Malena Erbe

Improving accuracy of genomic predictions within and between dairy cattle breeds with imputed high-density single nucleotide polymorphism panels.

Achieving accurate genomic estimated breeding values for dairy cattle requires a very large reference population of genotyped and phenotyped individuals. Assembling such reference populations has been achieved for breeds such as Holstein, but is challenging for breeds with fewer individuals. An alternative is to use a multi-breed reference population, such that smaller breeds gain some advantage in accuracy of genomic estimated breeding values (GEBV) from information from larger breeds. However, this requires that marker-quantitative trait loci associations persist across breeds. Here, we assessed the gain in accuracy of GEBV in Jersey cattle as a result of using a combined Holstein and Jersey reference population, with either 39,745 or 624,213 single nucleotide polymorphism (SNP) markers. The surrogate used for accuracy was the correlation of GEBV with daughter trait deviations in a validation population. Two methods were used to predict breeding values, either a genomic BLUP (GBLUP_mod), or a new method, BayesR, which used a mixture of normal distributions as the prior for SNP effects, including one distribution that set SNP effects to zero. The GBLUP_mod method scaled both the genomic relationship matrix and the additive relationship matrix to a base at the time the breeds diverged, and regressed the genomic relationship matrix to account for sampling errors in estimating relationship coefficients due to a finite number of markers, before combining the 2 matrices. Although these modifications did result in less biased breeding values for Jerseys compared with an unmodified genomic relationship matrix, BayesR gave the highest accuracies of GEBV for the 3 traits investigated (milk yield, fat yield, and protein yield), with an average increase in accuracy compared with GBLUP_mod across the 3 traits of 0.05 for both Jerseys and Holsteins. The advantage was limited for either Jerseys or Holsteins in using 624,213 SNP rather than 39,745 SNP (0.01 for Holsteins and 0.03 for Jerseys, averaged across traits). Even this limited and nonsignificant advantage was only observed when BayesR was used. An alternative panel, which extracted the SNP in the transcribed part of the bovine genome from the 624,213 SNP panel (to give 58,532 SNP), performed better, with an increase in accuracy of 0.03 for Jerseys across traits. This panel captures much of the increased genomic content of the 624,213 SNP panel, with the advantage of a greatly reduced number of SNP effects to estimate. Taken together, using this panel, a combined breed reference and using BayesR rather than GBLUP_mod increased the accuracy of GEBV in Jerseys from 0.43 to 0.52, averaged across the 3 traits.

Genome-based prediction of testcross values in maize

This is the first large-scale experimental study on genome-based prediction of testcross values in an advanced cycle breeding population of maize. The study comprised testcross progenies of 1,380 doubled haploid lines of maize derived from 36 crosses and phenotyped for grain yield and grain dry matter content in seven locations. The lines were genotyped with 1,152 single nucleotide polymorphism markers. Pedigree data were available for three generations. We used best linear unbiased prediction and stratified cross-validation to evaluate the performance of prediction models differing in the modeling of relatedness between inbred lines and in the calculation of genome-based coefficients of similarity. The choice of similarity coefficient did not affect prediction accuracies. Models including genomic information yielded significantly higher prediction accuracies than the model based on pedigree information alone. Average prediction accuracies based on genomic data were high even for a complex trait like grain yield (0.72–0.74) when the cross-validation scheme allowed for a high degree of relatedness between the estimation and the test set. When predictions were performed across distantly related families, prediction accuracies decreased significantly (0.47–0.48). Prediction accuracies decreased with decreasing sample size but were still high when the population size was halved (0.67–0.69). The results from this study are encouraging with respect to genome-based prediction of the genetic value of untested lines in advanced cycle breeding populations and the implementation of genomic selection in the breeding process.

Improving the accuracy of whole genome prediction for complex traits using the results of genome wide association studies.

Utilizing the whole genomic variation of complex traits to predict the yet-to-be observed phenotypes or unobserved genetic values via whole genome prediction (WGP) and to infer the underlying genetic architecture via genome wide association study (GWAS) is an interesting and fast developing area in the context of human disease studies as well as in animal and plant breeding. Though thousands of significant loci for several species were detected via GWAS in the past decade, they were not used directly to improve WGP due to lack of proper models. Here, we propose a generalized way of building trait-specific genomic relationship matrices which can exploit GWAS results in WGP via a best linear unbiased prediction (BLUP) model for which we suggest the name BLUP|GA. Results from two illustrative examples show that using already existing GWAS results from public databases in BLUP|GA improved the accuracy of WGP for two out of the three model traits in a dairy cattle data set, and for nine out of the 11 traits in a rice diversity data set, compared to the reference methods GBLUP and BayesB. While BLUP|GA outperforms BayesB, its required computing time is comparable to GBLUP. Further simulation results suggest that accounting for publicly available GWAS results is potentially more useful for WGP utilizing smaller data sets and/or traits of low heritability, depending on the genetic architecture of the trait under consideration. To our knowledge, this is the first study incorporating public GWAS results formally into the standard GBLUP model and we think that the BLUP|GA approach deserves further investigations in animal breeding, plant breeding as well as human genetics.

Predicting Genetic Values: a Kernel-Based Best Linear Unbiased Prediction with Genomic Data

Genomic data provide a valuable source of information for modeling covariance structures, allowing a more accurate prediction of total genetic values (GVs). We apply the kriging concept, originally developed in the geostatistical context for predictions in the low-dimensional space, to the high-dimensional space spanned by genomic single nucleotide polymorphism (SNP) vectors and study its properties in different gene-action scenarios. Two different kriging methods [“universal kriging” (UK) and “simple kriging” (SK)] are presented. As a novelty, we suggest use of the family of Matérn covariance functions to model the covariance structure of SNP vectors. A genomic best linear unbiased prediction (GBLUP) is applied as a reference method. The three approaches are compared in a whole-genome simulation study considering additive, additive-dominance, and epistatic gene-action models. Predictive performance is measured in terms of correlation between true and predicted GVs and average true GVs of the individuals ranked best by prediction. We show that UK outperforms GBLUP in the presence of dominance and epistatic effects. In a limiting case, it is shown that the genomic covariance structure proposed by VanRaden (2008) can be considered as a covariance function with corresponding quadratic variogram. We also prove theoretically that if a specific linear relationship exists between covariance matrices for two linear mixed models, the GVs resulting from BLUP are linked by a scaling factor. Finally, the relation of kriging to other models is discussed and further options for modeling the covariance structure, which might be more appropriate in the genomic context, are suggested.

Genome partitioning of genetic variation for milk production and composition traits in holstein cattle.

The objective of this study was to estimate the contribution of each autosome to genetic variation of milk yield, fat, and protein percentage and somatic cell score in Holstein cattle. Data on 2294 Holstein bulls genotyped for 39,557 autosomal markers were used. Three approaches were applied to estimate the proportion of genetic variance attributed to each chromosome. In two of them, marker-derived kinship coefficients were computed, using either marker genotypes observed on the whole genome or on subsets relative to each chromosome. Variance components were then estimated using residual maximum likelihood in method 1 or a regression-based approach in method 2. In method 3, genetic variances associated to each marker were estimated in a linear multiple regression approach, and then were summed up chromosome-wise. Generally, all chromosomes contributed to genetic variation. For most of the chromosomes, the amount of variance attributed to a chromosome was found to be proportional to its physical length. Nevertheless, for traits influenced by genes with very large effects a larger proportion of the genetic variance is expected to be associated with the chromosomes where these genes are. This is illustrated with the DGAT1 gene on BTA14 which is known to have a large effect on fat percentage in milk. The proportion of genetic variance for fat percentage associated with chromosome 14 was two to sevenfold (depending on the method) larger than would be predicted from chromosome size alone. Based on method 3 an approach is suggested to estimate the effective number of genes underlying the inheritance of the studied traits, yielding numbers between N ≈ 400 (for fat percentage) to N ≈ 900 (for milk yield). It is argued that these numbers are conservative lower bound estimates, but are in line with recent findings suggesting a highly polygenic background of production traits in dairy cattle.

A function accounting for training set size and marker density to model the average accuracy of genomic prediction.

Prediction of genomic breeding values is of major practical relevance in dairy cattle breeding. Deterministic equations have been suggested to predict the accuracy of genomic breeding values in a given design which are based on training set size, reliability of phenotypes, and the number of independent chromosome segments (). The aim of our study was to find a general deterministic equation for the average accuracy of genomic breeding values that also accounts for marker density and can be fitted empirically. Two data sets of 5′698 Holstein Friesian bulls genotyped with 50 K SNPs and 1′332 Brown Swiss bulls genotyped with 50 K SNPs and imputed to ∼600 K SNPs were available. Different k-fold (k = 2–10, 15, 20) cross-validation scenarios (50 replicates, random assignment) were performed using a genomic BLUP approach. A maximum likelihood approach was used to estimate the parameters of different prediction equations. The highest likelihood was obtained when using a modified form of the deterministic equation of Daetwyler et al. (2010), augmented by a weighting factor (w) based on the assumption that the maximum achievable accuracy is . The proportion of genetic variance captured by the complete SNP sets () was 0.76 to 0.82 for Holstein Friesian and 0.72 to 0.75 for Brown Swiss. When modifying the number of SNPs, w was found to be proportional to the log of the marker density up to a limit which is population and trait specific and was found to be reached with ∼20′000 SNPs in the Brown Swiss population studied.

Accuracy of whole-genome prediction using a genetic architecture-enhanced variance-covariance matrix.

Obtaining accurate predictions of unobserved genetic or phenotypic values for complex traits in animal, plant, and human populations is possible through whole-genome prediction (WGP), a combined analysis of genotypic and phenotypic data. Because the underlying genetic architecture of the trait of interest is an important factor affecting model selection, we propose a new strategy, termed BLUP|GA (BLUP-given genetic architecture), which can use genetic architecture information within the dataset at hand rather than from public sources. This is achieved by using a trait-specific covariance matrix (T), which is a weighted sum of a genetic architecture part (S matrix) and the realized relationship matrix (G). The algorithm of BLUP|GA (BLUP-given genetic architecture) is provided and illustrated with real and simulated datasets. Predictive ability of BLUP|GA was validated with three model traits in a dairy cattle dataset and 11 traits in three public datasets with a variety of genetic architectures and compared with GBLUP and other approaches. Results show that BLUP|GA outperformed GBLUP in 20 of 21 scenarios in the dairy cattle dataset and outperformed GBLUP, BayesA, and BayesB in 12 of 13 traits in the analyzed public datasets. Further analyses showed that the difference of accuracies for BLUP|GA and GBLUP significantly correlate with the distance between the T and G matrices. The new strategy applied in BLUP|GA is a favorable and flexible alternative to the standard GBLUP model, allowing to account for the genetic architecture of the quantitative trait under consideration when necessary. This feature is mainly due to the increased similarity between the trait-specific relationship matrix (T matrix) and the genetic relationship matrix at unobserved causal loci. Applying BLUP|GA in WGP would ease the burden of model selection.

Population Genomic Analyses Based on 1 Million SNPs in Commercial Egg Layers

Identifying signatures of selection can provide valuable insight about the genes or genomic regions that are or have been under selective pressure, which can lead to a better understanding of genotype-phenotype relationships. A common strategy for selection signature detection is to compare samples from several populations and search for genomic regions with outstanding genetic differentiation. Wrights fixation index, FST, is a useful index for evaluation of genetic differentiation between populations. The aim of this study was to detect selective signatures between different chicken groups based on SNP-wise FST calculation. A total of 96 individuals of three commercial layer breeds and 14 non-commercial fancy breeds were genotyped with three different 600K SNP-chips. After filtering a total of 1 million SNPs were available for FST calculation. Averages of FST values were calculated for overlapping windows. Comparisons of these were then conducted between commercial egg layers and non-commercial fancy breeds, as well as between white egg layers and brown egg layers. Comparing non-commercial and commercial breeds resulted in the detection of 630 selective signatures, while 656 selective signatures were detected in the comparison between the commercial egg-layer breeds. Annotation of selection signature regions revealed various genes corresponding to productions traits, for which layer breeds were selected. Among them were NCOA1, SREBF2 and RALGAPA1 associated with reproductive traits, broodiness and egg production. Furthermore, several of the detected genes were associated with growth and carcass traits, including POMC, PRKAB2, SPP1, IGF2, CAPN1, TGFb2 and IGFBP2. Our approach demonstrates that including different populations with a specific breeding history can provide a unique opportunity for a better understanding of farm animal selection.

Estimation of genetic parameters for novel functional traits in Brown Swiss cattle

The aim of this study was to estimate genetic parameters and accuracies of breeding values for a set of functional, behavior, and conformation traits in Brown Swiss cattle. These traits were milking speed, udder depth, position of labia, rank order in herd, general temperament, aggressiveness, milking temperament, and days to first heat. Data of 1,799 phenotyped Brown Swiss cows from 40 Swiss dairy herds were analyzed taking the complete pedigree into account. Estimated heritabilities were within the ranges reported in literature, with results at the high end of the reported values for some traits (e.g., milking speed: 0.42±0.06, udder depth: 0.42±0.06), whereas other traits were of low heritability and heritability estimates were of low accuracy (e.g., milking temperament: 0.04±0.04, days to first heat: 0.02±0.04). For most behavior traits, we found relatively high heritabilities (general temperament: 0.38±0.07, aggressiveness: 0.12±0.08, and rank order in herd: 0.16±0.06). Position of labia, arguably an indicator trait for pathological urovagina, was genetically analyzed in this study for the first time, and a moderate heritability (0.28±0.06) was estimated.

Accuracy of direct genomic values for functional traits in Brown Swiss cattle.

In this study, direct genomic values for the functional traits general temperament, milking temperament, aggressiveness, rank order in herd, milking speed, udder depth, position of labia, and days to first heat in Brown Swiss dairy cattle were estimated based on ~777,000 (777 K) single nucleotide polymorphism (SNP) information from 1,126 animals. Accuracy of direct genomic values was assessed by a 5-fold cross-validation with 10 replicates. Correlations between deregressed proofs and direct genomic values were 0.63 for general temperament, 0.73 for milking temperament, 0.69 for aggressiveness, 0.65 for rank order in herd, 0.69 for milking speed, 0.71 for udder depth, 0.66 for position of labia, and 0.74 for days to first heat. Using the information of ~54,000 (54K) SNP led to only marginal deviations in the observed accuracy. Trying to predict the 20% youngest bulls led to correlations of 0.55, 0.77, 0.73, 0.55, 0.64, 0.59, 0.67, and 0.77, respectively, for the traits listed above. Using a novel method to estimate the accuracy of a direct genomic value (defined as correlation between direct genomic value and true breeding value and accounting for the correlation between direct genomic values and conventional breeding values) revealed accuracies of 0.37, 0.20, 0.19, 0.27, 0.48, 0.45, 0.36, and 0.12, respectively, for the traits listed above. These values are much smaller but probably also more realistic than accuracies based on correlations, given the heritabilities and samples sizes in this study. Annotation of the largest estimated SNP effects revealed 2 candidate genes affecting the traits general temperament and days to first heat.