Assessing different cross-validation schemes for predicting novel traits using sensor data: an application to dry matter intake and residual feed intake using milk spectral data.

Feed efficiency is important for economic profitability of dairy farms; however, recording daily dry matter intakes (DMI) is expensive. Our objective was to investigate the potential use of milk mid-infrared (MIR) spectral data to predict proxy phenotypes for DMI based on different cross-validation schemes. We were specifically interested in comparisons between a model that included only MIR data (Model M1), a model that incorporated different energy sink predictors, such as body weight, body weight change, and milk energy (Model M2), and an extended model that incorporated both energy sinks and MIR data (Model M3). Models M2 and M3 also included various cow level variables (stage of lactation, age at calving, parity) such that any improvement in model performance from M2 to M3, whether through a smaller root mean squared error (RMSE) or a greater squared predictive correlation (R2), could indicate a potential benefit of MIR to predict residual feed intake. The data used in our study originated from a multi-institutional project on the genetics of feed efficiency in US Holsteins. Analyses were conducted on 2 different trait definitions based on different period lengths: averaged across weeks vs. averaged across 28-d. Specifically, there were 19,942 weekly records on 1,812 cows across 46 experiments or cohorts and 3,724 28-d records on 1,700 cows across 43 different cohorts. The cross-validation analyses involved 3 different k-fold schemes. First, a 10-fold cow-independent cross-validation was conducted whereby all records from any one cow were kept together in either training or test sets. Similarly, a 10-fold experiment-independent cross-validation kept entire experiments together whereas a 4-fold herd-independent cross-validation kept entire herds together in either training or test sets. Based on cow-independent cross-validation for both weekly and 28-d DMI, adding MIR predictors to energy sinks (Models M3 vs M2) significantly (P < 10-10) reduced average RMSE to 1.59 kg and increased average R2 to 0.89. However, adding MIR to energy sinks (M3) to predict DMI either within an experiment-independent or herd-independent cross-validation scheme seemed to demonstrate no merit (P > 0.05) compared with an energy sink model (M2) for either R2 or RMSE (respectively, 0.68 and 2.55 kg for M2 in herd-independent scheme). We further noted that with broader cross-validation schemes, i.e., from cow-independent to experiment-independent to herd-independent schemes, the mean and slope bias increased. Given that proxy DMI phenotypes for cows would need to be almost entirely generated in herds having no DMI or training data of their own, herd-independent cross-validation assessments of predictive performance should be emphasized. Hence, more research on predictive algorithms suitable for broader cross-validation schemes and a more earnest effort on calibration of spectrophotometers against each other should be considered.

Predicting daily milk yield for primiparous cows using data of within-herd relatives to capture genotype-by-environment interactions.

This study develops and illustrates a hybrid k-medoids, random forest, and support vector regression (K-R-S) approach for predicting the lactation curves of individual primiparous cows within a targeted environment using monthly milk production data from their dams and paternal siblings. The model simulation and evaluation were based on historical test-day (TD) milk production data from 2010 to 2016 for 260 Wisconsin dairy farms. Data from older paternal siblings and dams were used to create family units (n = 6,400) of individual calves, from which their future performance was predicted. Test-day milk yield (MY) records from 2010 to 2014 were used for model training, whereas monthly milk production records of Holstein calves born in 2014 were used for model evaluation. The K-R-S hybrid approach was used to generate MY predictions for 5 randomly selected batches of 320 primiparous cows, which were used to evaluate model performance at the individual cow level by cross-validation. Across all 5 batches, the mean absolute error and the root mean square error of the K-R-S predictions were lower (by 24.2 and 23.4%, respectively) than that of the mean daily MY of paternal siblings. The K-R-S predictions of TD MY were closer to actual values 74.2 ± 2.0% of the time, as compared with means of paternal siblings'. The correlation between actual TD MY and K-R-S predictions was greater (0.34 ± 0.01) than the correlation between the actual yield and the mean of paternal siblings (0.08 ± 0.01). The results of this study demonstrate the effectiveness of the K-R-S hybrid approach for predicting future first-lactation MY of dairy calves in management applications, such as milk production forecasting or decision-support simulation, using only monthly TD yields of within-herd relatives and in the absence of detailed genomic data.

Feeding behavior parameters and temporal patterns in mid-lactation Holstein cows across a range of residual feed intake values.

Residual feed intake (RFI) is a measurement of the difference between actual and predicted feed intake when adjusted for energy sinks; more efficient cows eat less than predicted (low RFI) and inefficient cows eat more than predicted (high RFI). Data evaluating the relationship between RFI and feeding behaviors (FB) are limited in dairy cattle; therefore, the objective of this study was to determine daily and temporal FB in mid-lactation Holstein cows across a range of RFI values. Mid-lactation Holstein cows (n = 592 multiparous; 304 primiparous) were enrolled in 17 cohorts at 97 ± 26 d in milk (± standard deviation), and all cows within a cohort were fed a common diet using automated feeding bins. Cow RFI was calculated as the difference between predicted and observed dry matter intake (DMI) after accounting for parity, days in milk, milk energy, metabolic body weight and change, and experiment. The associations between RFI and FB at the level of meals and daily totals were evaluated using mixed models with the fixed effect of RFI and the random effects of cow and cohort. Daily temporal FB analyses were conducted using 2-h blocks and analyzed using mixed models with the fixed effects of RFI, time, RFI × time, and cohort, and the random effect of cow (cohort). There was a positive linear association between RFI and DMI in multiparous cows and a positive quadratic relationship in primiparous cows, where the rate of increase in DMI was less at higher RFI. Eating rate, DMI per meal, and size of the largest daily meal were positively associated with RFI. Daily temporal analysis of FB revealed an interaction between RFI and time for eating rate in multiparous and primiparous cows. The eating rate increased with greater RFI at 11 of 12 time points throughout the day, and eating rate differed across RFI between multiple time points. There tended to be an interaction between RFI and time for eating time and bin visits in multiparous cows but not primiparous cows. Overall, there was a time effect for all FB variables, where DMI, eating time and rate, and bin visits were greatest after the initial daily feeding at 1200 h, increased slightly after each milking, and reached a nadir at 0600 h (6 h before feeding). Considering the relationship between RFI and eating rate, additional efforts to determine cost-effective methods of quantifying eating rate in group-housed dairy cows is warranted. Further investigation is also warranted to determine if management strategies to alter FB, especially eating rate, can be effective in increasing feed efficiency in lactating dairy cattle.

Predicting dry matter intake in mid-lactation Holstein cows using point-in-time data streams available on dairy farms.

Quantifying dry matter intake (DMI) in lactating dairy cows is important for determining feed efficiency; however, there are no methods for economically quantifying individual cow DMI on dairy farms where cows are group-fed. Attempts have been made to model DMI using cow factors, milk production, milk infrared spectra, and behavioral sensors with reasonable success. Other data streams are available on the farm that may contribute to DMI predictions. In this study, our objective was to model DMI with multiple linear regression using data from a single point-in-time that can easily be accessed on-farm. Candidate predictor variables included cow descriptors, milk yield and composition, milk fatty acid profile, and production and efficiency predicting transmitting abilities (PTA). Observations of DMI were obtained from 350 cows across 6 cohorts using individual feed bunks. The cow to bunk ratio was 2:1, with an overall bunk occupation rate of 32% throughout the day. The following models were developed sequentially with milk data obtained from a single morning milking and other data from the same day: model B (production, metabolic body weight, body condition score, lactation category, and week of lactation), model BC [model B + fatty acid (FA) content], model BY (model B + FA yield), model BPE (model B + production and efficiency PTA), model BYP (model BY + production PTA), model BYE (model BY + efficiency PTA), and model BYPE (model BY + production and efficiency PTA). Outcome variables predicted in these models were the DMI on the previous day or current day relative to the morning milk sample. The predictions for DMI on the previous day outperformed current day DMI in every model for which they were both determined. Addition of milk FA and PTA as candidate predictor variable types to the models resulted in enhanced predictive ability, with incremental enhancements when combined. The most robust model (BYPE) included cow descriptors, protein and FA yields, and PTA for milk and residual feed intake. Model BYPE described 21 to 32% more of the variation in DMI (based on concordance correlation coefficient) than when other common DMI models were applied to the same data set. Overall, reasonable performance of models including single point-in-time cow descriptors, milk and FA production, and production and efficiency PTA commonly available to dairy farmers through dairy herd improvement programs offer an opportunity for on-farm prediction of DMI, yet further improvement may be possible.

Effect of diet energy level and genomic residual feed intake on bred Holstein dairy heifer growth and feed efficiency.

The objective of this study was to determine growth, feed intake, and feed efficiency of postbred dairy heifers with different genomic residual feed intake (RFI) predicted as a lactating cow when offered diets differing in energy density. Postbred Holstein heifers (n = 128, ages 14-20 mo) were blocked by initial weight (high, medium-high, medium-low, and low) with 32 heifers per block. Each weight block was sorted by RFI (high or low) to obtain 2 pens of heifers with high and low genomically predicted RFI within each block (8 heifers per pen). Low RFI heifers were expected to have greater feed efficiency than high RFI heifers. Dietary treatments consisted of a higher energy control diet based on corn silage and alfalfa haylage [HE; 62.7% total digestible nutrients, 11.8% crude protein, and 45.6% neutral detergent fiber; dry matter (DM) basis], and a lower energy diet diluted with straw (LE; 57.0% total digestible nutrients, 11.7% crude protein, and 50.1% neutral detergent fiber; DM basis). Each pen within a block was randomly allocated a diet treatment to obtain a 2 × 2 factorial arrangement (2 RFI levels and 2 dietary energy levels). Diets were offered in a 120-d trial. Dry matter intake by heifers was affected by diet (11.0 vs. 10.0 kg/d for HE and LE, respectively) but not by RFI or the interaction of RFI and diet. Daily gain was affected by the interaction of RFI and diet, with low RFI heifers gaining more than high RFI heifers when fed LE (0.94 vs. 0.85 kg/d for low and high RFI, respectively), but no difference for RFI groups when fed HE (1.16 vs. 1.19 kg/d for low and high RFI, respectively). Respective feed efficiencies were improved for low RFI compared with high RFI heifers when fed LE (10.6 vs. 11.8 kg of feed DM/kg of gain), but no effect of RFI was found when fed HE (9.4 vs. 9.5 kg of DM/kg of gain for high and low RFI, respectively). No effect of RFI or diet on first-lactation performance through 150 DIM was observed. Based on these results, the feed efficiency of heifers having different genomic RFI may be dependent on diet energy level, whereby low RFI heifers utilized the LE diet more efficiently. The higher fiber straw (LE) diet controlled intake and maintained more desirable heifer weight gains. This suggests that selection for improved RFI in lactating cows may improve feed efficiency in growing heifers when fed to meet growth goals of 0.9 to 1.0 kg of gain/d.

Multiple-trait random regression modeling of feed efficiency in US Holsteins.

Residual feed intake (RFI) and feed saved (FS) are important feed efficiency traits that have been increasingly considered in genetic improvement programs. Future sustainability of these genetic evaluations will depend upon greater flexibility to accommodate sparsely recorded dry matter intake (DMI) records on many more cows, especially from commercial environments. Recent multiple-trait random regression (MTRR) modeling developments have facilitated days in milk (DIM)-specific inferences on RFI and FS, particularly in modeling the effect of change in metabolic body weight (MBW). The MTRR analyses, using daily data on the core traits of DMI, MBW, and milk energy (MilkE), were conducted separately for 2,532 primiparous and 2,379 multiparous US Holstein cows from 50 to 200 DIM. Estimated MTRR variance components were used to derive genetic RFI and FS and DIM-specific genetic partial regressions of DMI on MBW, MilkE, and change in MBW. Estimated daily heritabilities of RFI and FS varied across lactation for both primiparous (0.05-0.07 and 0.11-0.17, respectively) and multiparous (0.03-0.13 and 0.10-0.17, respectively) cows. Genetic correlations of RFI across DIM varied (>0.05) widely compared with FS (>0.54) within either parity class. Heritability estimates based on average lactation-wise measures were substantially larger than daily heritabilities, ranging from 0.17 to 0.25 for RFI and from 0.35 to 0.41 for FS. The partial genetic regression coefficients of DMI on MBW (0.11 to 0.16 kg/kg0.75 for primiparous and 0.12 to 0.14 kg/kg0.75 for multiparous cows) and of DMI on MilkE (0.45 to 0.68 kg/Mcal for primiparous and 0.36 to 0.61 kg/Mcal for multiparous cows) also varied across lactation. In spite of the computational challenges encountered with MTRR, the model potentially facilitates an efficient strategy for harnessing more data involving a wide variety of data recording scenarios for genetic evaluations on feed efficiency.

Comparison of methods to predict feed intake and residual feed intake using behavioral and metabolite data in addition to classical performance variables.

Predicting dry matter intake (DMI) and feed efficiency by leveraging the use of data streams available on farm could aid efforts to improve the feed efficiency of dairy cattle. Residual feed intake (RFI) is the difference between predicted and observed feed intake after accounting for body size, body weight change, and milk production, making it a valuable metric for feed efficiency research. Our objective was to develop and evaluate DMI and RFI prediction models using multiple linear regression (MLR), partial least squares regression, artificial neural networks, and stacked ensembles using different combinations of cow descriptive, performance, sensor-derived behavioral (SMARTBOW; Zoetis), and blood metabolite data. Data were collected from mid-lactation Holstein cows (n = 124; 102 multiparous, 22 primiparous) split equally between 2 replicates of 45-d duration with ad libitum access to feed. Within each predictive approach, 4 data streams were added in sequence: dataset M (week of lactation, parity, milk yield, and milk components), dataset MB (dataset M plus body condition score and metabolic body weight), dataset MBS (dataset MB plus sensor-derived behavioral variables), and dataset MBSP (dataset MBS plus physiological blood metabolites). The combination of 4 datasets and 4 analytical approaches resulted in 16 analyses of DMI and RFI, using variables averaged within cow across the study period. Additional models using weekly averaged data within cow and study were built using all predictive approaches for datasets M, MB, and MBS. Model performance was assessed using the coefficient of determination, concordance correlation coefficient, and root mean square error of prediction. Predictive models of DMI performed similarly across all approaches, and models using dataset MBS had the greatest model performance. The best approach-dataset combination was MLR-dataset MBS, although several models performed similarly. Weekly DMI models had the greatest performance with MLR and partial least squares regression approaches. Dataset MBS models had incrementally better performance than datasets MB and M. Within each approach-dataset combination, models with DMI averaged over the study period had slightly greater model performance than DMI averaged weekly. Predictive performance of all RFI models was poor, but slight improvements when using MLR applied to dataset MBS suggest that rumination and activity behaviors may explain some of the variation in RFI. Overall, similar performance of MLR, compared with machine learning techniques, indicates MLR may be sufficient to predict DMI. The improvement in model performance with each additional data stream supports the idea of integrating data streams to improve model predictions and farm management decisions.

Enteric methane, lactation performances, digestibility, and metabolism of nitrogen and energy of Holsteins and Jerseys fed 2 levels of forage fiber from alfalfa silage or corn silage.

Our objective was to determine the effects of replacing alfalfa silage (AS) neutral detergent fiber (NDF) with corn silage (CS) NDF at 2 levels of forage NDF (FNDF) on enteric methane (CH4), lactation performance, ruminal fluid characteristics, digestibility, and metabolism of N and energy in Holstein and Jersey cows. Twelve Holstein and 12 Jersey cows (all primiparous and mid-lactation) were used in a triplicated split-plot 4 × 4 Latin square experiment, where breed and diet formed the main and subplots, respectively. The 4 iso-nitrogenous and iso-starch dietary treatments were arranged as a 2 × 2 factorial with 2 levels of FNDF [19 (low FNDF, LF) and 24% (high FNDF, HF) of dry matter] and 2 sources of FNDF (70:30 and 30:70 ratio of AS NDF to CS NDF). Soyhull (non-forage NDF) and corn grain were respectively used to keep dietary NDF and starch content similar across diets. Total collection of feces and urine over 3 d was performed on 8 cows (1 Latin square from each breed). The difference in dry matter intake (DMI) between Holsteins and Jerseys was greater when fed AS than CS. Compared with Jerseys, Holstein cows had greater body weight (48%), DMI (34%), fat- and protein-corrected milk (FPCM; 31%) and CH4 production (22%; 471 vs. 385 g/d). However, breed did not affect CH4 intensity (g/kg of FPCM) or yield (g/kg of DMI), nutrient digestibility, and N partitioning. Compared with HF, LF-fed cows had greater DMI (10%), N intake (8%), and FPCM (5%), but they were 5% less efficient (both FPCM/DMI and milk N/intake N). Compared with HF, LF-fed cows excreted 11 and 17% less urinary N (g/d and % of N intake, respectively). In spite of lower (2.5%) acetate and higher (10%) propionate (mol/100 mol ruminal volatile fatty acids) LF-fed cows had greater (6%) CH4 production (g/d) than did HF-fed cows, most likely due to increased DMI, as affected mainly by the soyhulls. Compared with AS, CS-fed cows had greater DMI (7%) and FPCM (4%), but they were less efficient (5%), and CH4 yield (g/kg of DMI) was reduced by 8%. In addition, per unit of gross energy intake, CS-fed cows lost less urinary energy (15%) and CH energy (11%) than did AS-fed cows. We concluded that, in contrast to level and source of FNDF, breed did not affect digestive and metabolic efficiencies, and, furthermore, neither breed nor dietary treatments affected CH4 intensity. The tradeoff between CH4 and N losses may have implications in future studies assessing the environmental effects of milk production when approached from a whole-farm perspective.

Genomic prediction of residual feed intake in US Holstein dairy cattle.

Genomic selection is an important tool to introduce feed efficiency into dairy cattle breeding. The goals of the current research are to estimate genomic breeding values of residual feed intake (RFI) and to assess the prediction reliability for RFI in the US Holstein population. The RFI data were collected from 4,823 lactations of 3,947 Holstein cows in 9 research herds in the United States, and were pre-adjusted to remove phenotypic correlations with milk energy, metabolic body weight, body weight change, and for several environmental effects. In the current analyses, genomic predicted transmitting abilities of milk energy and of body weight composite were included into the RFI model to further remove the genetic correlations that remained between RFI and these energy sinks. In the first part of the analyses, a national genomic evaluation for RFI was conducted for all the Holsteins in the national database using a standard multi-step genomic evaluation method and 60,671 SNP list. In the second part of the study, a single-step genomic prediction method was applied to estimate genomic breeding values of RFI for all cows with phenotypes, 5,252 elite young bulls, 4,029 young heifers, as well as their ancestors in the pedigree, using a high-density genotype chip. Theoretical prediction reliabilities were calculated for all the studied animals in the single-step genomic prediction by direct inversion of the mixed model equations. In the results, breeding values were estimated for 1.6 million genotyped Holsteins and 60 million ungenotyped Holsteins, The genomic predicted transmitting ability correlations between RFI and other traits in the index (e.g., fertility) are generally low, indicating minor correlated responses on other index traits when selecting for RFI. Genomic prediction reliabilities for RFI averaged 34% for all phenotyped animals and 13% for all 1.6 million genotyped animals. Including genomic information increased the prediction reliabilities for RFI compared with using only pedigree information. All bulls had low reliabilities, and averaged to only 16% for the top 100 net merit progeny-tested bulls. Analyses using single-step genomic prediction and high-density genotypes gave similar results to those obtained from the national evaluation. The average theoretical reliability for RFI was 18% among the elite young bulls under 5 yr old, being lower in the younger generations of elite bulls compared with older bulls. To conclude, the size of the reference population and its relationship to the predicted population remain as the limiting factors in the genomic prediction for RFI. Continued collection of feed intake data is necessary so that reliabilities can be maintained due to close relationships of phenotyped animals with breeding stock. Considering the currently low prediction reliability and high cost of data collection, focusing RFI data collection on relatives of elite bulls that will have the greatest genetic contribution to the next generation will give more gains and profit.

Inclusion of herdmate data improves genomic prediction for milk-production and feed-efficiency traits within North American dairy herds.

Genomic data are widely available in the dairy industry and provide a cost-effective means of predicting genetic merit to inform selection decisions and increase genetic gains. As more dairy farms adopt genomic selection practices, dairy producers will soon have genomic data available on all of the animals within their herds. This is a very rich, but currently underused, source of information. Herdmates provide an excellent indication of how a selection candidate's genetics will perform within a given herd, noting that herdmates often include close relatives that share a similar environment. The study objective was to evaluate the utility of incorporating herdmate data into genomic predictions in a data set composed of 3,303 Holsteins from one herd in Canada and 6 herds throughout the United States. Within-herd prediction accuracy was assessed for milk-production and feed-efficiency traits determined from genomic best linear unbiased prediction under 4 different scenarios. Scenario 1 did not include herdmates in the training population. Scenarios 2 through 4 included herdmates in the training population, and scenarios 3 and 4 also included modeling of herd-specific marker effects. Leave-one-out cross validation was used to maximize the number of herdmates in the training population in scenarios 2 through 4, while maintaining constant training population size with scenario 1. Results from the present study reveal the importance of incorporating herdmate data into genomic evaluations. Inclusion of herdmates in the training population improved mean within-herd prediction accuracy for milk-production traits (± standard error) by 0.08 ± 0.03 (milk yield), 0.07 ± 0.03 (fat percentage), and 0.05 ± 0.01 (protein percentage) and feed-efficiency traits by 0.07 ± 0.02 (milk energy), 0.03 ± 0.02 (DMI), and 0.08 ± 0.01 (metabolic body weight). Modeling herd-specific marker effects further improved mean within-herd prediction accuracy for milk yield and energy by 0.03 ± 0.01 and 0.02 ± 0.01, respectively. Herds with higher within-herd heritability and low genomic correlation with the remaining herds benefitted most from the inclusion of herdmate data.

Effect of diet energy density and genomic residual feed intake on prebred dairy heifer feed efficiency, growth, and manure excretion.

The objective of this study was to determine the growth, feed efficiency, and manure excretion of prebred dairy heifers with differing predicted genomic residual feed intakes (RFI) when offered diets differing in energy density. Prebred Holstein heifers (n = 128, ages 4 to 8 mo) were blocked by weight (low, medium-low, medium-high, or high) with 32 heifers per block. Heifers in each weight block were grouped by RFI and randomly assigned to obtain 2 pens of high (HRFI) and 2 pens of low RFI (LRFI) heifers within each block (8 heifers/pen). Heifers with LRFI were hypothesized to have greater feed efficiency than HRFI heifers. Dietary treatments were a high-energy diet (HE; 66.6% total digestible nutrients, 14.0% crude protein, and 36.3% neutral detergent fiber, dry matter basis) and a low-energy diet (LE; 63.8% total digestible nutrients, 13.5% crude protein, and 41.2% neutral detergent fiber, dry matter basis). Each pen of heifers was randomly assigned to a treatment to obtain a 2 × 2 factorial arrangement (2 RFI levels × 2 diet energy densities). Diets were offered in a 120-d trial. Dry matter intake was not affected by diet, RFI, or their interaction. Average daily gain (ADG) was affected by diet, with heifers fed HE having greater ADG than heifers fed LE. In addition, RFI affected ADG, with LRFI heifers having greater ADG than HRFI heifers, whereas the interaction of RFI and diet was not significant. Feed efficiency was improved for heifers fed the HE diet, but it was not affected by RFI or the interaction of RFI and diet. Overall, feed efficiency of prebred heifers was not dependent on predicted genomic RFI, because the greater ADG of LRFI heifers was accompanied by slightly higher dry matter intake. Feed efficiency of heifers was reduced when heifers were fed the LE diet, but this resulted in more optimal ADG compared with the HE diet fed for ad libitum intake.

High-density genome-wide association study for residual feed intake in Holstein dairy cattle.

Improving feed efficiency (FE) of dairy cattle may boost farm profitability and reduce the environmental footprint of the dairy industry. Residual feed intake (RFI), a candidate FE trait in dairy cattle, can be defined to be genetically uncorrelated with major energy sink traits (e.g., milk production, body weight) by including genomic predicted transmitting ability of such traits in genetic analyses for RFI. We examined the genetic basis of RFI through genome-wide association (GWA) analyses and post-GWA enrichment analyses and identified candidate genes and biological pathways associated with RFI in dairy cattle. Data were collected from 4,823 lactations of 3,947 Holstein cows in 9 research herds in the United States. Of these cows, 3,555 were genotyped and were imputed to a high-density list of 312,614 SNP. We used a single-step GWA method to combine information from genotyped and nongenotyped animals with phenotypes as well as their ancestors' information. The estimated genomic breeding values from a single-step genomic BLUP were back-solved to obtain the individual SNP effects for RFI. The proportion of genetic variance explained by each 5-SNP sliding window was also calculated for RFI. Our GWA analyses suggested that RFI is a highly polygenic trait regulated by many genes with small effects. The closest genes to the top SNP and sliding windows were associated with dry matter intake (DMI), RFI, energy homeostasis and energy balance regulation, digestion and metabolism of carbohydrates and proteins, immune regulation, leptin signaling, mitochondrial ATP activities, rumen development, skeletal muscle development, and spermatogenesis. The region of 40.7 to 41.5 Mb on BTA25 (UMD3.1 reference genome) was the top associated region for RFI. The closest genes to this region, CARD11 and EIF3B, were previously shown to be related to RFI of dairy cattle and FE of broilers, respectively. Another candidate region, 57.7 to 58.2 Mb on BTA18, which is associated with DMI and leptin signaling, was also associated with RFI in this study. Post-GWA enrichment analyses used a sum-based marker-set test based on 4 public annotation databases: Gene Ontology, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways, Reactome pathways, and medical subject heading (MeSH) terms. Results of these analyses were consistent with those from the top GWA signals. Across the 4 databases, GWA signals for RFI were highly enriched in the biosynthesis and metabolism of amino acids and proteins, digestion and metabolism of carbohydrates, skeletal development, mitochondrial electron transport, immunity, rumen bacteria activities, and sperm motility. Our findings offer novel insight into the genetic basis of RFI and identify candidate regions and biological pathways associated with RFI in dairy cattle.

Genome-wide association analyses based on a multiple-trait approach for modeling feed efficiency.

Genome-wide association (GWA) of feed efficiency (FE) could help target important genomic regions influencing FE. Data provided by an international dairy FE research consortium consisted of phenotypic records on dry matter intakes (DMI), milk energy (MILKE), and metabolic body weight (MBW) on 6,937 cows from 16 stations in 4 counties. Of these cows, 4,916 had genotypes on 57,347 single nucleotide polymorphism (SNP) markers. We compared a GWA analysis based on the more classical residual feed intake (RFI) model with one based on a previously proposed multiple trait (MT) approach for modeling FE using an alternative measure (DMI|MILKE,MBW). Both models were based on a single-step genomic BLUP procedure that allowed the use of phenotypes from both genotyped and nongenotyped cows. Estimated effects for single SNP markers were small and not statistically important but virtually identical for either FE measure (RFI vs. DMI|MILKE,MBW). However, upon further refining this analysis to develop joint tests within nonoverlapping 1-Mb windows, significant associations were detected between either measure of FE with a window on each of Bos taurus autosomes BTA12 and BTA26. There was, as expected, no overlap between detected genomic regions for DMI|MILKE,MBW and genomic regions influencing the energy sink traits (i.e., MILKE and MBW) because of orthogonal relationships clearly defined between the various traits. Conversely, GWA inferences on DMI can be demonstrated to be partly driven by genetic associations between DMI with these same energy sink traits, thereby having clear implications when comparing GWA studies on DMI to GWA studies on FE-like measures such as RFI.

A 100-Year Review: Methods and impact of genetic selection in dairy cattle-From daughter-dam comparisons to deep learning algorithms.

In the early 1900s, breed society herdbooks had been established and milk-recording programs were in their infancy. Farmers wanted to improve the productivity of their cattle, but the foundations of population genetics, quantitative genetics, and animal breeding had not been laid. Early animal breeders struggled to identify genetically superior families using performance records that were influenced by local environmental conditions and herd-specific management practices. Daughter-dam comparisons were used for more than 30 yr and, although genetic progress was minimal, the attention given to performance recording, genetic theory, and statistical methods paid off in future years. Contemporary (herdmate) comparison methods allowed more accurate accounting for environmental factors and genetic progress began to accelerate when these methods were coupled with artificial insemination and progeny testing. Advances in computing facilitated the implementation of mixed linear models that used pedigree and performance data optimally and enabled accurate selection decisions. Sequencing of the bovine genome led to a revolution in dairy cattle breeding, and the pace of scientific discovery and genetic progress accelerated rapidly. Pedigree-based models have given way to whole-genome prediction, and Bayesian regression models and machine learning algorithms have joined mixed linear models in the toolbox of modern animal breeders. Future developments will likely include elucidation of the mechanisms of genetic inheritance and epigenetic modification in key biological pathways, and genomic data will be used with data from on-farm sensors to facilitate precision management on modern dairy farms.

The genetic and biological basis of feed efficiency in mid-lactation Holstein dairy cows.

The objective of this study was to identify genomic regions and candidate genes associated with feed efficiency in lactating Holstein cows. In total, 4,916 cows with actual or imputed genotypes for 60,671 single nucleotide polymorphisms having individual feed intake, milk yield, milk composition, and body weight records were used in this study. Cows were from research herds located in the United States, Canada, the Netherlands, and the United Kingdom. Feed efficiency, defined as residual feed intake (RFI), was calculated within location as the residual of the regression of dry matter intake (DMI) on milk energy (MilkE), metabolic body weight (MBW), change in body weight, and systematic effects. For RFI, DMI, MilkE, and MBW, bivariate analyses were performed considering each trait as a separate trait within parity group to estimate variance components and genetic correlations between them. Animal relationships were established using a genomic relationship matrix. Genome-wide association studies were performed separately by parity group for RFI, DMI, MilkE, and MBW using the Bayes B method with a prior assumption that 1% of single nucleotide polymorphisms have a nonzero effect. One-megabase windows with greatest percentage of the total genetic variation explained by the markers (TGVM) were identified, and adjacent windows with large proportion of the TGVM were combined and reanalyzed. Heritability estimates for RFI were 0.14 (±0.03; ±SE) in primiparous cows and 0.13 (±0.03) in multiparous cows. Genetic correlations between primiparous and multiparous cows were 0.76 for RFI, 0.78 for DMI, 0.92 for MBW, and 0.61 for MilkE. No single 1-Mb window explained a significant proportion of the TGVM for RFI; however, after combining windows, significance was met on Bos taurus autosome 27 in primiparous cows, and nearly reached on Bos taurus autosome 4 in multiparous cows. Among other genes, these regions contain ß-3 adrenergic receptor and the physiological candidate gene, leptin, respectively. Between the 2 parity groups, 3 of the 10 windows with the largest effects on DMI neighbored windows affecting RFI, but were not in the top 10 regions for MilkE or MBW. This result suggests a genetic basis for feed intake that is unrelated to energy consumption required for milk production or expected maintenance as determined by MBW. In conclusion, feed efficiency measured as RFI is a polygenic trait exhibiting a dynamic genetic basis and genetic variation distinct from that underlying expected maintenance requirements and milk energy output.

Use of genotype × environment interaction model to accommodate genetic heterogeneity for residual feed intake, dry matter intake, net energy in milk, and metabolic body weight in dairy cattle.

Feed efficiency in dairy cattle has gained much attention recently. Due to the cost-prohibitive measurement of individual feed intakes, combining data from multiple countries is often necessary to ensure an adequate reference population. It may then be essential to model genetic heterogeneity when making inferences about feed efficiency or selecting efficient cattle using genomic information. In this study, we constructed a marker × environment interaction model that decomposed marker effects into main effects and interaction components that were specific to each environment. We compared environment-specific variance component estimates and prediction accuracies from the interaction model analyses, an across-environment analyses ignoring population stratification, and a within-environment analyses using an international feed efficiency data set. Phenotypes included residual feed intake, dry matter intake, net energy in milk, and metabolic body weight from 3,656 cows measured in 3 broadly defined environments: North America (NAM), the Netherlands (NLD), and Scotland (SAC). Genotypic data included 57,574 single nucleotide polymorphisms per animal. The interaction model gave the highest prediction accuracy for metabolic body weight, which had the largest estimated heritabilities ranging from 0.37 to 0.55. The within-environment model performed the best when predicting residual feed intake, which had the lowest estimated heritabilities ranging from 0.13 to 0.41. For traits (dry matter intake and net energy in milk) with intermediate estimated heritabilities (0.21 to 0.50 and 0.17 to 0.53, respectively), performance of the 3 models was comparable. Genomic correlations between environments also were computed using variance component estimates from the interaction model. Averaged across all traits, genomic correlations were highest between NAM and NLD, and lowest between NAM and SAC. In conclusion, the interaction model provided a novel way to evaluate traits measured in multiple environments in which genetic heterogeneity may exist. This model allowed estimation of environment-specific parameters and provided genomic predictions that approached or exceeded the accuracy of competing within- or across-environment models.

Modeling genetic and nongenetic variation of feed efficiency and its partial relationships between component traits as a function of management and environmental factors.

Feed efficiency (FE), characterized as the fraction of feed nutrients converted into salable milk or meat, is of increasing economic importance in the dairy industry. We conjecture that FE is a complex trait whose variation and relationships or partial efficiencies (PE) involving the conversion of dry matter intake to milk energy and metabolic body weight may be highly heterogeneous across environments or management scenarios. In this study, a hierarchical Bayesian multivariate mixed model was proposed to jointly infer upon such heterogeneity at both genetic and nongenetic levels on PE and variance components (VC). The heterogeneity was modeled by embedding mixed effects specifications on PE and VC in addition to those directly specified on the component traits. We validated the model by simulation and applied it to a joint analysis of a dairy FE consortium data set with 5,088 Holstein cows from 13 research stations in Canada, the Netherlands, the United Kingdom, and the United States. Although no differences were detected among research stations for PE at the genetic level, some evidence was found of heterogeneity in residual PE. Furthermore, substantial heterogeneity in VC across stations, parities, and ration was observed with heritability estimates of FE ranging from 0.16 to 0.46 across stations.

Prediction of whole-genome risk for selection and management of hyperketonemia in Holstein dairy cattle.

Hyperketonemia (HYK), a common early postpartum health disorder characterized by elevated blood concentrations of ß-hydroxybutyrate (BHB), affects millions of dairy cows worldwide and leads to significant economic losses and animal welfare concerns. In this study, blood concentrations of BHB were assessed for 1,453 Holstein cows using electronic handheld meters at four time points between 5 and 18 days postpartum. Incidence rates of subclinical (1.2 ≤ maximum BHB ≤ 2.9 mmol/L) and clinical ketosis (maximum BHB ≥ 3.0 mmol/L) were 24.0 and 2.4%, respectively. Variance components, estimated breeding values, and predicted HYK phenotypes were computed on the original, square-root, and binary scales. Heritability estimates for HYK ranged from 0.058 to 0.072 in pedigree-based analyses, as compared to estimates that ranged from 0.071 to 0.093 when pedigrees were augmented with 60,671 single nucleotide polymorphism genotypes of 959 cows and 801 male ancestors. On average, predicted HYK phenotypes from the genome-enhanced analysis ranged from 0.55 mmol/L for first-parity cows in the best contemporary group to 1.40 mmol/L for fourth-parity cows in the worst contemporary group. Genome-enhanced predictions of HYK phenotypes were more closely associated with actual phenotypes than pedigree-based predictions in five-fold cross-validation, and transforming phenotypes to reduce skewness and kurtosis also improved predictive ability. This study demonstrates the feasibility of using repeated cowside measurement of blood BHB concentration in early lactation to construct a reference population that can be used to estimate HYK breeding values for genomic selection programmes and predict HYK phenotypes for genome-guided management decisions.

Genome-wide association mapping and pathway analysis of leukosis incidence in a US Holstein cattle population.

Bovine leukosis virus is an oncogenic virus that infects B cells, causing bovine leukosis disease. This disease is known to have a negative impact on dairy cattle production and, because no treatment or vaccine is available, finding a possible genetic solution is important. Our objective was to perform a comprehensive genetic analysis of leukosis incidence in dairy cattle. Data on leukosis occurrence, pedigree and molecular information were combined into multitrait GBLUP models with milk yield (MY) and somatic cell score (SCS) to estimate genetic parameters and to perform whole-genome scans and pathway analysis. Leukosis data were available for 11 554 Holsteins daughters of 3002 sires from 112 herds in 16 US states. Genotypes from a 60K SNP panel were available for 961 of those bulls as well as for 2039 additional bulls. Heritability for leukosis incidence was estimated at about 8%, and the genetic correlations of leukosis disease incidence with MY and SCS were moderate at 0.18 and 0.20 respectively. The genome-wide scan indicated that leukosis is a complex trait, possibly modulated by many genes. The gene set analysis identified many functional terms that showed significant enrichment of genes associated with leukosis. Many of these terms, such as G-Protein Coupled Receptor Signaling Pathway, Regulation of Nucleotide Metabolic Process and different calcium-related processes, are known to be related to retrovirus infection. Overall, our findings contribute to a better understanding of the genetic architecture of this complex disease. The functional categories associated with leukosis may be useful in future studies on fine mapping of genes and development of dairy cattle breeding strategies.

Short communication: Genetic correlation of bovine leukosis incidence with somatic cell score and milk yield in a US Holstein population.

Bovine leukosis (BL) is a retroviral disease caused by the bovine leukosis virus (BLV), which affects only cattle. Dairy cows positive for BL produce less milk and have more days open than cows negative for BL. In addition, the virus also affects the immune system and causes weaker response to vaccines. Heritability estimates of BL incidence have been reported for Jersey and Holstein populations at about 0.08, indicating an important genetic component that can potentially be exploited to reduce the prevalence of the disease. However, before BL is used in selection programs, it is important to study its genetic associations with other economically important traits such that correlated responses to selection can be predicted. Hence, this study aimed to estimate the genetic correlations of BL with milk yield (MY) and with somatic cell score (SCS). Data of a commercial assay (ELISA) used to detect BLV antibodies in milk samples were obtained from Antel BioSystems (Lansing, MI). The data included continuous milk ELISA scores and binary milk ELISA results for 11,554 cows from 112 dairy herds across 16 US states. Continuous and binary milk ELISA were analyzed with linear and threshold models, respectively, together with MY and SCS using multitrait animal models. Genetic correlations (posterior means ± standard deviations) between BL incidence and MY were 0.17 ± 0.077 and 0.14 ± 0.076 using ELISA scores and results, respectively; with SCS, such estimates were 0.20 ± 0.081 and 0.17 ± 0.079, respectively. In summary, the results indicate that selection for higher MY may lead to increased BLV prevalence in dairy herds, but that the inclusion of BL (or SCS as an indicator trait) in selection indexes may help attenuate this problem.