The priority to increase grain yield has led to the loss of alleles that benefit grain quality traits of wheat, such as grain protein content (GPC). Modern germplasms either lack the favourable alleles for GPC, or the trait is heterogeneous among grain harvested from a plot or from a single spike, which presents challenges for identifying genetic variation to improve grain quality.

The sequence of a global collection of A. E. Watkins wheat landraces shows that modern cultivars have emerged from only two of the seven ancestral groups; this collection offers the opportunity to identify unique GPC alleles. In this study, we used hyperspectral imaging (HSI) to screen single wheat grains from over 600 Watkins landraces, with 12 grains per genotype. We developed a Partial Least Squares Regression (PLSR) model with R-squared value of 0.90 and Root Mean Squared Error (RMSE) of 0.77 for predicting GPC from single wheat grains.

Using this model, we predicted GPC from two field trials of Watkins population grown at the John Innes Centre and Rothamsted Research in 2020 and 2022, respectively. The predicted single GPC varied between 8–20%. Using sequenced Watkins data of over 90 million genetic variants, we performed a genome-wide association study (GWAS) and identified 22 loci for GPC, with 5 prioritised for gene identification.

Within these 5 loci, we identified 14 haplotypes from Watkins HapMap with over 5% positive effect on GPC-related traits. Taking advantage of HSI’s ability to phenotype single grains, we performed a GWAS for single grain protein heterogeneity – a previously unexplored trait that has potential to select homogenous grains with high GPC.

Moreover, we used HSI and machine learning generated data to investigate loci for mean and single grain heterogeneity for grain weight, length, breadth, area, perimeter, and roundness. Our findings report novel landrace-derived alleles that influence grain quality and size, and they highlight the potential to combine HSI and machine learning with genomics to fast-track gene identification of key agronomic traits.