Unravelling genomic and environmental contributions to nitrogen-use efficiency in tropical maize through Bayesian causal forests

Nitrogen fertilisation constitutes a significant portion of maize production costs, and distinct crop genotypes exhibit varying efficiency in converting added nitrogen into yield. Breeding for nitrogen-use efficiency necessitates comprehension of both the average crop response to nitrogen and genotype and environment-specific variations.

This study employs a multivariate Bayesian causal forest to investigate the impact of optimal versus low nitrogen regimes on three maize traits (grain yield, plant height, and ear height) within a genotyped breeding panel. This approach represents one of the initial attempts to estimate the causal effect of nitrogen in a genotyped breeding population, utilizing a designed contrast within the potential-outcomes framework rather than relying on genotype-environment associations reported by genomic selection and association mapping.

The trial involved 614 single-cross hybrids cultivated under both nitrogen levels in adjacent plots, ensuring that nitrogen was assigned by design rather than being confounded with genotype or environment. The estimated propensity score was near 0.5, indicating a suitable causal reading. The leading principal components of the additive genomic relationship matrix were incorporated into the moderator set, allowing for the investigation of both baseline outcomes and treatment effects that vary with genomic background.

Genotypes demonstrated a significant influence on all three traits (grain yield increased by 0.59, with a 95% credible interval of 0.51 to 0.67; plant height increased by 0.095; and ear height increased by 0.044, in adjusted-phenotype units), with the heterogeneity of the effect predominantly attributed to environmental factors.

A small genotypic component and negligible genotype-by-environment interaction were identified. The posterior credible interval of the effect was broader than the spread among genotypes, indicating that differences in responsiveness among hybrids were only weakly resolved. Separate analyses of the two tree ensembles revealed that genomic background primarily determines the yield level, while climate influences the response to nitrogen, with the leading genomic contribution being a single axis.

The findings suggest that a multivariate causal forest with genomic moderators, interpreted through its two ensembles and reported with posterior uncertainty, serves as a suitable and transparent tool for studying nitrogen response in breeding material.