Integrating phenomic selection using single-kernel near-infrared spectroscopy and genomic selection for corn breeding improvement

Rafaela P. Graciano, Marco Antônio Peixoto, Kristen A. Leach, Noriko Suzuki, Jeffery L. Gustin, A. Mark Settles, Paul R. Armstrong & Márcio F. R. Resende Jr.

Theoretical and Applied Genetics; 26 February 2025; vol.138; article 60

Key message

Phenomic selection using intact seeds is a promising tool to improve gain and complement genomic selection in corn breeding. Models that combine genomic and phenomic data maximize the predictive ability.

Abstract

Phenomic selection (PS) is a cost-effective method proposed for predicting complex traits and enhancing genetic gain in breeding programs. The statistical procedures are similar to those utilized in genomic selection (GS) models, but molecular markers data are replaced with phenomic data, such as near-infrared spectroscopy (NIRS). However, the use of NIRS applied to PS typically utilized destructive sampling or collected data after the establishment of selection experiments in the field. Here, we explored the application of PS using nondestructive, single-kernel NIRS in a sweet corn breeding program, focusing on predicting future, unobserved field-based traits of economic importance, including ear and vegetative traits. Three models were employed on a diversity panel: genomic and phenomic best linear unbiased prediction models, which used relationship matrices based on SNP and NIRS data, respectively, and a combined model. The genomic relationship matrices were evaluated with varying numbers of SNPs. Additionally, the PS model trained on the diversity panel was used to select doubled haploid (DH) lines for germination before planting, with predictions validated using observed data. The findings indicate that PS generated good predictive ability (e.g., 0.46 for plant height) and distinguished between high and low germination rates in untested DH lines. Although GS generally outperformed PS, the model combining both information yielded the highest predictive ability, with higher accuracies than GS when low marker densities were used. This study highlights NIRS’s potential to achieve genetic gain where GS may not be feasible and to maintain/improve accuracy with SNP-based information while reducing genotyping costs.

See https://link.springer.com/article/10.1007/s00122-025-04843-w

Figure 2

NIRS data analysis for the diversity panel. A shows the correlation between each wavelength and the traits from 2021 for days to pollination (DTP), ear height (EH), ear width (EW), germination (GER), plant height (PH). B Principal component analysis (PCA) shows the amount of variance explained by the first two components, PC1 and PC2, based on marker data (left) and NIRS data averaged from 2019 and 2020. The group indicates if the genotypes have the sugary1 (su1) and/or shrunken2 (sh2) mutation genes