Classification approaches for sorting maize (Zea mayssubsp.mays) haploids using single-kernel near-infrared spectroscopy

Doubled haploids (DHs) are an important breeding tool for creating maize inbred lines. One bottleneck in the DH process is the manual separation of haploids from among the much larger pool of hybrid siblings in a haploid induction cross. Here, we demonstrate the ability of single-kernel near-infrared reflectance spectroscopy (skNIR) to identify haploid kernels. The skNIR is a high-throughput device that acquires an NIR spectrum to predict individual kernel traits. We collected skNIR data from haploid and hybrid kernels in 15 haploid induction crosses and found significant differences in multiple traits such as percent oil, seed weight, or volume, within each cross. The two kernel classes were separated by their NIR profile using Partial Least Squares Linear Discriminant Analysis (PLS-LDA). A general classification model, in which all induction crosses were used in the discrimination model, and a specific model, in which only kernels within a specific induction cross, were compared. Specific models outperformed the general model and were able to enrich a haploid selection pool to above 50% haploids. Applications for the instrument are discussed.