Potentiation of paclitaxel activity by the HSP90 inhibitor 17-allylamino-17-demethoxygeldanamycin in human ovarian carcinoma cell lines with high levels of activated AKT

Activation of the phosphatidylinositol-3-kinase (PI3K)/AKT survival pathway is a mechanism of cytotoxic drug resistance in ovarian cancer, and inhibitors of this pathway can sensitize to cytotoxic drugs. The HSP90 inhibitor 17-allylamino-17-demethoxygeldanamycin (17-AAG) depletes some proteins involved in PI3K/AKT signaling, e.g., ERBB2, epidermal growth factor receptor (EGFR), and phosphorylated AKT (p-AKT). 17-AAG and paclitaxel were combined (at a fixed 1:1 ratio of their IC₅₀) in four ovarian cancer cell lines that differ in expression of p-AKT, EGFR, and ERBB2. The EGFR-overexpressing A431 and KB epidermoid cell lines were also included. Combination indices (CI) were calculated using the median-effect equation and interpreted in the context of 17-AAG-mediated inhibition of PI3K signaling. Synergy was observed in IGROV-1- and ERBB2-overexpressing SKOV-3 ovarian cancer cells that express a high level of constitutively activated p-AKT [CI at fraction unaffected (fu)_0.5 = 0.50 and 0.53, respectively]. Slight synergy was observed in A431 cells (moderate p-AKT/overexpressed EGFR; CI at fu_0.5 = 0.76) and antagonism in CH1 (moderate p-AKT), HX62 cells (low p-AKT), and KB cells (low p-AKT/overexpressed EGFR; CI at fu₅₀ = 3.0, 3.5, and 2.0, respectively). The observed effects correlated with changes in the rate of apoptosis induction. 17-AAG induced a decrease in HSP90 client proteins (e.g., C-RAF, ERBB2, and p-AKT) or in downstream markers of their activity (e.g., phosphorylated extracellular signal-regulated kinase or p-AKT) in SKOV-3, IGROV-1, and CH1 cells at IC₅₀ concentrations. A non–growth-inhibitory concentration (6 nmol/L) reduced the phosphorylation of AKT (but not extracellular signal-regulated kinase) and sensitized SKOV-3 cells to paclitaxel. In conclusion, 17-AAG may sensitize a subset of ovarian cancer to paclitaxel, particularly those tumors in which resistance is driven by ERBB2 and/or p-AKT. [Mol Cancer Ther 2006;5(5):1197–208]

Novel agents that specifically inhibit molecular targets driving oncogenesis are proving to be exciting drugs for the treatment of cancer (1). However, in the first-line, single-agent setting, imatinib is the only small molecule that has thus far received regulatory approval for chronic myeloid leukemia and gastrointestinal stromal tumors (2). These tumors are largely driven by single molecular abnormalities, the BCR-ABL translocation, and constitutively activated mutated c-KIT, respectively. The more common solid tumors possess multiple abnormalities (3) and molecularly targeted agents may, therefore, (a) need to be used in selected subpopulations of patients or (b) need to inhibit multiple molecular targets. The monoclonal antibody trastuzumab is the only agent in which patient selection, based on amplification of ERBB2, has guided its usage (4). Preliminary data showing increased copy number of the epidermal growth factor receptor (EGFR) gene in responding colorectal cancer patients (5) may lead to selection of patients who will benefit from cetuximab, and the association of activating EGFR mutations in the kinase domain with efficacy of gefitinib or erlotinib in non–small cell lung cancer is also suggesting a way forward (6).

Clinical trials with monoclonal antibodies (targeting ERBB2, EGFR, and vascular endothelial growth factor) have indicated that the major effect on survival comes from combination schedules with chemotherapy (7). In contrast, large-scale randomized trials with small-molecule signaling inhibitors have mostly been negative and this has led to a more cautious approach to the development of combination therapy (7). This underscores the need to improve the predictive value of preclinical models by, for example, determining the molecular context for mechanisms of synergy and antagonism, and also emphasizes the importance of examining the effect of factors such as scheduling and dose. In this way, it may be possible to select patients likely to benefit from combination chemotherapy and to suggest schedules most likely to succeed.

17-Allylamino-17-demethoxygeldanamycin (17-AAG) inhibits HSP90, a molecular chaperone that is important for the posttranslational folding and stability of several client proteins, a number of which are components of signal transduction pathways that are considered important targets for cancer therapy (8–12). Key client proteins include those important to signaling through the RAS/RAF/MEK/extracellular signal-regulated kinase (ERK) and phosphahtidylinositol-3-kinase (PI3K)/AKT pathways e.g., ERBB2, EGFR, C-RAF, and AKT (13). 17-AAG, by virtue of its effects on multiple proteins, may therefore overcome resistance to agents that have more specific pathway effects (8). Indeed, clinical studies with 17-AAG have shown its potential as a single agent (9, 14, 15). However, there may be additional potential for its use in combination with conventional agents because some of the pathways inhibited by 17-AAG are also implicated in cytotoxic drug resistance, e.g., the PI3K/AKT survival pathway (16). The outcome of such combinations may be increased therapeutic benefit as a result of either their independent (additive) drug action or as a result of their synergistic interaction.

Combination studies of 17-AAG and cytotoxic agents in breast, BCR-ABL-positive leukemia, and non–small cell lung cancer have shown the potential for synergy in some tumor cell lines (17–19). For example, Munster et al. (17) studied the combination of 17-AAG with paclitaxel in four breast cancer cell lines with different levels of ERBB2 expression. Synergistic growth inhibition was observed in the two high ERBB2-expressing cell lines, BT-474 and SKBr-3, and this effect was associated with an earlier onset of apoptosis. In the other breast cell lines, sensitization at the level of apoptosis was observed irrespective of ERBB2 expression levels. Studies have shown that 17-AAG-mediated inhibition of PI3K/AKT signaling is an important component of its cytotoxic sensitizing property (20, 21). This pathway is oncogenic in many tumor types, driving processes such as cell growth and proliferation, and of particular relevance to cytotoxic drug therapy, generating strong antiapoptotic survival signals (22, 23). 17-AAG can inhibit AKT activity via at least two mechanisms: (a) depletion of ERBB2 or other ERBB family members and (b) preventing HSP90-dependent conformational stability of phosphorylated AKT (p-AKT; refs. 10, 13). Studies in mice showed that doses of 17-AAG that had no single-agent activity were sufficient to lead to a reduction of ERBB2 protein and p-AKT protein levels and to sensitization of BT-474 xenografts to paclitaxel (21).

The PI3K/Akt pathway is frequently highly activated in ovarian cancer, most commonly through mutational activation of PI3KCA or overexpression of AKT2 (24, 25). Overall, at least 40% of ovarian tumors possess an activated AKT2 (26). Altomare et al. (24) reported that 68% of ovarian carcinomas displayed elevated p-AKT using immunostaining of tissue microarrays. There was a close association between positive staining for p-AKT and positive staining for the phosphorylated forms of downstream substrates, mTOR and GSK3β. Inhibition of PI3K with LY294002 increased cisplatin- or paclitaxel-induced apoptosis in some human ovarian tumor cell lines (24, 27, 28). Taxanes and platinum-based drugs are part of standard chemotherapy for patients with ovarian cancer. In view of (a) the importance of the PI3K/AKT pathway in ovarian cancer, (b) the ability of 17-AAG to inhibit this pathway as well as other important oncogenic pathways, and (c) the ability of 17-AAG to potentiate the effect of paclitaxel in breast cancer models, we were interested in investigating whether 17-AAG could sensitize human ovarian tumor cell lines to these drugs either through effects on PI3K/AKT signaling and/or via other HSP90-dependent signaling pathways.

Four human ovarian tumor cell lines, SKOV-3, IGROV-1, CH1, and HX62, were chosen from a molecularly characterized panel of eight ovarian tumor cell lines because they displayed a range of molecular abnormalities leading to different levels of p-AKT (Ser⁴⁷³). We confirmed that SKOV-3 (overexpresses ERBB2) and IGROV-1 cells have a constitutively activated AKT (p-AKT) and also showed that CH1 and HX62 cells express lower p-AKT levels that are inducible and noninducible by serum, respectively. Continuous exposure of these cell lines to 17-AAG and paclitaxel led to synergistic growth inhibition in SKOV-3 and IGROV-1 cells but antagonistic activity in CH1 and HX62 cells. It was also possible to use very low or non–growth-inhibitory concentrations of 17-AAG to sensitize SKOV-3 and IGROV-1 cells to paclitaxel. A431 and KB epidermoid cell lines were included in some experiments as control lines with EGFR overexpression. Molecular studies are described that implicate the effect of 17-AAG on client proteins important for sustained PI3K/AKT signaling and evasion of apoptosis when challenged with paclitaxel in the observed synergy/sensitization outcomes.

CH1, HX62, PXN94 (Institute of Cancer Research, Surrey, United Kingdom; ref. 29), A2780 (30), and IGROV-1 (American Type Culture Collection, Manassas, VA) ovarian tumor cell lines and A431 and KB epidermoid tumor cell lines were cultured in DMEM without glutamine supplemented with 1% (v/v) nonessential amino acids, 2 mmol/L glutamine (Life Technologies, Ltd., Paisley, United Kingdom), and 10% heat-inactivated fetal bovine serum (PAA Laboratories, Somerset, United Kingdom; Invitrogen, Paisley, United Kingdom). SKOV-3 and OVCAR-4 ovarian tumor lines (American Type Tissue Collection) were cultured in RPMI medium without glutamine (Sigma-Aldrich, Gillingham, Dorset, United Kingdom) supplemented as described above. Cells were maintained in an incubator with a humidified atmosphere containing 5% CO₂ at 37°C.

17-AAG (kindly supplied by Dr. E. Sausville, National Cancer Institute, Washington, DC) and paclitaxel (Sigma-Aldrich, Poole, Dorset, United Kingdom) were dissolved in DMSO and ethanol, respectively, to give a stock solution of 10 mmol/L. A 20 mmol/L stock solution of carboplatin (Sigma-Aldrich) was made by dissolving in PBS. The stock solutions were stored for up to 3 months (17-AAG and carboplatin stored at −20°C and paclitaxel at 4°C). On the day of experiments, drug dilutions were made fresh by first diluting with unsupplemented medium to 0.2 mmol/L, thereby diluting the solvent concentration to 1%. Further drug dilutions were made in unsupplemented medium containing 1% solvent.

The following antibodies were used for immunoblotting: mouse monoclonal secondary antibodies (Amersham Biosciences, Buckinghamshire, United Kingdom) against C-RAF, HSP70 (Santa Cruz Biotechnology, Santa Cruz, CA), and glyceraldehyde-3-phosphate dehydrogenase (GAPDH; Chemicon Europe, Ltd., Hampshire, United Kingdom); rabbit polyclonal secondary antibodies (Amersham Biosciences) against ERBB2 (Upstate, NY), p-ERK, ERK, cyclin D1, p-AKT (Ser⁴⁷³), AKT, p-mTOR, mTOR, p-GSK3β (Ser⁹), GSK3β, and cleaved poly(ADP)ribose polymerase (PARP, Asp²¹⁴; New England Biolabs, Ltd., Hertfordshire, United Kingdom).

Inhibition of Cell Growth. Exponentially growing CH1, SKOV-3, IGROV-1, HX62, A431, and KB tumor cells were harvested and plated (2,000 per well in a volume of 0.16 mL) in 96-well plates (Corning, c/o Sigma-Aldrich, Gillingham, Dorset, United Kingdom) and incubated at 37°C, 5% CO₂ for ∼30 hours. A range of drug concentrations (prepared as described above) was then added (20 μL) to the wells either alone (plus 20 μL of solvent used for second drug used in the combination) or in combination (quadruplicate wells), giving a final volume of 200 μL/well. Control wells contained 20 μL of each relevant solvent (final concentration of each solvent was 0.1%). Incubation was for 72 hours (HX62, IGROV-1, A431, and KB) or 96 hours (SKOV-3 and CH1) sufficient to give approximately three to four population doublings. Cell growth was determined using an 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay as described previously (31).

Analysis. Most experiments were done combining both agents added together at a fixed ratio of the IC₅₀ of each individual drug. The effects of the combination were calculated for each experimental condition using an in-house spreadsheet based on the median-effect analysis method of Chou and Talalay (32). For each level of fraction unaffected (fu), a combination index (CI) was calculated according to the following equation: CI = (D)₁ / (D_f)₁ + (D)₂ / (D_f)₂ + [(D)₁ (D_f)₂ / (D_f)₁ (D_f)₂], where (D)₁ and (D)₂ are the concentrations of the combination required to produce a fu, and (D_f)₁ and (D_f)₂ are the concentrations of the individual drugs required to produce fu. Linear regression coefficients (r²) of median-effect plots were required to be >0.95 for the data to qualify for analysis. A histogram is obtained in which CI values are given at various fu values. Only values obtained on the near linear part of the fu/dose plot are valid (generally between fu_0.2 and fu_0.8 depending on shape of curve). These data can also be used to generate a best-fit CI at fu_0.25, fu_0.50, and fu_0.75 (combined drug doses in a ratio of their IC₅₀ that inhibit growth by 75%, 50%, and 25%, respectively). Occasionally, we rejected the CI at fu_0.25 or fu_0.75 when the actual data points at these fu values diverged from the fitted line. CI <1, CI ∼1, and CI >1 indicates synergy, additivity, and antagonism, respectively. A one-sample t test (two-tailed) was used to compare the CI at fu_0.50 with 1 (GraphPad Prism 4.00 software; GraphPad Software, Inc., San Diego, CA). The data are also presented as the average CI of mean fu_0.25, fu_0.5, and fu_0.75.

Further studies were done in some cell lines in which the dose-response curve of paclitaxel was determined in the presence of a fixed, low concentration of 17-AAG (at a concentration that gave a fu ≥0.90). The fu values and CI were then calculated and analyzed using the equation of Chou and Talalay as described above.

Exponentially growing cells (5 × 10⁴) were added to T25 flasks (Nunc A/S, Kamstrupvej, Denmark) in 4 mL medium. Approximately 30 hours later, drugs and/or solvent were added (total of 1 mL). Attached and detached cells were collected at 24, 48, and 72 hours and counted using a hemocytometer (Fisher Scientific, Loughborough, United Kingdom). Nuclei of detached cells were visually confirmed as being largely apoptotic after Geisma staining.

Adherent cells were grown to 30% to 50% confluence and then exposed to different agents, at various times and doses, as described in Results. After exposure, cells were harvested, washed with ice-cold PBS, resuspended in 150 μL cold lysis buffer [50 mmol/L Tris/150 mmol/L sodium chloride (pH 7.5), 1% IGEPAL CA-630, 0.1% SDS, 100 μg/mL phenylmethylsufonylfluoride, 10 μg/mL aprotinin, 20 μg/mL leupeptin, and 100 μmol/L sodium orthovanadate], left on ice for 15 minutes, and then centrifuged at 15,000 × g for 15 minutes at 4°C; the supernatant was stored at −70°C until required. Protein concentration for each lysate was determined using the BCA protein assay method (Sigma-Aldrich). Following solubilization of 80 to 100 μg protein in Laemmli buffer and boiling for 1 minute, the proteins were separated on precast 4% to 20% gradient Tris-glycine gels (Novex, Invitrogen, Paisley, United Kingdom) and transferred onto 0.45 μm nitrocellulose membranes (Invitrogen). These membranes were then probed with individual antibodies, using horseradish peroxidase–conjugated secondary antibodies, and visualized using the enhanced chemiluminescence plus detection system (Amersham Biosciences).

Exponentially growing cells (1 × 10⁶) were washed in ice-cold PBS and resuspended in 100 μmol/L PBS containing 0.1% (w/v) bovine serum albumin. Ten microliters of either rat anti-ERBB2 (clone ICR62-7.5 μg/mL) or rat anti-ERBB2 (clone ICR55-7.5 μg/mL), both kind gifts from Dr. Sue Eccles (The Institute of Cancer Research), were added, and cells were incubated for 2 hours on ice. Following two washing steps in ice-cold PBS, cells were incubated with 100 μL FITC-conjugated anti-rat antibody (Sigma; 1:1,000) on ice and in the dark for 1 hour. Finally, cells were washed twice in ice-cold PBS before fixing in 0.1% (w/v) paraformaldehyde in PBS (Sigma) in preparation for flow cytometric analysis using the FACSCalibur (Becton and Dickinson, Oxford, United Kingdom). The most homogenous cell population was distinguished and gated by forward scatter and side scatter variables, and at least 5,000 events were recorded. The geometric-mean-FITC-fluorescence in each sample was measured and assessed using WinMDI v2.4.³

This was converted to molecules of equivalent soluble fluorescein using calibrated fluorescent beads (DakoCytomation, Ltd., Copenhagen, Denmark).

A panel of eight human ovarian tumor cell lines (A2780, CH1, HX62, IGROV-1, OVCAR-4, PXN94, SKOV-3, and 41M) was characterized for a number of key signaling proteins. From this panel, a minipanel of four lines (SKOV-3, IGROV-1, CH1, and HX62) was chosen for this present study based on the following properties. SKOV-3 and IGROV-1 cells express the highest level of p-AKT (phosphorylated at Ser⁴⁷³) in the panel of eight lines (Fig. 1A). This activation was not sensitive to serum starvation and readdition and was therefore considered to be constitutive (data not shown) and in line with published data for SKOV-3 (24). The CH1 line was one of three lines (A2780, OVCAR-4, and CH1) to express a lower basal level of p-AKT that was nevertheless inducible by serum. The HX62 cell line was one of three lines (HX62, PXN94, and 41M) to have barely detectable levels of p-AKT (noninducible). These levels of p-AKT generally correlated well with the levels of the downstream effector, phospho-GSK3β (i.e., high basal levels in SKOV-3 and IGROV-1 cells compared with CH1 cells and barely detectable levels in HX62 cells; Fig. 1A). There are a number of points in growth factor downstream signaling that can contribute to constitutive AKT activation. For example, there is an increase in PIK3CA copy number in SKOV-3 cells (33). PTEN negatively regulates AKT through its phosphatase activity on PIP2/3, and whereas IGROV-1 cells are PTEN null (34), low expression has been reported in SKOV-3 (35). Our own data confirmed a virtually undetectable level of PTEN in these cell lines by Western blotting (Fig. 1A). However, levels were detectable in the other three cell lines but were lower than the positive control, OVCAR-4.⁴

SKOV-3 cells also overexpress ERBB2 (36) and we compared its surface expression with the other cell lines in our minipanel (Fig. 1B). IGROV-1 cells were shown to express lower levels of ERBB2 consistent with other reports and CH1 and HX62 cells expressed even lower levels.³ All lines expressed EGFR but at lower levels compared with the positive controls, KB and A431 epidermoid tumor cell lines (Fig. 1B). A431 cells express a moderate basal level of p-AKT (similar to CH1) that is inducible by serum, whereas KB cells express a very low level (noninducible; data not shown).

Inhibition of cell growth was measured using an MTT assay as the end point. CH1 cells were shown to be sensitive to paclitaxel (IC₅₀ = 6.6 nmol/L), whereas HX62, SKOV-3, and IGROV-1 cells were 6- to 10-fold less sensitive (Table 1). SKOV-3 cells were the most sensitive to 17-AAG and CH1 cells were the most resistant (IC₅₀ ranged from 66 to 1,570 nmol/L). The wide range of sensitivity to 17-AAG relates, at least in part, to the level of DT-diaphorase/NQO1 expression (37). DT-diaphorase reduces the benzoquinone moiety in 17-AAG to the more potent HSP90 inhibitor, 17-AAGH₂ (38). In the National Cancer Institute 60 cell line panel, the six ovarian cancer cell lines displayed activity of this bioreductive enzyme lower than the mean (0.65 μmol/min/mg protein) with values ranging from 0.053 to 0.45 μmol/min/mg protein.³ The values for SKOV-3 and IGROV-1 were 0.11 and 0.13 μmol/min/mg protein, respectively. Another cell line comparison study reported that CH1 cells have enzyme activity <0.002 μmol/min/mg protein, consistent with the low sensitivity of these cells to 17-AAG (37).

17-AAG can lead to cell cycle arrest in either G₁ or G₂-M depending on the cell lines (17, 39, 40). We observed G₁ arrest in CH1 and IGROV-1 cells (both p53 wild type; refs. 41, 42), and both a G₁ and G₂ arrest in the SKOV-3 (p53 null) and HX62 cells (p53 mutant; ref. 40; data not shown). A similar association between expression of wild-type p53– and 17-AAG-induced G₁ arrest was observed in some colon tumor cell lines (39, 40). Paclitaxel induced a G₂-M arrest in all of the cell lines (data not shown). All the cell lines are RB positive as determined by detection of p-RB (Ser⁷⁸⁰) protein by immunoblotting (data not shown).

SKOV-3, IGROV-1, CH1, and HX62 ovarian tumor cells were simultaneously and continuously exposed to increasing equitoxic concentrations of paclitaxel and 17-AAG (∼1:1 IC₅₀ ratios) and growth inhibition was measured. CI values at fu_0.50 and the average CI values of fu_{0.25,0.5,0.75} are given in Table 2. Values significantly less than 1 were obtained for SKOV-3 cells and IGROV-1 cells (mean CI at fu_0.5 = 0.53 and 0.50, respectively), demonstrating that the two drugs interact synergistically to inhibit growth (example plot given in Fig. 2A). In contrast, values significantly less than 1 were obtained for CH1 and HX62 cells, indicating antagonism (mean CI at fu_0.5 = 3.0 and 6.3, respectively). In the EGFR-overexpressing epidermoid KB and A431,17-AAG and paclitaxel were antagonistic and weakly synergistic, respectively (Table 2).

The order of addition of the two drugs was evaluated in CH1 cells in which antagonism had been observed when the drugs were added simultaneously. Cells were exposed to either paclitaxel or 17-AAG for 4 hours [IC₅₀ = 65 ± 6.2 and 6,000 nmol/L (n = 2), respectively] followed by drug washout and the addition of the other drug for 92 hours. Sequencing 17-AAG first still gave antagonism, whereas paclitaxel first gave additivity (Table 2).

SKOV-3 and IGROV-1 cells were exposed simultaneously to 17-AAG and carboplatin at a fixed ratio of their IC₅₀. This combination was antagonistic in SKOV-3 cells (CI at fu_0.50 = 2.3 ± 1.4). An example CI plot is given in Fig. 2B. The data for IGROV-1 cells suggested additivity/slight antagonism (CI at fu_0.50 = 1.3 ± 0.21).

SKOV-3 cells were exposed to a low, non–growth-inhibitory concentration of 17-AAG (6 nmol/L; 10% of IC₅₀) and increasing concentrations of paclitaxel. As might be expected, cell cycle phase distribution was not affected at this 17-AAG concentration (data not shown). The IC₅₀ for paclitaxel decreased 3-fold under these conditions (i.e., from 70 to 23 nmol/L; mean of two independent experiments; Fig. 3A). The CI values at fu_0.5 were 0.47 in each of the two experiments. These data clearly show that a concentration of 17-AAG, essentially devoid of growth inhibitory effects, can nonetheless sensitize SKOV-3 cells to paclitaxel. In similar experiments done with IGROV-1 cells, the paclitaxel IC₅₀ was reduced from 37 to 10 nmol/L (mean of two independent experiments) upon the addition of 100 nmol/L 17-AAG (a concentration that alone induces ∼5% growth inhibition). This effect was ascribed to synergy because the CI at fu_0.5 were 0.63 and 0.51.

SKOV-3 cells were exposed to 80 nmol/L 17-AAG or 50 nmol/L paclitaxel (approximate respective IC₅₀ values in the 96 hours MTT assay) or to the combination, and the attached and detached cells were counted at 24 hours. The individual agents did not significantly reduce the number of attached cells at this time but the combination led to an ∼50% reduction compared with controls (data not shown). The detached cell number for the individual drugs was similar to control at 24 hours (i.e., ∼8% of total cells), but ∼45% of cells were detached in the presence of both drugs (Fig. 4A). Furthermore, the detached cells displayed apoptotic morphology (data not shown) and cleaved PARP was clearly visible by immunoblotting (Fig. 4B). These data are consistent with the synergy observed in the MTT assay being attributable to an earlier induction of apoptosis. Similar synergy at the level of apoptosis was confirmed in IGROV-1 and A431 cells (Fig. 4A). In contrast, the combination of IC₅₀ concentrations of 17-AAG and paclitaxel in CH1 cells did not lead to a reduction in the attached cell number at 24 hours compared with the individual agents alone (data not shown). Although paclitaxel induced some detached apoptotic cells at 24 hours (∼22% of total), the coaddition of 17-AAG did not lead to any further increase (Fig. 4A). Furthermore, the amount of cleaved PARP was not increased in the total CH1 cell population treated with both agents compared with paclitaxel alone at 24 hours (Fig. 4B). Increasing the concentration of paclitaxel 3-fold (to 15 nmol/L) led to more detached apoptotic cells at 24 hours (∼32%) but coaddition of 17-AAG reduced this to ∼10% (Fig. 4A).

In experiments described earlier, 6 nmol/L 17-AAG (a non–growth-inhibitory concentration) sensitized SKOV-3 cells to paclitaxel using the MTT assay as an end point. As might be expected, 6 nmol/L 17-AAG alone did not lead to a reduction in attached cells compared with controls (up to 72 hours; Fig. 3B). However, when this concentration was combined with 50 nmol/L paclitaxel (∼IC₅₀), the number of attached cells was reduced at all time points compared with paclitaxel alone. Furthermore, the number of detached cells was higher at 24 hours compared with paclitaxel alone (Fig. 3C), ∼38% and ∼8%, respectively. This was associated with an increase in cleaved PARP in the total cell population (Fig. 3D).

The exposure time and concentration dependence of 17-AAG-mediated depletion of selected HSP90 client proteins and their downstream effectors was examined in SKOV-3 cells. An increase in inducible HSP70 was used as a sensitive marker of HSP90 inhibition (43, 44). Figure 5A shows the effect of 60 and 180 nmol/L 17-AAG (1 × IC₅₀ and 3 × IC₅₀, respectively) at 4, 8, 16, and 24 hours. The proteins most sensitive to HSP90 inhibition that we examined were ERBB2 and HSP70, markedly decreasing and increasing, respectively, at 4 hours following exposure to 60 nmol/L 17-AAG. C-RAF and p-AKT were decreased at the 16-hour time point (8 hours after exposure to 180 nmol/L 17-AAG). Although HSP90 maintains the phosphorylation status of AKT, p-AKT can also be reduced as a result of 17-AAG-induced reduction in ERBB signaling. Total AKT was not markedly reduced unless cells were exposed to 200 nmol/L 17-AAG for 24 hours (data not shown). p-ERK 1/2 was decreased at 8 hours and very noticeably at 24 hours following exposure to 60 nmol/L 17-AAG (Fig. 5A). This marker of C-RAF signaling was markedly depleted 4 hours following exposure to 180 nmol/L 17-AAG (3 × IC₅₀ concentration). Similar experiments were done on IGROV-1 cells exposed to equitoxic concentrations of 17-AAG. At the IC₅₀ (300 nmol/L 17-AAG), ERBB2 was reduced at 16 hours and C-RAF and p-AKT at 24 hours (later than in SKOV-3). At 3 × IC₅₀, both markers were reduced at 16 hours (data not shown). CH1 cells were exposed to 1.8 and 5.4 μmol/L 17-AAG (∼1 × IC₅₀ and 3 × IC₅₀, respectively) for 4, 8, 16, and 24 hours (Fig. 5B). An increase in HSP70 was observed after 4-hour exposure to 1.8 μmol/L. C-RAF reduction became more marked with time, and p-ERK and p-AKT were markedly reduced at 8 and 16 hours, respectively. The effects on all these markers were earlier (or more marked) at the higher concentration of 17-AAG.

Results described earlier in this article showed that at a very low or non–growth-inhibitory concentration of 17-AAG could sensitize IGROV-1 and SKOV-3 cells to paclitaxel. We therefore examined the effect of these low concentrations of 17-AAG on HSP90 inhibition and client protein reduction. IGROV-1 cells were exposed to 100 nmol/L 17-AAG (1/3 IC₅₀; ∼5% growth inhibition) and a reduction in ERBB2 was observed at 16 hours and a decrease in p-AKT (but not C-RAF) at 24 hours (data not shown). SKOV-3 cells exposed to 6 nmol/L 17-AAG (1/10 IC₅₀; non–growth inhibitory) for 4, 8, 16, and 24 hours had a markedly increased level of HSP70 at all time points, but C-RAF and p-ERK remained at control levels (Fig. 6A). However, there was a marked reduction in the level of ERBB2 at 16 hours and p-AKT at 24 hours (Fig. 6A). In contrast with effects in SKOV-3 cells, the effects of an equitoxic concentration of 17-AAG in CH1 cells (150 nmol/L 17-AAG; ∼1/10 of the IC₅₀) were less marked with only a small increase in HSP70 observed at 24 hours and no apparent changes to c-RAF, p-ERK, or p-AKT (data not shown).

In SKOV-3 cells, 50 nmol/L paclitaxel (∼IC₅₀) did not induce any apparent changes to the levels of HSP70, ERBB2, C-RAF, p-ERK, or p-AKT at 4, 8, 12, and 24 hours (Fig. 7; data not shown). Similar to the results described above, 80 nmol/L 17-AAG (∼IC₅₀) led to a marked increase in HSP70 and decreased ERBB2 at 8 hours and at later time points. The level of these proteins in cells coexposed to 50 nmol/L paclitaxel and 80 nmol/L 17-AAG were very similar to those for 17-AAG alone (Fig. 7). The 17-AAG-induced reduction in p-AKT at 12 and 24 hours was not affected by the coaddition of paclitaxel. Interestingly, the 17-AAG-induced reduction in C-RAF and p-ERK at these times was not observed in the combination (Fig. 7). Six-nanomolar 17-AAG is non–growth inhibitory; nevertheless, this paclitaxel-sensitizing concentration of 17-AAG is associated with increased HSP70 and decreased levels of ERBB2 and p-AKT but not of C-RAF or p-ERK. Similar changes in these proteins were observed when 50 nmol/L paclitaxel (IC₅₀) was added in addition to the 17-AAG (Fig. 6B).

The coaddition of 5 nmol/L paclitaxel (∼IC₅₀) and 1,800 nmol/L 17-AAG (∼IC₅₀) to CH1 cells did not affect the 17-AAG-induced increase in the level of HSP70 or reduction in p-AKT (Fig. 8). In contrast with the results with SKOV-3 cells, the combination did lead to a reduction in C-RAF and p-ERK that was similar to that seen with 17-AAG alone.

Ovarian cancer is frequently sensitive to paclitaxel and carboplatin; however, following relapse, treatment options become more limited (45). Some patients remain chemosensitive and can be retreated with the same therapy until the disease becomes highly refractory to conventional treatment. Novel molecularly targeted agents may offer promise for the future in the treatment of ovarian cancer at any stage of the disease. Inhibitors of HSP90, such as 17-AAG, interfere with signaling pathways that are central to tumor growth and its metastatic spread (8–13). Preclinical studies have shown that 17-AAG may be combined effectively with conventional agents in a number of preclinical models (17, 39) and these include the agents particularly relevant to ovarian cancer. We have now shown the potential for 17-AAG to be combined with paclitaxel in ovarian cancer. However, we emphasize the need to better understand issues such as mechanism of action and molecular context so that ultimately the combination can be targeted at those patients most likely to respond.

The combination of 17-AAG and paclitaxel was synergistic in the two cell lines with a constitutively activated AKT (IGROV-1 and SKOV-3). The ability of 17-AAG to reduce p-AKT levels is therefore central to our studies, and a comparison was made between the molecular effects of 17-AAG on HSP90 inhibition in the SKOV-3 and CH1 ovarian tumor cells, in which the synergy was not seen. 17-AAG mediated reductions in C-RAF, p-ERK, and p-AKT in both cell lines. Changes were also observed in SKOV-3 cells but not CH1 cells at low, non–growth-inhibitory concentrations of 17-AAG. However, these effects at low concentrations were confined to reductions in ERBB2 and p-AKT and not to C-RAF or p-ERK. Similar results to SKOV-3 cells were obtained in IGROV-1 cells. The fact that these low concentrations were all that was required to sensitize SKOV-3 and IGROV-1 cells to paclitaxel further implicates inhibition of PI3K/AKT signaling as a crucial mediator of 17-AAG-induced sensitization to paclitaxel. These data also suggest that although full activity of this pathway is not critical for proliferation in SKOV-3 and IGROV-1 cells, it is essential in mediating paclitaxel resistance in these cells. Our results confirm the importance of inhibiting the PI3K/AKT pathway as one method for overcoming resistance in ovarian cancer. In contrast to the results with SKOV-3 and IGROV-1 cells, CH1 cells, which have a lower basal level of p-AKT, are more sensitive to paclitaxel and are not sensitized further by the addition of 17-AAG. This may suggest that p-AKT is not a critical mediator of paclitaxel resistance in CH1 cells. In partial support of this hypothesis is the fact that although stable transfection of CH1 cells with ERBB2 (CH1:ERBB2) increased the level of p-AKT and sensitized them to 17-AAG (10-fold), the combination with paclitaxel remained antagonistic.⁵

However, there could be an alternative explanation for the different effects of the paclitaxel/17-AAG combination between cell lines based on the effect paclitaxel may have on 17-AAG-induced effect on cell signaling. Paclitaxel prevented the 17-AAG-induced reduction in C-RAF signaling in SKOV-3 cells when both agents were combined at their respective IC₅₀ concentrations. Similar results to SKOV-3 were obtained using A431 epidermoid cells in which the combination was synergistic (data not shown). In contrast, paclitaxel did not affect 17-AAG-mediated inhibition of C-RAF signaling in CH1 cells. Although the reasons are unclear, it is therefore possible that 17-AAG-mediated inhibition of ERK signaling in CH1 cells opposes the drug sensitization effects of reduced AKT signaling to an extent not seen in SKOV-3 cells.

The role that ERBB family member overexpression plays in mediating synergy when 17-AAG and paclitaxel are combined is unclear. ERBB2 is regarded as one of the most sensitive client proteins to HSP90 inhibition (13). Munster et al. (46) showed that although 17-AAG led to degradation of ERBB2 in a number of ERBB2-expressing cell lines, only in tumors in which it was overexpressed did depletion of ERBB2 seem to be linked to growth arrest and apoptosis. However, the situation may be different when 17-AAG is used to sensitize to cytotoxic drugs by inhibiting survival pathways. SKOV-3 cells overexpress ERBB2, and when these cells are exposed to 17-AAG, a rapid reduction in ERBB2 was observed followed by a reduction in p-AKT. However, IGROV-1 cells express EGFR and ERBB2 at average levels and exposure to low concentrations of 17-AAG also led to a rapid reduction in the level of ERBB2 and sensitization to paclitaxel. Interpretation of these data is complicated by the fact that the phosphorylation status of AKT is maintained by HSP90, making it difficult to conclude whether the observed effect on p-AKT is directly related to 17-AAG-mediated destabilization of p-AKT rather than as an indirect result of ERBB2 reduction. Both mechanisms could apply and further work is needed to deconvolute the importance of each. The effect of other ERBB family members, such as EGFR, on the outcome of this combination is unclear. The HX62 line expresses high levels of EGFR but has low levels of p-AKT (noninducible by serum) and antagonism was observed with 17-AAG and paclitaxel. The A431 and KB epidermoid cell lines overexpress EGFR; however, whereas A431 cells express a moderate basal level of p-AKT, KB cells express a very low basal level of p-AKT (not inducible). The combination of 17-AAG and paclitaxel was weakly synergistic and antagonistic in A431 and KB, respectively.

Taking all the cell line data together, we can summarize the results as follows: synergy was observed in SKOV-3 (overexpressed ERBB2, high p-AKT) and IGROV-1 (high p-AKT); weak synergism was seen in A431 (overexpressed EGFR, moderate p-AKT); and antagonism was observed in CH1 (moderate p-AKT), HX62 (overexpressed EGFR, low p-AKT), and KB cells (overexpressed EGFR, low p-AKT). A reasonable hypothesis that needs to be tested further is that 17-AAG and paclitaxel are synergistic in cell lines that have an activated AKT pathway mediating paclitaxel resistance, driven at least in part by ERBB (over)expression. However, the fact that CH1:ERBB2 cells (discussed earlier) were not sensitized to paclitaxel by 17-AAG illustrates the complexity of the mechanisms involved and/or the limitations of these types of model. Currently, we are attempting to stably knock down AKT1/2 in SKOV-3 cells with short hairpin RNA to provide an alternative experimental model.

It is important to address the question of pharmacologic relevance of our results, particularly as we have evaluated the combination in cell lines with very different intrinsic sensitivities to 17-AAG. SKOV-3 cells are relatively sensitive in that the IC₅₀ (66 nmol/L) is lower than the mean IC₅₀ in a panel of 26 human ovarian and colon tumor cell lines (220 nmol/L; ref. 37) or the National Cancer Institute 60 cell line panel (120 nmol/L).³ Furthermore, we found that as little as 6 nmol/L 17-AAG sensitized SKOV-3 cells to paclitaxel. On the other hand, CH1 cells are relatively resistant to 17-AAG (IC₅₀ = 1,600 nmol/L), which relates to very low DT-diaphorase activity (37). Median-effect analysis is usually done at a fixed ratio of the IC₅₀ of the two drugs and therefore the question arises as to the clinical significance of the concentrations used. The plasma levels at the recommended phase II dose on a weekly schedule of 450 mg/m²/wk are ∼120 nmol/L for at least 24 hours (14). Thus, the concentrations that sensitized SKOV-3 and IGROV-1 cells (∼10–100 nmol/L) are easily achievable. In contrast, the concentration of 17-AAG used in CH1 cells either alone or in combination was much higher and probably not achievable clinically. Novel inhibitors of HSP90 are being developed that are not metabolized by DT-diaphorase (10, 47). It will be important to evaluate such compounds in the human ovarian tumor cell line and xenograft models.

The combination of 17-AAG and carboplatin was antagonistic in SKOV-3 and additive/weakly antagonistic in IGROV-1 cells. Because these results contrast with those of paclitaxel, it can be proposed that the mechanism of action of the cytotoxic agent may be crucial in determining whether it can be combined successfully with 17-AAG. It has been suggested that 17-AAG-mediated inhibition of c-Jun-NH₂-kinase signaling can lead to antagonism when cisplatin and 17-AAG are combined (39). It will be necessary to further explore the effect of order of addition of these drugs and effects on other human ovarian tumor cell lines.

In summary, this study suggests that 17-AAG has the potential to be synergistic with paclitaxel in some human ovarian tumors. Importantly, we have shown that 17-AAG does not necessarily have to be given at a concentration that alone affects proliferation or induces apoptosis. This may, however, be restricted to a subset of tumors that are highly sensitive to 17-AAG-mediated reduction of ERBB2 and/or p-AKT levels. Further work will expand this study to other models of ovarian cancer, including additional isogenic lines engineered for stable knockdown or expression of proteins such as AKT and ERBB2, and to newer inhibitors of HSP90 so that we can inform subsequent clinical studies with HSP90 inhibitors in humans. Retreatment with paclitaxel and/or carboplatin comprises an important element of current therapy for relapsed ovarian cancer. Our data support the view that the efficacy of retreatment with paclitaxel, but not necessarily carboplatin, may be increased by coadministration of 17-AAG (or potentially other inhibitors of PI3K/AKT signaling). However, this benefit may be limited to patients whose tumors have a defined molecular signature. Our studies should ultimately help guide clinical investigators to select patients who are likely to benefit from this type of combined therapy.