Abstract
Human activity is causing an elevation of CO2 levels in the atmosphere, that are expected to rise from 420 ppm to approximately 1000 ppm by the end of the century. C3 plants, a major part of cultivated crops, are particularly affected by the rise of CO2 levels. Even though a stimulation of biomass production is expected under elevated CO2 (eCO2), this gain is met with a marked depletion of the plant mineral composition and an especially strong decline in nitrogen (N) content. This poses a major threat to crop quality and human nutrition, that we propose to start addressing through systems biology approaches. Promising hypotheses to explain this decline invoke a disruption of signalling pathways associated to N uptake and assimilation, motivating the investigation at the genomic scale of gene expression reprogramming in the roots of the model plant Arabidopsis thaliana under eCO2. To uncover the unknown regulators orchestrating such networks, we developed statistical methods for Gene Regulatory Network (GRN) inference, a challenging task hindered by high dimension and the scarcity of ground truth networks. Modelling transcriptional dependencies from gene expression data can be performed by regression-based techniques assuming that the expression variations of regulator genes hold descriptive and predictive power over the expression variations of their targets. We propose two novel approaches : (i) an extension of a Random Forest-based method, GENIE3, via permutation procedures assessing the significance of regulatory interactions that we include within a complete suite for GRN inference, and (ii) two integrative GRN inference methods based on sparse linear regression with stability selection and Random Forests, integrating Transcription Factor Binding Sites (TFBSs) with gene expression. We benchmark those methods against experimental gold standards, and show that they improve the biological relevance of inferred GRNs in Arabidopsis thaliana. We applied the first inference approach to a combinatorial transcriptomic dataset of root tissues under contrasted CO2 levels and nutritional conditions, and the second to the roots of plants exposed to a gradient of CO2 concentrations. The inferred GRNs provided candidate genes for the control of this response, and we demonstrate that some of them regulate growth stimulation under eCO2 without penalizing shoot nutrient content. Overall, our results indicate that key nitrate and iron nutrition genes and their known regulators are misregulated by rising CO2, and that pathways associated to high affinity nitrate transport systems are especially unfavorably altered. The last objective of this work was to leverage natural genetic variability to identify genes controlling the ionome response to eCO2. We confirmed a mineral content decline in three populations of Arabidopsis at different geographic scales, and showed that the variability in this response can be explained by genetic determinants in the world-wide panel via linear mixed models. We put forward another set of candidate genes, highly associated to iron, N and zinc depletion in the shoots under eCO2 that pave the way for designing plants with sustainable nutritional value for the near future.