Vol. 2, 199–205, February 2003 Molecular Cancer Therapeutics 199 Relevance Network between Chemosensitivity and Transcriptome in Human Hepatoma Cells1 Masaru Moriyama,2 Yujin Hoshida, topoisomerase II ␤ expression, whereas it negatively Motoyuki Otsuka, ShinIchiro Nishimura, Naoya Kato, correlated with expression of carboxypeptidases A3 Tadashi Goto, Hiroyoshi Taniguchi, and Z. Response to nimustine was associated with Yasushi Shiratori, Naohiko Seki, and Masao Omata expression of superoxide dismutase 2. Department of Gastroenterology, Graduate School of Medicine, Relevance networks identified several negative University of Tokyo, Tokyo 113-8655 [M. M., Y. H., M. O., N. K., T. G., H. T., Y. S., M. O.]; Cellular Informatics Team, Computational Biology correlations between gene expression and resistance, Research Center, Tokyo 135-0064 [S. N.]; and Department of which were missed by hierarchical clustering. Our Functional Genomics, Graduate School of Medicine, Chiba University, results suggested the necessity of systematically Chiba 260-8670 [N. S.], Japan evaluating the transporting systems that may play a major role in resistance in hepatoma. This may provide Abstract useful information to modify anticancer drug action in Generally, hepatoma is not a chemosensitive tumor, hepatoma. and the mechanism of resistance to anticancer drugs is not fully elucidated. We aimed to comprehensively Introduction evaluate the relationship between chemosensitivity and Hepatoma is a major cause of death even in developed gene expression profile in human hepatoma cells, by countries, and its incidence is increasing (1). Despite the using microarray analysis, and analyze the data by progress of therapeutic technique (2), the efficacy of radical constructing relevance networks. therapy is hampered by frequent recurrence and advance of In eight hepatoma cell lines (HLE, HLF, Huh7, Hep3B, the tumor (3). Although chemotherapy was initially expected PLC/PRF/5, SK-Hep1, Huh6, and HepG2), the baseline to provide a breakthrough for the treatment of advanced expression levels of 2300 genes were measured by hepatoma, the overall results were disappointing (4, 5). We cDNA microarray. The concentrations of eight cannot identify the few patients who will respond to chem- anticancer drugs (nimustine, mitomycin C, cisplatin, otherapy, and we do not have clear explanation of the mech- carboplatin, doxorubicin, epirubicin, mitoxantrone, and anism of resistance to treatment. 5-fluorouracil) needed for 50% growth inhibition were The development of hepatoma is a complicated process, examined and used as a measure of chemosensitivity. associated with numerous genetic factors. Large-scale tran- These data were combined and comprehensive pair- scriptional profiling allows us to obtain large amounts of data wise correlations between gene expression levels and about complicated biological processes. However, to interpret the 50% growth inhibition values were calculated. the result of DNA microarray experiments, we have to extract Significant correlations with significance were used to the biologically significant information from the many changes construct networks of similarity. that are random or not related to the phenomena studied. Fifty-two relations, including 42 genes, were Relevance networks offer a method to construct networks selected. Among them, nearly 20% were various types of similarity with various type of information, including mutual of transporters, and most of them negatively correlated information and the correlation status (6). In this study, we with chemosensitivity. Transporter associated with applied this methodology to provide new information on antigen processing 1 was associated with resistance to chemosensitivity in hepatoma. mitoxantrone, consistent with previous reports. Other transporters were not reported previously to associate Materials and Methods with chemosensitivity. Resistance to doxorubicin and Hepatoma Cells. Eight hepatoma cell lines (HLE, HLF, Huh7, its analogue, epirubicin, were positively correlated with Hep3B, PLC/PRF/5, SK-Hep1, Huh6, and HepG2) were ob- tained from RIKEN Cell Bank (Tsukuba, Japan). They were maintained using DMEM containing 10% fetal bovine serum. Received 12/2/02; accepted 12/2/02. Anticancer Drugs. We tested eight drugs used currently The costs of publication of this article were defrayed in part by the for the treatment of hepatoma in clinical practice: an antime- payment of page charges. This article must therefore be hereby marked 3 advertisement in accordance with 18 U.S.C. Section 1734 solely to indi- tabolite (5-FU ), two platinum anticancer agents (CDDP and cate this fact. 1 Supported in part by the Health Science Research Grants for Medical Frontier Strategy Research from the Ministry of Health, Labor, and Welfare of Japan, and by grants-in-aid for scientific research from the Ministry of 3 The abbreviations used are: 5-FU, 5-fluorouracil; ACNU, nimustine; Education, Culture, Sports, Science and Technology, Japan. ADM, doxorubicin; CBDCA, carboplatin; CDDP, cisplatin; CA1, carbonic 2 To whom requests for reprints should be addressed, at Department of anhydrase 1; CPA3, carboxypeptidase A3; CPZ, carboxypeptidase Z; Gastroenterology, Graduate School of Medicine, University of Tokyo, EpiADM, epirubicin; GI50, 50% growth inhibition activity of the drug; topo, 7-3-1 Hongo, Bunkyo-ku, Tokyo 113-8655, Japan. Phone: 81-3-3815- topoisomerase; MIT, mitoxantrone; MMC, mitomycin C; SOD, superoxide 5411, extension 33056; Fax: 81-3-3814-0021; E-mail: moriyamam- dismutases; TAP, transporter associated with antigen processing; TRG, [email protected]. T-cell receptor ␥ locus; TOP2B, DNA topoisomerase II ␤. Downloaded from mct.aacrjournals.org on October 1, 2021. © 2003 American Association for Cancer Research. 200 Network of Genes and Anticancer Drugs in Hepatoma CBDCA), two alkylating agents (ACNU and MMC), and three Relevance Networks. Using log-transformed expression (topo) II inhibitors (ADM, EpiADM, and MIT). CDDP, CBDCA, ratios of selected genes and the GI50 values of the antican- ADM, MMC, MIT, and 5-FU were purchased from Sigma Co. cer drugs, similarity metrics, R, were calculated in a com- (St. Louis, MO). ACNU was obtained from Sankyo Co. (To- prehensive pair-wise manner as a square of the Pearson kyo, Japan). EpiADM was obtained from Pharmacia Co. correlation coefficient with original sign (positive or negative). (Peapack, NJ). The correlation coefficient, r, was calculated by the following Measurement of Response to Drugs. The response of formula: each cell line to the anticancer drugs was determined as ϭ ͚ Ϫ Ϫ ͙͚ Ϫ 2 Ϫ 2 follows. Briefly, cells were grown in 24-well microplates r (xi xmean)(yi ymean)/ (xi xmean) (yi ymean) (IWAKI Glass, Chiba, Japan) and exposed to the drugs for where xi represents the log expression ratio (log2Cy5:Cy3) of 48 h. Cell growth was assessed by 3-(4,5-dimethylthiazol-2- gene x in cell line i, whereas yi is the log response (log10GI50) yl)-2,5-diphenyltetrazolium bromide assay. The drug con- to drug y in cell line i. To generate a reference distribution of centration required to inhibit cell growth by 50% in compar- R, measurements of expression ratios and GI50 values were ison with untreated controls was designated as the growth randomly permuted 100 times, and calculated Rs were re- inhibition activity of the drug (GI50). Data were averages of corded. On the basis of these random data, we set threshold three independent experiments. values of R for selection of genes with significance levels of RNA Extraction. Total cellular RNA was extracted using P Ͻ 0.001, P Ͻ 0.01, and P Ͻ 0.05. We considered that for an acid guanidine isothiocyanate phenol-chloroform method a given pair of genes, a relationship with R higher than the according to the manufacturer’s instructions (ISOGEN Rea- threshold value may represent a significant biological rela- gent; Nippon Gene Co., Tokyo, Japan). RNA was treated with tionship. We then connected these genes with each other to RNase-free DNase to remove contaminating genomic DNA, form clusters or relevance networks of related genes and and RNA integrity was confirmed by gel electrophoresis and drugs. For these calculations, we developed custom Perl ethidium bromide staining. Polyadenylic acid mRNA was codes (included in bioperl release 1.2).4 Finally, the networks obtained using the Oligotex-dT30 mRNA purification kit were visualized as graphs using a custom JAVA application. (TaKaRa Co., Tokyo, Japan). The functions of the selected genes were obtained from cDNA Microarray. We assessed gene expression using web-based databases, e.g., the National Center for Biotech- in-house cDNA microarray developed by ourselves in the nology Information-Locuslink,5 Gene Ontology,6 Kyoto En- Helix Research Institute (Chiba, Japan; Ref. 7). A total of cyclopedia of Genes and Genomes,7 and other linked data- 2300 named, sequence-validated human cDNAs (Research bases. Genetics) were spotted onto carbodiimide-coated glass slides using a robot (SPBIO-2000; Hitachi Software Engi- Results neering Co., Tokyo, Japan). GI50 Values of Anticancer Drugs. Table 1 summarizes the Microarray experiments were performed as described pre- GI50 values of eight anticancer drugs in eight hepatoma cell viously (8), and for each gene, expression was compared lines. Drugs and their analogues (CDDP and CBDCA; ADM with expression in normal human liver RNA (Clontech, Palo and EpiADM) showed relatively similar ranges and average Alto, CA) using the Cy5:Cy3 ratios. To control for