Multi-Model Elucidating of Nutritional Quality Contributions to Maize Kernel Test Weight and Regional Heterogeneity

This study systematically quantified the contribution rates and spatial heterogeneity of protein, starch, and fat—the three major nutritional components—to maize kernel test weight formation, and elucidated how genetic background, ecological region, and cultivation density interactively modulate both nutritional quality and test weight. The findings aim to establish a science-based foundation for region-specific optimization of maize quality and to advance an integrated “high-yield-high-qualityhigh-efficiency” production paradigm.

A nationwide field survey was conducted across four major maize-producing regions in China, encompassing 718 representative kernel samples from 77 leading cultivars grown under 24 distinct planting density gradients (37500-127500 plants/hm²). All samples were naturally air-dried to standardized moisture content (14% w.b.) prior to uniform physicochemical analysis. Protein, starch, and fat contents were determined using calibrated near-infrared reflectance spectroscopy (NIRS), and test weight was measured with a certified grain test weight instrument (ISO 7971-3 compliant). To dissect the complex determinants of test weight, we implemented a hierarchical analytical framework integrating: (ⅰ) multiple linear regression to estimate independent linear effects; (ⅱ) random forest modeling to capture nonlinear interactions and relative feature importance; and (ⅲ) structural equation modeling (SEM) to infer directional causal pathways among traits. Three-way ANOVA was further employed to assess the main and interactive effects of cultivar, ecological region, and cultivation density on test weight and each nutritional component.

Protein (β=8.406, P<0.001) and starch (β=6.413, P<0.001) emerged as statistically robust and biologically dominant drivers of test weight, accounting for 28% and 45% of the total explained variance in the random forest model, respectively—both exhibiting high path coefficient stability in SEM (standardized coefficients ≥0.72, P<0.001). In contrast, fat showed negligible explanatory power (2%), and its effect failed to reach statistical significance (P=0.09). Three-way ANOVA confirmed highly significant (P<0.001) main effects and two- and three-way interactions among cultivar, ecological region, and density for test weight, protein, and starch—indicating strong contextual dependency. Spatially, protein contributed most strongly in the Northeast spring maize region (43.9% of model variance), whereas starch dominated in the Huang-Huai-Hai summer maize region (52.9%). Critically, the synergistic contribution of protein and starch jointly explained 81.0% and 85.0% of the total model variance in these two regions, respectively. Structural equation modeling revealed a direct positive effect of protein on test weight, but an indirect negative effect stemming from the compensatory relationship between protein and starch accumulation, which underscores the physiological trade-off in kernel sink-filling.

Maize test weight formation was a biologically synergistic process driven by protein and starch, with fat playing no substantial role. Significant interactions existed among cultivar, ecological region, and density, with the same cultivar exhibiting distinct regulatory pathways under different ecological and cultivation conditions. Consequently, the Northeast region should prioritize high-protein cultivar selection and precise nitrogen management, while the Huang-Huai-Hai region should enhance carbon assimilation efficiency and regulate key starch-synthesis enzymes. All production areas should achieve a precise "cultivar-region-practice" matching strategy to synergistically improve maize yield and quality.