N.S. 0.005 N.S. 97.588

Goodness-of-fit statistic −2log L AIC Hosmer–Lemeshow c-value % correct predicted

27.841 35.841 0.000 1.000 99.7

–: not applicable; N.S.: not significant.

222.895 234.895 12.676 99.3 96.3

233.694 241.694 12.317 99.6 97.9

484.520 500.520 16.300 99.8 98.3

54

C.L. Marvig et al. / International Journal of Food Microbiology 192 (2015) 51–57

Table 2 Parameter estimates and the goodness-of-fit statistics for the growth/no growth models for Z. rouxii M-5-12-L. Variables

aw Eth pH Time a2w Eth2 pH2 Time2 aw • Eth aw • pH aw • Time Eth • pH Eth • Time pH • Time Constant

3D model—30 days

3D model—60 days

3D model—90 days

4D model

Estimate

St. dev.

Estimate

St. dev.

Estimate

St. dev.

Estimate

St. dev.

−1012.429 33.399 −365.556 – N.S. N.S. −1.769 – −28.074 547.556 – −5.204 – – 660.400

804.511 23.300 268.870 – N.S. N.S. 0.854 – 17.679 394.334 – 4.215 – – 537.643

−94.045 −3.273 −41.068 – N.S. −0.214 N.S. – 6.082 61.894 – N.S. – – 59.263

37.405 1.268 9.074 – N.S. 0.031 N.S. – 2.001 12.935 – N.S. – – 26.370

1157.973 −5.083 −48.379 – −882.162 −0.156 N.S. – 7.784 70.834 – N.S. – – −379.206

267.458 1.252 9.505 – 178.310 0.025 N.S. – 1.861 13.548 – N.S. – – 101.223

871.409 −3.210 −21.721 1.016 −572.665 −0.120 −1.187 −0.003 4.899 39.827 −0.891 N.S. N.S. −0.039 −354.621

145.686 0.660 5.433 0.138 94.336 0.012 0.344 0.000 0.958 5.945 0.147 N.S. N.S. 0.000 58.114

Goodness-of-fit statistic −2log L AIC Hosmer–Lemeshow c-value % correct predicted

147.660 161.66 0.146 99.6 96.9

341.669 353.669 30.200 98.6 95.7

277.899 291.899 250.992 98.7 97.9

967.717 991.717 63.260 99.1 96.1

–: not applicable; N.S.: not significant.

pH 2.5 (Fig. 3). The decrease in growth probability was only observed for Z. rouxii M-5-12-L, indicating a higher tolerance towards low pH for strain Z. rouxii MUCL 44388. It has previously been reported that pH values below 2.5 (Membré et al., 1999) or 2.8 (Rojo et al., 2014) were necessary to change the length of the lag-phase for strains of Z. rouxii even at low aw (0.788 and 0.744, respectively). Praphailong and Fleet (1997) suggested that the ability to initiate growth at low

pH is even increased by low aw. The present results show that at aw of 0.70 and in the presence of ethanol the growth of Z. rouxii is hardly influenced by lowering the pH down to 2.5. It is, therefore, not sufficient for the confectionary industry to use pH as a factor to eliminate growth of spoilage organisms unless organic acids are present in the product.

Fig. 1. Growth/no growth boundaries for 3D-models (thin lines) and 4D-models (bold lines) at pH 4.0 for (A) Z. rouxii M-5-12-L after 30 days, (B) Z. rouxii M-5-12-L after 90 days and (C) Z. rouxii MUCL 44388 after 90 days of incubation. Lines represent the logistic regression model predictions p = 0.9 (-), p = 0.5 (–), p = 0.1 (…). Growth zone is above line and nogrowth zone is below lines. (+) p = 1, (○) p = 0 and (Δ) p ϵ ]0,1[ with the measured percentage of wells showing growth indicated.

C.L. Marvig et al. / International Journal of Food Microbiology 192 (2015) 51–57

55

Fig. 2. Growth/no growth boundaries for 4D models at pH 4 for Z. rouxii M-5-12-L after (A) 30 days, (B) 60 days and (C) 90 days of incubation. Lines represent the logistic regression model predictions p = 0.9 (-), p = 0.5 (–), p = 0.1 (…).Growth zone is above line and no-growth zone is below lines. (+) p = 1, (○) p = 0 and (Δ) p ϵ ]0,1[ with the measured percentage of wells showing growth indicated.

Ethanol and aw had a large effect on the growth probability of the two strains (Fig. 4). For both strains an aw of 0.67 was necessary to avoid growth alone. With increasing concentrations of ethanol the G/NG boundaries shifted to higher aw values. In general, Z. rouxii M-512-L showed highest tolerance towards ethanol and at pH 4.0 and aw 0.80 the strain was able to grow in up to 14.5% ethanol (w/w, in water phase), while Z. rouxii MUCL 44388 only tolerated 4.7% ethanol at aw 0.80. However, the tolerance of Z. rouxii M-5-12-L towards ethanol was dependent on pH and at pH 2.5 the strain only tolerated 10% ethanol. For Z. rouxii MUCL 44388 the effect of aw and ethanol seemed unaffected by pH. The actual variable estimates cannot be compared directly to each other because the units of the aw, pH, ethanol and time values were not standardized. However, the observations described above are in

agreement with the interaction terms found significant (Tables 1 and 2). The Z. rouxii MUCL 44388 is mainly influenced by aw and ethanol, whereas most of the interaction terms for Z. rouxii M-5-12-L are significant, which means that all of the variables tested are relevant for the growth/no growth boundary of this strain. Only ethanol in combination with time and ethanol in combination with pH is not significant, which was also concluded from the graphs (Figs. 3 and 4). 3.3. Comparison of two Z. rouxii strains Two strains of Z. rouxii, which have previously been shown to have similar high tolerance towards single stress factors, such as low aw, ethanol and low pH (Marvig et al., 2014), were included in this study to test the variability when combinations of stress factors were used. The two

Fig. 3. Growth/no growth boundaries for 4D models after 90 days of incubation for (A) Z. rouxii M-5-12-L with 0% [w/w] ethanol (bold lines) and 9% [w/w] ethanol (thin lines)and (B) Z. rouxii MUCL 44388 with 0% [w/w] ethanol (bold lines) and 4.5% [w/w] ethanol (thin lines). Lines represent the logistic regression model predictions p = 0.9 (-), p = 0.5 (–), p = 0.1 (…).Growth zone is above line and no-growth zone is below lines.

56

C.L. Marvig et al. / International Journal of Food Microbiology 192 (2015) 51–57

Fig. 4. Growth/no growth boundaries for 4D models after 90 days of incubation for (A) Z. rouxii M-5-12-L and (B) Z. rouxii MUCL 44388 at pH 2.5 (bold lines) and pH 4.0 (thin lines). Lines represent the logistic regression model predictions p = 0.9 (-), p = 0.5 (–), p = 0.1 (…).Growth zone is above line and no-growth zone is below lines.

strains showed different tolerance towards the different stress factors when tested in combination and, especially, the ability to start growing in the presence of ethanol was different between the two strains. Z. rouxii M-5-12-L was generally the most tolerant of the two strains when more than one stress factor was applied and it showed very high ability to grow even at high ethanol concentrations. However, after 30 days of incubation and when no ethanol was present in the media Z. rouxii MUCL 44388 seemed to have the largest growth interval (Fig. 5). Inter-strain variability has previously been reported for Z. rouxii (Praphailong and Fleet, 1997; Vermeulen et al., 2012). The inter-strain variability observed in the present and previous studies illustrates the importance of selecting the correct model organism, but also the importance of selecting several strains of a given organism when developing G/NG models. 3.4. Effect of presence of sorbic acid Sorbic acid is commonly used by the confectionary industry and often at the maximum concentration allowed (1500 ppm, EU, 1129/ 2011, 2011). It has previously been shown that the effect of sorbic acid is very pH dependent and pH values below 5.4 were necessary to eliminate growth of Z. rouxii at aw-values above 0.80 and sorbic acid concentrations of 1500 ppm (Vermeulen et al., 2014). The purpose of the present study was to test if the presence of sorbic acid in only low concentrations was sufficient to avoid growth of Z. rouxii in sugar/fruit based fillings (pH below 4.0). In the presence of 250 ppm (on product basis) no growth was observed for both strains of Z. rouxii tested (results not shown). These results are in agreement with results of Praphailong and Fleet (1997) and Martorell et al. (2007), who found pH values of 3.0 and 4.0 to be sufficient to avoid growth of Z. rouxii when 250 mg l−1 and

Fig. 5. Growth/no growth boundaries for 4D models after 30 days of incubation for Z. rouxii M-5-12-L (thin lines) and Z. rouxii MUCL 44388 (bold lines) at pH 4.0. Lines represent the logistic regression model predictions p = 0.9 (-), p = 0.5 (–), p = 0.1 (…).Growth zone is above line and no-growth zone is below lines.

320 mg l−1 of sorbic acid were added, respectively. It should be noted that these results were based on incubations for only four weeks, which was shown by Vermeulen et al. (2012, 2014) not to be sufficient time to detect full probability of growth under these conditions. The typical aw, pH and ethanol concentrations of chocolate praline fillings are in the range of 0.64–0.90, 2.7–5.8 and 0–10.8% [w/w] of product basis, respectively. Sorbic acid is added to most of the industrial produced chocolate pralines in a concentration of 150–1500 ppm (Marvig et al., 2014; and unpublished results). Based on the model predictions it can be concluded that the aw, pH and ethanol levels in most cases are not sufficient alone to limit growth of spoilage organisms and caution needs to be taken to ensure microbiological stability of the products. Such caution is today taken by the industry by addition of sorbic acid. Since the present study shows that at pH values below 4.0 and aw below 0.80, low levels of sorbic acid (250 ppm) were sufficient to eliminate growth of spoilage organisms in sweet IMFs, it can be concluded that even if the industry cannot completely avoid use of preservatives, they can decrease it to low amounts by lowering the pH in the product. It would thus be interesting to include sorbic acid at several levels in future G/NG models to provide a tool for decreasing the use of sorbic acid in the confectionary industry. 3.5. Comparison with previous model for Z. rouxii MUCL 44388 The developed models for sweet IMFs by Vermeulen et al. (2012, 2014) showed that the most stringent conditions (pH 5.0 and aw 0.76) were not sufficient to achieve microbial stability of sweet IMFs and more than 10% of ethanol (w/w on waterphase) was needed to prevent growth of Z. rouxii MUCL 44388. In the present study, models for acidic sweet IMFs (pH 2.5–4.0, aw 0.65–0.80) found the growth boundary for Z. rouxii MUCL 44388 to be at aw 0.67 when no ethanol was added. Increasing levels of ethanol was necessary to reach the G/NG boundary at increasing aw levels and at pH 4.0 and aw 0.76 ethanol concentrations of 4.7% (w/w in water phase) were necessary to prevent growth. This concentration is much lower than concluded by Vermeulen et al. (2012, 2014), which might be due to the difference in pH (4.0 and 5.0 in the present study and the studies of Vermeulen et al. (2012, 2014), respectively). However, this is not expected to be the case since the influence of pH in both studies has been shown to be negligible. Another explanation could be the span of the G/NG boundary, which is much broader in the models of Vermeulen et al. (2012, 2014). The models were developed with relatively high inoculum sizes to ensure the observation of growth by optical density measurements. This inoculation size is not unrealistic since microbial levels up to 105 CFU/g have been found in fillings of chocolate pralines and sugar syrups used for their production (Marvig et al., 2014). Inoculation size is shown to affect the G/NG boundary, which shift towards more stringent conditions at higher inoculation levels (Koutsoumanis and Sofos, 2005; Steels et al., 2000).

C.L. Marvig et al. / International Journal of Food Microbiology 192 (2015) 51–57

It should be noted, that the developed models all need validation in real food products, where other variables and a different viscosity may influence the probability of growth. Because of the simple model system, the low viscosity and high inoculation level of the model medium, the growth probability is expected to be higher in the model than in real products and the model will be fail-safe. 4. Conclusion In this study G/NG models were developed for prediction of growth of Z. rouxii in acidic sweet IMFs incorporating the factors aw, pH, and ethanol. Time was successfully incorporated as a fourth factor. Two strains of Z. rouxii were tested showing the importance of including several organisms and strains when developing G/NG models. Together with the models developed by Vermeulen et al. (2012, 2014) these models provide a useful tool in the development of sweet and acidic sweet IMFs with lower content of preservatives, fat and sugar. References Agresti, A., 2002. Categorical Data Analysis. Wiley, New York. Dang, T.D.T., Vermeulen, A., Mertens, L., Geeraerd, A.H., Van Impe, J.F., Devlieghere, F., 2011. The importance of expressing antimicrobial agents on water basis in growth/ no growth interface models: a case study for Zygosaccharomyces bailii. Int. J. Food Microbiol. 145, 258–266. EU 1129/2011, 2011. Amending Annex II to Regulation 1333/2008/EC by Establishing a Union List of Food Additives. Fleet, G., 1992. Spoilage yeasts. Crit. Rev. Biotechnol. 12, 1–44. Gysemans, K.P.M., Bernaerts, K., Vermeulen, A., Geeraerd, A.H., Debevere, J., Devlieghere, F., Van Impe, J.F., 2007. Exploring the performance of logistic regression model types on growth/no growth data of Listeria monocytogenes. Int. J. Food Microbiol. 114, 316–331. Jermini, M., Geiges, O., Schmidt-Lorenz, W., 1987. Detection, isolation and identification of osmotolerant yeasts from high-sugar products. J. Food Prot. 50, 468–472. Koutsoumanis, K.P., Sofos, J.N., 2005. Effect of inoculum size on the combined temperature, pH and aw limits for growth of Listeria monocytogenes. Int. J. Food Microbiol. 104, 83–91.

57

Martorell, P., Fernández-Espinar, M.T., Querol, A., 2005. Molecular monitoring of spoilage yeasts during the production of candied fruit nougats to determine food contamination sources. Int. J. Food Microbiol. 101, 293–302. Martorell, P., Stratford, M., Steels, H., Fernández-Espinar, M.T., Querol, A., 2007. Physiological characterization of spoilage strains of Zygosaccharomyces bailii and Zygosaccharomyces rouxii isolated from high sugar environments. Int. J. Food Microbiol. 114, 234–242. Marvig, C.L., Kristiansen, R.M., Nielsen, D.S., 2014. Identification and characterisation of organisms associated with chocolate pralines and sugar syrups used for their production. Int. J. Food Microbiol. 185, 167–176. Membré, J.M., Kubaczka, M., Chéné, C., 1999. Combined effects of pH and sugar on growth rate of Zygosaccharomyces rouxii, a bakery product spoilage yeast. Appl. Environ. Microbiol. 65, 4921–4925. Mossel, D.A.A., Sand, F., 1968. Occurrence and prevention of microbial deterioration of confectionery products. Conserva 17, 23–32. Praphailong, W., Fleet, G., 1997. The effect of pH, sodium chloride, sucrose, sorbate and benzoate on the growth of food spoilage yeasts. Food Microbiol. 14, 459–468. Restaino, L., Bills, S., Tscherneff, K., Lenovich, L.M., 1983. Growth characteristics of Saccharomyces rouxii isolated from chocolate syrup. Appl. Environ. Microbiol. 45, 1614–1621. Rojo, M.C., Arroyo López, F.N., Lerena, M.C., Mercado, L., Torres, A., Combina, M., 2014. Effects of pH and sugar concentration in Zygosaccharomyces rouxii growth and time for spoilage in concentrated grape juice at isothermal and non-isothermal conditions. Food Microbiol. 38, 143–150. Smith, J.P., Daifas, D.P., El-Khoury, W., Koukoutsis, J., El-Khoury, A., 2004. Shelf life and safety concerns of bakery products—a review. Crit. Rev. Food Sci. Nutr. 44, 19–55. Steels, H., James, S.A., Roberts, I.N., Stratford, M., 2000. Sorbic acid resistance: the inoculum effect. Yeast 16, 1173–1183. Tokuoka, K., 1993. Sugar- and salt-tolerant yeasts. J. Appl. Bacteriol. 74, 101–110. Vermeulen, A., Daelman, J., Van Steenkiste, J., Devlieghere, F., 2012. Screening of different stress factors and development of growth/no growth models for Zygosaccharomyces rouxii in modified Sabouraud medium, mimicking intermediate moisture foods (IMF). Food Microbiol. 32, 389–396. Vermeulen, A., Devlieghere, F., Bernaerts, K., Van Impe, J., Debevere, J., 2007. Growth/no growth models describing the influence of pH, lactic and acetic acid on lactic acid bacteria developed to determine the stability of acidified sauces. Int. J. Food Microbiol. 119, 258–269. Vermeulen, A., Marvig, C.L., Daelman, J., Xhaferi, R., Nielsen, D.S., Devlieghere, F., 2014. Strategies to increase the stability of intermediate moisture foods towards Zygosaccharomyces rouxii: The effect of temperature, ethanol, pH and water activity, with or without the influence of organic acids. Food Microbiol. 1–7.