Optimizing marker density for maximizing the accuracy of genomic prediction and heritability estimates in three major North American and European spruce species

Abstract Genomic prediction, also called genomic selection (GS), is being increasingly used in tree breeding with aim to accelerate genetic gains by shortening the long breeding cycles. However, high genotyping costs remain a challenge. This study aimed to determine the optimal marker density in genome coverage, to maximize GS accuracy and precision of heritability estimates for growth and wood quality traits. Thousands of SNPs representative of the exome of three major spruce species were used: 18,275 SNPs for black spruce (representing 10,894 distinct gene loci), 11,328 SNPs for white spruce (8647 gene loci), and 116,765 SNPs for Norway spruce (20,695 gene loci). For each species, a similar experimental design was used with related full-sib families replicated on two sites, and GBLUP prediction models were developed. The effect of varying the number of SNPs was examined by re-sampling subsets from 500 to 100,000 SNPs. Results indicated that plateaus in heritability estimates were reached as the marker density increased, stabilizing between 4000 to 8000 SNPs for a spruce genome size of around 2000 centimogans, a trend consistent across all traits and species. Predictive ability and prediction accuracy both increased with the number of SNPs up to a similar level, beyond which further improvements were marginal. Such optimal marker density should be financially attainable for most spruce breeding programs, striking a balance between the need for maximizing accuracy and that for minimizing genotyping costs. These findings should support the further deployment of GS in conifer breeding programs, with high selection precision and by reducing the financial burden of very high-density SNP coverage, even for conifers characterized by large giga-genomes. Competing Interest Statement The authors have declared no competing interest. Data Availability Statement In order to comply with Intellectual Property Policies (IPP) of participating governmental and private organizations in this work, the supporting phenotyping and genotyping data is not deposited into the public domain. However, data will be shared upon motivated request to the corresponding author after consulting with participating governmental and private organizations involved in this work.