ABSTRACT: Transcriptomics is developing into an invaluable tool in toxicology. The aim of this study was, using a transcriptomics approach, to identify genes that respond similarly to chemicals (including drugs and industrial compounds) in both rat liver in vivo and in cultivated hepatocytes, which are up- or downregulated by many different test compounds. For this purpose, we analyzed Affymetrix gene array data from 162 compounds that were previously tested in a concentration-dependent manner in rat livers in vivo and rat hepatocytes cultivated in sandwich culture. These data were obtained from the Japanese Toxicogenomics (TGP) and North-Rhine-Westphalian (NRW) data sets, which consist of 138 and 29 compounds, respectively, and have 5 mutual compounds between them. The in vitro gene array data from the NRW data set were generated in the present study, while TGP is publicly available. For each of the data sets, the overlap between up- or down-regulated genes in vitro and in vivo was identified, further named in vitro-in vivo consensus genes. Interestingly, an overlap of in vivo-in vitro consensus genes was obtained between both data sets, which were 21-times (up-regulated genes) or 12-times (down-regulated genes) en-riched compared to random expectation. This is remarkable since the TGP and NRW data sets contained only five mutual compounds. Finally, the genes in the TGP and NRW overlap were used to identify the upregulated genes with the highest com-pound coverage, resulting in a 7-gene set of Cyp1a1, Ugt2b1, Cdkn1a, Mdm2, Aldh1a1, Cyp4a3, and Ehhad. This 7-gene set was then successfully tested with structural analogues of valproic acid that are not present in the TGP and NRW data sets. In conclusion, the 7-gene set identified in the present study responds similarly in vitro and in vivo and is induced by a wide range of different chemicals. Despite these results, transcriptomics with cultivated rat hepatocytes remains a challenge, because many genes are up- or downregulated by the culture conditions, respond differently to test compounds in vitro and in vivo, and shows a higher variability in the in vitro system compared to the corresponding in vivo data. The raw data for the TGP study are available at https://toxico.nibiohn.go.jp/ For the normalization of the entire set of expression arrays, the RMA algorithm was used. The results are summarized in the publicly available toxicotranscriptomics directory (http://wiki.toxbank.net/toxicogenomics-map/).
