Early-stage phenomic prediction in grapevine (Vitis spp.) using near-infrared (NIR) reflectance spectra

Abstract

Recent advances in high-throughput phenotyping, particularly near-infrared (NIR) spectroscopy, have prompted interest in phenomic selection as a powerful tool for breeding. NIR spectra capture tissue- and family-specific biochemical signatures and have been used to predict agriculturally relevant traits such as yield or fruit quality in a range of crop species. To explore the use of phenomic selection models for early-stage selection in grapevine, seeds and seedlings of 1,618 individuals from six families were evaluated using hyperspectral reflectance data. For seedlings, spectral signatures were collected under two irrigation treatments (well-watered and water-stressed) in the greenhouse.

Reflectance spectra were used to assess within- and among-population variation and to develop predictive models for early life-stage traits including germination success, germination rate, seedling morphology (above- and below ground-biomass, plant height, leaf number, and root-to-shoot ratio), and developmental shift (indicated by the first tendril). Using phenomic prediction models from seeds and seedlings, with five-fold cross-validation, we observed moderate to high (r = 0.3 – 0.8) predictive abilities in germination success, all morphological traits, and first tendril development (seed spectra only) across both irrigation treatments. Together, these results demonstrate that NIR reflectance data collected at the earliest stages of development can be used to predict seedling traits in grapevine. This work provides a proof of concept for phenomic prediction in grapevine seedlings and highlights the potential of low-cost, high-throughput spectral data to supplement genomic selection and accelerate decision-making in grapevine breeding programs.