Statistical model for classifying the feeding systems of Iberian pigs through Gas Chromatography ( GC-FID ) and Isotope Ratio Mass Spectrometry ( GCC-IRMS )

En el presente trabajo se han analizado un total de 734 muestras de tejido subcutáneo de cerdos ibéricos con distin­ tos tipos de alimentación de engorde (Bellota, Recebo, Cebo y Campo) a lo largo de tres años consecutivos, 2009­2011. Se han extraído los lípidos de la grasa subcutánea de rabadi­ lla, y después de su esterificación, se han analizado por Cro­ matografía de gases (GC­FID) y por Espectrometría de ma­ sas de relaciones isotópicas (GC­C­IRMS). Las medias de los ácidos grasos y de las relaciones isotópicas muestran que existen diferencias según el año y tipo de alimentación, factores que deberían tenerse en cuenta a la hora de clasifi­ car los animales. La aplicación de distintos modelos de pre­ dicción basados en análisis discriminante permite establecer un método para la clasificación de los animales según el tipo de alimentación, con un porcentaje de aciertos del 85% utili­ zando tres o cuatro categorías de clasificación (Bellota, Re­ cebo, Campo y/o Cebo) y del 91% utilizando sólo dos cate­ gorías, Cebo y Bellota. Este modelo podría sentar las bases para una clasificación adecuada del cerdo ibérico en función de su alimentación.


INTRODUCTION
The analysis of the fatty acid composition of the subcutaneous adipose tissue lipids of Iberian pigs by GCFID, was, until 2005, an effective and objective tool for the classification of carcasses based on the feed received by the animals during the final fattening period (Ruiz and Preton, 2001).In fact, it has been the only officially recognized method (B.O.E., 2004).The use of vegetable fats rich in oleic acid in formulated feed, which achieved a fatty acid profile in the animals' fat similar to animals which have consumed only grass and acorns, questioned the effectiveness of the GCFID to determine the source of fatty acids deposited in adipose tissue and Iberian pork products.The Official Quality standard eliminated this technique for the classification of carcasses (B.O.E., 2007a), but in practice, it is a method that is still used by the industrial sector, because the composition of the intramuscular fat from meat plays a decisive role in the drycured process of the quality parts and determines, for example, the number of days required for the salting and drying of hams (Cava and Andres, 2001).It also influences the consistency, color and fat oxidation, which are decisive factors in the quality of the meat, both fresh and drycured (Melgar et al., 1991;Ruiz et al., 2000;Ventanas et al., 2006;Ventanas et al., 1999;Gilles, 2009).

Gas chromatography-Combustion-Isotope ratio mass spectrometry
The determination of the 13 C/ 12 C isotope ratio (δ 13 C) of the fatty acids palmitic, stearic, oleic and linoleic using the technique GCCIRMS was made following the procedure described by Recio et al. (2010).We used an isotope ratio mass spectrometer with a continuous flow gas source, Hydra 2020® model of SERCON Ltd, equipped with an electromagnet, a combustion interface and a Nafion membrane to retain water from the combustion product.To separate and transfer FAMEs to the spectrometer, a gas chromatograph Agilent 7890A GC System was used, with a capillary column, 30 m × 0.25 mm ID and 0.25 µm thickness, using He as the carrier gas.The injector temperature was 280 °C and the detector was 300 °C.A combustion tube, comprised of a ceramic furnace with a copper oxide and platinum catalyst maintained at 860 °C, is used to obtain CO 2 products from the separated FAMEs.Water is removed from the combustion gases by passing them through a Nafion membrane and then CO 2 products are directed to the mass spectrometer.
Analyses were carried out in sets of 10 unknown samples, with standard samples at the beginning, middle and end of each series.As standards, three commercial FAMEs (Methylhexadecanoate, Methylheptadecanoate and Methylheneicosanoate from SigmaAldrich) were used, which were characterized by elemental analyzers of different national research centers coordinated by the stable isotope laboratory of the University of Salamanca.Additionally, a reference material of Iberian pig subcutaneous fat characterized (known values of isotopic ratios) by the stable isotope laboratory of the University of Salamanca and our own laboratory was used as a control.
The isotopic value obtained is expressed in terms of "δ" which represents the excess, typically heavy isotope, in a sample relative to a gas reference, ‰ units, referred to PDB (Pee Dee Belemnite;al., 1978).However, the use of certain formulated feed in the fattening of pigs can mask the results, so it has resulted in the separation of fatty acid methyl esters (FAMEs) by Gas Chromatography and subsequent combustion and mass spectrometry analysis of the 13 C/ 12 C isotope ratios of the products obtained, which create a characteristic isotope profile of each sample (RecioHernández, 2010).This technique, called GCCIRMS, is currently used for the detection of adulteration in vegetable oils (Spangenberg et al. 1998;Kelly et al., 1997, Kelly andRhodes, 2002) and wine (Regulation EEC, 1990).Although the method proposed for the Iberian pig indicates the determination of at least four isotopic ratios of the major FAMEs in the fat (oleic, palmitic, linoleic and stearic acids), which would be the minimum necessary to provide positive identification information in the case of an Iberian Acornfed pig (Bellota), some industries are currently using only the oleic 13 C/ 12 C isotope ratio for the differentiation of the Iberian pig feeding regime, applying an index which sets limit values for classifying the animals in the Bellota category.
In order to compare the efficacy of these techniques, jointly or separately, samples of subcutaneous adipose tissue from Iberian pigs were analyzed both by Isotope Ratio Mass Spectrometry (GCCIRMS) and by Gas Chromatography (GC FID).A statistical study comparing the results and five different models using Discriminant analysis for the classification of samples has been made, calculating the correct percentage on several assumptions based on the number of feeding categories considered.

Samples
Subcutaneous fat samples were taken on the day of slaughter from a total of 734 animals, belonging to 38 batches of Iberian pigs, reared and fattened in different farms in Extremadura, Andalusia and Salamanca.The feeding regime of these animals during the final fattening phase was known.Their classification, based on the Official method (B.O.E., 2007a), is shown in Table 1.A detailed description of the rearing systems and field information can be found at GarcíaCasco et al. (2013).The subcutaneous tissue sampling of slaughtered animals was performed following the established Official method (B.O.E., 2004), as well as the extraction and esterification of lipids.

Gas Chromatography
Gas Chromatography (GCFID) was carried out following the Official method (B.O.E., 2004).We used two Perkin Elmer chromatographs with autosamplers and a fused silica capillary column (30 m × 0.32 mm internal diameter and 0.25 µm film thickness).The injector temperature was kept at 4FA12I4: percentage of the twelve fatty acids and isotopic ratios of the four major fatty acids.
5FA12I4C: percentage of twelve fatty acids and four major fatty isotopic ratios, differentiating the campaign in the classification criteria.
The results were also compared with the classification of the samples based on the criteria established in the Official Method (B.O.E., 2007b), based on the values of the four major fatty acids (results coded as 0FAQS).

RESULTS AND DISCUSSION
Table 2 shows the mean values and standard deviations obtained from the percentage of the major fatty acids (GCFID) and its δ 13 C isotopic values obtained by GCCIRMS.As shown, the percentage of C16:0 and C18:0 is very similar in all lots of Bellota with no significant differences among the three lots after an ANOVA analysis and, therefore, no significance in the multiple range test (Table 3).There were significant differences among campaigns for the percentages of C18:1 and C18:2 in the ANOVA analysis, with the 2010 campaign mean (dry and scarce acorns) being lower when compared to the other two campaigns (Table 3).However, the values of C18:2 follow the opposite trend, with higher in lowincome campaigns as the 2010 and lower in 2011, which was a rainy year (NarváezRivas et al., 2009), with significant differences between them.The percentages of C16:0 and C18:0 increase progressively in batches of Recebo, Campo and Cebo, while the C18:1 decreases in the same direction.The exception is international reference data δ 13 C).A regression line of the three internal standards analyzed along with the unknown samples was used to normalize the measured values.The following Goodman and Brenna formula (1992) was applied to obtain the FAMEs isotopic value discounting the contribution of the methylating agent: δ R = ((M m × δ m ) (M met x δ met ))/M R , where δ R is the FAME isotope actual value, M m the number of moles of the measured species, δ m the isotopic value measured, M met the number of moles of C in methanol (1), δ met the isotopic value of the methanol used and M R the number of moles of C in the FAME.

Statistical analysis and Models
Statistical analysis of the data was carried out with Statgraphics Centurion xVI.I (2011) to calculate the mean values, standard deviations, ANOVA and Fisher´s LSD multiple range tests with a confidence level of 95%.
A linear discriminant analysis by Statgraphics Centurion xVI.I (2011) was applied for the prediction of the feeding of pigs where the same probability for all groups had been established.We studied a total of five different models of discriminant analysis, based on the values used as the basis of the model: 1FA4: percentage of the four major fatty acids obtained by GCFID 2FA12: percentage of the twelve fatty acids obtained by GCFID.
3I4: 13 C/ 12 C isotope ratios of the four major fatty acids.Statistical models show that the success rate increases as the number of variables included in the discriminant analysis increases, and dramatically reduces false positives in the category of Bellota.Comparing the results by considering only three categories (Bellota, Recebo and Cebo), increasing from 4 to 12 fatty acids (1FA4 models of Table 4 and 2FA12 in Table 5) is an improvement in the prediction, especially in the category of Recebo (from 34% to 53%), with an overall increase in the correct percentage of 8% (from 65% to 73%, Table 7).The results are not very different between the 2FA12 and 3I4 models (discriminant analysis with four isotopic ratios) of Table 5.However, by combining both techniques in the analysis (4FA12I4, Table 6) the correct percentage increases 10 points, reaching 82%, with results in each category of 91% in Cebo/Campo (formulated feed), 66% in Recebo and 83% in Bellota.If environmental factors, such as campaigns, are in the batch Recebo 2011, where the average value is higher in oleic acid (54.77%), close to the values obtained in the category of Bellota.The C18:2 shows great variability among campaigns and categories as it depends on the type of formulated feed used in the previous fattening stages.
For isotopic ratios there are no clear trends regarding the feeding categories since there are significant variations depending on the campaign (Table 3).So the 13 C/ 12 C mean isotope ratio obtained from the FAME of oleic acid, of the Bellota category in the campaigns of 2009 and 2011 are of greater magnitude (-27.1 and -28.1, respectively) while in the 2010 campaign the mean isotope ratio is -25.8‰,similar to that obtained in other categories in different campaigns (eg Recebo andCampo in 2011, Cebo andRecebo in 2010).These values indicate the difficulty of establishing a δ 13 C value that allows for classification based on the feeding system, because Bellota values of a dry campaign like 2010 are of the same order as those of Campo and Recebo of a rainy campaign like 2011 and even that of animals fed exclusively with formulated feed.
It is important to mention the high standard deviation within the same analytical value, feeding system and campaign of the great variety of samples that exist in the study.

Models of discriminant analysis
The prediction of the classification of samples based on the type of feed was carried out by a discriminant analysis with the five models previously described, which are differentiated by the values included in the analysis.The results were compared taken into account in the analysis (5FA12I4C, Table 6), the overall percentage of correct samples increased to 85% (94% in Cebo/Campo, 71% in Recebo and 85% in Bellota).Figure 1 shows the discriminant function model 4FA12I4, where it can be seen that there is a clear separation between   The Recebo category shows a wide range of variability, hence the difficulty for a correct prediction.Depending on the quantity and the quality of the acorns and the harvest, which is different among geographical zones, some animals are classified as Recebo or as Cebo/Campo, depending on the above factors.
The consideration of only two categories significantly increases the level of accuracy in the prediction model 5FA12I4C.
The use of this model complemented with actual field data, would classify pigs according to current regulations and would provide essential information for the correct labelling of products, with the consequent benefits for the consumer.

ACKNOWLEDGEMENTS
Samples that have made this work possible come from an RTA0826 INIA funded project and CC0831 from the Ministerio de Agricultura, Alimentación y Medio Ambiente (Spain).The (6%) are erroneously classified, with a drastic reduction in false positives to 6 samples (2%) in Bellota.In the case of animals from the Recebo category which presents increased difficulty of prediction, 55 animals were incorrectly classified (29%).In 18 samples out of 30 framed in the Cebo/ Campo category the oleic acid percentage was less than 51% and/or the isotopic ratio of oleic acid was higher than -23‰, ie below the normal rate category, which should be reclassified as Cebo/Campo despite qualifying it as Recebo.The false positives of 25 samples classified as Bellota (13%) belong to batches in which the weight gain obtained by pigs during the consumption of acorns and grass was very high, close to that required for the category of Bellota.

CONCLUSIONS
The discriminant analysis using the values of 12 fatty acids obtained by Gas Chromatography and four isotope ratios by CGCIRMS, as well as the inclusion of the environmental factor of each campaign, provide a necessary basis for a correct classification of subcutaneous adipose tissue samples of Iberian pigs according to the type of feed received during fattening.
This paper shows that combining both instrumental methods, GCFID and GCC 5,2 3,2

Figure 1
Scatterplot of 4FA12I4 discriminant analysis model according to the feeding category

Table 2 Mean values of % fatty acids and isotopic ratio by CG y GC-C-IRMS Campaign % by CG-FID a δ 13 C by GC-C-IRMS b
DeLgaDo-Chavero, e. Zapata-MárqueZ, J.M. garCía-CasCo anD a. pareDes-torronteras with the Official Method based on the percentage of the four major fatty acids.Tables 4, 5 and 6 show the prediction made and the percentage of success in each campaign and feeding type.Table7contains a summary of the overall rate of success of each model when considering four, three or two categories.When applied according to the Official classification values (B.O.E., 2007b) for each category, the overall accuracy of the results is 66%, 44% and 76% in the three categories considered (Cebo, Recebo and Bellota, Table4), although not distinguishable between Cebo and Campo.The number of false positives in the Bellota category (misclassification of a lower category in this one) is very high: 63 Recebo animals (33%) and 46 of Cebo/Campo (16%).

Table 3
Significant differences among campaigns in the same fatty acid and analytical technique groups of Bellota, Cebo and Campo, while the Recebo group is a mixture of Bellota and Campo, with a higher number of misclassifications in this category.The inclusion of the environmental factor of each campaign helps to increase the accuracy in the Recebo category.

Table 7 Correct percentage for all models according to the number of categories considered Correct percentage
analytical determinations were performed in the laboratory of the interprofessional ASICI supported by Quality standard extension studies (B.O.E., 2010).The authors thank Dr. Clemente Recio Hernández for his assistance with the implementation of the technical CGCIRMS. CENTROIDS