Drugging MYCN Oncogenic Signaling through the MYCN-PA2G4 Binding Interface

The MYC oncoproteins (MYCN, MYC, and MYCL) are short-lived transcription factors that are overexpressed or deregulated in more than half of all human cancers, making them important therapeutic targets. However, the design of MYC inhibitors has been hampered by the lack of globular functional MYC domains or deep protein “pockets” for drug binding. Moreover, MYC inhibitors carry a heightened potential for side-effects due to the dependency of most normal cells on transient MYC expression at entry into the cell cycle.

Neuroblastoma is the most common solid malignancy in early childhood and advanced disease accounts for a disproportionately high mortality when compared with other childhood cancer types (1). The MYCN oncogene is amplified and overexpressed in 25% of patients with this embryonal childhood cancer. Furthermore, transgenic MYCN (TH-MYCN) mice develop neuroblastoma with most of the characteristics of the human disease, making neuroblastoma a model of MYC-driven human cancer (2, 3). MYCN drives the initiation and progression of neuroblastoma through a transcriptional program regulated by multiple protein–protein and protein–DNA interactions in malignant cells (4). Protein–protein interaction sites involving oncoproteins have become a major focus for the discovery of cancer therapeutics, whereby oncogenic signal activation or oncoprotein stability can be inhibited by competitive binding of small molecules at these protein–protein interfaces (5).

We and others have demonstrated several mechanisms by which increased MYCN protein stability acts as a necessary cofactor for MYCN-driven tumorigenesis (6–8). Moreover, we find that many of the proteins that directly or indirectly stabilize MYCN are themselves transactivation targets of MYCN, creating a positive feedforward regulatory loop and a therapeutic vulnerability in neuroblastoma cells (7, 8). MYC and MYCN protein stability is tightly regulated via the ubiquitin-proteasome system in normal cells, where it is an important binding target of the E3 ligase, F-box, and WD40 domain protein 7 (Fbxw7; refs. 6, 9). Fbxw7-mediated MYCN proteolysis in neuroblastoma cells can be inhibited by the Aurora A (AurA) serine-threonine kinase, which blocks binding of Fbxw7 to MYCN, or by Polo-like kinase-1 (PLK1), which induces Fbxw7 autoubiquitination (6, 10, 11). Targeting MYCN through disruption of protein interaction partners remains a highly promising approach if suitable inhibitors can be found.

The proliferation-associated 2AG4 protein (PA2G4), or ErbB3 binding protein (EBP1), binds Fbxw7, sequestering it in the cytoplasm, hence stabilizing MYC in colorectal cancer cells (12). PA2G4 is a ubiquitously expressed DNA and RNA binding protein with roles in normal embryonal muscle and neural crest cell growth (13, 14). The PA2G4 gene transcribes two isoforms: the predominant long form, p48, and p42. Initiation of p48 translation starts at the first ATG codon, making p48 54 amino acids longer than p42. p42 translation begins at the third ATG codon, thus the protein has a truncated N-terminus, potentially resulting in its inability to translocate to the nucleus (15–17). Generally, p48 function acts as a tumor promoter whereas p42 acts as a potential tumor suppressor. Overexpression of p48 stimulates cell growth, prevents neurite outgrowth, enriches cells in S phase (15), and prevents apoptosis by interacting with AKT and HDM2 to regulate p53 (18, 19). Overexpression of p42 suppresses cell growth, initiates cell differentiation, and causes accumulation of cells in the G₁ phase of the cell cycle (20). Considerable evidence suggests that high levels of PA2G4-p48 drives cell proliferation and correlates with poor patient survival in several human cancers (16, 21). Here we show that PA2G4 is a transactivation target for MYCN in neuroblastoma cells, which directly binds to and stabilizes MYCN in a positive feedforward expression loop. Using mutant proteins and structural biologic approaches, we define a PA2G4–MYCN protein–protein interface that can be inhibited by a known PA2G4-binding compound, with significant antitumor effects in vitro and in vivo. Furthermore, we show that PA2G4 expression levels in neuroblastoma patient tumor tissues correlated with MYCN levels and poor prognosis.

MYCN and Input ChIP-seq fastq files were retrieved directly from the European Nucleotide Archive under the SRA accession number SRP073110 for four MYCN-amplified cell lines (KELLY, BE2C, NGP, and TET21N; ref. 22). Fastq files were first trimmed to remove poor quality reads and adapters (Q30 < 20, length < 20 bp) using the TrimGalore wrapper (Cutadapt/FastQC; ref. 23). Reads from the trimmed fastq's were then aligned to the reference human genome (hg38) using bowtie2 to produce sam files. Sam files were converted to the bam format and then sorted using samtools (view/sort). Duplicate reads were then removed from sorted bam files using picard tools (MarkDuplicates). Chromatin immunoprecipitation sequencing (ChIP-seq) peaks were then called using the MACS2 algorithm (callpeak; q < 0.05) from sorted bam files, followed by the generation of fold enrichment tracks (MYCN/Input) in bedgraph format. Peaks were called in individual cell lines and when combined as pseudoreplicates. Bedgraph files were then converted to the bigwig format using bedtools (bdg2bw). The IGV genome browser was then used to visualize fold enrichment peaks. Peaks identified by MACS2 were annotated using HOMER (annotatepeaks.pl), promoters were defined as being within −1000/+100 bp of the transcription start site (TSS). PWMtools was used to annotate canonical MYCN E-Boxes (CACGTG) across the human genome (hg38), and were visualized on IGV alongside fold enrichment tracks (24).

Neuroblastoma cell lines NBL-S were kindly provided by Professor Susan L. Cohn (Northwestern University, Chicago, IL), and SH-EP and SK-N-BE(2)C by Professor June Biedler (Memorial Sloan Kettering Cancer Center, New York, NY). Non-MYCN–amplified human neuroblastoma cell line SHEP MYCN-3 and SHEP-tet21N, which are genetically modified to overexpress or repress human MYCN cDNA when exposed to doxycycline, were generously supplied by Professor Jason Shohet (Texas Children's Cancer Center, Houston, TX). All other cell lines were purchased from ATCC. SK-N-AS, SK-N-DZ, SK-N-FI, SK-N-BE(2)-C, SH-EP, SH-SY5Y, LAN-1, IMR-32, NBLS, and HEK-293T were maintained in DMEM (Invitrogen, Life Technologies) with 10% FCS. CHP-134 and Kelly cells were cultured in RPMI media (Invitrogen, Life Technologies) with 10% FCS. MRC-5 and WI-38 lung fibroblast cells (ATCC) were cultured in Alpha-MEM media (Invitrogen, Life Technologies) with 10% FCS. Cell lines are passaged to a maximum of 10–15 times before being discarded. All cell lines used were authenticated by CellBank Australia, free from Mycoplasma, and cultured at 37°C with 5% CO₂ in a humidifier incubator.

The mRNA expression in human neuroblastoma tissues from the Cologne dataset (477 diverse patient samples) was analyzed from publically available gene expression databases (http://r2.amc.nl). Primary human neuroblastoma cDNA samples were synthesized from total RNA isolated from 40 patient tumors obtained at diagnosis from patients consecutively treated at the Sydney Children's Hospital following consenting approved by the Institutional Ethics Committee (approval number: HC12591, University of New South Wales, Sydney, Australia). The patients have given written informed consents.

Point mutations were generated by GenScript. pCMV6-PA2G4.p48-MYC Flag-tag was used as the template for the site mutagenesis of the six PA2G4 mutants. The six PA2G4 mutants used in the experiments were: mutant 1 Ser47->Ala; mutant 2 Glu50->Ala; mutant 4 Lys93->Tryptophan; mutant 5 Arg271->Ala; mutant 6 Arg272->Ala; and mutant 7 Arg281->Ala. All mutations were confirmed by direct sequencing. pShuttle-IRES-hrGFP- MYCN plasmid DNA was used as the template for the site mutagenesis of the four MYCN mutants.

Preparation of DNA from SK-N-BE(2)-C and Kelly cells (∼6 × 10⁷ or more cells) for ChIP assay was performed using the Chromatin Immunoprecipitation Assay Kit (Millipore) according to the manufacturer's instructions. The shearing of DNA was performed on the Bioruptor Pico (Diagenode) with 45 sets of 30 seconds per pulse. ChIP assays were performed with 10 μg anti-MYCN antibody (Santa Cruz Biotechnology) or a control mouse IgG antibody (Santa Cruz Biotechnology). DNA from was purified using a MiniElute PCR Purification Kit (Qiagen) and eluted with 40 μL of DNase/RNase-free water. Real-time PCR was performed with primers targeting the negative control region, which was 1200 bp upstream of TSS, PA2G4 gene promoter, PA2G4 intron 1 region, which contains two MYCN canonical E-boxes, and the ODC promoter region as a positive control. PA2G4 gene sequences containing putative MYCN-binding regions were identified by the anti-MYCN antibody, and fold enrichment was calculated by dividing the densitometric assessment of the PCR product from PA2G4 gene promoter and PA2G4 intron 1 region by the PCR product from the negative control region, relative to input.

Human PA2G4 was cloned into the PDONR221 gateway donor vector by BP recombination reaction. The PA2G4-PDONR221 or mCherry-PDONR221 (25) construct was combined with the other two entry clones, dβh-pDONRP4-P1R (26) and p3E-EGFPpA, as well as pDestTol2CG destination vector, using the Multisite Gateway System to generate the dβh:PA2G4 Tol2 and dβh:mCherry Tol2 transgenic constructs.

To mosaically overexpress PA2G4 or control mCherry in the Tg(dβh: EGFP-MYCN) transgenic fish (designated the “MYCN” transgenic line), a 1–2 nL of DNA solution containing 50 ng/μL of dβh:PA2G4 Tol2 or dβh:mCherry Tol2 DNA construct and 5 ng/μL of the Tol2 mRNA was coinjected into one-cell stage MYCN transgenic embryos. In the pDestTol2CG construct, there is a cmlc2:egfp-pA cassette. The embryos with successful integration of injected DNA constructs show the EGFP expression in the heart region. Such heart-EGFP and MYCN-EGFP double-positive embryos were sorted at day 2–5 and monitored every alternate week for EGFP/mCherry-expressing cell masses in the inter-renal gland region under fluorescence microscopy from 5 weeks onward. A Leica MZ10F Stereo fluorescence microscope was used to capture bright field and fluorescent images of tumor-bearing fish. Images were processed in Leica Application Suite X (v. 2.0.0), Adobe Photoshop, and Illustrator CC (v. 19.1.0) software. Kaplan–Meier analysis was applied to assess the rate of tumor development. Fish that died prior to evidence of external EGFP/mCherry-positive masses were censored. The log-rank statistical test was used to assess the difference between tumor onset by 11 wpf in the MYCN fish mosaically overexpressing mCherry and that in the MYCN line mosaically overexpressing PA2G4.

To further test the efficacy of WS6, another animal model, MYCN;EGFP transgenic fish, with established neuroblastomas were treated with 175.4 mg/kg of WS6 (5 μL) dissolved in vehicle (saline solution) or vehicle alone, for 7 consecutive days with a 26 G needle. After 7 days of treatment, fish were anesthetized and sacrificed for sectioning and pathology analyses. Bright-field and fluorescent images of treated tumor-bearing fish at day 1 and day 7 were taken using identical parameters on a Leica MZ10F fluorescence microscope. Images were processed with Zeiss Zen 2012, Adobe Photoshop, and Illustrator CS3 (Adobe) software. The acquired fluorescence images were quantified by measuring the EGFP-covered area using ImageJ software. Means of changes in tumor volume between WS6-treated and vehicle-treated groups were compared by Student t test (GraphPad prism 6). Bright-field images of hematoxylin and eosin or immunostained paraffin sections (3–5 μm each) were captured by an Olympus AX70 compound microscope equipped with an Olympus DP71 camera. Images were processed with Zeiss Zen 2012, Adobe Photoshop, and Illustrator CS3 (Adobe) software. The acquired hematoxylin and eosin or immunostained images were quantified by measuring the intensity of stained tumor using ImageJ software. Means of changes in staining intensity between WS6-treated and vehicle-treated groups were compared by Student t test (GraphPad prism 6).

All surface plasmon resonance (SPR) was conducted on a Biacore T200 (GE Healthcare) and analyzed using the BIAevaluation software package. Wild-type, triple mutant (R271A, R272A, and S47A) and single mutant (S47A or R272A) PA2G4 protein was biotinylated using standard procedures (Thermo Fisher Scientific) and purified using a Superdex 200 10/30 GL column. The resultant proteins were captured on immobilized streptavidin, which has been coupled to a CM5 chip (GE Healthcare) via amine coupling. Final amount of biotinylated protein captured was 3500RU. Flow cell 1 was derivitized and blocked for use as a reference. MycN oligopeptides (DHKALST and GGDHKALSTGEDTL) and mutant MycN oligopeptides (DHAALST, DHAALAT, and DHKALAT) were diluted in HBS+P buffer (0.01 mol/L HEPES pH 7.4, 0.15 mol/L NaCl, and 0.005% v/v Surfactant P20) from 100 to 0.8 μmol/L in a 2-fold dilution series and injected across the chip for 60 seconds at 30 μL/minute. Experiments were conducted in duplicate at least three independent times. For the competition assay, WS6 was added at a final concentration of 10 μmol/L to the MycN dilution series. The steady-state affinities (K_d) were calculated by plotting the steady-state binding levels against the concentration.

The molecular dynamics program NAMD (27) was used to minimize the docking models of PA2G4 with MYCN oligopeptides on a TELSA K20 GPU machine (NVIDIA). Each model was initially solvated with TIP3P water using the Solvate plugin within VMD version 1.9 (28) with the rotate to minimize volume selected, the boundary reduced to 1.8, and the box padding increased to 20 Å in all directions. Charges were neutralized with NaCl using the autoionize plugin within VMD version 1.9. Each structure was then minimized and equalized for 1 nanosecond under the CHARMM27 all-atom force field at 298K. Langevin dynamics were used with group pressure and langevin piston turned on. Trajectory snapshots were collected every picosecond. A production run was then completed 10 times for each alternate simulation (40 simulations in total) on this minimized structure for 10 nanoseconds using the same conditions as described above. Convergence was monitored by plotting root mean square deviation as a function of the entire simulation time. The first 2 nanoseconds of the production runs were calculated to be the equilibration stage, thus analyses were carried out on the final 8 nanoseconds of each production run. The H bond plugin of VMD was used to calculate the hydrogen bonds between the PA2G4 and the MYCN oligopeptide. Criteria for the formation of hydrogen bonds were that the distance between hydrogen donor (D) and hydrogen acceptor (A) atoms was less than the cut-off distance of 3.2 Å and the angle D-H-A was less than the cut-off angle of 42°. The average occupancy of the hydrogen bonds was calculated as a ratio between the number of the trajectory frames containing the hydrogen bond of interest, and the total number of frames.

SHEP PA2G4 stable–overexpressing cells, SHEP control cells expressing EV, or Kelly neuroblastoma were xenografted into BALB/c nude mice to assess the ability of PA2G4 to induce tumorigenesis. Cells were resuspended in PBS at a concentration of 6 × 10⁶ cells per 100 μL then mixed with Matrigel (BD Biosciences) basement membrane matrix at a ratio of 1:1. The cell/Matrigel solution was kept cold on ice and 200 μL was injected subcutaneously into the dorsal flank of BALB/c nude mice (6 weeks of age). Mice were monitored for signs of tumor formation for a period of 12 weeks postinjection, and then humanely euthanized by CO₂ asphyxiation. Kelly xenografted mice were treated with the maximum tolerated dose of 70 mg/kg WS6 when the mice developed 4 mm tumor and were treated for 5 consecutive days of each week for 3 weeks or when the tumor reached an ethical limit of 1,000 mm. Tumor tissues were excised, measured, and collected for morphologic and biochemical analysis.

To assess the capability of the PA2G4 inhibitor, WS6, to delay the growth of neuroblastoma tumors, TH-MYCN^+/+ homozygous mice (3 weeks of age) were treated with 85 mg/kg of WS6 (100–200 μL) dissolved in DMSO, or DMSO only as control, for 5 consecutive days per week for a maximum of 3 weeks intraperitoneal injection with a 27–29 G needle. Treatment was discontinued when tumors reached 1,000 mm or 3 weeks post treatment. All mice were humanely euthanized by CO₂ asphyxiation and tumors were excised, measured, and collected for morphologic, histologic, and biochemical analysis. This study was approved by the Animal Care and Ethics Committee, UNSW Sydney (Approval number: 18/72B).

To assess whether PA2G4 was required for tumor growth and progression, PA2G4 expression was suppressed with highly specific siRNAs targeting PA2G4 or PA2G4.p48 gene sequences, which were delivered by nonviral and nontoxic nanoparticles synthesized by chemists at the Australian Centre for NanoMedicine, UNSW Sydney. BE(2) cells (4 × 10⁶ in 100 μL PBS) were subcutaneously injected into the dorsal flank of BALB/c nude mice (6 weeks of age). Once tumors reached approximately 100–200 mm, mice were randomized into treatment groups to receive a local injection of nanoparticle–siRNA complexes directly into the tumor using a 0.5 mL syringe containing a 29 G needle. The solution injected into the tumor contained sterile saline with 200 μg of nanoparticles complexed to PA2G4 siRNA (40 μg) or control siRNA (40 μg; total volume 50 μL). Mice were injected once every 3 days for up to 8 injections or until the tumor reached 1,000 mm. The rate of tumor growth was measured twice weekly using digital calipers and recorded. This study was approved by the Animal Care and Ethics Committee, UNSW Sydney (Approval number: 14/152B).

All experiments included a minimum of three independent replicates. Statistical calculations were performed using GraphPad Prism 6 software. Results were compared using ANOVA among groups or two-sided unpaired t test for two groups and expressed as mean values with 95% confidence intervals. Graphical error bars for in vitro data represent the SEM, while error bars for in vivo data were calculated as SD. A P value of less than 0.05 was considered statistically significant.

We investigated whether a regulatory relationship existed between PA2G4 and MYCN expression in neuroblastoma cells with amplified and overexpressed MYCN. We first analyzed publicly available ChIP-seq data of genome-wide MYCN occupancy (22) and found that MYCN strongly bound a canonical E-Box proximal to the PA2G4 gene promoter (FE = 8.43082; P = 2.18162E⁻⁸³; Q = 2.3877E⁻⁷⁹; n = 4; Fig. 1A). We next found that MYCN-amplified neuroblastoma cells transiently transfected with MYCN-specific siRNA duplexes, displayed reduced MYCN and PA2G4 expression at the protein (Supplementary Fig. S1A), and mRNA level (Fig. 1B). Doxycycline treatment of a human neuroblastoma cell line with doxycycline-inducible MYCN overexpression (SH-EP MYCN3, SH-EP-TRE-MYCN), increased PA2G4 expression and showed decreased MYC levels (Fig. 1C; Supplementary Fig. S1B and S1C). Whereas doxycycline-induced repression of MYCN expression in SHEP-tet21N neuroblastoma cells correlated with repressed PA2G4 and a reciprocal increase in MYC levels. The modest decrease in PA2G4 expression observed, despite the abrogation of MYCN, may be due to the reciprocal elevation of c-MYC levels. We performed ChIP assays and demonstrated a 4–10-fold enrichment of MYCN binding at canonical E-Box regulatory sequences situated within 500 bp upstream of the PA2G4 TSS, and intron 1a and 1b of the PA2G4 gene in two neuroblastoma cell lines (Fig. 1D; Supplementary Fig. S1D). MYCN chromatin binding was abolished by transiently transfected MYCN-specific siRNAs. These data indicate that MYCN enhances PA2G4 gene transcription in neuroblastoma cells.

PA2G4 levels closely correlated with the high levels of MYCN protein expression across a panel of MYCN-amplified human neuroblastoma cell lines, but not in neuroblastoma cell lines that were nonamplified for MYCN (NBLS, SH-EP, and SK-N-FI; Supplementary Fig. S1E). Interestingly, two MYCN nonamplified cell lines, but with high MYC expression (SH-SY5Y and SK-N-AS), also had detectable levels of PA2G4, indicating a possible regulatory relationship between PA2G4 and MYC. Subcellular localization studies using immunoblotting of nuclear and cytoplasmic protein fractions from neuroblastoma cell lines and immunofluorescence showed that both MYCN and PA2G4 were colocalized in the nucleus (Supplementary Fig. S1F–S1H). Homozygous TH-MYCN^+/+ transgenic mice develop neuroblastoma in paravertebral and coeliac sympathetic ganglia at 6–7 weeks of age, after a precancer stage of neuroblast hyperplasia in the ganglia during the first 2 weeks of life (29). We found the PA2G4 expression pattern mirrored the high levels of MYCN produced by the MYCN transgene from neuroblastoma initiation in primary ganglia to later tumor formation (Supplementary Fig. S1I).

We next assessed whether PA2G4 participated in a positive feedforward expression loop with MYCN, as we have identified for other MYCN transactivation targets (7, 8, 30, 31). Following transient transfection of neuroblastoma cells with PA2G4-specific siRNAs, we found a marked reduction in both PA2G4 and MYCN protein (Fig. 2A), but not MYCN mRNA levels (Fig. 2B). Cycloheximide chase experiments assessing MYCN protein half-life demonstrated a 50% reduction following PA2G4 siRNA knockdown (Fig. 2C; Supplementary Fig. S2A), while PA2G4 overexpression caused a 2-fold increase in MYCN protein half-life (Fig. 2D; Supplementary Fig. S2B). The treatment of neuroblastoma cells with the proteasome inhibitor, MG132, resulted in increased MYCN protein levels, which were maintained at high levels even after PA2G4 knockdown, indicating that PA2G4 protects MYCN protein from ubiquitin-mediated proteasomal degradation (Fig. 2E). We next performed coimmunoprecipitation (co-IP) using a MYCN-specific antibody on neuroblastoma cell lysates after PA2G4 knockdown, and immunoblotted with an ubiquitin-specific antibody (Fig. 2F). Repression of PA2G4 expression was associated with an increase in MYCN ubiquitination. Studies in colorectal cancer cells suggested PA2G4 could bind directly to, and sequester Fbxw7 in the cytoplasm, thus indirectly increasing nuclear MYC protein stability (12). We investigated this potential mechanism of PA2G4 influencing MYCN protein stability indirectly by shifting the location of Fbxw7 toward the cytoplasm and possibly protecting nuclear proteins including MYCN. We measured nuclear and cytoplasmic Fbxw7 levels after modulating PA2G4 (Fig. 2G). We were unable to detect any change in Fbxw7 levels in the cytoplasm or nucleus in response to knocking down or overexpressing PA2G4, suggesting that PA2G4 is not involved in sequestering Fbxw7 to the cytoplasm in neuroblastoma cells. In-line with this, PA2G4 and Fbxw7 do not bind directly (Supplementary Fig. S2C). These findings are consistent with a model of PA2G4 directly binding to, and stabilizing MYCN, thereby protecting it from ubiquitin-mediated proteasomal degradation.

We hypothesized that the positive regulatory relationship between PA2G4 and MYCN expression may result from direct binding of PA2G4 to MYCN. We used co-IP to show that native PA2G4 directly bound native MYCN in two human MYCN-amplified neuroblastoma cell lines (Fig. 3A). A diarylamide-based compound, which binds PA2G4, WS6, had previously been identified in a screen for small molecules that stimulated pancreatic β islet cell growth (32). When we treated two human MYCN-amplified neuroblastoma cell lines with 1 μmol/L WS6 (Supplementary Fig. S3A) for only 1 hour, we showed that WS6 treatment completely blocked PA2G4-MYCN protein binding (Fig. 3A). Treatment of two MYCN-amplified neuroblastoma cell lines with 0.5–1 μmol/L WS6 reduced PA2G4 and MYCN expression levels (Fig. 3B), and decreased MYCN protein stability (Supplementary Fig. S3B). WS6 treatment of neuroblastoma cell lines reduced colony formation in vitro by 20-fold (Supplementary Fig. S3C). To determine the specificity of WS6 for PA2G4, we transfected and overexpressed PA2G4 in neuroblastoma cells, then treated the transfected cells with WS6. Overexpression of PA2G4 increased cell viability in the absence of WS6 and partially blocked the cytopathic effects of WS6, consistent with the hypothesis that the cytopathic effects of WS6 on neuroblastoma cells were caused in part by a reduction of PA2G4 levels (Fig. 3C; Supplementary Fig. S3D).

We next assessed the antitumor activity of WS6 in the homozygous TH-MYCN^+/+ mouse model of neuroblastoma. Mice with palpable intra-abdominal tumors at 3 weeks of age were treated with the MTD of 85 mg/kg WS6 intraperitoneally for 5 consecutive days per week for a maximum of 3 weeks, or until the tumor reached 1,000 mm. Treatment with WS6 caused a 2-fold reduction in tumor size and a corresponding decrease in MYCN and PA2G4 protein expression in vivo (Fig. 3D; Supplementary Fig. S3E). Moreover, athymic nude mice xenografted subcutaneously in the flank with MYCN-amplified Kelly neuroblastoma cells, then treated with the MTD of 70 mg/kg WS6 intraperitoneally for 5 consecutive days per week for a maximum of 3 weeks, also demonstrated an almost 2-fold reduction in tumor size (Fig. 3E). Adult MYCN transgenic zebrafish develop neuroblastoma in the inter-renal gland, an analog of the human adrenal gland (26). Intraperitoneal WS6 treatment of adult zebrafish with established neuroblastoma caused a 4-fold reduction in tumor size compared with solvent-treated controls and was accompanied by a marked reduction in PA2G4 and MYCN protein levels (Fig. 3F; Supplementary Fig. S3F).

To compare the relative potency of WS6 with other MYCN signal inhibitor molecules, we performed cytotoxicity treatment experiments in vitro against Kelly neuroblastoma cells and calculated IC₅₀ values for the USP7 inhibitor, P22077 (33), the FACT inhibitor, CBL0137 (8), and the AurA kinase inhibitor, MLN8237. The IC₅₀s for the ATP-competitive Aurora-A inhibitor, CD532, against a range of tumor cell lines have been published previously (34). The IC₅₀ comparisons (Supplementary Fig. S3G) showed WS6 had comparable single-agent potency to the USP7 inhibitor, the FACT inhibitor, and the ATP-competitive Aurora-A inhibitor, but not the AurA kinase inhibitor, which was more potent (35).

We next sought to better delineate the PA2G4–MYCN protein–protein interface. Using co-IP of transiently overexpressed GST-MYCN deletion mutant constructs and a full-length PA2G4-Flag vector in HEK-293T cells, we determined that the MYCN region bounds PA2G4. Only the MYCN deletion mutant consisting of amino acids 82–254 (MYCN-N2) bound strongly to PA2G4, while MYCN-N3 and -N4 had weak binding (Fig. 4A). Analysis of this MYCN region revealed a seven amino acid sequence (DHKALST, aa248–aa254) overlapping constructs MYCN-N2 and MYCN-N3, which was not repeated elsewhere in the MYCN protein. Computational molecular docking analysis of this MYCN oligopeptide (aa248–aa254) against the known PA2G4 crystal structure (36), predicted that this region of MYCN interacted with a specific surface pocket of PA2G4 (Fig. 4B). A blind dock across the whole surface of PA2G4 predicted that WS6 (in green) bound to PA2G4 at the same site as the seven amino acid MYCN oligopeptide. Furthermore, the modeling of an additional seven amino acids at the N- and C-termini suggested that this entire 14 amino acid MYCN sequence (aa246–aa259) contributed to binding.

To determine direct binding of these MYCN oligopeptides to PA2G4, we expressed and purified recombinant PA2G4 (aa9–aa394), then conducted SPR (Biacore T200) assays with the MYCN oligopeptides. Both the MYCN 7aa oligopeptide DHKALST (aa248–aa254) and the MYCN 14aa oligopeptide GGDHKALSTGEDTL (aa246–aa259) gave clear dose-response binding to immobilized PA2G4 with K_ds of 28.3 ± 0.73 μmol/L (n = 5) and 24.8 ± 0.61 μmol/L (n = 5), respectively (Fig. 4C). The molecular docking model further predicted PA2G4 amino acids Ser47, Glu50, Lys93, Arg271, Arg272, and Arg 281 were involved in hydrogen bonding with MYCN (Fig. 4D). When we compared the effect of transfected wild-type PA2G4 on MYCN protein half-life, to mutant PA2G4 at these seven sites, only transfected PA2G4 mutations at Ser47, Arg271, and Arg272 correlated with reduced MYCN stability (Fig. 4E; Supplementary Fig. S4A). Both the 7aa and 14aa MYCN oligopeptides also stabilized PA2G4 when measured by differential scanning fluorimetry (Supplementary Fig. S4B). Furthermore, where possible SPR experiments were conducted on the single mutants, giving extrapolated K_ds of 79.5 μmol/L and 175 μmol/L for the S47A and R272A mutations, respectively (Table 1; Supplementary Fig. S4C–S4E). SPR experiments showed that this triple mutant had an extrapolated K_d of >300 μmol/L (Supplementary Fig. S4D). Our molecular dynamics simulations suggested that MYCN amino acids Lys250 and Ser253 were involved in hydrogen bonds with PA2G4, for 3.2% and 18.8% of the simulation time, respectively. Equivalent molecular dynamics simulations with the MYCN oligopeptide residues mutated showed that MYCN oligopeptides DHAALST, DHKALAT, and DHAALAT were less energetically favorable than DHKALST, and SPR experiments showed each peptide had much higher extrapolated K_ds (Table 2). SPR competition experiments showed repression of MYCN oligopeptide DHKALST dose-response binding to immobilized PA2G4 when 10 μmol/L WS6 was preincubated with the MYCN oligopeptide, suggesting WS6 competed out the MYCN oligopeptide binding to PA2G4 (Fig. 4E). Next, we transiently transfected HEK293T cells with full-length wild-type MYCN or four individual point mutants of the DHKALST MYCN sequence, then performed co-IP with a MYCN antibody. Four individual amino acid mutations of MYCN-7aa sequence disrupted PA2G4 binding (Fig. 4F). Collectively, these data suggest that PA2G4 binds a 14aa MYCN sequence (aa246–aa259) through PA2G4 amino acid residues Ser47, Arg271, and Arg272, reducing MYCN ubiquitination and thus increasing its stability.

We demonstrated strong causal relationships between PA2G4 expression and other features of the malignant phenotype of neuroblastoma cells in vitro. Overexpression of PA2G4-p48 in neuroblastoma cells increased cell migration in the transwell migration assay, whereas PA2G4-p48 knockdown had the converse effect (Fig. 5A). Neuroblastoma cells are arrested in the pathway toward neuritic differentiation, which can be restored in vitro by treatment with 13-cis-retinoic acid, an agent used in treatment of the disease (37). Knockdown of PA2G4 increased the sensitivity of neuroblastoma cell lines to the neurite-forming effects of 13-cis-retinoic acid (Fig. 5B; Supplementary Fig. S5A) and markedly reduced in vitro colony formation (Fig. 5C). Cotransfected MYCN restored the effect of PA2G4 knockdown on neuroblastoma cell colony formation.

The PA2G4 gene transcribes two alternatively spliced mRNA isoforms encoding proteins of 48 and 42 kDa, which have been previously reported to behave, respectively, as an oncoprotein or a tumor suppressor protein in cancer cells (16, 17, 19, 38). Immunoblotting with a PA2G4 antibody demonstrated that both PA2G4-p48 and -p42 were expressed in neuroblastoma cells (Supplementary Fig. S5B). The knockdown of the PA2G4-p48 isoform reduced viability of MYCN-amplified neuroblastoma cell lines (Fig. 5D; Supplementary Fig. S5C). Whereas overexpression of PA2G4-p48 increased cell proliferation and viability, regardless of MYCN amplification status, in five neuroblastoma cell lines, but not in normal fibroblast cells (Supplementary Fig. S5D–S5F). In neuroblastoma cells, we found that PA2G4-p42 also acted as an oncoprotein, because overexpression of PA2G4-p42 increased both cell growth and colony formation (Supplementary Fig. S5G and S5H). Thus, both PA2G4 isoforms increased the malignant phenotype of neuroblastoma cells in vitro.

To assess the effect of PA2G4 on neuroblastoma tumorigenicity in vivo, we first used SH-EP neuroblastoma cells, which had no MYCN amplification or PA2G4 expression. SH-EP cells stably transfected with an empty vector (EV) control (SH-EP-EV) or a PA2G4 expression vector (SH-EP-PA2G4) were mixed with Matrigel and injected subcutaneously into the dorsal flank of athymic nude mice. SH-EP-PA2G4 cells were tumorigenic, whereas SH-EP-EV cells were not (Fig. 6A; Supplementary Fig. S6A and S6B). Excised tumor tissues from these mice exhibited increased PA2G4 protein and mRNA expression, but MYCN protein expression increases did not reach statistical significance (Fig. 6B; Supplementary Fig. S6C and S6D). We next used self-assembling cationic star polymers (39) mixed with siRNA as an in vivo nanoparticle delivery system targeting either total PA2G4, PA2G4-p48, or scrambled control, to assess the effect of inhibiting PA2G4 by direct injection of PA2G4 siRNA into established PA2G4-expressing neuroblastoma tumors. Tumor latency was significantly delayed in mice carrying xenografts injected with siRNA targeting total PA2G4 or PA2G4-p48 (Fig. 6C). Immunoblotting of protein lysates from excised residual tumor tissues for PA2G4 and MYCN expression demonstrated a decrease in expression of both proteins in vivo (Fig. 6D; Supplementary Fig. S6E). Significant decreases in PA2G4 and MYCN mRNA expression were also observed for both PA2G4 and PA2G4-p48 siRNA-treated tumors, compared with control siRNA-treated tumors (Supplementary Fig. S6F), confirming that the PA2G4-MYCN codependency signal is a significant driver of neuroblastoma tumorigenesis in vivo.

To assess the significance of PA2G4 and MYCN overexpression on neuroblastoma pathogenesis in another species, we transiently overexpressed a PA2G4 construct in the peripheral sympathetic nervous system under the control of the dβh promoter in Tg(dβh:EGFP-MYCN) transgenic zebrafish. A dbh-mCherry construct was transiently overexpressed in the MYCN transgenic fish to serve as a control for this experiment. We identified EGFP-positive masses at 5 weeks of age in the inter-renal gland of transgenic fish mosaically overexpressing PA2G4 (Fig. 6E). By 11 weeks of age, 79% of MYCN transgenic fish mosaically overexpressing PA2G4 had tumors (Fig. 6F), a rate much higher than in the MYCN fish mosaically overexpressing mCherry (32.5%; P < 0.0001). Taken together these results indicate that PA2G4 is a potent oncogenic cofactor for MYCN-driven neuroblastoma tumorigenesis across different species.

High mRNA expression of PA2G4 had a strong association with poor neuroblastoma patient prognosis in a publicly available R2 gene expression database (n = 477; Cologne dataset) of primary neuroblastoma samples (Fig. 6G; Supplementary Fig. S6G and S6H; refs. 40, 41). We also found a correlation between high PA2G4 mRNA expression and MYCN amplification in a second cohort of primary neuroblastoma samples using real-time PCR (n = 40, Sydney Children's Hospital; Supplementary Fig. S6I). High PA2G4 mRNA expression in the Cologne dataset correlated with MYCN expression, particularly in MYCN-amplified tumor tissues (Fig. 6H). Univariate comparisons showed that PA2G4 expression was higher in patients with other known predictors of poor prognosis, such as advanced clinical stage (3/4) and MYCN amplification (Fig. 6I). Multivariate analysis showed that high PA2G4 expression retained independent prognostic significance, when PA2G4 was compared with other prognostic factors in neuroblastoma (Supplementary Fig. S6J).

We next sought evidence of this association in other human MYCN-driven tumorigenesis. Publically available data from The Cancer Genome Atlas revealed a high frequency of PA2G4 amplification (16%) in the MYCN-driven human neuroendocrine prostate cancer (NEPC; Supplementary Fig. S6K; ref. 42), indicating PA2G4 may also be a driver in this malignancy. Together these data suggest that PA2G4 is an oncogenic cofactor in MYCN-driven human cancer, and, that there is a functional interdependency between MYCN and PA2G4.

Here we have identified a novel MYCN-binding protein, PA2G4, which protects MYCN from ubiquitin-mediated proteolysis in neuroblastoma cells. MYCN acts in a forward feedback expression loop with PA2G4, driving ever higher levels of each protein and consequently oncogenesis (Fig. 6J). This same regulatory relationship for PA2G4 and MYCN may extend to other human cancer types, such as NEPC. Surprisingly, this forward feedback regulatory relationship even operated in neuroblastoma cells with amplified MYCN, suggesting that strong oncoprotein instability barriers must exist in normal cells to prevent transformation by MYCN, making MYCN stability an attractive drug target.

PA2G4 was originally identified as a cell-cycle–regulated protein (human homolog of the mouse protein p38-2AG4) and then an ErbB3-binding protein (43, 44). Subsequent studies showed that PA2G4 belongs to a family of DNA/RNA binding proteins implicated in cell growth, apoptosis, and differentiation (45). The longer PA2G4 isoform, p48 (375 amino acids), was ubiquitously expressed in the cytoplasm and nucleus, and suppressed apoptosis (15). PA2G4-p48 has a known oncogenic function via a direct interaction with nuclear histone deacetylases causing transcriptional repression of tumor suppressor genes with effects on cell-cycle regulation and ribosomal synthesis (15, 16, 46). PA2G4-p48 overexpression in glioblastoma cells enhanced tumor growth in mouse xenograft models and contributed to oral tumorigenesis by activating podoplanin expression (21). Our data strongly supports the role of PA2G4 as an oncogene in MYCN-driven neuroblastoma in humans, mice, and zebrafish. However, PA2G4 may have MYCN-dependent and -independent oncogenic effects, as suggested by our finding that high PA2G4 expression in neuroblastoma tumor tissue had prognostic significance, which was independent of MYCN amplification. While our studies indicated nuclear and cytoplasmic PA2G4 expression in neuroblastoma cells, it will be important to assess whether PA2G4–MYCN complexes can bind DNA and directly influence transcription. Moreover, other proteins may also bind in this complex influencing MYCN stability, so that the comparative potency of the PA2G4-MYCN mechanism of maintaining MYCN stability to other MYCN stability factors needs to be determined (6–8, 30, 31, 33, 47–49).

Our structural analyses indicated that PA2G4 exhibited high affinity binding to a short amino acid MYCN sequence at MYC Box IIIb (aa246–259), protecting MYCN from proteasomal degradation. In combination with computational molecular dynamics, mutagenesis of both the PA2G4 protein and MYCN oligopeptides highlighted the important amino acids that give significant energy contributions to this protein–protein interface. This site is physically quite distinct from the well described phosphodegron within MYC Box I (aa 58–62) where the E3 ligase, Fbxw7, binds to direct MYC and MYCN toward proteasomal degradation in normal and malignant cells (6), but overlaps another putative MYC phosphodegron (aa 242–250) and MYC mutational hotspot in human lymphoma (50). Whether PA2G4 can interact directly or indirectly with other regions of MYCN, MYC, or other phosphodegron binding proteins such as AurA or Fbxw7, is unknown. Our studies indicate that WS6 directly competes with MYCN binding to PA2G4, thus, it is possible that WS6 will exhibit favorable synergy with inhibitors of AurA or other compounds affecting MYCN stability.

We have defined the amino acids responsible for PA2G4-MYCN binding, and show that this binding site can be competitively blocked with significant antitumor effects in vivo. Our findings have important implications for the design of anti-MYCN therapeutics. We saw no ill-effects in TH-MYCN or xenografted mice treated with WS6. WS6 displayed therapeutic effects, however, the compound itself is not drug-like, but it serves as a very useful positive control and chemical tool for future drug discovery campaigns targeting PA2G4. PA2G4 and MYCN expression levels were very low in normal fibroblasts. Moreover, PA2G4 expression in these normal cells affected neither cell proliferation nor viability, suggesting inhibitors of PA2G4 may have low toxicity in normal tissues. Loss of PA2G4 in knockout mice caused only mildly reduced fecundity and early growth, but adult mice appeared normal (51). In contrast, our experiments showed that WS6 caused significant cytopathic effects in vitro and in vivo in neuroblastoma cell lines and tissues, which correlated with reductions in MYCN and PA2G4 levels. The cytopathic action of WS6 in neuroblastoma cells was blocked by exogenous overexpression of PA2G4, indicating specificity for the target. The interaction between the MYCN amino acid sequence, GGDHKALSTGEDTL, and a pronounced surface crevice in PA2G4 defined by residues Ser47, Glu50, Lys93, Arg271, Arg272, and Arg281, represents a novel protein–protein interface for future anti-MYCN drug design. Small-molecule drug discovery projects aimed at large, flat, featureless protein surfaces for, what may be, transient protein–protein interactions have previously been seen as very challenging. However, multiple recent successes in the discovery and development phases of protein–protein interface inhibitor design, and the progress of several of these compounds into successful clinical trials, indicate that these problems can now be overcome (5). The IC₅₀ for WS6 in treated neuroblastoma cells was a relatively high average of 650 nmol/L; however, as a tool inhibitor compound, WS6 can provide the basis for medicinal chemistry and optimization for clinical use in future. Our data points to the exciting prospect of a novel MYCN signal inhibitor with WS6 properties, which competitively blocks PA2G4-MYCN binding and markedly lowers the level of both proteins, providing a completely new focus for the design of MYCN inhibitors.

M. Fischer has received speakers bureau honoraria from Janssen-Cilag and Novartis. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J. Koach, J.K. Holien, H. Massudi, G. Milazzo, J.E. Murray, G. Perini, M.W. Parker, B.B. Cheung, G.M. Marshall

Development of methodology: J. Koach, J.K. Holien, H. Massudi, O.C. Ciampa, M. Herath, G. Milazzo, J.A. McCarroll, L. Doughty, B.B. Cheung

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Koach, J.K. Holien, D.R. Carter, O.C. Ciampa, M. Herath, T. Lim, G. Milazzo, J.E. Murray, J.A. McCarroll, B. Keenan, B.W. Stevenson, M.A. Gorman, J.L. Bell, L. Doughty, S. Hüttelmaier, A. Oberthuer, M. Fischer, X. Zhang, S. Zhu, W.C. Gustafson, G.M. Marshall

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Koach, J.K. Holien, H. Massudi, O.C. Ciampa, M. Herath, T. Lim, J.A. Seneviratne, G. Milazzo, J.A. McCarroll, B. Liu, C. Mayoh, B. Keenan, B.W. Stevenson, J.L. Bell, L. Doughty, A. Oberthuer, A.J. Gifford, T. Liu, S. Zhu, G. Perini, M.W. Parker, B.B. Cheung, G.M. Marshall

Writing, review, and/or revision of the manuscript: J. Koach, J.K. Holien, H. Massudi, D.R. Carter, O.C. Ciampa, J.E. Murray, J.A. McCarroll, M.A. Gorman, L. Doughty, A. Oberthuer, M. Fischer, W.C. Gustafson, M. Haber, M.D. Norris, J.I. Fletcher, M.W. Parker, B.B. Cheung, G.M. Marshall

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Koach, J.K. Holien, O.C. Ciampa, J.A. Seneviratne, L. Doughty

Study supervision: J.K. Holien, S. Zhu, J.I. Fletcher, M.W. Parker, B.B. Cheung, G.M. Marshall

Other (interpretation of histopathologic data): A.J. Gifford

This work was supported by Program Grants (to G.M. Marshall, M.D. Norris, and Michelle Haber) from the National Health and Medical Research Council (NHMRC) Australia (APP1016699), Cancer Institute NSW (10/TPG/1-13), Cancer Council NSW (PG-11-06), and an Australia Postgraduate Research Award, UNSW Sydney, Australia (to J. Koach). This work was also supported by NHMRC Project grant APP1125171 and Neuroblastoma Australia (to G.M. Marshall and B.B. Cheung); Project grant APP571073; Senior Principal Research Fellowship to M.W. Parker, Cure Cancer Australia Foundation, Leukemia Foundation, and 5-point Foundation (Postdoctoral Fellowships to J.K. Holien). Infrastructure support from the NHMRC Independent Research Institutes Infrastructure Support Scheme and the Victorian State Government Operational Infrastructure Support Program to St Vincent's Institute is gratefully acknowledged. The authors thank the Sydney Children's Tumour Bank Network for providing samples and related clinical information for this study.

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.