A combination of gene expression ranking and co‐expression network analysis increases discovery rate in large‐scale mutant screens for novel Arabidopsis thaliana abiotic stress genes

Abstract

SummaryAs challenges to food security increase, the demand for lead genes for improving crop production is growing. However, genetic screens of plant mutants typically yield very low frequencies of desired phenotypes. Here, we present a powerful computational approach for selecting candidate genes for screening insertion mutants. We combined ranking of Arabidopsis thaliana regulatory genes according to their expression in response to multiple abiotic stresses (Multiple Stress [MST] score), with stress-responsive RNA co-expression network analysis to select candidate multiple stress regulatory (MSTR) genes. Screening of 62 T-DNA insertion mutants defective in candidate MSTR genes, for abiotic stress germination phenotypes yielded a remarkable hit rate of up to 62%; this gene discovery rate is 48-fold greater than that of other large-scale insertional mutant screens. Moreover, the MST score of these genes could be used to prioritize them for screening. To evaluate the contribution of the co-expression analysis, we screened 64 additional mutant lines of MST-scored genes that did not appear in the RNA co-expression network. The screening of these MST-scored genes yielded a gene discovery rate of 36%, which is much higher than that of classic mutant screens but not as high as when picking candidate genes from the co-expression network. The MSTR co-expression network that we created, AraSTressRegNet is publicly available at http://netbio.bgu.ac.il/arnet. This systems biology-based screening approach combining gene ranking and network analysis could be generally applicable to enhancing identification of genes regulating additional processes in plants and other organisms provided that suitable transcriptome data are available.

Changes in Gene Expression in Space and Time Orchestrate Environmentally Mediated Shaping of Root Architecture

“…This is an important finding since it shows that the scale and scope of gene activity is not simply the molecular function of a gene, but a consequence of its spatial location, timing of expression, and responsiveness to environmental conditions. It also helps to provide an explanation for the typically higher hit rate in identifying phenotypes in regulators identified in systems biology data set analysis using a reverse genetic approach (Ransbotyn et al, 2015); a higher level specificity in the measurement of expression levels, regulation, and location of genes means that investigators can determine more precise and accurate predictions of when perturbation of function might lead to a measurable phenotype.…”

Integrated omics analysis of specialized metabolism in medicinal plants

“…Microarray resource databases including ATTED‐II (Obayashi et al ., ; Aoki et al ., ), Genevestigator (Zimmermann et al ., , ), AtGenExpress (Kilian et al ., ; Goda et al ., ), OryzaExpress (Hamada et al ., ), BAR (Patel et al ., ), At‐TAX (Laubinger et al ., ), PRIMe (Akiyama et al ., ), CORNET (De Bodt et al ., , ), PathoPlant (Bulow et al ., ), PLEXdb (Dash et al ., ) and ePlant (Fucile et al ., ) are few of many databases developed to perform co‐expression analysis using thousands of microarray datasets from Arabidopsis and other model plants. These microarray resources have contributed to the functional characterization of numerous genes in Arabidopsis and other plants (Mao et al ., ; Yonekura‐Sakakibara and Saito, ; Amrine et al ., ; Costa et al ., ; Ransbotyn et al ., ).…”

Functional networks inference from rule-based machine learning models

“…Nevertheless, it is abstract enough to be applied to all kinds of biological data. It has been demonstrated that co-expression networks are able to effectively identify pathways and candidate biomarkers [6] or reveal gene modules representing a biological process perturbed in a disease [7], just to name a few examples, and the similarity-based approach remains the dominant method of functional network inference today, with many recent examples: [8–12]. …”