Development of a Rapid and Accurate Prediction Model for Whey Separation in Pasteurized Drinking Yogurt Caused by Long-term Ambient Storage

Abstract

Long-term ambient storage of pasteurized drinking yogurt (PDY) may result in excessive whey separation. This study aimed to develop a rapid and accurate prediction method of PDY whey separation when maintained at 25 °C for 45 and 90 days. Predictive models of whey separation were created by regression analysis based on the results of physical property measurements including the 48 hour Turbiscan Stability Index (TSI). The predictive model based on TSI showed a high coefficient of determination and high suitability not only when the pectin concentration in PDY was changed but also when the manufacturing conditions of PDY were changed. In addition, this model was superior in prediction accuracy and versatility as compared to models based on other physical property measurements. The advantage of the TSI model is that it allows for more rapid, accurate and versatile evaluation of spontaneous whey separation.

Introduction

In recent years, the demand for pasteurized yogurt has increased dramatically, particularly in the Asian market, with beverages leading the market. The unique advantages of pasteurized yogurt, such as fewer limitations on storage and distribution temperatures, longer shelf life as compared to refrigerated products, and the edibility related convenience popularize it among consumers.

There are, however, concerns regarding quality degradation associated with the long-term ambient storage of pasteurized yogurt. Specifically, the degradation of appearance, represented by whey separation, and the degradation of flavor, represented by odor deterioration, are typical forms of quality degradation. However, due to a lack of research in this topic there are few detailed reports related to quality deterioration parameters. The deterioration of appearance largely depends on excessive whey separation, occurring especially in PDY stored at room temperature. Lucey et al. (1999) predicted that the higher the storage temperature of acid milk beverages, the higher the thermal motion, and hence the higher the collision frequency between the protein particles, leading to faster coagulation. The results of their study showed that whey separation was actually higher at 25 °C than at 9 °C. Specifically in Japan, where low viscosity drinking yogurt is preferred, manufacturing PDY with low viscosity with an acceptable quality requires ingenuity of formulation and manufacturing methods to maintain PDY at long-term ambient storage without whey separation.

Yogurt is acidified by lactic acid produced by lactic acid bacteria in the fermentation process, thereby reducing the electrical repulsive force near the isoelectric point of casein which is approximately pH 4.6. As a result, hydrophobic interactions cause casein particles to aggregate (Lucey 2004) and form yogurt. Low viscosity drinking yogurt however exhibits a clear aqueous layer separated at the top as a result of the yogurt settling in the container during storage. Heat treatment of acidified yogurt also promotes the formation of casein aggregates. The physical stability of yogurt is generally increased by blending high methoxyl (HM) pectin (Glahn 1982, Glahn and Rolin 1994, Syrbe et al., 1998) to prevent the formation of such casein aggregates.

A study by Bockelmann and Bockelmann (1998) mentions certain requirements for the production of stable acid milk beverages, thereby, clarifying the production conditions of PDY. However, few studies have quantitatively evaluated the effect of manufacturing conditions such as post-fermentation pasteurization conditions on physical stability.

The most typical physical degradation of PDY observed over time is whey separation. In many cases, the graduated cylinder method has been used to observe whey separation (Towler, 1984, Amice-Quemeneur et al., 1995, Sejersen et al., 2007). In this method, drinking yogurt is placed in a graduated measuring cylinder and stored at varying storage temperatures for the shelf life duration and the amount of whey separated out is visually measured. However, this conventional method requires a great deal of time to assess PDY whey separation with a shelf life of several months or longer. Therefore, a method capable of accurately and rapidly predicting whey separation after long-term ambient storage is considered to be of great advantage for determining an appropriate shelf life and analyzing the factors affecting PDY stability.

Previously, indicators used to evaluate the stability of acidified milk beverages were measurement of viscosity and particle size (Jensen et al., 2010, Wu et al., 2014, Zhao et al., 2018, Yuliarti et al., 2019), measurement of sedimentation by centrifugation (Jensen et al., 2010, Nobuhara et al., 2014, Zhao et al., 2018), and measurement of zeta potential (Sejersen 2007, Nobuhara et al., 2014, Li et al., 2018, Zhao et al., 2018, Yuliarti et al., 2019). Recently, Turbiscan has been used to evaluate the stability of dispersions present in liquids, for example, in the food sector for monitoring the coagulation status of rennet milk (Blecker et al., 2012, Zhao et al., 2014), for evaluating the emulsification stability of emulsifiers in oil-in-water beverages (Raikos et al., 2017), for determining the precipitation rate of particles in acidified milk beverages (Sedlmeyer et al., 2004), and for monitoring the process of milk fermentation by lactic acid bacteria (Raikos et al., 2018, Ni and Raikos 2019). However, no instances of the use of Turbiscan for the rapid evaluation of stability of drinking yogurt during long-term storage have been reported.

This study aimed to create a model that allows rapid assessment of PDY whey separation after long-term ambient storage. Various quantifiable measurements such as viscosity measurement, particle size measurement, sedimentation measurement by centrifugation, and zeta potential measurement, along with dispersion stability measurement by Turbiscan, were performed to achieve this aim. Models capable of predicting whey separation by regression analysis were created based on these measurements. The accuracy and versatility of the models was assessed by applying them to evaluate PDY whey separation prepared under different manufacturing conditions, and the model with the highest accuracy and versatility was selected.

Materials and Methods

Preparation of fermented milk samples    Reconstituted skimmed milk samples were prepared by dissolving high heat skimmed milk powder (Morinaga milk industry Co., Ltd., Tokyo, Japan) in deionized water to 14.25% (w/w). Skimmed milk samples were warmed to 85 °C and homogenized in a homogenizer (Sanmaru machinery Co., Ltd., Shizuoka, Japan) at 15 MPa. This was followed by pre-heating at the specified temperature and holding time, 95 °C (360 s), 130 °C (2 s), 140 °C (2 s), and 145 °C (2 s), in the indirect heating sterilizer (plate heat exchanger and holding tube system, Morinaga engineering Co., Ltd., Tokyo, Japan) or 140.6 °C (5.5 s) in the direct heating sterilizer [infusion system (INF), Morinaga engineering Co., Ltd., Tokyo, Japan]. The pre-heating condition, 95 °C (360 s), is the general condition used for normal fermented milk samples. The temperatures (130 °C, 140 °C, and 145 °C) used for indirect heating are those generally used for UHT sterilization of milk samples. Direct heating at 140.6 °C (5.5 s) yielded the same F-value (at z = 10 °C) as indirect heating at 145 °C (2 s). After pre-heating, the milk was quickly cooled to 43 °C. The freeze dried starter culture (Express1.0, Chr. Hansen Holding A/S, Hoersholm, Denmark) was then inoculated at 0.01% (w/w) into the skimmed milk samples and stirred sufficiently, followed by fermentation until the pH reached 4.3 in an incubator at 43 °C. After fermentation, the fermented milk was agitated, cooled to 10 °C in a cooling bath, and refrigerated overnight at 10 °C.

Preparation of pectin solutions    Pectin solutions were prepared by dissolving HM pectin (Esterification degree: 70–75% (w/w), weight average molecular weight: 500 000, San-Ei Gen F.F.I., Inc., Osaka, Japan) to 0.67, 1.00, 1.33, 1.67 and 2.00% (w/w) in deionized water. The solutions were subjected to batch sterilization in a hot water bath at 90 °C or higher for 5 min. After sterilization, they were cooled immediately to 10 °C in a cooling bath.

Preparation of PDYs    PDYs were prepared by mixing the fermented milk samples and pectin solutions at a weight ratio of 7: 3 and stirring in a homomixer (PRIMIX Co., Ltd., Osaka, Japan) so that the milk solid non-fat part of PDY comprised of 9.5% (w/w) and the pectin concentration of PDY was 0.2, 0.3, 0.4, 0.5, 0.6% (w/w). The pH of the mixture was adjusted to 4.15 at 25 °C with a 10% (w/w) citric acid (Fuso chemical Co., Ltd., Osaka, Japan) solution. The mixture was then treated with one of the following methods:

1) Downstream-homogenization method The mixture was subjected to pasteurization under the following conditions using the indirect heating sterilizer (plate heat exchanger and holding tube system, Seika Co., Ltd., Tokyo, Japan). The pasteurization condition specified by law in Japan for pasteurization of yogurt is 75 °C for 900 s. The conditions [79.8 °C (300 s), 89.8 °C (30 s), and 101.5 °C (2 s)] selected were those calculated to yield the same F-value (at z = 10 °C) as the law specified condition. In addition, the lower F-value at 78 °C(30 s), and the higher F-value at 114.6 °C (2 s) were selected. After pasteurization, the mixture was cooled to 45 °C, then homogenized at 15 MPa in a homogenizer (SPX Inc., NC, USA). After homogenization, they were cooled to 25 °C.

2) Upstream-homogenization method The yogurt mixture was homogenized in a homogenizer (SPX Inc., NC, USA) at 15 MPa after the yogurt was heated to 45 °C. After homogenization, pasteurization was carried out in the same sterilizer as a downstream method at 89.8 °C for 30 s, and then cooled to 25 °C.

PDYs (45 mL) obtained by any of above methods were aseptically filled into 50 mL scaled sterile plastic bottles (Sarstedt K.K., Numbrecht, Germany) in a clean bench, and used as samples for all evaluations with the exception of the Turbiscan measurement, for which PDYs were dispensed into dedicated vials.

Whey separation measurement    At 45 and 90 days after storage at 25 °C in an incubator, the volume (mL) of the upper transparent layer in the bottle was measured visually, and then divided by the total volume (mL) to obtain the whey separation rate (% (v/v)).

Turbiscan measurement    In this study, Turbiscan (Turbiscan Tower, Formulaction Inc., Toulouse, France), a liquid dispersion stability evaluation device was used to evaluate PDY whey separation during storage at 25 °C. The device uses a near infrared light (λair = 880 nm) electron light emitting diode as a light source and two synchronized optical sensors to receive the light transmitted through the sample (180° for incident light) and the light backscattered by the sample (45° for incident light). On scanning, the optical measurement head scans the sample height and measures transmitted and backscattered light every 40 µm in a vertical direction (Mengual et al., 1999). Transmitted light measurement reveals primarily a change in the state of transparent samples, while backscattered light measurement reveals mainly a change in the state of non-transparent samples. In addition, using TowerSoft (Formulaction Inc., Toulouse, France), Turbiscan Stability Index (TSI) can be calculated from the measured transmitted and backscattered light.

TSI was used as an indicator of instability of PDY in this study because it is a summary of all the changes in the dispersions contained in the samples. TSI indicates that the higher the value, the lower the dispersion stability. TSI is calculated by:   

where t_max is the measurement point corresponding to the time t at which the TSI is calculated; z_min and z_max are lower and upper selected height limits respectively; N_h is the number of height positions in the selected zone of the scan (N_h = (z_max-z_min)/Δh, Δh is scanning height); BST is the considered signal (if T < 0.2%, adopt BS, otherwise, adopt T. BS is backscattering light intensity, T is transmission light intensity). TSI is a dimensionless index. Samples (PDYs immediately after manufacturing) for evaluation were filled at approximately 20 mL into a Turbiscan measuring vial (27 mm outside diameter) and measured continuously for 72 h, with an initial waiting time of 5 min and at 4 h intervals at 25 °C. TSI value at the upper part of the vial (the upper 20% (v/v) of the liquid level) was used as an indicator of the separation stability of PDY. Fig. 1. shows an example of the time-dependent change in TSI at 25 °C. Overall, TSI curves showed a tendency to shift to a linear change after a certain period of time while changing the slope. In this study, from the viewpoint of the stability evaluation during long-term storage, TSI_slope48 was used as a stability indicator, which is calculated by subtracting the TSI value at 44 h from the TSI value at 48 h, and dividing by the time interval. TSI_slope48 is calculated by:   

where TSI_48h is the TSI value after 48 h storage at 25 °C, and TSI_44h is the TSI value after 44 h storage at 25 °C. As a result of preliminary studies (data not shown), it was confirmed that there is variation in the time required to shift to a linear change depending on the samples, and found that if the time for calculating TSI_slope was too early, the long-term stability of the samples may not be evaluated accurately.

Fig. 1.

Time course of the Turbiscan Stability Index (TSI) of pasteurized drinking yogurts (PDYs) at different pectin concentrations.

Pectin concentration: ◊, 0.2%; □, 0.3%; △, 0.4%; ×, 0.5%; *, 0.6%.

Sedimentation measurement    Measurement of sedimentation has often been used as a method for evaluating the physical stability of yogurt. In this study, sedimentation was measured in accordance with the conventional sedimentation measurement method. Briefly, the PDY samples were stored at 25 °C for 48 h immediately after manufacturing, and were centrifuged (2 000 g × 20 min) using a centrifuge (Koki Holdings Co., Ltd., Tokyo, Japan) at 25 °C, the supernatant was removed; the remaining part was dried in a drying oven (Isuzu Seisakusho Co., Ltd., Niigata, Japan) at 60 °C for 48 h, and the mass (g) of the dried matter was weighed. Subsequently, the measured mass of the dried matter was divided by the total solid content (g) of the evaluation samples to obtain sedimentation ratio (% (w/w)). Total solid content of samples was determined using the water-solid analyzer (SMART5, CEM Inc., NC, USA).

Particle size measurement    Volume median diameter (µm) was measured with a laser diffraction/scattering particle size distribution analyzer (LA-950V2, Horiba Co., Ltd., Kyoto, Japan). Samples stored at 25 °C for 48 h immediately after manufacturing were mixed well before measurement.

Brookfield viscosity measurement    Brookfield viscosity (mPa·s) of samples was measured with a Brookfield viscometer (RB-80L, Toki Sangyo Co., Ltd., Tokyo, Japan) at 25 °C and a rotor speed of 60 rpm at 10 s intervals. The samples were stored at 25 °C for 48 h immediately after manufacturing before measurement.

Zeta potential measurement    Zeta potential (mV) of the samples was measured by electrophoretic light scattering method using a zeta potential/particle size measurement system (ELSZ-2000, Otuka Electronics Co., Ltd., Osaka, Japan). Samples were stored at 25 °C for 48 h immediately after manufacturing before measurement, then diluted 10-fold in milli-Q water. Refractive index, 1.33; viscosity, 0.89 mPa·s; dielectric constant, 78.3; and measured temperature, 25 °C were used as measurement parameters.

Statistical analysis    All experiments were repeated in triplicates, and the data are expressed as mean ± standard deviation. Tukey HSD test and Student's t-test were used to determine the significant difference. Prediction models of whey separation were determined by single regression analysis. The explanatory variables were “Brookfield viscosity”, “Particle size”, “Sedimentation”, “Zeta potential” and “TSI_slope48”. The samples made under different manufacturing conditions were analyzed to calculate the coefficient of determination (R²), mean absolute error (MAE), and root mean square error (RMSE) to enable assessment of the accuracy and versatility of the models for determination of the statistically significant models. These statistical analyses were performed using IBM SPSS Statistics ver.25 for windows (IBM Corp., NY, USA).

Results and Discussion

Effect of HM pectin concentration on physical stability of PDY    The electrical repulsion between casein micelles is known to be reduced in acidified milk (Lucey and Singh 1998a, Lucey 2004). Furthermore, the collapse of the hairy layer of κ casein reduces the steric repulsive force, causing the protein clusters to become unstable and aggregate (De Kruif 1998, 1999). Excessive aggregation can be prevented by addition of HM pectin, which is known to inhibit the occurrence of excessive aggregation by adsorbing around casein micelles and producing electrical repulsive and steric interference (Tromp et al., 2004). Meanwhile, the remaining pectin that is not adsorbed is designated as “free pectin”. In this study, in order to develop models for estimating whey separation after long-term storage at room temperature, PDYs with different HM pectin concentrations were prepared, and whey separation and various other physical property values were measured. The common manufacturing conditions were: pre-heating condition of 130 °C (2 s), pasteurization condition of 89.8 °C (30 s), and homogenization condition after fermentation (downstream homogenization at 15 MPa). Table 1 shows the values of various physical properties measured after storage at 25 °C for 48 h and the whey separation measured after storage at 25 °C for 45 and 90 days. The results are consistent with previously reported results for chilled yogurt (Lucey et al., 1999, Sejersen et al., 2007) and show that the higher the pectin concentration in PDY, the more significantly reduced the whey separation after long-term ambient storage. TSI_slope48 showed a tendency to decrease significantly with increasing pectin concentration at 25 °C after 48 h. Lesser TSI indicates more stability of the solution, which reflects the improved stability with pectin addition. The sedimentation also decreased significantly with the increase in the amount of pectin added, suggesting that sedimentation could also be an index for assessing stability. Brookfield viscosity showed a tendency to significantly increase as pectin concentration increased. This result is consistent with the results of the study done by Jensen et al. (2010), which indicates that the viscosity of the solution increases as the amount of free pectin increases. The addition of pectin, however, showed no significant difference in the particle size but the value tended to decrease, indicating that heat aggregation was inhibited. Zeta potential showed no significant changes with the addition of pectin. However, it was confirmed that the zeta potential showed a tendency to increase on the negative side depending on the amount of pectin added in the range of 0.2% (w/w) to 0.5% (w/w). This is consistent with the results of the study by Sejersen et al. (2007), which showed that negatively charged pectin is adsorbed on the surface of casein clusters that have a positive charge below the isoelectric point, and that the zeta potential shifts to the negative side.

Table 1. Physical properties of PDYs produced using varying pectin concentrations.

Pectin concentration	Brookfield viscosity	Particle size	Sedimentation	Zeta potential	TSIslope48	Whey separation
						45th day	90th day
(%)	(mPa·s)	(µm)	(%)	(mV)	(Dimensionless)
						(%)	(%)
0.2	9.68±0.90b	3.55±0.33a	24.62±1.89a	−8.65±1.28a	0.41±0.0836a	64.20±18.65a	76.62±14.30a
0.3	13.72±0.77b	3.58±0.53a	21.38±4.35a	−10.52±1.10a	0.21±0.042b	23.17±3.54b	44.27±12.94b
0.4	22.72±1.26b	3.21±0.18a	4.32±0.29b	−11.00±0.78a	0.06±0.012c	10.78±2.00bc	16.18±3.66c
0.5	41.62±5.76b	3.14±0.15a	1.35±0.03b	−11.15±1.85a	0.05±0.010c	2.04±0.07bc	3.87±0.34c
0.6	119.30±26.65a	3.01±0.07a	0.74±0.06b	−10.27±1.48a	0.01±0.002c	0.07±0.01c	0.18±0.01c

Whey separation was measured after being sorted at 25 °C for 45 and 90 days.

All data, except for whey separation, were measured after being kept at 25 °C for 48 h.

The samples were manufactured under the following conditions:

pre-heating, 130 °C 2 s; pasteurization, 89.8 °C 30 s; homogenization after fermentation, downstream 15 MPa.

The measured values are shown as mean ± SD.

Means within a row with different subscripts are significantly different (Tukey-HSD test (p < 0.05)).

Development of models to predict whey separation after long-term ambient storage    The results of single regression analysis using various physical property measurement values as explanatory variables and whey separation after each day of storage as objective variables are shown in Table 2. Linear approximation was performed after logarithmic conversion for Brookfield viscosity because it fit better, while otherwise the linear approximation was performed as it is. As a result of regression analysis, models with statistical significance were created except for the particle size (45 days) and the zeta potential (90 days). In the regression models for whey separation after 45 days, the models using TSI_slope48 and Brookfield viscosity as explanatory variables showed high R² (0.97 and 0.91). On the other hand, in the models using sedimentation and zeta potential as explanatory variables, the R² values were not so high (0.78 and 0.79), and the model using particle size as the explanatory variable showed the lowest R² (0.64). In the regression models for whey separation after 90 days, the models using TSI_slope48 and Brookfield viscosity maintained high R² (0.98 and 0.99). R² of the sedimentation and particle size models were improved when compared with R² at 45 days (0.93 and 0.82). On the other hand, R² of the zeta potential decreased to 0.67. These results indicate that whey separation with varying HM pectin concentrations could be well explained by TSI_slope48 and Brookfield viscosity. The reason why R² of sedimentation is low at 45 days may be due to the centrifugation conditions. The centrifugal conditions used in this study may have resulted in excessive external force, as compared with the natural whey separation phenomenon that occurs at rest. An improvement in R² of sedimentation at 90 days, when the change in whey separation seems to be stagnant, may be correlated with the excessively harsh centrifugation conditions. Table 3 shows the results of regression analysis using calculated TSI_slope at 16, 24, 48, and 72 h to verify whether the TSI_slope calculation time is appropriate. Similar to the result of our preliminary study (data not shown), TSI_slope16 with the calculation time of less than 24 h showed low R². Comparison of R² at TSIs_lope24 with R² at TSI_slope48, revealed that TSI_slope48 had a higher R², especially at 90 days. However, when TSI_slope48 and TSI_slope72 are compared, both show a sufficiently high R². This resulted in the TSI_slope calculation time to be set at 48 h. The regression equations obtained by using TSI_slope48 as an explanatory variable are shown below.   

  

where WS45 is whey separation of PDY after storage at 25 °C for 45 days; WS90 is whey separation of PDY after storage at 25 °C for 90 days.

Table 2. Simple regression models obtained for whey separation after being kept at 25 °C for 45 days and 90days with physical property data as explanatory variables.

Response variable	Explanatory variable	Coefficient	Intercept	R2	p-value
Whey Separation after being kept at 25°C for 45 days	Brookfield viscosity	−4.267	26728.000	0.91	0.012
	Particle size	0.821	−2.506	0.64	0.106
	Sedimentation	2.011	−0.010	0.78	0.047
	Zeta potential	0.234	2.618	0.79	0.043
	TSIslope48	1.593	−0.033	0.97	0.002
Whey Separation after being kept at 25°C for 90 days	Brookfield viscosity	−2.432	254.190	0.99	0.001
	Particle size	1.138	−3.472	0.82	0.033
	Sedimentation	2.672	0.002	0.93	0.009
	Zeta potential	0.263	2.993	0.67	0.091
	TSIslope48	1.954	−0.004	0.98	0.001

Brookfield viscosity is linearly approximated by logarithmic conversion of the data to provide a better fit, while the remaining data are linearly approximated as they are.

Table 3. Simple regression models obtained for whey separation after being kept at 25 °C for 45 days and 90days with TSIslope as explanatory variables with different measurement times.

Response variable	Explanatory variable	Coefficient	Intercept	R2	p-value
Whey Separation after being kept at 25 °C for 45 days	TSIslope16	2.720	−0.150	0.59	0.128
	TSIslope24	1.493	−0.057	0.97	0.003
	TSIslope48	1.593	−0.033	0.97	0.002
	TSIslope72	1.688	−0.023	0.97	0.002
Whey Separation after being kept at 25 °C for 90 days	TSIslope16	1.980	−0.114	0.75	0.057
	TSIslope24	1.759	−0.021	0.90	0.013
	TSIslope48	1.954	−0.004	0.98	0.001
	TSIslope72	2.076	0.008	0.99	<0.001

TSI_slope are calculated as the slopes between the TSI value at 16, 24, 48, 72 h and the TSI value 4 h before each time point.

Effect of different pre-heating conditions on physical stability of PDY    The accuracy and versatility of the created prediction models were confirmed by preparing and evaluating PDYs in which the manufacturing conditions (pre-heating condition, pasteurization condition and time of homogenization after fermentation) were changed. The pectin concentration was 0.3% (w/w) in each case. Data in the top section of Table 4 shows the measured values of the various physical properties of the samples and whey separation after storage at 25 °C for 45 and 90 days. As a result of changing the pre-heating conditions, whey separation after 45 days showed a tendency to increase significantly at 145 °C (2 s) as compared to 95 °C (360 s). However, after storage for 90 days, no significant difference was observed in any of the samples. TSI_slope48 and sedimentation showed changes related to whey separation, but Brookfield viscosity, particle size and zeta potential did not show any changes related to whey separation. The non-significant change in whey separation after 90 days suggests that pre-heating condition does not have a significant effect on PDY whey separation. This result is consistent with the results of the study by Lucey et al. (1999), who reported that pre-heating conditions did not significantly affect the stability of non-pasteurized yogurt. Usually, 90–95 °C for 5–10 min is selected as the heating condition for milk used to make yogurt. Under this condition, the mostly heat-denatured whey protein forms a complex with κ-casein. As a result, it is known that the yogurt structure after fermentation is strengthened and the water retention is also improved (IDF, 2018). However, in PDY that undergoes pasteurization and homogenization process after fermentation, the structure formed by pre-heating and fermentation is destroyed. Therefore, it can be inferred that the influence of pre-heating on structure formation is reduced. This suggests that the pre-heating process generally used for non-pasteurized yogurt production may not be required in PDY production for its physical stability.

Table 4. The physical property data of PDYs produced under different manufacturing conditions.

Condition of pre-heating	Condition of pasteurization	Condition of homogenization after fermentation	Brookfield viscosity (mPa·s)	Particle size (µm)	Sedimentation (%)	Zeta potential (mV)	TSIslope Dimensionless	Whey separation
								45th day (%)	90th day (%)
95°C 360s	89.8°C 30s	Downstream 15MPa	12.52±0.47bc	2.65±0.26b	7.80±0.96c	−7.43±0.04b	0.085±0.017b	16.15±2.63b	26.13±6.75a
(130°C 2s	89.8°C 30s	Downstream 15MPa	13.72±0.77ab	3.58±0.53a	21.38±4.35a	−10.52±1.10a	0.210±0.042a	23.17±3.54ab	44.27±12.94a)
140°C 2s	89.8°C 30s	Downstream 15MPa	14.06±0.44a	2.85±0.34ab	14.20±3.16abc	−10.76±0.94a	0.143±0.029ab	23.46±3.35ab	34.31±9.04a
145°C 2s	89.8°C 30s	Downstream 15MPa	13.08±0.28ab	3.09±0.29ab	17.90±5.25ab	−9.17±0.69a	0.228±0.046a	27.35±4.69a	42.67±9.35a
140.6°C 5.5s (INF)	89.8°C 30s	Downstream 15MPa	11.30±0.56c	2.60±0.16b	9.77±1.59bc	−10.42±1.23ab	0.100±0.020b	19.50±5.45ab	35.16±7.64a
130°C 2s	75°C 900s	Downstream 15MPa	16.06±0.37a	2.34±0.05c	3.65±0.53b	−11.02±1.65a	0.088±0.018c	10.18±2.36b	18.09±4.17c
130°C 2s	78.2°C 30s	Downstream 15MPa	15.72±1.23ab	2.31±0.26c	3.27±0.38b	−11.45±2.61a	0.070±0.014c	9.64±1.02b	16.70±1.64c
130°C 2s	79.8°C 300s	Downstream 15MPa	13.46±0.71c	2.59±0.23bc	5.94±0.99b	−10.59±1.58a	0.088±0.020c	11.43±1.63b	23.02±5.92bc
(130°C 2s	89.8°C 30s	Downstream 15MPa	13.72±0.77bc	3.58±0.53b	21.38±4.35a	−10.52±1.10a	0.210±0.042b	23.17±3.54b	44.27±12.94b)
130°C 2s	101.5°C 2s	Downstream 15MPa	12.72±0.77c	5.21±0.53a	33.77±7.22a	−10.23±1.31a	0.318±0.065ab	54.54±8.32a	75.77±10.75a
130°C 2s	114.6°C 2s	Downstream 15MPa	11.96±0.52c	5.37±0.20a	32.26±7.34a	−9.41±0.97a	0.328±0.067a	61.43±7.71a	81.49±9.26a
(130°C 2s	89.8°C 30s	Downstream 15MPa	13.72±0.77	3.58±0.53	21.38±4.35	−10.52±1.10	0.210±0.042	23.17±3.54	44.27±12.94)
130°C 2s	89.8°C 30s	Upstream 15MPa	20.10±1.33**	8.94±2.74	34.66±2.84*	−9.04±0.66	0.110±0.022*	10.53±0.59**	17.43±2.69*

The altered manufacturing conditions are pre-heating condition, pasteurization condition, and the timing of homogenization after fermentation. All samples are made with 0.3% pectin.

The measured values are shown as mean ± SD.

Significant differences are tested in groups with changing pre-heating condition, varying pasteurization condition and changing homogenization condition, respectively. For each group, the tests are performed by adding the data at 0.3% pectin in Table 1, as shown in brackets.

Means within a row with different subscripts are significantly different by Tukey-HSD test (p < 0.05).

Means with asterisk are significantly different by Student's t-test (*: P < 0.05, ** : p < 0.01).

Effect of different pasteurization conditions on physical stability of PDY    As a result of changing pasteurization conditions, whey separation after 45 and 90 days showed a tendency to increase significantly as pasteurization temperature was increased (presented in the middle section of Table 4). The relation between pasteurization temperature and holding time is established by the fact that when converted into F-value, 75 °C (900 s), 79.8 °C (300 s), 89.8 °C (30 s), and 101.5 °C (2 s) all have the same F-values. Comparison between the sample groups with the same F-value revealed that the higher the pasteurization temperature, the greater the increase in whey separation, while the holding time during pasteurization did not affect the whey separation. Pasteurization under low pH condition promotes excessive aggregation, resulting in an increase in particle size of protein clusters and significant whey separation. Hydrophobic interactions are strongly implicated in casein aggregation (McMahon and Brown 1984) and have been reported to be promoted by increasing temperature (Oakenfull and Fenwick 1977). Based on these results, it is considered that the hydrophobic interactions affected the result of this study, wherein the PDY stability decreased as the pasteurization temperature increased. In addition, comparison within the sample group with pasteurization temperature of 80 °C or less, revealed no significant differences in whey separation, regardless of the F-value; the samples at 75 °C (900 s) and 79.8 °C (300 s) had the same F-value while the sample at 78.2 °C (30 s) had an F-value one order of magnitude lower. This, therefore, proves that the effect of the holding time on whey separation is insignificant specifically in the pasteurization temperature range of 80 °C or less. TSI_slope48, sedimentation, and particle size showed a significant increase in relation to whey separation. On the other hand, the Brookfield viscosity showed a tendency to decrease significantly as pasteurization temperature increased. Usually, the larger the particle size of the protein clusters, the smaller the amount of pectin adsorbed around the casein molecules. Therefore, when the pectin concentration is sufficient, the viscosity of the solution increases due to an increase in free pectin not adsorbed on casein. However, in this study, the contradictory result was obtained. One possible reason for this result is that at the pectin concentration of 0.3% (w/w), even if the protein cluster size is about 5 µm, it may not be possible to cover the entire protein cluster. In that case, since free pectin is insufficient, no increase in viscosity derived from free pectin is observed. On the other hand, a protein cluster having a particle diameter of approximately 2 µm has a larger surface area of the entire clusters than that of approximately 5 µm, so that the pectin necessary to cover the entire clusters may be insufficient. As a result, it is possible that the viscosity increased due to the frictional force between the particles with partially reduced electrical repulsion. The results obtained in the present study are in compliance with Marozienea and Kruif (2000) who reported that at low levels of pectin concentration, a bridging aggregation occurs, in which pectin bridges protein clusters, and increases the viscosity of the solutions. No significant change in zeta potential was observed, but the zeta potential showed a tendency to decrease as the pasteurization temperature increased. This is consistent with the results of the study by Sejersen et al. (2007), which suggested that heat treatment may reduce zeta potential due to detachment of pectin from casein surface or rearrangement of pectin molecules. Normally, pasteurized yogurt is subjected to pasteurization in order to inactivate the starter culture derived lactic acid bacteria, and mold and yeast which are typically harmful microorganisms. Puhan (1979) states that heating conditions at 70 °C for 30–60 s is usually sufficient as a pasteurization condition, but that it is necessary to pasteurize at a temperature 10–15 °C higher when considering heat-resistant thermophilic lactic acid bacteria. Bockelmann and Bockelmann (1998) state that the pasteurization conditions used in commercially manufactured sterilized yogurt are often around 85–95 °C (15–30 s). In practice, the pasteurization temperature is determined according to the microbial level of the raw materials and the hygiene level of the manufacturing environment, and as evidenced by the results above it is necessary to take into consideration that the pasteurization temperature has a large effect on the physical stability, unlike the pre-heating process.

Effect of different homogenization conditions on physical stability of PDY    When the post-fermentation homogenization process was performed before pasteurization (upstream-homogenization), the whey separation after 45 and 90 days was significantly reduced as compared with the downstream-homogenization process (Data in the bottom section of Table 4). TSI_slope48 also showed a significant decrease, but sedimentation and Brookfield viscosity increased significantly. No significant changes were observed in the particle size and zeta potential, but the particle size showed a tendency to increase (p = 0.07). Interestingly, the present study showed that the whey separation was reduced despite the tendency of the particle size to increase. According to Stokes' law, it is expected that the larger the particle size, the faster the sedimentation velocity and the greater the amount of whey separation, however, this result did not follow this rule. In the case of downstream-homogenization, it can be presumed that the protein aggregates formed by fermentation and pasteurization are homogenized after pasteurization, and hence the protein cluster size decreases. In the case of upstream-homogenization, pasteurization process is performed after homogenization, so protein aggregation may have progressed slightly and the particle size was increased to approximately 9 µm. Although it is necessary to consider that the actual protein cluster has an irregular shape rather than a perfect spherical shape, Tromp et al. (2004) and Jensen et al. (2010) reported that a correlation exists between the particle size of protein clusters and the amount of pectin required to cover them. Furthermore, the adsorbed pectin increases as the protein particle size becomes smaller. The increasing particle size resulted in the pectin being able to cover the entire cluster, and a weak network being formed including an excess of free pectin. As a result, the Brookfield viscosity may have increased. However, this weak network is expected to be easily broken by severe external forces such as centrifugation, and therefore the measurement of sedimentation by centrifugation may not be suitable for the evaluation of whey separation during static storage. Lucey et al. (1998b) evaluated the whey separation on gelation of skimmed milk with glucono-δ-lactone using volumetric flasks, petri dishes, and low-speed centrifugation and concluded that low-speed centrifugation is not a useful method for quantifying whey separation in set gels. Since the study intended to evaluate spontaneous whey separation, centrifugation was performed at a low speed (100 g × 10 min), but did not give conclusive results. According to these inconclusive results, control of centrifugation conditions may not yield positive results in studies evaluating drinking yogurt. While evaluating the effect of changing the order of homogenization after fermentation of PDY, van Hooydonk et al. (1982) found that the stabilizing effect was not changed when downstream-homogenization alone or upstream and downstream are used in combination. McKenna (1987) also reported that downstream-homogenization was effective when the pasteurization temperature is 85 °C or higher at which casein aggregation occurs. The higher pectin concentrations used in these studies is responsible for the contradiction between the results of these studies and those of the present study. It is speculated that at lower pectin concentrations, the influence of the process may be more serious. However, due to other differences, such as the formula including the types of pectin, homogenization pressure, and the pasteurization temperature, further investigation of the responsible factors is warranted.

Verification of accuracy and versatility of prediction models    In order to verify their accuracy and versatility, the statistically significant prediction models developed in the present study were then applied to sample groups manufactured by changing the conditions. Fig. 2. shows the predicted and measured values of whey separation, R², MAE, and RMSE of applied models. As shown in Fig. 2.(a), the prediction model using TSI_slope48 showed the best accuracy with R²: 0.93, MAE: 4.9%, RMSE: 6.0% at 45 days, and with R²: 0.92, MAE: 7.6%, RMSE: 9.4% at 90 days. This showed that the model using TSI_slope48 was highly versatile and could predict whey separation with good accuracy even when pre-heating and pasteurization conditions and the order of homogenization after fermentation were changed. As shown in Fig. 2.(d), the prediction model based on sedimentation showed low accuracy with R²: 0.47, MAE: 9.5%, RMSE: 19.1% at 45 days, and with R²: 0.44, MAE: 14.1%, RMSE: 25.0% at 90 days. Sedimentation model significantly reduced the accuracy by failing to predict the whey separation when the order of homogenization after fermentation was changed. One of the features of the stability evaluation by Turbiscan, unlike the sedimentation measurement by centrifugation, is that the stability can be quantitatively evaluated under natural conditions that do not include external forces other than gravity. In addition, it is of great value that it is possible to predict the whey separation after 45 and 90 days at the 48-h stage, which is difficult to visually evaluate. The model based on Brookfield viscosity showed a high R² under different pectin concentration trial, but it was difficult to predict whey separation when pre-heating and pasteurization conditions were changed (Fig. 2.(c)). In addition, prediction models based on particle size (Fig. 2.(b)) and zeta potential (Fig. 2.(e)) were all difficult to predict with high accuracy. This suggests that although there is a certain correlation between PDY whey separation and Brookfield viscosity, particle size, and zeta potential, they do not independently control the stability of yogurt. In fact, in addition to the pre-heating and pasteurization conditions and the order of homogenization after fermentation, many other factors such as homogenization pressure, milk and sugar concentration, types of pectin, that can affect the physical stability of yogurt are complicatedly related. This therefore warrants the need for high versatility of prediction models which are also capable of rapid evaluation of stability, thereby, enabling the easy optimization of manufacturing conditions and assessment of the effect of the formulation on stability.

Fig. 2.

Predictive performance of each statistically significant regression model by comparing the predicted whey separation with (a) TSIslope48, (b) Particle size, (c) Brookfield viscosity, (d) Sedimentation, (e) Zeta potential and the observed whey separation when predictive models are applied to the samples produced under different manufacturing conditions. The circular plots indicate whey separation after being kept at 25 °C for 45 days. The triangular plots indicate whey separation after being kept at 25 °C for 90 days. Black symbols represent samples with changed pre-heating conditions. Green symbols represent samples with changed pasteurization conditions. Red symbols represent samples with upstream homogenization. The coefficient of determination (R²), mean absolute error (MAE), and the root mean square error (RMSE) are shown. The dashed line indicates 1:1 relation.

Conclusion

This study aimed to develop a prediction model for rapid and accurate prediction of whey separation in PDY caused by long-term ambient storage. Predictive models based on various quantifiable measurements were created. The results have demonstrated that whey separation of PDY after 45 and 90 days storage at 25 °C can be predicted accurately in 48 h by applying the predictive model created using TSI measurement by Turbiscan. This model is highly versatile than other models based on physical measurements, such as viscosity, particle size, sedimentation, and zeta potential, that have been used to evaluate the stability of drinking yogurt and can be accurately applied not only to the stability evaluation of pectin concentration, which is known to strongly affect the physical stability of PDY, but also to the stability evaluation when manufacturing conditions, such as heating conditions and the time of homogenization after fermentation are changed. TSI measurement using Turbiscan enabled quick and accurate evaluation of whey separation in the natural state of PDY, which would take a very long time with conventional methods such as the graduated cylinder method. A model with high versatility and high accuracy can be a powerful tool not only in selecting the optimal manufacturing conditions and the optimal formulation to enhance PDY stability, but also in elucidating the mechanism of action of destabilizing factors. The factors influencing the stability of PDY include the production conditions and the formulation, which have not been verified in this study, such as the homogenous pressure after fermentation, homogenous temperature, milk protein concentration, and sugar concentration. In practice, these factors are complexly interconnected and affect PDY stability. The model needs to be tested for changes in factors and parameters not investigated in this study to further verify the versatility of the model.

References

•  Amice-Quemeneur,  N.,  Haluk,  J.-P.,  Hardy,  J., and  Kravtchenko,  T.P. (1995). Influence of the Acidification Process on the Colloidal Stability of Acidic Milk Drinks Prepared from Reconstituted Nonfat Dry Milk. J. Dariy Sci., 78, 2683-2690.
•  Blecker,  C.,  Habib-Jiwan,  J.-M., and  Karoui,  R. (2012). Effect of heat treatment of rennet skim milk induced coagulation on the rheological properties and molecular structure determined by synchronous fluorescence spectroscopy and turbiscan. Food Chem., 135, 1809-1817.
•  Bockelmann,  B. and  Bockelmann,  I. (1998). “Long-life products: Heat-treated, aseptically packed: A guide to Quality”, SE: Tetra Pak. pp. 193-196.
•  De Kruif,  C.G. (1998). Supra-aggregates of Casein Micelles as a Prelude to Coagulation. J. Dairy Sci., 81, 3019-3028.
•  De Kruif,  C.G. (1999). Casein micelle interactions. Int. Dairy J., 9, 183-188.
•  Glahn,  P.-E. (1982). Hydrocolloid stabilization of protein suspensions at low pH. Prog. Food Nutr. Sci., 6, 171-177.
•  Glahn,  P.-E. and  Rolin,  C. (1994). Casein-Pectin Interaction in Sour Milk Beverages. Food Ingredients Europe Conference Proceedings, 252-258.
• IDF (2018). Heat Treatment of Milk - Overview. IDF Factsheet 001/2018-02. Brussels, Belgium: International Dairy Federation.
•  Jensen,  S.,  Rolin,  C., and  Ipsen,  R. (2010). Stabilisation of acidified skimmed milk with HM pectin. Food Hydrocolloids, 24, 291-299.
•  Li,  Y.,  Wang,  X.,  Meng,  Y.,  Zhang,  F.,  Shao,  Z., and  Hu,  L. (2018). Effect of the modified high methoxly pectin on the stability of the fermented milk beverage. Int. J. Food Prop., 21, 2075-2086.
•  Lucey,  J.A. and  Singh,  H. (1998a). Formation and physical properties of acid milk gels: a review. Food Res. Int., 30, 529-542.
•  Lucey,  J.A.,  Munro,  P.A., and  Singh,  H. (1998b). Whey separation in acid skim milk gels made with glucono-δ-lactone: Effects of heat treatment and gelation temperature. J. Texture Stud., 29, 413-426.
•  Lucey,  J.A.,  Tamehana,  M.,  Singh,  H., and  Munro,  P.A. (1999). Stability of model acid milk beverage: Effect of pectin concentration, storage temperature and milk heat treatment. J. Texture Stud., 30, 305-318.
•  Lucey,  J.A. (2004). Cultured dairy products: An overview of their gelation and texture properties. Int. J. Dairy Technol., 57, 77-84.
•  Marozienea,  A. and  de Kruif,  C.G. (2000). Interaction of pectin and casein micelles. Food Hydrocolloids, 14, 391-394.
•  McKenna,  A.B. (1987). Effects of Homogenization Pressure and Stabilizer Concentration on the Physical Stability of Longlife Laban. N.Z.J. Dairy Sci. Technol., 22, 167-174.
•  McMahon,  D.J. and  Brown,  R.J. (1984). Composition, Structure, and Integrity of Casein Micelles: A Review. J. Dairy Sci., 67, 499-512
•  Mengual,  O.,  Meunier,  G.,  Cayre,  I.,  Puech,  K., and  Snabre,  P. (1999). TURBISCAN MA2000: multiple light scattering measurement for concentrated emulsion and suspension instability analysis. Talanta, 50, 445-456.
•  Ni,  H. and  Raikos,  V. (2019). Lactic-acid bacteria fermentation-induced effects on microstructure and interfacial properties of oil-in-water emulsions stabilized by goat-milk proteins. LWT—Food Sci. Technol., 109, 70-76.
•  Nobuhara,  T.,  Matsumiya,  K.,  Nambu,  Y.,  Nakamura,  A.,  Fujii,  N., and  Matsumura,  Y. (2014). Stabilization of milk protein dispersion by soybean soluble polysaccharide under acidic pH conditions. Food Hydrocolloids, 34, 39-45.
•  Oakenfull,  D. and  Fenwick,  D.E. (1977). Thermodynamics and Mechanism of Hydrophobic Interaction. Aust. J. Chem., 30, 741-752.
•  Puhan,  Z. (1979). Heat Treatment of Cultured Dairy Products. J. Food Prot., 42, 890-894.
•  Raikos,  V.,  Duthie,  G., and  Ranawana,  V. (2017). Comparing the efficiency of different food-grade emulsifiers to form and stabilise orange oil-in-water beverage emulsions: influence of emulsifier concentration and storage time. Int. J. Food Sci. Technol., 52, 348-358.
•  Raikos,  V.,  Grant,  S.B.,  Hayes,  H., and  Ranawana,  V. (2018). Use of β-glucan from spent brewer's yeast as a thickner in skimmed yogurt: Physicochemical, textural, and structural properties related to sensory perception. J. Dairy Sci., 101, 5821-5831.
•  Sedlmeyer,  F.,  Brack,  M.,  Rademacher,  B., and  Kulozik,  U. (2004). Effect of protein composition and homogenisation on the stability of acidified milk drinks. Int. Dairy J., 14, 331-336.
•  Sejersen,  M.T.,  Salomonsen,  T.,  Ipsen,  R.,  Clark,  R.,  Rolin,  C., and  Engelsen,  S.B. (2007). Zeta potential of pectin-stabilised casein aggregates in acidified milk drinks. Int. Dairy J., 17, 302-307
•  Syrbe,  A.,  Bauer,  W.J., and  Klostermeyer,  H.. (1998). Polymer Science Concepts in Dairy Systems-An Overview of Milk Protein and Food Hydrocolloid Interaction. Int. Dairy J., 8, 179-193.
•  Towler,  C. (1984). Sedimentation in a Cultured Milk Beverage. N. Z. J. Dairy Sci. Technol., 19, 205-211.
•  Tromp,  R.H.,  de Kruif,  C.G.,  van Eijk,  M., and  Rolin,  C. (2004). On the mechanism of stabilisation of acidified milk drinks by pectin. Food Hydrocolloids, 18, 565-572.
•  van Hooydonk,  A.C.M.,  Smalbrink,  L., and  Hagedoorn,  H.G. (1982). De bereiding van drinkyoghurt. 1 De invloed van procesfactoren. Voedingsmiddelentechnologie, 15, 23-25.
•  Wu,  J.,  Du,  B.,  Li,  J., and  Zhang,  H. (2014). Influence of homogenisation and the degradation of stabilizer on the stability of acidified milk drinks stabilized by carboxymethylcellulose. LWT—Food Sci. Technol., 56, 370-376.
•  Yuliarti,  O.,  Mei,  K.H.,  Kam Xue Ting,  Z., and  Yi,  K. Y. (2019). Influence of combination carboxymethylcellulose and pectin on the stability of acidified milk drinks. Food Hydrocolloids, 89, 216-223.
•  Zhao,  L.,  Zhang,  S.,  Uluko,  H.,  Liu,  L.,  Lu,  J.,  Xue,  H.,  Kong,  F., and  Lv,  J. (2014). Effect of ultrasound pretreatment on rennet-induced coagulation properties of goat's milk. Food Chem., 165, 167-174.
•  Zhao,  R.-X.,  Qi,  J.-R.,  Liu,  Q.-R.,  Zeng,  W.-Q., and  Yang,  X.-Q. (2018). Fractionation and characterization of soluble soybean polysaccharide esterified of octenyl succinic anhydride and its effect as a stabilizer in acidified milk drinks. Food Hydrocolloids, 85, 215-221.