HSP90 Inhibitor 17-AAG Selectively Eradicates Lymphoma Stem Cells

Molecular chaperone proteins function to ensure the proper conformation of client proteins when cells experience stress or damage (1). Heat shock protein 90 (HSP90) is the most studied and well-known molecular chaperone that facilitates the maturation and stable conformation of several client proteins, including transcription factors hypoxia-inducible factor 1α (HIF1α) and p53, serine/threonine kinases (AKT, Raf-1, and Cdk4), receptor/non-receptor kinases (HER2, EGFR, Src family kinases), and steroid hormone receptors (androgen and estrogen; refs. 2–10). Because many of these client proteins significantly contribute to tumor growth and survival, abrogation of their function with a single inhibitor has been an attractive prospect, making HSP90 an attractive molecular target for drug discovery (11). One of the original and most studied HSP90 inhibitors is a derivative of the geldanamycin antibiotic, 17-N-allylamino-17-demethoxy geldanamycin (17-AAG; ref. 12). Through reversible binding to the ATP pocket of HSP90, 17-AAG potently disrupts its function, and ultimately induces tumor cell death.

Tumor stem cells from both glioma and acute myeloid leukemia (AML) have been shown to rely on the activity of HIF1α or 2α, respectively, for their maintenance (13–15). In particular, human AML CSCs are a rare population of CD34⁺CD38⁻ cells, and are thought to be responsible for the resistance of conventional therapies (13, 16, 17). Current efforts for targeting CSCs have focused on disruption of self-renewal (18). However, this approach may be hampered by the quiescent nature of CSCs. Therefore, disrupting genes that are required for both maintaining CSCs in a stem-like state as well as self-renewal might provide a more effective therapy. We have previously reported a strain of T-cell receptor transgenic mice (TGB), which spontaneously developed lymphoma with 100% incidence, due to an insertional mutation of the Emp2a gene (19). Using this mouse lymphoma model, we identified that a small subset of cells expressing both c-Kit and Sca1 are lymphoma CSCs, and represent HIF1α is a target for lymphoma CSC therapy (14).

17-AAG has been shown to target both the bulk population of cells and CSCs present within glioma and AML (13–15, 20). In this regard, we hypothesized that 17-AAG may be more suitable for targeting both lymphoma CSCs and bulk lymphoma cells simultaneously through the disruption of multiple HSP90 client proteins. Unfortunately, acquired resistance to 17-AAG has been observed in both glioblastoma and melanoma through reduced expression of NAD(P)H/quinone oxidoreductase 1 (NQO1), an enzyme responsible for activation of 17-AAG. In addition, induction of stress-response proteins, such as HSP27, and HSP70 by 17-AAG plays a large role in resistant cancer cells (21, 22). This induction relies on the transcription factor heat shock factor 1 (HSF1), which regulates expression of HSP70, HSP27, and HSP90 (23–26). Thus, targeting HSF1 in order to abrogate 17-AAG-induced heat shock response has been proposed as an anticancer strategy (27). This concept is supported by reports showing that HSF1 is required for the initiation of both lymphoma and RAS oncogene–induced tumors (28, 29).

In our study, we showed that low concentrations of 17-AAG preferentially eliminate the CSCs of both lymphoma and AML, but leaves the differentiated cancer cells unaffected due to their high levels of HSF1 expression. Knockdown of HSF1 abrogates the colony formation ability of lymphoma and AML CSCs through disrupting HSP90α-mediated HIF1α stability. In addition, this provides a new anti-CSC strategy by targeting HSF1 to eliminate aggressive malignancies.

TGB mice in the B10.BR background were maintained by heterozygous breeding pairs. B10.BR mice were used as recipients in lymphoma cell inoculation, and severe combined immunodeficiency (SCID) mice were used for human AML maintenance. Both lines were purchased from the National Cancer Institute. All mice were kept and used for the tumor implantation and treatment according to the procedures approved by the Unit of Laboratory Animal Facility (ULAM) at University of Michigan.

TGB lymphoma cells were maintained in 1% methylcellulose medium (MethoCult GF, 03434, Stem Cell Technologies). For some in vitro experiments, TGB lymphoma cells were transferred from 1% methylcellulose culture to an expansion medium: RPMI1640 medium supplemented with 10% FBS, interleukin (IL)-3, IL-6, and stem cell factor cytokines (R&D Systems), 100 μmol/L non-essential amino acids, 10 mmol/L HEPES, and 1 mmol/L sodium pyruvate. Human AML cells were derived from the enlarged spleen tissue of SCID mice injected with aggressive AML CSCs (AML-71) as described previously (14), and maintained in SCID mice via transplantation. Short-term expansion and colony formation assays of the AML CSCs were carried out similarly as for our lymphoma cells, except for a substitution of human cytokines for murine cytokines. Clinical subtypes and genetic characterization of AML cells by examining the appearance of the malignant cells and by using cytogenetics to characterize any underlying chromosomal abnormalities were authenticated by the University of Michigan Comprehensive Cancer Center.

17-AAG was purchased from LC Laboratories, The sources of rabbit polyclonal antibodies to the specific proteins are as follows: HSF1 (4356), HSP90 (4874), HSP70 (4874), phosphor-S21/S9 GSK3α/β (9331), phosphor-S473AKT (9271), and AKT (9272) were purchased from Cell Signaling. Mouse antibodies to GSK3α/β (0011-A; Santa Cruz) and β-actin (AC-15; Sigma- Aldrich) were used in the studies.

The lymphoma CSCs (c-Kit⁺Sca1⁺) are a rare population in TGB lymphoma tissue and express the surface markers c-Kit and Sca1. The CSCs were sorted out from primary spleen lymphoma tissue of TGB mice by fluorescence-activated cell sorting (FACS) technology, or isolated by auto-MACS microbeads (Miltenyi Biotech) twice, after staining with antibodies to c-Kit and Sca1 (both from Becton Dickinson Biosciences). For the detection of some downstream proteins requiring larger amounts of cells, the c-Kit⁺ fraction (c-Kit⁺Sca1⁺ subset combined with c-Kit⁺ Sca1⁻ subset) were used. For colony formation assays, lymphoma cells were seeded in 1% methylcellulose medium containing these murine cytokines in a 12-well plate. After 6 to 8 days of culture, colonies were counted, pooled, and then replated for consecutive rounds. For tumor inoculation and viral infection, the cultured lymphoma cells were expanded in the RPMI1640 medium as described above, 24 hours before these experiments.

The lentiviruses for short-hairpin RNA (shRNA) targeting HSF1 (Sh-HSF1) and scrambled control Sh-RNA were generated in HEK293FT cells in accordance with the lentiviral preparation protocol (Invitrogen). Lentiviruses were prepared in a 150 mm falcon culture dish, and 50 mL of media was collected and centrifuged at 550g for 10 minutes to remove cell debris. Viral particles obtained after centrifugation at 12,857g overnight were suspended with 2 mL of RPMI1640 medium containing 10% FBS, then aliquoted and stored at −80°C. Between 50 and 100 μL of concentrated virus per million cells was mixed with 100 μL culture medium and used for spinoculation in a U-bottom tube in a mixture containing 9 μg/mL polybrene. The spinoculation was carried out at 2,500 rpm, for 2 hours at 30°C. After spinoculation, the cell pellet was suspended gently and completely with fresh culture medium, seeded and cultured in 1% methylcellulose for another 5 to 7 days before use.

The sequence of Sh-HSF1 (5-GCTCATTCAGTTCCTGATC-3) has been previously validated (30), and was cloned into a Sh-lentiviral vector as described previously (14). The Sh-HSF1 sequence is shared by mouse and human. The constructs of hypoxia responsive element (HRE) or its mutant (HRE-Mt) EGFP reporter specific for HIF1α were described previously (14). The full-length coding region of Hsf1 cDNA was amplified by real time-PCR using mouse bone marrow mRNA as template, and cloned into a pcDNA vector with Flag tag fused into the C-terminus of Hsf1.

In general, sorted CSCs or cultured lymphoma cells were lysed with 1% Triton X-100 lysis buffer containing 20 mmol/L Tris-HCl, pH 7.4, 150 mmol/L NaCl, 40 mmol/L NaF, 1 mmol/L dithiothreitol, protease, and phosphatase inhibitor cocktail (Sigma). Supernatants of the lysate were used for Western blotting and resolved on a reducing and non-heated (95°C for 3 minutes) denaturing 10% SDS PAGE gel. Transfer membrane was blotted with primary antibody overnight at 4°C with shaking, and was linked with secondary antibody at room temperature for 5 hours.

The expanded lymphoma cells were seeded in the RPMI1640 culture medium at a cellular density of 2.5 × 10⁵ cells/mL in a 24-well plate. Next, the cells were treated with increasing concentrations of 17-AAG and cultured for 48 hours. At the end of the treatment, the MTT assay for cell viability determination was conducted according to the manufacture's protocol (Promega) and the plate was read at 490 nm.

Total RNAs from FACS sorted lymphoma CSCs and non-CSCs (c-Kit⁻Sca1⁻) were extracted using the RNeasy extraction kit (Qiagen). After reverse transcription of RNA to cDNA, RT-PCR using the converted cDNA as template was carried out in triplicate using SYBR Green PCR Master Mix. Following primers were used: Hsf1, F-TTTGACCAGGGCCAGTTTG, R-TTCGGAAGCCATACATGTTGAG; Hif1α,F-AGTCTAGAGATGCAGCAAGATCTC, R-TCATATCGAGGCTGTGTCGACTGA; Hsp90α, F-TGTTGCGGTACTACACATCTGC, R-GTCCTTGGTCTCACCTGTGATA; Hsp90β, F-CAAACAAGGAGATTTTCCTCCG, R-GCTGTCCAACTTAGAAGGGTC; Glut1, F-TGTGCTGTGCTCATGACCATC, R-ACGAGGAGCACCGTGAAGAT; β-actin, F-GGCTGTATTCCCCTCCATCG, R-CCAGTTGGTAACAATGCCATGT. Fold differences was calculated using the delta Ct method (31) with β-actin as a reference gene.

The cultured lymphoma and AML cells, after treatment with 17-AAG for 24 hours, were stained with surface markers of fluorescence-conjugated c-Kit and Sca1 (CD34 and CD38 for AML cells), and then Annexin-V for 15 minutes at room temperature before addition of 5 μmol/L DAPI. The apoptosis percentage of these CSCs and differentiated cells were then analyzed by flow cytometry.

B10.BR mice were inoculated with 5 × 10⁵ lymphoma cells through intraperitoneal injection. Seven days after tumor implantation, the mice were intraperitoneally (i.p.) injected with 17-AAG or vehicle [10% DMSO + 40% Cremophor EL:Ethanol (3:1; v/v) + 50% PBS] every other day for 3 weeks. At the cessation of treatment, mice were monitored up to 80 days after tumor cell injection. To determine the effects of 17-AAG on lymphoma initiation in vivo, secondary B10.BR recipient mice were implanted by intraperitoneal injection of 1 × 10⁵ lymphoma cells from the spleens of first round mice that had been treated with 17-AAG or vehicle. These mice were followed up until 160 days post tumor cell injection to monitor differences in tumor initiation between the mice.

We have previously shown that AKT-GSK3-β-catenin signaling is constitutively activated in the TGB lymphoma c-Kit⁻Sca1⁻ cells (non-CSCs), and that HIF1α plays a critical role in the maintenance of the lymphoma c-Kit⁺Sca1⁺ cells (CSCs; refs. 14, 19). These lymphoma CSCs can efficiently initiate lymphoma when small numbers are injected into immunocompetent recipient mice. Following tumor initiation, the mice ultimately succumb to the committed, highly proliferative lymphoma cells, which rapidly grow into a tumor mass in the spleen and metastasize into other organs, such as the liver and lung, leading to death within 1 to 2 months post-inoculation (14, 19). These findings prompted us to devise a drug therapy that could effectively target both CSCs and non-CSCs, for the treatment of advanced lymphoma in humans. Because the CSCs and non-CSCs harbor constitutively active HIF1α and AKT (2 client proteins of HSP90), respectively, we tested the HSP90 inhibitor 17-AAG for the simultaneous targeting of the 2 cell populations.

We first examined the expression levels of chaperones and the transcriptional factor HSF1 in the CSCs and non-CSCs. Using a FACSAria cell sorter (Becton Dickinson), 2 subsets of TGB lymphoma cells (c-Kit⁺Sca1⁺, c-Kit⁻Sca1⁻) were sorted from mouse spleen tissue enlarged with the TGB lymphoma cells (Fig. 1A). As previously described (14), the sorted c-Kit⁺Sca1⁺ cells initiated lymphoma with 50-fold lesser number of cells than the sorted c-Kit⁻Sca1⁻subset (Fig. 1B). Quantitative RT-PCR (qRT-PCR) showed that the mRNA levels of HIF1α and its target Glut1 were 4- and 8-fold higher, respectively, in the c-Kit⁺Sca1⁺ cells than the c-Kit⁻Sca1⁻ cells. This confirmed the hypothesis that the c-Kit⁺Sca1⁺ subset still maintained the unique marker of active HIF1α. The mRNA levels of chaperones HSP90α, HSP90β, and HSP70, and HSF1 were approximately 1-fold higher in the c-Kit⁻Sca1⁻ than in the c-Kit⁺Sca1⁺ cells (Fig. 1C). Consistent with mRNA levels, protein levels of HSP70 were higher in the c-Kit⁻Sca1⁻ cells. Strikingly, there were large differences in the protein levels of HSF1, HIF1α, and the activated status of the AKT-GSK3 pathway between the 2 cell populations (Fig. 1D and E). HSF1 had more than 10-fold higher expression in the c-Kit⁻Sca1⁻ population, compared with the c-Kit⁺ population. Consistent with our previous results (14, 19), stable HIF1α was exclusively expressed in the CSCs, while active AKT (pS473AKT) and GSK3 (pS21/9GSK3α/β) were present in higher amounts in the non-CSCs than CSCs. These distinctive expression patterns for stable HIF1α and the active AKT-GSK3 pathway supports our initial hypothesis for targeting CSCs and non-CSCs simultaneously with 17-AAG.

We next examined the ability of 17-AAG to suppress TGB lymphoma growth using the MTT cell viability assay. TGB lymphoma viability was inhibited with 17-AAG, with an IC₅₀ of 238 nmol/L (Fig. 2A). In contrast, c-Kit⁺Sca1⁺ lymphoma cells treated with 17-AAG were significantly more sensitive to HSP90 inhibition, with an IC₅₀ of 5.64 nmol/L (Fig. 2B–D). These substantial differences in the IC₅₀ indicate a profound sensitivity to 17-AAG within the CSC population.

Previous studies have shown that treatment with 17-AAG results in programmed cell death via apoptosis (32). To determine whether 17-AAG induced apoptosis in the CSCs and bulk tumor cells, we stained lymphoma cells with Annexin V, a surface marker of apoptotic cells, after 17-AAG treatment. As shown in Fig. 3A, low doses of 17-AAG (10 and 25 nmol/L) increased apoptosis of c-Kit⁺Sca1⁺ cells by 2- to 3-fold, as compared with the untreated control. However, 17-AAG failed to induce apoptosis of c-Kit⁻Sca1⁻ cells at these low concentrations.

As CSCs are best defined by their ability to self-renew and initiate tumor growth rather than by their surface marker expression, functional assays provide a more accurate assessment of drug-mediated therapeutic effects (33). The generation of spheroid colonies has been widely used as a manner to assay the presence of CSCs (34–36). Formation of these colonies is dependent on the residing stem cell population, and is reduced with each stage of differentiation. To assess whether 17-AAG affected CSC self-renewal, we plated lymphoma cells and treated them with increasing concentrations of 17-AAG. As a result, 10 nmol/L 17-AAG reduced the colony formation unit (CFU) of CSCs by more than 50% (Fig. 3B), consistent with the dose necessary for induction of apoptosis. These colonies were then replated in the absence of drug, to observe whether self-renewal was also disrupted. Cells from colonies previously treated with 10 nmol/L 17-AAG were dramatically reduced in their capacity to initiate secondary colonies, indicating that 17-AAG eliminated the self-renewal ability of the CSCs in vitro.

In order to observe whether 17-AAG interfered with HIF1α's transcriptional activity, we constructed a lentiviral reporter system that used HIF1α binding to specific hypoxia-inducible responsive elements (HRE) to control EGFP reporter expression (14). Lymphoma cells were infected with either this reporter, or with a reporter containing a mutant HRE (HRE-Mt) sequences (as a negative control for gating). Without treatment, 25.8% of HRE-EGFP positive cells were present in the c-Kit⁺Sca1⁺ subset, whereas only 0.786% were in the c-Kit⁻Sca1⁻ subset (Fig. 3C). This is consistent with the specific expression of stable HIF1α in the CSCs. With treatment, 10 nmol/L 17-AAG reduced HRE-EGFP expression in the CSCs by 2-fold, but had little effect on the HRE-EGFP expression in non-CSCs, showing that 17-AAG specifically suppressed HIF1α activity in the CSCs. In addition to the reduction of HIF1α transcriptional activity induced by 17-AAG, the mRNA levels of HIF1α and its target Glut1 were significantly decreased in a dose-dependent manner (Fig. 3D), indicating that HIF1α and its target transcription are controlled by HIF1α stability. These results show that 17-AAG specifically eliminates the CSCs through inhibition of the HSP90-mediated mRNA expression and transcriptional activity of HIF1α at concentrations, which fail to eliminate the more differentiated c-Kit⁻Sca1⁻ lymphoma cells.

To investigate whether HSP90 inhibition can reduce lymphoma initiation in vivo, we injected 5 × 10⁵ TGB lymphoma cells i.p. into recipient B10.BR mice. Seven days after injection, different doses of 17-AAG (5–40 mg/kg) or vehicle control were given to mice every other day for 3 weeks. Upon cessation of treatment, 4 mice from the 40 mg/kg 17-AAG or control group were euthanized and their spleens weighed for indication of lymphoma. Vehicle-treated mice exhibited enlarged spleens enriched in lymphoma cells that stained positive for Thy-1, a marker for T lymphoma cells (Fig. 4A–C). In addition, the lymphoma cells had metastasized into liver and lung tissue, as typically seen with this cancer (19). In contrast, spleens from mice that had received 17-AAG were dramatically smaller, with less infiltrating lymphoma cells in the spleen, and a lower metastatic spread into other organs, compared with the vehicle-treated control (Fig. 4C). In addition, 17-AAG–treated mice survived for significantly longer compared with mice that had received vehicle alone, showing the effectiveness of 17-AAG in suppressing the lymphoma in vivo (Fig. 4D).

A hallmark of CSC function is their ability to serially reestablish tumors in recipient mice. This CSC trait reflects their capacity for self-renewal in vivo, and provides a method for detecting chemical agents that can disrupt this process. To examine whether 17-AAG treatment impaired the self-renewal ability of TGB lymphoma CSCs in vivo, 1 × 10⁵ cells obtained from either 40 mg/kg 17-AAG- or vehicle-treated spleens were i.p. injected into 5 healthy secondary recipient B10.BR mice. All of the mice injected with vehicle-treated spleen cells eventually succumbed to lymphoma. In contrast, only 1 mouse that received 17-AAG–treated spleen cells died over the course of the study (Fig. 4E). These data show that 17-AAG effectively impairs the self-renewal of the lymphoma CSCs in vivo.

To understand the mechanism underlying 17-AAG resistance in the non-CSCs, we investigated the effect of silencing HSF1 on the growth of CSCs and non-CSCs, as HSF1 was more highly expressed in the differentiated population. When we treated the lymphoma cells with 50 nmol/L 17-AAG, a dose capable of killing nearly all of the CSCs, we found that the bulk differentiated cells were largely unaffected, as protein levels of pS473AKT, pS21GSK3α, and pS9GSK3β remained unchanged (Fig. 5A). Coinciding with reports in the literature, we also observed an increase in HSP70 in reaction to HSP90 inhibition with 100 nmol/L 17-AAG. However, 100 nmol/L 17-AAG could reduce the levels of these phosphorylated proteins in addition to reducing HSP90 and HSF1. This dramatic loss in cell viability was, in part, due to drug toxicity, as all of our controls (AKT, GSK3α, GSK3β, and β-actin) were sharply decreased and cell death was observed. This result indicates that 17-AAG does not affect HSP90-mediated AKT activation if HSF1 protein levels are very high, because high levels of HSF1 could maintain or further increase the transcription of HSP90 or HSP70 to compensate for HSP90 inhibition.

To observe the effect of HSF1 on the growth of CSCs and non-CSCs, we made a hairpin RNA (shRNA) targeting HSF1. In HEK293 cells, the Sh-HSF1 silencer specifically knocked down ectopically expressed HSF1, but not Lef1, by up to 80% within 24 hours after co-transfection (Fig. 5B). Nearly half (48.1%) of the TGB lymphoma cells were infected after transfection with the Sh-HSF1 lentivirus, as determined by the expression of an EGFP reporter within the shRNA construct (Fig. 5C). Quantitative RT-PCR revealed that Sh-HSF1 knocked down approximately 80% of the endogenous HSF1 mRNA in sorted EGFP-positive cells, relative to the mRNA levels of sorted cells expressing scrambled shRNA (Sh-Sr) (Fig. 5C and D). With the knockdown of HSF1, the mRNA of one of its target genes, HSP90α (37), was correspondingly reduced. The HSF1 silencer significantly inhibited cell growth and eliminated the colony formation capacity of the CSCs without the need of 17-AAG (Fig. 5E), showing that HSF1 is positioned upstream of HSP90 and can control the level of chaperone activity. This effect was not seen with the Sh-Sr, as it failed to affect colony formation of the CSCs. This result shows that low levels of HSF1 are required for maintenance of the CSCs.

In order to determine whether 17-AAG treatment could selectively eliminate hematopoietic CSCs other than our lymphoma CSCs, we used a mouse model of human AML that has been previously established (14). Similar to our lymphoma cells, AML CSCs stably express HIF1α for their maintenance (14). To see whether 17-AAG also selectively targeted AML CSCs, AML cells were treated with increasing concentrations of 17-AAG. A dose of 25 nmol/L 17-AAG increased the percentage of apoptotic CSCs (CD34⁺CD38⁻) nearly 3-fold, compared with vehicle-treated CSCs. In contrast, 25 nmol/L 17-AAG had little effect on the induction of apoptosis in the non-CSCs (CD34⁺CD38⁺; Fig. 6A and B). Although a higher dose (100 nmol/L) of 17-AAG induced apoptosis in both CSCs and non-CSCs, CSCs initiated apoptosis to a higher extent (35.5%) than non-CSCs (13.5%; Fig. 6A and B). Similar to lymphoma CSCs, a dramatically low dose (10 nmol/L) of 17-AAG greatly reduced the capacity for AML CSCs to form colonies (Fig. 6C). This colony reduction could be attributable to the inhibition of HSP90-mediated HIF1α stability by 17-AAG, as the mRNA levels of HIF1α were correspondingly reduced by 17-AAG treatment (Fig. 6D). Compared with normal mouse bone marrow (BM) cells, the AML cells also highly expressed HSF1, phosphorylated AKT, and GSK3, in a similar manner as our TGB lymphoma cells, indicating that AML cells possibly respond to 17-AAG treatment via a similar mechanism as the lymphoma cells (Fig. 6E). Accordingly, knockdown of HSF1 by Sh-HSF1 silencer abrogated the colony formation capacity of the AML CSCs, as compared with the Sh-Sr (Fig. 6F). Finally, to verify that the reduction in colony formation in AML CSCs after knockdown of HSF1 was through a reduction in HSP90, we sorted the lentivirus-infected AML cells by gating on EGFP-positive cells (Fig. 6G) and determined the mRNA levels of HSF1, as well as its target gene HSP90α. Knockdown of HSF1 dramatically reduced the mRNA levels of HSP90α (Fig. 6H), verifying the requirement for functional HSP90α and HSF1 in the maintenance of the AML CSCs. Taken together, these results show that low doses of 17-AAG selectively eliminate AML CSCs, possibly through the loss of HSP90-mediated HIF1α activity. In addition our results show that a functional HSF1 is required for AML CSC maintenance.

T-Cell lymphoma (represented by our TGB mouse model) and AML are fast-growing aggressive tumors that are difficult to treat, and have a high relapse rate for patients in remission. Therefore, discovery of new therapies capable of eliminating CSCs are needed to block tumor recurrence. In this study, both CSCs derived from lymphoma and AML were eliminated through 17-AAG mediated HSP90 inhibition. However, the bulk population of lymphoma cells, which highly express HSF1, are resistant to 17-AAG treatment. As summarized in our model (Fig. 7), low expression of HSF1 in CSCs renders them more susceptible to 17-AAG, as at physiological conditions, the low amount of HSF1 likely fails to induce HSP90α transcription in a sufficiently fast manner to compensate for the loss of HSP90α. Thus, 17-AAG-mediated HSP90 inhibition abrogates the function of its client HIF1α in CSC. In contrast, high amounts of HSF1 in non-CSCs confers greater resistance to 17-AAG, as the high levels of HSF1 protein compensate for HSP90 inhibition by stimulating HSP90α expression at physiological conditions. Thus, the knockdown of HSF1 eliminates the HSF1–HSP90α axis, which supports the activation of AKT-GSK3 pathway that is essential for non-CSC proliferation.

The chaperone HSP90 is required for the proper folding of its client proteins that play a critical role in cancer proliferation and survival, such as HIF1α, AKT, and HER2 (3, 4, 38–40). Because HSP90 expression is controlled by the homotrimer of HSF1 and/or the heterotrimer of HSF1 and HSF2 (30, 41, 42), successful HSP90 inhibition by 17-AAG naturally depends on the levels and activation of HSF1. Deletion of HSF1 in chicken B lymphocyte revealed that HSP90α expression is inducible and controlled by HSF1 (37), which is consistent with our results showing knockdown of HSF1 reduced the levels of HSP90α. HSP90β is constitutively expressed and does not respond to the HSF1 deletion (37). Although both of the chaperone isoforms are capable of binding their client proteins (43), whether one plays more of an essential role in their client's function over the other remains undefined. In terms of our CSC models, we predicate that HSP90α plays a critical role in the maintenance of HIF1α function.

Low doses of 17-AAG inhibit the maintenance and self-renewal of lymphoma CSCs through blockade of HIF1α activity, as evidenced by the loss of HIF1α transcription activity and the reduced expression of HIF1α and its target gene Glut1. In addition, this mechanism could be effective in AML CSCs, as HIF1α expression was reduced by 17-AAG treatment, and the expression patterns for the active AKT-GSK3 pathway, HSF1, and HIF1α were similar to that in our TGB lymphoma cells (14). Further, loss of HIF1α activity following 17-AAG treatment induced apoptosis at concentrations that were more than 20-fold lower than the IC₅₀ required for the elimination of the differentiated non-CSCs. This further illustrates the dependence of lymphoma CSCs on HIF1α for survival. In addition, significantly lower concentrations of 17-AAG were able to largely eliminate colony formation capacity of both lymphoma CSCs and AML CSCs, indicating an impairment of self-renewal in vitro. Although mice treated with 17-AAG had prolonged survival, a few mice still succumbed to lymphoma. This is not surprising, given that high expression of HSF1 in the rapidly proliferative tumor bulk can provide resistance to 17-AAG, showing the need for combination of conventional therapies with CSC-targeting agents. However, consistent with this self-renewal blockage in vitro, residual lymphoma cells present in spleens from 17-AAG–treated mice had a reduced ability for secondary tumor initiation, illustrating a direct inhibitory effect of 17-AAG on the lymphoma CSC's self-renewal in vivo.

The selective elimination of mouse lymphoma CSCs and human AML CSCs by low doses of 17-AAG provided validation for its potential therapeutic role in targeting CSCs. 17-AAG has been used to treat various types of hematological malignancies in clinical phase I trials. However, only partial responses were seen with 17-AAG therapy (44). In addition, the high doses of 17-AAG used in advanced solid tumors caused serious adverse events, including intestinal and liver toxicity. These reports emphasize the limitations seen with using high doses of 17-AAG (45).

We revealed that the high expression of HSF1 in non-CSCs of both lymphoma and AML helps promote resistance to 17-AAG and maintain their aggressive proliferation. Similarly, the low expression of HSF1 in TGB lymphoma CSCs, and possibly also in AML CSCs, is essential for CSC maintenance through regulation of HSP90α and their client proteins, such as HIF1α. Therefore, targeting HSF1 may also efficiently eradicate advanced lymphoma and AML, in addition to potentially reducing recurrence. These results are supported by reports showing that genetic deletion of HSF1 prevented initiation of oncogene-induced cancers and tumor suppressor p53 null-mediated lymphoma (28, 29). Efforts to develop HSF1 inhibitors are ongoing (46). Combined with a HSF1 inhibitor, 17-AAG may be a powerful therapy for eradicating hematological cancers via a broad inhibition of transcriptional factors that are required for the proliferation of differentiated cancer cells, and the maintenance of CSCs. In addition, the low dose of 17-AAG used for the elimination of these CSCs shows that it alone has the potential to prevent relapse or, in combination with conventional therapy, to treat relapsed or refractory hematological cancers. In addition, using a low dose of 17-AAG to target CSCs may lessen toxicity arising from off target effects, as many of the signaling pathways used by CSCs are shared with their normal counterparts (47). Finally, our work, together with reports from glioma (20) and chronic myeloid leukemia (48), further extend the strategy of HSP90 inhibition as a means for eliminating tumorigenic stem cells.

In summary, our studies show that low doses of 17-AAG selectively target lymphoma CSCs through HSP90α-mediated HIF1α inhibition. This selectivity is likely attributable to the expression status of HSF1 in the CSCs and non-CSCs subsets. Knockdown of HSF1 abrogates the activity of the 2 subsets, providing important considerations for future drug development in patients with malignant lymphoma/leukemia. The regulatory mechanisms among HSF1, HSP90α, and HIF1α that are required for the maintenance of the highly proliferative state of differentiated cancer cells and the quiescent state of CSCs is worthy of further investigation, for future development of cancer therapy.

No potential conflicts of interest were disclosed.

The authors thank Dr. Sami N. Malek, Internal Medicine, for assistance with the AML mouse model and Linhua Zou, General Surgery, for assistance with FACS. The authors also thank Sarah J. Conley, Internal Medicine, for assistance with the editing of the manuscript.

This work was financially supported by the NIH grants R21NS062391, 5R21CA164469, R01-CA120023, and R21-CA143474. Additional support was provided by a grant GM007767 from the NIGMS.

The contents of this article are solely the responsibility of the authors and do not necessarily represent the official views of the NIGMS.

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.