Chemical-instrumental-sensory parameters and chemometrics as tools to discriminate among the quality categories of dry-cured Iberian shoulder

La combinación de determinaciones físico­químicas y sensoriales con herramientas quimiométricas han sido utili­ zadas para identificar la calidad de paletas Ibéricas curadas. Dependiendo del grado de confinamiento y la alimentación, tres clases de paletas Ibéricas curadas fueron analizadas: i) a partir de cerdos criados en régimen intensividad y alimen­ tados con piensos comerciales (clase CON), ii) a partir de cerdos criados en régimen de extensividad y alimentados con piensos comerciales (clase ExT) y iii) a partir de cerdos criados en régimen de extensividad y alimentados con recur­ sos naturales (hierba y bellota) (clase MON, montanera). El análisis de componentes principales (PCA) permitió diferen­ ciar entre tres tipos de paletas Ibéricas curadas de acuerdo con su calidad. El modelo multivariante clasificatorio, Suave Modelado Independiente de Clases Analógicas (SIMCA), permitió caracterizar la calidad de paletas curadas cuya cla­ se se desconoce. Finalmente, las representaciones gráficas de Cooman ́s y Si.vs.Hi permitieron la clasificación de la cali­ dad de paletas Ibéricas curadas de calidad desconocida.


INTRODUCTION
Iberian drycured meat products, like shoulders or hams have a high acceptance among consumers due to their particular sensory qualities.Recently, consumer demand for natural and traditional foods, such as meat products from Iberian pigs, has risen due to the effects of increasing concern for the environment, protection of the ecosystem, animal welfare, food safety and food quality properties (Ruiz et al., 2002, Ventanas et al., 2005, Ventanas, 2006).The Iberian pig is a breed of great economic importance in Spain and Portugal.According to the Spanish regulation of quality for Iberian meat and meat products, Iberian pig products should be derived from: i) crossed Iberian (Ib) x Duroc (Du), pigs normally reared in a confinement regime (CON) and fattened with commercial feed, and ii) pure Iberian, usually pigs reared in an extensive regime ("Montanera") (MON), in which traditionally pigs are fed according to the freereared system during the final fattening period in an expanse of variable land area, using natural resources, grass (Quercus ilex) and acorns (Quercus suber), (García et al., 1991).Nevertheless, other pigs, normally pure Iberian, are fattened in a freerange system with commercial feeds, so that pigs may eat grass also, this feeding regime is known as "Extensive regime" (ExT) or campo.The Iberian cured products obtained from freerange pigs have gained wide consumer acceptance and have a high commercial value by virtue of their characteristic sensory quality.Moreover, these products can be considered healthy foods according to the new European regulation of "Healthy declarations" (CE regulation nº 432/2012) because of their high content in unsaturated fats and vitamins (B group), iron, selenium, etc. (JiménezColmenero et al., 2010).The quality of drycured products, like ham

Chemical-instrumental-sensory parameters and chemometrics as tools to discriminate among the quality categories of dry-cured Iberian shoulder
By A. Silva a, *, R. Reina a , J. García-Casco b and J. Ventanas a a Animal Source Foodstuffs Innovation Service (SiPA), University of Extremadura, Avda.Universidad s/n, 10003 Cáceres b INIA.Instituto Nacional de Investigación y Tecnología Agraria y Alimentaria, Zafra (Badajoz) A. SILVA, R. REINA, J. GARCÍACASCO AND J. VENTANAS end of the rearing period, as previously mentioned in the introduction.The CON feeding pigs were 50% Iberian and the other pigs of ExT and MON were of pure Iberian.The provision of grass in the ExT group was quite high.The grass and acorns in the MON group were high enough for the MON regime pigs to gain 5@.The shoulders were cured in the Señorio de Montanera (Badajoz, Spain) for 14, 17 and 19 months for the CON, ExT and MON regimes, respectively.Other information, which is important for the interpretation of the results, has not been provided.The shoulders were removed from carcasses and kept under refrigeration before being subjected to the ripening process.

Sample treatment
Conformational measurements were carried out in each product: length, perimeter and width.Next, with an electric saw, two transverse cuts to each piece were made according to Reina et al. (2012).Random slices from the main muscles (Brachiocephalicus) were used for the sensory analysis.The rest of the sample was minced and homogenized, without subcutaneous or inter muscular fat, to carry out the rest of the analysis.

Experimental design
The study was carried out with two sets (calibration set and test set) of drycured shoulders.The calibration set was used to build the general PCA and PCA models for the SIMCA analysis and constituted 30 drycured shoulder from the three batches (n=10) of pigs as previously described.
At the same time, a test set was used to validate the study and constituted 15 samples of unknown quality drycured shoulders CON0 (2) and CON1; CON quality drycured shoulders from the same and from the previous year calibration samples, respectively were chosen.ExT0 (2), ExT1 (2) and ExT2 (2): ExT quality drycured shoulder from extensive rearing pigs fed with normal, special 1 and special 2 feeding systems, respectively were also used.The special feeds were probably enriched with high oleic oils.Finally, MON0 (2), MON1 (2) and MON2 (2); MON quality drycured shoulder were selected from the same (MON0) and from the previous year (MON1 and MON2, both from different MON) as the calibration samples.

Physico-chemical analysis
Before sectioning the drycured shoulder, they were morphologically characterized by length, width and perimeter measurements.All measurements were made in triplicate.
The physicochemical analyses were divided in two sections: and shoulder, depends directly on genetics (Ib x Du or pure Iberian) and on feeding and rearing regimes: CON, ExT, MON.However, there are other processing factors that affect final product quality, like salting time, and thermohygrometric conditions of the curing process (Ventanas et al., 2007).Analytical methods capable of discriminating among different rearing systems are needed to avoid fraudulent practices in the market.In Spain, the official method for discriminating among feeding and rearing regimes was established based on four fatty acids (FAs) of the subcutaneous fat from pigs (BOE, 2004).However, in recent years, new formulation feed enriched with high contents of oleic acid, mainly from sunflower, has been produced in order to mimic the fatty acid profile of pigs reared in the Montanera regime.In the order to avoid fraudulent practices at the market and mislead the final consumer, the Spanish Ministry of Agriculture had supported a project to look for an alternative analytical method to the fatty acid methods.Some methods based on neofitadiene analysis, triacylglycerol analysis, near infrared spectrometry (NIR), tocopherol analysis and chemicalsensor had been studied in the fat or muscle of Iberian pigs [INIA project RTA 200800026].The same authors had studied different analytical techniques with chemometric tools to classify feeding regimes (Alonso et al., 2008, NarváezRivas et al., 2010, Gallardo et al., 2012) or food and beverages authentication (Jurado et al., 2007, GaleanoDíaz, et al., 2005) In the framework of the INIA project, the chemical composition and the sensory quality of drycured shoulder from Iberian pigs with three combinations of genetics and feeding and rearing regimes, which are found in the practice (IbxDu with CON regime, pure Iberian with ExT and MON regimes) were evaluated.The evaluation of the quality of these products is based on composition analysis using classic techniques and instrumental analysis (protein, fat, sodium chloride percentage, myoglobin, oxidation index, moisture, intramuscular fat, fatty acids and volatile compounds profile) and sensory analysis.This quality study had been made according to previous studies (Ventanas et al., 2007;Reina et al., 2012) The aim of this work is to combine analytical parameters (physicochemical and sensory parameters), related to the drycured shoulder quality, to create a data matrix and with chemometric tools, such as principal component analysis (PCA) and soft independent modelling of class analogy (SIMCA) providing the classification of known dry cured shoulder quality.

Animal and samples
The pigs used in this study were divided into three batches according to the feeding system at the CHEMICALINSTRUMENTALSENSORY PARAMETERS AND CHEMOMETRICS AS TOOLS TO DISCRIMINATE AMONG… mean of all panelist scores for each attribute of each drycured shoulder was calculated to perform the statistical analysis.

Data Analysis
Treatment of anomalous data was carried out using the Grubbs test (Grubbs, 1969), as recommended by ISO rules.Data obtained from each analysis of the 30 samples were used as variables and the statistical analysis was carried out by ANOVA analysis of Variance using the statistical software package SPSS Version 15.0 (SPSS, 2006).The data were then subjected to a principal component analysis (PCA), which is a data compression method to reduce the dimensionality of the original data matrix by constructing Principal Components (PCs) that are linear combinations of the original variables and samples.The first PCs collected almost all of the variability in the original data.The supervised pattern recognition method with soft independent modelling of class analogy (SIMCA) based on the description of individual categories by means of the principal component analysis independent of mathematical models was used for the classification of samples of the unknown quality drycured shoulder.The statistical package UNSCRAMBLER (Unscrambler software, CAMO) was used for the application of PCA and SIMCA.

RESULTS AND DISCUSSIONS
In this study, several variables were measured and were used to build the PCA models and the SIMCA analysis: conformational parameters, such as length, width and perimeter, compositional and physicochemical parameters, such as moisture, protein, intramuscular fat, sodium chloride, myoglobin content, TBAs, fatty acid composition, volatile compounds and sensory features.Fatty acid composition, volatile compounds and sensory features are shown in tables 1, 2 and 3, respectively.The measurement of many variables in different samples provided a large amount of data, so that a great deal of information was generated which is not easy to interpret.Fortunately, there are several tools that allow us to interpret this data, such as PCA, a multivariate technique that provides a determination of which aspect one sample is different from another.
The principle of PCA is finding the linear combinations of the initial variables that most contribute to making samples different from each other.These combinations are called principal components (PCs).Normally, the first PC carries the most information (more explained variance) and the second PC carries the maximum share of the residual information.The number of PCs is estimated by crossvalidation, where variables that have a large standard deviation (SD) are weighted by the software and the PCA calculation 1. Proximate Composition: Moisture and salt contents were determined according to the Association of Official Analytical Chemists (AOAC, 2000) (moisture reference 935.29; salt content reference 971.19).Protein content was determined by the Kjeldahl method (AOAC, 2000).Intramuscular total lipids were extracted and quantified with chloroform/methanol (2:1, v/v) according to the method described by Folch et al., (1957).Haem pigments were assessed following the method described by Hornsey (1956).
2. Instrumental analysis: Lipid oxidation was analyzed using the 2thiobarbituric acid (TBA) method of (Salih et al.,1987) using 2.5 g samples of drycured shoulder.TBAreactive substance (TBARS) values were calculated from a standard curve of malondialdehyde (MDA) and expressed as mg MDA kg -1 meat.The fatty acid methyl esters (FAMEs) of the intramuscular fat were obtained by acidic transesterification following the method described by (Martin et al., 2008).Results were expressed as mg per 100 g of sample and as a percentage of each fatty acid relative to total fatty acids.Volatile compounds were extracted using the solidphase micro extraction (SPME) (Supelco Bellefonte, PA) fiber coated with carboxenpoly (dimethylsiloxane) (75 µm thickness) and subsequently analyzed by gas chromatography coupled to mass spectrometry (GC/MS) (gas chromatograph HewlettPackard 6890 series II coupled to a mass selective detector HewlettPackard HP5973 A) following the method described by Jurado et al., (2009)

Sensory analysis
The samples were assessed by a trained panel of 12 members, using a quantitativedescriptive analysis method (QDA) (García et al., 1996) for seven different attributes (brightness, marbling, odor intensity, oiliness, flavor intensity, cured flavor, and rancid flavor).
mentioned above, the number of variables is large, so we have studied whether there are statistically significant differences among variables for the three batches.The absence of significant variables is not considered in the PCA, therefore some significant variables are grouped and related to each other to reduce the number of initial inputs.Therefore, the variables considered in the PCA were: myoglobin, oleic acid (C18:1), fatty acid n3, saturated fatty acids (SFA), monounsaturated fatty acids (MUFA),  , means with different superscript differ significantly A. SILVA, R. REINA, J. GARCÍACASCO AND J. VENTANAS the role of that variable to discriminate among different class models.On the other hand, there are sample results called "Si" (square root of the residual variance of the sample) which is the measurement of the distance of a sample to the modeled group, sometimes it is compared to the overall variation of the class (S0) and is the basis of the statistical criteria to decide the new sample belonging.The Hi (leverage) expresses how a difference in the sample can be considered from the other class members.The use of the graphical plots Si vs. Hi or Comman´s plot (Si vs. Si) samples can easily be classified.Finally, the "model distance" shows how different two models are from each other, normally a model distance larger than 3 or more, shows that two models are quite different.
In our case, a PCA model for each class or group was built with the same considered variables as in the previously mentioned general PCA in order to generate a SIMCA model.This SIMCA had a model distance larger than 3 for the three models, which means a good separation between models.Based on modeling power and discrimination power the most important variables were: myoglobin, C18:1, fatty acids n3, AA, LIP, AA/LIP, hexanal and brightness and oiliness.
The SIMCA analysis was validated with the test set.The samples of test set were classified according to the drycured shoulders of MON, ExT and CON classes.In Figure 2A, the Cooman´s plot of the CON and MON models are shown.The CON samples are exactly classified, except CON1 which is one sample from a previous year.In the case of MON samples, all are exactly classified at 80%.In Figure 2B, the Cooman´s plot to CON and ExT models are shown.The classification of CON is the same as in the previous case and the classification of the ExT model is accurate for normal feeds, the special feeds are classified as belonging to any In Figure 1, a Biplot PCA model is shown.In this plot, loadings and scores are plotted on the same graph.Three groups are clearly separated by quality category or class.The diagram area of the MON class is characterized by the variables: ratio AA/LIP, the peak area of 3methylbutanal, % MUFA and % C18:1, recognized in these high quality Iberian products (Ventanas, S. et al. 2007).The area of the ExT class is characterized by the variables: content of myoglobin, % of PUFA and % of fatty acid n3 (which are typical of the grass diet in pigs feeding in an extensive regime).Finally, the area of the CON class is characterized by SFA, LIP and hexanal in decreasing order of quality from MON to ExT to CON.
Once the analysis of PCA was interpreted, supervisedlearning pattern recognition techniques were applied, such as SIMCA, to find an operative classification role for sample discrimination.

Soft independent modeling class analogy (SIMCA)
The SIMCA is based on the evaluation of the principal components of each category, setting up a critical distance with probabilistic meaning and the calculation of the distance of each object from the model of each group.There are two steps involved in classification: modeling, building one PCA separate model for each class; classifying new samples, where unknown samples are then compared to the class models and if each sample fits each model, it is then determined whether the sample belongs to the corresponding class.
In SIMCA analysis there are variable results called "modelling power" (MP), which means the contribution of that variable to the class or group model; "discrimination model" (DM), which means classical and instrumental analysis, and sensory analysis provided a large amount of information.This information was interpreted by multivariate techniques, such as PCA, and three quality classes were clearly generated, considering only the most important variables, those that had statistically significant differences among the three groups.Therefore, a first SIMCA analysis based on a PCA model for three different quality categories had been developed to accurately classify dry cured Iberian shoulders from an unknown quality class, over all the samples at the same year than calibration samples.However, drycured shoulder from pigs reared in extensive regime and fed with special feeds were not classified in any quality class and these samples can be considered a new quality category, which was discriminated as a special category (SPE class).Further data on dry cured shoulders will be used to validate the future model with new samples.
Therefore the combinations of physicchemical parameters and chemometrics enable accurate classification of the unknown quality class of dry cured shoulder even when pigs are fed in a way that mimics the fattyacid profile of the montanera feedings pigs.

Figure 1 PCA
Figure 1 PCA Biplot shown the loadings and scores.