Discovery of Differentially Expressed Genes Associated with Paclitaxel Resistance Using cDNA Array Technology: Analysis of Interleukin (IL) 6, IL-8, and Monocyte Chemotactic Protein 1 in the Paclitaxel-resistant Phenotype¹

Paclitaxel is a novel anticancer agent with activity in a broad range of epithelial cancers, including carcinomas of ovary, breast, head, neck, bladder, lung, and prostate (1, 2, 3, 4). The efficacy of paclitaxel is limited by the development of drug resistance in a population of surviving malignant cells. Defined molecular mechanisms for acquired tumor cell resistance to paclitaxel include overexpression of the drug efflux pump MDR1³ and differential expression of β-tubulin isotypes or β-tubulin gene point mutations (5, 6, 7, 8). Recent data also suggest that p53 status and mitosis checkpoint control is important in determining the sensitivity of cells to paclitaxel (9, 10). In general, there is little ex vivo clinical data and limited understanding of which, if any, of these mechanisms influences paclitaxel resistance in humans. Whereas most drug resistance studies have focused on a few specific gene(s), the aim of this study was to define broader patterns of change by evaluating differential gene expression between paclitaxel-sensitive and -resistant cells.

Techniques commonly used to determine gene expression profiles include RT-PCR, Northern blot analysis, and RNase protection. However, these methods of necessity examine only one gene at a time. A more promising approach, cDNA array analysis, has recently developed and allows rapid evaluation of large numbers of expressed genes. This technique, involving the hybridization of cellular cDNA to high-density nucleic acid arrays, is well suited for evaluating large numbers of gene transcripts. (11, 12, 13).

To evaluate the patterns of gene expression in paclitaxel-sensitive and -resistant ovarian cancer cells, we established a paclitaxel-resistant human ovarian cancer cell line, SKOV-3_TR, from its paclitaxel-sensitive parent, SKOV-3, and compared mRNA levels in the two lines using cDNA array technology. Genes demonstrating changed expression were selected for further evaluation in several other paclitaxel- and/or doxorubicin-resistant cell lines, with particular emphasis on a collection of chemokine and cytokine genes overexpressed in the SKOV-3_TR cell line.

Human ovarian cancer cell line SKOV-3 and some of the other cancer cell lines used in this study were obtained from the American Type Culture Collection (Rockville, MD). Human multiple myeloma doxorubicin-resistant cell line 8226/DOX₄₀ (maintained in the presence of 2.8 μm doxorubicin) and its sensitive counterpart 8226/S were kindly provided by Dr. William S. Dalton (University of South Florida, H. Lee Moffitt Cancer Center, Tampa, FL; Ref. 14). Human breast tumor ADR-resistant cell line MCF-7/ADR (maintained in the presence of 1.7 μm ADR) and its parent cell line, MCF-7, were provided by Dr. Jack Chen (University of Texas, M. D. Anderson Cancer Center, Houston, TX). β-tubulin mutant ovarian carcinoma cell lines PTX 10 and PTX 22 and their sensitive parental line 1A9 were kindly provided by Dr. Tito Fojo (NIH, Bethesda, MD) (8). SKOV-3 and SKOV-3_TR were grown in RPMI 1640 supplemented with 10% fetal bovine serum and penicillin and streptomycin at 100 units/ml and 100 μg/ml, respectively (Life Technologies, Inc., Grand Island, NY). Cells were incubated at 37°C in 5% CO₂-95% air atmosphere and passaged twice every 6 days. All of the drugs were obtained from commercial sources. SKOV-3_TR was selected from SKOV-3 by stepwise increases in paclitaxel concentrations from 0.0001 μm to 0.3 μm.

In vitro cytotoxicity was measured using a tetrazolium-based semiautomated colorimetric (MTT) assay, as described previously (15). The absorbance values were normalized assigning the value of the parent line in media to 1.0 and the value of the no cell blank to 0. Experiments were performed in duplicate.

Poly(A)⁺ RNA was prepared from SKOV-3 and SKOV-3_TR cells using FastTrack mRNA isolation kit (Invitrogen, Carlsbad, CA) according to the manufacturer’s instructions. Contaminating DNA in the samples was removed by treatment with RNase-free DNase (Boehringer Mannheim, Indianapolis, IN). Probes were synthesized from SKOV-3 and SKOV-3_TR mRNA and reagents (CLONTECH) following the manufacturer’s protocol. [³²P]dATP [3.5 μl (10 mCi/ml) Amersham] was used to label each cDNA probe.

After reverse transcription, cDNA probes were purified using CHROMA SPIN-200 DEPC-H₂O column (CLONTECH). The labeled cDNAs were applied to the center of the gel-bed’s flat surface and allowed to absorb into the resin bed. Columns were then washed four times with 250 μl of deionzed H₂O. Incorporation of ³²P into probes was determined using a Geiger counter. Fractions showing the first-peak of highest counts were collected and stored at −70°C.

cDNA arrays (CLONTECH) contain 200–500-bp gene fragments that lack poly(A) tails, repetitive elements, or other highly homologous sequences. The gene fragments cover several groups of genes important in cell growth, differentiation, activation and death. Plasmid and bacteriophage DNAs are included as negative controls to confirm hybridization specificity, along with nine housekeeping cDNAs as positive controls for normalizing mRNA abundance. Two identical cDNA array membranes were hybridized individually with freshly labeled and purified cDNA complex probes made from SKOV-3 and SKOV-3_TR mRNAs. The arrays were prehybridized using ExpressHyb solution (CLONTECH) for 30 min at 68°C (16) with continuous agitation and were hybridized overnight at the same conditions. Arrays were washed for 20 min at 68°C with 2× SSC-1% SDS four times, then for 20 min at 68°C with 0.1× SSC-0.5% SDS two times. The damp membranes were immediately wrapped in plastic wrap, and they were exposed to Kodak BioMax MS X-ray film with an intensifying screen at −70°C.

Total RNA was extracted from cells using TRIzol Reagent (Life Technologies, Inc.) according to the manufacturer’s instructions. RNA was separated by electrophoresis in 1.2% agarose/formaldehyde gels (approximately 5 μg total RNA per lane), transferred to Hybond N-plus membranes (Amersham), and UV cross-linked. Different ³²P-labeled cDNA probes were used for hybridization. Membranes were washed at room temperature with 2× SSC-0.1% SDS for 15 min twice and at 65°C with 0.2× SSC-0.1% SDS for 15 min twice. Blots were exposed to X-ray film using an intensifying screen and later were hybridized with β-actin to confirm RNA amounts loaded into each lane. RT-PCR was performed using Titan One Tube RT-PCR Kit (Boehringer Mannheim) following the manufacturer’s protocol. cDNA clones, obtained from the American Type Culture Collection, were used as probes used for Northern hybridization.

IL-6 and IL-8 levels in culture supernatants were measured using separate quantitative immunoassay ELISA kits according to the manufacturer’s instructions (Genzyme, Boston, MA). Human MCP-1 levels in culture supernatants were determined using an ELISA kit according to manufacturer’s instructions (Endogen, Inc. Woburn, MA). The absorbance of each well was read using a BT 2000 Microkinetics Reader (Bio-Tek Instrument Inc.) at 450 nm. A standard curve was constructed to quantitate the cytokine concentrations in the controls and samples.

SKOV-3_TR was selected from its parent cell line by exposure to stepwise increases in paclitaxel concentrations (from 0.0001 μm to 0.3 μm paclitaxel). The resistant phenotype was found to be stable after 18 months of continuous culture in 0.3 μm paclitaxel. MTT cytotoxicity analysis shows that SKOV-3_TR cells are 100-fold more resistant to paclitaxel as compared with the sensitive parent cell line (IC₅₀s, 350 nmversus 3.5 nm, respectively). In addition, SKOV-3_TR also demonstrates resistance to vincristine and partial cross-resistance to doxorubicin (Fig. 1). Northern analysis reveals strong expression of MDR1 in SKOV-3_TR as compared with no detectable MDR1 expression in SKOV-3. There is no change in MRP expression, whereas lung-resistance-related protein shows a modest 2-fold increase in expression in SKOV-3_TR (Fig. 2; Table 1).

Human cDNA arrays were used to examine relative gene expression between SKOV-3 and SKOV-3_TR. Approximately 15% of the 588 genes were expressed at a level of visual detection on the cDNA arrays. Comparative expression analysis of SKOV-3 versus SKOV-3_TR in the two experiments revealed 18 genes displaying altered signal intensity (Fig. 3). Of the 18 differentially expressed genes, 6 were decreased in expression in SKOV-3_TR, whereas 12 showed increased expression in SKOV-3_TR including several cytokine and chemokine genes, namely IL-6, IL-8, MIP2α and MCP-1 (Table 2).

To confirm array results, gene expression was measured by Northern analysis and/or RT-PCR. Differential expression of 12 of the 18 genes was confirmed with >3-fold differences as determined by Northern analysis (Fig. 4) and/or RT-PCR in all cases. cDNA probes of 6 of the 18 genes were not confirmed. Five of the six genes demonstrated less than 2-fold difference in expression by Northern analysis, whereas cyclin A showed no diffferences in RNA expression by RT-PCR (Table 2).

Differential expression of these 12 genes in the drug-resistant phenotype was studied in several other lines, selected for resistance to paclitaxel or doxorubicin. After 3–6 months of continuous culture in paclitaxel, breast and ovarian cancer cell lines were 10- to 70-fold resistant to paclitaxel as compared with their parent lines. Northern analysis revealed overexpression of MDR1 in all of the paclitaxel-resistant cell lines as compared with their parental cells. MRP expression was not altered between resistant and sensitive cell lines (Fig. 2; Table 1). Northern analysis and RT-PCR evaluating expression of genes identified in the cDNA array analysis in these cell lines have again shown an association of these genes with the drug-resistant phenotype. For example, IL-8 was relatively overexpressed in the Taxol-resistant cell lines SW626_TR, CAOV-3_TR, and MDA435_TR, as compared with their paclitaxel-sensitive parental lines, and in the human multiple myeloma doxorubicin-resistant cell lines 8226/DOX₄₀ and 8226/MDR₁₀V (Fig. 5; Table 3). Interestingly, β-tubulin mutant ovarian carcinoma cell lines PTX10 and PTX22 overexpress the CIP2 mRNA but do not show differences in expression of the other genes identified in this study. Of note, MCF-7_TR does not differentially express any of the 12 genes whose expression is altered in the paclitaxel-resistant phenotype, SKOV-3_TR.

Subsequent analysis focused on a subset of the chemokine and cytokine genes identified through this study. Particular attention was given to the kinetics of gene induction and the verification that changes in gene expression were associated with protein expression. IL-6, IL-8 and MCP-1 concentrations in culture supernatants of resistant and sensitive cell lines were examined by ELISA. SKOV-3_TR, SW626_TR, CAOV-3_TR, NIH:OVCAR-3_TR, and MCF-7/ADR produced high levels of IL-6 (Table 4). In contrast, secretion of IL-6 by MCF-7_TR, MDA 435_TR, and 8226/DOX₄₀ cell lines were nearly equal to the sectetion of the cytokine by the sensitive parental cell lines. IL-8 and MCP-1 were overexpressed in all of the paclitaxel- and ADR-resistant cell lines, but the magnitude of overproduction on culture varied significantly between lines (Table 4). Analysis of this data demonstrated that the cytokine values were not normally distributed, and, hence, a Wilcoxon sum-rank test was performed. All three of the cytokine/chemokines were significantly elevated in the drug-resistant cell lines as compared with the drug-sensitive parental lines (IL-6 and IL-8: both, P = 0.04; MCP-1, P = 0.009).

To determine whether the induction of cytokine expression was a result of rapid induction by paclitaxel prior to the establishment of the drug-resistant phenotype, IL-6, IL-8, and MCP-1 concentrations were evaluated after the paclitaxel-sensitive parent cell lines were exposed to paclitaxel for 48 h. After exposure, there were only modest increases of cytokine expression, and less than that was seen in cells with a paclitaxel-resistant phenotype (Fig. 6).

To evaluate whether the overproduction of these cytokines in the chemotherapy-resistant cells is dependent on continuous drug exposure, the resistant cells lines were grown both in and out of either paclitaxel or ADR for 1 week. At this time, the supernatants were collected and an analysis of the cytokines was performed. Supernatants from resistant cell lines had nearly equivalent levels of IL-6, IL-8, and MCP-1 with or without concurrent paclitaxel/ADR exposure, which suggests that the drug-resistant cell lines constitutively secrete cytokines independent of drug exposure (data not shown).

To exclude the possibility that different cytokine concentrations represented differences in cell densities, the kinetics of changes in IL-6 concentration were evaluated in both parental and drug-resistant SKOV-3 plated at 1 × 10⁵ cells in fresh media with supernatants collected at different time points for IL-6 measurements. The data indicate IL-6 levels accumulated rapidly in SKOV-3_TR both in and out of paclitaxel at levels markedly above that of the SKOV-3 parental line. This secretion continued to increase over time up to 96 h (Fig. 7). MTT-evaluated cell growth of the SKOV-3 and SKOV-3_TR cell lines demonstrates nearly equivalent growth rates with the SKOV-3 cell line demonstrating slightly greater growth rates as compared with the drug-resistant daughter cell line (data not shown).

This study uses high density cDNA array technology to demonstrate the transcriptional differences between the paclitaxel-sensitive SKOV-3 parental line and the paclitaxel-resistant daughter line, SKOV-3_TR. The results reveal that 18 of the 588 transcripts screened have clearly discernible changes between the two cell lines. Northern and RT-PCR techniques confirm transcriptional changes in 12 of the 18 candidate genes. These findings suggest that cDNA array analysis offers the advantage of a low false-positive rate when compared with differential display technology. The latter technique is also labor-intensive and time-consuming (17, 18, 19).

Our analysis was confined to a small portion of the total cellular transcripts and, hence, does not represent a complete transcriptional profile of the SKOV-3_TR drug-resistant phenotype. Review of the 12 genes with altered expression identifies 2 genes, previously linked to the drug-resistant phenotype, ERBB-2 and VEGF (20, 21). Review of the 10 remaining genes reveals that 5 are chemokine or cytokine genes whereas the other 5 genes—CIP2 (22), RIZ (23), NKEFB (24), GADD45 (25), and fos-related antigen 1 (FRA-1) (26) have not been previously linked to drug resistance.

Changes in chemokine/cytokine gene expression were observed for IL-6, IL-8, MCP-1, and MIP2α, but not for IL-1 through -5, IL-7 IL-9, IL-10 through -13, IL-15, IL-17, IFN-γ, granulocyte-colony stimulating factor, granulocyte macrophage-colony stimulating factor, or macrophage-colony stimulating factor. These results are the first to demonstrate the association of MCP-1 secretion with the paclitaxel (and ADR) multidrug-resistant phenotype and add further evidence to the association of IL-6 and IL-8 secretion with clinical paclitaxel resistance. This association has been established in the past for both breast and ovarian carcinoma cell lines as well as a multiple myeloma cell line. The overproduction of these cytokines seems not to represent direct induction by drugs but rather to reflect constitutive production from cells expressing a stable multidrug-resistant phenotype even when cells are grown in the absence of paclitaxel. Essentially all of the drug-resistant cell lines overexpress IL-8 and MCP-1 chemokines with a subset also overexpressing IL-6, which suggests that these proteins may serve as important survival factors that promote the multidrug-resistant phenotype.

IL-6 is a multifunctional cytokine that has been shown to play an important role in immune host-defense function, hematopoiesis, and hepatocyte growth (27, 28). Other studies suggest a role for these cytokines in normal epithelial cell growth, including ovarian epithelium (29). IL-6 is also a growth factor for several tumors, most notably myeloma but also renal cell and cervical carcinomas (30, 31, 32). Also, elevated levels of IL-6 have been found in the serum or ascites of patients with different tumors. These levels are associated with poor prognosis in women with ovarian and breast carcinoma (33, 34) as well as in persons with other malignancies (35, 36). IL-6 is an inhibitor of cytotoxic chemotherapy-associated apoptosis in in vitro systems. For example, IL-6 secretion protects the hormone-resistant prostate tumor cells against the cytotoxic effects of cisplatin and etoposide and its neutralization sensitizes the cells to cytotoxicity (37, 38). It is possible that elevated levels of IL-6 in the tumor microenvironment lead to local chemotherapy resistance. Curiously, paclitaxel resistance has been associated with changes in proapoptotic molecules such as Bax as well the anti-apoptotic molecule BCL-2 (39, 40).

IL-8 has been shown to enhance angiogenesis and tumor metastasis (41, 42). Study of IL-8 demonstrates that this cytokine is overexpressed in malignant as compared with benign ovarian cysts (43). Increased IL-8 expression has also been linked to aggressive cases of ovarian cancer in preclinical models and women with ovarian cancer (44, 45, 46). Other investigators have demonstrated that short-term exposure of paclitaxel to ovarian cancer cell lines in vitro caused the induction of IL-8, but not that of IL-6, in a subset of ovarian cancer cell lines after 18 h of paclitaxel exposure. (47). Our results demonstrate modest induction of IL-8 by paclitaxel in a subset of both human breast and ovarian tumor cells, although the levels of cytokine overexpression are much higher than those in the short-term induction experiments. These data suggest that IL-8 may play a more general role in the multidrug resistance phenotype or the stress response of cells exposed to an antineoplastic agent.

MCP-1 is a low molecular weight chemotactic chemokine important in the migration of monocytic cells as well as acute and chronic inflammatory processes (48). Immunohistochemical studies by Negus et al. (49, 50) have demonstrated that ovarian carcinoma cells express MCP-1 in tumor blocks. These tumors characteristically have immune cell infiltrates, presumably recruited by MCP-1, and other chemokines or cytokines expressed by the tumor cells or surrounding stroma.

There exists the possibility that some of the differentially expressed genes represent epigenetic and/or stress responses to paclitaxel and, thus, are more accurately considered paclitaxel resistance-associated genes rather than genes directly responsible for the paclitaxel-resistant phenotype. Further evaluation of this hypothesis requires functional studies of the individual candidate genes in transfection experiments.

In summary, cDNA array technology provides a rapid and powerful technique for continuing to define the constellation of genes involved in paclitaxel resistance. It is conceivable that the sum total of genes involved in the paclitaxel-resistant phenotype within a single cell line is very large. Whereas many of these changes may be epigenetic, it is possible that unigenetic mechanisms of drug resistance are the exception rather than the rule. This is not to suggest that the expanding pool of ATP-binding cassette transporter pumps (MDR1 and the MRP family) are not important in clinical drug resistance, but rather that the cellular response to paclitaxel may be influenced by a spectrum of genes that control proliferation, checkpoint function, and apoptosis. This hypothesis, if confirmed by functional studies, will have broad implications related to the efforts currently underway to reverse multidrug resistance in the clinical setting.

We thank Drs. William Dalton and Tito Fojo for kindly providing several of the drug-resistant lines. We also thank Dr. Dan Haber for discussion and advice and Drs. Diane Finklestein and Paul Stark for statistical assistance.