HRMAS-NMR spectroscopy and multivariate analysis meat characterisation

Meat Science 92 (2012) 754–761

Contents lists available at SciVerse ScienceDirect

Meat Science journal homepage: www.elsevier.com/locate/meatsci

HRMAS-NMR spectroscopy and multivariate analysis meat characterisation Mena Ritota a, Lorena Casciani a, Sebastiana Failla b, Massimiliano Valentini a,⁎ a b

Agricultural Research Council-Research Centre for the Soil-Plant System, Instrumental Centre of Tor Mancina Strada della Neve Km 1, 00016 Monterotondo, Rome, Italy Agricultural Research Council-Research Centre for Meat Production and Genetic Improvement, Via Salaria 29, 00016 Monterotondo, Rome, Italy

a r t i c l e

i n f o

Article history: Received 11 April 2012 Accepted 21 June 2012 Keywords: NMR assignment PLS-DA Longissimus dorsi Semitendinosus Intelligent bucketing

a b s t r a c t 1 H-high resolution magic angle spinning-nuclear magnetic resonance spectroscopy was employed to gain the metabolic profile of longissimus dorsi and semitendinosus muscles of four different breeds: Chianina, Holstein Friesian, Maremmana and Buffalo. Principal component analysis, partial least squares projection to latent structure – discriminant analysis and orthogonal partial least squares projection to latent structure – discriminant analysis were used to build models capable of discriminating the muscle type according to the breed. Data analysis led to an excellent classification for Buffalo and Chianina, while for Holstein Friesian the separation was lower. In the case of Maremmana the use of intelligent bucketing was necessary due to some resonances shifting allowed improvement of the discrimination ability. Finally, by using the Variable Importance in Projection values the metabolites relevant for the classification were identified. © 2012 Elsevier Ltd. All rights reserved.

1. Introduction Meat is an essential food in the human diet since it is an excellent source of amino acids, fatty acids, vitamins, especially B group, and minerals such as iron, copper and zinc (Pearson & Gillett, 1999). In the western world its consumption has markedly increased in the last decades, and the quality demand increased accordingly. The term quality for meat encompasses several factors, above all hygiene, nutrition, technology and sensory (Verbeke, Van Oeckel, Warnants, Viaene, & Boucque, 1999). The most important aspects contributing to meat quality, thus determining its acceptability among consumers, are taste (Shahidi, Rubin, & Dsouza, 1986), texture (Maltin, Balcerzak, Tilley, & Delday, 2003), and juiciness (Cheng & Sun, 2008). The last two are determined by microscopic and macroscopic tissue organizations, while chemical composition defines the taste. Several aspects of meat have been extensively investigated, including the use of metabolomics for evaluating some quality related properties. Metabolomics takes into account the pattern of the low molecular weight species, and is defined as the systematic study of the unique chemical fingerprints that specific cellular processes leave behind. Many analytical techniques have been used in metabolomics, including nuclear magnetic resonance spectroscopy. The latter was employed for determining the metabolic pattern of a large number of foods, predominantly fruits and vegetables. Tomato (Le Gall, Colquhoun, Davis, Collins, & Verhoeyen, 2003), lettuce (Sobolev, Brosio, Gianferri, & Segre, 2005), mango (Duarte, Goodfellow, Gil, &

⁎ Corresponding author. Tel.: +39 06 90627203; fax: +39 06 9068309. E-mail address: [email protected] (M. Valentini). 0309-1740/$ – see front matter © 2012 Elsevier Ltd. All rights reserved. doi:10.1016/j.meatsci.2012.06.034

Delgadillo, 2005), juices (Spraul et al., 2009), and grape berries (Son et al., 2008) are examples. On the contrary, meat systems have been poorly investigated by NMR, most likely because of complicated and time demanding extraction and purification procedures. Bertram, characterized meat metabolome as a function of pre-slaughter exercise (Bertram, Oksbjerg, & Young, 2010) and heat stress (Straadt et al., 2010). Recently, high resolution magic angle spinning-nuclear magnetic resonance (HRMAS-NMR) has been proposed as a reliable system based for assessing the metabolic profile. It offers the opportunity of measuring samples without any chemical and/or physical preparation, by producing highly resolved spectra. While HRMAS-NMR has been extensively used in medicine and related fields (Rocha et al., 2010; Wang et al., 2003), its application in food characterisation is still in its infancy, and few studies have been reported (Valentini et al., 2011), mainly referring to fresh vegetables (Perez, Iglesias, Ortiz, Perez, & Galera, 2010; Ritota, Marini, Sequi, & Valentini, 2010) and processed materials (Ciampa, Renzi, Taglienti, Sequi, & Valentini, 2010; Consonni, Cagliani, & Cogliati, 2011; Shintu & Caldarelli, 2005, 2006). Brescia et al. (2002) reported a preliminary study on meat composition by HRMAS-NMR. Later, the characterisation of Apulian lamb meat according to geographical origin was determined, by combining HRMAS-NMR spectroscopy with other analytical techniques (Sacco, Brescia, Buccolieri, & Jambrenghi, 2005). Also Shintu used 1H-HRMASNMR spectroscopy for the identification of molecular markers in the traceability of dried beef (Shintu, Caldarelli, & Franke, 2007). Finally, Renou characterized meat samples according to the production site and feed, by combining 18O IRMS with NMR data (Renou et al., 2004). Here is reported the comprehensive assignment of the metabolic pattern determined by 1H-HRMAS-NMR. The multivariate NMR data

M. Ritota et al. / Meat Science 92 (2012) 754–761

analysis allowed the identification, to the best of our knowledge, for the first time, of the metabolites capable of discriminating longissimus dorsi and semitendinosus muscles according to breed.

755

the HSQC since the latter is more sensible to the optimization of the acquisition parameters. 2.3. Multivariate data analysis

2. Materials and methods 2.1. Meat samples 47 young bulls were used, 5 buffalo, 18 chianina, 15 maremmana and 9 Holstein. Longissimus dorsi (ld) and semitendinosus (st) muscles were sampled; 5 ld and 5 st for buffalo, 18 ld and 18 st for chianina, 15 ld and 12 st for maremmana and 9 ld and 9 st for Holstein Friesian. All animals were reared in the meat production and genetic improvement research center (CRA-PCM, Rome Italy) according to the national guide for animal care. Cattle were intensively reared in stalls and were fed hay and maize silage ad libitum and 800 g of concentrate/ 100 kg of live weight in the finishing period (3 months before slaughter). Animals were slaughtered at about 550 kg of liveweight according to EU legislation, Reg. CE n. 1099/2009. Samples were collected and frozen in liquid nitrogen prior to transportation to the laboratory, and then stored at −80 °C. 2.2. HRMAS-NMR measurements Samples were prepared by inserting ca. 10 mg of meat in a 4 mm HRMAS rotor with a 50 μL spherical insert: approximately 40 μL of D2O phosphate buffer, 0.01 M, pH 7.2 and containing 0.01% TSP, i.e. 3-(trimethylsilyl)-propionic-2,2,3,3-d4 acid sodium salt, were then added. HRMAS-NMR spectra were recorded at 298 K with a Bruker AVANCE spectrometer operating at a 1H frequency of 400.13 MHz, equipped with a 4 mm HRMAS dual channel probe head and spinning the rotor at 7 kHz. 1H NMR spectra were referenced to the TSP methyl groups signal at δ= 0.00 ppm. Similarly 13C NMR spectra were referenced to the TSP signal δ = 0.00 ppm. 1 H-HRMAS-NMR spectra were acquired by using a water suppression pulse sequence containing a NOE enhancement, with 32 K data points over a 4807 Hz spectral width and adding 256 transients. A recycle delay of 3 s, a NOE build up period of 150 ms and a 90° pulse length of 5.88 μs were used. NMR data were processed using ACD/ Spec Manager 8.00 software (Advanced Chemistry Development Inc., Toronto, Canada). Each 1H-HRMAS-NMR spectrum was FT transformed with 64 K data points, manually phased and base-lined, and a line broadening factor equal to 0.3 Hz was applied to the FID prior to FT. Assignment of most resonances in the proton spectrum was made by connectivity information obtained from 2D spectra and the use, as guidelines, of chemical shift data reported in the literature (Brescia et al., 2002; Graham, et al., 2010; Shintu et al., 2007; Valentini et al., 2004). 13 C-HRMAS-NMR spectra were acquired with the power-gated decoupling sequence, using a 30° flip angle pulse of 5.0 μs. Experiments were carried out using 64 K data points over a 22,123 Hz (~ 220 ppm) spectral width by adding 64 K transients with a recycle delay of 3 s. Each spectrum was FT transformed with 128 K data points and manually phased and base-lined, and a line broadening factor of 0.5 Hz was applied to the FID. 1 H– 1H-TOCSY experiments were collected in the phase-sensitive mode using time proportional phase incrementation (TPPI) for quadrature detection in the direct dimension, with a 4807 Hz spectral width in both dimensions, 100 ms of spin-lock time, 2 K data points in f2, and 1 K increments in f1, each with 32 scans. The water signal was suppressed. 1 H– 13C-HMQC spectra were acquired in TPPI phase-sensitive mode, with a 4807 Hz spectral width in f2 dimension and a 15,083 Hz spectral width in f1. 1 K data points in f2 and 256 increments in f1, each with 32 scans, were used. HMQC was preferred to

Each NMR spectrum was divided into intervals equal to 0.04 ppm by using the ACD bucketing or, when necessary, the intelligent bucketing. The spectral region from 4.63 to 5.20 ppm was neglected in order to remove any spurious effects of variability occurring during the suppression of HDO resonance. Also the regions containing only noise, usually addressed as dark regions, were ignored. Finally, to take into account variations in sample concentration, thus making meaningful comparisons between samples, whole spectrum intensity was normalized to 100. Each 1H NMR bucketed spectrum was considered as a row vector and placed into a matrix of n rows (spectra) corresponding to the number of samples for a number of columns (variables) equal to the number of buckets used. Each variable was the integral of the spectrum area at the specific chemical shift. Multivariate analysis was performed using MATLAB (version 7.1, The Mathworks, Natick, MA, USA), PLS_Toolbox (version 502, Eigenvector Research Inc.) and home-made algorithms. Data were auto-scaled, that is variance scaled to unity and mean sets to zero, in order to avoid assigning large loadings to the highest intensity signals. The whole data set was used as training set and cross validation, by means of the leave-one-out method, was used to estimate the predictive ability of the model. The paired samples t-test was used to compare group means in the scores matrix, which is generally used when comparing two means that are repeated measurements for the same samples, scores might be repeated across different measures or across time, or comparing paired samples, as in a two treatment randomized block design. Furthermore, one-way ANOVA was used to test differences in means of the scores matrix for statistical significance. 3. Results and discussion 3.1. 1H-HRMAS-NMR spectrum assignment The 1H-HRMAS-NMR spectra of the meat samples are dominated by the intense resonances of lactic acid and creatine and/or phosphocreatine. Several minor signals, such as fatty acids, amino acids, organic acids and nucleosides, are detectable, and are summarized in Table 1. The high field region, top left panel of Fig. 1, contains signals belonging to the aliphatic groups of amino, organic and fatty acids. Signals in the range from 0.90 to 1.10 ppm belong to methyl or methylene groups of valine, leucine, and isoleucine, and the correct assignment was possible based on cross peaks in the TOCSY spectrum (not shown). In this region one finds also a series of broad peaks. The signal at 0.91 ppm has in the TOCSY spectrum cross peaks with proton atoms at δ equal to 1.33, 1.61, and 2.26 ppm, these were assigned to saturated fatty acids, i.e. palmitic and stearic acids which are the most abundant in meat. The HMQC spectrum, bottom right part of Fig. 1, confirmed this assignment, since 1H at δ = 1.33 and 0.91 ppm correlated with 13C at 30.38 and 14.79 ppm, respectively. Between 1.15 and 1.21 ppm different doublets with coupling constants varying from 6.16 to 6.95 Hz are present in several samples. Even though the corresponding correlations are not always clearly visible in the TOCSY spectrum, they were assigned to different hydroxybutyric acids, which are known to be involved in amino acid synthesis and degradation. Cross peaks between the signal at δ = 1.32 ppm and those at 2.04, 2.79 and 5.36 ppm indicates the presence of unsaturated lipid chains, with oleic and linoleic being predominant (Wood et al., 2008). The broad peak at δ = 0.74 ppm was tentatively assigned to terminal CH3 of sterols. One of the most intense peak in this region belongs to the β-CH3 of lactic acid, δ = 1.33 ppm; it shows correlations with

756

M. Ritota et al. / Meat Science 92 (2012) 754–761

Table 1 Metabolites identified in the 1H HRMAS-NMR spectrum of meat sample. 1H and chemical shifts are reported by referring to TSP signal (δ = 0.00 ppm). Compound

Assignment

Carbohydrates β-Glucose (β-Glc)

CH-1 CH-2 CH-3 CH-4 CH-5 CH2-6,6′ α-Glucose (α-Glc) CH-1 CH-2 CH-3 CH-4 CH-5 CH2-6,6′

Organic acids Acetic acid Formic acid Fumaric acid Lactic acid (Lac) Pyruvic acid Amino acids Alanine (Ala) Arginine (Arg)

Aspartate (Asp) Glutamine (Gln)

Glutamate (Glu)

Isoleucine (Ile)

Leucine (Leu)

Lysine (Lys)

Metionine (Met)

Phenylalanine (Phe) Proline (Pro)

Serine (Ser) Threonine (Thr)

Tyrosine (Tyr)

Table 1 (continued) 13

1

H (ppm) Multiplicity 13C [J (Hz)] (ppm)

4.65 d [7.9] 3.25 3.49 3.40 3.43 3.89; 3.74 5.24 d [3.8] 3.54 3.72 3.44 3.82 3.79; 3.44

CH3 HCOOH α,β-CH_CH α-CH β-CH3 β-CH3

1.92 8.46 6.52 4.11 1.33 2.37

α-CH β-CH3 α-CH β-CH2 γ-CH2 δ-CH2 α-CH β,β′-CH2 α-CH β,β′-CH2 γ-CH α-CH β-CH β′-CH γ-CH2 α-CH β-CH γ-CH3 γ-CH γ′-CH δ-CH3 α-CH β-CH2 γ-CH δ-CH3 α-CH β-CH2 γ-CH2 δ-CH2 ε-CH2 α-CH β-CH2 γ-CH2 S-CH3 o-CH m-CH p-CH α-CH β-CH β′-CH γ-CH2 δ-CH2 α-CH β,β′-CH2 α-CH β-CH γ-CH3 CH-2,6, ring CH-3,5, ring

3.77 1.48 3.77 1.91 1.71 3.25 3.88 2.65 3.77 2.15 2.46 3.77 2.06 2.10 2.36 3.68 1.99 1.01 1.26 1.49 0.93 3.72 1.75 1.70 0.97 3.77 1.89 1.46 1.69 3.01 3.82 2.15 2.64 2.15 7.33 7.43 7.39 4.14 2.36 2.08 2.01 3.35 3.85 3.96 3.63 4.26 1.33 7.20 6.90

s s s q [6.9] d [6.9] s

69.03 20.71

d [7.2]

dd

m m m m

54.18 25.98 30.43 54.93 26.22 26.22 33.40

C

Compound

Assignment

1

Amino acids Valine (Val)

α-CH

3.63

β-CH γ-CH3 γ′-CH3

2.27 1.00 1.05

Fatty acids Saturated fatty acids C16 Palmitic acid CH2-3 (p) C18 Stearic acid CH2-2 CH2-4 ÷ CH2-15 (p) (s) CH2-4 ÷ CH2-17 (s) CH3-16 (p) CH3-18 (s) Monounsaturated fatty acids C18:1 CH2-2 (oleic acid) CH2-3 CH2-4,7 CH2-8 CH-9 CH-10 CH2-11 CH2-12,17 CH3-18 Polyunsaturated fatty acids C18:2 CH2-2 (linoleic acid) CH2-3 CH2-4,7 CH2-8 CH-9 CH-10 CH2-11 CH-12 CH-13 CH2-14 CH2-15,17 CH3-18 Other metabolites Choline Carnosine

d [7.2]

s d d

m d d [8.4] d [8.4]

19.30

N\CH3 CH-5, ring CH-3, ring CH to ring

CH′ to ring CH\COOH CH2\C_O CH2\NH2 Carnitine α-CH2 β-CH γ-CH2 N\CH3 Creatine and/or N\CH3 phosphocreatine N\CH2 Hydroxybutyric CH3\ acida Hypoxanthine NH\C_N NH\C_O ATP/ADP/AMP/ C1′H, ribose IMP C2′H, ribose C3′H, ribose C4′H, ribose CH2 CH-2, ring CH-8, ring Adenosine/inosine C1′H, ribose C2′H, ribose C3′H, ribose C4′H, ribose CH2 CH-2, ring CH-8, ring NAD Sterols

CH3

H (ppm) Multiplicity 13C [J (Hz)] (ppm)

m d [7.0] d [7.0]

1.62

24.49

2.26 1.32

33.40 28.95

0.91

14.79

2.27 1.64 1.32 2.04 5.34 5.34 2.04 1.32 0.94

33.40 24.49 28.95 26.72 130.64 130.64 26.72 28.95 14.79

2.27 1.64 1.32 2.04 5.34 5.34 2.79 5.34 5.34 2.04 1.32 0.94

33.40 24.49 28.95 26.72 130.64 130.64

3.19 7.10 8.11¸8.24 3.07

53.88 118.76 135.83 27.46

130.64 130.64 26.72 28.95 14.77

s s s dd [17.6; 8.6] 3.19 dd 4.49 m 2.68 m 3.24 m 2.46 m 4.57 m 3.43 m 3.23 s 3.04 s 3.94 s 1.15–1.21 d [6.16; 6.95] 8.19 s 8.22 s 6.15 d [5.9] 4.66 4.51 4.38 4.12 8.53 s 8.23 s 6.10 d [5.7] 4.44 4.29 3.93 8.24 8.31 8.72 8.28 0.74

s s Broad Broad Broad

27.46 54.93 31.17 34.88 43.05 72.00 70.51 54.18 36.37 54.18

M. Ritota et al. / Meat Science 92 (2012) 754–761 Table 1 (continued) Compound

Assignment

1

Other metabolites Taurine

S\CH2

3.26

m

N\CH2 NH\C_N

3.43 7.91

m s

Xanthine

H (ppm) Multiplicity 13C [J (Hz)] (ppm)

a The hydroxybutyric acid is present in different forms, here the range of chemical shift and coupling constants are reported.

the proton at 4.11 ppm and with 13C at 69.03 ppm, in TOCSY and HMQC spectra, respectively, both arising from α-CH. Alanine was assigned based on the correlation between signals at δ = 1.48 (β-CH3) and 3.77 ppm (α-CH). Also lysine, methionine and arginine were identified thanks to the correlations in the TOCSY spectrum, see bottom left of Fig. 1. Singlets at δ = 1.92 and 2.37 ppm were assigned to acetic and pyruvic acids, respectively. One can also recognize the patterns for glutamine and glutamate; multiplets at 2.06 and 2.10 ppm, β-CH and β′-CH, correlate in the TOCSY experiment with the signals at δ = 2.36 (γ-CH2) and 3.77 ppm (α-CH), respectively. The presence of glutamate is also confirmed by the HMQC spectrum. Resonances of glutamine are visible in the TOCSY spectrum: in fact, cross peaks of several multiplets at δ = 2.15, 2.46 and 3.77 ppm are present, and were assigned to β-CH2, γ-CH2, and α-CH, respectively. Signals at δ = 2.46, 3.43 and 4.57 were assigned to carnitine, α-CH2, γ-CH2 and β-CH, respectively. Correlations between these protons and the corresponding carbons in the HMQC spectrum supported this assignment, see bottom right panel of Fig. 1. The signal at δ

757

equal to 2.65 ppm shows a cross peak in the TOCSY experiment with a doublet of doublets at 3.88 ppm: they were assigned to aspartic acid, β,β′-CH2 and α-CH, respectively. The multiplet at 2.68 ppm arises from the CH2\C_O of carnosine, a dipeptide abundant in meat; all corresponding correlations were found in the TOCSY and HMQC spectra. Singlets arising from N\CH3 and N\CH2 at 3.04 and 3.94 ppm, respectively, arising from creatine and/or phosphocreatine were found. The assignments were confirmed by their correlations in the HMQC spectrum with the corresponding carbon atoms at 36.37 and 54.18 ppm, respectively. Once again, the typical signals from carnitine and carnosine were observed. Also taurine was identified thanks to its correlations at δ equal to 3.26 and 3.43 ppm. In the region from 3.00 to 4.00 ppm carbohydrate resonances strongly overlap with amino acid α-CH peaks. The most intense signals arise from the different isomeric forms of D-glucose, and all resonances of α-D and β-D-glucose were assigned. Peaks belonging to nucleosides and nucleotides were observed in this range: ATP, ADP, AMP, IMP, inosine and adenine are all species involved in meat metabolism and they differ in the NMR spectrum by small chemical shift variations. Doublets at δ = 6.10 and 6.15 ppm were assigned to the anomeric protons of the ribose moiety of the corresponding nucleosides. The singlet at 6.52 ppm arose from α,β-CH_CH of fumaric acid. Finally the doublet at 6.90 ppm, which correlates in the TOCSY spectrum with the signal at 7.20 ppm, was assigned to the CH-3,5, ring of tyrosine. The low field region from 7.10 to 9.00 ppm, top right panel of Fig. 1, is characterized by signals due to aromatic amino acids and nucleosides. The most intense resonances belong to carnosine: singlet at

Fig. 1. Representative NMR spectra for ld muscle of Friesian Holstein at zero aging time in phosphate/D2O buffer with 0.01% of 3-(trimethylsilyl)propionic-2,2,3,3-d4 acid sodium salt (TSP): on the top left panel expansion of the 1H-HRMAS-NMR region from 0.70 to 2.90 ppm; top right expansion of the low field region, from 7.10 to 9.00 ppm; TOCSY and HMQC are reported on the bottom, left and right, respectively.

758

M. Ritota et al. / Meat Science 92 (2012) 754–761

Table 2 Latent variables (LVs), RESSY2 (REsidual Sum of Squares Y), PRESS (PREdicted Sum of Squares), Q2, SSQX and SSQY (cumulative Sum of Squares on X and Y, respectively) for PLS-DA and OPLS-DA models and corresponding Fischer factor (F) and p-value (p) from ANOVA, for buffalo, chianina, Holstein Friesian and maremmana, from top to bottom. LV Buffalo PLS-DA 1 2 3 4 OPLS-DA 1 2 Chianina PLS-DA 1 2 3 4 OPLS-DA 1 2

RESSY2

PRESS

Q2

SSQX

SSQY

F

p

0.2098 0.0890 0.0456 0.0093

5.6706 5.2906 5.1901 6.0261

– 0.8172 0.9889 0.2677

0.3118 0.4896 0.6643

0.5804 0.8219 0.9088

11.0648 2.5479 0.7613

0.01044 0.14911 0.40835

0.1319 0.0734

5.6010 4.6779

– 0.0183

0.1878 0.6702

0.7361 0.8532

22.3192 1.0609

0.00149 0.33313

hypoxanthine, NH\C_N and NH\C_O, respectively. Both metabolites are present in traces in a few samples, since they are the end products of ATP degradation, thus being barely visible immediately after slaughter. The singlet at 8.46 ppm is not always resolved due to overlap with the intense carnosine peak; it was assigned to formate. Among the aromatic amino acids, signals at δ = 7.34, 7.45, and 7.39 ppm, which correlate to each other in the TOCSY spectrum, were recognized as the typical phenylalanine pattern. Singlets at 8.23 and 8.53 ppm were assigned to the CH-2, ring and CH-8, ring, respectively, of ATP and/or ADP and/or AMP and/or IMP, while the singlets at 8.24 and 8.31 ppm belong to the CH-2, ring and CH-8, ring of adenosine and/or inosine, respectively. Finally the broad signal at 8.72 ppm shows a cross peak in the TOCSY spectrum with the signal at 8.28 ppm; they most likely arise from NAD ring protons. 3.2. Multivariate NMR data analysis

−7

0.1138 0.2258 0.1838 0.1514

11.6487 11.1516 12.9639 13.6618

– 0.6807 −3.0241 0.4259

0.1622 0.3543

0.5484 0.6324

41.2847 3.1169

2.43 · 10 0.0865

0.2204 0.1749

11.1178 11.7729

– 0.45

0.3924

0.5591

43.1222

1.59· 10−7

10.6345 12.1116 11.4356 11.6088

– −2.044 0.4741 0.9675

0.2252

0.4335

12.2413

0.003

10.1054 12.0484

– −4.2248

0.4771

0.4452

12.8376

0.0025

17.7205 25.1864 22.0576 20.9745

– −87.7953 −12.0634 −0.397

0.2243

0.2461

8.1628

21.3445 20.6771

– 0.3197

0.2246 0.4202

0.3008 0.6448

10.7575 13.1074

Holstein Friesian PLS-DA 1 0.2833 2 0.1311 3 0.0773 4 0.0247 OPLS-DA 1 0.2774 2 0.126 Maremmana PLS-DA 1 0.3723 2 0.2506 3 0.1602 4 0.0937 OPLS-DA 1 0.3453 2 0.1754

0.008489

0.0031 0.0013

7.22 ppm, which has a cross peak in the HMQC spectrum at 118.76 ppm, was assigned to CH-5, ring, while the singlet at 8.11 ppm, correlating in the HMQC with carbon at 135.83, was assigned to the CH-5, ring. The singlet at 7.91 ppm was attributed to the NH\C_N of xanthine, and the singlets at 8.19 and 8.22 ppm to

3.2.1. Buffalo breed Explorative multivariate analysis, by means of principal component analysis (PCA) was performed on the entire dataset, composed of 5 ld and 5 st samples in order to reveal any inherent data clustering and to identify outliers, if present. Table S1 (Supplementary Data) reports the eigenvalues with the corresponding explained and cumulative variance percentages obtained from PCA for a number of factors equal to nine. A t-test was performed on the scores matrix in order to evaluate which PC contributed significantly to discrimination. According to the p-value, which represents the probability of being correct that the two populations are equivalent on the basis of their PC score, none of the PCs was useful in sample classification. PCA is not necessarily able to distinguish between ‘inter-class’ and ‘intraclass’ variances, when the latter dominates the former, so PLS-DA and OPLS-DA regression was used. Table 2 reports the number of Latent Variables (LVs), REsidual Sum of Squares Y (RESSY 2), PREdicted Sum of Squares (PRESS), Q 2 and cumulative Sum of Squares X and Y, SSQ and SSY, respectively, for the PLS-DA and OPLS-DA models. PRESS is a measurement of the predictive ability of the built model, Q 2 is the default parameter used in PLS-DA discriminations and focuses on how well the class label can be predicted from new data (Westerhuis et al., 2008), while SSQ and RESSY 2 are related to the fitting model's goodness. By using the SSQ and PRESS parameters and the Q 2 criterion as a guideline (Q 2 > 0.05) it is possible to evaluate the optimal number of components for a model with a good fitting, a high predictive ability and without overfitting. It was found that the best PLS-DA model consisted of three LVs, and ANOVA indicated that the LV1 was the most relevant variable. OPLS-DA analysis reduced the complexity of

Fig. 2. PLS-DA (left) and OPLS-DA (right) score plots obtained from the 1H-HRMAS-NMR data set of longissimus dorsi (cross) and semitendinosus (empty triangle) muscles of Buffalo.

M. Ritota et al. / Meat Science 92 (2012) 754–761

759

Fig. 3. PLS-DA (left) and OPLS-DA (right) scores plots obtained from the 1H-HRMAS-NMR data set of longissimus dorsi (cross) and semitendinosus (empty triangle) muscles of Chianina.

the model to two latent variables, thus leading to a better interpretation of loading matrix, since information is focused along a single axis. Also in this case, ANOVA pointed out that LV1 is the component with the higher discriminating ability. Fig. 2 reports the training data set plotted onto the space spanned by the first latent variable for PLSDA and OPLS-DA models, left and right panels, respectively. The latter, besides reducing the number of LVs to be included in the model, improved the predictive ability; in fact, in the corresponding scores plot ld and st muscles are better discriminated. This is confirmed by the lower PRESS value, when comparing the OPLS-DA model with the PLS-DA one. Based on the loading matrix, the metabolites contributing to the classification along LV1 axis were identified. They were chosen according to the VIP (Variable Importance in Projection) value, whose minimum was set equal to 0.8. It was found that the most relevant metabolites were valine, leucine, isoleucine, glutamate, glutamine, alanine and the fatty acids. All had positive weighting factors, that is their concentrations were higher in st muscles. On the other hand, carnosine, carnitine, taurine, α- and β-glucose, lactic acid and nucleosides/nucleotides, were higher in concentration in the ld muscle. Furthermore, other unassigned species contributed significantly to the classification along the LV1 axis; they were in the intervals 1.50–1.58 ppm, 4.02–4.06 ppm, 7.01–7.09 ppm and 7.97–8.09 ppm, the first one being higher in the semitendinosus muscle and the last three in the longissimus dorsi one. Metabolites responsible for

discrimination were found to be the same for both the PLS-DA and OPLS-DA models.

3.2.2. Chianina breed Results of PCA analysis performed on the Chianina dataset, composed of 18 ld and 18 st samples, for a number of factors equal to ten and the corresponding t-test are reported in Table S1 in the Supplementary Data. They showed that PC2 and PC3, having p-values equal to 0.0189 and 0.0004, respectively, are the most discriminating components. The PC3/PC2 scores plot (data not reported) showed a barely visible separation between the two datasets. In order to improve discriminating ability, PLS-DA and OPLS-DA were performed (Table 2). The best PLS-DA model, PRESS value as low as possible, was based on two latent variables, with LV1 being the most relevant for the discrimination (left panel of Fig. 3), as confirmed by ANOVA. OPLS-DA reduced the number of factors to be included in the model to one, nevertheless, it does not gain higher discriminating ability, as highlighted in the scores plot in the right panel of Fig. 3. Metabolites contributing to classification were alanine, glutamine, methionine, carnosine, carnitine, taurine, lactic acid, nucleosides/nucleotides, all present in higher quantity in semitendinosus muscle. Glutamate, αglucose, fatty acids and other unassigned signals present in five different spectral regions (5.52–5.68, 6.93–6.97, 7.45–7.49, 8.57–8.61 and 8.81–8.85 ppm) were higher in longissimus dorsi muscle.

Fig. 4. PLS-DA (left) and OPLS-DA (right) scores plots obtained from the 1H-HRMAS-NMR data set of longissimus dorsi (cross) and semitendinosus (empty triangle) muscles of Holstein Friesian.

760

M. Ritota et al. / Meat Science 92 (2012) 754–761

Fig. 5. PLS-DA (left) and OPLS-DA (right) scores plots obtained from the 1H-HRMAS-NMR data set of longissimus dorsi (cross) and semitendinosus (empty triangle) muscles of Maremmana.

3.2.3. Holstein Friesian This dataset was composed of 9 ld and 9 st samples. PCA results for a number of factors equal to ten are reported in Table S1 in the Supplementary Data, and show that this unsupervised method was unable to discriminate between the groups. PLS-DA and OPLS-DA were thus performed, and it was found the optimal models, in both cases depended on only one latent variable. Fig. 4 reports the PLS-DA and OPLS-DA scores plots, left and right panels, respectively. Nevertheless, none of them gave an appreciable separation. It was hypothesized that some NMR signals might undergo small variations in chemical shifts, most likely for pH or ionic strength variation, and in order to elucidate this issue the intelligent bucketing method was used. PCA was once again unable to discriminate samples (data not shown), as well as PLS-DA, and OPLS-DA models did not significantly improve the separation (data not shown). One possible reason is that for Holstein Friesian the ld and st muscles are very similar, at least for the metabolic pattern detectable by HRMAS-NMR spectroscopy.

model with two latent variables, with LV2 being the most discriminating, right panel of Fig. 5. However, neither PLS-DA nor OPLS-DA were able to discriminate the samples, as evident from the score plots. As in the case of Holstein Friesian, the intelligent bucketing strategy was used. PCA was once again unable to discriminate samples (data not shown). PLS-DA gave the best results for a one latent variable-based model, which was not able to gain suitable discrimination, as shown in the scores plot reported in the left panel of Fig. 6. On the contrary, OPLS-DA, whose model was based on two latent variables, achieved good separation, (right panel of Fig. 6). It was found that several metabolites were relevant for this discrimination: valine, leucine, isoleucine, glutamate, glutamine, methionine, lysine and arginine, acetate, pyruvate, hydroxybutyric derivatives, carnitine, taurine and the nucleosides/nucleotides, all being present in higher quantity in semitendinosus muscle. Carnosine and fatty acids were found to be higher in concentration in longissimus dorsi muscle.

3.2.4. Maremmana breed Dataset consisted of 15 ld and 12 st samples. It was found that PCA was not suitable for discrimination, (Table S1 in the Supplementary Data), most likely because the intra-class variance was higher than the inter-class one. PLS-DA resulted in one latent variable-based model, left panel of Fig. 5, while OPLS-DA gave the best fitting

High resolution magic angle spinning-nuclear magnetic resonance spectroscopy was used to gain the metabolic profile of longissimus dorsi and semitendinosus of four different breeds. One- and two-dimensional NMR spectra were performed directly on few milligram samples, without any chemical and/or physical manipulation, allowing the recognition of several compounds within a single NMR experiment.

4. Conclusion

Fig. 6. PLS-DA (left) and OPLS-DA (right) scores plots obtained from the intelligent bucketing of the 1H-HRMAS-NMR data set of longissimus dorsi (cross) and semitendinosus (empty triangle) muscles of Maremmana.

M. Ritota et al. / Meat Science 92 (2012) 754–761

The use of multivariate statistical methods, i.e. PCA, PLS-DA and OPLS-DA, allowed the classification of muscle type according to the breed. In the case of Buffalo and Chianina the discrimination capability was very high, while for the other two breeds the model's goodness and predictive ability was lower. Most of the metabolites responsible for classification were clearly identified. The results of the present research are preparatory for forthcoming papers dealing with the discrimination of the cattle breeding according to muscle type and of the cattle species according to muscle type. Moreover, the knowledge of the metabolic pattern will open up the possibility for the assessment of the metabolic modification occurring during post-mortem aging. Supplementary materials related to this article can be found online at http://dx.doi.org/10.1016/j.meatsci.2012.06.034. Acknowledgment This work was carried out within the QUALITEC project, co-funded by the Agricultural Research Council and the Ministero per le politiche agricole, alimentari e forestali. We thank Dr. D. Meo Zillo and Dr. M. Contò, CRA-PCM, for the assistance on the meat sample supply. References Bertram, H. C., Oksbjerg, N., & Young, J. F. (2010). NMR-based metabonomics reveals relationship between pre-slaughter exercise stress, the plasma metabolite profile at time of slaughter, and water-holding capacity in pigs. Meat Science, 84(1), 108–113. Brescia, M. A., Jambrenghi, A. C., Di Martino, V., Sacco, D., Giannico, F., Vonghia, G., et al. (2002). High resolution nuclear magnetic resonance spectroscopy (NMR) studies on meat components: potentialities and prospects. Italian Journal of Animal Science, 1, 151–158. Cheng, Q., & Sun, D. W. (2008). Factors affecting the water holding capacity of red meat products: A review of recent research advances. Critical Reviews in Food Science and Nutrition, 48(2), 137–159. Ciampa, A., Renzi, G., Taglienti, A., Sequi, P., & Valentini, M. (2010). Studies on coffee roasting process by means of nuclear magnetic resonance spectroscopy. Journal of Food Quality, 33(2), 199–211. Consonni, R., Cagliani, L. R., & Cogliati, C. (2011). NMR based geographical characterization of roasted coffee. Talanta, 88, 420–426. Duarte, I. F., Goodfellow, B. J., Gil, A. M., & Delgadillo, I. (2005). Characterization of mango juice by high-resolution NMR, hyphenated NMR, and diffusion-ordered spectroscopy. Spectroscopy Letters, 38(3), 319–342. Graham, S. F., Kennedy, T., Chevallier, O., Gordon, A., Farmer, L., Elliott, C., et al. (2010). The application of NMR to study changes in polar metabolite concentrations in beef longissimus dorsi stored for different periods postmortem. Metabolomics, 6(3), 395–404. Le Gall, G., Colquhoun, I. J., Davis, A. L., Collins, G. J., & Verhoeyen, M. E. (2003). Metabolite profiling of tomato (Lycopersicon esculentum) using H-1 NMR spectroscopy as a tool to detect potential unintended effects following a genetic modification. Journal of Agricultural and Food Chemistry, 51(9), 2447–2456. Maltin, C., Balcerzak, D., Tilley, R., & Delday, M. (2003). Determinants of meat quality: Tenderness. The Proceedings of the Nutrition Society, 62(2), 337–347. Pearson, A. M., & Gillett, T. A. (1999). Processed meats (3rd ed.). Gaithersburg: Aspen Publishers. Perez, E. M. S., Iglesias, M. J., Ortiz, F. L., Perez, I. S., & Galera, M. M. (2010). Study of the suitability of HRMAS NMR for metabolic profiling of tomatoes: Application to tissue differentiation and fruit ripening. Food Chemistry, 122(3), 877–887.

761

Renou, J. P., Bielicki, G., Deponge, C., Gachon, P., Micol, D., & Ritz, P. (2004). Characterization of animal products according to geographic origin and feeding diet using nuclear magnetic resonance and isotope ratio mass spectrometry. Part II: Beef meat. Food Chemistry, 86(2), 251–256. Ritota, M., Marini, F., Sequi, P., & Valentini, M. (2010). Metabolomic characterization of Italian sweet pepper (Capsicum annum L.) by means of HRMAS-NMR spectroscopy and multivariate analysis. Journal of Agricultural and Food Chemistry, 58(17), 9675–9684. Rocha, C. M., Barros, A. S., Gil, A. M., Goodfellow, B. J., Humpfer, E., Spraul, M., et al. (2010). Metabolic profile of human lung cancer tissue by 1H high resolution magic angle spinning (HRMAS) NMR spectroscopy. Journal of Proteome Research, 9(1), 319–332. Sacco, D., Brescia, M. A., Buccolieri, A., & Jambrenghi, A. C. (2005). Geographical origin and breed discrimination of Apulian lamb meat samples by means of analytical and spectroscopic determinations. Meat Science, 71(3), 542–548. Shahidi, F., Rubin, L. J., & Dsouza, L. A. (1986). Meat flavor volatiles — A review of the composition, techniques of analysis, and sensory evaluation. CRC Critical Reviews in Food Science and Nutrition, 24(2), 141–243. Shintu, L., & Caldarelli, S. (2005). High-resolution MAS NMR and chemometrics: Characterization of the ripening of Parmigiano Reggiano cheese. Journal of Agricultural and Food Chemistry, 53(10), 4026–4031. Shintu, L., & Caldarelli, S. (2006). Toward the determination of the geographical origin of emmental(er) cheese via high resolution MAS NMR: A preliminary investigation. Journal of Agricultural and Food Chemistry, 54(12), 4148–4154. Shintu, L., Caldarelli, S., & Franke, B. M. (2007). Pre-selection of potential molecular markers for the geographic origin of dried beef by HR-MAS NMR spectroscopy. Meat Science, 76(4), 700–707. Sobolev, A. P., Brosio, E., Gianferri, R., & Segre, A. L. (2005). Metabolic profile of lettuce leaves by high-field NMR spectra. Magnetic Resonance in Chemistry, 43(8), 625–638. Son, H. S., Kim, K. M., Van den Berg, F., Hwang, G. S., Park, W. M., Lee, C. H., et al. (2008). H-1 nuclear magnetic resonance-based metabolomic characterization of wines by grape varieties and production areas. Journal of Agricultural and Food Chemistry, 56(17), 8007–8016. Spraul, M., Schutz, B., Humpfer, E., Mortter, M., Schafer, H., Koswig, S., et al. (2009). Mixture analysis by NMR as applied to fruit juice quality control. Magnetic Resonance in Chemistry, 47, S130–S137. Straadt, I. K., Young, J. F., Bross, P., Gregersen, N., Oksbjerg, N., Theil, P. K., et al. (2010). NMR-based metabonomic investigation of heat stress in myotubes reveals a timedependent change in the metabolites. Journal of Agricultural and Food Chemistry, 58(10), 6376–6386. Valentini, M., Ritota, M., Cafiero, C., Cozzolino, S., Leita, L., & Sequi, P. (2011). The HRMAS-NMR tool in foodstuff characterisation. Magnetic Resonance in Chemistry, 49, S121–S125. Valentini, M., Vaccaro, A., Rehor, A., Napoli, A., Hubbel, J. A., & Tirelli, N. (2004). Diffusion NMR spectroscopy for the characterization of the size and interactions of colloidal matter: the case of vesicles and nanoparticles. Journal of the American Chemical Society, 126, 2142–2147. Verbeke, W., Van Oeckel, M. J., Warnants, N., Viaene, J., & Boucque, C. V. (1999). Consumer perception, facts and possibilities to improve acceptability of health and sensory characteristics of pork. Meat Science, 53(2), 77–99. Wang, Y., Bollard, M. E., KKeun, H., Antti, H., Beckonert, O., Ebbels, T. M., et al. (2003). Spectral editing and pattern recognition methods applied to high-resolution magic-angle spinning 1H nuclear magnetic resonance spectroscopy of liver tissue. Analytical Biochemistry, 323(1), 26–32. Westerhuis, J. A., Hoefsloot, H. C. J., Smit, S., Vis, D. J., Smilde, A. K., van Velzen, E. J. J., et al. (2008). Assessment of PLSDA cross validation. Metabolomics, 4, 81–89. Wood, J. D., Enser, M., Fisher, A. V., Nute, G. R., Sheard, P. R., Richardson, R. I., et al. (2008). Fat deposition, fatty acid composition and meat quality: A review. Meat Science, 78(4), 343–358.