Fig 1.
Concordance of azathioprine in rat primary hepatocytes and rat liver.
Azathioprine incubated at 4 μM for 24 hours in RPH (X-axis) is compared to rat liver 6 hours after administering a 30 mg/kg dose (Y-axis) using A) 9071 liver-expressed genes, B) GSA analysis of 1840 gene sets and C) 415 co-expression modules. The concordance is quantified using the Pearson correlation coefficient R and percent overlap among the highest 5% of differentially expressed features. Features in blue are in the top 5% for RPH only, those in green in the top 5% for rat liver only, those in pink in the top 5% for both models, and those in orange (only observed here for gene-level analysis) are in the top 5% for both models but induced in one model and repressed in the other. The Venn diagrams denote overlap among the top 5% of features.
Table 1.
Percentage of concordance variation explained by treatment characteristics when comparing two different treatments of same drug within TG-GATEs rat liver or RPH.
Fig 2.
Concordance of drug-induced effects in rat and mouse liver.
Pairs of experiments involving the same drug are compared via the Pearson R and percent overlap metrics, and average concordance is calculated between TG rat liver (reference set) and A) TG rat liver (within-source), B) DM rat liver (cross-source) and C) GEO mouse liver (cross-source and cross-species). Concordance is shown additively for 3 levels of constraint on dose and time differences: experiments at the same time point and doses within 5 fold (pink), time within 2 fold and doses within 10 fold (blue), no constraint on doses or time (green). In each case, experiments are compared to TG rat liver and the most concordant experiment selected within the level of constraint. Only experiments that have a match to TG rat liver at the most constrained level are considered at the other levels. Because concordance can only improve upon removing dose and time constraints, results are shown additively with the total bar height denoting the concordance achieved with no constraints. Numeric values represent the probability that the given level of concordance exceeds that of random pairs involving different drugs for the most stringent level considered (pink bars). Experiment pairs are separated into 3 ranges based on the level of transcriptional activity for the least perturbing of two treatments (lowest quartile have avg. abs. EG ≤ 0.28, highest quartile have avg. abs. EG > 0.46; thresholds selected using 3528 TG rat liver experiments).
Fig 3.
Comparison of in vitro vs. in vivo treatment effects for azathioprine.
Transcriptional effects after treatment of rat primary hepatocytes with 4 μM azathioprine for 24 hours were compared to 24 different in vivo azathioprine experiments in rat liver. The low, medium and high in vivo doses are denoted with separate lines, and concordance is assessed vs. time using 6 analysis methods (two concordance metrics and genes / GSA / module analysis). The circled point denotes the most concordant in vivo dose / time condition for each of the 6 analysis methods.
Fig 4.
Concordance of drug-induced effects in cultured hepatocytes.
Pairs of experiments involving the same drug are compared via the Pearson R and percent overlap metrics, and average concordance is calculated between A) TG rat liver vs. TG RPH, B) TG RPH vs. TG HPH and C) TG HPH vs. GEO HepG2. The data source shown in italics is the reference for each comparison, to which experiments from the other source are compared. For A), only the optimistic comparison is performed because dose / time are not comparable in vivo vs in vitro. Refer to Fig 2 caption for details.
Fig 5.
Transcriptional effects of cell culture in the context of rat liver perturbations achieved with drug treatment.
A) Distribution of average absolute eigengene scores for 4182 rat liver drug treatments from DM and TG, and 6 baseline expression comparisons of control samples (shown in red text in the inset). The solid and dashed vertical lines denote the median and 90th percentile values for avg. abs. EG. Rat liver histology sections: B) Unaffected liver. A central vein (asterisk) and small portal triad (arrowhead) are marked. Mild hepatocellular glycogen (recognized by cytoplasmic pallor) is apparent in the periportal region. C) Liver histology 1 day after the last dose of methapyrilene administered at 100 mg/kg for 29 days. Biliary hyperplasia (to the left of the arrows), correlating with increased alkaline phosphatase (ALP) and γ-glutamyltransferase (GGT) activities (S7 Table), and hepatocellular apoptosis/single cell necrosis (arrowheads) are noted. Hepatocytes in the right side of the image are hypertrophied and exhibit prominent anisonucleosis. D) Liver histology 1 day after bortezomib administration at 1 mg/kg. Acute hepatocellular necrosis (left of the arrows) and mild hepatocellular vacuolation (arrowheads) are noted. Images were extracted from the Open TG-GATES website (http://toxico.nibiohn.go.jp/english/).
Table 2.
Pairwise correlation of baseline gene expression in untreated liver or hepatocytes.
Fig 6.
Correlation analysis of transcriptional changes in hepatocyte cell culture.
Co-expression network analysis of expression changes when comparing mRNA between cultured hepatocytes vs. liver or immediately after hepatocyte isolation (vs. 0hrs RPH). Cells are colored on the Pearson correlation coefficient R obtained by comparing module scores for two conditions. Conditions are clustered to group those showing similar scores across 415 modules, revealing broad similarity following 24–48 hours in culture for RPH, HPH, MPH and HepG2 cells from various sources.