The use of principal component analysis (PCA) to characterize beef

Meat Science 56 (2000) 255±259

www.elsevier.com/locate/meatsci

The use of principal component analysis (PCA) to characterize beef G. Destefanis *, M.T. Barge, A. Brugiapaglia, S. Tassone Department of Animal Science, University of Turin, Via L. Da Vinci 44, 10095 Grugliasco, Torino, Italy Received 11 November 1999; received in revised form 27 March 2000; accepted 26 April 2000

Abstract Principal component analysis was performed to study the relationships between chemical, physical and sensory variables (n=18) measured on longissimus thoracis et lumborum of 79 young bulls from the following ethnic groups: hypertrophied Piemontese, normal Piemontese, Friesian, crossbred hypertrophied PiemonteseFriesian, Belgian Blue and White. The ®rst three PCs explained about 63% of total variability. Sensory characteristics, protein content, shear force and cooking losses resulted the most e€ective variables for the PC1, while hydroxyproline and ether extract content, as well as hue and lightness were useful to de®ne the PC2. The distribution of the objects on the axes of the ®rst two PCs allowed the identi®cation of two groups, the ®rst one including meats of the hypertrophied animals (Piemontese and Belgian Blue and White) the second one including normal Piemontese and Friesian. However, a considerable variability within groups was noted. The crossbreds were placed between the two previous groups. In conclusion, PCA proved to be a very e€ective procedure to obtain a synthetic judgement of meat quality. # 2000 Elsevier Science Ltd. All rights reserved.

1. Introduction Beef quality can be a€ected by several factors, like breed, rearing technique, transport and slaughtering of the animals and post mortem technological treatments of the carcasses. It follows that meat is a very heterogeneous product. In order to describe completely its qualitative characteristics, di€erent kinds of analyses are needed, chemical, physical and sensory. The statistical processing of such a large amount of heterogeneous data according to classical methods gives important information to study every single variable. However, being too analytical, it does not provide a global knowledge on the relationships among the di€erent variables, nor allow the grouping of samples with homogeneous characteristics. Therefore, it may be useful to have a few elements to synthetize the trend of some particular phenomenon. To meet this need we can make use of multivariate statistical methods, such as principal component analysis (PCA), which makes it possible to identify the most

* Corresponding author. Tel.: + 39-011-6708569; fax: +39-0116708563. E-mail address: [email protected] (G. Destefanis).

important directions of variability in a multivariate data matrix and to present the results in a graphical plot. The principal component analysis transforms the original variables into new axes, or principal components (PCs), which are orthogonal, so that the data presented in those axes are uncorrelated with each other; therefore, PCA expresses as much as possible of the total variation in the data in only a few principal components and each successively derived PC expresses decreasing amounts of the variation (Smith, 1991). As explained by Naes, Baardseth, Helgesen and Isakson (1996), the results concerning both variables (loading plot) and objects (score plot) are presented on an xy plane. The abscissa corresponds to the ®rst principal component, the ordinate corresponds to the second principal component. Samples to the right in the score plot have high values for variables placed to the right in the loading plot. The same holds for samples to the left, at the top, or at the bottom. Objects close together have similar characteristics; variables close together are positively correlated, while variables lying opposite to each other in the loading plot tend to have a negative correlation. The more a variable is away from the axis origin, the better it is represented on the considered plane. This paper reports the results of the PCA employed to analyse chemical, physical and sensorial data of meat from young bulls of di€erent ethnic groups.

0309-1740/00/$ - see front matter # 2000 Elsevier Science Ltd. All rights reserved. PII: S0309-1740(00)00050-4

256

G. Destefanis et al. / Meat Science 56 (2000) 255±259

Meat Science Association (Cross et al., 1978), were employed. Sensory characteristics concerned appearance of the raw meat and eating qualities of the cooked meat. These latter included: tenderness (ease of sinking, friability and residue after chewing), initial and sustained juiciness and overall acceptability. The steaks were cooked on a grill, preheated at 250 C, to an internal temperature of 70 C.

2. Materials and methods The data relate to samples of longissimus thoracis et lumborum taken between the 9th thoracic and 1st lumbar vertebra from the right side of 79 carcasses of young bulls, 7 days after slaughtering (Barge, Brugiapaglia, Destefanis & Mazzoco, 1993; Destefanis, Barge & Brugiapaglia, 1994, 1996; Destefanis, Brugiapaglia & Barge, 1993). All the animals were kept in similar rearing conditions at the Experimental Centre of Animal Science Department of Turin. The animals were tyingtype housed and fed with mixed grass hay and concentrate in order to meet the nutritive requirement of 1 kg daily gain. The slaughter procedure and the ageing of meat occurred in the same conditions. The animals belonged to the following ethnic groups: . hypertrophied Piemontese (HP; n=23); . normal Piemontese (NP; n=12); . hypertrophied PiemonteseFriesian crossbred (HPF; n=10); . Friesian (F; n=11); . Belgian Blue and White (BBW; n=23). A total of 18 variables were analysed: 1. pH, measured 24 h after slaughtering on the right side at the 13th t.v. level, by a Hanna HI 9025 pHmeter provided with an Ingold spear electrode and an automatic temperature compensator; 2. water, protein and ether extract content (AOAC, 1970); 3. hydroxyproline content (International Organization of standardization [ISO], 1978) and heat-solubility of collagen (2 h at 80 C; SoÈrensen, 1981); 4. lightness and hue, according to parameters of the Hunter system, by a Minolta CR 331C colorimeter (Boccard et al. 1981); 5. drip losses, on a 1.5 cm thick steak kept for 48 h at 5 C in a plastic container with a double bottom (LundstroÈm & Malmfors, 1985); 6. cooking losses, on a 3.0 cm thick steak, which was sealed in a polyethylene bag and kept for 30 min in a water bath at 70 C; 7. shear force (kg) on cylindrical cores 2.54 cm in diameter, taken in parallel to muscular ®bres, and obtained from the steaks previously employed for cooking losses analysis; the shear force was measured by an Instron 1011, equipped with a Warner Bratzler shear and calibrated on a slipping speed of 100 mm/min; 8. sensory analysis, performed by a trained panel on an 8-point structured scale, where 1 and 8 were respectively the minimum and the maximum score. Seven assessors, selected and trained for beef evaluation according to guidelines of American

The data were analysed with SPSS package (SPSS, 1997), after standardization of the variables to mean of zero and variance of one. 3. Results and discussion Table 1 shows means, standard deviations, coecient of variation of the variables. The coecient of variation of some variables, such as pH, water and protein content, lightness, is lower than 10%, while for some others, like drip losses, hydroxyproline content, Warner±Bratzler shear and, above all, ether extract, is higher than 30%. Table 2 shows the correlation coecients between the 18 variables. There are several signi®cant correlations among variables, chemical, physical or sensorial, determined on raw or cooked meat. For example, there is a high positive correlation between hydroxyproline content and Warner±Bratzler shear, as reported by De Smet, Claeys, Buysse, Lenaerts and Demeyer (1998). Warner± Bratzler shear also shows a high positive correlation with cooking losses, as observed by Silva, Patarata and Martins (1999). The parameters of sensory tenderness are negatively correlated with Warner±Bratzler shear, Table 1 Mean, standard deviation (S.D.) and coecient of variation (C.V.) of the variables

pH Water (W),% Protein (P),% Ether extract (E),% Hydroxyproline (Hy), mg/g Collagen solubility (Cs),% Lightness (L) Hue (H) Drip losses (Dl),% Cooking losses (Cl),% Warner±Bratzler shear (WB), kg Appearance (A)a Ease of sinking (Te)a Friability (Tf)a Residue (Tr)a Initial juiciness (Ji)a Sustained juiciness (Js)a Overall acceptability (Oa)a a

Mean

S.D.

C.V.

5.50 75.17 22.20 0.58 529.66 11.46 35.30 23.52 2.53 30.84 8.27 6.20 6.44 6.19 5.98 5.92 5.60 6.13

0.18 0.64 0.85 0.45 181.25 3.24 2.70 4.15 0.82 3.10 2.86 0.78 0.69 0.68 0.70 0.64 0.64 0.70

3.27 0.90 3.82 77.59 34.22 28.27 7.65 17.72 32.41 10.05 34.58 12.58 10.71 10.99 11.71 10.81 11.43 11.42

8-Point structured scale (1=minimum score; 8=maximum score).

G. Destefanis et al. / Meat Science 56 (2000) 255±259

257

Table 2 Correlation coecients between the meat quality variablesa

pH W P E Hy Cs L H Dl Cl WB A Te Tf Tr Ji Js Oa

pH

W

P

E

Hy

Cs

L

H

Dl

Cl

WB

A

Te

0.09 0.28* ÿ0.28* ÿ0.33** ÿ0.08 ÿ0.02 ÿ0.33** 0.01 ÿ0.38** ÿ0.26* 0.10 0.17 0.10 0.08 0.08 0.01 0.13

ÿ0.40** ÿ0.16 ÿ0.08 ÿ0.01 0.03 ÿ0.23* 0.18 0.15 ÿ0.01 ÿ0.003 ÿ0.16 ÿ0.17 ÿ0.19 ÿ0.08 ÿ0.09 ÿ0.13

ÿ0.56** ÿ0.55** ÿ0.03 0.34** ÿ0.47** ÿ0.07 ÿ0.64** ÿ0.63** 0.25* 0.27* 0.20 0.23* 0.03 ÿ0.004 0.21

0.59** 0.05 ÿ0.31** 0.40** 0.08 0.44** 0.42** ÿ0.42** ÿ0.11 ÿ0.09 ÿ0.13 ÿ0.004 ÿ0.01 ÿ0.09

0.16 ÿ0.48** 0.62** ÿ0.12 0.66** 0.72** ÿ0.33** ÿ0.26* ÿ0.22* ÿ0.24* ÿ0.05 ÿ0.02 ÿ0.22

ÿ0.02 ÿ0.03 ÿ0.10 ÿ0.01 ÿ0.03 ÿ0.19 0.01 0.06 ÿ0.02 0.05 0.03 0.07

ÿ0.21 0.25* ÿ0.45** ÿ0.55** 0.35** 0.19 0.19 0.20 ÿ0.06 ÿ0.02 0.22

ÿ0.13 0.65** 0.67** 0.07 ÿ0.19 ÿ0.10 ÿ0.10 ÿ0.03 0.08 ÿ0.14

0.03 ÿ0.11 0.02 ÿ0.02 ÿ0.03 ÿ0.02 ÿ0.13 ÿ0.15 ÿ0.01

0.73** ÿ0.18 ÿ0.36** ÿ0.31** ÿ0.32** ÿ0.12 ÿ0.07 ÿ0.34**

ÿ0.28* ÿ0.38** ÿ0.32** ÿ0.33** ÿ0.10 ÿ0.03 ÿ0.37**

0.24* 0.27* 0.33** 0.16 0.24* 0.31**

0.93** 0.91** 0.69** 0.66** 0.92**

Tf

Tr

Ji

Js

0.94** 0.72** 0.72** 0.70** 0.70** 0.93** 0.92** 0.91** 0.80** 0.79**



=P<0.01; P<0.05, levels of signi®cance. Water (W); protein (P); ether extract (E); hydroxyproline (Hy); collagen solubility (Cs); lightness (L); hue (H); drip losses (Dl); cooking losses (Cl); Warner±Bratzler shear (WB); appearance (A); ease of sinking (Te); friability (Tf); residue (Tr); initial juiciness (Ji); sustained juiciness (Js); overall acceptability (Oa). a

cooking losses and hydroxyproline content, in agreement with data reported by other authors (De Smet et al., 1998; Silva et al., 1999; Vestergaard, Oksbjerg & Henckel, 2000; Whipple, Koohmaraie, Dikeman & Crouse, 1990). pH result correlated with cooking losses, as observed by Guignot, Touraille and Monin (1992) in veal meat but, contrary to what was reported by these authors, is not correlated with eating characteristics. The results of the principal component analysis are presented in Table 3 for the six principal components (PC). The analysis shows that about 34% of the total variation is explained by the ®rst principal component, 54.5% by the ®rst two principal components and the 62.5% by the ®rst there principal components. In other words, 62.6% of the total variance in the 18 considered variables can be condensed into three new variables (PCs). Table 4 shows that the most important variables for the ®rst PC were eating characteristics (except for the sustained juiciness), protein content, shear force and cooking losses. Table 3 Results from the principal component analysis for the ®rst six principal components Component

Eigenvalues

% of variance

Cumulative variance,%

1 2 3 4 5 6

6.10 3.72 1.43 1.36 1.20 1.00

33.90 20.64 7.94 7.54 6.69 5.58

33.90 54.54 62.48 70.02 76.71 82.29

So, the ®rst PC is de®ned by the eating quality, one chemical parameter and two physical parameters. In fact, in the loading plot (Fig. 1) these variables are placed far from the origin of the ®rst PC. In particular, the eating characteristics placed to the right in the loading plot are close together and, therefore, positively correlated. The PC2 is characterized by two chemical (hydroxyproline and ether extract) and two physical (hue and lightness) parameters. These variables, placed on the left in the loading plot are positively correlated. Finally, water content and drip losses are important for the PC3. The score plot (Fig. 2) shows the location of the objects in the multivariate space of two ®rst principal component score vectors. It can be seen that the scores are substantially arranged in two groups: the ®rst one includes meats of hypertrophied animals and the second one includes meats of the normal Piemontese and Friesian. Indeed the meat of the hypertrophied Piemontese and Belgian Blue and White shows, in general, high protein content and lightness and good eating qualities; on the other side, the meat of normal Piemontese and Friesian shows high values of shear force, cooking losses, hydroxyproline and ether extract content. However, these two groups exhibit a certain variability, because some hypertrophied Piemontese and Belgian Blue and White animals, placed to the left in the score plot (Fig. 2), have poor eating qualities, while few normal Piemontese have fair eating characteristics. The crossbreds show a very heterogeneous distribution in the score plot, which makes it dicult to de®ne them on the basis of the considered variables. In fact some samples present good eating qualities, even better

258

G. Destefanis et al. / Meat Science 56 (2000) 255±259

than those of hypertrophied Piemontese and Belgian Blue and White animals, while others are similar to normal Piemontese and Friesian. On the whole the crossbreds are placed between the two previous groups.

On the basis of the overall results concerning di€erent analyses, PCA proved to be a very useful method to identify the most e€ective variables and to point out quickly the relationships among the variables themselves. In fact,

Table 4 Principal component loadings

pH Water Protein Ether extract Hydroxyproline Collagen solubility Lightness Hue Drip losses Cooking losses Warner±Bratzler shear Appearance Ease of sinking Friability Residue Initial juiciness Sustained juiciness Overall acceptability

PC1

PC2

PC3

PC4

PC5

PC6

0.29 ÿ0.14 0.55 ÿ0.42 ÿ0.58 ÿ0.008 0.42 ÿ0.43 ÿ0.007 ÿ0.66 ÿ0.67 0.43 0.86 0.84 0.85 0.66 0.62 0.87

ÿ0.31 ÿ0.16 ÿ0.53 0.55 0.64 0.10 ÿ0.42 0.61 ÿ0.17 0.54 0.57 ÿ0.10 0.36 0.43 0.41 0.59 0.63 0.42

ÿ0.08 0.71 ÿ0.45 ÿ0.03 ÿ0.11 ÿ0.28 0.30 0.004 0.63 0.23 ÿ0.01 0.29 0.01 0.03 0.05 0.02 0.05 0.07

ÿ0.33 ÿ0.45 0.21 ÿ0.15 ÿ0.01 ÿ0.40 0.34 0.52 ÿ0.04 0.11 0.08 0.61 ÿ0.07 ÿ0.02 0.05 ÿ0.16 ÿ0.03 ÿ0.06

ÿ0.47 ÿ0.29 0.02 0.42 0.08 0.36 0.41 ÿ0.04 0.45 ÿ0.09 ÿ0.21 ÿ0.29 0.08 0.09 0.06 ÿ0.16 ÿ0.17 0.07

ÿ0.33 0.29 ÿ0.06 ÿ0.33 0.03 0.67 0.20 ÿ0.07 ÿ0.36 0.06 ÿ0.04 0.23 ÿ0.10 ÿ0.20 ÿ0.05 0.02 0.08 0.002

Fig. 1. Plot of the ®rst two PC loading vectors. Water (W); protein (P); ether extract (E); hydroxyproline (Hy); collagen solubility (Cs); lightness (L); hue (H); drip losses (Dl); cooking losses (Cl); Warner±Bratzler shear (WB); appearance (A); ease of sinking (Te); friability (Tf); residue (Tr); initial juiciness (Ji); sustained juiciness (Js); overall acceptability (Oa).

G. Destefanis et al. / Meat Science 56 (2000) 255±259

259

Fig. 2. Plot of the ®rst two PC score vectors.

in comparison to the classical correlations, as those reported in Table 2, the PCA allows at ®rst sight the identi®cation of which variables are correlated to each other and in which direction. Concerning the arrangement of the samples in relation to the variables, a substantial di€erentiation was shown due to breed and/or muscular hypertrophy; in fact, it has been possible to discriminate groups of samples, indicating di€erences in meat characteristics. Moreover, the existence of considerable variability within groups can be remark shown. In conclusion, even if does not provide the analytical information derived from other statistical methods, PCA is a very e€ective procedure to obtain a synthetic judgement of meat quality. References Cross, H. P., Bernholdt, H. F., Dikeman, M. E., Greene, B. E., Moody, W. G., Staggs, R., & West, R. L. (1978). Guidelines for cookery and sensory evaluation of meat. Chicago: American Meat Science Association. Horwitz, W. (Ed.). Association of Analytical Chemists (1970). Ocial methods of analysis. (11th ed.). Washington: Association of Analytical Chemists Barge, M. T., Brugiapaglia, A., Destefanis, G., & Mazzocco, P. (1993). Proceedings of 39th International Congress of Meat Science Technology, Calgary, Alberta, File S5P02.WP. Boccard, R., Butcher, L., Casteels, E., Cosentino, E., Drans®eld, E.,

Hood, D. E., Joseph, R. L., MacDougall, D. B., Rhodes, D. N., SchoÈn, I., Tinbergen, B. J., & Touraille, C. (1981). Liv. Prod. Sci., 8, 385±397. De Smet, S., Claeys, E., Buysse, G., Lenaerts, C. & Demeyer, D. (1998). Proceedings of the 44th International Congress of Meat Science Technology, (pp. 288±289) Barcelona, Spain. Destefanis, G., Barge, M. T. & Brugiapaglia, A. (1994). Proceedings of the 40th International Congress of Meat Science Technology, The Hague, Netherlands, File S-IV.25. Destefanis, G., Barge, M. T. & Brugiapaglia, A. (1996). Proceedings 42nd International Congress of Meat Science and Technology, Lillehammer, Norway, p. 298±299. Destefanis, G., Brugiapaglia, A. & Barge, M. T. (1993). Proceedings of the 39th International Congress on Meat Science and Technology., Calgary, Alberta, Canada, File S2P06.WP. Guignot, F, Touraille, C. & Monin, G. (1992). Proceedings of the 38th International Congress of Meat Science and Technology, ClermontFerrand, France, p. 367±370. International Organisation for Standardization (1978). Meat and meat products, part II, determination of l(-)hydroxyproline content. ISO, 3496. LundstroÈm, K., & Malmfors, G. (1985). Meat Science, 15, 203±215. Naes, T., Baardseth, P., Helgesen, H., & Isakson, T. (1996). Meat Science, 43(No S), Sl35±Sl49. Silva, J. A., Patarata, L., & Martins, C. (1999). Meat Science, 52, 453±459. Smith, G. L. (1991). Analytical Proc., 28, 150±151. SoÈrensen S. E. (1981). Thesis in meat science and technology. The Royal Veterinary and Agricultural University, Dept. of Meat Science and Technology, Copenhagen. SPSS Inc. (1997). SPSS base 7.5 for windows. Chicago, IL: SPSS Inc. Vestergaard, M., Oksbjerg, N., & Henckel, P. (2000). Meat Science, 54, 177±185. Whipple, G., Koohmaraie, M., Dikeman, M. E., & Crouse, J. D. (1990). Journal of Animal Science, 68, 4193±4199.