Unsupervised identification of disease states from high‐dimensional physiological and histopathological profiles

• Published in: Molecular systems biology Vol. 15; no. 2; pp. e8636 - n/a
• Main Authors: Shimada, Kenichi, Mitchison, Timothy J
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 01-02-2019; EMBO Press; John Wiley and Sons Inc; Springer Nature
• Subjects: automated diagnosis; Biomarkers; Body weight; Body weight loss; Cachexia; Chemical and Drug Induced Liver Injury - diagnosis; Chemical and Drug Induced Liver Injury - genetics; Chemical and Drug Induced Liver Injury - physiopathology; Clustering; data mining; Datasets; Desensitization; Disease; EMBO10; EMBO24; EMBO42; Ferroptosis; Food analysis; Food consumption; Gene expression; Genotype & phenotype; Growth Differentiation Factor 15 - genetics; Health risks; Histology; Histopathology; Humans; Injuries; Insulin-like growth factor I; Kidney - drug effects; Kidney - pathology; Kidneys; Learning algorithms; Liver; Liver - drug effects; Liver - pathology; Machine learning; Metabolism; Metabolites; Pathology; Pathophysiology; Phenotypes; Physiology; Signal Transduction - drug effects; Tissues; Toxicity; toxicogenomics; Toxicological Phenomena - genetics; Toxicology; Toxins; Unsupervised Machine Learning; Weight loss; Xenobiotics; Xenobiotics - toxicity; ferroptosis; automated diagnosis; toxicogenomics; data mining; body weight loss
• ISSN: 1744-4292; 1744-4292
• DOI: 10.15252/msb.20188636

The liver and kidney in mammals play central roles in protecting the organism from xenobiotics and are at high risk of xenobiotic‐induced injury. Xenobiotic‐induced tissue injury has been extensively studied from both classical histopathological and biochemical perspectives. Here, we introduce a machine‐learning approach to analyze toxicological response. Unsupervised characterization of physiological and histological changes in a large toxicogenomic dataset revealed nine discrete toxin‐induced disease states, some of which correspond to known pathology, but others were novel. Analysis of dynamics revealed transitions between disease states at constant toxin exposure, mostly toward decreased pathology, implying induction of tolerance. Tolerance correlated with induction of known xenobiotic defense genes and decrease of novel ferroptosis sensitivity biomarkers, suggesting ferroptosis as a druggable driver of tissue pathophysiology. Lastly, mechanism of body weight decrease, a known primary marker for xenobiotic toxicity, was investigated. Combined analysis of food consumption, body weight, and molecular biomarkers indicated that organ injury promotes cachexia by whole‐body signaling through Gdf15 and Igf1, suggesting strategies for therapeutic intervention that may be broadly relevant to human disease.
Synopsis
Unsupervised characterization of physiological and histological changes in a toxicogenomic dataset revealed nine toxin‐induced disease states. In one of them, tolerance is induced due to xenobiotic defense induction and desensitization to ferroptosis.
Unsupervised characterization of physiological and histological changes in a toxicogenomic dataset revealed nine toxin‐induced disease states.
Disease states correspond to known tissue injury phenotypes and unknown phenotypes.
A late‐onset disease state was induction of tolerance, correlated with increase in xenobiotic metabolism and desensitization to ferroptosis.
Toxin‐induced body weight loss was explained by inter‐tissue communication via metabolites and hepatokines.
Graphical Abstract
Unsupervised characterization of physiological and histological changes in a toxicogenomic dataset revealed nine toxin‐induced disease states. In one of them, tolerance is induced due to xenobiotic defense induction and desensitization to ferroptosis.