MDM4 inhibition: a novel therapeutic strategy to reactivate p53 in hepatoblastoma

Hepatoblastoma (HB) is the most common pediatric liver malignancy. High-risk patients have poor survival, and current chemotherapies are associated with significant toxicities. Targeted therapies are needed to improve outcomes and patient quality of life. Most HB cases are TP53 wild-type; therefore, we hypothesized that targeting the p53 regulator Murine double minute 4 (MDM4) to reactivate p53 signaling may show efficacy. MDM4 expression was elevated in HB patient samples, and increased expression was strongly correlated with decreased expression of p53 target genes. Treatment with NSC207895 (XI-006), which inhibits MDM4 expression, or ATSP-7041, a stapled peptide dual inhibitor of MDM2 and MDM4, showed significant cytotoxic and antiproliferative effects in HB cells. Similar phenotypes were seen with short hairpin RNA (shRNA)-mediated inhibition of MDM4. Both NSC207895 and ATSP-7041 caused significant upregulation of p53 targets in HB cells. Knocking-down TP53 with shRNA or overexpressing MDM4 led to resistance to NSC207895-mediated cytotoxicity, suggesting that this phenotype is dependent on the MDM4-p53 axis. MDM4 inhibition also showed efficacy in a murine model of HB with significantly decreased tumor weight and increased apoptosis observed in the treatment group. This study demonstrates that inhibition of MDM4 is efficacious in HB by upregulating p53 tumor suppressor signaling.

www.nature.com/scientificreports/ apoptosis in order to cease the propagation of mutations 12 . Specifically, the p53 protein is known to function upstream of apoptosis through transcriptional activation of Bax and Puma and of cell cycle arrest and senescence through Cyclin dependent kinase inhibitor 1A (CDKN1A) 13 . The TP53 gene has been identified as the most frequently inactivated gene in cancer and is mutated in approximately 50% of all cancers 12 . Interestingly, hepatocellular malignancies have a paucity of TP53 mutations. Mutations in the TP53 gene were seen in only 31% of adult hepatocellular carcinoma (HCC) cases in a recent study profiling 363 HCC cases 14 . In comparison, HB cases have shown almost no mutations in this gene in single strand conformation polymorphism analysis and exome sequencing studies 15,16 . Notably, HB tumors have a very low mutation rate of about 2.9 mutations per tumor with recurrent mutations most commonly seen in the CTNNB1 gene that codes for the β-catenin protein 15 . Additionally, normal hepatocytes are not susceptible to p53-mediated apoptosis 17 . Therefore, targeting the major regulators of p53 function to reactivate the pathway is an attractive therapeutic strategy for TP53 wild-type HB. The major regulators of p53 expression and function are Murine double minute 2 (MDM2) and MDM4 (MDMX). MDM2 and MDM4 work together and independently to inhibit and degrade the p53 tumor suppressor protein. MDM2 is known to monoubiquitinate the p53 protein; however, polyubiquitination of p53 requires heterodimer formation between MDM2 and MDM4 18 . In addition, MDM4 can directly inhibit p53 transcriptional activity by binding the transactivation domain and by directly binding and degrading p21 12 . MDM4 amplification and overexpression have been shown in many cancer types, such as melanoma, breast cancer, glioma, and soft tissue sarcoma 12 . Copy number gain of the 1q32.1 chromosomal region was shown in cases of HB and HCC, including in 28 of 56 HB tumors in a study that showed that this was the most frequent allelic imbalance in HB tumors, and MDM4 has been proposed as the candidate oncogene in this amplicon [19][20][21] . Biomarker studies using immunoblotting have shown that expression of MDM4 in HCC correlates with a poor prognosis 22 . In addition, a study of 363 HCC cases showed that MDM4 expression and copy number were significantly increased in cases with wild-type TP53 but low expression of p53 targets, relative to other cases 14 . Otherwise, the literature to date in regards to MDM4 function in liver cancer are limited to in vitro studies in HCC cell lines 22 .
In this paper, we explore the efficacy of inhibition of MDM4, in comparison to targeting MDM2, as a possible therapeutic strategy for HB. We assess the effects of the established MDM4 inhibitor NSC207895 (XI-006), which is known to inhibit gene expression of MDM4 by physically binding its promoter and preventing transcription, induce apoptosis, and activate the p53 signaling pathway 23 , and the peptide dual MDM2 and MDM4 inhibitor ATSP-7041 in a range of in vitro and in vivo assays to show that targeting MDM4 in HB may be an effective treatment strategy for this disease.

MDM4 is expressed in HB patient samples and cell lines.
We first examined gene expression levels of MDM4 and MDM2 along with a panel of p53 target genes in an Affymetrix microarray dataset of 50 HB patients and six normal pediatric liver tissues described in a previous key publication 20 . These analyses showed that gene expression of 22 established p53 target genes used as a readout for p53 activity significantly correlated with both MDM4 (p = 0.000456) and MDM2 (p = 0.000567) expression in the tumor samples (Fig. 1a). Importantly, the expression profiles of these p53 target genes inversely correlated with MDM4 expression (correlation coefficient = -0.52578) but directly correlated with MDM2 expression (correlation coefficient = 0.52736). Further analysis of this dataset did not reveal a correlation between MDM2 and MDM4 expression (Fig. 1b, correlation coefficient = -0.1372, p = 0.3369) and risk group or presence of identified high risk molecular profiles 20,24 . In a second analysis of primary patient data, we examined levels of MDM4 gene expression in a cohort of 18 primary HB samples from Children's Oncology Group (COG) stage III (n = 12) and IV (n = 6) patients with quantitative reverse transcription polymerase chain reaction (qPCR) experiments in comparison to expression in four adjacent uninvolved liver samples. Of the 18 patients, there were eleven (61%) males and 7 (39%) females. The average age at diagnosis of these patients was 41.2 months. All patients presented were TP53 wild-type. In 16 of the 18 patient samples, MDM4 gene expression was higher than average expression in the uninvolved liver samples; in five samples expression was at least 5-fold higher and in four samples expression was at least 10-fold higher (Fig. 1c). Of note, all four samples with at least 10-fold higher expression were from stage IV patients. We also looked at MDM2 mRNA expression as a comparison in these samples; MDM2 expression was only elevated in one of the 18 patient samples compared to the uninvolved liver samples (Fig. 1c). In summary, this data clearly shows that MDM4 gene expression is increased in primary HB tissues and that this correlates with downregulation of the p53 tumor suppressor signaling pathway, supporting further study of MDM4 as a target for disease therapy.
We also analyzed levels of MDM2 and MDM4 expression in the HB cell lines HepG2 25 , Huh-6 26 , HepT1 27 , and B6-2 28 and the terminally differentiated liver cell line HepRG 29 with qPCR experiments. Importantly, according to the literature, all of these cell lines are wild-type for the TP53 gene 15,28,30 . The cell lines all showed clear MDM4 gene expression ( Supplementary Fig. 1a). Immunoblotting for MDM4 with the same five cell lines in comparison to the MCF-7 breast cancer cell line that is established to express high levels of MDM4 protein 31 also showed that they all express MDM4 protein ( Supplementary Fig. 1b,c). Taken together, these data showed clear gene and protein expression of MDM4 in HB cell lines.
MDM4 inhibition leads to apoptosis of HB cell lines. We treated HB cell lines HepG2, Huh-6, HepT1, and B6-2 with NSC207895, which has been shown to inhibit gene expression of MDM4 23 ; ATSP-7041, which is a stapled-peptide dual inhibitor of MDM2 and MDM4 32 ; or the established MDM2 inhibitor Nutlin-3a 33 and compared the responses of the cells to each agent with MTT assays. All four cell lines showed cytotoxic responses to inhibition of MDM4 with NSC207895 (IC 50 0.95 μM -3.03 μM) or ATSP-7041 (IC 50 1.47 μM -7.29 μM) (Fig. 2a-d). HepG2 cells were the most sensitive to both agents (IC 50  www.nature.com/scientificreports/ for NSC207895) while Huh-6 cells were the least sensitive to ATSP-7041 (7.29 μM) and B6-2 cells were the least sensitive to NSC207895 (3.03 μM). Nutlin-3a treatment also resulted in cell death but at much higher concentrations when compared to the other two inhibitors (IC 50 (Fig. 3b). Further, we  (AEN, ALDH4A1, CDKN1A,  DDB2, DUSP1, EDA2R, ESR1, FAS, FDXR, GADD45A, GADD45B, PANK1, PERP, PTCHD4, RPRM, RPS27L,  RRM2B, SPATA18, TRIAP1, TRIM22, WRAP53, ZMAT3), together with the optimal linear fit. (b) Dot plot representing expression of MDM4 and MDM2, together with the optimal linear fit. (c) Bar graph representing normalized mRNA expression of MDM2 and MDM4 analyzed with qPCR experiments with 18 patient samples in comparison to four adjacent uninvolved liver samples. Y-axis shown with a log2 scale. Error bars represent SD. Data shown in b are representative of at least three independent experiments performed with three replicate wells each time. Student's t test (two tailed) *P < 0.05, **P < 0.01, ***P < 0.001 representing comparison of each sample's expression data to all expression data of uninvolved liver samples. www.nature.com/scientificreports/ looked at changes in protein expression of Bax, Puma, p21 (protein product of CDKN1A), MDM4, MDM2, phosphorylated p53 (phospho-p53, serine 15), and p53 in HepG2 and HepT1 cells exposed to ATSP-7041 and NSC207895. MDM4 protein expression was not reduced in HepG2 and HepT1 cells treated with ATSP-7041, which is consistent with the mechanism of action of this drug (Fig. 3c). Protein expression of MDM2, phospho-p53, Puma, and p21 were increased in both cell lines (Fig. 3c). Total p53 protein levels were also increased in HepG2 cells but not in HepT1 cells (Fig. 3c). With NS207895 treatment, MDM4 expression was reduced in both cell lines, supporting the established mechanism of action of this agent (Fig. 3d) 23 . Levels of p53 were unchanged among treated and untreated HepG2 samples but slightly increased in treated HepT1 cells (Fig. 3d, Supplementary Fig. 3). Finally, levels of p21, Bax, Puma, phospho-p53, and MDM2 were increased in both cell lines (Fig. 3d).
Prolonged MDM4 inhibition slows proliferation of HB cell lines and activates p53 signaling. Because p53 controls pathways involved in cellular proliferation and senescence 13 , we investigated changes in cell proliferation that occurred with exposure to doses of ATSP-7041 and NSC207895 lower than the estimated IC 50 concentrations (estimated percent viability of cells exposed to these doses for 48 h shown in   (Fig. 4a,b). To assess whether MDM4 inhibition affected anchorage-independent growth, we treated cells grown on soft agar with the same low doses of ATSP-7041 or NSC207895 (0.05-0.3 μM). With both cell lines, MDM4 inhibition led to a clear decrease in the anchorage-independent growth ability of the cells (p < 0.001, Fig. 4c-f). Therefore, MDM4 inhibition blocked HB cell proliferation at low concentrations. In a second set of experiments, we examined changes in gene expression of p53 transcriptional targets Bax, Puma, CDKN1A, and MDM2 with prolonged exposure to the MDM4 inhibitors with qPCR experiments. Treatment of HepG2 and HepT1 cells with low doses of ATSP-7041 (0.1-0.5 μM for HepG2, 1-3 μM for HepT1) for Student's t test (two tailed) *P < 0.05, **P < 0.01, ***P < 0.001. (b) Bar graphs representing normalized mRNA expression of p53 targets Bax, Puma, CDKN1A, and MDM2 analyzed with qPCR experiments. RNA extracted from HepG2 and HepT1 cells after treatment with 10 μM NSC207895 for 4 or 8 h was compared to that from untreated cells (0 h). Error bars represent SD. Data shown are representative of at least three independent experiments performed with three replicate wells each time. Student's t test (two tailed) *P < 0.05, **P < 0.01, ***P < 0.001. (c) Protein lysis from cells treated with 10 μM ATSP-7041 for 24 h were compared to untreated cells (0 h). Immunoblotting was done with the indicated antibodies, including MDM4 (04-1555, Millipore). β-Actin immunoblotting was used as a loading control. Data shown are representative of at least three independent experiments. (d) Protein lysis from cells treated with 10 μM NSC207895 for 4 or 8 h were compared to untreated cells (0 h). Immunoblotting was done with the indicated antibodies, including MDM4 (04-1555, Millipore). β-Actin immunoblotting was used as a loading control. Data shown are representative of at least three independent experiments. Full length blots for data shown in c and d are presented in Supplementary Fig. 8 (Fig. 5b).
We then verified inhibition of MDM4 protein expression at these lower doses of NSC207895 for 48 h to show that effects on p53 signaling were occurring in an MDM4-dependent manner. Immunoblotting experiments with cells treated with 0.5 μM NSC207895 and ATSP-7041 for 48 h showed clear lower expression of MDM4 protein with NSC207895 treatment but not with ATSP-7041 treatment (Fig. 5c), which is expected since ATSP-7041 does not affect expression levels of MDM4. We also examined induction of PARP and Caspase-3 cleavage at the same 0.5 μM dose of the drugs at 48 h to see if cell death was present under these conditions. With these experiments, we saw clear induction of PARP cleavage but no change in Caspase-3 cleavage (Fig. 5c), indicating Figure 4. Inhibition of MDM4 leads to decreases in proliferation. (a) HepG2 and HepT1 cells were exposed to ATSP-7041 for 120 h and MTT assays were done at 24 h intervals to assess cell number. Data shown are representative of at least three independent experiments performed with three replicate wells each time. Error bars represent SD. Student's t test *P < 0.05, **P < 0.01, ***P < 0.001. (b) HepG2 and HepT1 cells were exposed to NSC207895 for 96 or 120 h and MTT assays were done at 24 h intervals to assess cell number. Data shown are representative of at least three independent experiments performed with three replicate wells each time. Error bars represent SD. Student's t test *P < 0.05, **P < 0.01, ***P < 0.001. (c, e) Cells were maintained in anchorageindependent conditions on a soft agar base with ATSP-7041 or vehicle for 2 to 3 weeks before staining with 500 μl of MTT solution for 4 h to visualize colonies. Images were captured by a VersaDoc Imaging System and colonies were counted with Quantity One software. Bar graphs in c represent colony numbers in indicated conditions. Data shown are representative of at least three independent experiments performed with three replicate wells each time. Error bars represent SD. Student's t test *P < 0.05, **P < 0.01, ***P < 0.001. (d, f) Cells were maintained in anchorage-independent conditions on a soft agar base with NSC207895 or vehicle for 2 to 3 weeks before staining with 500 μl of MTT solution for 4 h to visualize colonies. Images were captured by a VersaDoc Imaging System and colonies were counted with Quantity One software. Bar graphs in d represent colony numbers in indicated conditions. Data shown are representative of at least three independent experiments performed with three replicate wells each time. Error bars represent SD. Student's t test *P < 0.05, **P < 0.01, ***P < 0.001. www.nature.com/scientificreports/ that the cells do not fully undergo apoptosis. This supports our proliferation assays that indicate inhibition of proliferation with exposure of cells to low doses of the inhibitors for 48 h (Fig. 4a,b).

Scientific Reports
Cytotoxic effects of NSC207895 are dependent on MDM4 and p53 signaling. Since treatment of HB cells with NSC207895 led to increases in p53 signaling, we knocked-down TP53 in HepG2 and HepT1 cells with short hairpin RNA (shRNA) targeting TP53 and measured changes in the cytotoxic response to MDM4 inhibition with MTT assays. We reasoned that cells would become resistant to NSC207895 with decreased TP53 expression. Sufficient knock-down of p53 was verified with immunoblotting assays with each cell line with sh-TP53 compared to sh-Luc control (Fig. 6a). In both cell lines, knocking-down TP53 led to significant (p < 0.001) resistance to NSC207895 treatment, most markedly seen in HepT1 cells (IC 50 value of 1.75 μM (sh-Luc) versus > 10 μM (sh-TP53)) ( Fig. 6b . 6c). In addition, MDM2 was also significantly (p < 0.01) less upregulated with NSC207895 treatment with TP53 knock-down in HepG2 cells (11.76 fold change (sh-luc) versus 0.9 fold change (sh-TP53)) ( Fig. 6c). Taken together, these experiments show that the effects of NSC207895 on viability and on tumor suppressor target gene expression are largely dependent on the p53 signaling pathway. www.nature.com/scientificreports/ In a second set of validation experiments, we manipulated levels of MDM4 with knock-down and overexpression experiments. First, we tested the effects of MDM4 overexpression in the setting of NSC207895 treatment. Using the HepG2 and HepT1 cell lines, we established MDM4 overexpression with retroviral (HepT1) and lentiviral (HepG2) overexpression vectors (Fig. 7a). These cells were then treated with NSC207895, and their responses were compared to that of the vector control (vc) cells. The MDM4 overexpressing cells showed significant (p < 0.001) resistance to NSC207895-mediated cytotoxicity when compared to the vc cells (for HepG2, IC 50 value of 5.3 μM (vc) versus > 10 μM (MDM4); for HepT1 cells, IC 50 value of 1.19 μM (vc) versus 3.16 μM (MDM4)) ( Fig. 7c, Supplementary Table 3). Second, we knocked-down expression of MDM4 with lentiviral shRNA targeting MDM4 to see if the observed changes in cells phenocopied their responses to NSC207895. We established HepG2 and HepT1 cell lines with clear knock-down of MDM4 gene (74.5-88.4% knock-down, p < 0.001) and protein expression (Fig. 7b, Supplementary Fig. 4). We conducted MTT assays on five consecutive days to measure changes in proliferation. Indeed, both HepG2 and HepT1 cells with knocked-down MDM4 showed obvious (p < 0.05) decreases in proliferation compared to control cells expressing sh-GFP (Fig. 7d), which was consistent with the effects of NSC207895 on proliferation. Specifically, HepG2 sh-GFP control cells have a growth rate approximately twofold higher and HepT1 sh-GFP control cells have a growth rate approximately 1.5-fold higher than the corresponding cells with knocked-down MDM4 at the 120 h time point.
Finally, a previous study with NSC207895 proposed that the effects of the inhibitor were mediated through off-target effects on the genes EDIL3, FLOT1, HEG1, UTRN, KIF20A, IDH1, and GPSM2 in studies of Ewing sarcoma and osteosarcoma cell lines 34 . We examined gene expression of these targets in HepG2 and HepT1  www.nature.com/scientificreports/ placebo. We treated animals at a dose of 5 mg/kg every other day for three weeks (Fig. 8a). Treatment was started when animals reached defined criteria (see methods). This occurred from 17 to 35 days after implantation for the 10 animals in the study. With this study, weekly BLI, as well as magnetic resonance imaging (MRI) at a late time point, was conducted to monitor the in vivo growth of tumors and effects of MDM4 inhibition. When the animals were initially separated into treatment and placebo groups, BLI showed that the tumors were very comparable in size (Fig. 8b,  Supplementary Fig. 6a), and BLI measurements during the course of the treatment period showed the growth of tumors with NSC207895 or placebo treatment (Supplementary Fig. 6a). At euthanasia, tumor weights were obtained. The HepT1-derived tumors treated with NSC207895 were significantly smaller than the tumors treated with vehicle (NSC207895-treated tumors were 58.01% the weight of placebo-treated tumors, p = 0.0079) (Fig. 8c).
We then explored whether treatment with NSC207895 was leading to apoptosis by examining cell death in tissues with terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) assays. Tissues from animals treated with NSC207895 showed clear evidence of apoptosis, as indicated by TUNEL staining in the tumor, and quantification of this staining showed a significant (p = 0.0056) difference between the treatment and placebo groups (Fig. 8d,e). We also examined inhibition of proliferation in these tissues with Ki67 staining ( Supplementary Fig. 7c), and although the NSC207895 treated tumors showed a decrease in percent Ki67 positive area compared to the placebo treated tumors (Supplementary Fig. 6b), this did not reach significance. Taken together, this data showed that inhibition of MDM4 shows strong efficacy in an in vivo model of HB. Importantly, treatment with NSC207895 had minimal effects on normal liver tissue in the animals, as shown by TUNEL staining of the neighboring normal mouse liver that appears consistent between drug and placebo treated animals (Supplementary Fig. 6d). www.nature.com/scientificreports/

Discussion
The idea of targeting the negative regulators of p53 in order to reactivate activity of the tumor suppressor protein in cancer cells is a strategy that has been the focus of a substantial amount of research since the identification of MDM2 inhibitors 33 . This significant pharmaceutical effort to target MDM2 has led to nine current agents in clinical development at various stages 35 . Small molecule inhibitors of MDM2 have shown anti-tumor efficacy in preclinical studies but disappointing results in clinical trials with many patients showing toxicities that were not seen in prior mouse studies, including hematological effects, such as thrombocytopenia leading to hemorrhage, and increased incidence of TP53 mutations with prolonged exposure [35][36][37] . Blocking MDM4 regulation of p53 is a more uncharted strategy with no agents specifically targeting only this protein in clinical development. Importantly, the stapled peptide dual inhibitor of MDM2 and MDM4, ALRN-6924, is currently being tested in four active clinical trials for a range of pediatric and adult malignancies, and this peptide shows an improved safety profile from the previous MDM2 small molecule inhibitors 38 . Cancers that are characterized by overexpression of MDM4 are particularly likely to respond to such therapies. Many solid tumors, including gliomas, soft tissue sarcomas, head and neck squamous carcinomas, retinoblastomas, melanomas, breast cancers, and HBs show amplification of the MDM4 gene 12,19 . A previous study of 56 HB tumors identified gains of the 1q chromosome as the most frequent allelic imbalance in HB (28 of 56 tumors) 19 . Further, genomic amplification limited to the 1q32.1 region in which the MDM4 gene is located was noted in four tumors 19 . A second study of 46 HB tumors detected 1q gains in 34% of cases with these gains more frequently identified in high-risk cases 20 . In the present study, we examined levels of MDM4 expression in two cohorts of HB patient samples and showed that gene expression was increased in primary HB tissues, with the highest expression of MDM4 seen in the stage IV patients (Fig. 1). In addition, this work showed that MDM4 gene expression in patient samples significantly correlates with decreased expression of p53 target genes (Fig. 1a), supporting a strong relationship between MDM4 overexpression and p53 tumor suppressor pathway inhibition. In contrast, MDM2 gene expression increased with increased expression of p53 target genes (Fig. 1a), indicating that MDM2 is more a marker of p53 tumor suppressor activity and may not serve as the key regulator of p53 activity in HB. Interestingly, when looking at our HB cell lines, we found a discrepancy in mRNA versus protein expression of MDM4 ( Supplementary Fig. 1). We feel that this is due to well described post-translational modification of the MDM4 protein, including phosphorylation/dephosphorylation and ubiquitination 39 . In most of our immunoblotting experiments, we visualize two bands corresponding to MDM4, which we believe represent these post-translationally modified proteins or to several isoforms that are very close in size. Importantly, these bands are present with two different MDM4 primary antibodies and also in an MCF-7 control cell lysate that represents a cell line that highly expresses MDM4.
We then explored the efficacy of the chemical inhibitor NSC207895 in comparison to effects of the established MDM2 inhibitor, Nutlin-3a, and the stapled peptide dual inhibitor of MDM2 and MDM4, ATSP-7041. NSC207895 was identified in a promoter-based screen for compounds that block MDM4 gene expression 23 and has also been suggested to function as a DNA damaging agent 40 while Nutlin-3a was the most potent agent found in a screen of a chemical library for compounds that bind to the MDM2 protein and inhibit its interaction with p53 33 . ATSP-7041 is a progenitor of the clinical compound ALRN-6924 that specifically binds to and inhibits both major p53 regulators, MDM2 and MDM4 32 . All three inhibitors have been well characterized to upregulate transcription of p53 targets including Bax, Puma, CDKN1A, and MDM2 in TP53 wild-type cells 23,32,33,41,42 . Treatment of cell lines with ATSP-7041 and NSC207895 led to slowing of proliferation at low doses (Fig. 4) and apoptosis at higher doses (Fig. 2). Two prior studies in glioma and melanoma similarly showed anti-oncogenic effects with inhibition of MDM4 with shRNA targeting MDM4 or with SAH-p53-8, a stapled p53 helix that binds to MDM4 to block its interaction with endogenous, wild-type p53 43,44 . Notably, our IC 50 values for cytotoxicity with NSC207895 and ATSP-7041 treatment were as much as 24-fold lower than for MDM2 inhibition with Nutlin-3a (Fig. 2d). Subsequent efforts to optimize Nutlin-3a led to the creation of the second-generation MDM2 inhibitor RG7112 45 and the third-generation inhibitor RG7388 46 , and it is possible that treatment of HB cells with these more potent agents may lead to more effectiveness; however, the enthusiasm for targeting MDM2 in these patients is lessened by the known toxicities that have been described in phase I trials. It is also notable that sensitivity to NSC207895 and ATSP-7041 may be related to expression of MDM4 in that HepG2 was the most sensitive to both agents (Fig. 2d) and had the lowest levels of protein expression of MDM4 ( Supplementary  Fig. 1b,c). At the same time, HepT1 cells had the highest protein expression of MDM4 ( Supplementary Fig. 1b,c) and yet were not the most resistant to either drug (Fig. 2d). It is likely that simple readouts of MDM4 gene and protein expression are not sufficient to predict sensitivity to these drugs that affect a complex cascade of tumor suppressor signaling.
To verify that the effects of NSC207895 and ATSP-7041 were occurring predominantly through reactivation of p53 downstream signaling, we measured changes in expression of p53 targets Bax, Puma, CDKN1A/ p21, and MDM2 (Figs. 3, 5). Increases in gene and protein expression of these targets were seen in both cell lines examined. Interestingly, the two inhibitors showed their maximal effects on gene expression of p53 targets at different time points, NSC207895 at shorter time points of 4 and 8 h and ATSP-7041 at a longer time point of 24 h. We hypothesize that this is due to the different physical nature of the agents as NSC207895 is a small chemical compound and ATSP-7041 is a stapled peptide inhibitor. Further, levels of p53 protein were stable in HepG2 cells treated with NSC207895 (Fig. 3d, Supplementary Fig. 3), showing a unique characteristic of MDM4 inhibition as compared to ubiquitination effects associated with MDM2. On its own, MDM4 acts only to sequester the transcription factor activity of p53 without leading to its degradation 47 . Since the cells are not experiencing additional genotoxic stress with NSC207895, p53 degradation is not affected 47 . We anticipate that if we introduced genotoxic stress in combination with NSC207895 treatment, p53 protein levels would measurably increase. In the setting of NSC207895 treatment, inhibition of MDM4 only releases p53 from this inhibitory www.nature.com/scientificreports/ interaction without changing its protein levels. A key paper that defined the functional relationship between MDM2 and MDM4 similarly showed that knocking-down MDM4 did not affect protein expression of p53 unless MDM2 was also manipulated 48 .
Because of a previous report that suggested that the chemical inhibitor NSC207895 acts on alternative targets in an MDM4/p53-independent manner 34 , we further validated the specificity of the observed effects by knockingdown TP53 or overexpressing MDM4 to change the responsiveness of cells to the agent. HepG2 cells with sh-TP53 showed a shift in IC 50 value from 4.01 μM to 9.94 μM (Fig. 6b), indicating that residual p53 expression is enough to kill cells with MDM4 inhibition or, alternatively, NSC207895 is acting in a p53-independent manner in these cells in the absence of p53. Notably, HepT1 cells expressing sh-TP53 showed resistance to the inhibitor with greater than 50% survival exhibited at doses exceeding the estimated IC 50 value (Fig. 6b). Thus, the observed cytotoxicity predominantly depends on the MDM4-p53 axis. However, we cannot completely rule out minor off-target effects contributing to the observed consequences of exposure. These differences observed between cell lines is likely due to innate characteristics of these cells and their unique MDM4 and p53 signaling pathways. In the aforementioned study, seven genes were suggested to be alternative targets of NSC207895 that mediate the observed effects on cell survival, cell cycle arrest, and senescence, EDIL3, FLOT1, HEG1, UTRN, KIF20A, IDH1, and GPSM2. We examined effects of treatment with the agent on these targets in HepG2 and HepT1 cells with concomitant knock-down of TP53, in comparison to cells expressing control sh-Luc. In these experiments, we found that only KIF20A expression showed a consistent decrease in cells treated with NSC207895 no matter the levels of TP53 (Supplementary Fig. 5) and may contribute to the observed effects of the agent. The KIF20A protein is necessary for normal cleavage furrow ingression and cytokinesis during cell division 49 and has been identified as an oncogene in pancreatic cancer, gastric cancer, glioma, cervical squamous cell carcinoma, lung adenocarcinoma, and HCC [49][50][51][52][53][54][55][56][57] . In addition, a study showed that KIF20A knock-down in Huh-6 HB cells inhibited proliferation without inducing apoptosis 58 . Therefore, although it is clear that our results are predominantly dependent on inhibition of MDM4, it is possible that an effect on KIF20A is also contributing to the observed anti-tumor effects.
To date, much HB preclinical research has utilized subcutaneous models that do not recapitulate the tumor microenvironment of the liver that is so important to the phenotype of HB tumors. This paper represents the first instance that the HepT1 orthotopic xenograft mouse model has been described. Importantly, NSC207895 showed in vivo efficacy with this unique orthotopic xenograft model of HB (Fig. 8, Supplementary Fig. 6). Tissue samples from tumors treated with the inhibitor displayed cell death that was generally absent from placebotreated tumors (Fig. 8d,e). Quantification of this cell death showed a statistically significant difference between placebo and drug treated tumors (Fig. 8d). Similarly, inhibition of MDM4 with SAH-p53-8, an early progenitor of ATSP-7041 and ALRN-6924, showed efficacy in an animal model of melanoma 43 . Interestingly, overexpression of MDM4 in glioma and melanoma mouse models led to enhanced tumorigenesis and blocked the effectiveness of chemotherapies 43,44 , all suggesting that MDM4 may be oncogenic and lead to chemotherapy resistance in vivo.
Taken together, our in vitro and in vivo data from this study strongly supports further development and testing of agents to specifically inhibit MDM4 to reactivate the p53 tumor suppressor pathway. Currently, no targeted agents are approved for adjuvant use with HB patients. Only three cell lines are commercially available, making it difficult to obtain the preclinical data necessary to move agents into clinical trials. Importantly, in all in vitro and in vivo experiments in this study, ATSP-7041 and NSC207895 showed anti-tumor effects by reactivating p53 signaling leading to decreased cellular proliferation and apoptosis. Overall, this study supports further examination of MDM4 inhibition as a clinical strategy for high-risk HB patients. The key to this therapeutic approach is the mutation status of TP53, in that all malignancies with wild type TP53 may respond to MDM4 targeted agents. Therefore, although this study only deals directly with HB tumors, the conclusions may be applicable to many other cancers, including a majority of pediatric cancers and liver cancers. These results support the first clinical trial of ALRN-6924 in pediatric patients (ClinicalTrials.gov Identifier: NCT03654716), which specifically includes enrollment of patients with histological diagnosis of HB and wild-type TP53.

Methods
Patient samples. The patient samples employed in these studies were collected from patients after informed consent from either the patients or their guardians was obtained via an Institutional Review Board-approved tissue collection protocol. All experiments on patient tissue samples were performed in compliance with the Helsinki Declaration and were approved by the Baylor College of Medicine Institutional Review Board.
Gene expression of patient samples. For the cohort of 50 HB tumors and six normal pediatric liver tissues (Fig. 1a), RNA was extracted, mRNA expression was profiled by Affymetrix microarrays, and data was analyzed as described previously 20 . Twenty-two genes established to be transcriptional targets of p53 (AEN,  ALDH4A1, CDKN1A, DDB2, DUSP1, EDA2R, ESR1, FAS, FDXR, GADD45A, GADD45B, PANK1, PERP,  PTCHD4, RPRM, RPS27L, RRM2B, SPATA18, TRIAP1, TRIM22, WRAP53, ZMAT3), including 17 used as a readout for p53 activity in a key publication describing extensive profiling of HCC 14 , were further analyzed as a readout of p53 activity. For genes that had multiple probesets, the probeset that correlated most strongly with TP53 expression was used. Normalized expression profiles of each gene were then summed to get a p53-activity inference, and this inference was correlated with MDM4 and MDM2. MDM4 and MDM2 were also correlated with each other. Correlation was quantified with a standard correlation coefficient, and p values were corrected for the number of probesets tested per gene. For the cohort of 18 HB tumors and four adjacent uninvolved liver samples (Fig. 1b), RNA from frozen hepatic tumor and adjacent normal liver samples was isolated using the mirVana miRNA isolation kit (Ambion, Austin, TX, USA). Samples were treated with DNase 1 and eluted in nuclease-free water. RNA purity and quantity were determined using a spectrophotometer measuring absorb- www.nature.com/scientificreports/ ance at 260/280 nm. cDNA was generated from total RNA with the SuperScript III First-Strand Synthesis System for RT-PCR (Invitrogen). Taqman qPCR was done with TaqMan Universal Master Mix II (Applied Biosystems, Foster, CA, USA) and with the following primers (Applied Biosystems): MDM4 (Hs00910358_s1) and MDM2 (Hs00242813_m1). GAPDH (Hs02758991_g1) was used as an internal control in all qPCR experiments. All experiments were run on a StepOnePlus Real-Time PCR System (Applied Biosystems). Samples were normalized first to GAPDH and then to an average of the uninvolved liver samples using the ΔΔC T method.
Cells and culture conditions. The HepG2, Huh-6, and HepRG cell lines used in this study were com-  The membranes were then incubated with horseradish peroxidase (HRP)-conjugated anti-mouse (7076S) or anti-rabbit (7074S) secondary antibodies (Cell Signaling). Immunoblot band densities were determined with ImageJ (v1.46r, NIH, USA) as previously described 59 . Relative intensity levels were determined by dividing the band intensity of the total protein by the intensity of the loading control protein (β-Actin).
shRNA and overexpression constructs. www.nature.com/scientificreports/ tumor, the animal was euthanized, and the tumor was cut into small pieces (approximately 3 mm 3 ) and serially reimplanted into the livers of ten animals as has been done previously 28 . For this experiment, animals were added to treatment or placebo groups when they satisfied two criteria: (1) consecutive BLI scans with increasing flux values, indicating active tumor growth, and (2) flux value above a threshold of 7 × 10 6 p/s, indicating tumor burden. Mice underwent BLI beginning at 10 days after implantation and twice every week thereafter with the IVIS, and luminescence flux was recorded to assess tumor growth. From 17 to 35 days, animals satisfied criteria, and were alternately added to NSC207895 (n = 5) and placebo (DMSO/saline) (n = 5) treatment groups. NSC207895 was given at a dose of 5 mg/kg every other day for three weeks. At the conclusion of treatment, all animals were euthanized and tissues were harvested for immunohistochemistry. All animals in this study were monitored on a daily basis for signs of pain and distress.
In vivo MRI. MRI was performed on a 1.0 T permanent MRI scanner (M2 system, Aspect Technologies, Israel). A 35 mm volume coil was used for transmit and receive of radiofrequency (RF) signal. Mice were sedated using 3% isoflurane, setup on the MRI animal bed, and then maintained under anesthesia at 1-1.5% isoflurane delivered using a nose cone setup. Body temperature was maintained by circulating hot water through the MRI animal bed. Respiration rate was monitored using a pneumatically controlled pressure pad placed in the abdominal area underneath the animal. Tumor imaging for was performed using a T2-weighted fast-spin echo

Statistical analysis.
All values were presented as mean ± standard deviation (SD). Student's t-Test (twotailed) was used to determine statistical significance, and p values were indicated (p < 0.05 (*), p < 0.01 (**), p < 0.001 (***)). Kruskal Wallis test used to show significance of differences in tumor weights between treatment and placebo groups at time of euthanasia.