The current study, supported by the Angus Foundation, has found that, to date, the common genotype density has performed well in the genetic evaluation for body weight and carcass traits in Australian Angus cattle.
Imagine the potential of whole genome sequence (WGS), containing millions of genetic markers, for genomic prediction. Could it outperform the common platform we currently use?
Genome-based genetic evaluations, known as genomic prediction, have become a standard approach for estimating livestock breeding values. Genomic prediction can improve the rate of response to selection by shortening generation intervals and gaining more accuracy in predicting breeding value, especially for young animals and difficult-to-measure traits. The accuracy of the genomic prediction depends on two major factors; the number of DNA-tested animals recorded for the objective trait and the number of DNA markers used in genotyping. Current genomic evaluations use standardised genotyping arrays ranging from 10k to 700k in density, with 50k being the most common platform (Goddard et al. 2011).
The advent of next-generation sequencing technologies has made it feasible to obtain whole-genome sequence data, millions of genetic markers across the genome, at a reasonable cost. This data could revolutionize routine genetic evaluations. Moreover, genotype imputation, a common practice, can ensure reliable accuracy when obtaining WGS for animals genotyped with lower densities.
Whole genomic sequence may improve the accuracy of genomic prediction since it should include actual causal variants in the dataset instead of depending on the association between the QTLs and markers (Meuwissen et al. 2016). Our study aimed to examine the benefit of the sequence data for genomic prediction in Australian Angus beef cattle. Different genetic marker densities, including medium-density 50k, high-density 700k and WGS, were used to examine the potential improvement in prediction ability when increasing the marker density for economically important traits in Australian Angus cattle.
To obtain the whole-genome sequence, genotype imputation was performed. The medium-density 50k genotype samples were imputed to the whole-genome sequence level with a stepwise genotype imputation, from 50k to high-density 700k, then to WGS. Quality control filtered out those markers with low in imputation accuracy and minor allele frequency (MAF) < 0.05. This resulted in 44,827, 522,192, and 7,899,466 markers for each animal for 50k, high-density and whole-genome sequence, respectively, in the genotype dataset. For phenotypes, economically important traits studied were body weight traits, including birthweight (BW), weight at 400 days (YW) and weight at 600 days of age (FW). The carcass traits were carcass weight (CW), carcass intramuscular fat (CIMF), and carcass marbling score (CMAU).
Genomic best linear unbiased prediction (GBLUP) was used to compare genomic predictions with the three marker densities. The performance of genomic prediction was determined by prediction abilities, expressed as prediction accuracy and prediction bias. Prediction accuracy indicates the ability of a model to predict outcomes accurately. Meanwhile, prediction bias indicates the differences between the model’s outcomes and previously generated predictions. The 10-fold cross-validation was selected was selected to validate the prediction. The validation was designed to predict the next generation from the previous forward prediction, which was suitable for the beef cattle production system. Animals born in the last two years of the dataset were allocated to the validation group, and the others were assigned to a reference population. Individuals in the validation group were grouped according to the level of their relatedness with the reference set by a relationship value, which was extracted from a genomic relationship matrix. The prediction ability was compared between three densities of genotypes and was reported from the testing group and the subgroups according to the degree of relatedness.
The results obtained from this study revealed that there was no substantial improvement in the prediction when increasing the genotype density up to the WGS. Although there was no significant difference, for BW, a slight decrease was found as marker density increased. The highest accuracy was from the 50k for BW. The highest accuracies were obtained from the prediction of the HD density. The lowest accuracy for the growth traits was from the WGS. The accuracy of the subgroups followed the same pattern, with only small differences in accuracy and bias between the marker densities. Similarly, for carcass traits, there was no difference in prediction accuracy or bias when increasing the marker densities.
We concluded that with the current level of genetic diversity in the Australian Angus population the common 50k genotype density captured the inherent relationship structure in the population; there was no substantial improvement in the prediction accuracy found once the marker density increased.
Since the common genotype density accurately performed genomic prediction, implementation of the WGS should be carefully considered against its cost, particularly given the large size datasets and expensive computational analysis. Nevertheless, WGS-based analysis has been developed and optimised for practical processes. With a better understanding of genetics, these efforts can exploit WGS for better solutions in the near future.
Reference
Goddard M.E., Hayes B.J., Meuwissen T.H.E. (2011) J. Anim. Breed. Genet. 128: 409.
Meuwissen T., Hayes B., Goddard M. (2016) Animal Frontiers 6: 6.
Nantapong Kamprasert – University of New England, Armidale NSW