Genomic predictions for crossbred dairy cattle.

Genomic evaluations are useful for crossbred as well as purebred populations when selection is applied to commercial herds. Dairy farmers had already spent more than $1 million to genotype over 32,000 crossbred animals before US genomic evaluations became available for those animals. Thus, new tools were needed to provide accurate genomic predictions for crossbreds. Genotypes for crossbreds are imputed more accurately when the imputation reference population includes purebreds. Therefore, genotypes of 6,296 crossbred animals were imputed from lower-density chips by including either 3,119 ancestors or 834,367 genotyped animals in the reference population. Crossbreds in the imputation study included 733 Jersey × Holstein F1 animals, 55 Brown Swiss × Holstein F1 animals, 2,300 Holstein backcrosses, 2,026 Jersey backcrosses, 27 Brown Swiss backcrosses, and 502 other crossbreds of various breed combinations. Another 653 animals appeared to be purebreds that owners had miscoded as a different breed. Genomic breed composition was estimated from 60,671 markers using the known breed identities for purebred, progeny-tested Holstein, Jersey, Brown Swiss, Ayrshire, and Guernsey bulls as the 5 traits (breed fractions) to be predicted. Estimates of breed composition were adjusted so that no percentages were negative or exceeded 100%, and breed percentages summed to 100%. Another adjustment set percentages above 93.5% equal to 100%, and the resulting value was termed breed base representation (BBR). Larger percentages of missing alleles were imputed by using a crossbred reference population rather than only the closest purebred reference population. Crossbred predictions were averages of genomic predictions computed using marker effects for each pure breed, which were weighted by the animal's BBR. Marker and polygenic effects were estimated separately for each breed on the all-breed scale instead of within-breed scales. For crossbreds, genomic predictions weighted by BBR were more accurate than the average of parents' breeding values and slightly more accurate than predictions using only the predominant breed. For purebreds, single-trait predictions using only within-breed data were as accurate as multi-trait predictions with allele effects in different breeds treated as correlated effects. Crossbred genomic predicted transmitting abilities were implemented by the Council on Dairy Cattle Breeding in April 2019 and will aid producers in managing their breeding programs and selecting replacement heifers.

Determination of quantitative trait nucleotides by concordance analysis between quantitative trait loci and marker genotypes of US Holsteins.

Experimental designs that exploit family information can provide substantial predictive power in quantitative trait nucleotide discovery projects. Concordance between quantitative trait locus genotype as determined by the a posteriori granddaughter design and marker genotype was determined for 30 trait-by-chromosomal segment effects segregating in the US Holstein population with probabilities of <10-20 to accept the null hypotheses of no segregating gene affecting the trait within the chromosomal segment. Genotypes for 83 grandsires and 17,217 sons were determined by either complete sequence or imputation for 3,148,506 polymorphisms across the entire genome; 471 Holstein bulls had a complete genome sequence, including 64 of the grandsires. Complete concordance was obtained only for stature on chromosome 14 and daughter pregnancy rate on chromosome 18. For each quantitative trait locus, effects of the 30 polymorphisms with highest concordance scores for the analyzed trait were computed by stepwise regression for predicted transmitting abilities of 26,750 bulls with progeny test and imputed genotypes. Effects for stature on chromosome 11, daughter pregnancy rate on chromosome 18, and protein percentage on chromosome 20 met 3 criteria: complete or almost complete concordance, nominal significance of the polymorphism effect after correction for all other polymorphisms, and marker coefficient of determination >40% of total multiple-regression coefficient of determination for the 30 polymorphisms with highest concordance. An intronic variant marker on chromosome 5 at 93,945,738 bp explained 7% of variance for fat percentage and 74% of total multiple-marker regression variance but was concordant for only 24 of 30 families. The missense polymorphism Phe279Tyr in GHR at 31,909,478 bp on chromosome 20 was confirmed as the causative mutation for fat and protein concentration. For effect on fat percentage on chromosome 14, 12 additional missense polymorphisms were found that had almost complete concordance with the suggested causative polymorphism (missense mutation Ala232Glu in DGAT1). The only polymorphism found likely to improve predictive power for genomic evaluation of dairy cattle was on chromosome 15; that polymorphism had a frequency of 0.45 for the allele with economically positive effects on all production 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.

Genomic imputation and evaluation using high-density Holstein genotypes.

Genomic evaluations for 161,341 Holsteins were computed by using 311,725 of 777,962 markers on the Illumina BovineHD Genotyping BeadChip (HD). Initial edits with 1,741 HD genotypes from 5 breeds revealed that 636,967 markers were usable but that half were redundant. Holstein genotypes were from 1,510 animals with HD markers, 82,358 animals with 45,187 (50K) markers, 1,797 animals with 8,031 (8K) markers, 20,177 animals with 6,836 (6K) markers, 52,270 animals with 2,683 (3K) markers, and 3,229 nongenotyped dams (0K) with >90% of haplotypes imputable because they had 4 or more genotyped progeny. The Holstein HD genotypes were from 1,142 US, Canadian, British, and Italian sires, 196 other sires, 138 cows in a US Department of Agriculture research herd (Beltsville, MD), and 34 other females. Percentages of correctly imputed genotypes were tested by applying the programs findhap and FImpute to a simulated chromosome for an earlier population that had only 1,112 animals with HD genotypes and none with 8K genotypes. For each chip, 1% of the genotypes were missing and 0.02% were incorrect initially. After imputation of missing markers with findhap, percentages of genotypes correct were 99.9% from HD, 99.0% from 50K, 94.6% from 6K, 90.5% from 3K, and 93.5% from 0K. With FImpute, 99.96% were correct from HD, 99.3% from 50K, 94.7% from 6K, 91.1% from 3K, and 95.1% from 0K genotypes. Accuracy for the 3K and 6K genotypes further improved by approximately 2 percentage points if imputed first to 50K and then to HD instead of imputing all genotypes directly to HD. Evaluations were tested by using imputed actual genotypes and August 2008 phenotypes to predict deregressed evaluations of US bulls proven after August 2008. For 28 traits tested, the estimated genomic reliability averaged 61.1% when using 311,725 markers vs. 60.7% when using 45,187 markers vs. 29.6% from the traditional parent average. Squared correlations with future data were slightly greater for 16 traits and slightly less for 12 with HD than with 50K evaluations. The observed 0.4 percentage point average increase in reliability was less favorable than the 0.9 expected from simulation but was similar to actual gains from other HD studies. The largest HD and 50K marker effects were often located at very similar positions. The single-breed evaluation tested here and previous single-breed or multibreed evaluations have not produced large gains. Increasing the number of HD genotypes used for imputation above 1,074 did not improve the reliability of Holstein genomic evaluations.

Multibreed genomic evaluations using purebred Holsteins, Jerseys, and Brown Swiss.

Multibreed models are currently used in traditional US Department of Agriculture (USDA) dairy cattle genetic evaluations of yield and health traits, but within-breed models are used in genomic evaluations. Multibreed genomic models were developed and tested using the 19,686 genotyped bulls and cows included in the official August 2009 USDA genomic evaluation. The data were divided into training and validation sets. The training data set comprised bulls that were daughter proven and cows that had records as of November 2004, totaling 5,331 Holstein, 1,361 Jersey, and 506 Brown Swiss. The validation data set had 2,508 Holstein, 413 Jersey, and 185 Brown Swiss bulls that were unproven (no daughter information) in November 2004 and proven by August 2009. A common set of 43,385 single nucleotide polymorphisms (SNP) was used for all breeds. Three methods of multibreed evaluation were investigated. Method 1 estimated SNP effects separately within breed and then applied those breed-specific SNP estimates to the other breeds. Method 2 estimated a common set of SNP effects from combined genotypes and phenotypes of all breeds. Method 3 solved for correlated SNP effects within each breed estimated jointly using a multitrait model where breeds were treated as different traits. Across-breed genomic predicted transmitting ability (GPTA) and within-breed GPTA were compared using regressions to predict the deregressed validation data. Method 1 worked poorly, and coefficients of determination (R(2)) were much lower using training data from a different breed to estimate SNP effects. Correlations between direct genomic values computed using training data from different breeds were less than 30% and sometimes negative. Across-breed GPTA from method 2 had higher R(2) values than parent average alone but typically produced lower R(2) values than the within-breed GPTA. The across-breed R(2) exceeded the within-breed R(2) for a few traits in the Brown Swiss breed, probably because information from the other breeds compensated for the small numbers of Brown Swiss training animals. Correlations between within-breed GPTA and across-breed GPTA ranged from 0.91 to 0.93. The multibreed GPTA from method 3 were significantly better than the current within-breed GPTA, and adjusted R(2) for protein yield (the only trait tested for method 3) were highest of all methods for all breeds. However, method 3 increased the adjusted R(2) by only 0.01 for Holsteins, ≤0.01 for Jerseys, and 0.01 for Brown Swiss compared with within-breed predictions.

Use of the Illumina Bovine3K BeadChip in dairy genomic evaluation.

Genomic evaluations using genotypes from the Illumina Bovine3K BeadChip (3K) became available in September 2010 and were made official in December 2010. The majority of 3K-genotyped animals have been Holstein females. Approximately 5% of male 3K genotypes and between 3.7 and 13.9%, depending on registry status, of female genotypes had sire conflicts. The chemistry used for the 3K is different from that of the Illumina BovineSNP50 BeadChip (50K) and causes greater variability in the accuracy of the genotypes. Approximately 2% of genotypes were rejected due to this inaccuracy. A single nucleotide polymorphism (SNP) was determined to be not usable for genomic evaluation based on percentage missing, percentage of parent-progeny conflicts, and Hardy-Weinberg equilibrium discrepancies. Those edits left 2,683 of the 2,900 3K SNP for use in genomic evaluations. The mean minor allele frequencies (MAF) for Holstein, Jersey, and Brown Swiss were 0.32, 0.28, and 0.29, respectively. Eighty-one SNP had both a large number of missing genotypes and a large number of parent-progeny conflicts, suggesting a correlation between call rate and accuracy. To calculate a genomic predicted transmitting ability (GPTA) the genotype of an animal tested on a 3K is imputed to the 45,187 SNP included in the current genomic evaluation based on the 50K. The accuracy of imputation increases as the number of genotyped parents increases from none to 1 to both. The average percentage of imputed genotypes that matched the corresponding actual 50K genotypes was 96.3%. The correlation of a GPTA calculated from a 3K genotype that had been imputed to 50K and GPTA from its actual 50K genotype averaged 0.959 across traits for Holsteins and was slightly higher for Jerseys at 0.963. The average difference in GPTA from the 50K- and 3K-based genotypes across trait was close to 0. The evaluation system has been modified to accommodate the characteristics of the 3K. The low cost of the 3K has greatly increased genotyping of females. Prior to the availability of the 3K (August 2010), female genotyping accounted for 38.7% of the genotyped animals. In the past year, the portion of total genotypes from females across all chip types rose to 59.0%.

Genomic inbreeding and relationships among Holsteins, Jerseys, and Brown Swiss.

Genomic measures of relationship and inbreeding within and across breeds were compared with pedigree measures using genotypes for 43,385 loci of 25,219 Holsteins, 3,068 Jerseys, and 872 Brown Swiss. Adjustment factors allow genomic and pedigree relationships to match more closely within breeds and in multibreed populations and were estimated using means and regressions of genomic on pedigree relationships and allele frequencies in base populations. Correlations of genomic relationships with pedigree inbreeding were higher within each breed when an allele frequency of 0.5, rather than base population frequencies, was used, whereas correlations of average genomic relationships with average pedigree relationships and also reliabilities of genomic evaluations were higher using base population frequencies. Allele frequencies differed in the 3 breeds and were correlated by 0.65 to 0.67 when estimated from genotyped animals compared with 0.72 to 0.74 when estimated from breed base populations. The largest difference in allele frequency was between Holstein and the other breeds on chromosome Bos taurus autosome 4 near a gene affecting appearance of white skin patches (vitiligo) in humans. Each animal's breed composition was predicted very accurately with a standard deviation of <3% using regressions on genotypes at all loci or less accurately with a standard deviation of <6% using subsets of loci. Genomic future inbreeding (half an animal's mean genomic relationship to current animals of the same breed) was correlated by 0.75 to 0.94 with expected future inbreeding (half the average pedigree relationship). Correlations of both were slightly higher with parent averages than with genomic evaluations for net merit of young Holstein bulls. Thus, rates of increase in genomic and pedigree inbreeding per generation should be slightly reduced with genomic selection, in agreement with previous simulations. Genomic inbreeding and future inbreeding have been provided with individual genomic predictions since 2008. New methods to adjust pedigree and genomic relationship matrices so that they match may provide an improved basis for multibreed genomic evaluation. Positive definite matrices can be obtained by adjusting pedigree relationships for covariances among base animals within breed, whereas adjusting genomic relationships to match pedigree relationships can introduce negative eigenvalues. Pedigree relationship matrices ignore common ancestry shared by base animals within breed and may not approximate genomic relationships well in multibreed populations.

Differences among methods to validate genomic evaluations for dairy cattle.

Two methods of testing predictions from genomic evaluations were investigated. Data used were from the August 2006 and April 2010 official USDA genetic evaluations of dairy cattle. The training data set consisted of both cows and bulls that were proven (had own or daughter information) as of August 2006 and included 8,022, 1,959, and 1,056 Holsteins, Jerseys, and Brown Swiss, respectively. The validation data set consisted of bulls that were unproven as of August 2006 and were proven by April 2010 with 2,653, 411, and 132 Holsteins, Jerseys, and Brown Swiss for the production traits. Method 1 used the training animal's predicted transmitting ability (PTA) from August of 2006. Method 2 used the training animal's April 2010 PTA to estimate single nucleotide polymorphism effects. Both methods were tested using several regressions with the same validation animals. In both cases, the validation animals were tested using the deregressed April 2010 PTA. All traits that had genomic evaluations from the official USDA April 2010 genetic evaluations were tested. Results included bias, differences from expected regressions (calculated using selection intensities), and the coefficients of determination. The genomic information increased the predictive ability for most of the traits in all of the breeds. The 2 methods of testing resulted in some differences that would affect interpretation of results. The coefficient of determination was higher for all traits using method 2. This was the expected result as the data were not independent because evaluations of the validation bulls contributed to their sires' evaluations. The regression coefficients from method 2 were often higher than the regression coefficients from method 1. Many traits had regression coefficients that were higher than 2 standard deviations from the expected regressions when using method 2. This was partially due to the lack of independence of the training and validation data sets. Most traits did have some level of bias in the prediction equations, regardless of breed. The use of method 1 made it possible to evaluate the increased accuracy in proven first-crop bull evaluations by using genomic information. Proven first-crop bulls had an increase in accuracy from the addition of genomic information. It is advised to use method 1 for validation of genomic evaluations.

Selection and management of DNA markers for use in genomic evaluation.

To facilitate routine genomic evaluation, a database was constructed to store genotypes for 50,972 single nucleotide polymorphisms (SNP) from the Illumina BovineSNP50 BeadChip (Illumina Inc., San Diego, CA). Multiple samples per animal are allowed. All SNP genotypes for a sample are stored in a single row. An indicator specifies whether the genotype for a sample was selected for use in genomic evaluation. Samples with low call rates or pedigree conflicts are designated as unusable. Among multiple samples that qualify for use in genomic evaluation, the one with the highest call rate is designated as usable. When multiple samples are stored for an animal, a composite is formed during extraction by using SNP genotypes from other samples to replace missing genotypes. To increase the number of SNP available, scanner output for approximately 19,000 samples was reprocessed. Any SNP with a minor allele frequency of > or = 1% for Holsteins, Jerseys, or Brown Swiss was selected, which was the primary reason that the number of SNP used for USDA genomic evaluations increased. Few parent-progeny conflicts (< or = 1%) and a high call rate (> or = 90%) were additional requirements that eliminated 2,378 SNP. Because monomorphic SNP did not degrade convergence during estimation of SNP effects, a single set of 43,385 SNP was adopted for all breeds. The use of a database for genotypes, detection of conflicts as genotypes are stored, online access for problem resolution, and use of a single set of SNP for genomic evaluations have simplified tracking of genotypes and genomic evaluation as a routine and official process.

Genetic evaluations for mixed-breed populations.

An all-breed animal model was developed for routine genetic evaluations of US dairy cattle. Data sets from individual breeds were combined, and records from crossbred cows were included. About 1% of recent cows were first-generation crossbreds. The numbers of cows with records since 1960 ranged from 10 to 22 million for the 6 traits analyzed, which were milk, fat, protein, somatic cell score, productive life, and daughter pregnancy rate. Programs were modified to account for general heterosis, to group unknown parents separately by breed, to adjust variances separately by breed, and to adjust data to a 36-mo age equivalent instead of a mature equivalent. Convergence rate of the all-breed model was similar to that of the previous within-breed animal model. Estimated breed differences were similar to those obtained previously from phenotypic breed means or from studies of crossbred cows and their herd-mates. Genetic evaluations from the all-breed and within-breed systems had high correlations: >0.99 for recent Holsteins and slightly <0.99 for other breeds. Predicted transmitting abilities will be converted back to the within-breed bases for purebred animals and to the breed of sire base for crossbred animals so that most purebred breeders will not be affected by the change to a multibreed model. Evaluations of crossbred animals from the multibreed model can include accurate information for both parents. Reliabilities also increase for purebred relatives because of the additional crossbred records and in mixed breed herds because cows of other breeds are additional contemporaries. Another benefit of the multibreed model is that breed differences are routinely estimated and updated. More research and education may be needed on using the new evaluations in the design of breeding programs. Implementation is expected in May 2007.

Productive life including all lactations and longer lactations with diminishing credits.

Alternative measures of productive life (PL) were compared, and life expectancy factors were updated to replace estimates from 1993. Alternatives were proposed with extra credits for lactations longer than 10 mo and beyond 84 mo of age and for each calving so that an extremely long lactation would not receive more credits than multiple shorter lactations with dry periods between. Maximum credits per lactation of 10 mo (original PL), 12 mo, and unlimited were compared. The unlimited credits option either included or excluded a calf value equal to 2 mo of production and had credits given for all days either uniformly or based on lactation curves (diminishing credits). Standard lactation curves (first, second, and greater lactations) were estimated based on the test-day yields of Holstein cows remaining in lactation from a set of 903,579 lactation records. For the diminishing credits alternative, credit for a given day of a parity was derived using the predicted yield of the day proportional to the average daily yield of the first 305 d of second parity. Daily yields were deviations from a baseline of 13.62 kg. Heritabilities and genetic correlations were estimated by multitrait REML for alternative measures of PL, for longevity censored at various ages, and for yield traits and SCS in first parity. Data for REML analysis included records from 1,098,329 Holsteins born from 1994 through 1997 from 5,109 sires, and a relationship matrix among sires was included in the model. Lactations beyond 84 mo added little information. Heritability of PL was 0.073 with 10 mo, 0.069 with 12 mo, 0.068 and 0.067 with unlimited (uniform) lactation credits (with and without calf credits, respectively), and 0.070 with unlimited diminishing credits. Corresponding correlations among predicted transmitting abilities for PL and protein yield were 0.07, 0.06, 0.12, 0.23, and 0.09, all much lower than the 0.46 estimated in 1993. Heritability of PL with diminishing credits improved from 0.017 to 0.070 when censoring age increased from 36 to 96 mo. There was no further increase in heritability beyond 96 mo. Genetic correlation with the final PL was 0.87 when PL was censored at 36 mo, but the estimate increased steadily with the censoring age. The PL with diminishing credits, which was favorable in both economic and genetic aspects, was desirable in crediting cows for complete lactations.

Development of a national genetic evaluation for cow fertility.

A national fertility evaluation was developed based on pregnancy rate, which measures the percentage of nonpregnant cows becoming pregnant within each 21-d opportunity period. Data for evaluation are days open, which are calculated as date pregnant minus previous calving date. Date pregnant is determined from last reported breeding or from subsequent calving minus expected gestation length. Success or failure of last breeding can be confirmed by veterinary diagnosis or a report that the cow was sold because of infertility. Data are adjusted for parity and calving season within geographic region and time period and evaluated. Fertility records are considered complete at 250 d in milk, and lower and upper limits of 50 and 250 d are applied to days open. For calculation of genetic evaluations, days open are converted to pregnancy rate by the linear formula pregnancy rate = 0.25 (233 - days open). Evaluations are expressed as predicted transmitting ability for daughter pregnancy rate, and calculation is done with an animal model. Genetic correlations among several fertility measures and other evaluated traits were estimated from 3 large data sets. Correlation with days open was less for nonreturn rate than for days to first breeding, probably because nonreturn rate had lower heritability. Cow fertility was negatively correlated with yield but is a major component of longevity. Thus, recent selection for longevity may have slowed the long-term decline in fertility. Direct selection for fertility could halt or reverse the decline.