Canadian Dairy Commission

Over the past 100 yr, the range of traits considered for genetic selection in dairy cattle populations has progressed to meet the demands of both industry and society. At the turn of the 20th century, dairy farmers were interested in increasing milk production; however, a systematic strategy for selection was not available. Organized milk performance recording took shape, followed quickly by conformation scoring. Methodological advances in both genetic theory and statistics around the middle of the century, together with technological innovations in computing, paved the way for powerful multitrait analyses. As more sophisticated analytical techniques for traits were developed and incorporated into selection programs, production began to increase rapidly, and the wheels of genetic progress began to turn. By the end of the century, the focus of selection had moved away from being purely production oriented toward a more balanced breeding goal. This shift occurred partly due to increasing health and fertility issues and partly due to societal pressure and welfare concerns. Traits encompassing longevity, fertility, calving, health, and workability have now been integrated into selection indices. Current research focuses on fitness, health, welfare, milk quality, and environmental sustainability, underlying the concentrated emphasis on a more comprehensive breeding goal. In the future, on-farm sensors, data loggers, precision measurement techniques, and other technological aids will provide even more data for use in selection, and the difficulty will lie not in measuring phenotypes but rather in choosing which traits to select for.

Mastitis is a disease of major economic importance to the dairy cattle sector because of the high incidence of clinical mastitis and prevalence of subclinical mastitis and, consequently, the costs associated with treatment, production losses, and reduced animal welfare. Disease-recording systems compiling data from a large number of farms are still not widely implemented around the world; thus, selection for mastitis resistance is often based on genetically correlated indicator traits such as somatic cell count (SCC), udder depth, and fore udder attachment. However, in the past years, several countries have initiated collection systems of clinical mastitis, based on producers recording data in most cases. The large data sets generated have enabled researchers to assess incidence of this disease and to investigate the genetic background of clinical mastitis itself, as well as its relationships with other traits of interest to the dairy industry. The genetic correlations between clinical mastitis and its previous proxies were estimated more accurately and confirmed the strong relationship of clinical mastitis with SCC and udder depth. New traits deriving from SCC were also studied, with the most relevant findings being associated with mean somatic cell score (SCS) in early lactation, standard deviation of SCS, and excessive test-day SCC pattern. Genetic correlations between clinical mastitis and other economically important traits indicated that selection for mastitis resistance would also improve resistance against other diseases and enhance both fertility and longevity. However, milk yield remains negatively correlated with clinical mastitis, emphasizing the importance of including health traits in the breeding objectives to achieve genetic progress for all important traits. These studies enabled the establishment of new genetic and genomic evaluation models, which are more efficient for selection to mastitis resistance. Further studies that are potential keys for future improvement of mastitis resistance are deep investigation of the bacteriology of mastitis, identification of novel indicator traits and tools for selection, and development of a larger female reference population to improve reliability of genomic evaluations. These cutting-edge studies will result in a better understanding of the genetic background of mastitis resistance and enable a more accurate phenotyping and genetic selection to improve mastitis resistance, and consequently, animal welfare and industry profitability.

A nation-wide health recording system for dairy cattle was started in Canada in 2007. Eight diseases are recorded by producers on a voluntary basis, including mastitis, displaced abomasum, ketosis, milk fever, retained placenta, metritis, cystic ovaries and lameness. Mastitis is the most frequent and most recorded disease, which highlights the economic importance of this trait. A routine genetic evaluation system for mastitis resistance will be officially implemented in December 2013 for Holstein, Ayrshire and Jersey breeds. The model for estimation of breeding values for mastitis resistance is a multiple-trait linear animal model including mastitis, mean SCS in early lactation, standard deviation of SCS, excessive test-day SCC, fore udder attachment, udder depth and body condition score. EBVs for mastitis resistance are published as relative breeding values with a mean of 100 and a standard deviation of 5, where higher values are desirable.

Subclinical mastitis (SCM) causes economic losses for dairy producers by reducing milk production and leading to higher incidence of clinical mastitis and premature culling. The prevalence of SCM in first-lactation heifers is highest during early lactation. The objective of this study was to estimate genetic parameters for SCM in early lactation in first-parity Holsteins. Somatic cell count test-day records were collected monthly in 91 Canadian herds participating in the National Cohort of Dairy Farms of the Canadian Bovine Mastitis Research Network. Only the first test-day record available between 5 and 30 d in milk was considered for analysis. The final data set contained 8,518 records from first lactation Holstein heifers. Six alternative traits were defined as indicators of SCM, using various cutoff values of SCC, ranging from 150,000 to 400,000 cells/mL. Both linear and threshold animal models were used. Overall prevalence of SCM using the 6 traits ranged from 13 to 24%. Heritability estimates (standard error) from linear and threshold models ranged from 0.037 to 0.057 (0.015 to 0.018) and from 0.040 to 0.051 (0.017 to 0.020), respectively. We found strong genetic correlations (standard error) among alternative SCC traits, ranging from 0.90 to 0.99 (0.013 to 0.069), indicating that these 6 traits were genetically similar. Despite low heritability, based on estimated breeding values (EBV) predicted from both models, we noted exploitable genetic variation among sires. Higher EBV of SCM resistance corresponded to sires with a higher percentage of daughters without SCM. Based on a linear model (all 6 traits), percentage of daughters with SCM ranged from 5 to 13% and from 19 to 33% for the top 10% and worst 10% of 69 sires with minimum 20 daughters in at least 5 herds, respectively. Spearman's rank correlations among EBV of sires predicted from linear (from 0.75 to 0.95) and threshold (from 0.74 to 0.95) models were moderate to high, respectively. Very high rank correlations (0.98 to 0.99) between EBV predicted for the same trait from linear and threshold model indicated that reranking of sires based on model used was minimal. In conclusion, despite low heritability, we found utilizable genetic variation in early lactation of heifers. Hence, genetic selection to improve genetic resistance to SCM in early lactation of heifers was deemed possible.

A routine genetic evaluation for mastitis resistance will be officially implemented in Canada in August 2014 for Holstein, Ayrshire and Jersey breeds. The model is a multiple-trait linear animal model including mastitis, average SCS in early lactation, standard deviation of SCS, excessive test-day SCC, fore udder attachment, udder depth and body condition score. Genetic evaluations for clinical mastitis in first lactation as well as in second and later lactations are calculated and expressed as relative breeding values with a mean of 100 and a standard deviation of 5, where higher values are desirable. An index for Mastitis Resistance was developed that includes both the two clinical mastitis traits and the official SCS evaluation, with equal weights. Hyperketonemia or ketosis is one of the most frequent diseases in dairy cattle and the level of milk s-hydroxybutyrate (BHBA) is an indicator of subclinical ketosis. Heritability estimates for milk BHBA in Canadian Holstein cows were between 0.13 and 0.29. Higher milk BHBA in early lactation was genetically associated with a higher frequency of clinical ketosis and displaced abomasum. Milk BHBA can be routinely analyzed in milk samples at test-days, and, therefore, provides a potential alternative for breeding cows with a lower susceptibility to hyperketonemia.

In the Canadian dairy industry, there are currently over 80 traits routinely evaluated, and more are considered for potential selection. Particularly, in the last few years, recording has commenced for several new phenotypes required to introduce novel traits with high economic importance into the selection program. However, without a systematic estimation of the genetic correlations that exist among traits, the potential results of indirect selection are unknown. Therefore, 29 traits representative of the trait diversity for first lactation Canadian animals were selected. Their two-by-two genetic correlations were estimated from a dataset of 62 498 first lactation Holstein cows, using a Markov Chain Monte Carlo Gibbs sampling approach. The general tendencies among the groups of traits confirm that production traits are negatively correlated with fertility traits and that functional traits are positively correlated with one another. The association of udder depth with fertility and disease resistance has also been highlighted. This contribution offers a comprehensive overview of current estimates across traits and includes correlations with novel traits that constitute an original addition to the literature. These new estimates can be used for newly developed genomic evaluation models and possibly lead to more accurate estimations of the dairy cows’ overall genetic merit.

Interbull has introduced a new validation test, and provided corresponding software to detect non-zero time trends and outliers years, for estimates of genetic variance.  The test is applied separately for cows and AI sires, for all traits included in the Interbull MACE evaluation service.  In recent years, AI sires have been genomically preselected, using genotype-based evaluations when they were young calves.  Genomic preselection significantly changes the expectation of Mendelian sampling distributions for AI bulls.  The new Interbull test is applied to EBV computed without genotypes, which are biased by ignored genomic preselection effects.  The purposes of the present study were to apply the new validation test to Canadian data, firstly using official EBV submitted for MACE, and secondly using corrected EBV, after making adjustments to reduce preselection biases in the MS distributions of the most recent AI bulls.  For the main traits under selection in Canada, test results were a pass for official EBV, but a fail for bias-corrected EBV.  For bull populations with genomic preselection, biased EBV are expected to pass the test, while unbiased data are expected to fail.

Genetic evaluation was developed for resistance to metabolic disease traits in Canadian Ayrshire, Holstein and Jersey breeds, with the first official release scheduled for December 2016. The model is a 9-trait animal linear model including producer-recorded data on clinical ketosis ( CK ) and displaced abomasum ( DA ), sub-clinical ketosis ( SCK ) defined as a level of milk β-hydroxybutyrate, and 2 indictor traits: fat to protein ratio ( F:P ) and first lactation body condition score ( BCS ) from the conformation classification.  First and later (up to the 5 th ) lactations are considered as different (but correlated) traits.  Genetic parameters were estimated using a subset (records on 35,575 cows) of the Holstein data. Heritabilities for CK and DA ranged from 0.02 to 0.06. Higher heritabilities were estimated for SCK and indicator traits, from 0.08 (SCK in later lactations) to 0.30 (BCS). Genetic correlations of clinical disease traits between first and later lactations were strong (0.70 for CK and 0.79 for DA), correlations for SCK and F:P were 0.50 and 0.70, respectively.  First lactation CK was strongly correlated with DA (0.77) and SCK (0.68); lower correlations were estimated with BCS (-0.56) and F:P (0.42). Genetic links between DA in first and lactations and indicator traits were weaker.  EBVs for CK, DA and SCK are published as relative breeding values, with a mean of 100 and standard deviation of 5, where higher values are desirable. The overall Metabolic Disease Resistance ( MDR ) index includes SCK, CK and DA, with weights of 50%, 25% and 25%, respectively, and the component EBVs are the averages of first and later lactation EBV for each trait. The MDR index is standardized in the same way as EBVs for individual metabolic disease traits.

Methods presented previously to combine GEBV of young bulls and MACE solutions of ancestors were reviewed.  Variances required for GMACE could be assumed equal to the variances in regular MACE, or estimated from GMACE input data.  Equations to estimate variance were partitioned to explain extreme estimates that have been observed.  Variance estimation was subsequently improved and constraints were applied to avoid extreme variances in GMACE applications.  Subtracting the average difference between national GEBV and MACE parent average forced a null average for Mendelian Sampling estimates and removed inconsistencies among population scales.  This adjustment reduced or eliminated the majority of extreme genomic variance estimates.  The small number of remaining extremes were for traits with unusually low reliabilities of national GEBV.  Nearly all other estimates of genetic standard deviation (SD) were within the range 0.80-1.20 times the SD used for MACE.  Estimates outside this range were truncated to the edges of the range.  RMSE of local GEBV predictions, based on GMACE of data that included GEBV from only foreign countries, were reduced by these constraints on genomic variance estimates.  The use of robust variance estimates also reduced the bias of top young bull predictions, especially for traits with the largest biases.  Relative to the use of MACE variances, GMACE with robust genomic variances gave a slightly higher but similarly low maximum bias for SCS (20% versus 18%) and for all other traits the maximum bias was reduced, from 22% to 10% for protein yield, from 46% to 16% for stature, from 61% to 44% for longevity, and from 28% to 27% for fertility.

This paper addresses two aspects of the reliability in the frame ofinternational beef evaluation: connectedness between countries andincrease of bulls' reliability. Connectedness was measured by the potential bias between genetic levels of countries, while bulls' reliability was computed following MTEDC method. These methodswere applied to a Limousine dataset combining weaning weights of purebred calves from three countries (France, Ireland and UnitedKingdom). The levels of connectedness were low compared to values computed for dairy genetic evaluations. Merging together data and pedigree of these countries leads to a small increase of reliability for France and a much larger one for Ireland and United Kingdom. The results highlight the main interestsin Interbeef evaluation: providing EBVs of foreign bulls without local progeny and increasing EBV reliability of bulls already used in the country.

Editors: S. Niklitschek, J. Lama, C. Trejo, B. Wickham, M. Burke and C. Mosconi

Reliabilities of genomic GEBV are approximated nationally by country, and via GMACE at the international level.  In previous studies, GMACE reliabilities were sometimes lower than expected, relative to corresponding national values.  Reasons for misalignment were investigated in the present study, which revealed that two important data contributions were being ignored for the GMACE reliabilities; the effective daughter contributions (EDC) of a bull's maternal grand-sire, and the contribution of cow records for the dam.  The GMACE system was updated to properly incorporate maternal grand-sire EDC for both reliability approximation and for genomic variance estimation.  The information from cow records for the dam was added only for reliability approximation, and only if it was helpful to align approximate reliabilities, since records of the dam are otherwise excluded from the GMACE model.  These updates improved alignment of GMACE reliabilities with national values.  With only a few exceptions, GMACE reliabilities became consistently equal or higher than national values, which is the generally expected pattern of alignment.

Pro$ (pronounced Pro Dollars) was recently developed by Canadian Dairy Network (CDN) as a second national index that targets dairy producers who generate essentially all of their farm revenue from milk sales. Actual cow profitability data provided to producers by dairy herd improvement (DHI) agencies in Canada, namely CanWest DHI and Valacta, was used as the basis for deriving the new profit-based genetic selection index. Economic parameters used to calculate profitability for each cow are updated annually by economists to reflect changes in milk pricing as well as the associated expenses, including overhead, maintenance feed costs, marginal feed costs and quota opportunity costs. Data used was the accumulated profit to 6 years of age for 672,254 registered Holstein cows with known sire identification, born from January 2005 to September 2008. For cows not surviving to 6 years of age, accumulated profit to the date they left the herd was considered as lifetime profit. For each sire, the average accumulated profit of daughters to 6 years of age was computed.  A total of 830 sires with at least 100 daughters with profit data were used to conduct the two-step multiple trait regression analysis to determine the contribution of sire EBVs for three production, four major type, and eight functional traits in predicting the average daughter profit to 6 years of age. Adjusted R-squared of the Pro$ prediction equation was .6167, which can be applied to any dairy breed with the appropriate scaling factors. Relative to LPI, selection for Pro$ in Holsteins has an stronger expected response for Milk and Protein Yields as well as various functional traits, including Herd Life, while both indexes have similar selection responses for Fat Yield, Daughter Fertility, Mastitis Resistance and Rump. Effective August 2015, Pro$ will be available in the Holstein and Jersey breeds and will be expressed in dollar terms as a deviation from breed average.  For other dairy breeds, the research behind the development of Pro$ was used to modify the LPI formula effective August 2015 to better reflect expected average daughter profit from milk sales.

Lactoferrin (LF) and milk fat globule (MFG) are 2 biologically active components of milk with great economical and nutritional value in the dairy industry. The objectives of this study were to estimate (1) the heritability of mid-infrared (MIR)-predicted LF and MFG size (MFGS) and (2) the genetic correlations between predicted LF and MFGS with milk, fat, and protein yields, fat and protein percentages, and somatic cell score in first-parity Canadian Holstein cattle. A total of 109,029 test-day records from 22,432 cows and 1,572 farms for MIR-predicted LF and 109,212 test-day records from 22,424 cows and 1,559 farms for MIR-predicted MFGS were used in the analyses. Four separate 5-trait random regression test-day models were used. The models included days in milk, herd test date, and a polynomial regression on DIM nested in age-season of calving classes as fixed effects, random polynomial regressions on DIM nested in herd-year of calving, animal additive genetic and permanent environment classes, and a residual effect. Regression curves were modeled using orthogonal Legendre polynomials of order 4 for the fixed age-season of calving effect and of order 5 for the random effects. Moderate overall heritability estimates of 0.34 and 0.46 were estimated for the MIR-predicted LF and MIR-predicted MFGS, respectively. These heritability estimates were similar to the ones estimated for the direct measure of MFGS in a previous study. The genetic correlations between predicted MFGS and fat percentage (0.53) and between predicted LF and protein percentage (0.41) were both moderate and positive. Predicted LF and somatic cell score showed a weaker correlation (0.06) compared with other studies. The moderate genetic correlation between MIR-predicted MFGS and fat percentage and between MIR-predicted LF and protein percentage suggests that MIR predictions of MFGS and LF are not simply a function of the amount of fat and protein percentage, respectively, in the milk (i.e., the prediction equations are not simply predicting fat or protein percentages). Thus, these MIR-predicted values may provide additional information for selecting for fine milk components in Holstein cattle.

Sire evaluations from MACE are used as input for national genomic evaluations. The MACE results are based on traditional evaluation models ignoring genotypes, at both the national and international levels. The exclusion of genotypes is to avoid a cyclical and repeated double-counting of genomic information between national and international systems. Ignoring the genotypes, however, has the consequence of introducing bias in the MACE results, because the effects of genomic preselection are not included in the MACE estimated breeding values of genomically preselected sires. The bias problem is especially relevant for most recent AI bulls, the young sires of most interest in current breeding programs. Current and future methods are discussed, which could be used to reduce genomic preselection biases in MACE, while still generating suitable MACE proofs that can be used as input to national genomic evaluation systems without double-counting the genomic information.

International standards do not exist for the approximation of national genomic reliabilities, which are used as input data for the GMACE international genomic evaluation system.  The focuses of the present study were to develop a method of reducing differences among the national reliabilities approximated by different countries, to apply GMACE using modified national reliabilities, and to use cross-validations tests to determine if GMACE results could be measurably improved.  A non-linear international regression model was applied to the average national reliabilities provided by countries for use in GMACE.  Residuals of prediction for the average national reliabilities were smaller, indicating greater consistency among the approximations of different countries, for protein and stature relative to traits more difficult to evaluate, such as mastitis, stillbirths and cow conception rate.  GMACE input reliabilities were modified by subtracting either some or all of the average prediction error for each combination of  trait by country.  The impacts of modifying the national reliabilities on GMACE results were relatively small.  Predictability of national genomic evaluations by GMACE with only foreign genomic data as input, was essentially the same using either modified or unmodified national reliabilities.  However, the international reliabilities produced by GMACE were more consistent if national reliabilities were modified as input and then the modifications were reversed for the GMACE reliability output.  The approach was to essentially apply an international standardization of reliability on the way into GMACE and then a de-standardization back to each of the original national scales of expression on the way out.

Estimation of genomic variances for GMACE is similar to the estimation of genetic variances for MACE, and is based on REML equations that include a prediction error variance term.  For GMACE, the prediction error variances must be approximated, and this approximation was improved in the present study to better reflect statistical covariance between the national GEBV of young bull and MACE EBV of parents.  In previous estimation, some parental information was being excluded, allowing a simple approximation to work well, but the simple approximation did not work well after including all parental information.  Use of any approximation can bias genomic variance estimates, and such bias would adversely affect conversions of genomic information from the evaluation scales of GEBV to non-GEBV countries.  This bias was minimized by scaling, to eliminate across-country average difference between estimated variance from GMACE relative to MACE.  This MACE-neutral scaling of genomic variance estimates does not affect bull comparisons among GEBV countries, because it does not alter relative genomic variances.  However, it should improve comparisons between GEBV and non-GEBV countries.  The new estimates of genomic variances were very similar to previous estimates from the same data.  Some individual estimates changed, but the rankings of countries from high to low variance were nearly identical as before.  As such, all GMACE bull rankings were nearly identical to previous rankings, with correlations higher than 0.997 for all 5 traits studied and all country scales.  Standard deviations of GMACE predicted breeding values were also very similar, within 1% of the previous in almost all cases.

Genomic variances have been estimated and used in GMACE since 2011, to adjust for differences among countries in the scaling of young bull genomic evaluations relative to progeny-tested bulls.  Interbull has implemented validation tests for national genomic evaluations, which countries must pass in order to participate in GMACE, and the sharing of data and knowledge among countries for genomic evaluations has also increased.  Each of these factors can improve consistency of genomic results among countries, and may reduce the need for genomic variance adjustments in GMACE.  Cross-validation tests have been used previously to compare GMACE results when using versus not using genomic variance adjustments, and have shown clear advantages for including genomic variance adjustments.  When repeated on current data for the present study, however, the cross-validation results no longer showed this clear advantage.  Genomic variance adjustments were helpful for some traits and countries but not for others.  On balance across all traits and countries, there was no longer a clear advantage either way.  The international sharing of data and knowledge, combined with genomic validation tests of Interbull are likely helping to reduce differences among countries in the relative scaling of genomic versus progeny-test evaluations within the same country.

Since the implementation of genomic evaluations in 2009, Canadian Dairy Network (CDN) has used SNP genotypes to verify the reported parents if genotyped and, when missing or incorrect, to discover the animal's sire and/or dam based on all existing SNP genotypes. To date, for breeds with official genomic evaluations in Canada, CDN has over 1.4M genotypes including 1.2M Holstein, 163,000 Jersey, 29,000 Brown Swiss, 6,000 Ayrshire and 3,000 Guernsey. These genotypes involve 23 different genotype panels, including low (3K-30K), medium (44K-140K) and high (over 600K) density. For parentage analysis, a list of 2,683 SNP in common from the 3K and 50K genotype panels are used as the basis for parentage verification, parentage discovery and for identifying families of genetically identical animals. Using the list of SNP proposed for inclusion in GenoEx-PSE for parentage verification (200) and parentage discovery (additional 675 or 354), it was concluded that the 200 SNP recommended by ISAG for parentage verification performed very well compared to the SNP routinely used  by CDN for dairy cattle breeds in Canada.  It was also concluded that parentage discovery using either set of additional 675 or 354 SNP also provided accurate results. To avoid a possible misuse of the additional SNP for parentage discovery, the reduced set of 354 SNP, selected from only 10 chromosome, are recommended for GenoEx-PSE due to the higher level of imputation error and lower accuracy of GEBV estimation compared to results based on the additional 675 SNP.

The objective of this study was to estimate genetic correlations between body condition score (BCS) and calving traits using random regression animal models. Calving traits were a) calving ease (CE) scored from 1=unassisted to 4=surgery and b) calf survival (CS) scored from 0=dead to 1=alive. The data analyzed included first parity Ayrshire BCS records collected between 2001 and 2008 by field staff in herds from Québec. BCS observations were available from 100 days before the calving to 335 after the calving. Calving records were extracted for herds with at least one BCS record. Data included 9,944 BCS observations; 12,011 CE records and 11,600 CS records. (Co)variances were estimated by REML using 2 two-traits models. For BCS, regression curve of genetic and permanent environmental effect were modelled using Legendre polynomials of order 3. For calving traits, no covariance between maternal and direct effects was assumed. The genetic correlation between the maternal effect of CE and the BCS during the 100 days before and after calving ranged between -0.40 and -0.25; a low BCS seemed to increase the chance of the cow to calf with difficulty. For direct CE and maternal and direct CS, the highest correlations with BCS occurred in mid and late lactation. The genetic correlations between BCS and direct and maternal CS ranged from 0.2 to 0.4 and the genetic correlation between BCS and direct CE was around 0.6 at 200 days in milk. It indicated that the ability of the cow to recover its body reserves after the postpartum period would increase the chance of the calf to born easily and to survive