Sie sind auf Seite 1von 7

ARTICLE IN PRESS

LWT 41 (2008) 15281534 www.elsevier.com/locate/lwt

Research Note

pH spatial distribution model during ripening of Camembert cheese


S. Liu, V.M. Puri
Department of Agricultural and Biological Engineering, Penn State University, University Park, PA, USA Received 8 June 2007; received in revised form 20 September 2007; accepted 24 September 2007

Abstract During ripening of Camembert cheese, a soft cheese, the pH values continually change, which impacts the growth of Listeria monocytogenes. In this study, a pH distribution model suitable for the ripening phase of Camembert cheese was developed and veried. An experimental trend-based statistical model for pH using normalized time and normalized pH as variables was developed to determine the evolution of pH. The pH model showed good agreement with the mean pH measured values, i.e., the pH model was able to capture the magnitudes and trends sufciently. The R2 values for top surface, center, innerouter side surface, and bottom surface regions mean measured and pH model-predicted values were 0.97, 0.95, 0.99, and 0.99, respectively. r 2007 Swiss Society of Food Science and Technology. Published by Elsevier Ltd. All rights reserved.
Keywords: Modeling; Measurement; pH; Ripening; Camembert cheese

1. Introduction Listeria monocytogenes is an important foodborne pathogen that causes listeriosis, a disease with a high mortality rate (George, Richardson, & Peck, 1996). Outbreaks of listeriosis have been associated with soft cheese. L. monocytogenes can grow between 0.4 and 50 1C and at pH values between 4.5 and 7.0 with no growth at pH 4.0 and lower (Farber & Peterkin, 1991). Mathematical models can be either empirical or mechanistic (France & Thornley, 1984; McMeekin & Ross, 2002). Mechanistic models are preferred because they are derived to represent the biochemical processes. However, if the mechanism governing the process is either unknown or partially understood, mathematical functions have to be used empirically, and models suitability is evaluated from its ability to t experimental data based on statistical criteria. Empirical models can evolve to more mechanistic models as additional information concerning the system becomes available (Baranyi & Roberts, 1995). However, no models were found that could predict the spatialtemporal pH distribution in foods, especially in
Corresponding author. University of Nebraska, 143 Filley Hall, Lincoln, NE 68503, USA. Tel.: +1 472 326 8938. E-mail address: liushaowei@hotmail.com (S. Liu).

Camembert cheese. Therefore, the objectives of this study were to develop and verify an experimental trend-based statistical model for predicting the spatial distributions of pH during ripening of Camembert cheese. 2. Materials and methods Since the materials and methods are identical to those given in Liu and Puri (2005), for completeness, these are briey summarized. Pasteurized whole milk (Penn State University Creamery) and rennet (Chr. Hansen Inc., Milwaukee, WI) were used throughout the study for making Camembert cheese. In this study, cylindrical Camembert cheese blocks (diameter 108 mm and height 32 mm) were manufactured. They are also the dimensions of commercially sold Camembert cheese blocks. A commercial starter culture and Pencillium candidum (Chr. Hansen Inc.) were used to make Camembert cheese. Un-ionized salt was used to brine the cheese in this study. The step-by-step process of making Camembert cheese is given in Liu and Puri (2005). Typical dimensions of cylindrical Camembert cheese blocks were 108 mm in diameter and 32 mm in height manufactured in this study. The cheese blocks shape and size used in this study correspond to the commercially sold Camembert cheese. For each set of experiments, three

0023-6438/$34.00 r 2007 Swiss Society of Food Science and Technology. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.lwt.2007.09.010

ARTICLE IN PRESS
S. Liu, V.M. Puri / LWT - Food Science and Technology 41 (2008) 15281534 1529

batches of Camembert cheese were manufactured (Fig. 1), eight cheese blocks corresponding to 8 measurement days (1, 5, 10, 15, 20, 25, 30, and 35) were made for each batch. To verify angular symmetry, three cross-sections 1201 apart were used for measurements. Three sections of each block were used to measure the pH during ripening of cheese. Camembert cheese forms a 45 mm deep crust on the surface after 23 days of ripening. Therefore, the closest measurement of pH value to the surface that was practically feasible was 6 mm below the surface (Fig. 1). The pH values were measured on day 1, 5, 10, 15, 20, 25, 30, and 35 of ripening. Camembert cheese is distributed for consumption after 45 weeks of ripening (Radhakrishnan, 2000). For one set, three batches of Camembert cheese were made. Eight cheese blocks corresponding to 8 measurement

days were made for each batch. Three sections 1201 apart of each block were used to measure the pH values during ripening of cheese. So, the effect of different sections within Camembert cheese could be evaluated. There were 16 measurement locations in each section of cheese (Fig. 1). Therefore, for each cheese block, 48 locations of pH values were measured. The detail of one set of experimental design for cheese is shown in Fig. 2. 3. Statistical analysis Totally, three sets of Camembert cheese were made in this research. The pH values from set 1 were used for model development, from set 2 for verication of normalized pH model, and from set 3 for pH model verication, respectively. There were nine replicates of pH values for each set. The reliability of the model predictions was assessed by quantifying the accuracy and bias of the model predictions. A number of statistics were used to evaluate the general goodness-of-t of each model. Proportion of variation accounted for (R2) was calculated as 1 RMSE=S2 , where RMSE is the residual mean square y error and S 2 is the total variance of the y-variable. Average y accuracy of the estimates was assessed using the accuracy factor, AF of Ross (1996), as rened by Baranyi, Pin, and Ross (1999): 2v3 uP u n 6u lnC k lnM k 2 7 6tk1 7 7, AF exp6 6 7 n 4 5 where Ck is the calculated value for observation number k, Mk is the measured value for observation number k, and n is the number of observations.
SET

1 5 9 13

2
6

3 7 11 15

4
8 12

T TM BM B r

10 14

16

CM

SM

Fig. 1. Grid for measurement locations of pH values in a cross section (T, top; TM, top middle; BM, bottom middle; B, bottom; C, center; CM, center middle; SM, surface middle; S, surface).

BATCH 1

BATCH 2

BATCH 3

BLOCK 1
SECTION OB SECTION OB SECTION OB SECTION OB SECTION OA

BLOCK 4

BLOCK 5

BLOCK 8 BLOCK 8 BLOCK 8 BLOCK 8

BLOCK 1

BLOCK 4 BLOCK 4 BLOCK 4 BLOCK 4

BLOCK 5 BLOCK 5 BLOCK 5 BLOCK 5

BLOCK 8

BLOCK 2
SECTION OC

BLOCK 3

BLOCK 6

BLOCK 7
SECTION OA

BLOCK 2 BLOCK 2 BLOCK 2 BLOCK 2

BLOCK 3

BLOCK 6

BLOCK 7 BLOCK 7 BLOCK 7 BLOCK 7

SECTION OC SECTION OA SECTION OA SECTION OA SECTION OA

SECTION OB

SECTION OB

SECTION OC

SECTION OB

SECTION OA

SECTION OC

Location 16

Location 16 Location 16 Location 16 Location 16

Location 16 Location 16 Location 16 Location 16

Location 16

Fig. 2. Experimental design for one set of Camembert cheese. (Block 1: day 1, sections OA, OB, and OC; block 2: day 5, sections OA, OB, and OC; block 3: day 10, section OA, OB, and OC, y)

Location 1

Location 1 Location 1 Location 1 Location 1

Location 1 Location 1 Location 1 Location 1

Location 1

..

..

..

..

ARTICLE IN PRESS
1530 S. Liu, V.M. Puri / LWT - Food Science and Technology 41 (2008) 15281534

4. Development of pH model 4.1. Data sets for model development Experimental data of pH recorded at 16 locations were used for model development (Fig. 1). Since the pH values were similar in the center region, the pH values in locations 6, 7, 10, 11 were averaged as representative of the center region. Similarly, locations 1, 2, 3, 4 were averaged as representative of the top surface region, and locations 13, 14, 15, 16 as representative for the bottom surface (Fig. 3). In addition, the pH values in locations 5 and 9 being similar to locations 8 and 12 were averaged as representative of innerouter side surface regions (Liu & Puri, 2005). In order to generate a model covering as a wide range of pH as possible, the pH data were normalized during model development. 4.2. pH curves for model development The measured pH values were analyzed and coupled with statistical analysis to develop a mathematical model for pH distribution during ripening of Camembert cheese. To get a generalized model for pH distribution and to reduce the effect of different batch and set of pH values, the normalized variables were used to construct pH model as shown in Eqs. (1) and (2): x y Day Day1 , Day35 Day1 pH pH1 , pH35 pH1 (1) (2)

day 1 (=4.5) (average of day 1 measured values), pH35 is the pH values on day 35 (=7.1) (average of day 35 measured values), and pH is the value for any day from day 2 to day 34. Normalized pH curves for center, top surface, bottom surface, and innerouter side surface locations from set 1 are shown in Fig. 4. The normalized pH model is developed based on these data for pH distribution in Camembert cheese. 4.3. Mathematical pH model development During the entire process, there are three stages of ripening. In the rst stage (stage I, day 1), the water was drained. In the second stage (stage II), Camembert cheese was ripened in an incubator from day 2 to day 15 at temperature of 12 1C and relative humidity of 93%. In the third stage (stage III), Camembert cheese was ripened further in a refrigerator for 3 weeks (day 15day 35) at temperature of 7 1C and relative humidity of 85%. Based on measured data and statistical analysis (mean values), all pH curves (Fig. 4) could be divided into three distinct stages based on the trends and rates of normalized variables: stage I from x Day1 Day1 11 0 Day35 Day1 35 1 Day15 Day1 15 1 to x 0:41, Day35 Day1 35 1 Day15 Day1 15 1 0:41 Day35 Day1 35 1 Day30 Day1 30 1 to x 0:85, Day35 Day1 35 1

stage II from x

where, Day1 is the beginning of ripening time (1 day), Day35 is the end of ripening time (35 days), and Day is any ripening time from day 2 to day 34; pH1 is the pH values on

Z
6.67mm
1 5

6mm
2 6

14mm
3 7 4 8

Top Surface Region

32mm Inner Region

9 13

10 14

Center region

Outside Surface region


12 16

11 15

Bottom Surface Region

C (center) 6 20

CM

SM S (Surface)

34 48 54mm

Fig. 3. pH measurement locations in Camembert cheese cross-section grouped according to statistically similar regions (Liu & Puri, 2005).

ARTICLE IN PRESS
S. Liu, V.M. Puri / LWT - Food Science and Technology 41 (2008) 15281534
1.0 0.8 0.6 y 0.4 0.2 0.0 0 0.2 0.4 x Set 1-Top surface (location 1, 2, 3, 4) 1.0 0.8 0.6 y 0.4 0.2 0.0 0 0.2 0.4 0.6 0.8 1 x Set 1-Center (6, 7, 10, 11) y Average 1.0 0.8 0.6 0.4 Average 0.2 0.0 0 0.2 0.4 0.6 0.8 1 x Set 1 Bottom surface (location 13, 14, 15, 16) 0.6 0.8 1 Average y 1.0 0.8 0.6 0.4 0.2 0.0 0 0.2 0.4 x Set 1-Inner-outer Side surface (location 5, 8, 9, 12) 0.6 0.8 1 Average

1531

Fig. 4. Typical normalized average pH values during ripening of Camembert cheese (x (DayDay1)/(Day35Day1), y (pHpH1)/(pH35pH1), where Day1 is the beginning of ripening time (1 day), Day35 is the end of ripening time (35 days), Day is any ripening time from day 2 to day 34; pH1 is the pH values on day 1 ( 4.5) (average of day 1 measured values), pH35 is the pH values on day 35 ( 7.1) (average of day 35 measured values), and pH is the value for any day from day 2 to day 34).

stage III from x

Day30 Day1 30 1 0:85 Day35 Day1 35 1 Day35 Day1 35 1 to x 1. Day35 Day1 35 1

Table 1 Verication of normalized pH model: R2 values, accuracy factor and residual mean square obtained when tting the pH models to the normalized pH data for set 2 Top surface R2 values Residual mean square error Accuracy factor 0.98 0.31 1.26 Side surface 0.99 0.21 1.21 Center surface 0.98 0.26 1.18 Bottom surface 0.99 0.24 1.11

The reason for division of three stages is: for stage I and stage III, the pH curves were well represented and tted with a linear function, and stage II was tted with a sigmoid function during the ripening phase of Camembert cheese. So one regression model was developed for stage I and stage III and a sigmoid function for stage II. The form of straight line used was y ax+b, where a and b are regression coefcients. Stage II from x 0.41 to 0.85 was represented with a sigmoid function: y c 1 ekxx0 d, where c, d and k are regression coefcients and x0 0.41. Based on the normalized pH data (Table 1), the parameters calculated using Origion software (Microcal Software, Nothampton, MA) are a 0.965, b 0.0041, c 0.49, d 0.42, k 10.2. The parameters for the three batches (Fig. 2) were not signicantly different (P40.05). Therefore, the mean values for these parameters can be used in the implementation of the normalized pH model. Accordingly, the overall

normalized pH model is as follows:


& y 0:976x 0:0035 0:441 e
9:7x0:41

0pxp0:41; 0:40

0:85pxp1;

0:41oxo0:85:

(3) 4.4. Verication of normalized pH model The pH data recorded for the second set (i.e., set 2) of Camembert cheeses were used for model evaluation. The data ts obtained with the model were evaluated statistically taking into account tting behavior and goodness-oft. There is not a single, simple statistical method to

ARTICLE IN PRESS
1532 S. Liu, V.M. Puri / LWT - Food Science and Technology 41 (2008) 15281534

evaluate similarities and differences between nonlinear models, and to deal with the question of which model should be used. Usually, a number of procedures are used to obtain an overall view of model behavior and choose the one that is more consistent for most of the test performed. The main statistical procedures used for model comparison are residual analysis and tests for goodness-of-t. The goodness-of-t was evaluated using the RMSE (Table 2) that takes into account the number of parameters contained in each model. The error of prediction is given as RMSE which provides a measure to judge the goodness-oft of the polynomial model. The lower the value of RMSE, the better is the t as reported by Sutherland, Bayliss, and Robert (1994). For set 2 when using the developed Eq. (3),
Table 2 Verication of pH model: R2 values, accuracy factor and residual mean square error obtained when comparing pH values calculated using Eq. (4) with the pH data for set 3 Top surface R2 values Residual mean square error Accuracy factor 0.9 0.23 1.20 Side surface 0.99 0.22 1.23 Center surface 0.95 0.33 1.24 Bottom surface 0.99 0.27 1.17

RMSE for top surface, center region, innerouter side surface, and bottom surface regions were 0.31, 0.21, 0.24, and 0.26, respectively; while, the R2 values were 0.98, 0.99, 0.98, and 0.99, respectively. The average RMSE across the four curves was the smallest with side surface followed by center and bottom surface, whereas the highest average RMSE was observed around the curved top surface. The results showed that the model was a good t of normalized measured pH values during ripening of Camembert cheese (Fig. 5). A similar trend was observed when the accuracy factor was used (Table 2), i.e., the accuracy factor values were all less than 1.26. The larger the accuracy factor value, the less accurate is the average estimate. The accuracy factor values of 1.111.26 indicate that the calculated values were accurate. 4.5. Verication of pH model Eqs. (1) and (2) were substituted into Eq. (3) to obtain a relationship between ripening time in days and pH values given in Eq. (4).
& pH 0:075 day 4:44 1:141 e4:260:285 day 5:54 1odayp15; 30pdayo35; 15odayo30:

(4)

1.0 0.8 0.6 y y 0.4 0.2 0.0 0 0.2 0.4 0.6 0.8 1 x Set 2-Top surface 1.0 0.8 0.6 y 0.4 0.2 0.0 0 0.2 0.4 0.6 0.8 1 x Set 2-Center y

1.0 0.8 0.6 0.4 0.2 0.0 0 0.2 0.6 0.8 x Set 2-Inner-outer side surface 0.4 1

1.0 0.8 0.6 0.4 0.2 0.0 0 0.2 0.6 x Set 2-Bottom surface 0.4 0.8 1

Fig. 5. Typical normalized predicted and experimental pH values during ripening of Camembert cheese for set 2 (J: experimental data, : predicted data) (x (DayDay1)/(Day35Day1), y (pHpH1)/(pH35pH1), where Day1 is the beginning of ripening time (1 day), Day35 is the end of ripening time (35 days), Day is any ripening time from day 2 to day 34; pH1 is the pH values on day 1 ( 4.5) (average of day 1 measured values), pH35 is the pH values on day 35 ( 7.1) (average of day 35 measured values), and pH is the value for any day from day 2 to day 34).

ARTICLE IN PRESS
S. Liu, V.M. Puri / LWT - Food Science and Technology 41 (2008) 15281534 1533

7.5 7.0 6.5 pH pH 0 10 20 Day Set 3 Top surface 7.5 7.0 6.5 pH 6.0 5.5 5.0 4.5 4.0 0 10 20 Day Set 3 Center 30 40 pH 30 40 6.0 5.5 5.0 4.5 4.0

8.0 7.5 7.0 6.5 6.0 5.5 5.0 4.5 4.0 0 10 20 Day Set 3 Inner-outer side surface 7.5 7.0 6.5 6.0 5.5 5.0 4.5 4.0 0 10 20 Day Set 3 Bottom surface 30 40 30 40

Fig. 6. Comparisons of predicted and experimental pH values during ripening of Camembert cheese (&: experimental data, : predicted data).

The pH data predicted by the experimental trend-based Eq. (4) for set 3 of Camembert cheeses were compared with measured pH values. The data ts obtained with the model were evaluated statistically taking into account tting behavior and goodness-of-t. In general, the pH model was able to capture the magnitudes and trends sufciently. The R2 values for all sets were greater than 0.95. The results showed the model was a good t of measured pH values during ripening of Camembert cheese. Predicted curves and experimental curves for set 3 are shown in Fig. 6. The R2 values for top surface, center region, innerouter side surface, and bottom surface regions between measured and pH predicted values were 0.97, 0.99, 0.95, and 0.99, respectively; while, the RMSE values were 0.23, 0.22, 0.33, and 0.27, respectively. Most statistical tests revealed a general trend with some differences between different locations in Camembert cheese in their goodness-of-t. Generally, all pH values proved valid for the predicted model showing the high R2 values, low RMSE values, and good accuracy factors. 5. Conclusions In this study, based on pH values in Camembert cheese during ripening, an experimental trends-based mathematical model was developed and veried using normalized variables. The results showed that the model was a good t

for measured pH values during ripening of Camembert cheese. In general, the pH model was able to capture the magnitudes and trends sufciently. The model t the innerouter side surface and bottom surface regions better than top surface and center surface regions. The R2 values for top surface, center, innerouter side surface, and bottom surface regions mean measured and pH modelpredicted values were 0.97, 0.95, 0.99, and 0.99, respectively. The veried pH distribution model can be used to estimate the pH distribution during ripening of Camembert cheese. Acknowledgments This research was partially supported by USDA Milk Safety grant no. 2001-34163-10544 and the Pennsylvania Agricultural Experiment Station. References
Baranyi, J., Pin, C., & Ross, T. (1999). Validating and comparing predictive models. International Journal of Microbiology, 48, 159166. Baranyi, J., & Roberts, T. A. (1995). Mathematics of predictive food microbiology. International Journal of Microbiology, 26(2), 199218. Farber, J. M., & Peterkin, P. I. (1991). Listeria monocytogenes, a foodborne pathogen. Microbiology Reviews, 55, 476511. France, J., & Thornley, J. H. M. (1984). Mathematical models in agriculture. London: Butterworths.

ARTICLE IN PRESS
1534 S. Liu, V.M. Puri / LWT - Food Science and Technology 41 (2008) 15281534 Radhakrishnan, S. (2000). Survival and growth modeling of Listeria monocytogenes during ripening of Camembert cheese. MS thesis, Penn State University, University Park, PA. Ross, T. (1996). Indices for performance evaluation of predictive models in food microbiology. Journal of Applied Bacteriology, 81, 501508. Sutherland, J. P., Bayliss, A. J., & Robert, T. A. (1994). Predictive modeling of growth of Staphylococcus aureus: The effects of temperature, pH, and sodium chloride. International Journal of Food Microbiology, 21, 217236. George, S. M., Richardson, L. C., & Peck, M. W. (1996). Predictive models of effect of temperature, pH and acetic and lactic acids on the growth of Listeria monocytogenes. International Journal of Food Microbiology, 32(12), 7390. Liu, S., & Puri, V. M. (2005). Spatial pH distributions during ripening of Camembert cheese. Transactions of ASAE, 48(1), 279285. McMeekin, T. A., & Ross, T. (2002). Predictive microbiology: Providing a knowledge-based framework for change management. International Journal of Food Microbiology, 78(12), 133153.

Das könnte Ihnen auch gefallen