Journal of Animal Science

Dalvit, C.;De Marchi, M.;Zanetti, E.;Cassandro, M.

doi: 10.2527/jas.2008-1682pmid: 19717776

ABSTRACT The genetic variability and presence of population substructures in 4 native Northern Italian sheep breeds, Alpagota, Brogna, Foza, and Lamon, undergoing in situ conservation, and 1 widespread Italian breed, Bergamasca, were studied by investigating 19 microsatellite markers. The breeds showed considerable genetic variability in terms of number of alleles and heterozygosity, with the exception of Alpagota, which was the least variable (0.607). Nevertheless, a significant deficit of heterozygotes was observed in each breed due to rather increased levels of inbreeding or to the presence of population substructures, probably caused by increased genetic variation in the founder populations. The analyses evidenced clear genetic differentiation (FST = 0.085), reduced levels of admixture, and presence of private alleles among the breeds, confirming their genetic uniqueness. In particular, according to Reynolds genetic distances, Alpagota was the most differentiated, perhaps because it had been bred mostly in a rather isolated area. Loss of any of the investigated breeds would result in a loss of genetic diversity ranging from 0.5 to 1.6% of the total observed gene diversity. Results supported the decision to safeguard these breeds as important reservoirs of genetic diversity and suggested breeding and mating practices to maintain variability and to overcome within-breed substructures. INTRODUCTION In Italy, sheep breeding has always played an important role, mainly in less developed, rural, and mountain areas, where the production systems are interrelated with local traditions and native breeds (Rancourt et al., 2006). In recent decades, the focus on production has remarkably changed sheep management, emphasizing the breeding of a small number of very productive and selected cosmopolitan breeds. This trend has also been observed in cattle breeding and is the consequence of socio-economic changes. In fact, the continuous improvement of crop productivity and the subsequent low price of cereals led farmers to feed animals with grain instead of exploiting pastures and to change the production systems in favor of more productive and highly selected breeds (Mendelsohn, 2003). Therefore, the intensification of production systems and the success of industrial breeding have led farmers to abandon several autochthonous genotypes (Taberlet et al., 2008). In the northeastern part of Italy, the Veneto region, the sheep population decreased drastically from 101,170 animals in 1953 (Montanari, 1954) to 34,734 animals in 1991, and several native breeds disappeared (Pastore and Fabbris, 1999). Nowadays, only 4 native breeds are still reared in Veneto: Alpagota (ALP), Brogna (BRO), Foza (FOZ), and Lamon (LAM). They are mainly found on farms located in mountainous and hilly areas, and due to their good adaptive traits and to their environmental role in exploiting marginal pastures, these breeds represent an important reservoir of genetic resources. The aim of this paper was to study the genetic variability and presence of population substructures in these 4 native breeds using microsatellite markers. In fact, several studies have already indicated the presence of significant inbreeding in many native sheep breeds (Pariset et al., 2003; Rendo et al., 2004; Santos-Silva et al., 2008), suggesting that even domestic breeds belonging to species such cattle, sheep, and goat could be considered endangered (Taberlet et al., 2008). MATERIALS AND METHODS Research protocols followed the guidelines stated in the Guide for the Care and Use of Agricultural Animals in Agricultural Research and Teaching (FASS, 1999). Breed Description The 4 Venetian native breeds (ALP, BRO, FOZ, and LAM) analyzed in the present paper are historically tied to mountain communities of the Veneto Italian region and have always been used for the production of traditional dairy or meat products. The following is a brief description of their main features. The ALP is a small-sized sheep that originated in the Alpago mountains, and nowadays about 1,500 animals are enrolled in the herd book; it is listed in the FAO global databank for farm animal genetic resources where it is classified as endangered. It is a multipurpose breed (meat, milk, and wool), but now it is bred only for meat production due to its purported high quality. For these reasons, ALP lamb is 1 of the 270 presidia in the world recognized by the Slow Food Foundation (2008), and its meat is sold to the best restaurants of the Veneto region. According to Di Stasio (1980), the good adaptive traits to the mountain environment exhibited by ALP are due to the greater frequency of blood group A with respect to other sheep breeds in which the B group is more frequent. The BRO breed has been known since the 12th century and is native to the Lessini mountains, although its origin are uncertain; at present the herd book accounts for about 1,300 animals. It is not present in the FAO databank, but according to the population size, it can be classified as endangered. The BRO is a multipurpose breed (meat, milk, and wool), and its milk has been traditionally transformed to produce typical cheeses. The animals are medium to small in size and are very prolific, with a twinning rate of 58% as reported by Pastore and Fabbris (1999). The first evidence of the existence of the LAM breed traces to the 19th century; it was described as a rustic breed with low mortality rates and was the primary economic resource of local farmers (Pastore, 2002). Today, about 300 animals are enrolled in the herd book, and according to the FAO global databank for farm animal genetic resources, it is classified as endangered. The LAM sheep are well adapted to adverse conditions and are maintained in a semi-wild state without the use of any housing throughout the year. Its traditional management provided for flock migration from hill to plain in winter, from plain to hill in spring, and from hill to mountains in summer (Pastore, 2002). It is a multipurpose breed (meat, milk, and wool), but nowadays is kept only for meat, in particular for the production of a typical smoked meat. The FOZ breed origins are uncertain, but it is native to the Asiago Mountains; nowadays the situation of FOZ is very critical because its population size is estimated at only about 100 animals. The FOZ breed was traditionally reared for the good quality of its wool and was typically reared in small herds either transhumance or farm flock (Pastore, 2002). Bergamasca (BER) is a native breed from the Lombardy region, which is widespread mainly in northern Italy. This breed specializes in meat production and is widely used also in crossing with other breeds. In the present work, it was used as a reference population. Animal Sampling A total of 170 individual blood samples were collected from the 5 sheep breeds involved in the study: ALP (n = 38), BRO (n = 38), FOZ (n = 33), LAM (n = 16), and BER (n = 45). Samples from ALP, BRO, FOZ, and LAM were collected on the experimental farm where they are maintained according to an in situ conservation program. All of the animals present in the conservation nucleus were sampled to estimate the degree of variability and the population structure of the conserved animals at the beginning of the conservation activities. The conservation scheme provided for the choice of founder animals according to the breed standards; founders belonged to different farms and were moved to the conservation nucleus on the experimental farm. Blood samples of BER were collected in several flocks and were used as a reference population. Blood samples were collected from each animal in 5-mL vacutainer tubes containing sodium citrate as anticoagulant and stored at −20°C until analyses were performed. The DNA extraction was carried out employing the Gentra System PUREGENE DNA purification kit (Gentra System, Minneapolis, MN) starting from 300 µL of whole blood. Amplification and Genotyping of Microsatellite Markers A panel of 19 microsatellite markers was established according to ISAG/FAO Standing Committee (2004) recommendations and to previous studies (Baumung et al., 2006) to investigate highly polymorphic markers spread throughout the genome. The loci studied were OarAE54, OarFCB20, URB58, McM527, INRA23, TGLA53, MAF65, OarCP49, MAF214, HSC, INRA63, McM42, OarAE119, OarAE129, ILSTS087, OarFCB304, OarCP34, OarCP20, and CSRD247 (Table 1). Details on the amplification protocol and primer annealing temperatures can be found in Dalvit et al. (2008). Allele size was determined with a CEQ 8000 Genetic Analysis System (Beckman Coulter, Fullerton, CA). Microsatellite markers with corresponding fragment size, chromosomal location, total number of detected alleles (TNA), allelic richness (AR), and gene diversity (GD) Table 1. Microsatellite markers with corresponding fragment size, chromosomal location, total number of detected alleles (TNA), allelic richness (AR), and gene diversity (GD) Locus  Fragment size  Chromosome  TNA  AR  GD  CSRD274  213–259  14  22  8.0  0.779  ILSTS87  138–178  6  19  8.8  0.814  OarCP20  68–102  21  15  5.5  0.745  OarCP34  100–128  3  12  6.5  0.788  OarFCB304  150–198  19  20  7.3  0.702  McM527  171–189  5  10  7.0  0.738  OarAE54  122–156  25  14  8.2  0.797  OarFCB20  92–122  2  16  8.9  0.836  URB58  161–209  13  18  8.0  0.808  OarAE129  137–157  5  8  4.7  0.651  OarAE119  147–185  19  13  7.7  0.770  INRA63  162–210  14  21  7.4  0.785  HSC  263–299  20  19  9.3  0.835  McM42  82–106  9  9  5.1  0.634  INRA23  196–224  1  15  9.0  0.834  MAF214  185–261  16  13  4.4  0.538  MAF65  121–143  15  12  7.5  0.821  OarCP49  69–119  17  22  8.1  0.793  TGLA53  140–168  12  15  7.9  0.846  Mean      15.4  7.3  0.764  Locus  Fragment size  Chromosome  TNA  AR  GD  CSRD274  213–259  14  22  8.0  0.779  ILSTS87  138–178  6  19  8.8  0.814  OarCP20  68–102  21  15  5.5  0.745  OarCP34  100–128  3  12  6.5  0.788  OarFCB304  150–198  19  20  7.3  0.702  McM527  171–189  5  10  7.0  0.738  OarAE54  122–156  25  14  8.2  0.797  OarFCB20  92–122  2  16  8.9  0.836  URB58  161–209  13  18  8.0  0.808  OarAE129  137–157  5  8  4.7  0.651  OarAE119  147–185  19  13  7.7  0.770  INRA63  162–210  14  21  7.4  0.785  HSC  263–299  20  19  9.3  0.835  McM42  82–106  9  9  5.1  0.634  INRA23  196–224  1  15  9.0  0.834  MAF214  185–261  16  13  4.4  0.538  MAF65  121–143  15  12  7.5  0.821  OarCP49  69–119  17  22  8.1  0.793  TGLA53  140–168  12  15  7.9  0.846  Mean      15.4  7.3  0.764  View Large Table 1. Microsatellite markers with corresponding fragment size, chromosomal location, total number of detected alleles (TNA), allelic richness (AR), and gene diversity (GD) Locus  Fragment size  Chromosome  TNA  AR  GD  CSRD274  213–259  14  22  8.0  0.779  ILSTS87  138–178  6  19  8.8  0.814  OarCP20  68–102  21  15  5.5  0.745  OarCP34  100–128  3  12  6.5  0.788  OarFCB304  150–198  19  20  7.3  0.702  McM527  171–189  5  10  7.0  0.738  OarAE54  122–156  25  14  8.2  0.797  OarFCB20  92–122  2  16  8.9  0.836  URB58  161–209  13  18  8.0  0.808  OarAE129  137–157  5  8  4.7  0.651  OarAE119  147–185  19  13  7.7  0.770  INRA63  162–210  14  21  7.4  0.785  HSC  263–299  20  19  9.3  0.835  McM42  82–106  9  9  5.1  0.634  INRA23  196–224  1  15  9.0  0.834  MAF214  185–261  16  13  4.4  0.538  MAF65  121–143  15  12  7.5  0.821  OarCP49  69–119  17  22  8.1  0.793  TGLA53  140–168  12  15  7.9  0.846  Mean      15.4  7.3  0.764  Locus  Fragment size  Chromosome  TNA  AR  GD  CSRD274  213–259  14  22  8.0  0.779  ILSTS87  138–178  6  19  8.8  0.814  OarCP20  68–102  21  15  5.5  0.745  OarCP34  100–128  3  12  6.5  0.788  OarFCB304  150–198  19  20  7.3  0.702  McM527  171–189  5  10  7.0  0.738  OarAE54  122–156  25  14  8.2  0.797  OarFCB20  92–122  2  16  8.9  0.836  URB58  161–209  13  18  8.0  0.808  OarAE129  137–157  5  8  4.7  0.651  OarAE119  147–185  19  13  7.7  0.770  INRA63  162–210  14  21  7.4  0.785  HSC  263–299  20  19  9.3  0.835  McM42  82–106  9  9  5.1  0.634  INRA23  196–224  1  15  9.0  0.834  MAF214  185–261  16  13  4.4  0.538  MAF65  121–143  15  12  7.5  0.821  OarCP49  69–119  17  22  8.1  0.793  TGLA53  140–168  12  15  7.9  0.846  Mean      15.4  7.3  0.764  View Large Statistical Analysis Number of alleles per locus, allelic frequencies, and observed and expected heterozygosity were calculated using Genetix version 4.05.2 (Belkhir et al., 1996–2004). Exact tests for deviation from Hardy-Weinberg equilibrium (Guo and Thompson, 1992) were applied using a Markov Chain Monte Carlo simulation (100 batches, 5,000 iterations per batch, and a dememorization number of 10,000) as implemented in GENEPOP version 3.4 (Raymond and Rousset, 1995). A test for population differentiation was performed, as implemented in GENEPOP 3.4; for each locus, an unbiased estimate of Fisher's exact test was computed to verify if the allelic distribution was different among breeds. The Fstat 2.9.3 software (Goudet, 1995) was employed in calculations of allelic richness (an estimation of mean number of alleles per locus corrected by sample size), gene diversity (Nei, 1987), and estimation of Wright's fixation index (Weir and Cockerham, 1984). Molecular coancestry coefficients within and between breeds were measured according to Caballero and Toro (2002) using MolKin 3.0 (Gutiérrez et al., 2005); to avoid bias, because of unequal sample sizes, 100 samples with exactly 50 individuals per breed were generated with a bootstrap procedure. To help set conservation priorities, Molkin 3.0 (Gutiérrez et al., 2005) was used to quantify the contribution of each analyzed population to the diversity of the whole data set using the method proposed by Caballero and Toro (2002). Because BER is not a Veneto native breed involved in the conservation scheme but has been used as reference population, its data were not included in the approach for setting conservation priorities. The genetic distances of Reynolds (DR) among breeds were estimated, a neighbor-joining consensus tree was reconstructed, and tree robustness was evaluated by bootstrapping over loci (1,000 replicates) using the PHYLIP package (Felsenstein, 1993–2002); the dendrogram was depicted using the software package TreeView version 1.6.6 (Page, 2001). Reynolds distance is the most suited for relatively closely related populations such as breeds in Europe that diverged during short time periods; in fact, in this case, the amount of mutation is negligible and the main factor to describe genetic variability is random drift (Eding and Laval, 1999). Moreover, genetic distances among individuals were estimated as the proportion of shared alleles (DPS) using MICROSAT 1.5 (Minch et al., 1998); individual distances were represented by a neighbor-joining tree and displayed using MEGA 4 (Tamura et al., 2007). To study population structure and to detect the most likely number of clusters (K) in the data set, the software STRUCTURE version 2.2 (Pritchard et al., 2000) was used. The analysis involved an admixture model with correlated allele frequencies as suggested by several authors (Pritchard et al., 2000; Vicente et al., 2008; Zuccaro et al., 2008). To choose the appropriate number of inferred clusters to model the data, 2 to 10 inferred clusters were performed with 50 independent runs each. All analyses used a burn-in period of 30,000 iterations and then 150,000 iterations for data collection. The best number of clusters fitting the data was established by plotting the Ln Pr(X|K) over the 50 independent runs for each K, as suggested by Pritchard et al. (2000). To find optimal alignments of independent runs, the computer program CLUMPP 1.1 (Jakobsson and Rosenberg, 2007) was used; the output obtained was used directly as input by the cluster visualization program DISTRUCT (Rosenberg, 2004). RESULTS Genetic Variability at Microsatellite Loci The total number of alleles detected in the entire data set was 293, and all markers were polymorphic in each breed. Information about the variability of the investigated loci is shown in Table 1. The largest number of alleles was found at loci CSRD274 and OarCP49 (22) and the smallest at locus OarAE129 (8). Global mean number of alleles per locus was 15.4, and allelic richness, an estimate of number of alleles per locus corrected by sample size, was 7.3. The gene diversity across loci was 0.764, ranging from 0.846 (TGLA53) to 0.651 (OarAE129). Private alleles were detected in each breed and for each marker; they represented 30% of the total number of observed alleles. However, allele frequencies were small and exceeded 0.10 in few cases. Bergamasca showed the largest number of private alleles with increased frequencies, displaying 4 alleles with frequencies greater than 0.10; ALP and BRO exhibited 2 private alleles each with frequencies greater than 0.15, whereas FOZ and LAM displayed private alleles with rather small frequencies. Breed Variability and Differentiation The genetic variability of each breed was studied, first, in terms of number of observed alleles and allelic richness, as shown in Table 2. Bergamasca showed the largest number of alleles per locus (9.6), followed by BRO (9.3), and LAM the least (7.1). On the other hand, allelic richness was very similar among breeds, with the exception of ALP, which exhibited the least value (5.3). Bergamasca also exhibited considerable heterozygosity (0.722), whereas in the native breeds, it was generally less, except for LAM (0.736). In general, all breeds showed heterozygosity estimates less than expected; a highly significant (P < 0.001) departure from Hardy-Weinberg equilibrium was observed in each of the breeds. This divergence was reflected in the FIS index within breed, which was always positive and rather large, especially in ALP and BRO, which are the least variable breeds according to observed heterozygosity. Another way to measure within-breed diversity is the estimation of molecular coancestry, a measure of relatedness among individuals. Molecular coancestry estimates varied from 0.236 (BER) to 0.318 (ALP), as shown in Table 2. Number of analyzed samples, total number of alleles (TNA), allelic richness (AR), expected (H. exp.) and observed (H. obs.) heterozygosity, within-breed heterozygote deficiency (FIS), and within-breed molecular coancestry (fij) for each breed analyzed Table 2. Number of analyzed samples, total number of alleles (TNA), allelic richness (AR), expected (H. exp.) and observed (H. obs.) heterozygosity, within-breed heterozygote deficiency (FIS), and within-breed molecular coancestry (fij) for each breed analyzed Breed  Sample size  TNA  AR  H. exp. ± SD  H. obs. ± SD  FIS  fij ± SD  Bergamasca  45  9.6  6.3  0.782 ± 0.090  0.722 ± 0.149  0.078***  0.236 ± 0.008  Alpagota  38  7.5  5.3  0.699 ± 0.187  0.607 ± 0.265  0.134***  0.318 ± 0.011  Brogna  38  9.3  6.1  0.764 ± 0.087  0.675 ± 0.164  0.118***  0.246 ± 0.012  Foza  33  9.1  6.2  0.770 ± 0.845  0.698 ± 0.186  0.095***  0.250 ± 0.011  Lamon  16  7.1  6.2  0.796 ± 0.063  0.736 ± 0.230  0.078***  0.239 ± 0.008  Breed  Sample size  TNA  AR  H. exp. ± SD  H. obs. ± SD  FIS  fij ± SD  Bergamasca  45  9.6  6.3  0.782 ± 0.090  0.722 ± 0.149  0.078***  0.236 ± 0.008  Alpagota  38  7.5  5.3  0.699 ± 0.187  0.607 ± 0.265  0.134***  0.318 ± 0.011  Brogna  38  9.3  6.1  0.764 ± 0.087  0.675 ± 0.164  0.118***  0.246 ± 0.012  Foza  33  9.1  6.2  0.770 ± 0.845  0.698 ± 0.186  0.095***  0.250 ± 0.011  Lamon  16  7.1  6.2  0.796 ± 0.063  0.736 ± 0.230  0.078***  0.239 ± 0.008  ***P < 0.001. View Large Table 2. Number of analyzed samples, total number of alleles (TNA), allelic richness (AR), expected (H. exp.) and observed (H. obs.) heterozygosity, within-breed heterozygote deficiency (FIS), and within-breed molecular coancestry (fij) for each breed analyzed Breed  Sample size  TNA  AR  H. exp. ± SD  H. obs. ± SD  FIS  fij ± SD  Bergamasca  45  9.6  6.3  0.782 ± 0.090  0.722 ± 0.149  0.078***  0.236 ± 0.008  Alpagota  38  7.5  5.3  0.699 ± 0.187  0.607 ± 0.265  0.134***  0.318 ± 0.011  Brogna  38  9.3  6.1  0.764 ± 0.087  0.675 ± 0.164  0.118***  0.246 ± 0.012  Foza  33  9.1  6.2  0.770 ± 0.845  0.698 ± 0.186  0.095***  0.250 ± 0.011  Lamon  16  7.1  6.2  0.796 ± 0.063  0.736 ± 0.230  0.078***  0.239 ± 0.008  Breed  Sample size  TNA  AR  H. exp. ± SD  H. obs. ± SD  FIS  fij ± SD  Bergamasca  45  9.6  6.3  0.782 ± 0.090  0.722 ± 0.149  0.078***  0.236 ± 0.008  Alpagota  38  7.5  5.3  0.699 ± 0.187  0.607 ± 0.265  0.134***  0.318 ± 0.011  Brogna  38  9.3  6.1  0.764 ± 0.087  0.675 ± 0.164  0.118***  0.246 ± 0.012  Foza  33  9.1  6.2  0.770 ± 0.845  0.698 ± 0.186  0.095***  0.250 ± 0.011  Lamon  16  7.1  6.2  0.796 ± 0.063  0.736 ± 0.230  0.078***  0.239 ± 0.008  ***P < 0.001. View Large The breeds showed considerable genetic differentiation; in fact, the FST index was equal to 0.085 (0.066 to 0.107; 99% confidence interval). Moreover, the test for population differentiation revealed a highly significant (P < 0.001) difference in the distribution of allelic frequencies among the breeds. To test if the large FST estimate was due to the presence of BER, which is not a native Venetian breed, it was computed again after the removal of this breed; the estimate was even greater as FST was equal to 0.097 (0.070 to 0.128, 99% confidence interval). The neighbor-joining tree constructed on DR (Figure 1) showed a clear distinction between ALP and BER and the other 3 local breeds, which appeared to be more similar. Figure 1. View largeDownload slide Representation of neighbor-joining Reynolds' genetic distance among Alpagota (ALP), Brogna (BRO), Foza (FOZ), Lamon (LAM), and Bergamasca (BER) breeds, based on 1,000 replicates (numbers in nodes are percentage bootstrap values). Figure 1. View largeDownload slide Representation of neighbor-joining Reynolds' genetic distance among Alpagota (ALP), Brogna (BRO), Foza (FOZ), Lamon (LAM), and Bergamasca (BER) breeds, based on 1,000 replicates (numbers in nodes are percentage bootstrap values). Results obtained with the Caballero and Toro (2002) approach to set up conservation priorities are illustrated in Table 3. The removal of 1 breed from the data set resulted in a loss of genetic diversity in the population, which ranged from –0.51 to –1.64% when ALP and FOZ, respectively, were removed. The greatest loss of between-breed diversity was found by removing ALP (−3.25%); on the other hand, its removal resulted in a contribution to the within-breed diversity (0.27%). On the contrary, removal of BRO gave a modest contribution to the between-breed diversity (0.08%) and a loss in the within-breed diversity (−1.13%). Loss or gain of genetic diversity (GD, in %) in the population when one breed is removed according to the approach of Caballero and Toro (2002) Table 3. Loss or gain of genetic diversity (GD, in %) in the population when one breed is removed according to the approach of Caballero and Toro (2002) Breed  GD  Within breeds  Between breeds  Loss (−)/Gain (+)  All breeds  0.813        Alpagota  0.806  +2.740  −3.245  −0.505  Brogna  0.801  −1.126  +0.081  −1.045  Foza  0.796  −0.889  −0.752  −1.641  Lamon  0.801  −0.538  −0.531  −1.069  Breed  GD  Within breeds  Between breeds  Loss (−)/Gain (+)  All breeds  0.813        Alpagota  0.806  +2.740  −3.245  −0.505  Brogna  0.801  −1.126  +0.081  −1.045  Foza  0.796  −0.889  −0.752  −1.641  Lamon  0.801  −0.538  −0.531  −1.069  View Large Table 3. Loss or gain of genetic diversity (GD, in %) in the population when one breed is removed according to the approach of Caballero and Toro (2002) Breed  GD  Within breeds  Between breeds  Loss (−)/Gain (+)  All breeds  0.813        Alpagota  0.806  +2.740  −3.245  −0.505  Brogna  0.801  −1.126  +0.081  −1.045  Foza  0.796  −0.889  −0.752  −1.641  Lamon  0.801  −0.538  −0.531  −1.069  Breed  GD  Within breeds  Between breeds  Loss (−)/Gain (+)  All breeds  0.813        Alpagota  0.806  +2.740  −3.245  −0.505  Brogna  0.801  −1.126  +0.081  −1.045  Foza  0.796  −0.889  −0.752  −1.641  Lamon  0.801  −0.538  −0.531  −1.069  View Large Population Structure The structure of the breeds was analyzed using a Bayesian approach that inferred the number of clusters (K) present in the population, permitting detection of differences among breeds and hidden substructures within breeds. The Ln Pr(X|K) increased sharply from K = 2 to K = 5 and reached a plateau without showing a significant decrease from K = 6 to K = 10, as illustrated in Figure 2. The greatest mean Ln Pr(X|K) over the 50 independent runs performed for each value of K was found at K = 7; at K = 5 and K = 6, estimated Ln Pr(X|K) were more consistent across runs leading to some difficulties in identifying the most probable number of clusters in the population. Results of STRUCTURE analyses are shown in Figure 3 for K ranging from 5 to 8. It can be noticed that, apart from the number of clusters, a clear distinction among the 5 breeds was possible and only a small level of admixture was detectable. In particular, ALP and BER formed 2 distinct clusters with high proportion of membership (0.876 and 0.890, respectively, considering K = 7), and admixture was nearly absent. On the other hand, BRO, FOZ, and LAM exhibited a more complex structure with proportion of membership split into 2 or more clusters and an underlying presence of within breed substructures; nevertheless, admixture seemed to be negligible even among these breeds. It is worth mentioning that, even at K > 8, ALP and BER still evidenced increased membership to only 1 cluster, whereas BRO, FOZ, and LAM presented a more and more complex structure. Figure 2. View largeDownload slide Estimated posterior probabilities [Ln Pr(X|K)] for different numbers of inferred clusters (K). Figure 2. View largeDownload slide Estimated posterior probabilities [Ln Pr(X|K)] for different numbers of inferred clusters (K). Figure 3. View largeDownload slide Graphical representation of the estimated membership fractions of individuals of the breeds analyzed in each of the K inferred clusters, for K = 6 to K = 8. Color version available in the online pdf. Figure 3. View largeDownload slide Graphical representation of the estimated membership fractions of individuals of the breeds analyzed in each of the K inferred clusters, for K = 6 to K = 8. Color version available in the online pdf. The neighbor-joining tree of individuals, based on the proportion of shared alleles (Figure 4), showed that individuals grouped together very well by breed. Some confusion was present only between FOZ and LAM animals, which, in some cases, grouped together. It is interesting to see that the tree could be separated into 2 groups, one composed of BRO, FOZ, and LAM groups and one composed of ALP and BER. Figure 4. View largeDownload slide Neighbor-joining tree based on the proportion of shared alleles of Alpagota (ALP = ●), Brogna (BRO = ■), Foza (FOZ = ▲), Lamon (LAM = ∆), and Bergamasca (BER = ○). Figure 4. View largeDownload slide Neighbor-joining tree based on the proportion of shared alleles of Alpagota (ALP = ●), Brogna (BRO = ■), Foza (FOZ = ▲), Lamon (LAM = ∆), and Bergamasca (BER = ○). DISCUSSION The 4 local breeds analyzed in this study are the only native Venetian breeds that survived extinction and are still reared on small farms. Nevertheless, to maintain them and to increase their population size, it has been necessary to set up a conservation flock on an experimental farm with financing support of the Veneto regional government. Reasons for conserving animal genetic resources have already been underlined by several authors (Mendelsohn, 2003; Rege and Gibson, 2003), and in our case, the conservation of ALP, BRO, FOZ, and LAM is carried out for 3 main reasons. 1) Their important environmental role in exploiting mountain and marginal pastures. In fact, the abandonment of pastures results in a landscape change in favor of the expansion of woodlands, as observed by Giupponi et al. (2006) in a study of the land use changes in the Belluno Venetian province. This loss of landscape diversity is negative for the aesthetic-recreational appeal of the area because it has been shown that tourism activities favor maintaining a more pleasant landscape and avoiding excessive afforestation (Giupponi et al., 2006). This question is of outstanding importance as tourism plays an important role in the economy of the Veneto region. In addition, landscape changes also have negative effects on the habitats of species of naturalistic interests (Giupponi et al., 2006). 2) The cultural value in that these breeds are used for the production of typical foods linked to old traditions; authors have usually referred to this aspect as existence value (Rege and Gibson, 2003). 3) The interesting adaptive traits representing genetic stocks that could transmit favorable traits to new breed crosses. Thanks to these features, these breeds have been enrolled in an in situ conservation program aiming to increase the population size, to maintain genetic variability, to organize matings, and to avoid crossbreeding, which was rather common in small flocks. In fact, in situ preservation of endangered breeds should be preferred to ex situ or cryo-conservation because in situ preservation promotes sustainable management of genetic resources and permits conservation of all features carried by a specific breed (Taberlet et al., 2008). Moreover, in situ conservation allows animals to be kept in the original area of production. This aspect is important considering the link between Venetian sheep breeds and their typical production. Product valorization is also a step toward the sustainability of these breeds in case Italian or European government subsidies are not available in the future. An example of the profitability of rearing local and less productive breeds in marginal areas, when there is a strong link among breed, territory, and product, has been examined by Pretto et al. (2009) in a study of the native Italian Burlina cattle breed. The first step toward safeguarding of ALP, BRO, LAM, and FOZ was the choice of the founder animals to rear in the conservation nucleus, according to their phenotypic traits. It is worth mentioning that, for the FOZ and LAM breeds, the situation was extremely critical because their population size is extremely small. Nevertheless, these were the 2 breeds showing the greatest variability and the least, but significant, deficit of heterozygotes. This lack of heterozygotes could be caused by the presence of population substructures, rather than by increased levels of inbreeding. In fact, results obtained with STRUCTURE evidenced a complex substructure with proportion of membership split into more than 1 cluster, at least for K = 7. The observed substructures could be due to the choice of founder animals that were collected on different farms and were probably genetically and geographically distant; to remove such structures, more generations of selected matings within the conservation flock are needed. On the other hand, molecular coancestry estimates were slightly less than were found in some Spanish sheep breeds (Legaz et al., 2008) and significantly less when compared with Spanish breeds considered as endangered (Álvarez et al., 2005). The BRO breed presented a similar situation, even if it seemed to retain decreased genetic diversity. In particular, BRO showed proportion of membership split almost equally into 2 clusters, probably due to a clear genetic distinction among founders. The ALP breed showed the least observed heterozygosity, the greatest FIS, and the greatest molecular coancestry estimates. Comparable levels of variability were observed in the autochthonous Chiapas breed from Mexico (Quiroz et al., 2008) and in the Italian Sarda population (Pariset et al., 2003). In this case, the significant deficit of heterozygotes seemed to be caused mainly by inbreeding because STRUCTURE did not detect any substructure in ALP. This hypothesis was also supported by the fact that ALP has been reared mainly in a rather isolated area and its breeding did not provide any transhumance activity; in this way exchange of genetic material and crosses with other breeds were more difficult. Confirmation of the genetic distinctness and isolation of ALP could be seen as well in the neighbor-joining tree based on DR distances, where BER seemed to be the closest breed. The breeds analyzed in this study showed clear genetic differentiation because 8.5% of the genetic variability was explained by differences among breeds; similar levels of differentiation were detected by Baumung et al. (2006) in an analysis of Austrian sheep breeds. Usually, native breeds coming from the same geographic area showed much smaller estimates of genetic differentiation as presented by other authors in Portuguese (FST = 0.049, Santos-Silva et al., 2008), northern Spanish (FST = 0.029, Rendo et al., 2004), and Alpine (FST = 0.057, Dalvit et al., 2008) sheep breeds. Moreover, greater proportions of private alleles were detected in each breed; in particular, BER showed the largest number of private alleles with increased allele frequencies, probably because of its different origins. Among the 4 native breeds, ALP showed private alleles with greater frequencies, confirming its uniqueness and isolation. The greater level of genetic differentiation, the clear distinction among breeds and individuals detected using DR and DPS distances, and the decreased level of admixture evidenced among breeds are important factors that support conservation of ALP, BRO, FOZ, and LAM as separate breeds with unique genetic features. Results of the analysis to set up conservation priorities, carried out according to Caballero and Toro (2002), also supported this hypothesis. Removal of any of the 4 local breeds would result in a loss of genetic diversity. Extinction of ALP would result in a rather small loss, because of its high inbreeding, as shown by the within-breed diversity. In fact, the approach by Caballero and Toro (2002) suggests that within- and between-subpopulation variability should be included in conservation decisions; ignoring the within breed variability will, in fact, favor inbred populations and populations with extreme allele frequencies (Toro and Caballero, 2005). Extinction of any of the other 3 breeds would instead result mainly in loss of within-breed variability. It is worth mentioning that the loss of diversity caused by removal of any of the studied breeds appeared small, being on average nearly 1%; however, this estimate was large compared with literature results. Fabuel et al. (2004), Glowatzki-Mullis et al. (2008), and Thirstrup et al. (2008) found smaller estimates in studies of Iberian pig, Swiss goat, and Danish horse breeds, respectively. In conclusion, the present study highlighted the importance of the in situ conservation program set up to safeguard native sheep breeds. In fact, these breeds are an important reservoir of genetic variation and of genetic uniqueness, and their extinction would cause a loss of diversity. Our study will be useful in outlining further development of the in situ conservation scheme; particular attention should be paid to organizing matings to minimize inbreeding, especially in ALP, which is already suffering from a lack of heterozygosity. The presence of substructures observed in BRO, FOZ, and LAM was probably due to fragmentation in isolated small flocks before their inclusion in the conservation flock. This population structure should be overcome in the next generations by rotating the rams and favoring mating between genetically distinct animals. In addition, to achieve sustainable management of the described genetic resources, the production of these breeds should be valorized by taking into account their value as reservoirs of unique diversity, as suggested by Taberlet et al. (2008). LITERATURE CITED Álvarez I. Gutiérrez J. P. Royo L. J. Fernández I. Gómez E. Arranz J. J. Goyache F. 2005. Testing the usefulness of the molecular coancestry information to assess genetic relationships in livestock using a set of Spanish sheep breeds. J. Anim. Sci.  83: 737– 744. [PubMed] Google Scholar CrossRef Search ADS PubMed  Baumung R. Cubric-Curik V. Schwend K. Achmann R. Sölkner J. 2006. Genetic characterisation and breed assignment in Austrian sheep breeds using microsatellite markers information. J. Anim. Breed. Genet.  123: 265– 271. [PubMed] Google Scholar CrossRef Search ADS PubMed  Belkhir, K., P. Borsa, L. Chikhi, N. Raufaste, and F. Bonhomme 1996–2004. GENETIX 4.05, logiciel sous Windows TM pour la génétique des populations. Laboratoire Génome, Populations, Interactions. CNRS UMR 5000,  Université de Montpellier II, Montpellier, France. http://www.genetix.univ-montp2.fr/genetix/genetix.htm Accessed Nov. 5, 2008. Caballero A. Toro M. A. 2002. Analysis of genetic diversity for the management of conserved subdivided populations. Conserv. Genet.  3: 289– 299. Google Scholar CrossRef Search ADS   Cabrio, G. 1965. Pecore Venete.  Ramella Artigrafiche ed., Biella, Italy. Dalvit C. Saccà E. Cassandro M. Gervaso M. Pastore E. Piasentier E. 2008. Genetic diversity and variability in Alpine sheep breeds. Small Rumin. Res.  80: 45– 51. Google Scholar CrossRef Search ADS   Di Stasio, L. 1980. Indagine genetica su tre razze ovine di montagna in via di estinzione. In Proc. Le possibilità delle colture e degli allevamenti nei territori alpini,  St. Vincent, Italy. Eding, J. U., and G. Laval 1999. Measuring genetic uniqueness in livestock. Pages 33–58 in Genebanks and the Conservation of Farm Animal Genetic Resources.  J. K. Oldenbroek ed. Institute for Animal Science and Health, Lelystad, the Netherlands. Fabuel E. Barragán C. Silió L. Rodriguez M. C. Toro M. A. 2004. Analysis of genetic diversity and conservation priorities in Iberian pigs based on microsatellite markers. Heredity  93: 104– 113. [PubMed] Google Scholar CrossRef Search ADS PubMed  FASS 1999. Guide for Care and Use of Agricultural Animals in Agricultural Research and Teaching.  1st rev. ed. Fed. Anim. Sci. Soc., Savoy, IL. Felsenstein, J. 1993–2002. Phylogeny Inference Package (PHYLIP). Genomes Sciences,  Department of Genetics, University of Washington, Seattle, WA. http://evolution.gs.washington.edu/phylip.html Accessed Nov. 5, 2008. Giupponi C. Ramanzin M. Sturaro E. Fuser S. 2006. Climate and land use changes, bidiversity and agro-environmental measures in the Belluno province, Italy. Environ. Sci. Policy  9: 163– 173. Google Scholar CrossRef Search ADS   Glowatzki-Mullis M. Muntwyler J. Bäumle E. Gaillard C. 2008. Genetic diversity measures of Swiss goat breeds as decision-making support for conservation policy. Small Rumin. Res.  74: 202– 211. Google Scholar CrossRef Search ADS   Goudet J. 1995. FSTAT (version 2.9.3): A computer programme to calculate F-statistics. J. Hered.  8: 485– 486. Google Scholar CrossRef Search ADS   Guo S. W. Thompson E. A. 1992. Performing the exact test of Hardy-Weinberg proportion for multiple alleles. Biometrics  48: 361– 372. [PubMed] Google Scholar CrossRef Search ADS PubMed  Gutiérrez J. P. Royo L. J. Álvarez I. Goyache F. 2005. MolKin v2.0: A computer program for genetic analysis of populations using molecular coancestry information. J. Hered.  96: 718– 721. [PubMed] Google Scholar CrossRef Search ADS PubMed  ISAG/FAO Standing Committee 2004. Secondary Guidelines for Development of National Farm Animal Genetic Resources Management Plans. Measurement of Domestic Animal Diversity (MoDAD): Recommended Microsatellite Markers.  http://dad.fao.org/cgi-bin/getblob.cgi?sid=ca53b91a6f7c80be8e7066f4a5066dc1,50006220 Accessed Nov. 4, 2008. Jakobsson M. Rosenberg N. A. 2007. CLUMPP: A cluster matching and permutation program for dealing with label switching and multimodality in analysis of population structure. Bioinformatics  23: 1801– 1806. [PubMed] Google Scholar CrossRef Search ADS PubMed  Legaz E. Álvarez I. Royo L. J. Fernández I. Gutiérrez J. P. Goyache F. 2008. Genetic relationship between Spanish Assaf (Assaf.E) and Spanish native dairy sheep breeds. Small Rumin. Res.  80:39–44. Mendelsohn R. 2003. The challenge of conserving indigenous domesticated animals. Ecol. Econ.  45: 501– 510. Google Scholar CrossRef Search ADS   Minch, E., A. Ruiz-Linares, D. B. Goldstein, M. W. Feldman, and L. L. Cavalli-Sforza 1998. Microsat2: A computer program for calculating various statistics on microsatellite allele data. Department of Genetics,  Stanford University, Stanford, CA. http://hpgl.stanford.edu/projects/microsat/ Accessed Nov. 4, 2008. Montanari, V. 1954. L'allevamento del bestiame nel Veneto Friuli-Venezia Giulia. Pages 140–146 Ministero dell'Agricoltura e delle Foreste,  Ispettorato Agrario Compartimentale delle Venezie ed, Italy. Nei, M. 1987. Molecular Evolutionary Genetics.  Columbia University Press, New York, NY. Page R. D. M. 2001. TREEVIEW: An application to display phylogenetic trees on personal computers. Comput. Appl. Biosci.  12: 357– 358. Pariset L. Savarese M. C. Cappuccio I. Valentini A. 2003. Use of microsatellites for genetic variation and inbreeding analysis in Sarda sheep flocks of central Italy. J. Anim. Breed. Genet.  120: 425– 432. Google Scholar CrossRef Search ADS   Pastore, E. 2002. Le razze ovine autoctone del Veneto.  Veneto Agricoltura ed., Legnaro (PD), Italy. Pastore, E., and L. Fabbris 1999. L'allevamento ovi-caprino nel Veneto, analisi e prospettive future di un settore ricco di storia.  Veneto Agricoltura ed., Legnaro (PD), Italy. Pretto D. De Marchi M. Dalvit C. Cassandro M. 2009. Comparing profitability of Burlina and Holstein Friesian cattle breeds. Ital. J. Anim. Sci.  In press. Pritchard J. K. Stephens M. Donnelly P. 2000. Inference of population structure using multilocus genotype data. Genetics  155: 945– 959. [PubMed] Google Scholar PubMed  Quiroz J. Martinez A. M. Zaragoza L. Perezgrovas R. Vega-Pla J. L. Delgado J. V. 2008. Genetic characterization of the autochthonous sheep populations from Chiapas, Mexico. Livest. Sci.  116: 156– 161. Google Scholar CrossRef Search ADS   Rancourt M. Fois N. Lavin M. P. Tchakerian E. Vallerand F. 2006. Mediterranean sheep and goats production: An uncertain future. Small Rumin. Res.  62: 167– 179. Google Scholar CrossRef Search ADS   Raymond M. Rousset F. 1995. GENEPOP (version 1.2): Population genetics software for exact tests and ecumenicism. J. Hered.  86: 248– 249. Google Scholar CrossRef Search ADS   Rege J. E. O. Gibson J. P. 2003. Animal genetic resources and economic development: Issues in relation to economic evaluation. Ecol. Econ.  45: 319– 330. Google Scholar CrossRef Search ADS   Rendo F. Iriondo M. Jugo B. M. Mazón L. I. Aguirre A. Vicario A. Estonba A. 2004. Tracking diversity and differentiation in six sheep breeds from the North Iberian Peninsula through DNA variation. Small Rumin. Res.  52: 195– 202. Google Scholar CrossRef Search ADS   Rosenberg N. A. 2004. Distruct: A program for the graphical display of population structure. Mol. Ecol. Notes  4: 137– 138. Google Scholar CrossRef Search ADS   Santos-Silva F. Ivo R. S. Sousa M. C. O. Carolino M. I. Ginja C. Gama L. T. 2008. Assessing genetic diversity and differentiation in Portuguese coarse-wool sheep breeds with microsatellite markers. Small Rumin. Res.  78: 32– 40. Google Scholar CrossRef Search ADS   Slow Food Foundation 2008. Presidio Slow Food. A logo for the Italian Slow Food Presidia.  http://www.presidislowfood.it/pdf/italian_slowfood_presidia.pdf Accessed Apr. 8, 2009. Taberlet P. Valentini A. Rezael H. R. Naderi S. Pompanon F. Negrini R. Ajmone-Marsan P. 2008. Are cattle, sheep, and goats endangered species? Mol. Ecol.  17: 275– 284. [PubMed] Google Scholar CrossRef Search ADS PubMed  Tamura K. Dudley J. Nei M. Kumar S. 2007. MEGA4: Molecular Evolutionary Genetics Analysis (MEGA) software version 4.0. Mol. Biol. Evol.  24: 1596– 1599. [PubMed] Google Scholar CrossRef Search ADS PubMed  Thirstrup J. P. Pertoldi C. Loeschcke V. 2008. Genetic analysis, breed assignment and conservation priorities of three native Danish horse breeds. Anim. Genet.  39: 496– 505. [PubMed] Google Scholar CrossRef Search ADS PubMed  Toro M. A. Caballero A. 2005. Characterization and conservation of genetic diversity in subdivided populations. Philos. Trans. R. Soc. B  360: 1367– 1378. [PubMed] Google Scholar CrossRef Search ADS   Vicente A. A. Carolino M. I. Sousa M. C. O. Ginja C. Silva F. S. Martinez A. M. Vega-Pla J. L. Carolino N. Gama L. T. 2008. Genetic diversity in native and commercial breeds of pigs in Portugal assessed by microsatellites. J. Anim. Sci.  86: 2496– 2507. [PubMed] Google Scholar CrossRef Search ADS PubMed  Weir B. S. Cockerham C. C. 1984. Estimating F-statistics for the analysis of population structure. Evolution  38: 1358– 1370. Google Scholar PubMed  Zuccaro A. Bordonaro S. Criscione A. Guastella A. M. Perrotta G. Blasi M. D'Urso G. Marletta D. 2008. Genetic diversity and admixture analysis of Sanfratellano and three other Italian horse breeds assessed by microsatellite markers. Animal  2: 991– 998. Google Scholar CrossRef Search ADS PubMed  American Society of Animal Science This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/), which permits non-commercial reuse, distribution, and reproduction in any medium, provided the original work is properly cited. For commercial re-use, please contact [email protected]

Forni, S.;Gianola, D.;Rosa, G. J. M.;de los Campos, G.

doi: 10.2527/jas.2008-1514pmid: 19684262

ABSTRACT A Bayesian model for quantitative genetic analysis of longitudinal traits is presented. It connects the model known as the Kalman filter (KF) with the standard mixed model of quantitative genetics. The KF model can be implemented easily in a Bayesian framework because, under standard prior assumptions, all fully conditional posterior distributions have closed forms. An analysis of beef cattle growth data including comparisons with a standard multivariate model was performed to assess applicability of the KF to animal breeding. Models were compared using the deviance information criterion and the Bayes factor. Models in which a KF acted on additive genetic and maternal effects were favored by the deviance information criterion, although KF did not describe residual (co)variance adequately. The Bayes factor did not provide conclusive evidence in favor of a specific model. Fitting KF to longitudinal traits provides estimates of genetic value for a whole range of time points, assuming that there are genetic differences through time between and within individuals. Different models embedding the KF in a mixed model were demonstrated to provide a more parsimonious (co)variance structure than a standard multitrait specification for the quantitative genetic analysis of longitudinal data. INTRODUCTION Genetic studies often involve traits measured in the same individual at different times. Different approaches to the genetic analysis of time-series data have been reviewed (e.g., Harvey, 1993; Lindsey, 1993; Shumway and Stoffer, 2000). Methods commonly used include 1) treating phenotypic values at different time points as repeated measurements of the same trait (repeatability model); 2) treating phenotypes as different traits and using multivariate analysis; and 3) using regression (linear or nonlinear) with fixed and random coefficients (e.g., Meyer and Hill, 1997; Schaeffer, 2004). In many instances, the assumption in 1 does not hold because it implies constant correlations between pairs of measures. On the other hand, a full multivariate model (MM) can produce an excessively parameterized specification. Although conceptually appealing, application of random regression models has been hampered by problems regarding the number of parameters to be estimated, poor fit of polynomial approximations, uncertain predictions at the beginning and end of trajectories, and high computational demands (Varona et al., 1997; Meyer, 2005; Forni et al., 2007). State-space models or dynamic linear models provide a flexible way of accounting for sources of random variation that may change over time. An example of a well-known dynamic model is the Kalman filter (KF; Kalman, 1960). An essential difference between this model and the conventional linear model is that, in the former, unobserved quantities evolve with time. By incorporating such a filtering structure into a standard linear genetic model, environmental and genetic effects can be inferred via state equations relating effects at time t to those at past time points. Thus, a time-variant feature can be incorporated for random effects, giving rise to a (co)variance structure with fewer parameters than in a classical multivariate analysis. In this paper, a Bayesian model for genetic analysis of longitudinal data connecting the KF with the standard mixed models procedure is presented. An example of application to cattle growth data are provided to demonstrate the advantages of this specification. MATERIALS AND METHODS Animal Care and Use Committee approval was not obtained for this study because data were obtained from an existing database (Lôbo et al., 2007). KF A brief description of KF follows. Let yt (t = 1, 2, …, T) denote observed values of a Gaussian process for scalars or vectors. These observed values at time t depend on unobservable quantities θt, scalar or vector valued. A linear relationship between observed values and θt is assumed in the observation equation:  ytftθtεt [1] where ft is some function, and εt is an observation error, normally distributed with mean zero and known (co)variance matrix: The dynamic feature of KF is incorporated through a model that describes how the state θt propagates in time because of external influences, such as input or noise. This model or state equation is  θtgtθt−1δt [2] where gt is a known quantity and δt is the state equation error, or random shock. This error is assumed independent of the observation error and is normally distributed as The relationships specified through ft and gt may or may not change with time; likewise for the (co)variance matrices and In state-space models, the estimation of unobservable or hidden states is referred to as filtering when only data up to time t, and not beyond, are used for inferring θt. From a Bayesian perspective, the KF can be thought of as an updating procedure consisting of forming a prior guess and then adding a correction to this guess. The mean of the prior distribution of θt is gtθt−1 and the correction is the random shock δt. A more detailed description of KF is given in Sallas and Harville (1981), and Meinhold and Singpurwalla (1983). KF: Application to Genetic Analysis of Longitudinal Traits The mixed linear model commonly used in quantitative genetics postulates that the data and other random components are multivariate normally distributed, and that location effects and data are linearly related. Let yt denote records for some potentially infinite-dimensional character, such as growth or milk yield, measured at times t = 1, 2, …, T, in each of n individuals. Under the additive genetic effects model, a simple way of expressing the vector of phenotypic values at each time t is as a linear combination of time-dependent nuisance parameters (βt), additive genetic effects (at), and residual effects (et):  ytXβtZatet [3] Above, the genetic and residual effects are assumed to be independently distributed as and respectively, where A is a known q × q relationship matrix; In is an identity matrix of order n; and are the variances of additive genetic and residual effects at time t, and X and Z are known incidence matrices that relate βt and at, respectively, to the observations. Assuming that there are no missing observations, all genetic and residual effects can be stacked into vectors a and e, such that and Here, G0 is a T × T (co)variance matrix between genetic effects, and R0 is a T × T (co)variance matrix between residual effects. For simplicity, just one single random factor (at), besides the residuals, will be considered next. However, the model can be extended to accommodate other sources of variation, as shown in the next section. Assuming that genetic and residual effects are time dependent, but the β is not, the KF structure postulates that the former change over time according to  [4a] and  [4b] where Iq is an identity matrix of order q, and are rates of change of genetic and residual effects, respectively, at time t relative to time t − 1, and and are independently distributed random shocks following the normal distributions and respectively. Equation 4a implies that the additive variance at time t has 2 components: one related to additive effects at time t − 1 and another to an additive genetic random noise. Genetic or residual noises observed at different times are uncorrelated. Environmental correlations across time points are accommodated through Eq. 4b, and more complex covariance structures can be imposed as needed. Iterating backward,  [5a] and likewise  [5b] Combining assumptions of the standard model 3 with those of the state 4 leads to the joint distribution  [6] where δa and δe above represent the vectors of all additive and residual random shocks, respectively; and are additive and residual variances at time t = 1 and Δa and Δe are diagonal matrices of order (T − 1) × (T − 1), with typical elements and respectively. The genetic or residual variances at any time point can be deduced from Eq. 5a and 5b. The relationships between variances at time t and t − 1 are   [7a] and  [7b] where and stand for the diagonal elements of Δa and Δe, as noted above. Covariances between genetic and residual effects at any 2 time points i and j (i < j) are obtained using λ as regression coefficients and the variance estimates as predictors:  [8a] and  [8b] This leads to the joint distribution  [9] where is the (T × T) matrix of genetic (co)variances, and is the (T × T) matrix of residual (co)variances under KF. This distribution is equivalent to the joint distribution of genetic and residual effects in a standard MM. With KF, the algorithm to estimate (co)variance matrices is different, but genetic values are similarly estimated. The KF introduces a time-variant structure for random effects, and all covariances are expressible in terms of variance components and “regression” coefficients (λ). A prediction equation for phenotypic records is not presented, but KF allows the prediction of genetic values, residual effects, or any random variable at all phenotypic trajectories using Eq. 2. The KF postulates a linear relationship between phenotypic records or random variables in different time points. In a Bayesian framework, the marginal distribution of yet-to-be-observed data can be accessed from the joint distribution of the data and the explanatory variables. In the context of animal population dynamics, Walters (1986) interpreted the rate of change (λ) as a survival rate. Here, it can be viewed as a “permanence” rate of additive and residual effects. It is assumed that and are not subject specific, but this can be relaxed in a more complex hierarchical model. In addition, if there is interest in inferring (co)variances between β over time, these parameters can be treated as random variables, and the filtering model can be extended accordingly. In the quantitative genetic context, KF is consistent with the idea that phenotypes at different time points are influenced by different sets of genes that may or may not overlap between time points. Furthermore, the (co)variance structure between measures can be modeled with a smaller number of parameters than in a standard multitrait model. For instance, with t = 10, a standard multiple trait mixed model has 110 dispersion parameters; KF has 20, plus 18 λ-coefficients. KF: Bayesian Implementation for Genetic Analysis Adopting a hierarchical representation, the joint posterior density of all unknowns is assumed to have the form  [10] where Λa and Λe are vectors containing the genetic and residual rates of changes, respectively. The conditional density of the data given β and the genetic effects, can be represented as  [11] Conjugate prior densities were adopted to simplify Markov chain Monte Carlo sampling. Omitting the dependence on hyperparameters in the notation, the following prior distributions were assumed:  [12]  [13]  [14] and  [15] where β0 and contain prior means and (co)variances for nuisance effects; and are prior means for the genetic and residual rates of change, respectively, and and represent prior guesses of their variances. Further, independently scaled inverted chi-square distributions were used as priors for variance components:   [16]  [17]  [18] and  [19] where the S are scale parameters and the v are degrees of freedom. The fully conditional posterior distributions of each unknown can be ascertained from the joint density (Eq. 10) by retaining the parts varying with the parameters of interest and treating the remaining as known; a detailed description of conditional posterior distributions for multivariate mixed models is given in Sorensen and Gianola (2002). Given the following mixed model equations,  [20] Let C1 represent the coefficient matrix and stand for the solution vector. The fully conditional distribution of elements of θ is  [21] where θi can be a scalar or a vector; “else” means all other parameters, and is the ith diagonal element of C1 or a submatrix thereof. If θi is a scalar, drawing samples from this distribution does not require inversion of matrices; this is convenient when working with a large θ, but can have poor mixing. The state equations in Eq. 4 are analogous to a linear regression model. Let  [22a] and  [22b] and be the coefficient matrix and solution, respectively, of linear system Eq. 22a, and and be the corresponding items in Eq. 22b. The fully conditional posterior distributions of the rates of changes at each time t are   [23a] and  [23b] Finally, the fully conditional posterior distributions of the additive and residual variances at time t = 1, and of the (T − 1) random shock variances are  [24]  [25]  [26] and  [27] The scale and degrees of freedom parameters are  [28]  [29]  [30] and  [31] A Bayesian implementation of KF can be based entirely on Gibbs sampling (Casella and George, 1992) because all fully conditional posterior distributions have a closed form under our assumptions. The computing scheme can be outlined as follows: sample all location parameters (a, β) using Eq. 21; sample genetic and residual rates of changes using Eq. 23a and 23b, respectively; compute genetic and residual random shocks using Eq. 4a and 4b; sample genetic and residual variances at time t = 1, from Eq. 24 and 25, respectively; obtain the (T − 1) samples corresponding to the diagonal elements of the variance matrices using Eq. 26 and 27; and update the additive genetic and residual (co)variance matrices using Eq. 7a and 8a for and Eq. 7b and 8b for Application to Cattle Growth Data An analysis of cattle growth data was conducted to illustrate application of KF to modeling longitudinal traits in quantitative genetics. Different models, with the filtering structure applied to one or more explanatory variables, were contrasted. Data. Data consisted of BW records of Brazilian Nelore cattle, provided by ANCP (Associação Nacional de Criadores e Pesquisadores). Recording began in 1987, and a genetic improvement program has been operational since (Lôbo et al., 2007). Animals were weighed regularly at 90-d intervals from birth to 540 d of age. Six equally spaced periods were set, and the record of each animal within a period was linearly interpolated to its midpoint. Mean ages at each interval were 45, 135, 225, 315, 405, and 495 d. To account for nuisance effects, a “contemporary group” variable was created by combining the factors sex of animal, herd, year and season of birth [rainy or dry; see Forni et al. (2007)]. From the entire data set, records were extracted for animals without missing records and that belonged to a contemporary group with at least 10 animals. After edits, BW records of 6,856 animals, progeny of 326 sires and 3,898 dams, were available. Hence, 41,136 BW were used in the analysis. Pedigree information was obtained for all available generations, resulting in a total of 15,946 animals in the additive relationship matrix. Models. Nine models were investigated. The standard multitrait model (model 1) was  ytXβtZatWmtet [32] where yt denotes the vector of data at each time point t = 1, 2, …, 6; βt represents the effects of 154 contemporary groups; at stands for the vector of additive genetic effects at time t; mt denotes the vector of maternal effects, assumed to be environmental; et represents the vector of residual effects; and X, Z, and W are known incidence matrices. The additive, maternal, and residual effects were assumed to be independently distributed random vectors, following the normal distributions and respectively. Here, and are 6 × 6 matrices each. This model has 63 dispersion parameters. Note that, if the data were not equally spaced, the effect of age could be accommodated in βt. Model 2 was a KF, with filtering structures as described in Eq. 4, applied to the genetic, maternal, and residual effects. The 3 different state equations can be represented as  [33] where ut stands for either additive, maternal, or residual effects, with each one having time-specific rates of change and (co)variance structures as defined in Eq. 7 and 8. This model has 33 dispersion parameters, including the λ. Model 3 was like model 2, but all λ were set equal to 1, leading to 18 dispersion parameters and is equivalent to an autoregressive model of order 1. This assumption makes the variances of each effect increase with time. For these data, however, this may not be a strong restriction, because measurements represent early growth, and the variance is expected to increase with the mean. Preliminary results showed that a KF would not be needed for all effects. Thus, models with the filter applied to just 1 or 2 effects were fitted as well. Model 4 was as in Eq. 32, but with the KF applied only to additive genetic effects; model 5 had a filter for maternal effects only; model 6 assumed a KF only for residual effects. Model 7 had a filter for both additive and maternal effects; model 8 had the KF structure for additive and residual effects and, finally, model 9 assumed KF for maternal and residual effects. In all models, except for model 3, rates of change were treated as unknown. The prior distributions used were        and  where is the phenotypic variance when t is maximum. In the multitrait model, the prior distributions of (co)variance matrices were  f(G0) = f(M0) = f(R0) ~ IW(S, 7), where S is a diagonal matrix containing Model Comparison. The models were compared using 2 Bayesian measures. The first one was the deviance information criterion (DIC) proposed by Spiegelhalter et al. (2002):  [34] where is the posterior mean of parameters θi under model Mi,  [35] is the posterior expectation (under Mi) of the deviance, and  [36] is the deviance evaluated at the posterior means. A measure of the effective number of parameters in a model is the difference between Eq. 35 and 36. Models with a smaller DIC are favored because this indicates a better fit and a lesser degree of model complexity. The second approach for comparing models was the Bayes factor (BF). It is defined as the ratio between marginal densities of the data under 2 competing models:  [37] The BF provides a measure of how much the data modify, from before posterior, the odds in favor of model i against j. The marginal densities were estimated using the harmonic mean of the likelihood values, as proposed by Newton and Raftery (1994):  [38] where stands for a draw from the posterior distribution, and m is the number of samples. Gibbs Sampling. With the Gibbs sampler described previously, chains of 1,100,000 iterations were run, and samples were drawn every 100 iterations. The last 10,000 samples were used for estimating features of marginal distributions. Standard tests, proposed by Geweke (1992) and Gelman and Rubin (1992), were performed to assess convergence. RESULTS AND DISCUSSION Posterior means of deviances, deviances evaluated at posterior means of the parameters, effective number of parameters, and DIC values, expressed as differences from the DIC for model 1, are presented in Table 1. The multiple trait model (model 1) had the smallest deviance (best fit), but had a greater DIC than models 2, 3, 4, 5, and 7 because of its greater complexity. A KF structure for genetic and maternal effects was best, in the DIC sense, mainly because of its deceased complexity. When a KF was imposed on residual effects (models 6, 8, and 9), DIC values were greater than for the other models. In model 3, variances increase linearly with time, and this restriction seems to hold because data represented only early growth. In this model, the covariance between effects at times t and t + 1 is forced to be equal to the variance of such effect at time t; this can be a strong restriction in other circumstances. Posterior means of deviances, deviances at the posterior means of the parameters, and deviance information criterion (DIC) Table 1. Posterior means of deviances, deviances at the posterior means of the parameters, and deviance information criterion (DIC) Model1  Posterior means of deviances  Deviances at the posterior means  pD2  DIC3  1 (Multitrait)  52,418  32,402  20,015  —  2 (KF = a, m, e)  53,115  34,649  18,466  −853  3 (KF = a, m, e, λ = 1)4  53,312  37,164  16,148  −2,973  4 (KF = a)  52,590  34,585  18,005  −1,838  5 (KF = m)  52,469  33,530  18,938  −1,026  6 (KF = e)  53,105  31,057  22,047  2,719  7 (KF = a, m)  52,626  35,893  16,733  −3,074  8 (KF = a, e)  53,115  33,280  19,835  516  9 (KF = m, e)  53,108  31,870  21,238  1,913  Model1  Posterior means of deviances  Deviances at the posterior means  pD2  DIC3  1 (Multitrait)  52,418  32,402  20,015  —  2 (KF = a, m, e)  53,115  34,649  18,466  −853  3 (KF = a, m, e, λ = 1)4  53,312  37,164  16,148  −2,973  4 (KF = a)  52,590  34,585  18,005  −1,838  5 (KF = m)  52,469  33,530  18,938  −1,026  6 (KF = e)  53,105  31,057  22,047  2,719  7 (KF = a, m)  52,626  35,893  16,733  −3,074  8 (KF = a, e)  53,115  33,280  19,835  516  9 (KF = m, e)  53,108  31,870  21,238  1,913  1KF = Kalman filter; a = additive effect; m = maternal effect; e = residual effects. 2pD = effective number of parameters. 3Expressed as differences from model 1. 4λ = rate of change. View Large Table 1. Posterior means of deviances, deviances at the posterior means of the parameters, and deviance information criterion (DIC) Model1  Posterior means of deviances  Deviances at the posterior means  pD2  DIC3  1 (Multitrait)  52,418  32,402  20,015  —  2 (KF = a, m, e)  53,115  34,649  18,466  −853  3 (KF = a, m, e, λ = 1)4  53,312  37,164  16,148  −2,973  4 (KF = a)  52,590  34,585  18,005  −1,838  5 (KF = m)  52,469  33,530  18,938  −1,026  6 (KF = e)  53,105  31,057  22,047  2,719  7 (KF = a, m)  52,626  35,893  16,733  −3,074  8 (KF = a, e)  53,115  33,280  19,835  516  9 (KF = m, e)  53,108  31,870  21,238  1,913  Model1  Posterior means of deviances  Deviances at the posterior means  pD2  DIC3  1 (Multitrait)  52,418  32,402  20,015  —  2 (KF = a, m, e)  53,115  34,649  18,466  −853  3 (KF = a, m, e, λ = 1)4  53,312  37,164  16,148  −2,973  4 (KF = a)  52,590  34,585  18,005  −1,838  5 (KF = m)  52,469  33,530  18,938  −1,026  6 (KF = e)  53,105  31,057  22,047  2,719  7 (KF = a, m)  52,626  35,893  16,733  −3,074  8 (KF = a, e)  53,115  33,280  19,835  516  9 (KF = m, e)  53,108  31,870  21,238  1,913  1KF = Kalman filter; a = additive effect; m = maternal effect; e = residual effects. 2pD = effective number of parameters. 3Expressed as differences from model 1. 4λ = rate of change. View Large Model contrasts using the BF are shown in Table 2. Kass and Raftery (1995) suggested that BF values favoring a model greater than 1 but smaller than 3 should not be worth “more than a bare mention.” On this basis, the data did not strongly favor any specific model. The estimator of the marginal likelihood used to compute the BF is known to be numerically unstable (Newton and Raftery, 1994), but this was not observed here. Figure 1 gives values of the harmonic mean of the likelihood as the Gibbs sampler progressed, by model. Despite some jumps at earlier stages of sampling, the estimates stabilized quickly and the ranking of the models remained consistent throughout the process. Evidence provided by the data in favor of the model specified in a row against the model specified in a column Table 2. Evidence provided by the data in favor of the model specified in a row against the model specified in a column Item  Bayes factor    Model 1  Model 2  Model 3  Model 4  Model 5  Model 6  Model 7  Model 8  Model 9  Model 1    0.708  1.828  1.181  1.066  0.633  1.230  0.692  0.673  Model 2  1.412    2.582  1.668  1.506  0.895  1.737  0.978  0.950  Model 3  0.547  0.387    0.646  0.583  0.347  0.673  0.379  0.368  Model 4  0.846  0.599  1.547    0.902  0.536  1.041  0.586  0.570  Model 5  0.938  0.664  1.714  1.108    0.594  1.153  0.649  0.631  Model 6  1.578  1.117  2.884  1.864  1.682    1.940  1.093  1.062  Model 7  0.813  0.576  1.486  0.961  0.867  0.515    0.563  0.547  Model 8  1.444  1.022  2.639  1.706  1.540  0.915  1.776    0.972  Model 9  1.486  1.052  2.716  1.755  1.584  0.942  1.827  1.029    Item  Bayes factor    Model 1  Model 2  Model 3  Model 4  Model 5  Model 6  Model 7  Model 8  Model 9  Model 1    0.708  1.828  1.181  1.066  0.633  1.230  0.692  0.673  Model 2  1.412    2.582  1.668  1.506  0.895  1.737  0.978  0.950  Model 3  0.547  0.387    0.646  0.583  0.347  0.673  0.379  0.368  Model 4  0.846  0.599  1.547    0.902  0.536  1.041  0.586  0.570  Model 5  0.938  0.664  1.714  1.108    0.594  1.153  0.649  0.631  Model 6  1.578  1.117  2.884  1.864  1.682    1.940  1.093  1.062  Model 7  0.813  0.576  1.486  0.961  0.867  0.515    0.563  0.547  Model 8  1.444  1.022  2.639  1.706  1.540  0.915  1.776    0.972  Model 9  1.486  1.052  2.716  1.755  1.584  0.942  1.827  1.029    View Large Table 2. Evidence provided by the data in favor of the model specified in a row against the model specified in a column Item  Bayes factor    Model 1  Model 2  Model 3  Model 4  Model 5  Model 6  Model 7  Model 8  Model 9  Model 1    0.708  1.828  1.181  1.066  0.633  1.230  0.692  0.673  Model 2  1.412    2.582  1.668  1.506  0.895  1.737  0.978  0.950  Model 3  0.547  0.387    0.646  0.583  0.347  0.673  0.379  0.368  Model 4  0.846  0.599  1.547    0.902  0.536  1.041  0.586  0.570  Model 5  0.938  0.664  1.714  1.108    0.594  1.153  0.649  0.631  Model 6  1.578  1.117  2.884  1.864  1.682    1.940  1.093  1.062  Model 7  0.813  0.576  1.486  0.961  0.867  0.515    0.563  0.547  Model 8  1.444  1.022  2.639  1.706  1.540  0.915  1.776    0.972  Model 9  1.486  1.052  2.716  1.755  1.584  0.942  1.827  1.029    Item  Bayes factor    Model 1  Model 2  Model 3  Model 4  Model 5  Model 6  Model 7  Model 8  Model 9  Model 1    0.708  1.828  1.181  1.066  0.633  1.230  0.692  0.673  Model 2  1.412    2.582  1.668  1.506  0.895  1.737  0.978  0.950  Model 3  0.547  0.387    0.646  0.583  0.347  0.673  0.379  0.368  Model 4  0.846  0.599  1.547    0.902  0.536  1.041  0.586  0.570  Model 5  0.938  0.664  1.714  1.108    0.594  1.153  0.649  0.631  Model 6  1.578  1.117  2.884  1.864  1.682    1.940  1.093  1.062  Model 7  0.813  0.576  1.486  0.961  0.867  0.515    0.563  0.547  Model 8  1.444  1.022  2.639  1.706  1.540  0.915  1.776    0.972  Model 9  1.486  1.052  2.716  1.755  1.584  0.942  1.827  1.029    View Large Figure 1. View largeDownload slide Estimates of the marginal likelihood obtained with the harmonic mean as sample size increased. Figure 1. View largeDownload slide Estimates of the marginal likelihood obtained with the harmonic mean as sample size increased. The model comparisons based on DIC indicate that embedding a KF into the standard MM may be useful for modeling random effects in longitudinal analysis of quantitative traits. Model 7 had dynamic additive and maternal effects and has a smaller number of parameters than the standard specification. Highly parameterized models impose a heavy computational burden, which increases with the number of measures. A multiple-trait model contains T(T + 1)/2 dispersion parameters. A random regression model with r coefficients comprises r(r + 1)/2 dispersion parameters. The KF specification encloses T + (T − 1) dispersion parameters. Even if r is smaller than T, the order of (co)variance matrices in a random regression model can be greater than in KF. In the example presented, only 6 time points were considered, but studies involving growth, lactation curves, or biomarkers over time may involve more intense recording. In these cases, more parsimonious models such as KF could be helpful, although when the data set comprises a large number of measures, such as test day milk yield, KF requires the choice of a subset of the available data points. The subset should represent the entire trajectory. In such situations, models that reduce data dimensionality, such as random regression models, can be more cost-effective. Conceptually compared with random regressions, KF does not require a choice of orthogonal functions and is not susceptible to the “edge-of-range” problems frequently observed when fitting polynomial regressions (Meyer, 2005). The fact that the models differed in performance, at least with respect to DIC, does not imply that they would work differently in predicting future data or additive genetic values. For example, suppose that 10% of the individuals appearing in the additive relationship matrix in the cattle growth data are selected as parents based on posterior means of their additive effects; genetic effects favoring faster growth are typically desirable. Table 3 shows the percentage of individuals selected in a standard MM analysis that would also be selected if other models were applied for selection at each of the 6 time points considered. A high degree of overlap was observed. However, even if 80%, say, of individuals selected as parents were the same for all models, selection decisions would differ. This has an important implication in animal breeding. Percentage of individuals that would be selected (considering the top 10% of estimated additive genetic effects) in common with a standard multiple trait and Kalman filter model Table 3. Percentage of individuals that would be selected (considering the top 10% of estimated additive genetic effects) in common with a standard multiple trait and Kalman filter model Model1  Time point    1  2  3  4  5  6  2 (KF = a, m, e)  95  90  92  91  89  88  3 (KF = a, m, e, λ = 1)2  81  80  76  79  78  82  4 (KF = a)  88  84  86  85  87  88  5 (KF = m)  94  91  91  92  92  93  6 (KF = e)  88  87  85  84  87  85  7 (KF = a, m)  91  87  88  88  90  90  8 (KF = a, e)  80  78  76  78  77  81  9 (KF = m, e)  89  86  87  86  90  87  Model1  Time point    1  2  3  4  5  6  2 (KF = a, m, e)  95  90  92  91  89  88  3 (KF = a, m, e, λ = 1)2  81  80  76  79  78  82  4 (KF = a)  88  84  86  85  87  88  5 (KF = m)  94  91  91  92  92  93  6 (KF = e)  88  87  85  84  87  85  7 (KF = a, m)  91  87  88  88  90  90  8 (KF = a, e)  80  78  76  78  77  81  9 (KF = m, e)  89  86  87  86  90  87  1KF = Kalman filter; a = additive effect; m = maternal effect; e = residual effects. 2λ = rate of change. View Large Table 3. Percentage of individuals that would be selected (considering the top 10% of estimated additive genetic effects) in common with a standard multiple trait and Kalman filter model Model1  Time point    1  2  3  4  5  6  2 (KF = a, m, e)  95  90  92  91  89  88  3 (KF = a, m, e, λ = 1)2  81  80  76  79  78  82  4 (KF = a)  88  84  86  85  87  88  5 (KF = m)  94  91  91  92  92  93  6 (KF = e)  88  87  85  84  87  85  7 (KF = a, m)  91  87  88  88  90  90  8 (KF = a, e)  80  78  76  78  77  81  9 (KF = m, e)  89  86  87  86  90  87  Model1  Time point    1  2  3  4  5  6  2 (KF = a, m, e)  95  90  92  91  89  88  3 (KF = a, m, e, λ = 1)2  81  80  76  79  78  82  4 (KF = a)  88  84  86  85  87  88  5 (KF = m)  94  91  91  92  92  93  6 (KF = e)  88  87  85  84  87  85  7 (KF = a, m)  91  87  88  88  90  90  8 (KF = a, e)  80  78  76  78  77  81  9 (KF = m, e)  89  86  87  86  90  87  1KF = Kalman filter; a = additive effect; m = maternal effect; e = residual effects. 2λ = rate of change. View Large Quantitative genetic studies are frequently concerned with variation between individuals, whereas other areas focus on within-subject covariances. A KF for longitudinal traits provides estimates of genetic values for the entire range of time points, assuming that time-specific random effects are affecting variation between and within individuals. In polynomial regressions, the dispersion parameters are static, and genetic covariance functions reflect variation in functions of time. Structured antedependence models were also suggested for the genetic analysis of longitudinal traits (Jaffrézic et al., 2004). Breeding values were modeled as standard autoregressive variables without a constraint on the regression coefficients (i.e., they do not have to be between −1 and 1), and a polynomial function of time was used to model the variances of shocks. To relate these variances with the genetic (co)variance matrix, a Cholesky decomposition is necessary, and as implemented, the T × T covariance matrix is required. Piepho and Ogutu (2007) showed how to formulate state-space models with a known transition matrix in a mixed model using maximum likelihood. They argued that, when there are unknowns in the transition matrix (e.g., rates of change in the state equations), it is not always possible to obtain a mixed model representation that is easy to fit. Here, it was demonstrated that incorporating a KF in a mixed linear model under a Bayesian framework is feasible when there are unknown parameters in the transition matrix. Higher order regressions could be considered if first-order processes in the state equations are inadequate. If necessary, restrictions imposed on the (co)variance matrices could be also lifted, facilitating interpretation of effect × time interactions arising in complex designs for repeated measures. Different models embedding the KF in a mixed model were demonstrated to provide a more parsimonious (co)variance structure than a standard multitrait specification for the quantitative genetic analysis of longitudinal data. However, a formal comparison between KF and other approaches would be of interest. LITERATURE CITED Casella G. George I. E. 1992. Explaining the Gibbs sampler. Am. Stat.  46: 167– 174. Forni S. Piles M. Blasco A. Varona L. Oliveira H. N. Lôbo R. B. Albuquerque L. G. 2007. Analysis of beef cattle longitudinal data applying a non-linear model. J. Anim. Sci.  85: 3189– 3197. [PubMed] Google Scholar CrossRef Search ADS PubMed  Gelman A. Rubin D. B. 1992. Inference from iterative simulation using multiple sequences. Stat. Sci.  7: 457– 472. Google Scholar CrossRef Search ADS   Geweke, J. 1992. Evaluating the accuracy of sampling-based approaches to the calculation of posterior moments (with discussion). Pages 169–193 in Bayesian Statistics 4.  Oxford Univ. Press, Oxford, UK. Harvey, A. C. 1993. Time Series Models.  MIT Press, Cambridge, MA. Jaffrézic F. Venot E. Laloe D. Vinet A. Renand G. 2004. Use of structured antedependence models for the genetic analysis of growth curves. J. Anim. Sci.  82: 3465– 3473. [PubMed] Google Scholar CrossRef Search ADS PubMed  Kalman R. E. 1960. A new approach to linear filtering and predictive problems. J. Basic Eng.  82: 35– 45. Google Scholar CrossRef Search ADS   Kass R. E. Raftery A. E. 1995. Bayes factors. J. Am. Stat. Assoc.  90: 773– 795. Google Scholar CrossRef Search ADS   Lindsey, J. K. 1993. Models for Repeated Measurements. Oxford Statistical Science Series.  Clarendon Press, Oxford, UK. Lôbo, R. B., L. A. F. Bezerra, P. S. Barros, C. U. Magnabosco, L. G. Albuquerque, J. A. G. Bergmann, R. D. Sainz, and H. N. Oliveira 2007. Avaliação Genética de Touros e Matrizes da Raça Nelore: Sumário 2007. Associação Nacional de Criadores e Pesquisadores,  Ribeirão Preto, São Paulo, Brazil. Meinhold R. J. Singpurwalla N. D. 1983. Understanding the Kalman filter. Am. Stat.  37: 123– 127. Meyer K. 2005. Advances in methodology for random regression analyses. Aust. J. Exp. Agric.  45: 847– 858. Google Scholar CrossRef Search ADS   Meyer K. Hill W. G. 1997. Estimation of genetic and phenotypic covariance functions for longitudinal data by Restricted Maximum Likelihood. Livest. Prod. Sci.  47: 185– 200. Google Scholar CrossRef Search ADS   Newton M. A. Raftery A. E. 1994. Approximate Bayesian inference by the weighted likelihood bootstrap (with discussion). J. R. Stat. Soc. B  56: 1– 48. Piepho H. P. Ogutu J. O. 2007. Simple state-space models in a mixed model framework. Am. Stat.  61: 224– 232. Google Scholar CrossRef Search ADS   Sallas W. M. Harville D. A. 1981. Best linear recursive estimation for mixed linear models. J. Am. Stat. Assoc.  76: 860– 869. Google Scholar CrossRef Search ADS   Schaeffer L. R. 2004. Application of random regression models in animal breeding. Livest. Prod. Sci.  86: 35– 45. Google Scholar CrossRef Search ADS   Shumway, R. H., and D. S. Stoffer 2000. Times Series Analysis and Its Applications.  Springer-Verlag, New York, NY. Google Scholar CrossRef Search ADS   Sorensen, D., and D. Gianola 2002. Likelihood, Bayesian, and MCMC Methods in Quantitative Genetics.  Springer-Verlag, New York, NY. Google Scholar CrossRef Search ADS   Spiegelhalter D. J. Best N. G. Carlin B. P. van der Linde A. 2002. Bayesian measures of model complexity and fit. J. R. Stat. Soc. B  64: 538– 639. Google Scholar CrossRef Search ADS   Varona L. Moreno C. García Cortés L. A. Altarriba J. 1997. Multiple trait genetic analysis of underlying biological variables of production functions. Livest. Prod. Sci.  47: 201– 209. Google Scholar CrossRef Search ADS   Walters, C. 1986. Adaptive Management of Renewable Resources.  Macmillan, New York, NY. American Society of Animal Science

Forni, S.;Gianola, D.;Rosa, G. J. M.

doi: 10.2527/jas.2008-1515pmid: 19684263

ABSTRACT A set of analyses using a multiple-trait model (model 1) and dynamic models for the evaluation of beef cattle growth is presented. All models contained additive direct and maternal environmental effects, as well as contemporary groups as nuisance parameters. The predictive ability of models at different parts of the growth trajectory was compared. Body weight records of 6,856 Nelore animals taken at 6 different ages (birth to 540 d) were used. Different models embedding a Kalman filter (KF) into a mixed model representation were fitted. Model 2 assumed that additive, maternal, and residual effects changed over time according to a linear autoregressive process. Model 3 was similar to model 2, but all regression coefficients were set to 1. In model 4, KF was applied only to direct genetic and maternal environmental effects. A leave-one-out cross-validation check was used to assess the predictive ability of models. Estimates of additive variance were similar in the analysis with models 1, 3, and 4 for all ages. Posterior means of maternal components increased slightly after birth and decreased after 135 d of age. Posterior means of additive rates of change were close to 1 at almost all time points, irrespective of the model. The posterior means of residual rates of change, which varied from 0.096 to 0.529, did not support the restrictions that regression coefficients were equal to 1 imposed by model 3. Estimates of additive and maternal correlations obtained with dynamic models were larger than those from a multivariate model. Model 3 produced different phenotypic correlations. Models 2 and 4 had better predictive ability than the multivariate specification. Model 3 predicted the data very poorly, and errors increased markedly with age. The KF can be a useful tool for structuring (co)variance matrices without reducing dimensionality. This model provided accurate predictions and plausible estimates of (co)variance components. Moreover, KF is a flexible specification, because a multivariate structure can be used for some random effects, whereas a dynamic feature can be incorporated for others. INTRODUCTION Many traits of interest in quantitative genetics consist of points measured at a series of times on a subject. These traits can change up or down between time points. Models and methods for analyzing records taken along some time scale have deserved attention in both evolutionary genetics and animal breeding (Hill, 1998). Longitudinal records of livestock have been analyzed either as repeated measurements of the same trait (repeatability model), or different correlated traits, or function-valued traits described by linear or nonlinear regressions (Mrode, 1996; Meyer and Hill, 1997; Varona et al., 1997). Repeatability models may not account correctly for the covariance structure between time points, and multiple-trait models may have an excessive number of parameters. Regression models present problems related to poor polynomial approximations, and covariance functions reflect variation in regression coefficients. A hierarchical scheme embedding the Kalman filter (KF; Kalman, 1960) into a mixed model representation that allows random effects to account for changes over time between and within individuals (Forni et al., 2009; this issue). Comparisons between KF and a standard multitrait model with respect to goodness of fit favored the former. This article presents a set of analyses using a multiple-trait model and the KF to evaluate the growth of beef cattle. The focus is on comparing the predictive ability of the models at different points in the growth trajectory. MATERIALS AND METHODS Animal Care and Use Committee approval was not obtained for this study because data were obtained from an existing database (Lôbo et al., 2007). The data set included records for BW taken from birth to 540 d of age in Nelore beef cattle from Brazil, provided by ANCP (Associação Nacional de Criadores e Pesquisadores). Body weight records of 6,856 animals, progeny of 326 sires and 3,898 dams, at each of 6 equally spaced time points were analyzed. Pedigree information was obtained for all available generations, resulting in a total of 15,946 animals in the additive relationship matrix. A detailed description of data edits is given in Forni et al. (2009). Models Let yt denote the BW records at times t = 1, 2, …, 6. A standard multiple-trait model (model 1) was fitted:  ytXβtZatWmtet [1] where βt represents time-specific effects of 154 contemporary groups; at stands for the vector of additive genetic effects at time t; mt denotes a vector of maternal effects at time t, assumed to be environmental; et represents the vector of residuals; and X, Z, and W are known incidence matrices. The additive, maternal, and residual effects were assumed to be independently distributed random vectors, following the normal distributions and respectively. Here, A is the additive relationship matrix; I stands for an identity matrix of appropriate order; G0 is a 6 × 6 (co)variance matrix between additive effects operating at the 6 time points, M0 is a 6 × 6 (co)variance matrix between maternal effects, and R0 is a matrix of the same dimensions, containing (co)variances between residual effects. This model has 63 dispersion parameters. Body weights at all times were assumed to be influenced by maternal effects, because previous analyses of the same data indicated that substantial maternal effects on BW until weaning are carried over through maturity (Forni et al., 2007). Assuming that genetic, maternal, and residual effects are time-dependent random variables, a KF postulates that these evolve over time according to  [2] where ut represents either additive, maternal, or residual effects, each one having time-specific rates of change and denotes a vector of random variables, called random shocks. The additive, maternal, and residual random shocks were assumed to be independently distributed, following the normal distributions and respectively; Δa, Δm, and Δe are 6 × 6 diagonal matrices with elements and (t = 1, 2, …, 6). Combining assumptions of the standard model (Eq. 1) with those in Eq. 2 leads to the joint distribution   [3] where δa, δm, and δe are the vectors of additive, maternal, and residual random shocks at all times other than 1. With Eq. 2, random effects at any time can be calculated given the random shocks and the rates of change The genetic, maternal, and residual variances at any time t are defined by   [4] Moreover, covariances between any 2 time points t and t′ (t < t′) are  [5] The preceding indicates that KF introduces a time-variant structure for random effects, and that covariances can be expressed in terms of variance components and rates of change. Therefore, the (co)variance structure is modeled with a smaller number of parameters than in a multiple-trait analysis. A model with KF describing additive, maternal, and residual effects (model 2) was used. This model has 33 dispersion parameters, including the rates of change. A third model (model 3) was considered similar to model 2, but with all regression coefficients set equal to 1, leading to 18 dispersion parameters. In this model, the variances of each effect are forced to increase with time. Figure 1 shows phenotypic SD and CV for different ages. Although the relative variation is similar, the SD clearly increase with age. A fourth model (model 4), in which KF was applied only for genetic direct and environmental maternal effects, was fitted. A standard multitrait structure was assumed for the residual (co)variance matrix, resulting in 43 dispersion parameters. Figure 1. View largeDownload slide Phenotypic SD (left axis) and CV (right axis) for BW records at 6 different ages. Figure 1. View largeDownload slide Phenotypic SD (left axis) and CV (right axis) for BW records at 6 different ages. Model Comparison The performance of the 4 models was evaluated in terms of predictive ability, using a leave-one-out cross-validation approach (Gelfand et al., 1992). When data have a multilevel structure, it is proper to cross-validate by leaving out data individually or in clusters. The leave-one-out cross-validation has 2 main advantages. First, all other data are used in the estimation of parameters. Therefore, estimation errors tend to be smaller. In a k-fold cross-validation, the variance of the estimates is reduced as k is increased. Leave-one-out is the most efficient data partition to reduce estimation errors. The second advantage is avoiding bias introduced by relying on any particular division of the data set. Bias can be introduced in random partitioning of animal breeding data because of family structures. Let yit represent the record of animal i at time t. Each observed value yit was compared with its prediction obtained using all data points y_it other than yit. The posterior predictive density of conditionally on y_it and on model Mr (r = 1, 2, 3, and 4) is   [6] where θr is the vector of parameters of model Mr, and p(θr | y_it, Mr) is the posterior density of θr based on y_it. With T data points, there are T posterior predictive densities (Eq. 6). Note that yit does not enter into the prediction of If the model holds, the observed value yit should be a random sample from its respective posterior predictive density The observed value can be compared with its posterior predictive density as an outlier diagnostic. Such comparison, when carried out for different models, may disclose which model predicts observed values better. Here, the average squared difference between observed value yit and predicted values was used as the predictive ability diagnostic:   [7] where was estimated by averaging over Markov chain Monte Carlo (MCMC) samples, that is, drawing θr from p(θr | y_it, Mr) and from as given under the integral sign in Eq. 6. The conditional distributions of records for different animals are independent, given θr, so that   [8] where yi is the vector of data pertaining to animal i (i = 1, 2, …, n). Considering Eq. 1 and the assumptions regarding residual (co)variance, the distribution of the data vector for each animal is normal, with density  [9] where and are vectors of nuisance parameters, genetic, and maternal effects, respectively, acting on animal i, and Xi, Zi, and Wi are incidence matrices. The conditional expectation of a single observation was   [10] where and as before, are the nuisance parameters, genetic and maternal effects on time t; xi, zi, and wi are incidence vectors; represents the phenotypic mean at time t′; is the residual variance at time t′; and is the residual covariance between times t and t′. Following Eq. 10, the expectation of a yet-to-be-observed value at time t is   [11] where and are estimated expectations at time t, respectively, inferring without including yit; is the residual at time t′ for animal i, inferred without yit; is the estimate of residual variance at time t′; and is the estimated residual covariance between times t and t′, both obtained without including yit′ in the analysis. The calculations involved in Eq. 6 and 7 are analytically intractable and computationally demanding. Further, estimation of requires an MCMC evaluation of the integral:  [12] With the Bayes theorem,  [13] Substituting in Eq. 12 yields  [14] Suppose m samples can be obtained from the distribution Let the samples be For large m, a consistent estimator of the posterior expectation is given by the ratio   [15] Inference and MCMC Implementation Normal prior distributions were assumed for nuisance parameters and random variables. In the multitrait model, the prior distributions of (co)variance matrices were scaled inverted Wishart. In KF models, independent scaled inverted chi-square distributions were used as priors for the variance components. The priors and fully conditional distributions are presented in Forni et al. (2009). Chains of 1,100,000 samples were run and the first 100,000 were discarded as burn-in; 1 out of each 100 successive samples was retained. The last 10,000 samples kept were used to estimate features of marginal distributions. Convergence was assessed by visual inspection of trace plots, as well as by the diagnostic tests of Geweke (1992) and Gelman and Rubin (1992). Autocorrelations between samples and estimates of Monte Carlo error (Geyer, 1992) were calculated. RESULTS AND DISCUSSION (Co)variance Components The lower and upper bounds of the highest posterior density intervals of size 95% of variance components by age and model are presented in Table 1. All marginal densities were sharp and posterior means are plotted in Figure 2. These values are in the range of estimates reported in other studies of Nelore growth data (e.g., Albuquerque and Meyer, 2001a,b). The estimates of additive variance obtained with models 1, 3, and 4 were fairly similar for all ages, but estimates from model 2 (KF applied to a, m, and e) were distinct. Conversely, models 1, 2, and 4 yielded close estimates of maternal and residual variances, whereas estimates from model 3 (KF applied to a, m, and e, but with = 1) were the most different. The phenotypic variances estimated with models 1, 2, and 4 were similar. Model 2 yielded larger estimates of additive variance and smaller estimates of residual variance, markedly so for the last age at recording (495 d). Posterior means of all variance components at 45 d were very similar; differences appeared afterward and increased with age. The increases in variance components reflect restrictions imposed by the alternative models. Model 3 yielded estimates that increased more sharply than others. However, the strong restrictions imposed by this model did not result in additive variance estimates much different from those obtained using other models, as noted. Lower and upper bounds of the highest posterior density intervals of 95% probability (HPD95%) for variance components of BW, by model and age Table 1. Lower and upper bounds of the highest posterior density intervals of 95% probability (HPD95%) for variance components of BW, by model and age Age, d  Model 1 (multitrait)    Genetic  Maternal  Residual  Phenotypic  45  (83 : 103)  (58 : 70)  (101 : 116)  (257 : 275)  135  (100 : 125)  (69 : 86)  (107 : 126)  (297 : 319)  225  (163 : 218)  (100 : 128)  (107 : 126)  (527 : 569)  315  (209 : 279)  (135 : 177)  (287 : 338)  (655 : 706)  405  (264 : 351)  (135 : 177)  (328 : 392)  (791 : 854)  495  (326 : 434)  (152 : 203)  (408 : 485)  (966 : 1,042)    Model 2 (KF = a, m, e)1  Age, d  Model 1 (multitrait)    Genetic  Maternal  Residual  Phenotypic  45  (83 : 103)  (58 : 70)  (101 : 116)  (257 : 275)  135  (100 : 125)  (69 : 86)  (107 : 126)  (297 : 319)  225  (163 : 218)  (100 : 128)  (107 : 126)  (527 : 569)  315  (209 : 279)  (135 : 177)  (287 : 338)  (655 : 706)  405  (264 : 351)  (135 : 177)  (328 : 392)  (791 : 854)  495  (326 : 434)  (152 : 203)  (408 : 485)  (966 : 1,042)    Model 2 (KF = a, m, e)1  Age, d  Genetic  Maternal  Residual  Phenotypic  45  (83 : 104)  (49 : 60)  (96 : 110)  (241 : 260)  135  (160 : 186)  (71 : 88)  (68 : 80)  (313 : 338)  225  (311 : 365)  (121 : 153)  (136 : 162)  (600 : 649)  315  (437 : 513)  (135 : 175)  (150 : 181)  (764 : 829)  405  (493 : 588)  (183 : 236)  (191 : 236)  (923 : 1,002)  495  (738 : 852)  (210 : 274)  (158 : 206)  (1,171 : 1,271)    Model 3 (KF = a, m, e, = 1)2  Age, d  Genetic  Maternal  Residual  Phenotypic  45  (83 : 104)  (49 : 60)  (96 : 110)  (241 : 260)  135  (160 : 186)  (71 : 88)  (68 : 80)  (313 : 338)  225  (311 : 365)  (121 : 153)  (136 : 162)  (600 : 649)  315  (437 : 513)  (135 : 175)  (150 : 181)  (764 : 829)  405  (493 : 588)  (183 : 236)  (191 : 236)  (923 : 1,002)  495  (738 : 852)  (210 : 274)  (158 : 206)  (1,171 : 1,271)    Model 3 (KF = a, m, e, = 1)2  Age, d  Genetic  Maternal  Residual  Phenotypic  45  (71 : 93)  (48 : 59)  (103 : 120)  (237 : 256)  135  (161 : 194)  (104 : 122)  (253 : 279)  (543 : 571)  225  (236 : 278)  (153 : 174)  (404 : 437)  (824 : 859)  315  (334 : 384)  (221 : 249)  (598 : 640)  (1,191 : 1236)  405  (415 : 472)  (282 : 314)  (791 : 838)  (1,531 : 1,582)  495  (513 : 577)  (350 : 385)  (1,032 :1,088)  (1,944 : 2,002)    Model 4 (KF = a, m)  Age, d  Genetic  Maternal  Residual  Phenotypic  45  (71 : 93)  (48 : 59)  (103 : 120)  (237 : 256)  135  (161 : 194)  (104 : 122)  (253 : 279)  (543 : 571)  225  (236 : 278)  (153 : 174)  (404 : 437)  (824 : 859)  315  (334 : 384)  (221 : 249)  (598 : 640)  (1,191 : 1236)  405  (415 : 472)  (282 : 314)  (791 : 838)  (1,531 : 1,582)  495  (513 : 577)  (350 : 385)  (1,032 :1,088)  (1,944 : 2,002)    Model 4 (KF = a, m)  Age, d  Genetic  Maternal  Residual  Phenotypic  45  (67 : 88)  (60 : 71)  (102 : 118)  (245 : 264)  135  (106 : 126)  (73 : 91)  (98 : 116)  (295 : 316)  225  (206 : 252)  (126 : 153)  (196 : 230)  (560 : 603)  315  (282 : 337)  (136 : 172)  (240 : 284)  (704 : 752)  405  (362 : 448)  (184 : 234)  (259 : 315)  (872 : 935)  495  (455 : 557)  (200 : 261)  (330 : 394)  (1,055 : 1,142)  Age, d  Genetic  Maternal  Residual  Phenotypic  45  (67 : 88)  (60 : 71)  (102 : 118)  (245 : 264)  135  (106 : 126)  (73 : 91)  (98 : 116)  (295 : 316)  225  (206 : 252)  (126 : 153)  (196 : 230)  (560 : 603)  315  (282 : 337)  (136 : 172)  (240 : 284)  (704 : 752)  405  (362 : 448)  (184 : 234)  (259 : 315)  (872 : 935)  495  (455 : 557)  (200 : 261)  (330 : 394)  (1,055 : 1,142)  1KF = Kalman filter; a = additive effect; m = maternal effect, e = residual effect. 2 = rate of change. View Large Table 1. Lower and upper bounds of the highest posterior density intervals of 95% probability (HPD95%) for variance components of BW, by model and age Age, d  Model 1 (multitrait)    Genetic  Maternal  Residual  Phenotypic  45  (83 : 103)  (58 : 70)  (101 : 116)  (257 : 275)  135  (100 : 125)  (69 : 86)  (107 : 126)  (297 : 319)  225  (163 : 218)  (100 : 128)  (107 : 126)  (527 : 569)  315  (209 : 279)  (135 : 177)  (287 : 338)  (655 : 706)  405  (264 : 351)  (135 : 177)  (328 : 392)  (791 : 854)  495  (326 : 434)  (152 : 203)  (408 : 485)  (966 : 1,042)    Model 2 (KF = a, m, e)1  Age, d  Model 1 (multitrait)    Genetic  Maternal  Residual  Phenotypic  45  (83 : 103)  (58 : 70)  (101 : 116)  (257 : 275)  135  (100 : 125)  (69 : 86)  (107 : 126)  (297 : 319)  225  (163 : 218)  (100 : 128)  (107 : 126)  (527 : 569)  315  (209 : 279)  (135 : 177)  (287 : 338)  (655 : 706)  405  (264 : 351)  (135 : 177)  (328 : 392)  (791 : 854)  495  (326 : 434)  (152 : 203)  (408 : 485)  (966 : 1,042)    Model 2 (KF = a, m, e)1  Age, d  Genetic  Maternal  Residual  Phenotypic  45  (83 : 104)  (49 : 60)  (96 : 110)  (241 : 260)  135  (160 : 186)  (71 : 88)  (68 : 80)  (313 : 338)  225  (311 : 365)  (121 : 153)  (136 : 162)  (600 : 649)  315  (437 : 513)  (135 : 175)  (150 : 181)  (764 : 829)  405  (493 : 588)  (183 : 236)  (191 : 236)  (923 : 1,002)  495  (738 : 852)  (210 : 274)  (158 : 206)  (1,171 : 1,271)    Model 3 (KF = a, m, e, = 1)2  Age, d  Genetic  Maternal  Residual  Phenotypic  45  (83 : 104)  (49 : 60)  (96 : 110)  (241 : 260)  135  (160 : 186)  (71 : 88)  (68 : 80)  (313 : 338)  225  (311 : 365)  (121 : 153)  (136 : 162)  (600 : 649)  315  (437 : 513)  (135 : 175)  (150 : 181)  (764 : 829)  405  (493 : 588)  (183 : 236)  (191 : 236)  (923 : 1,002)  495  (738 : 852)  (210 : 274)  (158 : 206)  (1,171 : 1,271)    Model 3 (KF = a, m, e, = 1)2  Age, d  Genetic  Maternal  Residual  Phenotypic  45  (71 : 93)  (48 : 59)  (103 : 120)  (237 : 256)  135  (161 : 194)  (104 : 122)  (253 : 279)  (543 : 571)  225  (236 : 278)  (153 : 174)  (404 : 437)  (824 : 859)  315  (334 : 384)  (221 : 249)  (598 : 640)  (1,191 : 1236)  405  (415 : 472)  (282 : 314)  (791 : 838)  (1,531 : 1,582)  495  (513 : 577)  (350 : 385)  (1,032 :1,088)  (1,944 : 2,002)    Model 4 (KF = a, m)  Age, d  Genetic  Maternal  Residual  Phenotypic  45  (71 : 93)  (48 : 59)  (103 : 120)  (237 : 256)  135  (161 : 194)  (104 : 122)  (253 : 279)  (543 : 571)  225  (236 : 278)  (153 : 174)  (404 : 437)  (824 : 859)  315  (334 : 384)  (221 : 249)  (598 : 640)  (1,191 : 1236)  405  (415 : 472)  (282 : 314)  (791 : 838)  (1,531 : 1,582)  495  (513 : 577)  (350 : 385)  (1,032 :1,088)  (1,944 : 2,002)    Model 4 (KF = a, m)  Age, d  Genetic  Maternal  Residual  Phenotypic  45  (67 : 88)  (60 : 71)  (102 : 118)  (245 : 264)  135  (106 : 126)  (73 : 91)  (98 : 116)  (295 : 316)  225  (206 : 252)  (126 : 153)  (196 : 230)  (560 : 603)  315  (282 : 337)  (136 : 172)  (240 : 284)  (704 : 752)  405  (362 : 448)  (184 : 234)  (259 : 315)  (872 : 935)  495  (455 : 557)  (200 : 261)  (330 : 394)  (1,055 : 1,142)  Age, d  Genetic  Maternal  Residual  Phenotypic  45  (67 : 88)  (60 : 71)  (102 : 118)  (245 : 264)  135  (106 : 126)  (73 : 91)  (98 : 116)  (295 : 316)  225  (206 : 252)  (126 : 153)  (196 : 230)  (560 : 603)  315  (282 : 337)  (136 : 172)  (240 : 284)  (704 : 752)  405  (362 : 448)  (184 : 234)  (259 : 315)  (872 : 935)  495  (455 : 557)  (200 : 261)  (330 : 394)  (1,055 : 1,142)  1KF = Kalman filter; a = additive effect; m = maternal effect, e = residual effect. 2 = rate of change. View Large Figure 2. View largeDownload slide Marginal posterior means of variance components for BW, by age and model. Figure 2. View largeDownload slide Marginal posterior means of variance components for BW, by age and model. The shed lines of Figure 3 show the marginal posterior means of the dispersion parameters for random shocks. Dashed lines depict the percentage of variances attributable to random shocks within each effect. Estimates of variance of additive and maternal shocks from different models agreed well for all ages. However, the posterior means of residual shock variances at the first and last ages at measurements are fairly different. The relative contribution of additive genetic shocks to total genetic variation decreased with time; total genetic variation also includes the previous variance and λ coefficients, as shown in Eq. 4. At the second time point analyzed (135 d), the additive shock variance represented more than 50% of the total additive variance. In this case, greater additive shocks are indicative of greater genetic potential for faster growth between 2 time points. In a KF model, predicted shock can be used as a selection criterion for increasing growth rate at early ages. Combining information provided by additive effects and additive shocks can be useful for modifying growth curve shapes. However, growth rates and BW are strongly correlated (Forni et al., 2007), which introduces constraints on selective forces. Figure 3. View largeDownload slide Marginal posterior means of random shocks variance (solid line) and percentage of variances attributable to random shocks (dashed line) for BW, by age and model. Figure 3. View largeDownload slide Marginal posterior means of random shocks variance (solid line) and percentage of variances attributable to random shocks (dashed line) for BW, by age and model. Marginal posterior means of heritability and of the relative contribution of maternal effects to variance are displayed in Figure 4. Heritability estimates obtained with models 1, 3, and 4 are within the range described in the literature for Nelore cattle (Mercadante et al., 1995). The estimates obtained with model 2, however, were larger than reported estimates. As discussed before, model 2 produced larger estimates of additive variance than other models. The proportions of phenotypic variances attributable to maternal effects were similar for all models. These increased slightly after birth and decreased after 135 d of age, as also reported by Meyer (1992a) and Albuquerque and Meyer (2001b). A similar pattern has been described for additive maternal effects on cattle growth (Meyer et al., 1993; Eler et al., 1995). Difficulties in separating additive and environmental maternal effects using field data are well known (Willham, 1980; Meyer, 1992b). When only 1 of these effects (genetic or permanent environment) is considered, most of the maternal variation is likely to be taken into account (Meyer, 1992a). Figure 4. View largeDownload slide Marginal posterior means of heritability and the relative contribution of maternal effects to variance in BW, by age and model. Figure 4. View largeDownload slide Marginal posterior means of heritability and the relative contribution of maternal effects to variance in BW, by age and model. Features of marginal densities of regression coefficients (i.e., rates of change of random effects in KF models) are shown in Table 2. The marginal densities for ages 135 and 495 d, obtained with model 2, are shown in Figure 5. These were unimodal and symmetric, suggesting reasonable and precise inference. Posterior means of additive rates of change were close to 1 for almost all points, irrespective of the model. The posterior means of residual rates of change, which varied from 0.096 to 0.529, clearly did not support the restrictions imposed by model 3 Features of marginal posterior distributions of rates of change, by age for models 2 and 4 Table 2. Features of marginal posterior distributions of rates of change, by age for models 2 and 4   Model 2 (KF = a, m, e)1    Genetic  Maternal  Residual  Age, d  Mean  HPD95%2  Mean  HPD95%  Mean  HPD95%  135  0.910  (0.82 : 1.00)  0.660  (0.57 : 0.75)  0.096  (0.05 : 0.14)  225  1.244  (1.18 : 1.30)  1.058  (0.98 : 1.13)  0.450  (0.37 : 0.53)  315  1.063  (1.02 : 1.11)  0.805  (0.73 : 0.87)  0.372  (0.31 : 0.42)  405  0.990  (0.95 : 1.03)  0.979  (0.92 : 1.04)  0.529  (0.46 : 0.59)  495  1.145  (1.11 : 1.19)  0.920  (0.86 : 0.98)  0.346  (0.28 : 0.41)    Model 4 (KF = a, m)    Model 2 (KF = a, m, e)1    Genetic  Maternal  Residual  Age, d  Mean  HPD95%2  Mean  HPD95%  Mean  HPD95%  135  0.910  (0.82 : 1.00)  0.660  (0.57 : 0.75)  0.096  (0.05 : 0.14)  225  1.244  (1.18 : 1.30)  1.058  (0.98 : 1.13)  0.450  (0.37 : 0.53)  315  1.063  (1.02 : 1.11)  0.805  (0.73 : 0.87)  0.372  (0.31 : 0.42)  405  0.990  (0.95 : 1.03)  0.979  (0.92 : 1.04)  0.529  (0.46 : 0.59)  495  1.145  (1.11 : 1.19)  0.920  (0.86 : 0.98)  0.346  (0.28 : 0.41)    Model 4 (KF = a, m)    Genetic  Maternal  Residual  Age, d  Mean  HPD95%  Mean  HPD95%  Mean  HPD95%  135  0.628  (0.52 : 0.72)  0.640  (0.56 : 0.71)  —  —  225  1.150  (0.70 : 1.23)  1.054  (0.98 : 1.12)  —  —  315  0.973  (0.91 : 1.03)  0.792  (0.73 : 0.86)  —  —  405  1.026  (0.97 : 1.08)  0.981  (0.92 : 1.05)  —  —  495  1.005  (0.94 : 1.07)  0.894  (0.83 : 0.95)  —      Genetic  Maternal  Residual  Age, d  Mean  HPD95%  Mean  HPD95%  Mean  HPD95%  135  0.628  (0.52 : 0.72)  0.640  (0.56 : 0.71)  —  —  225  1.150  (0.70 : 1.23)  1.054  (0.98 : 1.12)  —  —  315  0.973  (0.91 : 1.03)  0.792  (0.73 : 0.86)  —  —  405  1.026  (0.97 : 1.08)  0.981  (0.92 : 1.05)  —  —  495  1.005  (0.94 : 1.07)  0.894  (0.83 : 0.95)  —    1KF = Kalman filter; a = additive effect; m = maternal effect; e = residual effect. 2HPD95% = highest posterior density interval with 95% probability constant. View Large Table 2. Features of marginal posterior distributions of rates of change, by age for models 2 and 4   Model 2 (KF = a, m, e)1    Genetic  Maternal  Residual  Age, d  Mean  HPD95%2  Mean  HPD95%  Mean  HPD95%  135  0.910  (0.82 : 1.00)  0.660  (0.57 : 0.75)  0.096  (0.05 : 0.14)  225  1.244  (1.18 : 1.30)  1.058  (0.98 : 1.13)  0.450  (0.37 : 0.53)  315  1.063  (1.02 : 1.11)  0.805  (0.73 : 0.87)  0.372  (0.31 : 0.42)  405  0.990  (0.95 : 1.03)  0.979  (0.92 : 1.04)  0.529  (0.46 : 0.59)  495  1.145  (1.11 : 1.19)  0.920  (0.86 : 0.98)  0.346  (0.28 : 0.41)    Model 4 (KF = a, m)    Model 2 (KF = a, m, e)1    Genetic  Maternal  Residual  Age, d  Mean  HPD95%2  Mean  HPD95%  Mean  HPD95%  135  0.910  (0.82 : 1.00)  0.660  (0.57 : 0.75)  0.096  (0.05 : 0.14)  225  1.244  (1.18 : 1.30)  1.058  (0.98 : 1.13)  0.450  (0.37 : 0.53)  315  1.063  (1.02 : 1.11)  0.805  (0.73 : 0.87)  0.372  (0.31 : 0.42)  405  0.990  (0.95 : 1.03)  0.979  (0.92 : 1.04)  0.529  (0.46 : 0.59)  495  1.145  (1.11 : 1.19)  0.920  (0.86 : 0.98)  0.346  (0.28 : 0.41)    Model 4 (KF = a, m)    Genetic  Maternal  Residual  Age, d  Mean  HPD95%  Mean  HPD95%  Mean  HPD95%  135  0.628  (0.52 : 0.72)  0.640  (0.56 : 0.71)  —  —  225  1.150  (0.70 : 1.23)  1.054  (0.98 : 1.12)  —  —  315  0.973  (0.91 : 1.03)  0.792  (0.73 : 0.86)  —  —  405  1.026  (0.97 : 1.08)  0.981  (0.92 : 1.05)  —  —  495  1.005  (0.94 : 1.07)  0.894  (0.83 : 0.95)  —      Genetic  Maternal  Residual  Age, d  Mean  HPD95%  Mean  HPD95%  Mean  HPD95%  135  0.628  (0.52 : 0.72)  0.640  (0.56 : 0.71)  —  —  225  1.150  (0.70 : 1.23)  1.054  (0.98 : 1.12)  —  —  315  0.973  (0.91 : 1.03)  0.792  (0.73 : 0.86)  —  —  405  1.026  (0.97 : 1.08)  0.981  (0.92 : 1.05)  —  —  495  1.005  (0.94 : 1.07)  0.894  (0.83 : 0.95)  —    1KF = Kalman filter; a = additive effect; m = maternal effect; e = residual effect. 2HPD95% = highest posterior density interval with 95% probability constant. View Large Figure 5. View large Download slide Marginal posterior densities of rates of change of random effects for BW at different ages (model 2: Kalman filter applied to genetic, maternal, and residual effects). Figure 5. View large Download slide Marginal posterior densities of rates of change of random effects for BW at different ages (model 2: Kalman filter applied to genetic, maternal, and residual effects). Marginal posterior means and SD of genetic and maternal correlations are shown in Table 3; posterior means and SD of residual and phenotypic correlations are presented in Table 4. Estimates agree with results reported in the literature (see Mercadante et al., 1995; Lôbo et al., 2000). Both genetic and maternal correlation estimates decreased with increasing distance between time points. Estimates of additive and maternal correlations obtained with KF models were larger than those from the multiple-trait model. Model 3 produced very small phenotypic correlations, close to zero between the most distant time points. In this model, the covariance between time points t and t + 1 is forced to be equal to the variance at time t. These restrictions probably distorted (co)variance estimates. Marginal posterior means and SD of genetic correlations (above diagonal) and maternal effect correlations (below diagonal) for BW, by model Table 3. Marginal posterior means and SD of genetic correlations (above diagonal) and maternal effect correlations (below diagonal) for BW, by model Age, d  Model 1 (multitrait)    45  135  225  315  405  495  45    0.35 (0.03)  0.34 (0.04)  0.28 (0.04)  0.26 (0.04)  0.26 (0.04)  135  0.34 (0.03)    0.63 (0.02)  0.54 (0.03)  0.50 (0.03)  0.50 (0.03)  225  0.29 (0.03)  0.57 (0.03)    0.66 (0.03)  0.58 (0.03)  0.56 (0.03)  315  0.22 (0.04)  0.44 (0.03)  0.54 (0.03)    0.76 (0.02)  0.70 (0.02)  405  0.21 (0.04)  0.37 (0.04)  0.41 (0.04)  0.64 (0.03)    0.77 (0.02)  495  0.21 (0.04)  0.40 (0.04)  0.43 (0.04)  0.56 (0.03)  0.66 (0.03)      Model 2 (KF = a, m, e)1  Age, d  45  135  225  315  405  495  45    0.55 (0.03)  0.44 (0.03)  0.33 (0.02)  0.28 (0.02)  0.24 (0.02)  135  0.67 (0.02)    0.80 (0.02)  0.61 (0.02)  0.51 (0.02)  0.44 (0.02)  225  0.60 (0.02)  0.89 (0.01)    0.75 (0.02)  0.64 (0.02)  0.54 (0.02)  315  0.53 (0.02)  0.80 (0.01)  0.90 (0.01)    0.84 (0.01)  0.72 (0.02)  405  0.48 (0.02)  0.74 (0.01)  0.83 (0.01)  0.93 (0.01)    0.85 (0.01)  495  0.47 (0.02)  0.70 (0.01)  0.78 (0.01)  0.87 (0.01)  0.94 (0.01)      Model 3 (KF = a, m, e, = 1)2  Age, d  45  135  225  315  405  495  45    0.69 (0.01)  0.57 (0.01)  0.48 (0.01)  0.42 (0.01)  0.38 (0.01)  135  0.68 (0.02)    0.83 (0.01)  0.69 (0.01)  0.62 (0.01)  0.55 (0.01)  225  0.56 (0.02)  0.83 (0.01)    0.83 (0.01)  0.74 (0.01)  0.67 (0.01)  315  0.48 (0.02)  0.70 (0.01)  0.85 (0.01)    0.89 (0.01)  0.80 (0.01)  405  0.43 (0.01)  0.63 (0.01)  0.76 (0.01)  0.90 (0.01)    0.90 (0.01)  495  0.39 (0.01)  0.57 (0.01)  0.69 (0.01)  0.81 (0.01)  0.90 (0.01)      Model 4 (KF = a, m)  Age, d  45  135  225  315  405  495  45    0.57 (0.03)  0.46 (0.02)  0.35 (0.02)  0.29 (0.02)  0.25 (0.02)  135  0.51 (0.04)    0.81 (0.01)  0.61 (0.02)  0.51 (0.02)  0.43 (0.02)  225  0.42 (0.03)  0.82 (0.02)    0.75 (0.02)  0.63 (0.02)  0.54 (0.02)  315  0.35 (0.03)  0.69 (0.02)  0.84 (0.02)    0.84 (0.01)  0.71 (0.02)  405  0.32 (0.02)  0.62 (0.02)  0.75 (0.02)  0.90 (0.01)    0.85 (0.01)  495  0.28 (0.02)  0.55 (0.02)  0.67 (0.02)  0.81 (0.01)  0.90 (0.01)    Age, d  Model 1 (multitrait)    45  135  225  315  405  495  45    0.35 (0.03)  0.34 (0.04)  0.28 (0.04)  0.26 (0.04)  0.26 (0.04)  135  0.34 (0.03)    0.63 (0.02)  0.54 (0.03)  0.50 (0.03)  0.50 (0.03)  225  0.29 (0.03)  0.57 (0.03)    0.66 (0.03)  0.58 (0.03)  0.56 (0.03)  315  0.22 (0.04)  0.44 (0.03)  0.54 (0.03)    0.76 (0.02)  0.70 (0.02)  405  0.21 (0.04)  0.37 (0.04)  0.41 (0.04)  0.64 (0.03)    0.77 (0.02)  495  0.21 (0.04)  0.40 (0.04)  0.43 (0.04)  0.56 (0.03)  0.66 (0.03)      Model 2 (KF = a, m, e)1  Age, d  45  135  225  315  405  495  45    0.55 (0.03)  0.44 (0.03)  0.33 (0.02)  0.28 (0.02)  0.24 (0.02)  135  0.67 (0.02)    0.80 (0.02)  0.61 (0.02)  0.51 (0.02)  0.44 (0.02)  225  0.60 (0.02)  0.89 (0.01)    0.75 (0.02)  0.64 (0.02)  0.54 (0.02)  315  0.53 (0.02)  0.80 (0.01)  0.90 (0.01)    0.84 (0.01)  0.72 (0.02)  405  0.48 (0.02)  0.74 (0.01)  0.83 (0.01)  0.93 (0.01)    0.85 (0.01)  495  0.47 (0.02)  0.70 (0.01)  0.78 (0.01)  0.87 (0.01)  0.94 (0.01)      Model 3 (KF = a, m, e, = 1)2  Age, d  45  135  225  315  405  495  45    0.69 (0.01)  0.57 (0.01)  0.48 (0.01)  0.42 (0.01)  0.38 (0.01)  135  0.68 (0.02)    0.83 (0.01)  0.69 (0.01)  0.62 (0.01)  0.55 (0.01)  225  0.56 (0.02)  0.83 (0.01)    0.83 (0.01)  0.74 (0.01)  0.67 (0.01)  315  0.48 (0.02)  0.70 (0.01)  0.85 (0.01)    0.89 (0.01)  0.80 (0.01)  405  0.43 (0.01)  0.63 (0.01)  0.76 (0.01)  0.90 (0.01)    0.90 (0.01)  495  0.39 (0.01)  0.57 (0.01)  0.69 (0.01)  0.81 (0.01)  0.90 (0.01)      Model 4 (KF = a, m)  Age, d  45  135  225  315  405  495  45    0.57 (0.03)  0.46 (0.02)  0.35 (0.02)  0.29 (0.02)  0.25 (0.02)  135  0.51 (0.04)    0.81 (0.01)  0.61 (0.02)  0.51 (0.02)  0.43 (0.02)  225  0.42 (0.03)  0.82 (0.02)    0.75 (0.02)  0.63 (0.02)  0.54 (0.02)  315  0.35 (0.03)  0.69 (0.02)  0.84 (0.02)    0.84 (0.01)  0.71 (0.02)  405  0.32 (0.02)  0.62 (0.02)  0.75 (0.02)  0.90 (0.01)    0.85 (0.01)  495  0.28 (0.02)  0.55 (0.02)  0.67 (0.02)  0.81 (0.01)  0.90 (0.01)    1KF = Kalman filter; a = additive effect; m = maternal effect; e = residual effect. 2 = rate of change. View Large Table 3. Marginal posterior means and SD of genetic correlations (above diagonal) and maternal effect correlations (below diagonal) for BW, by model Age, d  Model 1 (multitrait)    45  135  225  315  405  495  45    0.35 (0.03)  0.34 (0.04)  0.28 (0.04)  0.26 (0.04)  0.26 (0.04)  135  0.34 (0.03)    0.63 (0.02)  0.54 (0.03)  0.50 (0.03)  0.50 (0.03)  225  0.29 (0.03)  0.57 (0.03)    0.66 (0.03)  0.58 (0.03)  0.56 (0.03)  315  0.22 (0.04)  0.44 (0.03)  0.54 (0.03)    0.76 (0.02)  0.70 (0.02)  405  0.21 (0.04)  0.37 (0.04)  0.41 (0.04)  0.64 (0.03)    0.77 (0.02)  495  0.21 (0.04)  0.40 (0.04)  0.43 (0.04)  0.56 (0.03)  0.66 (0.03)      Model 2 (KF = a, m, e)1  Age, d  45  135  225  315  405  495  45    0.55 (0.03)  0.44 (0.03)  0.33 (0.02)  0.28 (0.02)  0.24 (0.02)  135  0.67 (0.02)    0.80 (0.02)  0.61 (0.02)  0.51 (0.02)  0.44 (0.02)  225  0.60 (0.02)  0.89 (0.01)    0.75 (0.02)  0.64 (0.02)  0.54 (0.02)  315  0.53 (0.02)  0.80 (0.01)  0.90 (0.01)    0.84 (0.01)  0.72 (0.02)  405  0.48 (0.02)  0.74 (0.01)  0.83 (0.01)  0.93 (0.01)    0.85 (0.01)  495  0.47 (0.02)  0.70 (0.01)  0.78 (0.01)  0.87 (0.01)  0.94 (0.01)      Model 3 (KF = a, m, e, = 1)2  Age, d  45  135  225  315  405  495  45    0.69 (0.01)  0.57 (0.01)  0.48 (0.01)  0.42 (0.01)  0.38 (0.01)  135  0.68 (0.02)    0.83 (0.01)  0.69 (0.01)  0.62 (0.01)  0.55 (0.01)  225  0.56 (0.02)  0.83 (0.01)    0.83 (0.01)  0.74 (0.01)  0.67 (0.01)  315  0.48 (0.02)  0.70 (0.01)  0.85 (0.01)    0.89 (0.01)  0.80 (0.01)  405  0.43 (0.01)  0.63 (0.01)  0.76 (0.01)  0.90 (0.01)    0.90 (0.01)  495  0.39 (0.01)  0.57 (0.01)  0.69 (0.01)  0.81 (0.01)  0.90 (0.01)      Model 4 (KF = a, m)  Age, d  45  135  225  315  405  495  45    0.57 (0.03)  0.46 (0.02)  0.35 (0.02)  0.29 (0.02)  0.25 (0.02)  135  0.51 (0.04)    0.81 (0.01)  0.61 (0.02)  0.51 (0.02)  0.43 (0.02)  225  0.42 (0.03)  0.82 (0.02)    0.75 (0.02)  0.63 (0.02)  0.54 (0.02)  315  0.35 (0.03)  0.69 (0.02)  0.84 (0.02)    0.84 (0.01)  0.71 (0.02)  405  0.32 (0.02)  0.62 (0.02)  0.75 (0.02)  0.90 (0.01)    0.85 (0.01)  495  0.28 (0.02)  0.55 (0.02)  0.67 (0.02)  0.81 (0.01)  0.90 (0.01)    Age, d  Model 1 (multitrait)    45  135  225  315  405  495  45    0.35 (0.03)  0.34 (0.04)  0.28 (0.04)  0.26 (0.04)  0.26 (0.04)  135  0.34 (0.03)    0.63 (0.02)  0.54 (0.03)  0.50 (0.03)  0.50 (0.03)  225  0.29 (0.03)  0.57 (0.03)    0.66 (0.03)  0.58 (0.03)  0.56 (0.03)  315  0.22 (0.04)  0.44 (0.03)  0.54 (0.03)    0.76 (0.02)  0.70 (0.02)  405  0.21 (0.04)  0.37 (0.04)  0.41 (0.04)  0.64 (0.03)    0.77 (0.02)  495  0.21 (0.04)  0.40 (0.04)  0.43 (0.04)  0.56 (0.03)  0.66 (0.03)      Model 2 (KF = a, m, e)1  Age, d  45  135  225  315  405  495  45    0.55 (0.03)  0.44 (0.03)  0.33 (0.02)  0.28 (0.02)  0.24 (0.02)  135  0.67 (0.02)    0.80 (0.02)  0.61 (0.02)  0.51 (0.02)  0.44 (0.02)  225  0.60 (0.02)  0.89 (0.01)    0.75 (0.02)  0.64 (0.02)  0.54 (0.02)  315  0.53 (0.02)  0.80 (0.01)  0.90 (0.01)    0.84 (0.01)  0.72 (0.02)  405  0.48 (0.02)  0.74 (0.01)  0.83 (0.01)  0.93 (0.01)    0.85 (0.01)  495  0.47 (0.02)  0.70 (0.01)  0.78 (0.01)  0.87 (0.01)  0.94 (0.01)      Model 3 (KF = a, m, e, = 1)2  Age, d  45  135  225  315  405  495  45    0.69 (0.01)  0.57 (0.01)  0.48 (0.01)  0.42 (0.01)  0.38 (0.01)  135  0.68 (0.02)    0.83 (0.01)  0.69 (0.01)  0.62 (0.01)  0.55 (0.01)  225  0.56 (0.02)  0.83 (0.01)    0.83 (0.01)  0.74 (0.01)  0.67 (0.01)  315  0.48 (0.02)  0.70 (0.01)  0.85 (0.01)    0.89 (0.01)  0.80 (0.01)  405  0.43 (0.01)  0.63 (0.01)  0.76 (0.01)  0.90 (0.01)    0.90 (0.01)  495  0.39 (0.01)  0.57 (0.01)  0.69 (0.01)  0.81 (0.01)  0.90 (0.01)      Model 4 (KF = a, m)  Age, d  45  135  225  315  405  495  45    0.57 (0.03)  0.46 (0.02)  0.35 (0.02)  0.29 (0.02)  0.25 (0.02)  135  0.51 (0.04)    0.81 (0.01)  0.61 (0.02)  0.51 (0.02)  0.43 (0.02)  225  0.42 (0.03)  0.82 (0.02)    0.75 (0.02)  0.63 (0.02)  0.54 (0.02)  315  0.35 (0.03)  0.69 (0.02)  0.84 (0.02)    0.84 (0.01)  0.71 (0.02)  405  0.32 (0.02)  0.62 (0.02)  0.75 (0.02)  0.90 (0.01)    0.85 (0.01)  495  0.28 (0.02)  0.55 (0.02)  0.67 (0.02)  0.81 (0.01)  0.90 (0.01)    1KF = Kalman filter; a = additive effect; m = maternal effect; e = residual effect. 2 = rate of change. View Large Marginal posterior means and SD of residual correlations (above diagonal) and phenotypic correlations (below diagonal) for BW, by model Table 4. Marginal posterior means and SD of residual correlations (above diagonal) and phenotypic correlations (below diagonal) for BW, by model   Model 1 (multitrait)  Age, d  45  135  225  315  405  495  45    0.35 (0.01)  0.32 (0.01)  0.26 (0.01)  0.23 (0.01)  0.25 (0.01)  135  0.36 (0.02)    0.62 (0.01)  0.51 (0.01)  0.45 (0.01)  0.47 (0.01)  225  0.31 (0.02)  0.64 (0.02)    0.64 (0.01)  0.53 (0.01)  0.53 (0.01)  315  0.26 (0.02)  0.53 (0.02)  0.68 (0.01)    0.73 (0.01)  0.65 (0.01)  405  0.22 (0.03)  0.46 (0.02)  0.55 (0.02)  0.74 (0.01)    0.73 (0.01)  495  0.24 (0.02)  0.48 (0.02)  0.53 (0.02)  0.65 (0.02)  0.73 (0.01)      Model 2 (KF = a, m, e)1  Age, d  45  135  225  315  405  495  45    0.46 (0.01)  0.37 (0.01)  0.32 (0.01)  0.29 (0.01)  0.28 (0.01)  135  0.65 (0.01)    0.74  0.60 (0.01)  0.53 (0.01)  0.51 (0.01)  225  0.52 (0.01)  0.80 (0.01)    0.75 (0.01)  0.64 (0.01)  0.59 (0.01)  315  0.43 (0.01)  0.66 (0.01)  0.82 (0.01)    0.81 (0.01)  0.72 (0.01)  405  0.37 (0.01)  0.57 (0.01)  0.72 (0.01)  0.87 (0.01)    0.81 (0.01)  495  0.32 (0.01)  0.50 (0.01)  0.63 (0.01)  0.76 (0.01)  0.88 (0.01)      Model 3 (KF = a, m, e, = 1)2  Age, d  45  135  225  315  405  495  45    0.67 (0.01)  0.54 (0.01)  0.45 (0.01)  0.40 (0.01)  0.35 (0.01)  135  0.11 (0.02)    0.81  0.68 (0.01)  0.60 (0.01)  0.53 (0.01)  225  0.04 (0.01)  0.32 (0.03)    0.83 (0.01)  0.74 (0.01)  0.65 (0.01)  315  0.01 (0.01)  0.11 (0.01)  0.35 (0.02)    0.88 (0.01)  0.78 (0.01)  405  0.01 (0.01)  0.05 (0.01)  0.16 (0.02)  0.46 (0.02)    0.89 (0.01)  495  0.01 (0.01)  0.02 (0.01)  0.06 (0.01)  0.18 (0.02)  0.38 (0.03)      Model 4 (KF = a, m)  Age, d  45  135  225  315  405  495  45    0.43 (0.01)  0.36 (0.01)  0.29 (0.01)  0.26 (0.01)  0.25 (0.01)  135  0.28 (0.02)    0.71  0.58 (0.01)  0.51 (0.01)  0.49 (0.01)  225  0.26 (0.02)  0.53 (0.02)    0.72 (0.01)  0.60 (0.01)  0.57 (0.01)  315  0.22 (0.02)  0.43 (0.02)  0.57 (0.02)    0.79 (0.01)  0.70 (0.01)  405  0.18 (0.02)  0.37 (0.02)  0.40 (0.02)  0.63 (0.02)    0.80 (0.01)  495  0.24 (0.02)  0.45 (0.02)  0.45 (0.02)  0.55 (0.02)  0.61 (0.02)      Model 1 (multitrait)  Age, d  45  135  225  315  405  495  45    0.35 (0.01)  0.32 (0.01)  0.26 (0.01)  0.23 (0.01)  0.25 (0.01)  135  0.36 (0.02)    0.62 (0.01)  0.51 (0.01)  0.45 (0.01)  0.47 (0.01)  225  0.31 (0.02)  0.64 (0.02)    0.64 (0.01)  0.53 (0.01)  0.53 (0.01)  315  0.26 (0.02)  0.53 (0.02)  0.68 (0.01)    0.73 (0.01)  0.65 (0.01)  405  0.22 (0.03)  0.46 (0.02)  0.55 (0.02)  0.74 (0.01)    0.73 (0.01)  495  0.24 (0.02)  0.48 (0.02)  0.53 (0.02)  0.65 (0.02)  0.73 (0.01)      Model 2 (KF = a, m, e)1  Age, d  45  135  225  315  405  495  45    0.46 (0.01)  0.37 (0.01)  0.32 (0.01)  0.29 (0.01)  0.28 (0.01)  135  0.65 (0.01)    0.74  0.60 (0.01)  0.53 (0.01)  0.51 (0.01)  225  0.52 (0.01)  0.80 (0.01)    0.75 (0.01)  0.64 (0.01)  0.59 (0.01)  315  0.43 (0.01)  0.66 (0.01)  0.82 (0.01)    0.81 (0.01)  0.72 (0.01)  405  0.37 (0.01)  0.57 (0.01)  0.72 (0.01)  0.87 (0.01)    0.81 (0.01)  495  0.32 (0.01)  0.50 (0.01)  0.63 (0.01)  0.76 (0.01)  0.88 (0.01)      Model 3 (KF = a, m, e, = 1)2  Age, d  45  135  225  315  405  495  45    0.67 (0.01)  0.54 (0.01)  0.45 (0.01)  0.40 (0.01)  0.35 (0.01)  135  0.11 (0.02)    0.81  0.68 (0.01)  0.60 (0.01)  0.53 (0.01)  225  0.04 (0.01)  0.32 (0.03)    0.83 (0.01)  0.74 (0.01)  0.65 (0.01)  315  0.01 (0.01)  0.11 (0.01)  0.35 (0.02)    0.88 (0.01)  0.78 (0.01)  405  0.01 (0.01)  0.05 (0.01)  0.16 (0.02)  0.46 (0.02)    0.89 (0.01)  495  0.01 (0.01)  0.02 (0.01)  0.06 (0.01)  0.18 (0.02)  0.38 (0.03)      Model 4 (KF = a, m)  Age, d  45  135  225  315  405  495  45    0.43 (0.01)  0.36 (0.01)  0.29 (0.01)  0.26 (0.01)  0.25 (0.01)  135  0.28 (0.02)    0.71  0.58 (0.01)  0.51 (0.01)  0.49 (0.01)  225  0.26 (0.02)  0.53 (0.02)    0.72 (0.01)  0.60 (0.01)  0.57 (0.01)  315  0.22 (0.02)  0.43 (0.02)  0.57 (0.02)    0.79 (0.01)  0.70 (0.01)  405  0.18 (0.02)  0.37 (0.02)  0.40 (0.02)  0.63 (0.02)    0.80 (0.01)  495  0.24 (0.02)  0.45 (0.02)  0.45 (0.02)  0.55 (0.02)  0.61 (0.02)    1KF = Kalman filter; a = additive effect; m = maternal effect; e = residual effect. 2 = rate of change. View Large Table 4. Marginal posterior means and SD of residual correlations (above diagonal) and phenotypic correlations (below diagonal) for BW, by model   Model 1 (multitrait)  Age, d  45  135  225  315  405  495  45    0.35 (0.01)  0.32 (0.01)  0.26 (0.01)  0.23 (0.01)  0.25 (0.01)  135  0.36 (0.02)    0.62 (0.01)  0.51 (0.01)  0.45 (0.01)  0.47 (0.01)  225  0.31 (0.02)  0.64 (0.02)    0.64 (0.01)  0.53 (0.01)  0.53 (0.01)  315  0.26 (0.02)  0.53 (0.02)  0.68 (0.01)    0.73 (0.01)  0.65 (0.01)  405  0.22 (0.03)  0.46 (0.02)  0.55 (0.02)  0.74 (0.01)    0.73 (0.01)  495  0.24 (0.02)  0.48 (0.02)  0.53 (0.02)  0.65 (0.02)  0.73 (0.01)      Model 2 (KF = a, m, e)1  Age, d  45  135  225  315  405  495  45    0.46 (0.01)  0.37 (0.01)  0.32 (0.01)  0.29 (0.01)  0.28 (0.01)  135  0.65 (0.01)    0.74  0.60 (0.01)  0.53 (0.01)  0.51 (0.01)  225  0.52 (0.01)  0.80 (0.01)    0.75 (0.01)  0.64 (0.01)  0.59 (0.01)  315  0.43 (0.01)  0.66 (0.01)  0.82 (0.01)    0.81 (0.01)  0.72 (0.01)  405  0.37 (0.01)  0.57 (0.01)  0.72 (0.01)  0.87 (0.01)    0.81 (0.01)  495  0.32 (0.01)  0.50 (0.01)  0.63 (0.01)  0.76 (0.01)  0.88 (0.01)      Model 3 (KF = a, m, e, = 1)2  Age, d  45  135  225  315  405  495  45    0.67 (0.01)  0.54 (0.01)  0.45 (0.01)  0.40 (0.01)  0.35 (0.01)  135  0.11 (0.02)    0.81  0.68 (0.01)  0.60 (0.01)  0.53 (0.01)  225  0.04 (0.01)  0.32 (0.03)    0.83 (0.01)  0.74 (0.01)  0.65 (0.01)  315  0.01 (0.01)  0.11 (0.01)  0.35 (0.02)    0.88 (0.01)  0.78 (0.01)  405  0.01 (0.01)  0.05 (0.01)  0.16 (0.02)  0.46 (0.02)    0.89 (0.01)  495  0.01 (0.01)  0.02 (0.01)  0.06 (0.01)  0.18 (0.02)  0.38 (0.03)      Model 4 (KF = a, m)  Age, d  45  135  225  315  405  495  45    0.43 (0.01)  0.36 (0.01)  0.29 (0.01)  0.26 (0.01)  0.25 (0.01)  135  0.28 (0.02)    0.71  0.58 (0.01)  0.51 (0.01)  0.49 (0.01)  225  0.26 (0.02)  0.53 (0.02)    0.72 (0.01)  0.60 (0.01)  0.57 (0.01)  315  0.22 (0.02)  0.43 (0.02)  0.57 (0.02)    0.79 (0.01)  0.70 (0.01)  405  0.18 (0.02)  0.37 (0.02)  0.40 (0.02)  0.63 (0.02)    0.80 (0.01)  495  0.24 (0.02)  0.45 (0.02)  0.45 (0.02)  0.55 (0.02)  0.61 (0.02)      Model 1 (multitrait)  Age, d  45  135  225  315  405  495  45    0.35 (0.01)  0.32 (0.01)  0.26 (0.01)  0.23 (0.01)  0.25 (0.01)  135  0.36 (0.02)    0.62 (0.01)  0.51 (0.01)  0.45 (0.01)  0.47 (0.01)  225  0.31 (0.02)  0.64 (0.02)    0.64 (0.01)  0.53 (0.01)  0.53 (0.01)  315  0.26 (0.02)  0.53 (0.02)  0.68 (0.01)    0.73 (0.01)  0.65 (0.01)  405  0.22 (0.03)  0.46 (0.02)  0.55 (0.02)  0.74 (0.01)    0.73 (0.01)  495  0.24 (0.02)  0.48 (0.02)  0.53 (0.02)  0.65 (0.02)  0.73 (0.01)      Model 2 (KF = a, m, e)1  Age, d  45  135  225  315  405  495  45    0.46 (0.01)  0.37 (0.01)  0.32 (0.01)  0.29 (0.01)  0.28 (0.01)  135  0.65 (0.01)    0.74  0.60 (0.01)  0.53 (0.01)  0.51 (0.01)  225  0.52 (0.01)  0.80 (0.01)    0.75 (0.01)  0.64 (0.01)  0.59 (0.01)  315  0.43 (0.01)  0.66 (0.01)  0.82 (0.01)    0.81 (0.01)  0.72 (0.01)  405  0.37 (0.01)  0.57 (0.01)  0.72 (0.01)  0.87 (0.01)    0.81 (0.01)  495  0.32 (0.01)  0.50 (0.01)  0.63 (0.01)  0.76 (0.01)  0.88 (0.01)      Model 3 (KF = a, m, e, = 1)2  Age, d  45  135  225  315  405  495  45    0.67 (0.01)  0.54 (0.01)  0.45 (0.01)  0.40 (0.01)  0.35 (0.01)  135  0.11 (0.02)    0.81  0.68 (0.01)  0.60 (0.01)  0.53 (0.01)  225  0.04 (0.01)  0.32 (0.03)    0.83 (0.01)  0.74 (0.01)  0.65 (0.01)  315  0.01 (0.01)  0.11 (0.01)  0.35 (0.02)    0.88 (0.01)  0.78 (0.01)  405  0.01 (0.01)  0.05 (0.01)  0.16 (0.02)  0.46 (0.02)    0.89 (0.01)  495  0.01 (0.01)  0.02 (0.01)  0.06 (0.01)  0.18 (0.02)  0.38 (0.03)      Model 4 (KF = a, m)  Age, d  45  135  225  315  405  495  45    0.43 (0.01)  0.36 (0.01)  0.29 (0.01)  0.26 (0.01)  0.25 (0.01)  135  0.28 (0.02)    0.71  0.58 (0.01)  0.51 (0.01)  0.49 (0.01)  225  0.26 (0.02)  0.53 (0.02)    0.72 (0.01)  0.60 (0.01)  0.57 (0.01)  315  0.22 (0.02)  0.43 (0.02)  0.57 (0.02)    0.79 (0.01)  0.70 (0.01)  405  0.18 (0.02)  0.37 (0.02)  0.40 (0.02)  0.63 (0.02)    0.80 (0.01)  495  0.24 (0.02)  0.45 (0.02)  0.45 (0.02)  0.55 (0.02)  0.61 (0.02)    1KF = Kalman filter; a = additive effect; m = maternal effect; e = residual effect. 2 = rate of change. View Large Predictive Ability Figure 6 shows the average squared differences between predicted and observed values. Smaller values represent better predictive ability. The models performed very similarly at the first time point, but differences appeared afterward. The absolute values of errors of all models increased with age. Models 2 and 4 had better predictive ability than the multitrait specification. These models also showed better goodness of fit when the effective number of parameters was considered (Forni et al., 2009). Model 3 was the worst in predicting the data and the errors increased with age. This model had the greatest deviance in a previous analysis (Forni et al., 2009). The predictions of model 2 were the most accurate. Box plot squared errors indicated that errors from model 3 were the most distant from zero, as displayed in Figure 7. The box plots of model 3 errors were the most spread and had the largest outliers. The multiple-trait model (model 1) produced larger errors than the dynamic models 2 and 4 in the leave-one out cross-validation, which might favor overparameterized models according to Efron and Tibshirani (1993). Model 2 had the narrowest error distributions and the closest medians to zero, for all ages. In a previous analysis, model 3 had better goodness of fit than models 1 and 2 when the effective number of parameters was used as a penalty (Forni et al., 2009). However, it is clear that model 3 is not a good choice for prediction of missing or future data. Here, the same amount of information was available at all time points. However, few records available at some points, mainly at the edges of recording, are typical in animal breeding data. The predictive performance of future records is an important matter for livestock breeding. This is an example in which a model that reproduces observed data well (model 3) is poor at predicting future records. Figure 6. View largeDownload slide Average squared differences between observed and predicted BW, by age and model. Figure 6. View largeDownload slide Average squared differences between observed and predicted BW, by age and model. Figure 7. View largeDownload slide Box plots of posterior means of squared differences between observed and predicted BW, by age and model. Figure 7. View largeDownload slide Box plots of posterior means of squared differences between observed and predicted BW, by age and model. The KF was a useful tool for structuring (co)variance matrices via a parsimonious specification, and it fitted the data well. Here, KF provided accurate predictions and plausible estimates of (co)variance components. Moreover, KF gives rise to a class of flexible models because the multivariate structure can be used for some random effects, whereas the dynamic feature can be incorporated for others. The analysis of longitudinal traits that do not follow an easy-to-model trajectory (i.e., with many up and down segments) may benefit from use of a KF, because polynomial approximations are not required. In addition, KF may be helpful when estimated genetic covariance matrices have a high probability of being nonpositive definite, which increases with the number of traits and the absolute value of the correlations. LITERATURE CITED Albuquerque L. G. Meyer K. 2001a. Estimates of covariance functions for growth from birth to 630 days of age in Nelore cattle. J. Anim. Sci.  79: 2776– 2789. [PubMed] Google Scholar CrossRef Search ADS   Albuquerque L. G. Meyer K. 2001b. Estimates of direct and maternal genetic effects for weights from birth to 600 days of age in Nelore cattle. J. Anim. Breed. Genet.  118: 83– 92. Google Scholar CrossRef Search ADS   Efron, B., and R. J. Tibshirani 1993. An Introduction to the Bootstrap.  Chapman and Hall, New York, NY. Google Scholar CrossRef Search ADS   Eler J. P. Van Vleck L. D. Ferraz J. B. Lobo R. B. 1995. Estimates of variances due to genetic and maternal effects for growth traits of Nelore cattle. J. Anim. Sci.  73: 3253– 3258. [PubMed] Google Scholar CrossRef Search ADS PubMed  Forni S. Piles M. Blasco A. Varona L. Oliveira H. N. Lôbo R. B. Albuquerque L. G. 2007. Analysis of beef cattle longitudinal data applying a non-linear model. J. Anim. Sci.  85: 3189– 3197. [PubMed] Google Scholar CrossRef Search ADS PubMed  Gelfand, A. E., K. Dey, and H. Chang 1992. Model determination using predictive distributions with implementation via sampling-based methods. Pages 147–167 in Bayesian Statistics 4.  Oxford Univ. Press, Oxford, UK. Gelman A. Rubin D. B. 1992. Inference from iterative simulation using multiple sequences. Stat. Sci.  7: 457– 472. Google Scholar CrossRef Search ADS   Geweke, J. 1992. Evaluating the accuracy of sampling-based approaches to the calculation of posterior moments (with discussion). Pages 169–193 in Bayesian Statistics 4.  Oxford Univ. Press, Oxford, UK. Geyer C. J. 1992. Practical Markov chain Monte Carlo. Stat. Sci.  7: 473– 511. Google Scholar CrossRef Search ADS   Hill, W. G. 1998. Inferences from evolutionary biology to livestock breeding. In Proc. 6th World Cong.  Genet. Appl. Livest. Prod., Armidale, Australia (CD-ROM). Kalman R. E. 1960. A new approach to linear filtering and predictive problems. J. Basic Eng.  82: 35– 45. Google Scholar CrossRef Search ADS   Lôbo, R. B., L. A. F. Bezerra, P. S. Barros, C. U. Magnabosco, L. G. Albuquerque, J. A. G. Bergmann, R. D. Sainz, and H. N. Oliveira 2007. Avaliação Genética de Touros e Matrizes da Raça Nelore: Sumário 2007. Associação Nacional de Criadores e Pesquisadores, Ribeirão Preto,  São Paulo, Brazil. Lôbo R. N. B. Madalena F. E. Vieira A. R. 2000. Average estimates of genetic parameters for beef and dairy cattle in tropical regions. Anim. Breed. Abstr.  68: 433– 462. Mercadante M. E. Z. Lôbo R. B. de los Reyes A. 1995. Parámetros genéticos para características de crecimiento en cebuínos de carne: Una revisión. Arch. Latinoam. Prod. Anim.  3: 45– 89. Meyer K. 1992a. Variance components due to direct and maternal effects for growth traits of Australian beef cattle. Livest. Prod. Sci.  31: 179– 204. Google Scholar CrossRef Search ADS   Meyer K. 1992b. Bias and sampling covariances of estimates of variance components due to maternal effects. Genet. Sel. Evol.  24: 487– 509. Google Scholar CrossRef Search ADS   Meyer K. Carrick M. J. Donnelly B. J. 1993. Genetic parameters for growth traits of Australian beef cattle from a multibreed selection experiment. J. Anim. Sci.  71: 2614– 2622. [PubMed] Google Scholar CrossRef Search ADS PubMed  Meyer K. Hill W. G. 1997. Estimation of genetic and phenotypic covariance functions for longitudinal or “repeated” records by Restricted Maximum Likelihood. Livest. Prod. Sci.  47: 185– 200. Google Scholar CrossRef Search ADS   Mrode, R. A. 1996. Linear Models for the Prediction of Animal Breeding Values.  CAB Int., Wallingford, UK. Varona L. Moreno C. García Cortés L. A. Altarriba J. 1997. Multiple trait genetic analysis of underlying biological variables of production functions. Livest. Prod. Sci.  47: 201– 209. Google Scholar CrossRef Search ADS   Willham R. L. 1980. Problems in estimating maternal effects. Livest. Prod. Sci.  7: 405– 418. Google Scholar CrossRef Search ADS   American Society of Animal Science

Pabiou, T.;Fikse, W. F.;Näsholm, A.;Cromie, A. R.;Drennan, M. J.;Keane, M. G.;Berry, D. P.

doi: 10.2527/jas.2008-1510pmid: 19717761

ABSTRACT The objective of this study was to estimate genetic parameters for the weights of different wholesale cuts, using an experimental and a commercial data set. The experimental and commercial data sets included 413 and 635 crossbred Belgian Blue, Charolais, Limousin, Angus, Holstein, and Simmental animals, respectively. Univariate analyses using a mixed linear animal model with relationships were undertaken to estimate the heritability of cold carcass weight, carcass conformation and fat, and the cut weights, whereas a series of bivariate analyses was used to estimate the phenotypic and genetic correlations between carcass weight, carcass conformation, carcass fat, and the major primal cuts. Heritability estimates for cold carcass weight in both data sets were moderate (>0.48), whereas heritability estimates for carcass conformation and fat grading were greater in the commercial data set (>0.63) than in the experimental study (>0.33). Across both data sets, heritability estimates for wholesale cut weight in the forequarter varied from 0.03 to 0.79, whereas heritability estimates of carcass cut weight in the hindquarter varied from 0.14 to 0.86. Heritability estimates for cut weights expressed as a proportion of the entire carcass weight varied from 0.04 to 0.91. Genetic correlations were strong among the different carcass cut weights within the experimental and the commercial studies. Genetic correlations between the weights of selected carcass cuts and carcass weight were moderate to high (minimum 0.45; maximum 0.88) in both data sets. Positive genetic correlations were observed in the commercial data set between the different wholesale cut weights and carcass conformation, whereas these were positive and negative in the experimental data set. Selection for increased carcass weight will, on average, increase the weight of each cut. However, the genetic correlations were less than unity, suggesting a benefit of more direct selection on high value cuts. INTRODUCTION In Ireland, as in most other countries, the value farmers generally receive for each carcass is predominantly based on carcass weight, carcass conformation, and carcass fat score. In the European Union, the EUROP classification system, as implemented by the European Council regulations 1208/81 and 2930/81, is currently used to assign a conformation and fat grade to each carcass (Department of Agriculture and Food, 2004). The conformation classification system uses the letters E (excellent), U, R, O, P (poor) to describe the development of the carcass profiles with particular emphasis on the round, back, and shoulder. The carcass fat classification system uses the scale 1 (low), 2, 3, 4, and 5 (very high) to measure the amount of fat on the outside of the carcass and in the thoracic cavity. Three subdivisions (+, =, −) can be accounted for in each carcass conformation or fat class. Differences in retail value exist between different parts of the carcass (Morris et al., 1999). Farmers should logically be rewarded for producing a larger quantity of these high value cuts, and the current EUROP grading system, measuring the overall conformation and fat, may not be reflecting these differences within carcasses. Active selection for individual wholesale cut weight is currently limited by a lack of routinely collected phenotypic data to estimate breeding values, a lack of sufficient studies on the genetic parameters for carcass cut weights, as well as a lack of knowledge on the phenotypic and genetic correlations between carcass cut weights and other routinely measured traits. Cundiff et al. (1969) published moderate to high heritability estimates for some wholesale cut weights such as round (0.68), loin (0.48), rib (0.44), and chuck (0.49). Brackelsberg et al. (1971) reported similar heritability estimates. Despite most Irish farmers not currently being paid for individual primal weight, advances in technologies such as mechanical grading may facilitate future estimation of primal weight, which may subsequently lead to payment on these estimates. Therefore, to be proactive, as well as to ensure that the quality of Irish beef carcasses does not deteriorate, it is important that the effect of current selection practices on individual cut weights, especially high value cut weights, is quantified. The objective of this study was to estimate, in the Irish context of across breed genetic evaluation, genetic parameters for weight of different wholesale beef cuts and to determine their correlations with the currently recorded traits of carcass weight, carcass conformation, and carcass fat. Carcass cut data used in the present study originated from 2 sources, which included an experimental herd and a commercial retailer. MATERIALS AND METHODS Data used in the present study were obtained from pre-existing databases. Hence, animal care and use committee approval was not necessary for this study. Two databases on carcass cut weight were used in the present study. The first database originated from a series of experiments conducted using a research herd during recent years, and the second database was made available by an Irish supermarket chain. Pedigree information was extracted from the Irish Cattle Breeding Federation database. Experimental Data Set Data from 578 carcasses cut into primal weights from animals slaughtered between 2005 and 2008 were made available from the Teagasc beef research center in Dunsany, Co. Meath, Ireland. These data will be referred to as the experimental data. All of the animals were processed in the same factory and the cutting methods were supervised by the same Teagasc researcher. Animals without a known sire (n = 158) were detected and discarded from the analysis. Within the remaining data set, 94% of animals were crossbred (26 were purebred animals; i.e., at least 28/32 purebred). Almost all of the sires (n = 411) were purebred males (Holstein: n = 89; Belgian Blue: n = 85; Charolais: n = 72; Angus: n = 62; Friesian: n = 41; Limousin: n = 34; Simmental: n = 28), and 84% of dams (n = 346) were crossbred females, where the most prevalent breeds represented were Holstein (n = 161), Limousin (n = 59), and Simmental (n = 37). The animals originated from 7 different experiments that investigated the performance of different finishing diets, as well as animals of divergent genetic merit for growth rate and of different genetic backgrounds (Table 1). Contemporary group was defined as experimental treatment (n = 8) by slaughter date (n = 11). There were 12 contemporary groups with at least 6 animals; the data for 7 animals were discarded because they were in small contemporary groups. Contemporary groups were composed of steers or bulls. The animals slaughtered were bulls (n = 73) or steers (n = 340). The average slaughter age of the bulls and steers was 459 and 762 d, respectively. Age of the dam was grouped into 6 categories: 2 to 4.5 yr old (n = 53), 4.5 to 6 yr old (n = 62), 6 to 7.5 yr old (n = 90), 7.5 to 9 yr old (n = 51), ≥9 yr old (n = 68), and missing data (n = 49). Heterosis and recombination loss coefficients were computed using the formula of Van der Werf and De Boer (1989):  heterozygosis = Pd (1 − Ps) + Ps (1 − Pd);  recombination loss = Ps (1 − Ps) + Pd (1 − Pd), Table 1. Description of the different experiments used in the experimental data set Exp.  n1  Sex2  DOS3  Sire breed4  Dam breed4  Description  A  43  S  Feb. 7, 2005  BB, CH, HO, LM, SI  FR, HO, LM, SI  The spring-born animals spent their first winter indoors on a grass-silage and barley-based concentrate. These animals then grazed on a rotational grazing, paddock-based system before being housed indoors for the second winter where they were fed on ad libitum grass silage and 6 kg of barley-based concentrate.  B5  18  S  Apr. 4, 2005  AA, BB, HO, LM  FR, HO  The spring-born steers grazed pasture during the grazing seasons either side of a store winter period and were finished during their second winter on barley-based concentrate diets and ad libitum grass silage.    24  S  Apr. 11, 2005    D6  73  B  Jun. 27, 2006  BB, CH, FR, HO, LM, SI  AA, BB, CH, FR, HE, HO, LM, SI  Following purchase after weaning, bulls were given ad libitum access to a barley-based concentrate diet in which grass silage was offered at 1 kg of DM·animal−1 daily.  E  6  S  Mar. 23, 2007  CH, SI  CH, LM  The spring-born suckled calves grazed with their dam until weaning before being housed indoors for the first winter period and offered grass silage ad libitum plus 1 kg of concentrates each daily. At the end of the first winter indoor period, the animals were turned out to pasture and rotationally grazed. The animals were housed indoors on slats for the second winter and offered grass silage ad libitum and 4.5 kg of a barley-based concentrate in one single feed daily.  F  11  S  Mar. 23, 2007  CH, SI  CH, LM  Same description as in Exp. E until the second winter during which the animals were housed on slats with access to out-wintering pad (wood chips) and offered maize silage ad libitum and 4.5 kg of a high protein concentrate in one single morning feed daily.  G6  34  S  Apr. 13, 2007  BB, CH  AA, CH  Following purchase after weaning and housing over their first winter, steers were grazed in 2 batches under a rotational grazing, paddock-based system. Steers were finished indoors during the second winter with ad libitum access to a barley-based concentrate and 1 kg of DM grass silage per animal per day.    33  S  Apr. 27, 2007  FR, LM, SI  FR, HE, HO, LM, SI  H7  42  S  Feb. 6, 2008  AA, BB  FR, HO  Animals were purchased at 2 to 6 wk of age. Following weaning at approximately 10 wk of age, animals were grazed at pasture until their first housing at winter when they were offered ad libitum grass silage plus 1.5 kg of concentrates daily. All animals were grazed during the second grazing season and were finished indoors during their second winter on a total mixed ration of 70% concentrates + 30% grass silage.    45  S  Apr. 2, 2008  FR, HO      44  S  Apr. 30, 2008        40  S  Jun. 11, 2008      Exp.  n1  Sex2  DOS3  Sire breed4  Dam breed4  Description  A  43  S  Feb. 7, 2005  BB, CH, HO, LM, SI  FR, HO, LM, SI  The spring-born animals spent their first winter indoors on a grass-silage and barley-based concentrate. These animals then grazed on a rotational grazing, paddock-based system before being housed indoors for the second winter where they were fed on ad libitum grass silage and 6 kg of barley-based concentrate.  B5  18  S  Apr. 4, 2005  AA, BB, HO, LM  FR, HO  The spring-born steers grazed pasture during the grazing seasons either side of a store winter period and were finished during their second winter on barley-based concentrate diets and ad libitum grass silage.    24  S  Apr. 11, 2005    D6  73  B  Jun. 27, 2006  BB, CH, FR, HO, LM, SI  AA, BB, CH, FR, HE, HO, LM, SI  Following purchase after weaning, bulls were given ad libitum access to a barley-based concentrate diet in which grass silage was offered at 1 kg of DM·animal−1 daily.  E  6  S  Mar. 23, 2007  CH, SI  CH, LM  The spring-born suckled calves grazed with their dam until weaning before being housed indoors for the first winter period and offered grass silage ad libitum plus 1 kg of concentrates each daily. At the end of the first winter indoor period, the animals were turned out to pasture and rotationally grazed. The animals were housed indoors on slats for the second winter and offered grass silage ad libitum and 4.5 kg of a barley-based concentrate in one single feed daily.  F  11  S  Mar. 23, 2007  CH, SI  CH, LM  Same description as in Exp. E until the second winter during which the animals were housed on slats with access to out-wintering pad (wood chips) and offered maize silage ad libitum and 4.5 kg of a high protein concentrate in one single morning feed daily.  G6  34  S  Apr. 13, 2007  BB, CH  AA, CH  Following purchase after weaning and housing over their first winter, steers were grazed in 2 batches under a rotational grazing, paddock-based system. Steers were finished indoors during the second winter with ad libitum access to a barley-based concentrate and 1 kg of DM grass silage per animal per day.    33  S  Apr. 27, 2007  FR, LM, SI  FR, HE, HO, LM, SI  H7  42  S  Feb. 6, 2008  AA, BB  FR, HO  Animals were purchased at 2 to 6 wk of age. Following weaning at approximately 10 wk of age, animals were grazed at pasture until their first housing at winter when they were offered ad libitum grass silage plus 1.5 kg of concentrates daily. All animals were grazed during the second grazing season and were finished indoors during their second winter on a total mixed ration of 70% concentrates + 30% grass silage.    45  S  Apr. 2, 2008  FR, HO      44  S  Apr. 30, 2008        40  S  Jun. 11, 2008      1Number of animals. 2S = steer; B = bull. 3Date of slaughter. 4Main sire (dam) breeds (breed fraction ≥50% of total breed fraction); AA: Aberdeen Angus, BB: Belgian Blue, CH: Charolais, FR: Friesian, HO: Holstein, HE: Hereford, LM: Limousin, SI: Beef Simmental. 5Cummins et al. (2007). 6A. M. Clarke (Teagasc Beef Research Center, Dunsany, Co. Meath, Ireland), M. J. Drennan, M. McGee (Teagasc Beef Research Center), D. A. Kenny (Teagasc Beef Research Center), R. D. Evans (Irish Cattle Breeding Federation, Bandon, Co. Cork, Ireland), and D. P. Berry, unpublished data. 7Campion et al. (2008). Table 1. Description of the different experiments used in the experimental data set Exp.  n1  Sex2  DOS3  Sire breed4  Dam breed4  Description  A  43  S  Feb. 7, 2005  BB, CH, HO, LM, SI  FR, HO, LM, SI  The spring-born animals spent their first winter indoors on a grass-silage and barley-based concentrate. These animals then grazed on a rotational grazing, paddock-based system before being housed indoors for the second winter where they were fed on ad libitum grass silage and 6 kg of barley-based concentrate.  B5  18  S  Apr. 4, 2005  AA, BB, HO, LM  FR, HO  The spring-born steers grazed pasture during the grazing seasons either side of a store winter period and were finished during their second winter on barley-based concentrate diets and ad libitum grass silage.    24  S  Apr. 11, 2005    D6  73  B  Jun. 27, 2006  BB, CH, FR, HO, LM, SI  AA, BB, CH, FR, HE, HO, LM, SI  Following purchase after weaning, bulls were given ad libitum access to a barley-based concentrate diet in which grass silage was offered at 1 kg of DM·animal−1 daily.  E  6  S  Mar. 23, 2007  CH, SI  CH, LM  The spring-born suckled calves grazed with their dam until weaning before being housed indoors for the first winter period and offered grass silage ad libitum plus 1 kg of concentrates each daily. At the end of the first winter indoor period, the animals were turned out to pasture and rotationally grazed. The animals were housed indoors on slats for the second winter and offered grass silage ad libitum and 4.5 kg of a barley-based concentrate in one single feed daily.  F  11  S  Mar. 23, 2007  CH, SI  CH, LM  Same description as in Exp. E until the second winter during which the animals were housed on slats with access to out-wintering pad (wood chips) and offered maize silage ad libitum and 4.5 kg of a high protein concentrate in one single morning feed daily.  G6  34  S  Apr. 13, 2007  BB, CH  AA, CH  Following purchase after weaning and housing over their first winter, steers were grazed in 2 batches under a rotational grazing, paddock-based system. Steers were finished indoors during the second winter with ad libitum access to a barley-based concentrate and 1 kg of DM grass silage per animal per day.    33  S  Apr. 27, 2007  FR, LM, SI  FR, HE, HO, LM, SI  H7  42  S  Feb. 6, 2008  AA, BB  FR, HO  Animals were purchased at 2 to 6 wk of age. Following weaning at approximately 10 wk of age, animals were grazed at pasture until their first housing at winter when they were offered ad libitum grass silage plus 1.5 kg of concentrates daily. All animals were grazed during the second grazing season and were finished indoors during their second winter on a total mixed ration of 70% concentrates + 30% grass silage.    45  S  Apr. 2, 2008  FR, HO      44  S  Apr. 30, 2008        40  S  Jun. 11, 2008      Exp.  n1  Sex2  DOS3  Sire breed4  Dam breed4  Description  A  43  S  Feb. 7, 2005  BB, CH, HO, LM, SI  FR, HO, LM, SI  The spring-born animals spent their first winter indoors on a grass-silage and barley-based concentrate. These animals then grazed on a rotational grazing, paddock-based system before being housed indoors for the second winter where they were fed on ad libitum grass silage and 6 kg of barley-based concentrate.  B5  18  S  Apr. 4, 2005  AA, BB, HO, LM  FR, HO  The spring-born steers grazed pasture during the grazing seasons either side of a store winter period and were finished during their second winter on barley-based concentrate diets and ad libitum grass silage.    24  S  Apr. 11, 2005    D6  73  B  Jun. 27, 2006  BB, CH, FR, HO, LM, SI  AA, BB, CH, FR, HE, HO, LM, SI  Following purchase after weaning, bulls were given ad libitum access to a barley-based concentrate diet in which grass silage was offered at 1 kg of DM·animal−1 daily.  E  6  S  Mar. 23, 2007  CH, SI  CH, LM  The spring-born suckled calves grazed with their dam until weaning before being housed indoors for the first winter period and offered grass silage ad libitum plus 1 kg of concentrates each daily. At the end of the first winter indoor period, the animals were turned out to pasture and rotationally grazed. The animals were housed indoors on slats for the second winter and offered grass silage ad libitum and 4.5 kg of a barley-based concentrate in one single feed daily.  F  11  S  Mar. 23, 2007  CH, SI  CH, LM  Same description as in Exp. E until the second winter during which the animals were housed on slats with access to out-wintering pad (wood chips) and offered maize silage ad libitum and 4.5 kg of a high protein concentrate in one single morning feed daily.  G6  34  S  Apr. 13, 2007  BB, CH  AA, CH  Following purchase after weaning and housing over their first winter, steers were grazed in 2 batches under a rotational grazing, paddock-based system. Steers were finished indoors during the second winter with ad libitum access to a barley-based concentrate and 1 kg of DM grass silage per animal per day.    33  S  Apr. 27, 2007  FR, LM, SI  FR, HE, HO, LM, SI  H7  42  S  Feb. 6, 2008  AA, BB  FR, HO  Animals were purchased at 2 to 6 wk of age. Following weaning at approximately 10 wk of age, animals were grazed at pasture until their first housing at winter when they were offered ad libitum grass silage plus 1.5 kg of concentrates daily. All animals were grazed during the second grazing season and were finished indoors during their second winter on a total mixed ration of 70% concentrates + 30% grass silage.    45  S  Apr. 2, 2008  FR, HO      44  S  Apr. 30, 2008        40  S  Jun. 11, 2008      1Number of animals. 2S = steer; B = bull. 3Date of slaughter. 4Main sire (dam) breeds (breed fraction ≥50% of total breed fraction); AA: Aberdeen Angus, BB: Belgian Blue, CH: Charolais, FR: Friesian, HO: Holstein, HE: Hereford, LM: Limousin, SI: Beef Simmental. 5Cummins et al. (2007). 6A. M. Clarke (Teagasc Beef Research Center, Dunsany, Co. Meath, Ireland), M. J. Drennan, M. McGee (Teagasc Beef Research Center), D. A. Kenny (Teagasc Beef Research Center), R. D. Evans (Irish Cattle Breeding Federation, Bandon, Co. Cork, Ireland), and D. P. Berry, unpublished data. 7Campion et al. (2008).where Ps and Pd are the proportions of genes of the primary breed (most prevalent breed) in the sire and dam, respectively. Recombination loss was derived from the heterozygosity of the parental gametes, representing a within-gamete epistatic loss effect (Van der Werf and De Boer, 1989). Cold carcass weight (hereon in referred to as carcass weight), as well as carcass conformation and carcass fat grade, scored using the EUROP classification system (Department of Agriculture and Food, 2004), is recorded for each animal slaughtered in Ireland. In the present study, the EUROP classification grades were transformed to a 15-point scale as outlined by Hickey et al. (2007). Each carcass was cut into forequarter and hindquarter by a section between the 5th and 6th ribs, giving an 8-rib hindquarter and a 5-rib forequarter (Gerrard and Mallion, 1977). The right side of each carcass was cut into 23 different primal cuts: 11 taken in the forequarter and 12 in the hindquarter. The fat trimming procedure aimed to remove all possible fat from the cuts. Using the ratio of total carcass weight over the right side carcass weight, the weight of the cuts measured from the right side of the carcass was extrapolated to a weight taken from the whole carcass. The kidney and pelvic fat were removed before carcass weighing. The total meat weight was defined as the sum of the primal cuts and lean trimmings weights, and the proportion of the total meat weight over the cold carcass weight defined the meat percentage. The total fat and total bone weights were created to respectively sum the carcass fat and bones. Not all of the different cuts recorded were retained for estimation of variance components, and some were grouped together into combined primal cuts. Figure 1 illustrates the location of the retained cuts. Figure 1. View largeDownload slide Location of various beef cuts (Gerrard and Mallion, 1977; Jones et al., 2004). Figure 1. View largeDownload slide Location of various beef cuts (Gerrard and Mallion, 1977; Jones et al., 2004). The retained forequarter cuts were the fore shin, brisket, ribs 1 (nearer to the head) to 5, ribs 6 to 13, shoulder, chuck, and neck. The flank was left attached to ribs 6 to 13. The 2 sets of ribs were summed as ribs. The shoulder was the sum of the clod cut and the whole outside shoulder muscles (blade steak, braising muscle, chuck tender, and leg of mutton cut). The chuck was cut from the first to the sixth thoracic vertebrae. For the analysis, the chuck and the neck cuts were grouped as chuck. Two overall variables were also investigated: forequarter meat weight, summing the primal cut, combined primal cuts, and lean trimmings weights from the forequarter; and total forequarter weight, representing the total meat, fat, and bone weight from the forequarter. The retained hindquarter cuts were the cap of ribs, cube roll, strip-loin, rump, tail of rump, round, heel, and hind shin. The cube roll and the cap of ribs were cut between the fifth and the eleventh thoracic vertebrae and were summed as rib roast. The strip-loin is part of the LM cut between the 10th thoracic vertebrae and the rump. The sum of the rump and tail of rump cuts was labeled sirloin. The fillet, also known as the tenderloin, is the M. psoas, inside the loin area. The round is the main part of the hind leg, summing the silverside, topside, knuckle, and salmon cuts. The hind shin and heel cuts were grouped as hind shin. Two overall variables were also investigated: hindquarter meat weight, summing the primal cuts, combined primal cuts, and lean trimmings weights from the hindquarter; and total hindquarter weight, representing the total meat, fat, and bone weight from the hindquarter. Primal cut weights and combined primal cut weight in the experimental data set will be referred to in the rest of this paper as the wholesale cut weight in the experimental data set. Commercial Data Set A total of 3,501 carcasses cut into primal cuts from purebred and crossbred animals slaughtered between 1999 and 2005 were made available by an Irish supermarket chain. These data will be referred to as the commercial data. All of the animals were processed through the same meat processing plant. Animals with unknown sire (n = 2,502), as well as animals lacking information on herd before slaughter (n = 16), were removed. Additionally, animals slaughtered at less than 12 mo of age (n = 2) were discarded. Age of the dam was grouped into 4 categories: 2 to 3 yr old (n = 130), 4 to 7 yr old (n = 314), ≥8 yr old (n = 72), and missing data (n = 119). Heterosis and recombination loss were computed using the formula of Van der Werf and De Boer (1989) as described previously. Contemporary groups of slaughter were generated using the iterative algorithm of Crump et al. (1997) parameterized by the minimum (30 d) and maximum (120 d) span of a group, and the minimum number of records (n = 4) per group. The composition of contemporary groups was based on finishing herd, date of slaughter, and intervals between consecutive slaughter dates as the variables of interest. First, consecutive animals (ranked on slaughter date) are put into groups according to their slaughter dates and the minimum span of days defined in the parameter file. This step is then repeated considering the start and end slaughter date of the groups and the minimum span defined in the parameter file. Second, contemporary groups are created by reading the groups created previously and clustering consecutive groups according to the maximum span and the minimum records required per group. This step is then repeated considering the maximum span and the minimum records required per group in the parameter file. As a result, 315 animals were discarded from the analysis because of the inability to assign them to a contemporary group of sufficient size. After further restrictions were applied on the weight of the individual cuts (see below), a total of 83 contemporary groups were created, which included 635 animals from 91 sires in 41 different herds. These animals consisted of heifers (n = 575), bulls (n = 26), or steers (n = 34) and were mostly crossbred animals (n = 621 crossbred and n = 14 purebred animals). The sires of these animals were 96% purebred, mainly represented by Belgian Blue (n = 386), Limousin (n = 110), and Charolais (n = 83). The dams were 98% crossbred, where the most prevalent breeds represented were Holstein (n = 353), Limousin (n = 133), Charolais (n = 36), and Simmental (n = 36). Because the average slaughter age of the heifers was 21.5 mo and 66% were born in early spring (January to March), the overall rearing system can be described as a 21-mo-old heifer production system from spring-born calves as outlined by Keane et al. (2008). Heifers after their first winter are fed grazed grass after which they are finished indoors over a 2-mo period on a finishing diet consisting of concentrates and ad libitum grass silage. Cold carcass weight was recorded as described above for the experimental data set. Due to the recent storing (from 2001) of EUROP classification in the national database, conformation and fat grading, transformed to a 15-point scale as outlined by Hickey et al. (2007), were available for only a limited number of animals. The carcasses were trimmed of excessive fat, and the fat depth left averaged 5 mm when measured along the LM. Additional to the routinely recorded carcass traits, information on individual primal cuts was also made available. The primal cutting procedure used on these animals generated 14 different cuts, 7 taken in the 5-rib forequarter, 5 in the 8-rib hindquarter, and 2 from both locations. Not all of these different cuts were retained for estimation of variance components, and some were grouped together into combined primal cuts. The location of the retained cuts can be identified in Figure 1. Primal cuts retained for the analysis from the forequarter were the chuck, a portion of the shoulder labeled the blade, flat ribs, brisket, and flank. The blade is a combined primal cut gathering the clod, the braising muscle, and the chuck tender. The flat rib cut represented only part of the rib set and was taken from ribs 1 to 5. The retained hindquarter primal cuts were rib roast, strip-loin, sirloin, fillet, and round. Due to occasional retail demand, T-bones steaks, whole strip-loin and fillet weights were only available on a reduced data set, the T-bone steak being cut through the strip-loin and the fillet. Two cuts, the lean trimmings and the diced beef, were generated from the forequarter and the hindquarter. In addition, the sum of the primal cuts, combined primals, diced beef, and lean trimmings weights defined the total retail product weight and the proportion of the total retail product weight over the cold carcass weight defined the retail product percentage. Thus, the retail product weight consisted of total meat weight (describing the meat part of the cuts) and total dressing fat weight (describing the variable fat weight left on the cuts). The sum of the primal cuts and combined primal cuts within the forequarter and hindquarter will be referred to as forequarter wholesale cut weight and hindquarter wholesale cut weight, respectively. Within carcass trait, observations greater than ±4 SD from the mean estimated within sex by breed groups were set to missing. If cold carcass weight or one of the major cuts (chuck, brisket, blade, rib-roast, sirloin, and round) was missing, the animal was removed from the analysis (n = 31). Primal cut weights and combined primal cut weight in the commercial data set will be referred to as the wholesale cut weight in the commercial data set. Analysis Despite the similarities observed in the cutting procedures between the experimental and the commercial data sets, (co)variance components were estimated within each data set separately to account for potential differences in the traits. Model building for fixed effects was done using PROC GLM (SAS Inst. Inc., Cary, NC) for data sets and (co)variance components were estimated in ASREML (Gilmour et al., 2006). The choice of fixed effects was based on the data available. The models were generated for each data set separately based on backward elimination (P > 0.05) of factors that were not associated with the dependent variable; significance was based on the F-test. Two-way interactions were also tested for associations with the dependent variable. For the experimental and the commercial data set, the overall mixed linear model can be written as  y = Xb + Zu + ZQg + e, where y is the matrix of records, b is the matrix of fixed effects, u is the matrix of random effects, g is the matrix of breed groups, e is the vector of residual effects, and the X, Z, and Q matrices are the respective design matrices. The mixed linear animal model used in the experimental data set included contemporary group and dam age, included as class effects, and heterosis and age at slaughter centered within sex, both treated as continuous variables, as well as the breed group effect. Factors that did not affect (P > 0.05) any of the traits investigated included whether the animal was a singleton or not, and recombination loss. The effect of the sex of the animal was confounded with the contemporary group. Relationships among animals were accounted for using a relationship matrix. A total of 8,300 animals were included in the pedigree file, and unknown ancestors were included as phantom groups of the Belgian Blue, Charolais, Friesian, Holstein, Limousin, Angus, Simmental, and unknown breeds in the pedigree file. The mixed linear animal model used in the commercial data included the class effects of contemporary group, sex of the animal, and the fixed regression of age at slaughter, which was included as a quadratic regression, as well as in a 2-way interaction with sex of the animal and breed group effect. Heterosis (continuous variable) and recombination loss of the animal (continuous variable), whether the animal was a singleton or not, and age of the dam at the birth of the animal, did not significantly affect any of the traits analyzed (P > 0.05). Relationships among animals were accounted for using a relationship matrix. A total of 6,250 animals were included in the relationship matrix, where unknown ancestors were included as phantom groups of breeds: Belgian Blue, Charolais, Friesian, Holstein, Limousin, and unknown breed in the pedigree file. In a separate series of analyses, carcass weight was included as a covariate in the model to investigate whether the distribution of carcass cuts was heritable. For the experimental and the commercial data sets, heritability estimates were obtained from single trait analyses. The coefficient of genetic variation (CVg) for each trait was calculated as the genetic SD divided by the mean (Houle, 1992). As multitrait (3 × 3 and more) analyses failed to converge, a series of bivariate analyses was used to calculate the correlations between carcass weight, carcass conformation, carcass fat, chuck, shoulder, brisket, rib roast, strip-loin, sirloin, round, and fillet. Fore shin, hind shin, ribs, and flank were not included in the matrix given their relatively low importance for the industry or due to convergence difficulty (ribs). The resulting genetic covariance matrix was bended using the procedure (unweighted option) of Jorjani et al. (2003) to ensure that it was positive definite. RESULTS Experimental Data The average cold carcass weight across all animals was 337 kg, and the average total meat weight was 230 kg, giving a meat percentage of 68% of the total cold carcass weight (Table 2). The average EUROP conformation and fat grades in the animals in the present study corresponded respectively to R− (i.e., good muscle development), and 3= (i.e., fleshy, almost everywhere covered with fat with the exception of the round and shoulder). The total forequarter weight represented on average 54% of the carcass weight, and the heaviest cut of the forequarter was the ribs (35 kg; 17% of the retail cut weight). The round cut was the largest cut in the hindquarter (48 kg; 24% of the retail cut weight), and the smallest was the tenderloin, averaging 6 kg (3% of the retail cut weight). The average weight for the total hindquarter was 155 kg (46% of the cold carcass weight). Overall mean, phenotypic SD (σp), heritability (h2), and coefficient of genetic variation (CVg) for carcass traits and cut weight of 413 bulls and steers in the experimental data set Table 2. Overall mean, phenotypic SD (σp), heritability (h2), and coefficient of genetic variation (CVg) for carcass traits and cut weight of 413 bulls and steers in the experimental data set Trait  Mean  σp  h2 (SE)  CVg, %  Abattoir carcass measure           Cold carcass wt, kg  337  29.07  0.48 (0.21)  6.0   Carcass conformation (scale 1 to 15)  7  1.14  0.45 (0.21)  10.5   Carcass fat (scale 1 to 15)  7  1.21  0.33 (0.18)  10.1  Forequarter           Fore shin, kg  5  0.66  0.39 (0.19)  8.1   Brisket, kg  10  1.74  0.25 (0.19)  8.6   Ribs,1 kg  35  4.42  0.03 (0.15)  2.2   Chuck,2 kg  28  4.33  0.83 (0.24)  14.2   Shoulder, kg  28  3.44  0.79 (0.23)  11.1   Forequarter meat wt,3 kg  120  12.28  0.62 (0.23)  8.0   Total forequarter wt,4 kg  181  16.77  0.42 (0.20)  6.0  Hindquarter           Rib roast, kg  10  1.35  0.14 (0.16)  5.3   Strip-loin, kg  11  1.43  0.49 (0.22)  9.0   Sirloin, kg  13  1.87  0.67 (0.22)  11.6   Round, kg  48  5.43  0.86 (0.23)  10.6   Fillet, kg  6  0.64  0.29 (0.20)  6.0   Hind shin, kg  9  0.98  0.73 (0.22)  9.1   Hindquarter meat wt,3 kg  110  10.82  0.70 (0.23)  8.2   Total hindquarter wt,4 kg  155  12.83  0.57 (0.21)  6.3  Total carcass measure           Kidney and pelvic fat, kg  10  3.49  0.09 (0.17)  10.9   Total meat wt,5 kg  230  22.35  0.68 (0.23)  8.0   Meat percentage,6 %  68  0.03  0.50 (0.21)  2.6   Total fat wt, kg  41  9.41  0.27 (0.18)  12.0   Total bone wt, kg  65  5.59  0.75 (0.22)  7.5  Trait  Mean  σp  h2 (SE)  CVg, %  Abattoir carcass measure           Cold carcass wt, kg  337  29.07  0.48 (0.21)  6.0   Carcass conformation (scale 1 to 15)  7  1.14  0.45 (0.21)  10.5   Carcass fat (scale 1 to 15)  7  1.21  0.33 (0.18)  10.1  Forequarter           Fore shin, kg  5  0.66  0.39 (0.19)  8.1   Brisket, kg  10  1.74  0.25 (0.19)  8.6   Ribs,1 kg  35  4.42  0.03 (0.15)  2.2   Chuck,2 kg  28  4.33  0.83 (0.24)  14.2   Shoulder, kg  28  3.44  0.79 (0.23)  11.1   Forequarter meat wt,3 kg  120  12.28  0.62 (0.23)  8.0   Total forequarter wt,4 kg  181  16.77  0.42 (0.20)  6.0  Hindquarter           Rib roast, kg  10  1.35  0.14 (0.16)  5.3   Strip-loin, kg  11  1.43  0.49 (0.22)  9.0   Sirloin, kg  13  1.87  0.67 (0.22)  11.6   Round, kg  48  5.43  0.86 (0.23)  10.6   Fillet, kg  6  0.64  0.29 (0.20)  6.0   Hind shin, kg  9  0.98  0.73 (0.22)  9.1   Hindquarter meat wt,3 kg  110  10.82  0.70 (0.23)  8.2   Total hindquarter wt,4 kg  155  12.83  0.57 (0.21)  6.3  Total carcass measure           Kidney and pelvic fat, kg  10  3.49  0.09 (0.17)  10.9   Total meat wt,5 kg  230  22.35  0.68 (0.23)  8.0   Meat percentage,6 %  68  0.03  0.50 (0.21)  2.6   Total fat wt, kg  41  9.41  0.27 (0.18)  12.0   Total bone wt, kg  65  5.59  0.75 (0.22)  7.5  1The sum of ribs numbered 1 to 5 and 6 to 13. 2The sum of chuck and neck cuts. 3The meat weight for the fore/hindquarter; weight of wholesale cuts and lean trimmings. 4Total forequarter/hindquarter weight; wholesale cuts, lean trimmings, fat, and bones. 5The sum of the forequarter and hindquarter meat weights. 6The total meat weight to the cold carcass weight. View Large Table 2. Overall mean, phenotypic SD (σp), heritability (h2), and coefficient of genetic variation (CVg) for carcass traits and cut weight of 413 bulls and steers in the experimental data set Trait  Mean  σp  h2 (SE)  CVg, %  Abattoir carcass measure           Cold carcass wt, kg  337  29.07  0.48 (0.21)  6.0   Carcass conformation (scale 1 to 15)  7  1.14  0.45 (0.21)  10.5   Carcass fat (scale 1 to 15)  7  1.21  0.33 (0.18)  10.1  Forequarter           Fore shin, kg  5  0.66  0.39 (0.19)  8.1   Brisket, kg  10  1.74  0.25 (0.19)  8.6   Ribs,1 kg  35  4.42  0.03 (0.15)  2.2   Chuck,2 kg  28  4.33  0.83 (0.24)  14.2   Shoulder, kg  28  3.44  0.79 (0.23)  11.1   Forequarter meat wt,3 kg  120  12.28  0.62 (0.23)  8.0   Total forequarter wt,4 kg  181  16.77  0.42 (0.20)  6.0  Hindquarter           Rib roast, kg  10  1.35  0.14 (0.16)  5.3   Strip-loin, kg  11  1.43  0.49 (0.22)  9.0   Sirloin, kg  13  1.87  0.67 (0.22)  11.6   Round, kg  48  5.43  0.86 (0.23)  10.6   Fillet, kg  6  0.64  0.29 (0.20)  6.0   Hind shin, kg  9  0.98  0.73 (0.22)  9.1   Hindquarter meat wt,3 kg  110  10.82  0.70 (0.23)  8.2   Total hindquarter wt,4 kg  155  12.83  0.57 (0.21)  6.3  Total carcass measure           Kidney and pelvic fat, kg  10  3.49  0.09 (0.17)  10.9   Total meat wt,5 kg  230  22.35  0.68 (0.23)  8.0   Meat percentage,6 %  68  0.03  0.50 (0.21)  2.6   Total fat wt, kg  41  9.41  0.27 (0.18)  12.0   Total bone wt, kg  65  5.59  0.75 (0.22)  7.5  Trait  Mean  σp  h2 (SE)  CVg, %  Abattoir carcass measure           Cold carcass wt, kg  337  29.07  0.48 (0.21)  6.0   Carcass conformation (scale 1 to 15)  7  1.14  0.45 (0.21)  10.5   Carcass fat (scale 1 to 15)  7  1.21  0.33 (0.18)  10.1  Forequarter           Fore shin, kg  5  0.66  0.39 (0.19)  8.1   Brisket, kg  10  1.74  0.25 (0.19)  8.6   Ribs,1 kg  35  4.42  0.03 (0.15)  2.2   Chuck,2 kg  28  4.33  0.83 (0.24)  14.2   Shoulder, kg  28  3.44  0.79 (0.23)  11.1   Forequarter meat wt,3 kg  120  12.28  0.62 (0.23)  8.0   Total forequarter wt,4 kg  181  16.77  0.42 (0.20)  6.0  Hindquarter           Rib roast, kg  10  1.35  0.14 (0.16)  5.3   Strip-loin, kg  11  1.43  0.49 (0.22)  9.0   Sirloin, kg  13  1.87  0.67 (0.22)  11.6   Round, kg  48  5.43  0.86 (0.23)  10.6   Fillet, kg  6  0.64  0.29 (0.20)  6.0   Hind shin, kg  9  0.98  0.73 (0.22)  9.1   Hindquarter meat wt,3 kg  110  10.82  0.70 (0.23)  8.2   Total hindquarter wt,4 kg  155  12.83  0.57 (0.21)  6.3  Total carcass measure           Kidney and pelvic fat, kg  10  3.49  0.09 (0.17)  10.9   Total meat wt,5 kg  230  22.35  0.68 (0.23)  8.0   Meat percentage,6 %  68  0.03  0.50 (0.21)  2.6   Total fat wt, kg  41  9.41  0.27 (0.18)  12.0   Total bone wt, kg  65  5.59  0.75 (0.22)  7.5  1The sum of ribs numbered 1 to 5 and 6 to 13. 2The sum of chuck and neck cuts. 3The meat weight for the fore/hindquarter; weight of wholesale cuts and lean trimmings. 4Total forequarter/hindquarter weight; wholesale cuts, lean trimmings, fat, and bones. 5The sum of the forequarter and hindquarter meat weights. 6The total meat weight to the cold carcass weight. View Large Heritability of cold carcass weight and total meat weight was 0.48 and 0.68, respectively. In the forequarter, the shoulder and the chuck had the greatest heritability (0.79 and 0.83, respectively); the least heritability estimate was for the ribs (0.03). The heritability for the forequarter meat weight was 0.62. In the hindquarter, the round cut was most heritable (0.86), whereas the least heritability estimates were for the rib roast (0.14) and the fillet (0.29). The heritability for the hindquarter meat weight was 0.70. The CVg of the cut weights across the carcass varied from 2.2% (ribs) to 14.2% (chuck). Phenotypically, carcass weight and conformation were positively associated with the different wholesale cut weights, whereas the phenotypic correlations between carcass fat and the wholesale cut weight were all close to zero (Table 3). Few genetic correlations with carcass weight, conformation, and fat score were more than twice their respective SE. However, carcass weight was positively genetically correlated with the different cut weights, whereas carcass fat score was negatively genetically correlated with the different cuts. With the exception of the moderate genetic correlation between the brisket and the rib roast (0.38), the genetic correlations among the different wholesale cut weights were generally strong and positive (≥0.47). Genetic (above the diagonal) and phenotypic1 (below the diagonal) correlations (SE in parentheses) between various carcass traits and cuts of bulls/steers from the Teagasc beef research center experimental data set Table 3. Genetic (above the diagonal) and phenotypic1 (below the diagonal) correlations (SE in parentheses) between various carcass traits and cuts of bulls/steers from the Teagasc beef research center experimental data set Item  CCW  CCON  CFAT  Chuck  Shoulder  Brisket  Rib roast  Strip-loin  Sirloin  Round  Fillet  CCW    −0.24 (0.37)  −0.14 (0.39)  0.75 (0.12)  0.88 (0.09)  0.63 (0.27)  0.70 (0.36)  0.47 (0.25)  0.87 (0.10)  0.84 (0.08)  0.83 (0.14)  CCON  0.36    0.35 (0.35)  0.30 (0.14)  −0.04 (0.30)  −0.24 (0.45)  0.09 (0.55)  −0.20 (0.41)  0.30 (0.24)  0.25 (0.26)  −0.12 (0.46)  CFAT  0.21  0.05    −0.26 (0.31)  −0.37 (0.31)  −0.59 (0.48)  −0.45 (0.67)  −0.59 (0.35)  −0.41 (0.31)  −0.45 (0.29)  −0.29 (0.43)  Chuck  0.69  0.31  −0.04    0.70 (0.13)  0.47 (0.28)  0.86 (0.23)  0.78 (0.14)  0.82 (0.09)  0.83 (0.09)  0.84 (0.16)  Shoulder  0.70  0.31  −0.04  0.57    0.68 (0.23)  0.79 (0.27)  0.68 (0.16)  0.93 (0.07)  0.83 (0.09)  0.79 (0.11)  Brisket  0.53  0.34  −0.06  0.49  0.48    0.38 (0.56)  0.52 (0.15)  0.70 (0.20)  0.78 (0.19)  0.79 (0.23)  Rib roast  0.56  0.4  −0.03  0.47  0.55  0.43    0.81 (0.11)  0.80 (0.38)  0.79 (0.27)  0.70 (0.10)  Strip-loin  0.65  0.41  −0.00  0.56  0.55  0.46  0.54    0.81 (0.11)  0.78 (0.12)  0.67 (0.24)  Sirloin  0.65  0.34  −0.11  0.60  0.63  0.52  0.50  0.60    0.93 (0.05)  0.83 (0.09)  Round  0.77  0.45  −0.07  0.65  0.70  0.59  0.57  0.66  0.72    0.93 (0.12)  Fillet  0.66  0.34  −0.06  0.52  0.60  0.50  0.59  0.56  0.68  0.69    Item  CCW  CCON  CFAT  Chuck  Shoulder  Brisket  Rib roast  Strip-loin  Sirloin  Round  Fillet  CCW    −0.24 (0.37)  −0.14 (0.39)  0.75 (0.12)  0.88 (0.09)  0.63 (0.27)  0.70 (0.36)  0.47 (0.25)  0.87 (0.10)  0.84 (0.08)  0.83 (0.14)  CCON  0.36    0.35 (0.35)  0.30 (0.14)  −0.04 (0.30)  −0.24 (0.45)  0.09 (0.55)  −0.20 (0.41)  0.30 (0.24)  0.25 (0.26)  −0.12 (0.46)  CFAT  0.21  0.05    −0.26 (0.31)  −0.37 (0.31)  −0.59 (0.48)  −0.45 (0.67)  −0.59 (0.35)  −0.41 (0.31)  −0.45 (0.29)  −0.29 (0.43)  Chuck  0.69  0.31  −0.04    0.70 (0.13)  0.47 (0.28)  0.86 (0.23)  0.78 (0.14)  0.82 (0.09)  0.83 (0.09)  0.84 (0.16)  Shoulder  0.70  0.31  −0.04  0.57    0.68 (0.23)  0.79 (0.27)  0.68 (0.16)  0.93 (0.07)  0.83 (0.09)  0.79 (0.11)  Brisket  0.53  0.34  −0.06  0.49  0.48    0.38 (0.56)  0.52 (0.15)  0.70 (0.20)  0.78 (0.19)  0.79 (0.23)  Rib roast  0.56  0.4  −0.03  0.47  0.55  0.43    0.81 (0.11)  0.80 (0.38)  0.79 (0.27)  0.70 (0.10)  Strip-loin  0.65  0.41  −0.00  0.56  0.55  0.46  0.54    0.81 (0.11)  0.78 (0.12)  0.67 (0.24)  Sirloin  0.65  0.34  −0.11  0.60  0.63  0.52  0.50  0.60    0.93 (0.05)  0.83 (0.09)  Round  0.77  0.45  −0.07  0.65  0.70  0.59  0.57  0.66  0.72    0.93 (0.12)  Fillet  0.66  0.34  −0.06  0.52  0.60  0.50  0.59  0.56  0.68  0.69    1CCW: cold carcass weight; CCON: carcass conformation; CFAT: carcass fat. All SE of the phenotypic correlations were ≤0.06. View Large Table 3. Genetic (above the diagonal) and phenotypic1 (below the diagonal) correlations (SE in parentheses) between various carcass traits and cuts of bulls/steers from the Teagasc beef research center experimental data set Item  CCW  CCON  CFAT  Chuck  Shoulder  Brisket  Rib roast  Strip-loin  Sirloin  Round  Fillet  CCW    −0.24 (0.37)  −0.14 (0.39)  0.75 (0.12)  0.88 (0.09)  0.63 (0.27)  0.70 (0.36)  0.47 (0.25)  0.87 (0.10)  0.84 (0.08)  0.83 (0.14)  CCON  0.36    0.35 (0.35)  0.30 (0.14)  −0.04 (0.30)  −0.24 (0.45)  0.09 (0.55)  −0.20 (0.41)  0.30 (0.24)  0.25 (0.26)  −0.12 (0.46)  CFAT  0.21  0.05    −0.26 (0.31)  −0.37 (0.31)  −0.59 (0.48)  −0.45 (0.67)  −0.59 (0.35)  −0.41 (0.31)  −0.45 (0.29)  −0.29 (0.43)  Chuck  0.69  0.31  −0.04    0.70 (0.13)  0.47 (0.28)  0.86 (0.23)  0.78 (0.14)  0.82 (0.09)  0.83 (0.09)  0.84 (0.16)  Shoulder  0.70  0.31  −0.04  0.57    0.68 (0.23)  0.79 (0.27)  0.68 (0.16)  0.93 (0.07)  0.83 (0.09)  0.79 (0.11)  Brisket  0.53  0.34  −0.06  0.49  0.48    0.38 (0.56)  0.52 (0.15)  0.70 (0.20)  0.78 (0.19)  0.79 (0.23)  Rib roast  0.56  0.4  −0.03  0.47  0.55  0.43    0.81 (0.11)  0.80 (0.38)  0.79 (0.27)  0.70 (0.10)  Strip-loin  0.65  0.41  −0.00  0.56  0.55  0.46  0.54    0.81 (0.11)  0.78 (0.12)  0.67 (0.24)  Sirloin  0.65  0.34  −0.11  0.60  0.63  0.52  0.50  0.60    0.93 (0.05)  0.83 (0.09)  Round  0.77  0.45  −0.07  0.65  0.70  0.59  0.57  0.66  0.72    0.93 (0.12)  Fillet  0.66  0.34  −0.06  0.52  0.60  0.50  0.59  0.56  0.68  0.69    Item  CCW  CCON  CFAT  Chuck  Shoulder  Brisket  Rib roast  Strip-loin  Sirloin  Round  Fillet  CCW    −0.24 (0.37)  −0.14 (0.39)  0.75 (0.12)  0.88 (0.09)  0.63 (0.27)  0.70 (0.36)  0.47 (0.25)  0.87 (0.10)  0.84 (0.08)  0.83 (0.14)  CCON  0.36    0.35 (0.35)  0.30 (0.14)  −0.04 (0.30)  −0.24 (0.45)  0.09 (0.55)  −0.20 (0.41)  0.30 (0.24)  0.25 (0.26)  −0.12 (0.46)  CFAT  0.21  0.05    −0.26 (0.31)  −0.37 (0.31)  −0.59 (0.48)  −0.45 (0.67)  −0.59 (0.35)  −0.41 (0.31)  −0.45 (0.29)  −0.29 (0.43)  Chuck  0.69  0.31  −0.04    0.70 (0.13)  0.47 (0.28)  0.86 (0.23)  0.78 (0.14)  0.82 (0.09)  0.83 (0.09)  0.84 (0.16)  Shoulder  0.70  0.31  −0.04  0.57    0.68 (0.23)  0.79 (0.27)  0.68 (0.16)  0.93 (0.07)  0.83 (0.09)  0.79 (0.11)  Brisket  0.53  0.34  −0.06  0.49  0.48    0.38 (0.56)  0.52 (0.15)  0.70 (0.20)  0.78 (0.19)  0.79 (0.23)  Rib roast  0.56  0.4  −0.03  0.47  0.55  0.43    0.81 (0.11)  0.80 (0.38)  0.79 (0.27)  0.70 (0.10)  Strip-loin  0.65  0.41  −0.00  0.56  0.55  0.46  0.54    0.81 (0.11)  0.78 (0.12)  0.67 (0.24)  Sirloin  0.65  0.34  −0.11  0.60  0.63  0.52  0.50  0.60    0.93 (0.05)  0.83 (0.09)  Round  0.77  0.45  −0.07  0.65  0.70  0.59  0.57  0.66  0.72    0.93 (0.12)  Fillet  0.66  0.34  −0.06  0.52  0.60  0.50  0.59  0.56  0.68  0.69    1CCW: cold carcass weight; CCON: carcass conformation; CFAT: carcass fat. All SE of the phenotypic correlations were ≤0.06. View Large Using the model including carcass weight as a covariate, heritability for the forequarter cuts ranged from 0.04 (ribs) to 0.65 (chuck); the heritabilities for forequarter meat weight and total forequarter weight were 0.39 and 0.51, respectively. Heritability estimates for the different hindquarter cuts ranged from 0.08 (fillet) to 0.61 (round); heritability estimates for hindquarter meat weight and total hindquarter weight were 0.68 and 0.66, respectively. Commercial Data Average cold carcass weight was 290 kg, and the total retail product weight averaged 192 kg, which gave a retail product percentage of 66% of the total cold carcass weight (Table 4). The average EUROP conformation and fat grades in the animals in the present study corresponded approximately to R+ and 3, respectively. Only 3 carcass conformation classes (conformation O, R, and U) were represented in the data set with 70% of the animals graded as class R. Five carcass fat classes (equivalent to class 1, 2, 3, 4, and 4L in the EUROP scale) were represented in the data set with 65% of the animals residing in class 3. Within the forequarter, the chuck and the blade each made up 16% of the forequarter primal weight, which averaged 38 kg (20% of total meat weight). The round cut made up the major proportion (56%) of the hindquarter primal weight, whereas the tenderloin averaged 5 kg (6% of hindquarter primal weight) and represented the smallest proportion of the hindquarter cuts. The average weight of the hindquarter meat was 77 kg (40% of the total meat weight). The trimmings and diced beef represented 28% of the cold carcass weight (43% of the total meat weight). Number of observations (n), overall mean, phenotypic SD (σp), heritability (h2), and coefficient of genetic variation (CVg) for the carcass traits from an Irish commercial data set Table 4. Number of observations (n), overall mean, phenotypic SD (σp), heritability (h2), and coefficient of genetic variation (CVg) for the carcass traits from an Irish commercial data set Trait  n  Mean  σp  h2 (SE)  CVg, %  Abattoir carcass measure             Cold carcass wt, kg  635  290  20.66  0.59 (0.20)  5.5   Carcass conformation (scale 1 to 15)  345  9  1.23  0.78 (0.27)  12.5   Carcass fat (scale 1 to 15)  345  6  1.14  0.63 (0.26)  14.6  Forequarter             Blade,1 kg  635  12  1.06  0.61 (0.20)  7.1   Chuck, kg  635  13  1.46  0.41 (0.20)  7.3   Brisket, kg  635  8  1.12  0.47 (0.18)  10.0   Flat ribs,2 kg  628  5  0.63  0.28 (0.15)  7.1   Flank, kg  451  2  0.28  0.37 (0.26)  10.0   Forequarter wholesale cuts wt,3 kg  635  38  6.86  0.46 (0.18)  6.4  Hindquarter             Rib roast, kg  635  8  0.94  0.40 (0.19)  7.6   Strip-loin, kg  523  11  1.05  0.41 (0.22)  6.2   Sirloin, kg  635  10  0.95  0.55 (0.20)  7.3   Round, kg  635  43  3.88  0.42 (0.19)  5.9   Fillet, kg  520  5  0.45  0.62 (0.20)  7.9   Hindquarter wholesale cuts wt,3 kg  635  77  6.86  0.34 (0.20)  5.2  Other weights             Total lean trimmings,4 kg  635  64  6.36  0.46 (0.18)  6.8   Total dice beef,4 kg  634  18  2.17  0.74 (0.19)  10.7  Total carcass measure              Total retail product wt,5 kg  635  192  16.34  0.54 (0.19)  6.2   Retail product percentage,6 %  635  66  3.47  0.86 (0.17)  4.9  Trait  n  Mean  σp  h2 (SE)  CVg, %  Abattoir carcass measure             Cold carcass wt, kg  635  290  20.66  0.59 (0.20)  5.5   Carcass conformation (scale 1 to 15)  345  9  1.23  0.78 (0.27)  12.5   Carcass fat (scale 1 to 15)  345  6  1.14  0.63 (0.26)  14.6  Forequarter             Blade,1 kg  635  12  1.06  0.61 (0.20)  7.1   Chuck, kg  635  13  1.46  0.41 (0.20)  7.3   Brisket, kg  635  8  1.12  0.47 (0.18)  10.0   Flat ribs,2 kg  628  5  0.63  0.28 (0.15)  7.1   Flank, kg  451  2  0.28  0.37 (0.26)  10.0   Forequarter wholesale cuts wt,3 kg  635  38  6.86  0.46 (0.18)  6.4  Hindquarter             Rib roast, kg  635  8  0.94  0.40 (0.19)  7.6   Strip-loin, kg  523  11  1.05  0.41 (0.22)  6.2   Sirloin, kg  635  10  0.95  0.55 (0.20)  7.3   Round, kg  635  43  3.88  0.42 (0.19)  5.9   Fillet, kg  520  5  0.45  0.62 (0.20)  7.9   Hindquarter wholesale cuts wt,3 kg  635  77  6.86  0.34 (0.20)  5.2  Other weights             Total lean trimmings,4 kg  635  64  6.36  0.46 (0.18)  6.8   Total dice beef,4 kg  634  18  2.17  0.74 (0.19)  10.7  Total carcass measure              Total retail product wt,5 kg  635  192  16.34  0.54 (0.19)  6.2   Retail product percentage,6 %  635  66  3.47  0.86 (0.17)  4.9  1Part of the shoulder. 2Ribs numbered 1 to 5. 3The weight of wholesale cuts for the fore/hindquarter. 4Cut taken in the fore- or in the hindquarter. 5The sum of the forequarter, hindquarter, and other weights. 6Total retail product weight to cold carcass weight. View Large Table 4. Number of observations (n), overall mean, phenotypic SD (σp), heritability (h2), and coefficient of genetic variation (CVg) for the carcass traits from an Irish commercial data set Trait  n  Mean  σp  h2 (SE)  CVg, %  Abattoir carcass measure             Cold carcass wt, kg  635  290  20.66  0.59 (0.20)  5.5   Carcass conformation (scale 1 to 15)  345  9  1.23  0.78 (0.27)  12.5   Carcass fat (scale 1 to 15)  345  6  1.14  0.63 (0.26)  14.6  Forequarter             Blade,1 kg  635  12  1.06  0.61 (0.20)  7.1   Chuck, kg  635  13  1.46  0.41 (0.20)  7.3   Brisket, kg  635  8  1.12  0.47 (0.18)  10.0   Flat ribs,2 kg  628  5  0.63  0.28 (0.15)  7.1   Flank, kg  451  2  0.28  0.37 (0.26)  10.0   Forequarter wholesale cuts wt,3 kg  635  38  6.86  0.46 (0.18)  6.4  Hindquarter             Rib roast, kg  635  8  0.94  0.40 (0.19)  7.6   Strip-loin, kg  523  11  1.05  0.41 (0.22)  6.2   Sirloin, kg  635  10  0.95  0.55 (0.20)  7.3   Round, kg  635  43  3.88  0.42 (0.19)  5.9   Fillet, kg  520  5  0.45  0.62 (0.20)  7.9   Hindquarter wholesale cuts wt,3 kg  635  77  6.86  0.34 (0.20)  5.2  Other weights             Total lean trimmings,4 kg  635  64  6.36  0.46 (0.18)  6.8   Total dice beef,4 kg  634  18  2.17  0.74 (0.19)  10.7  Total carcass measure              Total retail product wt,5 kg  635  192  16.34  0.54 (0.19)  6.2   Retail product percentage,6 %  635  66  3.47  0.86 (0.17)  4.9  Trait  n  Mean  σp  h2 (SE)  CVg, %  Abattoir carcass measure             Cold carcass wt, kg  635  290  20.66  0.59 (0.20)  5.5   Carcass conformation (scale 1 to 15)  345  9  1.23  0.78 (0.27)  12.5   Carcass fat (scale 1 to 15)  345  6  1.14  0.63 (0.26)  14.6  Forequarter             Blade,1 kg  635  12  1.06  0.61 (0.20)  7.1   Chuck, kg  635  13  1.46  0.41 (0.20)  7.3   Brisket, kg  635  8  1.12  0.47 (0.18)  10.0   Flat ribs,2 kg  628  5  0.63  0.28 (0.15)  7.1   Flank, kg  451  2  0.28  0.37 (0.26)  10.0   Forequarter wholesale cuts wt,3 kg  635  38  6.86  0.46 (0.18)  6.4  Hindquarter             Rib roast, kg  635  8  0.94  0.40 (0.19)  7.6   Strip-loin, kg  523  11  1.05  0.41 (0.22)  6.2   Sirloin, kg  635  10  0.95  0.55 (0.20)  7.3   Round, kg  635  43  3.88  0.42 (0.19)  5.9   Fillet, kg  520  5  0.45  0.62 (0.20)  7.9   Hindquarter wholesale cuts wt,3 kg  635  77  6.86  0.34 (0.20)  5.2  Other weights             Total lean trimmings,4 kg  635  64  6.36  0.46 (0.18)  6.8   Total dice beef,4 kg  634  18  2.17  0.74 (0.19)  10.7  Total carcass measure              Total retail product wt,5 kg  635  192  16.34  0.54 (0.19)  6.2   Retail product percentage,6 %  635  66  3.47  0.86 (0.17)  4.9  1Part of the shoulder. 2Ribs numbered 1 to 5. 3The weight of wholesale cuts for the fore/hindquarter. 4Cut taken in the fore- or in the hindquarter. 5The sum of the forequarter, hindquarter, and other weights. 6Total retail product weight to cold carcass weight. View Large Heritability of cold carcass weight and total meat weight was 0.59 and 0.54, respectively. Heritability estimates for conformation and fat grading were also high (0.78 and 0.63 for carcass conformation and fat grade, respectively). In general, heritability estimates of the different joints in the forequarter and hindquarter were all moderate, ranging from 0.28 (flat ribs) to 0.62 (fillet). The CVg of the cut weights across the carcass varied from 5.9% (round) to 10.0% (brisket, flank). The phenotypic and genetic correlations between the cold carcass weight and the various wholesale cut weights were moderately to strongly positive (Table 5); the phenotypic correlations with carcass weight ranged from 0.48 (brisket) to 0.77 (round and blade), whereas the genetic correlations with carcass weight ranged from 0.45 (chuck) to 0.67 (tenderloin). The phenotypic and genetic correlations between carcass conformation and the different cuts were all positive. The phenotypic correlations between carcass fat and the different wholesale weights tended to be negative or close to zero, whereas the genetic correlations were mostly negative, but not different from zero. Among the different wholesale cut weights, all phenotypic correlations were positive and moderate, ranging from 0.44 to 0.76. The genetic correlations between the cuts were also positive, but stronger than their respective phenotypic correlations, ranging from 0.35 to 0.87. Genetic (above the diagonal) and phenotypic1 (below the diagonal) correlations (SE in parentheses) between various carcass traits and cuts estimated from the commercial data set Table 5. Genetic (above the diagonal) and phenotypic1 (below the diagonal) correlations (SE in parentheses) between various carcass traits and cuts estimated from the commercial data set Item  CCW  CCON  CFAT  Chuck  Blade  Brisket  Rib roast  Strip-loin  Sirloin  Round  Fillet  CCW    −0.15 (0.20)  0.38 (0.27)  0.45 (0.23)  0.63 (0.14)  0.49 (0.22)  0.66 (0.17)  0.54 (0.21)  0.59 (0.15)  0.62 (0.17)  0.67 (0.14)  CCON  0.20    −0.20 (0.31)  0.06 (0.37)  0.16 (0.29)  0.41 (0.29)  0.20 (0.35)  0.62 (0.25)  0.27 (0.28)  0.20 (0.34)  0.11 (0.29)  CFAT  0.22  −0.16    −0.10 (0.41)  −0.18 (0.33)  −0.19 (0.33)  −0.59 (0.44)  −0.25 (0.45)  −0.11 (0.35)  −0.34 (0.39)  −0.31 (0.35)  Chuck  0.65  0.28  −0.06    0.77 (0.16)  0.63 (0.24)  0.75 (0.15)  0.44 (0.33)  0.51 (0.24)  0.68 (0.21)  0.74 (0.18)  Blade  0.77  0.30  −0.06  0.66    0.87 (0.11)  0.82 (0.08)  0.46 (0.26)  0.81 (0.08)  0.85 (0.10)  0.85 (0.07)  Brisket  0.48  0.26  −0.23  0.48  0.59    0.82 (0.09)  0.74 (0.19)  0.74 (0.17)  0.73 (0.17)  0.71 (0.15)  Rib roast  0.60  0.26  −0.09  0.52  0.69  0.60    0.66 (0.15)  0.66 (0.14)  0.78 (0.11)  0.77 (0.12)  Strip-loin  0.68  0.36  0.07  0.52  0.57  0.44  0.54    0.41 (0.30)  0.35 (0.32)  0.46 (0.26)  Sirloin  0.71  0.36  −0.02  0.56  0.69  0.50  0.57  0.57    0.77 (0.12)  0.45 (0.16)  Round  0.77  0.42  −0.16  0.63  0.76  0.59  0.63  0.65  0.73    0.78 (0.12)  Fillet  0.68  0.24  −0.12  0.63  0.67  0.56  0.61  0.53  0.57  0.74    Item  CCW  CCON  CFAT  Chuck  Blade  Brisket  Rib roast  Strip-loin  Sirloin  Round  Fillet  CCW    −0.15 (0.20)  0.38 (0.27)  0.45 (0.23)  0.63 (0.14)  0.49 (0.22)  0.66 (0.17)  0.54 (0.21)  0.59 (0.15)  0.62 (0.17)  0.67 (0.14)  CCON  0.20    −0.20 (0.31)  0.06 (0.37)  0.16 (0.29)  0.41 (0.29)  0.20 (0.35)  0.62 (0.25)  0.27 (0.28)  0.20 (0.34)  0.11 (0.29)  CFAT  0.22  −0.16    −0.10 (0.41)  −0.18 (0.33)  −0.19 (0.33)  −0.59 (0.44)  −0.25 (0.45)  −0.11 (0.35)  −0.34 (0.39)  −0.31 (0.35)  Chuck  0.65  0.28  −0.06    0.77 (0.16)  0.63 (0.24)  0.75 (0.15)  0.44 (0.33)  0.51 (0.24)  0.68 (0.21)  0.74 (0.18)  Blade  0.77  0.30  −0.06  0.66    0.87 (0.11)  0.82 (0.08)  0.46 (0.26)  0.81 (0.08)  0.85 (0.10)  0.85 (0.07)  Brisket  0.48  0.26  −0.23  0.48  0.59    0.82 (0.09)  0.74 (0.19)  0.74 (0.17)  0.73 (0.17)  0.71 (0.15)  Rib roast  0.60  0.26  −0.09  0.52  0.69  0.60    0.66 (0.15)  0.66 (0.14)  0.78 (0.11)  0.77 (0.12)  Strip-loin  0.68  0.36  0.07  0.52  0.57  0.44  0.54    0.41 (0.30)  0.35 (0.32)  0.46 (0.26)  Sirloin  0.71  0.36  −0.02  0.56  0.69  0.50  0.57  0.57    0.77 (0.12)  0.45 (0.16)  Round  0.77  0.42  −0.16  0.63  0.76  0.59  0.63  0.65  0.73    0.78 (0.12)  Fillet  0.68  0.24  −0.12  0.63  0.67  0.56  0.61  0.53  0.57  0.74    1CCW: cold carcass weight; CCON: carcass conformation; CFAT: carcass fat. All SE of the phenotypic correlations were ≤0.08. View Large Table 5. Genetic (above the diagonal) and phenotypic1 (below the diagonal) correlations (SE in parentheses) between various carcass traits and cuts estimated from the commercial data set Item  CCW  CCON  CFAT  Chuck  Blade  Brisket  Rib roast  Strip-loin  Sirloin  Round  Fillet  CCW    −0.15 (0.20)  0.38 (0.27)  0.45 (0.23)  0.63 (0.14)  0.49 (0.22)  0.66 (0.17)  0.54 (0.21)  0.59 (0.15)  0.62 (0.17)  0.67 (0.14)  CCON  0.20    −0.20 (0.31)  0.06 (0.37)  0.16 (0.29)  0.41 (0.29)  0.20 (0.35)  0.62 (0.25)  0.27 (0.28)  0.20 (0.34)  0.11 (0.29)  CFAT  0.22  −0.16    −0.10 (0.41)  −0.18 (0.33)  −0.19 (0.33)  −0.59 (0.44)  −0.25 (0.45)  −0.11 (0.35)  −0.34 (0.39)  −0.31 (0.35)  Chuck  0.65  0.28  −0.06    0.77 (0.16)  0.63 (0.24)  0.75 (0.15)  0.44 (0.33)  0.51 (0.24)  0.68 (0.21)  0.74 (0.18)  Blade  0.77  0.30  −0.06  0.66    0.87 (0.11)  0.82 (0.08)  0.46 (0.26)  0.81 (0.08)  0.85 (0.10)  0.85 (0.07)  Brisket  0.48  0.26  −0.23  0.48  0.59    0.82 (0.09)  0.74 (0.19)  0.74 (0.17)  0.73 (0.17)  0.71 (0.15)  Rib roast  0.60  0.26  −0.09  0.52  0.69  0.60    0.66 (0.15)  0.66 (0.14)  0.78 (0.11)  0.77 (0.12)  Strip-loin  0.68  0.36  0.07  0.52  0.57  0.44  0.54    0.41 (0.30)  0.35 (0.32)  0.46 (0.26)  Sirloin  0.71  0.36  −0.02  0.56  0.69  0.50  0.57  0.57    0.77 (0.12)  0.45 (0.16)  Round  0.77  0.42  −0.16  0.63  0.76  0.59  0.63  0.65  0.73    0.78 (0.12)  Fillet  0.68  0.24  −0.12  0.63  0.67  0.56  0.61  0.53  0.57  0.74    Item  CCW  CCON  CFAT  Chuck  Blade  Brisket  Rib roast  Strip-loin  Sirloin  Round  Fillet  CCW    −0.15 (0.20)  0.38 (0.27)  0.45 (0.23)  0.63 (0.14)  0.49 (0.22)  0.66 (0.17)  0.54 (0.21)  0.59 (0.15)  0.62 (0.17)  0.67 (0.14)  CCON  0.20    −0.20 (0.31)  0.06 (0.37)  0.16 (0.29)  0.41 (0.29)  0.20 (0.35)  0.62 (0.25)  0.27 (0.28)  0.20 (0.34)  0.11 (0.29)  CFAT  0.22  −0.16    −0.10 (0.41)  −0.18 (0.33)  −0.19 (0.33)  −0.59 (0.44)  −0.25 (0.45)  −0.11 (0.35)  −0.34 (0.39)  −0.31 (0.35)  Chuck  0.65  0.28  −0.06    0.77 (0.16)  0.63 (0.24)  0.75 (0.15)  0.44 (0.33)  0.51 (0.24)  0.68 (0.21)  0.74 (0.18)  Blade  0.77  0.30  −0.06  0.66    0.87 (0.11)  0.82 (0.08)  0.46 (0.26)  0.81 (0.08)  0.85 (0.10)  0.85 (0.07)  Brisket  0.48  0.26  −0.23  0.48  0.59    0.82 (0.09)  0.74 (0.19)  0.74 (0.17)  0.73 (0.17)  0.71 (0.15)  Rib roast  0.60  0.26  −0.09  0.52  0.69  0.60    0.66 (0.15)  0.66 (0.14)  0.78 (0.11)  0.77 (0.12)  Strip-loin  0.68  0.36  0.07  0.52  0.57  0.44  0.54    0.41 (0.30)  0.35 (0.32)  0.46 (0.26)  Sirloin  0.71  0.36  −0.02  0.56  0.69  0.50  0.57  0.57    0.77 (0.12)  0.45 (0.16)  Round  0.77  0.42  −0.16  0.63  0.76  0.59  0.63  0.65  0.73    0.78 (0.12)  Fillet  0.68  0.24  −0.12  0.63  0.67  0.56  0.61  0.53  0.57  0.74    1CCW: cold carcass weight; CCON: carcass conformation; CFAT: carcass fat. All SE of the phenotypic correlations were ≤0.08. View Large Using the model including carcass weight as covariate, heritability for the forequarter cuts ranged from 0.34 (flat ribs) to 0.91 (blade); heritability of forequarter wholesale cut weight was 0.69. The heritability of hindquarter cuts ranged from 0.31 (rib roast) to 0.72 (round); the heritability of hindquarter wholesale cut weight was 0.55. DISCUSSION The objective of this study was to use commercial and experimental data to estimate genetic parameters for different beef wholesale cut weights and to determine their correlations with the currently recorded carcass weight, carcass conformation, and carcass fat. The 2 crossbred populations used in this study gave a fair representation of the types of animals on Irish beef farms. Evans et al. (2007) showed a high interdependency between dairy and beef herds in Ireland, where, in 2005, 58% of the calves born were beef crosses or beef-dairy crosses. The proportion of animals with unknown sires in the commercial data set also reflects reality in that there is a low level of sire recording at calf registration. The Irish Legislation by Statutory Instrument S.I.276/1999 (transposed from European Regulation 1760/2000 on identification and registration of bovine) specifies the compulsory recording, on the maternal side, of the breed and identification of the dam, and on the paternal side, of only the breed of the sire. Both data sets used in the present study were relatively small. However, across the literature reviewed for the genetic analysis of primal cuts, the populations studied were also of limited size and comparable with both of our data sets; Cantet et al. (2003) used 474 Angus animals, Cundiff et al. (1969) used 287 Hereford-, Angus-, and Shorthorn-crossed animals, and Brackelsberg et al. (1971) used 257 Hereford- and Angus-sired animals. Other studies using larger data sets focused on overall retail meat, fat, and bone yields and did not present estimates for the primal cuts; Shackelford et al. (1995) used 2,762 purebred and composite animals, Koch et al. (1982) studied 2,453 steers of various biological backgrounds, and Morris et al. (1999) used 1,962 animals from 3 large multi-breed breeding experiments. The current beef genetic evaluation system for carcass traits in Ireland is across breed, and the breeding objective includes a positive economic weight on carcass weight and carcass conformation, but a negative economic weight on carcass fat (Evans et al., 2007). The breeding objective also includes a negative economic weight on cow mature BW (Amer et al., 2001). Trait Means and Heritability Estimates The 2 data sets, commercial and experimental, used in the present study were different in origin in that the majority of the experimental data set was composed of steers (82% of animals), whereas the commercial data were made up of predominantly heifers (91% of animals). This was reflected in differences in observed average carcass weight between the 2 data sets: 337 and 290 kg observed in the experimental and commercial data sets, respectively. The heritability estimates observed in the experimental and the commercial data set for cold carcass weight were similar and are in accordance with the mean estimate of 0.40 reported by Rios Utrera and Van Vleck (2004) after an extensive review of heritability estimates for carcass traits across 56 studies. The large heritability estimates observed in the commercial data set for carcass conformation and fat, albeit with large SE, may be due to the poor distribution of the data in that data set. However, the heritabilities of EUROP carcass conformation and fat grading vary considerably between populations. Using a large data set on Irish crossbred cattle, Hickey et al. (2007) reported heritability estimates ranging from 0.04 to 0.36, and from 0.00 to 0.24 for conformation and fat, respectively, across 8 breed groups. Eriksson et al. (2003), using 2 distinct purebred populations of Swedish Charolais and Hereford, reported heritability estimates of 0.34 (Charolais) and 0.22 (Hereford) for carcass conformation, and 0.38 (Charolais) and 0.27 (Hereford) for carcass fat grading. The meat percentage and the retail product percentage were similar across the experimental (68%) and commercial (66%) data sets, reflecting differences in cutting procedures between the experimental and the commercial data set; the commercial cutting procedure applied a more severe cutting procedure on the individual cuts with the objective of neat presentation of the cuts on the supermarket shelves. These estimates were also consistent with values reported in the literature, which vary from 66 to 68% (Koch et al., 1982; Shackelford et al., 1995; Morris et al., 1999). The heritabilities for total meat weight in the experimental data set (0.68) and for total retail product weight in the commercial data set (0.54) were similar to those found by Koch et al. (1982; 0.58), Shackelford et al. (1995; 0.67), and Morris et al. (1999; 0.48), despite some differences in the definition of the trait; sum of roast and steak meat (Koch et al., 1982), sum of the weight of the boneless, totally trimmed retail cuts and 20% fat lean trim (Shackelford et al., 1995), or carcass components weight trimmed of fat (Morris et al., 1999). Between the 2 data sets, the individual cuts were generally heavier in the experimental data set than in the commercial data set, reflecting i) the difference in the representation of different sexes in the data sets and ii) the difference in cutting procedure (i.e., cutting and subcutaneous and seam fat trimming) as noted previously. The average forequarter primal weight differed between the experimental and commercial data sets, at 106 kg (31% of carcass weight) and 38 kg (13% of carcass weight), respectively. The difference was mainly attributable to the different cutting procedures adopted, as well as the different carcass weights, in the 2 data sets. In the commercial data set, the carcass cutting was driven by retailer demand and so, to a certain extent, by the cooking habits. Three categories of beef cuts exist in Ireland (Board Bia, 2008), which include the roast cuts (part of the chuck, brisket), the pot roast or braising cuts (part of the chuck, flat ribs, and flank), and the casserole cuts (blade). The remaining parts of the forequarter (neck, part of the shoulder muscles, part of the rib set, and part of the flank) are categorized as lean trimmings or “dice and stew beef,” to be sold as diced (stir fry) and ground beef. The total lean trimmings and total dice beef can be taken from the forequarter and the hindquarter. The heritability of total forequarter weight in the experimental data set (0.42) is similar to the estimate of 0.49 reported by Brackelsberg et al. (1971). The moderate to high heritability estimates in the present study for the different forequarter wholesale weights also corroborates previous estimates in other studies. Cundiff et al. (1969) reported heritability estimates ranging from 0.34 to 0.49 for the chuck and from 0.38 to 0.44 for the rib; Brackelsberg et al. (1971) calculated a moderate heritability (0.42) for the composite cut called chuck and rib. However, large differences in heritability estimates were observed between the experimental and commercial data sets for the same traits (for example, brisket, chuck, round, fillet). These observed discrepancies may be due to several factors such as i) population specific genetic parameters and ii) possible differences in genetic parameters between sexes; Crews and Kemp (2001) reported large differences in additive genetic variance for the LM area between bulls and heifers, iii) possible differences in the cutting methods even for well located cuts such as the chuck or brisket, and iv) possible confounding effects between genetics and unknown environmental effects. In addition, relatively large SE were observed for the heritability estimates of most traits in this study. Average hindquarter meat weight (sum of the wholesale cuts) was slightly greater in the experimental data set (97 kg; 29% of the cold carcass weight) than in the commercial data set (77 kg; 27% of the cold carcass weight). These small differences could be due to differences observed in the populations (breed, sex), but are more likely due to differences in cutting practices between the 2 data sets as described previously. The heritability estimate for total hindquarter weight (0.57) in the experimental data set is similar to the heritability of 0.57 reported by Brackelsberg et al. (1971) and 0.46 reported by Cantet et al. (2003) for the same trait. The generally high heritability estimates for the wholesale cuts located in the hindquarter in the present study agree with previous estimates. Cundiff et al. (1969) reported heritability estimates ranging from 0.07 to 0.48 for the loin and from 0.42 to 0.68 for the round, whereas Brackelsberg et al. (1971) reported a high heritability (0.81) for the composite cut called round and loin. The CVg for carcass weight in both data sets (6.0% for the experimental data set and 5.5% for the commercial data set) was similar to the average of 4.4% calculated from the results of Hickey et al. (2007) across 8 cattle populations in Ireland. The CVg for carcass conformation and fat in both of our studies were greater (>10.1%) than those calculated from the results presented by Hickey et al. (2007), where the average CVg for both carcass conformation and fat was 8.0%. The CVg for the different wholesale cut weights (2.2 to 14.2%) is consistent with the CVg reported in other studies for other performance traits such as growth rate (Arthur et al., 2001; 4.6 to 7.2%), feed intake (Arthur et al., 2001; 6.9 to 7.2%), and weaning weight (Phocas and Laloë, 2004; 6.0 to 8.0%). From the experimental study, the heritability estimate for total bone weight (0.75) was greater than heritability estimates reported in the literature. Shackelford et al. (1995) reported a heritability of 0.62 for carcass bones, whereas Koch et al. (1982) and Morris et al. (1999) reported heritability estimates of 0.57 and 0.51, respectively, for the same overall trait. For total fat weight, the low heritability observed in our experimental data set (0.27), although in agreement with the estimate reported by Morris et al. (1999; 0.30), is less than the heritability estimates reported by Shackelford et al. (1995; 0.65) and Koch et al. (1982; 0.47). In the experimental and commercial data sets, little change in the heritability estimates of the cuts was observed when carcass weight was included as a covariate in the model. The decision to report the heritability estimates for carcass cut weights was made because farmers will be paid on the yield of each cut and because the cut weights will be included in an overall breeding objective that accounts for potential unfavorable correlated responses in traits such as mature size. Relationship Among Carcass Traits The phenotypic correlations among carcass conformation, carcass fat, and carcass weight across the 2 data sets are consistent with those cited by Hickey et al. (2007), ranging from 0.17 to 0.38. The genetic correlations observed in the experimental data set between carcass weight and EUROP conformation and fat grading were negative (−0.24 to −0.14) and different from the positive, albeit weak, genetic correlations (0.11 to 0.26) reported by Hickey et al. (2007). Nonetheless, large SE were associated with all genetic correlations estimated in the present study, reflecting the limited sample size. As a result, the genetic correlation estimates were not statistically different from zero. The moderate to strong positive genetic correlations between carcass weight and the various wholesale cut weights in the present study were not unexpected given the part whole relationship between the different cuts and carcass weight. Therefore, selection for greater carcass weight will increase the weight of each of the cuts. Selection for increased carcass conformation will also be associated with an increase in individual wholesale cut weights with the exception of the shoulder, brisket, and strip-loin cuts as estimated from the experimental data, although the SE of the correlations were large. Carcass conformation and carcass fat tended to be more positively genetically correlated with the different wholesale cuts in the commercial data set compared with the experimental data set, thus reflecting the difference in the fat trimming procedure applied to the cuts in the 2 data sets. The genetic correlations between the different wholesale cut weights were moderate to strongly positive, in agreement with Brackelsberg et al. (1971), who also documented moderate to strong genetic correlations between the studied cuts (round and loin cut, chuck and rib cut, and round, loin and rib cut) ranging from 0.16 to 1.00. Cundiff et al. (1969) observed strong genetic correlations between 4 beef cuts (minimum genetic correlation of 0.72), namely, the round, loin, rib, and chuck. The results from our study show that direct selection on a primal beef cut would result in indirect positive genetic gain in all of the cuts, although some of the correlations were less than unity. The existence of moderate to large heritability estimates, albeit with large SE, and large CVg suggests that genetic selection for individual carcass cut weight may be fruitful. Genetic correlations among all beef wholesale cut weights were moderate to strongly positive, although some were less than unity, albeit sometimes with large SE, indicating a potential benefit of placing more emphasis on some greater value cuts to increase genetic gain in carcass value. This is further substantiated by the strong positive genetic correlations between the carcass weight and the cuts, implying i) that selection for increased carcass weight will, on average, increase the weight of each cut and ii) a benefit of more direct selection on high value cuts. Further research is to be undertaken on the feasibility of using routinely collected carcass digital images to predict weights of individual beef carcass cuts and to subsequently investigate the feasibility of genetically selecting for these traits to improve carcass value. LITERATURE CITED Amer P. R. Simm G. Keane M. G. Diskin M. G. Wickham B. W. 2001. Breeding objective for beef cattle in Ireland. Livest. Prod. Sci.  67: 223– 239. Google Scholar CrossRef Search ADS   Arthur P. F. Renand G. Krauss D. 2001. Genetic and phenotypic relationship among different measures of growth and feed efficiency in young Charolais bulls. Livest. Prod. Sci.  68: 131– 139. Google Scholar CrossRef Search ADS   Board Bia 2008. Simple steps to perfect beef.  http://www.bordbia.ie/Consumers/About_Food/Meat/Beef/Know_Your_Cuts/ Accessed June 9, 2008. Brackelsberg P. O. Kline E. A. Willham R. L. Hazel L. N. 1971. Genetic parameters for selected beef-carcass traits. J. Anim. Sci.  33: 13– 17. [PubMed] Google Scholar CrossRef Search ADS PubMed  Campion, B., M. G. Keane, D. A. Kenny, and D. P. Berry 2008. Effect of breed and genetic merit on feed intake and growth rate in beef × dairy and dairy steers. Page 49 (Abstract) in Proc.  Agricultural Research Forum, Tullamore, Ireland. Teagasc, Carlow, Ireland. Cantet R. J. C. Pedro Steibel J. Birchmeier A. N. Santa Coloma L. F. 2003. Bayesian estimation of genetic parameters for growth and carcass traits of grass-fed beef cattle by Full Conjugate Gibbs. Arch. Tierzucht  46: 435– 443. Crews D. H. Kemp R. A. 2001. Genetic parameters for ultrasound and carcass measures of yield and quality among replacement and slaughter beef cattle. J. Anim. Sci.  79: 3008– 3020. [PubMed] Google Scholar CrossRef Search ADS PubMed  Crump R. E. Wray N. R. Thompson R. Simm G. 1997. Assigning pedigree beef performance records to contemporary groups taking account of within-herd calving patterns. Anim. Sci.  65: 193– 198. Google Scholar CrossRef Search ADS   Cummins B. Keane M. G. O'Kiely P. Kenny D. A. 2007. Effects of breed type, silage harvest date and pattern of offering concentrates on intake, performance and carcass traits of finishing steers. Ir. J. Agric. Food Res.  46: 149– 168. Cundiff L. V. Gregory K. E. Koch R. M. Dickerson G. E. 1969. Genetic variation in total and differential growth of carcass components in beef cattle. J. Anim. Sci.  29: 233– 244. [PubMed] Google Scholar CrossRef Search ADS PubMed  Department of Agriculture and Food 2004. Expenditure review of beef carcass classification scheme.  The Department of Agriculture and Food, Dublin, Ireland. Eriksson S. Näsholm A. Johansson K. Philipsson J. 2003. Genetic analysis of field recorded growth and carcass traits for Swedish beef cattle. Livest. Prod. Sci.  84: 53– 62. Google Scholar CrossRef Search ADS   Evans, R. D., T. Pabiou, A. R. Cromie, F. Kearney, and B. Wickham 2007. Genetic improvement in the Irish suckler beef herd: Industry expectation and experience so far. Page 7 in Proc. Interbull Technical Workshop, Paris, France, March 9–10, 2007,  Bulletin No. 36. http://www-interbull.slu.se/bulletins/framesida-pub.htm Accessed Dec. 6, 2009. Gerrard, F., and F. J. Mallion 1977. Beef and beef and veal cutting. Page 293 in The Complete Book of Meat. Virtue and Company Limited.  The Trinity Press, Worcester, UK. Gilmour, A. R., B. J. Gogel, B. R. Cullis, and R. Thompson 2006. ASReml User Guide Release 2.0.  VSN Int. Ltd., Hemel Hempstead, UK. Hickey J. M. Keane M. G. Kenny D. A. Cromie A. R. Veerkamp R. F. 2007. Genetic parameters for EUROP carcass traits within different groups of cattle in Ireland. J. Anim. Sci.  85: 314– 321. [PubMed] Google Scholar CrossRef Search ADS PubMed  Houle D. 1992. Comparing evolvability and variability of quantitative traits. Genetics  130: 195– 204. [PubMed] Google Scholar PubMed  Jones, S. J., A. Guru, V. Singh, B. Carpenter, C. R. Calkins, and D. Johnson 2004. Bovine Myology and Muscle Profiling.  http://bovine.unl.edu Accessed May 22, 2008. Jorjani H. Klei L. Emanuelson U. 2003. A simple method for weighted bending of genetic (co)variance matrices. J. Dairy Sci.  86: 677– 679. [PubMed] Google Scholar CrossRef Search ADS PubMed  Keane, M. G., E. G. O'Riordan, and P. O'Kiely 2008. System of heifer production. Page 38 in Dairy calf-to-beef production systems.  Teagasc Publication. Teagasc, Oak Park, Carlow, Ireland. Koch R. M. Cundiff L. V. Gregory K. E. 1982. Heritability and genetic, environmental and phenotypic correlations of carcass traits in a population of diverse biological types and their implications in selection programs. J. Anim. Sci.  55: 1319– 1329. Google Scholar CrossRef Search ADS   Morris C. A. Cullen N. G. McCall D. G. 1999. Genetic and phenotypic relationships among carcass measurements in beef cattle. N. Z. J. Agric. Res.  42: 415– 421. Google Scholar CrossRef Search ADS   Phocas F. Laloë D. 2004. Genetic parameters for birth and weaning traits in French specialized beef cattle breeds. Livest. Prod. Sci.  89: 121– 128. Google Scholar CrossRef Search ADS   Rios Utrera A. Van Vleck L. D. 2004. Heritability estimates for carcass traits of cattle: A review. Genet. Mol. Res.  3: 380– 394. [PubMed] Google Scholar PubMed  Shackelford S. D. Cundiff L. V. Gregory K. E. Koohmaraie M. 1995. Predicting beef carcass cutability. J. Anim. Sci.  73: 406– 413. [PubMed] Google Scholar CrossRef Search ADS PubMed  Strydom P. E. Smith M. F. 2005. Predicting yields of high priced trimmed beef cuts by means of carcass weight and visual assessments of fat cover and conformation. S. Afr. J. Anim. Sci.  35: 195– 205. Van Der Werf J. H. J. De Boer W. 1989. Influence of non-additive effects on estimation of genetic parameters in dairy cattle. J. Dairy Sci.  72: 2606– 2614. Google Scholar CrossRef Search ADS   American Society of Animal Science

Elzo, M. A.;Riley, D. G.;Hansen, G. R.;Johnson, D. D.;Myer, R. O.;Coleman, S. W.;Chase, C. C.;Wasdin, J. G.;Driver, J. D.

doi: 10.2527/jas.2008-1553pmid: 19684277

ABSTRACT The influence of additive and nonadditive genetic effects and temperament on 4 postweaning feed intake and growth traits was evaluated in a group of 581 bull, heifer, and steer calves born in 3 Florida herds in 2006 and 2007. Calves had breed compositions ranging from 100% Angus (A) to 100% Brahman (B). They were randomly allocated to 24 pens each year by herd (Brooksville, Gainesville, Marianna, FL), sire group (A, 3/4 A 1/4 B, Brangus, 1/2 A 1/2 B, 1/4 A 3/4 B, and B), and sex (bull, heifer, and steer) in a GrowSafe automated feeding facility at Marianna. Calves were fed a concentrate diet during the 21-d adjustment and the 70-d trial periods. Individual feed intakes were recorded daily, and BW, chute scores, and exit velocities were recorded every 2 wk. Traits were phenotypic daily residual feed intake (RFI), mean daily feed intake (DFI), mean daily feed conversion ratio (FCR), and postweaning BW gain. Phenotypic RFI was computed as the difference between actual and expected feed intakes. Calves were assigned to 3 RFI groups: high (RFI greater than 0.9 kg of DM/d), low (RFI less than −0.9 kg of DM/d), and medium (RFI between mean ± 0.9 kg of DM/d; SD = 1.8 kg of DM/d). The mixed model included the fixed effects of contemporary group (herd-year-pen), RFI group (except when trait was RFI), age of dam, sex of calf, age of calf, B fraction of calf, heterozygosity of calf, mean chute score, and mean exit velocity. Brahman fraction and heterozygosity of calf were nested within sex of calf for RFI and within RFI group for DFI, FCR, and postweaning BW gain. Random effects were sire and residual. Feed efficiency tended to improve (decreased RFI) as the B fraction increased. However, calves required larger amounts of feed per kilogram of BW gain (larger FCR) as the B fraction increased. Postweaning BW gain tended to decrease as the B fraction increased. Temperament traits were unimportant for all traits except exit velocity for DFI, suggesting perhaps a lack of variation for temperament traits in this herd, or that calves became accustomed to the level of handling pre- and postweaning, thus decreasing behavioral differences among them. INTRODUCTION In an era of dwindling resources and escalating costs of production, efficiency of feed utilization has become an essential component of beef cattle production systems. Particular attention has been devoted to phenotypic daily residual feed intake (RFI; Koch et al., 1963), defined as the difference between the actual feed intake of an animal and the expected feed intake over a time period. For growing cattle, expected feed intake is usually computed as a regression of intake on ADG and metabolic midweight (Archer et al., 1997; Arthur et al., 2001a,b; Nkrumah et al., 2004, 2007). This makes RFI phenotypically independent of ADG and metabolic midweight by design (Kennedy et al., 1993; Crews, 2005). Largely because of this property, RFI has become the preferred measure of efficiency of feed utilization. Research on RFI and postweaning growth in beef cattle has been concentrated in temperate regions and has used either purebred (Archer et al., 1997; Arthur et al., 2001a,b; Schenkel et al., 2004) or crossbred (Basarab et al., 2003; Nkrumah et al., 2004,2007) Bos taurus cattle. Breed has been found to affect both growth and RFI (Nkrumah et al., 2004; Schenkel et al., 2004), and temperament scores have been associated with growth in beef cattle (Burrow and Dillon, 1997; Voisinet et al., 1997). Studies involving RFI and its association with breed composition, postweaning growth, and temperament using B. taurus and Bos indicus cattle in subtropical and tropical environments may yield substantially different results from those obtained for B. taurus breeds in temperate regions. Thus, the objectives of this research were to assess the effect of breed composition and indicators of temperament [chute score (CS) and exit velocity (EV)] on RFI, daily feed intake (DFI), feed conversion ratio (FCR), and postweaning growth, and to estimate genetic parameters for these traits in a group of bulls, heifers, and steers ranging from 100% Angus (A) to 100% Brahman (B). MATERIALS AND METHODS Research protocols were approved by the University of Florida Institutional Animal Care and Use Committee. Animals and Preconditioning Animals (n = 581) were from 3 Florida herds of beef cattle located in Brooksville (n = 100), Gainesville (n = 388), and Marianna (n = 93), Florida. Calves were born in 2006 (n = 330) and 2007 (n = 251). Established standards for animal care and use were followed. There were 31 bulls, 317 heifers, and 233 steers from 6 breed groups: A (n = 153), 3/4 A 1/4 B (n = 66), Brangus (5/8A 3/8 B; n = 107), 1/2 A 1/2 B (n = 115), 1/4 A 3/4 B (n = 49), and B (n = 91). Calves from the 3 locations were born between December and March and were weaned in August or early September in 2006 and 2007. After weaning, calves grazed on bahiagrass (Paspalum notatum) pastures and received a preconditioning diet for 3 to 6 wk in preparation for their feed intake trial at the Feed Efficiency Facility of the Institute of Food and Agricultural Sciences of the University of Florida (IFEF), equipped with GrowSafe technology (GrowSafe Systems Ltd., Airdrie, Alberta, Canada), in Marianna. Heifers from Brooksville received a preconditioning cottonseed- and soybean meal-based ration (1.8 kg/d; 14% CP; 488 Pellet Medicated Weaning Ration, Lakeland Animal Nutrition, Lakeland, FL), bahiagrass hay, and a free-choice mineral (Brooksville Research Mineral, Lakeland Animal Nutrition). Calves from Gainesville were fed concentrate (1.6 kg to 3.6 kg/d; 14.0% CP; 488 Pellet Medicated Weaning Ration, Lakeland Animal Nutrition; and soyhull pellets), bahiagrass hay, and a free-choice mineral (UF University Special Hi-Cu Mineral, University of Florida, Animal Science Department, Gainesville). Calves from Marianna were given 2.5 kg of a mixed feed (75% soyhull pellets and 25% corn gluten pellets; 15.1% CP). They also had access to free-choice Tifton 85 bermudagrass hay and a complete mineral (Southern States, Marianna, FL). Subsequently, calves were moved to the IFEF in September (Gainesville and Marianna) and October (Brooksville). Management, Nutrition, and Data Collection at the IFEF Calves were randomly allocated to 24 pens at the IFEF within herd (Brooksville, Gainesville, and Marianna) by sire group [A, 3/4 A 1/4 B, Brangus (5/8 A 3/8 B), 1/2 A 1/2 B, 1/4 A 3/4 B, and B] by sex (bull, heifer, and steer) subclass in 2006 and in 2007. The IFEF has 24 pens (108 m2/pen) equipped with 2 GrowSafe feed nodes in each pen. Calves were identified with passive, half-duplex, transponder ear tags (Allflex USA Inc., Dallas-Fort Worth, TX) before entering the IFEF facility. The mean stocking rate was 16.9 calves per pen. Each pen had 2 GrowSafe nodes; thus, the mean stocking rate per node was 8.5 calves per GrowSafe node, and the range was from 7.5 to 10 calves per GrowSafe node. Animals were fed a concentrate diet ad libitum twice daily. The 2006 concentrate diet was composed of whole corn, soybean hulls, corn gluten feed, cottonseed hulls, and a protein, vitamin, and mineral supplement (FRM, Bainbridge, GA). The 2007 diet had greater fiber content (chopped bermudagrass instead of soybean hull pellets). The concentrate had a DM, CP, NEm, and NEg of 91.2%, 17.3%, 1.7 mcal/kg of DM, and 1.2 mcal/kg of DM in 2006, and 90.0%, 14.1%, 1.5 mcal/kg of DM, and 0.9 mcal/kg of DM in 2007. The pretrial adjustment period was 21 d, and the trial period lasted 70 d. GrowSafe software recorded individual feed intake in real time. Body weights and temperament traits were taken every 2 wk. Traits Traits were phenotypic RFI, mean DFI, mean daily FCR, and postweaning BW gain (PWG). All intake traits were converted to a DM basis before analyses. Individual feed disappearance and node attendance data were recorded using GrowSafe Data Acquisition software (GrowSafe Systems Ltd.) and DFI were compiled using the Process Intakes routine. Individual DFI data were excluded from the analysis because of equipment failure or when the proportion of daily feed assigned to individual animals (leakage) at any feeding station was less than 94% for a given day. Individual pens constituted a feeding station (2 GrowSafe nodes). A total of 3,220 feeding station days were produced during the study, with 19 and 28 d excluded from the data set in 2006 and 2007, respectively. The GrowSafe system accounted for 99.53 ± 1.49% (range = 75.12 to 100%) of the feed delivered to the nodes being assigned to individual animals, showing the software to be robust. Feed disappearance when comparing feed delivery equipment with the GrowSafe system (2006 only) was within 98.52% agreement between the 2 weighing systems, indicating highly accurate data. Animals with individual means for feed intake 3 SD below or above the sample mean were considered outliers and were removed from the analysis. Phenotypic daily RFI was computed as the difference between actual average DFI and expected DFI (Koch et al., 1963; Archer et al., 1997; Arthur et al., 2001a,b) during the 70-d postweaning feeding trial. Expected DFI was estimated as a linear regression of average DFI on ADG and metabolic midweight. Thus, expected DFI was computed across all pens, years, breed groups, and sexes of calves. The proportion of the variation for average DFI explained by this model (i.e., R2 value) was 30%. This R2 value computed across subclasses (pens, years, breed groups, and sexes of calves) was less than 50% of R2 values estimated within subclasses (steers within years: Basarab et al., 2003; bulls within test groups: Schenkel et al., 2004). The smaller R2 value obtained here was likely because the regression of DFI on ADG and metabolic midweight was done across pens, years, breed groups, and sexes of calves. Average daily gain was computed as the regression of calf BW on test day, using BW taken every 2 wk at the IFEF. Midweight was computed as the sum of the regression estimate for initial BW plus the regression estimate for ADG times 35 d. Metabolic midweight was equal to the estimated midweight to the power of 0.75. Total DFI was the sum of all measurements of feed intake for each animal measured by the GrowSafe system for a given day, and DFI was the average DFI over the 70-d trial period. Mean daily FCR was computed as the ratio of DFI to ADG. Postweaning BW gain was computed as the difference between the BW of each calf at the end and at the beginning of the 70-d trial. Mean CS was the average of the 6 CS measured every 2 wk (1 = docile; 2 = restless; 3 = nervous; 4 = flighty; 5 = aggressive; 6 = very aggressive; Beef Improvement Federation, 2002) during the 70-d trial period. Similarly, mean EV was the average of the 6 EV (velocity out of the chute in meters per second; Burrow and Dillon, 1997; Curley et al., 2006) measurements taken every 2 wk at the IFEF. Statistical Analysis Calves were assigned to 3 groups according to their RFI values (Nkrumah et al., 2004). The RFI groups were high (calf RFI > mean + 0.5 SD), medium (calf RFI between mean ± 0.5 SD; SD = 1.8 kg), and low (calf RFI < mean − 0.5 SD). Feed efficiency increases from high to medium to low RFI groups; thus, calves in the low RFI group were the most efficient and calves in the high RFI group were the least efficient. Allocation of calves to low, medium, and high RFI groups was done using the complete data set. Numbers of calves per RFI group appeared to depend on their breed composition. Thus, a log-linear analysis (Stokes et al., 2000; Agresti, 2002) of calf frequencies per breed group by RFI group was conducted to test the hypothesis of independence between breed group and RFI group. Computations were performed using the CATMOD procedure (SAS Inst. Inc., Cary, NC). Traits were analyzed individually using mixed models. Fixed effects were contemporary group (herd-year-pen; herd = Brooksville, Gainesville, and Marianna; year = 2006 and 2007; pen = 1 to 24), RFI group (except when trait was RFI), age of dam (1 = 3 yr, 2 = 4 yr, and 3 = 5 yr and older), sex of calf (1 = bull, 2 = heifer, and 3 = steer), age of calf, B fraction of calf, heterozygosity of calf (i.e., probability of A and B alleles at 1 locus in the calf), mean CS, and mean EV. Brahman fraction and heterozygosity of calf were nested within sex of calf for RFI, and within RFI group for the other traits. Preliminary models for DFI, FCR, and PWG contained the B fraction and heterozygosity nested within sex of calf and within RFI group. The B fraction and heterozygosity nested within sex of calf were nonsignificant for these traits and thus were dropped from the final models. Random effects were sire and residual. Sires effects were assumed to have a mean of zero, a common variance of and to be uncorrelated. Similarly, residual effects were assumed to have a mean of zero, a common variance of and to be uncorrelated. Mixed model analyses were carried out using the MIXED procedure of SAS. Sex of calf least squares means for RFI and for differences between sexes for the 6 breed groups defined here were computed using the expected B fraction of each group (0.0 for A, 0.25 for 3/4 A 1/4 B, 0.375 for Brangus, 0.5 for 1/2 A 1/2 B, 0.75 for 1/4 A 3/4 B, and 1.0 for B) and the expected heterosis of each group (0.0 for A, 0.5 for 3/4 A 1/4 B, 0.469 for Brangus, 1.0 for 1/2 A 1/2 B, 0.5 for 1/4 A 3/4 B, and 0.0 for B). Similarly, RFI group least squares means for DFI, FCR, and PWG, and differences between RFI groups for the 6 breed groups were computed using the expected B fractions and heterosis for each breed group. Bonferroni t-tests were used to compare pairs of least squares means. Restricted maximum likelihood estimates of heritabilities for, and genetic and phenotypic correlations among, RFI, DFI, FCR, and PWG were obtained using ASREML (Gilmour et al., 1999). Genetic parameters were estimated using 2-trait analyses (RFI-DFI, RFI-FCR, RFI-PWG, DFI-FCR, DFI-PWG, and FCR-PWG) because of the small size of the data set (655 calves with records). The same model was used for all traits. Fixed effects were contemporary group, age of dam, sex of calf, age of calf, B fraction of calf nested within sex of calf, heterozygosity of calf nested within sex of calf, mean CS, and mean EV. Random effects were calf and residual. Calf effects were assumed to have a mean of zero, and variance equal to the relationship matrix times (direct product) the 2 × 2 variance-covariance matrix of additive genetic effects for each pair of traits analyzed. For example, the elements of the additive genetic variance-covariance matrix for the analysis of RFI and DFI were addvar(RFI), addcov(RFI, DFI), addcov(DFI, RFI), and addvar(DFI), where addvar is additive genetic variance, and addcov is additive genetic covariance. The relationship matrix included 655 calves with records and all known ancestors. The total number of animals in the relationship matrix was 1,712 (655 calves, 71 sires of calves, 464 dams of calves, and 522 other ancestors). Residual effects were assumed to have a mean of zero, a common variance of and to be uncorrelated. RESULTS AND DISCUSSION Description of Data Table 1 contains a description of the data (number of calves, mean, SD) by trait for each breed group of calf × RFI group subclass and for the complete data set. A total of 144 calves were allocated to the high RFI group, 262 calves to the medium RFI group, and 175 calves to the low RFI group. Numbers of calves per breed group ranged from 14 (1/4 A 3/4 B) to 37 (1/2 A 1/2 B) for the high RFI group, from 23 (B) to 72 (A) for the medium RFI group, and from 8 (1/4 A 3/4 B) to 51 (A) for the low RFI group. The log-linear analysis of calf frequencies per breed group × RFI group yielded a highly significant likelihood ratio (P < 0.0001), indicating that frequencies of calves per breed group × RFI group categories were not independent. Brahman had a greater proportion of calves in the low RFI group (51.6%; Figure 1) than did other breed groups (16.3 to 33.3%; Figure 1). Greater fractions of calves in the medium RFI group existed in Brangus (58.9%) and 1/4 A 3/4 B (55.1%) than in other breed groups, with B having the least percentages (25.3%; Figure 1). Brangus (18.7%), A (19.6%), and B (23.1%) had smaller percentages of calves in the high RFI group than did the remaining 3 crossbred groups (28.6% to 32.2%). Numbers of calves, means, and SD per breed group of calf by residual feed intake group and total Table 1. Numbers of calves, means, and SD per breed group of calf by residual feed intake group and total       Trait2        RFI, kg of DM/d  DFI, kg of DM/d  FCR, kg of DM·d−1/kg of BW gain·d−1  PWG, kg  Breed group  RFI group1  N  Mean  SD  Mean  SD  Mean  SD  Mean  SD  Angus  High  30  2.34  1.14  11.72  1.38  10.43  2.39  81.61  23.79  Angus  Medium  72  −0.10  0.50  9.45  1.11  7.36  1.56  95.16  22.70  Angus  Low  51  −1.70  0.83  7.61  1.39  6.81  2.53  84.38  29.28  3/4 A 1/4 B3  High  22  2.33  1.33  12.38  1.13  9.69  2.81  94.55  21.60  3/4 A 1/4 B  Medium  33  −0.02  0.45  9.93  1.34  7.16  1.45  101.79  21.99  3/4 A 1/4 B  Low  11  −2.34  1.80  7.79  2.15  6.49  2.17  89.59  22.38  Brangus  High  20  2.95  1.88  12.54  2.00  10.08  2.09  89.41  19.96  Brangus  Medium  63  −0.11  0.58  9.80  0.97  6.93  1.21  103.41  20.66  Brangus  Low  24  −1.58  0.57  8.23  1.12  6.14  0.99  95.91  22.65  1/2 A 1/2 B  High  37  2.42  1.34  12.26  1.51  11.05  2.51  82.80  23.94  1/2 A 1/2 B  Medium  44  −0.13  0.56  9.70  1.36  7.77  2.21  94.41  27.25  1/2 A 1/2 B  Low  34  −1.92  0.64  7.61  1.07  6.70  1.36  84.30  18.38  1/4 A 3/4 B  High  14  2.51  1.10  11.52  1.22  10.96  2.05  76.98  16.08  1/4 A 3/4 B  Medium  27  −0.16  0.44  9.52  1.00  7.41  1.22  94.07  14.52  1/4 A 3/4 B  Low  8  −1.35  0.25  8.20  0.67  6.86  0.76  86.70  14.75  Brahman  High  21  2.24  1.01  10.93  1.39  11.24  2.31  69.65  17.84  Brahman  Medium  23  −0.04  0.58  8.57  1.02  9.53  2.75  69.84  25.62  Brahman  Low  47  −2.21  0.89  6.36  1.28  6.94  1.73  70.54  25.6  Total  High  144  2.45  1.32  11.94  0.13  10.60  0.20  82.78  1.87  Total  Medium  262  −0.10  0.52  9.56  0.07  7.50  0.11  95.52  1.49  Total  Low  175  −1.89  0.89  7.40  0.11  6.71  0.14  82.66  1.94  Total  Total  581  −0.01  1.83  9.50  0.09  8.03  0.10  88.49  1.04        Trait2        RFI, kg of DM/d  DFI, kg of DM/d  FCR, kg of DM·d−1/kg of BW gain·d−1  PWG, kg  Breed group  RFI group1  N  Mean  SD  Mean  SD  Mean  SD  Mean  SD  Angus  High  30  2.34  1.14  11.72  1.38  10.43  2.39  81.61  23.79  Angus  Medium  72  −0.10  0.50  9.45  1.11  7.36  1.56  95.16  22.70  Angus  Low  51  −1.70  0.83  7.61  1.39  6.81  2.53  84.38  29.28  3/4 A 1/4 B3  High  22  2.33  1.33  12.38  1.13  9.69  2.81  94.55  21.60  3/4 A 1/4 B  Medium  33  −0.02  0.45  9.93  1.34  7.16  1.45  101.79  21.99  3/4 A 1/4 B  Low  11  −2.34  1.80  7.79  2.15  6.49  2.17  89.59  22.38  Brangus  High  20  2.95  1.88  12.54  2.00  10.08  2.09  89.41  19.96  Brangus  Medium  63  −0.11  0.58  9.80  0.97  6.93  1.21  103.41  20.66  Brangus  Low  24  −1.58  0.57  8.23  1.12  6.14  0.99  95.91  22.65  1/2 A 1/2 B  High  37  2.42  1.34  12.26  1.51  11.05  2.51  82.80  23.94  1/2 A 1/2 B  Medium  44  −0.13  0.56  9.70  1.36  7.77  2.21  94.41  27.25  1/2 A 1/2 B  Low  34  −1.92  0.64  7.61  1.07  6.70  1.36  84.30  18.38  1/4 A 3/4 B  High  14  2.51  1.10  11.52  1.22  10.96  2.05  76.98  16.08  1/4 A 3/4 B  Medium  27  −0.16  0.44  9.52  1.00  7.41  1.22  94.07  14.52  1/4 A 3/4 B  Low  8  −1.35  0.25  8.20  0.67  6.86  0.76  86.70  14.75  Brahman  High  21  2.24  1.01  10.93  1.39  11.24  2.31  69.65  17.84  Brahman  Medium  23  −0.04  0.58  8.57  1.02  9.53  2.75  69.84  25.62  Brahman  Low  47  −2.21  0.89  6.36  1.28  6.94  1.73  70.54  25.6  Total  High  144  2.45  1.32  11.94  0.13  10.60  0.20  82.78  1.87  Total  Medium  262  −0.10  0.52  9.56  0.07  7.50  0.11  95.52  1.49  Total  Low  175  −1.89  0.89  7.40  0.11  6.71  0.14  82.66  1.94  Total  Total  581  −0.01  1.83  9.50  0.09  8.03  0.10  88.49  1.04  1High = RFI > mean + 0.5 SD; medium = RFI between mean ± 0.5 SD; low = RFI < mean − 0.5 SD; SD = 1.8 kg. 2RFI = residual feed intake; DFI = mean daily feed intake; FCR = mean daily feed conversion ratio; PWG = postweaning BW gain. 3A = Angus; B = Brahman. View Large Table 1. Numbers of calves, means, and SD per breed group of calf by residual feed intake group and total       Trait2        RFI, kg of DM/d  DFI, kg of DM/d  FCR, kg of DM·d−1/kg of BW gain·d−1  PWG, kg  Breed group  RFI group1  N  Mean  SD  Mean  SD  Mean  SD  Mean  SD  Angus  High  30  2.34  1.14  11.72  1.38  10.43  2.39  81.61  23.79  Angus  Medium  72  −0.10  0.50  9.45  1.11  7.36  1.56  95.16  22.70  Angus  Low  51  −1.70  0.83  7.61  1.39  6.81  2.53  84.38  29.28  3/4 A 1/4 B3  High  22  2.33  1.33  12.38  1.13  9.69  2.81  94.55  21.60  3/4 A 1/4 B  Medium  33  −0.02  0.45  9.93  1.34  7.16  1.45  101.79  21.99  3/4 A 1/4 B  Low  11  −2.34  1.80  7.79  2.15  6.49  2.17  89.59  22.38  Brangus  High  20  2.95  1.88  12.54  2.00  10.08  2.09  89.41  19.96  Brangus  Medium  63  −0.11  0.58  9.80  0.97  6.93  1.21  103.41  20.66  Brangus  Low  24  −1.58  0.57  8.23  1.12  6.14  0.99  95.91  22.65  1/2 A 1/2 B  High  37  2.42  1.34  12.26  1.51  11.05  2.51  82.80  23.94  1/2 A 1/2 B  Medium  44  −0.13  0.56  9.70  1.36  7.77  2.21  94.41  27.25  1/2 A 1/2 B  Low  34  −1.92  0.64  7.61  1.07  6.70  1.36  84.30  18.38  1/4 A 3/4 B  High  14  2.51  1.10  11.52  1.22  10.96  2.05  76.98  16.08  1/4 A 3/4 B  Medium  27  −0.16  0.44  9.52  1.00  7.41  1.22  94.07  14.52  1/4 A 3/4 B  Low  8  −1.35  0.25  8.20  0.67  6.86  0.76  86.70  14.75  Brahman  High  21  2.24  1.01  10.93  1.39  11.24  2.31  69.65  17.84  Brahman  Medium  23  −0.04  0.58  8.57  1.02  9.53  2.75  69.84  25.62  Brahman  Low  47  −2.21  0.89  6.36  1.28  6.94  1.73  70.54  25.6  Total  High  144  2.45  1.32  11.94  0.13  10.60  0.20  82.78  1.87  Total  Medium  262  −0.10  0.52  9.56  0.07  7.50  0.11  95.52  1.49  Total  Low  175  −1.89  0.89  7.40  0.11  6.71  0.14  82.66  1.94  Total  Total  581  −0.01  1.83  9.50  0.09  8.03  0.10  88.49  1.04        Trait2        RFI, kg of DM/d  DFI, kg of DM/d  FCR, kg of DM·d−1/kg of BW gain·d−1  PWG, kg  Breed group  RFI group1  N  Mean  SD  Mean  SD  Mean  SD  Mean  SD  Angus  High  30  2.34  1.14  11.72  1.38  10.43  2.39  81.61  23.79  Angus  Medium  72  −0.10  0.50  9.45  1.11  7.36  1.56  95.16  22.70  Angus  Low  51  −1.70  0.83  7.61  1.39  6.81  2.53  84.38  29.28  3/4 A 1/4 B3  High  22  2.33  1.33  12.38  1.13  9.69  2.81  94.55  21.60  3/4 A 1/4 B  Medium  33  −0.02  0.45  9.93  1.34  7.16  1.45  101.79  21.99  3/4 A 1/4 B  Low  11  −2.34  1.80  7.79  2.15  6.49  2.17  89.59  22.38  Brangus  High  20  2.95  1.88  12.54  2.00  10.08  2.09  89.41  19.96  Brangus  Medium  63  −0.11  0.58  9.80  0.97  6.93  1.21  103.41  20.66  Brangus  Low  24  −1.58  0.57  8.23  1.12  6.14  0.99  95.91  22.65  1/2 A 1/2 B  High  37  2.42  1.34  12.26  1.51  11.05  2.51  82.80  23.94  1/2 A 1/2 B  Medium  44  −0.13  0.56  9.70  1.36  7.77  2.21  94.41  27.25  1/2 A 1/2 B  Low  34  −1.92  0.64  7.61  1.07  6.70  1.36  84.30  18.38  1/4 A 3/4 B  High  14  2.51  1.10  11.52  1.22  10.96  2.05  76.98  16.08  1/4 A 3/4 B  Medium  27  −0.16  0.44  9.52  1.00  7.41  1.22  94.07  14.52  1/4 A 3/4 B  Low  8  −1.35  0.25  8.20  0.67  6.86  0.76  86.70  14.75  Brahman  High  21  2.24  1.01  10.93  1.39  11.24  2.31  69.65  17.84  Brahman  Medium  23  −0.04  0.58  8.57  1.02  9.53  2.75  69.84  25.62  Brahman  Low  47  −2.21  0.89  6.36  1.28  6.94  1.73  70.54  25.6  Total  High  144  2.45  1.32  11.94  0.13  10.60  0.20  82.78  1.87  Total  Medium  262  −0.10  0.52  9.56  0.07  7.50  0.11  95.52  1.49  Total  Low  175  −1.89  0.89  7.40  0.11  6.71  0.14  82.66  1.94  Total  Total  581  −0.01  1.83  9.50  0.09  8.03  0.10  88.49  1.04  1High = RFI > mean + 0.5 SD; medium = RFI between mean ± 0.5 SD; low = RFI < mean − 0.5 SD; SD = 1.8 kg. 2RFI = residual feed intake; DFI = mean daily feed intake; FCR = mean daily feed conversion ratio; PWG = postweaning BW gain. 3A = Angus; B = Brahman. View Large Figure 1. View largeDownload slide Percentage of calves by breed group of calf and residual feed intake (RFI) group. High RFI group = RFI > mean + 0.5 SD; medium RFI group = RFI between mean ± 0.5 SD; low RFI group = RFI < mean − 0.5 SD; SD = 1.8 kg; A = Angus; B = Brahman. Figure 1. View largeDownload slide Percentage of calves by breed group of calf and residual feed intake (RFI) group. High RFI group = RFI > mean + 0.5 SD; medium RFI group = RFI between mean ± 0.5 SD; low RFI group = RFI < mean − 0.5 SD; SD = 1.8 kg; A = Angus; B = Brahman. Mean RFI was similar across breed groups for the high (from 2.24 kg for B to 2.95 kg for Brangus), medium (from −0.16 kg for 1/4 A 3/4 B to −0.02 for 3/4 A 1/4 B), and low RFI groups (from −2.34 kg for 3/4 A 1/4 B to −1.35 kg for 1/4 A 3/4 B). Mean DFI and mean FCR decreased from the high to the low RFI group for all breed groups of calves (mean DFI of 11.94, 9.56, and 7.40 kg, and mean FCR of 10.60, 7.50, and 6.71 kg of feed/kg of BW gain for the high, medium, and low RFI groups for the complete data set). Means for PWG tended to be greater for the medium RFI group than for the high and low RFI groups, and averaged 82.78 kg for the high RFI group, 95.52 kg for the medium RFI group, and 82.66 kg for the low RFI group over the complete data set. Daily RFI Daily RFI was affected by contemporary group (P < 0.0001), sex of calf (P < 0.003), and B fraction of calf nested within sex of calf (P < 0.0001; Table 2). Age of dam, age of calf, heterosis of calf within sex of calf, mean CS, and mean EV were not important sources of variation. Significance values (F-test) for fixed effects Table 2. Significance values (F-test) for fixed effects   Trait1  Effect  RFI  DFI  FCR  PWG  Herd-year-pen  <0.0001  <0.0002  <0.0001  <0.0001  Age of dam  0.41  0.12  0.65  0.28  Sex of calf  0.003  <0.0001  <0.0001  <0.0001  Age of calf  0.06  0.01  0.0004  0.04  RFI group    <0.0001  0.78  0.09  Brahman fraction nested within sex of calf  <0.0001        Heterosis nested within sex of calf  0.24        Brahman fraction nested within RFI group    0.0009  0.0073  0.0004  Heterosis nested within RFI group    0.0002  0.22  0.20  Mean chute score  0.39  0.42  0.11  0.33  Mean exit velocity  0.89  0.0012  0.34  0.31    Trait1  Effect  RFI  DFI  FCR  PWG  Herd-year-pen  <0.0001  <0.0002  <0.0001  <0.0001  Age of dam  0.41  0.12  0.65  0.28  Sex of calf  0.003  <0.0001  <0.0001  <0.0001  Age of calf  0.06  0.01  0.0004  0.04  RFI group    <0.0001  0.78  0.09  Brahman fraction nested within sex of calf  <0.0001        Heterosis nested within sex of calf  0.24        Brahman fraction nested within RFI group    0.0009  0.0073  0.0004  Heterosis nested within RFI group    0.0002  0.22  0.20  Mean chute score  0.39  0.42  0.11  0.33  Mean exit velocity  0.89  0.0012  0.34  0.31  1RFI = residual feed intake; DFI = mean daily feed intake; FCR = mean daily feed conversion ratio; PWG = postweaning BW gain. Empty cells indicate effects that were not modeled for the respective traits. View Large Table 2. Significance values (F-test) for fixed effects   Trait1  Effect  RFI  DFI  FCR  PWG  Herd-year-pen  <0.0001  <0.0002  <0.0001  <0.0001  Age of dam  0.41  0.12  0.65  0.28  Sex of calf  0.003  <0.0001  <0.0001  <0.0001  Age of calf  0.06  0.01  0.0004  0.04  RFI group    <0.0001  0.78  0.09  Brahman fraction nested within sex of calf  <0.0001        Heterosis nested within sex of calf  0.24        Brahman fraction nested within RFI group    0.0009  0.0073  0.0004  Heterosis nested within RFI group    0.0002  0.22  0.20  Mean chute score  0.39  0.42  0.11  0.33  Mean exit velocity  0.89  0.0012  0.34  0.31    Trait1  Effect  RFI  DFI  FCR  PWG  Herd-year-pen  <0.0001  <0.0002  <0.0001  <0.0001  Age of dam  0.41  0.12  0.65  0.28  Sex of calf  0.003  <0.0001  <0.0001  <0.0001  Age of calf  0.06  0.01  0.0004  0.04  RFI group    <0.0001  0.78  0.09  Brahman fraction nested within sex of calf  <0.0001        Heterosis nested within sex of calf  0.24        Brahman fraction nested within RFI group    0.0009  0.0073  0.0004  Heterosis nested within RFI group    0.0002  0.22  0.20  Mean chute score  0.39  0.42  0.11  0.33  Mean exit velocity  0.89  0.0012  0.34  0.31  1RFI = residual feed intake; DFI = mean daily feed intake; FCR = mean daily feed conversion ratio; PWG = postweaning BW gain. Empty cells indicate effects that were not modeled for the respective traits. View Large Among sex of calf effects, only heifers differed from steers (1.24 ± 0.36 kg of DM/d; P < 0.0006), indicating that heifers were less efficient than steers. Comparison with estimates from other studies was not possible because published RFI studies used either male calves (bulls, steers, or both; Arthur et al., 2001b; Nkrumah et al., 2004; Schenkel et al., 2004; Wang et al., 2006) or bulls and heifers, but they did not report sex differences for RFI (Archer et al., 1997; Arthur et al., 2001a). Differences between bulls and steers were nonsignificant, likely because of the small number of bulls in the data set. Nkrumah et al. (2004) found bulls to be more efficient than steers in a group of crossbred cattle composed of various B. taurus breeds including A, Charolais, Galloway, Hereford, and Holstein. Regression estimates for B breed effects nested within sex of calf were nonsignificant for bulls and steers but were negative (−1.29 ± 0.28 kg of DM/d; P < 0.0001) for heifers. Because B effects were expressed as deviations from A, this indicates that feed efficiency improved (decreased RFI) as the B fraction increased from A to B. The B regression estimate for bulls was also negative, albeit nonsignificant (P < 0.25). Perhaps a similar trend would have been found had more bulls been represented in this data set. Comparable studies were unavailable; however, Nkrumah et al. (2004) found nonsignificant differences among B. taurus crossbred cattle sired by A, Charolais, and University of Alberta crossbred bulls. Schenkel et al. (2004) reported RFI differences between Limousin, A, Simmental, Hereford, and Blonde d'Aquitaine and Charolais bulls ranging from 0.07 kg/d (as fed; Hereford − Charolais) to 0.55 kg/d (as fed; A − Charolais). Significance values were not reported in Schenkel et al. (2004). When deviated from A, these breed differences ranged from −0.38 kg/d (as fed; Simmental − A) to −1.08 kg/d (as fed; Blonde d'Aquitaine − A). Estimates of breed differences were smaller than the regression estimates here for both bulls (−2.14 ± 1.86 kg of DM/d; P < 0.25) and heifers (−1.29 ± 0.28 kg of DM/d; P < 0.0001). Although the data set here is small, these results may be an indication that breed differences for RFI may be larger between B and B. taurus breeds than among B. taurus breeds. Table 3 presents least squares means of RFI differences between sexes for the 6 breed groups in this study. Least squares means differences between bulls and heifers were negative within and across breed groups, indicating that bulls were more efficient than heifers, but were significant within only 3 breed groups: 3/4 A 1/4 B (−1.41 ± 0.51 kg of DM/d; P < 0.02), Brangus (−1.46 ± 0.39 kg of DM/d; P < 0.0008), and 1/4 A 3/4 B (−1.84 ± 0.69 kg of DM/d; P < 0.03), and overall (−1.34 ± 0.55 kg of DM/d; P < 0.05). Least squares means differences between bulls and steers within breed groups were mostly negative and nonsignificant, except for 1/4 A 3/4 B (−1.68 ± 0.69 kg of DM/d; P < 0.05). Contrarily, differences between heifers and steers were positive (heifers were less efficient than steers) for all breed groups, except for B, and were significant for A (1.24 ± 0.36 kg of DM/d; P < 0.002), 3/4 A 1/4 B (0.86 ± 0.17 kg of DM/d; P < 0.0001), and Brangus (0.69 ± 0.15 kg of DM/d; P < 0.0001). Heifers were also less efficient overall (0.61 ± 0.14 kg of DM/d; P < 0.0001). Differences between sex of calf least squares means for residual feed intake1 Table 3. Differences between sex of calf least squares means for residual feed intake1 Breed group  Bulls − heifers  Bulls − steers  Heifers − steers  Angus  −0.23 ± 1.52 (1.0)  1.01 ± 1.52 (1.0)  1.24 ± 0.36 (0.002)  3/4 A 1/4 B2  −1.41 ± 0.51 (0.02)  −0.55 ± 0.49 (0.81)  0.86 ± 0.17 (<0.0001)  Brangus  −1.46 ± 0.39 (0.0008)  −0.77 ± 0.37 (0.12)  0.69 ± 0.15 (<0.0001)  1/2 A 1/2 B  −3.00 ± 1.63 (0.34)  −2.12 ± 1.65 (0.61)  0.48 ± 0.38 (0.64)  1/4 A 3/4 B  −1.84 ± 0.69 (0.03)  −1.68 ± 0.69 (0.05)  0.16 ± 0.21 (1.0)  Brahman  −1.07 ± 2.38 (1.0)  −1.24 ± 2.38 (1.0)  −0.18 ± 0.35 (1.0)  All  −1.34 ± 0.55 (0.05)  −0.73 ± 0.53 (0.51)  0.61 ± 0.14 (0.0001)  Breed group  Bulls − heifers  Bulls − steers  Heifers − steers  Angus  −0.23 ± 1.52 (1.0)  1.01 ± 1.52 (1.0)  1.24 ± 0.36 (0.002)  3/4 A 1/4 B2  −1.41 ± 0.51 (0.02)  −0.55 ± 0.49 (0.81)  0.86 ± 0.17 (<0.0001)  Brangus  −1.46 ± 0.39 (0.0008)  −0.77 ± 0.37 (0.12)  0.69 ± 0.15 (<0.0001)  1/2 A 1/2 B  −3.00 ± 1.63 (0.34)  −2.12 ± 1.65 (0.61)  0.48 ± 0.38 (0.64)  1/4 A 3/4 B  −1.84 ± 0.69 (0.03)  −1.68 ± 0.69 (0.05)  0.16 ± 0.21 (1.0)  Brahman  −1.07 ± 2.38 (1.0)  −1.24 ± 2.38 (1.0)  −0.18 ± 0.35 (1.0)  All  −1.34 ± 0.55 (0.05)  −0.73 ± 0.53 (0.51)  0.61 ± 0.14 (0.0001)  1Least squares means differences in kilograms of DM per day; numbers in parentheses refer to levels of probability as determined by a Bonferroni t-test. 2A = Angus; B = Brahman. View Large Table 3. Differences between sex of calf least squares means for residual feed intake1 Breed group  Bulls − heifers  Bulls − steers  Heifers − steers  Angus  −0.23 ± 1.52 (1.0)  1.01 ± 1.52 (1.0)  1.24 ± 0.36 (0.002)  3/4 A 1/4 B2  −1.41 ± 0.51 (0.02)  −0.55 ± 0.49 (0.81)  0.86 ± 0.17 (<0.0001)  Brangus  −1.46 ± 0.39 (0.0008)  −0.77 ± 0.37 (0.12)  0.69 ± 0.15 (<0.0001)  1/2 A 1/2 B  −3.00 ± 1.63 (0.34)  −2.12 ± 1.65 (0.61)  0.48 ± 0.38 (0.64)  1/4 A 3/4 B  −1.84 ± 0.69 (0.03)  −1.68 ± 0.69 (0.05)  0.16 ± 0.21 (1.0)  Brahman  −1.07 ± 2.38 (1.0)  −1.24 ± 2.38 (1.0)  −0.18 ± 0.35 (1.0)  All  −1.34 ± 0.55 (0.05)  −0.73 ± 0.53 (0.51)  0.61 ± 0.14 (0.0001)  Breed group  Bulls − heifers  Bulls − steers  Heifers − steers  Angus  −0.23 ± 1.52 (1.0)  1.01 ± 1.52 (1.0)  1.24 ± 0.36 (0.002)  3/4 A 1/4 B2  −1.41 ± 0.51 (0.02)  −0.55 ± 0.49 (0.81)  0.86 ± 0.17 (<0.0001)  Brangus  −1.46 ± 0.39 (0.0008)  −0.77 ± 0.37 (0.12)  0.69 ± 0.15 (<0.0001)  1/2 A 1/2 B  −3.00 ± 1.63 (0.34)  −2.12 ± 1.65 (0.61)  0.48 ± 0.38 (0.64)  1/4 A 3/4 B  −1.84 ± 0.69 (0.03)  −1.68 ± 0.69 (0.05)  0.16 ± 0.21 (1.0)  Brahman  −1.07 ± 2.38 (1.0)  −1.24 ± 2.38 (1.0)  −0.18 ± 0.35 (1.0)  All  −1.34 ± 0.55 (0.05)  −0.73 ± 0.53 (0.51)  0.61 ± 0.14 (0.0001)  1Least squares means differences in kilograms of DM per day; numbers in parentheses refer to levels of probability as determined by a Bonferroni t-test. 2A = Angus; B = Brahman. View Large Mean DFI and Mean FCR Most effects in the model significantly affected DFI: contemporary group, sex of calf, age of calf, RFI group, B fraction nested within RFI group, heterosis nested within RFI group, and mean EV (P-values ranging from 0.01 to 0.0001; Table 2). Only contemporary group, sex of calf, age of calf, and B fraction nested within RFI group were significant for FCR (Table 2). Bulls consumed more feed per day (0.78 ± 0.27 kg of DM/d; P < 0.004) but required less feed per kilogram of BW gain (−1.14 ± 0.40 kg of DM·d−1/kg of BW gain·d−1; P < 0.004) than steers. On the other hand, heifers consumed slightly less feed per day (−0.51 ± 0.11 kg of DM/d; P < 0.0001) but required more feed per kilogram of BW gain (1.14 ± 0.17 kg of DM·d−1/kg of BW gain·d−1; P < 0.0001) than steers. Differences in DFI were greater between calves in the high and low RFI groups (3.96 ± 0.33 kg of DM/d; P < 0.0001) than between calves in the medium and low RFI groups (1.55 ± 0.28 kg of DM/d; P < 0.0001). However, only calves in the high RFI group required more feed per kilogram of BW gain (1.71 ± 0.48 kg of DM·d−1/kg of BW gain·d−1; P < 0.004). Although calves in the medium RFI group tended to consume more feed than those in the low RFI group, the difference was not significant. Regression estimates of DFI on the B fraction of calves were negative and significant for all RFI groups (high: −0.97 ± 0.38 kg of DM/d; P < 0.01; medium: −0.90 ± 0.33 kg of DM/d; P < 0.007; low: −0.99 ± 0.31 kg of DM/d; P < 0.001), indicating that DFI decreased as the B fraction of calves increased. Conversely, regression estimates of heterosis effects for DFI were positive for all RFI groups and were significant for the high (1.01 ± 0.35 kg of DM/d; P < 0.004) and medium (1.09 ± 0.32 kg of DM/d; P < 0.0009) RFI groups. Heterosis estimates suggest that nonadditive effects would increase feed consumption in crossbred animals. Regression estimates for FCR were positive for B breed effects nested within RFI groups and were significant only for B breed effects in the high (1.41 ± 0.52 kg of DM·d−1/kg of BW gain·d−1; P < 0.008) and medium (1.29 ± 0.47 kg of DM·d−1/kg of BW gain·d−1; P < 0.007) groups. These estimates suggest that calves with a greater B fraction required more feed per kilogram of BW gain (less efficient). Heterosis effects were positive and nearly significant for the high RFI group (0.92 ± 0.51 kg of DM·d−1/kg of BW gain·d−1; P < 0.07), suggesting that crossbred calves in the high RFI group tended to require more feed per kilogram of BW gain than calves (less efficient) in the low RFI group. Analyses failed to detect strong associations between temperament measures and feed efficiency traits. Mean CS regression estimates for DFI and FCR were nonsignificant. The regression of FCR on mean EV was nonsignificant. Only the regression estimate of DFI on mean EV was negative (−0.29 ± 0.09 kg of DM/d; P < 0.001), suggesting that calves that consumed more feed were slower out of the chute, perhaps indicating a more docile temperament (Curley et al., 2006). Pairwise differences between DFI least squares means for the high, medium, and low RFI groups (Table 4) were all positive (P < 0.0001), indicating that calves in the high RFI group consumed more feed than those in the medium RFI group, which in turn consumed more feed than calves in the low RFI group. Differences within breed groups were above 2.3 kg of DM/d between the high and medium RFI groups, 3.9 kg of DM/d and above between the high and low RFI groups, and above 1.5 kg of DM/d between the medium and low RFI groups. These values were similar to the overall least squares differences between the high and medium (2.35 ± 0.14 kg of DM/d; P < 0.0001), high and low (4.11 ± 0.17 kg of DM/d; P < 0.0001), and medium and low RFI groups (1.75 ± 0.13 kg of DM/d; P < 0.0001). Differences between residual feed intake (RFI) group least squares means for daily feed intake1 Table 4. Differences between residual feed intake (RFI) group least squares means for daily feed intake1 Breed group  High − medium RFI group  High − low RFI group  Medium − low RFI group  Angus  2.41 ± 0.30  3.96 ± 0.33  1.55 ± 0.28  3/4 A 1/4 B2  2.36 ± 0.17  4.14 ± 0.20  1.79 ± 0.15  Brangus  2.36 ± 0.15  4.14 ± 0.18  1.78 ± 0.13  1/2 A 1/2 B  2.31 ± 0.31  4.33 ± 0.35  2.03 ± 0.32  1/4 A 3/4 B  2.32 ± 0.19  4.15 ± 0.22  1.83 ± 0.19  Brahman  2.34 ± 0.33  3.98 ± 0.33  1.63 ± 0.31  All  2.35 ± 0.14  4.11 ± 0.17  1.75 ± 0.13  Breed group  High − medium RFI group  High − low RFI group  Medium − low RFI group  Angus  2.41 ± 0.30  3.96 ± 0.33  1.55 ± 0.28  3/4 A 1/4 B2  2.36 ± 0.17  4.14 ± 0.20  1.79 ± 0.15  Brangus  2.36 ± 0.15  4.14 ± 0.18  1.78 ± 0.13  1/2 A 1/2 B  2.31 ± 0.31  4.33 ± 0.35  2.03 ± 0.32  1/4 A 3/4 B  2.32 ± 0.19  4.15 ± 0.22  1.83 ± 0.19  Brahman  2.34 ± 0.33  3.98 ± 0.33  1.63 ± 0.31  All  2.35 ± 0.14  4.11 ± 0.17  1.75 ± 0.13  1Least squares means differences in kilograms of DM per day (Bonferroni t-test); P < 0.0001 for all differences. 2A = Angus; B = Brahman. View Large Table 4. Differences between residual feed intake (RFI) group least squares means for daily feed intake1 Breed group  High − medium RFI group  High − low RFI group  Medium − low RFI group  Angus  2.41 ± 0.30  3.96 ± 0.33  1.55 ± 0.28  3/4 A 1/4 B2  2.36 ± 0.17  4.14 ± 0.20  1.79 ± 0.15  Brangus  2.36 ± 0.15  4.14 ± 0.18  1.78 ± 0.13  1/2 A 1/2 B  2.31 ± 0.31  4.33 ± 0.35  2.03 ± 0.32  1/4 A 3/4 B  2.32 ± 0.19  4.15 ± 0.22  1.83 ± 0.19  Brahman  2.34 ± 0.33  3.98 ± 0.33  1.63 ± 0.31  All  2.35 ± 0.14  4.11 ± 0.17  1.75 ± 0.13  Breed group  High − medium RFI group  High − low RFI group  Medium − low RFI group  Angus  2.41 ± 0.30  3.96 ± 0.33  1.55 ± 0.28  3/4 A 1/4 B2  2.36 ± 0.17  4.14 ± 0.20  1.79 ± 0.15  Brangus  2.36 ± 0.15  4.14 ± 0.18  1.78 ± 0.13  1/2 A 1/2 B  2.31 ± 0.31  4.33 ± 0.35  2.03 ± 0.32  1/4 A 3/4 B  2.32 ± 0.19  4.15 ± 0.22  1.83 ± 0.19  Brahman  2.34 ± 0.33  3.98 ± 0.33  1.63 ± 0.31  All  2.35 ± 0.14  4.11 ± 0.17  1.75 ± 0.13  1Least squares means differences in kilograms of DM per day (Bonferroni t-test); P < 0.0001 for all differences. 2A = Angus; B = Brahman. View Large Similarly, least squares estimates of pairwise differences for FCR between the high, medium, and low RFI groups were positive for all breed groups, suggesting a decreasing level of feed efficiency between the 3 RFI groups. Least squares means differences for FCR between the high and medium RFI groups were less for purebred groups than for crossbred groups, and ranged from 1.24 ± 0.43 kg of DM·d−1/kg of BW gain·d−1 (P < 0.01) for A to 2.67 ± 0.45 kg of DM·d−1/kg of BW gain·d−1 (P < 0.0001) for 1/2 A 1/2 B. Differences between FCR least squares means for the high and low RFI calf groups ranged from 1.71 ± 0.48 kg of DM·d−1/kg of BW gain·d−1 (P < 0.001) for A to 3.14 ± 0.51 kg of DM·d−1/kg of BW gain·d−1 (P < 0.0001) for 1/2 A 1/2 B, and B had the second largest difference (2.78 ± 0.49; P < 0.0001). Differences between FCR least squares means of the medium and low RFI groups were smaller than differences between the high and medium, and high and low RFI groups, and were significant only for Brangus (0.60 ± 0.20 kg of DM·d−1/kg of BW gain·d−1; P < 0.007), 1/4 A 3/4 B (0.94 ± 0.27 kg of DM·d−1/kg of BW gain·d−1; P < 0.002), and B (1.40 ± 0.46 kg of DM·d−1/kg of BW gain·d−1; P < 0.008). The pattern of differences among RFI groups suggests that calves with greater B fractions needed more feed per kilogram of BW gain, and thus were less efficient in terms of FCR than A calves. PWG Factors affecting PWG (Table 2) were contemporary group (P < 0.0001), sex of calf (P < 0.0001), age of calf (P < 0.04), and B fraction within sex of calf (P < 0.0004). Age of dam was unimportant for PWG. As expected, bulls had greater PWG (20.00 ± 4.28 kg; P < 0.0001) than steers, and heifers had less PWG (−17.02 ± 1.79 kg; P < 0.0001) than steers. Estimates of RFI group differences for PWG were significant only for the high vs. low groups (11.25 ± 5.29 kg; P < 0.03), indicating that less efficient calves gained more BW during the 70-d trial. To achieve these BW gains, less efficient calves in the high RFI group consumed more feed than those in the more efficient low RFI group (1.71 ± 0.48 kg of DM/d; P < 0.0004). Brahman breed effects for PWG within RFI group were all negative, indicating that PWG decreased as the fraction of B in calves increased. Estimates of B regression effects decreased in absolute value from the high (−18.81 ± 5.98 kg; P < 0.005) to the medium (−16.18 ± 5.28 kg; P < 0.002) to the low (−14.82 ± 4.85 kg; P < 0.002) RFI group. Thus, there were smaller PWG differences between B and A among more efficient calves. Heterosis estimates within RFI group for PWG were all positive but were significant only for the medium RFI group (10.59 ± 5.88 kg; P < 0.04). Regression estimates of PWG on mean CS and on mean EV were nonsignificant, indicating that neither temperament measurement was associated with PWG in these cattle. Estimates of differences between least squares means for PWG of the high and medium RFI groups were all nonsignificant. Least squares means differences between the high and low RFI groups tended to decrease as the fraction of B in calves increased. However, differences between the high and low RFI groups were important only for 3/4 A 1/4 B (10.18 ± 3.14 kg; P < 0.0004), Brangus (9.97 ± 2.79 kg; P < 0.001), and 1/4 A 3/4 B (9.19 ± 3.55 kg; P < 0.03). These results suggest that differences in PWG between the high and low RFI groups may be more evident in calves with greater A than B fractions. A somewhat different pattern existed for the medium vs. low RFI groups. Least squares means differences between the medium and low RFI groups for PWG were smaller and nonsignificant for A and B (about 6 kg) but were larger and significant for the other 4 breed groups (10.26 ± 2.43 kg, P < 0.0001 for 3/4 A 1/4 B; 9.89 ± 2.11 kg, P < 0.0001 for Brangus; 13.17 ± 5.10 kg, P < 0.03 for 1/2 A 1/2 B; and 9.58 ± 3.55 kg, P < 0.004 for 1/4 A 3/4 B), perhaps suggesting a larger amount of variation for PWG among calves in crossbred groups. Considering all calves in the data set, PWG differences were significant between the high and low RFI groups (9.55 ± 2.65 kg, P < 0.0006) and between the medium and low RFI groups (9.24 ± 2.00 kg; P < 0.0001). Heritabilities, Genetic Correlations, and Phenotypic Correlations Table 5 shows REML estimates of heritabilities on the diagonal, genetic correlations above the diagonal, and phenotypic correlations below the diagonal. Heritability estimates from different pairwise analyses involving a given trait were similar. Thus, the estimate of heritability for RFI reported here is from the RFI-DFI analysis, and the heritabilities for DFI, FCR, and PWG are those from the RFI-DFI, RFI-FCR, and RFI-PWG analyses. These parameter estimates should be viewed with caution because of the small number of animals with records in the data set. The heritability estimate for RFI (0.19 ± 0.11) was less than that estimated by Arthur et al. (2001a) in Australia (0.39 ± 0.03; 1,180 A bulls and heifers; 15 mo of age), Arthur et al. (2001b) in France (0.39 ± 0.04; 792 Charolais bulls; 9 mo of age), and Schenkel et al. (2004) in Canada (0.38 ± 0.07; 0.07 = average SE of 10 pairwise analyses; 2,284 Charolais, Limousin, Simmental, Hereford, A, and Blonde d'Aquitaine bulls; 9 mo of age). Numbers of animals in these 3 studies were larger than here. The heritability estimate for RFI may increase as the numbers of records from the B-A population in Florida increase and additional genetic variation from the B-A population in Florida is accounted for. However, the estimate of heritability obtained here suggests that selection for RFI is feasible in Florida, although it would be advisable to substantially increase the number of calves evaluated for RFI at feed efficiency facilities each year. Estimates of heritabilities (diagonal) for, and genetic correlations (above diagonal) and phenotypic correlations (below diagonal) among residual feed intake (RFI), mean daily feed intake (DFI), mean feed conversion ratio (FCR), and postweaning BW gain (PWG) Table 5. Estimates of heritabilities (diagonal) for, and genetic correlations (above diagonal) and phenotypic correlations (below diagonal) among residual feed intake (RFI), mean daily feed intake (DFI), mean feed conversion ratio (FCR), and postweaning BW gain (PWG) Item  RFI  DFI  FCR  PWG  RFI  0.19 ± 0.11  0.73 ± 0.13  0.09 ± 0.38  0.58 ± 0.28  DFI  0.89 ± 0.01  0.42 ± 0.13  −0.05 ± 0.31  0.88 ± 0.12  FCR  0.55 ± 0.03  0.37 ± 0.04  0.24 ± 0.11  −0.50 ± 0.23  PWG  0.15 ± 0.04  0.41 ± 0.04  −0.57 ± 0.03  0.40 ± 0.13  Item  RFI  DFI  FCR  PWG  RFI  0.19 ± 0.11  0.73 ± 0.13  0.09 ± 0.38  0.58 ± 0.28  DFI  0.89 ± 0.01  0.42 ± 0.13  −0.05 ± 0.31  0.88 ± 0.12  FCR  0.55 ± 0.03  0.37 ± 0.04  0.24 ± 0.11  −0.50 ± 0.23  PWG  0.15 ± 0.04  0.41 ± 0.04  −0.57 ± 0.03  0.40 ± 0.13  The heritability estimate was 0.42 ± 0.13 for DFI and 0.24 ± 0.11 for FCR. The estimate of heritability for DFI was comparable with those obtained for A in Australia (0.39 ± 0.03; Arthur et al., 2001a), for Charolais in France (0.48 ± 0.04; Arthur et al., 2001b), and for 6 B. taurus breeds in Canada (0.44 ± 0.06; 0.06 = average SE of 10 pairwise analyses; Schenkel et al., 2004). The estimate of heritability for FCR was similar to the estimate obtained in Australia (0.29 ± 0.04; Arthur et al., 2001a) but was less than estimated from France (0.46 ± 0.04; Arthur et al., 2001b) and Canada (0.37 ± 0.06; Schenkel et al., 2004). The heritability estimate for PWG was moderate (0.40 ± 0.13). This estimate is within the range of parameters reported for cattle of B. indicus × B. taurus ancestry (Kriese et al., 1991; Davis, 1993), and was similar to estimates of heritability of postweaning ADG in feed efficiency studies (Arthur et al., 2001a,b; Schenkel et al., 2004). Estimates of genetic correlations (Table 5) were positive between RFI and DFI (0.73 ± 0.13) and between RFI and PWG (0.58 ± 0.28), indicating that less efficient calves based on RFI had greater BW gains during the 70-d postweaning trial. The genetic correlation between RFI and FCR was close to zero (0.09 ± 0.38). The genetic correlation estimate between RFI and DFI was similar to, and the one between RFI and FCR was much smaller than, those obtained in feed efficiency studies with B. taurus cattle (Arthur et al., 2001a,b; Schenkel et al., 2004). The positive genetic correlation between RFI and PWG disagreed with the low and negative (Arthur et al., 2001b) and the near zero (Schenkel et al., 2004) estimates of genetic correlations reported between RFI and ADG. The positive estimate of genetic correlation between RFI and PWG here suggests that selection for more efficient animals (i.e., animals with reduced RFI) will reduce PWG, which may be economically undesirable. However, this estimate may change substantially as additional RFI and PWG from the A-B multibreed population of Florida is collected in future years. Estimates of genetic correlations were near zero between DFI and FCR (−0.05 ± 0.31), were positive between DFI and PWG (0.88 ± 0.12), and were negative between FCR and PWG (−0.50 ± 0.23). These genetic correlation estimates indicate that calves that consumed more feed during the 70-d trial gained more BW and required approximately the same amount of feed per kilogram of BW gain. Estimates of genetic correlations between DFI and FCR were less than, and those between DFI and PWG and between FCR and PWG were in agreement with, values obtained with B. taurus breeds (Arthur et al., 2001b; Schenkel et al., 2004). Phenotypic correlations had the same sign as genotypic correlations, except for the correlation between DFI and FCR (positive phenotypic, negative genotypic; Table 5). Estimates agreed with those reported by Arthur et al. (2001a,b) and Schenkel et al. (2004), except for the greater value between RFI and PWG (0.15 ± 0.04). The Pearson partial correlation coefficient (correlation coefficient corrected for contemporary group, sex of calf, B fraction of calf, and heterosis of calf) between RFI and PWG was zero, suggesting that the low positive value of the phenotypic correlation obtained here may be a computational issue. Perhaps with a larger data set the phenotypic correlation computed would also approach zero. View Large Table 5. Estimates of heritabilities (diagonal) for, and genetic correlations (above diagonal) and phenotypic correlations (below diagonal) among residual feed intake (RFI), mean daily feed intake (DFI), mean feed conversion ratio (FCR), and postweaning BW gain (PWG) Item  RFI  DFI  FCR  PWG  RFI  0.19 ± 0.11  0.73 ± 0.13  0.09 ± 0.38  0.58 ± 0.28  DFI  0.89 ± 0.01  0.42 ± 0.13  −0.05 ± 0.31  0.88 ± 0.12  FCR  0.55 ± 0.03  0.37 ± 0.04  0.24 ± 0.11  −0.50 ± 0.23  PWG  0.15 ± 0.04  0.41 ± 0.04  −0.57 ± 0.03  0.40 ± 0.13  Item  RFI  DFI  FCR  PWG  RFI  0.19 ± 0.11  0.73 ± 0.13  0.09 ± 0.38  0.58 ± 0.28  DFI  0.89 ± 0.01  0.42 ± 0.13  −0.05 ± 0.31  0.88 ± 0.12  FCR  0.55 ± 0.03  0.37 ± 0.04  0.24 ± 0.11  −0.50 ± 0.23  PWG  0.15 ± 0.04  0.41 ± 0.04  −0.57 ± 0.03  0.40 ± 0.13  The heritability estimate was 0.42 ± 0.13 for DFI and 0.24 ± 0.11 for FCR. The estimate of heritability for DFI was comparable with those obtained for A in Australia (0.39 ± 0.03; Arthur et al., 2001a), for Charolais in France (0.48 ± 0.04; Arthur et al., 2001b), and for 6 B. taurus breeds in Canada (0.44 ± 0.06; 0.06 = average SE of 10 pairwise analyses; Schenkel et al., 2004). The estimate of heritability for FCR was similar to the estimate obtained in Australia (0.29 ± 0.04; Arthur et al., 2001a) but was less than estimated from France (0.46 ± 0.04; Arthur et al., 2001b) and Canada (0.37 ± 0.06; Schenkel et al., 2004). The heritability estimate for PWG was moderate (0.40 ± 0.13). This estimate is within the range of parameters reported for cattle of B. indicus × B. taurus ancestry (Kriese et al., 1991; Davis, 1993), and was similar to estimates of heritability of postweaning ADG in feed efficiency studies (Arthur et al., 2001a,b; Schenkel et al., 2004). Estimates of genetic correlations (Table 5) were positive between RFI and DFI (0.73 ± 0.13) and between RFI and PWG (0.58 ± 0.28), indicating that less efficient calves based on RFI had greater BW gains during the 70-d postweaning trial. The genetic correlation between RFI and FCR was close to zero (0.09 ± 0.38). The genetic correlation estimate between RFI and DFI was similar to, and the one between RFI and FCR was much smaller than, those obtained in feed efficiency studies with B. taurus cattle (Arthur et al., 2001a,b; Schenkel et al., 2004). The positive genetic correlation between RFI and PWG disagreed with the low and negative (Arthur et al., 2001b) and the near zero (Schenkel et al., 2004) estimates of genetic correlations reported between RFI and ADG. The positive estimate of genetic correlation between RFI and PWG here suggests that selection for more efficient animals (i.e., animals with reduced RFI) will reduce PWG, which may be economically undesirable. However, this estimate may change substantially as additional RFI and PWG from the A-B multibreed population of Florida is collected in future years. Estimates of genetic correlations were near zero between DFI and FCR (−0.05 ± 0.31), were positive between DFI and PWG (0.88 ± 0.12), and were negative between FCR and PWG (−0.50 ± 0.23). These genetic correlation estimates indicate that calves that consumed more feed during the 70-d trial gained more BW and required approximately the same amount of feed per kilogram of BW gain. Estimates of genetic correlations between DFI and FCR were less than, and those between DFI and PWG and between FCR and PWG were in agreement with, values obtained with B. taurus breeds (Arthur et al., 2001b; Schenkel et al., 2004). Phenotypic correlations had the same sign as genotypic correlations, except for the correlation between DFI and FCR (positive phenotypic, negative genotypic; Table 5). Estimates agreed with those reported by Arthur et al. (2001a,b) and Schenkel et al. (2004), except for the greater value between RFI and PWG (0.15 ± 0.04). The Pearson partial correlation coefficient (correlation coefficient corrected for contemporary group, sex of calf, B fraction of calf, and heterosis of calf) between RFI and PWG was zero, suggesting that the low positive value of the phenotypic correlation obtained here may be a computational issue. Perhaps with a larger data set the phenotypic correlation computed would also approach zero. View Large Conclusion This study found significant B breed effects nested within sex of calf for RFI. These B effects differed depending on the sex of the calf (negative and nonsignificant for bulls, negative and significant for heifers, and positive and nonsignificant for steers). Additional data will be required to clarify these sex differences, particularly the estimate obtained for steers. Nevertheless, results here indicated that B cattle tended to be more efficient than A cattle for postweaning growth under subtropical conditions in Florida. Brahman breed effects nested within RFI groups were negative for DFI, indicating that as the fraction B of calves increased, they tended to consume less feed. In contrast, B estimates for FCR were positive, suggesting that as the B fraction increased, the amount of feed needed per kilogram of BW gain increased, making calves with a greater B fraction less efficient than calves with greater A fractions. Last, B estimates for PWG were all negative, indicating a decrease in PWG as the fraction of B in calves increased. As indicated above, results for FCR contradicted results for RFI. This is another aspect that will need to be reevaluated as new data are collected in Florida. Heritabilities for all traits were reasonable. Similarly, estimates of genetic and phenotypic correlations seemed reasonable, considering the size and complexity of the multibreed data set used here. Estimates of genetic parameters here confirm that genetic variability existed and that selection for RFI would be feasible in the Florida A-B multibreed population. However, feed efficiency data need to continue to be collected in increasing numbers per year to compute accurate populational genetic parameters for multiple traits. Considering the high cost of obtaining feed efficiency information, it would seem reasonable to link feed efficiency facilities across the Southern region to improve the accuracy of genetic evaluation of animals, particularly sires, to increase the rate of progress attributable to selection. Heterosis was not a significant factor for RFI, FCR, or PWG, but it increased DFI. The P-values for RFI and FCR were less than 0.22 (Table 2); thus, their heterosis estimates may differ substantially in a different data set. Again, a larger data set would be needed to clarify the role of heterosis for feed efficiency traits in the Florida cattle population. Temperament was not an important factor for any of the traits, with the exception of EV for DFI, where animals that consumed more feed exited the chute more slowly. This may be an indication of better temperament, or it may simply suggest that animals that ate more were more sluggish coming out of the chute. Further, calves from the 3 locations were worked though the chute frequently both before and during their stay at IFEF. Perhaps calves became accustomed to this level of management, and consequently behavioral differences among calves decreased, and so did the potential impact of temperament on feed efficiency and BW gain traits. Temperament traits will continue to be collected and reevaluated with larger accumulated data sets in Florida. LITERATURE CITED Agresti, A. 2002. Categorical Data Analysis.  2nd ed. John Wiley and Sons, Hoboken, NJ. Google Scholar CrossRef Search ADS   Archer J. A. Arthur P. F. Herd R. M. Parnell P. F. Pitchford W. S. 1997. Optimum postweaning test for measurement of growth rate, feed intake, and feed efficiency in British breed cattle. J. Anim. Sci.  75: 2024– 2032. [PubMed] Google Scholar CrossRef Search ADS PubMed  Arthur P. F. Archer J. A. Johnston D. J. Herd R. M. Richardson E. C. Parnell P. F. 2001a. Genetic and phenotypic variance and covariance components for feed intake, feed efficiency, and other postweaning traits in Angus cattle. J. Anim. Sci.  79: 2805– 2811. [PubMed] Google Scholar CrossRef Search ADS   Arthur P. F. Renand G. Krauss D. 2001b. Genetic and phenotypic relationships among different measures of growth and feed efficiency in young Charolais bulls. Livest. Prod. Sci.  68: 131– 139. Google Scholar CrossRef Search ADS   Basarab J. A. Price M. A. Aalhus J. L. Okine E. K. Snelling W. M. Lyle K. L. 2003. Residual feed intake and body composition in young growing cattle. Can. J. Anim. Sci.  83: 189– 204. Google Scholar CrossRef Search ADS   Beef Improvement Federation 2002. Guidelines for Uniform Beef Improvement Programs.  8th ed. http://www.beefimprovement.org/library/06guidelines.pdf Accessed Jul. 12, 2008. Burrow H. M. Dillon R. D. 1997. Relationships between temperament and growth in a feedlot and commercial carcass traits of Bos indicus crossbreds. Aust. J. Exp. Agric.  37: 407– 411. Google Scholar CrossRef Search ADS   Crews D. H.Jr. 2005. Genetics of efficient feed utilization and national cattle evaluation: A review. Genet. Mol. Res.  4: 152– 165. [PubMed] Google Scholar PubMed  Curley K. O.Jr. Paschal J. C. Welsh T. H.Jr. Randel R. D. 2006. Technical note: Exit velocity as a measure of cattle temperament is repeatable and associated with serum concentration of cortisol in Brahman bulls. J. Anim. Sci.  84: 3100– 3103. [PubMed] Google Scholar CrossRef Search ADS PubMed  Davis G. P. 1993. Genetic parameters for tropical beef cattle in Northern Australia: A review. Aust. J. Agric. Res.  44: 179– 198. Google Scholar CrossRef Search ADS   Gilmour, A. R., B. R. Cullis, S. J. Welham, and R. Thompson 1999. ASREML Reference Manual.  NSW Agriculture, Orange, Australia. Kennedy B. W. van der Werf J. H. J. Meuwissen T. H. E. 1993. Genetic and statistical properties of residual feed intake. J. Anim. Sci.  71: 3239– 3250. [PubMed] Google Scholar CrossRef Search ADS PubMed  Koch R. M. Swiger L. A. Chambers D. Gregory K. E. 1963. Efficiency of feed use in beef cattle. J. Anim. Sci.  22: 486– 494. Google Scholar CrossRef Search ADS   Kriese L. A. Bertrand J. K. Benyshek L. L. 1991. Genetic and environmental growth trait parameter estimates for Brahman and Brahman-derivative cattle. J. Anim. Sci.  69: 2362– 2370. [PubMed] Google Scholar CrossRef Search ADS PubMed  Nkrumah J. D. Basarab J. A. Price M. A. Okine E. K. Ammoura A. Guercio S. Hansen C. Li C. Benkel B. Murdoch B. Moore S. S. 2004. Different measures of energetic efficiency and their phenotypic relationships with growth, feed intake, and ultrasound and carcass merit in hybrid cattle. J. Anim. Sci.  82: 2451– 2459. [PubMed] Google Scholar CrossRef Search ADS PubMed  Nkrumah J. D. Crews D. H.Jr. Basarab J. A. Price M. A. Okine E. K. Wang Z. Li C. Moore S. S. 2007. Genetic and phenotypic relationships of feeding behavior and temperament with performance, feed efficiency, ultrasound, and carcass merit of beef cattle. J. Anim. Sci.  85: 2382– 2390. [PubMed] Google Scholar CrossRef Search ADS PubMed  Schenkel F. S. Miller S. P. Wilton J. W. 2004. Genetic parameters and breed differences for feed efficiency, growth, and body composition traits of young beef bulls. Can. J. Anim. Sci.  84: 177– 185. Google Scholar CrossRef Search ADS   Stokes, M. E., C. S. Davis, and G. G. Koch 2000. Categorical Data Analysis Using the SAS System.  2nd ed. SAS Inst. Inc., Cary, NC. Voisinet B. D. Grandin T. O'Conner S. F. Tatum J. D. Deesing M. J. 1997. Bos indicus-cross feedlot cattle with excitable temperaments have tougher meat and a higher incidence of borderline dark cutters. Meat Sci.  46: 367– 377. Google Scholar CrossRef Search ADS PubMed  Wang Z. Nkrumah J. D. Li C. Basarab J. A. Goonewardene L. A. Okine E. K. Crews D. H.Jr. Moore S. S. 2006. Test duration for growth, feed intake, and feed efficiency in beef cattle using the GrowSafe system. J. Anim. Sci.  84: 2289– 2298. [PubMed] Google Scholar CrossRef Search ADS PubMed  American Society of Animal Science

Lancaster, P. A.;Carstens, G. E.;Jr., D. H. Crews,;Jr., T. H. Welsh,;Forbes, T. D. A.;Forrest, D. W.;Tedeschi, L. O.;Randel, R. D.;Rouquette, F. M.

doi: 10.2527/jas.2009-2041pmid: 19717782

ABSTRACT The objective of this study was to characterize residual feed intake (RFI) and to estimate phenotypic and genetic correlations with performance and ultrasound carcass traits in growing heifers. Four postweaning feed efficiency trials were conducted using 468 Brangus heifers. The complete Brangus pedigree file from Camp Cooley Ranch (Franklin, TX), which included 31,215 animals, was used to generate genetic parameter estimates. The heifer progeny from 223 dams were sired by 36 bulls, whereas the complete pedigree file contained 1,710 sires and 8,191 dams. Heifers were individually fed a roughage-based diet (ME = 1.98 Mcal/kg of DM) using Calan gate feeders for 70 d. Heifer BW was recorded weekly and ultrasound measures of 12th- to 13th-rib fat thickness (BF) and LM area (LMA) obtained at d 0 and 70. Residual feed intake (RFIp) was computed as actual minus predicted DMI, with predicted DMI determined by linear regression of DMI on mid-test BW0.75 (MBW) and ADG with trial, trial × MBW, and trial × ADG as random effects. Overall means for ADG, DMI, and RFI were 1.01 (SD = 0.15), 9.51 (SD = 1.02), and 0.00 (SD = 0.71) kg/d, respectively. Stepwise regression analysis revealed that inclusion of gain in BF and final LMA into the base model increased the R2 (0.578 vs. 0.534) and accounted for 9% of the variation in DMI not explained by MBW and ADG (RFIp). Residual feed intake and carcass-adjusted RFI (RFIc) were strongly correlated phenotypically and genetically with DMI and FCR, but not with ADG or MBW. Gain in BF was phenotypically correlated (P < 0.05) with RFIp (0.22), but not with FCR or RFIc; however, final BF was genetically correlated (P < 0.05) with RFIp (0.36) and RFIc (0.39). Gain in LMA was weakly phenotypically correlated with FCR, but not with RFIp or RFIc; however, gain in LMA was strongly genetically correlated with RFIp (0.55) and RFIc (0.77). The Spearman rank correlation between RFIp and RFIc was high (0.96). These results suggest that adjusting RFI for ultrasound carcass composition traits will facilitate selection phenotypically independent of growth, body size, and carcass composition; however, genetic relationships may still exist between RFI and carcass composition. INTRODUCTION Net returns of integrated beef cattle production systems are heavily dependent on the costs of feed inputs relative to the value of outputs. Given that feed inputs are the largest variable costs associated with producing beef, selection programs to improve profitability of beef production systems should focus on reducing feed inputs without compromising economically relevant output traits such as carcass merit. Considerable genetic variation for feed efficiency exists in beef cattle, yet limited genetic progress has been achieved due to the costs of labor and equipment to acquire feed intake data compared with output traits. Traditional ratio-based efficiency traits like feed conversion ratio (G:F) are confounded by variation due to maturity patterns and are strongly associated with production traits (e.g., growth). Recent studies have demonstrated that residual feed intake (RFI), which quantifies variation in feed intake beyond that expected to meet energy requirements for maintenance and production (Koch et al., 1963), is an appropriate trait to use in selection programs to improve feed efficiency due to its independence of growth and maturity patterns (Arthur et al., 2001a,c; Schenkel et al., 2004). Residual feed intake has been shown to be correlated with heat production, methane production, and digestibility, indicating that more efficient RFI phenotypes have less maintenance energy requirements and methane emissions, and improved diet digestibility (Herd and Arthur, 2009). However, several studies have also demonstrated that RFI is positively correlated with rate of fat deposition such that cattle with favorable RFI phenotypes have leaner carcasses (Nkrumah et al., 2004, 2007) and less empty body fat (Basarab et al., 2003), which may lead to unfavorable correlated responses in meat quality and heifer fertility traits (Arthur et al., 2005). Opportunities to improve the efficiency of beef production systems through selection by exploiting genetic variation in feed efficiency is dependent upon the existence of genetic variation, as well as knowledge of genetic associations with other economically relevant traits such as growth, carcass quality, and fertility. Few studies have examined the genetic associations between RFI and composition of growth in growing heifers. Thus, the objectives of this study were to characterize feed efficiency traits and estimate phenotypic and genetic correlations with performance and ultrasound carcass composition traits in growing Brangus heifers. MATERIALS AND METHODS All animal care and use procedures were in accordance with guidelines for use of Animals in Agricultural Teaching and Research and as approved by the Texas A&M University Institutional Animal Care and Use Committee. Animals and Management A total of 468 purebred Brangus heifers from Camp Cooley Ranch in Franklin, TX, were used in this study. Four postweaning trials were conducted in 4 consecutive years at the O. D. Butler, Jr. Animal Science Complex in College Station, TX, to measure performance and feed efficiency of heifers. Upon arrival, heifers were blocked by BW, randomly assigned to 20 pens (6 animals per pen), and adapted to the experimental diet for 24 d. Heifers were fed ad libitum twice daily, and individual feed intakes were measured weekly for 70 d (Archer et al., 1997) using Calan gate feeders (American Calan Inc., Northwood, NH). Heifers were weighed at 7-d intervals and real-time ultrasound measurements of 12th- to 13th-rib fat thickness (BF), LM area (LMA), and percent intramuscular fat (IM) obtained at the start and end of each trial by an Ultrasound Guidelines Council field certified technician using an Aloka 500-V instrument with a 17-cm, 3.5-MHz transducer (Corometrics Medical Systems Inc., Wallingford, CT). Images were collected and analyzed by the Beef Image Analysis Pro software (Designer Genes Inc., Harrison, AR) or sent to the National Centralized Ultrasound Processing laboratory (Ames, IA). Diet ingredient samples were collected weekly and composited by weight at the end of each trial. Moisture analysis was conducted by drying in a forced-air oven for 48 h at 105°C (AOAC, 1995), and chemical analysis was conducted by an independent laboratory (Cumberland Valley Analytical Services Inc., Hagerstown, MD). Metabolizable energy and NE concentrations of the trial diets were computed from the chemical analysis using the Cornell Net Carbohydrate and Protein System (version 5.0, Cornell University, Ithaca, NY). Ingredient and chemical composition of the diet fed to heifers are presented in Tables 1 and 2, respectively. Ingredient composition of diet fed to Brangus heifers in the 4 trials Table 1. Ingredient composition of diet fed to Brangus heifers in the 4 trials Item  Value, as-fed %  Chopped alfalfa  35.00  Pelleted alfalfa  15.00  Dry-rolled corn  20.95  Cottonseed hulls  21.50  Molasses  7.00  Salt  0.40  Vitamin E1  0.14  Trace mineral2  0.02  Item  Value, as-fed %  Chopped alfalfa  35.00  Pelleted alfalfa  15.00  Dry-rolled corn  20.95  Cottonseed hulls  21.50  Molasses  7.00  Salt  0.40  Vitamin E1  0.14  Trace mineral2  0.02  1Vitamin E contained 44,000 IU/kg of product. 2Trace mineral contained minimum 19.0% Zn, 7.0% Mn, 4.5% Cu, 4,000 mg/kg of Fe, 2,300 mg/kg of I, 1,000 mg/kg of Se, and 500 mg/kg of Co. View Large Table 1. Ingredient composition of diet fed to Brangus heifers in the 4 trials Item  Value, as-fed %  Chopped alfalfa  35.00  Pelleted alfalfa  15.00  Dry-rolled corn  20.95  Cottonseed hulls  21.50  Molasses  7.00  Salt  0.40  Vitamin E1  0.14  Trace mineral2  0.02  Item  Value, as-fed %  Chopped alfalfa  35.00  Pelleted alfalfa  15.00  Dry-rolled corn  20.95  Cottonseed hulls  21.50  Molasses  7.00  Salt  0.40  Vitamin E1  0.14  Trace mineral2  0.02  1Vitamin E contained 44,000 IU/kg of product. 2Trace mineral contained minimum 19.0% Zn, 7.0% Mn, 4.5% Cu, 4,000 mg/kg of Fe, 2,300 mg/kg of I, 1,000 mg/kg of Se, and 500 mg/kg of Co. View Large Chemical composition of diet fed to Brangus heifers in the 4 trials Table 2. Chemical composition of diet fed to Brangus heifers in the 4 trials Item  Trial 1  Trial 2  Trial 3  Trial 4  DM, %  87.49  89.36  88.05  86.81  CP, % of DM  12.57  13.16  12.48  12.58  NDF, % of DM  43.04  43.75  44.97  45.87  ME,1 Mcal/kg of DM  2.03  2.00  1.93  1.96  Item  Trial 1  Trial 2  Trial 3  Trial 4  DM, %  87.49  89.36  88.05  86.81  CP, % of DM  12.57  13.16  12.48  12.58  NDF, % of DM  43.04  43.75  44.97  45.87  ME,1 Mcal/kg of DM  2.03  2.00  1.93  1.96  1Metabolizable energy content computed using Cornell Net Carbohydrate and Protein System. View Large Table 2. Chemical composition of diet fed to Brangus heifers in the 4 trials Item  Trial 1  Trial 2  Trial 3  Trial 4  DM, %  87.49  89.36  88.05  86.81  CP, % of DM  12.57  13.16  12.48  12.58  NDF, % of DM  43.04  43.75  44.97  45.87  ME,1 Mcal/kg of DM  2.03  2.00  1.93  1.96  Item  Trial 1  Trial 2  Trial 3  Trial 4  DM, %  87.49  89.36  88.05  86.81  CP, % of DM  12.57  13.16  12.48  12.58  NDF, % of DM  43.04  43.75  44.97  45.87  ME,1 Mcal/kg of DM  2.03  2.00  1.93  1.96  1Metabolizable energy content computed using Cornell Net Carbohydrate and Protein System. View Large Computation of Traits Growth rates of individual heifers were modeled by linear regression of BW on day of trial using a regression model (PROC GLM, SAS Inst. Inc., Cary, NC), and the regression coefficients used to compute ADG, initial and final BW, and metabolic BW (MBW; mid-test BW0.75). Moisture analyses of diet ingredient samples were used to compute average daily DMI from feed intake data. Residual feed intake (RFIp) was computed as actual DMI minus expected DMI to meet growth and maintenance energy requirements (Koch et al., 1963). Expected DMI was derived from linear regression of DMI on MBW and ADG using a mixed model (PROC MIXED, SAS) with trial and trial × independent variable interactions as random effects and the variance component option used for the (co)variance matrix structure (St-Pierre, 2001). Feed conversion ratio (FCR) was computed as the ratio of daily DMI to ADG. To determine if individual-animal variation in carcass composition (ultrasound traits) affected the derivation of expected DMI, a 2-step approach was used (Arthur et al., 2003; Lancaster et al., 2009). First, stepwise regression analysis was performed (PROC REG, SAS) to determine the order in which ultrasound carcass composition traits should be included in the base model that included MBW and ADG. With the order derived from stepwise regression analysis, ultrasound traits were then sequentially added to the base model, and the resulting change in coefficient of determination used to determine their relative importance to account for additional variation in DMI. Four models were used to evaluate methods of combining data from multiple trials and the inclusion of ultrasound traits to compute RFI as described by Lancaster et al. (2009). In model 1, DMI was regressed on MBW, ADG, and ultrasound traits across the 4 trials without trial included as an independent variable to serve as a base model to compare the R2 of models 2, 3, and 4. Model 2 included trial as a fixed effect, whereas model 3 included trial and trial × independent variable interactions as random effects to account for potential variation in mean DMI and the differential relationships of DMI with MBW, ADG, and ultrasound traits due to trial. Model 4 was similar to model 3 in that it included trial and trial × independent variable interactions as random effects; however, to compute the R2, an adjusted DMI trait was computed using only the fixed effects and the residual from the complete model and then regressed on MBW and ADG. Model 4 was used to evaluate the amount of variation in DMI explained by the fixed effects of MBW and ADG (adjusted DMI) after accounting for the variation in DMI due to the random effects of trial and trial × independent variable interactions. The coefficient of multiple determination was compared between models by testing the extra sums of squares (Neter et al., 1996). Results from these 4 analyses were used to evaluate the inclusion of ultrasound measurements of carcass composition to calculate expected DMI. An additional RFI trait (RFIc) was computed from expected DMI adjusted for carcass composition, as well as MBW and ADG. The (co)variance structure of the G matrix for the above mixed models was evaluated according to Littell et al. (2006). The candidate covariance structures evaluated were variance components, compound symmetry, first-order autoregressive, Toeplitz, and unstructured. Based on Bayesian information criteria, the variance components structure proved to provide the best estimation of (co)variance structure. Statistical Analysis All performance, feed efficiency, and ultrasound carcass composition traits were adjusted to remove the random effect of trial using a random model (PROC MIXED, SAS). Phenotypic correlation coefficients among adjusted performance, feed efficiency, and ultrasound carcass composition traits were generated (PROC CORR, SAS). To further characterize RFI, heifers were classified into low, medium, and high RFI groups that were <0.5, ±0.5, and >0.5 SD, respectively, from the mean RFIp of 0.00 ± 0.71 kg/d, and a linear model (PROC MIXED, SAS) was used to examine the fixed effect of RFIp group on performance, feed efficiency, and ultrasound composition traits. Comparisons of least squares means between RFIp groups were performed using Tukey's post hoc test. Components of (co)variance that led to genetic parameters were estimated using an animal model and REML. The complete Brangus pedigree file from Camp Cooley included 31,215 animals. The 468 heifers with feeding records were sired by 36 bulls and were out of 223 dams, whereas the complete pedigree file contained 1,710 sires and 8,191 dams. Residual feed intake was defined in 2 ways for the genetic analyses. Base RFI (RFIp) was the residual term from the phenotypic linear regression of DMI on ADG and MBW, and RFIc was the residual term from the phenotypic linear regression of DMI on ADG, MBW, gain in subcutaneous rib fat, and gain in 12th-rib LMA. The multiple linear regression approach was adopted here to be consistent with other recent reports (e.g., Basarab et al., 2003; Schenkel et al., 2004). The adjustment of RFI for body composition using gains in ultrasound subcutaneous fat depth and LMA was nearly equivalent to using final (i.e., single measures at the end of trial) fat depth and LMA. Heritability estimates were obtained initially for all traits using univariate models that included the fixed effect of feeding contemporary group (trail × pen) and random direct animal effects. Genetic (co)variance components and correlation (Rg) estimates were obtained using bivariate models to reduce computational demand. All genetic models were fit using ASREML, which utilizes an average information algorithm and convergence was obtained within 10 rounds of iteration. Fixed and random effects in the bivariate models were identical to those used for the initial univariate models. Due to the relatively small number of feeding records, some bivariate models did not converge because estimates were near the boundary of parameter space (i.e., |Rg| ~1.0) or, in the case of models involving gain in subcutaneous fat depth, because gain in fat was nearly invariate. RESULTS AND DISCUSSION Descriptive statistics are presented in Table 3 for the 4 trials, and means and phenotypic variances for the trial-adjusted variables are presented in Table 4. The initial age of the heifers at the start of the trials averaged 231 ± 12 d across the 4 trials and ranged from 226 d in trial 1 to 236 d in trial 2. Numerically, heifers in trial 1 had heavier initial BW and less ADG and DMI compared with heifers in trials 2, 3, and 4, although the variation in ADG (CV = 13 to 17%), DMI (CV = 9 to 12%), and FCR (CV = 10 to 15%) were similar across the 4 trials. Performance, feed intake, and feed efficiency traits of the Brangus heifers were similar to previously published studies using growing heifers. Arthur et al. (1997) reported that ADG ranged from 1.01 ± 0.01 to 1.21 ± 0.01 kg/d, DMI ranged from 9.68 ± 0.08 to 10.96 ± 0.11 kg/d, and FCR ranged from 8.4 ± 0.1 to 10.7 ± 0.1 kg of DMI/kg of BW gain across 5 trials with British breeds of heifers fed a roughage-based diet. Arthur et al. (2003) reported means and SE for DMI and ADG of 9.2 ± 1.2 and 1.19 ± 0.19 kg/d, respectively, in Angus heifers fed a roughage-based diet. Summary statistics (±SE) of performance, feed efficiency, and ultrasound composition traits for Brangus heifers in each of the 4 trials Table 3. Summary statistics (±SE) of performance, feed efficiency, and ultrasound composition traits for Brangus heifers in each of the 4 trials Trait1  Trial 1  Trial 2  Trial 3  Trial 4  No. animals  114  115  119  120  Initial age, d  225.8 ± 9.1  236.0 ± 10.7  235.6 ± 14.6  228.3 ± 11.7  Initial BW, kg  285.1 ± 28.0  268.5 ± 23.8  267.8 ± 25.8  264.4 ± 26.9  Final BW, kg  345.8 ± 31.2  342.9 ± 28.9  337.7 ± 29.0  339.7 ± 30.0  ADG, kg/d  0.90 ± 0.15  1.06 ± 0.16  1.00 ± 0.13  1.08 ± 0.17  DMI, kg/d  9.10 ± 1.11  9.47 ± 1.04  9.92 ± 1.06  9.53 ± 0.88  FCR, feed/gain  10.26 ± 1.54  9.04 ± 1.31  10.02 ± 1.00  9.01 ± 1.21  RFIp, kg/d  0.00 ± 0.75  0.00 ± 0.68  0.00 ± 0.70  0.00 ± 0.66  Initial BF, cm  0.44 ± 0.15  0.38 ± 0.15  0.37 ± 0.09  0.43 ± 0.13  Initial LMA, cm2  57.58 ± 7.15  62.67 ± 7.27  52.76 ± 7.16  52.39 ± 7.64  Final BF, cm  0.65 ± 0.17  0.68 ± 0.18  0.66 ± 0.17  0.66 ± 0.19  Final LMA, cm2  70.96 ± 7.94  79.42 ± 7.40  62.51 ± 7.18  63.48 ± 8.55  Trait1  Trial 1  Trial 2  Trial 3  Trial 4  No. animals  114  115  119  120  Initial age, d  225.8 ± 9.1  236.0 ± 10.7  235.6 ± 14.6  228.3 ± 11.7  Initial BW, kg  285.1 ± 28.0  268.5 ± 23.8  267.8 ± 25.8  264.4 ± 26.9  Final BW, kg  345.8 ± 31.2  342.9 ± 28.9  337.7 ± 29.0  339.7 ± 30.0  ADG, kg/d  0.90 ± 0.15  1.06 ± 0.16  1.00 ± 0.13  1.08 ± 0.17  DMI, kg/d  9.10 ± 1.11  9.47 ± 1.04  9.92 ± 1.06  9.53 ± 0.88  FCR, feed/gain  10.26 ± 1.54  9.04 ± 1.31  10.02 ± 1.00  9.01 ± 1.21  RFIp, kg/d  0.00 ± 0.75  0.00 ± 0.68  0.00 ± 0.70  0.00 ± 0.66  Initial BF, cm  0.44 ± 0.15  0.38 ± 0.15  0.37 ± 0.09  0.43 ± 0.13  Initial LMA, cm2  57.58 ± 7.15  62.67 ± 7.27  52.76 ± 7.16  52.39 ± 7.64  Final BF, cm  0.65 ± 0.17  0.68 ± 0.18  0.66 ± 0.17  0.66 ± 0.19  Final LMA, cm2  70.96 ± 7.94  79.42 ± 7.40  62.51 ± 7.18  63.48 ± 8.55  1FCR = feed conversion ratio; RFIp = residual feed intake from base model; BF = 12th- to 13th-rib fat thickness; LMA = LM area. View Large Table 3. Summary statistics (±SE) of performance, feed efficiency, and ultrasound composition traits for Brangus heifers in each of the 4 trials Trait1  Trial 1  Trial 2  Trial 3  Trial 4  No. animals  114  115  119  120  Initial age, d  225.8 ± 9.1  236.0 ± 10.7  235.6 ± 14.6  228.3 ± 11.7  Initial BW, kg  285.1 ± 28.0  268.5 ± 23.8  267.8 ± 25.8  264.4 ± 26.9  Final BW, kg  345.8 ± 31.2  342.9 ± 28.9  337.7 ± 29.0  339.7 ± 30.0  ADG, kg/d  0.90 ± 0.15  1.06 ± 0.16  1.00 ± 0.13  1.08 ± 0.17  DMI, kg/d  9.10 ± 1.11  9.47 ± 1.04  9.92 ± 1.06  9.53 ± 0.88  FCR, feed/gain  10.26 ± 1.54  9.04 ± 1.31  10.02 ± 1.00  9.01 ± 1.21  RFIp, kg/d  0.00 ± 0.75  0.00 ± 0.68  0.00 ± 0.70  0.00 ± 0.66  Initial BF, cm  0.44 ± 0.15  0.38 ± 0.15  0.37 ± 0.09  0.43 ± 0.13  Initial LMA, cm2  57.58 ± 7.15  62.67 ± 7.27  52.76 ± 7.16  52.39 ± 7.64  Final BF, cm  0.65 ± 0.17  0.68 ± 0.18  0.66 ± 0.17  0.66 ± 0.19  Final LMA, cm2  70.96 ± 7.94  79.42 ± 7.40  62.51 ± 7.18  63.48 ± 8.55  Trait1  Trial 1  Trial 2  Trial 3  Trial 4  No. animals  114  115  119  120  Initial age, d  225.8 ± 9.1  236.0 ± 10.7  235.6 ± 14.6  228.3 ± 11.7  Initial BW, kg  285.1 ± 28.0  268.5 ± 23.8  267.8 ± 25.8  264.4 ± 26.9  Final BW, kg  345.8 ± 31.2  342.9 ± 28.9  337.7 ± 29.0  339.7 ± 30.0  ADG, kg/d  0.90 ± 0.15  1.06 ± 0.16  1.00 ± 0.13  1.08 ± 0.17  DMI, kg/d  9.10 ± 1.11  9.47 ± 1.04  9.92 ± 1.06  9.53 ± 0.88  FCR, feed/gain  10.26 ± 1.54  9.04 ± 1.31  10.02 ± 1.00  9.01 ± 1.21  RFIp, kg/d  0.00 ± 0.75  0.00 ± 0.68  0.00 ± 0.70  0.00 ± 0.66  Initial BF, cm  0.44 ± 0.15  0.38 ± 0.15  0.37 ± 0.09  0.43 ± 0.13  Initial LMA, cm2  57.58 ± 7.15  62.67 ± 7.27  52.76 ± 7.16  52.39 ± 7.64  Final BF, cm  0.65 ± 0.17  0.68 ± 0.18  0.66 ± 0.17  0.66 ± 0.19  Final LMA, cm2  70.96 ± 7.94  79.42 ± 7.40  62.51 ± 7.18  63.48 ± 8.55  1FCR = feed conversion ratio; RFIp = residual feed intake from base model; BF = 12th- to 13th-rib fat thickness; LMA = LM area. View Large Overall summary statistics, heritability, and phenotypic variance of trial-adjusted performance, feed efficiency, and ultrasound composition traits in Brangus heifers Table 4. Overall summary statistics, heritability, and phenotypic variance of trial-adjusted performance, feed efficiency, and ultrasound composition traits in Brangus heifers Trait1  n  Mean  SD  Heritability    Initial age, d  468  231.4  11.7  —  —  Initial BW, kg  468  271.4  26.1  0.07 ± 0.10  94.5  Final BW, kg  468  341.5  29.7  0.35 ± 0.15  267.9  ADG, kg/d  468  1.01  0.15  0.21 ± 0.12  0.02  DMI, kg/d  468  9.51  1.02  0.48 ± 0.14  0.92  FCR, feed/gain  468  9.55  1.27  0.29 ± 0.12  1.54  RFIp, kg/d  468  0.00  0.71  0.47 ± 0.13  0.54  RFIc, kg/d  453  0.00  0.68  0.42 ± 0.13  0.52  Final BF, cm  459  0.66  0.18  0.36 ± 0.15  0.03  Final LMA, cm2  461  69.09  7.76  0.11 ± 0.09  44.76  Final IM, %  467  6.76  0.69  0.67 ± 0.14  0.52  Trait1  n  Mean  SD  Heritability    Initial age, d  468  231.4  11.7  —  —  Initial BW, kg  468  271.4  26.1  0.07 ± 0.10  94.5  Final BW, kg  468  341.5  29.7  0.35 ± 0.15  267.9  ADG, kg/d  468  1.01  0.15  0.21 ± 0.12  0.02  DMI, kg/d  468  9.51  1.02  0.48 ± 0.14  0.92  FCR, feed/gain  468  9.55  1.27  0.29 ± 0.12  1.54  RFIp, kg/d  468  0.00  0.71  0.47 ± 0.13  0.54  RFIc, kg/d  453  0.00  0.68  0.42 ± 0.13  0.52  Final BF, cm  459  0.66  0.18  0.36 ± 0.15  0.03  Final LMA, cm2  461  69.09  7.76  0.11 ± 0.09  44.76  Final IM, %  467  6.76  0.69  0.67 ± 0.14  0.52  1FCR = feed conversion ratio; RFIp = residual feed intake from base model; RFIc = residual feed intake from composition-adjusted model; BF = 12th- to 13th-rib fat thickness; LMA = LM area; IM = intramuscular fat. View Large Table 4. Overall summary statistics, heritability, and phenotypic variance of trial-adjusted performance, feed efficiency, and ultrasound composition traits in Brangus heifers Trait1  n  Mean  SD  Heritability    Initial age, d  468  231.4  11.7  —  —  Initial BW, kg  468  271.4  26.1  0.07 ± 0.10  94.5  Final BW, kg  468  341.5  29.7  0.35 ± 0.15  267.9  ADG, kg/d  468  1.01  0.15  0.21 ± 0.12  0.02  DMI, kg/d  468  9.51  1.02  0.48 ± 0.14  0.92  FCR, feed/gain  468  9.55  1.27  0.29 ± 0.12  1.54  RFIp, kg/d  468  0.00  0.71  0.47 ± 0.13  0.54  RFIc, kg/d  453  0.00  0.68  0.42 ± 0.13  0.52  Final BF, cm  459  0.66  0.18  0.36 ± 0.15  0.03  Final LMA, cm2  461  69.09  7.76  0.11 ± 0.09  44.76  Final IM, %  467  6.76  0.69  0.67 ± 0.14  0.52  Trait1  n  Mean  SD  Heritability    Initial age, d  468  231.4  11.7  —  —  Initial BW, kg  468  271.4  26.1  0.07 ± 0.10  94.5  Final BW, kg  468  341.5  29.7  0.35 ± 0.15  267.9  ADG, kg/d  468  1.01  0.15  0.21 ± 0.12  0.02  DMI, kg/d  468  9.51  1.02  0.48 ± 0.14  0.92  FCR, feed/gain  468  9.55  1.27  0.29 ± 0.12  1.54  RFIp, kg/d  468  0.00  0.71  0.47 ± 0.13  0.54  RFIc, kg/d  453  0.00  0.68  0.42 ± 0.13  0.52  Final BF, cm  459  0.66  0.18  0.36 ± 0.15  0.03  Final LMA, cm2  461  69.09  7.76  0.11 ± 0.09  44.76  Final IM, %  467  6.76  0.69  0.67 ± 0.14  0.52  1FCR = feed conversion ratio; RFIp = residual feed intake from base model; RFIc = residual feed intake from composition-adjusted model; BF = 12th- to 13th-rib fat thickness; LMA = LM area; IM = intramuscular fat. View Large The phenotypic variances for growth and composition traits were similar to those previously reported in growing cattle (Arthur et al., 1997; Robinson and Oddy, 2004; Schenkel et al., 2004). Korver et al. (1991) reported phenotypic variances for DMI (0.42 kg/d) and RFI (0.27 kg/d) in dairy heifers fed ad libitum forage that were less than those for the Brangus heifers; however, their reported phenotypic variance for FCR (4.32 kg of DMI/kg of BW gain) was substantially greater than the FCR phenotypic variance found in our study. In contrast, Schenkel et al. (2004) and Hoque et al. (2006) reported that phenotypic variances for DMI (1.27 and 1.30 kg/d, respectively) and RFI (0.95 and 1.04 kg/d, respectively) were greater, and phenotypic variances for FCR (0.51 and 0.89 kg of DMI/kg of BW gain, respectively) less in British/European crossbred and Japanese Black bulls, respectively, compared with variance estimates for Brangus heifers in our study. In finishing steers, Robinson and Oddy (2004) reported phenotypic variances of 2.35 kg/d, 2.09 kg of DMI/kg of BW gain, and 0.76 kg/d for DMI, FCR, and RFI, respectively. The results from models used to compute expected DMI are presented in Table 5. The base regression model that excluded the effect of trial (model 1) explained 42.6% of the variation in DMI, which was less than expected based on the R2 of the 4 individual trials (R2 ranged from 0.433 to 0.572). Including trial as a fixed effect in model 2 increased (P < 0.05) the R2 to 0.546. Most previous studies have included trial as a fixed variable (Arthur et al., 2001a,c, 2003; Schenkel et al., 2004), although the effect of trial on variation in DMI was not reported. St-Pierre (2001) indicated that ignoring the effect of trial, as well as trial × independent variable interactions, when performing regression across multiple trials, would lead to biased estimates of regression coefficients. This would then result in a biased estimate of the residual variance (SD of RFI). In addition, trial is fundamentally a random variable because inference is to be made about future trials. In our study, including trial and trial × independent variable (MBW and ADG) interactions as random effects in model 3 resulted in an improved (P < 0.05) R2 compared with model 1 (0.555 vs. 0.426, respectively), but not compared with model 2 (0.555 vs. 0.546, respectively), suggesting that including random interactions of trial and independent variables had minimal impact in explaining variation in DMI due to MBW and ADG across the 4 trials. This likely reflects the relative similarity of cattle type and management and environmental conditions among the 4 trials. Percentage of variation explained (R2) by different models to predict DMI Table 5. Percentage of variation explained (R2) by different models to predict DMI   Model number1    1  2  3  4  Item  F1+e1  F2+e2  F1+R+e3  F1−R+e3  Regression2           Base model (BM; ADG and MBW)  0.426  0.546  0.555  0.534   BM + BFg  0.477  0.555  0.597  0.576   BM + fLMA  0.446  0.548  0.566  0.541   BM + BFg + fLMA  0.508  0.575  0.602  0.578    Model number1    1  2  3  4  Item  F1+e1  F2+e2  F1+R+e3  F1−R+e3  Regression2           Base model (BM; ADG and MBW)  0.426  0.546  0.555  0.534   BM + BFg  0.477  0.555  0.597  0.576   BM + fLMA  0.446  0.548  0.566  0.541   BM + BFg + fLMA  0.508  0.575  0.602  0.578  1F1 = fixed effects of indicated variables; F2 = fixed effects of indicated variables + fixed effect of trial; R = random effects of trial and trial by independent variable interactions; e1 = random trial and uncontrolled error; e2 = random uncontrolled error and error associated with fixed interactions of trial and independent variables; e3 = random uncontrolled error. 2MBW = mid-test BW0.75; BFg = gain in 12th- to 13th-rib fat thickness; fLMA = final LM area. View Large Table 5. Percentage of variation explained (R2) by different models to predict DMI   Model number1    1  2  3  4  Item  F1+e1  F2+e2  F1+R+e3  F1−R+e3  Regression2           Base model (BM; ADG and MBW)  0.426  0.546  0.555  0.534   BM + BFg  0.477  0.555  0.597  0.576   BM + fLMA  0.446  0.548  0.566  0.541   BM + BFg + fLMA  0.508  0.575  0.602  0.578    Model number1    1  2  3  4  Item  F1+e1  F2+e2  F1+R+e3  F1−R+e3  Regression2           Base model (BM; ADG and MBW)  0.426  0.546  0.555  0.534   BM + BFg  0.477  0.555  0.597  0.576   BM + fLMA  0.446  0.548  0.566  0.541   BM + BFg + fLMA  0.508  0.575  0.602  0.578  1F1 = fixed effects of indicated variables; F2 = fixed effects of indicated variables + fixed effect of trial; R = random effects of trial and trial by independent variable interactions; e1 = random trial and uncontrolled error; e2 = random uncontrolled error and error associated with fixed interactions of trial and independent variables; e3 = random uncontrolled error. 2MBW = mid-test BW0.75; BFg = gain in 12th- to 13th-rib fat thickness; fLMA = final LM area. View Large The use of model 4 to remove the random effects of trial and trial by independent variable interactions on DMI resulted in an R2 of 0.534, which is more similar to the average (R2 = 0.529) of the individual trials than the R2 of models 2 or 3. Thus, the R2 from model 4 provides a more realistic estimate of the variation in DMI explained by the fixed independent variables MBW and ADG alone. Based on these analyses, we conclude that the inclusion of trial and trial × independent variable interactions as random effects in the regression model (model 3) to estimate expected DMI is the most appropriate method to compute RFI of individual animals when data across multiple trials are combined. Moreover, adjusting DMI to remove the random effects of trial and trial × independent variable interactions (model 4) will provide a better estimate of the variation in DMI explained by the independent variables of interest, MBW and ADG. Results from stepwise regression analysis revealed that the order of inclusion of ultrasound carcass composition traits in the regression model for expected DMI was gain in BF followed by final LMA. Final BF and gain in LMA were not selected as significant variables by the regression procedure. Gain in BF explained the largest amount of additional variation (P < 0.05; 4 percentage units) in DMI based on model 4. Arthur et al. (2003) and Basarab et al. (2003) reported that inclusion of carcass fat traits increased the R2 of linear regression models for expected DMI by 2 to 4 percentage units. Final LMA was a significant source of variation in the model for expected DMI, although the increase in R2 (0.20) due to inclusion of this trait was minimal. The final regression model used to compute carcass-adjusted expected DMI included gain in BF, final LMA, MBW, and ADG as fixed effects, and trial and trial × independent variable interactions as random effects. In this study, the R2 of the base model (0.534) used to compute RFI was less than in previous studies (Arthur et al., 2003; Basarab et al., 2003; Schenkel et al., 2004; Baker et al., 2006; Lancaster et al., 2009) that reported R2 ranging from 0.68 to 0.82. Most studies that have characterized RFI in beef cattle have used medium (2.4 Mcal/kg of DM) to high (2.80 Mcal/kg of DM) energy diets, whereas a decreased energy diet (1.98 Mcal/kg of DM) was fed to growing heifers in this study. Crews et al. (2003b) found that the phenotypic variance for RFI was larger in steers fed a low-energy barley silage-based diet during the growing phase than for RFI measured in the same steers fed a high-energy barley grain-based diet during the finishing phase. However, when evaluating the literature, studies using low-energy roughage-based diets (Arthur et al., 2001a,b; Brown, 2005) report similar SD of RFI (range of 0.74 to 0.84 vs. range of 0.66 to 0.88) compared with studies using high-energy grain-based diets (Basarab et al., 2003; Nkrumah et al., 2004,2007). In addition, breed type, sex, or both may have influenced the low R2 of the base regression of the Brangus heifers. Lancaster et al. (2009) reported an R2 of 0.76 for base model RFIp in Angus bulls fed a high-energy corn silage-based diet compared with an R2 of 0.22 for base model RFIp in Brahman heifers fed the same diet in the same facility reported by Ribeiro et al. (2006). However, Arthur et al. (2003) reported that the R2 for the base model RFI was 0.68 for Angus heifers compared with 0.70 for Angus bulls, suggesting that sex may have little influence on the amount of variation in DMI explained by MBW and ADG. Despite the reduced R2 of the base model RFIp in this study, inclusion of gain in BF into the carcass-adjusted model explained a similar amount of additional variation in DMI as in previous studies (Arthur et al., 2003; Basarab et al., 2003; Lancaster et al., 2009). The inclusion of gain in BF and final LMA explained 9% of the variation in DMI not accounted for by MBW and ADG. Arthur et al. (2003) reported that inclusion of fat depth explained 5.7 and 12.4% of the variation in DMI of Angus heifers and bulls, respectively. Similarly, Basarab et al. (2003) reported that inclusion of gain in BF and gain in IM explained 9 to 12.5% of the variation in DMI not accounted for by MBW and ADG in finishing steers. In 2 review papers, Herd et al. (2004) and Richardson and Herd (2004) reported that differences in body composition, particularly fat deposition, accounted for 5% of the variation in feed intake between calves from low and high RFI selection lines after a single generation of selection. Heritability estimates for performance, feed intake, efficiency, and carcass composition traits are presented in Table 4. The heritability estimates for growth traits in this study were small to moderate and similar to estimates reported by Arthur et al. (2001a,b,c), Schenkel et al. (2004), and Nkrumah et al. (2007) in British and Continental European beef cattle. The heritability estimate for ADG (0.21) was slightly less than that reported in a review by Koots et al. (1994a; 0.31). The heritability estimate for final ultrasound BF was consistent with those reported by Koots et al. (1994a), Arthur et al. (2001a), and Schenkel et al. (2004), although the heritability for final LMA of the Brangus heifers used in this study (0.11) was less than the estimates from Koots et al. (1994a), Arthur et al. (1997,2001a), Schenkel et al. (2004), and Nkrumah et al. (2007), which ranged from 0.27 to 0.43. The heritability estimate for final ultrasound IM was similar to that found in crossbred steers (Nkrumah et al., 2007), but greater than estimates of 0.30 in Angus and Hereford steers (Kemp et al., 2002), 0.52 in Simmental heifers (Crews et al., 2003a), and 0.40 in Angus heifers (MacNeil and Northcutt, 2008). The heritability estimates for DMI (0.48), FCR (0.29), and RFIp (0.47) from the current study were moderate and consistent with recent estimates reported for British and Continental European beef cattle (Arthur et al., 2001a,b; Schenkel et al., 2004; Nkrumah et al., 2007), which ranged from 0.39 to 0.54 for DMI, 0.29 to 0.46 for FCR, and 0.21 to 0.39 for RFIp. In a review of published estimates, Koots et al. (1994a) reported that weighted heritability estimates for DMI and FCR were 0.34 and 0.32, respectively. Slightly decreased heritability estimates for FCR and RFIp were found for Japanese Black bulls (Hoque et al., 2006), Hereford bulls (Herd and Bishop, 2000), and Holstein dairy heifers (Korver et al., 1991). Few studies have reported heritability estimates for RFI corrected for variation in carcass composition. In this study, the heritability estimate for RFI adjusted for variation in ultrasound carcass traits (RFIc) was similar to that for RFIp. Likewise, Schenkel et al. (2004) found that heritability estimates were similar for RFI with and without correction for variation in carcass composition. In Brown Swiss- and Holstein-sired bull calves, Jensen et al. (1992) found that adjusting RFI for variation in dissectible carcass fat increased the heritability estimate from 0.28 to 0.36. Phenotypic and genetic correlations among growth and feed efficiency traits are presented in Table 6. Dry matter intake was strongly correlated, phenotypically and genetically, with ADG (0.57 and 0.56) and MBW (0.61 and 0.77), which is consistent with previous results from Arthur et al. (2001a,b) and Schenkel et al. (2004) for growing bulls, and Korver et al. (1991) for Holstein dairy heifers. In agreement with previous studies, a strong negative phenotypic correlation between FCR and ADG was found, and the genetic correlations with FCR were moderately negative for ADG (−0.36) and MBW (−0.29). The phenotypic correlation between FCR and DMI (0.15) was less, but the corresponding genetic correlation (0.60) was in agreement with that found in Charolais bulls (Arthur et al., 2001b). Smaller genetic correlations between FCR and DMI were reported by Arthur et al. (2001a), Schenkel et al. (2004), Hoque et al. (2006), and Nkrumah et al. (2007), which ranged from 0.28 to 0.39. A meta-review of earlier published data (Koots et al., 1994b) found that weighted mean genetic correlations among FCR and ADG, yearling BW, and feed intake were −0.67, −0.60, and 0.71, respectively. These results indicate that correlated genetic responses to selection for improved FCR would be increased for growth rate and yearling BW, and reduced for feed intake, thereby limiting the utility of this trait to improve production efficiency of integrated beef production systems due to unfavorable responses in mature BW and maintenance energy requirements of the breeding herd (Herd and Bishop, 2000). Additionally, selection using ratio-based traits can result in divergent and unpredictable genetic responses of the component traits if the genetic variances of the component traits are different (Gunsett, 1984). In fact, Bishop et al. (1991) found that Angus progeny from single-generation selection for reduced FCR tended to have more desirable FCR due to greater ADG, but similar feed intakes compared with progeny selected for high FCR. Phenotypic (below diagonal) and genetic (±SE) correlations (above diagonal) among performance and feed efficiency traits in Brangus heifers Table 6. Phenotypic (below diagonal) and genetic (±SE) correlations (above diagonal) among performance and feed efficiency traits in Brangus heifers Trait1  MBW  ADG  DMI  FCR  RFIp  RFIc  MBW    0.99 ± 0.08  0.77 ± 0.15  −0.29 ± 0.32  0.33 ± 0.29  0.32 ± 0.30  ADG  0.35*    0.56 ± 0.22  −0.36 ± 0.31  0.04 ± 0.32  0.01 ± 0.34  DMI  0.61*  0.57*    0.60 ± 0.26  0.85 ± 0.08  0.86 ± 0.09  FCR  0.10*  −0.71*  0.15*    0.93 ± 0.09  0.94 ± 0.09  RFIp  0.00  0.00  0.70*  0.59*    NC2  RFIc  0.00  0.00  0.67*  0.56*  0.97*    Trait1  MBW  ADG  DMI  FCR  RFIp  RFIc  MBW    0.99 ± 0.08  0.77 ± 0.15  −0.29 ± 0.32  0.33 ± 0.29  0.32 ± 0.30  ADG  0.35*    0.56 ± 0.22  −0.36 ± 0.31  0.04 ± 0.32  0.01 ± 0.34  DMI  0.61*  0.57*    0.60 ± 0.26  0.85 ± 0.08  0.86 ± 0.09  FCR  0.10*  −0.71*  0.15*    0.93 ± 0.09  0.94 ± 0.09  RFIp  0.00  0.00  0.70*  0.59*    NC2  RFIc  0.00  0.00  0.67*  0.56*  0.97*    1MBW = mid-test BW0.75; FCR = feed conversion ratio; RFIp = residual feed intake from base model; RFIc = residual feed intake from composition-adjusted model. 2NC indicates that the model would not converge. *Correlations are different from zero at P < 0.05. View Large Table 6. Phenotypic (below diagonal) and genetic (±SE) correlations (above diagonal) among performance and feed efficiency traits in Brangus heifers Trait1  MBW  ADG  DMI  FCR  RFIp  RFIc  MBW    0.99 ± 0.08  0.77 ± 0.15  −0.29 ± 0.32  0.33 ± 0.29  0.32 ± 0.30  ADG  0.35*    0.56 ± 0.22  −0.36 ± 0.31  0.04 ± 0.32  0.01 ± 0.34  DMI  0.61*  0.57*    0.60 ± 0.26  0.85 ± 0.08  0.86 ± 0.09  FCR  0.10*  −0.71*  0.15*    0.93 ± 0.09  0.94 ± 0.09  RFIp  0.00  0.00  0.70*  0.59*    NC2  RFIc  0.00  0.00  0.67*  0.56*  0.97*    Trait1  MBW  ADG  DMI  FCR  RFIp  RFIc  MBW    0.99 ± 0.08  0.77 ± 0.15  −0.29 ± 0.32  0.33 ± 0.29  0.32 ± 0.30  ADG  0.35*    0.56 ± 0.22  −0.36 ± 0.31  0.04 ± 0.32  0.01 ± 0.34  DMI  0.61*  0.57*    0.60 ± 0.26  0.85 ± 0.08  0.86 ± 0.09  FCR  0.10*  −0.71*  0.15*    0.93 ± 0.09  0.94 ± 0.09  RFIp  0.00  0.00  0.70*  0.59*    NC2  RFIc  0.00  0.00  0.67*  0.56*  0.97*    1MBW = mid-test BW0.75; FCR = feed conversion ratio; RFIp = residual feed intake from base model; RFIc = residual feed intake from composition-adjusted model. 2NC indicates that the model would not converge. *Correlations are different from zero at P < 0.05. View Large Phenotypically, both RFI traits were strongly correlated with DMI, but not with ADG or MBW, such that heifers with low phenotypic RFIp had similar ADG and final BW, but consumed 15% less (P < 0.01) DMI than heifers with high RFIp (Table 7). This was expected because the use of linear regression to compute expected DMI for RFI forces this trait to be phenotypically independent of the component traits. However, Kennedy et al. (1993) demonstrated that computing RFI by regression does not guarantee genetic independence of RFI from its component traits. In our study, RFIp was genetically independent of ADG, but was positively correlated with MBW (0.33), suggesting that selection for favorable RFI may tend to reduce body size, although these genetic correlations should be interpreted with caution given their large SE, which likely reflects the relatively small size of the data set. Arthur et al. (2001b) and Herd and Bishop (2000) reported that RFIp was genetically independent of ADG, but positively correlated with MBW (0.32 and 0.22, respectively), whereas Schenkel et al. (2004) found that RFIp was genetically independent of ADG, but negatively correlated with MBW (−0.17). Arthur et al. (2005) found that Angus females generated from divergent selection for RFI over 1 to 2.5 generations had similar cow BW and calf weaning BW, demonstrating that genetically correlated responses of growth traits to selection for RFI in this study were minimal. The computation of RFI from genetic regression has been suggested as an alternative to RFIp for use in selection programs (Crews, 2005) to ensure that RFI is genetically independent of component traits. Nkrumah et al. (2007) demonstrated that RFI computed by genetic regression was independent of both ADG and BW, and that it was strongly correlated phenotypically (0.97) and genetically (0.92) with RFI computed with phenotypic regression. Performance, feed efficiency, and ultrasound composition traits of heifers with low (<0.5 SD), medium (±0.5 SD), and high (>0.5 SD) residual feed intake (RFI; from base model) Table 7. Performance, feed efficiency, and ultrasound composition traits of heifers with low (<0.5 SD), medium (±0.5 SD), and high (>0.5 SD) residual feed intake (RFI; from base model) Item  Low RFI  Medium RFI  High RFI  SE  P-value  Trait1             No. of animals  150  176  142  —  —   Initial BW, kg  272.4  271.0  271.0  3.1  0.87   Final BW, kg  342.8  340.7  341.2  3.5  0.81   ADG, kg/d  1.01  1.01  1.01  0.02  0.86   DMI, kg/d  8.76a  9.48b  10.34c  0.09  0.01   FCR, feed/gain  8.75a  9.52b  10.42c  0.13  0.01   RFIp, kg/d  −0.78a  −0.01b  0.83c  0.04  0.01   RFIc, kg/d  −0.73a  −0.01b  0.76c  0.04  0.01  Initial composition trait             12th- to 13th-rib fat thickness, cm  0.42  0.40  0.39  0.02  0.08   LM area, cm2  57.53a  56.30ab  55.15b  0.85  0.02   Intramuscular fat, %  2.88  2.97  3.03  0.07  0.09  Final composition trait             12th- to 13th-rib fat thickness, cm  0.65  0.66  0.68  0.02  0.45   LM area, cm2  69.77  68.88  68.65  0.92  0.43   Intramuscular fat, %  3.38  3.49  3.50  0.08  0.15  Gain in composition trait             12th- to 13th-rib fat thickness, cm  0.23a  0.25a  0.29b  0.01  0.01   LM area, cm2  12.19  12.55  13.47  0.78  0.24   Intramuscular fat, %  0.50  0.53  0.49  0.06  0.77  Item  Low RFI  Medium RFI  High RFI  SE  P-value  Trait1             No. of animals  150  176  142  —  —   Initial BW, kg  272.4  271.0  271.0  3.1  0.87   Final BW, kg  342.8  340.7  341.2  3.5  0.81   ADG, kg/d  1.01  1.01  1.01  0.02  0.86   DMI, kg/d  8.76a  9.48b  10.34c  0.09  0.01   FCR, feed/gain  8.75a  9.52b  10.42c  0.13  0.01   RFIp, kg/d  −0.78a  −0.01b  0.83c  0.04  0.01   RFIc, kg/d  −0.73a  −0.01b  0.76c  0.04  0.01  Initial composition trait             12th- to 13th-rib fat thickness, cm  0.42  0.40  0.39  0.02  0.08   LM area, cm2  57.53a  56.30ab  55.15b  0.85  0.02   Intramuscular fat, %  2.88  2.97  3.03  0.07  0.09  Final composition trait             12th- to 13th-rib fat thickness, cm  0.65  0.66  0.68  0.02  0.45   LM area, cm2  69.77  68.88  68.65  0.92  0.43   Intramuscular fat, %  3.38  3.49  3.50  0.08  0.15  Gain in composition trait             12th- to 13th-rib fat thickness, cm  0.23a  0.25a  0.29b  0.01  0.01   LM area, cm2  12.19  12.55  13.47  0.78  0.24   Intramuscular fat, %  0.50  0.53  0.49  0.06  0.77  a–cMeans in the same row with unlike superscripts are different at P < 0.05. 1FCR = feed conversion ratio; RFIp = residual feed intake from base model; RFIc = residual feed intake from composition-adjusted model. View Large Table 7. Performance, feed efficiency, and ultrasound composition traits of heifers with low (<0.5 SD), medium (±0.5 SD), and high (>0.5 SD) residual feed intake (RFI; from base model) Item  Low RFI  Medium RFI  High RFI  SE  P-value  Trait1             No. of animals  150  176  142  —  —   Initial BW, kg  272.4  271.0  271.0  3.1  0.87   Final BW, kg  342.8  340.7  341.2  3.5  0.81   ADG, kg/d  1.01  1.01  1.01  0.02  0.86   DMI, kg/d  8.76a  9.48b  10.34c  0.09  0.01   FCR, feed/gain  8.75a  9.52b  10.42c  0.13  0.01   RFIp, kg/d  −0.78a  −0.01b  0.83c  0.04  0.01   RFIc, kg/d  −0.73a  −0.01b  0.76c  0.04  0.01  Initial composition trait             12th- to 13th-rib fat thickness, cm  0.42  0.40  0.39  0.02  0.08   LM area, cm2  57.53a  56.30ab  55.15b  0.85  0.02   Intramuscular fat, %  2.88  2.97  3.03  0.07  0.09  Final composition trait             12th- to 13th-rib fat thickness, cm  0.65  0.66  0.68  0.02  0.45   LM area, cm2  69.77  68.88  68.65  0.92  0.43   Intramuscular fat, %  3.38  3.49  3.50  0.08  0.15  Gain in composition trait             12th- to 13th-rib fat thickness, cm  0.23a  0.25a  0.29b  0.01  0.01   LM area, cm2  12.19  12.55  13.47  0.78  0.24   Intramuscular fat, %  0.50  0.53  0.49  0.06  0.77  Item  Low RFI  Medium RFI  High RFI  SE  P-value  Trait1             No. of animals  150  176  142  —  —   Initial BW, kg  272.4  271.0  271.0  3.1  0.87   Final BW, kg  342.8  340.7  341.2  3.5  0.81   ADG, kg/d  1.01  1.01  1.01  0.02  0.86   DMI, kg/d  8.76a  9.48b  10.34c  0.09  0.01   FCR, feed/gain  8.75a  9.52b  10.42c  0.13  0.01   RFIp, kg/d  −0.78a  −0.01b  0.83c  0.04  0.01   RFIc, kg/d  −0.73a  −0.01b  0.76c  0.04  0.01  Initial composition trait             12th- to 13th-rib fat thickness, cm  0.42  0.40  0.39  0.02  0.08   LM area, cm2  57.53a  56.30ab  55.15b  0.85  0.02   Intramuscular fat, %  2.88  2.97  3.03  0.07  0.09  Final composition trait             12th- to 13th-rib fat thickness, cm  0.65  0.66  0.68  0.02  0.45   LM area, cm2  69.77  68.88  68.65  0.92  0.43   Intramuscular fat, %  3.38  3.49  3.50  0.08  0.15  Gain in composition trait             12th- to 13th-rib fat thickness, cm  0.23a  0.25a  0.29b  0.01  0.01   LM area, cm2  12.19  12.55  13.47  0.78  0.24   Intramuscular fat, %  0.50  0.53  0.49  0.06  0.77  a–cMeans in the same row with unlike superscripts are different at P < 0.05. 1FCR = feed conversion ratio; RFIp = residual feed intake from base model; RFIc = residual feed intake from composition-adjusted model. View Large A greater phenotypic correlation was found between FCR and RFIp (0.59), such that heifers with less RFIp had 16% less (P < 0.05) FCR compared with heifers with greater RFIp (Table 7). Similarly, Arthur et al. (2001a,b), Schenkel et al. (2004), Nkrumah et al. (2004, 2007), and Hoque et al. (2006) reported increased phenotypic (range of 0.52 to 0.76) correlations of RFIp with FCR in growing bulls and steers. The corresponding genetic correlations between FCR and RFIp from these studies ranged from 0.62 to 0.85, which are less than found in our study (0.94). These data indicate that selection for improved RFI will result in an improvement in gross feed efficiency (FCR). Phenotypic and genetic correlations between feed efficiency and ultrasound traits are shown in Table 8. Phenotypic correlations between ultrasound traits and RFIp were weak for final BF and gain in BF (0.12 and 0.22, respectively). Heifers with low RFIp had similar final BF, but gained more BF (P < 0.01) during the trial than heifers with increased RFIp as initial BF tended to be greater (P = 0.08) for low RFIp heifers. The moderate genetic correlation between RFIp and final BF (0.36) indicates that selection for favorable RFI will reduce subcutaneous fat depots. Weak to moderate phenotypic and genetic correlations between RFIp and rib fat depth have also been reported in growing bulls (Arthur et al., 2001c; Schenkel et al., 2004) and steers (Basarab et al., 2003; Nkrumah et al., 2004, 2007). Phenotypic and genetic correlations between feed efficiency and ultrasound composition traits in Brangus heifers1 Table 8. Phenotypic and genetic correlations between feed efficiency and ultrasound composition traits in Brangus heifers1 Trait  MBW  ADG  DMI  FCR  RFIp  RFIc  Phenotypic correlation               Final BF  0.35*  −0.03  0.23*  0.21*  0.12*  −0.05   Gain in BF  0.22*  0.23*  0.35*  0.00  0.22*  −0.02   Final LMA  0.47*  0.01  0.18*  0.13*  −0.05  0.01   Gain in LMA  0.03  0.17*  0.12*  −0.09*  0.05  0.06   Final IM  −0.09*  −0.11*  −0.02  0.10*  0.08  0.06   Gain in IM  −0.02  −0.01  −0.05  −0.04  −0.04  −0.04  Genetic correlation (±SE)               Final BF  0.36 ± 0.34  0.12 ± 0.39  0.40 ± 0.25  0.28 ± 0.32  0.36 ± 0.26  0.39 ± 0.27   Gain in BF  NC2  NC  NC  NC  NC  NC   Final LMA  0.04 ± 0.50  0.41 ± 0.60  0.05 ± 0.41  −0.32 ± 0.45  −0.04 ± 0.39  0.03 ± 0.41   Gain in LMA  0.04 ± 0.50  0.80 ± 0.60  NC  0.22 ± 0.56  0.55 ± 0.24  0.77 ± 0.32   Final IM  −0.33 ± 0.29  −0.29 ± 0.31  0.05 ± 0.22  0.08 ± 0.26  0.17 ± 0.21  0.15 ± 0.22   Gain in IM  0.02 ± 0.44  −0.38 ± 0.50  −0.26 ± 0.38  −0.17 ± 0.40  −0.26 ± 0.36  −0.22 ± 0.38  Trait  MBW  ADG  DMI  FCR  RFIp  RFIc  Phenotypic correlation               Final BF  0.35*  −0.03  0.23*  0.21*  0.12*  −0.05   Gain in BF  0.22*  0.23*  0.35*  0.00  0.22*  −0.02   Final LMA  0.47*  0.01  0.18*  0.13*  −0.05  0.01   Gain in LMA  0.03  0.17*  0.12*  −0.09*  0.05  0.06   Final IM  −0.09*  −0.11*  −0.02  0.10*  0.08  0.06   Gain in IM  −0.02  −0.01  −0.05  −0.04  −0.04  −0.04  Genetic correlation (±SE)               Final BF  0.36 ± 0.34  0.12 ± 0.39  0.40 ± 0.25  0.28 ± 0.32  0.36 ± 0.26  0.39 ± 0.27   Gain in BF  NC2  NC  NC  NC  NC  NC   Final LMA  0.04 ± 0.50  0.41 ± 0.60  0.05 ± 0.41  −0.32 ± 0.45  −0.04 ± 0.39  0.03 ± 0.41   Gain in LMA  0.04 ± 0.50  0.80 ± 0.60  NC  0.22 ± 0.56  0.55 ± 0.24  0.77 ± 0.32   Final IM  −0.33 ± 0.29  −0.29 ± 0.31  0.05 ± 0.22  0.08 ± 0.26  0.17 ± 0.21  0.15 ± 0.22   Gain in IM  0.02 ± 0.44  −0.38 ± 0.50  −0.26 ± 0.38  −0.17 ± 0.40  −0.26 ± 0.36  −0.22 ± 0.38  1MBW = mid-test BW0.75; FCR = feed conversion ratio; RFIp = residual feed intake from the base model; RFIc = residual feed intake from the composition-adjusted model; BF = 12th- to 13th-rib fat thickness; LMA = LM area; IM = intramuscular fat. 2NC indicates that the model did not converge. *Correlations are different from zero at P < 0.05. View Large Table 8. Phenotypic and genetic correlations between feed efficiency and ultrasound composition traits in Brangus heifers1 Trait  MBW  ADG  DMI  FCR  RFIp  RFIc  Phenotypic correlation               Final BF  0.35*  −0.03  0.23*  0.21*  0.12*  −0.05   Gain in BF  0.22*  0.23*  0.35*  0.00  0.22*  −0.02   Final LMA  0.47*  0.01  0.18*  0.13*  −0.05  0.01   Gain in LMA  0.03  0.17*  0.12*  −0.09*  0.05  0.06   Final IM  −0.09*  −0.11*  −0.02  0.10*  0.08  0.06   Gain in IM  −0.02  −0.01  −0.05  −0.04  −0.04  −0.04  Genetic correlation (±SE)               Final BF  0.36 ± 0.34  0.12 ± 0.39  0.40 ± 0.25  0.28 ± 0.32  0.36 ± 0.26  0.39 ± 0.27   Gain in BF  NC2  NC  NC  NC  NC  NC   Final LMA  0.04 ± 0.50  0.41 ± 0.60  0.05 ± 0.41  −0.32 ± 0.45  −0.04 ± 0.39  0.03 ± 0.41   Gain in LMA  0.04 ± 0.50  0.80 ± 0.60  NC  0.22 ± 0.56  0.55 ± 0.24  0.77 ± 0.32   Final IM  −0.33 ± 0.29  −0.29 ± 0.31  0.05 ± 0.22  0.08 ± 0.26  0.17 ± 0.21  0.15 ± 0.22   Gain in IM  0.02 ± 0.44  −0.38 ± 0.50  −0.26 ± 0.38  −0.17 ± 0.40  −0.26 ± 0.36  −0.22 ± 0.38  Trait  MBW  ADG  DMI  FCR  RFIp  RFIc  Phenotypic correlation               Final BF  0.35*  −0.03  0.23*  0.21*  0.12*  −0.05   Gain in BF  0.22*  0.23*  0.35*  0.00  0.22*  −0.02   Final LMA  0.47*  0.01  0.18*  0.13*  −0.05  0.01   Gain in LMA  0.03  0.17*  0.12*  −0.09*  0.05  0.06   Final IM  −0.09*  −0.11*  −0.02  0.10*  0.08  0.06   Gain in IM  −0.02  −0.01  −0.05  −0.04  −0.04  −0.04  Genetic correlation (±SE)               Final BF  0.36 ± 0.34  0.12 ± 0.39  0.40 ± 0.25  0.28 ± 0.32  0.36 ± 0.26  0.39 ± 0.27   Gain in BF  NC2  NC  NC  NC  NC  NC   Final LMA  0.04 ± 0.50  0.41 ± 0.60  0.05 ± 0.41  −0.32 ± 0.45  −0.04 ± 0.39  0.03 ± 0.41   Gain in LMA  0.04 ± 0.50  0.80 ± 0.60  NC  0.22 ± 0.56  0.55 ± 0.24  0.77 ± 0.32   Final IM  −0.33 ± 0.29  −0.29 ± 0.31  0.05 ± 0.22  0.08 ± 0.26  0.17 ± 0.21  0.15 ± 0.22   Gain in IM  0.02 ± 0.44  −0.38 ± 0.50  −0.26 ± 0.38  −0.17 ± 0.40  −0.26 ± 0.36  −0.22 ± 0.38  1MBW = mid-test BW0.75; FCR = feed conversion ratio; RFIp = residual feed intake from the base model; RFIc = residual feed intake from the composition-adjusted model; BF = 12th- to 13th-rib fat thickness; LMA = LM area; IM = intramuscular fat. 2NC indicates that the model did not converge. *Correlations are different from zero at P < 0.05. View Large The genetic correlation between RFIp and final LMA was close to zero, whereas the gain in LMA was 0.55 (albeit with increased SE). In this study, heifers with reduced RFIp had similar final LMA and gain in LMA during the trial compared with heifers with increased RFIp; however, heifers with smaller RFIp had larger (P < 0.05) initial LMA (57.53 vs. 55.15 cm2) than heifers with increased RFIp, which may have contributed to the genetic correlation between RFIp and gain in LMA. Previous studies have reported weak phenotypic (range −0.10 to 0.14) and genetic (range −0.17 to 0.09) correlations between RFIp and final LMA that were generally close to zero (Arthur et al., 2001a; Carstens et al., 2002; Arthur et al., 2003; Nkrumah et al., 2004; Schenkel et al., 2004). In finishing steers, Basarab et al. (2003) reported that RFIp was not phenotypically correlated with gain in ultrasound LMA; however, the proportions of carcass lean tissue (−0.21) and empty body protein (−0.14; P = 0.09) were negatively correlated with RFIp. Reductions in the proportion of carcass fat may be a favorable correlated response to selection for RFI in progeny destined for the feedlot, but may be an unfavorable response in replacement females due to the positive association between body fat stores and reproductive efficiency (DeRouen et al., 1994). Angus cows selected for small RFI for 1 to 2.5 generations had smaller rib fat depths at the start of the breeding seasons than cows selected for increased RFI (Arthur et al., 2005). However, no differences were detected in pregnancy, calving, or weaning rates between cows divergently selected for RFI. Final BF and gain in BF were not phenotypically correlated with RFIc, which is expected given that inclusion of gain in BF in the linear regression model to compute RFIc adjusts for variation due to body fat. Basarab et al. (2003) found that RFI adjusted for gain in ultrasound BF and IM fat during the trial was not phenotypically correlated with carcass or empty body fat content in steers fed a high-grain diet. Likewise, Schenkel et al. (2004) reported that RFI adjusted for ultrasound BF depth was not genetically correlated with BF in bulls. In our study, a moderate genetic correlation between final BF and RFIc remained after adjusting for variation in gain in BF, although the SE associated with this genetic correlation was large, making it difficult to interpret whether the estimate was indeed different from zero. Final IM was weakly phenotypically correlated with FCR, but not RFIp or RFIc such that more efficient heifers (as measured by FCR) had less final IM. Gain in IM was not correlated with any of the feed efficiency traits measured in this study. Moreover, IM traits were not genetically correlated with any of the feed efficiency traits, and final IM was similar between heifers with low and high RFI phenotypes (Table 7). Few studies have reported genetic correlations between RFI and IM. Previous studies (Carstens et al., 2002; Basarab et al., 2003; Nkrumah et al., 2004; Schenkel et al., 2004) have reported no phenotypic correlation between ultrasound measurements of IM and FCR or RFI. Robinson and Oddy (2004) reported that RFI had genetic correlations of 0.72 and 0.48 with rump fat and rib fat, respectively, but a reduced genetic correlation with IM fat (0.22). In contrast, Basarab et al. (2003) and Nkrumah et al. (2007) reported positive phenotypic correlations between carcass marbling score and RFIp in finishing steers (0.15 and 0.17, respectively). In addition, Nkrumah et al. (2007) reported that ultrasound IM, but not carcass marbling score, was genetically correlated with RFIp (0.32 ± 0.29); however, the large SE makes it difficult to discern whether this estimate is different from zero. In general, more efficient cattle may be slightly leaner as measured by BF, but there appears to be little or no relationship between RFI and IM. Strong Pearson (0.97) and Spearman rank (0.96) correlation coefficients between RFIp and RFIc were found in this study. Other studies have reported weaker rank correlations (range 0.87 to 0.92) between RFI computed from base and carcass-adjusted models in cattle fed high-energy diets (Basarab et al., 2003; Lancaster et al., 2005). As described previously, relationships of DMI and FCR with RFIc were similar to those with RFIp, suggesting that selection for improved RFIc would result in similar corresponding changes in feed intake and efficiency as selection for improved RFIp. In these data, inclusion of carcass composition traits in the regression model used to derive expected DMI marginally increased R2 by 4 to 5%; however, RFIc is forced to be phenotypically independent of changes in body composition, whereas RFIp is not (Basarab et al., 2003; Schenkel et al., 2004). Identifying a feed efficiency trait that facilitates reductions in feed inputs without affecting growth or other value-determining traits (e.g., carcass composition) has the capability to improve profitability of beef production systems. Compared with the other feed efficiency traits examined, RFI has considerable potential for use in selection programs because it is less affected by differences in rate and composition of growth. In addition, RFI can easily be computed to be phenotypically independent of carcass composition traits. However, carcass-adjusted RFI may not be genetically independent of carcass composition traits, and further research is needed to evaluate these genetic relationships. Inclusion of carcass composition traits in the computation of RFI appears to have minimal impact on rank (i.e., accurate selection) of growing animals compared with use of performance traits (MBW and ADG) alone. However, due to the stronger relationships between RFI and carcass composition traits found in cattle fed high-grain diets, inclusion of carcass composition traits to compute RFI may be useful to reduce the potential impact of selection for RFI on carcass quality grade of progeny. Furthermore, due to its strong relationship with FCR, RFI can be used as a selection tool to improve gross feed efficiency without influencing cow mature size. LITERATURE CITED AOAC 1995. Official Methods of Analysis.  16th ed. Assoc. Off. Anal. Chem., Washington, DC. Archer J. A. Arthur P. F. Herd R. M. Parnell P. F. Pitchford W. S. 1997. Optimum postweaning test for measurement of growth rate, feed intake, and feed efficiency in British breed cattle. J. Anim. Sci.  75: 2024– 2032. [PubMed] Google Scholar CrossRef Search ADS PubMed  Arthur P. F. Archer J. A. Herd R. M. Richardson E. C. Exton S. C. Wright J. H. Dibley K. C. P. Burton D. A. 1997. Genetic and phenotypic variation in feed intake, feed efficiency and growth in beef cattle. Proc. Assoc. Adv. Anim. Breed. Genet.  12: 234– 237. Arthur P. F. Archer J. A. Johnston D. J. Herd R. M. Richardson E. C. Parnell P. F. 2001a. Genetic and phenotypic variance and covariance components for feed intake, feed efficiency, and other postweaning traits in Angus cattle. J. Anim. Sci.  79: 2805– 2811. [PubMed] Google Scholar CrossRef Search ADS   Arthur P. F. Herd R. M. Archer J. A. 2003. Should measures of body composition be included in the model for residual feed intake in beef cattle? Proc. Assoc. Adv. Anim. Breed. Genet.  15: 306– 309. Arthur P. F. Herd R. M. Wilkins J. F. Archer J. A. 2005. Maternal productivity of Angus cows divergently selected for post-weaning residual feed intake. Aust. J. Exp. Agric.  45: 985– 993. Google Scholar CrossRef Search ADS   Arthur P. F. Renand G. Krauss D. 2001b. Genetic and phenotypic relationships among different measures of growth and feed efficiency in young Charolais bulls. Livest. Prod. Sci.  68: 131– 139. Google Scholar CrossRef Search ADS   Arthur P. F. Renand G. Krauss D. 2001c. Genetic parameters for growth and feed efficiency in weaner versus yearling Charolais bulls. Aust. J. Agric. Res.  52: 471– 476. Google Scholar CrossRef Search ADS   Baker S. D. Szasz J. I. Klein T. A. Kuber P. S. Hunt C. W. Glaze J. B.Jr. Falk D. Richard R. Miller J. C. Battaglia R. A. Hill R. A. 2006. Residual feed intake of purebred Angus steers: Effects on meat quality and palatability. J. Anim. Sci.  84: 938– 945. [PubMed] Google Scholar CrossRef Search ADS PubMed  Basarab J. A. Price M. A. Aalhus J. L. Okine E. K. Snelling W. M. Lyle K. L. 2003. Residual feed intake and body composition on young growing cattle. Can. J. Anim. Sci.  83: 189– 204. Google Scholar CrossRef Search ADS   Bishop M. D. Davis M. E. Harvey W. R. Wilson G. R. VanStavern B. D. 1991. Divergent selection for postweaning feed conversion in Angus beef cattle. II. Genetic and phenotypic correlations and realized heritability estimate. J. Anim. Sci.  69: 4360– 4367. [PubMed] Google Scholar CrossRef Search ADS PubMed  Brown, E. G. 2005. Sources of biological variation in residual feed intake in growing and finishing steers. PhD Diss. Texas A&M Univ., College Station. Carstens G. E. Theis C. M. White M. B. Welsh T. H.Jr. Warrington B. G. Randel R. D. Forbes T. D. A. Lippke H. Greene L. W. Lunt D. K. 2002. Residual feed intake in beef steers: I. Correlations with performance traits and ultrasound measures of body composition. Proc. Western Section Am. Soc. Anim. Sci.  53: 552– 555. Crews D. H.Jr. 2005. Genetics of efficient feed utilization and national cattle evaluation: A review. Genet. Mol. Res.  4: 152– 165. [PubMed] Google Scholar PubMed  Crews D. H.Jr. Pollak E. J. Weaber R. L. Quaas R. L. Lipsey R. J. 2003a. Genetic parameters for carcass traits and their live animal indicators in Simmental cattle. J. Anim. Sci.  81: 1427– 1433. [PubMed] Google Scholar CrossRef Search ADS   Crews D. H.Jr. Shannon N. H. Genswein B. M. A. Crews R. E. Johnson C. M. Kendrick B. A. 2003b. Genetic parameters for net feed efficiency of beef cattle measured during postweaning growing versus finishing periods. Proc. West. Sec. Am. Soc. Anim. Sci.  54: 125– 128. DeRouen S. M. Franke D. E. Morrison D. G. Wyatt W. E. Coombs D. F. White T. W. Humes P. E. Greene B. B. 1994. Prepartum body condition and weight influences on reproductive performance of first-calf beef cows. J. Anim. Sci.  72: 1119– 1125. [PubMed] Google Scholar CrossRef Search ADS PubMed  Gunsett F. C. 1984. Linear index selection to improve traits defined as ratios. J. Anim. Sci.  59: 1185– 1193. Google Scholar CrossRef Search ADS   Herd R. M. Arthur P. F. 2009. Physiological basis for residual feed intake. J. Anim. Sci.  87: E64– E71. [PubMed] Google Scholar CrossRef Search ADS PubMed  Herd R. M. Bishop S. C. 2000. Genetic variation in residual feed intake and its association with other production traits in British Hereford cattle. Livest. Prod. Sci.  63: 111– 119. Google Scholar CrossRef Search ADS   Herd R. M. Oddy V. H. Richardson E. C. 2004. Biological basis for variation in residual feed intake in beef cattle. 1. Review of potential mechanisms. Aust. J. Exp. Agric.  44: 423– 430. Google Scholar CrossRef Search ADS   Hoque M. A. Arthur P. F. Hiramoto K. Oikawa T. 2006. Genetic relationship between different measures of feed efficiency and its component traits in Japanese Black (Wagyu) bulls. Livest. Sci.  99: 111– 118. Google Scholar CrossRef Search ADS   Jensen J. Mao I. L. Andersen B. B. Madsen P. 1992. Phenotypic and genetic relationships between residual energy intake and growth, feed intake, and carcass traits of young bulls. J. Anim. Sci.  70: 386– 395. [PubMed] Google Scholar CrossRef Search ADS PubMed  Kemp D. J. Herring W. O. Kaiser C. J. 2002. Genetic and environmental parameters for steer ultrasound and carcass traits. J. Anim. Sci.  80: 1489– 1496. [PubMed] Google Scholar CrossRef Search ADS PubMed  Kennedy B. W. van der Werf J. H. J. Meuwissen T. H. E. 1993. Genetic and statistical properties of residual feed intake. J. Anim. Sci.  71: 3239– 3250. [PubMed] Google Scholar CrossRef Search ADS PubMed  Koch R. M. Swiger L. A. Chambers D. Gregory K. E. 1963. Efficiency of feed use in beef cattle. J. Anim. Sci.  22: 486– 494. Google Scholar CrossRef Search ADS   Koots K. R. Gibson J. P. Smith C. Wilton J. W. 1994a. Analyses of published genetic parameter estimates for beef production traits. 1. Heritability. Anim. Breed. Abstr.  62: 309– 338. Koots K. R. Gibson J. P. Wilton J. W. 1994b. Analyses of published genetic parameter estimates for beef production traits. 2. Phenotypic and genetic correlations. Anim. Breed. Abstr.  62: 826– 853. Korver S. van Eekelen E. A. M. Vos H. Nieuwhof G. J. van Arendonk J. A. M. 1991. Genetic parameters for feed intake and feed efficiency in growing heifers. Livest. Prod. Sci.  29: 49– 59. Google Scholar CrossRef Search ADS   Lancaster P. A. Carstens G. E. Ribeiro F. R. B. Tedeschi L. O. Crews D. H.Jr 2009. Characterization of feed efficiency traits and relationships with feeding behavior and ultrasound carcass traits in growing bulls. J. Anim. Sci.  87: 1528– 1539. Google Scholar CrossRef Search ADS PubMed  Lancaster P. A. Schilling B. R. Carstens G. E. Brown E. G. Craig T. M. Lunt D. K. 2005. Correlations between residual feed intake and carcass traits in finishing steers administered anthelmintic treatments. J. Anim. Sci.  83( Suppl. 1): 263. (Abstr.) Littell, R. C., G. A. Milliken, W. A. Stroup, R. D. Wokfinger, and O. Schabenberger 2006. SAS for mixed models.  2nd ed. SAS Institute Inc., Cary, NC. MacNeil M. D. Northcutt S. L. 2008. National cattle evaluation system for combined analysis of carcass characteristics and indicator traits recorded using ultrasound in Angus cattle. J. Anim. Sci.  86: 2518– 2524. Google Scholar CrossRef Search ADS PubMed  Neter, J., M. H. Kutner, C. J. Nachtsheim, and W. Wasserman 1996. Applied linear regression models.  3rd ed. Irwin, Chicago, IL. Nkrumah J. D. Basarab J. A. Price M. A. Okine E. K. Ammoura A. Guercio S. Hansen C. Li C. Benkel B. Murdoch B. Moore S. S. 2004. Different measures of energetic efficiency and their phenotypic relationships with growth, feed intake, and ultrasound and carcass merit in hybrid cattle. J. Anim. Sci.  82: 2451– 2459. [PubMed] Google Scholar CrossRef Search ADS PubMed  Nkrumah J. D. Basarab J. A. Wang Z. Li C. Price M. A. Okine E. K. Crews D. H.Jr. Moore S. S. 2007. Genetic and phenotypic relationships of feed intake and measures of efficiency with growth and carcass merit of beef cattle. J. Anim. Sci.  85: 2711– 2720. [PubMed] Google Scholar CrossRef Search ADS PubMed  Ribeiro F. R. B. Carstens G. E. Lancaster P. A. Tedeschi L. O. Fernandes M. H. M. R. 2006. Evaluation of feed efficiency traits in growing Brahman heifers and relationship with body composition ultrasound traits and feeding behavior. J. Anim. Sci.  84( Suppl. 1): 123. (Abstr.) Richardson E. C. Herd R. M. 2004. Biological basis for variation in residual feed intake in beef cattle. 2. Synthesis of results following divergent selection. Aust. J. Exp. Agric.  44: 431– 440. Google Scholar CrossRef Search ADS   Robinson D. L. Oddy V. H. 2004. Genetic parameters for feed efficiency, fatness, muscle area and feeding behavior of feedlot finished beef cattle. Livest. Prod. Sci.  90: 255– 270. Google Scholar CrossRef Search ADS   Schenkel F. S. Miller S. P. Wilton J. W. 2004. Genetic parameters and breed differences for feed efficiency, growth, and body composition traits of young beef bulls. Can. J. Anim. Sci.  84: 177– 185. Google Scholar CrossRef Search ADS   St-Pierre N. R. 2001. Invited review: Integrating quantitative findings from multiple studies using mixed model methodology. J. Dairy Sci.  84: 741– 755. [PubMed] Google Scholar CrossRef Search ADS PubMed  American Society of Animal Science

Chung, K. Y.;Johnson, B. J.

doi: 10.2527/jas.2008-1645pmid: 19684257

ABSTRACT Melengestrol acetate (MGA) has been used in the United States for nearly 40 yr to enhance feedlot heifer performance, yet unequivocal studies have not been conducted to discover the mechanism of action. Our hypothesis was that MGA may induce various populations of muscle-derived cells (MDC) to the adipogenic pathway in both a bovine and murine cell culture model. To determine this, MDC were digested from the semimembranosus muscle tissue of six 14-mo-old crossbred steers. The addition of insulin, oleic acid, and ciglitizone (IOC) with cultured bovine MDC resulted in morphological differences compared with control cultures. Multilocular lipid droplets stained with Oil Red O were seen not only in single MDC, but also in fused myotubes. An increase (P < 0.05) in relative PPARγ messenger RNA (mRNA) levels was measured in MDC incubated with IOC. However, myogenin mRNA levels in MDC incubated with IOC were repressed (P < 0.05) compared with nontreated MDC. Cultures of MDC treated with 10 μM insulin, 10 μM oleic acid, 10 μM ciglitizone, 10 nM estradiol-17β (E2), and 10 nM MGA resulted in cultures with highly distributed lipid droplets not only in single cells, but also in the multinucleated myotubes. Relative C/EBPβ and PPARγ mRNA levels in total RNA isolated from MDC treated with MGA increased (P < 0.05) compared with control cultures. Estradiol treatment had no effect (P > 0.05) on these mRNA levels. The addition of both E2 and MGA to MDC increased (P < 0.05) C/EBPβ mRNA levels and tended (P = 0.06) to increase the PPARγ mRNA level. There was no difference (P > 0.10) in relative myogenin mRNA among the control, E2, and MGA treatments. Relative C/EBPβ, PPARγ, and myogenin mRNA levels were investigated in murine C2C12, C3H 10T 1/2, and 3T3-L1 cells. Treatment of cultures with 10 nM MGA increased C/EBPβ levels (P < 0.05) in C2C12 myoblasts and tended (P = 0.08) to increase C/EBPβ levels in 3T3-L1 preadipocytes. These data indicate that populations of cells are present in postnatal skeletal muscle that, under the appropriate stimuli in a culture model, express adipogenic genes and accumulate lipids. In addition, the synthetic progestogen MGA appeared to upregulate the genes necessary for conversion to the adipogenic pathway. INTRODUCTION The adipogenic transcriptional factors PPARγ and C/EBP are highly involved not only in the differentiation of adipocytes (Umek et al., 1991), but also in the transdifferentiation of myoblasts to adipocytes (Hu et al., 1995). Increased expression of C/EBP restricted mitotic growth of preadipocytes and promoted differentiation of preadipocytes (Umek et al., 1991). A specific C/EBP antibody and nucleic acid hybridization analysis was conducted previously on liver, lung, adipose, intestine, and placenta tissues (Birkenmeier et al., 1989), but not on the muscle tissue. Several in vitro studies have indicated a role for PPARγ and C/EBPα in inducing adipocyte numbers in the bovine and porcine skeletal muscle (Torii et al., 1998; Poulos and Hausman, 2006). These 2 adipogenic transcriptional factors increased under the transdifferentiation process of myoblasts when exposed to thiazolidinediones (TZD). However, these factors are not expressed when myoblasts are treated under optimal myogenic differentiation conditions (Hu et al., 1995; Kook et al., 2006; Singh et al., 2007). Therefore, PPARγ and C/EBP can be used as specific markers for indicating myoblast transdifferentiation. Thiazolidinediones and long-chain fatty acids (LCFA) are commonly used for activating differentiation of preadipocytes. Grimaldi et al. (1997) reported that treatment of C2C12 cells with TZD or LCFA reduced myogenic gene expression but induced adipogenic gene expression. Our objectives were to compare the effects of melengestrol acetate (MGA) on messenger RNA (mRNA) levels for key adipogenic genes in both bovine primary culture and mouse immortalized cell line. Thus, we tried to determine whether they inhibit or stimulate the conversion of muscle cells to adipose tissue cells, respectively. MATERIALS AND METHODS All experimental procedures were approved by the Kansas State University Institutional Animal Care and Use Committee. Bovine Muscle-Derived Cell Isolation Bovine muscle-derived cells (MDC) were isolated as described previously (Johnson et al., 1998). Muscle-derived cells were isolated from six 14-mo-old crossbred steers. Using an aseptic technique, we collected approximately 500 g of semimembranosus muscle tissue and transported it to the cell culture laboratory. Cell isolation procedures were done in the sterile environment of a culture hood. The muscle pieces were dissected from connective tissue, blood vessels, and adipose tissue and were then passed through a sterile meat grinder. The ground muscle was incubated with 0.1% pronase (Calbiochem, La Jolla, CA) in Earl's Balanced Salt Solution (Sigma, St. Louis, MO) for 1 h at 37°C, with mixing every 10 min. After incubation, the mixture was centrifuged at 1,500 × g for 4 min at room temperature. The resulting pellet was suspended in PBS (Invitrogen, Grand Island, NY; 140 mM NaCl, 1 mM KH2PO4, 3 mM KCl, 8 mM Na2HPO4) and the suspension was centrifuged at 500 × g for 10 min at room temperature. The supernatant was collected and centrifuged at 1,500 × g for 10 min at room temperature to pellet the mononucleated cells. The PBS wash and differential centrifugations were repeated 2 more times. The resulting mononucleated cell preparation was suspended in cold (4°C) Dulbecco's modified Eagle's medium (DMEM; Invitrogen) containing 10% fetal bovine serum (FBS; Invitrogen) and 10% (vol/vol) dimethylsulfoxide (Sigma). Cells were frozen in liquid nitrogen for use in future MDC studies. Bovine MDC Culture Cell culture plates (2 cm2) were precoated with 1:10 (vol/vol) diluted Matrigel (reduced growth factor form; BD Biosciences, Bedford, MA). After a 48-h incubation in DMEM containing 10% FBS at 37°C and 5% CO2, the cultures were rinsed 3 times with serum-free DMEM, and DMEM containing 10% FBS was added. When cultures approached approximately 80% confluence, phenol red-free DMEM (Invitrogen, Grand Island, NY) containing 3% horse serum was added alone (control) or with insulin (10 μM), oleic acid (100 μM), and ciglitizone (10 μM; IOC; Sigma). Cultures that were induced to transdifferentiate were maintained with IOC and the indicated additions of estradiol-17β (E2), trenbolone-17β (TBOH), or MGA for 168 h. The culture medium was replaced once after 72 h. Morphological Analysis Oil Red O and hematoxylin staining was conducted to confirm accumulated lipid droplets in the differentiated bovine MDC. After aspirating off the media, the cells were fixed with 10% neutral buffered formalin (30 mM NaH2PO4∙H2O, 54.6 mM NaHPO4, 40% formalin) for 10 min at room temperature and then washed in flowing water. The fixed cells were treated with 100% propylene glycol for 2 min and washed in water. After washing, the cells were stained with 0.5% Oil Red O solution (2:3 vol/vol mixture of 0.5% Oil Red O in isopropyl alcohol and distilled water) in darkness for 20 min and washed with 60% propylene glycol for 1 min. The cells were stained by Harris' hematoxylin in darkness for 3 min and mounted in glycerol. Transdifferentiated bovine satellite cells were identified by the presence of Oil Red O-stained lipid droplets in the cytosol and myotubes. RNA Isolation and cDNA Synthesis At 168 h, total RNA was isolated using the tri-reagent (Sigma). The RNA was quantified by absorbance at 260 nm. Quality of the RNA was determined by agarose gel electrophoresis and by optical density measurements. Subsequently, 1 μg of total RNA was reverse transcribed to produce first-strand cDNA, using TaqMan reverse transcription reagents and MultiScribe (Applied Biosystems, Foster City, CA) according to the protocol provided by the manufacturer. Real-Time Quantitative PCR from Cultured Bovine MDC, C2C12, C3H 10T 1/2, and 3T3-L1 Cells Real-time quantitative PCR was used to measure the quantity of C/EBPβ, PPARγ, stearoyl-CoA desaturase (SCD), and myogenin gene expression relative to the quantity of 18S ribosomal RNA (rRNA) in total RNA isolated from cultured bovine MDC and cells of mouse origin (see below). Measurement of the relative quantity of cDNA was performed using TaqMan Universal PCR Master Mix (Applied Biosystems), 900 nM of the appropriate forward and reverse primers, and TaqMan detection probe, Assays-on-Demand Gene Expression Products (Applied Biosystems), and 1 μM cDNA mixture. The forward and reverse primers used for C/EBPβ, PPARγ, SCD, and myogenin are shown in Table 1. Commercially available eukaryotic 18S rRNA and Assays-on-Demand primer and probes were used as an endogenous control (Applied Biosystems; GeneBank accession number X03205) and for gene expression of the mouse-origin cell line. The ABI Prism 7000 detection system and thermal cycling variables (Applied Biosystems) recommended by the manufacturer (50 cycles of 15 s at 95°C and 1 min at 60°C) were used to perform the assay. The endogenous 18S rRNA control was used to normalize the expression of C/EBPβ, PPARγ, SCD, and myogenin. Relative abundance of the PPARγ, C/EBPβ, SCD, myogenin, and G-protein-coupled receptor (GPR) 40, GPR41, and GPR43 genes were normalized with the 18S endogenous control by using the change in cycle threshold (ΔCT) method; CT values were expressed in relative units and relative whole numbers multiplied by a value of 107. Forward and reverse primers for real-time PCR for adipogenic gene messenger RNA Table 1. Forward and reverse primers for real-time PCR for adipogenic gene messenger RNA Marker gene: gene no.  Primer  C/EBPβ: NM_176788  Forward: 5′-CCA GAA GAA GGT GGA GCA ACT G-3′    Reverse: 5′-TCG GGC AGC GTC TTG AAC-3′  PPARγ: NM_181024  Forward: 5′-ATC TGC TGC AAG CCT TGG A-3′    Reverse: 5′-TGG AGC AGC TTG GCA AAG A-3′  Stearoyl-CoA desaturase: AB 075020  Forward: 5′-TGC CCA CCA CAA GTT TTC AG-3′    Reverse: 5′-GCC AAC CCA CGT GAG AGA AG-3′  Myogenin: BC 118336  Forward: 5′-AGA AGG TGA ATG AAG CCT TCG A-3′    Reverse: 5′-GCA GGC GCT CTA TGT ACT GGA T-3′  Marker gene: gene no.  Primer  C/EBPβ: NM_176788  Forward: 5′-CCA GAA GAA GGT GGA GCA ACT G-3′    Reverse: 5′-TCG GGC AGC GTC TTG AAC-3′  PPARγ: NM_181024  Forward: 5′-ATC TGC TGC AAG CCT TGG A-3′    Reverse: 5′-TGG AGC AGC TTG GCA AAG A-3′  Stearoyl-CoA desaturase: AB 075020  Forward: 5′-TGC CCA CCA CAA GTT TTC AG-3′    Reverse: 5′-GCC AAC CCA CGT GAG AGA AG-3′  Myogenin: BC 118336  Forward: 5′-AGA AGG TGA ATG AAG CCT TCG A-3′    Reverse: 5′-GCA GGC GCT CTA TGT ACT GGA T-3′  View Large Table 1. Forward and reverse primers for real-time PCR for adipogenic gene messenger RNA Marker gene: gene no.  Primer  C/EBPβ: NM_176788  Forward: 5′-CCA GAA GAA GGT GGA GCA ACT G-3′    Reverse: 5′-TCG GGC AGC GTC TTG AAC-3′  PPARγ: NM_181024  Forward: 5′-ATC TGC TGC AAG CCT TGG A-3′    Reverse: 5′-TGG AGC AGC TTG GCA AAG A-3′  Stearoyl-CoA desaturase: AB 075020  Forward: 5′-TGC CCA CCA CAA GTT TTC AG-3′    Reverse: 5′-GCC AAC CCA CGT GAG AGA AG-3′  Myogenin: BC 118336  Forward: 5′-AGA AGG TGA ATG AAG CCT TCG A-3′    Reverse: 5′-GCA GGC GCT CTA TGT ACT GGA T-3′  Marker gene: gene no.  Primer  C/EBPβ: NM_176788  Forward: 5′-CCA GAA GAA GGT GGA GCA ACT G-3′    Reverse: 5′-TCG GGC AGC GTC TTG AAC-3′  PPARγ: NM_181024  Forward: 5′-ATC TGC TGC AAG CCT TGG A-3′    Reverse: 5′-TGG AGC AGC TTG GCA AAG A-3′  Stearoyl-CoA desaturase: AB 075020  Forward: 5′-TGC CCA CCA CAA GTT TTC AG-3′    Reverse: 5′-GCC AAC CCA CGT GAG AGA AG-3′  Myogenin: BC 118336  Forward: 5′-AGA AGG TGA ATG AAG CCT TCG A-3′    Reverse: 5′-GCA GGC GCT CTA TGT ACT GGA T-3′  View Large Mouse-Origin Cell Culture Myoblasts (C2C12; American Type Culture Collection, Manassas, VA), mesenchymal precursor cells (C3H 10T 1/2), and preadipocytes (3T3-L1) were cultured in DMEM containing 10% FBS until cell confluence was reached. Cells were removed from the culture flask with 0.05% trypsin (Invitrogen) neutralized by adding DMEM containing 10% FBS. Cells were plated from passages 3 to 8 at 1,000 cells/cm2 onto 12-well tissue culture plates for Oil Red O staining, and at 2,000 cells/cm2 onto 6-well plates for gene expression assays. Plates were incubated for 48 h to reach approximately 80% confluence and were then changed with phenol red-free DMEM containing 3% horse serum without (control) or with IOC. The C3H 10 T 1/2 cells were incubated 48 h after treatment with 5-azacytidine to induce myogenic determination. Cultures that were induced to transdifferentiate were maintained with IOC and the indicated additions of E2, TBOH, or MGA for 120 h. The medium was changed once at 48 h. Statistical Analysis Data were analyzed as a completely randomized design using the MIXED model (SAS Inst. Inc., Cary, NC). Differences between the control and treatments were determined using the LSD procedure. Unless otherwise noted, means were considered different at P < 0.05, and trends were considered at P < 0.10. RESULTS Bovine MDC The cultures in these experiments were treated with insulin, oleic acid, and IOC to induce transdifferentiation in bovine MDC cultures. Addition of IOC to MDC produced morphological differences compared with the control (Figure 1A and 1B). Multilocular lipid droplets stained with Oil Red O were located not only in single bovine MDC, but also in fused multinucleated myotubes. Real-time reverse transcription-PCR analysis was used to analyze adipogenic and myogenic gene expression during MDC differentiation. Relative PPARγ mRNA levels in total RNA isolated from MDC incubated with IOC were increased (P < 0.05) compared with nontreated MDC. However, mRNA levels of myogenin (a transcription factor inducing the myogenic process) were repressed (P < 0.05) in the MDC incubated with IOC compared with nontreated MDC (Figure 1C). Figure 1. View largeDownload slide Oil Red O- and hematoxylin-stained bovine muscle-derived cells (MDC) treated with insulin (10 μM), oleic acid (100 μM), and ciglitizone (Sigma, St. Louis, MO; 10 μM; IOC; B) or with no treatment (Cont; A) in Dulbecco's modified Eagle's medium (Invitrogen, Grand Island, NY) containing 3% horse serum. Relative C/EBPβ, PPARγ, and myogenin messenger RNA (mRNA) levels in total RNA isolated from transdifferentiation-inducing bovine MDC cultures (C). Bars are means ± SE relative to the control. *Means differ from the control (P < 0.05). Values are the means of 6 culture dishes delivered from 6 animals. Color version available in the online PDF. Figure 1. View largeDownload slide Oil Red O- and hematoxylin-stained bovine muscle-derived cells (MDC) treated with insulin (10 μM), oleic acid (100 μM), and ciglitizone (Sigma, St. Louis, MO; 10 μM; IOC; B) or with no treatment (Cont; A) in Dulbecco's modified Eagle's medium (Invitrogen, Grand Island, NY) containing 3% horse serum. Relative C/EBPβ, PPARγ, and myogenin messenger RNA (mRNA) levels in total RNA isolated from transdifferentiation-inducing bovine MDC cultures (C). Bars are means ± SE relative to the control. *Means differ from the control (P < 0.05). Values are the means of 6 culture dishes delivered from 6 animals. Color version available in the online PDF. Muscle-derived cell cultures maintained with IOC were used as a positive control (Figure 2A). Treating cultures with 10 nM E2 (Figure 2B) and 20 nM TBOH (Figure 2C) induced (morphologically) a small distribution of lipid droplets in MDC as compared with the control cultures. However, cultures treated with 10 nM MGA had a larger distribution of lipid droplets, not only in a single cell, but also in the multinucleated myotubes relative to the control cultures (Figure 2D). Relative C/EBPβ and PPARγ mRNA levels in total RNA isolated from MDC treated with MGA were increased (P < 0.05) compared with the control culture (Figure 3). Although no difference was observed compared with the control, the addition of both E2 and MGA to MDC increased the C/EBPβ mRNA level (P < 0.05) and tended (P = 0.06) to increase the PPARγ mRNA level (Figure 3). There was no difference in relative myogenin mRNA levels in the control, E2, and MGA treatments (P > 0.10). The SCD gene did not differ among treatments (P > 0.10). This may be because the SCD gene was typically expressed in the late phase of adipogenic differentiation in MDC. Time course differences may cause variation in SCD gene expression because in vitro studies usually take a shorter time to perform than in vivo ones. Figure 2. View largeDownload slide Bovine muscle-derived cells (MDC) stained with Oil Red O and hematoxylin. The positive control contained insulin (10 μM), oleic acid (100 μM), and ciglitizone (Sigma, St. Louis, MO; 10 μM; IOC; A). Treatments: 10 nM estradiol-17β (IOC + E2; B); 20 nM trenbolone-17β (IOC + TBOH; C); 10 nM melengestrol acetate (IOC + MGA; D). Color version available in the online PDF. Figure 2. View largeDownload slide Bovine muscle-derived cells (MDC) stained with Oil Red O and hematoxylin. The positive control contained insulin (10 μM), oleic acid (100 μM), and ciglitizone (Sigma, St. Louis, MO; 10 μM; IOC; A). Treatments: 10 nM estradiol-17β (IOC + E2; B); 20 nM trenbolone-17β (IOC + TBOH; C); 10 nM melengestrol acetate (IOC + MGA; D). Color version available in the online PDF. Figure 3. View largeDownload slide Relative C/EBPβ, PPARγ, stearoyl-CoA desaturase (SCD), and myogenin messenger RNA (mRNA) levels in total RNA isolated from transdifferentiation-inducing bovine muscle-derived cell (MDC) cultures with IOC [insulin (10 μM), oleic acid (100 μM), and ciglitizone (Sigma, St. Louis, MO; 10 μM)], 10 nM estradiol-17β (E2), 10 nM melengestrol acetate (MGA), and 10 nM MGA + 10 nM E2 (M + E). Cultures were established as described in the culture methods, and at 48 h, the culture received 10% fetal bovine serum (Invitrogen, Grand Island, NY) and Dulbecco's modified Eagle's medium (Invitrogen) containing IOC with E2, MGA, and E2 + MGA, and without treatment. After a change of medium at 72 h, total RNA was isolated and relative mRNA was determined in 168-h cultures. Bars are means ± SE relative to the control. *Means differ from the control (P < 0.05). **Means differ from the control (P = 0.06). Values are the means of 6 culture dishes delivered from 6 animals. Figure 3. View largeDownload slide Relative C/EBPβ, PPARγ, stearoyl-CoA desaturase (SCD), and myogenin messenger RNA (mRNA) levels in total RNA isolated from transdifferentiation-inducing bovine muscle-derived cell (MDC) cultures with IOC [insulin (10 μM), oleic acid (100 μM), and ciglitizone (Sigma, St. Louis, MO; 10 μM)], 10 nM estradiol-17β (E2), 10 nM melengestrol acetate (MGA), and 10 nM MGA + 10 nM E2 (M + E). Cultures were established as described in the culture methods, and at 48 h, the culture received 10% fetal bovine serum (Invitrogen, Grand Island, NY) and Dulbecco's modified Eagle's medium (Invitrogen) containing IOC with E2, MGA, and E2 + MGA, and without treatment. After a change of medium at 72 h, total RNA was isolated and relative mRNA was determined in 168-h cultures. Bars are means ± SE relative to the control. *Means differ from the control (P < 0.05). **Means differ from the control (P = 0.06). Values are the means of 6 culture dishes delivered from 6 animals. Murine-Origin Cells Insulin, oleic acid, and ciglitizone were added to C2C12 mouse-origin myoblasts to determine their effects on the transdifferentiation process, as measured by morphological and genetic differences. Addition of IOC to cultured C2C12 cells resulted in morphological differences compared with control cultures (Figure 4A and 4B). Multilocular lipid droplets stained with Oil Red O were located not only in single C2C12 cells, but also in fused multinucleated myotubes. Real-time reverse transcription-PCR analysis was used to analyze adipogenic and myogenic processes during C2C12 cell differentiation. Relative PPARγ mRNA levels in total RNA isolated from C2C12 cells incubated with IOC did not differ (P = 0.19) from nontreated C2C12 myoblasts. However, relative myogenin mRNA levels in total RNA isolated from C2C12 myoblasts incubated with IOC were decreased (P < 0.05) compared with nontreated C2C12 myoblasts (Figure 4C). The mesenchymal precursor cells, C3H 10T 1/2, followed a pattern similar to that of mouse-origin myoblasts. Insulin, oleic acid, and ciglitizone addition to C3H 10T 1/2 mesenchymal precursor cells increased (P < 0.05) relative PPARγ mRNA levels, as measured by real-time PCR (Figure 5A). However, there was no difference in C/EBPβ, PPARγ, and myogenin mRNA levels in 3T3-L1 mouse preadipocytes during the final differentiation (P > 0.10; Figure 5B). Figure 4. View largeDownload slide Mouse-origin myoblast C2C12 cells stained with Oil Red O and hematoxylin. Control (Cont; A). Insulin (10 μM), oleic acid (100 μM), and ciglitizone (Sigma, St. Louis, MO; 10 μM; IOC; B). Relative PPARγ and myogenin messenger RNA (mRNA) levels in total RNA isolated from transdifferentiation-inducing bovine muscle-derived cell cultures (C). Bars are means ± SE relative to the control obtained from triplicate cultures. *Means differ from the control (P < 0.05). Color version available in the online PDF. Figure 4. View largeDownload slide Mouse-origin myoblast C2C12 cells stained with Oil Red O and hematoxylin. Control (Cont; A). Insulin (10 μM), oleic acid (100 μM), and ciglitizone (Sigma, St. Louis, MO; 10 μM; IOC; B). Relative PPARγ and myogenin messenger RNA (mRNA) levels in total RNA isolated from transdifferentiation-inducing bovine muscle-derived cell cultures (C). Bars are means ± SE relative to the control obtained from triplicate cultures. *Means differ from the control (P < 0.05). Color version available in the online PDF. Figure 5. View largeDownload slide Relative C/EBPβ, PPARγ, and myogenin messenger RNA (mRNA) levels in total RNA isolated from transdifferentiation-inducing mouse-origin mesenchymal precursor C3H 10T 1/2 cells (A) and preadipocyte 3T3-L1 cells (B). Bars are means ± SE relative to the control obtained from triplicate cultures. *Means differ from the control (P < 0.05). Cont = control; IOC = insulin (10 μM), oleic acid (100 μM), and ciglitizone (Sigma, St. Louis, MO; 10 μM). Figure 5. View largeDownload slide Relative C/EBPβ, PPARγ, and myogenin messenger RNA (mRNA) levels in total RNA isolated from transdifferentiation-inducing mouse-origin mesenchymal precursor C3H 10T 1/2 cells (A) and preadipocyte 3T3-L1 cells (B). Bars are means ± SE relative to the control obtained from triplicate cultures. *Means differ from the control (P < 0.05). Cont = control; IOC = insulin (10 μM), oleic acid (100 μM), and ciglitizone (Sigma, St. Louis, MO; 10 μM). The C2C12 myoblast cells maintained with IOC were used as a positive control (Figure 6A). Treating cultures with 10 nM MGA increased (P < 0.05) relative C/EBPβ mRNA levels in total RNA isolated from C2C12 myoblasts (Figure 6A). Melengestrol acetate-treated 3T3-L1 cells tended to increase relative C/EBPβ mRNA levels compared with the control; however, there was no effect of MGA treatment in the 10T 1/2 mesenchymal precursor cells (Figure 6B and 6C). Figure 6. View largeDownload slide Relative C/EBPβ, PPARγ, and myogenin messenger RNA (mRNA) levels in total RNA isolated from transdifferentiation-inducing mouse-origin myoblast C2C12 cells (A), mesenchymal precursor C3H 10T 1/2 cells (B), and preadipocyte 3T3-L1 cells (C) with the IOC, 10 nM estradiol-17β (E2), 10 nM melengestrol acetate (MGA), and 10 nM MGA + 10 nM E2 (M + E). Cultures were established as described in the culture methods, and at 48 h, the culture received 10% fetal bovine serum (Invitrogen, Grand Island, NY) and Dulbecco's modified Eagle's medium (Invitrogen) containing ciglitizone [Sigma, St. Louis, MO; IOC: insulin (10 μM), oleic acid (100 μM), and ciglitizone (10 μM)] with E2, MGA, and M + E, and without treatment. After a change of medium at 72 h, total RNA was isolated and relative mRNA were determined in 168-h cultures. Bars are means ± SE relative to the control obtained from triplicate cultures. *Means differ from the control (P < 0.05). **Means differ from the control (P = 0.08). Figure 6. View largeDownload slide Relative C/EBPβ, PPARγ, and myogenin messenger RNA (mRNA) levels in total RNA isolated from transdifferentiation-inducing mouse-origin myoblast C2C12 cells (A), mesenchymal precursor C3H 10T 1/2 cells (B), and preadipocyte 3T3-L1 cells (C) with the IOC, 10 nM estradiol-17β (E2), 10 nM melengestrol acetate (MGA), and 10 nM MGA + 10 nM E2 (M + E). Cultures were established as described in the culture methods, and at 48 h, the culture received 10% fetal bovine serum (Invitrogen, Grand Island, NY) and Dulbecco's modified Eagle's medium (Invitrogen) containing ciglitizone [Sigma, St. Louis, MO; IOC: insulin (10 μM), oleic acid (100 μM), and ciglitizone (10 μM)] with E2, MGA, and M + E, and without treatment. After a change of medium at 72 h, total RNA was isolated and relative mRNA were determined in 168-h cultures. Bars are means ± SE relative to the control obtained from triplicate cultures. *Means differ from the control (P < 0.05). **Means differ from the control (P = 0.08). DISCUSSION Melengestrol acetate, an orally active synthetic progestin, has been fed to enhance heifer performance in the United States for nearly 40 yr. The endocrine mechanism of MGA in heifers is via inhibition of the preovulatory surge of LH (Imwalle et al., 2002). Several heifer studies have indicated that treatment with MGA along with E2 and trenbolone acetate reduces feed efficiency and ribeye area but increases carcass fatness and backfat thickness (Hutcheson et al. 1993; Mader and Lechtenberg, 2000; Macken et al., 2003). Treatment with MGA reduces bovine satellite cell proliferation, and this may explain reductions in carcass muscling (Sissom et al., 2006). These data also indicated that MGA enhances IGF concentrations in both mouse and bovine satellite cells. Several implant studies have demonstrated that anabolic implants not only enhance proliferation of bovine satellite cells (Johnson et al., 1998), but also induce the expression of IGF-I in bovine satellite cells (Kamanga-Sollo et al., 2004). However, it has been reported that trenbolone acetate and E2 implants have a negative impact on marbling score (Roeber et al., 2000) and fatty acid composition (Duckett et al., 1999). A study by Smith et al. (2007) confirmed that anabolic implants affected not only fat thickness, but also marbling scores. Interestingly, the numbers of intramuscular adipocytes per gram in implanted cattle were greater than those in nonimplanted cattle. These data indicated that anabolic implants affected the proliferation of multipotent precursor cells in the LM (sixth to ninth rib) area compared with nonimplanted cattle. Our data indicated that cells treated with MGA, alone or combination with E2, had similar or greater mRNA levels of adipogenic transcription factors. The MGA treatment may attenuate E2 effects on mRNA levels of key adipogenic factors in the bovine MDC model. In vitro transdifferentiation of MDC was induced with 3% horse serum combined with IOC. Long-chain fatty acids and TZD have been used as key regulators for adipose differentiation or lipid homeostasis (Grimaldi et al., 1997; Kliewer et al., 1997). Thiazolidinediones are well defined as specific ligands that bind to the nuclear transcriptional factor PPARγ. Grimaldi et al. (1997) reported that TZD- and LCFA-treated myoblasts had less formation of multinucleated myotubes and less expression of myogenic genes in the mouse cell model. Treatment with unsaturated LCFA, such as linoleic, linolenic, or arachidonic acid, has been reported to promote lipid accumulation processes in the mouse myoblast (Grimaldi et al., 1997). However, few studies have shown the functional activation of oleic acid, even though oleic acid constitutes the largest percentage of fatty acids in the bovine adipose tissue (Chung et al., 2006). The previous report also indicated that the addition of MUFA was significantly positively correlated not only with intramuscular fat content, but also with melting point in high-marbling cattle. Thus, we hypothesized that oleic acid may be a critical factor in enhancing the adipogenic pathway in bovine MDC. Investigation of a suitable transdifferentiation mixture for bovine MDC showed the oleic acid-containing mixture to be more effective in the accumulation of lipid droplets in bovine myoblasts (data not shown). We also demonstrated the adipogenic and myogenic transcription factors to be highly involved in transdifferentiation from myoblast to adipoblast. In the current study, relative myogenin, C/EBP, PPAR, and SCD mRNA levels were used to determine myogenic or adipogenic development of MDC. The relative expression level of myogenin, used as a muscle-specific marker, was low in MDC and C2C12 myoblast cultures, but was not different in 10T 1/2 cells. These current findings are similar to results of the mouse myoblast transdifferentiation study by Grimaldi et al. (1997). Those authors reported that LCFA and TZD inhibited not only the formation of multinucleated myotubes, but also the expression of myogenic genes. However, LCFA and TZD enhanced the expression of adipogenic markers such as adipocyte lipid-binding protein and fatty acid transporter. The transcriptional factor C/EBPβ was expressed at an early stage of adipogenic differentiation (Wu et al., 1995). The adipocyte-specific transcriptional factor PPARγ is highly expressed during differentiation of preadipocytes (Tontonoz et al., 1994) and is highly involved in the transdifferentiation process of MDC (Grimaldi et al., 1997). Recently, research has demonstrated that LCFA may specifically affect the cell-surface protein GPR40 (Briscoe et al., 2003). In addition, GPR41 or GPR43 was specifically affected by short-chain fatty acids such as acetate or propionate, and may act as cell-surface signaling ligands in the adipose tissue (Brown et al., 2003; Hong et al., 2005). Thus, fatty acids, such as oleic acid, can regulate signaling cascades at the cell-surface level, and this cascade controls the adipogenic or myogenic pathways of the myoblast in the muscle. Our research indicates that cells are present in postnatal skeletal muscle that, under the appropriate stimuli in a culture model, will differentiate into adipocytes as determined by lipid accumulation. These data lead us to believe we could possibly “turn on” this process in the finishing animal during the feeding period, thereby enhancing the marbling process. In addition, anabolic steroids such as E2 and TBOH appeared to attenuate this conversion. However, the synthetic progestin MGA appeared to upregulate genes necessary for the transdifferentiation of MDC to adipocytes. Interestingly, anecdotal industry reports suggest that feeding MGA to finishing heifers may improve marbling scores. However, mouse-origin immortalized cell lines responded differentially to MGA, E2, and TBOH. One possible reason is the different fatty acid composition between the different species. Bovine fat has increased MUFA relative to fat in other species. This may cause each cell type to have a different sensitivity to different fatty acids. More research is needed to determine if these changes observed in vitro will occur in the finishing animal. LITERATURE CITED Birkenmeier E. H. Gwynn B. Haward S. Jerry J. Gordon J. I. Landschulz W. H. McKnight S. L. 1989. Tissue-specific expression, developmental regulation, and genetic mapping of the gene encoding CCAAT/enhancer binding protein. Genes Dev.  3: 1146– 1156. [PubMed] Google Scholar CrossRef Search ADS PubMed  Briscoe C. P. Tadayyon M. Andrews J. L. Benson W. G. Chambers J. K. Eilert M. M. Ellis C. Elshourbagy N. A. Goetz N. A. Minnick A. S. Murdock P. R. Sauls H. R.Jr. Shabon U. Spinage L. D. Strum J. C. Szekeres P. G. Tan K. B. Way J. M. Ignar D. M. Wilson S. Muir A. I. 2003. The orphan G protein-coupled receptor GPR40 is activated by medium and long chain fatty acids. J. Biol. Chem.  278: 11303– 11311. [PubMed] Google Scholar CrossRef Search ADS PubMed  Brown A. J. Goldsworthy S. M. Barnes A. A. Eilert M. M. Tcheang L. Daniels D. Muir A. I. Wigglesworth M. J. Kinghorn I. Fraser N. J. Pike N. B. Strum J. C. Steplewski K. M. Murdock P. R. Holder J. C. Marshall F. H. Szekeres P. G. Wilson S. Ignar D. M. Foord S. M. Wise A. Dowell S. J. 2003. The orphan G protein-coupled receptors GPR41 and GPR43 are activated by propionate and other short chain carboxylic acids. J. Biol. Chem.  278: 11312– 11319. [PubMed] Google Scholar CrossRef Search ADS PubMed  Chung K. Y. Lunt D. K. Choi C. B. Chae S. H. Rhoades R. D. Adams T. H. Booren B. Smith S. B. 2006. Lipid characteristics of subcutaneous adipose tissue and M. longissimus thoracis of Angus and Wagyu steers fed to U.S. and Japanese endpoints. Meat Sci.  73: 431– 442. Google Scholar CrossRef Search ADS   Duckett S. K. Wagner D. G. Owens F. N. Dolezal H. G. Gill D. R. 1999. Effect of anabolic implants on beef intramuscular lipid content. J. Anim. Sci.  77: 1100– 1104. [PubMed] Google Scholar CrossRef Search ADS PubMed  Grimaldi P. A. Teboul L. Inadera H. Gaillard D. Amri E. Z. 1997. Trans-differentiation of myoblasts to adipoblasts: Triggering effects of fatty acids and thiazolidinediones. Prost. Leuk. and Esst. Fatty Acids  57: 71– 75. Google Scholar CrossRef Search ADS   Hong Y.-H. Nishimura Y. Hishikawa D. Tsuzuki H. Miyahara H. Gotoh C. Choi K. Feng D. D. Chen C. Lee H. Katoh K. Rho S. Sasaki S. 2005. Acetate and propionate short chain fatty acids stimulate adipogenesis via GPCR43. Endocrinology  146: 5092– 5099. [PubMed] Google Scholar CrossRef Search ADS PubMed  Hu E. Tontonoz P. Spiegelman B. M. 1995. Transdifferentiation of myoblasts by the adipogenic transcription factors PPARγ and C/EBPα. Proc. Natl. Acad. Sci. USA  92: 9856– 9860. [PubMed] Google Scholar CrossRef Search ADS   Hutcheson D. P. Rains J. R. Paul J. W. 1993. The effects of different implant and feed additive strategies on performance and carcass characteristics in finishing heifers: A review. Prof. Anim. Sci.  9: 132– 137. Imwalle D. B. Fernandez D. L. Schillo K. K. 2002. Melengestrol acetate blocks the preovulatory surge of luteinizing hormone, the expression of behavioral estrus, and ovulation in beef heifers. J. Anim. Sci.  80: 1280– 1284. [PubMed] Google Scholar CrossRef Search ADS PubMed  Johnson B. J. Halstead N. White M. E. Hathaway M. R. DiCostanzo A. Dayton W. R. 1998. Activation state of muscle satellite cells isolated from steers implanted with a combined trenbolone acetate and estradiol implant. J. Anim. Sci.  76: 2779– 2786. [PubMed] Google Scholar CrossRef Search ADS PubMed  Kamanga-Sollo E. Pampusch M. S. Xi G. White M. E. Hathaway M. R. Dayton W. R. 2004. IGF-I mRNA levels in bovine satellite cell cultures: Effects of fusion and anabolic steroid treatment. J. Cell. Physiol.  201: 181– 189. [PubMed] Google Scholar CrossRef Search ADS PubMed  Kliewer S. A. Sundseth S. S. Jones S. A. Brown P. J. Wisely G. B. Koble C. S. Devchand P. Wahli W. Willson T. M. Lenhard J. M. Lehmann J. M. 1997. Fatty acids and eicosanoids regulate gene expression through direct interactions with peroxisome proliferator-activated receptors α and γ. Proc. Natl. Acad. Sci. USA  94: 4318– 4323. [PubMed] Google Scholar CrossRef Search ADS   Kook S. H. Choi K. C. Son Y. O. Lee K. Y. Hwang I. H. Lee H. J. Chang J. S. Choi I. H. Lee J. C. 2006. Satellite cells isolated from adult Hanwoo muscle can proliferate and differentiate into myoblasts and adipose-like cells. Mol. Cells  22: 239– 245. [PubMed] Google Scholar PubMed  Macken C. N. Milton C. T. Klopfenstein T. J. Dicke B. D. McClellan D. E. 2003. Effects of final implant type and supplementation of melengestrol acetate on finishing feedlot heifer performance, carcass characteristics, and feeding economics. Prof. Anim. Sci.  19: 159– 170. Mader T. L. Lechtenberg K. F. 2000. Growth-promoting systems for heifer calves and yearlings finished in the feedlot. J. Anim. Sci.  78: 2485– 2496. [PubMed] Google Scholar CrossRef Search ADS PubMed  Poulos S. P. Hausman G. J. 2006. A comparison of thiazolidinedione-induced adipogenesis and myogenesis in stromal-vascular cells from subcutaneous adipose tissue or semitendinosus muscle of postnatal pigs. J. Anim. Sci.  84: 1076– 1082. [PubMed] Google Scholar CrossRef Search ADS PubMed  Roeber D. L. Cannell R. C. Belk K. E. Miller R. K. Tatum J. D. Smith G. C. 2000. Implant strategies during feeding: Impact on carcass grades and consumer acceptability. J. Anim. Sci.  78: 1867– 1874. [PubMed] Google Scholar CrossRef Search ADS PubMed  Singh N. K. Chae H. S. Hwang I. H. Yoo Y. M. Ahn C. N. Lee S. H. Lee H. J. Park H. J. Chung H. Y. 2007. Transdifferentiation of porcine satellite cell to adipoblasts with ciglitizone. J. Anim. Sci.  85: 1126– 1135. [PubMed] Google Scholar CrossRef Search ADS PubMed  Sissom E. K. Reinhardt C. D. Johnson B. J. 2006. Melengestrol acetate alters muscle cell proliferation in heifers and steers. J. Anim. Sci.  84: 2950– 2958. [PubMed] Google Scholar CrossRef Search ADS PubMed  Smith K. R. Duckett S. K. Azain M. J. Sonon R. N.Jr. Pringle T. D. 2007. The effect of anabolic implants on intramuscular lipid deposition in finished beef cattle. J. Anim. Sci.  85: 430– 440. [PubMed] Google Scholar CrossRef Search ADS PubMed  Tontonoz P. Hu E. Spiegelman B. M. 1994. Stimulation of adipogenesis in fibroblasts by PPARγ2, a lipid-activated transcription factor. Cell  79: 1147– 1156. [PubMed] Google Scholar CrossRef Search ADS PubMed  Torii S. I. Kawada T. Matsuda K. Matui T. Ishihara T. Yano H. 1998. Thiazolidinedione induces the adipose differentiation of fibroblast-like cells resident within bovine skeletal muscle. Cell Biol. Int.  22: 421– 427. [PubMed] Google Scholar CrossRef Search ADS PubMed  Umek R. M. Friedman A. D. McKnight S. L. 1991. CCAAT-enhancer binding protein: A component of a differentiation switch. Science  251: 288– 292. [PubMed] Google Scholar CrossRef Search ADS PubMed  Wu Z. Xie Y. Bucher N. L. R. Farmer S. R. 1995. Conditional ectopic expression of C/EBPβ in NIH-3T3 cells includes PPAR and stimulates adipogenesis. Genes Dev.  9: 2350– 2363. [PubMed] Google Scholar CrossRef Search ADS PubMed  American Society of Animal Science

Cánovas, A.;Estany, J.;Tor, M.;Pena, R. N.;Doran, O.

doi: 10.2527/jas.2009-2091pmid: 19684259

ABSTRACT The objectives of this study were 1) to determine whether selection toward less subcutaneous fat thickness at constant intramuscular fat content in pigs is related to tissue-specific changes in the expression of lipogenic enzymes acetyl-CoA carboxylase (ACC), stearoyl-CoA desaturase (SCD), and Δ6-desaturase (Δ6d); and 2) to investigate tissue specific distribution of the porcine ACC, SCD, and Δ6d. The study was conducted on 20 purebred Duroc barrows. Ten animals were from a group selected for decreased subcutaneous fat thickness at constant intramuscular fat content (experimental group). The other 10 animals were from the unselected (control) group. Distribution of ACC, SCD, and Δ6d was investigated in semimembranosus muscle (SM), subcutaneous adipose tissue (SA), liver (L), kidney (K), heart (H), diaphragm (D), rectus capitis muscle (RCM), and abdominal fat (AF). The enzyme expression was studied in 10 animals in the case of SM and SA and in 4 animals in the case of other tissues. The following expression pattern was established for ACC: SM ≤ H = K ≤ D < RCM < L < AF = SA, whereas the expression patterns for SCD and Δ6d proteins were SM < H < RCM < D < L < K < AF = SA and RCM = SM = D < L ≤ H < SA < K < AF, respectively. Expression of ACC and SCD proteins was less in subcutaneous adipose tissue of the experimental animals when compared with the control group (P < 0.001). However, no difference (P > 0.1) in ACC and SCD protein expression between the control and experimental groups was observed in SM. Expression of Δ6d protein did not differ between the control and experimental groups for SA (P = 0.47) or SM (P = 0.31). There was a positive relationship between muscle SCD protein expression and intramuscular fat content (r = 0.48, P < 0.05). Intramuscular fat content did not correlate with ACC or Δ6d protein expression (P = 0.23 and P = 0.80, respectively). We conclude that SCD might be an effective potential biomarker for intramuscular fat deposition. INTRODUCTION An increase in carcass lean content has been considered a major objective of the pig industry. This has been achieved via selective breeding against subcutaneous fat thickness (McPhee and Trout, 1995; Hermesch et al., 2000). However, selection toward leaner carcasses has also resulted in a reduction in intramuscular fat (IMF), which is known to contribute to eating quality of pork (Ellis et al., 1996; Verbeke et al., 1999; Ruiz et al., 2000). Therefore, a challenge is to produce pigs with less subcutaneous fat thickness without reduction of IMF below the level required for optimum eating quality. It has been suggested that fat deposition in different depots might be regulated by different mechanisms (Gardan et al., 2006; Gondret et al., 2008) and that the lipogenic enzyme stearoyl-CoA desaturase (SCD) plays the key role in this process (Da Costa et al., 2004; Doran et al., 2006). Contribution of other lipogenic enzymes to the tissue-specific regulation of fat deposition in pigs remains unclear. Moreover, the mechanisms regulating IMF and subcutaneous fat deposition are largely unknown. Knowledge of these mechanisms would allow identification of physiological candidate genes that could be used for developing of tests for evaluation of effectiveness of genetic selection or dietary manipulations. The objectives of the present study were 1) to investigate whether selection toward less backfat thickness at constant IMF content in pigs is related to tissue-specific changes in the expression of the lipogenic enzymes acetyl-CoA carboxylase (ACC), SCD, and Δ6-desaturase (Δ6d), and 2) to characterize tissue distribution of ACC, SCD, and Δ6d. This would contribute to our knowledge about possible involvement of different tissues in fat deposition in pigs. MATERIALS AND METHODS All the experimental procedures were approved by the Ethics Committee for Animal Experimentation of the University of Lleida. Animals and Sample Collection The experiments were conducted on 20 purebred Duroc barrows, which were selected from previously established experimental and control groups (Reixach et al., 2008,2009). Ten of these animals were randomly chosen from a genetic group selected for decreased backfat thickness at constant BW and IMF content (experimental group). The other 10 animals were randomly chosen from the unselected (control) group. The experimental group differed from the control group in backfat thickness but not in IMF content and composition (Reixach et al., 2008). Mid-parent BLUP breeding values for backfat depth (based on approximately 37,000 records) and IMF content in gluteus medius (based on 3,000 records) were used to separate newborn litters into 2 groups differing in backfat depth but not in IMF content. This 1-generation selection process was repeated in 4 consecutive batches. Pigs in the present manuscript were obtained from the third batch. The groups were constituted according to the mid-parent (litter) breeding values for backfat thickness and BW at 180 d and for IMF content in the gluteus medius muscle adjusted for carcass weight, which were predicted using the model described in Solanes et al. (2009). Litters in the experimental group were selected against backfat thickness while maintaining IMF content and BW to the values most similar to those in the control group. Linear programming was used to select the litters satisfying the constraints above. At the age of 11 wk, 2 barrows per litter were taken at random and moved to a finishing facility until the age of 200 d. During the test period, pigs had ad libitum access to a commercial diet (Esporc, Ruidarenes, Girona, Spain; Table 1). Feed analyses were performed in triplicate. Dry matter was determined by oven-drying at 100 to 102°C for 24 h; ash was determined by muffle-heating at 550°C until constant weight. Crude protein was analyzed by the Kjeldahl method (AOAC, 2000), crude lipid was determined by Soxhlet fat analysis (AOAC, 2000), and crude fiber was analyzed by acid and alkaline digestion with a Dosi-Fiber extractor (Selecta, Barcelona, Spain; AOAC, 2000). Analyses of fatty acid composition were performed as described below for tissue samples after extracting the total lipids by the method of Hanson and Olley (1963). Composition of the diet fed during the finishing period Table 1. Composition of the diet fed during the finishing period Item  Amount  DM, g/kg  893.2  CP, g/kg of DM  164.5  Crude lipid, g/kg of DM  66.1  Crude fiber, g/kg of DM  61.6  Ash, g/kg of DM  60.0  Fatty acid composition, mg/g of fatty acids     C12:0  1.2   C14:0  12.2   C16:0  208.8   C16:1  25.2   C18:0  60.1   C18:1  353.2   C18:2  293.5   C18:3  1.8   C20:0  28.3   C20:1  8.0   C20:2  5.4   C20:4  2.3   SFA  310.6   MUFA  386.4   PUFA  303.0  Item  Amount  DM, g/kg  893.2  CP, g/kg of DM  164.5  Crude lipid, g/kg of DM  66.1  Crude fiber, g/kg of DM  61.6  Ash, g/kg of DM  60.0  Fatty acid composition, mg/g of fatty acids     C12:0  1.2   C14:0  12.2   C16:0  208.8   C16:1  25.2   C18:0  60.1   C18:1  353.2   C18:2  293.5   C18:3  1.8   C20:0  28.3   C20:1  8.0   C20:2  5.4   C20:4  2.3   SFA  310.6   MUFA  386.4   PUFA  303.0  Pigs were performance-tested at around 180 and 200 d of age. This included the measurement of BW and backfat thickness. The backfat thickness was determined ultrasonically at 5 cm off the midline at the position of the last rib (Piglog 105, SFK-Technology, Herlev, Denmark). Pigs were slaughtered at 203 d in a commercial abattoir. Carcass backfat thickness at 6 cm off the midline between the third and fourth last ribs and carcass lean percentage were measured by an on-line ultrasound automatic scanner (AutoFOM, SFK-Technology, Herlev, Denmark). Carcass lean content was estimated on the basis of 35 measurements of AutoFOM points by using the official approved equation (Decision 2001/775/CE, 2001). The differences that have been observed for backfat thickness in live animals (at 200 d) and for carcasses (at 203 d) are related to the use of different measurement techniques. Samples from 10 pigs from the same slaughtering batch (each originated from a different litter) were collected. Intramuscular fat content was measured as described previously (Bosch et al., 2009). Immediately after slaughter, samples of semimembranosus muscle (SM) and subcutaneous adipose tissue were collected at the level of the third and fourth ribs. In addition, samples of liver, kidney, heart, diaphragm, rectus capitis muscle, and abdominal fat were collected from 4 randomly chosen animals and were used to investigate tissue distribution of lipogenic enzymes. The animals used in these experiments were the first 4 pigs on the slaughter line. Three of these animals were from the experimental group, and 1 animal was from the control group. All the tissue samples were snap-frozen and stored at −80°C until analyzed. It has been previously demonstrated that these storage conditions do not affect expression of lipogenic enzymes (Doran et al., 2006). View Large Table 1. Composition of the diet fed during the finishing period Item  Amount  DM, g/kg  893.2  CP, g/kg of DM  164.5  Crude lipid, g/kg of DM  66.1  Crude fiber, g/kg of DM  61.6  Ash, g/kg of DM  60.0  Fatty acid composition, mg/g of fatty acids     C12:0  1.2   C14:0  12.2   C16:0  208.8   C16:1  25.2   C18:0  60.1   C18:1  353.2   C18:2  293.5   C18:3  1.8   C20:0  28.3   C20:1  8.0   C20:2  5.4   C20:4  2.3   SFA  310.6   MUFA  386.4   PUFA  303.0  Item  Amount  DM, g/kg  893.2  CP, g/kg of DM  164.5  Crude lipid, g/kg of DM  66.1  Crude fiber, g/kg of DM  61.6  Ash, g/kg of DM  60.0  Fatty acid composition, mg/g of fatty acids     C12:0  1.2   C14:0  12.2   C16:0  208.8   C16:1  25.2   C18:0  60.1   C18:1  353.2   C18:2  293.5   C18:3  1.8   C20:0  28.3   C20:1  8.0   C20:2  5.4   C20:4  2.3   SFA  310.6   MUFA  386.4   PUFA  303.0  Pigs were performance-tested at around 180 and 200 d of age. This included the measurement of BW and backfat thickness. The backfat thickness was determined ultrasonically at 5 cm off the midline at the position of the last rib (Piglog 105, SFK-Technology, Herlev, Denmark). Pigs were slaughtered at 203 d in a commercial abattoir. Carcass backfat thickness at 6 cm off the midline between the third and fourth last ribs and carcass lean percentage were measured by an on-line ultrasound automatic scanner (AutoFOM, SFK-Technology, Herlev, Denmark). Carcass lean content was estimated on the basis of 35 measurements of AutoFOM points by using the official approved equation (Decision 2001/775/CE, 2001). The differences that have been observed for backfat thickness in live animals (at 200 d) and for carcasses (at 203 d) are related to the use of different measurement techniques. Samples from 10 pigs from the same slaughtering batch (each originated from a different litter) were collected. Intramuscular fat content was measured as described previously (Bosch et al., 2009). Immediately after slaughter, samples of semimembranosus muscle (SM) and subcutaneous adipose tissue were collected at the level of the third and fourth ribs. In addition, samples of liver, kidney, heart, diaphragm, rectus capitis muscle, and abdominal fat were collected from 4 randomly chosen animals and were used to investigate tissue distribution of lipogenic enzymes. The animals used in these experiments were the first 4 pigs on the slaughter line. Three of these animals were from the experimental group, and 1 animal was from the control group. All the tissue samples were snap-frozen and stored at −80°C until analyzed. It has been previously demonstrated that these storage conditions do not affect expression of lipogenic enzymes (Doran et al., 2006). View Large Fatty Acid Analysis Once defrosted, SM and subcutaneous adipose tissue samples were freeze-dried and thoroughly homogenized by mixing with sand using a glass stirring rod. Due to the small sample size, DM was calculated as the weight difference before and after freeze-drying. Fatty acid composition analysis was performed in duplicate by quantitative determination of the individual fatty acids by gas chromatography. Fatty acid methyl esters were directly obtained by transesterification using a solution of 20% boron trifluoride in methanol (Rule et al., 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. The analytical column was coated with a 0.20-µm film. The oven temperature program was increased from 150 to 225°C (by 7°C per min). The injector and detector temperatures were 250°C (Tor et al., 2005). Fatty acid quantification was carried out via normalization of the area under appropriate picks after adding 1,2,3-tripentadecanoylglycerol into each sample as an internal standard. Intramuscular fat content in SM and gluteus muscles was calculated as a sum of individual fatty acids expressed as triglyceride equivalents (AOAC, 2000) on a dry tissue basis (Bosch et al., 2009). Isolation of Microsomal and Cytosolic Fractions Expression of SCD and Δ6d proteins was analyzed in microsomal fraction; ACC protein expression was analyzed in cytosolic fraction. Microsomes and cytosol were isolated by differential centrifugation with Ca2+ precipitation (Schenkman and Cinti, 1978) with minor modifications. In brief, approximately 10 g of frozen tissue (2 g for liver) was homogenized in 20 mL of cold sucrose buffer (10 mM Tris-HCl, 250 mM sucrose, pH 7.4) and centrifuged at 12,000 × g for 10 min at 4°C. The supernatant (or infranatant in the case of adipose tissue) was collected and mixed with 8 mM CaCl2 to facilitate sedimentation of the microsomal fraction. Microsomes were obtained by centrifugation at 25,000 × g for 35 min at 4°C. The supernatant (cytosolic fraction) was collected and the remaining microsomal pellet was resuspended in a buffer containing 10 mM Tris-HCl, 250 mM KCl (pH 7.4), and inhibitors of proteolytic enzymes (1.5 μM antipain, 1.5 μM pepstatin, and 2 μM leupeptin; Sigma, Dorset, UK). Total microsomal and cytosolic protein content was determined by the Bradford method using BSA as the standard (Bradford, 1976). Protein Expression Expression of the lipogenic enzymes was analyzed by western blotting. Microsomal and cytosolic proteins (6 μg) were separated by SDS-PAGE and electroblotted onto a nitrocellulose membrane as described previously (Nicolau-Solano et al., 2006). The membrane was incubated with 1 of the following antibodies: goat polyclonal anti-human ACC (Santa Cruz Biotechnology Inc., Santa Cruz, CA), rabbit polyclonal anti-bovine adipose tissue SCD (custom-made at the University of Bristol, Bristol, UK), or rabbit polyclonal anti-Δ6d IgG (Sigma Genosys Ltd., Cambridge, UK). Incubation with primary antibody was followed by incubation with secondary antibody that was horseradish peroxide-linked donkey anti-rabbit IgG for SCD and Δ6d (GE Healthcare, Amersham, Bucks, UK) or donkey anti-goat IgG for ACC (Santa Cruz Biotechonology). Blots were developed using an Enhanced Chemiluminescent Reagent (GE Healthcare) and quantified using the ImageQuant program (Molecular Dynamics, Sunnyvale, CA). A microsomal or cytosolic preparation from 1 particular animal was present on all the blots (the reference sample). Intensity of the reference sample signal was taken as 100 arbitrary units, and the intensities of other samples on the blot were expressed as fractions of the reference sample. In the case of the genetic selection effects study, we have used subcutaneous adipose tissue and SM reference samples for adipose tissue and muscle Western blots, respectively. In the case of between-tissue enzyme expression comparison, the reference sample was prepared from liver and has been continuously used on all the relevant blots. All gels and blots were done in duplicate. The duplicate samples were run on different blots. The average intensity of the duplicates was calculated. Statistical Analyses The effect of selection for reduced backfat thickness at restrained BW and IMF content on lipogenic enzyme expression was analyzed by comparing mean values between the control and experimental groups for each tissue using a t-test. Regression analysis was used to test the association between fat content and expression of the lipogenic enzymes. A regression × group interaction was included to test if associations differed across the groups. Results were considered statistically significant at P < 0.05. Data were analyzed with SAS (SAS Inst. Inc., Cary, NC) using PROC MIXED procedure. No significant diversion from normality according to the Shapiro-Wilk test was found (P < 0.05). Between-tissue differences in enzyme expression were analyzed by 1-way ANOVA with 8 levels and post-hoc comparison of means by the Tukey test. RESULTS The results of meat quality traits analysis in pigs from experimental and control groups are shown in Table 2. Backfat thickness in the experimental group was less (P < 0.05) at all time points (i.e., in live animals at 180 d, live animals at 200 d, and in carcasses at 203 d by 15.3, 14.4, and 13.2%, respectively), whereas IMF content in the gluteus muscles did not change. Fatty acid composition of the SM and subcutaneous adipose tissue for the control and experimental groups did not differ (Table 3). Backfat thickness, BW, and carcass characteristics in the control (n = 10) and experimental (n = 10) groups Table 2. Backfat thickness, BW, and carcass characteristics in the control (n = 10) and experimental (n = 10) groups Trait  Group  SEM    Control  Experimental1    Live measurement at 180 d         Age, d  177.5  179.4     BW, kg  110.6  107.0  3.2   Backfat thickness,2 mm  19.0a  16.1b  1.0  Live measurement at 200 d         Age, d  199.5  201.4     BW, kg  125.7  121.3  3.3   Backfat thickness,2 mm  21.9a  18.3b  1.1  Carcass measurement         Carcass weight, kg  96.9  94.1  2.9   Backfat thickness,3 mm  28.8a  25.0b  1.3   Lean percentage, % of fresh tissue  41.8  44.8  1.3   Intramuscular fat in GM,4 % of fresh tissue  5.0  5.0  0.4   Intramuscular fat in SM,4 % of fresh tissue  2.2  1.7  0.3  Trait  Group  SEM    Control  Experimental1    Live measurement at 180 d         Age, d  177.5  179.4     BW, kg  110.6  107.0  3.2   Backfat thickness,2 mm  19.0a  16.1b  1.0  Live measurement at 200 d         Age, d  199.5  201.4     BW, kg  125.7  121.3  3.3   Backfat thickness,2 mm  21.9a  18.3b  1.1  Carcass measurement         Carcass weight, kg  96.9  94.1  2.9   Backfat thickness,3 mm  28.8a  25.0b  1.3   Lean percentage, % of fresh tissue  41.8  44.8  1.3   Intramuscular fat in GM,4 % of fresh tissue  5.0  5.0  0.4   Intramuscular fat in SM,4 % of fresh tissue  2.2  1.7  0.3  a,bMeans within a row with different superscript differ (P < 0.05). 1Pigs in the experimental group were selected for decreased backfat depth at restrained intramuscular fat content. 2Backfat thickness on live animal was measured ultrasonically at 5 cm off the midline at the position of the last rib using the Piglog technology (SFK-Technology, Herlev, Denmark). 3Backfat thickness of carcasses was determined at 6 cm off the midline between the third and fourth last ribs using the AutoFOM automatic scanner (SFK-Technology). 4GM = gluteus muscle; SM = semimembranosus muscle. View Large Table 2. Backfat thickness, BW, and carcass characteristics in the control (n = 10) and experimental (n = 10) groups Trait  Group  SEM    Control  Experimental1    Live measurement at 180 d         Age, d  177.5  179.4     BW, kg  110.6  107.0  3.2   Backfat thickness,2 mm  19.0a  16.1b  1.0  Live measurement at 200 d         Age, d  199.5  201.4     BW, kg  125.7  121.3  3.3   Backfat thickness,2 mm  21.9a  18.3b  1.1  Carcass measurement         Carcass weight, kg  96.9  94.1  2.9   Backfat thickness,3 mm  28.8a  25.0b  1.3   Lean percentage, % of fresh tissue  41.8  44.8  1.3   Intramuscular fat in GM,4 % of fresh tissue  5.0  5.0  0.4   Intramuscular fat in SM,4 % of fresh tissue  2.2  1.7  0.3  Trait  Group  SEM    Control  Experimental1    Live measurement at 180 d         Age, d  177.5  179.4     BW, kg  110.6  107.0  3.2   Backfat thickness,2 mm  19.0a  16.1b  1.0  Live measurement at 200 d         Age, d  199.5  201.4     BW, kg  125.7  121.3  3.3   Backfat thickness,2 mm  21.9a  18.3b  1.1  Carcass measurement         Carcass weight, kg  96.9  94.1  2.9   Backfat thickness,3 mm  28.8a  25.0b  1.3   Lean percentage, % of fresh tissue  41.8  44.8  1.3   Intramuscular fat in GM,4 % of fresh tissue  5.0  5.0  0.4   Intramuscular fat in SM,4 % of fresh tissue  2.2  1.7  0.3  a,bMeans within a row with different superscript differ (P < 0.05). 1Pigs in the experimental group were selected for decreased backfat depth at restrained intramuscular fat content. 2Backfat thickness on live animal was measured ultrasonically at 5 cm off the midline at the position of the last rib using the Piglog technology (SFK-Technology, Herlev, Denmark). 3Backfat thickness of carcasses was determined at 6 cm off the midline between the third and fourth last ribs using the AutoFOM automatic scanner (SFK-Technology). 4GM = gluteus muscle; SM = semimembranosus muscle. View Large Fatty acid composition of semimembranosus muscle and subcutaneous adipose tissue in the control (n = 10) and experimental (n = 10) groups Table 3. Fatty acid composition of semimembranosus muscle and subcutaneous adipose tissue in the control (n = 10) and experimental (n = 10) groups Fatty acid, mg/g of DM1  Semimembranosus muscle  Subcutaneous adipose tissue    Control  Experimental2  SEM  Control  Experimental2  SEM  Total SFA  28.95  22.20  3.27  240.32  247.09  6.20   C14:0  1.17  0.89  0.14  10.83  10.37  0.25   C16:0  18.75  14.21  2.11  157.81  159.01  3.89   C18:0  8.91  7.02  1.03  70.47  76.44  2.41   C20:0  0.11  0.08  0.02  1.22  1.26  0.05  Total MUFA  37.72  26.17  4.34  340.43  324.67  8.03   C16:1  2.52  1.47  0.30  17.87  16.88  0.81   C18:1  34.65  24.32  4.12  315.91  301.41  7.21   C20:1  0.55  0.38  0.07  6.65  6.38  0.18  Total PUFA  13.26  12.69  0.85  143.68  139.25  4.61   C18:2  10.14  9.56  0.72  125.49  122.13  4.19   C18:3  0.46  0.42  0.06  9.80  9.13  0.32   C20:2  0.41  0.35  0.12  6.62  6.33  0.12   C20:4  2.24  2.35  0.08  1.77  1.67  0.10  Fatty acid, mg/g of DM1  Semimembranosus muscle  Subcutaneous adipose tissue    Control  Experimental2  SEM  Control  Experimental2  SEM  Total SFA  28.95  22.20  3.27  240.32  247.09  6.20   C14:0  1.17  0.89  0.14  10.83  10.37  0.25   C16:0  18.75  14.21  2.11  157.81  159.01  3.89   C18:0  8.91  7.02  1.03  70.47  76.44  2.41   C20:0  0.11  0.08  0.02  1.22  1.26  0.05  Total MUFA  37.72  26.17  4.34  340.43  324.67  8.03   C16:1  2.52  1.47  0.30  17.87  16.88  0.81   C18:1  34.65  24.32  4.12  315.91  301.41  7.21   C20:1  0.55  0.38  0.07  6.65  6.38  0.18  Total PUFA  13.26  12.69  0.85  143.68  139.25  4.61   C18:2  10.14  9.56  0.72  125.49  122.13  4.19   C18:3  0.46  0.42  0.06  9.80  9.13  0.32   C20:2  0.41  0.35  0.12  6.62  6.33  0.12   C20:4  2.24  2.35  0.08  1.77  1.67  0.10  1DM determined by freeze-drying to constant weight. 2Pigs in the experimental group were selected for decreased backfat depth at restrained intramuscular fat content. View Large Table 3. Fatty acid composition of semimembranosus muscle and subcutaneous adipose tissue in the control (n = 10) and experimental (n = 10) groups Fatty acid, mg/g of DM1  Semimembranosus muscle  Subcutaneous adipose tissue    Control  Experimental2  SEM  Control  Experimental2  SEM  Total SFA  28.95  22.20  3.27  240.32  247.09  6.20   C14:0  1.17  0.89  0.14  10.83  10.37  0.25   C16:0  18.75  14.21  2.11  157.81  159.01  3.89   C18:0  8.91  7.02  1.03  70.47  76.44  2.41   C20:0  0.11  0.08  0.02  1.22  1.26  0.05  Total MUFA  37.72  26.17  4.34  340.43  324.67  8.03   C16:1  2.52  1.47  0.30  17.87  16.88  0.81   C18:1  34.65  24.32  4.12  315.91  301.41  7.21   C20:1  0.55  0.38  0.07  6.65  6.38  0.18  Total PUFA  13.26  12.69  0.85  143.68  139.25  4.61   C18:2  10.14  9.56  0.72  125.49  122.13  4.19   C18:3  0.46  0.42  0.06  9.80  9.13  0.32   C20:2  0.41  0.35  0.12  6.62  6.33  0.12   C20:4  2.24  2.35  0.08  1.77  1.67  0.10  Fatty acid, mg/g of DM1  Semimembranosus muscle  Subcutaneous adipose tissue    Control  Experimental2  SEM  Control  Experimental2  SEM  Total SFA  28.95  22.20  3.27  240.32  247.09  6.20   C14:0  1.17  0.89  0.14  10.83  10.37  0.25   C16:0  18.75  14.21  2.11  157.81  159.01  3.89   C18:0  8.91  7.02  1.03  70.47  76.44  2.41   C20:0  0.11  0.08  0.02  1.22  1.26  0.05  Total MUFA  37.72  26.17  4.34  340.43  324.67  8.03   C16:1  2.52  1.47  0.30  17.87  16.88  0.81   C18:1  34.65  24.32  4.12  315.91  301.41  7.21   C20:1  0.55  0.38  0.07  6.65  6.38  0.18  Total PUFA  13.26  12.69  0.85  143.68  139.25  4.61   C18:2  10.14  9.56  0.72  125.49  122.13  4.19   C18:3  0.46  0.42  0.06  9.80  9.13  0.32   C20:2  0.41  0.35  0.12  6.62  6.33  0.12   C20:4  2.24  2.35  0.08  1.77  1.67  0.10  1DM determined by freeze-drying to constant weight. 2Pigs in the experimental group were selected for decreased backfat depth at restrained intramuscular fat content. View Large Effect of Selection on Expression of Lipogenic Enzymes To determine whether the reduction in subcutaneous fat thickness in experimental animals is related to inhibition of lipogenic enzyme expression, the expression of ACC, SCD, and Δ6d proteins was analyzed. Figure 1 shows that there was a decrease in the expression of ACC and SCD proteins in the subcutaneous adipose tissue from the experimental group. The ACC protein level was about 60% less (P < 0.001; Figure 1A) and SCD expression was about 50% less (P < 0.0001, Figure 1B) than the control group. No differences between control and experimental groups were observed in the case of Δ6d protein expression (P = 0.47; Figure 1C). There was no relationship between the ACC, SCD, and Δ6d protein expression and the content of SFA, MUFA, and PUFA, respectively, in adipose tissue for the whole set of animals (control plus experimental groups, data not shown). However, a regression × group interaction analysis showed that selection against backfat thickness triggered a change in the association pattern between SCD protein expression and subcutaneous adipose tissue MUFA content. Thus, there was a negative relationship between SCD protein expression and MUFA content in the control (r = −0.68, P < 0.05), but not in the experimental group (r = 0.48, P = 0.20; Figure 2). Figure 1. View largeDownload slide Representative blots and expression of acetyl-CoA carboxylase (ACC), stearoyl-CoA desaturase (SCD), and Δ6-desaturase (∆6d) proteins in subcutaneous adipose tissue of pigs in the control and experimental groups. Pigs in the experimental group were selected for decreased backfat depth at restrained intramuscular fat content. Bars represent average of measurements for 10 animals. All measurements were done in duplicate. Error bars represent SEM. a,bMeans without a common letter differ (P < 0.001). Figure 1. View largeDownload slide Representative blots and expression of acetyl-CoA carboxylase (ACC), stearoyl-CoA desaturase (SCD), and Δ6-desaturase (∆6d) proteins in subcutaneous adipose tissue of pigs in the control and experimental groups. Pigs in the experimental group were selected for decreased backfat depth at restrained intramuscular fat content. Bars represent average of measurements for 10 animals. All measurements were done in duplicate. Error bars represent SEM. a,bMeans without a common letter differ (P < 0.001). Figure 2. View largeDownload slide Relationship between stearoyl-CoA desaturase (SCD) protein expression and MUFA content (mg/g of DM) in subcutaneous adipose tissue of pigs in the control (▲, P < 0.05) and experimental groups (■, P = 0.20). Pigs in the experimental group were selected for decreased backfat depth at restrained intramuscular fat content. Figure 2. View largeDownload slide Relationship between stearoyl-CoA desaturase (SCD) protein expression and MUFA content (mg/g of DM) in subcutaneous adipose tissue of pigs in the control (▲, P < 0.05) and experimental groups (■, P = 0.20). Pigs in the experimental group were selected for decreased backfat depth at restrained intramuscular fat content. Results of ACC, SCD, and Δ6d proteins expression analysis in SM are presented in Figure 3. There were no differences (P > 0.1) between the control and experimental groups for the enzymes investigated. Moreover, no regression × group interaction of IMF, SFA, MUFA, and PUFA content on ACC, SCD, and Δ6d protein expression was found (P > 0.05, data not shown). When the results were analyzed as 1 data set, a positive relationship was found between SCD protein expression and IMF content (r = 0.48, P < 0.05; Figure 4A). Similar relationship was also found between SCD expression and the product of SCD-catalyzed reaction, MUFA (r = 0.53, P < 0.05; Figure 4B); and SCD expression and C18:1/C18:0 ratio (r = 0.61, P < 0.01; Figure 4C). No relationship was observed between IMF content and expression of ACC (P = 0.23) or Δ6d (P = 0.80) in muscle. Figure 3. View largeDownload slide Representative blots and expression of acetyl-CoA carboxylase (ACC), stearoyl-CoA desaturase (SCD), and Δ6-desaturase (∆6d) proteins in semimembranosus muscle of pigs from control and experimental groups. Pigs in the experimental group were selected for decreased backfat depth at restrained intramuscular fat content. Bars represent average of measurements for 10 animals. All measurements were done in duplicate. Error bars represent SEM. aMeans without a common letter differ (P < 0.05). Figure 3. View largeDownload slide Representative blots and expression of acetyl-CoA carboxylase (ACC), stearoyl-CoA desaturase (SCD), and Δ6-desaturase (∆6d) proteins in semimembranosus muscle of pigs from control and experimental groups. Pigs in the experimental group were selected for decreased backfat depth at restrained intramuscular fat content. Bars represent average of measurements for 10 animals. All measurements were done in duplicate. Error bars represent SEM. aMeans without a common letter differ (P < 0.05). Figure 4. View largeDownload slide Relationship between stearoyl-CoA desaturase (SCD) protein expression and intramuscular fat (IMF) content (A), MUFA content (B), and C18:1/C18:0 ratio (C) in semimembranosus muscle of pigs of the control (▲) and experiment (◆) groups. Figure 4. View largeDownload slide Relationship between stearoyl-CoA desaturase (SCD) protein expression and intramuscular fat (IMF) content (A), MUFA content (B), and C18:1/C18:0 ratio (C) in semimembranosus muscle of pigs of the control (▲) and experiment (◆) groups. Tissue Distribution of ACC, SCD, and Δ6d Proteins Although subcutaneous and intramuscular adipose tissues are the most important fat depots in terms of meat quality traits, understanding the mechanism regulating the whole body fat distribution is important for designing strategies for manipulation of fat partitioning. In this study, we have investigated tissue-specific distribution of the key lipogenic enzymes, ACC SCD, and Δ6d, catalyzing the biosynthesis of SFA, MUFA, and PUFA, respectively. Immunoreactive bands for all 3 proteins were detected in the liver, subcutaneous adipose tissue, abdominal fat, rectus capitis muscle, SM, diaphragm, heart, and kidney samples. Expression profiles of ACC, SCD, and Δ6d proteins are presented in Figure 5. Expression of ACC protein was the greatest in subcutaneous adipose tissue and abdominal fat, with a decrease in liver, followed by rectuscapitis muscle; the smallest level occurred in SM, heart, and kidney (P < 0.05; Figure 5A). Similar to ACC, the greatest SCD expression level was observed in subcutaneous adipose tissue and abdominal fat. In contrast to ACC, the next organ with the greatest SCD expression was kidney, followed by liver, diaphragm, heart, rectus capitis muscle, and SM (P < 0.05; Figure 5B). The greatest Δ6d protein expression was observed in abdominal fat, kidney, subcutaneous adipose tissue, liver, and heart. The smallest Δ6d protein level was observed in diaphragm, rectus capitis muscle, and SM (P < 0.05; Figure 5C). Figure 5. View largeDownload slide Expression profile of acetyl-CoA carboxylase (ACC), stearoyl-CoA desaturase (SCD), and Δ6-desaturase (∆6d) proteins in pig tissues. Each bar represents means of 4 pigs. All measurements were done in duplicate. Error bars represent SEM. a–gMeans without a common letter differ (P < 0.05). Preparations from liver were used as reference samples in all the cases. The intensity of the signal of the reference sample for each enzyme was taken as 100 arbitrary units. The intensity of the signals of other samples on the same blot has been calculated as a fraction of the reference sample. Representative blots of tissue-specific distribution of ACC, SCD, and ∆6d proteins are given under corresponding graphs. SB = subcutaneous adipose tissue; SM = semimembranosus muscle. Figure 5. View largeDownload slide Expression profile of acetyl-CoA carboxylase (ACC), stearoyl-CoA desaturase (SCD), and Δ6-desaturase (∆6d) proteins in pig tissues. Each bar represents means of 4 pigs. All measurements were done in duplicate. Error bars represent SEM. a–gMeans without a common letter differ (P < 0.05). Preparations from liver were used as reference samples in all the cases. The intensity of the signal of the reference sample for each enzyme was taken as 100 arbitrary units. The intensity of the signals of other samples on the same blot has been calculated as a fraction of the reference sample. Representative blots of tissue-specific distribution of ACC, SCD, and ∆6d proteins are given under corresponding graphs. SB = subcutaneous adipose tissue; SM = semimembranosus muscle. DISCUSSION It has been demonstrated that that reduction in IMF negatively affects meat juiciness, tenderness, and flavor (Fernandez et al., 1999), and therefore increasing IMF content without affecting backfat thickness is one of the challenges of the pig industry. Positive correlation between IMF and pork eating quality has been found in a range of IMF from 0.7 to 4.5% (Goransson et al., 1992; Eikelenboom et al., 1996). In spite of a large number of reports supporting the positive impact of IMF on eating quality of pork, some reported a lack of correlation between IMF and perceived juiciness, which might be related to the age and eating habits of the particular group of consumers involved in this study (Ventanas et al., 2007). In the present study, maximum differences in the backfat thickness between the control and experimental groups were observed at 180 d. This is consistent with the results of our previous larger study (188 and 172 pigs for control and experimental groups, respectively), which demonstrated that the difference in backfat depth between the 2 groups was already significant at 120 d, reached the maximum at around 180 d, and remained relatively stable, or even decreased, by 210 d (Reixach et al., 2009). The present study demonstrated that selection for reduced subcutaneous fat in pigs with constant IMF is accompanied by significant decreases of ACC and SCD protein expression in subcutaneous adipose tissue but not in muscle. Stearoyl-CoA desaturase is the enzyme involved in the biosynthesis of MUFA from SFA (Enoch et al., 1976), whereas ACC catalyzes the first step in the SFA biosynthesis. Acetyl-CoA carboxylase is considered to be a rate-limiting enzyme of lipogenesis in animal tissues, and in pig tissues in particular (Scott et al., 1981). Tissue-specific responses of porcine ACC and SCD have been previously reported in dietary trials. Thus, Doran et al. (2006) established that a reduced protein diet increases the expression of SCD protein (and to a less extent ACC protein) in pig LM, but not in subcutaneous adipose tissue. The reasons for tissue-specific changes in lipogenic enzyme expression are not clear. This could be related, at least in part, to variations in the level of transcription factors regulating the enzyme expression. It is known that IMF fat is a later-maturing tissue when compared with subcutaneous fat (Gardan et al., 2006), and there are significant morphological and metabolic differences between these 2 depots, including differences in transcription factors level (Gardan et al., 2006; Gondret et al., 2008). The other possible reason for tissue-specific responses of the porcine ACC and SCD could be tissue-specific expression of ACC and SCD isoforms. It is known that more than 1 ACC and SCD isoforms exist in mice, rats, and some other species (Thiede et al., 1986; Miyazaki and Ntambi, 2003; Miyazaki et al., 2003). It has also been demonstrated that SCD isoforms are tissue-specific, distributed with SCD1 being the predominant isoforms in liver and adipose tissue (Ntambi et al., 1988; Kim et al., 2002), whereas SCD2, SCD3, and SCD4 have been found in brain, skin, and heart, respectively (Kaestner et al., 1989; Zheng et al., 2001; Miyazaki et al., 2003). So far, 2 SCD isoforms have been reported in pigs: SCD1, which is preferentially expressed in subcutaneous adipose tissue (Ren et al., 2004); and SCD5, which has been recently reported to be expressed at very high levels in pig brain (Lengi and Corl, 2008). The SCD isoform spectrum in pig muscles and other tissues remains unknown. In the present study, a decrease in the expression of adipose tissue SCD in the experimental group was not accompanied by changes in the amount of the product of SCD catalyzed reaction, namely C16:1 and C18:1, although there was a trend toward decrease in the amount of these fatty acids. A lack of differences might be related to large between-individual variations (especially in the case of C18:1) and to a relatively small number of animals. In the present study, the selection for reduced backfat thickness with constant IMF was not accompanied by any change in the expression of muscle or subcutaneous fat Δ6d protein. Delta-6 desaturase is one of the enzymes that catalyzes the conversion of the essential fatty acids (linoleate and α-linolenate) into long-chain PUFA in animal tissues (Cho et al., 1999). The fact that we did not observe any changes in Δ6d in this study indicates that biosynthesis of PUFA might have less input in the regulation of fat partitioning in pig when compared with the biosynthesis of MUFA and SFA. Fatty acid composition and fat content depend on not only the rate of de novo lipogenesis in a particular tissue, but also on several other factors, including the rate of fatty acid transport from other lipogenic sites. In most mammals, lipogenesis occurs predominantly in the liver and adipose tissue (Girard et al., 1994). In pigs, the major site of fatty acid metabolism is subcutaneous adipose tissue, which has the greatest expression and activity of the key lipogenic enzymes (O'Hea and Leveille, 1969). The input of other tissues in regulation of fat deposition in pigs remains unclear, and tissue-specific distribution of lipogenic enzymes is unknown. In this study, we have characterized the distribution of ACC, SCD, and Δ6d proteins in 8 tissues from organs with diverse physiological functions. Immunoreactive bands corresponding to all 3 enzymes investigated were detected in liver, subcutaneous adipose tissue, abdominal fat, rectuscapitis muscle, SM, diaphragm, heart, and kidney. The greatest expression of the lipogenic enzymes was found in subcutaneous adipose tissue and abdominal fat, which is consistent with the key role of these enzymes in lipid biosynthesis and other processes that take place in adipose tissue (Ntambi and Miyazaki, 2004). Interestingly, we have also observed greater levels of SCD and Δ6d proteins in kidney (when compared with the liver, rectus capitis muscle, diaphragm, SM, and heart). Moreover, a greater level of ACC, SCD, and Δ6d proteins was also observed in the liver (when compared with the diaphragm, rectus capitis muscle, SM, and heart). Tissue-specific distribution of lipogenic enzymes is well known in other species and might be related to tissue-specific distribution of particular transcription factors and gene-specific promoter signals (Kim and Tae, 1994; Raclot and Oudart, 1999). Results of this study contribute to our understanding of the mechanisms regulating whole body fatty acid metabolism and partitioning in pigs. Favorable scenarios for IMF content increase during genetic selection are expected as long as predicted breeding values based on IMF records are available (Solanes et al., 2009). At the present time, IMF evaluation in live animals is mainly conducted by ultrasound, which is expensive and not very accurate. Evaluation of IMF in carcasses can be performed by gas chromatography or similar techniques that are time-consuming and also expensive. Several DNA polymorphisms have been considered for developing of genetic tests, but the known polymorphisms only explain a small percentage of variations in IMF. Therefore, it would be beneficial to identify reliable biomarkers for rapid IMF evaluation. From the results of the present study, we conclude that SCD might be effective potential biomarker for fat deposition and partition in pigs. Further validation of the strength of the relationship between the lipogenic enzyme expression and fat content in a larger pig population is required. LITERATURE CITED AOAC 2000. Official Methods of Analysis.  17th ed. Assoc. Off. Anal. Chem., Gaithersburg, MD. Bosch L. Tor M. Reixach J. Estany J. 2009. Estimating intramuscular fat content and fatty acid composition in live and post-mortem samples in pigs. Meat Sci.  82: 432– 437. Google Scholar CrossRef Search ADS PubMed  Bradford M. M. 1976. A rapid and sensitive method for the quantitation of microgram quantities of protein utilizing the principle of protein-dye binding. Anal. Biochem.  72: 248– 254. [PubMed] Google Scholar CrossRef Search ADS PubMed  Cho H. P. Nakamura M. T. Clarke S. D. 1999. Cloning, expression, and nutritional regulation of the mammalian delta-6 desaturase. J. Biol. Chem.  274: 471– 477. [PubMed] Google Scholar CrossRef Search ADS PubMed  Da Costa N. McGillivray C. Bai Q. Wood J. D. Evans G. Chang K. C. 2004. Restriction of dietary energy and protein induces molecular changes in young porcine skeletal muscles. J. Nutr.  134: 2191– 2199. [PubMed] Google Scholar CrossRef Search ADS PubMed  Decision 2001/775/CE. 2001. Methods for grading pig carcasses in Spain. Off. J. Eur. Commun. 291:50–51. Doran O. Moule S. K. Teye G. A. Whittington F. M. Hallett K. G. Wood J. D. 2006. A reduced protein diet induces stearoyl-CoA desaturase protein expression in pig muscle but not in subcutaneous adipose tissue: Relationship with intramuscular lipid formation. Br. J. Nutr.  95: 609– 617. [PubMed] Google Scholar CrossRef Search ADS PubMed  Eikelenboom G. Hoving-Bolink A. H. Van der Wal P. G. 1996. The eating quality of pork. 2. The influence of intramuscular fat. Fleischwirtschaft  76: 517– 518. Ellis M. Webb A. J. Avery P. J. Brown I. 1996. The influence of terminal sire genotype, sex, slaughter weight, feeding regimen and slaughter hose on growth performance, and carcass and meat quality in pigs and on the organoleptic properties of fresh pork. J. Anim. Sci.  74: 521– 530. Google Scholar CrossRef Search ADS   Enoch H. G. Catala A. Strittmatter P. 1976. Mechanisms of rat microsomal stearoyl CoA desaturase. J. Biol. Chem.  251: 5095– 5103. [PubMed] Google Scholar PubMed  Fernandez X. Monin G. Talmant A. Mourot J. Lebret B. 1999. Influence of intramuscular fat content on the quality of pig meat—1. Composition of the lipid fraction and sensory characteristics of m. longissimus lumborum. Meat Sci.  53: 59– 65. Google Scholar CrossRef Search ADS PubMed  Gardan D. Gondret F. Louveau I. 2006. Lipid metabolism and secretary function of porcine intramuscular adipocytes compared with subcutaneous and perirenal adipocytes. Am. J. Physiol. Endocrinol. Metab.  291: E372– E380. [PubMed] Google Scholar CrossRef Search ADS PubMed  Girard J. Perdereau D. Foufelle F. Prip-Buus C. Ferre P. 1994. Regulation of lipogenic enzyme gene expression by nutrients and hormones. FASEB J.  8: 36– 42. [PubMed] Google Scholar CrossRef Search ADS PubMed  Gondret F. Guitton N. Guillerm-Regost C. Louveau I. 2008. Regional differences in porcine adipocytes isolated from skeletal muscle and adipose tissues as identified by a proteomic approach. J. Anim. Sci.  86: 2115– 2125. [PubMed] Google Scholar CrossRef Search ADS PubMed  Goransson, A., G. von Seth, and E. Tornberg 1992. The influence of intramuscular fat content on the eating quality of pork. Page 245 in 38th Int. Congr.  Meat Sci. Technol., 23–28 August, Clermont-Ferrand, France. Hanson S. W. F. Olley J. 1963. Application of the method of lipid extraction to tissue homogenate. Biochem. J.  89: 101– 102. Hermesch S. Luxford B. G. Graser H.-U. 2000. Genetic parameters for lean meat yield, meat quality, reproduction and feed efficiency traits for Australian pigs: 2. Genetic relationships between production, carcase and meat quality traits. Livest. Prod. Sci.  65: 249– 259. Google Scholar CrossRef Search ADS   Kaestner K. H. Ntambi J. M. Kelly T. J.Jr. Lane M. D. 1989. Differentiation-induced gene expression in 3T3–L1 preadipocytes. A second differentially expressed gene encoding stearoyl-CoA desaturase. J. Biol. Chem.  264: 14755– 14761. [PubMed] Google Scholar PubMed  Kim H. J. Miyazaki M. Ntambi J. M. 2002. Dietary cholesterol opposes PUFA-mediated repression of the stearoyl-CoA desaturase-1 gene by SREBP-1 independent mechanism. J. Lipid Res.  43: 1750– 1757. [PubMed] Google Scholar CrossRef Search ADS PubMed  Kim K. H. Tae H. J. 1994. Pattern and regulation of acetyl-CoA carboxylase gene expression. J. Nutr.  124: 1273S– 1283S. [PubMed] Google Scholar PubMed  Lengi A. J. Corl B. A. 2008. Comparison of pig, sheep and chiken SCD5 homologs: Evidence for an early gene duplication event. Comp. Biochem. Physiol. B Biochem. Mol. Biol.  150: 440– 446. [PubMed] Google Scholar CrossRef Search ADS PubMed  McPhee C. P. Trout G. R. 1995. The effects of selection for lean growth and the halothane allele on carcass and meat quality of pigs transported long and short distances to slaughter. Livest. Prod. Sci.  42: 55– 62. Google Scholar CrossRef Search ADS   Miyazaki M. Jacobson M. J. Man W. C. Cohen P. Asilmaz E. Friedman J. M. Ntambi J. M. 2003. Identification and characterization of murine SCD4, a novel hart-specific stearoyl-CoA desaturase isoform regulated by leptin and dietary factors. J. Biol. Chem.  278: 33904– 33911. Google Scholar CrossRef Search ADS PubMed  Miyazaki M. Ntambi M. 2003. Role of stearoyl-conenzyme A desaturase in lipid metabolism. Prostaglandins Leukot. Essent. Fatty Acids  68: 113– 121. [PubMed] Google Scholar CrossRef Search ADS PubMed  Nicolau-Solano S. I. McGivan J. D. Whittington F. M. Nieuwhof G. J. Wood J. D. Doran O. 2006. Relationship between the expression of hepatic but not testicular 3β-hydroxysteroid dehydrogenase with androstenone deposition in pig adipose tissue. J. Anim. Sci.  84: 2809– 2817. [PubMed] Google Scholar CrossRef Search ADS PubMed  Ntambi J. M. Buhrow S. A. Kaestner K. H. Christy R. J. Sibley E. Kelly T. J.Jr. Lane M. D. 1988. Differentiation-induced gene expression in 3T3–L1 preadipocytes. Characterization of a differentially expressed gene encoding stearoyl-CoA desaturase. J. Biol. Chem.  263: 17291– 17300. [PubMed] Google Scholar PubMed  Ntambi J. M. Miyazaki M. 2004. Regulation of stearoyl-CoA desaturases and role in metabolism. Prog. Lipid Res.  43: 91– 104. [PubMed] Google Scholar CrossRef Search ADS PubMed  O'Hea E. K. Leveille G. A. 1969. Significance of adipose tissue and liver as sites of fatty acid synthesis in the pig and the efficiency of utilization of various substrates for lipogenesis. J. Nutr.  99: 338– 344. [PubMed] Google Scholar CrossRef Search ADS PubMed  Raclot T. Oudart H. 1999. Selectivity of fatty acids on lipid metabolism and gene expression. Proc. Nutr. Soc.  58: 633– 646. [PubMed] Google Scholar CrossRef Search ADS PubMed  Reixach, J., M. Tor, C. Díaz, and J. Estany 2008. Between and within breed variation for lean growth and intramuscular fat content and fatty acid composition in pigs.  Page 9.13 in 54th Int. Congr. Meat Sci. Technol., Cape Town, South Africa. (Abstr.) Reixach, J., M. Tor, and J. Estany 2009. Respuesta a la seleccion contra grasa dorsal a grasa intramuscular constante en cerdos Duroc. Pages 120–122 in XIII Jornada sobre Produccion Animal.  M. Joy, J. H. Calvo, C. Calvete, M. A. Latorre, I. Casasus, A. Bernues, B. Panea, A. Sanz, and J. Balcells ed. AIDA, Zaragoza, Spain. Ren J. Knorr C. Huang L. Brenig B. 2004. Isolation and molecular characterisation of the porcine stearoyl-CoA desaturase gene. Gene  340: 19– 30. [PubMed] Google Scholar CrossRef Search ADS PubMed  Ruiz J. Ventanas J. Cava R. Andres A. Garcia C. 2000. Texture and appearance of dry-cured ham as affected by fat content and fatty acid composition. Food Res. Int.  33: 91– 95. Google Scholar CrossRef Search ADS   Rule D. C. MacNeil M. D. Short R. E. 1997. Influence of sire growth potential, time on feed, and growing-finishing strategy on cholesterol and fatty acids of the ground carcass and longissimus muscle of beef steers. J. Anim. Sci.  75: 1525– 1533. [PubMed] Google Scholar CrossRef Search ADS PubMed  Schenkman J. B. Cinti D. L. 1978. Preparation of microsomes with calcium. Methods Enzymol.  52: 83– 89. [PubMed] Google Scholar CrossRef Search ADS PubMed  Scott A. R. Gornelius S. G. Mersmann H. J. 1981. Effect of age on lipogenesis and lipolysis in lean and obese swine. J. Anim. Sci.  53: 505– 511. Google Scholar CrossRef Search ADS   Solanes F. X. Reixach J. Tor M. Tibau J. Estany J. 2009. Genetic correlations and expected response for intramuscular fat content in a Duroc pig line. Livest. Sci.  123: 63– 69. Google Scholar CrossRef Search ADS   Thiede M. A. Ozols J. Strittmatter P. 1986. Construction and sequence of cDNA for rat liver stearoyl coenzyme A desaturase. J. Biol. Chem.  261: 13230– 13235. [PubMed] Google Scholar PubMed  Tor M. Estany J. Francesch F. Cubillo M. D. 2005. Comparison of fatty acid profiles of edible meat, adipose tissues and muscles between cocks and capons. Anim. Res.  54: 413– 424. Google Scholar CrossRef Search ADS   Ventanas S. Ruiz J. Garcia C. Ventanas J. 2007. Preference and juiciness of Iberian dry-cured loin as affected by intramuscular fat content, crossbreeding and rearing system. Meat Sci.  77: 324– 330. Google Scholar CrossRef Search ADS PubMed  Verbeke W. Van Oeckel M. J. Warnants N. Viaene J. Boucque Ch. V. 1999. Consumer perception, facts and possibilities to improve acceptability of health and sensory characteristics of pork. Meat Sci.  53: 77– 90. Google Scholar CrossRef Search ADS PubMed  Zheng Y. Prouty S. M. Harmon A. Sundberg J. P. Stenn K. S. Parimoo S. 2001. Scd3—A novel gene of the stearoyl-CoA desaturase family with restricted expression in skin. Genomics  71: 182– 191. [PubMed] Google Scholar CrossRef Search ADS PubMed  Footnotes 1 We acknowledge Teresa Giró, Anna Ñaco, and Laura Frutos of the Department of Animal Production, University of Lleida, Spain, for assistance with sampling and laboratory analyses. A. Cánovas is the recipient of a PhD scholarship from INIA. This research was partially funded by the Spanish Ministry of Education and Science, Spain (CICYT grant AGL2006-01243). We thank the staff of Selección Batallé for cooperation in this experiment. Lipogenic enzyme expression was analyzed at the facilities of the University of Bristol, UK. American Society of Animal Science

Hemmings, K. M.;Parr, T.;Daniel, Z. C. T. R.;Picard, B.;Buttery, P. J.;Brameld, J. M.

doi: 10.2527/jas.2009-2067pmid: 19684280

ABSTRACT The contractile and associated metabolic characteristics of muscles are determined by their myosin heavy chain (MHC) isoform expression. In large mammals, the level of MHCIIB expression, which is associated with fast glycolytic-type muscle fibers, has not been fully characterized. In this study, quantitative reverse transcription-PCR and SDS-PAGE methodologies were developed for the analyses of adult ovine MHC isoform expression and used to characterize MHC expression in 3 skeletal muscles [LM, semitendinosus, and supraspinatus) from 66-d-old lambs. Three MHC isoforms (MHCI, MHCIIA, and MHCIIX) were detected at both the protein and messenger RNA levels in all 3 muscles, with greater proportions of type II than type I MHC. The expression of MHCIIB could not be detected at the protein level in any of the muscles and was detectable (in semitendinosus muscle) only at the messenger RNA level by using semiquantitative reverse transcription-PCR, indicating that MHCIIX is the predominant fast glycolytic fiber type in the sheep muscles studied. The methodologies developed are suitable for studying fiber type transformations at the molecular level, as well as allowing analyses of very small samples, including biopsies, when histochemical analysis may not be possible. INTRODUCTION Skeletal muscles are dynamic in their fiber type composition and can change as an animal develops and ages (Picard et al., 2002) or in response to external factors, such as nutrition, growth promoters, or exercise (Lefaucheur and Gerrard, 2000). Muscle fibers are often classified based on their contractile and metabolic properties, and the different myosin heavy chain (MHC) isoforms form the basis of the contractile fiber type (Lefaucheur and Gerrard, 2000) because it is this component of the myosin molecule that has enzymatic activity, hydrolyzing ATP (Wagner and Giniger, 1981). Fibers are named according to the MHC isoforms expressed within them, and 4 adult isoforms (MHCI, MHCIIA, MHCIIX, and MHCIIB) are expressed in the skeletal muscle of many species, with each isoform originating from a different gene (Weiss and Leinwand, 1996). The conventional method for characterizing muscle fiber type is the histochemical measurement of myosin adenosine triphosphatase activity after preincubation at differing pH (Brooke and Kaiser, 1970). In sheep, 3 major fiber types (types I, IIA, and IIB) have been identified using this method (Peinado et al., 2004), with type I fibers being slow and types IIA and IIB being fast. With immunohistochemical methods, it has not been possible to distinguish between the MHCIIX and MHCIIB isoforms (Picard et al., 1999), and it is thought that fibers characterized histochemically as MHCIIB in sheep probably contain the MHCIIX isoform or both MHCIIX and MHCIIB (Picard et al., 1999; Greenwood et al., 2007), as observed in cattle (Picard et al., 1999). In sheep muscle, MHCIIB has been detected at the messenger RNA (mRNA) level (Vuocolo et al., 2007), but it is not clear whether the protein is expressed. Interestingly, MHCIIB mRNA can be detected in certain human muscle fibers (Horton et al., 2001), but not the protein. The objective of the current study was to characterize the expression of adult MHC isoforms in sheep skeletal muscles by using PCR and SDS-PAGE methodologies. MATERIALS AND METHODS Sheep studies were approved by the University of Nottingham Local Ethical Review Committee and were carried out in accordance with the UK Home Office guidelines [Animals (Scientific Procedures) Act, 1986]. Sample Collection and Preparation Six Mule × Charolais wether lambs were weaned at 53 ± 2 d and slaughtered at 66 ± 2 d by electrical stunning and severing the carotid arteries. Newly weaned lambs were used because a change in nutrition had previously been shown to affect muscle fiber type composition (Greenwood et al., 2006); therefore, we expected fiber type transformations to be occurring, resulting in differences in protein and mRNA expression of the MHC isoforms. Samples of LM (from the 12th rib), and a complete transverse of the midsection of the supraspinatus (SS) and semitendinosus (STM) skeletal muscles were taken within 5 min of slaughter. Three cattle muscles, cutaneus trunci and the diaphragma from a 19-mo-old Charolais bull (Picard et al., 1999) and the longissimus thoracis from a 15-mo-old Blonde d'Aquitaine bull, were taken within 1 h of slaughter at the experimental INRA abattoir (Theix, France); each muscle had previously been shown to express MHCIIA and MHCIIX, MHCI and MHCIIA, and MHCI, MHCIIA, MHCIIX, and MHCIIB, respectively (Picard and Cassar-Malek, 2009). All samples were immediately frozen in liquid nitrogen, crushed, and stored at −80°C for protein or mRNA expression analysis. Analysis of MHC Protein Expression Muscle samples were homogenized in 10 vol of extraction buffer containing 0.5 M sodium chloride, 20 mM sodium pyrophosphate, 50 mM Tris, 1 mM EDTA, and 1 mM dithiothreitol, incubated on ice for 10 min, and then centrifuged at 2,500 × g for 10 min at 4°C. The supernatant was mixed with an equal volume of 87% (vol/vol) glycerol and stored at −20°C. Protein concentration was determined (Bradford, 1976) and samples were adjusted to equal concentrations, and then mixed with an equal volume of loading buffer [4% (wt/vol) SDS, 125 mM Tris, pH 6.8, 20% (vol/vol; 87%) glycerol, 10% (vol/vol) β-mercaptoethanol (β-Me), and 0.02% (wt/vol) pyronin Y], incubated at room temperature for 10 min, and then incubated at 70°C for 10 min. Four micrograms of protein was loaded per well and samples were run in duplicate on 0.75-mm-thick gels, using the Protean II system (Bio-Rad, Hemel Hempstead, UK); compositions of separating and stacking gels are shown in Table 1. The upper running buffer contained 100 mM Tris, 150 mM glycine, 0.1% (wt/vol) SDS, and 10 mM β-Me; the lower buffer was one-half this concentration without β-Me. Gels were run at a constant 70 V for 30 h at 4°C, and then fixed in 30% (vol/vol) ethanol and 5% (vol/vol) acetic acid, followed by staining in 25% (vol/vol) isopropanol, 10% (vol/vol) acetic acid, and 0.2% (wt/vol) Coomassie Blue R250. Gels were destained in a 30% ethanol (vol/vol) and 5% acetic acid (vol/vol) solution, as described previously (Picard et al., 1999). Gel images were captured and proportions of the different MHC bands were determined by densitometry (Image Quant, Amersham Biosciences/GE Healthcare Europe GmbH, Saclay, France). Gel composition for 4× 0.75-mm gels1 Table 1. Gel composition for 4× 0.75-mm gels1 Item  Separating gel  Stacking gel  40% acrylamide, mL  3.92  0.98  2% bisacrylamide, mL  1.57  0.39  2 M Tris-hydrochloride, pH 8.8, mL  2.0    1 M Tris-hydrochloride, pH 6.8, mL    0.7  1 M glycine, mL  2.0    100 mM EDTA, mL    0.4  87% glycerol, mL  6.9  3.45  10% SDS  0.8  0.4  N,N,N′,N′ tetramethylethylenediamine, μL  10  5  10% ammonium persulfate, μL  200  100  Total volume, mL  17.4  6.43  Item  Separating gel  Stacking gel  40% acrylamide, mL  3.92  0.98  2% bisacrylamide, mL  1.57  0.39  2 M Tris-hydrochloride, pH 8.8, mL  2.0    1 M Tris-hydrochloride, pH 6.8, mL    0.7  1 M glycine, mL  2.0    100 mM EDTA, mL    0.4  87% glycerol, mL  6.9  3.45  10% SDS  0.8  0.4  N,N,N′,N′ tetramethylethylenediamine, μL  10  5  10% ammonium persulfate, μL  200  100  Total volume, mL  17.4  6.43  1Adapted from Picard et al. (2007). View Large Table 1. Gel composition for 4× 0.75-mm gels1 Item  Separating gel  Stacking gel  40% acrylamide, mL  3.92  0.98  2% bisacrylamide, mL  1.57  0.39  2 M Tris-hydrochloride, pH 8.8, mL  2.0    1 M Tris-hydrochloride, pH 6.8, mL    0.7  1 M glycine, mL  2.0    100 mM EDTA, mL    0.4  87% glycerol, mL  6.9  3.45  10% SDS  0.8  0.4  N,N,N′,N′ tetramethylethylenediamine, μL  10  5  10% ammonium persulfate, μL  200  100  Total volume, mL  17.4  6.43  Item  Separating gel  Stacking gel  40% acrylamide, mL  3.92  0.98  2% bisacrylamide, mL  1.57  0.39  2 M Tris-hydrochloride, pH 8.8, mL  2.0    1 M Tris-hydrochloride, pH 6.8, mL    0.7  1 M glycine, mL  2.0    100 mM EDTA, mL    0.4  87% glycerol, mL  6.9  3.45  10% SDS  0.8  0.4  N,N,N′,N′ tetramethylethylenediamine, μL  10  5  10% ammonium persulfate, μL  200  100  Total volume, mL  17.4  6.43  1Adapted from Picard et al. (2007). View Large RNA Extraction and cDNA Synthesis Total RNA was extracted using Trizol reagent according to the protocol of the manufacturer (Invitrogen, Paisley, UK). Total RNA was treated with deoxyribonuclease, and then first-strand cDNA was generated from 0.5 µg of total RNA by using random primers and Moloney murine leukemia virus reverse transcriptase in a 25-μL final volume as described by the manufacturer (Promega, Southampton, UK). Testing Primer and Probe Specificity The PCR was carried out on cDNA by using the primers shown in Figure 1. Reactions were performed in a 50-μL reaction volume of PCR Gold Buffer (Applied Biosystems, Warrington, UK), 1.5 mM magnesium chloride, 0.2 mM nucleotide mix, 0.25 µM forward and reverse primers, and 1.25 U of AmpliTaq Gold DNA Polymerase (Applied Biosystems); these were incubated at 94°C for 10 min, 40 cycles of 94°C for 30 s, 60°C (MHCIIA) or 58°C (MHCI and MHCIIX) for 30 s, and 72°C for 60 s, followed by 72°C for 5 min. Amplicons were run on a 1% (wt/vol) agarose gel and bands of the correct size (MHCI: 321 bp; MHCIIA: 282 bp; MHCIIX: 332 bp) were cloned into the pGEM-T Easy vector, transfected into JM109 competent cells (Promega), and the identity of inserts was confirmed by sequencing. To test for interference between templates in real-time PCR, vectors containing MHC isoform sequences were linearized by using SacI restriction endonuclease (New England Biolabs, Hitchin, UK), adjusted to 1 × 10−1 ng/µL, and serially diluted before being mixed 1:1 with water (control) or 1 × 10−2 ng/µL of DNA of each of the other clones. Real-time PCR primers and probes were designed using Primer Express software (version 1.5, Applied Biosytems) and are also shown in Figure 1. Real-time PCR reactions (in triplicate) were set up in 1× Universal Master Mix (Applied Biosystems) containing 0.3 μM of each forward and reverse primer and 0.2 μM of probe in a final volume of 12.5 μL in a 384-well plate; they were run on a Lightcycler 480 PCR machine (Roche, Burgess Hill, UK): 50°C for 2 min, 95°C for 10 min, and then 40 cycles of 95°C for 15 s, 60°C for 1 min, and 72°C for 1 s (data acquisition). Crossing point values were calculated using the second derivative maximum method (Lightcycler 480 Software Version 1.2, Roche). Figure 1. View largeDownload slide Variability in myosin heavy chain (MHC) isoform cDNA at the 5′ end. ClustalW (http://www.ebi.ac.uk/clustalw) alignment of ovine partial-length MHCI, MHCIIA, and MHCIIX cDNA sequences (GenBank accession numbers AB058898, AB058896, and AB058897). Underlined nucleotides indicate the position of forward and reverse primers used to generate the positive controls. Real-time PCR primers are both underlined and in bold, and dual-labeled probes (5′ 6-carboxyfluorescein and 3′ 6-carboxytetramethylrhodamine) are in bold. *Indicates identical nucleotides among all 3 sequences. Figure 1. View largeDownload slide Variability in myosin heavy chain (MHC) isoform cDNA at the 5′ end. ClustalW (http://www.ebi.ac.uk/clustalw) alignment of ovine partial-length MHCI, MHCIIA, and MHCIIX cDNA sequences (GenBank accession numbers AB058898, AB058896, and AB058897). Underlined nucleotides indicate the position of forward and reverse primers used to generate the positive controls. Real-time PCR primers are both underlined and in bold, and dual-labeled probes (5′ 6-carboxyfluorescein and 3′ 6-carboxytetramethylrhodamine) are in bold. *Indicates identical nucleotides among all 3 sequences. Skeletal Muscle MHC mRNA Expression and Statistical Analysis First-strand cDNA generated from LM, STM, and SS muscles was diluted 1:4, and from this, a pool of cDNA was generated for each muscle and a dilution series made and used as a standard curve; individual samples were further diluted 1:16 for analyses. Real-time PCR reactions using the real-time PCR primers and probes described above were carried out in 1× GeneAmp Fast PCR Master Mix (Applied Biosystems) containing 0.3 μM of each forward and reverse primer and 0.2 μM of probe, in a final volume of 25 μL. Reactions were carried out in duplicate on a 96-well plate run on a 7500T fast PCR machine (Applied Biosystems): 95°C for 20 s, and then 40 cycles of 95°C for 1 s and 60°C for 20 s; fluorescence was detected in real time. Threshold cycle (CT) values were calculated using SDS 2.2.2 software (Applied Biosystems). Relative standard curves were generated for each MHC isoform primer and probe set to calculate PCR efficiency. The CT values (y-axis) were plotted against log10 ng equivalent RNA (x-axis), and PCR reaction efficiencies (E) were calculated from the standard curve as 10(−1/slope) − 1 (Čikoš et al., 2007). An average CT value was obtained for each primer and probe set for each sample, and this was used to calculate the relative expression ratio (rER):  Because of the nature of the analysis, any one of the primer and probe sets could be set as the MHC control gene, with the expression of each other MHC target gene being calculated relative to this. The rER of the MHC control gene was set at a constant value of 1. The sum of rER values for the different primer and probe sets was calculated and the relative contribution of each primer and probe set rER was determined as a percentage. Data were expressed as percentage of total adult MHC expression. For method comparison of each isoform in different muscles, data were analyzed by 2-way ANOVA using Genstat (Lawes Agricultural Trust, Hertfordshire, UK). RESULTS AND DISCUSSION MHC Protein Expression Polyacrylamide gel electrophoresis represents a direct approach for typing muscles based on the protein expression of the MHC isoforms, but the conditions of electrophoresis have to be optimized for different species. In sheep skeletal muscle, up to 4 adult MHC isoforms have been reported to be separated by gel electrophoresis (Maier et al., 1992; Sayd et al., 1998; Zhu et al., 2006), but interpretation of each MHC isoform band migration is inconsistent between studies. Because cattle and sheep are phylogenetically close, it is likely that their MHC isoforms will migrate similarly (Sayd et al., 1998). The separation of the 4 adult MHC isoforms in cattle skeletal muscle was first described in a study by Picard et al. (1999) and was confirmed recently in specific bulls (Picard and Cassar-Malek, 2009). Electrophoretic separation using an acrylamide gradient (Picard et al., 1999) was very difficult to reproduce, so the method used herein was a modification of the method used by Talmadge and Roy (1993), which provides better resolution and reproducibility than the previous method (Picard et al., 2007). The diaphragma and cutaneus trunci cattle muscles had previously been characterized as containing only MHCI and MHCIIA or MHCIIA and MHCIIX, respectively (Picard et al., 1994). The longissimus thoracis sample used had 4 bands of MHC, characterized as MHCIIB < MHCIIX < MHCIIA < MHCI, in order of migration (Picard and Cassar-Malek, 2009). The sheep STM, SS, and LM muscles were chosen because we had previously shown (using histochemistry) that they varied in fiber type composition (Sazili et al., 2005). When the migration of sheep MHC isoforms was compared with that of cattle, it appeared that 3 major isoforms were expressed, corresponding to MHCI, MHCIIA, and MHCIIX in cattle (Figure 2). No MHCIIB band was observed in any of the samples. It is possible that the MHCIIX and MHCIIB isoforms may have comigrated, as previously observed in pigs, where 4 isoforms were expressed but only 3 bands could be separated (Bee et al., 1999). Alternatively, MHCIIB protein may not be present at a detectable level in any of the muscles studied. It was suggested previously that, where 2 MHC isoforms are coexpressed, the expression of the minor isoform can be detected only when its staining is greater than 1% of the predominant band (Bottinelli et al., 1994); therefore, it could be possible that the minor isoform, MHCIIB, was expressed at a more reduced level than was detectable and therefore could not be observed using the gel electrophoresis method. At the protein level, the proportion of fast MHC isoforms (MHCIIA and MHCIIX combined) was greater than that of slow isoforms (MHCI) in both LM and STM, whereas in SS, the proportions were similar (Table 2). In LM and STM, MHCIIX was the most abundant isoform, whereas in the SS it was MHCI. Figure 2. View largeDownload slide Myosin heavy chain (MHC) isoform protein expression. (A) Electrophoretic separation of MHC isoforms in (1) bovine longissimus thoracis muscle and (2) ovine supraspinatus (SS) muscle, indicating a lack of MHCIIB expression in ovine muscle. (B) Separation in SS (2), semitendinosus (3), and LM (4) ovine skeletal muscles and in bovine cutaneus trunci (1) and diaphragma (5) muscles, which contain only MHCIIA and MHCIIX or MHCI and MHCIIA, respectively. Figure 2. View largeDownload slide Myosin heavy chain (MHC) isoform protein expression. (A) Electrophoretic separation of MHC isoforms in (1) bovine longissimus thoracis muscle and (2) ovine supraspinatus (SS) muscle, indicating a lack of MHCIIB expression in ovine muscle. (B) Separation in SS (2), semitendinosus (3), and LM (4) ovine skeletal muscles and in bovine cutaneus trunci (1) and diaphragma (5) muscles, which contain only MHCIIA and MHCIIX or MHCI and MHCIIA, respectively. Percentage of myosin heavy chain (MHC) expression as determined by real-time PCR [messenger RNA (mRNA)] and electrophoretic separation (protein)1,2 Table 2. Percentage of myosin heavy chain (MHC) expression as determined by real-time PCR [messenger RNA (mRNA)] and electrophoretic separation (protein)1,2   Muscle and method        LM  STM  SS    P-value  Item, %  mRNA  Protein  mRNA  Protein  mRNA  Protein  SED  Method  Muscle  Method × muscle  MHCI  7.9  8.3  13.8  5.3  26.9  48.5  6.5  0.247  <0.001  0.009  MHCIIA  29.4  25.9  24.2  19.1  23.4  16.5  3.0  0.006  0.003  0.725  MHCIIX  62.7  65.8  61.9  75.6  49.7  35.0  6.3  0.852  <0.001  0.014    Muscle and method        LM  STM  SS    P-value  Item, %  mRNA  Protein  mRNA  Protein  mRNA  Protein  SED  Method  Muscle  Method × muscle  MHCI  7.9  8.3  13.8  5.3  26.9  48.5  6.5  0.247  <0.001  0.009  MHCIIA  29.4  25.9  24.2  19.1  23.4  16.5  3.0  0.006  0.003  0.725  MHCIIX  62.7  65.8  61.9  75.6  49.7  35.0  6.3  0.852  <0.001  0.014  1Data are means (n = 6). 2STM = semitendinosus muscle; SS = supraspinatus muscle; SED of the mean for the method × muscle interaction. View Large Table 2. Percentage of myosin heavy chain (MHC) expression as determined by real-time PCR [messenger RNA (mRNA)] and electrophoretic separation (protein)1,2   Muscle and method        LM  STM  SS    P-value  Item, %  mRNA  Protein  mRNA  Protein  mRNA  Protein  SED  Method  Muscle  Method × muscle  MHCI  7.9  8.3  13.8  5.3  26.9  48.5  6.5  0.247  <0.001  0.009  MHCIIA  29.4  25.9  24.2  19.1  23.4  16.5  3.0  0.006  0.003  0.725  MHCIIX  62.7  65.8  61.9  75.6  49.7  35.0  6.3  0.852  <0.001  0.014    Muscle and method        LM  STM  SS    P-value  Item, %  mRNA  Protein  mRNA  Protein  mRNA  Protein  SED  Method  Muscle  Method × muscle  MHCI  7.9  8.3  13.8  5.3  26.9  48.5  6.5  0.247  <0.001  0.009  MHCIIA  29.4  25.9  24.2  19.1  23.4  16.5  3.0  0.006  0.003  0.725  MHCIIX  62.7  65.8  61.9  75.6  49.7  35.0  6.3  0.852  <0.001  0.014  1Data are means (n = 6). 2STM = semitendinosus muscle; SS = supraspinatus muscle; SED of the mean for the method × muscle interaction. View Large MHC mRNA Expression Real-time PCR represents a quantitative method for the analysis of MHC isoform mRNA transcripts. It has greater sensitivity compared with the gel electrophoresis method and can generally detect as few as 50 to 500 copies of a transcript (Bustin, 2000). Only 3 partial-length cDNA sequences had been identified for sheep MHC isoforms, corresponding to types MHCI, MHCIIA, and MHCIIX. Therefore, the real-time PCR primers and probes were designed to detect both MHCIIX and MHCIIB combined. Because of the great degree of homology among the MHC isoforms, it was necessary to test the specificity of primers and probes designed for real-time PCR. Sheep skeletal muscle total RNA was used to produce partial-length MHC isoform cDNA to act as positive controls (Figure 1). The MHC mRNA sequences are most variable in the 5′ and 3′ untranslated region, whereas the rest of the sequence is very similar. Therefore, to distinguish between MHC transcripts, PCR primers were designed within the 5′ untranslated region for MHCI and MHCIIX isoforms, but for MHCIIA, the primers were generic and would be expected to amplify all fast MHC isoforms (MHCIIA, MHCIIX, and MHCIIB; Figure 1). Specific PCR primers for sheep MHCIIB transcripts could not be designed because no sheep MHCIIB cDNA sequence was identified in the nucleic acid databases. The products from each primer set PCR reaction were cloned, and vectors containing partial-length MHCI, MHCIIA, and MHCIIX cDNA were identified by sequencing. Serial dilutions of each of the isolated MHC cDNA (MHCI, MHCIIA, or MHCIIX) were generated, to which fixed equal concentrations of the other 2 MHC cDNA were added separately. When real-time PCR was carried out using the primer and probe set corresponding to the serially diluted MHC cDNA isoform, no interference in amplification was observed (Figure 3). Figure 3. View largeDownload slide Specificity of myosin heavy chain (MHC) isoform real-time PCR. Amplification of MHCI (A), MHCIIA (B), and MHCIIX (C) template cDNA using corresponding primer and probe sets. Each dilution was spiked with a fixed quantity of the other 2 templates to test for interference (the dashed line represents the predicted observation if an interference occurred). CP = crossing point values. Figure 3. View largeDownload slide Specificity of myosin heavy chain (MHC) isoform real-time PCR. Amplification of MHCI (A), MHCIIA (B), and MHCIIX (C) template cDNA using corresponding primer and probe sets. Each dilution was spiked with a fixed quantity of the other 2 templates to test for interference (the dashed line represents the predicted observation if an interference occurred). CP = crossing point values. Two methods are generally used to analyze real-time PCR data: the standard curve method and the ΔΔCT method. The method used herein to calculate MHC proportions was an adaptation of the CT difference method of the gene expression used by Schefe et al. (2006), using an internal standard curve generated from sample cDNA to determine PCR efficiency (Čikoš et al., 2007). This method calculates the relative quantities of mRNA transcripts, allowing proportions of the different transcripts to be determined. It does not require a housekeeping gene because differences in the quantity of RNA in the PCR reaction and reverse transcription efficiency are taken into account because measurements for the different isoforms are carried out on the same cDNA sample. The pig MHCIIX and MHCIIB sequences (GenBank accession numbers AB025262 and AB025261) are identical within the region to which sheep MHCIIX primers and probes were designed and are similar in other species; thus, if MHCIIB mRNA is expressed in sheep muscles, then the MHCIIX primers and probes are likely to detect it as well as MHCIIX. However, MHCIIB was detected previously in sheep muscle by using primers designed to the cattle MHCIIB mRNA sequence (Vuocolo et al., 2007). In that study, 4 adult MHC isoforms were detected in sheep LM skeletal muscle, with MHCIIX being the predominant MHC isoform expressed and MHCIIB being expressed in extremely small amounts (less than 1% of total adult MHC expression). In the present study, the 3 sheep muscles used to characterize MHC isoform protein expression were also examined for MHC isoform mRNA expression by using real-time PCR (Table 2). The SS had the greatest proportion of MHCI mRNA, whereas the proportions in LM and STM were similar. The proportion of fast transcripts (MHCIIA and MHCIIX combined) was greater than the proportion of slow transcripts (MHCI) in all 3 muscles examined, and MHCIIX mRNA made up the greatest proportion of the MHC isoform transcripts. As described above, we predicted that the MHCIIX primers and probes would also detect MHCIIB mRNA. The analysis of MHC protein expression had failed to detect any MHCIIB. Therefore, to determine whether sheep MHCIIB mRNA could be detected in the muscles studied, semiquantitative reverse transcription-PCR was carried out using published primers (Vuocolo et al., 2007) previously shown to detect MHCIIB transcripts in sheep LM muscle, whereas the level of MHCIIX + MHCIIB combined was determined using the real-time PCR primers described in Figure 1. Expression of MHCIIX + MHCIIB combined was consistently greater than MHCIIB alone, and, in several instances, the MHCIIB primers failed to generate a product (Figure 4); MHCIIB mRNA was barely detectable in the LM and SS muscles, but was detected in some of the STM muscles. Hence, MHCIIB expression was extremely low, suggesting that the predominant MHC isoform in fibers previously characterized as type MHCIIB (fast glycolytic) was actually MHCIIX. Figure 4. View largeDownload slide Semiquantitative PCR for myosin heavy chain (MHC) IIB and MHCIIB and MHCIIX (35 cycles) in (A) LM, (B) semitendinosus, and (C) supraspinatus muscles of the 6 lambs studied (1 to 6). Figure 4. View largeDownload slide Semiquantitative PCR for myosin heavy chain (MHC) IIB and MHCIIB and MHCIIX (35 cycles) in (A) LM, (B) semitendinosus, and (C) supraspinatus muscles of the 6 lambs studied (1 to 6). The measures of percentages of MHCI protein and mRNA (Table 2) were similar in LM and STM muscles, but were quite different in SS muscle (P = 0.009, muscle × method interaction). Likewise, the percentages of MHCIIX protein and mRNA (Table 2) were similar in LM muscle, but were different in STM and SS muscles (P = 0.014, muscle × method interactions). Differences between muscles (P = 0.003) and methods (P = 0.006) were observed for percentage of MHCIIA, but no interaction. This was due to LM having a greater percentage of MHCIIA than the other 2 muscles, and the percentage of MHCIIA mRNA being slightly greater than protein. Although histochemical analyses of fiber type were not carried out here, previous studies in humans and cattle have shown good agreement between the electrophoretic separation method used herein and immunohistochemical determination of fiber type composition (Serrano et al., 2001; Picard et al., 2007). Differences between protein and mRNA measures of MHC isoforms have been observed by other groups and are suggested to occur when fibers are transforming (Andersen and Schiaffino, 1997), or they have been used as evidence that expression of MHC isoforms might be regulated at the translational level (Horton et al., 2001). We suggest that the data presented herein provide evidence for transcriptional regulation of the MHC genes because the 2 measures were similar in the LM. Thus, the difference in expression in the SS muscle is likely to represent a transition in MHC expression; this may be explained by the fact that the muscle samples were from newly weaned lambs, which have altered nutrition, a factor that has previously been shown to alter fiber type (Greenwood et al., 2006). Although we cannot rule out the effect of age, we suggest that the transitions are more likely to be an effect of weaning. In conclusion, we have developed both PAGE and quantitative real-time PCR methodologies to detect sheep MHC isoforms that will be useful for investigating fiber type transformations at the molecular level, as well as for allowing fiber type determination on very small samples, such as biopsies, where histochemical analyses may not be possible. Although metabolic activity generally correlates with the contractile fiber type, a recent report indicates that the 2 components may be uncoupled (Park et al., 2009); therefore, combining the methodologies described in this paper with measures of metabolic enzyme activity would allow a more complete classification of muscle fiber type. In the current study, expression of the fastest isoform, MHCIIB, was observed only at the mRNA level in a subset of samples. This agrees with studies in other large mammals, including humans, in which MHCIIB mRNA transcripts have been detected, but not the protein (Horton et al., 2001). However, this is different from what has been reported for pigs (Lefaucheur et al., 2004) and rodents (Vadászová et al., 2006), in which both the mRNA and protein can be detected. Expression of the MHCIIB isoform at the protein level has been demonstrated recently in bovine skeletal muscle, but only in certain breeds of bulls (Picard and Cassar-Malek, 2009). Hence, there are some fundamental differences between species in terms of muscle biology, at least in terms of fiber type, which means caution is required when relating results from studies in one species to those in another. LITERATURE CITED Andersen J. L. Schiaffino S. 1997. Mismatch between myosin heavy chain mRNA and protein distribution in human skeletal muscle fibers. Am. J. Physiol. Cell Physiol.  272: C1881– C1889. Google Scholar CrossRef Search ADS   Bee G. Solomon M. B. Czerwinski S. M. Long C. Pursel V. G. 1999. Correlation between histochemically assessed fiber type distribution and isomyosin and myosin heavy chain content in porcine skeletal muscles. J. Anim. Sci.  77: 2104– 2111. [PubMed] Google Scholar CrossRef Search ADS PubMed  Bottinelli R. Canepari M. Reggiani C. Stienen G. J. M. 1994. Myofibrillar ATPase activity during isometric contraction and isomyosin composition in rat single skinned muscle-fibers. J. Physiol.  481: 663– 675. [PubMed] Google Scholar CrossRef Search ADS PubMed  Bradford M. M. 1976. Rapid and sensitive method for quantitation of microgram quantities of protein utilizing principle of protein-dye binding. Anal. Biochem.  72: 248– 254. [PubMed] Google Scholar CrossRef Search ADS PubMed  Brooke M. H. Kaiser K. K. 1970. Three “myosin adenosine triphosphatase” systems: The nature of their pH lability and sulfhydryl dependence. J. Histochem. Cytochem.  18: 670– 672. [PubMed] Google Scholar CrossRef Search ADS PubMed  Bustin S. A. 2000. Absolute quantification of mRNA using real-time reverse transcription polymerase chain reaction assays. J. Mol. Endocrinol.  25: 169– 193. [PubMed] Google Scholar CrossRef Search ADS PubMed  Čikoš Š. Bukovská A. Koppel J. 2007. Relative quantification of mRNA: Comparison of methods currently used for real-time PCR data analysis. BMC Mol. Biol.  8: 113. [PubMed] Google Scholar CrossRef Search ADS PubMed  Greenwood P. L. Gardner G. E. Hegarty R. S. 2006. Lamb myofibre characteristics are influenced by sire estimated breeding values and pastoral nutritional system. Aust. J. Agric. Res.  57: 627– 639. Google Scholar CrossRef Search ADS   Greenwood P. L. Harden S. Hopkins D. L. 2007. Myofibre characteristics of ovine longissimus and semitendinosus muscles are influenced by sire breed, gender, rearing type, age and carcass weight. Aust. J. Exp. Agric.  47: 1137– 1146. Google Scholar CrossRef Search ADS   Horton M. J. Brandon C. A. Morris T. J. Braun T. W. Yaw K. M. Sciote J. J. 2001. Abundant expression of myosin heavy-chain IIB RNA in a subset of human masseter muscle fibres. Arch. Oral Biol.  46: 1039– 1050. [PubMed] Google Scholar CrossRef Search ADS PubMed  Lefaucheur L. Gerrard D. 2000. Muscle fiber plasticity in farm mammals. J. Anim. Sci.  77 ( E Suppl.): 1– 19. Google Scholar CrossRef Search ADS   Lefaucheur L. Milan D. Ecolan P. Le Callennec C. 2004. Myosin heavy chain composition of different skeletal muscles in Large White and Meishan pigs. J. Anim. Sci.  82: 1931– 1941. [PubMed] Google Scholar CrossRef Search ADS PubMed  Maier A. McEwan J. C. Dodds K. G. Fischman D. A. Fitzsimons R. B. Harris A. J. 1992. Myosin heavy-chain composition of single fibers and their origins and distribution in developing fascicles of sheep tibialis cranialis muscles. J. Muscle Res. Cell Motil.  13: 551– 572. [PubMed] Google Scholar CrossRef Search ADS PubMed  Park S. K. Gunawan A. M. Scheffler T. L. Grant A. L. Gerrard D. E. 2009. Myosin heavy chain isoform content and energy metabolism can be uncoupled in pig skeletal muscle. J. Anim. Sci.  87: 522– 531. [PubMed] Google Scholar CrossRef Search ADS PubMed  Peinado B. Latorre R. Váquez-Autón J. M. Poto A. Ramírez G. López-Albors O. Moreno F. Gil F. 2004. Histochemical skeletal muscle fibre types in the sheep. Anat. Histol. Embryol.  33: 236– 243. [PubMed] Google Scholar CrossRef Search ADS PubMed  Picard B. Barboiron C. Chadeyron D. Jurie C. 2007. Une technique d'électrophorèse appliquée à la séparation des isoforms de chaînes lourdes de myosine du muscle squelettique de bovin. Cah. Techn. INRA  62: 17– 24. Picard B. Barboiron C. Duris M. P. Gagniére H. Jurie C. Geay Y. 1999. Electrophoretic separation of bovine muscle myosin heavy chain isoforms. Meat Sci.  53: 1– 7. Google Scholar CrossRef Search ADS PubMed  Picard B. Cassar-Malek I. 2009. Evidence for expression of IIb myosin heavy chain isoform in some skeletal muscles of Blonde d'Aquitaine bulls. Meat Sci.  82: 30– 36. Google Scholar CrossRef Search ADS PubMed  Picard B. Lefaucheur L. Berri C. Duclos M. J. 2002. Muscle fibre ontogenesis in farm animal species. Reprod. Nutr. Dev.  42: 415– 431. [PubMed] Google Scholar CrossRef Search ADS PubMed  Picard B. Robelin J. Pons F. Geay Y. 1994. Comparison of the fetal development of fiber types in four bovine muscles. J. Muscle Res. Cell Motil.  15: 473– 486. [PubMed] Google Scholar CrossRef Search ADS PubMed  Sayd T. Mera T. Martin V. Laville E. 1998. Spatial distribution of myosin heavy chain isoforms and lactate dehydrogenase M-4 in the limb musculature of two crossbred lambs. Comp. Biochem. Physiol. B Biochem. Mol. Biol.  120: 153– 163. [PubMed] Google Scholar CrossRef Search ADS PubMed  Sazili A. Q. Parr T. Sensky P. L. Jones S. W. Bardsley R. G. Buttery P. J. 2005. The relationship between slow and fast myosin heavy chain content, calpastatin and meat tenderness in different ovine skeletal muscles. Meat Sci.  69: 17– 25. Google Scholar CrossRef Search ADS PubMed  Schefe J. H. Lehmann K. E. Buschmann I. R. Unger T. Funke-Kaiser H. 2006. Quantitative real-time RT-PCR data analysis: Current concepts and the novel “gene expression's CT difference” formula. J. Mol. Med.  84: 901– 910. [PubMed] Google Scholar CrossRef Search ADS PubMed  Serrano A. L. Pérez M. Lucía A. Chicharro J. L. Quiroz-Rothe E. Rivero J. L. L. 2001. Immunolabelling, histochemistry and in situ hybridization in human skeletal muscle fibres to detect myosin heavy chain expression at the protein and mRNA level. J. Anat.  199: 329– 337. [PubMed] Google Scholar CrossRef Search ADS PubMed  Talmadge R. J. Roy R. R. 1993. Electrophoretic separation of rat skeletal-muscle myosin heavy-chain isoforms. J. Appl. Physiol.  75: 2337– 2340. [PubMed] Google Scholar CrossRef Search ADS PubMed  Vadászová A. Hudecová S. Križanová O. Soukup T. 2006. Levels of myosin heavy chain mRNA transcripts and protein isoforms in the fast extensor digitorum longus muscle of 7-month-old rats with chronic thyroid status alterations. Physiol. Res.  55: 707– 710. [PubMed] Google Scholar PubMed  Vuocolo T. Byrne K. White J. McWilliam S. Reverter A. Cockett N. E. Tellam R. L. 2007. Identification of a gene network contributing to hypertrophy in callipyge skeletal muscle. Physiol. Genomics  28: 253– 272. [PubMed] Google Scholar CrossRef Search ADS PubMed  Wagner P. D. Giniger E. 1981. Hydrolysis of ATP and reversible binding to F-actin by myosin heavy-chains free of all light-chains. Nature  292: 560– 562. [PubMed] Google Scholar CrossRef Search ADS PubMed  Weiss A. Leinwand L. A. 1996. The mammalian myosin heavy chain gene family. Annu. Rev. Cell Dev. Biol.  12: 417– 439. [PubMed] Google Scholar CrossRef Search ADS PubMed  Zhu M. J. Ford S. P. Means W. J. Hess B. W. Nathanielsz P. W. Du M. 2006. Maternal nutrient restriction affects properties of skeletal muscle in offspring. J. Physiol.  575: 241– 250. [PubMed] Google Scholar CrossRef Search ADS PubMed  American Society of Animal Science

Shin, J.;Li, B.;Davis, M. E.;Suh, Y.;Lee, K.

doi: 10.2527/jas.2009-2124pmid: 19717768

ABSTRACT The objective of this research was to further characterize the promoter regions of the bovine and porcine fatty acid-binding protein 4 (FABP4) genes relative to those of other mammals. The DNA sequences of FABP4 promoters for the human, mouse, cow, pig, and dog were obtained from the genomic database of the National Center for Biotechnology Information and also from the sequencing of bovine and porcine genomic DNA clones obtained by 5′ PCR racing of genomic DNA. Sequence alignments of these FABP4 promoters using the basic local alignment search tool of the National Center for Biotechnology Information revealed 3 highly conserved promoter regions across the species. Two computational bioinformatics databases and the literature identified the conserved transcription factor-binding sites for C/EBP, adapter primer-1, and boxes of CAAT and TATA in the first conserved proximal promoter region, a direct repeat 1-type PPAR responsive element in the second distal conserved region, and another PPAR responsive element in the third distal conserved promoter region of FABP4 in all 5 mammals. Five new short interspersed repetitive elements (SINE) in the bovine FABP4 promoter and 2 new SINE in the porcine were found, but these SINE did not disrupt the 3 conserved regions, indicating that important regulatory elements are maintained regardless of evolutionary pressure. In conclusion, the conserved cis-acting elements, especially the 2 key adipocyte transcription factors C/EBP and PPAR, may contribute greatly to adipogenic regulation and adipose tissue-specific expression of FABP4 in these mammals. This helps to further characterize and decipher important cis-acting elements that are important for adipocyte development in adipose and muscle tissue. INTRODUCTION Fatty acid-binding protein 4 (FABP4) belongs to a family of 9 currently identified as cytoplasmic fatty acid-binding proteins. The role of FABP4 in fatty acid trafficking has been identified in animal fat tissues by showing involvement of FABP4 in efflux and influx of fatty acids in the adipocyte in response to anabolic and catabolic conditions, respectively (Bernlohr et al., 1984; Armstrong et al., 1990; Shen et al., 1999; Hertzel et al., 2006; Vural et al., 2008). Recent studies on the linkage analysis of fatness traits for FABP4 loci in the pig and cow have suggested an association between animal marbling and FABP4, suggesting that the FABP4 gene is a strong candidate gene for marbling among adipose-specific genes (Estellé et al., 2006; Wibowo et al., 2008). Understanding the regulatory role of FABP4 in lipid metabolism and regulation of FABP4 gene expression in domestic animals may contribute to improving product efficiency and quality, such as fattening or marbling. Tissue-specific promoters generally contain cis-acting elements, where tissue-specific transcription factors bind and regulate the tissue-specific expression of genes. There has previously been interest in finding fat-specific regulatory regions in both the human and mouse FABP4 promoter (Ross et al., 1990; Rival et al., 2004). The −5.4 kb of the FABP4 5′-flanking promoter is essential and sufficient to direct the adipose-specific expression of transgenes in mice (Ross et al., 1990; Graves et al., 1992) and to regulate differentiation-dependent gene expression in vitro (Christy et al., 1989; Tontonoz et al., 1994b). Furthermore, PPARγ2, a member of the PPAR family at 5.4 kb upstream of the promoter, has an important role in regulating the differentiation- or adipose tissue-specific expression of the FABP4 gene (Graves et al., 1991,1992; Tontonoz et al., 1994b). However, the bovine and porcine FABP promoters have not yet been studied. Therefore, the aims of the current study were to clone, sequence, and align the FABP4 promoter regions among mammals. In addition, the major conserved regions were identified, and the transcription-binding protein elements in these regions were further related to the regulation of FABP4 expression in these animals. MATERIALS AND METHODS Animal care and use procedures were approved by The Ohio State University Institutional Animal Care and Use Committee. Genomic DNA Isolation Porcine (Landrace; n = 1) and bovine (Angus; n = 1) genomic DNA was isolated from blood buffy coats. The white blood cell layers were harvested and digested with buffer containing 0.1 M MgCl2, 0.02 M EDTA, 0.5% SDS, 0.01 M Tris, pH 8.0, and 0.1 mg/mL of protease K at 55°C overnight. The digested cells were centrifuged at 13,200 × g for 10 min at room temperature. The DNA from the supernatant was purified by double phenol-chloroform extractions and ethanol precipitation. The genomic DNA was dissolved in Tris-EDTA buffer. The amount of genomic DNA was quantified by spectrophotometric analysis, and the quality of genomic DNA was checked for high molecular weight by 0.7% agarose gel electrophoresis. Genomic DNA Libraries To obtain the unknown sequences of the pig and bovine FABP4 promoter by 5′ PCR racing of genomic DNA, a Universal GenomeWalker kit (Clontech Laboratory Inc., Mountain View, CA) was used, with methods described previously (Siebert et al., 1995). To generate pig and bovine genomic DNA libraries containing a known sequence of adapter, 2.5 µg of genomic DNA was digested with 1 of 4 blunt-end restriction enzymes, namely, DraI, EcoRV, PvuII, or StuI. After purification of the digested genomic DNA, each digested genomic DNA was ligated with 48 bp of adapter to provide the sequence for the forward primer of PCR amplification for the gene of interest. PCR Amplification, Cloning, and Sequencing Five nanograms of each genomic DNA library containing adapter was used as a template for PCR amplification of the pig and bovine FABP4 promoters. The sense adapter primers 1 and 2 were provided from a commercial kit (Clontech Laboratory), and the gene-specific antisense primers for the porcine and bovine FABP4 promoters are listed in Table 1. The conditions for long-distance PCR in an MJ Research Thermocycler (MJ Research Inc., South San Francisco, CA) were as follows: denaturation for 15 s at 94°C, and annealing and elongation for 5 min at 68°C with 40 cycles. The primary PCR products from each library were separated in 0.8% agarose gels to visualize the bands. To confirm amplification of the primary PCR products from the specific genomic DNA, 1 μL of primary PCR product was used as the template for a nested PCR reaction using the adapter primer 2 and gene-specific primer 2. The conditions for the nested PCR were as follows: denaturation for 15 s at 94°C, and annealing and elongation for 3 min at 66°C with 30 cycles. The resulting secondary PCR products were separated on 0.8% agarose gels. The specific bands with the expected sizes were excised from the gel, and the DNA fragments in the gels were purified using a gel extraction kit (Qiagen, Valencia, CA). The purified PCR products were cloned into a pCR 2.1 vector by using a TOPO-TA cloning kit (Invitrogen, Carlsbad, CA). For the long PCR product cloned into the pCR 2.1 vector, the PCR products were subcloned into a pBluscript vector (Invitrogen) and sequenced at the Sequencing Core Facility in the Comprehensive Cancer Center at The Ohio State University. Primer sequences used for 5′ PCR racing of genomic DNA and quantitative real-time PCR to amplify sequences of bovine and porcine fatty acid-binding protein 4 (FABP4) promoters Table 1. Primer sequences used for 5′ PCR racing of genomic DNA and quantitative real-time PCR to amplify sequences of bovine and porcine fatty acid-binding protein 4 (FABP4) promoters Primer1  Sequence  Reference  mCYC-F  5′-GGT GGA GAG CAC CAA GAC AGA-3′  Li et al. (2009)  mCYC-R  5′-GCC GGA GTC GAC AAT GAT G-3′     mPPARγ-F  5′-CCG AAG AAC CAT CCG ATT GAA-3′     mPPARγ-R  5′-GCC CAA ACC TGA TGG CAT T-3′     bCYC-F  5′-GTG GTC ATC GGT CTC TTT GG-3′     bCYC-R  5′-CAC CGT AGA TGC TCT TAC CTC-3′     bPPARγ-F  5′-CTG TGA AGT TCA ACG CAC TGG A-3′     bPPARγ-R  5′-GGT TCA GCT TGA GCT GCA GCT-3′     pCYC-F  5′-GGA TAA TTT TGT GGC CTT GGC-3′  Li et al. (2007)  pCYC-R  5′-ACT GGG AGC CAT TGG TGT CT-3′     pPPARγ-F  5′-TAG ATG ACA GCG ACC TGG CGA-3′  Li et al. (2007)  pPPARγ-R  5′-AGC AGC TTA GCA AAG AGC TGG-3′     p-rR1  5′-GCA CTC TAG GAT TAT TCT TCA AGG T-3′     p-rR2  5′-CAA GGT GAG AAA GAA GCC GTA ACC TT-3′     p-rR3  5′-TGT ATT CAA CTG AGC AGT ATC TGC ATC-3′     p-rR4  5′-CAG GAC TGT TGC TTA GAG ATA TAG ACT-3′     p-rR5  5′-CAT CCT CAG CAG CAC TTG ATG TTG TC-3′     p-F6  5′-TCT ACT GGG AAT GTA AAA TAG TGC TG-3′     p-rR7  5′-TTG AGA CAT GCT AGT GTA ACT GAG AC-3′     p-rR8  5′-GTA CAG CAA GGG GAT CAA GTT ATC CT-3′     b-rR1  5′-AAT GGT TGC ATA TGT TAA CGG CAC AGT-3′     b-rR2  5′-TCA TTC AGT CTG AAC TGT TGC ATA GAC-3′     AP-F1  5′-GTA ATA CGA CTC ACT ATA GGG C-3′  Clontech2  AP-F2  5′-ACT ATA GGG CAC GCG TGG T-3′  Clontech  Primer1  Sequence  Reference  mCYC-F  5′-GGT GGA GAG CAC CAA GAC AGA-3′  Li et al. (2009)  mCYC-R  5′-GCC GGA GTC GAC AAT GAT G-3′     mPPARγ-F  5′-CCG AAG AAC CAT CCG ATT GAA-3′     mPPARγ-R  5′-GCC CAA ACC TGA TGG CAT T-3′     bCYC-F  5′-GTG GTC ATC GGT CTC TTT GG-3′     bCYC-R  5′-CAC CGT AGA TGC TCT TAC CTC-3′     bPPARγ-F  5′-CTG TGA AGT TCA ACG CAC TGG A-3′     bPPARγ-R  5′-GGT TCA GCT TGA GCT GCA GCT-3′     pCYC-F  5′-GGA TAA TTT TGT GGC CTT GGC-3′  Li et al. (2007)  pCYC-R  5′-ACT GGG AGC CAT TGG TGT CT-3′     pPPARγ-F  5′-TAG ATG ACA GCG ACC TGG CGA-3′  Li et al. (2007)  pPPARγ-R  5′-AGC AGC TTA GCA AAG AGC TGG-3′     p-rR1  5′-GCA CTC TAG GAT TAT TCT TCA AGG T-3′     p-rR2  5′-CAA GGT GAG AAA GAA GCC GTA ACC TT-3′     p-rR3  5′-TGT ATT CAA CTG AGC AGT ATC TGC ATC-3′     p-rR4  5′-CAG GAC TGT TGC TTA GAG ATA TAG ACT-3′     p-rR5  5′-CAT CCT CAG CAG CAC TTG ATG TTG TC-3′     p-F6  5′-TCT ACT GGG AAT GTA AAA TAG TGC TG-3′     p-rR7  5′-TTG AGA CAT GCT AGT GTA ACT GAG AC-3′     p-rR8  5′-GTA CAG CAA GGG GAT CAA GTT ATC CT-3′     b-rR1  5′-AAT GGT TGC ATA TGT TAA CGG CAC AGT-3′     b-rR2  5′-TCA TTC AGT CTG AAC TGT TGC ATA GAC-3′     AP-F1  5′-GTA ATA CGA CTC ACT ATA GGG C-3′  Clontech2  AP-F2  5′-ACT ATA GGG CAC GCG TGG T-3′  Clontech  1m = mouse; b = bovine; p = porcine; r = 5′ PCR racing of genomic DNA; AP = commercial adapter primer; CYC = cyclophilin; F = forward; R = reverse. 2Clontech Laboratory Inc. (Mountain View, CA). View Large Table 1. Primer sequences used for 5′ PCR racing of genomic DNA and quantitative real-time PCR to amplify sequences of bovine and porcine fatty acid-binding protein 4 (FABP4) promoters Primer1  Sequence  Reference  mCYC-F  5′-GGT GGA GAG CAC CAA GAC AGA-3′  Li et al. (2009)  mCYC-R  5′-GCC GGA GTC GAC AAT GAT G-3′     mPPARγ-F  5′-CCG AAG AAC CAT CCG ATT GAA-3′     mPPARγ-R  5′-GCC CAA ACC TGA TGG CAT T-3′     bCYC-F  5′-GTG GTC ATC GGT CTC TTT GG-3′     bCYC-R  5′-CAC CGT AGA TGC TCT TAC CTC-3′     bPPARγ-F  5′-CTG TGA AGT TCA ACG CAC TGG A-3′     bPPARγ-R  5′-GGT TCA GCT TGA GCT GCA GCT-3′     pCYC-F  5′-GGA TAA TTT TGT GGC CTT GGC-3′  Li et al. (2007)  pCYC-R  5′-ACT GGG AGC CAT TGG TGT CT-3′     pPPARγ-F  5′-TAG ATG ACA GCG ACC TGG CGA-3′  Li et al. (2007)  pPPARγ-R  5′-AGC AGC TTA GCA AAG AGC TGG-3′     p-rR1  5′-GCA CTC TAG GAT TAT TCT TCA AGG T-3′     p-rR2  5′-CAA GGT GAG AAA GAA GCC GTA ACC TT-3′     p-rR3  5′-TGT ATT CAA CTG AGC AGT ATC TGC ATC-3′     p-rR4  5′-CAG GAC TGT TGC TTA GAG ATA TAG ACT-3′     p-rR5  5′-CAT CCT CAG CAG CAC TTG ATG TTG TC-3′     p-F6  5′-TCT ACT GGG AAT GTA AAA TAG TGC TG-3′     p-rR7  5′-TTG AGA CAT GCT AGT GTA ACT GAG AC-3′     p-rR8  5′-GTA CAG CAA GGG GAT CAA GTT ATC CT-3′     b-rR1  5′-AAT GGT TGC ATA TGT TAA CGG CAC AGT-3′     b-rR2  5′-TCA TTC AGT CTG AAC TGT TGC ATA GAC-3′     AP-F1  5′-GTA ATA CGA CTC ACT ATA GGG C-3′  Clontech2  AP-F2  5′-ACT ATA GGG CAC GCG TGG T-3′  Clontech  Primer1  Sequence  Reference  mCYC-F  5′-GGT GGA GAG CAC CAA GAC AGA-3′  Li et al. (2009)  mCYC-R  5′-GCC GGA GTC GAC AAT GAT G-3′     mPPARγ-F  5′-CCG AAG AAC CAT CCG ATT GAA-3′     mPPARγ-R  5′-GCC CAA ACC TGA TGG CAT T-3′     bCYC-F  5′-GTG GTC ATC GGT CTC TTT GG-3′     bCYC-R  5′-CAC CGT AGA TGC TCT TAC CTC-3′     bPPARγ-F  5′-CTG TGA AGT TCA ACG CAC TGG A-3′     bPPARγ-R  5′-GGT TCA GCT TGA GCT GCA GCT-3′     pCYC-F  5′-GGA TAA TTT TGT GGC CTT GGC-3′  Li et al. (2007)  pCYC-R  5′-ACT GGG AGC CAT TGG TGT CT-3′     pPPARγ-F  5′-TAG ATG ACA GCG ACC TGG CGA-3′  Li et al. (2007)  pPPARγ-R  5′-AGC AGC TTA GCA AAG AGC TGG-3′     p-rR1  5′-GCA CTC TAG GAT TAT TCT TCA AGG T-3′     p-rR2  5′-CAA GGT GAG AAA GAA GCC GTA ACC TT-3′     p-rR3  5′-TGT ATT CAA CTG AGC AGT ATC TGC ATC-3′     p-rR4  5′-CAG GAC TGT TGC TTA GAG ATA TAG ACT-3′     p-rR5  5′-CAT CCT CAG CAG CAC TTG ATG TTG TC-3′     p-F6  5′-TCT ACT GGG AAT GTA AAA TAG TGC TG-3′     p-rR7  5′-TTG AGA CAT GCT AGT GTA ACT GAG AC-3′     p-rR8  5′-GTA CAG CAA GGG GAT CAA GTT ATC CT-3′     b-rR1  5′-AAT GGT TGC ATA TGT TAA CGG CAC AGT-3′     b-rR2  5′-TCA TTC AGT CTG AAC TGT TGC ATA GAC-3′     AP-F1  5′-GTA ATA CGA CTC ACT ATA GGG C-3′  Clontech2  AP-F2  5′-ACT ATA GGG CAC GCG TGG T-3′  Clontech  1m = mouse; b = bovine; p = porcine; r = 5′ PCR racing of genomic DNA; AP = commercial adapter primer; CYC = cyclophilin; F = forward; R = reverse. 2Clontech Laboratory Inc. (Mountain View, CA). View Large Bioinformatics and Sequence Analysis Bioinformatics and sequence analysis of mammal FABP4 promoters were performed as described previously (Deiuliis et al., 2008; Shin et al., 2008; Lee et al., 2009). A sequence scanner (version 1.0) was used to generate a chromatogram of cloned bovine and porcine FABP4 sequences [approximately −1,189 to −4,519 bp (FJ884068), and approximately −1 to −4,748 bp (FJ884069), individually] by using an Applied Biosystems (Foster City, CA) genetic analyzer instrument. In addition, long distal promoter sequences of FABP4 in the mouse (approximately −1 to −5,545; NT 078380), human (approximately −1 to −9,600; NT 008183), dog (approximately −1 to −13,494; NW 876284), cow (approximately −1 to −1,188 and −4,520 to −10,730; NW 001493222), and pig (approximately −4,749 to −9,582; NW 001886208) were obtained from the 5′-upstream region of individual mammal FABP4 shown in the genome database at the National Center for Biotechnology Information (NCBI). Homology analysis was performed using align 2 sequences in the basic local alignment search tool (BLAST) at NCBI (Tatusova and Madden, 1999). The parameters for the Nucleotide BLAST program were described as follows: reward for a match (1); penalty for a mismatch (−2); penalties for open gap (5) and extension gap (2); gap × dropoff 50 expect 10.0 ward size 11 filter. Next, alignment and comparison of the promoter sequences among the 5 mammals were carried out using the NCBI ClustalX (version 2.0) and GeneDoc software (version 2.7). To predict putative elements of transcription-binding factors on FABP4 promoters, 2 databases from TFSEARCH software (Searching Transcription Factor Binding Sites, version 1.3; http://www.cbrc.jp/research/db/TFSEARCH.html) and from TRANSFAC software (version 6.0) of the Transcription Element Search System (TESS; http://www.cbil.upenn.edu/cgi-bin/tess/tess) were used. Western Blot Western blot analysis was performed as described previously (Li et al., 2009). Briefly, tissue protein samples (white adipose, heart, muscle, spleen, lung, liver, and kidney) were isolated from the mouse (FVB; n = 4; 70 d old), cow (Angus; n = 4; 450 to 600 kg), and pig (Landrace; n = 4; 100 to 120 kg). Frozen tissue samples were lysed in 60 mM Tris-HCl (pH 6.8), 1% SDS, and 1:100 dilution of proteinase inhibitor cocktail (Sigma, St. Louis, MO). The protein lysates were centrifuged at 12,000 × g for 3 min at 4°C. Equal amounts of each protein lysate were mixed with Laemmli buffer (Bio-Rad, Hercules, CA) and boiled for 2 min at 95°C before a discontinuous SDS-PAGE was used. After electrophoresis, the proteins were transferred to polyvinylidene fluoride membranes (Amersham Biosciences, Piscataway, NJ), and the membranes were incubated in blocking buffer [150 mM NaCl, 10 mM Tris-HCl, pH 8.0, and 0.05% Tween 20 (TBST) with 5% nonfat dry milk]. The membranes were probed with a goat antibody against FABP4 (R&D systems, Minneapolis, MN) and β-actin (Cell Signaling Technology, Danvers, MA) at 1:2,000, and with a mouse antibody against α-tublin at 1:3,000 (The Developmental Studies Hybridoma Bank, Iowa City, IA) in TBST with 5% nonfat dry milk overnight at 4°C. Membranes were washed 6 times with TBST and incubated for 1 h with horseradish peroxidase-conjugated donkey anti-goat at 1:3,000 (Bio-Rad) and horseradish peroxidase-conjugated goat anti-mouse at 1:5,000 (Cell Signaling Technology). After washing 5 times in TBST, the proteins were detected with an enhanced chemiluminescence system (ECL Plus, GE Healthcare Life Sciences, Piscataway, NJ) and visualized on Hyperfilm (GE Healthcare Life Sciences). Total RNA Isolation and Quantitative Real Time-PCR Adipose tissue, heart, muscle, spleen, lung, liver, and kidney of the mouse (FVB; n = 4), cow (Angus; n = 4), and pig (Landrace; n = 4) were harvested after slaughter of animals at The Ohio State University Meat Science Laboratory, located in the Department of Animal Sciences Building (Columbus). Tissues were immediately snap-frozen in liquid nitrogen and homogenized using a Tissuemiser homogenizer (Fisher Scientific, Pittsburgh, PA). Total RNA isolation and quantitative real time-PCR (qRT-PCR) were performed as described previously (Shin et al., 2008). Total RNA from the tissue was isolated using Trizol (Invitrogen) following the instructions of the manufacturer, and RNA quality was assessed by electrophoresis (1% agarose gel). Reverse transcription-PCR was performed using 1 μg of total RNA and Moloney murine leukemia virus reverse transcriptase (Invitrogen). Reverse transcription conditions for cDNA amplification were 65°C for 5 min, 37°C for 52 min, and 70°C for 15 min. The primer sequences used for qRT-PCR are shown in Table 1. Quantitative real time-PCR using murine, porcine, and bovine PPARγ-forward and PPARγ-reverse primer sets was performed to measure the relative levels of PPARγ gene expression in various tissues. The murine, porcine, and bovine cyclophilin genes served as housekeeping genes. All primers for qRT-PCR that span the genomic introns for qRT-PCR could avoid amplification of contaminated genomic DNA during the PCR reactions. Quantitative real time-PCR was performed using AmpliTaq Gold polymerase (Applied Biosystems), and SYBR Green was used as the detection dye. Conditions for qRT-PCR were 95°C for 10 min and 40 cycles of 94°C for 15 s, 60°C for 40 s, 72°C for 30 s, and 82°C for 32 s. Quantitative real-time PCR was performed in duplicate on 25-μL reactions using an ABI 7300 real-time PCR instrument (Applied Biosystems). The relative level of target gene expression, as determined by ABI software (Applied Biosystems), was calculated using the comparative 2−ΔΔCt method for relative quantification (Livak and Schmittgen, 2001). Statistical Analysis Results are presented as means ± SEM. Comparisons of PPARγ messenger RNA (mRNA) expression among various tissues of the mouse, cow, and pig were performed by one-way ANOVA, followed by the Tukey test at P ≤ 0.05. All statistical analyses were performed with Minitab software (version 15.0; Minitab Inc., State College, PA). RESULTS AND DISCUSSION In general, the lipids in adipocytes originate from either dietary fat or de novo synthesis of fatty acids in adipose and liver, depending on the species. The hydrophobic fatty acids inside adipocytes are chaperoned by FABP4 (Coe and Bernlohr, 1998). Adipose tissue-specific gene expression of FABP4 was demonstrated in mammalian species (Lee et al., 2003; Stejskal and Karpisek, 2006; Li et al., 2007; Soliman et al., 2007). However, the tissue-specific expression of porcine and bovine FABP4 at the protein level has not been reported. In the current study, Western blot analysis of FABP4 in adipose, heart, muscle, spleen, lung, liver, and kidney tissue showed the predominant expression of FABP4 protein in adipose tissue of the mouse, cow, and pig (Figure 1). These results led us to perform a comparative analysis of FABP4 promoters to identify conserved regulatory elements in the promoters regulating expression of FABP4 in these species. Generally, the highly conserved nucleotide sequences in promoter regions are more likely to be functionally important to regulate transcription of a gene in a similar manner in different species (Dieterich et al., 2005). Figure 1. View largeDownload slide Western blot analysis represents tissue distribution of fatty acid-binding protein 4 (FABP4) protein expression for the mouse, cow, and pig. Beta-actin was used as a reference for FABP4 protein in the mouse, whereas α-tublin was used as a reference for FABP4 protein expression of the cow and pig. WF = white adipose tissue; Hrt = heart; Mus = muscle; Spl = spleen; Lu = lung; Li = liver; Kid = kidney. Figure 1. View largeDownload slide Western blot analysis represents tissue distribution of fatty acid-binding protein 4 (FABP4) protein expression for the mouse, cow, and pig. Beta-actin was used as a reference for FABP4 protein in the mouse, whereas α-tublin was used as a reference for FABP4 protein expression of the cow and pig. WF = white adipose tissue; Hrt = heart; Mus = muscle; Spl = spleen; Lu = lung; Li = liver; Kid = kidney. Expression of the FABP4 gene is extremely low in preadipocytes but is dramatically induced in adipocytes during differentiation in the mouse and pig (Chmurzyńska, 2006; Li et al., 2007, 2009; Deiuliis et al., 2008), and is referred to as being “adipocyte specific.” In addition, FABP4 mRNA is expressed exclusively in the adipose tissues of most mammalian species, including the human, rat, mouse, and pig (Bernlohr et al., 1985; Stejskal and Karpisek, 2006; Li et al., 2007), and is referred to as being “adipose tissue specific.” For these reasons, the FABP4 gene has been widely used as an adipose and adipocyte marker gene in studies on obesity and adipocyte development in humans, mice, and food animals (Chavey et al., 2006; Li et al., 2007, 2009). For the present comparative analysis, the FABP4 promoter sequences for the human, mouse, cow, pig, and dog available from NCBI were downloaded. However, the relatively large sequences for bovine and porcine FABP4 promoters comparable with the −5.4 kb of the mouse FABP4 promoter were not available in the NCBI genome database. Therefore, 5′ PCR racing of genomic DNA for the cow and pig was performed and several clones were obtained, covering the nucleotide sequences (total of 3,331 bp, located at −4,519 to −1,189 bp) of the bovine 5′-flanking FABP4 promoter and sequences (total of 4,748 bp, located at −4,748 to −1 bp) of the porcine FABP4 promoter (Figure 2A). These sequences were reported in GenBank [bovine (FJ884068) and porcine (FJ884069)]. Recently, the NCBI was updated to include the sequences for bovine and porcine FABP4 promoters. The promoter sequences we obtained by 5′ racing of genomic DNA are almost identical to the sequences reported by the NCBI (Figure 2). In addition, our promoter sequences were able to fill the 100-bp gap of missing nucleotide sequences (NNNN…..NNNN) in porcine FABP4 promoter regions at −830 to −731 and the 355-bp gap of missing sequences in bovine FABP4 promoter regions at −1,730 to −1,376 that were reported in GenBank [bovine (FJ884068) and porcine (FJ884069)]. As shown in Figure 3, the sequences of FABP4 promoters for the 5 mammalian species were available for alignment by using BLAST at NCBI. Figure 2. View largeDownload slide Cloning of porcine and bovine nucleotides of fatty acid-binding protein 4 (FABP4) promoters, alignment, and schematic comparison of the sequence homology of mouse FABP4 promoter with other mammal FABP4 promoter regions. (A) Cloning of porcine and bovine nucleotides of FABP4 promoters by 5′ PCR racing of individual genomic DNA. Restriction enzymes are DraI, EcoRV, and StuI. The arrows indicate the 5′ racing direction for cloning of bovine and porcine FABP4 promoters in 5′ flanking regions. Individual primers for cloning the bovine and porcine promoter regions are shown in Table 1. (B) The murine FABP4 promoter sequence (approximately −1 to 5,545 bp) was aligned with that of the human (approximately −1 to 9,592 bp), cow (approximately −1 to 10,609 bp), pig (approximately −1 to 9,582 bp), and dog (approximately −1 to 13,494 bp). The small boxes indicate conserved regions of FABP4 promoter among species. (C) Schematic representation of highly conserved FABP4 promoter regions between the mouse and human, cow, dog, and pig, individually. Seven numbered boxes indicate putative short interspersed nucleotide elements in the promoter regions of the cow and pig. Figure 2. View largeDownload slide Cloning of porcine and bovine nucleotides of fatty acid-binding protein 4 (FABP4) promoters, alignment, and schematic comparison of the sequence homology of mouse FABP4 promoter with other mammal FABP4 promoter regions. (A) Cloning of porcine and bovine nucleotides of FABP4 promoters by 5′ PCR racing of individual genomic DNA. Restriction enzymes are DraI, EcoRV, and StuI. The arrows indicate the 5′ racing direction for cloning of bovine and porcine FABP4 promoters in 5′ flanking regions. Individual primers for cloning the bovine and porcine promoter regions are shown in Table 1. (B) The murine FABP4 promoter sequence (approximately −1 to 5,545 bp) was aligned with that of the human (approximately −1 to 9,592 bp), cow (approximately −1 to 10,609 bp), pig (approximately −1 to 9,582 bp), and dog (approximately −1 to 13,494 bp). The small boxes indicate conserved regions of FABP4 promoter among species. (C) Schematic representation of highly conserved FABP4 promoter regions between the mouse and human, cow, dog, and pig, individually. Seven numbered boxes indicate putative short interspersed nucleotide elements in the promoter regions of the cow and pig. Figure 3. View largeDownload slide Sequences and multialignment of the first conserved proximal fatty acid-binding protein 4 (FABP4) promoters of 5 mammals, and putative or reported transcription binding sites on their FABP4 promoters. The abbreviations under dotted empty boxes indicate the myocyte-specific enhancer-binding factor DNA binding site (MEF-2); C/EBPα, C/EBPβ, or C/EBPδ; activation protein-1 (AP-1); fat-specific enhancer 1 (FSE1); CAAT box; and TATA box. The black arrow under the ATG codons indicates the start sites of transcription of FABP4 in all 5 mammals. Figure 3. View largeDownload slide Sequences and multialignment of the first conserved proximal fatty acid-binding protein 4 (FABP4) promoters of 5 mammals, and putative or reported transcription binding sites on their FABP4 promoters. The abbreviations under dotted empty boxes indicate the myocyte-specific enhancer-binding factor DNA binding site (MEF-2); C/EBPα, C/EBPβ, or C/EBPδ; activation protein-1 (AP-1); fat-specific enhancer 1 (FSE1); CAAT box; and TATA box. The black arrow under the ATG codons indicates the start sites of transcription of FABP4 in all 5 mammals. To perform the alignment of the 5 FABP4 promoter sequences and to identify highly homologous regions of the FABP4 promoter among mammalian species, sequences of mouse FABP4 promoters were chosen to compare with those of the human, bovine, pig, and dog individually because the mouse FABP4 promoter was studied previously. In addition, all possible cross-comparisons with all 5 species were made to locate evolutionarily conserved regions of the promoters. As shown in Figure 2B, nucleotide alignment of all FABP4 promoters revealed a total of 5 conserved promoter regions. Among them, the second region was specifically conserved between the mouse and cow, and the fourth region was conserved among the mouse, pig, and dog. However, the first, third, and fifth regions were conserved in all species, indicating that these conserved regions may contain important cis-regulatory elements regulating FABP4 gene expression in animals. To find the important regulatory elements among the 5 species, 2 computational bioinformatics databases (TFSEARCH and TRANSFAC software) were used. Studies relative to FABP4 and its promoter were also widely reviewed. As shown in Figure 3, five major transcription factor-binding elements—C/EBP, activation protein-1, fat-specific enhancer 1 (FSE1), the CAAT box, and the TATA box—were found in the first conserved region of the FABP4 promoters. Gene expression of C/EBPα is positively associated with adipocyte cell differentiation in the mouse (Christy et al., 1989), cow (Ohsaki et al., 2007), pig (Lee et al., 1998), and human (Chiche et al., 2009). The C/EBP are master transcription factors that not only play a regulatory role in the differentiation of preadipocytes into adipocytes (Birkenmeier et al., 1989), but also induce adipogenic genes, including FABP4 (MacDougald and Lane, 1995). Mouse models with knockout of the C/EBPα gene and overexpression of the dominant negative form of C/EBPα in adipose tissue have shown a complete loss of fat accumulation in adipose tissue, further supporting the importance of C/EBPα in adipose development (Wang et al., 1995; Moitra et al., 1998). In humans, an SNP was found and reported at −89 (T to C) of the FABP4 promoter region (Tuncman et al., 2006), resulting in reduced binding activity of C/EBPα to the element of the promoter and, consequently, reduced transcription activity from the FABP4 promoter. Interestingly, this C/EBPα binding site (CCAAAGTTGAGAAATTT), located between 140 and 155 bp of the first conserved region, is 100% homologous in all 5 species, further supporting the importance of conservation of the C/EBPα binding site for the regulation of FABP4 expression. Indeed, a mutation in the C/EBP binding site of the murine FABP promoter region decreased CAT reporter gene expression during preadipocyte differentiation (Herrera et al., 1989). However, the proximal FABP4 promoter (−168 to 21 bp) containing the C/EBP binding site showed ubiquitous expression of the CAT reporter in transgenic mice (Ross et al., 1990), indicating that the C/EBP element is not sufficient for adipose-specific expression of FABP4. In support of this, a search of GEO Profiles in the NCBI database for the human and mouse (http://www.ncbi.nlm.nih.gov/geo/; GDS181 for the human and GDS182 for the mouse) revealed that C/EBPα is expressed in many tissues, including adipose, liver, intestine, lung, and muscle tissue (Su et al., 2002). Taken together, conservation of the C/EBP element in FABP4 promoters has an important role in the upregulation of FABP4 expression during adipocyte differentiation but contributes less to adipose tissue-specific expression of FABP4. The putative cis-acting elements in the third conserved distal promoter region were 2 PPAR binding elements (PPRE), the C/EBP binding element, nuclear factor-conserved lympokine element 0a, heat shock factor, and pituitary-specific transcription factor 1a (Figure 4). Among them, it is worth focusing on and discussing the conserved PPRE sites to predict the role of FABP4 promoters for the pig, cow, and dog because 1) the PPRE is known to be an important cis-acting element of the mouse and human FABP4 for its tissue-specific expression, and 2) PPARγ promotes expression of the adipogenic genes that are involved in lipid metabolism in their adipose tissues. A meta-analysis of PPRE sites (Lemay and Hwang, 2006; Heinäniemi et al., 2007) showed a consensus sequence (AGGTCAAAGGTCA) of direct repeat 1 (DR1)-type PPRE. In the present study, the consensus sequence of the DR1-type PPRE was highly conserved in all 5 mammalian species. Compared with the mouse sequences, 83% homology was shown in the human sequences, 92% homology was shown in the cow, and 100% homology was shown in the pig and dog. Furthermore, many other adipogenic genes, such as perilipin (Shimizu et al., 2006), fatty acid transporter proteins (Frohnert et al., 1999), human peroxisomal fatty acyl-CoA oxidase (Varanasi et al., 1996), lipoprotein lipase (Schoonjans et al., 1996), phosphoenolpyruvate carboxykinase (Tontonoz et al., 1995), acyl-CoA synthetase (Schoonjans et al., 1995), and malic enzyme (Castelein et al., 1994), contain the consensus sequence of DR1-type PPRE. This strongly suggests that the consensus PPRE may be one of the key elements for adipose tissue-specific expression of FABP4 in the pig, cow, and dog. Figure 4. View largeDownload slide Sequences and multialignment of the third distal conserved fatty acid-binding protein 4 (FABP4) promoters of 5 mammals, and putative or reported transcription factor binding sites on their FABP4 promoters. The abbreviations under dotted empty boxes indicate the protein binding sites of nuclear factor-conserved lympokine element 0a (NF-CLE0a); heat shock factor (HSF); pituitary-specific transcription factor 1a (POU1F1a); C/EBPα; and direct repeat 1-type PPAR binding element (DR1-PPRE). The black arrow under DR1-PPRE indicates the binding direction of their binding proteins. Figure 4. View largeDownload slide Sequences and multialignment of the third distal conserved fatty acid-binding protein 4 (FABP4) promoters of 5 mammals, and putative or reported transcription factor binding sites on their FABP4 promoters. The abbreviations under dotted empty boxes indicate the protein binding sites of nuclear factor-conserved lympokine element 0a (NF-CLE0a); heat shock factor (HSF); pituitary-specific transcription factor 1a (POU1F1a); C/EBPα; and direct repeat 1-type PPAR binding element (DR1-PPRE). The black arrow under DR1-PPRE indicates the binding direction of their binding proteins. Analysis of the fifth distal conserved promoter region of FABP4 revealed several conserved cis-regulatory elements, including adipocyte regulatory element (ARE)6, ARE7, T (thymus) cell-specific factor-2α, ARE2, globin transcription factor-1, and ARE4 (Figure 5). Among them, ARE7, a PPAR binding site, is an important cis-regulatory element for adipose tissue-specific expression of murine FABP4 (Graves et al., 1992; Tontonoz et al., 1994a,b). Here, conservation of the ARE7 (TGAACTCTGATCC) site indicates that the PPRE sites may contribute to the adipose tissue-specific regulation of FABP4 in the human, pig, cow, and dog. Figure 5. View largeDownload slide Sequences and multialignment of the fifth distal conserved fatty acid-binding protein 4 (FABP4) promoters of 5 mammals, and putative or reported transcription binding sites on their FABP4 promoters. The abbreviations under dotted empty boxes indicate the transcription factor binding sites of T (thymus) cell-specific factor-2α (TCF-2α; Ho et al., 1990) and globin transcription factor-1 (GATA-1). TBF-2a is the transcription factor binding site of TCF-2α. A fat-specific enhancer bearing 2 response elements was shown as adipocyte regulatory element (ARE)6, ARE2, and ARE7. The black arrows under ARE sites indicate the binding direction of their binding proteins. Figure 5. View largeDownload slide Sequences and multialignment of the fifth distal conserved fatty acid-binding protein 4 (FABP4) promoters of 5 mammals, and putative or reported transcription binding sites on their FABP4 promoters. The abbreviations under dotted empty boxes indicate the transcription factor binding sites of T (thymus) cell-specific factor-2α (TCF-2α; Ho et al., 1990) and globin transcription factor-1 (GATA-1). TBF-2a is the transcription factor binding site of TCF-2α. A fat-specific enhancer bearing 2 response elements was shown as adipocyte regulatory element (ARE)6, ARE2, and ARE7. The black arrows under ARE sites indicate the binding direction of their binding proteins. To investigate the relationship between PPARγ expression and the predominant expression of FABP4 in fat tissue, qRT-PCR for PPARγ was performed in various tissues of the mouse, cow, and pig (Figure 6). As expected, the gene expression of PPARγ was significantly predominant in fat tissue of the mouse (P < 0.05) when compared with the PPARγ gene expression in other tissues. Likewise, the porcine and bovine PPARγ mRNA were abundantly expressed in white adipose tissue (P < 0.05). This suggests that porcine and bovine PPARγ may contribute to adipose tissue-specific expression of FABP4 in the pig and cow. The positive association of FABP4 expression with PPARγ expression during adipocyte differentiation in the pig (Ding et al., 1999; Samulin et al., 2008) and cow (Soliman et al., 2007) and induction of FABP4 expression by treatments with PPARγ agonists (Hausman et al., 2008) further supported the involvement of PPARγ in regulation of FABP4 expression. Figure 6. View largeDownload slide The gene expression patterns of PPARγ in various tissues of the mouse (FVB), cow (Angus), and pig (Landrace). (A) Predominant messenger RNA (mRNA) expression of PPARγ in mouse adipose tissue (n = 4). (B) Abundant mRNA expression of PPARγ in bovine adipose, spleen, and lung tissues (n = 4). (C) Abundant mRNA expression of PPARγ in porcine adipose, spleen, and lung tissue (n = 4). Quantitative real-time PCR was performed to show the relative gene expression. Cyclophilin (CYC) was used as a reference for PPARγ gene expression in the mouse, cow, and pig. The bars indicate relative values of gene expression, representing means and SEM. Letters above the bars (a,b) indicate statistical significance of gene expression among the various tissues by one-way ANOVA. WF = white adipose tissue; Hrt = heart; Mus = muscle; Spl = spleen; Lu = lung; Li = liver; Kid = kidney. Figure 6. View largeDownload slide The gene expression patterns of PPARγ in various tissues of the mouse (FVB), cow (Angus), and pig (Landrace). (A) Predominant messenger RNA (mRNA) expression of PPARγ in mouse adipose tissue (n = 4). (B) Abundant mRNA expression of PPARγ in bovine adipose, spleen, and lung tissues (n = 4). (C) Abundant mRNA expression of PPARγ in porcine adipose, spleen, and lung tissue (n = 4). Quantitative real-time PCR was performed to show the relative gene expression. Cyclophilin (CYC) was used as a reference for PPARγ gene expression in the mouse, cow, and pig. The bars indicate relative values of gene expression, representing means and SEM. Letters above the bars (a,b) indicate statistical significance of gene expression among the various tissues by one-way ANOVA. WF = white adipose tissue; Hrt = heart; Mus = muscle; Spl = spleen; Lu = lung; Li = liver; Kid = kidney. In the human, mouse, pig, and cow, PPARγ is produced in 2 isoforms, PPARγ1 and PPARγ2, by alternative promoter use and alternative splicing (Tontonoz et al., 1994a,b; Elbrecht et al., 1996; Sundvold et al., 1997; Houseknecht et al., 1998). Quantitative real time-PCR analysis for total PPARγ mRNA expression showed considerably greater PPARγ expression in the spleen and lung tissue in the pig and cow, which is in agreement with previous findings that the gene expression of PPARγ is abundant in the adipose tissue, spleen, and lung tissue of the pig and cow (Sundvold et al., 1997; Houseknecht et al., 1998). However, PPARγ2 is selectively expressed in adipose tissue but is very low in the spleen and lung, in which PPARγ1 is a dominant form in the pig and cow (Sundvold et al., 1997; Houseknecht et al., 1998). Studies on the functional role of the PPARγ isoforms have revealed that knockdown of both PPARγ1 and PPARγ2 expression in 3T3-L1 preadipocytes completely inhibits adipogenesis and FABP4 expression (Ren et al., 2002), suggesting an important role of PPARγ in induction of FABP4 expression. In addition, exogenously expressed PPARγ2 in these cells restored adipogenesis, but not by PPARγ1, indicating an adipogenic capacity of PPARγ2. Furthermore, the transcription activation capacity of PPARγ1 is much less than that of PPARγ2 (Werman et al., 1997). The adipose tissue-specific protein expression of FABP4 in the pig and cow (Figure 1), even though total PPARγ mRNA expression was relatively high in the spleen and lung (Figure 6B and 6C), may be largely due to the predominant expression of PPARγ2 in adipose tissue. Other possible explanations include 1) different modulation of chromatin structures to expose the promoter region of FABP4, 2) the cooperative regulation of the FABP4 promoter by factors binding to PPARγ, such as a coactivator (PPARγ coactivator 1 alpha, PGC-1α; Hondares et al., 2006) and a repression factor (chicken ovalbumin upstream promoter-transcription factor 1, COUP-TF1; Brodie et al., 1996; Tsai and Tsai, 1997; Brandebourg and Hu, 2005), and 3) different levels of PPAR ligands among tissues. Previous studies using transgenic mice, expressing the CAT reporter gene under the control of different sizes of mouse FABP4 promoter, revealed that −5.4 kb of the FABP promoter, including ARE6 and ARE7 at the fifth conserved region, was sufficient for adipose tissue-specific expression of the FABP4 gene. In support of this, adipose tissue-specific expression of target genes in numerous transgenic mice models was achieved using −5.4 kb of mouse FABP promoter (Lee et al., 2003; Takasawa et al., 2008; Wang et al., 2008). This likely indicates that murine ARE6 and ARE7 could be the main cis-acting elements regulating the predominant adipose tissue-specific expression of FABP4. In addition, the PPRE site at the third conserved region in the human increased the activity of the luciferase reporter gene by treatments of PPAR agonists such as rosiglitazone (Wurch et al., 2002; Rival et al., 2004). This indicates that the consensus DR1-type PPRE of the human FABP4 promoter that is located in the third conserved distal area (approximately −5.2 to −5.1 kb) may play an additional role in adipose tissue-specific expression of FABP4. The positive associated expression of FABP4 with PPARγ during adipocyte differentiation in the pig (Ding et al., 1999; Li et al., 2007; Samulin et al., 2008) and cow (Soliman et al., 2007; Taniguchi et al., 2008) has been identified. In this regard, 2 conserved PPAR binding sites, the ARE6 and ARE7 elements at the fifth conserved region and the DR1-type PPRE at the third conserved region, may be important in regulating the FABP4 gene in these species, although it has not been determined which site of the PPRE is required for adipose tissue-specific expression of porcine and bovine FABP4. Alignment of the bovine and porcine FABP4 promoters with the FABP4 promoters of other species revealed not only the conserved homologous area, but also a diverse area including possible insertion sequences in the promoter area. As shown in Figure 2C, a BLAST search of these sequences against the bovine genome database at NCBI revealed that 5 different regions, located at −2,724 to −1,666 bp (total of 1,058 bp), −4,578 to −4056 bp (521 bp), −5,807 to −5,388 bp (420 bp), −6,196 to −5,908 bp (378 bp), and −7,929 to −7,725 bp (205 bp), are found repetitively throughout the entire bovine genome. The first insertion (1,058 bp, located at −2,724 to −1,666 bp of the bovine FABP4 promoter region) was shown to distribute 3,308 BLAST hits on the query sequence in the entire bovine genome. Likewise, the second insertion was shown to distribute 14,335 BLAST hits, the third distributed 14,838 BLAST hits, the fourth was 14,968 BLAST hits, and the fifth was 11,876 BLAST hits. In the porcine FABP4 promoter, 2 new repetitive sequences were found at −1,542 to −1,314 bp (285 bp, with 2,243 BLAST hits) and −4,678 to −3,437 bp (1,242 bp, with 12,563 BLAST hits; Figure 2C). In general, a considerable number of animal genomes contain long interspersed repetitive elements (Girardot et al., 2006) and short interspersed repetitive elements (SINE; DeCerbo and Carmichael, 2005). Because SINE should have at least 104 to 105 copies per genome (Sheikh et al., 2002), the 7 regions of repeated DNA sequences in bovine and pig FABP4 promoters can be considered as unreported new SINE. The insertion of these SINE resulted in relatively long promoters, as shown by the location of the conserved areas at the far upstream region of the bovine and porcine FABP4 promoters, compared with the mouse FABP4 promoter. Like Alu sequences in the human, these repetitive sequences may stabilize genome sequences or may inversely allow recombination of genomes for evolutionary diversity. In addition, a BLAST search of these insertion sequences against the genome databases of other species did not reveal homologous sequences, indicating that these sequences are unique to the bovine and porcine genomes. However, insertion of these SINE did not disrupt the conserved homologous sequences in the FABP4 promoters, suggesting the conservation of important cis-regulatory elements for the regulation of FABP4 expression across the animal species examined. In the current study, we cloned porcine and bovine sequences of the FABP4 5′-flanking promoter regions, finding missing sequences shown in the BLAST genome database at NCBI. The comparative analysis of FABP4 promoters of all 5 mammals examined revealed 3 highly conserved promoter regions (first proximal, third distal, and fifth distal). Each conserved promoter region contains differentiation regulatory elements, hormone regulatory elements, and fat-specific regulatory elements. Among them, a conserved PPAR binding element such as DR1-type PPRE in the third distal promoter region of FABP4 that may govern fat-specific expression of FABP4 was found unexpectedly across the 5 mammals. In addition, ARE7, a necessary PPAR binding element for fat-specific expression of FABP4 in the mouse, was highly conserved in the fifth distal promoter region of FABP4 among the 5 species. This suggests that PPAR binding sites may also play an important role in adipose-specific expression in these species. This study provides a new bioinformatics approach for comparative analysis of promoters and to identify putative conserved transcription factor binding sites. LITERATURE CITED Armstrong M. K. Bernlohr D. A. Storch J. Clarke S. D. 1990. The purification and characterization of a fatty acid binding protein specific to pig (Sus domesticus) adipose tissue. Biochem. J.  267: 373– 378. [PubMed] Google Scholar CrossRef Search ADS PubMed  Bernlohr D. A. Angus C. W. Lane M. D. Bolanowski M. A. Kelly T. J.Jr. 1984. Expression of specific mRNAs during adipose differentiation: Identification of an mRNA encoding a homologue of myelin p2 protein. Proc. Natl. Acad. Sci. USA  81: 5468– 5472. [PubMed] Google Scholar CrossRef Search ADS   Bernlohr D. A. Bolanowski M. A. Kelly T. J.Jr. Lane M. D. 1985. Evidence for an increase in transcription of specific mRNAs during differentiation of 3T3-L1 preadipocytes. J. Biol. Chem.  260: 5563– 5567. [PubMed] Google Scholar PubMed  Birkenmeier E. H. Gwynn B. Howard S. Jerry J. Gordon J. I. Landschulz W. H. McKnight S. L. 1989. Tissue-specific expression, developmental regulation, and genetic mapping of the gene encoding CCAAT/enhancer binding protein. Genes Dev.  3: 1146– 1156. [PubMed] Google Scholar CrossRef Search ADS PubMed  Brandebourg T. D. Hu C. Y. 2005. Regulation of differentiating pig preadipocytes by retinoic acid. J. Anim. Sci.  83: 98– 107. [PubMed] Google Scholar CrossRef Search ADS PubMed  Brodie A. E. Manning V. A. Hu C. Y. 1996. Inhibitors of preadipocyte differentiation induce COUP-TF binding to a PPAR/RXR binding sequence. Biochem. Biophys. Res. Commun.  228: 655– 661. [PubMed] Google Scholar CrossRef Search ADS PubMed  Castelein H. Gulick T. Declercq P. E. Mannaerts G. P. Moore D. D. Baes M. I. 1994. The peroxisome proliferator activated receptor regulates malic enzyme gene expression. J. Biol. Chem.  28: 26754– 26758. Chavey C. Boucher J. Monthouël-Kartmann M. N. Sage E. H. Castan-Laurell I. Valet P. Tartare-Dekert S. Van Obberghen E. 2006. Regulation of secreted protein acidic and rich in cysteine during adipose conversion and adipose tissue hyperplasia. Obesity (Silver Spring)  14: 1890– 1897. [PubMed] Google Scholar CrossRef Search ADS PubMed  Chiche F. Le Guillou M. Chetrite G. Lasnier F. Dugail I. Carpene C. Moldes M. Feve B. 2009. The antidepressant phenelzine alters differentiation of cultured human and mouse preadipocytes. Mol. Pharmacol.  https://doi.org/10.1124/mol.108.052563. [PubMed] Chmurzyńska A. 2006. The mutigene family of fatty acid-binding proteins (FABPs): Function, structure and polymorphism. J. Appl. Genet.  47: 39– 48. [PubMed] Google Scholar CrossRef Search ADS PubMed  Christy R. J. Wang V. W. Ntambi J. M. Geiman D. E. Landschulz W. H. Friedman A. D. Nakabeppu Y. Kelly T. J. Lane M. D. 1989. Differentiation-induced gene expression in 3T3-L1 preadipocytes: CCAAT/enhancer binding protein interacts with and activates the promoters of two adipocyte-specific genes. Genes Dev.  3: 1323– 1335. [PubMed] Google Scholar CrossRef Search ADS PubMed  Coe N. R. Bernlohr D. A. 1998. Physiological properties and functions of intracellular fatty acid-binding proteins. Biochim. Biophys. Acta  1931: 287– 306. Google Scholar CrossRef Search ADS   DeCerbo J. Carmichael G. G. 2005. SINEs point to abundant editing in the human genome. Genome Biol.  6: 216. [PubMed] Google Scholar CrossRef Search ADS PubMed  Deiuliis J. A. Shin J. Bae D. Azain M. J. Barb R. Lee K. 2008. Developmental, hormonal, and nutritional regulation of porcine adipose triglyceride lipase (ATGL). Lipids  43: 215– 225. [PubMed] Google Scholar CrossRef Search ADS PubMed  Dieterich C. Grossmann S. Tanzer A. Ropcke S. Arndt P. F. Stadler P. F. Vingron M. 2005. Comparative promoter region analysis powered by CORG. BMC Genomics  6: 24. [PubMed] Google Scholar CrossRef Search ADS PubMed  Ding S. T. McNeel R. L. Mersmann H. J. 1999. Expression of porcine adipose transcripts: Tissue distribution and differentiation in vitro and in vivo. Comp. Biochem. Physiol. B Biochem. Mol. Biol.  123: 307– 318. [PubMed] Google Scholar CrossRef Search ADS PubMed  Elbrecht A. Chen Y. Cullinan C. A. Hayes N. Leibowitz M. Moller D. D. Berger J. 1996. Molecular cloning, expression and characterization of human peroxisome proliferator activated receptors γ1 and γ2. Biochem. Biophys. Res. Commun.  224: 431– 437. [PubMed] Google Scholar CrossRef Search ADS PubMed  Estellé J. Pérez-Enciso M. Mercadé A. Varona L. Alves E. Sánchez A. Folch J. M. 2006. Characterization of the porcine FABP5 gene and its association with the FAT1 QTL in an Iberian by Landrace cross. Anim. Genet.  37: 589– 591. [PubMed] Google Scholar CrossRef Search ADS PubMed  Frohnert B. I. Hui T. Y. Bernlohr D. A. 1999. Identification of a functional peroxisome proliferator-responsive element in the murine fatty acid transport protein gene. J. Biol. Chem.  274: 3970– 3977. [PubMed] Google Scholar CrossRef Search ADS PubMed  Girardot M. Guibert S. Laforet M. P. Gallard Y. Larroque H. Oulmouden A. 2006. The insertion of a full-length Bos taurus LINE element is responsible for a transcriptional deregulation of the Normande Agouti gene. Pigment Cell Res.  19: 346– 355. [PubMed] Google Scholar CrossRef Search ADS PubMed  Graves R. A. Tontonoz P. Platt K. A. Ross S. R. Spiegelman B. M. 1992. Identification of a fat cell enhancer: Analysis of requirements for adipose tissue-specific gene expression. J. Cell. Biochem.  49: 2119– 2124. Google Scholar CrossRef Search ADS   Graves R. A. Tontonoz P. Ross S. R. Spiegelman B. M. 1991. Identification of a potent adipocyte-specific enhancer: Involvement of an NF-1-like factor. Genes Dev.  5: 428– 437. [PubMed] Google Scholar CrossRef Search ADS PubMed  Hausman G. J. Poulos S. P. Pringle T. D. Azain M. J. 2008. The influence of thiazolidinediones on adipogenesis in vitro and in vivo: Potential modifiers of intramuscular adipose tissue deposition in meat animals. J. Anim. Sci.  86( E Suppl.): E236– E243. [PubMed] Google Scholar CrossRef Search ADS PubMed  Heinäniemi M. Oskari Uski J. Degenhardt T. Carlberg C. 2007. Meta-analysis of primary target genes of peroxisome proliferator-activated receptors. Genome Biol.  8: R147. [PubMed] Google Scholar CrossRef Search ADS PubMed  Herrera R. Ro H. S. Robinson G. S. Xanthopoulos K. G. Spiegelman B. M. 1989. A direct role for C/EBP and the AP-I-binding site in gene expression linked to adipocyte differentiation. Mol. Cell. Biol.  9: 5331– 5339. [PubMed] Google Scholar CrossRef Search ADS PubMed  Hertzel A. V. Smith L. A. Berq A. H. Cline G. W. Shulman G. I. Scherer P. E. Bernlohr D. A. 2006. Lipid metabolism and adipokine levels in fatty acid-binding protein null and transgenic mice. Am. J. Physiol. Endocrinol. Metab.  290: E814– E823. [PubMed] Google Scholar CrossRef Search ADS PubMed  Ho I. C. Bhat N. K. Gottschalk L. R. Lindsten T. Thompson C. B. Papas T. S. Leiden J. M. 1990. Sequence-specific binding of human Ets-1 to the T cell receptor α gene enhancer. Science  250: 814– 818. [PubMed] Google Scholar CrossRef Search ADS PubMed  Hondares E. Mora O. Yubero P. Rodriguez de la Concepción M. Iglesias R. Giralt M. Villarroya F. 2006. Thiazolidinediones and rexinoids induce peroxisome proliferator-activator (PGC)-1α gene transcription: An autoregulatory loop controls PGC-1α expression in adipocytes via peroxisome proliferator-activated receptor-γ coactivation. Endocrinology  147: 2829– 2838. [PubMed] Google Scholar CrossRef Search ADS PubMed  Houseknecht K. L. Bidwell C. A. Portocarrero C. P. Spurlock M. E. 1998. Expression and cDNA cloning of porcine peroxisome proliferator-activated receptor γ (PPARγ). Gene  225: 89– 96. [PubMed] Google Scholar CrossRef Search ADS PubMed  Lee K. Hausman G. J. Dean R. G. 1998. Expression of C/EBPα, β, and δ in fetal and postnatal subcutaneous adipose tissue. Mol. Cell. Biochem.  178: 269– 274. [PubMed] Google Scholar CrossRef Search ADS PubMed  Lee K. Shin J. Latshaw J. D. Suh Y. Serr J. 2009. Cloning of adipose triglyceride lipase complementary deoxyribonucleic acid in poultry and expression of adipose triglyceride lipase during development of adipose in chickens. Poult. Sci.  88: 620– 630. [PubMed] Google Scholar CrossRef Search ADS PubMed  Lee K. Vellena J. A. Moon Y. S. Kim K. H. Lee S. Kang C. Sul H. S. 2003. Inhibition of adipogenesis and development of glucose intolerance by soluble preadipocyte factor-1 (Pref-1). J. Clin. Invest.  111: 453– 461. [PubMed] Google Scholar CrossRef Search ADS PubMed  Lemay D. G. Hwang D. H. 2006. Genome-wide identification of peroxisome proliferator response elements using integrated computational genomics. J. Lipid Res.  47: 1583– 1587. [PubMed] Google Scholar CrossRef Search ADS PubMed  Li B. Shin J. Lee K. 2009. Interferon-stimulated gene ISG12b1 inhibits adipogenic differentiation and mitochondrial biogenesis in 3T3-L1 cells. Endocrinology  150: 1217– 1224. [PubMed] Google Scholar CrossRef Search ADS PubMed  Li B. Zerby H. N. Lee K. 2007. Heart fatty acid binding protein is upregulated during porcine adipocyte development. J. Anim. Sci.  85: 1651– 1659. [PubMed] Google Scholar CrossRef Search ADS PubMed  Livak K. J. Schmittgen T. D. 2001. Analysis of relative gene expression data using real-time quantitative PCR and the 2−ΔΔC(T) method. Methods  25: 402– 408. [PubMed] Google Scholar CrossRef Search ADS PubMed  MacDougald O. A. Lane M. D. 1995. Adipocyte differentiation. When precursors are also regulators. Curr. Biol.  5: 618– 621. [PubMed] Google Scholar CrossRef Search ADS PubMed  Moitra J. Mason M. M. Olive M. Krylov D. Gavrilova O. Marcus-Samuels B. Feigenbaum L. Lee E. Aoyama T. Eckhaus M. Reitman M. L. Vinson C. 1998. Life without white fat: A transgenic mouse. Genes Dev.  12: 3168– 3181. [PubMed] Google Scholar CrossRef Search ADS PubMed  Ohsaki H. Sawa T. Sasazaki S. Kano K. Taniquchi M. Mukai F. Mannen H. 2007. Stearoyl-CoA desaturase mRNA expression during bovine adipocyte differentiation in primary culture derived from Japanese Black and Holstein cattle. Comp. Biochem. Physiol. A Mol. Integr. Physiol.  148: 629– 634. [PubMed] Google Scholar CrossRef Search ADS PubMed  Ren D. Collingwood T. N. Rebar E. J. Wolffe A. P. Camp H. S. 2002. PPARγ knockdown by engineered transcription factors: Exogenous PPARγ2 but not PPARγ1 reactivates adipogenesis. Genes Dev.  16: 27– 32. Google Scholar CrossRef Search ADS PubMed  Rival Y. Stennevin A. Puech L. Rouquette A. Cathala C. Lestienne F. Dupont-Passelaigue E. Patoiseau J. F. Wurch T. Junquéro D. 2004. Human adipocyte fatty acid-binding protein (aP2) gene promoter-driven reporter assay discriminates nonlipogenic peroxisome proliferator-activated receptor γ ligands. J. Pharmacol. Exp. Ther.  311: 467– 475. [PubMed] Google Scholar CrossRef Search ADS PubMed  Ross S. R. Graves R. A. Greenstein A. Platt K. A. Shyu H. L. Mellovitz B. Spiegelman B. M. 1990. A fat-specific enhancer is the primary determinant of gene expression for adipocyte P2 in vivo. Proc. Natl. Acad. Sci. USA  87: 9590– 9594. [PubMed] Google Scholar CrossRef Search ADS   Samulin J. Berget I. Lien S. Sundvold H. 2008. Differential gene expression of fatty acid binding proteins during porcine adipogenesis. Comp. Biochem. Physiol. B Biochem. Mol. Biol.  151: 147– 152. [PubMed] Google Scholar CrossRef Search ADS PubMed  Schoonjans K. Peinado-Onsurbe J. Lefebvre A. M. Heyman R. A. Briqqs M. Deeh S. Staels B. Auwerx J. 1996. PPARα and PPARγ activators direct a distinct tissue-specific transcriptional response via a PPRE in the lipoprotein lipase gene. EMBO J.  15: 5336– 5348. [PubMed] Google Scholar PubMed  Schoonjans K. Watanabe M. Suzuki H. Mahfoudi A. Krey G. Wahli W. Grimaldi P. Staels B. Yamamoto T. Auwerx J. 1995. Induction of the acyl-coenzyme A synthetase gene by fibrates and fatty acids is mediated by a peroxisome proliferator response element in the C promoter. J. Biol. Chem.  270: 19269– 19276. Google Scholar CrossRef Search ADS PubMed  Sheikh F. G. Mukhopadhyay S. S. Gupta P. 2002. PstI repeat: A family of short interspersed nucleotide element (SINE)-like sequences in the genomes of cattle, goat, and buffalo. Genome  45: 44– 50. [PubMed] Google Scholar CrossRef Search ADS PubMed  Shen W. J. Sridhar K. Bernlohr D. A. Kraemer F. B. 1999. Interaction of rat hormone-sensitive lipase with adipocyte lipid-binding protein. Proc. Natl. Acad. Sci. USA  96: 5528– 5532. [PubMed] Google Scholar CrossRef Search ADS   Shimizu M. Yamashita D. Yamaguchi T. Hirose F. Osumi T. 2006. Aspects of the regulatory mechanisms of PPAR functions: Analysis of a bidirectional response element and regulation by sumoylation. Mol. Cell. Biochem.  286: 33– 42. [PubMed] Google Scholar CrossRef Search ADS PubMed  Shin J. Lim S. Latshaw J. D. Lee K. 2008. Cloning and expression of delta-like protein 1 messenger ribonucleic acid during development of adipose and muscle tissues in chickens. Poult. Sci.  87: 2636– 2646. [PubMed] Google Scholar CrossRef Search ADS PubMed  Siebert P. D. Chenchik A. Kellogg D. E. Lukyanov K. A. Lukyanov S. A. 1995. An improved method for walking in uncloned genomic DNA. Nucleic Acids Res.  23: 1087– 1088. [PubMed] Google Scholar CrossRef Search ADS PubMed  Soliman M. Kimura K. Ahmed M. Yamaji D. Matsushita Y. Okamatsu-Ogura Y. Makondo K. Saito M. 2007. Inverse regulation of leptin mRNA expression by short- and long-chain fatty acids in cultured bovine adipocytes. Domest. Anim. Endocrinol.  33: 400– 409. [PubMed] Google Scholar CrossRef Search ADS PubMed  Stejskal D. Karpisek M. 2006. Adipocyte fatty acid binding protein in a Caucasian population: A new marker of metabolic syndrome? Eur. J. Clin. Invest.  36: 621– 625. [PubMed] Google Scholar CrossRef Search ADS PubMed  Su A. I. Cooke M. P. Ching K. A. Hakak Y. Walker J. R. Wiltshire T. Orth A. P. Vega R. G. Sapinoso L. M. Moqrich A. Patapoutian A. Hampton G. M. Schultz P. G. Hogenesch J. B. 2002. Large-scale analysis of the human and mouse transcriptomes. Proc. Natl. Acad. Sci. USA  99: 4465– 4470. [PubMed] Google Scholar CrossRef Search ADS   Sundvold H. Brzozowska A. Lien S. 1997. Charaterisation of bovine peroxisome proliferator-activated receptors γ1 and γ2: Genetic mapping and differential expression of the two isoforms. Biochem. Biophys. Res. Commun.  239: 857– 861. [PubMed] Google Scholar CrossRef Search ADS PubMed  Takasawa K. Kubota N. Terauchi Y. Kadowaki T. 2008. Impact of increased PPARγ activity in adipocytes in vivo on adiposity, insulin sensitivity and the effects of rosiglitazone treatment. Endocr. J.  55: 767– 776. [PubMed] Google Scholar CrossRef Search ADS PubMed  Taniguchi M. Guan L. L. Zhang B. Dodson M. V. Okine E. Moore S. S. 2008. Adipogenesis of bovine perimuscular preadipocytes. Biochem. Biophys. Res. Commun.  366: 54– 59. [PubMed] Google Scholar CrossRef Search ADS PubMed  Tatusova T. A. Madden T. L. 1999. BLAST 2 sequences, a new tool for comparing protein and nucleotide sequences. FEMS Microbiol. Lett.  174: 247– 250. [PubMed] Google Scholar CrossRef Search ADS PubMed  Tontonoz P. Graves R. A. Budavari A. I. Erdjument-Bromage H. Lui M. Hu E. Tempst P. Spiegelman B. M. 1994a. Adipocyte-specific transcription factor ARF6 is a heterodimeric complex of two nuclear hormone receptors, PPARγ and RXRα. Nucleic Acids Res.  22: 5628– 5634. [PubMed] Google Scholar CrossRef Search ADS   Tontonoz P. Hu E. Devine J. Beale E. G. Spiegelman B. M. 1995. PPARγ2 regulates adipose expression of the phosphoenolpyruvate carboxykinase gene. Mol. Cell. Biol.  15: 351– 357. [PubMed] Google Scholar CrossRef Search ADS PubMed  Tontonoz P. Hu E. Graves R. A. Budavari A. I. Spiegelman B. M. 1994b. mPPAR γ2: Tissue-specific regulator of an adipocyte enhancer. Genes Dev.  15: 1224– 1234. Google Scholar CrossRef Search ADS   Tsai S. Y. Tsai M. J. 1997. Chick ovalbumin upstream promoter-transcription factors (COUP-TFs): Coming of age. Endocr. Rev.  18: 229– 240. Google Scholar PubMed  Tuncman G. Erbay E. Hom X. De Vivo I. Campos H. Rimm E. B. Hotamisligil G. S. 2006. A genetic variant at the fatty acid-binding protein aP2 locus reduces the risk for hypertriglyceridemia, type 2 diabetes, and cardiovascular disease. Proc. Natl. Acad. Sci. USA  103: 6970– 6975. [PubMed] Google Scholar CrossRef Search ADS   Varanasi U. Chu R. Huang Q. Castellon R. Yeldandi A. V. Reddy J. K. 1996. Identification of a peroxisome proliferator-responsive element upstream of the human peroxisomal fatty acyl coenzyme A oxidase gene. J. Biol. Chem.  271: 2147– 2155. [PubMed] Google Scholar CrossRef Search ADS PubMed  Vural B. Atalar F. Ciftci C. Demirkan A. Suslevici-Duman B. Gunay D. Akpinar B. Saqbas E. Ozbek U. Buyukdevrim A. S. 2008. Presence of fatty-acid-binding protein 4 expression in human epicardial adipose tissue in metabolic syndrome. Cardiovasc. Pathol.  17: 392– 398. [PubMed] Google Scholar CrossRef Search ADS PubMed  Wang M. Y. Grayburn P. Chen S. Ravazzola M. Orci L. Unger R. H. 2008. Adipogenic capacity and the susceptibility to type 2 diabetes and metabolic syndrome. Proc. Natl. Acad. Sci. USA  105: 6139– 6144. [PubMed] Google Scholar CrossRef Search ADS   Wang N. D. Finegold M. J. Bradley A. Ou C. N. Abdelsayed S. V. Wilde M. D. Taylor L. R. Wilson D. R. Darlington G. J. 1995. Impaired energy homeostasis in C/EBPα knockout mice. Science  269: 1108– 1112. [PubMed] Google Scholar CrossRef Search ADS PubMed  Werman A. Hollenberg A. Solanes G. Bjorbaek C. Vidal-Puig A. J. Flier J. S. 1997. Ligand-independent activation domain in the N terminus of peroxisome proliferator-activated receptor γ (PPARγ). Differential activity of PPARγ1 and -2 isoforms and influence of insulin. J. Biol. Chem.  272: 20230– 20235. [PubMed] Google Scholar CrossRef Search ADS PubMed  Wibowo T. A. Gaskins C. T. Newberry R. C. Thorqaard G. H. Michal J. J. Jianq Z. 2008. Genome assembly anchored QTL map of bovine chromosome 14. Int. J. Biol. Sci.  4: 406– 414. [PubMed] Google Scholar CrossRef Search ADS PubMed  Wurch T. Junquero D. Delhon A. Pauwels J. 2002. Pharmacological analysis of wild-type α, γ, and δ subtypes of the human peroxisome proliferator-activated receptor. Naunyn Schmiedebergs Arch. Pharmacol.  365: 133– 140. [PubMed] Google Scholar CrossRef Search ADS PubMed  Footnotes 1 This work was supported by an Ohio Agricultural Research and Development Center (Columbus) grant to K. Lee. We are grateful to Michelle Milligan (The Ohio State University) for editing this manuscript. American Society of Animal Science