Agents that trigger cell differentiation are highly efficacious in treating certain cancers, but such approaches are not generally effective in most malignancies. Compounds such as DMSO and hexamethylene bisacetamide (HMBA) have been used to induce differentiation in experimental systems, but their mechanisms of action and potential range of uses on that basis have not been developed. Here, we show that HMBA, a compound first tested in the oncology clinic over 25 years ago, acts as a selective bromodomain inhibitor. Biochemical and structural studies revealed an affinity of HMBA for the second bromodomain of BET proteins. Accordingly, both HMBA and the prototype BET inhibitor JQ1 induced differentiation of mouse erythroleukemia cells. As expected of a BET inhibitor, HMBA displaced BET proteins from chromatin, caused massive transcriptional changes, and triggered cell-cycle arrest and apoptosis in Myc-induced B-cell lymphoma cells. Furthermore, HMBA exerted anticancer effects in vivo in mouse models of Myc-driven B-cell lymphoma. This study illuminates the function of an early anticancer agent and suggests an intersection with ongoing clinical trials of BET inhibitor, with several implications for predicting patient selection and response rates to this therapy and starting points for generating BD2-selective BET inhibitors. Cancer Res; 76(8); 2376–83. ©2016 AACR.
The idea of tumor cell differentiation was conceptualized when Charlotte Friend showed that the solvent DMSO was capable of inducing differentiation of virus-induced mouse erythroleukemia (MEL) cells (1). Because the FDA had banned DMSO for human use, Tanaka and colleagues (2) initiated a search for polar compounds that would be able to induce MEL cell differentiation at lower concentrations. The most effective compound was hexamethylene bisacetamide (HMBA; ref. 3). It was developed as far as to the clinic where it showed acceptable toxicity profiles in phase I clinical trials and activity against one quarter of patients with acute myelogenous leukemia (AML) in phase II trials. However, likely due to low response rates, HMBA was not further developed after 1992 (4–7). Parallel to the development of HMBA, Richon and colleagues (8) also discovered the target of a much more potent MEL differentiation agent, suberoylanilide hydroxamic acid (SAHA). They demonstrated that SAHA is a potent inhibitor of histone deacetylases (HDAC), which spawned active investigations and clinical development. SAHA, or vorinostat, is now FDA-approved for treatment of cutaneous T-cell leukemia (9) and other HDAC inhibitors are approved for use in other tumor types. The target of HMBA remained elusive though (10) because Richon and colleagues (8) could not demonstrate that it inhibited HDACs.
The function of BET bromodomain proteins (BRD2, BRD3, BRD4, and BRDT) is to bind acetylated proteins, most notably histones, and thereby regulate transcription (11). BET bromodomain inhibitors (BETi) block this binding, resulting in broad transcriptional changes (12–16). Recently, we demonstrated that vorinostat triggers transcriptional changes that are overlapping with those induced by BETi in murine Myc-induced lymphoma cells (17). Given our data that BETi and HDACi can induce similar transcriptional changes (17), we hypothesized that the target of HMBA could be one or several BET bromodomain proteins.
Materials and Methods
The (+)-enantiomer of JQ1 was purchased from Cayman chemicals. HMBA, C646, and CPTH2 were purchased from Sigma-Aldrich. Vorinostat was purchased from Selleck Biochemicals.
Cell culture and mouse experiments
The B-cell lymphoma cell lines λ663, λ820, Eμ239, and Eμ580 were established 2006 from single cell suspensions of tumors arising in λ-Myc or Eμ-Myc transgenic mice by serial passage in culture (17) and were grown in RPMI-1640 supplemented with 10% FBS, glutamine, 50 μmol/L β-mercaptoethanol, and antibiotics. The only mean of authentication of these lines were by qRT-PCR of Myc (human and mouse for λ-Myc or Eμ-Myc transgenic mice, respectively). The human Burkitt lines Akata and Daudi were obtained from Erik Lundgren (Umeå University, Umeå, Sweden, Sweden) and were authenticated by Myc overexpression and p53 mutation status in 2008 as previously described (18). The MEL cells were a kind gift from Dr. Olle Heby (Umeå University) and were cultured in DMEM supplemented with 20 % FBS and antibiotics. They were only authenticated in this study by showing their differentiation to erythrocytes (Fig. 1).
All animal experiments were performed in accordance with regional/local animal ethics committee approval (approval number 287/2011 and 193/2014). C57BL/6 mice were transplanted with 200,000 λ2749 lymphoma cells (17). White blood cell (WBC) counts were measured by blood sampling from vena saphena. When WBC counts were above the normal range (6–15 cells/nL), mice were treated with bi-daily intraperitoneal injections of HMBA or left untreated.
One million cells per mL were lysed and stained for 30 minutes at 37°C in modified Vindelöv's solution (20 mmol/L Tris, 100 mmol/L NaCl, 1 μg/mL 7-AAD, 20 μg/mL RNase, and 0.1 % NP40 adjusted to pH 8.0) followed by analysis of DNA content using the FL3 channel (linear mode and cell cycle) or FL3 channel (logarithmic mode and apoptosis).
RNA preparation and analyses
λ820 (in biologic replicates) were cultured in the absence or presence of 5 mmol/L HMBA for 24 hours. RNA was prepared using the NucleoSpin RNA II Kit (Macherey-Nagel). The analysis of gene expression was performed using the Illumina BeadChip system at the Genomics Core Facility, University of Gothenburg, Sweden. Mouse Ref-8 v2 Beadchip Arrays (Illumina) were used following the manufacturer's protocol. Primary data were collected from the BeadChips using the Illumina BeadArray Reader (HiScan) and analyzed using the supplied software. Data normalization was performed by cubic spline normalization using Illumina's GenomeStudio software equipped with the necessary modules. Before further analysis, FORCEPOS adjustment (+25) was performed to avoid negative data arising during normalization. To compare data with those generated previously from JQ1-treated or vorinostat-treated cells (17), all data from compound-treated cells were normalized to their respective experimental controls (untreated cells for HMBA, 0.1% DMSO-treated cells for JQ1 and vorinostat). Full datasets are available as Supplementary Dataset S1. Clustering and visualization of genes was performed using Spotfire. Gene-set enrichment analysis was done using the Qlucore software.
For quantitative reverse transcriptase PCR (qRT-PCR), total RNA was prepared from cultured cells or tumors. cDNA was prepared from 500 ng total RNA using the iScript cDNA Synthesis Kit (Bio-Rad). qRT-PCR was performed using KAPA SYBR FAST ABI Prism 2X qPCR Master Mix (Kapa Biosystems, Inc.). Data analyses were performed by comparing ΔΔCt values, using Ubiquitin (Ub) as the reference gene and with a control sample set as relative expression 1.
Chromatin immunoprecipitation (ChIP) was carried out using the manufacturer's protocol (SimpleChIP Plus Enzymatic Chromatin IP Kit; Cell Signaling Technology). Briefly, cells left untreated or treated with HMBA for 18 hours were fixed with formaldehyde and lysed; chromatin was partially digested by micrococcal nuclease and then sonicated. Lysates were quantified and 10 μg of chromatin was loaded per immunoprecipitation and the rest was used as input. ChIP was carried out using antibodies directed against Brd2 (rabbit; Bethyl Laboratories), Brd3 (mouse; Santa Cruz Biotechnology), or Brd4 (rabbit; Bethyl Laboratories), where anti-rabbit and anti-mouse IgG was used as an isotype control. Purified and reverse cross-linked chromatin was analyzed by quantitative PCR using primers designed against, Cd74 (17), Srm (19), and Vpreb3.
HMBA at five different concentrations (5, 0.5 mmol/L, 50, 5, 0.5 μmol/L, and 50 nmol/L) was profiled against 10 different bromodomain (BRD) protein fragments and rat liver HDAC. For all assays, reference compounds were used to validate the assays. For BRD proteins, binding of recombinant human proteins to biotinylated and acetylated histone peptides was detected using the AlphaScreen detection method as previously described (20). For HDAC, a fluorimetric method was used that had been previously described (21).
Molecular docking of BRD2 bromodomain 2 (PDB# 2E3K) to JQ1 or HMBA was performed using the Swissdock server (www.swissdock.ch/). Input ligands were imported from the ZINC database. The resulting docking results were visualized in Chimera.
For crystallization, His-tagged BRD3 bromodomain 2 (BRD3 BD2) was expressed in E. coli and purified on a Ni-NTA column as previously described (20). The His-tag was removed by TEV protease cleavage followed by an additional purification on Ni-NTA column. BRD3 BD2 was concentrated to 42.1 mg/mL in 20 mmol/L HEPES, 500 mmol/L NaCl, 5 % glycerol, pH 7.5. Crystallizations were then done using a mosquito robot (TTP Labtech). Before crystallization, the buffer was changed using a micron centrifugal filter with a cutoff value of 10 kDa. The crystallization buffer contained 10 mmol/L HEPES pH 7.5 and 300 mmol/L NaCl. Crystals were grown by vapor diffusion at 4°C from a sitting drop consisting of 150 nL protein (10 mg/mL) and 150 nL reservoir solution containing 19% PEG 6000, 0.1 mol/L HEPES pH 7.0. The crystals appeared after a few days and had a size of 60 × 60 × 15 μm. For soaking, the crystal was transferred to a drop containing 19 % PEG 6000, 0.1 mol/L HEPES pH 7.0 and 50 mmol/L HMBA for 20 hours. The crystal was then transferred to a cryoprotectant solution containing 19 % PEG6000, 0.1 mol/L HEPES pH 7.0, 25 % ethylene glycol and 50 mmol/L HMBA and flash-frozen in liquid nitrogen.
A BRD3 BD2 dataset was collected to 1.9 Å at 100 K at beam line I911-3, MAX IV laboratory (λ = 1.0000 Å). One hundred and sixty-five images in total, each with 1° rotation, were collected. All data were integrated and scaled using XDS and Aimless softwares. The structure of BRD3-BD2 in complex with HMBA was determined by rigid body refinement of an in-house apo structure to 1.5 Åresolution. Ligand conformations were created using software from Schrödinger, LLC. After initial anatomic refinement of the protein coordinates in Refmac5 the coordinates of HMBA were fitted to the electron density using the Coot software. CCP4-style restraints for HMBA were generated using the Jligand program. The structures were refined to convergence using Refmac5. The protein structure quality was checked using the Molprobity server. Hydrogen atoms were added in the riding positions. Water molecules were added to positive difference density peaks more than 5 SDs above the mean and present in 2m|Fo|-d|Fc| map at the 1 sigma level.
Accession codes: Coordinates and structure factors have been deposited in the Protein Data Bank under accession code 5A7C.
Cells were washed twice in PBS, dissolved in water and freeze-thawed three times. After a 30-min centrifugation at 15,000 × g the absorbance spectrum of the supernatant was measured on a Nanodrop spectrophotometer. Whole blood from a mouse was diluted and processed the same way as the cells and was used a positive control. Presence of hemoglobin in the sample was evident by the presence of peaks at 414, 540 and 576 nm.
The bars shown represent the mean ± SD. The Student t test were performed using GraphPad Prism (GraphPad Software). Statsitically significant differences are indicated with *, P < 0.05; **, P < 0.01; ***, P < 0.001; or ****, P < 0.0001.
HMBA cause MEL cell differentiation by acting as a selective BD2 inhibitor
JQ1 is a potent BETi (16), which binds to bromodomain 1 and 2 (BD1 and BD2) in all BET bromodomain proteins. Consistent with the hypothesis that HMBA could be a BET inhibitor, molecular docking in silico shows that HMBA fits the same pocket in BD2 of BRD2 as JQ1 (Supplementary Fig. S1A). Because HMBA is a small polar molecule that potentially could fit many pockets, we also measured binding of bromodomain-containing proteins to acetylated histone peptides in the presence of increasing concentrations of HMBA or known ligands. This demonstrated that HMBA could displace BD2 in BRD2, BRD3, and BRD4, with the highest affinity to BRD3 (Table 1 and Fig. 1A). On the other hand, as previously shown (8), HMBA does not inhibit HDACs (Table 1).
|BRD assay .||Reference Cpd .||IC50 (mol/L) Reference .||IC50 (mol/L) HMBA .|
|BRD assay .||Reference Cpd .||IC50 (mol/L) Reference .||IC50 (mol/L) HMBA .|
NOTE: Assays were performed in duplicates over a 10-fold dilution series with 5 mmol/L being the highest concentration. Inhibition was observed in four assays; in the rest, an IC50 value was not reached (NR).
To assess whether HMBA's interactions with BET proteins were specific, we soaked crystals of the second bromodomain of BRD3 with HMBA and determined the X-ray crystal structure of the complex. Electron density maps revealed unambiguous density for bound HMBA (Fig. 1B; Supplementary Fig. S1B). There are four molecules of BRD3 BD2 in the asymmetric unit. HMBA is fully visible in the electron density for subunit A and is bound in the acetyl-lysine pocket of BRD3 BD2 (Fig. 1B). For the other three subunits the acetamide binding to Asn391 and the conserved water molecules is very similar, but the tail and the second acetamide are poorly visible in the electron density, indicating flexibility. Furthermore, in subunit A there are two water molecules from a symmetry-related molecule interacting with the second acetamide, which could explain why the entire ligand is only visible in the electron density for this subunit. The surrounding crystallographic environment is not the same for the other subunits. When superimposing all subunits, it becomes clear that the linker and the second acetamide are flexible (Supplementary Fig. S1C and S1D). Of interest is also that HMBA makes a hydrogen bond with His395 in one of the subunits. Because this position is conserved in the BC-loop of the BD2 domain of all BET proteins, and as BD1 has an Asp at the corresponding position (20), it is tempting to speculate that this is the reason behind the apparent BD2 selectivity of HMBA.
BET proteins belong to a larger family of bromodomain-containing proteins, the BRD family (20). Given HMBA's small size, it requires 3 to 5 mmol/L in concentration to exhibit most effects in cell culture. Hence, it was of importance to investigate whether HMBA could bind more proteins than the BD2 domain of BET proteins. Analyzing representative proteins of different branches of the BRD family tree, we found that the bromodomain of the histone acetyltransferase EP300 was also inhibited from binding acetylated histone (Table 1).
Next, we evaluated the effects of HMBA and JQ1 on mouse erythroleukemia (MEL) cells. We reasoned that if HMBA induced differentiation by blocking BET proteins then the prototype BET inhibitor JQ1 should also induce differentiation of MEL. Indeed, treatment of MEL cells with either 5 mmol/L HMBA or 0.1 μmol/L JQ1 resulted in formation of red cells, indicative that hemoglobin has been produced (Fig. 1C). Confirming this, lysates of HMBA- or JQ1-treated MEL cells exhibited the same absorbance spectra as did mouse blood (Fig. 1D) and qRT-PCR analysis showed that both JQ1 and HMBA induce the expression of the hemoglobin-encoding gene Hba-a1 (Fig. 1E). These data suggest that BET inhibition is sufficient to trigger differentiation of MEL cells. However, inhibiting EP300 with the selective inhibitor C646 (22), or histone acetyltransferase GCN5 with the selective inhibitor CPTH2 (23), does not result in differentiation (Fig. 1C). This suggests that HMBA's effect on differentiation is mediated by inhibiting BET proteins and not by inhibiting acetyltransferases in general.
HMBA kills Myc-induced lymphoma cells in vitro and in vivo
In the clinic, the therapeutic effects of HMBA were only modest. Because we previously demonstrated that BETi such as JQ1 and RVX2135 arrest and kill Myc transgenic murine lymphoma cells (17), we were interested to investigate whether Myc-overexpressing cells would be sensitive to HMBA. To that end, cultured λ820 and λ663 cells, developed from λ-Myc transgenic mice, were treated with HMBA. Similar to the effects seen with other BETi (17), HMBA induced a cell-cycle arrest at 24 hours followed by apoptosis of all cells by 48 hours (Fig. 2A and data not shown). We also treated two lymphoma lines developed from Eμ-Myc transgenic mice (24), and two human Burkitt lymphoma cell line. We observed robust antiproliferative effects after 24 hours in the murine lines (Fig. 2A) whereas the effects in Burkitt lymphoma cell line were slower (Fig. 2B).
Next, we performed microarray analyses of λ820 cells treated with HMBA and compared the data with those of previously performed expression profiles of JQ1 or the HDACi vorinostat (SAHA)–treated cells (17). As shown in Fig. 3A, unsupervised hierarchical clustering analysis of the 2-fold up- and downregulated genes (by all compounds, Supplementary Dataset S1) shows that HMBA clusters more closely with JQ1 than with vorinostat. Moreover, 56 % of genes suppressed and 68 % of genes induced (>2-fold) by JQ1 are also regulated in the same way by HMBA, whereas that similarity was lower for vorinostat-regulated genes (36% similarity of both induced and suppressed genes, Fig. 3B). In our previous study, it was demonstrated that HDACi and BETi regulate a few genes in each other's opposite directions (17). Indeed, the expressions of genes like Cd74, which are induced by vorinostat but repressed by JQ, were also repressed by HMBA (Fig. 3A, inset and Fig. 3C). Furthermore, ingenuity pathway analyses of 2-fold up- and downregulated genes showed high similarities between JQ1 and HMBA. Of the 10 pathways most affected by both compounds, six were of lymphocyte development or signaling (e.g., B-cell development shown in Supplementary Fig. S2), and four were of cell-cycle regulation (e.g., Supplementary Fig. S3). Accordingly, investigation of the MutSigDB database and follow-up gene-set enrichment analysis showed that genes repressed by HMBA were enriched in genes induced in S-phase by E2f (Supplementary Fig. S4), as previously shown for JQ1 (25). Finally, a ChIP assay experiment demonstrated that Brd2 and Brd4, but not Brd3, was clearly displaced by HMBA from the regulatory regions of the Spermidine synthase (Srm) gene (Fig. 3D), a gene whose expression was suppressed by HMBA (Supplementary Dataset S1). Taken together, the data strongly suggest that HMBA is a BETi also in Myc-induced lymphoma cells.
HMBA was never shown to be efficacious in mouse models of cancer, but was moved into clinical trials based on overwhelming in vitro data. Having established that HMBA kills Myc-induced lymphoma cells in vitro, we were interested in investigating whether antitumoral activity could be observed in this tumor type. We transplanted mice with λ2749 cells, which originally developed in a λ-Myc transgenic mouse and has been maintained since by serial transplantation in syngenic recipient B6 mice. Mice transplanted with these cells develop a lethal lymphoma that can be monitored by measuring whether the WBC count is above the normal range (6–15 cells/nL; ref.17). Treatment with the BETi RVX2135 results in a rapid reduction of WBC (17) and prolonged survival, making this an optimal model to assess effects of HMBA in vivo. As seen in Fig. 4A, treatment of mice with HMBA carrying λ2749 cells resulted in a normalization of WBC, and this translated to a prolonged survival of these mice (Fig. 4B). To investigate whether HMBA exhibited activity against other Myc-driven B-cell malignancies, λ663 cells were injected into mice. The λ663 cells do not display traditional growth pattern with growth primarily in lymph nodes and spleen. Instead the mice transplanted develop hind leg paralysis. Labeling the cells with luciferase demonstrated that the lymphoma cells in these mice indeed propagate in the bone marrow (Fig. 4C). Although the disease was not ameliorated by HMBA, the progression was significantly blunted (Fig. 4C). Taken together, we here demonstrate for the first time that HMBA has activity against Myc-driven lymphoma in vivo.
In this report, we have established that HMBA has the ability to inhibit at least four members of the BRD family. Our data also suggest starting points for synthesis of BD2-selective BET inhibitors. Identification of the targets is important because HMBA continues to be used as a differentiation agent and continues to be instrumental for identifying mechanisms of transcriptional elongation (26). Our data suggest that HMBA can be exchanged for more selective and potent BET bromodomain inhibitors in the study of differentiation.
More than 25 years ago, HMBA was used in patients suffering from many different malignancies, most notably AML. Current clinical investigators could learn from the HMBA trials that certain toxicities, for example, thrombocytopenia, need to be managed, and, unfortunately, few patients are likely to respond. The presented data suggest that MYC overexpression could be a biomarker of response to BET inhibitors. This was not known when the HMBA trials were conducted, so it is plausible that the trials would have generated better results if patients suffering from Myc-driven diseases had been recruited. Re-evaluating HMBA as a drug to treat Myc-induced cancers such as multiple myeloma, glioblastoma, neuroblastoma, and small-cell lung cancer is ongoing in animal models in our laboratory.
Although not formally proven here, our data also suggest that BET inhibition could be the reason for the induction of differentiation by DMSO. Indeed, DMSO is generally advised against as solvent of compounds for use in bromodomain protein–binding assays or in crystallization studies, because it can bind and inhibit bromodomains by displacing essential water molecules (27). It is tempting to speculate that BET protein inhibition could explain why high concentrations (>1%) of DMSO have many biologic effects such as inducing differentiation or killing tumor cells or blocking inflammation. Speculatively, it could also explain why cultured cells need time to recover after cryopreservation—they have been soaked in 10% of a BET inhibitor.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Conception and design: L.M. Nilsson, J.A. Nilsson
Development of methodology: L.M. Nilsson
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L.C. Green, S.V. Muralidharan, D. Demir, M. Welin, J. Bhadury, B. Walse
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L.M. Nilsson, L.C. Green, S.V. Muralidharan, M. Welin, D.T. Logan, B. Walse, J.A. Nilsson
Writing, review, and/or revision of the manuscript: L.M. Nilsson, L.C. Green, M. Welin, D.T. Logan, B. Walse, J.A. Nilsson
Study supervision: D.T. Logan, J.A. Nilsson
The authors thank the MAX IV Laboratory for access to the I911-3 beam line and Sofia Nordstrand and the animal house personnel for animal care.
This work was supported by grants (J.A. Nilsson) from the Swedish Cancer Society (#2012/355), the Swedish Research Council (#2012-2771), the BioCARE Strategic Grants for Cancer Research, Västra Götalandsregionen (#ALFGBG-434331), and the Knut & Alice Wallenberg Foundation (#2014.0080). J. Bhadury. and S.V. Muralidharan are supported by PhD student research grants from Assar Gabrielsson Foundation Wilhelm & Martina Lundgren's foundation and Sahlgrenska University Hospital, and D. Demir was supported by MD student laboratory rotation grants from Maj Bülow Foundation and Göteborgs Läkaresällskap.
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.