Expression-based biomarker for acceleration of liver toxicity may potentially allow for early identification of liver cancer risk. These findings from a machine learning study were published in Scientific Reports.

Researchers from the University of Illinois obtained gene expression data from male rats obtained through Affymetrix Microarray Chips. Rats were exposed to 42 chemical compounds at low-, intermediate-, and high-dose levels for 3, 6, 9, and 24 hours repeated at 4, 8, 15, and 29 days. A total of 20 rats were used for each compound with 5 biological replicates for each duration of exposure. Livers were harvested and RNA was isolated.

Data from ethinyl estradiol (EE) exposure was used for this analysis as it causes hepatocellular carcinoma in rats. High-dose EE exposure resulted in alterations to alkaline phosphatase (day 4) and total bilirubin levels (week 2). Body weight, liver weight, and triglycerides were found to be altered by day 4 in exposed rates compared with control rats. Among all EE exposures, liver necrosis was the only apical alteration across all treatments.

Continue Reading

Clinical indicators of liver damage were observed earlier among high-dose treatments. Among the high-dose treatment, 1387 differentially expressed genes in 8 clusters were identified. Clusters 1 through 4 contained genes that were upregulated among later time points, cluster 5 genes were downregulated at later time points, cluster 6 genes were related with liver toxicity, and clusters 7 and 8 genes were upregulated at earlier time points.

A principal component analysis identified cluster 6 as the major contributor for the 24-hour EE high-dose exposure. Among this expression profile, 10 genes were chosen as they were associated with liver necrosis and differed significantly from controls. These 10 genes performed best under multiple statistical and machine learning approaches and included the genes selenocysteine lyase (SCLY), dermcidin (DCD), ester hydrolase C11orf54 homolog (RGD1309534), solute carrier family 23 member 1 (SLC23A1), betaine-homocysteine S-methyltransferase 2 (BHMT2), triokinase and FMN cyclase (TKFC), sterol regulatory element binding transcription factor 1 (SREBF1), actin binding LIM protein family member 3 (ABLIM3), exostosin-like glycosyltransferase 1 (EXTL1), and cytochrome P450 family 39 subfamily A member 1 (CYP39A1).

This study was limited by not relating these results to data obtained from humans.

These results suggest this set of 10 biomarkers may allow for prediction of future carcinogenic effects of long-term harmful exposure and predict toxic effects among humans that may eventually lead to liver cancer.

Disclosure: Multiple authors declared affiliations with industry. Please refer to the original article for a full list of disclosures.


Smith BP, Auvil LS, Welge M, et al. Identification of early liver toxicity gene biomarkers using comparative supervised machine learning. Sci Rep. 2020;10(1):19128. doi:10.1038/s41598-020-76129-8

This article originally appeared on Gastroenterology Advisor