Gene expression correlating with response to paclitaxel in ovarian carcinoma xenografts

Introduction

Evaluation of the response to chemotherapeutic regimens is often empirical, mainly based on changes in tumor mass rather than on molecular effects elicited by the drugs. The tumor's intrinsic biological characteristics together with many other variables are most likely to underlie differences in sensitivity. Thus, a given treatment may affect multiple pathways and elicit complex molecular responses, which are presumably different than cancer response to therapy. Tubulin binding agents are an important class of clinically used antineoplastic compounds. Among them, Taxol® (paclitaxel) is a microtubule-stabilizing agent that blocks cell division by interfering with the function of the mitotic spindle through inhibition of microtubule dynamics (1–3). Recently, a variety of cellular and molecular effects of paclitaxel have been described. These include induction of cytokines, tumor suppressor genes, and activation of signal transduction pathways (4). Paclitaxel is particularly important in the therapy of ovarian carcinomas, but the efficacy is hampered by the relapsing cancers resistant to chemotherapy (5). Proposed mechanisms relevant to paclitaxel resistance include overexpression of MDR1 (P-glycoprotein drug efflux pump; Ref. 6), point mutations (7), and differential expression of β-tubulin isotypes (8). In addition, mitotic checkpoint control (9) and p53 status (10, 11) might also contribute to the sensitivity to paclitaxel (4). Despite such findings, there is little evidence that these mechanisms influence the responsiveness of paclitaxel in clinical settings, where the development of resistance may involve alterations of multiple genes, prior to or occurring under the treatment.

In general, analyses have focused on the contribution of individual genes to drug response. Only recently, the molecular profiling of tumor cells exhibiting different responsiveness to anticancer drugs has been made possible by microarray technologies (12–14). Molecular analyses of the response to treatment has been mainly explored in vitro and with the drug tested at high concentration. In vitro studies do not take into account the host environment in contributing to a given response, which is of clearly greater relevance to clinical settings. Host metabolism certainly influences pharmacokinetics, pharmacodynamics, molecular response, and, ultimately, drug sensitivity. The use of xenograft models allows the study of the in vivo behavior of human tumor and hence evaluate their response to chemotherapy under the influence of host factors (15, 16).

In the present study, we sought to examine the molecular events elicited by paclitaxel in vivo in models of human ovarian carcinoma. By investigating the effects on gene expression after paclitaxel was given to mice bearing xenografted tumors, we have found that the type and magnitude of expression changes revealed by microarray correlated with the response to treatment. Results show the modulation of genes as a function of time of exposure to the drug, and while several genes are affected in the responsive xenografts, considerably less changes are detectable in the nonresponsive xenografts. Gene expression alterations occurring shortly after paclitaxel administration (24 h) allowed us to distinguish the responding xenografts from the nonresponsive xenografts without knowing their mechanism of paclitaxel resistance.

We believe that this information might propose a useful approach to response prediction in drug development studies and assist in the design of novel strategies for the treatment of cancer.

Materials and Methods

Animals

Female NCr-nu/nu mice were obtained from the Animal Production Colony, National Cancer Institute-Frederick Cancer Research and Development Center (Frederick, MD). The mice were 8–10 weeks old and had a mean body weight of 23 g (SD = 2). Throughout this study, nude mice were housed in filtered-air laminar flow cabinets and manipulated following aseptic procedures. Procedures involving animals and their care were conducted in conformity with the institutional guidelines that are in compliance with national (Decreto Legge No. 116, Gazzetta Ufficiale, Suppl. 40, Feb. 18, 1992; Circolare No. 8, Gazzetta Ufficiale, July 1994) and international laws and policies (European Economic Community Council Directive 86/609, Official Journal Legislation 358.1, Dec. 12, 1987; Guide for the Care and Use of Laboratory Animals, U.S. National Research Council, 1996).

Ovarian Carcinoma Models

The 1A9 and 1A9PTX22 cell lines were kindly provided by Dr. Tito Fojo (NIH, Bethesda, MD). 1A9 was originally derived from A2780 human ovarian carcinoma cells (17), and 1A9PTX22 was subsequently selected as a paclitaxel-resistant variant by exposure of 1A9 to the drug (7). IGROV1 cells originally established from a previously untreated surgical specimen (18) were obtained from the National Cancer Institute Tumor Repository (Frederick, MD). The cell lines were grown in vitro in RPMI 1640 supplemented with 10% fetal bovine serum and 5 mm l-glutamine. HOC18, HOC22, HOC94/2, and MNB-PTX1 xenografts were established from ovarian carcinoma patients and maintained in nude mice as described previously (19). HOC18 and HOC22 were from patients who had not received paclitaxel-based chemotherapy, whereas HOC94/2 and MNB-PTX1 were established from paclitaxel-refractory cancers (20, 21).

Drug Treatment of Tumor Xenografts

Tumor xenografts were obtained by injecting 10 × 10⁶ cells in 200 μl suspension (1A9, 1A9PTX22, and IGROV1) or by transplanting tumor fragments (HOC18, HOC22, HOC94/2, and MNB-PTX1) into the flanks of nude mice. The growing tumor masses were measured with a Vernier caliper, and the estimates of tumor weights were calculated by the formula: tumor weight = (length × width²) / 2. When tumor weight reached 150–300 mg, nude mice were randomized to receive paclitaxel or vehicle or remained untreated (Fig. 1). Paclitaxel (kindly provided by Indena, Milan, Italy) was prepared in a mixture containing 50% Cremophor EL (Sigma-Aldrich, Steinheim, Germany) and 50% ethanol and further diluted with 5% glucose in water immediately before administration. Paclitaxel was given by i.v. injection at the single dose of 60 mg/kg. This treatment did not affect the health status of the mice as evaluated by lack of body weight loss. Mice were used either for tumor tissue collection and gene expression analysis or to assess tumor responsiveness to the treatment (Fig. 1).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Experimental design flow chart. Nude mice were implanted s.c. with the tumor of interest. Tumor-bearing mice were randomized at tumor volume 150–300 mg to receive paclitaxel (60 mg/kg i.v.), vehicle, or no treatment. Groups of at least six mice were used for evaluating therapeutic response (see Fig. 2A). Additional groups of four mice were sacrificed before (0 h) and 4 and 24 h after treatment to collect tumor tissues for gene expression analyses (see Fig. 2B). Examples of cDNA microarray hybridization for tumor xenografts 1A9 # 877 and 1A9PTX22 # 995.

Collection of Tumor Tissue for Gene Expression Analysis

To study the molecular response to paclitaxel, we collected samples before and 4 and 24 h after the administration of paclitaxel or the corresponding vehicle (Fig. 1). For each of the xenograft models, the tumors were harvested from 4 mice/group and bulk tissues were snap frozen in liquid nitrogen immediately after collection and later used for cDNA microarray analysis (only 1A9 and 1A9PTX22) and for Northern analysis (all the xenograft models).

Expression Analysis by cDNA Microarrays

Total RNA and Antisense RNA Frozen tissue was directly added to TRIzol reagent (Life Technologies, Inc., Gaithersburg, MD) and homogenized with a tissue grinder. Total RNA was then isolated following the manufacturer's instructions and 3–4 μg were amplified to obtain antisense RNA (aRNA), essentially following the Eberwine protocol (22) with some modifications (23). First-strand cDNA synthesis was obtained in a reverse transcription reaction primed by oligo-(deoxythymidylate)₂₄-T7 (5′-GGC CAG TGA ATT GTA ATA CGA CTC ACT ATA GGG AGG CGG-(T)₂₄-3′), incubating the reaction at 42°C for 60 min in the presence of SuperScript II. Double-stranded cDNA was then synthesized at 16°C for 2 h in the presence of Escherichia coli DNA polymerase I, E. coli DNA ligase, and RNase H followed by 5 min with T4 DNA polymerase (all the enzymes were from Life Technologies). Double-stranded cDNA was extracted with phenol/chloroform/isoamyl alcohol (25:24:1) and precipitated with ethanol in the presence of 0.1 g of linear acrylamide. aRNA was obtained by in vitro transcription with the T7 MEGAscript kit (Ambion, Austin, TX) following the manufacturer's instructions. The reactions were allowed to take place for 5 h at 37°C. To recover the amplified aRNA, an extraction with phenol/chloroform/isoamyl alcohol (25:24:1) followed by a precipitation in ethanol was done. aRNAs were aliquot appropriately and stored at −80°C for future use.

Both total RNA and aRNA concentrations were determined by evaluating the absorbance at 260 nm and by gel electrophoresis that also served to control for the quality of the samples.

Two-Color Fluorescent Hybridization We fluorescent labeled 3–4 μg of aRNA in a 40 μl reverse transcription reaction employing 400 units of SuperScript II in the presence of 6 μg of random primer oligonucleotides [pd(N)6] and 4 pmol of cyanine 3-dUTP (Cy3) or cyanine 5-dUTP (Cy5; Enzo Diagnostics, Farmingdale, NY). Reaction products were purified by repeated washes with 10 mm Tris (pH 7.6)-1 mm EDTA in Microcon YM-30 columns (Amicon/Millipore, Bedford, MA). The last wash was done by combining the appropriate experimental target with the reference and concentrating the samples down to 14 μl in volume. Human CotI DNA (1 μg) and yeast tRNA (4 μg) were added to the purified and labeled samples before heat denaturation. Hybridization to the cDNAs microarrayed onto glass slide was carried out at 65°C for 14–16 h. Washes lasting 3–4 min each were done sequentially with 2× SSC and 0.1% SDS, 1× SSC, 0.5× SSC, and 0.05× SSC.

Cy5/Cy3 Fluorescence Ratio Evaluation Fluorescence images and measurements of Cy3 and Cy5 were collected by scanning the hybridized slides at 10 μm resolution on a GenePix 4000 microarray scanner (Axon Instruments, Union City, CA) at variable photomultiplier tube voltage to obtain maximal signal intensities with <1% probe saturation (Fig. 1). Resulting TIFF images were evaluated using the GenePix 3.0 software (Axon Instruments; Refs. 24, 25) to quantify the extent of hybridization to any given cDNA spotted onto the slide (Cy5/Cy3 fluorescence ratio). Spots showing obvious defects were excluded from the analysis. The generated raw data files were entered in a Web-based relational database maintained by the Center for Information Technology, National Cancer Institute (Bethesda, MD). The expression ratios for the spots on each array were normalized by subtracting the median ratio for the same array. Data were filtered to exclude spots with a size of <25 μm and spots with an intensity of less than twice the background or <250 units in both red and green channels.

For each of the 32 treated tumors, at least two (up to four) independent hybridizations were done, in which the target and reference were labeled by exchanging the fluors. The reference/control aRNA was from pooled equal amounts of total RNAs of tissue xenografts harvested from four untreated mice at the time of randomization.

Data Analysis Comparison between groups was performed as follows. For each time point (4 and 24 h) and treatment (paclitaxel and vehicle), xenografted tissues (1A9 and 1A9PTX22) from four mice were evaluated against the untreated tumors by means of the Hs-ATC-6.5k-4p human cDNA microarray manufactured at the National Cancer Institute Microarray Facility (http://nciarray.nci.nih.gov). On the Hs-ATC-6.5k-4p glass arrays, there are 6947 spotted cDNAs consisting of 5275 distinct clones, with 5097 representing unique UniGene clusters (as for Build 160 Homo sapiens) including 432 expressed sequence tags and 4665 named genes. The additional 178 clones (mostly unknown) are not present in UniGene.

The panel of genes that contribute to the expression profiles shown in Figs. 2B and 3 was selected based on “consistency of the observation.” For each treated mouse/tumor xenograft, only the cDNAs, the Cy5/Cy3 fluorescence ratios of which reverted when target and reference were labeled by exchanging the fluors, were considered significant (to account for dye effects), and among them, only those with a ratio of 1.5 or greater were classified as outlier cDNAs. To account for biological fluctuation among tumor xenografts from different mice and to overcome vehicle-mediated effects, a cDNA/gene was then selected out as relevant (Fig. 2B), only when the following two criteria were simultaneously met: (a) it was an outlier in at least three of the four paclitaxel-treated mice and (b) it was not an outlier in at least three of the four vehicle-treated mice. The selected cDNAs/genes were then manually assigned to functional groups (Fig. 3).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Effect of paclitaxel on 1A9 and 1APTX22 xenografts. 1A9 or 1A9PTX22 were injected s.c. into nude mice. Tumor-bearing mice were treated with paclitaxel (60 mg/kg i.v.). Details of the experimental design are described in Fig. 1. A, antitumor effect against tumor xenografts. Points, median RTW for six mice; arrows, day of paclitaxel administration. B, cDNA microarray analysis of xenografts taken 4 and 24 h after paclitaxel administration (4 mice/group). Colored images, graphical presentation of modulation of gene expression. Color saturation is directly proportional to the magnitude of the changes in the level of expression evaluated against untreated tumors: yellow, increased expression; light blue, decreased expression. Tumor xenografts are identified by the numbers on top of each column, which are averaged Cy5/Cy3 values originating from at least two independent hybridizations for each treated tumor.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Changes in gene expression caused by paclitaxel treatment. A, restricted to sensitive 1A9 xenografts; B, common to both 1A9 and 1A9PTX22 xenografts; C, restricted to resistant 1A9PTX22 xenografts. Gray bars, increased expression level; black bars, decreased expression level.

As internal validation, self-hybridization and reverse fluorochrome experiments were performed. Briefly, Hs-ATC-6.5k-4p human cDNA microarray from independent glass printing sets were hybridized to the same target by arbitrarily mixing Cy5 and Cy3 labeled aRNA (the aRNA used as the reference throughout the study). Results were evaluated and analyzed to define the likelihood to find a Cy5/Cy3 ratio for each individual spot altered in three of four slides. We found that none of the arrayed cDNAs were altered when the cutoff for Cy5/Cy3 ratio was set at 1.5 or greater. Thus, such a value was chosen as the limit to exclude variation due to experimental procedures.

Expression Analysis by Northern Blot

Ten to 12 μg of total RNA from xenografted tissues were electrophoresed through an agarose-formaldehyde denaturing gel and transferred onto nylon membranes. The filters were hybridized overnight at 42°C with ³²P-labeled cDNA dissolved in 50% formamide, 5% dextrane sulfate, 5× saline-sodium phosphate-EDTA [0.75 m NaCl, 0.05 m NaH₂PO₄, 5 mm EDTA (pH 7.4)], 1× Denhardt's solution (BSA, Ficoll, and polyvinylpyrrolidone, 0.2 mg/ml each), 1% SDS, and 100 μg/ml denatured salmon sperm. Filters were then washed twice at 65°C with 0.2× SSC and 0.1% SDS. Probes for CDKN1A (p21/Waf1/Cip1/Sdi1) and topoisomerase 2 α (TOP2A; 26) were kindly supplied by Dr. M. Broggini (Mario Negri Institute for Pharmacological Research, Milan, Italy). To account for the amount of RNA being analyzed, the filters were hybridized overnight essentially following the protocol of Church and Gilbert (27) with a ³²P-labeled oligonucleotide specific for 18S rRNA and washed thrice at 50°C (28). For each tumor model, at least three tissues for each time point were analyzed (before and 4 and 24 h after paclitaxel delivery). The intensities of the bands on the autoradiographic film were evaluated by densitometry analysis using the Gel-Pro Analyzer software (Media Cybernetics, Silver Spring, MD). The ratio between the density of the mRNA of interest and the corresponding rRNA was calculated, and it was arbitrarily assumed that in untreated xenografts, this ratio corresponded to 100% expression.

Cell Cycle Studies

To study the cell cycle perturbation in response to paclitaxel, we collected 1A9 xenografted tumor tissues before and at different time intervals after the administration of paclitaxel (4 and 24 h and 2, 4, and 6 days). Immediately after collection, the tumors were dissociated with 200 units/ml collagenase (Sigma Chemical Co., St. Louis, MO) and the cell suspensions were fixed in 70% ethanol and kept at 4°C until staining. DNA flow cytometric analysis was performed as described previously (29). Briefly, the fixed cells were centrifuged, washed in PBS, and stained with 2 ml of propidium iodide (PI) solution containing 10 μg/ml PI in PBS + 25 μl RNase (10,000 units) overnight in the dark. The cell cycle phase perturbations induced by paclitaxel were evaluated by using a FACS Calibur instrument (Becton Dickinson, Sunnyvale, CA). The fluorescence pulses of PI were detected using a bandpass filter at 620 nm. Monoparametric DNA analysis was performed on at least 20,000 cells for each sample and analyzed with Cell Quest software.

Tumor Response Evaluation

Mice were examined thrice a week. The end point of the experiments occurred when tumors reached a weight of ∼2 g. Tumor weights were normalized in the different groups by obtaining the relative tumor weight (RTW) calculated by the formula: RTW = W_t / W₀, where W_t is the tumor weight at any day of measurement and W₀ is the tumor weight at the start of treatment. The median RTW (n = 6) for all groups was used to plot graphs evaluating the efficacy of treatment. The %T/C (where T and C are median RTW values for paclitaxel-treated and vehicle controls, respectively) was calculated for all the measurements and the lowest value was considered as optimal growth inhibition in Table 1 (20). Net log cell kill (NLCK) was calculated as [(T − C) − (duration of treatment) × 0.301 / doubling time] and growth delay (GD) was calculated as [(T − C) / C × 100], where T and C are median times to reach 1 g, respectively, for treated (paclitaxel, n = 6) and controls (vehicle, n = 6). TR₅₀ (tumor regression) indicates the percentage of mice with >50% reduction of tumor mass.

Results

Overall Microarray Analysis of Tumor Xenografts

Forty tumor tissue samples were collected and about 0.5 million measurements of cDNA expression were made. The results were reproducible in repeated hybridization experiments (data not shown), demonstrating the applicability of such technology and procedures to tissues obtained ex vivo from mice bearing xenografted tumors. Throughout the whole study, analyses were conducted against reference aRNA from xenograft tumors of untreated mice (Fig. 1). For each of the tumor xenografts collected after paclitaxel or vehicle administration, only the Cy5/Cy3 fluorescence ratios, the values of which reverted, were accounted for. The 1.5 fluorescence ratio was chosen as the cutoff value on the basis of preliminary setup experiments, indicating this as the limit to exclude variation due to experimental procedures (see “Materials and Methods” for details). To overcome the biological variability among tumor tissues from different mice, we evaluated four treated xenografts for each experimental group (Fig. 1). We have selected only those gene expression alterations simultaneously present in three of four (75%) paclitaxel-treated mice but not affected by the vehicle.

Changes in Gene Expression after Paclitaxel Administration and Treatment Outcome in 1A9 Xenografts

Mice bearing the 1A9 ovarian carcinoma xenograft were treated with 60 mg/kg of paclitaxel given by single i.v. injection. The evaluation of tumor response showed that paclitaxel inhibited tumor growth in all the mice (T/C = 12%, NCLK = 2.5, GD = 110%) with a >50% reduction of the tumor mass (TR₅₀) in half of them (Fig. 2A; Table 1). This is within the expected range of response to paclitaxel on the part of this model (30).

The evaluation of molecular response, by way of cDNA microarray analysis of the 1A9 tumor xenografts collected shortly after such a treatment, indicates that paclitaxel induced several changes. The level of mRNA expression was altered in a variety of genes (78 cDNAs) and in a time-dependent fashion (Figs. 2B and 3). Four hours after paclitaxel administration, 26 cDNAs were affected. Of those, about 30% remained clearly altered, most of them involved in signal transduction and transcriptional regulation (Figs. 2B and 3). At longer time after paclitaxel administration, the number of changes increased, most alterations (61 cDNAs) becoming evident by 24 h (Figs. 2B and 3). At this time, changes mainly affected genes involved in proliferation/differentiation, apoptosis, cell metabolism, protein biosynthesis, and trafficking (Fig. 3).

Modulation of genes involved in transcriptional response affected both activators and repressors and was already detectable 4 h after paclitaxel administration. The zinc finger protein early growth response 3 (EGR3) was up-regulated, but only transiently, as it returned to normal expression level by 24 h. Instead, the expression of the zinc finger protein SLUG decreased and remained low throughout the 24 h. Similarly, the histone deacetylase HDAC3, the sprouty homologue SPRY4, the G protein GNA11, and the LLR family member SSP29/APRIL were all induced at 4 h and remained overexpressed throughout the 24 h. In addition, by 24 h, both c-fos and the c-fos interacting upstream transcription factor USF2, together with the ADP-ribosylation factor ARF1, the transcriptional regulator TBR1 and the activating transcription factor ATF2, were all down-regulated (Fig. 3).

Distinct genes involved in cell cycle regulation and cell proliferation/differentiation were affected at either 4 or 24 h after paclitaxel delivery. By 4 h, there was an increase in the expression of Nek1 [a protein kinase related to fungal cell cycle regulator NIMA (never in mitosis A)] and a decrease in the expression level of the cell division control CDC2-like protein kinases CLK-1 and CLK-2 and of the dominant-negative helix-loop-helix proteins ID2 and ID3. Their modulation was transient, and as time progressed, they all returned to normal expression level. Meanwhile, at 24 h, the protein kinases CDC2 (CDK1/p34cdc2) and TOP2A were down-regulated and the cyclin-dependent kinase (cdk) inhibitor CDKN1A (p21/Waf1/Cip1/Sdi1), the transforming growth factor-β (TGFβ) superfamily member PLAB, and the type IV protein phosphatase PRL1/PTP4A1 were up-regulated (Fig. 3).

Genes involved in apoptosis were affected 24 h after paclitaxel administration but not yet at 4 h. The expression level of the negative controller of cell growth PIG8 increased; conversely, the bcl2 family member BNIP3 was reduced (Fig. 3).

The diminished expression of several IFN-inducible genes such as G1P3, ISG15, IFI27, IFITM1, and IFI16 was observed 24 h after paclitaxel. At that time, numerous genes involved in metabolism of fatty acids and sterol were also down-regulated. These included for example the dimethylallyltranstransferase/farnesyl diphosphate synthase FDPS, the isopentenyl-diphosphate δ isomerase IDI1, the cholesterol esterase lipase-A LIPA, and the sterol C5 desaturase SC5D. An exception was the lipoprotein lipase LPL, the expression of which decreased 4 h after paclitaxel delivery but was back to normal by 24 h (Fig. 3). The expression of several genes involved in protein biosynthesis and trafficking also decreased at 24 h, exceptions being the two ribosomal proteins RPS21 and RPS29 that were overexpressed (Fig. 3).

Changes in Gene Expression after Paclitaxel Administration and Treatment Outcome in 1A9PTX22 Xenografts

Paclitaxel (60 mg/kg i.v.) given to 1A9PTX22-bearing mice did not achieve significant tumor response (T/C = 70%, NCLK = 0.2, GD = 10%, TR₅₀ = 0; Fig. 2A; Table 1). This confirms the nonresponse to paclitaxel expected for this model (31).

Accordingly, the microarray analysis of the 1A9PTX22 tumor xenografts collected shortly after paclitaxel administration showed that such a treatment caused only minimal perturbation in gene modulation. Changes in the expression level occurred in a few genes at both 4 and 24 h (9 and 7 cDNAs, respectively) after paclitaxel administration. Overall, only 40% (6 cDNAs) seemed selectively altered in 1A9PTX22 xenografts (Fig. 3C), while about 60% (10 cDNAs) were affected also in 1A9 tumors (Fig. 3B).

Gene expression alterations induced by paclitaxel specifically in 1A9PTX22 xenografts included the overexpression of the death domain CRADD adaptor and the down-regulation of transcription factor TFAP4 and of TCL6 occurring at 4 h. Overexpression of succinate dehydrogenase SDHD, phosphatidylcoline transfer protein PCTP, and SEC61G secretory protein was seen at 24 h (Fig. 3C).

Modulation affecting 1A9PTX22-resistant and 1A9-sensitive xenografts included the overexpression of the phosphogluconate dehydrogenase PGD and the down-regulation of ID2 and ID3 occurring at 4 h. In 1A9PTX22, the diminished expression of the IFN-inducible protein IFI27 and the glucose transporter SCL2A3/GLUT3 was observed at 4 h while it occurred at 24 h in 1A9 xenografts. The expression of ARF1 and the major histocompatibility complex class I HLA-B decreased at 24 h for both xenograft models. Interestingly, PK428 serine/threonine protein kinase, the expression of which decreased in 1A9PTX22-resistant tumors 24 h after paclitaxel administration, was instead up-regulated in 1A9-responsive tumors at both 4 and 24 h (Fig. 3B).

Gene Expression Changes Validated by Northern Blot

CDKN1A and TOP2A, the expression of which increased and decreased, respectively, after paclitaxel treatment in 1A9-responsive tumors (Fig. 4A), were chosen as representative genes to validate the microarray results. For this purpose, 1A9 tissue xenografts (n = 12) were evaluated by Northern blot analysis. As shown in Fig. 4B, the Northern analysis of 1A9 xenografts confirmed that (a) 24 h after paclitaxel administration, CDKN1A expression increased in three of four tumors whereas TOP2A decreased in all four tumors, (b) these changes were not yet evident at 4 h, and (c) the magnitude of the modulation paralleled the microarray results for the xenografts taken singularly (CDKN1A expression: 877 > 821 > 827 and TOP2A expression 904 < 821 = 827 = 877; Figs. 4, A and B, and 5). No changes were detectable in vehicle-treated 1A9 xenografts (data not shown). These observations are in accordance with a G₂-M block induced in 1A9 tumor by the treatment. This was evident at 24 h after paclitaxel administration as shown by the DNA histograms in Fig. 4C.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

CDKN1A and TOP2A expression and cell cycle perturbation following paclitaxel administration. 1A9 or 1A9PTX22 were injected s.c. into nude mice. Tumor-bearing mice were randomized and treated with paclitaxel (60 mg/kg i.v.). Details of the experimental design are described in Fig. 1 and in Materials and Methods. A, CDKN1A and TOP2A expression by cDNA microarray analysis of 1A9 and 1A9PTX22 xenografts taken 4 and 24 h after paclitaxel administration. For each treated tumor (identified by Mouse #), the level of CDKN1A and TOP2A expression is presented in gray shaded boxes. Color saturation is proportional to the magnitude of the change in expression evaluated by cDNA microarray against untreated tumors (black, no difference). For those xenografts where change in expression is detectable, the Cy5/Cy3 values calculated by averaging the Cy5/Cy3 ratios from 2 up to 4 independent hybridizations are shown. B, CDKN1A and TOP2A expression by Northern blot analysis of 1A9 xenografts. Nylon membranes with 10–12 μg of total RNA (from xenografted tissues collected before or 4 and 24 h after paclitaxel administration) were hybridized with ³²P-labeled CDKN1A, TOP2A, and 18s rRNA probes. The autoradiographic signals are shown (for densitometric analysis, see Fig. 5). C, cell cycle analysis of 1A9 xenografts. DNA flow cytometric analysis was performed for 1A9 xenografted tumor tissues. Representative cell cycle phase distribution of tumors collected before and 4 and 24 h after paclitaxel administration.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5.

Effect of paclitaxel on CDKN1A and TOP2A modulation in a panel of ovarian carcinoma xenografts. Experiments were performed as described in Fig. 1. For each ovarian carcinoma model, at least three xenografted tumor tissues harvested before and 24 h after paclitaxel (60 mg/kg i.v.) were evaluated for CDKN1A, TOP2A, and 18S rRNA expression by Northern blot analysis. The intensities of the bands on the autoradiographic film were evaluated by densitometry analysis. The ratio between the density of the mRNA of interest and the corresponding rRNA for each tissue was calculated. Columns, percentages of expression for each tumor 24 h after paclitaxel administration in respect to the corresponding untreated xenografts (closed bars, 100% expression). X axis labels: numbers under each column, number of the tumor xenograft; open bars, sensitive tumor models; striped bars, resistant tumor models; asterisks, below the limit of detection in both treated and untreated.

In contrast to sensitive 1A9 tumors, microarray data showed that in 1A9PTX22-resistant xenografts, neither CDKN1A nor TOP2A expression was affected by paclitaxel (Fig. 4A). Concordantly, Northern blots of paclitaxel-treated 1A9PTX22 xenografts showed no changes in the expression of TOP2A (Fig. 5). The expression level of CDKN1A was below the detection capacity of the system in both untreated and treated tissues.

Gene Modulation Is Confirmed in a Panel of Ovarian Carcinoma Xenografts Treated with Paclitaxel

To extend our findings, we studied the effect of paclitaxel on CDKN1A and TOP2A modulation in a panel of ovarian carcinoma xenografts exhibiting different sensitivity to the treatment. Paclitaxel (60 mg/kg i.v.) or vehicle was given to tumor-bearing mice. Similar to 1A9, therapeutic efficacy was achieved for HOC18, HOC22, and IGROV1 xenografts (Table 1). Conversely, but in accordance with 1A9PTX22, the treatment with paclitaxel did not significantly affect the growth of HOC94/2 and MNB-PTX1 xenografts (Table 1).

Northern analyses revealed that by 24 h, the expression of TOP2A decreased in HOC18, HOC22, and IGROV1 (Fig. 5), all tumors highly responsive to the treatment with paclitaxel (Table 1). Concomitantly, CDKN1A expression increased in HOC18 and IGROV1 (Fig. 5). CDKN1A was below the level of detection in both untreated and treated HOC22 tissues. No changes of CDKN1A and TOP2A expression were observed in HOC94/2 and MNB-PTX1 tumors not responding to paclitaxel treatment (Fig. 5).

Discussion

This study has analyzed at the transcriptional level human ovarian carcinoma xenografts undergoing treatment with paclitaxel as a way of understanding the in vivo molecular consequences of drug treatment. We have found that paclitaxel can affect the level of expression of a variety of genes and that such an effect occurred mostly 24 h after its delivery in those tumors responding to the treatment. These findings show that the therapeutic effectiveness of paclitaxel correlates with the dynamic modulation of gene expression occurring shortly after its administration and also suggest that evaluating the quantity of the changes and/or the change affecting specific gene types postchemotherapy in vivo might be a novel strategy to define/study pharmacodynamic end points.

To study drug responsiveness of cancer cells, most studies have used cell lines in vitro often exposed to high drug concentrations, not always achievable in the plasma of patients (4). In this study, 60 mg/kg of paclitaxel were given, a therapeutic relevant nontoxic dose for mice bearing tumors (32), and by analyzing the gene expression profile of tumor xenografts from treated mice, this study takes into account the pharmacokinetics and pharmacodynamics of the drug. After the administration of such a dose, paclitaxel rapidly reaches the tumors with a peak at 4 h and is still detectable 24 h later (data not shown). Different gene expression kinetics were observed 4 and 24 h after paclitaxel administration that might either reflect the direct effects of the drug or result from downstream effects mediated by earlier responding genes.

It is widely accepted that the cytotoxicity of tubulin binding agents is due to their effect on the mitotic spindle, resulting in aberrant mitosis, aneuploidy, mitotic block, and apoptosis (1–3, 33). However, the underlying biochemical events leading to cell death after perturbation of tubulin are far from understood and a matter of current investigation. Our study shows that the up-modulation and down-modulation of genes involved in cell cycle regulation, cell proliferation, and apoptosis following paclitaxel administration are consistent with an altered proliferation state in the growth-inhibited tumors. Specifically, at 4 h, we observed transient down-regulation of ID2 and ID3 involved in coordination of proliferation and differentiation in mammalian cells. It has been shown that by binding pRb pocket proteins, ID2 is able to abolish the S-phase entry block (34). Conversely, the dual-specificity kinase Nek1, closely related to a cell cycle regulator that controls initiation of mitosis in Aspergillus nidulans (35), was transiently up-regulated. In 1A9PTX22 xenografts, paclitaxel caused only the transient down-regulation of ID2 and ID3. It seems therefore that 1A9 but not 1A9PTX22 cells have to deal with conflicting signals involved in controlling cell cycle progression and mitotic regulation. As a possible consequence, the attempt by 1A9 cells to traverse and complete the cell cycle is not going to succeed; while in the absence of conflicting signals, cell growth will resume. This is supported by the perturbation of the necessary growth regulatory machinery that later occurred in 1A9 but not in 1A9PTX22 tumors. Specifically, 24 h after paclitaxel administration, different genes involved in cell growth regulation continued to be affected in 1A9 xenografts. The relevance of such findings is supported by the cell cycle perturbation studies where a G₂-M block in 1A9-sensitive xenograft occurs 24 h after paclitaxel administration. The expression of CDC2 and TOP2A decreased, whereas the expression of CDKN1A increased. The TGFβ superfamily member PLAB (36), previously suggested to have a growth inhibitory influence, also increased. CDKN1A inhibits cdk activity in such a way that the phosphorylation of G₁ and G₂ cdk critical substrates may be prevented (37). On the other end, CDC2, the primary mitotic kinase, is required to progress through G₂ and enter mitosis (38), where it also regulates the mitotic spindle and the movement of chromosomes. The repression of TOP2A, which is described to be up-regulated during the G₂ phase of the cell cycle, could certainly be a yet distinct mechanism of maintaining G₂ arrest and prevent mitosis. Moreover, the physical association of TOP2A/CDC2 contributes to the formation of precondensed chromosomes (39). Changes in genes associated with apoptosis are also evident 24 h after treatment. The PIG8 (40) apoptosis inducer was up-regulated. Conversely, the expression of BNIP3 was reduced. BNIP3 interacts with bcl2 and with E1B virus 19-kDa protein and counteract the initiation of caspase cascade leading to apoptosis (41), and its down-regulation in response to apoptotic stimuli has been reported (42). Altogether, these alterations imply a complex interplay among the different effects related to the action of paclitaxel.

Paclitaxel has been recently described to affect cellular function not previously associated with microtubules, such as cytokine release (43, 44) and transcription regulation (45). However, these in vitro effects were seen with high doses of paclitaxel. Our results were obtained in vivo, with a clinically relevant dose of paclitaxel (32). In our study, modulation of genes involved in the regulation of transcription affected both transcriptional repression and activation in 1A9-responsive tumors. Four hours after paclitaxel treatment, HDAC3 was induced and remained overexpressed throughout the 24 h. Meanwhile, at 24 h, ATF2, the motif of which responsible for stimulation of transcription is localized within the histone acetyl transferase domain (46), was instead down-regulated. By impairing access to chromatin and DNA through histone acetylation/deacetylation changes, a mechanism to trigger a general repression of transcription may be occurring in the cellular response to paclitaxel. This might partly explain the general decrease in levels of gene expression 24 h after paclitaxel administration (Fig. 3), which is not seen in 1A9PTX22. On the other hand, the decrease in expression of the transcription corepressor SLUG (47) could suggest a more finely tuned transcriptional activity. Transcription through TGFβ-responsive promoters and the c-fos-responsive element was also affected by paclitaxel as indicated by the diminished expression of MADH2/SMAD2 and both c-fos and its interacting coactivator USF2 24 h after the administration of paclitaxel. Beside its involvement in transcription, HDAC3 contributes to the accumulation of cells in G₂-M phase (48), so its increased expression may also be important in perturbing cell cycle progression. Other implications of the unexpected link of paclitaxel action to transcriptional activation through modulation of transcriptional repressor mechanism might be an interaction between taxane and histone deacetylase inhibitor drugs, of which a number are entering early clinical trials (49, 50).

Finally, an action of paclitaxel with the cellular response to stimuli mediated by transmembrane signaling systems is suggested by its effects on the small GTPase ARF1 and the tyrosine kinase receptor inhibitor SPRY4, both part of the ras-mediated signal transduction pathway (51, 52), and on the guanine nucleotide binding protein GNA11 (53). It is also worth noting the up-regulation of the transcription coactivator EGR3 4 h after paclitaxel administration. As EGR1 activates the PRL1 promoter (54), EGR3 might well play a role in the tyrosine phosphatase PRL1/PTP4A1 increase seen at 24 h.

The link to modulation of ARF1 is also of interest in view of the important role of ARF1 in maintaining the normal secretory function of the cell through its regulation of vesicle secretion and migration from the endoplasmic reticulum to the Golgi apparatus. Vesicles are associated with the microtubule network (55) and our data suggest that perturbation of the microtubule network by paclitaxel may induce a feedback/adaptive response by modulating in part the formation of the vesicle coat processing mechanism. It is worth noting that CDC2 kinase can directly phosphorylate dynein (minus-end-directed microtubule motor; 56) and its involvement in membrane organelle movement during interphase has been suggested. Thus, CDC2 kinase down-regulation by paclitaxel might also be implicated in such phenomenon. Other signaling effects are suggested by our data. The expression of several IFN-inducible genes was reduced 24 h after paclitaxel treatment of 1A9-responsive xenografts. Overexpression of INF-responsive genes has been recently associated with resistance to paclitaxel in tumor cell lines (57). These findings support the possible involvement of mediators of INF signaling in responsiveness to paclitaxel.

Several genes involved in metabolism appear also to be affected by paclitaxel. Interestingly, the expression of genes involved in sterol/cholesterol and fatty acid/glycerolipid biosynthesis was diminished after treatment. While the effect of a leaf and steam extract of Taxus baccata on circulating cholesterol level has been reported (58), the relevance of this metabolic pathway to the mechanism of action of paclitaxel deserves further investigation.

The molecular effects elicited by paclitaxel documented herein reflect genetic aspects of the response. Certainly, epigenetic mechanisms such as phosphorylation of proteins linked to cell survival or activation of cell signaling pathways (4) are of importance to the drug's effect but would not manifest as changes in gene expression. Although the genes we analyzed are only a fraction of those present in the whole genome, the investigation provides useful information on the transcriptional response evoked by paclitaxel. As with any microarray-based strategy, results are hypothesis generating and point to areas where additional experiments are required to assess the importance of the changes.

Surprisingly, only a relatively small number of genes were expressed differently in treated and untreated xenografts. This might be due to the strict criteria imposed by our selection, which may have resulted in the exclusion of some genes. On the other hand, it should have avoided experimental artifacts and biological fluctuations, thus ending with only a few genes relevant to the pharmacological consequences of paclitaxel and not dependent on the experimental model. The latter likelihood derives support from results on a panel of human ovarian carcinoma xenografts where, although only two genes were examined, a strong correlation between the sensitivity to paclitaxel and the induction of CDKN1A and the decreased expression of TOP2A emerged (Fig. 5). These findings not only serve as one form of validation for the microarray analysis results for 1A9 and 1APTX22 but also and most importantly extend the significance of the observations to ovarian cancer, suggesting that gene expression changes following drug delivery might have the potential to distinguish responsive tumors from not responsive ones. To this end, our findings argue against a role of p53 status in predicting the in vivo response to paclitaxel. p53 has been described as having different roles in the sensitivity to paclitaxel depending on the experimental model used (10, 11). All the responsive ovarian carcinoma xenografts used here express wild-type p53 (26, 59, 60) and paclitaxel treatment elicited a clear up-regulation of p53-responsive genes (CDKN1A and PIG8). Conversely, such an up-regulation was not seen in HOC94/2 and MNB-PTX1 xenografts, both using wild-type p53 but resistant to paclitaxel, or in the resistant 1A9PTX22 expressing mutated p53 (Refs. 26, 60 and data not shown).

It is noteworthy that our observations suggest that the differences in responsiveness as well as the effects on gene expression cannot be easily attributed to the MDR phenotype, to impaired paclitaxel binding to β-tubulin, or to altered distribution of the β-tubulin isotypes expression. The whole panel of ovarian carcinoma xenograft models used here is negative for the MDR phenotype mediated by the 170-kDa P-glycoprotein drug efflux pump overexpression (7, 21, 26, 59). A point mutation in β-tubulin isotype M40 has been described for 1A9PTX22 cells (7) and might be relevant to determine its resistance to paclitaxel. On the other hand, HOC94/2- and MNB-PTX1-resistant xenografts express wild-type β-tubulin (21). In addition, β-tubulin isotypes expression distribution is similar in all the models used herein (7, 21).

To evaluate the patterns of gene expression in ovarian cancer, we used in vivo models derived from ovarian carcinoma patients sensitive and resistant to paclitaxel. Some disadvantages of using an in vivo animal model must be considered. Drug metabolism and distribution in the mice may be different from that of humans. However, the identical genetic background of the mice, all bearing the same tumor and receiving the same drug dose, ensures the reproducibility of the molecular pharmacology studies. In addition, the possibility of comparing tumor tissues before and after treatment allowed us to study molecular modifications along with the treatment.

In conclusion, this study shows the usefulness of mouse xenograft human tumor models to study the molecular basis for altered gene expression in response to a certain treatment (in this case, paclitaxel). Our results measure the molecular consequences of the therapy rather than the initial molecular differences in sensitive or resistant tumors (61). Molecular changes are potentially more selective and are detectable much sooner compared with the empirical long-term evaluation of tumor growth inhibition. Therefore, in addition to informing about potential approaches to modulating sensitivity to paclitaxel, they might serve as efficient pharmacodynamic molecular end points for evaluating drug activity shortly after drug administration. The relationships between antitumoral paclitaxel activity and gene expression levels are only correlative at this stage, but they generate hypotheses that are worth investigating.