Leveraging transcriptomics-based approaches to enhance genomic prediction: integrating SNPs and gene networks for cotton fibre quality improvement

Abstract

Cultivated cotton plants are the world’s largest source of natural fibre, where yield and quality are key traits for this renewable and biodegradable commodity. The Gossypium hirsutum cotton genome contains ~80K protein-coding genes, making precision breeding of complex traits a challenge. This study tested approaches to improving the genomic prediction (GP) accuracy of valuable cotton fibre traits to help accelerate precision breeding. With a biology-informed basis, a novel approach was tested for improving GP for key cotton fibre traits with transcriptomics of key time points during fibre development, namely, fibre cells undergoing primary, transition, and secondary wall development. Three test approaches included weighting of SNPs in DE genes overall, in target DE gene lists informed by gene annotation, and in a novel approach of gene co-expression network (GCN) clusters created with partial correlation and information theory (PCIT) as the prior information in GP models. The GCN clusters were nucleated with known genes for fibre biomechanics, i.e., fasciclin-like arabinogalactan proteins, and cluster size effects were evaluated. The most promising improvements in GP accuracy were achieved by using GCN clusters for cotton fibre elongation by 4.6%, and strength by 4.7%, where cluster sizes of two and three neighbours proved most effective. Furthermore, the improvements in GP were due to only a small number of SNPs, in the order of 30 per trait using the GCN cluster approach. Non-trait-specific biological time points, and genes, were found to have neutral effects, or even reduced GP accuracy for certain traits. As the GCN clusters were generated based on known genes for fibre biomechanics, additional candidate genes were identified for fibre elongation and strength. These results demonstrate that GCN clusters make a specific and unique contribution in improving the GP of cotton fibre traits. The findings also indicate that there is room for incorporating biology-based GCNs into GP models of genomic selection pipelines for cotton breeding to help improve precision breeding of target traits. The PCIT-GCN cluster approach may also hold potential application in other crops and trees for enhancing breeding of complex traits.