Genetic correlations of intramuscular fat content and fatty acid composition among muscles and with subcutaneous fat in Duroc pigs 1

There is an increasing interest in including intramuscular fat (IMF) content and fatty acid composition, particularly oleic acid (C18:1) content, in the selection objectives of pig lines for quality pork markets. These traits are costly and can be measured in more than 1 location, so knowing their correlation structure across muscles and with subcutaneous fat (SF) is necessary for developing optimum sampling and recording schemes. We analyzed the genetic and phenotypic correlations of IMF content and composition among 3 of the most relevant muscles (LM, gluteus medius muscle [GM], and semimembranosus muscle [SM]) and with the fatty acid composition of SF. All genetic correlations were positive but variable. For IMF, the genetic correlation between GM and LM was 0.68, and for fatty acids, the genetic correlation ranged from 0.62 for C18:1 to 0.82 for total PUFA. Genetic correlations of GM and LM with SM were much lower: 0.13 to 0.19 for IMF and 0.10 to 0.54 for fatty acids. Correlations for fatty acid composition in SF with GM and LM were moderate to high (0.29–0.53 and 0.43–0.75, respectively) but were null with SM. The expected responses for IMF in the 3 muscles and for C18:1 in each muscle and in SF to selection on records taken from only a single muscle or SF were estimated. Selection for IMF and C18:1 in GM is expected to lead to positive responses in IMF and C18:1 in LM and vice versa, although this can entail genetic lags of 20 to 45% in the muscle not directly selected for. Selection for C18:1 in SF is more effective for C18:1 in LM than in GM and of very limited value for IMF. In conclusion, the genetic correlations of IMF content and fatty acid composition among muscles and with SF, although positive, are variable enough to influence the genetic evaluation scheme for IMF and fat quality. They also indicate that GM and LM can be used alternatively for selection purposes.


INTRODUCTION
Intramuscular fat content (IMF) and fatty acid composition affect both the organoleptic and nutritional properties of pork and its derivatives (Wood et al., 2004).Particularly, oleic acid (C18:1) content has become an appreciated trait because of its association with flavor, technological properties, and health ben-efits (Toldrá, 2002;Christophersen and Haug, 2011;Jiménez-Colmenero et al., 2010).The strong economic importance of dry-cured ham in the Mediterranean area, where hams containing higher levels of C18:1 are premium-paid, together with the increased demand of healthy sources of meat, has triggered the interest of including IMF and fatty acid composition in the breeding goal of the pig lines producing for those markets.Because these traits are difficult and costly to measure, their genetic evaluation is usually based on indirect assessments (Jeremiah, 1998;Newcom et al., 2002Newcom et al., , 2005) ) or on a limited number of records taken either on a single muscle (Ntawubizi et al., 2010;Ros-Freixedes et al., 2012) or from the subcutaneous fat (SF; Fernández et al., 2003;Hofer et al., 2006;Gjerlaug-Enger et al., 2011).However, it is known that the pattern of fatty acid deposition may differ between IMF and SF (Duran-Montgé et al., 2008;Sellier et al., 2010;Bosch et al., 2012), across muscles (Sharma et al., 1987;Leseigneur-Meynier and Gandemer, 1991;Kim et al., 2008), and even among locations within a specific tissue (Faucitano et al., 2004).Therefore, to develop adequate recording and genetic evaluation schemes for IMF and fatty acid composition traits, there is a need to know the correlation structure of these traits across valuable muscles and with SF.The objective of this study is to estimate the genetic correlation of IMF and fatty acid content across 3 economically relevant muscles (the loin and 2 muscles from the ham) and with SF.The expected response for IMF and C18:1 in each muscle and SF to selection on records from only 1 of them is assessed.

MATERIALS AND METHODS
All experimental procedures were approved by the Ethics Committee for Animal Experimentation of the University of Lleida.

Animals and Sample Collection
Data from a purebred Duroc line (Selección Batallé, Riudarenes, Girona, Spain) were used for the analyses (Solanes et al., 2009;Ros-Freixedes et al., 2012).The line was completely closed in 1991 and since then it has been selected for an index including BW, backfat thickness (BT), and IMF.The data set used for the estimation of genetic parameters consisted of 111,305 pedigree-connected pigs, from which 102,915 had at least 1 recorded trait (Table 1).Pigs with records were born from 1996 to 2011.At approximately 75 d of age, pigs were moved to the fattening units, where they were penned by sex (8 to 12 pigs/ pen) until slaughter.All pigs were performance-tested at an average age of 180 d for BW and BT.Backfat thickness was ultrasonically measured at 5 cm off the midline at the position of the last rib (Piglog 105; SFK-Technology, Herlev, Denmark).During the test period, pigs had ad libitum access to commercial diets.Since 2002, 1,204 of the purebred barrows used for producing dry-cured ham were taken for recording IMF and C18:1.These barrows were raised in 15 batches until slaughter at around 210 d of age.From 160 d onward, the barrows were fed a commercial pelleted finishing diet (Esporc, Riudarenes, Girona, Spain) with an average composition of 17.4% CP, 6.01% fiber, and 6.44% fat (C16:0, 20.4%; C18:0, 6.9%; C18:1, 33.9%; and C18:2, 30.1%).At the end of the finishing period, all barrows were slaughtered in the same commercial slaughterhouse at approximately 125 kg of BW.Immediately after slaughter, a sample of SF (n = 333) and semimembranosus muscle (SM; n = 198) was collected.After chilling for about 24 h at 2°C, samples of gluteus medius muscle (GM; n = 1,204) from the left side ham and LM at the level of the third and fourth ribs (n = 318) were also collected.Samples of SF were collected at the same location than either the LM (n = 203) or the GM (n = 130) samples.Samples were immediately vacuum packaged and stored at -20°C until required for IMF and C18:1 determination.A summary of the population characteristics and number of records, sires, dams, and litters used for each analyzed trait is given in Table 1.

Fat Analysis
After muscle samples were completely defrosted and vacuum drip losses were eliminated, the dissected muscle, trimmed of subcutaneous and intermuscular fat, was minced.A representative aliquot from the pulverized freeze-dried muscle was used for fat analysis.Intramuscular fat content and composition was determined in duplicate by quantitative determination of the individual fatty acid by gas chromatography (Bosch et al., 2009).Fatty acid methyl esters were directly obtained by transesterification using a solution of 20% boron trifluoride in methanol (Rule, 1997).Methyl esters were determined by gas chromatography using a capillary column SP2330 (30 m × 0.25 mm; Supelco, Bellefonte, PA) and a flame ionization detector with helium as carrier gas.
Runs were made with a constant column-head pressure of 172 kPa.The oven temperature program increased from 150 to 225°C at 7°C per min and injector and detector temperatures were both 250°C.The quantification was performed through area normalization after adding into each sample 1,2,3-tripentadecanoylglycerol as internal standard.Intramuscular fat content was calculated as the sum of each individual fatty acid expressed as triglyceride equivalents (AOAC, 1997).Total SFA, MUFA, and PUFA fatty acid content as well as individual C18:1 were expressed as their percentage relative to total fatty acids in IMF.Fatty acids were identified by comparing their relative retention times with those of the external standard and confirmed by comparing their mass spectra to the computer library of the gas-liquid chromatography-mass spectrometry database Wiley 275 K and NBS 75 K (Agilent Technologies, Wilmington, DE).Fatty acids were analyzed on a simple quadrupole instrument (GC/MSD 6890N-5973N; Agilent Technologies, Wilmington, DE) equipped with an electron ionization source using the same temperature program as described above.Scanned mass range of fatty acid was 35 to 450 m/z and the scanning rate was 3.46 scans/s.Fatty acid profiles of SF were analyzed following the same procedure.Means and SD by tissue are shown in Table 1.

Estimation of Genetic Parameters
Genetic parameters for IMF and C18:1 in GM, LM, SM, and SF were estimated fitting 4-trait multivariate models, where BW and BT were the 2 first traits and IMF or fatty acids content in 2 different tissues the other 2. In matrix notation, the model was in which y i is the vector of observations for trait i; b i , a i , c i , and e i are the vectors of systematic, additive genetic, litter, and residual effects, respectively; and X i , Z i , and W i are the known incidence matrices that relate b i , a i , and c i with y i , respectively.Systematic effects for BW and BT were the batch (1,226 levels), gender (3 levels: males, females, and castrates), and age at measurement as a covariate.The model for IMF and fatty acids content included only the batch (15 levels) and age at measurement.Because there were only 1.2 piglets/litter with records on IMF and fatty acids content in LM, SM, and SF, litter was dropped from the model for these 2 traits.Genetic correlations between IMF and C18:1 in different tissues were estimated fitting 6-trait (or 5-trait) multivariate models including, besides BW and BT, IMF and C18:1 in 2 different tissues (only IMF in 1 muscle if the other tissue was SF).The genetic parameters were estimated in a Bayesian framework using Gibbs sampling with the TM software (Legarra et al., 2011).Observed phenotypes and missing records imputed by data augmentation were assumed to be conditionally normally distributed as follows: , , in which R was the variance or covariance matrix.Sorting records by trait and pig within trait, R could be written as R 0 Ä I, with R 0 being the n × n residual variance or covariance matrix between the n traits analyzed and I being an identity matrix of appropriate order.Flat priors were used for b i and residual variance or covariance components.
Additive genetic and litter values, conditional on the associated variance or covariance components, were both assumed multivariate normally distributed with mean 0 and with variance or covariance G Ä A and C Ä I, respectively, in which A was the numerator relationship matrix, G was the n × n genetic relationship matrix between the n traits, and C was the 2 × 2 variance or covariance matrix between litter effects of BW and BT.The matrix A was calculated using all the pedigree information.Flat priors were used for additive and litter variance or covariance components.Statistical inferences (means and highest posterior density intervals at 95% of probability [HPD95]) were derived from the samples of the marginal posterior distribution using a unique chain of 1,000,000 iterations, where the first 500,000 were discarded and 1 sample out of 100 iterations retained.Statistics of marginal posterior distributions and the convergence diagnostics were obtained using the BOA package (Smith, 2005).Convergence was tested using the Z-criterion of Geweke (Geweke, 1992) and visual inspection of convergence plots.

Prediction of Expected Responses
The expected genetic responses for IMF and C18:1 were evaluated in a simulated breeding program based on records on either IMF or C18:1 or on both simultaneously, taken from a particular tissue.For a given scenario, we assumed that records from only 1 of the tissues were available.Intramuscular fat and C18:1 were assumed to have the same economic weight when both traits were included in the selection objective.The simulated breeding program was a simplified version of that described in Ros-Freixedes et al. (2012).A population of 40 boars and 400 sows randomly mated was maintained on discrete generations.We assumed that 3 individuals per sire family were slaughtered to determine IMF or C18:1 or both.In each generation, 25% of males and 50% of females were selected based on 3 half-sib plus pedigree records.Selection response was predicted deterministically by using the program SelAction (Rutten et al., 2002).The program accounts for reduction in variance due to selection (Bulmer, 1971) and corrects selection intensities for finite population size and for the correlation between index values of family members (Meuwissen, 1991).

RESULTS
The posterior mean of the genetic variance and the posterior mean and HPD95 of the heritability of IMF in GM, LM, and SM as well as of the genetic correlations among them and with BT are shown in Table 2.The corresponding posterior means and HPD95 for fatty acid composition in GM, LM, SM, and SF are given in Table 3.The heritability of IMF in the 3 muscles was high, particularly for LM.Although they had wide HPD95 (due to the low number of pigs with data on these traits), all of them showed 95% probability of being greater than 0.30.The heritabilities of C18:1, SFA, MUFA, and PUFA in the 3 muscles were of similar magnitude than those for IMF.In GM and LM, the heritability estimates for SFA were the lowest and those for PUFA were the highest.The heritabilities estimated in SF tended to be lower than those estimated in the muscles for all fatty acids.The genetic variance of fatty acids was much higher in SM than in GM, LM, and SF.
The genetic correlation between IMF in GM and LM was high (0.68), but it decreased to approximately 0.15 for that between them and SM.Unlike for GM and LM, the HPD95 for the genetic correlation between IMF in SM and IMF in GM and LM included null and negative values, thereby indicating very little evidence of correlation between them.Similarly, BT was positively correlated to IMF in GM and LM (approximately 0.40) but uncorrelated to IMF in SM.The phenotypic correlations showed the same trends but were lower in magnitude than the genetic correlations.For all fatty acid traits, the highest genetic correlations were also found between GM and LM (0.62 to 0.82).The genetic correlations of GM and LM with SM were also positive but more moderate (0.29 to 0.44 and 0.10 to 0.54, respectively).However, the genetic correlations of LM with SF were consistently higher (0.43 to 0.75) than those of GM with SF (0.29 to 0.53).No evidence of genetic correlation between SM and SF was found.
The genetic parameters for C18:1 adjusted for IMF are shown in Table 4. Adjusted estimates did not relevantly differ from the unadjusted estimates reported in Table 2. Including IMF of the involved muscles as covariates only slightly decreased the correlations among muscles but increased those between muscles and SF.Including IMF as additional traits in the multivariate model did not have any systematic effect on the genetic parameters.
The posterior mean and HPD95 of the genetic correlations of C18:1 in GM, LM, SM, and SF with IMF in the 3 muscles are given in Table 5.The IMF content of GM and LM were moderately correlated with the C18:1 content in the same muscles (0.47-0.52), except for IMF in GM with C18:1 in LM (0.24).The genetic correlations between C18:1 and IMF were much lower when SM was involved (ranging from 0.14 to 0.37), although C18:1 and IMF in SM were highly correlated (0.69).The IMF content in any of the 3 muscles was uncorrelated with C18:1 in SF.
The expected responses for IMF and C18:1 in the 3 sampled muscles and SF to selection on records from different tissues are shown in Table 6.The correlated response in IMF (or C18:1) in GM to selection for the same trait in LM, and vice versa, was 0.6 to 0.7 times the direct response obtained in the sampled muscle.For GM and LM, selection for C18:1 (or IMF) led to a correlated 1 Mean of the posterior density and, in parentheses, highest posterior density interval at 95% of probability.

DISCUSSION
Three economically relevant muscles were considered in this study, 2 of them located in the ham (GM and SM) and 1 in the loin (LM).Sampling of central LM for chemical analysis is laborious and depreciates the loin as a primal cut.Instead, a big sample of GM can be easily obtained on the cutting line from the superior edge of the ham at no cost.Because of this, GM has been frequently used as the reference muscle in studies conducted under field conditions (Casellas et al., 2010;Ros-Freixedes et al., 2012, 2013).It is also feasible to sample SM from its exposed surface at no cost, but this sampling scheme has the limitation that it allows obtaining only small off-line samples.Since SF samples are much easier to obtain than muscle samples, SF has been often used as the reference tissue where to determine the fatty acid profile, for both research and genetic evaluation purposes (Fernández et al., 2003;Hofer et al., 2006;Gjerlaug-Enger et al., 2011).Although alternative non-destructive methods can be used in substitution of chemical determinations, such as near-infrared technology (González-Martín et al., 2002, 2005), the nature of the problem still persists and the correlation structure between target and measured muscles for IMF content and fatty acid composition still needs to be known.The present study investigates the genetic implications of using alternative muscles or SF for phenotyping IMF and fatty acid composition in pigs.
The estimates of the heritability were slightly higher than those previously reported for IMF, C18:1, MUFA, and PUFA (Suzuki et al., 2006;Casellas et al., 2010;Sellier et al., 2010) but similar for SFA.Among muscles, GM and LM showed high correlations between them for IMF and fatty acid content but not with SM, the correlations of which were much lower, particularly for IMF and SFA.An explanation for this result is that SM is subjected to greater sampling errors.To avoid depreciation of the ham, SM was sampled by cutting a small slice from the exposed surface of the carcass at the slaughterhouse.In contrast, a much bigger sample of GM and LM was obtained from the ham and the loin retail cuts, respectively.As a result, samples from GM and LM are likely more representative of the whole muscle than the small slices of SM.This result would confirm that sampling can be a critical factor for an adequate interpretation of the correlations across muscles (Bosch et al., 2009).
The genetic correlations of fatty acid content between GM and LM were higher than those between them and SF, in line with the results of Cánovas et al. (2009), who found different expression patterns between IMF and SF.The only exception was the correlation between SFA in LM and SF.In general, the correlations of fatty acid composition between LM and SF were higher than those between GM or SM and SF.This can be attributed to the fact that SF samples were mostly collected at the same anatomical location as LM samples, thereby suggesting that SF composition correlates better to the IMF composition of an adjacent muscle.In line with this, the remaining SF samples were taken at the same location as GM samples, and consequently, SF showed a higher correlation with GM than with SM.Note, however, that, due to the low number of samples at each location, genetic parameters for SF are based on pooled estimates at both locations.An additional source of sampling error may be incurred by sampling SF across fat layers.Although it is known that fatty acid composition differs between SF layers, its effect on the estimates of genetic parameters is likely small.For the main fatty acids, Suzuki et al. (2006) found that the correlation between the inner and outer SF layers was very high, from 0.84 to 0.96.The correlation structure of IMF fat content and composition with BT and SF composition has practical implications.On one hand, it indicates that there is room for improving IMF content independently from overall fatness (Tribout et al., 2004;Solanes et al., 2009;Ros-Freixedes et al., 2013) but, on the other hand, that measuring fatty acid content in SF can be a good criterion for improving IMF traits only in certain retail cuts.Therefore, regarding C18:1, SF (as measured in this study at the level of the third and fourth ribs) could be a good criterion for loin but not for ham.
The IMF content is known to affect fatty acid composition, being positively related to SFA and MUFA and negatively related to PUFA (Wood et al., 2008;Ros-Freixedes and Estany, 2014).Using C18:1 as an example, genetic parameters were adjusted for IMF of the involved muscles, including them either as covariates in the respective models or as additional traits in a multivariate approach.In general, the estimates based on variance or covariance adjusted for IMF as a covariate were lower than those obtained when adding IMF as additional traits.Although the interpretation of this result is not straightforward, what is important here is that the differences of both approaches with the unadjusted estimates are minor, particularly in terms of HPD95.
Results in the literature regarding the correlation of IMF and fatty acid composition among tissues are scarce but in line with those obtained here.Rauw et al. (2012) reported a phenotypic correlation of IMF between GM and LM higher than ours (0.69), but in contrast, for the correlation among the main fatty acids, their estimates were below our lower HPD95 limit, with values below 0.38.A genetic correlation of 0.65 between IMF in GM and LM and much lower ones with BT (0.36-0.38) were found by Hernández-Sánchez et al. ( 2013) using genomic marker information.These estimates were similar to ours.The phenotypic correlations reported by Yang et al. (2010) between LM and SF in a White Duroc × Erhualian cross were in the same range of values than ours (their values were included in our HPD95), with the exception of SFA, which were lower.Cameron and Enser (1991) reported much lower values for C18:1 (0.19) but more moderate for the main SFA and PUFA (0.31-0.54), using data from Duroc and Landrace.These latter results are in contrast with those obtained by Suzuki et al. (2006) in Duroc for the genetic correlation of MUFA and SFA between LM and SF (approximately 0.70).For PUFA, this genetic correlation was as low as approximately 0.18.Although part of the discrepancies among estimates may be explained by the age of the pigs, much younger in Cameron and Enser (1991) compared to other works, and part by the relatively high SE associated to them, they provide sufficient evidence indicating that the pattern of fat deposition can differ widely across muscles and fat tissues.
Low correlations between muscles have also been found for other meat quality traits.Huff-Lonergan et al. (2002), in Large White, reported phenotypic correlations of 0.47 and 0.30 between LM and SM for pH and color (relative lightness) at 24 h postmortem, respectively.Similarly, Gjerlaug-Enger et al. (2010) reported high genetic correlations (approximately 0.8) between ultimate pH in GM and LM, both in Landrace and Duroc, but the estimates between these muscles and gluteus profundus were only in the range of 0.10 to 0.55.The phenotypic correlations among these 3 muscles did not exceed 0.5.As in our study, correlations were positive but moderate in magnitude.
It has been shown that there is room for improving IMF and fatty acid composition of pork through genetic selection (Ros-Freixedes et al., 2012).This involves setting up a feasible routine of recording these data on a commercial basis.The definition of an optimum design for such schemes requires knowing the correlation structure of IMF and fatty acid composition among target and sampled tissues.One of the main costs of sampling is the depreciation cost, which is likely to occur if measures are taken from the inner side of a high value retail cut such as loin.For its sampling simplicity, an alternative is to sample a portion of GM from the superior edge of the hams.Although it implies an opportunity cost with respect to LM, the target muscle, our results indicate that selection based on GM still leads to acceptable genetic gains in LM, both for IMF and C18:1.In some cases, however, selecting for C18:1 in SF can be a good criterion to increase C18:1 in LM without increasing IMF, at least if SF is taken at the same location as LM.However, in general, C18:1 in SF is of very limited value for improving IMF 1 In each generation 25% of males and 50% of females were selected based on 3 half-sib plus pedigree records.
3 Genetic SD units (× 100). 4Same economic weights for both traits in the selection objective.
or its fatty acid composition.A full description of the consequences of alternative selection and sampling schemes must take into account both the economic value of each muscle and its relative proportion in the carcass as well as the genetic variation of IMF and fatty acid composition traits within each of them (Faucitano et al., 2004).
In conclusion, the genetic correlations of IMF and fatty acid composition across muscles and fat tissues, although positive, are variable enough to influence the genetic evaluation schemes for IMF and fat quality.The results obtained indicate that, in terms of genetic response, GM and LM can be used alternatively as the reference muscle for selection purposes.Moreover, they also reveal that using fatty acid composition of SF as selection criterion should cause more changes in LM than in GM but not in IMF.

Table 1 .
Description of the data set used in the analyses

Table 4 .
Heritability (diagonal, in bold), genetic correlations (above diagonal), and phenotypic correlations (below diagonal) for muscular oleic acid (C18:1) content adjusted for intramuscular fat (IMF) content and C18:1 in subcutaneous fat (SF).Adjustment for IMF was performed either adding IMF of the corresponding muscles as covariates in the model for C18:1 or as additional traits in a multivariate analysis 1

Table 6 .
Direct (bold)and correlated (not bold) expected genetic response for intramuscular fat (IMF) and oleic acid (C18:1) content in a given tissue to selection on records taken on different muscles or subcutaneous fat (SF) 1