Global Targeting of Subcellular Heat Shock Protein-90 Networks for Therapy of Glioblastoma

Despite a better understanding of cancer genes (1), the survival rates of many human tumors have only marginally improved over several decades. This likely reflects the extraordinary molecular and genetic heterogeneity of the transformed cell (2), which, combined with the multilayer redundancy of cancer signaling pathways (3), promotes cellular adaptation and multidrug resistance (4). These challenges are exemplified by glioblastomas, a group of heterogeneous and highly malignant primary brain tumors with survival rates that rarely exceed 12 to 15 months after diagnosis (5). Although progress has been made in uncovering core pathways often deregulated in glioblastoma (6), this has not translated into better therapeutic options in the clinic (7–9), pressing the need for fresh approaches in the management of these patients (10).

Especially for such recalcitrant malignancies, drug discovery platforms are now evolving beyond traditional, single-gene targeting, to simultaneously disabling multiple signaling circuitries of disease maintenance (11). An ideal candidate for such “network-oriented” drug discovery is the molecular chaperone heat shock protein-90 (Hsp90; ref. 12), the ATPase-directed protein folding of which orchestrates multiple intersecting pathways of cellular adaptation, survival, and drug resistance (13). Accordingly, several small molecule Hsp90 inhibitors have been developed and are currently undergoing clinical testing (14). However, it has become increasingly clear that Hsp90 homeostasis is not arranged randomly in cells, but is spatially organized in multiple, semiautonomous subcellular compartments. In this context, Hsp90 and Hsp90-like chaperones oversee transcriptional responses in the nucleus (15), control client protein stability and maturation in the cytosol (16), maintain secretory pathways at the endoplasmic reticulum (17), and preserve mitochondrial integrity, especially in tumors (18), by suppressing cyclophilin D (CypD)–dependent permeability transition (19). Whether this subcellular compartmentalization of the chaperoning network is important for Hsp90-based therapy is currently unknown. However, it is intriguing that none of the structurally diverse Hsp90 inhibitors developed thus far has the ability to accumulate in mitochondria (20), suggesting that current therapeutic strategies do not encompass the globality of Hsp90 homeostasis, and this might affect their anticancer efficacy in the clinic (14).

In this study, we asked whether global targeting of Hsp90 homeostasis in its multiple subcellular compartments, including the mitochondria (18), could better fulfill the concept of “network inhibition” in controlling complex and traditionally recalcitrant tumors, i.e., glioblastoma. We found that a cell- and mitochondria-permeable peptidomimetic inhibitor of Hsp90 ATPase activity, Shepherdin (18, 21), induced the collapse of mitochondria and loss of Hsp90 client proteins in the cytosol, providing strong, single-agent antiglioma activity in vivo, without systemic or organ toxicity.

Twelve surgically resected samples of WHO grade 4 glioblastomas containing adjacent normal brain were collected and analyzed anonymously as discarded tissue by immunohistochemistry. The patient population comprised both males and females 27 to 79 years of age.

Human glioblastoma cell lines LN229 (mutant p53, wild-type PTEN), U87 (wild-type p53, mutant PTEN), U251 (mutant p53), or normal fetal human astrocytes were purchased from the American Type Culture Collection or ScienCell Research Laboratories. Patient-derived, primary cultures of glioblastoma cells (GS620, GS48, and AS515) were established as Mycoplasma-free from surgically resected WHO grades 3 and 4 glioblastomas. The glial origin of these cultures was confirmed by staining for α-glial fibrillary acidic protein (Dako), whereas antibodies against endothelial cell markers, CD31 (PharMingen) or factor VIII (Dako), or neuronal neurofilament proteins 70, 160, and 200 (all from Progen), were unreactive.

Antibodies to LC3, Beclin-1 (1:1,000; CST, Inc.), CypD, COX-IV (1:1,000; Calbiochem), cytochrome c (1:1,000; Clontech), Hsp70 (1:1,000; Abcam), Bcl-2 (1:1,000; CST), survivin (1:1,000; NOVUS Biologicals), XIAP (1:1,000; BD), Ser 473-phosphorylated Akt (1:500; CST), and Akt (1:1,000; CST) were used.

The cell- and mitochondria-permeable peptidomimetic Hsp90 inhibitor, Shepherdin, was characterized previously (18, 21), and synthesized in the reverse orientation employing all d-amino acids as follows: free/biotin-X-KKWKMRRNQFWVKVQRLFACGSSHK-CONH₂ (the cell-permeable Antennapedia helix III sequence is underlined; X, hexanoic acid spacer). A cell-permeable scrambled peptidomimetic was also synthesized with d-amino acids in the reverse orientation as follows: free/biotin-X-KKWKMRRNQFWVKVQRGHSFCALKS-CONH₂, and used as a control (18, 21). All peptidomimetics were dissolved in water and buffered to pH 7.4.

The various cell types were seeded in triplicate onto 96-well plates at 2 × 10³ cells/well, treated with Shepherdin or scrambled peptidomimetic (0–100 μmol/L) for up to 24 hours, and analyzed for metabolic activity by a 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay, as described (21). For determination of apoptosis, control or Shepherdin-treated glioblastoma cells (1 × 10⁶) were labeled for Annexin V plus propidium iodide (BD Bioscience), or alternatively, for DEVDase (Asp-Glu-Val-Asp, caspase) activity/propidium iodide (CaspaTag, Intergen), by multiparametric flow cytometry (Becton Dickinson), as described (18, 21). To quantify changes in mitochondrial membrane potential, glioblastoma cells were labeled for 10 minutes in the dark with the mitochondrial membrane potential–sensitive fluorescent dye, JC-1 (10 μg/mL; Molecular Probes), washed, and analyzed for modulation of red/green (FL-2/FL-1) fluorescence ratio using flow cytometry (18). In some experiments, glioblastoma cells were transfected with nontargeting or CypD-directed SMARTPool small interfering RNA (siRNA; Dharmacon; ref. 18), using OligofectAMINE 2000, confirmed for protein knockdown by Western blotting, and analyzed for cell viability by MTT or mitochondrial membrane potential. For characterization of autophagy, glioblastoma cells were transfected with an light chain 3-green fluorescent protein (LC3-GFP) cDNA by LipofectAMINE, treated with Shepherdin or scrambled peptidomimetic, and analyzed by fluorescence microscopy. A cell was scored as autophagic when exhibiting a punctate GFP labeling with >10 LC3-GFP dots/cell. An average of 200 cells was counted in four to six independent fields per condition.

All experiments involving animals were approved by the Institutional Animal Care and Use Committee. U87 glioblastoma cells stably transfected with a luciferase expression plasmid (U87-Luc) were suspended in sterile PBS (pH 7.2), and stereotactically implanted (1 × 10⁵) in the right cerebral striatum of immunocompromised CB17 severe combined immunodeficiency (SCID)/beige female mice (Charles River Laboratories). Animals with established tumors were randomized into two groups (n = 10; five animals/group), and started on day 8 after implantation on sterile vehicle (PBS at pH 7.2), or Shepherdin (50 mg/kg as daily i.p. injections) for 17 consecutive days. Tumor growth was assessed weekly after i.p. injection of 58 mg/kg of d-luciferin by bioluminescence imaging using a Xenogen In vivo Imaging System. In some experiments, animals were harvested on day 21 and subjected to histologic analysis. In other experiments, mice carrying established intracranial U87 gliomas (n = 6; three animals/group) were stereotactically injected with a single dose (0.5 mg) of vehicle or Shepherdin on day 8. On day 20, all animals were sacrificed and brain samples were harvested for histologic analysis.

Patient-derived tissue specimens were stained with H&E, or with an antibody to the mitochondrial Hsp90 chaperone, TRAP-1 (BD Biosciences), as described (18). Immunoreactive cells were scored semiquantitatively for staining intensity as 0, negative; 1, low; 2, medium; and 3, high; with the percentage of TRAP-1 expression in the cell population as 0, no expression; 1, <1%; 2, 1% to 9%; 3, 10% to 50%; and 4, >50%, as described (22). Brain samples from various animal groups were stained with H&E or with antibodies to Ki-67 (Zymed) or CD31 (Becton Dickinson), as described (20). For in situ determination of apoptosis, brain sections were analyzed for terminal deoxynucleotidyl transferase–mediated dUTP nick-end labeling (TUNEL; Roche). Images were collected on an Olympus microscope with an on-line charge-coupled device camera. Microvessel density in CD31-stained sections was determined using an image analysis algorithm (S. CO LifeScience Co., Garching, Germany), as described (23). In some experiments, cryosections of glioblastoma xenografts treated with vehicle or Shepherdin were stained with a 1:1,000 dilution of streptavidin-Texas red (Amersham Biosciences), and intratumoral accumulation of Shepherdin was visualized by fluorescence microscopy.

Data were analyzed by two-sided unpaired t tests using a GraphPad software package (Prism 4.0) for Windows. P = 0.05 was considered statistically significant. Values are expressed as means of triplicate or duplicate experiments. For preclinical studies, the mean and SEM were calculated with mice as the data units.

We began this study by examining a potential differential expression of one of the cytoprotective mitochondrial Hsp90 chaperones, TRAP-1 (18), in human gliomas. Immunohistochemical analysis of human grade 4 glioblastomas revealed that TRAP-1 was highly expressed in the tumor cell population (Fig. 1A and B). Adjacent normal astrocytes did not contain TRAP-1 (Fig. 1A and B), although a low level of TRAP-1 expression was detected in neurons (data not shown). TRAP-1 staining under these conditions appeared as a punctate, perinuclear reactivity, with strong to moderate signal intensity in the various glioblastoma cases examined, whereas a control, nonbinding IgG was negative (Fig. 1A). Consistent with these results, TRAP-1 was abundantly expressed in a panel of human glioblastoma cell lines, but was barely detectable in normal human fetal astrocytes (Fig. 1C).

Treatment of glioblastoma cells with Shepherdin, a cell- and mitochondria-permeable inhibitor of Hsp90 ATPase activity (18, 21), triggered sudden loss of mitochondrial membrane potential (Fig. 2A) and discharge of cytochrome c in the cytosol (Fig. 2B), two hallmarks of mitochondrial permeability transition (19). In addition, Shepherdin-treated glioblastoma cells exhibited biochemical markers of autophagy, with decreased expression of the autophagy-associated gene, Beclin-1, and the appearance of a lipidated, faster migrating form of the ubiquitin-like protein, LC3 (Fig. 2C), which is involved in autophagosome formation (24). This was associated with a punctate pattern of LC3-GFP labeling (Fig. 2D), characteristic of autophagy (24), in Shepherdin-treated cells. In contrast, a cell-permeable scrambled peptidomimetic did not affect mitochondrial membrane potential (Fig. 2A), or cytochrome c release (Fig. 2B), and did not induce markers of autophagy in glioblastoma cells (Fig. 2D).

To determine the mechanism of mitochondrial dysfunction induced by Shepherdin (19), we next acutely ablated CypD, a proapoptotic component of the permeability pore, in glioblastoma cell types. Transfection of LN229 cells with CypD-directed siRNA efficiently silenced the expression of endogenous CypD, whereas a control, nontargeting siRNA was ineffective (Fig. 3A). Under these conditions, CypD knockdown partially reversed Shepherdin-induced loss of mitochondrial membrane potential (Fig. 3B), and significantly reduced Shepherdin-mediated glioblastoma cell killing (Fig. 3C). In contrast, a nontargeting siRNA (Fig. 3A) had no effect on mitochondrial membrane potential (Fig. 3B), or tumor cell killing by Shepherdin (Fig. 3C).

Consistent with irreversible mitochondrial damage under these conditions, treatment of glioblastoma cell lines (Fig. 4A), or patient-derived glioblastoma cultures (Fig. 4B) with Shepherdin caused rapid (∼3 hours), concentration-dependent, and complete loss of metabolic activity, irrespective of p53 or PTEN status. In contrast, Shepherdin did not affect normal human fetal astrocytes (Fig. 4A), and a control, scrambled peptidomimetic was ineffective in normal or glioblastoma cell types (Fig. 4A and B). Shepherdin-mediated tumor cell killing under these conditions had the hallmarks of apoptosis, with extensive cellular labeling for Annexin V (Fig. 4C, top), and prominent DEVDase activity (i.e., caspase; Fig. 4C, bottom), by multiparametric flow cytometry.

In parallel experiments, Shepherdin treatment inhibited Hsp90 chaperone function in the cytosol of glioblastoma cells, resulting in concentration-dependent degradation of multiple cytoprotective Hsp90 client proteins, including Akt and Ser473-phosphorylated Akt, IAP proteins, XIAP and survivin, and Bcl-2 (Fig. 4D). In contrast, low-dose Shepherdin did not affect the cellular levels of Hsp70, whereas higher concentrations led to a weak reduction of Hsp70 levels that did not achieve statistical significance (Fig. 4D). In control experiments, a scrambled peptidomimetic had no effect on Annexin V labeling, expression of DEVDase activity (Fig. 4C), or cytosolic levels of Hsp90 client proteins (Fig. 4D) in glioblastoma cells.

Stereotactic implantation of U87-Luc glioblastoma cells in the right cerebral striatum of immunocompromised SCID/beige mice gave rise to exponentially growing tumors following bioluminescence imaging (Fig. 5A). Systemic administration of Shepherdin in these mice suppressed intracranial glioma growth in vivo (Fig. 5A), and significantly prolonged animal survival, even after treatment suspension, as compared with the vehicle group (Fig. 5B). Brain specimens of treated animals revealed extensive intraglioma accumulation of Shepherdin following fluorescence imaging in vivo, whereas samples from control mice were unreactive (Fig. 5C). Systemic treatment with Shepherdin was not associated with systemic or organ toxicity, and did not result in animal weight loss throughout treatment (Fig. 5D). In contrast, animals in the vehicle group exhibited progressive weight loss due to advanced disease (Fig. 5D). Histologically, intracranial gliomas growing in vehicle-treated mice exhibited an elevated mitotic index, negligible apoptosis, and extensive angiogenesis (Fig. 6A and B). In contrast, brain sections collected from Shepherdin-treated mice revealed significant inhibition of glioblastoma cell proliferation, increased apoptosis, and suppression of angiogenesis in vivo (Fig. 6A and B).

In the normal brain, the neurons contain low but detectable levels of endogenous Hsp90s in the mitochondria; therefore, we next asked (in this study) whether delivery of Shepherdin directly into the brain provided antiglioma activity without toxicity. In these experiments, a single stereotactic administration of Shepherdin was sufficient to suppress intracranial glioma growth in implanted SCID/beige mice, whereas animals in the control group exhibited exponential intracranial glioma growth (Supplementary Fig. S1A). Histologic examination of brain samples collected from Shepherdin-treated mice was unremarkable compared with control animals receiving vehicle (Supplementary Fig. S1B). Conversely, stereotactic delivery of Shepherdin significantly inhibited glioblastoma cell proliferation in vivo (Supplementary Fig. S1B and C).

In this study, we have shown that global subcellular targeting of the Hsp90 networks in cytosol and mitochondria with Shepherdin (18, 21) provides strong, single-agent activity in preclinical models of glioblastoma, one of the most recalcitrant human malignancies (5). Mechanistically, Shepherdin induced a simultaneous collapse of mitochondrial integrity and degradation of multiple Hsp90 client proteins in the cytosol, triggering tumor cell killing by apoptosis and autophagy, without affecting normal astrocytes. When given systemically or stereotactically to mice, Shepherdin-based therapy was remarkably well tolerated at a preclinically effective dosage of 50 mg/kg/daily, comparable with the therapeutic range of other conventional or targeted anticancer agents, and potently inhibited intracranial glioma growth via induction of apoptosis, suppression of tumor cell proliferation, and reduction of angiogenesis in vivo.

Although Hsp90-based therapy (12, 14) has been intensely pursued as a paradigm of network-oriented drug discovery (11), the clinical results with these agents have thus far been inferior to the expectations, producing only small gains in cancer patients, and often at a cost of significant toxicity (14). Although several factors might contribute to these results, it is possible that the inability of current Hsp90 inhibitors to target the multiple subcellular pools of the chaperone, especially a cytoprotective fraction in the mitochondria that antagonizes CypD-initiated permeability transition (20), might reduce their clinical efficacy. In this context, Shepherdin provides a conceptually and structurally unique Hsp90 ATPase inhibitor, capable of accumulating in both the cytosol (21) and in the mitochondria (18), and thus is suited to globally inhibit chaperone homeostasis in its multiple subcellular fractions (13).

Consistent with these predictions, Shepherdin induced sudden and complete collapse of mitochondrial integrity, triggering CypD-dependent permeability transition, and massive induction of apoptosis in glioblastoma cells. Mechanistically, the “mitochondriotropic” properties of Shepherdin depend on its NH₂-terminus Antennapedia helix III cell-penetrating sequence (18). Although it is unclear how this region promotes the transfer of cargo across biological membranes, including mitochondria (25), Antennapedia-conjugated Shepherdin has been shown to readily distribute throughout all submitochondrial compartments in tumor cells, and to physically associate with Hsp90 chaperones inside the organelle (18). In addition to triggering apoptosis, mitochondrial dysfunction has been linked to the induction of autophagy (26), and accordingly, Shepherdin-treated glioblastoma cells exhibited biochemical markers of autophagy, including loss of Beclin-1 and posttranslational processing of the LC3 ubiquitin ligase. Although the implications of autophagy for tumor progression are still being elucidated, this process may be important as a cell death mechanism in glioblastoma (24), efficiently activated in response to stress stimuli (27).

The second phenotype induced by Shepherdin involved acute inhibition of Hsp90 chaperoning in the cytosol (12, 13), with concomitant loss of multiple cytoprotective client proteins, including Bcl-2, XIAP, survivin, and Akt. In addition to lowering the cellular antiapoptotic threshold, thus further aiding in antiglioma activity, this response might contribute to the antiangiogenic activity of Shepherdin observed here. Accordingly, critical angiogenesis regulators, including HIF-1α (28) and Erk (29), are recognized client proteins of Hsp90, and their acute downregulation after chaperone inhibition might contribute to impaired blood vessel formation. Alternatively, it is possible that angiogenic stimulation might recruit Hsp90s to the mitochondria of proliferating, but not quiescent, endothelial cells, thus making them susceptible to Shepherdin-induced killing. Conversely, one important difference with non-subcellularly targeted Hsp90 inhibitors is that Shepherdin treatment did not increase the levels of Hsp70 in glioblastoma cells (12). Although the basis for this response remains to be elucidated, the lack of Hsp70 modulation under these conditions might be therapeutically beneficial, as induction of this chaperone by conventional Hsp90 inhibitors has been linked to the suppression of apoptosis and reduced anticancer activity (30).

The strategy of comprehensive subcellular targeting of Hsp90 networks proposed here may offer tangible advantages over current Hsp90 antagonists. One of the most studied agents in this class, the benzoquinone ansamycin derivative 17-allylamino demethoxygeldanamycin (17-AAG; refs. 12, 14) has shown modest single-agent activity in glioblastoma models (22, 31), often further attenuated by resistance mechanisms (32). Instead, because of its direct, “mitochondriotoxic” mechanism of action, Shepherdin indistinguishably kills apoptosis-sensitive or -resistant glioblastoma cell types, irrespective of survival mechanisms, i.e., expression of Bcl-2, which typically reduces the efficacy of apoptosis-based therapeutics, including 17-AAG (33). Second, Shepherdin exerts its antiglioma activity independently of p53 and regardless of PTEN status, thus providing broader therapeutic prospects for genetically heterogeneous tumors. Third, Shepherdin-based therapy is remarkably well tolerated in vivo. This likely reflects the differential distribution of Hsp90s in the mitochondria of tumors, including glioblastomas (this study), but not in most normal tissues (18). Intriguingly, despite the fact that normal neurons are among the few tissues that contain low levels of Hsp90s in mitochondria (18), stereotactical injection of Shepherdin in the brain was well tolerated and exerted potent antiglioma activity without detectable local or systemic toxicity in vivo. This suggests that, in normal tissues, the lower affinity with which Hsp90 binds ATPase-directed antagonists (34), including Shepherdin (21), might further contribute to the overall safety of this agent in vivo.

In summary, these data establish the rationale for the pursuit and development of a novel class of global Hsp90 inhibitors (18), including Shepherdin, capable of simultaneously disabling the multiple, subcellularly compartmentalized pools of the chaperone. Such an approach may more adequately fulfill the concept of network-oriented drug discovery, and offer fresh therapeutic approaches for the management of heterogeneous and otherwise recalcitrant human tumors, including glioblastomas.

No potential conflicts of interest were disclosed.

We thank Drs. Eric Baehrecke and Claire-Marie Sauvageot for reagents, and Kathryn Chase and Neil Aronin for providing the stereotactical frames.

Grant Support: NIH grants CA78810, CA90917 and CA118005 (D.C. Altieri), and Deutsche Forschungsgemeinschaft grant, Si 1546/1-1 (M.D. Siegelin).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.