HSF1 Is Essential for Myeloma Cell Survival and A Promising Therapeutic Target

Myeloma is a plasma cell neoplasm of the bone marrow that is characterized by the sustained secretion of large amounts of immunoglobulin (Ig). Myeloma cells therefore rely on intracellular protein homoeostasis mechanisms for survival (1). This is highlighted by the clinical success of proteasome inhibitors that disable the ubiquitin–proteasome pathway (2). Interestingly, one of the mechanisms by which bortezomib induces apoptosis is through the activation and overloading of a different protein homoeostasis pathway, the unfolded protein response (UPR; refs. 3–5). It can therefore be envisaged that myeloma would be susceptible to therapies targeting various protein handling mechanisms (6–7). Indeed, there is current interest in the inhibition of these types of targets for myeloma, such as IRE1α in the UPR (8).

Another avenue for inducing proteotoxic stress is to target the heatshock response (HSR) and molecular chaperones. Although tanespimycin (17-AAG), a HSP90 inhibitor, showed convincing preclinical antimyeloma efficacy, only modest responses were observed in clinical trials (9–10). This is attributable to the upregulation of antiapoptotic HSP 70 protein 1 (HSP72) and other HSPs following tanespimycin treatment by heatshock factor 1 (HSF1), the master transcription factor and regulator of the HSR (11–14). A similar induction of HSPs and HSF1 activation is also observed following bortezomib treatment, suggesting that activation of the HSR limits the efficacy of a number of current myeloma drugs (6, 15).

In addition to regulating HSP expression, HSF1 is indispensable for oncogenic transformation and cancer cell survival (16, 17). Additional studies have implicated HSF1 in other aspects of tumor progression such as cell migration and angiogenesis (18–20). Furthermore, higher levels of nuclear HSF1 are found in breast, kidney, and oral cancers compared with normal tissues, with links to poor patient prognosis and increased metastasis (19, 21–22). Extensive ChIP-seq experiments in breast cancer revealed that HSF1 promotes oncogenesis and cell survival through regulating a cancer-specific transcriptional program (23). Not only is HSF1 strongly bound to its target genes in breast and colon tumors with high nuclear HSF1 levels, but the cancer-specific HSF1 gene expression profile is also a prognostic indicator for poor outcome (23).

With accumulating evidence for HSF1 and its transcriptional program in maintaining the malignant phenotype as well as its role in protein homeostasis, inhibition of HSF1 looks to be a promising therapeutic strategy (24). Although several inhibitors of HSF1 have been reported, many of these have major limitations and none have progressed to the clinic (25). Our own drug discovery efforts have resulted in a highly potent and selective inhibitor of the HSF1 stress pathway, CCT251236, which shows therapeutic activity at well-tolerated doses in human ovarian cancer xenografts (26). In a similar fashion to our own phenotypic screen (26), Yoon and colleagues (27) have also identified an inhibitor of the HSF1 pathway, KRIBB11, which is well-tolerated in vivo. Both these compounds facilitate the in vitro and preclinical exploration of targeting HSF1-mediated transcription for myeloma therapy.

The role of the HSF1 pathway in hematologic cancers is relatively unexplored compared with solid cancers. Given the evidence for HSF1 in mediating protein homeostasis and oncogenesis, we hypothesized that it may be a good myeloma therapeutic target. Here, we describe the prognostic significance of HSF1 expression and demonstrate that shRNA-mediated knockdown of HSF1 in human myeloma cell lines (HMCL) leads to a downregulation of global protein synthesis, activation of the UPR and caspase-mediated cell death. Using CCT251236 and KRIBB11 as tool compounds, we show antimyeloma activity in a human myeloma xenograft model and a potential therapeutic window for HSF1 pathway inhibition using primary patient-derived myeloma cells and peripheral blood mononuclear cells (PBMC).

Expression data from CD138⁺ plasma cells (n = 262), collected from relapsed patients enrolled in APEX, SUMMIT and CREST trials were examined (GEO accession GSE9782; ref. 28). Data for newly diagnosed patients were obtained from the following clinical trials: Myeloma IX (n = 258; GSE21349), Total Therapy 2/3 (n = 559; GSE2658) and HOVON/GMMG-HD4 (n = 320; GSE19784). Patients were separated into high and low HSF1 expression (Affymetrix probeset 202244_at) using the partitioning around medoids algorithm in R. Kaplan–Meier overall survival (OS) curves were generated and log-rank tests carried out using the R survival package. HR and 95% confidence intervals (CI) were computed using a univariate Cox proportional hazards model in SPSS (IBM). To evaluate the impact of HSF1 target gene expression on OS, expression of genes in the cancer-specific HSF1 signature (23) was analyzed (456 genes; 793 probesets). Probesets with <5 samples with expression values >200 were removed. Data were min–max normalized and probeset intensities with low variance (<200) were discarded. Hierarchical clustering was performed on the remaining 359 probesets (Ward method) and heatmaps were generated using the R hclust and gplots packages. Further analysis was performed on RNA-seq data from CD138⁺ plasma cells from the MMRF CoMMpass trial (NCT01454297).

RPMI-8226, NCI-H929, U266, HEK293T/17, and HS-5 were purchased from the ATCC. KMS-11 and MOLP-8 were a kind gift from Professor H. Johnsen (Aarhus University Hospital, Denmark). GFP-tagged bone marrow stromal cells, HS-5-GFP, were generated as previously described (29). HEK293T/17 and HS-5-GFP cells were cultured in DMEM containing GlutaMAX and 10% FBS (Life Technologies). All other cells were cultured in RPMI1640 containing GlutaMAX and 10% FBS. All cells tested negative for Mycoplasma by PCR, were authenticated by STR analysis and were used two passages after thawing.

PBMCs were obtained from healthy donors. Patient primary myeloma cells were isolated from bone marrow aspirates by density gradient centrifugation using Ficoll-Paque Premium (GE Healthcare) according to the manufacturer's instructions. CD138⁺ cells were purified using CD138 Microbeads (Miltenyi Biotec) to a purity of >95% and used immediately. All procedures were performed following informed consent. Approval for these studies was obtained from the Royal Marsden Hospital Review Board (CCR4238) and the Health Research Authority National Research Ethics Service Committee (14/YH/1317).

CCT251236 was synthesized as described previously (26). Compounds were purchased as described: KRIBB11 (Merck Millipore), bortezomib (Cambridge Bioscience), Z-VAD-FMK and puromycin (InvivoGen), pactamycin, and tunicamycin (Sigma). Plasmids were purchased or obtained as described: HSF1 shRNA pLKO.1 plasmids (Thermo Fisher Scientific; TRCN0000007480, TRCN0000007484), pLKO.1 empty vector, pLKO.1 GFP shRNA, pCMV-R8.72 lentiviral packaging and pCMV-VSV-G envelope plasmid (Addgene; plasmid ID 10878, 30323, 22036 and 8454).

Plasmids were propagated in bacterial cultures and purified using the PureLink HiPure Plasmid Filter Maxiprep Kit. A total of 2 × 10⁷ HEK293T/17 cells were transfected using the calcium phosphate method with a mixture of 16 μg pCMV-R8.74, 5 μg pCMV-VSV-G, and 20 μg of pLKO.1 empty vector, pLKO.1 GFP shRNA or HSF1 shRNA pLKO.1 and 125 mmol/L CaCl₂ (Sigma) in 500 μL nuclease-free water and 500 μL HEPES-buffered saline pH 7.05. Conditioned medium was collected at 48 to 72 hours post-transduction and concentrated using the Lenti-X Concentrator Kit (Clontech) according to the manufacturer's instructions. Up to 1 mL of concentrated virus-containing media and 8 μg/mL polybrene (Sigma) were used for the transduction of 5 × 10⁶ cells.

RNA was recovered from HMCLs using the RNeasy Mini Kit (Qiagen) and cDNA was synthesized from 50ng RNA using qSCRIPT cDNA Super Mix (Quanta Biosciences) according to manufacturer's instructions. cDNA was diluted 1:10 in water in 25-μL reaction volume with SYBR® Green Master Mix. The quantitative-PCR (q-PCR) cycling conditions were 15 seconds at 95°C and 1 minute at 60°C using the 7500 Fast Real-Time PCR System and analysis was carried out using the software (Applied Biosystems). The comparative C_t method was used for the relative quantitation of cDNA, where GAPDH was used as the housekeeping gene. PCR primers used: HSP27 forward (300 nmol/L) 5′-CTGATGAAGGGGAAGCAGG-3′ and reverse (300 nmol/L) 5′GACGACTTTCTGTTGCTGGG-3′, CHOP forward (300 nmol/L) 5′-TGGAAATGAAGAGGAAGAATCAAAA-3′ and reverse (900 nmol/L) 5′-CAGCCAAGCCAGAGAAGCA-3′; GAPDH forward (500 nmol/L) 5′-GAAGGTGAAGGTCGGAGTC-3′ and reverse (500 nmol/L) 5′-GAAGATGGTGATGGGATTTC-3′. Positive controls for CHOP qPCR were generated from cells treated with 10 μg/mL tunicamycin for 4 hours.

Proteins were harvested from cells with RIPA lysis buffer supplemented with protease inhibitor cocktail (Roche), PMSF and 1 mmol/L Na₃VO₄. Excised xenograft tumors were suspended in CHAPS lysis buffer in hard tissue grinding reinforced tubes with stainless steel beads (Precellys) and processed using a Precellys 24 tissue homogenizer. Insoluble material and fatty tissue were removed by centrifugation at 16,000 × g for 10 minutes at 4°C, twice.

Proteins were separated by SDS-PAGE on NuPAGE Bis-Tris gels (Life Technologies) and transferred onto PVDF membranes using the iBlot Dry Blotting System (Life Technologies). Membranes were blocked with 5% milk or BSA in TBS with 0.1% Tween-20 and incubated with primary antibody overnight. Primary antibodies used were HSF1 (Cell Signaling Technology, #4356), HSF1 phospho-Ser326 (Abcam, 76076), HSP72 (Enzo Life Science, ADI-SPA-810), HSP27 (Cell Signaling Technology, #2402) PARP (Cell Signaling Technology, #9542), eIF2α (Cell Signaling Technology, #9722), eIF2α phospho-Ser51 (Cell Signaling Technology, #9721) and GAPDH (Santa Cruz Biotechnology, 25778). Membranes were incubated with HRP-linked anti-rabbit IgG (Cell Signaling Technology, #7072) or anti-mouse IgG (Cell Signaling Technology, #7076) for 1 hour. ECL proteins were detected using Primer Western Blotting Detection Reagent (GE Healthcare) and visualized on Kodakκ BioMax XAR Film.

For HSF1 knockdown studies, proteins were collected 72 hours post-transduction. For CCT251236 and KRIBB11 treatment studies, proteins were collected as indicated. Quantitative densitometry analysis of protein bands was carried out using ImageJ software.

Ig light chain enzyme-linked immunosorbent assay (ELISA) was performed as described previously (3) and quantified using the lambda (λ) or kappa (κ) ELISA Quantitation Kit (Bethyl Laboratories) according to the manufacturer's instructions. In vitro secreted light chain was measured from conditioned media (1:100 dilution) collected from 3 × 10⁶ cells following 24-hour culture. Intracellular light chain was measured from 500 ng cell lysate protein collected. Plasma samples from mice bearing xenograft tumors and age-matched nontumor bearing mice (as controls) were tested at 1:100 dilution. For HSF1 knockdown studies, cells were plated 72 hours posttransduction.

A total of 3 × 10⁵ cells were plated following 72 hours transduction for 4 hours and treated with 10 μmol/L puromycin for 10 minutes before whole-cell lysis. Cells pretreated with 10 μmol/L pactamycin, a ribosome inhibitor, were used as a control. SDS-PAGE and Western blotting was performed on lysates as described previously.

The growth-inhibitory response to CCT251236 and KRIBB11 were measured using WST-1 reagent (Roche) or CellTiter 96 Aqueous One Solution Cell Proliferation Assay (MTS; Promega), respectively. The Caspase-Glo 3/7 luminescent assay (Promega) was used to measure caspase activity. Absorbance was measured on an Epoch Microplate Spectrophotometer (Bio-Tek) and luminscence on a Mithras LB940 Microplate Reader (Berthold Technologies) according to the manufacturer's instructions. Cells were plated at 2 × 10⁵ or 5 × 10³ cells per well for 48 to 96 hours in 96-well plates. Cells treated with 5 nmol/L bortezomib for 24 hours were used as positive controls for caspase activity.

Cells in Annexin V Binding Buffer (BD Biosciences) were stained with 2.5 μg/mL propidium iodide (PI) Staining Solution and 5 μL Annexin V-APC for 15 minutes and analyzed a BD LSR II flow cytometer. For HSF1 knockdown studies, cells were stained 72 hours following transduction. For rescue of cell death, cells were treated with 50 μmol/L Z-VAD-FMK for 24 hours before staining. In HSF1 pathway inhibitor studies, 3 × 10⁵ cells were treated with 48-hour GI₂₅ or GI₅₀ concentrations of CCT251236 or KRIBB11. For patient cell studies, 5 × 10⁵ CD138⁺ cells were cocultured with 1 × 10⁴ HS-5-GFP cells and treated with CCT251236 or KRIBB11 for 48 hours before staining with Annexin V-APC and DAPI (BD Biosciences). GFP-positive and -negative cells were gated and analyzed separately to distinguish between stromal HS-5-GFP and myeloma cells.

Myeloma xenograft tumors were established subcutaneously in female NOD/SCIDγc^null mice (Charles River). A total of 5 × 10⁶ H929 cells in 100% Matrigel (BD Biosciences) were injected in a single flank. Mice with tumors that reached a mean diameter of 0.6 to 0.7 mm were randomly grouped for initiation of treatment (day 0) with daily intraperitoneal (i.p.) administration of KRIBB11 at 65 mg/kg or vehicle (10% dimethylacetamide, 50% polyethylene glycol-300, 40% sterile water; ref. 27), or CCT251236 at 20 mg/kg orally, or vehicle (10% DMSO, 90% 2-hydroxypropyl-β-cyclodextrin). Tumor volumes and mouse body weights were determined at regular intervals. At the end of treatment (day 18–19) 16 hours after the final dose, plasma samples were taken and tumors removed, weighed, and extracted for proteins. The study was performed in accordance with UK Home Office regulations under the Animals Scientific Procedures Act 1986 and in accordance with UK National Cancer Research Institute guidelines (30).

HSF1 expression correlates with malignancy and poor prognosis in a number of solid tumors (18–19, 21–23). To explore the pathobiologic relevance of HSF1 in myeloma, we first assessed HSF1 expression levels in HMCLs. HSF1 expression by Western blot varied across the cell panel, with cell lines expressing higher HSF1 levels also generally showing higher expression of Ser326 HSF1 phosphorylation, an indicator of HSF1 transcriptional activity (ref. 31; Supplementary Fig. S1). Expression of HSP72 and HSP27 that are regulated by HSF1 showed considerable variation between HMCLs.

To determine whether HSF1 is a clinically relevant prognostic factor in myeloma, we looked for the impact of HSF1 expression on patient survival using publicly available gene expression profiling (GEP) datasets. HSF1 mRNA levels were assessed in CD138⁺ plasma cells isolated from newly-diagnosed (Supplementary Fig. S2) and relapsed patients (Fig. 1A and B). In the relapsed dataset (28), high HSF1 expression correlated with poorer OS (Fig. 1B; log rank P = 0.009; Cox regression P = 0.014; HR, 1.749; 95% CI, 1.098–2.785). Although this was not observed in the GEP of newly diagnosed patient datasets (Supplementary Fig. S2A–S2C), we found that high HSF1 expression was significantly associated with poorer OS in an independent dataset derived from RNA-sequencing of newly diagnosed patients (Supplementary Fig. S2D). These findings suggest that HSF1 stress response proteins have an important role in myeloma.

As HSF1 regulates oncogenic processes by driving a specific transcriptional program, we used the HSF1 cancer signature (HSF1 Ca-Sig) of 456 genes identified by Mendillo and colleagues (23) to determine whether differential expression of these HSF1 target genes can also be an indicator for patient prognosis. Hierarchical clustering of relapsed patient samples, based on expression of the HSF1 Ca-Sig, separated the patients into two groups with distinct patterns of HSF1 target gene expression (Fig. 1C). Not only did group 1 patients have a higher expression of HSF1 than group 2 (Fig. 1D), but they also performed worse in terms of OS compared with group 2 (Fig. 1E; log rank P = 4.04 × 10⁻⁷). These results suggest HSF1 regulates the expression of a subset of HSF1 Ca-Sig genes that impact the clinical outcome of myeloma patients.

To study the cellular function of HSF1 in myeloma, we established stable HSF1 knockdown models using RPMI-8226 and KMS-11 HMCLs. We achieved 80% to 90% depletion of HSF1 expression using two independent HSF1-targeted shRNAs (sh1 and sh2) in both HMCLs following 72 hours transduction as compared with empty vector and GFP-targeted shRNA-transduced (shGFP) controls (Fig. 2A). The downregulation of HSF1 phospho-Ser326, HSP72 and HSP27 was also observed, giving confidence that this model is useful for assessing both the functionality of HSF1 and its downstream transcriptional targets.

Knockdown of HSF1 has previously been shown to reduce the viability of various cancer cell lines (16, 32). It was, therefore, of interest to assess myeloma cell survival following HSF1 silencing. Knockdown of HSF1 led to a marked reduction in cell viability (>40% decrease) compared with shGFP controls (Fig. 2B). This was accompanied by an increase in caspase-3/7 activity (Fig. 2C) and the cleavage of PARP (Fig. 2D), which is a downstream substrate of caspase 3. Concurrent treatment with a pan-caspase inhibitor (z-VAD-FMK) partially rescued PARP cleavage (Fig. 2D) and decreased the population of dead cells showing Annexin V and PI-positive staining (Fig. 2E). These data demonstrate that the depletion of HSF1 expression is associated with caspase-mediated apoptosis, indicating a dependence of myeloma cells on HSF1 for survival.

To evaluate the feasibility of HSF1 as a therapeutic target, two chemically distinct inhibitors of the HSF1 pathway were employed: CCT251236, our novel potent and selective small molecule inhibitor of the HSF1 pathway (26), and KRIBB11, a pharmacologically distinct inhibitor of the HSF1 pathway previously described (27). Both CCT251236 and KRIBB11 led to a growth inhibitory effect in HMCLs (Supplementary Fig. S3A). Using concentrations of CCT251236 and KRIBB11 corresponding to the 48-hour GI₂₅ and GI₅₀ (Supplementary Fig. S3B), pharmacological inhibition of the HSF1 pathway led to a concentration and time-dependent increase in caspase-3/7 activity (Fig. 3A) that was accompanied by a decrease in cell viability (Fig. 3B).

To demonstrate on-pathway activity, levels of HSP27 (HSPB1) mRNA expression were assessed by qPCR. A time-dependent decrease in HSPB1 mRNA expression was observed following treatment (Fig. 3C) that corresponded to the downregulation of HSP27 protein expression (Fig. 3D). Downregulation of HSP72 expression and an increase in PARP cleavage were also observed in a time-dependent fashion (Fig. 3D). Similar results were obtained in human KMS-11 and H929 cells in response to pharmacologic inhibition of the HSF1 pathway (Supplementary Fig. S4).

Given that HSF1 and the HSPs play a pivotal role in protein folding, the consequence of HSF1 depletion on the Ig production capacity of myeloma cells was assessed using an Ig light chain ELISA. The ELISA was performed on cell supernatants and whole-cell lysates to detect secreted and intracellular levels of Ig light chains from RPMI-8226 and KMS-11 cells. Loss of HSF1 was associated with a significant downregulation of secreted Ig light chains compared with shGFP controls (Fig. 4A; Supplementary Fig. S5). Interestingly, no significant differences were observed in the intracellular levels of Ig light chains. We had previously observed that silencing of HSC70 and HSP72 reduced the secretion but increased the retention of Ig light chains in HMCLs (33). The lack of intracellular Ig accumulation in this model suggested that knocking down HSF1, which in turn alters the expression of several hundred HSF1-regulated genes, reduces overall Ig light-chain synthesis.

To assess whether protein synthesis is impaired by HSF1 knockdown, a protein translation assay based on puromycin incorporation into elongating polypeptide chains at the ribosome was used. Knockdown of HSF1 decreased the relative level of puromycin incorporation compared with shGFP-transduced cells (Fig. 4B; Supplementary Fig. S5B). The level of puromycin incorporation observed was comparable with the positive control treated with pactamycin, a ribosome inhibitor. Interestingly, one of the signaling cascades that can block protein translation is activation of the PERK arm of the UPR. Specifically, eukaryotic initiation factor 2α (EIF2α) is phosphorylated at Ser 51 by PERK (34). Consistent with this, knockdown of HSF1 (Fig. 4C) as well as pharmacologic inhibition of the HSF1 pathway using CCT251236 or KRIBB11 (Fig. 4D; Supplementary Fig. S5C) led to a marked increase in phospho-EIF2α. The transcript of CHOP, a downstream transcriptional target of PERK activation, was also upregulated by both agents (Fig. 4E; Supplementary Fig. S5D). These data suggest an additional downstream effect of HSF1 knockdown on other cellular protein homeostasis pathways.

To gain insight into the potential therapeutic index of agents targeting HSF1, the antimyeloma activity of CCT251236 and KRIBB11 was further assessed in vitro using CD138⁺ cells isolated from myeloma patient bone marrows. A concentration-dependent decrease in cell viability was observed in these primary cells following 48-hour treatment (Fig. 5A). We also examined the cytotoxicity of both compounds in PBMC and the bone marrow stromal cell line (BMSC), HS-5. Both CCT251236 and KRIBB11 decreased PBMC and HS-5 BMSC viability in a concentration-dependent fashion, but this was seen to a much lesser extent than in RPMI-8226 myeloma cells (Fig. 5B and C). Importantly, when tested under conditions of the coculture of RPMI-8226 with HS-5 BMSCs, there was still significant antimyeloma activity of CCT251236. Although as seen with other effective myeloma therapies, the bone marrow microenvironment offers some protection (Fig. 5C).

These results demonstrate in vitro antimyeloma activity of HSF1 pathway inhibitors on patient primary myeloma cells and strongly suggest a potential therapeutic window for a clinical candidate HSF1 inhibitor. Next, as single-agent activity was observed with HSF1 pathway inhibitors in in vitro models, we tested the activity in vitro of CCT251236 or KRIBB11 combined with bortezomib, a proteasome inhibitor that is a current myeloma first-line therapy. CCT251236 and KRIBB11 at subtoxic concentrations potentiated the antimyeloma effect of bortezomib in HMCLs (Supplementary Fig. S6), supporting the further study of HSF1 pathway inhibitors with proteasome inhibitors.

Yoon and colleagues (27) previously demonstrated that KRIBB11 administered at 50 mg/kg i.p. daily was well-tolerated by mice and inhibited the growth of HCT-116 human colon cancer xenograft tumors. On the basis of this work, a pilot study of KRIBB11 at 50 mg/kg i.p. daily was similarly well tolerated in mice carrying subcutaneous H929 human myeloma xenografts, but the effects on tumor growth were marginal (data not shown). We therefore increased the dose to 65 mg/kg KRIBB11 i.p. daily. At this slightly higher dose, we observed a significant decrease in tumor volume and tumor weight in KRIBB11-treated mice compared with vehicle-treated controls at end of therapy (day 19; Fig. 6A and B). The dose schedule was well-tolerated with no loss in body weight over the treatment period (Supplementary Fig. S7A), indicating a therapeutic index. The therapeutic benefit was associated with a significant decrease in serum κ Ig light chain as well as a significant reduction in HSP27 protein expression in treated tumors compared with controls (Fig. 6C and D; Supplementary Fig. S7B). Similar results were observed with subcutaneous H929 human myeloma xenografts with treated with 20 mg/kg orally CCT251236 (Fig. 6E and F; Supplementary Fig. S7C and S7D). These data indicate that inhibition of the HSF1 pathway has antimyeloma efficacy in a human myeloma xenograft model and that HSP27 expression may be a suitable pharmacodynamic (PD) biomarker of efficacy.

Our study provides both in vitro and in vivo evidence for the pro-oncogenic role of HSF1 in myeloma and the tractability of the HSF1 pathway as a therapeutic target. This is the first report of a significant association between HSF1 expression and the survival of myeloma patients, which adds to the array of studies in solid tumors showing the prognostic impact of HSF1 expression and clinical outcome (19, 21–22, 35). Furthermore, the difference in survival of patients exhibiting differential expression of a subset HSF1 target genes that was previously reported as prognostic for other cancers (23) supports the notion that HSF1 drives a transcriptional program in maintaining the cancer phenotype with effects on patient outcome.

Clearly, the above prognostic ability of HSF1 and its target genes is insufficient evidence to infer essentiality in myeloma or to conclude that the HSF1 pathway is a good target for therapeutic intervention in this disease. However, we also show here, employing both shRNA knockdown and the use of two chemically distinct HSF1 pathway inhibitors, that HSF1 is indeed required for myeloma cell survival: Because either genetic or pharmacological perturbation induces myeloma cell death. Such complementary, orthogonal use of two technically distinct perturbation approaches is accepted as an important means of derisking target validation, building confidence in the target-of-interest and the potential for small-molecule tractability (36).

The caspase-mediated cell death observed following HSF1 knockdown in RPMI-8226 and KMS-11 HMCLs clearly suggests that HSF1 and the genes in the HSF1 transcriptional program have a pro-survival and/or anti-apoptotic function in myeloma. Our findings are supported by observations by Heimberger and colleagues (37) where 72-hour HSF1 shRNA-mediated silencing in two other HMCLs (MM1.s, INA-6) also led to loss in cell viability. Shah and colleagues (15) however, did not observe a decrease in cell viability to the same extent following 48-hour HSF1 siRNA-mediated knockdown, suggesting that a sustained and prolonged loss of HSF1 is required for an anti-myeloma effect. Myeloma is therefore another example where HSF1 enables cell proliferation or suppresses cell death in the oncogenic context, as shown using Hsf1^−/− MEFs transformed with Pdgf-b or C-myc (16). The sensitivity of HMCLs, patient primary cells, as well as a tumor xenograft, to two chemically distinct small-molecule compounds that inhibit the HSF1 stress pathway, CCT251236 and KRIBB11, is an encouraging sign for the potential efficacy of inhibitors of the HSF1 pathway for myeloma patients.

Given the challenge of therapeutically targeting transcription factors such as HSF1 with small molecules (25), other more chemically tractable targets within the HSR previously have been investigated in myeloma, namely HSP90 and HSP70. A number of HSP90 inhibitors have been tested both pre-clinically and/or clinically, including tanespimycin (38). However, the induction of antiapoptotic HSP72 expression following HSP90 inhibition in myeloma and other solid tumors has been reported to limit their effects (10, 38–39). The cytoprotective action of HSP70 family proteins are illustrated by our findings where HSP72 siRNA knockdown enhanced sensitivity to tanespimycin (14). Targeting one HSP70 isoform alone, however, is insufficient, as we showed that silencing the constitutively expressed HSP70 isoform (HSC70) led to the upregulation of HSP72 and that dual knockdown of both HSP70 isoforms induced heightened caspase-mediated cell death compared with silencing either isoform alone (13, 33). Indeed, we and others have reported that small-molecule HSP70 inhibitors that target both isoforms, such as VER155008 and MAL3-101, increase the antimyeloma effect of HSP90 inhibition (33, 40). Besides HSP70 family proteins, the compensatory upregulation of antiapoptotic HSP27 has also been reported following tanespimycin treatment in HMCLs (41). Therefore, we propose that targeting the broad downstream HSF1 transcriptional program may be a more effective therapeutic strategy than inhibiting HSPs individually.

Additional compounds, notably triptolide and KNK437, have been described as inhibitors of HSF1-mediated HSP72 expression and used to demonstrate the antimyeloma efficacy of pharmacologic HSF1 pathway blockade (37, 42). However, as these compounds are reported to have general transcription inhibitor characteristics, we used two novel HSF1 pathway inhibitors as tools to explore the response of myeloma models to pharmacological inhibition of the HSF1 pathway.

CCT251236, a bisamide compound, was derived from an initial hit compound that suppressed the induction of HSF1-mediated HSP72 expression in a cell-based phenotypic screen (26). We showed by extensive profiling that CCT251236 possesses no detectable off-target kinase activity, making it superior to compounds from another chemical series, 4,6-disubstituted pyrimidines, which were also discovered in the same screen (43). KRIBB11 was a hit compound from a cell-based screen designed to identify molecules that attenuated heatshock element (HSE)–driven luciferase reporter signals (27). Although the off-target effects of KRIBB11 are unknown, Yoon and colleagues (27) demonstrated that KRIBB11 directly interacts with HSF1 and does not attenuate NFκB promoter activity. Furthermore, the similar phenotypic changes observed following KRIBB11 treatment and HSF1 knockdown eases potential concerns over KRIBB11 off-target activity at concentrations used in our current work. Indeed, the growth-inhibitory activity of KRIBB11 on HMCLs has also been demonstrated in MM1.s cells (44). The more potent growth inhibitory activity of CCT251236 compared with KRIBB11 can be attributed to the much lower cellular potency of KRIBB11 in the inhibition of HSF-driven luciferase activity (EC₅₀ = 1.2 μmol/L) compared with that of CCT251236 in the cellular inhibition of HSP72 expression (EC₅₀ = 19 nmol/L; refs. 26–27).

Given the challenges in discovering selective inhibitors of HSF1 (25, 45), compounds such as CCT251236 or ones recently described by Vilaboa and colleagues (46) are steps in the right direction toward developing a potent and selective HSF1 pathway inhibitor for clinical use. Of note, one of the compounds reported, IHSF058 (46), causes depletion of HSF1 protein—as we also observe here with CCT251236 and KRIBB11, albeit at 24 to 48 hours when cell death is underway. Understanding the mechanism and significance of this HSF1 depletion requires further study.

Importantly, the therapeutic activity of CCT261236 and KRIBB11 in our xenograft models as well as the in vitro cell killing effects in HMCLs observed that was concomitant with a decrease HSP72 and HSP27 expression, provides additional evidence that HSF1 pathway inhibition is a viable means to suppress the cytoprotective and prosurvival actions of HSF1, the HSPs and also other HSF1-regulated genes. The further development of compounds with improved potency and pharmacokinetic properties, as well as additional work identifying robust biomarkers that correlate with response will be very important. Furthermore, the use of additional preclinical myeloma models that more closely recapitulate human disease, together with refinement of optimally efficacious doses and schedules, will be essential in determining whether HSF1 pathway inhibitors merit progression into the clinic.

Another important question that remains to be answered is whether targeting the HSF1 pathway would be a universal approach for myeloma patients, as with proteasome inhibition, or whether particular genetically or clinically defined subgroup of patients would benefit more from such therapy. Additional studies are required to examine whether factors—such as expression of HSF1, the transcriptional program it regulates (including the HSF1 Ca-Sig derived from other cancers (ref. 23) or a signature that could be determined specifically in myeloma) or indeed Ig production—are able to predict sensitivity to HSF1 pathway inhibition. These studies could potentially provide insights into myeloma subtypes that might best respond to HSF1-targeted therapy. Combining HSF1 pathway inhibitors with other modulators of protein handling pathways should also be explored further, in line with the idea of using combinatorial proteotoxic stress targeted therapy for myeloma (1, 6). Indeed, initial evidence reported here showing the combinatorial benefit with bortezomib supports this approach. Furthermore, the induction of CHOP expression and EIF2α phosphorylation following HSF1 knockdown or pharmacologic HSF1 pathway inhibition suggests that PERK inhibitors may also potentiate HSF1 inhibitors.

Despite the clinical progress made with novel agents for myeloma in the past decade, patients continue to relapse (47). There is therefore an unmet need for new classes of myeloma therapeutics. With increasing evidence for intraclonal genetic heterogeneity in myeloma (48), targeting fundamental cellular stress support pathways may be an alternative and potentially be more effective than targeting specific genetic lesions or oncogenic pathways. Thus, inhibiting one target with multiple downstream effects is an attractive proposition.

In summary, our work demonstrates that HSF1 pathway activity is essential for myeloma cell survival. More importantly, our findings provide proof-of-concept evidence to support the further development of highly potent and selective HSF1 pathway inhibitors as an additional and promising therapeutic option for myeloma.

N.E.A. Chessum and K. Jones report receiving commercial research grants from Battle Against Cancer Investment Trust and Cancer Research Technology Pioneer Fund. P. Workman reports receiving commercial research grants from Battle Against Cancer Investment Trust and Cancer Research Technology Pioneer Fund, holds ownership interest (including patents) in Chroma Therapeutics, and is a consultant/advisory board member for Astex Therapeutics, Chroma Therapeutics, Nextech Invest, and Nuevolution. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J.H.L. Fok, S. Hedayat, M.D. Bright, N.E.A. Chessum, K. Jones, G.J. Morgan, P. Workman, F.E. Davies

Development of methodology: J.H.L. Fok, S. Hedayat, L. Zhang, L.I. Aronson, M.D. Bright, E. De Billy, P. Workman, F.E. Davies

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.H.L. Fok, S. Hedayat, F. Mirabella, C. Pawlyn, J.J. Keats, G.J. Morgan, S.A. Eccles, P. Workman, F.E. Davies

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.H.L. Fok, S. Hedayat, L.I. Aronson, F. Mirabella, C. Pawlyn, C.P. Wardell, J.J. Keats, E. De Billy, C.S. Rye, P. Workman

Writing, review, and/or revision of the manuscript: J.H.L. Fok, S. Hedayat, C. Pawlyn, J.J. Keats, K. Jones, S.A. Eccles, P. Workman, F.E. Davies

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.H.L. Fok, S. Hedayat, L.I. Aronson, P. Workman

Study supervision: S. Hedayat, M.D. Bright, G.J. Morgan, P. Workman, F.E. Davies

Other (designed and synthesized one of the chemical compounds used in the study): C.S. Rye

Other (contributed toward the synthetic chemistry design and synthesis of the active compound): N.E.A. Chessum

This work was supported by Cancer Research UK (grant number C20826/A12103). J.H.L. Fok was funded by a studentship from Cancer Research UK. F.E. Davies was a Cancer Research UK Senior Cancer Research Fellow. P. Workman is a Cancer Research UK Life Fellow. We acknowledge NHS funding to the NIHR Biomedical Research Centre at the Royal Marsden Hospital and the Institute of Cancer Research, Cancer Research UK funding to the Cancer Research UK Cancer Therapeutics Unit and funding from Cancer Research Technology Pioneer Fund and the Battle Against Cancer Investment Trust for the project leading to the discovery of CCT251236 and other HSF1 pathway inhibitors.

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.