Multivariate analysis of the cotton seed ionome reveals integrated genetic signatures of abiotic stress response

Heat and drought represent two of the most common abiotic stresses that plants encounter in modern agricultural production systems and can cause significant economic losses. As climate change continues to increase the frequency and severity of these conditions, the development of stress-resilient cultivars becomes pivotal to sustaining crop yields. To meet these challenges, a better understanding of the genetic control of physiological responses to these environmental conditions is needed. In light of this, we evaluated an upland cotton (Gossypium hirsutum L.) recombinant inbred line (RIL) mapping population under water-limited and well-watered conditions in a hot, arid environment across three years. Ionomic profiling, the rapid quantification of elemental concentrations in a given sample, was used to phenotype seed subsamples from the population. Additionally, soil samples taken from throughout the entire field site were also assayed to better model the soil elemental heterogeneity and account for this variability in subsequent analyses. The elements profiled in seeds exhibited moderate to high heritabilities as well as strong phenotypic and genotypic correlations between elements. Both types of correlations maintained their strength and direction despite the imposed irrigation regimes. Quantitative trait loci (QTL) mapping results from a Bayesian classification method identified multiple genomic regions where QTL for individual elements colocalized, suggesting that genetic control of the ionome is highly interrelated. To more fully exploit this shared genetic architecture, multivariate QTL mapping was implemented among groups of biochemically related elements. This analysis revealed both additional as well as pleiotropic QTL responsible for the coordinated control of phenotypic variation for elemental accumulation in seeds. To further leverage these data to gain insight into the physiological status of the plants, machine learning algorithms that utilized only ionomic data were used to predict the irrigation under which RIL lines were grown. The best performing method, which was support vector machines, produced a prediction accuracy of 97.7% providing empirical evidence that ionome can capture the physiological status of the plant. Taken together, these results demonstrate the extent to which the seed ionome is genetically interrelated and predictive of plant physiological response to adverse environmental conditions.