The current study was undertaken to evaluate various quality attributes of buffalo meat and to characterize it using principal component analysis (PCA). Ten different muscles each from six female Murrah buffaloes were analyzed for 22 variables including physico-chemical, structural and sensory attributes. The coefficients of variation of different parameters were found to range from 2 to 55 per cent. PCA transformed the variables into seven principal components (PCs) which explained more than 76 per cent of the total variability. PC1 comprised of sensory attributes (excluding appearance), myofibril fragmentation index, shear force, collagen content and collagen solubility. PC2 was characterized by a* and chroma. Loading plots of the first two PCs revealed high correlation between most of the eating quality attributes. Shear force, myofibril fragmentation index and collagen content formed another group of highly correlated variables. The study has revealed that PCA can be effectively used for interpretation of large amount of data generated in studies like quality profiling of buffalo muscles.
India harbors the largest proportion (56.7 per cent) of world’s buffalo population (FAOSTAT, 2014). Buffaloes are potentially suited for red meat production in India due to the absence of any religious or social discrimination. Buffalo meat also has desirable nutritional attributes, like leanness and content of conjugated linoleic acid, vitamins and minerals (Rao and Kowale, 1991). Buffalo meat has been a valuable livestock product for both domestic trade and export. Meat quality attributes can be characterized by various physical, chemical, sensory, nutritional and microbiological parameters. Thus, comprehensive evaluation of buffalo meat quality characteristics can generate a large set of complex data which may be very complicated to interpret. Though all characteristics could be relevant to the buffalo meat quality, identifying a smaller set of variables that can adequately explain the total variation in the data would be a very useful approach. Classical methods of statistical processing of large set of data provide an important methodology to study every single variable. However, these methods may not provide holistic information on the relationships between the variables selected and also does not allow grouping of samples with homogenous characteristics.
Principal component analysis (PCA) is one of the most basic methods of data compression developed to analyze large data matrices (Naes et al., 1996). PCA explains the variance-covariance structure of a large set of data through few linear combinations of the variables. The general objectives of PCA are data reduction and interpretation (Johnson and Wichern, 2007). PCA linearly transforms the original set of variables into a substantially smaller set of uncorrelated variables that represents the whole information in the original set of variables. The linear composites (principal components, PCs) are ordered with respect to their variations, so that the first few PCs account for most of the variation present in the original variables (Duntemann, 1989). In order to interpret the new composite variables, we need to study the directions of different components. These directions represent the relationship between the principal components and the original variables. The plots of such directions are two or three dimensional scale plots, called PC loading plots. In the loading plots, variables close together are positively correlated, while variable lying opposite to each other tend to have negative correlation. The more a variable is away from the axis origin; the better it is appreciated in the considered plane (Naes et al., 1996).
The present study was undertaken to apply PCA to analyze various physico-chemical, structural and sensory attributes of meat from mature female Murrah buffaloes. As per our understanding, this is the first report of application of PCA for buffalo meat characterization.
Materials and Methods
Six female Murrah buffaloes from Kerala Veterinary and Animal Sciences University Buffalo Farm, Mannuthy, Kerala were utilized in this study. All animals were in the age group of four to six years. They were reared intensively under similar management practices with occasional periods of grazing. The animals were slaughtered at the Meat Technology Unit, Kerala Veterinary and Animal Sciences University, Mannuthy after 12-24 h fasting as per scientific slaughter procedures. Ante-mortem and post-mortem inspections were conducted for each animal. The carcasses were electrically stimulated (100-110 V, 1.5 to 2 min) and the following ten muscles were immediately harvested from each carcass by hot deboning, viz. serratus ventralis cervicis, supraspinatus, infraspinatus, triceps brachii, longissimus thoracis et lumborum, psoas major,vastus lateralis, rectus femoris, semimembranosus and biceps femoris. Separable fat and the connective tissue were removed. Each muscle was packed in high density poly ethylene pouches and aged for 72 h at 2-4ºC (Samsung Digital Inverter Technology, India). After ageing, each muscle was portioned parallel to the direction of muscle fibers for sensory evaluation, determination of Warner-Bratzler shear force and for analysis of physico-chemical and structural attributes. The muscle portions were again packed in HDPE pouches and transferred to deep freezer and maintained at -18ºC until further analysis which took place within one week of freezer storage. The samples were thawed at 4±1ºC for 12 h before assessment of the parameters. The samples were analyzed for the following variables.
The sensory evaluation of each buffalo muscle was conducted by a semi-trained panel (n=10) consisting of faculty and post-graduate students from the Department of Livestock Products Technology, College of Veterinary and Animal Sciences, Mannuthy. They were briefly told about the nature of the experiment without disclosing the identity of samples. Meat samples used were cut into approximately equal sizes (1.5 x 1.5 x 1.9 cm) and were cooked by indirect pressure cooking in small stainless steel boxes. Cooking was done under high flame till the first whistle and then kept to cook under low flame for 30 min. All panelists received two cubes each of cooked beef muscles coded with three digit random numbers along with a score card. The panelists were asked to rate the samples for appearance, tenderness, juiciness, flavor, amount of connective tissue and overall acceptability on an eight point hedonic scale with 8 and 1 respectively representing maximum and minimum sensory scores (AMSA, 1983). Two fore-noon sessions were scheduled with a gap of 30-45 min between the sessions. Panelists were provided with filtered water to cleanse their palate between samples during sensory evaluation.
Data recorded were analyzed statistically by using Principal Component Analysis (PCA) for identifying the underlying structure of the variables. PCA with varimax rotation was used for identifying unrelated components in the PCA. Data analysis was done by the dimension reduction procedure of SPSS Software (Version 21.0) (Snedecor and Cochran, 1994).
Results and Discussion
Mean, standard error and coefficient of variation of different variables of buffalo meat are shown in Table1.
Table 1: Mean, standard error and coefficients of variation (C.V) of variables
|Mean||Std. Error||Coefficient of Variation|
|Warner-Bratzler shear force (N)||35.47||1.23||27.00|
|Myofibril fragmentation index||792.54||9.51||09.00|
|Water holding capacity||0.34||0.00||10.00|
|Cooking loss (per cent)||26.37||0.82||24.00|
|Sarcomere length (µm)||15.42||0.18||09.00|
|Fiber diameter (µm)||36.84||1.14||24.00|
|Fat content (per cent fresh weight)||1.98||0.13||50.00|
|Collagen content (per cent fresh weight)||0.54||0.02||32.00|
|Collagen solubility (per cent of collagen)||3.26||0.13||31.00|
|Drip loss (per cent)||3.28||0.23||55.00|
|Amount of connective tissue||4.76||0.10||16.00|
Table 2: Correlation coefficients of physico-chemical and sensory attributes of buffalo meat
WBSF-Warner-Bratzler shear force, MFI-Myofibril fragmentation index, WHC-Water holding capacity, CL-Cooking loss, SL-Sarcomere length, FB-Fiber diameter, FC-Fat content, CC-Collagen content, CS-Collagen solubility, DL-Drip loss, AP-Appearance, TD-Tenderness, JU-Juiciness, FL-Flavour, ACT-Amount of connective tissue, OA-Overall acceptability, H-Hue and CH-Chroma, L*-lightness a*- redness and b*-yellowness:**p<0.01and*p<0.05
The coefficient of variation of some variables like pH, L*, MFI, WHC, SL, appearance, flavour, juiciness and overall acceptability was less than 10 per cent, while CC, CS, DL and FC showed more than 30 per cent coefficient of variation. Highest coefficient of variation was recorded for DL (55 per cent) followed by fat content (50 per cent). Similar observations were reported by the Destefanis et al. (2000) and Kopuzlu et al. (2011) for beef. Correlation coefficients between the variables are shown in the Table 2. WBSF and MFI were significantly (p<0.01) and negatively correlated to CS and all the sensory attributes of buffalo meat, whereas they were positively correlated with CC.
Similar observations for buffalo meat have been reported by Rajagopal and Oommen (2015). Silva et al. (1999), Whipple et al. (1990); Rhee et al. (2004); Destefanis et al. (2000) reported similar results with beef. The a* and b* values were significantly (p<0.01) correlated with hue and chroma. Karamucki et al. (2006) reported significant correlation for a* and b* with hue and chroma of pig longissimus muscle. Fat content was significantly (p<0.01) correlated to flavour, juiciness and hue of buffalo meat. Juiciness of muscle has an important fat component (Savell and Cross, 1988). Sasaki et al. (2006) reported that tender bovine muscles were also perceived as juicy.
Results of PCA of the 22 variables which gave seven PCs are represented in Table 3. Out of the variables, the seven PCs were extracted using the Kaiser criterion (Johnson and Wichern, 2007) to determine the number of components, retaining only those components which had Eigen values greater than unity.
Table 3: Results from the PCA for the first seven principal components
|Components||Eigen values||Per cent variance||Percent cumulative Variance|
Scree plot can also be used to illustrate the various PCs and to decide the actual number of components to be included in the analysis: components having Eigen values up to the point “bend of elbow” are usually considered (Fig. 1). The seven PCs could explain a cumulative variance of 76.03 per cent of the total variability of the 22 variables considered in the study. Similar observations were reported by the Destefanis et al. (2000) and Kopuzlu et al. (2011) for beef.
Fig. 1: Scree plot
The first PC accounted for 24.40 per cent of the variation. In a study to characterize beef from Holstein Friesian young bulls using PCA, Kopuzlu et al. (2011) observed the first PC to be explaining 28.66 per cent of variation. Destefanis et al. (2000) observed that the first PC accounted for 33.90 per cent of variation in beef. The first PC was represented by significantly high component loading of WBSF, MFI, CC, CS and sensory attributes of buffalo meat except appearance (Table 4). The first PC seemed to explain the maximum of eating quality attributes in buffalo meat. The second PC explained 12.46 per cent of total variance with high component loading of a* and chroma. Third PC explained 10.23 per cent of total variance and showed high component loading for SL, FB and sensory appearance. The fourth PC accounted for 8.84 per cent for total variability, with higher loading for b*, hue and FC of buffalo meat (Table 4).
The loading plot for the first two PCs is shown in Fig. 2. Thus, all the eating quality attributes are closely clustered and placed away from the origin on the right hand side of the plot. The WBSF, MFI, CC and CS are closely placed and situated away from the origin on the left hand side of the plot.
Table 4: Principal component loadings for the first seven PCs
|Attributes||PC 1||PC 2||PC 3||PC 4||PC 5||PC 6||PC 7|
WBSF-Warner-Bratzler shear force, MFI-Myofibril fragmentation index, WHC-Water holding capacity, CL-Cooking loss, SL-Sarcomere length, FB-Fiber diameter, FC-Fat content, CC-Collagen content, CS-Collagen solubility, DL-Drip loss, AP-Appearance, TD- Tenderness, JU-Juiciness, FL-Flavour, ACT-Amount of connective tissue, OA-Overall acceptability, H-Hue, CH-Chroma, L*-lightness, a*- redness and b*-yellowness.
Fig. 2: Component loading plot
Destefanis et al. (2000) also reported closely located eating quality attributes on the right side and hydroxyproline content and WBSF on the left side of the PC loading plot for beef. FC, SL and FB are also closely situated but were closer to the origin. The loading plot of the buffalo meat attributes presents some significant findings. The PCs are interpreted according to the correlations between each attribute and each PC, thus measurements close to each other are positively correlated, measurements separated 180° are negatively correlated, whereas those separated by 90° are independent (Kopuzlu et al., 2011). It points out that one parameter among MFI, CC or WBSF may explain the variation between muscles with respect to these attributes. These attributes were negatively correlated with the sensory scores as these two groups were placed opposite to each other. Similarly, as eating quality attributes were closely correlated, one attribute among these, say sensory tenderness may be used to explain the sensory variability in these muscles. The CS and sensory amount of connective tissue were only moderately correlated. The sensory attributes were placed away from pH and colour characteristics. The structural and compositional characteristics including FB, SL and FC were not sufficient to explain the variability between the muscles. Some physico-chemical attributes like CL, WHC and pH were also inadequate to explain the variability. Destefanis et al. (2000) stated that variables in the loading plot that lie close together are positively correlated while those lying opposite to each other tend to have negative correlation.
The study has shown that PCA can be effectively used for a holistic judgment of buffalo meat quality by identifying the groups of variables that determine the nature and extent of variability in buffalo meat by data reduction and visualization. However, for more analytical information to estimate any effects and explain differences between treatments, traditional statistical methods may be applied.