The Discovery of a Novel Antimetastatic Bcl3 Inhibitor

Despite significant advances in cancer drug therapy in recent years, treating tumor metastasis remains an area of significant unmet medical need. It is widely estimated that metastasis accounts for cancer deaths in around 90% of patients (1), yet historically treating metastasis has been considered a difficult challenge in part due to its association with late-stage disease and the absence of suitably targeted therapies. However, recent advances in understanding of this complex multistage process, and some of the key molecular pathways potentiating the dissemination and colonization of tumor cells, offers potential for early therapeutic intervention of metastasis in high-risk (short latency) disease as well as the possibility to target existing metastases in established disease (1, 2). Previous success in the development of new therapeutic agents targeting metastasis has been mainly restricted to agents that target existing disseminated lesions and establishment of the tumor microenvironment to support them. For example, bevacizumab (3), eribulin mesylate (4), and trastuzumab emtansine (5) showed initial promise in clinical trials in advanced breast cancer, yet offered only modest improvements to long-term survival benefits compared with established chemotherapy. The strategy of using antimetastatic drugs to prevent the initial or ongoing dissemination of disease as early as possible, possibly in conjunction with existing chemotherapy, has gained traction in recent years. The successes of the receptor activator of NF-κB inhibitory antibody and osteoporosis therapy denosumab (6), and the widely used bisphosphonates (7), in specifically reducing skeletal-related events are notable although only limited to metastasis to the bone, whereas experimental agents targeting Rho/Rac-mediated cell migration/invasion for example are effective at reducing metastasis in the preclinical setting but exhibit significant systemic toxicity in vivo (8). To ensure successful translation into a clinical setting, therefore, such interventions must overcome the dual problems of nonselective toxicity and cellular plasticity, for example when tumor cells adopt alternative modes of migration to overcome pharmacologic inhibitors of cell motility (9, 10).

Recently, we have described a nonredundant role for the NF-κB facilitator protein, Bcl3, in the multimodal migration of metastatic breast cancer cells in vivo (11), whereby suppression of Bcl3 inhibits all forms of compensatory cell migration, thus targeting metastatic seeding in vivo. Originally described as a key oncogenic translocation in B-cell lymphomas (12), Bcl3 has proven roles in a variety of hematologic (13) and solid tumor cell types including breast (14, 15), colorectal (16), nasopharyngeal (17), ovarian (18), and prostate (19) cancers, and glioblastomas (20, 21). In addition, a recent study has implicated Bcl3 in progression of metastatic colorectal cancer through the maintenance of cancer stem cells (22). The fact that mouse models with constitutive Bcl3 deficiency are viable with only minor immunologic defects (14, 23), and the role of Bcl3 in induction of immune checkpoint PD-L1 expression (18), further substantiates the potential role of Bcl3 as a novel target for therapeutic intervention.

To date, there have been no reported small-molecule Bcl3 inhibitors, and mechanistic studies on Bcl3 biology have been limited to gene expression regulatory tools such as siRNA.

Inhibition of protein–protein interactions (PPIs) as a therapeutic strategy has emerged as a new frontier within drug discovery in recent years. It has long been recognized that specific PPIs within cellular signaling pathways can drive disease progression. Recently, the previously held consensus that PPIs were rather large, flat, and featureless interfaces not amenable to small-molecule drug design has been challenged by the concept of key hotspot residues responsible for therapeutically relevant PPIs (24). The clinically approved microtubule inhibitors such as taxanes and vinca alkaloids (25), and Bcl2 homology domain (BH3) mimetic Venetoclax (26) provide further validation for this concept.

Here, we sought to devise a strategy to discover selective small-molecule inhibitors of Bcl3 with antimetastatic properties. Our studies focused on a key PPI between the ankyrin repeat domain of Bcl3 and its regulatory protein partner p50. The importance of this Bcl3–p50 interaction in driving Bcl3 function has been demonstrated through generation of p50 mutants (27) and an ankyrin repeat mutant of Bcl3 via site-directed mutagenesis (ANKM123), which when transfected into HEK293 cells lacks the ability to bind to p50 and mediate NF-κB signaling (28). Here, we report a successful application of molecular modeling and in silico screening in the identification of JS6, the first small-molecule PPI inhibitor of Bcl3 and p50, with selective in vivo activity against metastatic progression in models of triple-negative breast cancer (TNBC).

All molecular modeling studies were performed on a MacPro dual 2.66GHz Xeon running Ubuntu 9. All protein crystal structures were downloaded from the PDB data bank (http://www.rcsb.org/). Hydrogen atoms were added to the protein, and the ionization of residues was set appropriate to pH 7.4, using molecular operating environment (MOE; ref. 29), and minimized keeping all the heavy atoms fixed until a root mean square deviation gradient of 0.05 kcal mol/Å was reached. Protein complexes were built with MOE and minimized using the Amber 99 forcefield until a root mean square deviation gradient of 0.05 kcal mol/Å was reached. Molecular dynamics (MD) were performed using Gromacs 4.5 (30) on the Bcl3–p50 complex using Gromos 96 forcefield and constant number, pressure, and temperature working environment. The simulation was conducted at 300 K, 1 atmosphere, and time step of 0.002 ps. The minimized structure was solvated in a periodic dodecahedron simulation box using spc216 water molecules, providing a minimum of 9 Å of water between the protein surface and any periodic box edge. Following minimization, the entire system was equilibrated for 100 ps followed by a production phase of 5 ns. The compound library was downloaded from the SPECS website (http://www.specs.net) and used the Conformation Import function in MOE. The pharmacophore query was generated within MOE. The docking simulations were performed using GLIDE (31), saving three poses per ligand. Rescore was performed using PLANTS 1.1 (32) and Flexx (33) using default parameters.

All chemical reagents and solvents were purchased from Sigma-Aldrich and used without further purification. Reaction progress was monitored by thin-layer chromatography using pre-prepared silica gel plates (Merck Kieselgel 60F₂₅₄), visualized with UV light (254 or 366 nm). Final compound purification was carried out using flash column chromatography (silica gel 40–60 μm, Merck) using the appropriate eluent mixture, or using the automated column chromatography Interchim PuriFlash 4000 system. ¹H NMR and ¹³C NMR spectra were recorded using a Bruker AVANCE (500 MHz) spectrometer autocalibrated to the deuterated solvent reference peak. Chemical shifts (δ) are given in ppm relative to tetramethylsilane, and coupling constants (J values) are in Hz. Signal multiplicities are given as singlet (s), doublet (d), triplet (t), quartet (q), and multiplet (m). Mass spectrometry was performed on a Bruker Daltonics microTOF instrument in positive mode, using electrospray ionization.

To a stirred suspension of anthranilic acid (1; 1.0 g, 7.29 mmol) in pyridine (10 mL) was added 2-fluorobenzoyl chloride (2; 2.54 g, 16.0 mmol). The reaction was stirred at room temperature for 5 hours, and then poured slowly into a 10% solution of sodium carbonate (50 mL). The resulting precipitate was collected by filtration under reduced pressure, and the crude powder was washed three times with hexane to obtain the intermediate 2-(2-fluorophenyl)-4H-benzo[d][1,3]oxazin-4-one (3) after drying, which was used without further purification.

N,N-diisopropylamine (0.72 mL, 0.536 g, 4.15 mmol) and 2-morpholinoethanamine (0.60 mL, 0.594 g, 4.56 mmol) were added to a stirring solution of 2-(2-fluorophenyl)-4H-benzo[d][1,3]oxazin-4-one (3; 0.5 g, 2.07 mmol) in DMF (10 mL) at room temperature. After stirring for 18 hours, the reaction mixture was diluted with water (50 mL), extracted with ethyl acetate (50 mL), washed three times with brine (3 × 50 mL), dried (MgSO₄), and concentrated in vacuo to give crude 2-fluoro-N-(2-((2-morpholinoethyl)-carbamoyl)phenyl)benzamide (JS6). Purification by column chromatography using chloroform:methanol (9:1) as eluent provided pure JS6 as a white powder in 75% yield, with spectroscopic and analytical data corresponding to the purchased sample (www.specs.net). ¹H NMR (CDCl₃) δ 2.50 (4H, t, J = 4.5 Hz, 2xCH₂N), 2.61 (2H, t, J = 5.9 Hz, CH₂N), 3.54 (2H, m, CH₂NH), 3.72 (4H, t, J = 4.5 Hz, 2xCH₂O), 6.98 (1H, m, ArH), 7.14 (1H, t, J = 7.5 Hz, ArH), 7.19 (1H, ddd, J = 11, 8.5, 1.2 Hz, ArH), 7.28 (1H, d, J = 7.5 Hz, ArH), 7.51 (3H, m, ArH), 8.04 (1H, td, J = 7.5, 1.8 Hz, ArH), 8.74 (1H, d, J = 8.0 Hz, NH), 11.80 (1H, d, J = 7.0 Hz,, NH). ¹⁹F NMR (CDCl3) δ 112.48. ¹³C NMR (CDCl₃) δ: 36.01 (CH₂NH), 53.28 (2xCH₂N-morph), 56.64 (CH₂N), 66.94 (2xCH₂O), 116.52 (d, ²J_C-F = 23.38 Hz, ArCH), 121.97 (ArC), 122.44 (ArCH), 122.76 (d, ²J_C-F = 12.38 Hz, ArC), 123.39 (ArCH), 124.62 (d, ³J_C-F = 3.63Hz, ArCH), 126.66 (ArCH), 131.49 (d, ⁴J_C-F = 2.25Hz, ArCH), 132.29 (ArCH), 133.37 (d, ³J_C-F = 8.89Hz, ArCH), 139.00 (ArC), 160.33 (d, ¹J_C-F = 249.63 Hz, ArC-F), 162.30 (C = O), 168.66 (C = O). MS (ESI⁺) 372.2 [M+H]⁺ (C₂₀H₂₂FN₃O₃).

JS6 was prepared for cell culture and in vivo assays by diluting powdered compound in DMSO to a concentration of 100 mg/mL and diluted immediately before use in either cell culture medium (in vitro) or sterile water (in vivo).

The human embryonic kidney cell line HEK-293 (ATCC CRL-1573) was maintained in DMEM (Invitrogen), and the human breast cancer cell lines MDA-MB-231 (TNBC-ATCC HTB-26), MDA-MB-436 (TNBC-ATCC HTB-130), SKBR3 (ATCC HTB-30), BT474 (ATCC HTB-20), and HCC1954 (HER2-CRL-2338) were maintained in Hybi-Care or McCoy's 5a media (HTCC) or RPMI (Invitrogen) in the absence of antibiotics at 37°C according to ATCC guidance. The murine mammary tumor cell line 4T1.2 (TNBC) was a kind gift from R. Anderson (Peter MacCallum Cancer Centre, Australia). Plasmids containing either wild-type (WT) Bcl3 or the Bcl3 ANK M123 nonbinding mutant sequences contained within a FLAG-tagged pcDNA 3.1 backbone vector were gifted from Dr. Alain Chariot (Interdisciplinary Cluster for Applied Genoproteomics, University of Liège, Belgium). In order to achieve Bcl3 knockdown in cell lines, respective cells were transfected with ON-Target plus SMART pool (Dharmacon, Thermo Fisher Scientific) or scrambled siRNA control (32 nmol/L) using lipofectamine RNAiMAX (Invitrogen) according to the manufacturer's protocol. To generate cells overexpressing either Bcl3 WT or Bcl3 ANK M123, cells were transfected using Lipofectamine LTX reagent (Invitrogen). Stably transfected cells were selected by the addition of neomycin (G418, Sigma) to the culture medium for at least 7 days.

Nondenatured cell lysate was diluted with TBS (Calbiochem) supplemented with 0.5% v/v Tween (Sigma) to a concentration of 0.5 to 1 μg/μL, and 100 μL was added onto ANTI-Flag–coated flat-bottom ELISA plates (Sigma). Wells were cultivated at 37°C for an hour followed by 3 × 200-μL washes with TBS/Tween. Primary antibodies, either Bcl3 (1:30 Santa Cruz Biotechnology, sc-185) for indirect ELISA or p50 for sandwich ELISA (1:200 Abcam, ab 7549), were covered from light for an hour at room temperature. Following three 200-μL washes with TBS/Tween, alkaline phosphatase–conjugated secondary antibody was incubated and wells were washed with 3 × 200-μL TBS/Tween. pNPP solution (Santa Cruz Biotechnology) was added (50 μL/well) and cultivated for an hour covered from light at room temperature. The reaction was stopped by addition of 3N NaOH (20 μL/well), and the colorimetric changes were measured at 405 nm using a plate reader.

Adherent cell lines were cultured in 60-mm dishes and then cells were rinsed with ice-cold PBS, scraped off in 1-mL ice-cold PBS, and spun at 4°C. Cell pellets were resuspended in 400 μL of prechilled lysis buffer (20 mmol/L HEPES pH 7.9, 100 mmol/L NaCl, 20% glycerol, 1 mmol/L MgCl₂, 0.5 mmol/L EDTA, 0.1 mmol/L EGTA, and 1% NP-40). The suspension was placed on ice for 30 to 45 minutes and insoluble material pelleted for 10 minutes at 4°C. One hundred microliter protein A-sepharose bead (GE Healthcare Life Sciences) suspension was added to 200 μL of lysates, agitated at 4°C for 30 to 45 minutes, then spun for 3 minutes, and the supernatant was transferred, discarding pelleted beads. Ten microliter of Bcl3 primary antibody and 5 μL of p50 antibody were added to separate aliquots of 100-μL lysate and incubated overnight at 4°C under agitation. Following centrifugation for 3 minutes, beads were washed three times in lysis buffer and respun. The pellet was resuspended in 5x Laemmli buffer and boiled at 95°C for 5 minutes. Twenty-five micrograms of protein was loaded per well of Western blots and incubated with antibodies against Bcl3 (1:200; Proteintech; 23959-1-AP) or p50 (1:500; Cell Signaling Technology; MA 3035) in 5% w/v nonfat milk powder in PBS/Tween. Membranes were incubated in the primary antibody solution overnight at 4°C on a roller.

Cells were seeded into clear-bottom black-walled 96-well plates (Corning Inc.) in antibiotic-free culture media (1–2 × 10⁵ cells/mL). After 24 hours, cells were transfected with NF-κB reporter plasmid (34) as previously described (35). Briefly the NF-κB reporter plasmid pLTRX-Luc (10 ng/well) was used with pcDNA3.1-LacZ plasmid (10 ng/well) to normalize for transfection efficiency. For positive and negative controls respectively, 10 ng of pGL3control or pGL3basic was transfected in place of the reporter plasmid. After 48 hours after transfection, cells were lysed using Glo-lysis buffer and analyzed using Beta-glo and Bright-glo according to manufacturer's protocol (Promega). Luminescence relative to LacZ was read using a Flurostar Optima plate reader (BMG Labtech) and displayed as relative light units.

Single-cell timelapse, Boyden chamber, and wound-healing (scratch) assays were performed as previously described (15). For in situ Boyden chamber assays, cells were seeded in low-serum media in an 8 μmol/L pore chamber coated with Matrigel Basement Membrane Matrix (BD Biosciences; BioCoat). The cell insert was placed into a well with 750 μL of complete growth media containing 10% of serum of a 24-well cell culture insert companion plate (BD Biosciences). Three hundred fifty microliters of cells (2 × 10⁵ cells/mL) was resuspended in normal growth media containing 0.1% serum and plates incubated for 24 hours at 37°C and 5% CO₂. Cells were fixed with 70% ice-cold ethanol and stained with Harris' hematoxylin (Sigma) and 0.5% filtered Eosin (Sigma) for 2 minutes. Membranes were mounted on slides with glycerol gelatin.

CellTiter-Blue cell viability assay (Promega) was performed in 96-well plates using 100 μL of media and 20 μL CellTiter Blue reagent for an hour at 37°C in 5% CO₂.

Cells were seeded at low density (250 cells/mL) in 6-well format in complete growth medium and incubated for 10 days. Quantification of colonies involved aspiration of culture media and washing the cells gently with PBS, followed by fixation of colonies with glutaraldehyde (6% v/v) and staining with crystal violet (0.5% w/v) for 15 minutes at room temperature.

All animal procedures were carried out in accordance with the UK Home Office in compliance with the Animals Scientific Procedures Act of 1986 under the UK HO project license number 3003433. Athymic nude (Hsd:Athymic Nude-Foxn1nu) mice were obtained from Envigo Laboratories. Balb/c-SCID and Balb/c (AnNCrl) mice were obtained from Charles River. Animals were acquired at 6 to 8 weeks of age and maintained in individually ventilated cages (Allentown Inc.) with a 12-hour day/night cycle. Mice received a Teklad global 19% protein extruded rodent diet (Envigo) and water ad libitum. Bilateral subcutaneous orthotopic xenografts were performed with 1 × 10⁶ MDA-MB-436, MDA-MB-231-Luc, and HCC1954-Luc cells suspended in 100-μL RPMI and 100-μL Matrigel (Thermo Fisher Scientific). Bilateral subcutaneous orthotopic allografts were performed with 5 × 10⁵ 4T1.2 cells suspended in 100-μL RPMI. Mice were measured two to three times weekly for tumors- with digital callipers, and volume was calculated using the formula: volume = (length × width²). Experimental metastases were established by i.v. injection of 2 × 10⁵ MDA-MD-231-Luc cells into athymic nudes. Cells were suspended in 100 μL of RPMI media and injected via the tail vein of 8-week-old female mice. Metastatic progression to distal organs was assessed through bioluminescence imaging with the IVIS Spectrum In Vivo Imaging System (Perkin Elmer) and histologic analysis of hematoxylin and eosin–stained paraffin-embedded sections. Prior to imaging, an intraperitoneal (i.p.) injection of 100-μL D-luciferin was administered to each animal. The mice were then anesthetized with 2.5% isoflurane, oxygen mix, and imaged with the charge-coupled IVIS camera device selecting an exposure time of 2 minutes. Luminescence signal was measured through region of interest selection and quantified as total flux. Mice were treated in vivo daily with either JS6 or DMSO control via i.p. injection in sterile water.

In all statistical comparisons, unpaired Student t tests were applied assuming equal variance.

NF-κB transcription factor complexes can both induce and repress gene expression by binding to DNA sequences known as κB elements, leading to the regulation of numerous genes controlling processes such as apoptosis, cell adhesion, proliferation, immunity, and inflammation (35). For this reason, inhibition of the NF-κB signaling pathway has been pursued as a therapeutic strategy for a variety of pathologies, including cancer, although to date there is no selective NF-κB pathway inhibitor approved for human use, largely due to toxicity concerns associated with inhibiting global NF-κB pathway activation (36). For this reason, we sought to develop a more selective approach to pathway inhibition, based on targeting the NF-κB pathway coactivator Bcl3 and its interacting protein partners, given our previous work demonstrating the viability of Bcl3 knockdown in mouse models (14). Bcl3 is known to modulate transcription through binding to the NF-κB family proteins p50 and p52; these key PPIs provided the starting point for our computational design work.

The crystal structure of the ankyrin repeat domain (seven repeats) of Bcl3 was solved previously at a resolution of 1.86 Å (PDB code 1K1A) spanning from residues 119 to 359 (37). The N- and C- terminal domains have not been crystallized to date; however, numerous mutagenesis studies have shown that the ankyrin repeat of Bcl3 is sufficient for interaction with its binding partners (38–41). The structurally most similar member of the IκB family, IκBα, has only six ankyrin repeats, but shares 35% sequence identity over these repeats. The crystal structure of Bcl3 in a complex with its binding partners is not yet available; therefore, we used the structure of IκBα crystallized in a complex with p65/p50 heterodimers as a template for construction of our model within the MOE platform (29). Initially, we prepared the Bcl3/p50 complex by aligning and superposing the crystal structure of the Bcl3 ankyrin repeat (PDB: 1K1A) to the corresponding domain of IκBα (PDB: 1NFI; ref. 42), crystallized in complex with p50. Removal of the IKBα portion revealed the putative complex between Bcl3 and p50. Two molecules of Bcl3 have been observed to bind a single p50 homodimer (43); therefore, a second molecule of Bcl3 was added using pseudo-dyad symmetry of the p50 homodimer as published previously (Fig. 1A; ref. 44). The model was then refined using a short MD simulation (5 ns) to relax the contacts between Bcl3 and p50 (30). The Bcl3/p50 dimer interface exhibits an extensive network of hydrophobic interactions and hydrogen bonds within ankyrin repeats 4 and 7. Importantly, we observed a hydrophobic binding pocket within ankyrin repeats 6 and 7. This observation was of particular interest as IкBα does not share structural similarity within this area; therefore, it can serve as a unique target on Bcl3. The interacting residues within the binding pocket contain p50 residues (mainly Lys 275, Asp 297, Ser 299, Pro 300, Thr 301, Asp 302, Val 303, His 304, and Arg 305) interacting with Bcl3 residues from the ankyrin repeat 6 (Met 298, Tyr 299, Ser 300, Gly 301, Ser 302, Ser 303, His 306, Ser 307) and unique ankyrin repeat seven (Asn 331, Cys 332, His 333, Asn 334, Asp 335, Val 340, and Arg 342).

The final state of the MD simulation was energy minimized and used as the input protein structure in a virtual screening protocol. As a binding site for molecular docking, we selected the interface between p50 and ankyrin repeats 6 and 7 of Bcl3, corresponding to Pro300 and Thr301 of p50 (Fig. 1B). The first step of the virtual screening was the filtration of the SPECS library of compounds (∼360,000 structures) through a pharmacophore query developed on the interaction between p50 and Bcl3 in the selected pocket (Fig. 1C). The resulting 27,013 structures were refined by selecting the molecules with a molecular weight between 200 and 450, and clustering the remaining compounds based on a minimum of 95% identity (Tanimoto index). Molecular docking, using GLIDE (31), was then performed on the 16,058 remaining structures. Docking results were rescored using the scoring functions of two other types of docking software (Flexx and PLANTS), and the 121 compounds that were present within the top 10% of all the different scoring ranking were visually inspected (42, 43). This process led to the selection of 10 molecules (denoted as JS1-JS10) that appear to fit the pocket in a very similar way as the corresponding p50 residues (Fig. 1D).

The 10 shortlisted compounds from the computational screening and docking studies were purchased from the SPECS library and assessed for their ability to inhibit Bcl3:p50 PPI using an ELISA-based assay (Fig. 2). Initially, we assessed the effect of the virtual hit compounds on cellular Bcl3 levels. HEK293 cells overexpressing FLAG-tagged Bcl3 (Supplementary Fig. S1A and S1B) were incubated with each of the test compounds at 10 μmol/L or vehicle control for 24 hours, comparing with untreated HEK293 cells expressing either a dominant negative FLAG-tagged Bcl3 ANK mutant (28) or empty vector control (-Bcl3). ELISA assay results on cell lysates, using an immobilized FLAG-tag antibody and a specific Bcl3 detection antibody, are shown (Fig. 2A). None of the test compounds caused a significant decrease in cellular Bcl3 levels compared with untreated control.

Next, we investigated the ability of the candidate compounds to specifically interfere with the PPI between Bcl3 and p50, using an NF-κB (p50) antibody–based sandwich ELISA assay to detect p50:Bcl3–FLAG complexes (Fig. 2B). The ELISA data showed that 4 of the 10 compounds identified from computational screening significantly inhibited p50:Bcl3 PPI, with compound 6 (JS6) most effectively inhibiting the interaction, equivalent to the ANK binding mutant negative control. The specific reduction in p50:Bcl3 PPI was confirmed by an immunoprecipitation assay of the breast cancer cell line MDA-MB-231 pretreated with JS6 for either 30 minutes or 2 hours using polyclonal antibodies to endogenous Bcl3 and p50 (Fig. 2C). A reduction in p52:Bcl3 PPI was also observed following 2-hour incubation with JS6, but not at 30 minutes, confirming the modeling prediction that both p50 and p52 interactions are disrupted, but suggesting a marginal difference in kinetics of inhibition for the two NF-κB subunits (Supplementary Fig. S1C). Treatment with JS6 for 24 hours was also found to be noncytotoxic when tested in nontumorigenic HEK293 (Fig. 2D) and breast MCF10A cells (Supplementary Fig. 1D), whereas breast cancer MDA-MB-231 TNBC cells exhibited a small but statistically insignificant reduction in cell viability according to the CellTiter-Glo luminescent cell viability endpoint assay (Fig. 2D and E). The noninhibitory compound JS1, which was confirmed to have no inhibitory effect on p50:Bcl3 PPI by coimmunoprecipitation assay (Supplementary Fig. S1D), was used in subsequent assays as a negative control for JS6.

Having established the ability of compound 6 (JS6) to block Bcl3:p50 binding interactions, we proceeded to investigate the effects of JS6 on downstream cellular events. To determine the effects of JS6 on Bcl3-dependent NF-κB signaling, we made use of an NF-κB–responsive luciferase reporter assay (35), following JS6 treatment (10 μmol/L, 24 hours) of the following human Bcl3 overexpressing cell lines: HEK293; HEK293 co-overexpressing NF-kB1/p52; and the breast cancer cell line MDA-MB-231 (Fig. 3). Overexpression of Bcl3 in all cell lines resulted in an expected increase in NF-κB reporter activity, which was sequestered by coadministration of JS6 in a dose-dependent manner (Fig. 3A–F). Overexpression of NF-kB1/p52 in HEK293 cells, which lack endogenous p52, augmented Bcl3-mediated NF-kB activity (Supplementary Fig. S1E). This p52-mediated response was also sequestered by JS6 (Fig. 3C) but was not affected by the noninhibitory compound JS1 (Supplementary Fig. S1F). Half-maximal responses to JS6 were obtained in the range of 45 to 710 nmol/L across the three cell lines.

To confirm the specificity of JS6 on Bcl3-mediated transcription, we compared the responses of nine genes previously identified to be either positively or negatively regulated by NF-κB/Bcl3 (45–47) by treating MDA-MB-231 cells with JS6 or Bcl3-specific siRNA (Supplementary Fig. S1F). The transcriptional signature of the nine gene panel from JS6-treated cells closely resembled that of the Bcl3-siRNA control, with seven of the nine genes exhibiting similar amplitudes of response (Fig. 3G). Furthermore, JS6 led to a small but significant decrease in Cdc42 (Supplementary Fig. S1G), which we had recently identified as a novel Bcl3 gene target mediating the promigratory effects of Bcl3 in breast cancer cells (11).

To establish whether these transcriptional responses to JS6 manifested in a concomitant reduction in multimodal cell migration, thus phenocopying the effects of Bcl3-siRNA previously observed in breast cancer cells (11, 14, 15), we investigated the effect of JS6 in single-cell (transwell) assay (Fig. 4A; Supplementary Fig. S2A–S2C), single-cell (mesenchymal) assay (Fig. 4C; Supplementary Fig. S1D), and collective sheet (wound-healing) assay (Fig. 4D–F). JS6 significantly suppressed all modes of cell migration. Single-cell motility, the preferred migratory state of MDA-MB-231 and MDA-MB-436 cells (11), was most profoundly affected by JS6, with a half-maximal dose of 310 nmol/L relative to the dominant negative Bcl3 ANK-mutant and siRNA-mediated knockdown of Bcl3 expression (Fig. 4A–C). Similar responses were observed with collective sheet migration of mesenchymal-like TNBC cells and the HER2-positive cell line BT474 (Fig. 4D–F). The noninhibitory compound JS1 failed to demonstrate effects on cell migration in these cell lines (Fig. 4D–F). This supports the previously described nonredundant role for Bcl3 in multimodal cell motility phenotypes (11) and confirms efficacy of JS6 in both TNBC- and HER2-positive breast cancer cells.

Consistent with its role in cell migration, we and others have previously demonstrated that Bcl3 plays a key role in the metastatic progression of breast cancer (14, 15). Thus, following establishment of the Bcl3–p50-inhibitory activity of JS6, leading to suppression of cellular NF-κB signaling and antimigratory phenotypes, we wished to investigate the potential for this compound to disrupt the metastatic process in vivo, using murine models of metastatic breast cancer. Given the requirement for larger (gram scale) quantities of JS6 for in vivo work at this stage, we additionally developed a straightforward two-step synthetic route to JS6 from commercially available starting materials (Fig. 5A). This allowed analytical and spectroscopic comparison between synthetic and commercially available samples for further authentication and validation of results. In addition, the synthetically prepared sample was assessed in the in vitro migration assay described above, showing a comparable activity to the commercially procured sample.

Reaction of anthranilic acid (1) with 2-fluorobenzoyl chloride (2) in basic solvent (pyridine) at room temperature gives the intermediate 2-(2-fluorophenyl)-4H-benzo[d][1,3]oxazin-4-one (3) in 65% yield. Nucleophilic ring opening of intermediate (3) with 2-morpholinoethan-1-amine and N,N-di(isopropyl)ethylamine (DIPEA) in dimethylformamide (DMF) at room temperature gave pure 2-fluoro-N-(2-((2-morpholinoethyl)-carbamoyl)phenyl)benzamide (JS6) in 75% yield following final purification using column chromatography.

Using an established model of experimental metastasis, MDA-MB-231 cells were inoculated intravenously into immunocompromised mice. JS6 (3.5 mg/kg) or vehicle control (1% DMSO) was given as a single daily i.p. injection for 10 days immediately following tumor cell inoculation and scanned for luciferase activity over the following 49 days (Fig. 5B). Twenty-eight days after inoculation of tumor cells, the majority of JS6-treated mice were essentially devoid of detectable lung metastases (Fig. 5C and D) representing an 83% reduction in overall thoracic tumor burden in the JS6-treated cohort compared with untreated mice. After a further 28 days, a minority of JS6-treated mice (31% compared with 92% untreated) subsequently developed thoracic lesions that were on average 41% smaller than extant untreated tumors (Fig. 5E and F).

This reduction in the establishment and subsequent expansion of disseminated tumors was also observed in breast tumors seeded orthotopically. Two TNBC cell lines, MDA-MB-436 (Fig. 5G–J) and MDA-MB-231 (Supplementary Fig. S3A–S3D), and an HER2-positive breast cancer cell line HCC1954 (Supplementary Fig. S3F and S3G) were inoculated into the mammary fat pads of recipient mice receiving daily doses of 3.5 mg/kg JS6 for up to 60 days following tumor seeding. Systemic JS6 treatment resulted in a 66% reduction in MDA-MB-436 tumor volume (Fig. 5H and I), a 38% reduction in MDA-MB-231 tumor volume, and a 48% reduction in HCC1954 tumor burden (Supplementary Fig. S2) with no overt signs of toxicity (Supplementary Fig. S4). This reduction in tumor growth correlated with a decrease in tumor cell turnover index as measured by the relative extent of apoptosis (cleaved caspase 3) and mitosis (phospho-histone H3) within the tumors (Fig. 5J; Supplementary Fig. S3D), with the major component of this response being the effect of JS6 on apoptosis rates.

The effects of JS6 on experimental seeding and clonal expansion of tumor cells in vivo in xenograft models were confirmed in an immune-competent murine model of spontaneous metastasis (Fig. 6A). The metastatic murine mammary carcinoma cell line 4T1.2 (48) was orthotopically transplanted into the mammary fatpads of recipient Balb/c mice and JS6 administered by intraperitoneal route at 3.5 and 10 mg/kg daily for the duration of the experiment. Mammary tumor growth was reduced by 58% (±3%) and 87% (±5%) respectively, consistent with the reduction in tumor growth seen in human TNBC xenografts. Moreover, spontaneous metastatic tumor burden arising from the primary orthotopic tumor and disseminated to the lungs was also inhibited by JS6 administration (Fig. 6C), with a 38% (±2%) reduction in metastatic tumor burden at 3.5 mg/kg and a 93% (±4%) reduction in metastatic tumors at 10 mg/kg JS6. This demonstrates the efficacy of JS6 on inhibiting spontaneous tumor metastasis, equivalent to the prevention of circulating tumor cell dissemination observed in xenograft models (Fig. 5). Furthermore, the average size of metastatic lesions detected in lung tissues was significantly affected by JS6, with an approximate 50% reduction in the size of 10 mg/kg JS6-treated tumors (Fig. 6D). Moreover, the demonstration of JS6 efficacy in murine mammary tumors in vivo confirms tumor specificity and the lack of toxicity observed in mice treated daily with JS6 (Supplementary Fig. S4).

The effect of JS6 on tumor seeding and clonal expansion was confirmed in colony formation assays in vitro (Fig. 6E; Supplementary Fig. S2E). JS6 exhibited a dose-dependent inhibition of colony formation and growth in MDA-MB-231 and MDA-MB-436 cell lines. Significant inhibition of colony formation in vitro was observed in the micromolar range of JS6, which likely reflects a requirement to maximally load cells with inhibitor in order to affect long-term tumor cell proliferation and is consistent with the relative impact of JS6 on tumor metastasis in vivo. These data combined are consistent with previous observations of the effects of Bcl3 inhibition on cancer stem–like cells in colorectal cancer cell lines (22) and confirm antitumor effects of a novel Bcl3 inhibitor which affects in vivo tumor cell seeding and colonization.

In this article, we report the anthranilic acid derivative JS6 as the first small-molecule Bcl3 inhibitor in metastatic breast cancer models. JS6 was identified via virtual library screening of a computational Bcl3/p50 protein–protein interface model (Fig. 1). We provide evidence of its efficacy as a Bcl3 (ankyrin repeat domain)/p50 inhibitor in vitro (Figs. 2 and 3), and as an inhibitor of downstream Bcl3 cellular functions and metastatic progression in animal models in vivo (Figs. 5 and 6). Consistent with the growing evidence of Bcl3′s role in a variety of solid tumors (14–21), this study highlights a dual role for Bcl3 in tumor progression, promoting both tumor growth and metastasis. Although this likely reflects the pleiotropic effects of NF-kB signaling more widely, it remains to be determined whether the impact of inhibiting Bcl3 is mediated primarily through one or both of the NF-kB canonical/noncanonical pathways or indeed whether there are alternative functions of Bcl3 as yet to be determined. What is clear however is that targeting Bcl3 in this manner is well tolerated (Supplementary Fig. S4; refs. 14, 23).

These data show that it is possible to use a multimodular computational screening approach to identify a relatively high proportion of functional hit compounds, with in vivo efficacy. The in vivo data presented are consistent with the concept of JS6 as a lead compound with optimal activity in the prevention of metastatic outgrowth, as opposed to shrinkage of established metastases. Although it is intuitive that a drug candidate should have optimal activity at the early stages of metastatic disease, this does present some particular challenges in moving forward toward clinical development. First, the activity of JS6 (or a related analog) in clinic development might be optimally targeted to cancer types characterized by early metastatic progression. These might include short latency cancer types such as triple-negative and HER2-positive breast cancer, or other cancer types where Bcl3 plays a key role and where metastatic progression occurs within a few months rather than years (e.g., pancreatic and nasopharyngeal cancer). The requirement for “short latency,” poor prognosis tumor types reflects the realities, costs, and timescales of clinical development.

The compound discovered in this study (JS6; molecular weight = 371.4) presents a number of appealing molecular properties from a drug discovery and development standpoint. As shown in Fig. 5A, JS6 can be readily synthesized in two high-yielding chemical steps at room temperature from commercially available and inexpensive starting materials. This is an important consideration when progressing new drug candidates into full preclinical development, where multigram quantities of pure material are needed for ongoing in vivo efficacy and toxicology studies. JS6 is fully compliant with the common “rules of thumb” used by pharmaceutical chemists to guide drug candidate selection, such as the Lipinski rule-of-five for prediction of oral activity (49). In addition, JS6 does not present solubility issues within the in vitro and in vivo delivery vehicles used to date. The utility of the morpholine group as a popular option for improving water solubility of drug candidates (either as free base or salt form) is well established in drug discovery laboratories (50). Finally, the chemical structure of JS6 provides no obvious toxicological flags (e.g., functional groups associated with metabolic instability or genotoxicity) or promiscuous artifact groups known as pan-assay interference compounds (51). Overall, the molecular properties of JS6 provide a basis for efforts to expand this work toward a first-in-class clinical candidate for patient administration.

J. Soukupová reports a patent for WO2015014972 A1 issued to Cardiff University. R.W.E. Clarkson, A.D. Westwell, and A. Brancale report grants and personal fees from Tiziana Life Sciences, grants from Cancer Research Wales, other from Breast Cancer Research Aid, and nonfinancial support from Life Sciences Research Network and Wales Cancer Research Centre during the conduct of the study; in addition, they have a patent for WO2015014972 A1, "2-Benzoylaminobenzamide derivatives as bcl-3 inhibitors," issued to Cardiff University. No disclosures were reported by the other authors.

J. Soukupová: Formal analysis, investigation, methodology. C. Bordoni: Investigation, methodology. D.J. Turnham: Investigation, methodology. W.W. Yang: Investigation, methodology. G. Seaton: Investigation. A. Gruca: Investigation, methodology. R. French: Investigation, methodology. K.Y. Lee: Investigation. A. Varnava: Investigation. L. Piggott: Formal analysis, investigation, project administration. R.W.E. Clarkson: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing. A.D. Westwell: Conceptualization, formal analysis, supervision, investigation, methodology, writing–original draft, writing–review and editing. A. Brancale: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing.

This work was supported by Cardiff University, through a President's Scholarship PhD Studentship to J. Soukupová, and through infrastructure support to the School of Pharmacy and Pharmaceutical Sciences (A. Brancale and A.D. Westwell) and European Cancer Stem Cell Research Institute/School of Biosciences (R.W.E. Clarkson). The authors thank Tiziana Life Sciences for PhD studentship support to D. Turnham and W.W. Yang; and postdoctoral funding to A. Gruca. Funding for a PhD studentship to C. Bordoni was provided by Cancer Research Wales. Additional running costs were supported by donations from the Breast Cancer Research Aid (UK charity number 1166674). They also acknowledge infrastructure support from the Wales Cancer Research Centre (to L. Piggott) and the Life Science Research Network Wales.

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.