High-throughput hyperspectral phenotyping and transcriptomics reveal expression networks associated with nitrogen-limitation-induced senescence in sorghum

Abstract

Abstract Background Sorghum ( Sorghum bicolor ) is a versatile C4 crop used for food and feed and as biomass for bioproducts and energy. Improving nitrogen use efficiency (NUE) in sorghum is important because fertilizer is costly and excessive fertilizer use has negative environmental impacts. Leaf senescence mediates nutrient recycling, but its dynamic progression is difficult to quantify at scale. We evaluated whether visible-near-infrared hyperspectral imaging can provide high-throughput measures of N-limitation-induced senescence in sorghum and link these phenotypes to gene expression. Sorghum Tx430 plants were grown under four N treatments (6, 9, 12, and 15 mM), imaged from vegetative growth through grain fill, and destructively sampled for RNA-seq at four developmental stages. Results A supervised support vector machine with a radial basis function kernel classified pixels from a hyperspectral image of sorghum plants grown under different N levels into green leaf, yellow leaf, dry leaf, stalk, panicle, and background classes with 0.93 accuracy. We defined the senescence ratio as the sum of yellow and dry leaf areas divided by the green leaf area and computed it across multiple growth stages and nitrogen levels. The senescence ratio did not differ among N treatments during vegetative growth, but it declined with increasing N during boot, anthesis, and grain fill, indicating earlier senescence under N limitation. Among the genes whose expression positively correlated with senescence ratio were 13 putative transcription factors, including SbiRTX430.02G247100, a WRKY1/ZAP1 homolog and a WRKY4 homolog. Gene regulatory network analysis of the top 1% of genes associated with SbiRTX430.02G247100 showed enrichment for processes associated with leaf senescence and chlorophyll catabolism. In contrast, the network associated with the WRKY4 homolog was enriched for autophagy-related terms. Conclusions Our study shows that automated hyperspectral imaging is highly effective for monitoring dynamic plant phenotypes, such as stress-induced senescence, that are difficult to visually score with the naked eye. Here, nitrogen deficiency served as the stress condition. Still, this approach supports large-scale phenotypic data collection for any such stressor and enables analyses with greater statistical power, yielding more robust conclusions and the potential for new insights that can be applied to engineering and breeding better crops.

Plant phenotyping relevance

ソルガムの動的な老化表現型を高スループットに取得する hyperspectral imaging と、SVMによる画像分類・senescence ratio算出が研究の中心であり、植物状態の定量化手法を実証している。

Code and data availability

The paper's availability statement points to a public GitHub repository containing the authors' image-processing, machine-learning classification, transcriptomic analysis, and figure-generation scripts. The 148 GB hyperspectral image data is only promised 'upon acceptance' (not yet public), and the RNA-seq deposit is a