Fig. 3. Boxplots of the ALT and SDH data from the training samples with 48 h exposures included. The Boxplots were performed using R statistical software. The median, 25%, and 75% quartiles are illustrated in the plots as well as the range. (a) Blood was drawn from each experimental animal and was analyzed to determine the alanine aminotransferase (ATL) level. ALT measurements were log transformed (base 10) and plotted for each dose and time combination (as one column). Generally, there were four animals for each condition (time and dose combination). There are more animals for 48 h time line, with 7 animals for 2,000 mg/kg level and 26 animals for the control (0 mg/kg). (b) Blood was drawn from each experimental animal and was analyzed to determine the sorbitol dehydrogenase (SDH) enzyme level. SDH measurements were log transformed (natural logarithm) and plotted for each dose and time combination (as one column). Generally, there were four animals for each condition (time and dose combination), unless we lost an animal. There are more animals for the 48 h time line, with 8 animals for 50 mg/kg level and 27 animals for the control (0 mg/kg).

Fig. 4. Visualization of the rat blood APAP training set samples and patterns of gene expression. (a) PCA was performed using Partek v 6.0 analytical software on the gene expression data from the rat blood APAP training set using the 152 genes selected by the DME ANOVA model. The samples were projected into 3D space using the top three PCs. The total amount of variability captured by the PCs is 81.6%. Blue circles represent samples dosed with subtoxic levels of APAP. Red circles denote the samples dosed with toxic levels of APAP. Rat no. 3336 dosed with 2,500 mg/kg APAP for 12 h is shown. (b) The microarray data from the rat blood APAP training set were analyzed (see Materials and Methods) to extract the gene expression patterns (graphs 1-8) that distinguish between subtoxic and toxic doses of the agent. The analysis was performed using intra-groups consisting of a given dose/time point's four biological replicates and two fluor-flip hybridizations, nine inter-groups separating each dose/time point, a signal-to-noise ratio (S/N) = 3.0, correlation r value = 0.64, and magnitude of the fold change = 1.5. The x axis represents the time/dose point of the sample. Open symbols, samples dosed with a subtoxic amount of APAP (150 mg/kg); filled symbols, samples dosed with a toxic amount of APAP (shaded, 1,500 m/kg; black, 2,500 mg/kg). Exposure durations: triangle, 6 h; circle, 12 h; square, 24 h. The y axis signifies the log base 2-fold change of the gene expression ratio intensity values.

Fig. 5. PCA and visualization of the rat APAP training and test data sets. (a) PCA on the gene expression data from the 248 genes selected from the eight patterns extracted by EPIG from the blood training set was performed using Partek v 6.0 analytical software. The samples were projected into 3D space using the top three PCs. The total amount of variability captured by the PCs is 70.4% (PC1 = 35.9%; PC2 = 28.2%; PC3 = 6.27%). Green circles, samples dosed with subtoxic levels of APAP; purple circles, the samples dosed with toxic levels of APAP; red circles, blood blinded samples predicted as subtoxic/nontoxic dosed; blue circles, blinded blood samples predicted as dosed with a toxic amount of APAP. (b) PCA of the APAP blood training set and the blinded rat liver samples was performed using Partek v 6.0 analytical software on the gene expression data from the 248 genes selected from the eight patterns extracted by EPIG from the blood training set. The samples were projected into 3D space using the top three PCs. The total amount of variability captured by the PCs is 68.3% (PC1 = 35.5%; PC2 = 26.9%; PC3 = 5.87%). Green circles, blood samples dosed with subtoxic levels of APAP; purple circles, samples dosed with toxic levels of APAP; red circles, blinded liver samples predicted as subtoxic/nontoxic dosed; blue circles, blinded liver samples predicted as dosed with a toxic amount of APAP.

Fig. 6. Hierarchical cluster analysis of peripheral blood cell gene expression data from five human APAP overdose victims and three control individuals. (a) For 66 of the 270 genes discriminating between subtoxic/nontoxic and toxic exposure to APAP in rat blood, human orthologous genes are present on the human Agilent chip. Shown is a hierarchical cluster analysis of human data with those 66 orthologous genes. Clustering of the samples (columns) using the gene expression data (log base 10 ratio values) from the 66 orthologous genes was performed with the average linkage agglomerative clustering method and cosine correlation as the dissimilarity measure. The length of the branches of the dendrogram represents the amount of similarity between samples or clusters of samples. The colors in the heat map matrix denote the differential expression relative to the control sample used for hybridization. Red, up-regulation; green, down-regulation; black, no change; gray, missing data. The numbers represent the patient identifier. (b) The numbers on the branches denote the cosine correlation between samples or clusters of samples.

Fig. 7. Hierarchical cluster analysis of the gene expression data from APAP-treated training samples, samples exposed to a compendium of hepatotoxicants, or samples treated with LPS. Two-way hierarchical cluster analysis (genes in the columns and arrays as the rows) on the APAP-treated training samples, samples from rat blood exposed to hepatotoxicants (1), or samples treated with LPS (2) was performed using the gene expression data (log base 10 ratio values) from the 270 genes acquired from the union of the predictors from each prediction approach. The average linkage agglomerative clustering method was used with cosine correlation as the dissimilarity measure. The length of the branches of the dendrogram represents the amount of similarity between samples or clusters of samples. The colors in the heat map matrix denote the differential expression relative to the control sample used for hybridization. Red, up-regulation; green, down-regulation; black, no change. (a) The two-way hierarchical clustering heat map containing all of the samples and the 270 genes. (b) A selected node of clustered samples, genes and the heat map. The red rectangle indicates some of the APAP-toxic dosed samples (24 h duration), the green rectangle signifies the N-nitrosmorpholine 300 mg/kg for 48 h exposed samples, and the purple rectangle represents the LPS exposed samples. (c) A selected node of clustered samples, genes and the heat map. The red rectangle indicates some of the APAP-toxic dosed samples (24 h duration), the yellow rectangle signifies Diquat dibromide 25 mg/kg for 24 h exposed samples, and the black rectangle represents 1,2 dichlorobenze 1,500 mg/kg for 24 h exposed samples.

1. Lobenhofer EK, Boorman GA, Phillips KL, Heinloth AN, Malarkey DE, Blackshear PE, Houle C, Hurban P (2006) Toxicol Pathol 34:921-928.

2. Fannin RD, Auman JT, Bruno ME, Sieber SO, Ward SM, Tucker CJ, Merrick BA, Paules RS (2005) Physiol Genomics 21:92-104.