Construction of genomic prediction models for leaf protein content in Nicotiana tabacum

With its high soluble protein content, large biomass yield, and ease of cultivation, tobacco leaves show strong potential as a novel protein source for livestock. However, the genetic basis underlying leaf protein content remains poorly understood, necessitating the use of genomic prediction models to screen germplasm resources and accelerate the improvement of this trait in future breeding programs. To address this, we analyzed 2517 tobacco germplasm accessions from the Chinese National Tobacco Germplasm Resource Bank, which represent broad genetic diversity, to investigate the genetic architecture of leaf protein content and construct genomic prediction models. Tobacco leaf protein content exhibited a moderate heritability of 0.16, and association analysis identified a significant peak that explained approximately 1% of the phenotypic variance. We further evaluated the performance of 16 mainstream genomic prediction models using five-fold cross-validation. Among these models, best linear unbiased prediction (rrBLUP) model achieved the highest prediction accuracy (0.87). In addition, rrBLUP required less computational time and resources compared with other models, highlighting its stability and efficiency. Field validation (Longshan County, Hunan Province, 111°37′45″E, 27°30′52″N) confirmed the robustness and accuracy of our genomic selection model. Overall, our results demonstrate that genomic prediction can enable rapid screening of tobacco germplasm resources and substantially enhance the efficiency of developing high-protein varieties. • Tobacco shows great potential in protein sources for livestock. • The variation of protein content appears to be driven by multiple genes with some being detectable. • Genomic selection is a powerful tool for accelerating breeding in tobacco leaf protein content.