Genetic discovery could reduce fertilizer use in corn

Scientists from New York University, Taiwan University and the U.S. Department of Agriculture have developed a novel approach to identify genes that regulate nitrogen use efficiency (NUE) in corn. Using gene regulatory networks (GRNs) and artificial intelligence, the researchers validated a set of genes that allow them to more accurately predict how the plant responds to nitrogen in the field. The method also revealed potential targets for gene editing and breeding programs.

Nitrogen use efficiency is essential to increase agricultural productivity and reduce the environmental impacts of excess fertilizers. However, the genetic mechanisms that control this process in crops such as corn are still poorly understood.

The team conducted a nitrogen-applied experiment on B73 corn seedlings, monitoring the expression of thousands of genes at 10 time points, from 0 to 120 minutes. The analysis revealed a temporal cascade of nitrogen response, with genes activated in an orderly fashion—from nitrate transport to metabolic processes such as glutamine synthesis.

Using these data, the scientists applied the Just-In-Time (JIT) model to cluster genes according to the timing of their activation. This allowed them to identify 4.600 nitrogen-responsive genes, 50% of which had not been reported in previous single-point studies. The temporal response also showed a correlation with nitrogen accumulation in plants, confirmed by ¹⁵N isotope assays.

Among the genes identified, 453 stood out with simultaneous response in roots and leaves, including 55 transcription factors. These factors coordinate the activation of complex gene networks. To validate their functions, the authors tested 23 of these factors in a protoplast-based gene perturbation assay (TARGET), adapted for corn cells. On average, each factor regulated about 4.000 genes.

Most of these validated targets correspond to genes identified in the experiment with whole plants, which reinforces the relevance of the in vitro model. Overlap analysis indicated MYB34 and MYBR3 as the main regulators in leaves. Their versions in Arabidopsis (AtDIV1) also participate in the regulation of the response to nitrogen. Mutant plants with loss of AtDIV1 function exhibit greater efficiency in nitrogen use.

Based on these validations, the researchers constructed high-confidence regulatory networks by applying time-series inference methods (Dynamic Factor Graphs - DFG). The networks were refined with precision/recall analysis, retaining robust interactions between approximately 200 factors and 700 genes.

These networks revealed, for example, that the gene Knotted1 (KN1), known for its role in leaf development, also responds rapidly to nitrogen application and indirectly regulates 63% of its targets through other factors. Network walk analysis revealed that KN1 connects to genes linked to cell proliferation and nitrogen metabolism.

The next step used machine learning with the XGBoost algorithm to associate gene expression with field performance. Data from 331 genes conserved between maize and Arabidopsis, whose responses to nitrogen occurred in parallel in both species, were used. Models based on these genes statistically significantly outperformed models with random genes in predicting nitrogen use efficiency traits.

To identify the most relevant modules, the authors calculated a “NUE regulon score” by combining the weights of the learning models with the gene networks. Again, MYB34 and MYBR3 topped the ranking. The target genes of these factors were compared with their orthologs regulated by AtDIV1, revealing 24 genes in maize and 23 in Arabidopsis with expression capable of accurately predicting nitrogen use efficiency.

The practical application was validated on a second dataset of 137 elite corn lines from China, showing that the MYB34/MYBR3 regulon genes effectively predicted traits such as ear diameter and length.

This study demonstrates that combining temporal analysis, functional validation and machine learning can accelerate gene discovery for crop improvement. The developed methodology allows transferring knowledge from model species to agricultural crops, even with 160 million years of evolutionary divergence between Arabidopsis and maize.

More information can be found at doi.org/10.1093/plcell/koaf093