Mean and variance heterogeneity loci impact kernel compositional traits in maize

Maize (Zea mays) kernel composition is critical for food, feed, and industrial applications. Improving traits such as starch, protein, oil, fiber, and ash requires understanding their genetic basis. We conducted genome-wide association studies (GWAS) and variance genome-wide association studies (vGWAS) analyses using 954 inbred lines from the USDA-ARS North Central Regional Plant Introduction Station collection to identify loci influencing both trait means and variability. We detected 10 significant single nucleotide polymorphisms (SNPs) associated with five kernel traits, some of which colocalized with known genes such as waxy1 and gras7. vGWAS uncovered additional loci not detected by standard GWAS, highlighting its value as a complementary tool. Genomic selection models, including ridge-regression best linear unbiased prediction, reproducing kernel Hilbert space, and random forest, achieved moderate prediction accuracies (0.41-0.55), with parametric and semi-parametric models showing less prediction bias. Although our dataset was derived from unreplicated genebank seed, key findings, particularly for protein and starch, were consistent with results from replicated field trials, supporting the utility of genebank-derived high-quality samples for initial genomic analysis. These results highlight the potential for using existing seed resources and high-throughput phenotyping to identify candidate loci and prioritize traits for future replicated validation.