Effect of errors in pedigree on the accuracy of estimated breeding value for carcass traits in Korean Hanwoo cattle

Objective This study evaluated the effect of pedigree errors (PEs) on the accuracy of estimated breeding value (EBV) and genetic gain for carcass traits in Korean Hanwoo cattle. Methods The raw data set was based on the pedigree records of Korean Hanwoo cattle. The animals’ information was obtained using Hanwoo registration records from Korean animal improvement association database. The record comprised of 46,704 animals, where the number of the sires used was 1,298 and the dams were 38,366 animals. The traits considered were carcass weight (CWT), eye muscle area (EMA), back fat thickness (BFT), and marbling score (MS). Errors were introduced in the pedigree dataset through randomly assigning sires to all progenies. The error rates substituted were 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, and 80%, respectively. A simulation was performed to produce a population of 1,650 animals from the pedigree data. A restricted maximum likelihood based animal model was applied to estimate the EBV, accuracy of the EBV, expected genetic gain, variance components, and heritability (h2) estimates for carcass traits. Correlation of the simulated data under PEs was also estimated using Pearson’s method. Results The results showed that the carcass traits per slaughter year were not consistent. The average CWT, EMA, BFT, and MS were 342.60 kg, 78.76 cm2, 8.63 mm, and 3.31, respectively. When errors were introduced in the pedigree, the accuracy of EBV, genetic gain and h2 of carcass traits was reduced in this study. In addition, the correlation of the simulation was slightly affected under PEs. Conclusion This study reveals the effect of PEs on the accuracy of EBV and genetic parameters for carcass traits, which provides valuable information for further study in Korean Hanwoo cattle.


INTRODUCTION
Pedigrees are essential tools in the livestock breeding industry, because they provide ances tral information and knowledge for predicting progeny performance. A pedigree record contains the performance records of individuals and their progeny, and each domestic animal species has traits that are of economic value. For examples, meat and milk traits in cattle, sheep and goats [1]. Carmen [2] reported that pedigree was initially used in cattle breeding and other domestic animals. Henceforth, it becomes the principal breeding tool in the livestock sector. The importance of pedigree records in livestock breeding cannot be over emphasized, because the accuracy of selection depends on the superiority and size of the performance records that are available. Pedigree and performance records have been pre viously used for evaluation of genetic improvement and selection of animals with the highest genetic merit [3]. More so, the pedigree information can explain the genetic differences between and within individuals, creating an essential tech nique to evaluate parameters such as inbreeding, generation interval, estimated breeding value (EBV), heritability, and effective population size. These parameters could be utilized for proper selection and maintenance of healthy and geneti cally superior animals [4]. Pedigree and performance records have been earlier used to evaluate the genetic merit of animals in Hanwoo breeding scheme with proper selection of proven bulls [5]. Lee et al [6,7] also used Hanwoo pedigrees to iden tify the major loci associated with carcass weight (CWT) and intramuscular fat. On the other hand, Long et al [8] stated that the use of raw data for estimation of EBV of livestock could produce biased estimates when pedigree contains errors. In this case, these estimates would be an inaccurate genetic evaluation and slower genetic progress in that population.
Harder et al [9] described two types of pedigree errors (PEs), which could affect the EBV and genetic gain in a dairy cattle population. One of them is missing pedigree information (unknown parents), whereas the other is mistaken pedigree information (wrong parents). They added that, the proportion of wrong paternity decreased the estimates of genetic para meters. Previous studies by Israel and Weller [10], Christensen et al [11], and Gelderman et al [12] showed the consequences of PE or incorrect sire information in estimation of genetic parameters, for example, decreased value of parent trans mitting ability for a cow and her relatives, reduced EBV, h 2 , and genetic gain for meat and milk traits in cattle popula tions. In addition, biased estimates of EBV and genetic gain for both bulls and cows have been reported [10,1315]. Al though, the effect of PEs on the accuracy of EBV and genetic estimates might not be available in Korean Hanwoo cattle. As a result, the knowledge of PEs on the accuracy of genetic parameters (EBVs, genetic gain, and h 2 ) would be useful in the Hanwoo beef industry. Therefore, the aim of this study was to assess the effect of PEs on the accuracy of EBV and genetic gain for carcass traits.

Raw data
The raw data set was based on the pedigree records of Korean Hanwoo cattle, which were the offspring of Korean proven bulls with different dam lines. The animals' records were ob tained using Hanwoo registration records from Korean animal improvement association database [16]. The record comprised of 46,704 animals, where the number of the sires used was 1,298 and the dams were 38,366 animals. The pedigree record consists of the performance and progeny test data sets of con temporary and ancestral relatives. The traits considered were CWT, eye muscle area (EMA), back fat thickness (BFT), and marbling score (MS). The measurements of carcass traits were in accordance with animal product grading service in South Korea [17]. Information about the production and breeding systems of Korean Hanwoo cattle are in accordance with Kim et al [18].

Pedigree errors
Errors were introduced in the pedigree dataset through ran domly assigning sires to all progenies born between 2000 and 2013. This method of changing sire records resulted in wrong sire information. SampleBy function in doBy package of the R software package was used for making PEs. For each gen eration, the pedigree dataset was substituted by error rates of 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, and 80% as the true parent previously described in Oliehoek and Bijma [19].

Simulated data
A simulation was performed to produce one replicate popu lation of 1,650 animals using QMSim software package [20]. The number of sires was 150 whereas the dams were 1,500. For this replicate, the abovementioned method of misiden tification and error rates were introduced into the simulated data. More so, the effect of PE was assessed up to 12 genera tions with their averages. A single trait with a heritability of 0.4 and phenotypic variance of 2,071.47 were simulated in our study. The replacement ratios were 0.5 for sire and 0.4 for dam per generation.

Statistical analyses
A restricted maximum likelihood (REML) based animal model was applied to estimate variance and covariance com ponents of the studied traits using ASReml 4.0 software package [21]. The model included fixed effects of farm loca tion (2), year of birth (3), and season of birth (7). A linear covariate of slaughter age was also fitted in the model. The mixedmodel equation of the animal model used in the study was:  where y, b, u, and e, are the vectors of phenotypes (lists of traits), fixed effects, random effects and residual errors, re spectively, and X and Z are the design matrices.

Breeding value estimation
The similar animal model was also used for estimation of breeding values using Henderson's BLUP method [22] as implemented in ASReml 4.0. The accuracies of EBV estima tion under different error levels of the pedigree were then calculated for all studied traits. Pearson's correlation was used for each trait to assess the influence of PEs on the prediction www.ajas.info

Expected genetic gain
The expected genetic gain from the selection was calculated using the following equation.
where ∆G yr = expected genetic gain/yr, i = intensity of selection, r IH = accuracy, G A = genetic standard deviation, L = generation interval.
In the equation the term i, L were set as 0.80, 5.5, r IH = es timates in Table 3 per each trait, G A = square root of σ 2 a per each trait. Then, the expected genetic gain per year for each trait was obtained from the studied dataset.

Estimates of variances and heritability
The variance components as well as h 2 estimates for carcass traits were estimated using a single trait animal model in equations (1), whereas the equation for h 2 estimates was as follows, where σ 2 a , additive genetic variance; σ 2 p , phenotypic variance; h 2 , heritability.

RESULTS
The average carcass traits per year are presented in Table 1. All the traits showed inconsistency between the years. The CWT, EMA, BFT, and MS showed an increase by 65.61 kg, 4.96 cm 2 , 1.76 mm, and 0.79, respectively. Table 2 illustrates the overall mean, standard deviation, minimum, maximum and coefficient of variation for carcass traits. The average val ues for CWT, EMA, BFT, and MS were 342 kg, 78.76 cm 2 , 8.63 mm, and 3.31, respectively. Figure 1 shows the accuracy of EBV for simulated data under PE scenario. The graphical presentations of the accu racy of EBVs for each trait under PE situation with raw data are shown in Figure 2 to 5. Table 3 indicates the EBV accu racy for carcass traits under PE scenario, and the result shows that all the studied traits were affected by errors. The correla tions of the simulated data under PE are presented in Table  4. The result indicates that, the correlation of the carcass trait was slightly decreased as errors were introduced in the pedi gree. Table 5 shows the expected genetic gain for carcass traits. In this study, the expected genetic gain for carcass traits were 3.13 kg, 0.80 cm 2 , 0.34 mm, and 0.17 for CWT, EMA, BFT, and MS with no PE (0%), and that deemed to decline con stantly as PEs were increased gradually in the dataset. With a 5% PE, the decline in the traits were 0.34 kg, 0.03 cm 2 , 0.01 mm, and 0.01 for CWT, EMA, BFT, and MS with respect to that estimate at 0% PE. This result was followed by the highest values of 2.85 kg, 0.73 cm 2 , 0.26 mm, and 0.17 decline at 80% PE. Table 6 presents the variance components and h 2 estimates for carcass traits. For CWT, the estimated h 2 with no PE was 0.36 in this study. However, this h 2 decreased consistently as more PE introduced in the dataset and reached to as low as 0.03 at 80% of PE. A very similar negative effect of PE on h 2 of EMA was observed as well, where h 2 with no error was estimated as high as 0.42 and as low as 0.05 at 80% PE. The presence of PE equally affected the h 2 of BFT and MS. In this regard, BFT h 2 was reduced from 0.48 with no PE to 0.11 with 80% of PE, whereas for MS, such decrease in h 2 was from 0.58 to a negligible heritability (0.00).

DISCUSSION
In this study, we assessed the effect of PE on the accuracy of EBVs, genetic gain, and h 2 estimates of carcass traits. In gen   [26] in Japanese black (Wagyu). Addi tionally, we need to consider the influences from the breed under study or genotypeenvironment interactions on ani mals that could introduce variations in growth and carcass traits, as suggested earlier by Fabrizio et al [27]. With a dataset including PE, some parameters were found to be influenced in our study (Table 3). We observed that PE had negative associations with the accuracy of EBV of the studied traits. In this case, more errors in the pedigree also reduced the evaluation accuracy in those traits noticeably and that could be a great disadvantage to selection responses in the breeding program. Our results are also in accordance with Israel [10], Ron et al [13], and Bovenhuis and Van Arendonk [28] that similarly reported lower accuracy in EBV with the presence of errors in the pedigree. All traits in this study showed similar trends with respect to PE, even though there were differences in magnitudes of influences on each trait evaluation. Long et al [8] in this regard, agreed with our re sults by showing a reduction in accuracy of EBV for litter size, BFT and average daily gain in swine. Banos et al [29] also reported a 9% milk yield decrease in bull due to PE in their study.
The correlations of the simulated data under PE were slightly reduced in our study as shown in Table 4. Our results indicat ed that the influence of PE on the correlation was somewhat lower. This is probably because our simulated population was not as complicated as the raw dataset. Similar reductions in response between simulated BV and EBV were observed by the study of Van Arendonk et al [30].
We found that the expected genetic gain in animals was also largely reduced due to PEs as presented in Table 5. Our estimated genetic gain and its reduction due to PE is also comparable to other studies by Long et al [8], Israel [10], Christensen et al [11], and Bovenhuis and Van Arendonk [28]. A report on reduced genetic gain, as Angeln dairy cattle population by Sanders et al [31], due to either erroneous or missing sire information was also in agreement with this study. Van Arendonk et al [30] also reported lower genetic gain in a closed pigbreeding nucleus with introduction of errors in the pedigree. The variance components and h 2 estimates for carcass traits are illustrated in Table 6. For all the studied traits, the h 2 esti mates were negatively affected as more errors were introduced in the pedigree. The h 2 being affected by PE also indicated that such errors could reduce the selection accuracy at the same time. For this reason, h 2 is very important to selection for polygenic traits, because selection accounts for those animals with the best breeding values to become the parents of the next generation. In order to increase selection accuracy, we need good information about the candidates for selec tion because the only information available is the phenotypic records, which is the strength of the relationship between phenotypic values and breeding values (i.e., h 2 ). Therefore, when h 2 is low, the phenotypic values mostly reveal little about the underlying breeding values, and it is difficult to determine which animals have the best breeding values to become the potential parents [32]. Our study is also comparable to those reported earlier by Senneke et al [33] who observed a reduc tion in h 2 estimates for birth and weaning weights in Herford cattle. Previous studies by Gelderman et al [12], and Parlato and Van Vleck [34] also reported decreased h 2 estimates in milk fat and milk yield in both cattle and buffalo popula tions under erroneous pedigree.

CONCLUSION
The accuracy of EBV, genetic gain, and heritability estimates for the studied traits were affected by introduction of errors in the pedigree. On the other hand, the result of the correla tion estimates of simulated data for carcass trait was slightly decreased as errors were introduced in the pedigree. As a result, PEs had a negative effect on the overall estimates, which