Modeling heterogeneous (co)variances from adjacent-SNP groups improves genomic prediction for milk protein composition traits

A breed-of-origin of alleles model that includes crossbred data improves predictive ability for crossbred animals in a multi-breed population

BACKGROUND

Recently, crossbred animals have begun to be used as parents in the next generations of dairy and beef cattle systems, which has increased the interest in predicting the genetic merit of those animals. The primary objective of this study was to investigate three available methods for genomic prediction of crossbred animals. In the first two methods, SNP effects from within-breed evaluations are used by weighting them by the average breed proportions across the genome (BPM method) or by their breed-of-origin (BOM method). The third method differs from the BOM in that it estimates breed-specific SNP effects using purebred and crossbred data, considering the breed-of-origin of alleles (BOA method). For within-breed evaluations, and thus for BPM and BOM, 5948 Charolais, 6771 Limousin and 7552 Others (a combined population of other breeds) were used to estimate SNP effects separately within each breed. For the BOA, the purebreds' data were enhanced with data from ~ 4K, ~ 8K or ~ 18K crossbred animals. For each animal, its predictor of genetic merit (PGM) was estimated by considering the breed-specific SNP effects. Predictive ability and absence of bias were estimated for crossbreds and the Limousin and Charolais animals. Predictive ability was measured as the correlation between PGM and the adjusted phenotype, while the regression of the adjusted phenotype on PGM was estimated as a measure of bias.

RESULTS

With BPM and BOM, the predictive abilities for crossbreds were 0.468 and 0.472, respectively, and with the BOA method, they ranged from 0.490 to 0.510. The performance of the BOA method improved as the number of crossbred animals in the reference increased and with the use of the correlated approach, in which the correlation of SNP effects across the genome of the different breeds was considered. The slopes of regression for PGM on adjusted phenotypes for crossbreds showed overdispersion of the genetic merits for all methods but this bias tended to be reduced by the use of the BOA method and by increasing the number of crossbred animals.

CONCLUSIONS

For the estimation of the genetic merit of crossbred animals, the results from this study suggest that the BOA method that accommodates crossbred data can yield more accurate predictions than the methods that use SNP effects from separate within-breed evaluations.

Fitting Genomic Prediction Models with Different Marker Effects among Prefectures to Carcass Traits in Japanese Black Cattle

We fitted statistical models, which assumed single-nucleotide polymorphism (SNP) marker effects differing across the fattened steers marketed into different prefectures, to the records for cold carcass weight (CW) and marbling score (MS) of 1036, 733, and 279 Japanese Black fattened steers marketed into Tottori, Hiroshima, and Hyogo prefectures in Japan, respectively. Genotype data on 33,059 SNPs was used. Five models that assume only common SNP effects to all the steers (model 1), common effects plus SNP effects differing between the steers marketed into Hyogo prefecture and others (model 2), only the SNP effects differing between Hyogo steers and others (model 3), common effects plus SNP effects specific to each prefecture (model 4), and only the effects specific to each prefecture (model 5) were exploited. For both traits, slightly lower values of residual variance than that of model 1 were estimated when fitting all other models. Estimated genetic correlation among the prefectures in models 2 and 4 ranged to 0.53 to 0.71, all <0.8. These results might support that the SNP effects differ among the prefectures to some degree, although we discussed the necessity of careful consideration to interpret the current results.

Genomic prediction in Nordic Red dairy cattle considering breed origin of alleles

This study investigated the reliability of genomic prediction (GP) using breed origin of alleles (BOA) approach in the Nordic Red (RDC) population, which has an admixed population structure. The RDC population consists of animals with varying degrees of genetic materials from the Danish Red (RDM), Swedish Red (SRB), Finnish Ayrshire (FAY), and Holstein (HOL) because bulls have been used across the breeds. The BOA approach was tested using 39,550 RDC animals in the reference population and 11,786 in the validation population. Deregressed proofs (DRP) of milk, fat and protein were used as response variable for GP. Direct genomic breeding values (DGV) for animals in the validation population were calculated with (BOA model) or without (joint model) considering breed origin of alleles. The joint model assumed homogeneous marker effects and a single set of marker effects were estimated, whereas BOA model assumed heterogeneous marker effects, and different sets of marker effects were estimated across the breeds. For the BOA approach, we tested scenarios assuming both correlated (BOA_cor) and uncorrelated (BOA_uncor) marker effects between the breeds. Additionally, we investigated GP using a standard Illumina 50K chip and including SNP selected from imputed whole-genome sequencing (50K+WGS). We also studied the effect of estimating (co)variances for genome regions of different sizes to exploit the information of the genome regions contributing to the (co)variance between the breeds. Region sizes were set as 1 SNP, a group of 30 or 100 adjacent SNP, or the whole genome. Reliability of DGV was measured as squared correlations between DGV and DRP divided by the reliability of DRP. Across the 3 traits, in general, RS30 and RS100 SNP yielded the highest reliabilities. Including WGS SNP improved reliabilities in almost all scenarios (0.297 on average for 50K and 0.307 on average for 50K+WGS). The BOA_uncor (0.233 on average) was inferior to the joint model (0.339 on average), but the reliabilities obtained using BOA_cor (0.334 on average) in most cases were not significantly different from those obtained using the joint model. The results indicate that both including additional whole-genome sequencing SNP and dividing the genome into fixed regions improve GP in the RDC. The BOA models have the potential to increase the reliability of GP, but the benefit is limited in populations with a high exchange of genetic material for a long time, as is the case for RDC.

Genomic prediction using a reference population of multiple pure breeds and admixed individuals

BACKGROUND

In dairy cattle populations in which crossbreeding has been used, animals show some level of diversity in their origins. In rotational crossbreeding, for instance, crossbred dams are mated with purebred sires from different pure breeds, and the genetic composition of crossbred animals is an admixture of the breeds included in the rotation. How to use the data of such individuals in genomic evaluations is still an open question. In this study, we aimed at providing methodologies for the use of data from crossbred individuals with an admixed genetic background together with data from multiple pure breeds, for the purpose of genomic evaluations for both purebred and crossbred animals. A three-breed rotational crossbreeding system was mimicked using simulations based on animals genotyped with the 50 K single nucleotide polymorphism (SNP) chip.

RESULTS

For purebred populations, within-breed genomic predictions generally led to higher accuracies than those from multi-breed predictions using combined data of pure breeds. Adding admixed population's (MIX) data to the combined pure breed data considering MIX as a different breed led to higher accuracies. When prediction models were able to account for breed origin of alleles, accuracies were generally higher than those from combining all available data, depending on the correlation of quantitative trait loci (QTL) effects between the breeds. Accuracies varied when using SNP effects from any of the pure breeds to predict the breeding values of MIX. Using those breed-specific SNP effects that were estimated separately in each pure breed, while accounting for breed origin of alleles for the selection candidates of MIX, generally improved the accuracies. Models that are able to accommodate MIX data with the breed origin of alleles approach generally led to higher accuracies than models without breed origin of alleles, depending on the correlation of QTL effects between the breeds.

CONCLUSIONS

Combining all available data, pure breeds' and admixed population's data, in a multi-breed reference population is beneficial for the estimation of breeding values for pure breeds with a small reference population. For MIX, such an approach can lead to higher accuracies than considering breed origin of alleles for the selection candidates, and using breed-specific SNP effects estimated separately in each pure breed. Including MIX data in the reference population of multiple breeds by considering the breed origin of alleles, accuracies can be further improved. Our findings are relevant for breeding programs in which crossbreeding is systematically applied, and also for populations that involve different subpopulations and between which exchange of genetic material is routine practice.

Genetic analysis on infrared-predicted milk minerals for Danish dairy cattle

A group of milk components that has shown potential to be predicted with milk spectra is milk minerals. Milk minerals are important for human health and cow health. Having an inexpensive and fast way to measure milk mineral concentrations would open doors for research, herd management, and selective breeding. The first aim of this study was to predict milk minerals with infrared milk spectra. Additionally, milk minerals were predicted with infrared-predicted fat, protein, and lactose content. The second aim was to perform a genetic analysis on infrared-predicted milk minerals, to identify QTL, and estimate variance components. For training and validating a multibreed prediction model for individual milk minerals, 264 Danish Jersey cows and 254 Danish Holstein cows were used. Partial least square regression prediction models were built for Ca, Cu, Fe, K, Mg, Mn, Na, P, Se, and Zn based on 80% of the cows, selected randomly. Prediction models were externally validated with 8 herds based on the remaining 20% of the cows. The prediction models were applied on a population of approximately 1,400 Danish Holstein cows with 5,600 infrared spectral records and 1,700 Danish Jersey cows with 7,200 infrared spectral records. Cows from this population had 50k imputed genotypes. Prediction accuracy was good for P and Ca, with external R² ≥ 0.80 and a relative prediction error of 5.4% for P and 6.3% for Ca. Prediction was moderately good for Na with an external R² of 0.63, and a relative error of 18.8%. Prediction accuracies of milk minerals based on infrared-predicted fat, protein, and lactose content were considerably lower than those based on the infrared milk spectra. This shows that the milk infrared spectrum contains valuable information on milk minerals, which is currently not used. Heritability for infrared-predicted Ca, Na, and P varied from low (0.13) to moderate (0.36). Several QTL for infrared-predicted milk minerals were observed that have been associated with gold standard milk minerals previously. In conclusion, this study has shown infrared milk spectra were good at predicting Ca, Na, and P in milk. Infrared-predicted Ca, Na, and P had low to moderate heritability estimates.

Vitamin B12 and transcobalamin in bovine milk: Genetic variation and genome-wide association with loci along the genome

In human nutrition, bovine milk is an essential source of bioavailable vitamin B₁₂ and B₁₂-binding proteins, including transcobalamin. In this study, we estimated genetic parameters for milk content of vitamin B₁₂ and transcobalamin using milk samples from 341 and 663 Danish Holstein cows, respectively. Additionally, we conducted whole-genome association analysis to identify SNP and genes associated with vitamin B₁₂ and transcobalamin. Our results indicated moderate to high heritability for vitamin B₁₂ (0.37 ± 0.18) and transcobalamin (0.61 ± 0.13) content in the Danish Holstein. With a significance threshold of -log₁₀ P-value > 5.87, significant associations were detected between SNP in Bos taurus autosome (BTA)17 and the log-transformed transcobalamin content of milk; no significant association was detected for vitamin B₁₂. The significant region in BTA17 was imputed to full sequence for further fine mapping, and the SNP with the most significant associations to transcobalamin were assigned to the transcobalamin 2 (TCN2) gene.

Factors influencing milk osteopontin concentration based on measurements from Danish Holstein cows

Our objective was to determine the content of the bioactive protein osteopontin (OPN) in bovine milk and identify factors influencing its concentration. OPN is expressed in many tissues and body fluids, with by far the highest concentrations in milk. OPN plays a role in immunological and developmental processes and it has been associated with several milk production traits and lactation persistency in cows. In the present study, we report the development of an enzyme linked immunosorbent assay (ELISA) for measurement of OPN in bovine milk. The method was used to determine the concentration of OPN in milk from 661 individual Danish Holstein cows. The median OPN level was determined to 21.9 mg/l with a pronounced level of individual variation ranging from 0.4 mg/l to 67.8 mg/l. Breeding for increased OPN in cow's milk is of significant interest, however, the heritability of OPN in milk was found to be relatively low, with an estimated value of 0.19 in the current dataset. The variation explained by the herd was also found to be low suggesting that OPN levels are not affected by farm management or feeding. Interestingly, the concentration of OPN was found to increase with days in milk and to decrease with parity.

Multi-trait single-step genomic prediction accounting for heterogeneous (co)variances over the genome

Widely used genomic prediction models may not properly account for heterogeneous (co)variance structure across the genome. Models such as BayesA and BayesB assume locus-specific variance, which are highly influenced by the prior for (co)variance of single nucleotide polymorphism (SNP) effect, regardless of the size of data. Models such as BayesC or GBLUP assume a common (co)variance for a proportion (BayesC) or all (GBLUP) of the SNP effects. In this study, we propose a multi-trait Bayesian whole genome regression method (BayesN0), which is based on grouping a number of predefined SNPs to account for heterogeneous (co)variance structure across the genome. This model was also implemented in single-step Bayesian regression (ssBayesN0). For practical implementation, we considered multi-trait single-step SNPBLUP models, using (co)variance estimates from BayesN0 or ssBayesN0. Genotype data were simulated using haplotypes on first five chromosomes of 2200 Danish Holstein cattle, and phenotypes were simulated for two traits with heritabilities 0.1 or 0.4, assuming 200 quantitative trait loci (QTL). We compared prediction accuracy from different prediction models and different region sizes (one SNP, 100 SNPs, one chromosome or whole genome). In general, highest accuracies were obtained when 100 adjacent SNPs were grouped together. The ssBayesN0 improved accuracies over BayesN0, and using (co)variance estimates from ssBayesN0 generally yielded higher accuracies than using (co)variance estimates from BayesN0, for the 100 SNPs region size. Our results suggest that it could be a good strategy to estimate (co)variance components from ssBayesN0, and then to use those estimates in genomic prediction using multi-trait single-step SNPBLUP, in routine genomic evaluations.

Genetic analysis of orotic acid predicted with Fourier transform infrared milk spectra

Fourier transform infrared spectral analysis is a cheap and fast method to predict milk composition. A not very well studied milk component is orotic acid. Orotic acid is an intermediate in the biosynthesis pathway of pyrimidine nucleotides and is an indicator for the metabolic cattle disorder deficiency of uridine monophosphate synthase. The function of orotic acid in milk and its effect on calf health, health of humans consuming milk or milk products, manufacturing properties of milk, and its potential as an indicator trait are largely unknown. The aims of this study were to determine if milk orotic acid can be predicted from infrared milk spectra and to perform a large-scale phenotypic and genetic analysis of infrared-predicted milk orotic acid. An infrared prediction model for orotic acid was built using a training population of 292 Danish Holstein and 299 Danish Jersey cows, and a validation population of 381 Danish Holstein cows. Milk orotic acid concentration was determined with nuclear magnetic resonance spectroscopy. For genetic analysis of infrared orotic acid, 3 study populations were used: 3,210 Danish Holstein cows, 3,360 Danish Jersey cows, and 1,349 Dutch Holstein Friesian cows. Using partial least square regression, a prediction model for orotic acid was built with 18 latent variables. The error of the prediction for the infrared model varied from 1.0 to 3.2 mg/L, and the accuracy varied from 0.68 to 0.86. Heritability of infrared orotic acid predicted with the standardized prediction model was 0.18 for Danish Holstein, 0.09 for Danish Jersey, and 0.37 for Dutch Holstein Friesian. We conclude that milk orotic acid can be predicted with moderate to good accuracy based on infrared milk spectra and that infrared-predicted orotic acid is heritable. The availability of a cheap and fast method to predict milk orotic acid opens up possibilities to study the largely unknown functions of milk orotic acid.

Genome-wide association study on Fourier transform infrared milk spectra for two Danish dairy cattle breeds

Background

Infrared spectral analysis of milk is cheap, fast, and accurate. Infrared light interacts with chemical bonds present inside the milk, which means that Fourier transform infrared milk spectra are a reflection of the chemical composition of milk. Heritability of Fourier transform infrared milk spectra has been analysed previously. Further genetic analysis of Fourier transform infrared milk spectra could give us a better insight in the genes underlying milk composition. Breed influences milk composition, yet not much is known about the effect of breed on Fourier transform infrared milk spectra. Improved understanding of the effect of breed on Fourier transform infrared milk spectra could enhance efficient application of Fourier transform infrared milk spectra. The aim of this study is to perform a genome wide association study on a selection of wavenumbers for Danish Holstein and Danish Jersey. This will improve our understanding of the genetics underlying milk composition in these two dairy cattle breeds.

Results

For each breed separately, fifteen wavenumbers were analysed. Overall, more quantitative trait loci were observed for Danish Jersey compared to Danish Holstein. For both breeds, the majority of the wavenumbers was most strongly associated to a genomic region on BTA 14 harbouring DGAT1. Furthermore, for both breeds most quantitative trait loci were observed for wavenumbers that interact with the chemical bond C-O. For Danish Jersey, wavenumbers that interact with C-H were associated to genes that are involved in fatty acid synthesis, such as AGPAT3, AGPAT6, PPARGC1A, SREBF1, and FADS1. For wavenumbers which interact with –OH, associations were observed to genomic regions that have been linked to alpha-lactalbumin.

Conclusions

The current study identified many quantitative trait loci that underlie Fourier transform infrared milk spectra, and thus milk composition. Differences were observed between groups of wavenumbers that interact with different chemical bonds. Both overlapping and different QTL were observed for Danish Holstein and Danish Jersey.

Reliability of genomic prediction for milk fatty acid composition by using a multi-population reference and incorporating GWAS results

Background

Large-scale phenotyping for detailed milk fatty acid (FA) composition is difficult due to expensive and time-consuming analytical techniques. Reliability of genomic prediction is often low for traits that are expensive/difficult to measure and for breeds with a small reference population size. An effective method to increase reference population size could be to combine datasets from different populations. Prediction models might also benefit from incorporation of information on the biological underpinnings of quantitative traits. Genome-wide association studies (GWAS) show that genomic regions on Bos taurus chromosomes (BTA) 14, 19 and 26 underlie substantial proportions of the genetic variation in milk FA traits. Genomic prediction models that incorporate such results could enable improved prediction accuracy in spite of limited reference population sizes. In this study, we combine gas chromatography quantified FA samples from the Chinese, Danish and Dutch Holstein populations and implement a genomic feature best linear unbiased prediction (GFBLUP) model that incorporates variants on BTA14, 19 and 26 as genomic features for which random genetic effects are estimated separately. Prediction reliabilities were compared to those estimated with traditional GBLUP models.

Results

Predictions using a multi-population reference and a traditional GBLUP model resulted in average gains in prediction reliability of 10% points in the Dutch, 8% points in the Danish and 1% point in the Chinese populations compared to predictions based on population-specific references. Compared to the traditional GBLUP, implementation of the GFBLUP model with a multi-population reference led to further increases in prediction reliability of up to 38% points in the Dutch, 23% points in the Danish and 13% points in the Chinese populations. Prediction reliabilities from the GFBLUP model were moderate to high across the FA traits analyzed.

Conclusions

Our study shows that it is possible to predict genetic merits for milk FA traits with reasonable accuracy by combining related populations of a breed and using models that incorporate GWAS results. Our findings indicate that international collaborations that facilitate access to multi-population datasets could be highly beneficial to the implementation of genomic selection for detailed milk composition traits.

Electronic supplementary material

The online version of this article (10.1186/s12711-019-0460-z) contains supplementary material, which is available to authorized users.

Genomic Prediction Using Multi-trait Weighted GBLUP Accounting for Heterogeneous Variances and Covariances Across the Genome

Implicit assumption of common (co)variance for all loci in multi-trait Genomic Best Linear Unbiased Prediction (GBLUP) results in a genomic relationship matrix (G) that is common to all traits. When this assumption is violated, Bayesian whole genome regression methods may be superior to GBLUP by accounting for unequal (co)variance for all loci or genome regions. This study aimed to develop a strategy to improve the accuracy of GBLUP for multi-trait genomic prediction, using (co)variance estimates of SNP effects from Bayesian whole genome regression methods. Five generations (G1-G5, test populations) of genotype data were available by simulations based on data of 2,200 Danish Holstein cows (G0, reference population). Two correlated traits with heritabilities of 0.1 or 0.4, and a genetic correlation of 0.45 were generated. First, SNP effects and breeding values were estimated using BayesAS method, assuming (co)variance was the same for SNPs within a genome region, and different between regions. Region size was set as one SNP, 100 SNPs, a whole chromosome or whole genome. Second, posterior (co)variances of SNP effects were used to weight SNPs in construction of G matrices. In general, region size of 100 SNPs led to highest prediction accuracies using BayesAS, and wGBLUP outperformed GBLUP at this region size. Our results suggest that when genetic architectures of traits favor Bayesian methods, the accuracy of multi-trait GBLUP can be as high as the Bayesian method if SNPs are weighted by the Bayesian posterior (co)variances.