Biomarker genes for model-based prediction of drought-stress perception levels in rice.

Abstract

Background Drought is a global challenge that severely restricts crop yields and threatens food security. Plants respond to drought stress by modulating gene expression before visible phenotypic changes occur. However, most studies of drought resistance have examined phenotypes after drought treatment, with little emphasis on how severely the plants were perceiving drought-stress conditions before the appearance of stress symptoms. We therefore developed drought-stress biomarkers (DSBMs) to detect drought-stress perception levels based on gene expression profiles by performing time-series transcriptome analysis and phenotypic analysis of rice (Oryza sativa) under drought conditions in the growth chamber. Results Time-series RNA-seq of the drought-susceptible rice cultivar IR64 revealed drastic changes in the transcriptome after 4-6 days of drought treatment in plants grown in pot culture mimicking drought conditions in the field, particularly for genes related to photosynthesis. Among the differentially expressed genes, we selected 23 DSBM genes that consistently responded to drought stress. Rehydration immediately reset the changes in expression of these DSBM genes, indicating that their expression changes reflect current drought-stress perception levels, but not stress memories. Responses of DSBM genes tended to be conserved among rice accessions, irrespective of the rice subpopulation (such as indica, aus, and japonica). We developed a machine learning model using the expression levels of DSBM genes trained by the time-series RNA-seq data for IR64. This model successfully predicted the drought-stress perception levels of various rice accessions, representing the probability of exposure to drought treatment, with an accuracy of 75%. Extreme root architecture traits, such as the largest root surface area, narrowest crown root diameter, and largest ratio of deep rooting, influenced the predicted drought-stress perception levels. Conclusion We identified DSBM genes and developed a machine learning model as a robust tool for assessing drought-stress perception levels in rice. Monitoring and predicting drought-stress perception levels should contribute to more efficient crop management and breeding schemes. Furthermore, our dataset would serve as a resource for further understanding the mechanisms of drought resistance in rice.

Plant phenotyping relevance

イネの乾燥ストレス知覚レベルという植物状態を、遺伝子発現バイオマーカーと機械学習で推定する手法を開発し、複数アクセッションで検証しているため、方法論が中心である。

Code and data availability

The paper deposits only RNA-seq transcriptome data (GSE288412, GSE288615, GSE288713), which are molecular omics deposits rather than plant-phenotyping measurements. The phenotypic traits (plant height, shoot dry weight, tiller number, root architecture traits) and the machine-learning model/code have no stated public,

No evidence-backed public reproduction asset is currently recorded.