Comparison of performance records and national breeding values as input into international genetic evaluation

The purpose of this investigation was to compare accuracy and precision of variance components and breeding values for international genetic evaluations based on national breeding values or animal performance records. A conventional progeny test scheme was simulated for 3 countries. True breeding values and observations were generated specific to production environments. Two production environments were considered, and both balanced and unbalanced distribution of production environments over countries were considered. True breeding values for both production environments were generated as bivariate normal deviates, and low (0.70) and high (0.90) genetic correlations between performance in production environments were considered. Each cow had an observation in one country only. Performance records were generated as the sum of the true breeding value, a contemporary group effect, and a random residual. Eight generations of data were simulated, and the entire simulated data set was used to compare 3 methods for international genetic evaluation: 1) multiple-trait across-country evaluation based on national predicted breeding values of bulls (Mace), 2) international genetic evaluation across country using performance records, and 3) international genetic evaluation across production environment using performance records. Estimated genetic parameters were biased for all models in this study. Genetic correlations between countries were generally more biased for Mace than for the across-country analyses using performance records. Bias in within-country genetic variances was smaller for Mace. Even genetic parameters obtained with the international evaluation across production environment using performance records were biased, despite the fact that this model was closest to the true, simulated model. The root mean square error of predicted breeding values was similar between models for most of the situations considered. The difference between models was largest when the distribution of production environments over countries was unbalanced and the genetic correlation between performance in production environments was low (0.70). Using breeding values obtained with the across-production environment international genetic evaluation based on performance records will increase the response to selection.     

Estimation of genetic correlations among countries in international dairy sire evaluations with structural models

The increase in the number of participating countries and the lack of genetic ties between some countries has lead to statistical and computational difficulties in estimating the genetic (co)variance matrix needed for international sire evaluation of milk yield and other traits. Structural models have been proposed to reduce the number of parameters to estimate by exploiting patterns in the genetic correlation matrix. Genetic correlations between countries are described as a simple function of unspecified country characteristics that can be mapped in a space of limited dimensions. Two link functions equal to the exponential of minus the Euclidian distance between the coordinates of two countries and the exponential of minus the square of this Euclidian distance were used for the study on international simulated and field data. On simulated data, it was shown that structural models might allow an easier estimation of genetic correlations close to the border of the parameter space. This is not always possible with an unstructured model. On milk yield data, genetic correlations obtained from 22 countries for structural models based on 2 and 7 dimensions, respectively, were analyzed. Only a structural model with a large number of axes gave reasonable estimates of genetic correlations compared with correlations obtained for an unstructured model: 76.7% of correlations deviated by less than 0.030. Such a model reduces the number of parameters from 231 genetic correlations to 126 coordinates. On foot angle data, large deviations were observed between genetic correlations estimated with an unstructured model and correlations estimated with a structural model, regardless of the number of axes taken into account.     

Effectiveness of national and regional sire evaluations in predicting future-daughter milk yield

National and regional bull evaluations were compared for ability to predict standardized milk yield of future daughters. Correlations between evaluations and first-, second-, and third-parity yields of future daughters were calculated within herd-year-month group. Mean correlations with predicted yield of future daughters across the United States were higher for national (0.109, 0.111, and 0.082 for first, second, and third parities, respectively) than for Northeast (0.098, 0.085, and 0.061) Holstein evaluations; corresponding correlations for future Northeast daughters were similar. Bull evaluations based on the first 5 parities of daughters that first calved through 1991 from either California, North Central, Northeast, or Southeast regions as well as from the entire United States were compared with standardized milk yields of daughters that calved later. Correlations with first-, second-, and third-parity yields of future daughters were higher (from 0.001 to 0.011) for national than for regional evaluations. National evaluations were better predictors of future-daughter yield, especially for California and the Southeast. Evaluations based on only first parity were slightly better than those based on the first 5 parities in predicting first-parity yield for 3 of 4 regions but were far less useful in predicting second-or third-parity yield regardless of region. Regional evaluations included fewer bulls because of limited numbers of daughters in each region. The top 100 bulls for genetic merit for milk yield based on regional rankings were inferior to the top 100 bulls based on national ranking by 25 to 173 kg. Reliance on regional rather than national evaluations would reduce current US genetic gains.     

Assessment of the value of international genetic evaluations for yield in predicting domestic breeding values for foreign Holstein bulls

International genetic evaluations are a valuable source of information for decisions about the importation of (the semen of) foreign bulls. This study analyzed data from 6 countries (Australia, Canada, Italy, France, the Netherlands, and the United States) and compared international evaluations for production traits of foreign bulls (i.e., when no national daughter information was available) to their national breeding values in August 2009, which were based only on domestic daughters’ data. A total of 821 bulls with highly reliable estimated breeding values (EBV) for milk, fat, and protein yield were analyzed. No evidence of systematic over- or underestimation was found in most of the countries analyzed. Observed correlations between national and international evaluations were close to 0.9 and, for most countries, generally close to their expected values (calculated from national and international EBV reliabilities). In Italy, however, higher differences between observed and expected correlations and significant mean differences between EBV for more than one trait were observed in bulls progeny-tested in the United States and in other European countries (with differences up to 33.1% of the genetic standard deviation). These results were probably induced by a relatively recent change in the model for national evaluation. The findings in this study reflect a conservative estimate of the real value of international evaluations, as changes in methodologies in either the national or the international evaluations decreased the ability of past international evaluations to predict current national evaluations. Nevertheless, our results indicate that international evaluations based on foreign information for Holstein bulls were reasonably accurate predictors of the future national breeding values based only upon domestic daughters.     

Identification of factors that cause genotype by environment interaction between herds of Holstein cattle in seventeen countries

Currently, the International Bull Evaluation Service calculates international dairy sire evaluations using the multiple-trait across country evaluation procedure. This method depends implicitly on political boundaries between countries, because the input data are national evaluations from each participating country. Therefore, different countries are treated as different production environments. The goal of this study was to identify factors that describe the production system on each farm. Such factors could be used to group herds across countries for borderless genetic evaluations. First lactation milk records of Holstein cows calving between January 1, 1990 and December 31, 1997 in Australia, Austria, Belgium, Canada, Czech Republic, Estonia, Finland, Germany, Hungary, Ireland, Israel, Italy, The Netherlands, New Zealand, South Africa, Switzerland, and the USA were used in this study. Thirteen genetic, management, and climatic variables were considered as potential indicators of production environments: peak milk yield, persistency, herd size, age at first calving, seasonality of calving, standard deviation of milk yield, culling rate, days to peak yield, fat to protein ratio, sire PTA milk, percentage of North American Holstein genes, maximum monthly temperature, and annual rainfall. Herds were grouped into quintiles based on herd averages for each of these variables. Genetic correlations for lactation milk yield between quintiles were significantly less than one for maximum monthly temperature, sire PTA milk, percent North American Holstein genes, herd size, and peak milk yield. The variables can be used to group herds into similar production environments, regardless of country borders, for the purpose of accounting for genotype by environment interaction in international dairy sire evaluation.     

Prediction of genetic correlations and international breeding values for missing traits

Prediction of genetic merit for missing traits is possible by combining available indicator traits. Indicator traits were combined using genetic correlations obtained from multiple regression equations of estimated genetic correlations among available indicator traits on variables explaining production circumstances and trait definitions. This prediction of missing traits was closer to actual breeding values than breeding values for any of the indicator traits. This was verified by evaluating clinical mastitis in each of the Nordic countries as a missing trait. The derived methodology was used to predict breeding values for clinical mastitis in the United States for local and international bulls with an average reliability of 43%.     

Ignoring genotype by environment interaction in the genetic evaluation of dairy cattle reduces accuracy but may increase selection intensity

Genotype by environment interaction (G×E) may exist for traits that are expressed in different environments. The G×E is often ignored in the genetic evaluation of selection candidates. We hypothesized that genetic gain in 2 environments is always higher when the true value of the genetic correlation (r_g) between traits expressed in different environments is considered in the genetic evaluation. We tested this hypothesis by stochastic simulation of dairy cattle breeding programs in a mainstream and a niche environment. The r_g was varied from 0 to 1 in steps of 0.1. We simulated the following 3 scenarios: 1Trait_1Index, 2Traits_1Index, and 2Traits_2Indices. The G×E was ignored in the genetic evaluation in the scenario with 1Trait and included in scenarios with 2Traits. Selection was based on the mainstream selection index in both environments in scenarios with 1Index. Selection in the mainstream environment was based on the mainstream selection index and selection in the niche environment was based on the niche selection index in the scenario with 2Indices. With moderate G×E (r_g between 0.6 and 0.9), the highest genetic gain was achieved in the niche environment by selecting for the mainstream selection index and ignoring G×E. At lower r_g, the highest genetic gain was achieved when considering G×E and selecting for the niche selection index. For the mainstream environment, it was never an advantage to ignore G×E. Therefore, although our hypothesis was confirmed in most cases, there were cases where ignoring G×E was the better option, and using the correct evaluation led to inferior genetic gain. The results of the current study can be used in animal breeding programs that encompass multiple environments.     

Genetic parameters for conformation traits in herds that differ in mean final score and completeness of pedigree and performance data

The objectives of this study were to assess differences in the heritability of type (conformation) traits between herds that differ in mean final score and completeness of pedigree and performance data and to estimate genetic correlations among these environments. Measurement of subjective characteristics, such as conformation traits, may be more difficult in herds with poor management conditions, and genetic evaluation of sires using data from such herds could lead to inaccurate selection decisions. Furthermore, missing pedigree data is a significant problem in many herds, and a lack of phenotypic data from maternal relatives may reduce the effectiveness of animal model evaluation systems. These hypotheses were examined using type classification scores of 1,728,836 first-parity Holstein cows (from 54,223 sires) that calved from 1993 to 2002 in 24,207 US dairy herds. These data included 480,927 records from progeny test daughters, but only 254,891 (47%) were from dams that had valid sire identification, and only 132,953 (28%) were from dams that had also been classified. Herds were grouped into quartiles by mean classification score, percentage of known maternal grandsires, and percentage of classified dams. Estimated heritability of final score was 0.20 in herds with mean score <74.5, 0.17 in herds with <25% known maternal grandsires, and 0.19 in herds with <18% classified dams. Conversely, estimates were 0.39 in herds with mean score >78.7, 0.35 in herds with 100% known maternal grandsires, and 0.37 in herds with >71% classified dams. Estimated genetic correlations between quartiles ranged from 0.86 to 0.95. Based on this study, it appears that improvements in animal identification and data collection in progeny test herds would lead to greater accuracy and stability of genetic evaluations for conformation traits in US Holstein cattle.     

Prediction accuracies and genetic parameters for test-day traits from genomic and pedigree-based random regression models with or without heat stress interactions

The aim of this study was to compare genetic (co)variance components and prediction accuracies of breeding values from genomic random regression models (gRRM) and pedigree-based random regression models (pRRM), both defined with or without an additional environmental gradient. The used gradient was a temperature-humidity index (THI), considered in statistical models to investigate possible genotype by environment (G×E) interactions. Data included 106,505 test-day records for milk yield (MY) and 106,274 test-day records for somatic cell score (SCS) from 12,331 genotyped Holstein Friesian daughters of 522 genotyped sires. After single nucleotide polymorphism quality control, all genotyped animals had 40,468 single nucleotide polymorphism markers. Test-day traits from recording years 2010 to 2015 were merged with temperature and humidity data from the nearest weather station. In this regard, 58 large-scale farms from the German federal states of Berlin-Brandenburg and Mecklenburg-West Pomerania were allocated to 31 weather stations. For models with a THI gradient, additive genetic variances and heritabilities for MY showed larger fluctuations in dependency of DIM and THI than for SCS. For both traits, heritabilities were smaller from the gRRM compared with estimates from the pRRM. Milk yield showed considerably larger G×E interactions than SCS. In genomic models including a THI gradient, genetic correlations between different DIM × THI combinations ranged from 0.26 to 0.94 for MY. For SCS, the lowest genetic correlation was 0.78, estimated between SCS from the last DIM class and the highest THI class. In addition, for THI × THI combinations, genetic correlations were smaller for MY compared with SCS. A 5-fold cross-validation was used to assess prediction accuracies from 4 different models. The 4 different models were gRRM and pRRM, both modeled with or without G×E interactions. Prediction accuracy was the correlation between breeding values for the prediction data set (i.e., excluding the phenotypic records from this data set) with respective breeding values considering all phenotypic information. Prediction accuracies for sires and for their daughters were largest for the gRRM considering G×E interactions. Such modeling with 2 covariates, DIM and THI, also allowed accurate predictions of genetic values at specific DIM. In comparison with a pRRM, the effect of a gRRM with G×E interactions on gain in prediction accuracies was stronger for daughters than for sires. In conclusion, we found stronger effect of THI alterations on genetic parameter estimates for MY than for SCS. Hence, gRRM considering THI especially contributed to gain in prediction accuracies for MY.     

Comparison between a sire model and an animal model for genetic evaluation of fertility traits in Danish Holstein population

Comparisons between a sire model, a sire-dam model, and an animal model were carried out to evaluate the ability of the models to predict breeding values of fertility traits, based on data including 471,742 records from the first lactation of Danish Holstein cows, covering insemination years from 1995 to 2004. The traits in the analysis were days from calving to first insemination, calving interval, days open, days from first to last insemination, number of inseminations per conception, and nonreturn rate within 56 d after first service. The correlations between sire estimated breeding value (EBV) from the animal model and the sire-dam model were close to 1 for all the traits, and those between the animal model and the sire model ranged from 0.95 to 0.97. Model ability to predict sire breeding value was assessed using 4 criteria: 1) the correlation between sire EBV from 2 data subsets (DATA_A and DATA_B); 2) the correlation between sire EBV from training data (DATA_A or DATA_B) and yield deviation from test data (DATA_B or DATA_A) in a cross-validation procedure; 3) the correlation between the EBV of proven bulls, obtained from the whole data set (DATA_T) and from a reduced set of data (DATA_C1) that contained only the first-crop daughters of sires; and 4) the reliability of sire EBV, calculated from the prediction error variance of EBV. All criteria used showed that the animal model was superior to the sire model for all the traits. The sire-dam model performed as well as the animal model and had a slightly smaller computational demand. Averaged over the 6 traits, the correlations between sire EBV from DATA_A and DATA_B were 0.61 (sire model) versus 0.64 (animal model), the correlations between EBV from DATA_T and DATA_C1 for proven bulls were 0.59 versus 0.67, the correlations between EBV and yield deviation in the cross-validation were 0.21 versus 0.24, and the reliabilities of sire EBV were 0.42 versus 0.46. Model ability to predict cow breeding value was measured by the reliability of cow EBV, which increased from 0.21 using the sire model to 0.27 using the animal model. All the results suggest that the animal model, rather than the sire model, should be used for genetic evaluation of fertility traits.     

Comparison of single-trait to multi-trait national evaluations for yield,health, and fertility

Flexible software was designed to replace the current animal model programs used for national genetic evaluations. Model improvements included (1) multi-trait processing, (2) multiple fixed class and regression variables, (3) differing models for different traits, (4) random regressions, and (5) foreign data included using pseudo-records. Computational improvements included (6) parallel processing, (7) renumbering class variables to equation numbers within the program so that estimated effects are output with original identification numbers, and (8) reliability computed within the same program. When applied to 3 fertility traits of 27,971,895 cows and heifers, the new model used daughter pregnancy rate as a correlated trait to improve heifer and cow conception rate evaluations for older animals and in herd-years where records are missing, and also added information from crossbreds. When applied to 7 traits and 76,846,327 lactation records of 30,064,300 cows, gains in accuracy were small for yield and somatic cell score, moderate for daughter pregnancy rate, and larger for productive life for recent bulls compared with single-trait evaluations. For very old bulls, multi-trait gains were also large for protein because lactation records were available only for milk and fat. Multi-trait productive life was computed with exact rather than approximate methods; however, correlated information from conformation was excluded, reducing advantages of the new model over the previous software. Estimates of breed differences, inbreeding depression, and heterosis were similar to previous estimates; new estimates were obtained for conception rates. Predictions were compared by truncating 4 yr of data, and genetic trend validation was applied to all breed–trait combinations. The estimates of trend account for increases in inbreeding across time. Incorporation of foreign data gave correlations above 0.98 for new with previous evaluations of foreign Holstein bulls, but lower for other breeds. The 7-trait model required 35 GB of memory and 3 d to converge using 7 processors. The new software was implemented for fertility traits in 2013 and is scheduled for implementation with yield, somatic cell score, and productive life in 2014. Further revision of the models and software may be needed in the near future to account for genomic preselection.     

Characterization of best linear unbiased estimates generated from national genetic evaluations of reproductive performance,survival, and milk yield in dairy cows

Genetic evaluations decompose an observed phenotype into its genetic and nongenetic components; the former are termed BLUP with the solutions for the systematic environmental effects in the statistical model termed best linear unbiased estimates (BLUE). Geneticists predominantly focus on the BLUP and rarely consider the BLUE. The objective of this study, however, was to define and quantify the association between 8 herd-level characteristics and BLUE for 6 traits in dairy herds, namely (1) age at first calving, (2) calving to first service interval (CFS), (3) number of services, (4) calving interval (CIV), (5) survival, and (6) milk yield. Phenotypic data along with the fixed and random effects solutions were generated from the Irish national multi-breed dairy cow fertility genetic evaluations on 3,445,557 cows; BLUE for individual contemporary groups were collapsed into mean herd-year estimates. Data from 5,707 spring-calving herds between the years 2007 and 2016 inclusive were retained; association analyses were undertaken using linear mixed multiple regression models. Pearson coefficient correlations were used to quantify the relationships among individual trait herd-year BLUE, and transition matrices were used to understand the dynamics of mean herd BLUE estimates over years. Based on the mean annual trends in raw, BLUP, and BLUE, it was estimated that BLUE were associated with at least two-thirds of the improvement in CIV and milk production over the past 10 yr. Milk recording herds calved heifers for the first time on average 15 d younger, had an almost 2 d longer CFS but 2.3 d shorter CIV than non-milk-recording herds. Larger herd sizes were associated with worse BLUE for both CFS and CIV. Expanding herds and herds that had the highest proportion of cows born on the farm itself, on average, calved heifers younger and had shorter CIV. By separating the raw performance of a selection of herds into their respective BLUE and BLUP, it was possible to identify herds with inferior management practices that were being compensated by superior genetics; similarly, herds were identified with superior BLUE, but because of their inferior genetic merit, were not reaching their full potential. This suggests that BLUE could have a pivotal role in a tailored decision support tool that would enable producers to focus on the most limiting factor hindering them from achieving their maximum performance.     

Genetic analysis of the Israeli Holstein dairy cattle population for production and nonproduction traits with a multitrait animal model

Milk, fat, and protein production, somatic cell score (SCS), and female fertility in the Israeli Holstein dairy cattle population were analyzed using a multitrait animal model (AM) with parities 1 through 5 as separate traits. Female fertility was measured as the inverse of the number of inseminations to conception in percent. Variance components were estimated using both the repeatability AM and multitrait AM. The multitrait heritabilities for individual parities were greater than the heritabilities from the repeatability AM, and heritabilities decreased with an increase in parity number. Heritabilities were higher for production traits, lower for SCS, and lowest for female fertility. The genetic correlations were higher than the environmental correlations. Genetic correlations between parities decreased with an increase in the difference in parity number, but all were greater than 0.5. The environmental correlations were higher for production traits, lower for SCS, and close to zero for female fertility. In the analysis of the complete milk recorded population, genetic trends from the repeatability and multitrait models were very similar. The genetic trend for SCS was economically unfavorable until 1993, and favorable since then. The genetic trend for female fertility was close to zero, but the annual environmental trend was -0.2%. The multitrait lactation model is an attractive compromise between repeatability lactation models, which do not account for maturing trends across parities, and test-day models, which are much more demanding computationally.     

Short communication: Genotyping of cows to speed up availability of genomic estimated breeding values for direct health traits in Austrian Fleckvieh (Simmental) cattle—Genetic and economic aspects

The aim of this study was to quantify the impact of genotyping cows with reliable phenotypes for direct health traits on annual monetary genetic gain (AMGG) and discounted profit. The calculations were based on a deterministic approach using ZPLAN software (University of Hohenheim, Stuttgart, Germany). It was assumed that increases in reliability of the total merit index (TMI) of 5, 15, and 25 percentage points were achieved through genotyping 5,000, 25,000, and 50,000 cows, respectively. Costs for phenotyping, genotyping, and genomic estimated breeding values vary between €150 and €20 per cow. The gain in genotyping cows for traits with medium to high heritability is more than for direct health traits with low heritability. The AMGG is increased by 1.5% if the reliability of TMI is 5 percentage points higher (i.e., 5,000 cows genotyped) and 6.53% higher AMGG can be expected when the reliability of TMI is increased by 25 percentage points (i.e., 50,000 cows genotyped). The discounted profit depends not only on the costs of genotyping but also on the population size. This study indicates that genotyping cows with reliable phenotypes is feasible to speed up the availability of genomic estimated breeding values for direct health traits. But, because of the huge amount of valid phenotypes and genotypes needed to establish an efficient genomic evaluation, it is likely that financial constraints will be the main limiting factor for implementation into breeding program such as Fleckvieh Austria.