Genetic parameters of calving ease trait in Korean Holstein cattle using an animal-maternal model and a sire-maternal grandsire model

1. Introduction

The calving ease (CE) trait of Korean Holstein cattle has gained interest in recent years due to its adverse effects on farm profits. Numerous reports on dairy cattle point to the adverse effects of CE (Dematawewa & Berger, 1997; Carnier et al., 2000; Mee, 2004), ranging from high production costs to the loss of animals and many other long-term animal health and fertility problem concerns. Aside from the stress experienced by the calf and dam during a difficult birth, the dam’s subsequent fertility and production performance can be substantially impacted (Buckley et al., 2003). The CE trait is generally considered a combined interaction between direct and maternal genetic effects (Philipsson, 1976; Meijering, 1984). The direct component is expressed as the genetic ability of a calf to be born easily and is primarily influenced by its body size. The maternal component indicates how easily a female calf can give birth when she becomes a dam and is mainly influenced by the pelvic size of the mature cow. Due to the biological aspects of these two components and their antagonistic relationships (Thompson et al., 1981; Meijering, 1986), each of the genetic effects demands consideration in CE improvement programs. Earlier reports on their relationships suggest that even though female calves are likely to be born easily due to their small size, they are also likely to encounter undesired CE as dams because of their reduced pelvic sizes, which is a negative impact of the small body size at birth.

Despite the appeal of improving CE and evidence of genetic variances (Varona et al., 1999; Carnier et al., 2000), genetic evaluations become challenging due to the choice of statistical evaluation models for this categorical trait. Earlier reports on CE applied different approaches to choosing evaluation models. Many earlier studies applied linear models (Eriksson et al., 2004; Jamrozik et al., 2005) and threshold models (Varona et al., 1999; Ramirez-Valverde et al., 2001). Threshold models are often considered more suitable for CE due to their categorical nature (Gianola, 1982). In contrast, linear models are easy to implement, although they violate some critical statistical assumptions for a quantitative trait. Some linear model-based studies performed logistic transformations of their observations through Snell scoring (Snell, 1964; Tong et al., 1977; Mujibi & Crews, 2009) before analysis to improve the underlying distribution of the data. However, arguments supporting the linear model fit have some practical scenarios, such as, in a population with relatively small contemporary or sire group sizes, where such a model could perform better (Phocas & Laloe, 2003). Our initial observation of the Korean Holstein population also suggests that the data fit the latter scenario more closely. However, developing a statistical model of CE, especially with a target for national-level implementation, may require more effort and will differ from that of other cattle populations. Eaglen et al. (2012) emphasized that a statistical model for a maternally affected trait such as CE should reflect a meticulous balance between the model’s predictive ability and computational feasibility. On the other hand, CE in dairy cattle is often evaluated as a calf trait rather than a dam trait, offering significant advantages for genetic evaluation. This approach allows both male and female calves from the current generation to contribute through traits such as birth weight and body size. Additionally, it facilitates direct estimation of the service sire’s genetic influence on CE, which is crucial for selecting dairy bulls. In contrast, assessing CE as a dam trait restricts evaluations to female animals, focusing primarily on maternal characteristics like pelvic size.

CE is one of the least investigated traits in the Korean Holstein population, and only a few reports have been published on its genetic merit (Lee, 2002; Alam et al., 2017). Also, a comparative analysis of linear animal-maternal (AMAT) and linear sire-maternal grandsire (SMGS) models for CE evaluation has not been performed on Korean Holsteins. Therefore, this study aimed to investigate CE genetic parameters through multiple genetic evaluation models and assess their performance in the Korean Holstein population.

2. Materials and methods

3. Results

Table 2 presents the (co)variances and genetic parameter estimates for variances in the direct and maternal CE components for each parity analysis of the two datasets using the AMAT model. The direct heritability ( ${\text{h}}_{\text{d}}^2)$ estimates were relatively low (< 1%) within and between the parity data sets (0.002 ± 0.002 to 0.006 ± 0.002). However, the maternal heritability ( ${\text{h}}_{\text{m}}^2$ ) estimates varied broadly across the parity levels, where ${\text{h}}_{\text{m}}^2$ from P2 and P3 were relatively higher (0.197 ± 0.015 to 0.354 ± 0.024). Within each parity, dataset differences showed some variability in ${\text{h}}_{\text{m}}^2$ , especially for calves from higher parities. Also, compared to direct heritability, maternal heritability appeared larger in P2 and P3 calves. Genetic correlation estimates between direct and maternal effects ( ${\text{r}}_{\text{dm}}$ ) in the AMAT model were highly variable and demonstrated noticeable inconsistencies. For P1 calves, the correlation estimates varied from a moderately negative (DATA₇: -0.39 ± 0.57) to a positive correlation range (DATA_T: 0.20 ± 1.79). For P3 calves, ${\text{r}}_{\text{dm}}$ estimates varied largely in the two dataset analyses (DATA_T: -0.85 ± 0.85, DATA₇: -0.03 ± 0.65). However, P2 calf associations were moderately negative and similar, ranging between -0.34 and -0.46.

Table 3 presents the genetic parameter estimates of the SMGS model using the two different datasets. Like the AMAT model, direct heritability ( ${\text{h}}_{\text{d}}^2)$ estimates from the SMGS model were generally low for all datasets across the parity levels (0.004 to 0.008). Maternal heritability ( ${\text{h}}_{\text{m}}^2$ ) estimates using this model also appeared very similar to the ${\text{h}}_{\text{d}}^2$ values, which were different from those in the AMAT model. The ${\text{h}}_{\text{m}}^2$ values estimated by the SMGS model from DATA_T and DATA₇ were as small as the direct heritability estimates, and their differences were negligible across animals from all parity levels. The ${\text{r}}_{\text{dm}}$ estimates were moderately antagonistic, ranging between -0.36 ± 0.00 and -0.60 ± 0.00 across the parities and datasets. These correlation estimates were relatively more consistent across datasets than those from AMAT model evaluations.

Table 4 provides Pearson’s and Spearman rank correlations for common sires’ EBVs between DATA_T and DATA₇ from AMAT and SMGS analyses. The correlation of sires’ direct EBV varied widely between 0.13 and 0.91 in the three parity records, whereas their range was medium to high (0.59 to 0.86) for the maternal EBV. Overall, the strength of the correlation between sire EBVs and the sire ranking in the datasets decreased gradually in higher-parity animals. The consistency of the direct and maternal EBVs calculated from DATA_T and DATA₇ was also higher in the SMGS model than in the AMAT model. The sire direct EBV and sire rank were noticeably different using two datasets, as reflected by their low-rank correlation (15%).

Table 5 gives the sire reliability of EBV (R²) for direct and maternal effects based on DATA_T and DATA₇ using AMAT, SMGS, and AMAT*. The AMAT* model integrates direct and maternal variance components from the SMGS model. It uses the parameters estimated by the SMGS model as the final variance approximations for EBV estimation in the AMAT model. The sire reliabilities for the three parity animals varied considerably in the AMAT model analyses using DATA_T (direct: 0.29–0.60, maternal: 0.17–0.86) and DATA₇ (direct: 0.14–0.28; maternal: 0.21–0.88). These estimates were more consistent for both DATA_T (direct: 0.11–0.30, maternal: 0.14–0.30) and DATA₇ (direct: 0.21–0.33, maternal: 0.23–0.24) in the SMGS model. The R² estimate from AMAT* demonstrated similarities to R² values from the SMGS model. In P3, sire’s direct reliability in AMAT analysis also varied noticeably over different data sizes (0.14 to 0.60) compared to SMGS (0.15 to 0.21) and AMAT* (0.14 to 0.19) analyses. Overall, AMAT model-based sire reliabilities showed higher variability between the two datasets than SMGS model-based estimates.

4. Discussion

The knowledge of various sources of genetic effects, such as direct and maternal effects and their relationship, is the key to improving CE in any dairy cattle. This study demonstrated that heritability (h²) estimates ranged from very small to low across parities. A previous study using a SMGS model in Korean Holstein and first parity animals (Alam et al., 2017) revealed direct h² (0.11) and maternal h² (0.05) estimates that were larger than any model estimates in this study. Such a difference could be attributed to the different models and datasets between the previous report and the present study. A study by Salimi et al. (2017) in Iranian Holstein cattle reported small direct h² (0.02) and maternal h² (0.002) estimates, which also provides support for the low h² values in this study. Another animal model study on Iranian Holsteins (Ghiasi et al., 2014) also suggested similar lower h² estimates (direct: 0.041, maternal: 0.012). Similar low heritability estimates for direct (0.03) and maternal CE (0.02), like in our estimation range, were reported by Eaglen et al. (2012). The estimates reported by Eaglen and Bijma (2009) in Dutch Holstein-Friesian partially agree with ours (direct h²: ∼0.08 and maternal h²: ∼0.04). Our estimates were also slightly lower than reports on beef breeds using the linear model (Phocas & Laloe, 2003; Mujibi & Crews, 2009). Differences among reports are mainly due to differences in fitted factors, model types, trait definitions, and breeds. Some studies (Luo et al., 1999; Jamrozik et al., 2005) showed that linear models yielded lower estimates than threshold models. Salimi et al. (2017) argued that an underestimation of their population’s direct and maternal components could be caused by the large phenotypic variances (or residual variances) compared to genetic variances due to recording methods and herd management practices. The previous study in Korean Holstein using the SMGS model by Alam et al. (2017) fitted HYS as a random effect compared to the fixed effect of HY and YS in this study.

The genetic correlation estimates between direct and maternal effects varied across parities and models. Although these correlation estimates were mainly negative, the AMAT model estimate for P1 appeared positive with a high standard error, indicating the possibility of estimation error. In P3 calves, the inconsistency in correlation estimates regarding dataset differences was apparent. However, the SMGS model exhibited greater consistency across datasets and parities. Overall, moderately antagonistic relationships (-0.36 to -0.60) between genetic effects were identified using the SMGS model. Previous reports provided good overall support for our observed negative correlation estimates. Like this study, Alam et al. (2017) also studied first parity Korean Holstein calves and reported an ${\text{r}}_{\text{dm}}$ estimate of -0.68 for service-sire CE, further supporting the findings in the present study. Comparable estimates (-0.41 to -0.43) were also reported by Salimi et al. (2017) and Ghiasi et al. (2014) in Iranian Holstein. Similarly, the animal model correlation estimates (-0.04 to -0.44) by Eaglen and Bijma (2009) coincided with the data in our study. More dairy cattle reports using SMGS models (Luo et al., 1999; Wiggans et al., 2003; Hickey et al., 2007) on ${\text{r}}_{\text{dm}}$ also suggested a range of -0.08 to -0.47, consistent with the present report. While a study in Charolais cattle (Mujibi & Crews, 2009) extended support by a correlation estimate of -0.27, some disagreements in Holstein cattle reported a near absence of correlation (i.e., weakly negative to weakly positive) (Steinbock et al., 2003; Hansen et al., 2004). The variations across reports could be due to differences in breeds or populations, such as beef cattle, in which correlation estimates are often strongly negative (Robinson, 1996). A possible estimation bias is also likely for the genetic covariance between direct and maternal effects (Koch, 1972; Willham, 1980; Meyer, 1992). The present report found noticeable differences between AMAT and SMGS models for genetic parameter estimates. The inconsistency in AMAT model estimates could indicate problems in the present data structure, where dams and their daughters are somewhat confounded to herds. This is because dams and their female progenies rarely change herds in the Korean dairy production system. Using an SMGS model with such datasets could help to efficiently separate herd effects from genetic components and, therefore, reduce estimation bias originating from the confounding data structure.

The stability of sire EBV using AMAT and SMGS models was assessed by correlations between EBV obtained from parity-level DATA_T and DATA₇ datasets. We found an overall higher stability of sire EBV using the SMGS model than the AMAT model. The AMAT model also performed poorly with P3 animals. Our results disagreed with the CE study findings of Ramirez-Valverde et al. (2001) in beef cattle, which showed higher stability with a linear animal model than a linear SMGS model. Sun et al. (2009) also compared animal models with various sire models in Danish Holstein and reported a relative superiority of their animal models over sire models for fertility traits. However, such differences in genetic models could be related to differences in traits for evaluation, data structure, and breeds of interest across studies. Our study showed that the present animal model analyses suffered from variance component estimations and varied across data subsets and parities.

We also compared the expected reliability of sire EBV between direct and maternal effects. Our comparison of EBV reliability for sires suggested that the SMGS model could provide direct EBV as reliably as the AMAT model. Even though the reliability of maternal EBV using the SMGS model was relatively smaller than that of the AMAT model for the full dataset (DATA_T), there was greater consistency across parity-level datasets in the SMGS model. More support for SMGS-based EBV reliability was found from a similar AMAT* model. In contrast, Sun et al. (2009) showed that their animal model provided higher reliability for fertility traits than sire models.

Generally, an animal model is expected to provide higher EBV accuracy (or reliability) as it allows all relatives to contribute to evaluating an animal. In contrast, sire models use less information on relatives. The confounded data structure in this study could limit an accurate estimation of animal EBV in the Korean Holstein population using an AMAT model. However, due to the advantages of animal models, AMAT could be a model choice for national routine evaluations as it would allow for the inclusion of contributions from all relatives related to the animal. An SMGS model, based on the stability of predictions in this study, could be appealing for genetic parameter estimations in Korean Holstein. Although, in theory, a simple sire model is considered inferior to an animal model due to its lack of consideration for all relatives and its estimation bias for EBV (Schaeffer, 1983; Sun et al., 2009) this model can significantly reduce computational demands. An SMGS model, which adequately incorporates genetic relationships among males and females, could reduce the bias associated with a simple sire model (Everett et al., 1979; Schaeffer, 1983). This study showed that AMAT model-based parameter estimates were less stable, which could be associated with the existing dataset limitations described earlier. However, the SMGS model could minimize some bias related to data structures, as suggested by the greater stability of its estimates.

The present study identified multiple challenges regarding estimating variance components and genetic parameters of CE by AMAT and SMGS models. While the SMGS model had superior overall performance, its limited scope regarding animal inclusion (i.e., dam and progeny) for genetic evaluation poses a significant challenge for large-scale national evaluations. Conversely, despite its slightly inferior genetic parameter estimates, the AMAT model offers a comprehensive inclusion of animals, which is crucial for a robust national evaluation. To leverage the strengths of both models in CE evaluation, we suggest using the genetic (co)variance components estimated from the SMGS model within the framework of the AMAT model. This approach would allow us to derive animal EBVs that benefit from the SMGS model’s robust genetic parameter estimates while maintaining the population coverage of the AMAT model. This combined approach could offer a more accurate and comprehensive tool for the Korean national breeding program, effectively balancing the accuracy of EBV and practical applicability.

On the other hand, the SMGS model evaluation can still be valuable for the improvement of CE in Korean Holsteins. EBVs based on the SMGS model can assist in selecting service sires to improve calving ease in heifers, particularly in commercial herds. The SMGS model EBVs can assist breeding programs by focusing on superior maternal lines by identifying sires that can produce easy-calving calves and contribute to a future generation of dams with the desired CE. SMGS-EBVs can also be helpful for the cross-validation of AMAT-EBVs to improve the overall accuracy of EBVs by detecting any discrepancies between animal EBVs from both models. A greater alignment of AMAT-EBVs with those from the SMGS model can suggest the robustness of predictions of the national evaluation. Also, an effective comparison of AMAT and SMGS EBVs can allow breeders to identify the over-representation of sires’ genetics in the population (i.e., similar sires appear at the top of ranking from both models) and help decide to reduce the overuse of a small sire group through selection of a wider range of sires with better performances.

5. Conclusion

In this study, we compared the genetic parameters of the direct and maternal genetic components of CE in P1–P3 Holstein calves using AMAT and SMGS models. Two datasets were used to compare the genetic parameter estimates: one included all calves, and the other included only herd-years with a minimum of seven calves within each parity. Variance components and genetic parameters showed notable influences of model and dataset differences. The direct heritability estimates were mainly very low across parities and datasets. The maternal heritability values varied from low to moderate across parity levels and datasets in the AMAT model, whereas low and consistent values were found in SMGS analyses. The genetic correlation between direct and maternal genetic effects varied largely (-0.85 to 0.20) across parities in the AMAT model using the complete dataset. Genetic correlation estimates were primarily moderate and negative using the SMGS model. The EBV of sires and their ranking were more stable in the SMGS model than in the AMAT model. P3 sire EBVs were the least stable in this study. The expected reliability of sires (with more than 100 calves) was more inconsistent in the AMAT model than in the SMGS model. Overall, SMGS model estimates were more consistent in terms of genetic parameters, reliability of sire EBV, and ranking of sires. We suggest that due to the relative advantage of AMAT, this model can be implemented for national routine evaluations of CE in Korea. In contrast, the SMGS model could be suitable for evaluating CE genetic parameters. This comparative study of linear AMAT and SMGS models was a first in the Korean Holstein population. Our findings are expected to assist future CE evaluation of Korean Holsteins.