Impact of rare and low-frequency sequence variants on reliability of genomic prediction in dairy cattle

The effect of marker types and density on genomic prediction and GWAS of key performance traits in tetraploid potato

Genomic prediction and genome-wide association studies are becoming widely employed in potato key performance trait QTL identifications and to support potato breeding using genomic selection. Elite cultivars are tetraploid and highly heterozygous but also share many common ancestors and generation-spanning inbreeding events, resulting from the clonal propagation of potatoes through seed potatoes. Consequentially, many SNP markers are not in a 1:1 relationship with a single allele variant but shared over several alleles that might exert varying effects on a given trait. The impact of such redundant "diluted" predictors on the statistical models underpinning genome-wide association studies (GWAS) and genomic prediction has scarcely been evaluated despite the potential impact on model accuracy and performance. We evaluated the impact of marker location, marker type, and marker density on the genomic prediction and GWAS of five key performance traits in tetraploid potato (chipping quality, dry matter content, length/width ratio, senescence, and yield). A 762-offspring panel of a diallel cross of 18 elite cultivars was genotyped by sequencing, and markers were annotated according to a reference genome. Genomic prediction models (GBLUP) were trained on four marker subsets [non-synonymous (29,553 SNPs), synonymous (31,229), non-coding (32,388), and a combination], and robustness to marker reduction was investigated. Single-marker regression GWAS was performed for each trait and marker subset. The best cross-validated prediction correlation coefficients of 0.54, 0.75, 0.49, 0.35, and 0.28 were obtained for chipping quality, dry matter content, length/width ratio, senescence, and yield, respectively. The trait prediction abilities were similar across all marker types, with only non-synonymous variants improving yield predictive ability by 16%. Marker reduction response did not depend on marker type but rather on trait. Traits with high predictive abilities, e.g., dry matter content, reached a plateau using fewer markers than traits with intermediate-low correlations, such as yield. The predictions were unbiased across all traits, marker types, and all marker densities >100 SNPs. Our results suggest that using non-synonymous variants does not enhance the performance of genomic prediction of most traits. The major known QTLs were identified by GWAS and were reproducible across exonic and whole-genome variant sets for dry matter content, length/width ratio, and senescence. In contrast, minor QTL detection was marker type dependent.

Transmission ratio distortion regions in the context of genomic evaluation and their effects on reproductive traits in cattle

Transmission ratio distortion (TRD), which is a deviation from Mendelian expectations, has been associated with basic mechanisms of life such as sperm and ova fertility and viability at developmental stages of the reproductive cycle. In this study different models including TRD regions were tested for different reproductive traits [days from first service to conception (FSTC), number of services, first service nonreturn rate (NRR), and stillbirth (SB)]. Thus, in addition to a basic model with systematic and random effects, including genetic effects modeled through a genomic relationship matrix, we developed 2 additional models, including a second genomic relationship matrix based on TRD regions, and TRD regions as a random effect assuming heterogeneous variances. The analyses were performed with 10,623 cows and 1,520 bulls genotyped for 47,910 SNPs, 590 TRD regions, and several records ranging from 9,587 (FSTC) to 19,667 (SB). The results of this study showed the ability of TRD regions to capture some additional genetic variance for some traits; however, this did not translate into higher accuracy for genomic prediction. This could be explained by the nature of TRD itself, which may arise in different stages of the reproductive cycle. Nevertheless, important effects of TRD regions were found on SB (31 regions) and NRR (18 regions) when comparing at-risk versus control matings, especially for regions with allelic TRD pattern. Particularly for NRR, the probability of observing nonpregnant cow increases by up to 27% for specific TRD regions, and the probability of observing stillbirth increased by up to 254%. These results support the relevance of several TRD regions on some reproductive traits, especially those with allelic patterns that have not received as much attention as recessive TRD patterns.

Haplotype blocks for genomic prediction: a comparative evaluation in multiple crop datasets

In modern plant breeding, genomic selection is becoming the gold standard for selection of superior genotypes. The basis for genomic prediction models is a set of phenotyped lines along with their genotypic profile. With high marker density and linkage disequilibrium (LD) between markers, genotype data in breeding populations tends to exhibit considerable redundancy. Therefore, interest is growing in the use of haplotype blocks to overcome redundancy by summarizing co-inherited features. Moreover, haplotype blocks can help to capture local epistasis caused by interacting loci. Here, we compared genomic prediction methods that either used single SNPs or haplotype blocks with regards to their prediction accuracy for important traits in crop datasets. We used four published datasets from canola, maize, wheat and soybean. Different approaches to construct haplotype blocks were compared, including blocks based on LD, physical distance, number of adjacent markers and the algorithms implemented in the software "Haploview" and "HaploBlocker". The tested prediction methods included Genomic Best Linear Unbiased Prediction (GBLUP), Extended GBLUP to account for additive by additive epistasis (EGBLUP), Bayesian LASSO and Reproducing Kernel Hilbert Space (RKHS) regression. We found improved prediction accuracy in some traits when using haplotype blocks compared to SNP-based predictions, however the magnitude of improvement was very trait- and model-specific. Especially in settings with low marker density, haplotype blocks can improve genomic prediction accuracy. In most cases, physically large haplotype blocks yielded a strong decrease in prediction accuracy. Especially when prediction accuracy varies greatly across different prediction models, prediction based on haplotype blocks can improve prediction accuracy of underperforming models. However, there is no "best" method to build haplotype blocks, since prediction accuracy varied considerably across methods and traits. Hence, criteria used to define haplotype blocks should not be viewed as fixed biological parameters, but rather as hyperparameters that need to be adjusted for every dataset.

Weighted kernels improve multi-environment genomic prediction

Crucial to variety improvement programs is the reliable and accurate prediction of genotype's performance across environments. However, due to the impactful presence of genotype by environment (G×E) interaction that dictates how changes in expression and function of genes influence target traits in different environments, prediction performance of genomic selection (GS) using single-environment models often falls short. Furthermore, despite the successes of genome-wide association studies (GWAS), the genetic insights derived from genome-to-phenome mapping have not yet been incorporated in predictive analytics, making GS models that use Gaussian kernel primarily an estimator of genomic similarity, instead of the underlying genetics characteristics of the populations. Here, we developed a GS framework that, in addition to capturing the overall genomic relationship, can capitalize on the signal of genetic associations of the phenotypic variation as well as the genetic characteristics of the populations. The capacity of predicting the performance of populations across environments was demonstrated by an overall gain in predictability up to 31% for the winter wheat DH population. Compared to Gaussian kernels, we showed that our multi-environment weighted kernels could better leverage the significance of genetic associations and yielded a marked improvement of 4-33% in prediction accuracy for half-sib families. Furthermore, the flexibility incorporated in our Bayesian implementation provides the generalizable capacity required for predicting multiple highly genetic heterogeneous populations across environments, allowing reliable GS for genetic improvement programs that have no access to genetically uniform material.

Impact of linkage disequilibrium heterogeneity along the genome on genomic prediction and heritability estimation

BACKGROUND

Compared to medium-density single nucleotide polymorphism (SNP) data, high-density SNP data contain abundant genetic variants and provide more information for the genetic evaluation of livestock, but it has been shown that they do not confer any advantage for genomic prediction and heritability estimation. One possible reason is the uneven distribution of the linkage disequilibrium (LD) along the genome, i.e., LD heterogeneity among regions. The aim of this study was to effectively use genome-wide SNP data for genomic prediction and heritability estimation by using models that control LD heterogeneity among regions.

METHODS

The LD-adjusted kinship (LDAK) and LD-stratified multicomponent (LDS) models were used to control LD heterogeneity among regions and were compared with the classical model that has no such control. Simulated and real traits of 2000 dairy cattle individuals with imputed high-density (770K) SNP data were used. Five types of phenotypes were simulated, which were controlled by very strongly, strongly, moderately, weakly and very weakly tagged causal variants, respectively. The performances of the models with high- and medium-density (50K) panels were compared to verify that the models that controlled LD heterogeneity among regions were more effective with high-density data.

RESULTS

Compared to the medium-density panel, the use of the high-density panel did not improve and even decreased prediction accuracies and heritability estimates from the classical model for both simulated and real traits. Compared to the classical model, LDS effectively improved the accuracy of genomic predictions and unbiasedness of heritability estimates, regardless of the genetic architecture of the trait. LDAK applies only to traits that are mainly controlled by weakly tagged causal variants, but is still less effective than LDS for this type of trait. Compared with the classical model, LDS improved prediction accuracy by about 13% for simulated phenotypes and by 0.3 to ~ 10.7% for real traits with the high-density panel, and by ~ 1% for simulated phenotypes and by - 0.1 to ~ 6.9% for real traits with the medium-density panel.

CONCLUSIONS

Grouping SNPs based on regional LD to construct the LD-stratified multicomponent model can effectively eliminate the adverse effects of LD heterogeneity among regions, and greatly improve the efficiency of high-density SNP data for genomic prediction and heritability estimation.

On the use of whole-genome sequence data for across-breed genomic prediction and fine-scale mapping of QTL

Background

Whole-genome sequence (WGS) data are increasingly available on large numbers of individuals in animal and plant breeding and in human genetics through second-generation resequencing technologies, 1000 genomes projects, and large-scale genotype imputation from lower marker densities. Here, we present a computationally fast implementation of a variable selection genomic prediction method, that could handle WGS data on more than 35,000 individuals, test its accuracy for across-breed predictions and assess its quantitative trait locus (QTL) mapping precision.

Methods

The Monte Carlo Markov chain (MCMC) variable selection model (Bayes GC) fits simultaneously a genomic best linear unbiased prediction (GBLUP) term, i.e. a polygenic effect whose correlations are described by a genomic relationship matrix (G), and a Bayes C term, i.e. a set of single nucleotide polymorphisms (SNPs) with large effects selected by the model. Computational speed is improved by a Metropolis–Hastings sampling that directs computations to the SNPs, which are, a priori, most likely to be included into the model. Speed is also improved by running many relatively short MCMC chains. Memory requirements are reduced by storing the genotype matrix in binary form. The model was tested on a WGS dataset containing Holstein, Jersey and Australian Red cattle. The data contained 4,809,520 genotypes on 35,549 individuals together with their milk, fat and protein yields, and fat and protein percentage traits.

Results

The prediction accuracies of the Jersey individuals improved by 1.5% when using across-breed GBLUP compared to within-breed predictions. Using WGS instead of 600 k SNP-chip data yielded on average a 3% accuracy improvement for Australian Red cows. QTL were fine-mapped by locating the SNP with the highest posterior probability of being included in the model. Various QTL known from the literature were rediscovered, and a new SNP affecting milk production was discovered on chromosome 20 at 34.501126 Mb. Due to the high mapping precision, it was clear that many of the discovered QTL were the same across the five dairy traits.

Conclusions

Across-breed Bayes GC genomic prediction improved prediction accuracies compared to GBLUP. The combination of across-breed WGS data and Bayesian genomic prediction proved remarkably effective for the fine-mapping of QTL.

Imputation for sequencing variants preselected to a customized low-density chip

The sequencing variants preselected from association analyses and bioinformatics analyses could improve genomic prediction. In this study, the imputation of sequencing SNPs preselected from major dairy breeds in Denmark-Finland-Sweden (DFS) and France (FRA) was investigated for both contemporary animals and old bulls in Danish Jersey. For contemporary animals, a two-step imputation which first imputed to 54 K and then to 54 K + DFS + FRA SNPs achieved highest accuracy. Correlations between observed and imputed genotypes were 91.6% for DFS SNPs and 87.6% for FRA SNPs, while concordance rates were 96.6% for DFS SNPs and 93.5% for FRA SNPs. The SNPs with lower minor allele frequency (MAF) tended to have lower correlations but higher concordance rates. For old bulls, imputation for DFS and FRA SNPs were relatively accurate even for bulls without progenies (correlations higher than 97.2% and concordance rates higher than 98.4%). For contemporary animals, given limited imputation accuracy of preselected sequencing SNPs especially for SNPs with low MAF, it would be a good strategy to directly genotype preselected sequencing SNPs with a customized SNP chip. For old bulls, given high imputation accuracy for preselected sequencing SNPs with all MAF ranges, it would be unnecessary to re-genotype preselected sequencing SNPs.

A multi-breed reference panel and additional rare variants maximize imputation accuracy in cattle

Background

During the last decade, the use of common-variant array-based single nucleotide polymorphism (SNP) genotyping in the beef and dairy industries has produced an astounding amount of medium-to-low density genomic data. Although low-density assays work well in the context of genomic prediction, they are less useful for detecting and mapping causal variants and the effects of rare variants are not captured. The objective of this project was to maximize the accuracies of genotype imputation from medium- and low-density assays to the marker set obtained by combining two high-density research assays (~ 850,000 SNPs), the Illumina BovineHD and the GGP-F250 assays, which contains a large proportion of rare and potentially functional variants and for which the assay design is described here. This 850 K SNP set is useful for both imputation to sequence-level genotypes and direct downstream analysis.

Results

We found that a large multi-breed composite imputation reference panel that includes 36,131 samples with either BovineHD and/or GGP-F250 genotypes significantly increased imputation accuracy compared with a within-breed reference panel, particularly at variants with low minor allele frequencies. Individual animal imputation accuracies were maximized when more genetically similar animals were represented in the composite reference panel, particularly with complete 850 K genotypes. The addition of rare variants from the GGP-F250 assay to our composite reference panel significantly increased the imputation accuracy of rare variants that are exclusively present on the BovineHD assay. In addition, we show that an assay marker density of 50 K SNPs balances cost and accuracy for imputation to 850 K.

Conclusions

Using high-density genotypes on all available individuals in a multi-breed reference panel maximized imputation accuracy for tested cattle populations. Admixed animals or those from breeds with a limited representation in the composite reference panel were still imputed at high accuracy, which is expected to further increase as the reference panel expands. We anticipate that the addition of rare variants from the GGP-F250 assay will increase the accuracy of imputation to sequence level.