Genome‐wide association and genomic prediction for biomass yield in a genetically diverse Miscanthus sinensis germplasm panel phenotyped at five locations in Asia and North America

Abstract

To improve the efficiency of breeding of Miscanthus for biomass yield, there is a need to develop genomics‐assisted selection for this long‐lived perennial crop by relating genotype to phenotype and breeding value across a broad range of environments. We present the first genome‐wide association (GWA) and genomic prediction study of Miscanthus that utilizes multilocation phenotypic data. A panel of 568 Miscanthus sinensis accessions was genotyped with 46,177 single nucleotide polymorphisms (SNPs) and evaluated at one subtropical and five temperate locations over 3 years for biomass yield and 14 yield‐component traits. GWA and genomic prediction were performed separately for different years of data in order to assess reproducibility. The analyses were also performed for individual field trial locations, as well as combined phenotypic data across groups of locations. GWA analyses identified 27 significant SNPs for yield, and a total of 504 associations across 298 unique SNPs across all traits, sites, and years. For yield, the greatest number of significant SNPs was identified by combining phenotypic data across all six locations. For some of the other yield‐component traits, greater numbers of significant SNPs were obtained from single site data, although the number of significant SNPs varied greatly from site to site. Candidate genes were identified. Accounting for population structure, genomic prediction accuracies for biomass yield ranged from 0.31 to 0.35 across five northern sites and from 0.13 to 0.18 for the subtropical location, depending on the estimation method. Genomic prediction accuracies of all traits were similar for single‐location and multilocation data, suggesting that genomic selection will be useful for breeding broadly adapted M. sinensis as well as M. sinensis optimized for specific climates. All of our data, including DNA sequences flanking each SNP, are publicly available. By facilitating genomic selection in M. sinensis and Miscanthus × giganteus, our results will accelerate the breeding of these species for biomass in diverse environments.

Training Population Optimization for Genomic Selection in Miscanthus

“…Our analysis of the observed traits suggests that subpopulation structure contributed substantially to the prediction accuracies within each panel (Figure 2), and this finding is consistent with previous studies conducted by ourselves and others (Clark et al 2019b;Crossa et al 2016;Wientjes et al 2015). When we previously performed genomic selection on the Msi panel using a larger set of 46,177 genome-wide SNPs, we observed decreases in prediction accuracy after accounting for subpopulation structure (Clark et al 2019b). Outside of Msi, Crossa et al (2016) reported a decrease of 15-20% when using PCs of markers in the GS model to account for population structure present in two collections of wheat landrace accessions, while Wientjes et al (2015) observed that using population information among three cattle breeds for prediction gave prediction accuracies up to 30% higher than those from a GS model including breed as a fixed effect.…”

“…Our analysis of the observed traits suggests that subpopulation structure contributed substantially to the prediction accuracies within each panel (Figure 2), and this finding is consistent with previous studies conducted by ourselves and others (Clark et al 2019b;Crossa et al 2016;Wientjes et al 2015). When we previously performed genomic selection on the Msi panel using a larger set of 46,177 genome-wide SNPs, we observed decreases in prediction accuracy after accounting for subpopulation structure (Clark et al 2019b).…”

“…Given the rapid LD decay rate that we observed in both the Msi and Msa diversity panels (where pairwise LD decayed to nominal levels at distances less than 3 kb, Figure S8A-C), we conclude that the markers we used for GS did not provide sufficient density to tag enough causal variants; consequently, the observed prediction accuracies might be lower than expected with a denser marker set. This conclusion is in part justified given that prediction accuracies tended to be higher when Clark et al (2019b) performed GS on the same Msi diversity panel using 46,177 SNPs. Thus, one avenue for future research would be to redo this study with a larger number of genome-wide markers.…”

“…By leveraging genomic and phenotypic data in a training set, GS is capable of predicting genomic-estimated breeding values (GEBVs) of unphenotyped related accessions based solely on genome-wide marker data (Meuwissen et al 2001;Jarquín et al 2017). Given that previous GS studies in Msi obtained moderate-to high prediction accuracies for biomass and cell wall composition (Slavov et al 2014;Davey et al 2017;Clark et al 2019b), GS is theoretically well-suited to utilize the genetic diversity of existing Msi and Msa diversity panels and select genetically distinct M•g clones with superior biofuel capacity.…”

Genome biology of the paleotetraploid perennial biomass crop Miscanthus

“…10c, e), which indicates that the most productive miscanthus genotype currently grown is the product of more than one cycle of introgression from M. sinensis into M. sacchariflorus. Hybrids between M. sacchariflorus and M. sinensis are frequently highly vigorous and high-yielding, regardless of whether they are diploid, triploid, or tetraploid 46,47 . Thus, understanding how prior introgression of M. sinensis alleles into a primarily M. sacchariflorus genetic background affects the yield potential of subsequent interspecific hybrids will be important for optimizing breeding strategies.…”