Identification of a small-molecule compound that inhibits homodimerization of oncogenic NAC1 protein and sensitizes cancer cells to anticancer agents

Nucleus accumbens–associated protein-1 (NAC1) is a transcriptional repressor encoded by the NACC1 gene, which is amplified and overexpressed in various human cancers and plays critical roles in tumor development, progression, and drug resistance. NAC1 has therefore been explored as a potential therapeutic target for managing malignant tumors. However, effective approaches for effective targeting of this nuclear protein remain elusive. In this study, we identified a core unit consisting of Met7 and Leu90 in NAC1's N-terminal domain (amino acids 1–130), which is critical for its homodimerization and stability. Furthermore, using a combination of computational analysis of the NAC1 dimerization interface and high-throughput screening (HTS) for small molecules that inhibit NAC1 homodimerization, we identified a compound (NIC3) that selectively binds to the conserved Leu-90 of NAC1 and prevents its homodimerization, leading to proteasomal NAC1 degradation. Moreover, we demonstrate that NIC3-mediated down-regulation of NAC1 protein sensitizes drug-resistant tumor cells to conventional chemotherapy and enhances the antimetastatic effect of the antiangiogenic agent bevacizumab both in vitro and in vivo. These results suggest that small-molecule inhibitors of NAC1 homodimerization may effectively sensitize cancer cells to some anticancer agents and that NAC1 homodimerization could be further explored as a potential therapeutic target in the development of antineoplastic agents.

Nucleus accumbens-associated protein-1 (NAC1) is a transcriptional repressor encoded by the NACC1 gene, which is amplified and overexpressed in various human cancers and plays critical roles in tumor development, progression, and drug resistance. NAC1 has therefore been explored as a potential therapeutic target for managing malignant tumors. However, effective approaches for effective targeting of this nuclear protein remain elusive. In this study, we identified a core unit consisting of Met 7 and Leu 90 in NAC1's N-terminal domain (amino acids 1-130), which is critical for its homodimerization and stability. Furthermore, using a combination of computational analysis of the NAC1 dimerization interface and high-throughput screening (HTS) for small molecules that inhibit NAC1 homodimerization, we identified a compound (NIC3) that selectively binds to the conserved Leu-90 of NAC1 and prevents its homodimerization, leading to proteasomal NAC1 degradation. Moreover, we demonstrate that NIC3-mediated down-regulation of NAC1 protein sensitizes drug-resistant tumor cells to conventional chemotherapy and enhances the antimetastatic effect of the antiangiogenic agent bevacizumab both in vitro and in vivo. These results suggest that small-molecule inhibitors of NAC1 homodimerization may effectively sensitize cancer cells to some anticancer agents and that NAC1 homodimerization could be further explored as a potential therapeutic target in the development of antineoplastic agents.
Overexpression or mutation of oncogenes and loss of tumorsuppressor genes are important events in cancer development and progression (1,2). Because transcription is a critical control point in the production of many proteins, alterations of oncogenic or tumor suppressor proteins often involve dysregulated transcription (3). Indeed, certain transcription factors have been found to be in excess amount or overactive in various types of cancer and contribute to survival, unrestrained replication, and apoptosis escape in cancers (4). Therefore, inhibiting the activity of oncogenic transcription factors is considered as a promising strategy in anticancer treatment. However, many of those transcription factors seen in human cancers are considered "undruggable" due to their large protein-protein interaction interfaces and their lack of deep protein pockets and intracellular localization. For drugging oncogenic transcription factors, several new approaches have been developed to overcome the undruggable nature of these transcription factors (5). For instance, the BRD4-based small-molecule bromodomain inhibitor (JQ1) can directly block MYC transcription and Mycdependent transcriptional target genes; the conformationdisrupting Aurora A inhibitors can destabilize the complex of Aurora A with Myc and promote the degradation of Myc protein (6,7). Two compounds, IS3-295 and S3I-201, were reported to exert their antitumor activity partially through blocking the Stat3-mediated DNA-binding activity (8,9). Molecular docking and synthesis of novel quinazoline analogues were shown to inhibit NF-B-dependent gene transcription (10).
It is known that exposure of the hydrophobic interface of a dimeric protein may lead to conformational change, causing destabilization and degradation of a protein via proteasomal or autophagic pathways (24 -26). In this study, we identified a core unit consisting of Met 7 and Leu 90 in the N-terminal domain (amino acids 1-130) of NAC1, which is critical for its dimerization and stability. To test our hypothesis that inhibiting the dimerization of NAC1 can destabilize NAC1 protein and promote its degradation, we searched for chemicals able to inhibit the homodimerization of NAC1 using a combination approach of computational analysis of the dimerization interface and high-throughput screening. Here, we report a compound, NIC3, that has the ability to selectively bind with the conserved Leu 90 of NAC1 and to inhibit NAC1 dimerization, resulting in proteasomal degradation of the NAC1 protein. We further assessed the therapeutic potential of NIC3 in combination therapy. We showed both in vitro and in vivo that down-regulation of NAC1 protein by NIC3 significantly overcame tumor cell resistance to conventional chemotherapy and enhanced antimetastatic efficacy of the antiangiogenic agent bevacizumab. The results of this study not only underscore the potential of NAC1 as an anticancer target but also demonstrate the therapeutic benefits of the small-molecule inhibitors of NAC1 dimerization in cancer treatment.

Analysis of dimerization domain and residues of NAC1
The conserved BTB/POZ domain is essential for NAC1 dimerization, which plays important roles in tumor development (11); however, no evidence has been provided about the mechanisms and biologic consequences of NAC1 dimerization. To study this, we first assessed the association of the two forms of NAC1 proteins tagged with either V5 or Myc epitopes by co-IP after their coexpression in HEK-293T cells. Fig. 1A demonstrates the association of V5-NAC1 with Myc-NAC1 in the cells. To further assess the interaction between two tagged NAC1 proteins, we used bacterially expressed GST-NAC1 pro-tein in an in vitro dimerization assay. Fig. 1B shows that with increasing concentrations of disuccinimidyl suberate (DSS), a noncleavable bivalent chemical cross-linker that is commonly used to detect direct protein-protein interactions, the intensity of NAC1 monomers was reduced gradually accompanied by the appearance of the expected NAC1 homodimers. NAC1 homodimers could not be detected in the bacterially expressed NAC1(⌬N130) protein (BTB/POZ domain deletion) (Fig. 1C).
To identify the sites in BTB/POZ domain required for association between the two tagged NAC1 proteins, contacts between atoms of the protein were examined using the Protein Contacts module in Molecular Operating Environment (MOE). Among them, one dimerization core unit consisting of four residues, Met 7 and Leu 90 from one chain and the same residues from another chain, appears to be crucial for NAC1 dimerization (Fig. S1, A and B). To further delineate the role of binding residues (Met 7 and Leu 90 ) in dimerization, using a site-directed mutagenesis method we generated two different tagged NAC1 mutants in which Leu 90 sites were altered to Arg. We found that single Leu 90 mutation decreased NAC1 dimer formation, and double Leu 90 mutations further decreased NAC1 dimerization (Figs. 1D and S2A). Fig. S2B shows that with increasing concentrations of DSS the intensity of NAC1 monomers was reduced gradually and was accompanied by the appearance of the expected NAC1 homodimers; NAC1 homodimers could not be detected in the NAC1(S91A) protein. Introduction of Leu 90 mutation led to a significant reduction of NAC1 protein amount in HEK-293T cells, and reduction of NAC1 protein caused by Leu 90 mutation could be rescued by the proteasome inhibitor MG132 (Fig. 1E). The pulse-chase experiments demonstrated that NAC1 mutant (L90A) had a shorter half-life than WT NAC1 (4.6 h versus 20 min) (Fig. 1F). Notably, ubiquitination of the NAC1 mutant (L90A) was significantly increased as compared with WT NAC1 (Fig. S2C). These results suggest that the dimerization core unit of NAC1 consisting of Leu 90 and Met 7 is critical for the formation of stable NAC1 dimers; disrupting this core unit would suppress NAC1 dimerization and destabilize this protein.

Identification of NIC3 as an inhibitor of NAC1 dimerization
To search for small-molecule compounds able to target the dimeric interface of NAC1, we screened ϳ200,000 compounds using an in silico docking approach ( Fig. 2A). Of the 50 hits, one compound, NIC3 (Fig. 2B), was chosen for the current study based on the molecular docking showing that the nitrogen atoms in NIC3 form hydrogen bonds with the critical dimerization core residue Leu 90 , and this confers this compound druglike properties (Figs. 2, C-E, and S3). Validation of NIC3 as an inhibitor of NAC1 dimerization was carried out using co-IP assay. We showed that NIC3 could abolish co-IP of ectopic NAC1 protein tagged with V5 or Myc epitopes (Fig. 2F). Using the purified NAC1 protein (GST-NAC1) and nondenaturing PAGE analysis, we also showed the inhibition of formation of NAC1 dimerization by NIC3 in vitro (Fig. 2G). In these experiments, we incubated the bacterially expressed proteins with increasing concentrations of NIC3 and demonstrated that NAC1 dimerization was suppressed in a dosedependent manner.

Inhibitor targeting NAC1 dimerization
To determine how NIC3 docks into NAC1, we prepared several chemical probes (Fig. S4). Structure-activity relationship studies revealed that nitrogen atoms are essential for inhibition of the NAC1 dimerization as the NIC3 treatment was able to dissociate the dimeric NAC1 into monomers, whereas analogue 1 could not inhibit the formation of NAC1 dimerization and decrease NAC1 level (Fig. 3, A-C). Based on this information, we synthesized a positive probe (PP) and a negative probe (NP) with biotin and an ester bond linker (Fig. 3D). We found that the biotin-tagged positive probe possessed the ability to effectively inhibit NAC1 expression, whereas the biotin-tagged negative probe did not have this ability at the same concentration (Fig. 3E), and reduction of NAC1 protein caused by PP could be rescued by the proteasome inhibitor MG132 (Fig. 3F). Furthermore, NAC1 dimerization was inhibited, and ubiquiti-nation of the NAC1 protein was significantly increased by PP (Fig. 3, G and H). In co-IP experiments, we detected the association of NAC1 with PP but not with NP (Fig. 4A). To determine whether NIC3 directly binds to NAC1, we generated a bacterially expressed GST-NAC1 protein. The purified GST-NAC1 was incubated with increasing concentrations of PP bound to biotin-GSH particles followed by immunoblotting with biotin antibody. As shown in Fig. 4B, the intensity of GST-NAC1 increased with increasing concentrations of PP. However, NIC3 lost its inhibitory effect on NAC1 dimerization when the nitrogen atoms were replaced with oxygen atoms (Fig. 3B). Molecular docking demonstrated that Leu 90 , Ser 91 , and Met 92 form hydrogen bonds with the nitrogen atoms of NIC3 (Fig. 2, C-E). To determine which of these residues are critical for the binding with NIC3, we mutated these leucine, serine, and methionine residues of Inhibitor targeting NAC1 dimerization NAC1 to alanine residues ( Fig. 4, C and D). Among these mutants, we found that L90A mutant completely lost the ability to bind to NIC3, whereas S91A and M92A retained binding to NIC3 in a manner similar to WT NAC1 (Fig. 4, C and D). To determine the effects of those mutations on NAC1 structure, we calculated and compared the backbone root-mean-square deviations between the WT NAC1 and each of the mutants. None of the mutants showed significant structural changes, suggesting that neither L90A nor other NAC1 mutants tested in the studies have significant effects on the NAC1 monomer structure as compared with the WT (Fig. S1, C and D). These results imply that NIC3 can efficiently and specifically dock into NAC1 and that Leu 90 of NAC1 is essential for its interaction with NIC3.

NIC3 accelerates the proteasomal degradation of NAC1 protein
As the dimerization of NAC1 contributed to its stability ( Fig.  1), we next wanted to know whether inhibiting NAC1 dimerization by NIC3 would facilitate proteasomal degradation of NAC1 protein. In these experiments, we first treated HeLa cells with a series of concentrations of NIC3 for different periods of time and then analyzed the level of endogenous NAC1 protein using immunoblotting. Fig. 5, A and B, show that NIC3 treatment caused a reduction of NAC1 protein in a dose-and timedependent manner. Similar effects of NIC3 on the NAC1 level were observed in SKOV3 cells (Fig. S2D). The mRNA level of NAC1 was not affected by NIC3 treatment, as measured by quantitative RT-PCR (Fig. 5C). NIC3 treatment also reduced the level of ectopic V5-tagged NAC1 in HEK-293T cells (Fig.  5D). The reduction of NAC1 protein in the NIC3-treated cells could be rescued by the proteasome inhibitor MG132 (Fig. 5E). Fig. 5F shows that the turnover of NAC1 protein was greatly facilitated in the cells treated with NIC3 as compared with that in the control cells (t1 ⁄ 2 , 53 versus 560 min), as measured by pulse-chase experiments. Additionally, NIC3 treatment enhanced the ubiquitination of NAC1 protein (Figs. 5G and S2C). To confirm these findings, we examined the effects of NIC3 on the NAC1-regulated proteins HIF-1␣ and Gadd45-interacting pro- A, flowchart of the number of candidate compounds identified at each step, resulting in 50 candidate compounds after virtual screening. B, the structure of NIC3. C, 3D binding mode of the receptor-ligand complex with ligand colored in cyan, receptor atoms in green, and backbone in magenta. D, the tentative surfaces of the binding sites on NAC1 of which exposed, polar, and hydrophobic areas are colored in red, magenta, and green, respectively. E, 2D binding mode of the same complex. F, HEK-293T cells were transfected with a V5-NAC1 plasmid and Myc-NAC1 plasmid for 12 h and then treated with increasing concentrations of NIC3 for another 24 h. Cell lysates were immunoprecipitated with an anti-Myc antibody followed by immunoblotting (IB) for V5. G, purified GST-NAC1 protein was incubated with increasing concentrations of NIC3. Cell lysates were resolved by nondenaturing PAGE and analyzed by Western blotting using an anti-GST antibody.  Inhibitor targeting NAC1 dimerization tein 1 (16, 20, 23). We found that NIC3 treatment indeed affected HIF-1␣ and Gadd45-interacting protein 1 (Fig. S9), further supporting the role of NIC3 in accelerating the ubiquitin-proteasomal degradation of NAC1 protein via prohibiting its dimerization.

NIC3 overcomes chemoresistance in vitro and in vivo
Conventional chemotherapy is an important part of treatment for many cancers, but its effectiveness is often limited by acquired drug resistance (27). Previous studies showed that high expression of NAC1 was closely associated with chemotherapy resistance and tumor recurrence (11,22). Therefore, we determined the effects of NIC3 on tumor-cell sensitivity to cisplatin and Adriamycin, two commonly used anticancer drugs, in drug-resistant cancer cell lines. NIC3 itself did not show any cytotoxicity in the cancer cell lines until the concentration reached 100 -1000 M (Fig. S5). To test whether NIC3 can sensitize drug-resistant tumor cells to cisplatin or Adriamycin, we used the sublethal concentration of 20 M, which can inhibit NAC1 expression significantly. The human MCF-7/ ADR (Adriamycin-resistant) and HeLa/DDP (cisplatin-resis-tant) cell lines were used in these experiments. We observed that combination treatment of NIC3 with Adriamycin or cisplatin significantly augmented the cytotoxic effect in the resistant cancer cell lines (Fig. 6, A-D). The increased cytotoxicity seen in the cells treated with the combination of cisplatin or Adriamycin with NIC3 was associated with enhanced apoptosis, as evidenced by elevated levels of cleaved PARP (Fig. 6, E  and F). We also showed that the combination of cisplatin or Adriamycin with NIC3 resulted in decreased cell proliferation and increased apoptotic cell death in the human ovarian cancer cell line SKOV3 (Fig. S6, A-C). By contrast, NIC3 did not show a sensitizing effect in ES-2, an ovarian cancer cell line that has low intrinsic NAC1 expression (Fig. S7, A and B). Furthermore, NIC3 can also reverse drug resistance in an animal xenograft model. In mice bearing MCF-7/ADR cells (inoculated subcutaneously), NIC3 treatment alone had no effect on tumor growth; Adriamycin treatment alone showed a moderate inhibitory effect on the growth of the xenografted tumors (Fig. 6G). In contrast, the combination of NIC3 with Adriamycin significantly suppressed the growth of the xenografted tumors  ( Figs. 6G and S8A). Concomitantly, as compared with either vehicle control, NIC3, or Adriamycin alone, the combination treatment of NIC3 with Adriamycin caused a significant increase in the terminal deoxynucleotidyltransferase dUTP nick end labeling (TUNEL) staining-positive cells and a decrease in Ki67, a proliferation marker (Fig. 6H), and in NAC1 protein expression (Figs. 6H and S8B). No significant changes in body weight were observed in the mice treated with the combination therapy (Fig. 6G). These results suggest that NIC3 is synergistic with cisplatin or Adriamycin in suppressing cancer cell growth, which is associated with induction of apoptotic cell death in drug-resistant tumor cells both in vitro and in vivo.

NIC3 potentiates the antimetastatic effect of bevacizumab in a metastatic breast cancer model
Bevacizumab is an antiangiogenic agent used in treating metastatic breast cancer (28). However, hypoxia environment or cancer stem cell enrichment induced by bevacizumab may limit its antimetastatic efficacy (29 -31). We and other groups previously showed that silencing NAC1 could mitigate the hypoxic microenvironment (20) and regulate stem cell differentiation (32,33). Therefore, we next determined whether NIC3 could reinforce the antimetastatic effect of bevacizumab. An experimental metastatic breast cancer model established by intravenous injection of luciferase-labeled MDA-MB-231 cells was used. We observed that the combination treatment of bevacizumab with NIC3 significantly reduced the lung and bone disseminations of the tumor-bearing mice as compared with either control, NIC3, or bevacizumab alone, as evidenced by the luciferase expression in the thoracic cavity and in the hind limbs (Fig. 7, B and E). H&E staining of the sections corresponding to this region confirmed the decrease of lung metastasis (Fig. 7, C and D). Additionally, the enhanced antimetastatic effect of this combination treatment significantly extended the

Inhibitor targeting NAC1 dimerization
survival of the tumor-bearing mice (Fig. 7F). These observations suggest that NIC3 could be used as a sensitizer in antimetastatic therapy.

Discussion
Because of the critical role of NAC1 in oncogenesis, therapeutic targeting of this protein has been explored as a potential strategy for cancer treatment. For example, to inactivate NAC1, Wu et al. (34) developed a series of self-inhibitory peptides that block NAC1 dimerization by rebinding at the dimerization interface. Although this method might be an effective approach to inhibiting NAC1, the biological effects of these peptides and physicochemical challenges in the application of those peptides in vivo remain obscure. Thus, nonpeptidic molecules capable of targeting NAC1 appear to be better candidates. It is known that NAC1 homodimerization is critical for tumor cell proliferation and survival; however, the mechanism by which NAC1 dimerization occurs and its biologic consequences are unclear (11). In the current study, we first focused on the molecular basis of NAC1 protein dimerization (Figs. 1 and S1). We show that this type of homodimerization is important for the stability of NAC1 protein (Figs. 1 and S2C). Furthermore, using computational analysis of the dimerization interface and highthroughput screening approach, we identified a small-molecule compound, NIC3, that is able to inhibit NAC1 dimerization via targeting the critical dimerization core residues and can promote proteasome-dependent degradation of NAC1 protein (Figs. 2, F and G; 5, F and G; and S2, D and E). These results are consistent with the concept that exposure of the hydrophobic interface of a dimeric protein often leads to its conformational change and destabilization (24,25). We demonstrate that degradation of NAC1 protein caused by NIC3 is specifically based on its properties as an inhibitor of NAC1 homodimerization but is not a consequence of an off-target effect. Using biotinlabeled NIC3 as a probe, we show that NIC3 directly targets NAC1 protein, and through both computational and experimental studies we further show that NIC3 inhibits NAC1 homodimerization through docking into critical residues such

Inhibitor targeting NAC1 dimerization
as Leu 90 , thereby preventing the interaction between the two NAC1 molecules (Fig. 4).
A number of oncogenes such as kinases are valuable druggable targets. However, there are oncogenic proteins that are regarded as undruggable, and most of these are transcriptional factors, including MYB, MYC, and NF-B (35,36). Targeting nuclear proteins has been considered challenging, mainly due to their relatively larger interaction surfaces and lack of a druggable binding pocket (5). Nevertheless, based on new concepts in drug development and better understanding of the interaction surfaces, a number of so-called undruggable molecules have now been successfully targeted (37)(38)(39)(40)(41)(42). The results reported here provide another example of targeting a nuclear factor through its hydrophobic pockets and demonstrate that computational analysis of the hydrophobic dimerization interface to identify the core units is a feasible approach to searching for small-molecule compounds for inhibiting homodimerization of the target proteins.
Cancer metastasis and therapeutic resistance are the major clinical challenges that account for cancer-related deaths. We show that NIC3 can sensitize drug-resistant tumor cells to chemotherapeutic drugs such as Adriamycin and cisplatin and enhance the antimetastatic efficacy of the antiangiogenic agent bevacizumab. We and others have previously shown that NAC1 contributes to chemoresistance through tumor-suppressor Gadd45 inactivation (16), autophagic survival response (17), cellular senescence escape (18), and cancer cell cytokinesis (19). Thus, targeting NAC1 may weaken those survival advantages in tumor cells treated with anticancer drugs. Indeed, the combination treatment of NIC3 with Adriamycin or bevacizumab shows more potent antitumor efficacy than Adriamycin or bevacizumab alone in drug-resistant tumors both in vitro and in vivo. NAC1 has previously been shown to promote a hypoxic microenvironment through hypoxia-induced up-regulation of HIF-1␣ (20) and to maintain stemness of stem cells (32,33). Antiangiogenic agents such as bevacizumab may induce hypoxia in the tumor microenvironment and increase the population of cancer stem cells that favors metastatic spread (29 -31). Down-regulation of NAC1 through preventing its dimerization by small-molecule inhibitors such as NIC3 may mitigate the hypoxic tumor environment and reduce the population of cancer stem cells.
Whether or not NIC3 can be further developed as a therapeutic agent for treatment of human cancer remain to be investigated. For instance, the pharmacokinetic properties of this compound are currently unknown. Bioisosteric replacements and scaffold hopping are useful approaches in drug design as they facilitate optimization of pharmacokinetic properties and activity (43). Whether bioisosterism and scaffold hopping such as substitution of a lipophilic group with a hydrophilic group or ethylenediamine with piperazine or homopiperazine may improve pharmacokinetics, solubility, metabolic stability, binding ability, or simplification of the synthetic route of NIC3 requires further studies.
Collectively, the findings reported here uncover not only a biologically critical but also a therapeutically exploitable role for NAC1 homodimerization. The inhibitor of NAC1 homodimerization that we identified, NIC3, shows potent effects on sensitizing drug-resistant tumor cells to chemotherapy and reinforcing the antimetastatic efficacy of the antiangiogenic agent bevacizumab. Thus, small-molecule inhibitors of NAC1 homodimerization may have great potential to be used as therapeutic agents in cancer treatment and may warrant further investigation.

Cell lines and culture
The human ovarian cancer cell lines SKOV3 and ES-2; the human cervical cancer cell lines HeLa and HeLa/DDP (the cisplatin-resistant cell line); the human breast cancer cell lines MDA-MB-231, MCF-7, and MCF-7/ADM (the Adriamycinresistant cell line); and the human embryonic kidney HEK-293T cells were purchased from ATCC (Manassas, VA). The identities of these cell lines were verified by short tandem repeat analysis. HeLa, HeLa/DDP, ES-2, MDA-MB-231, MCF-7, and MCF-7/ADM cell lines were cultured in Dulbecco's modified Eagle's medium supplemented with 10% heatinactivated fetal bovine serum, 100 units/ml penicillin, and 100 mg/ml streptomycin. The SKOV3 cell line was cultured in RPMI 1640 medium supplemented with 10% heat-inactivated fetal bovine serum, 100 units/ml penicillin, and 100 mg/ml streptomycin. Cells were cultured at 37°C in a humidified atmosphere of 20% O 2 , 5% CO 2 (normoxia) or 1% O 2 , 5% CO 2 (hypoxia). All cultures were monitored routinely and found to be free of contamination by Mycoplasma or fungi. Cell lines did not surpass 10 passages between thawing and use.

Molecular docking
The 2D structures of the ligands were drawn in ChemBi-oDraw 2013 and converted to 3D in MOE 2014 through energy minimization. Protein structures were downloaded from the Protein Data Bank (PDB code 3GA1) and prepared with the Structure Preparation workflow in MOE to correct structural Inhibitor targeting NAC1 dimerization errors such as missing atoms or nonstandard atom names. Then, the protonation state of the proteins and the orientation of the hydrogens were optimized by LigX at a pH of 7 and temperature of 300 K. Prior to docking, the force field of AMBER10: EHT and the implicit solvation model of Reaction Field (R-field) were selected. The docking workflow followed the "induced fit" protocol of MOE-Dock, in which the side chains of the receptor pocket were allowed to move according to ligand conformations, with a constraint on their positions. The weight used for tethering side-chain atoms to their original positions was 10. For each ligand, all docked poses were ranked by London dG scoring first, and then a force field refinement was carried out on the top 30 poses followed by a rescoring of GBVI/WSA dG, respectively.

Chemical synthesis
Syntheses of the PP and NP were carried out according to Fig.  S4.

Plasmids and plasmid transfection
V5-NAC1 plasmid was described previously (20). The Myc-NAC1 plasmid was generated by replacing V5 tag with Myc tag. To generate the mutants of NAC1, we replaced the leucine site Leu 90 , serine site Ser 91 with arginine, and methionine site Met 92 with leucine. Site-directed mutagenesis was performed using the QuikChange kit (Stratagene). Transfection of the plasmids was carried out using Lipofectamine 2000 (Invitrogen) according to the manufacturer's protocol.

Quantitative real-time PCR
Total RNA was prepared using TRIzol reagent (Roche Applied Science). First-strand cDNA was synthesized using an Omniscript reverse transcription kit (Qiagen) with random primers. Quantitative RT-PCR was performed on ABI 7500 using Brilliant II SYBR Green QPCR Master Mix (Stratagene) and primers. After 40 cycles, data were collected and analyzed using the ABI 7500 software.

Immunoblotting and coimmunoprecipitation
Proteins (10 -20 g) were resolved by SDS-PAGE and then transferred to PVDF membrane (Bio-Rad). Membranes were incubated with primary antibodies in 3% BSA at 4°C overnight followed by incubation with secondary antibodies at room temperature for 1 h. The protein signals were detected by ECL method. For co-IP, appropriate antibodies were first incubated with protein A/G beads (Santa Cruz Biotechnology) at 4°C overnight, and then cell lysates were incubated with the protein A/G beads at 4°C for 6 h. At the end of incubation, the beads were washed three times with radioimmune precipitation assay buffer, and the immunoprecipitates were eluted with SDS buffer and then subjected to immunoblotting.

Nondenaturing PAGE
Purified NAC1 was incubated with increasing amounts of DSS (Thermo Scientific) or different concentrations of NIC3 before mixing with an equal volume of sample buffer (100 mmol/liter Tris, pH 8.0, 20% glycerol, 0.005% bromphenol blue, 2% Triton X-100, and 100 mmol/liter DTT) followed by incu-bation at room temperature for 30 min. The cross-linking reaction was stopped by addition of 1 l of 200 mM Tris-HCl, pH 7.4, and incubated for 15 min at room temperature to quench the cross-linker. After centrifugation at 11,000 ϫ g for 10 min, the supernatants were separated by electrophoresis on a 12% Tris/glycine polyacrylamide gel followed by transfer to PVDF membrane for Western blot analysis as described previously (44).

Clonogenic assay
Cells subjected to different treatments were plated in 35-mm tissue culture dishes (numbers of cells varied with different cell lines to generate single colonies). Following incubation at 37°C in a humidified atmosphere containing 5% CO 2 , 95% air for 12 days, cells were stained with 1% methylene blue in 50% methanol, and colonies (Ͼ30 cells) were counted.

Histology
For immunohistochemistry analysis of xenograft tumors, tumor specimens were fixed in 4% paraformaldehyde and stained with H&E. For in situ determination of cell proliferation or apoptosis, sample slides were incubated with anti-Ki67 antibody (1:200 dilution) and subjected to TUNEL.

Animal experiments
Animal maintenance and experimental procedures were approved by the Institutional Animal Care and Use Committee of Soochow University. Female severe combined immunodeficient (SCID) mice (5 weeks old, female) were inoculated subcutaneously with MCF-7/ADM cells (2 ϫ 10 6 cells/mouse, mixed (1:1 volume) with BD Matrigel). Two weeks after inoculation, the tumor-bearing mice were divided into four groups (five mice per group): 1) control, 2) Adriamycin, 3) NIC3, and 4) Adriamycin ϩ NIC3. Adriamycin (5 mg/kg) was given intraperitoneally q3d six times. NIC3 (25 mg/kg) was given via tail vein q3d six times. Tumor volumes were determined by measuring the length (L) and the width (W) of the tumors and calculating using the formula V ϭ L ϫ W 2 /2. At the end of the experiment (on day 27), the mice were euthanized, and tumors were surgically dissected. The tumor specimens were fixed in 4% paraformaldehyde for histopathologic examination. In the experimental metastasis mouse model, luciferase-labeled MDA-MB-231 human breast tumor cells (1 ϫ 10 6 ) were injected into the tail vein of SCID mice. One week after injection, the mice were divided into four groups (five mice per group): 1) control, 2) bevacizumab, 3) NIC3, and 4) bevaci-Inhibitor targeting NAC1 dimerization zumab ϩ NIC3. Bevacizumab (5 mg/kg) was given intraperitoneally q3d six times. NIC3 (25 mg/kg) was given via tail vein q3d for 2 weeks. Mice were imaged for luciferase expression on a weekly basis for 7 weeks via a Xenogen IVIS bioluminescence imager. Mice were injected with 5 l/g of body weight of 30 mg/ml Luciferin-D (Gold Biotechnology, catalogue number LUCK-1G) in PBS 5 min prior to imaging. Photon flux was calculated using region of interest measurements of the ventral thoracic area for lung and bone metastases. At the experiment end point, mice were euthanized, and lung tissue was harvested for ex vivo analysis and subsequent histology.

Statistical analysis
Statistical analyses were performed using Microsoft Excel software and GraphPad Prism. The results are presented as mean Ϯ S.D. from at least three independent experiments. The p values for comparisons between experimental groups were obtained by Student's t test. All statistical tests were two-sided. *, p Ͻ 0.05; **, p Ͻ 0.01.