Chemosensitivity profile of cancer cell lines and identification of genes determining chemosensitivity by an integrated bioinformatical approach using cDNA arrays Free

Predicting the chemosensitivity of individual patients is important to improve the efficacy of cancer chemotherapy. An approach to this end is to understand the genes that determine the chemosensitivity of cancer cells. Many genes have been described that determine the sensitivity to multiple drugs, including drug transporters (1–3) and metabolizing enzymes (4–6). Genes determining the sensitivity to specific drugs have also been reported. For example, increased activities of γ-glutamyl hydrolase (7) and dihydrofolate reductase (8) are resistant factors for methotrexate; increased activities of thymidylate synthase (9), metallothionein (10), and cytidine deaminase (11) are resistant factors for 5-fluorouracil (5-FU), cisplatin, 1-β-d-arabinofuranosylcytosine, respectively; and increased activity of NQO1 (12) is a sensitive factor for mitomycin C (MMC). However, the chemosensitivity of cancer cells is not determined by a handful of genes. These genes are not sufficient to explain the variation of the chemosensitivity of cancer cells.

Recently, attempts were made to predict the chemosensitivity of cancers using genome-wide expression profile analyses, such as cDNA microarray and single nucleotide polymorphisms (13–18). For example, Scherf et al. (18) and Zembutsu et al. (15) reported the analysis of genes associated with sensitivity to anticancer drugs in a panel of human cancer cell lines and in human cancer xenografts, respectively. Tanaka et al. (17) presented prediction models of anticancer efficacy of eight drugs using real-time PCR expression analysis of 12 genes in cancer cell lines and clinical samples. We also analyzed chemosensitivity-related genes in 39 human cancer cell lines (JFCR-39; ref. 19) and validated the association of some of these genes to chemosensitivity using additional cancer cell lines (20). These genes can be used as markers to predict chemosensitivity. Moreover, some of these genes may directly determine the chemosensitivity of cancer cells.

In the present study, we established a new panel of 45 human cancer cell lines (JFCR-45) derived from tumors from three different organs: breast, liver, and stomach. Using JFCR-45, we attempted to analyze the heterogeneity of chemosensitivity in breast, liver, and stomach cancers. We assessed their sensitivity to 53 anticancer drugs and developed a database of chemosensitivity. Then, we analyzed gene expression in 42 human cancer cell lines using cDNA arrays and stored them in the gene expression database. Using these two databases, we extracted genes whose expression was correlated to chemosensitivity. We further screened them to identify genes that could change the sensitivity to anticancer drugs using an in vitro gene transfection assay.

We established a panel of JFCR-45 that included a portion of JFCR-39 and the 12 stomach cancer cell lines described previously (19, 20). They consist of the following cell lines: breast cancer cells HBC-4, BSY1, HBC-5, MCF-7, MDA-MB-231, KPL-3C (21), KPL-4, KPL-1, T-47D (22), HBC-9, ZR-75-1 (23), and HBC-8; liver cancer cells HepG2, Hep3B, Li-7, PLC/PRF/5, HuH7, HLE, HLF (24), HuH6 (25), RBE, SSP-25 (26), HuL-1 (27), and JHH-1 (28); and stomach cancer cells St-4, MKN1, MKN7, MKN28, MKN45, MKN74, GCIY, GT3TKB, HGC27, AZ521 (29), 4-1ST, NUGC-3, NUGC-3/5-FU, HSC-42, AGS, KWS-1, TGS-11, OKIBA, ISt-1, ALF, and AOTO. The AZ521 cell line was obtained from the Cell Resource Center for Biomedical Research, Institute of Development, Aging and Cancer, Tohoku University (Sendai, Japan). The 4-1ST, OKIBA, and AOTO cell lines were provided by Dr. Tokuji Kawaguchi (Department of Pathology, Cancer Institute, Japanese Foundation for Cancer Research, Tokyo, Japan). All cell lines were cultured in RPMI 1640 (Nissui Pharmaceutical, Tokyo, Japan) with 5% fetal bovine serum, penicillin (100 units/mL), and streptomycin (100 μg/mL) at 37°C under 5% CO₂.

Growth inhibition experiments were done to assess the chemosensitivity to anticancer drugs. Growth inhibition was measured by determining the changes in the amounts of total cellular protein after 48 hours of drug treatment using a sulforhodamine B assay. The GI₅₀ values, which represent 50% growth inhibition concentration, were evaluated as described before (30, 31). Several experiments were done to determine the median GI₅₀ value for each drug. Absolute values were then log transformed for further analysis.

Actinomycin D, 5-FU, tamoxifen, cytarabine, radicicol, melphalan, 6-mercaptopurine, 6-thioguanine, and colchicine were purchased from Sigma (St. Louis, MO). The anticancer agents in clinical use were obtained from the company specified in parentheses, and those under development were kindly provided by the company specified as described below: aclarubicin and neocarzinostatin (Yamanouchi Pharmaceutical, Tokyo, Japan); oxaliplatin (Asahi Kasei, Tokyo, Japan), HCFU (Nihon Schering, Osaka, Japan); doxifluridine (Chugai Pharmaceutical, Tokyo, Japan); toremifene, bleomycin, and estramustine (Nippon Kayaku, Tokyo, Japan); daunorubicin and pirarubicin (Meiji, Tokyo, Japan); doxorubicin, epirubicin, MMC, vinorelbine, and l-asparaginase (Kyowa Hakko Kogyo, Tokyo, Japan); peplomycin, etoposide, NK109, and NK611 (Nippon Kayaku); vinblastine, vincrinstine, IFN-γ, and 4-hydroperoxycyclophosphamide (Shionogi, Tokyo, Japan); carboplatin and cisplatin (Bristol-Myers Squibb, New York, NY); mitoxantrone and methotrexate (Wyeth Lederie Japan, Tokyo, Japan); cladribine (Janssen Pharmaceutical, Titusville, NJ); amsacrine (Pfizer Pharmaceutical, formerly Warner Lambert, Plymouth, MI); camptothecin, irinotecan, and SN-38 (Yakult, Tokyo, Japan); paclitaxel (Bristol-Myers Squibb); docetaxel and topotecan (Aventis Pharma, Strasbourg, France); IFN-α (Sumitomo Pharmaceutical, Osaka, Japan); IFN-β (Daiichi Pharmaceutics, Tokyo, Japan); gemcitabine (Eli Lilly Japan, Kobe, Japan); E7010 and E7070 (Eisai, Tokyo, Japan); dolastatine 10 (Teikoku Hormone MFG, Tokyo, Japan); and TAS103 (Taiho Pharmaceutical Co., Tokyo, Japan).

Expression profiles of 3,537 genes in 42 human cancer cell lines were examined using Atlas Human 3.6 Array (BD Biosciences Clontech, Inc., Franklin Lakes, NJ) in duplicates. Experiments were done according to the manufacturer's instructions. Briefly, cell lines were harvested in log phase. Total RNA was extracted with TRIzol reagent (Invitrogen, Inc., Carlsbad, CA) and purified with Atlas Pure Total RNA Labeling System. Purified total RNAs were converted to ³²P-labeled cDNA probe by SuperScript II (Invitrogen). cDNA probe was hybridized to the Atlas Array overnight at 68°C and washed. Hybridized array was detected with PhosphorImager (Molecular Dynamics, Inc. Sunnyvale, CA). Scanned data were transformed to the numerical value with Atlas Image 2.0 software (BD Biosciences Clontech) and normalized by dividing by the value of 90% percentile of all genes in each experiment. Then, the intensities of the genes were defined by the average of intensities of duplicate results. The genes whose expression levels differed more than twice between the duplicates were eliminated from subsequent analysis. When the intensities of gene expression in both arrays were below the threshold value, they were given the value of threshold and were used for analysis. We determined the values of threshold of the normalized data as 30% of the value of 90% percentile. Then, log₂ was calculated for each expression value.

Hierarchical clustering using average linkage method was done by “Gene Spring” software (Silicon Genetics, Inc. Redwood, CA). Pearson correlation coefficients were used to determine the degree of similarity.

The genes whose expressions were observed in >50% of all cell lines examined were selected for the correlation analysis. The degree of similarity between chemosensitivity and gene expression were calculated using the following Pearson correlation coefficient formula:

where x_i is the log expression data of the gene x in cell i, y_i is the log sensitivity (∣log₁₀GI₅₀∣) of cell i to drug y, x_m is the mean of the log expression data of the gene x, and y_m is the mean sensitivity (∣log₁₀GI₅₀∣) of drug y. A significant correlation was defined as P < 0.05.

Candidate genes related to the chemosensitivity were cloned into the vector pcDNA3.1/myc-His A (Invitrogen). Transfection of HT1080 cells with the plasmid DNA was carried out using LipofectAMINE Plus reagent (Invitrogen). The transfection efficiency was monitored by green fluorescent protein fluorescence. The fluorescence of green fluorescent protein was observed in >90% of the green fluorescent protein–transfected HT1080 (data not shown). Twenty-four hours after the transfection, proper concentrations of MMC were added and the cells were treated for 24 hours. Efficacies of anticancer drugs were determined by measuring the growth inhibition. Cell growth was measured by following [³H]thymidine incorporation. [³H]thymidine (0.067 MBq) was added to each well and incubated at 37°C for 45 minutes. Cells were washed with prewarmed PBS(−) and fixed with 10% TCA on ice for 2 hours. After fixing, cells were washed with 10% TCA and lysed with 0.1% SDS-0.2 N NaOH solution. After incubation at 37°C, the lysed mixture was neutralized with 0.25 mol/L acetic acid solution. [³H]thymidine incorporated into the cells was determined using scintillation counter. All experiments, except for interleukin (IL)-18, were done four times.

Sensitivity to 53 drugs was assessed as described in Materials and Methods. The known modes of actions and the value of ∣log₁₀GI₅₀∣ of 53 anticancer drugs in each of the 45 cell lines are summarized in Table 1. The ∣log₁₀GI₅₀∣ indicated here is the median value of multiple experiments. The chemosensitivity of the cell lines differed even among those derived from the same organ. These data were stored in a chemosensitivity database. Figure 1 shows the classification of the anticancer drugs by hierarchical clustering analysis based on chemosensitivity, ∣log₁₀GI₅₀∣, of JFCR-45. As shown, the 53 drugs were classified into several clusters, each consisting of drugs with similar modes of action [e.g., one cluster included topoisomerase (topo) I inhibitors, such as camptothecin, topotecan, and SN-38]. The second cluster comprised tubulin binders, including taxanes and Vinca alkaloids. 5-FU and its derivatives were also clustered into a single group. These results indicated that our system using JFCR-45 was able to classify the drugs based on their modes of action, which is in agreement with previous findings using NCI-60 and JFCR-39 (18, 19, 32).

Using a cDNA array, we examined the expression of 3,537 genes in 42 cell lines of JFCR-45. Based on these expression profiles, hierarchical clustering was done. In a few experiments, cell lines derived from the same organ were clustered into a group (Fig. 2). Breast cancer cell lines, except KPL-4, formed one cluster. Liver and stomach cancer cell lines clustered separately from the breast cancer cell lines and formed tissue-specific subclusters. However, four stomach cancer cell lines, AOTO, ISt-1, TGS-11, and HGC27, were intercalated into a cluster of liver cancer cell lines. These results suggested that the established cell lines maintained characteristics of their organ of origin as far as the gene expression profile was concerned.

To investigate genes that may be involved in chemosensitivity, we integrated the two databases and did a correlation analysis between gene expression and drug sensitivity. Comprehensive calculations for the Pearson correlation coefficients were done on the expression of 3,537 genes and sensitivity to 53 drugs in 42 cell lines. We selected genes that satisfied the following criteria: showing a P of correlation < 0.05 between the expression of the gene and its sensitivity to a certain drug and being significantly expressed in >50% of the cell lines. We examined the data for the distribution by scatter graph analysis and removed those data showing a highly non-normal distribution. The higher the expression of the gene showing positive correlation, the higher the sensitivity was to the drug (i.e., this gene was a sensitive candidate gene). In contrast, genes that showed a negative correlation with chemosensitivity were resistant candidate genes. Consequently, different sets of genes were extracted with respect to each of the 53 drugs. Table 2 shows sets of genes whose expression was correlated with the sensitivity of 42 cell lines to MMC, paclitaxel, vinorelbine, and SN-38. As for MMC, 20 genes were extracted as sensitive genes and 10 genes were extracted as resistant candidate genes. Some of these genes (such as JUN, EMS1, and NMBR) are related to cell growth, whereas others included various types of genes (such as SOD1, PELP1, SFRS9, etc.). Similarly, many sensitive and resistant candidate genes were extracted with the other drugs tested. We further applied a Pearson correlation analysis to the cell lines originating from the same organ. Genes whose expressions were correlated with the MMC sensitivity in 10 breast cancer, 12 liver cancer, and 20 stomach cancer cell lines are shown in Table 3. As described previously (19, 20), these genes may predict chemosensitivity.