Development of a multi-breed heifer pregnancy genetic evaluation in beef cattle

Abstract

Heifer fertility represents a primary influence on the profitability of a beef cow-calf enterprise. Reproductive rates determine the number of calves born and thus influence the amount of beef product produced at the commercial level driving income for cow-calf operators. Heifer fertility then is an economically relevant trait, though in most cases pregnancy data are cumbersome, untimely to collect, and are considered a rare phenotype in national cattle evaluations (NCE). Despite this, there are successful examples of existing evaluations for heifer pregnancy (HP) across several beef breed associations. These HP genetic evaluations typically rely on categorical exposure (1 = exposed; 0 = not exposed) and pregnancy outcome (1 = pregnant; 0 = not pregnant) data and involve the use of threshold animal models (TM) to convert these binary observations to an underlying normally distributed range of values known as liabilities. These liabilities are then expressed as a percentage that predicts the likelihood of a bull's daughters becoming pregnant and giving birth as two-year-olds in the form of an expected progeny difference (EPD). However, despite these existing HP genetic evaluations, little improvement in the genetic trends in HP has been observed. Perhaps the reason for meager improvement in genetic trend is seedstock producers are not placing enough emphasis on HP, or with pregnancy rates already at or near 90% there is an assumption there is no need for genetic improvement. Additionally, though TM have been successfully implemented in genetic evaluations of HP, a common challenge with the methodology is the inability to evaluate data from contemporary groups that all have the same observation. Even more important is that TM are not supported in some software used for single-step genomic evaluation, such as BOLT by Theta Solutions. Because of these challenges, this study investigated the development of a multi-breed genetic evaluation for HP by performing a series of HP evaluations using TM, linear animal model (LM), and random regression model (RRM) methods. This study used HP data collected on heifers from 1974 to 2020 provided by the International Genetic Solutions (IGS) genetic evaluation, sourced from 9 partner breed associations. Because each breed organization may have its own nuanced definition of HP or differences in how data are reported, inconsistencies in HP data need to be investigated. For example, the American Simmental Association (ASA) does not have an upload format for producers to report HP data but instead uses a system of logic converting whole herd reporting (WHR) codes into HP phenotypes. The first study described the framework for how the ASA converted productivity, culling, and enrollment codes into HP phenotypes. It then evaluated the relative proportions of reasons why heifers/cows were culled. The proportion of heifers culled due to reproductive failure using this method of establishing HP phenotypes was 14%, which is consistent with the national average. The summary statistics for HP observations were cohesive with other HP observations reported to IGS partner breed organizations. Evaluating the effectiveness of these created phenotypes were investigated in the second study. Using data from the American Gelbvieh Association, the Red Angus Association of America, the North American Limousin Foundation, the American Shorthorn Association, and the Canadian Limousin Association, the second study estimated variance components, breed effects, and heterosis effects using LM and TM evaluation methods. Evaluations of HP were performed first within breed before a multibreed population was developed. The average heritability estimate across evaluations performed on 7 different breed groups for HP using LM methods was 0.026, with a minimum value of 0 and a maximum of 0.084. The average heritability for HP using TM methods was 0.17, with a minimum of 0.07 and a maximum of 0.28. Breed populations were then combined into a single multi-breed population, and the same stepwise procedure of incorporating heterosis and breed effects as fixed effects was used to generate variance components and fixed effect solutions. The heritability estimates in this multi-breed population were 0.023 and 0.088 using LM and TM methods, respectively. Heritability estimates did not change as additional fixed effects of breed and heterosis were fit. There were no statistically meaningful breed effects; however, heterosis results in a 17.2% increase (P<0.05) in the probability of HP when maximum heterosis is achieved. Results from this statistical method suggested that LM and TM may be performing equivalently for estimating HP breeding values in within-breed populations; however, in a multi-breed population, results were inconsistent, suggesting perhaps the model was over-specified with breed effects. These results suggest that LM as the model type within a genetic evaluation may be an alternative evaluation method for HP due to its simplicity, ability to use all available information, and support in modern genetic evaluation software programs. Due to being relatively simple to collect and economically important for beef producers, the third study performed a series of evaluations for age at first calving (AFC), which also served as an important investigation as AFC was a potential age covariate in HP evaluations. Models were implemented using single-breed populations and then combined into a larger multi-breed population so heterosis and breed effects could be estimated. The heritability estimates of AFC for Simmental and Red Angus were 0.19 ± 0.01 and 0.14 ± 0.01, respectively. These results demonstrate AFC in days is lowly to moderately heritable. However, when evaluating the genetic trend for both breeds the results seemed incongruous as AFC was sharply increasing over time. Many beef producers mass mate heifers at a single fixed breeding date. As a result, older heifers in a CG will not have the ability to have a younger AFC compared their younger counterparts in the same CG if conception occurs on the same day. To account for this systematic management influence which may be creating a disadvantage in some heifers, age differential (DIFF) was included to account for age differences prior to first exposure and was defined as the difference in days between an individual's birth date and the earliest birth date of an animal in a defined contemporary group. In addition to including DIFF as a fixed effect, accounting for heifer body weight prior to breeding was also considered, and subsequent bivariate animal models of AFC that included yearling weight (YW) were performed. Two bivariate multi-trait animal models for AFC and YW with random additive genetic and residual effects and fixed effects of contemporary group, breed proportion, and retained hybrid vigor were used. When DIFF was not included as a fixed effect, the additive, residual, and phenotypic variances for AFC were 126.1, 456.8, and 582.9 d2, respectively, and the genetic correlation between AFC and YW was 0.36 ± 0.02. When DIFF was included as a fixed effect, the additive, residual, and phenotypic variances for AFC were 10.0, 326.0, and 336.0 d2, respectively. The genetic correlation between AFC and YW was 0.19 ± 0.04. In the absence of DIFF, the heritability estimates for AFC and YW were 0.22 ± 0.01 and 0.44 ± 0.01, respectively, but were 0.03 ± 0.003 and 0.44 ± 0.01 respectively, when DIFF was included. Age differential had a significant effect on AFC at –0.86 (P < 0.0001). The low additive genetic variance of AFC, when accounting for DIFF, suggests that the influence of a female's age going into a fixed breeding date explains much of the variation in AFC. Because of the potential drawbacks associated with LM and TM evaluations of HP, the fourth study investigated alternative definitions of HP using RRM evaluation methods. Two fertility traits evaluated using RRM were proposed; the first being the evaluation of heifer pregnancy by calving week (HPcw), which regresses a binary calving event on the week a heifer calved within her contemporary groups calving window, and the second being the linear evaluation of binary HP which regresses HP on an age covariate such as age at first exposure (AFE) or yearling age (YAGE). In all evaluation methods, Legendre polynomials were used as the base function and observed heritability estimates at different age ranges were transformed from the (co)variances estimated for the intercept and linear term of HPcw or HP. Within the HPcw evaluations, two separate age covariates were proposed as additional fixed effects, with the first being age at first calving (AFC), and the second being AFE. Heritability estimates for HPcw fitting AFC as a fixed effect ranged from 0.39 to 0.56, though this is assuredly from AFC being a biased age estimate. Observed heritability estimates for HPcw across 10 weeks, fitting AFE as a fixed effect ranged from 0.010 to 0.20, which are more realistic and consistent with literature estimates compared to observed HPcw heritability estimates fitting AFC as an age covariate. For the HP evaluation regressing HP on YAGE, heritability estimates ranged from 0.01 to 0.14, suggesting that up to 14% of the variation in HP across ages could be attributed to differences in additive genetics. For the evaluation regressing HP on AFE, heritability estimates were 0 or near zero, so this evaluation method likely requires additional scrutiny. Differences in heifer age covariate and trait definition for the evaluation of HP provided expanded opportunities for the development of national cattle evaluations using RRM. The potential advantages of utilizing RRM in evaluations of categorical or single observation data are that it allows the use of all available data in a dataset and is more adapted to single-step genomic evaluation software systems. Because of this, RRM may be the preferred evaluation method for HP or related fertility traits, though this requires additional testing in global databases. Results from previous studies suggest there are options for evaluating HP in a multi-breed NCE, but no single method is ideal. While LM evaluations validate well, there is low variance in the EBV for the populations evaluated due to low heritability. The TM evaluations validate well and have reasonable predictions, but they cannot appropriately utilize all available data and are not supported by some modern genetic evaluation software programs. The potential of RRM evaluation methods is evident; however, further testing of this methodology must be performed before this approach can be considered.