Strategies for single nucleotide polymorphism (SNP) genotyping to enhance genotype imputation in Gyr (Bos indicus) dairy cattle: Comparison of commercially available SNP chips

Abstract

Genotype imputation is widely used as a cost-effective strategy in genomic evaluation of cattle. Key determinants of imputation accuracies, such as linkage disequilibrium patterns, marker densities, and ascertainment bias, differ between Bos indicus and Bos taurus breeds. Consequently, there is a need to investigate effectiveness of genotype imputation in indicine breeds. Thus, the objective of the study was to investigate strategies and factors affecting the accuracy of genotype imputation in Gyr (Bos indicus) dairy cattle. Four imputation scenarios were studied using 471 sires and 1,644 dams genotyped on Illumina BovineHD (HD-777K; San Diego, CA) and BovineSNP50 (50K) chips, respectively. Scenarios were based on which reference high-density single nucleotide polymorphism (SNP) panel (HDP) should be adopted [HD-777K, 50K, and GeneSeek GGP-75Ki (Lincoln, NE)]. Depending on the scenario, validation animals had their genotypes masked for one of the lower-density panels: Illumina (3K, 7K, and 50K) and GeneSeek (SGGP-20Ki and GGP-75Ki). We randomly selected 171 sires as reference and 300 as validation for all the scenarios. Additionally, all sires were used as reference and the 1,644 dams were imputed for validation. Genotypes of 98 individuals with 4 and more offspring were completely masked and imputed. Imputation algorithms FImpute and Beagle v3.3 and v4 were used. Imputation accuracies were measured using the correlation and allelic correct rate. FImpute resulted in highest accuracies, whereas Beagle 3.3 gave the least-accurate imputations. Accuracies evaluated as correlation (allelic correct rate) ranged from 0.910 (0.942) to 0.961 (0.974) using 50K as HDP and with 3K (7K) as low-density panels. With GGP-75Ki as HDP, accuracies were moderate for 3K, 7K, and 50K, but high for SGGP-20Ki. The use of HD-777K as HDP resulted in accuracies of 0.888 (3K), 0.941 (7K), 0.980 (SGGP-20Ki), 0.982 (50K), and 0.993 (GGP-75Ki). Ungenotyped individuals were imputed with an average accuracy of 0.970. The average top 5 kinship coefficients between reference and imputed individuals was a strong predictor of imputation accuracy. FImpute was faster and used less memory than Beagle v4. Beagle v4 outperformed Beagle v3.3 in accuracy and speed of computation. A genotyping strategy that uses the HD-777K SNP chip as a reference panel and SGGP-20Ki as the lower-density SNP panel should be adopted as accuracy was high and similar to that of the 50K. However, the effect of using imputed HD-777K genotypes from the SGGP-20Ki on genomic evaluation is yet to be studied.

Revealing misassembled segments in the bovine reference genome by high resolution linkage disequilibrium scan

“…Milanesi et al [9], comparing imputation accuracies in Simmental cattle using different reference genome assemblies, found segments poorly imputed in one reference assembly that in the other mapped to a different chromosome with low imputation error rates, arguing that a cluster of consecutive markers with a high percentage of imputation errors may be evidence of misassembled segments in the reference genome. Also, studying strategies for genotype imputation in Gyr ( B. indicus ) dairy cattle, Boison and colleagues [8] found poor imputation accuracies for certain segments of the genome. One of the segments ranged from 44.8 to 45.3 Mb on chromosome 1.…”

“…Using a higher density SNP array (Illumina® BovineHD BeadChip, BovineHD hereafter) to test imputation in Fleckvieh ( Bos taurus ), Pausch et al [6] identified 5,039 out of 599,535 SNPs (0.89 %) exhibiting poor imputation performance. Poor imputation accuracies of neighboring SNPs were also reported in Bos indicus , namely Nellore beef cattle [7] and Gyr dairy cattle [8]. Since the imputation process depends on LD blocks to infer missing marker genotypes, an assembly error would be the main reason for poor genotype imputation [9].…”

Red Sindhi cattle in Brazil: population structure and distribution

“…With the application of a R 2 threshold of 0.15 for the maximum correlation between markers, following the example of the study of Boison et al (2015), the minimum crossvalidation error was also detected for the inferred number of subpopulations equal to six. Same results were found with more permissive thresholds and consequently bigger numbers of markers used.…”

“…This was implemented with the command "--indep-pairwise 50 10 0.15" within the PLINK software (Purcell et al, 2007). Boison et al (2015), in their study with the zebu Gyr cattle in Brazil, indicated that a 100 kb distance between markers corresponded to an R 2 minimum of 0.15 for linkage disequilibrium. In the present study, the minimum R 2 (0.15) was chosen as the threshold for eliminating linkage disequilibrium in the set of markers used on the determination of ancestry proportions.…”

Genome-Wide Association Study for Milk Production and Reproduction Traits in Russian Holstein Cattle Population

“…Goddard [6], 300,000 SNPs are required for interbreed analysis in Bos taurus. The number of SNPs of about 50,000 may be quite enough for a single breed, such as Holstein, however the preference is given to the samples of several breeds and high density SNP panels for more precise mapping of QTL and improved reliability of genomic evaluations [9][10][11].…”