The carcass traits of 72,969 Hanwoo born from 2001 through 2011 (cows: 43,955, bulls: 5,849, steers: 23,165) at eight herd recording centers (so called Hanwoo Saupdan) were recorded at 75 beef packing plants from 2002 through 2013. The carcass traits, including grades, measurements and characteristic scores of animals aged 16 to 80 months at slaughter, were recorded by the official graders according to the Animal Products Grade Service (APGS) based on the APGS reference index. The pedigree data of animals were provided by the Korea Animal Improvement Association (KAIA) and the structure of pedigree information was shown by Park et al. (2013).

At slaughter all internal organs and heads were removed, as well as portions of the tail and legs. At 24 h post-slaughter, the carcasses were weighed and evaluated by an official grader for back fat thickness and ribeye area. As shown in Figure 1, backfat thickness was evaluated in terms of thickness of fat over the longissimus muscle (LM) measured perpendicular to the outside surface at a point two thirds of the length of the rib eye from its chine bone end (Park et al., 2002). The area of the rib eye was determined at the surface using a grid at the same point.

Marbling score measured marbling in the rib eye area, while texture was based on the water folding capacity and elasticity of the rib eye area in the grade decision region and maturity considered ossification of cartilage in the left semiconductor backbone thorn promontory. The scales of each score for carcass characteristics could be summarized as marbling (1 = devoid and 9 = abundant), lean meat color (1 = brightly cherry red and 7 = extremely dark red), fat color (1 = white and 7 = dark yellow), texture (1 = proper exudate and good elasticity, and 3 = very high or low exudate and poor elasticity) and maturity (1 = youthful and 9 = mature).

The yield index (Park et al., 2002) represents the retail cut percent, which is predicted from three variables: carcass weight (kg), ribeye area (cm²), and back fat thickness (mm). The correction factors of the variables were as follows: 0.625 for back fat thickness, 0.130 for ribeye area, and 0.024 for carcass weight. In the case of Hanwoo carcasses, a compensating factor of 3.23 was added to the function for yield index as follows:

The beef grading system of Korea reveals two types of information: quality grade and yield grade. Yield grades are primarily stratified by yield index, with A, B, and C indicating yield index ≥67.50, 62.00≤yield index <67.50, and yield index <62.00, respectively. For quality grade, a provisional grade is given according to the degrees of marbling (1 = devoid and 9 = abundant). Marbling scores of 8 and 9, 6 and 7, 4 and 5, 2 and 3, and 1 were roughly categorized as 1++, 1+, 1, 2, and 3, respectively, in the quality grades. The lean meat color, fat color, texture, and maturity of carcass were also evaluated. The quality grade of carcass decreased from the provisional quality grade in the case of a pale and dark color of meat, dark yellow color of fat, very high or low exudate, poor elasticity, and high maturity. The grades for quality and yield were converted to scores of 1 to 6 (for 1++, 1+, 1, 2, 3, and D) and 1 to 4 (for A, B, C, and D), respectively, for this study. Furthermore, the grades were treated as continuous quantitative variables similar to other carcass scores. The full descriptions of the 13 traits are summarized in Table 1. Data for both extremes (0.5%) of each measured trait were removed prior to analysis.

Components of (co)variance, heritabilities, and genetic correlations for the carcass traits were estimated by the restricted maximum likelihood (REML) method with the Multi-Trait Animal model using the Wombat program (Meyer, 2013). The average information matrix was used to approximate sampling errors of (co)variance components, heritabilities, and genetic correlations (Meyer, 2013). The parameters were estimated repeatedly by models with seven or less carcass traits for easy computation until all genetic (co)variances for the 13 carcass traits were obtained.

Fixed effects considered in all models included sex (cow, bull, and steer), age of animal at slaughter (continuous variable: 16 to 80 mo), and birth year (12) within the herd recording center (8). For data originating from the herd recording centers as consulting service providers to farms, birth year within herd recording centers were defined as a contemporary group. The average number of heads slaughtered per year was 6.7, and at least 50 herds were required to set up a herd recording center (Hanwoo Saupdan) according to the rules set by the government. Consequently, it was easy to secure proper numbers of observations in contemporary groups. To eliminate outliers from the data, all animals aged 16 mo or less were excluded, as were bulls and steers older than 36 mo, and cows aged above 80 mo. Age at slaughter was considered a covariate in the mixed linear models of all traits. Fixed covariable age month was fitted to the ordinary quadratic regression without intercept (Meyer, 2013).

Many fluctuations in beef price have occurred in the Korean market in recent years owing to changes in feed price, disease, beef consumption, and import volume of beef. When price traits (carcass price [CP] and price/kg) were dependent variables, slaughter year (n = 12) was added as a fixed effect in the models so that changes in beef price in the market could be taken into account instead of standardization of price per animal (McHugh et al., 2011). Consideration of the effects of slaughter year on contemporary groups was not necessary since animals had different ages at slaughter (Figure 2).

It is common to obtain discrete and/or continuous responses from the observations of animals. Most of the quantitative attribute traits shown in Table 1 were continuous and the evaluated scores of carcass characteristics (including carcass grades) seemed to be categorical and have multinomial distributions. For efficiency of parameter estimation, Liang et al. (1992) proposed a multivariate extension of quasi-likelihood known as generalized estimating equations to estimate the regression coefficients without completely specifying the joint distribution of the multivariate responses, including categorical variable. We assumed that all traits had multivariate normal distribution. If a represents the vector of animal genetic effects for the traits and e represents the vector of residual effects, the random effects can be assumed to follow normal distributions with zero means and the following distribution parameters:

where, A is the numerator relationship matrix; G is a matrix with order 13, the covariance matrix of the genetic effect between traits; I is the identity matrix of the appropriate dimension in each case, and R is the covariance matrix of residual effects between the traits. The residuals effects include all effects except additive genetic effects, such as dominance, maternal effects and environmental effects. The correlations of residual effects could be obtained from the covariances.

The statistics for all 14 observations based on individual carcasses are shown in Table 2. Lee et al. (2008) reported that means of carcass weight (kg), back fat thickness (mm), LM area (cm²), yield index, marbling score, meat color, fat color, texture and maturity were 419.75±39.43, 4.75±1.50, 71.50±11.21, 68.96±1.29, 1.25±0.50, 6.00±0.00, 4.00±0.00, 2.00±0.00, and 4.00±0.82, respectively, for Hanwoo bulls. Figure 2 shows the differences of carcass weight for cows, bulls and steers along with their ages. It could allow us to conjecture the growth of Hanwoo breed in these days. The carcass weights of Hereford, Angus, Brangus, Beefmaster, Bonsmara, and Romosinuano ranged from 331 to 366 kg (Ríos-Utrera et al., 2005; Casas et al., 2010). Even though Hanwoo are considered a lighter breed, their size was found to be similar (average carcass weight = 357.1 kg) to that of typical beef breeds. The average backfat thickness and LM area of crossbreeds ranged from 9.5 to 13 mm and 81.6 to 87 cm² (Casas et al., 2010), while the means in this study differed slightly, being 11.3 mm and 81.1 cm², respectively. The results of this study and previous studies showed slight differences in the locations at which the traits were measured. As shown in Figure 1, the backfat thickness and LM area were measured between the last thoracic vertebrae and the first lumbar. However, in the United States, ribeye area is measured as the surface area on the cut space of the ribeye muscle between the 12th and 13th ribs. Furthermore, fat thickness is measured at 75% of the length of the longissimus dorsi muscle from the split chine bone as the amount of fat opposite the rib eye at the cut space (http://www.thebeefsite.com). Nevertheless, the results indicated a numerical increase of 85 kg in carcass weight and 10.6 cm² in LM area relative to the results (n = 20,881) reported for Hanwoo by Park et al. (2002), which represents a significant improvement in beef production for such a relatively short period.

Both yield and quality grades were converted to numeric scores for this study, and the means of marbling score (4.24) and quality grade (2.93) indicated the third level (grade 1) of quality. Furthermore, the means of the yield index and yield grade (66.78 and 1.78, respectively) indicated second level (B) grades. The USDA and Korean yield grades showed differences in cutability, with the Korean yield grades being 34.2% for A, 53.9% for B, and 11.5% for C, while the USDA yield grades were 15.3% for grade 1, 71.7% for grade 2 and 3, and 13.0% for grade 4 and 5 (Moore et al., 2012). The distributions of Korean quality grades were 7.5% for 1++, 18.9% for 1+, 29.7% for 1, 29.3% for 2, and 14.3% for 3. The USDA quality distributions (Moore et al., 2012) were 2.1% for Prime, 58.9% for Choice, 32.6% for Select, 5.1% for Standard, 0.9% for Commercial, and 0.3% for Utility. The proportion of USDA Prime was lower than that of Korean 1++, which are the best grades in terms of quality, but are measured differently. Conversely, the proportion of cattle categorized as USDA Choice was higher than that of cattle categorized as Korean 1+, which are the second levels of quality grade in both countries. Roast beef without spice is widely consumed by Koreans, whereas marinated steaks with spice are popular in the United States. This difference in taste may lead to the quality grades in Korea being more sensitive to flavor and tenderness than the USDA quality grades, while the difference in the proportions of grades may reflect consumer preference and consumption patterns of both countries.

Heritabilities were presented separately for quantity and quality attributes. The quantity traits, which included carcass measurements (fat thickness, muscle area and carcass weight), yield index, yield grade, and CP, are presented in the first half of Table 4. Our estimated heritability for LM area was 0.23±0.02, which lower than the mean heritability estimate of 0.41 from 36 estimates (of the previous studies) reported by Utrera and Van Vleck (2006) and 0.33 from 3 estimates found in Hanwoo (Lee, 2004; Roh et al., 2004; Hwang et al., 2008). The heritability estimate for backfat thickness was 0.20±0.018, which is smaller than the values of 0.25 and 0.63 reported by Gregory et al. (1995) and Riley et al. (2002), respectively. Hwang et al. (2008) and Roh et al. (2004) reported 0.44 and 0.42, respectively in Hanwoo. Utrera and Van Vleck (2006) reported a mean heritability estimate of 0.41 for the adjusted backfat thickness based on 36 estimates. The estimated carcass weight heritability was 0.28±0.019 and similar to the results of the previous studies in Hanwoo (Lee, 2004; Roh et al., 2004; Hwang et al., 2008), while those of studies in other breeds (Gregory et al., 1995; Wulf et al., 1996; Riley et al., 2002) ranged from 0.15 to 0.55. Utrera and Van Vleck (2006) reported a mean heritability of 0.42 for carcass weight adjusted to a constant age. Previous studies based on small datasets primarily consisted of higher heritability values (Brackelsberg et al., 1971; Benyshek, 1981; Davis and Simmen, 2000; Nephawe et al., 2004; Smith et al., 2007; Nogi et al., 2011; Mao et al., 2013). It was assumed that small environmental or residual variation occurred in the limited or controlled experiments. In the present study, the heritabilities of the three measurement traits (carcass weight, backfat thickness, and LM area) were moderate. The records of beef cattle raised on 1,647 farms were collected from packing plants throughout the country, which may have led to greater variation associated with environmental or other unknown effects. However, similar estimates of 0.20, 0.24, and 0.27 for adjusted fat thickness, LM area and hot carcass weight, respectively, were reported by Ríos-Utrera et al. (2005).

Yield index, which is often referred to as the lean index, reflects cutability and is a function of carcass measurements including backfat thickness, LM area, and carcass weight. Our heritability estimates for yield grade and index were lower than previous estimates for yield grade of 0.30±0.08, 0.46±0.18, 0.54±0.19, 0.46±0.17, 0.85±0.14, and 0.76±0.26 reported by Ríos-Utrera et al. (2005), Davis and Simmen (2000), Pariacote et al. (1998), Smith et al. (2007), Wheeler et al. (2001), and Wulf et al. (1996), respectively.

The CP was determined based on both the weight and quality of the carcass. The heritability of CP was 0.29±0.24, which was lower than the value of 0.53 for carcasses (Brackelsberg et al., 1971) and 0.34 for cattle price at auction (McHugh et al., 2011). The genetic effects of a price trait can be considered a composite economic index of economic traits. Accordingly, the prices cited above could have similar properties as composite indices, even at different stages of lifetime.

The genetic parameters associated with the carcass characteristics, quality grade, and price per kg are presented in the second half of Table 4. The heritability of marbling score (0.28±0.021) was similar or lower than the estimates from studies reported by Davis and Simmen (2000), Ríos-Utrera et al. (2005) and Nephawe et al. (2004) (0.27±0.17, 0.40±0.09, 0.46±0.06, respectively), in which they agreed in general that marbling had a significant effect on carcass quality and was moderately heritable. The heritability estimates for marbling score in Hanwoo by Lee (2004), Hwang et al. (2008) and Roh et al. (2004) were ranged from 0.45 to 0.54, which were higher than the estimates in this study and the estimates for other breeds mentioned above.

The heritability estimates for lean color (0.34) reported by Pratt et al. (2013) and meat color (0.19) reported by Dinkel and Busch (1973) were much higher than the estimates in the present study (both 0.06 in Table 4). The heritabilities of redness, yellowness and lightness were 0.33, 0.28, and 0.21, respectively, in a study conducted by Bonfatti et al. (2013), and 0.29, 0.28 and 0.09 reported by Pratt et al. (2013). The carcass Commission Internationale de l’Eclairage color data were obtained using chroma meters, then converted for analysis (Francis and Clydesdale, 1975; Hunter and Harold, 1987). However, the scores of meat and fat color were assigned based on visual observation by graders using visual help kits, which may have caused increased variation and reduced accuracy in the present study.

The heritability of texture score (0.14±0.016) was lower than 0.29 with a standard error of ±0.17 for lean firmness in a study by Dinkel and Busch (1973) and 0.26±0.08 for taste panel tenderness in an investigation by Nephawe et al. (2004). The heritability of maturity score was very high in the present study (0.41±0.031) relative to the estimates reported by Smith et al. (2007; 0.10±0.10 and 0.00±0.08 for skeletal and lean maturity, respectively). In the present study, age at slaughter was included in the model as a covariate, which may have led to age effects on maturity scores, reduced residual variation, and high heritability.

The heritabilities of carcass quality grade reported by the USDA ranged from 0.25 to 0.29 (Dinkel and Busch, 1973; Davis and Simmen, 2000), which were similar to the results of the present study. Quality grades and price/kg are both related to carcass value, and the heritability of price/kg was similar to the quality grade in the present study (0.24 vs 0.26 in Table 4).

The average standard error (SE) associated with the genetic correlations among 13 traits was 0.079, ranging from 0.003 to 0.161, which was smaller than the values reported by Bergen et al. (2006) (0.11, ranging from 0.03 to 0.22) and Smith et al. (2007) (0.30, ranging from 0.01 to 0.50).

Most studies conducted to estimate genetic parameters for carcass traits have used data from a limited number of producers or experimental stations. Further, reducing variations caused by environmental factors or other sources in those farms has resulted in higher estimates of heritability (Nephawe et al., 2004; Ríos-Utrera et al., 2005; Bergen et al., 2006; Nogi et al., 2011). Conversely, field records in the present study were collected by APGS and from herd recording centers (Hanwoo Saupdan) throughout Korea, which are similar to dairy record processing centers in the United States. Carcass records were collected by APGS using nationwide animal identification data tagged by the Hanwoo Traceability System and pedigree information from the KAIA. The data provided in Table 2 included cows, bulls and steers. Differences in animal weight according to sex and age distribution of the animals are presented in Figure 2. Most bulls and steers were slaughtered within 36 months (average = 25.8 and 30.5 months, respectively), whereas cows were primarily used for reproduction and therefore commonly raised for over 80 months (average = 48.7 months). All of the above records were used to evaluate bulls for AI. The procedures for bull selection are similar to the national procedures, including the progeny test of dairy bulls, operated by the National Agricultural Cooperative Federation (NACF).

Even though evaluation of models was outside the scope of the present study, careful consideration was given to identification of proper models for analysis of field data. Koots et al. (1994) reviewed many studies of beef cattle and showed that there was no difference between heritability estimates when an animal model or other models were used or between estimates that did or did not include maternal effects. More than 70% of farmers raised less than 20 head of cattle, and the average herd size was 23.1 head (Statistics Korea, 2013). The herd recording centers operated by the NACF provide technical services, including mating and feeding plans with the same schemes, to member farms under its control. In addition, the NACF collects animal information regarding LM area and body weights of live cattle. Therefore, the recording centers (n = 8) were considered the primary fixed effect to account for environmental effects instead of farms. Lee et al. (2000) reported that genetic evaluation models of Hanwoo with changing fixed effect groups did not lead to significant differences in the ranking of sires for slaughter weight. They suggested that this indicated that carcass traits were not influenced by environments as much as growth traits. No pre-adjustments, partitions or restrictions for age on carcass traits, carcass weight and fat thickness (Lee et al., 2000; Bergen et al., 2006; McHugh et al., 2011) were adopted for the analysis conducted in the present study; rather, we attempted to adjust the models to account for such effects. Setting constants for factors or traits with field data results in exclusion or adjustment of records because animals are slaughtered at various ages or sizes. Age at slaughter determined by farmers influences almost all carcass traits. Lee et al. (2000) suggested including slaughter age in the model as a covariate for estimation of genetic parameters of Hanwoo carcasses. Manipulation of measurement traits for analysis of characteristic traits may lead farmers to ignore their potential to aid in improving carcass quality through selection. Measurement traits such as carcass weight, LM area and fat thickness may be related to carcass characteristic traits such as marbling and texture. Simultaneous analysis that considers genetic relationships of both measurement and characteristic traits with a multi-trait model may help address the issue, but no previous studies appraising genetic parameter estimation or genetic evaluation were found in a literature search.

Changes in the average breeding values of carcass traits for selection of cows carried out mainly in herd recording centers may reflect the selection response. The genetic trends in carcass weight, LM area, marbling and texture for animals are shown in Figure 3. Smith et al. (2007) reported that tenderness traits were emphasized by the current beef cattle industry, while marbling and texture of carcass are the major traits affecting quality grade and price of carcass. The results of the present study showed that carcass weight and LM area were important to yield grade and price, and that marbling and texture were important carcass characteristics influencing quality grade and price. Further, those measurements and traits were genetically and closely related to each other. Even though backfat thickness affected yield grade, it did not influence quality grade and price as much as expected. As shown in Figure 3, the increase (over 8 kg) in breeding value of carcass weight from birth year 2000 through 2009 partly explained the increase of carcass weight in the phenotypic results reported by Park et al. (2002). The mean breeding value of marbling score decreased by −0.2 during the same period. These findings indicate that emphasis was placed on growth, while no attention was given to selection of quality characteristics. The drastic change in genetic trends observed in 2009 and 2010 reflects shifts in emphasis from growth to quality.