Inhibition of Hsp90 Function by Ansamycins Causes Retinoblastoma Gene Product-dependent G₁ Arrest¹

Ansamycin antibiotics are natural products derived from Streptomyces hygroscopicus that have profound effects on eukaryotic cells. These drugs,HA⁴and GM, bind tightly to a pocket in the protein chaperone, Hsp90(1). Hsp90 plays a role in protein refolding in cells exposed to environmental stress and is required for the conformational maturation of several important signaling proteins (2),including steroid receptors (3–5), the serine kinase Raf(6, 7), and the tyrosine kinase v-Src (8). The binding of ansamycins to Hsp90 has been shown to inhibit protein refolding and to cause the proteasome-dependent degradation of a select group of cellular proteins (9, 10). These include proteins known to require Hsp90 function (4, 7, 10–12) as well as several transmembrane tyrosine kinases, including ErbB family members(13–16) and the insulin-like growth factor receptor(9).

The ansamycins were originally isolated on the basis of their ability to revert v-src-transformed fibroblasts (17). Subsequently, they were shown to have antiproliferative effects on cells transformed with a number of oncogenes, particularly those encoding tyrosine kinases (18). Inhibition of cell growth is associated with apoptosis and, in certain cellular systems, with induction of differentiation(19).⁵These findings have led to the development of a GM derivative currently in Phase I clinical trials. The spectrum of ansamycin targets suggests that these drugs disrupt multiple key regulatory pathways. However, the induction of apparently complete reversion of the transformed phenotype in certain malignant cells (20) and the occasional profound differentiation of treated tumor cell lines(19) ⁵ made us question whether the effects of ansamycins should be attributed to global inhibition of Hsp90 housekeeping function. Instead, we considered whether occupancy of the pocket exposed a regulatory function of this chaperone.

The ansamycin-binding pocket in the NH₂-terminus of Hsp90 is highly conserved and has weak homology to the ATP-binding site of DNA gyrase (1, 21). This pocket has been shown to bind ATP and ADP with low affinity and to have weak ATPase activity (22, 23). Studies on the effects of binding to the pocket have led to different conclusions regarding the effects of ansamycins on Hsp90 function. High concentrations of drug were shown to prevent the binding of Hsp90 to several of its protein partners (7, 10, 24). However, in other systems, the drug has been shown to stabilize the Hsp90-refolded protein complex and prevent ATP-dependent release of the mature protein. The stable complex is subsequently degraded in the proteasome (25). Occupancy of the pocket could then have two potential consequences: (a) it could result in global pharmacological inhibition of Hsp90 functions; and (b) it could determine whether Hsp90 protein partners are folded or degraded,depending on the nature of the ligand.

To address the possibility that the binding of ansamycins to Hsp90 regulates specific signaling pathways, we investigated the mechanism whereby they inhibit cellular proliferation. We found that HA effectively arrests the growth of a variety of tumor cell lines in G₁ phase. G₁ block is associated with accumulation of hypophosphorylated RB and loss of G₁cyclin-associated kinase activities. This inhibition involved a rapid decrease in the expression of cyclin D, cyclin E, cdk4, and cdk6. In addition, a late induction of p27^Kip1 was also seen. However, the decline in cyclin D1 preceded other observed cell cycle effects, suggesting that the cyclin D-RB pathway is the primary target of the drug. Indeed, treatment of Rb-negative tumor cell lines with HA failed to induce G₁ arrest, demonstrating that the effects of ansamycins on G₁ are RB dependent. These results suggest that ansamycins selectively affect mitogenic signaling pathways upstream of RB, including those that control the levels of cyclin D. Furthermore, Hsp90 may play a role in regulating these pathways under certain physiologic conditions.

The human breast cancer cell lines MB-MDA 468, MCF-7, and the human colon carcinoma cell line, Colo 205, were obtained from ATCC. MB-MDA 468 and MCF-7 cells were maintained in DME-F12 media and Colo 205 cells in RPMI; both media were supplemented with 5% FCS (BRL), 2 mm glutamine, and 50 units/ml each of penicillin and streptomycin. All cells were incubated at 37°C in 5%CO₂. Cells were treated for 24 h with 500 ng/ml HA(Life Technologies, Inc.) dissolved in DMSO. Rapamycin (Sigma) was dissolved in DMSO and used at a final concentration of 250 ng/ml. For synchrony experiments, cells were treated with 400 ng/ml nocodazole(Sigma) or 1 μg/ml aphidicolin (Sigma) for 14 h, washed, and replated in media containing either DMSO or HA.

Nuclei were isolated for flow cytometry assays as described previously(13), stained with ethidium bromide, and analyzed using a Becton Dickinson fluorescence-activated cell sorter. Statistical data were obtained using Multicycle program software.

Treated cells were harvested, washed with PBS, and lysed in NP40 lysis buffer [50 mM Tris (pH 7.4), 1% NP40, 150 mM NaCl, 40 mM NaF, 1 mm Na₃VO₄, 1 mmphenylmethylsulfonyl fluoride, and 10 μg/ml each of leupeptin,aprotinin, and soybean trypsin inhibitor] for 30 min on ice. Lysates were centrifuged at 15,000 × g for 10 min to pellet debris, and the protein concentration of the supernatant was determined by bicinchoninic acid protein assay (Pierce).

Equal amounts of total protein were resolved by SDS-PAGE and transferred onto Immobilon polyvinylidene difluoride membranes(Millipore) by electroblotting. Blots were blocked overnight in 5%nonfat milk in TBS-T [0.1% Tween-20 TBS, 10 mm Tris (pH 7.4), and 150 mm NaCl] at 4°C and subsequently probed with antibody raised against the protein of interest. Anti-cyclin D1,cyclin D3, cyclin E, cyclin A, p27^Kip1, cdk2, cdk4, and cdk6 antibodies were obtained from Santa Cruz Biotechnology and anti-RB antibodies purchased from Pharmingen. After incubation with horseradish peroxidase-conjugated secondary antibodies, proteins were detected using chemiluminescence (Amersham). All quantitations were performed using the Gel Doc 1000 and Molecular Analyst software from Bio-Rad.

For cdk4- and cdk6-associated in vitro kinase assays,treated cells were resuspended in lysis buffer [50 mmHEPES-KOH (pH 7.5), 150 mm NaCl, 1 mm EDTA, 2.5 mm EGTA, 1 mm DTT, 0.1% Tween-20, 10%glycerol, 10 mm β-glycerophosphate, 1 mm NaF,0.1 mm Na₃VO₄, 0.2 mmphenylmethylsulfonyl fluoride, and 10 μg/ml each of leupeptin and aprotinin], sonicated and centrifuged at 15,000 × g for 10 min. Anti-cdk4 or anti-cdk6 antibodies preincubated with protein G-Sepharose (Pharmacia) were then incubated with 450 μg of total lysate for 1 h at 4°C. Immune complexes were washed four times with lysis buffer, two times with kinase buffer (250 mmHEPES-KOH, 50 mm MgCl₂, 5 mm DTT,12.5 mm EGTA, 50 mm β-glycerophosphate, 5 mm NaF, and 0.5 mmNa₃VO₄) and incubated in 30 μl of kinase buffer containing 0.2 μg of full-length GST-RB (QED Bioscience,Inc.), 10 μCi [γ-³²P]ATP, and 300 μmLi-ATP for 10 min at 30°C. The reaction was stopped by the addition of SDS-PAGE sample buffer and boiled for 5 min. Proteins were resolved on SDS-PAGE, transferred onto nitrocellulose membrane, and exposed to autoradiography film.

For cyclin E in vitro kinase assay, 100 μg of NP40 lysate were incubated for 1 h at 4°C with cyclin E antibodies preincubated with protein A-Sepharose (Pharmacia). Immune complexes were washed four times with lysis buffer, two times with kinase buffer[20 mm Tris (pH 7.4), 7.5 mM MgCl₂, and 1 mM DTT] and incubated in 40 μl of kinase buffer containing 2 μg of histone H1, 10 μCi of [γ-³²P]ATP, and 300μ m Li-ATP for 10 min at 37°C. The reaction was stopped by the addition of SDS-PAGE sample buffer and boiled for 5 min. Proteins were resolved on SDS-PAGE, transferred onto Immobilon, and exposed to autoradiography film.

HA treatment potently inhibited cellular proliferation in 12 cell lines derived from a variety of tumors including breast, colon, and prostate(data not shown). In certain cellular systems, growth inhibition was found to be accompanied by differentiation. This raised the possibility that HA exerts its effects by inhibiting specific signaling pathways rather than by generally disrupting chaperone function. As such, the mechanism by which HA inhibited growth was investigated. The effects of HA on cell cycle progression were examined in the human colon carcinoma cell line Colo 205 and the human breast cancer cell line MCF7. In both Colo 205 and MCF7, 24 h of drug treatment resulted in growth arrest and an accumulation of cells in G₁ as well as a loss of cells in S phase, indicating that HA induces G₁ arrest(Fig. 1). This HA-induced G₁ arrest was also observed in 10 additional cell lines tested, demonstrating the generality of the effect (data not shown). An increase in cells with 4N DNA content was also observed after drug treatment. This accumulation, however, was lost by longer treatment times⁶and likely represents a lagging population.

A key regulator of the G₁-S transition is the retinoblastoma gene product (RB). Phosphorylation of RB results in its inactivation and allows traversal through G₁ into S phase. Immunoblot analysis of lysates from HA-treated Colo 205 cells revealed an accumulation of the hypophosphorylated form of RB, consistent with an arrest in G₁ phase. A decrease in RB phosphorylation was observed by 2 h of treatment. In addition, a decline in the overall level of RB was also detected by 12 h (Fig. 2).

The phosphorylation of RB is catalyzed, at least in part, by the D-type cyclins and their associated catalytic partners, cdk4 and cdk6, as well as by cyclin E and its catalytic partner, cdk2. The observed accumulation of hypophosphorylated RB, thus, could be attributable to an inhibition of cyclin D/cdk4/6 and/or cyclin E/cdk2-associated kinase activities. To address this possibility, immune complexes were isolated from HA- or DMSO-treated Colo 205 cells using anti-cdk4, anti-cdk6, or anti-cyclin E antibodies. The complexes were then assayed for in vitro kinase activity using GST-RB as substrate for cyclin D/cdk4/cdk6-associated kinases and histone H1 for cyclin E/cdk2-associated kinase. Both cdk4- and cdk6-associated kinase activities declined by 50% within 30 min of HA addition when compared with levels in DMSO-treated cells and were undetected by 24 h of HA treatment (Fig. 3,A). Cyclin E-associated kinase activity was down-regulated later, declining 30% within 2 h after drug addition (Fig. 3 B). Thus, HA appears to induce accumulation of hypophosphorylated RB by inhibiting G₁ cyclin-associated kinase activities.

Regulation of the G₁ cyclin-associated kinases has been shown to occur at several levels, including complex assembly, cyclin and cdk expression, and induction of cdk inhibitors (26). To determine the mechanism by which cyclin D- and E-associated kinase activities are inhibited, the effects of HA on the expression of their protein components were examined. Treatment with HA was found to down-regulate expression of the G₁ cyclins. A decline in cyclin D1 protein level was detected within 30 min of HA treatment and a 50% reduction by 4 h (Fig. 4,A). Cyclin D3 and cyclin E protein expression were also down-regulated,with lowered levels initially observed at 2 h after addition of HA(Fig. 4,A). A reduction of 50% was seen for both cyclins D3 and E by 4 h. In contrast to its effects on the G₁cyclins, HA did not detectably inhibit expression of cyclin A (Fig. 4 A).

In addition to the G₁ cyclins, HA treatment also affected the expression of their associated cdks. Lowered cdk4 and cdk6 expression was first detected 2 h after HA addition (Fig. 4,B). cdk4 protein levels were reduced by 50% in 6 h,whereas cdk6 protein levels fell to 50% by 4 h. The total level of cdk2 protein, however, remained unchanged, and instead, HA caused an accumulation of the faster migrating, active form of the protein by 12 h after drug treatment (Fig. 4 B). Despite this accumulation of the active form of cdk2, E/cdk2-associated kinase activity is reduced, indicating that the cyclin E subunit is the rate-limiting component.

Induction of cdk inhibitors has been shown to accompany cell cycle arrest in response to various antiproliferative stimuli. As such, the level of p27^Kip1 was examined in HA-treated cells. An initial increase in the expression of p27^Kip1 was observed by 24 h after HA treatment. Further induction of the protein was detected when treatment was extended, with a 7-fold increase seen by 48 h (Fig. 4 C). p21^Cip1 protein expression was undetectable in Colo 205 cells by Western blot analysis. Together,these results demonstrate that HA inhibits G₁cyclin-associated kinase activities by down-regulating expression of the G₁ cyclins and cdk4 and cdk6. Induction of p27^Kip1 occurs after the observed declines in kinase activity and is likely a consequence and not a cause of kinase inhibition and G₁ arrest.

The observed loss of cyclin D1 protein occurred rapidly and preceded that of cyclin E. To determine whether this loss is a primary cause or result of G₁ arrest, HA was added to Colo 205 cells released from aphidicolin block. Cyclin D1 expression declined within 30 min, suggesting that HA down-regulates cyclin D1 independently of G₁ arrest (Fig. 4,D). Together, these findings raise the possibility that the primary downstream target of HA is cyclin D. Several studies have shown that cyclin D regulates G₁ progression by inducing the phosphorylation of RB(27–33), and that, in the absence of RB, cyclin D-associated kinase activity is dispensable (34–39). Inhibition of cyclin E-associated kinase, however, has been shown to cause G₁ arrest independently of RB status(40). Thus, if HA works by selectively inhibiting pathways that affect cyclin D-associated kinase activity, it would not affect G₁ progression in cells lacking functional RB. In four such cell lines, MB-MDA 468, BT-549, DU145, and DU4475, HA treatment inhibited cell growth but failed to induce a G₁ block and instead, caused an accumulation of cells with 4N DNA content(Fig. 5 & data not shown). This is not attributable to a general defect in G₁-S regulation in Rb-deficient cells because rapamycin induced a G₁ arrest in MDA 468 cells (Fig. 6).

To further establish that G₁ progression is unaffected by HA in Rb-negative cell lines, MB-MDA 468 cells were arrested in mitosis with nocodazole and then released into media containing either HA or DMSO. The Rb-wild type cell line, Colo 205, was treated similarly for comparison. After release into HA, Colo 205 cells progressed from mitosis into G₁ and arrested there (Fig. 7,A). In contrast, the Rb-negative MB-MDA 468 cells progressed through G₁, and the majority of cells were in S and G₂-M 24 h after release (Fig. 7 B). These results demonstrate that HA induces G₁ arrest in an RB-dependent manner. Furthermore, these findings imply that the protein targets of HA affect G₁ progression solely by regulating pathways upstream of RB.

Hsp90 is an abundant and ubiquitous chaperone, but its roles in cellular metabolism are not completely defined (41). On one hand, Hsp90 appears to play an important role in the refolding of cellular proteins and thus, in protecting cells from environmental stress. On the other hand, under normal conditions, it appears necessary for the conformational maturation of several classes of proteins involved in transducing or mediating growth signals. These include the steroid receptors kinases (3–5), cdk4(42, 43), Raf kinase (6, 7), and certain tyrosine kinases (8).

The NH₂-terminal of Hsp90 contains a highly conserved pocket that has homology to the DNA gyrase ATPase (1, 21)and binds ATP and ADP with low affinity (22, 23). Several natural products derived from microorganisms, including the ansamycins and radicicol, have apparently evolved as high-affinity ligands for this pocket (1). Occupancy of this pocket by these compounds prevents maturation of Hsp90 protein substrates and leads to their proteasome-dependent degradation (9, 10). The function of these molecules in the microorganism is unknown, but they have been speculated to protect the host from other microorganisms.

In cancer cells, ansamycins induce the degradation of several proteins important for maintaining the transformed phenotype. These include ErbB family and insulin-like growth factor receptor tyrosine kinases(9, 13–16), Raf (7, 11), and the steroid receptors (4, 10, 12). Treatment of various tumor cell lines with ansamycins results in growth inhibition and subsequently, in cell death. However, a drug that inhibits general chaperone housekeeping function and causes the destruction of so many key signaling proteins would be expected to cause rapid, extensive cytotoxicity. This is in apparent contradiction to the induction of differentiation in tumor cell lines and selective toxicity for tumor cells observed in animal models (44).

In this study, we set out to examine whether ansamycins inhibit cell growth by disrupting specific signaling pathways or general chaperone function. We found that HA caused most tumor cell lines to undergo growth arrest in G₁ phase of the cell cycle. This inhibition was accompanied by hypophosphorylation of RB and correlated with decreases in D-type cyclins and cyclin E-associated kinase activities. The mechanism of inhibition of G₁cyclin-associated kinase activities appeared multifactorial, consistent with disruption of multiple pathways responsible for mitogenic signaling. HA treatment was found to be accompanied by down-regulation of D-type cyclins, cdk4, cdk6, and cyclin E as well as a late induction in p27^Kip1.

However, the decrease in cyclin D1 preceded that of other G₁ regulators. In addition, loss of cyclin D1 was the cell cycle effect most consistently observed in cell lines tested. This decrease was not a result of G₁ arrest, occurring with rapid kinetics even when HA was added to cells synchronized at the G₁-S interface. Furthermore, down-regulation of cyclin D was also observed in Rb-negative cells that do not arrest in G₁ in response to HA.⁶ These data, instead,suggest that ansamycins cause G₁ arrest by selectively inhibiting pathways required for D-type cyclin-associated kinase activity. This proved to be the case. The most recognized substrate of D-type cyclin-associated kinases is RB (26–33, 45). We show that when treated with HA, tumor cells that lack functional RB failed to arrest in G₁. Furthermore, RB-negative cells synchronized with nocodazole were able to traverse G₁ and progressed into S phase in the presence of ansamycins.

Two surprising conclusions are implied by these data. The results suggest that the signaling proteins targeted by HA, including steroid receptors, many transmembrane tyrosine kinases, and Raf, affect G₁ progression solely by regulating pathways upstream of RB. Inhibition of cyclin E-associated kinase activity and other late G₁ targets, as well as induction of p27^Kip1,would be expected to cause G₁ arrest irrespective of RB status (40, 46, 47). Thus, rapamycin, lovastatin, and overexpression of p27^Kip1 induce G₁ block in cells lacking RB, in contrast to HA (46, 48).

The data also imply that Hsp90 specifically regulates elements of pathways responsible for induction of cyclin D-associated kinase activity. One may speculate that under certain environmental conditions or stresses, occupancy of the Hsp90 pocket leads to degradation of these elements, hypophosphorylation of RB, and G₁ arrest. Recently, certain stresses have been shown to cause a G₁block mediated by reduction in cyclin D expression (49, 50). Moreover, the mitogenic effects of growth factor-receptor tyrosine kinases and estradiol activation of estrogen receptor have been shown to converge at the level of induction of cyclin D(51). These findings together further suggest that the effects of multiple environmental stimuli are integrated at the level of cyclin D expression.

The reduction of cyclin D expression by ansamycins would seem sufficient to cause G₁ block. However, it is not necessarily the only consequence of inhibition of these Hsp90-dependent pathways. Our data does not rule out the possibility that these pathways regulate other determinants of cyclin D-associated kinase activity, such as the expression of cdk4, cdk6, and members of the p16^INK4a family. In fact, Hsp90 is involved in the maturation of cdk4 (42, 43), and under certain conditions it may be a direct target of ansamycins. The inhibition of Hsp90 function, thus, may result in the coordinate down-regulation of several pathways required for cyclin D kinase activity. The reduction in cyclin D expression by ansamycins, nevertheless, appears to be one of their major modes of action. In previous work, we showed that the half-life of D-type cyclins is not affected by ansamycins, and therefore these proteins are unlikely to be direct targets of the drug(52). Growth factor-stimulated translation of D-type cyclins has been shown to occur via a phosphatidylinositol-3-kinase/Akt pathway (52). Ansamycins inhibit this pathway(52), in part by degrading upstream transmembrane tyrosine kinases, and likely down-regulate D-type cyclin expression in this manner.

An ansamycin, 17-allylaminogeldanamycin, is currently in Phase I clinical trials in patients with advanced cancer. This drug may be most useful in cancers dependent on protein targets that are especially sensitive to its action, such as the HER2 tyrosine kinase in breast cancer and androgen receptor in prostate cancer. The data presented here suggest that a primary consequence of degradation of the targets is G₁ arrest associated with decreased cyclin D expression and hypophosphorylation of RB. These may prove to be useful markers of drug effectiveness in patients whose tumors are accessible for analysis. Furthermore, G₁ progression is unaffected by drug in tumors that lack wild-type RB. Instead, these cells accumulate in the G₂-M phase of the cell cycle and undergo apoptosis. The mechanism of this phenomenon is under investigation, but it seems likely that the clinical response of tumors with wild-type RB will be fundamentally different from those that lack this protein.

We thank Dr. Andrew Koff for helpful discussions, Diane Domingo for technical assistance with flow cytometry, and members of the Rosen lab for critical reading of the manuscript.