Genomic index selection provides a pragmatic framework for setting and refining multi-objective breeding targets in Miscanthus

Background: Miscanthus has potential as a biomass crop but the development of varieties that are consistently superior to the natural hybrid M. times giganteus has been challenging, presumably because of strong G timesE interactions and poor knowledge of the complex genetic architectures of traits underling biomass productivity and climatic adaptation. While linkage and association mapping studies are starting to generate long lists of candidate regions and even individual genes, it seems unlikely that this information can be translated into effective marker-assisted selection for the needs of breeding programmes. Genomic selection has emerged as a viable alternative, and prediction accuracies are moderate across a range of phenological and morphometric traits in Miscanthus, though relatively low for biomass yield per se. Methods: We have previously proposed a combination of index selection and genomic prediction as a way of overcoming the limitations imposed by the inherent complexity of biomass yield. Here we extend this approach and illustrate its potential to simultaneously achieve multiple breeding targets in the absence of a priori knowledge about their relative economic importance, while also monitoring correlated selection responses for non-target traits. We evaluate two hypothetical scenarios of increasing biomass yield by 20% within a single round of selection. In the first scenario, this is achieved in combination with delaying flowering by 44 days (roughly 20%), whereas in the second, increased yield is targeted jointly with reduced lignin (-5%) and increased cellulose (+5%) content, relative to current average levels in the breeding population. Key Results: In both scenarios, the objectives were achieved efficiently (selection intensities corresponding to keeping the best 20% and 4% of genotypes, respectively). However, the outcomes were strikingly different in terms of correlated responses, and the relative economic values (i.e., value per unit of change in each trait compared to that for biomass yield) of secondary traits included in selection indices varied considerably. Conclusions: Although these calculations rely on multiple assumptions, they highlight the need to evaluate breeding objectives and explicitly consider correlated responses in silico, prior to committing extensive resources. The proposed approach is broadly applicable for this purpose and can readily incorporate high-throughput phenotyping data as part of integrated breeding platforms.