| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Department of Animal Science, University of Padova, Viale dell’Università 16, 35020 Legnaro, Padova, Italy
1 Corresponding author: riccardo.dalzotto{at}unipd.it
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
Key Words: milk coagulation properties • repeatability • reproducibility • mid-infrared spectroscopy
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
-casein (Mariani et al., 1976) and β-lactoglobulin genes (Kübarsepp et al., 2005), titratable acidity (Summer et al., 2002) and pH (Lindström et al., 1984), calcium content (Tervala et al., 1985; Kübarsepp et al., 2005), and breed (De Marchi et al., 2007). Because exploitable additive genetic variation seems to exist for MCP (Ikonen et al., 1997, 1999; Cassandro et al., 2008), genetic improvement of MCP could be an effective way to enhance the efficiency of cheese production (Ikonen, 2000). Currently, assessment of MCP can be performed through milk coagulation meters, which provide measures of milk rennet clotting time (RCT, min) and curd firmness (a30, mm). In addition to the fact that knowledge on reproducibility and repeatability of MCP measures is scarce, these methods have some unfavorable features: measurement is time consuming (31 min per analysis) and skilled personnel are required. For these reasons, the routine assessment of MCP for all cows involved in milk recording programs is practically unfeasible and, therefore, sire evaluation programs would need to be limited to a random sample of offspring per sire with unfavorable effects on accuracy of EBV.
Because of the high throughput (potentially hundreds of samples/hour; Foss, 2007), ease of use, and reduced cost of analysis (Soyeurt et al., 2006), mid-infrared reflectance spectroscopy (MIRS) may be a possible alternative technique for the assessment of MCP, if the accuracy of the measurements is satisfactory. Mid-infrared reflectance spectroscopy has been implemented in the measurement of milk protein content (Etzion et al., 2004) and milk composition (Lynch et al., 2006), the assessment of fatty acid contents in milk (Soyeurt et al., 2006), the prediction of sensory texture and chemical traits of processed cheese (Karoui et al., 2006; Fagan et al., 2007), and the quantification of acetone in milk for the detection of subclinical ketosis (Heuer et al., 2001). Currently, however, studies on the application of MIRS techniques for the measurement of MCP are not available.
The aims of this study were to estimate the repeatability of MCP measured by using a computerized renneting meter (CRM) to evaluate the reproducibility of MCP measures under different storage conditions of milk samples and to investigate the applicability of MIRS as a new technique for the assessment of MCP.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
Immediately after collection, each individual milk sample was partitioned into 8 subsamples of 50 mL each. Four subsamples of each cow (PS subsamples) were processed according to International Committee for Animal Recording procedures (ICAR, 2006) and combined with preservative (Bronopol, Knoll Pharmaceuticals, Nottingham, UK); the remaining 4 sub-samples (NP subsamples) were stored with no preservative. All PS subsamples of a cow were refrigerated at 4°C: 2 subsamples for 6 h (PS0: 1 subsample used for CRM analysis and 1 used for MIRS assessment), 1 subsample for 4 d (PS4, for MIRS assessment), and 1 subsample for 8 d (PS8, for MIRS assessment) following collection. Two NP subsamples were refrigerated at 4°C for 6 h (NP0): 1 subsample used for CRM analysis and 1 for MIRS assessment, whereas the remaining 2 NP were frozen and stored at –18°C for MIRS analysis carried out 4 (NP4) or 8 d (NP8) after collection, respectively. Milk samples of 4 cows did not coagulate within 31 min, were classified as not coagulating according to the approach of Ikonen et al. (1999), and were discarded from the study. Several environmental factors (e.g., parity, lactation stage, season, feeding, and management practices) can lead to noncoagulated samples as reported by Ikonen et al. (2004) and Tyrisevä et al. (2004). Studies comparing individual milk samples have shown that MCP is also affected by physical and chemical parameters such as titratable acidity (Formaggioni et al., 2001), SCC (Politis and Ng-Kwai-Hang, 1988), protein and casein contents, and calcium and phosphorus concentrations (Summer et al., 2002).
Analysis of MCP
In this study, analysis of MCP using a CRM (Polo Trade, Monselice, Italy) was considered the reference method for the assessment of RCT and a30. The CRM is based on the swing of a pendulum driven by an electromagnetic field. During the milk coagulation process, differences in the electromagnetic field are recorded. After the addition of the clotting enzyme, coagulation takes place and the swing of the pendulum becomes smaller because of the enhanced curd firmness (Figure 1
). This tool is often used as reference method to monitor MCP (Ikonen et al., 2004; Cassandro et al., 2008).
|
). Only NP0 and PS0 milk samples were analyzed through CRM and each analysis was performed in duplicate. The pH was measured once before heating milk samples, and not in samples with added preservative.
MIRS
A total of 474 milk samples (6 samples per cow: NP0, NP4, NP8, PS0, PS4, and PS8) were analyzed using MIRS. All MIRS analyses were carried out at the Milk Quality Lab of Veneto Agricoltura Institute (Thiene, Italy) using a MilkoScan FT120 (Foss, 2007), which scans the medium infrared region from 4,000 to 900 cm–1 and uses an interferometer. Spectra are generated from the resulting interferogram by means of fast Fourier transformations (Figure 2). The spectra were automatically recorded in duplicate in a database (Foss, 2007). The resulting spectral data were stored as log (1/R), where R are reflectance values, using WINISI II version 1.02 software (InfraSoft International, Port Matilda, PA).
|
 |
where yijklm is a RCT or a30 measure provided by CRM for NP0 and PS0 samples; µ is the intercept; Hi is the fixed effect of herd (i = 1, ..., 3); Dj is the fixed effect of the stage of lactation (j = 1: <100 d; j = 2: from 100 to 200 d; j = 3: >200 d); COWk:ij is the random effect of the cow (k = 1, ..., 79); Pl is the random effect of the treatment (l = 1: samples with no addition of preservative; l = 2: samples added with Bronopol preservative); P x COW is the random effect of the interaction between treatment and cow; and eijklm is a random residual. Animal (COW), treatment (P), and P x COW effects, as well as residuals, were assumed to be independently and normally distributed with a mean of zero and variance 
COW2, P2, PxCOW2, and e2, respectively; REML was used as the method of estimation of variance components.
Reproducibility and repeatability were computed according to the guidelines suggested by the International Organization for Standardization (ISO, 1994a). In this study, definition of reproducibility differed from the standard one and aimed to evaluate the agreement between measures of MCP obtained using CRM under different preservation conditions of milk samples. Evaluation of reproducibility was based on 2 parameters: 1) reproducibility (RD), defined as the value below which the absolute difference between 2 measures of MCP, obtained from the analysis of NP0 and PS0 subsamples of the same cow, respectively, and which is expected to lie within a probability of 95% (ISO, 1994a); and 2) coefficient of reproducibility (RD%), which is an indicator of the degree of agreement between 2 single measures of MCP for the same milk sample preserved with no addition or with addition of Bronopol preservative.
These parameters were computed as
and
Two statistical parameters were estimated to evaluate the repeatability of measures of MCP provided by CRM: 1) repeatability (RT), defined as the value below which the absolute difference between 2 single measures of MCP obtained by analyzing repeatedly, under the same conditions and within a short period, the same milk subsample (NP0 or PS0), and which is expected to lie with a probability of 95% (ISO, 1994a); and 2) coefficient of repeatability (RT%), which is an indicator of the degree of agreement between repeated measures of MCP obtained within a short period using CRM on the same milk subsample. Functions of estimated variance components were
and
MIRS Calibration Equations
The software used for deriving the calibration was WINISI software, version 2. Calibration equations for MIRS were obtained using, as reference data, MCP measured by CRM on NP0 subsamples and 6 sets of data including spectra of NP0, NP4, NP8, PS0, PS4, and PS8 subsamples, respectively. Analysis of these sets of data provided 6 models to predict MCP of NP0 from spectra measured on NP0, NP4, NP8, PS0, PS4, and PS8 subsamples, respectively. To optimize calibration accuracy, the data were processed to a variety of derivative transformations using common mathematical treatments and scatter correction (Martens and Naes, 1989). The math treatment used was 0,0,1,1, which refers to derivative, gap, smooth, smooth 2, respectively. A derivative treatment of 0 means no derivative was used, a gap of 0 is the gap over which a derivative would be calculated, smooth 1 is the number of data points smoothed, and smooth 2 is a second smoothing operation, which is nearly always set to 1.
All calibration regression models were estimated by modified partial least squares regression (Williams and Norris, 2001). Cross validation was performed during model development, wherein one-fourth of the population samples at a time were temporarily removed from the calibration set.
Samples with GH (Mahlanobis distance) and T (Student’s t-distribution) values greater than 2.5 and 10, respectively, were discarded, and the cycle was performed a second time, so the program removed outliers twice before completing the final calibration. The critical ‘T’ is chosen by the operator to eliminate samples from the regression model. A value of 3 is conservative, 2 is liberal, and 2.5 is recommended. The critical ‘H’ is chosen by the operator to eliminate samples with spectral distances too far from the population mean. A value of 10 is recommended. The critical ‘T’ is chosen by the operator to eliminate samples with spectral distance too far from the population mean. The recommended value is 10 (Paulsen and Singh, 2004).
To assess the efficiency of the calibration equations, several statistical parameters were estimated: standard error of cross-validation (SECV) and cross-validation coefficient of determination (RCV2) according to Haaland and Thomas (1988) and Martens and Naes (1989). To evaluate the efficiency of calibration equations for MCP prediction, the ratio of SECV to the standard deviation of reference data (RPD) and the ratio of SECV to the range of reference data were computed (RER; Williams and Norris, 2001). If RPD is >2, the calibration equation is considered good. On the other hand, if RPD ratio is <1.5, the predictions are of poor quality and the equation cannot be used in practice (Sinnaeve et al., 1994). The RER is useful for assessing the practical utility of predictive models. Models with an RER of <3 have little practical utility; RER values between 3 and 10 indicate limited to good practical utility, and >10 indicates that the model has a high utility value (Williams and Norris, 2001).
|
RESULTS AND DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
. On average, RCT of samples preserved with the addition of Bronopol was slightly lower than that of samples with no preservative, whereas the average a30 of PS samples exhibited a 10% decrease compared with that of NP samples. Variation of RCT and a30 was slightly affected by addition of preservative to milk samples: standard deviations of MCP were 6% smaller in PS than in NP samples and ranges of variation of RCT and a30 were similar for samples with added Bronopol and for milk without preservative. In general, RCT values observed in this study were similar to those reported by Summer et al. (1999) and Cassandro et al. (2008) for Italian Holsteins, but greater than those reported by Tyrisevä et al. (2004) and by Ikonen et al. (2004) for Finnish Ayrshire cows. These figures indicate a moderate cheese-processing aptitude of Holstein cattle milk when compared with MCP of milk from other dairy breeds (De Marchi et al., 2007).
|
. For RCT, the correlation between samples without and with addition of preservative was >0.95 and greater than that for a30 (r = 0.76; P < 0.01), indicating that measures of RCT were less sensitive to different preserving conditions of milk samples than were measures of a30 and that the ranking of cows based on measured RCT would have been similar under different preserving circumstances of samples.
|
Reproducibility and Repeatability of MCP Determined by CRM
Estimates of reproducibility and repeatability for RCT and a30 are presented in Table 3. In this study, definition of reproducibility aimed to evaluate the agreement between measures of MCP obtained for the same sample under different preserving conditions of milk, rather than to evaluate the consistency of measures provided by different laboratories or operators. For measures of RCT, the estimated RD [i.e., the value below which the true absolute difference between 2 single measures, obtained with the same method of analysis applied to the same sample, but with different preserving conditions of milk (with or without addition of Bronopol) and different time of analysis], which is expected to lie within a 95% probability, was 2.5 min (i.e., 16% of average RCT). For a30, the estimated RD, being 43% of the trait average, was less favorable than for RCT.
|
The degree of agreement between repeated measures of MCP was different for RCT and a30. The estimated RT was 13 and 35% of the trait average for RCT and a30, respectively. For RCT, the estimate of RT% was 0.96 and was in full agreement with the value of analytical repeatability reported by Caroli et al. (1990) for measures of RCT and a30 obtained using a formagraph. The estimated RT% for a30 was unfavorable (RT% = 0.77) and not consistent with previous literature estimates (Caroli et al., 1990). Based on the estimated parameters, assessment of a30 with CRM should be performed with replicated analyses.
Prediction of MCP using MIRS
An example of a mid-infrared spectrum is shown in Figure 2. Curley et al. (1998) assigned protein to specific bands in the mid-infrared spectra of raw milk in the 1,570 to 1,550 cm–1 range. These results were confirmed by Etzion et al. (2004), Hewavitharana and Brakel (1997), and by Luginbühl (2002). Lipids are known to contribute to the region 3,000 to 2,800 cm–1, which accounts for the observed peaks at 2,928 and 2,855 cm–1 (Lefèvre and Subirade, 2000). The peak at 1,748 cm–1 can also be attributed to lipids.
Statistics of prediction results for RCT and a30 are presented in Tables 4 and 5. Calibration equations were estimated using CRM measures of MCP for samples without preservative as reference data and spectra measured at 0, 4, and 8 d from collection of milk samples. Hence, 6 different calibration equations were developed.
|
|
and 5). Samples that have GH and T values greater than 2.5 and 10, respectively, may be classified as outliers. According to these criteria, a fraction of samples, which ranged from 5 to 14% of total data, were classified as outliers.
For reference samples, mean (SD) of RCT (Table 4) ranged from 14.54 (3.59) to 14.81 (4.04) min for calibrations based on spectra measured on milk with no preservative and from 14.64 (3.43) to 14.72 (3.44) min for calibrations based on spectra of samples combined with Bronopol (data not shown). Therefore, variability of coefficients of variation for the 6 sets of reference samples was small and ranged from 23 to 27%.
The preferred model for predicting RCT had a SECV value of 1.80 min and the corresponding RER value was 9.22. The accuracy of each model can be evaluated using the coefficients of determination (R2cv) between the predicted and measured values, as stated by Williams (2003). A value for R2cv between 0.50 and 0.65 indicates that discrimination between high and low values can be made. A value for R2cv between 0.66 and 0.81 indicates approximate quantitative predictions, whereas a value for R2cv between 0.82 and 0.90 reveals good predictions. Models having a value for R2cv >0.91 are considered to be excellent (Williams, 2003). The RCT model (R2cv = 0.73) allowed approximate quantitative predictions for RCT values, whereas for a30, the model (R2cv = 0.45) did not allow any type of discrimination.
Moreover, high values for R2cv and RPD parameters are required for potential use of predictive equations. For RPD, values >2 indicate good calibrations and values <1.5 indicate unreliable predictions (Karoui et al., 2006). In this study, models with greater values of R2cv also had greater values of RPD. This result is in agreement with results of Soyeurt et al. (2006), who observed a tight relationship between R2cv and RPD in a study aimed at estimating fatty acid content in cow’s milk using MIRS.
As a general tendency, prediction equations for RCT (Table 4) based on spectra of samples preserved with Bronopol exhibited values of validation parameters more favorable than those estimated for calibrations using spectra of milk with no preservative. Values for RPD were very close to or greater than 1.5 for all predictive equations based on spectra of milk with addition of preservative, whereas, for calibrations based on samples with no addition of preservative, only the equation developed from spectra of samples providing reference data (i.e., NP0 subsamples) had RPD >1.5. The only equation with RPD close to 2 and R2cv >70% was that based on spectra measured on PS4 subsamples. Values for RER provided indications consistent with those from inspection of RPD estimates, suggesting that calibrations based on samples preserved with Bronopol had better practical utility. Although some of these results indicate that MIRS may have a role in the prediction of MCP, it is important to note that a critical aspect of this study was the development of prediction equations on the basis of reference samples collected in only 3 herds and over a very short period. Consequently, conclusive evidence on the effectiveness of MIRS as a predictive tool for MCP requires an enlargement of the variability of milk sampling circumstances.
Models for predicting a30 (Table 5) were unsatisfactory for both unpreserved and Bronopol-preserved milk samples. The R2cv values were low, ranging from 0.31 to 0.35 and from 0.35 to 0.45, respectively; the estimated RPD were lower than 1.5 for all predictive equations; and RER values were less (from 5.51 to 7.56) than 10. As reported by Williams and Norris (2001), RER values of between 3 and 10 indicate limited to good practical utility.
It should also be noted that the loadings incorporated into the best RCT and a30 models were 8 and 4, respectively (Tables 4 and 5). This is a relatively low number of loadings, which may provide high model robustness. To investigate a molecular basis for the prediction of MCP, the model loadings were examined (data not shown). The first 2 loadings of the models accounted for greater than 70% of the variation in the spectral data; however, it was difficult to assign functional groups to individual peaks. In fact, using spectra pretreated with a first derivative step, interpretation of the loadings associated with these models was difficult. This result was because the observed peaks and valleys did not follow the raw spectral pattern. Nevertheless, the regions of the spectra that seem to be important in predicting RCT and a30 were associated with the vibration of the C–H, C–O bonds (from 1,200 to 1,000 cm–1), amide II (1,540, and 3,900 to 3,500 cm–1), and lipids (from 2,900 to 1,700 cm–1). These results can be explained by the effects of milk protein and fat content on milk coagulation. Protein level significantly affects coagulation rate, including firming times (Fagan et al., 2008). Castillo et al. (2003) stated that increasing the protein content of milk decreases the curd firming time, whereas Dalgleish (1980) found that increasing the casein concentration by ultrafiltration resulted in a firmer final curd. Bastian et al. (1991) found that, although fat content had no effect on RCT, it did affect curd firming, with greater fat levels associated with firmer gels.
|
CONCLUSIONS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
|
ACKNOWLEDGEMENTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
Received for publication October 12, 2007. Accepted for publication June 11, 2008.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONCONCLUSIONSACKNOWLEDGEMENTSREFERENCES |
|---|
-casein A and B variant in the production of the Parmigiano-Reggiano cheese. Sci. Tecn. Latt. Cas. 27:208–227.This article has been cited by other articles:
|
|
 |
|
  A. Cecchinato, M. De Marchi, L. Gallo, G. Bittante, and P. Carnier Mid-infrared spectroscopy predictions as indicator traits in breeding programs for enhanced coagulation properties of milk J Dairy Sci, October 1, 2009; 92(10): 5304 - 5313. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
