C/ebpα represses the oncogenic Runx3–Myc axis in p53-deficient osteosarcoma development

Osteosarcoma (OS) is characterized by TP53 mutations in humans. In mice, loss of p53 triggers OS development, and osteoprogenitor-specific p53-deleted mice are widely used to study the process of osteosarcomagenesis. However, the molecular mechanisms underlying the initiation or progression of OS following or parallel to p53 inactivation remain largely unknown. Here, we examined the role of transcription factors involved in adipogenesis (adipo-TFs) in p53-deficient OS and identified a novel tumor suppressive molecular mechanism mediated by C/ebpα. C/ebpα specifically interacts with Runx3, a p53 deficiency-dependent oncogene, and, in the same manner as p53, decreases the activity of the oncogenic axis of OS, Runx3-Myc, by inhibiting Runx3 DNA binding. The identification of a novel molecular role for C/ebpα in p53-deficient osteosarcomagenesis underscores the importance of the Runx-Myc oncogenic axis as a therapeutic target for OS.


INTRODUCTION
Osteosarcoma (OS) is a malignant bone tumor of mesenchymal cell origin, and there are few effective targeted therapies for the treatment of this malignancy. In humans, the high frequency of genetical inactivation of TP53 in sporadic OS [1]; in 95% [2] or 98% [3] of pediatric OS, and its germline mutation in Li-Fraumeni syndrome, which shows a high incidence of OS [4], support the role of p53 as the critical tumor suppressor in osteosarcomagenesis. In mice, loss of p53 in osteoprogenitor or mesenchymal stromal cells (MSCs) is sufficient for OS development, which has been well-documented using the Osterix (Osx)/Sp7-Cre; p53 fl/fl mouse line [5,6]. However, the molecular mechanisms underlying the initiation or progression of OS following or parallel to p53 inactivation remain largely unknown.
In a previous study, we identified the p53 deficiency-dependent oncogenic role of Runx3, a member of Runx family of transcription factors (TFs), in OS development [3]. In p53-deficient human and mouse OS, Runx3 is markedly upregulated and aberrantly upregulates Myc, the crucial oncogenic TF in OS [7], via the Runx consensus genomic element mR1 in the Myc promoter [3]. Because the DNA-binding ability of Runx3 is inhibited by p53 through a specific protein-protein interaction, Runx3 is considered a p53 deficiency-dependent oncogenic TF in OS [3]. Concurrent loss of p53 and dysregulation of Myc, a pivotal tumor-promoting force in humans and mice [8], is mediated by the oncogenic Runx3. A better understanding of this fundamental mechanism could be achieved by identifying the tumor suppressive factors that inhibit the oncogenic mediator, Runx3.
In the present study, we examined the expression of TFs involved in adipogenesis (adipo-TFs) in p53-null OS cells. The differentiation of MSCs into adipocytes or osteocytes is thought to be mutually exclusive and tightly regulated by TFs [9][10][11]. Analysis of Osx-Cre; p53 fl/fl mice (herein OS mice) and the TARGET (Therapeutically Applicable Research to Generate Effective Treatments) cohort, which possesses genomic alterations in the TP53 gene, showed that adipo-TFs are not uniformly downregulated during osteosarcomagenesis. Some adipo-TFs serve as prognostic factors in human OS patients and their expression correlates with OS tumorigenicity, suggesting that they play tumor suppressive roles in OS. Among them, a CCAAT/enhancer-binding protein, C/ebpα, was identified as the most favorable prognostic factor in human p53-deficient OS. Similar to Pparγ, C/ebpα is a master regulator of adipogenesis [10,12], and its function as a tumor suppressor gene has been reported in human cancers [13]. Repression of C/ebpα promotes the tumorigenic potential of OS cells in humans and mice.
Here, we provide a novel molecular mechanism in which C/ebpα acts as a tumor suppressor in p53-deficient osteosarcomagenesis. C/ebpα specifically interacts with Runx3, a p53 deficiencydependent oncogene, and decreases the activity of the oncogenic axis of OS, Runx3-Myc, by inhibiting Runx3 DNA binding. The anti-OS effects of C/ebpα examined all targeted Runx3, and inhibiting Runx3 reversed the functional defects of C/ebpα.

RESULTS
C/ebpα is a favorable prognostic factor and is downregulated in p53-deficient osteosarcomagenesis p53 inactivation is critical for the development and progression of OS in both humans and mice. The TARGET cohort possessing TP53 alterations [3] and the transcriptome of bone marrow-MSCs (BM-MSCs, or simply MSCs herein) and OS tissues obtained from OS mice were used for assessment. We compared the transcriptome of OS tissues to that of BM-MSCs (Fig. 1A) and analyzed changes in the expression of 45 adipo-TFs ( Fig. 1B and Supplementary Fig.  1A). Of five adipo-TFs with significantly favorable or poor prognostic value in the human OS cohort ( Fig. 1C and Supplementary Fig. 1B), C/ebpα, a favorable prognostic factor, was markedly downregulated in OS tissues that developed from OS mice, whereas the expression of the other four factors showed a moderate increase or decrease (Fig. 1B, D and Supplementary  Fig. 1A). C/EBPα was different from other family members, including C/EBPβ, C/EBPɤ, C/EBPδ, C/EBPε, and C/EBPζ, which showed no prognostic value in the p53-deficient human OS cohort (Fig. 1E). Based on these findings, the present study focused on the tumor suppressive function of C/ebpα. C/ebpα is a tumor suppressor in p53-deficient osteosarcomagenesis in humans and mice The tumor suppressive potential of C/ebpα in vivo was assessed by homozygous deletion of C/ebpα in OS mice, which shortened the lifespan and accelerated OS formation in OS mice ( Fig. 2A, B). C/ ebpα depletion accelerated the onset of OS (Fig. 2B) and shortened the time from onset to death (Fig. 2C). Comparison of gene expression between BM-MSCs and OS cells isolated from OS tissues (mOS cells) showed differences in C/ebpα expression that were distinct from those of the other four prognostic factors (Fig. 2D). In OS mice, the tumorigenic potential of clonal mOS cells was inversely proportional to the expression level of C/ebpα (Fig. 2E). Exogenous expression of C/ebpα reduced the tumorigenicity of the mOS-1 cells listed in Fig. 2D (Fig. 2F). The p53-deficient human OS cell lines MG-63 and HS-Os-1, which endogenously express C/EBPα and have no tumorigenic potential in immunodeficient mice, became tumorigenic upon deletion of C/EBPα (Fig. 2G, H). Furthermore, OS cells isolated from OS tissues of OS; C/ebpα fl/fl mice showed higher proliferating capacity and lower apoptosis levels than OS cells of OS mice ( Supplementary  Fig. 2). These results indicate that C/ebpα is a tumor suppressor in osteosarcomagenesis in humans and mice.
C/ebpα depletion upregulates Myc and increases the tumorigenicity of p53-deficient MSCs Cells of origin of OS are present in BM-MSCs carrying oncogenic gene alterations such as an inactive p53 [14][15][16]. To evaluate the tumor suppressive role of C/ebpα in the development of p53-deficient OS, we isolated MSCs from OS mice and OS; C/ebpα fl/fl mice, and compared their tumorigenic potential. Deletion of C/ebpα strongly increased the tumorigenic potential of MSCs from OS mice (OS;C/ebpα fl/fl MSCs vs. OS MSCs) (Fig. 3A). Loss of p53 resulted in upregulation of Myc in MSCs as demonstrated previously [3] (Supplementary Fig. 3). The tumorigenicity of OS MSCs was weak, but it was dependent on the Myc (Fig. 3A, B; Supplementary Fig. 3). This was confirmed by the finding that both OS; Myc fl/+ MSCs and OS; Runx3 fl/fl MSCs lacking Runx3, which upregulates Myc in the absence of p53 [3], showed little tumorigenic potential (Fig. 3A, B). OS MSCs formed small tumors, some of which contained an osteoid matrix, whereas OS; C/ebpα fl/fl MSCs formed bigger and malignant tumors with higher levels of Myc expression in immunodeficient mice (Fig. 3C).  (Fig. 3F), and C/EBPα and MYC expression was inversely correlated in the human OS cohort (TARGET) (Fig. 3G). These findings suggest that C/ebpα acts as a tumor suppressor by downregulating Myc in p53-null OS development.
Myc is suppressed by C/ebpα in various cell types, and different underlying mechanisms have been proposed [13]. C/ebpα represses E2Fs, which positively regulate Myc [17][18][19]. In addition, C/ebpα inhibits Rb-phosphorylation through an inhibitory proteinprotein interaction with CDK2 and CDK4, thereby indirectly inhibiting E2Fs [20,21]. In the present study, we evaluated the contribution of E2Fs and Rb-phosphorylation to the Myc upregulation caused by loss of C/ebpα in the p53-deficient setting. Treatment with HLM006474, an E2F inhibitor [22], downregulated Myc as well as E2F4, a positive control for HLM006474 treatment in the presence of C/ebpα, but not Myc, which was upregulated in the absence of C/ebpα ( Supplementary  Fig. 4A). This suggests that although E2Fs upregulate Myc, the upregulation of Myc in the absence of C/ebpα in p53-deficient MSCs is independent from E2Fs. There is no significant correlation between inactivation of Rb and tumorigenicity of mOS cells in the development of p53-deficient OS [3]; consistently, we did not find an inverse association between Rb phosphorylation and Cebpα/ CEBPα expression in mOS cells ( Supplementary Fig. 4B, C), HS-Os-1 cells ( Supplementary Fig. 4D), and MSCs ( Supplementary  Fig. 4E, F). Even in the presence of p53, C/ebpα deletion did not promote Rb phosphorylation in MSCs ( Supplementary Fig. 4F). These results suggest that C/ebpα targets other factors essential for Myc regulation in the p53-null setting that are not E2Fs or Rb.
C/ebpα attenuates DNA binding of oncogenic Runx3 to mR1 in the absence of p53 C/ebpα physically interacts with Runx3, as it does with Runx1 [23][24][25] (Fig. 4A). In OS MSCs, the endogenous interaction of C/ ebpα with Runx3 is stronger than that with Runx2 (Fig. 4B). The amount of Runx1 protein was low and no clear endogenous binding to C/ebpα was detected in OS MSCs (date not shown), in line with the smaller contribution of Runx1 to tumorigenesis in OS mice [3]. In mOS cells, exogenously expressed C/ebpα inhibited the binding of Runx3 to mR1, an essential Runx consensus site in the Myc promoter for aberrant upregulation of Myc in the absence  [3] with or without endogenous C/ebpα (left) and their tumorigenicity evaluated by allograft using nude mice (n = 4) (right). **p < 0.01; *p < 0.05. of p53 [3], as revealed by EMSA (Fig. 4C, D) and ChIP (Fig. 4E). In the absence of Runx3 or mR1, disruption of C/ebpα did not increase Myc and the tumorigenicity of OS MSCs, whereas in the presence of both, loss of C/ebpα increased Myc and tumorigenicity (Fig. 4F, G). These results indicate that C/ebpα attenuates Myc upregulation by blocking Runx3 DNA binding to mR1, thereby suppressing the tumorigenicity of OS MSCs.
Loss of C/ebpα does not affect Myc regulation and osteosarcomagenesis in the presence of p53 C/ebpα suppresses oncogenic Runx3, which upregulates Myc via mR1 only in the absence of p53 [3]. Consistent with this finding, in the presence of p53, loss of C/ebpα did not upregulate Myc in ST2 cells and MSCs (Fig. 5A, B), and inducible C/ebpα did not affect Myc expression levels in ST2 cells ( Supplementary Fig. 5). As shown in Fig. 2A, osteoprogenitor-specific deletion of C/ebpα promoted p53-deficient OS development in vivo, whereas deletion of C/ebpα alone resulted in no tumor formation in mice and no difference in lifespan from control mice (Fig. 5C). In addition, MSCs isolated from Osx-Cre; C/ebpα fl/fl mice had no tumorigenicity in immunodeficient mice (data not shown). All these findings suggest that C/ebpα functions as a tumor suppressor in OS development only in the absence of p53. As revealed by micro-CT (µCT) analysis, on the other hand, deletion of C/ebpα by Osx-Cre significantly increased trabecular and cortical bone formation in the adult femur, a common site of OS, although the trend was not observed for all parameters examined (Fig. 5D, E).
The Runx inhibitor AI-10-104 effectively cured p53and C/ ebpα-deficient OS The tumor suppressor C/ebpα targets Runx3 in the development of p53-deficient OS. Treatment with the Runx inhibitor AI-10-104, which inhibits the interaction between Cbfβ and Runx proteins [26], downregulated Myc in both p53-and C/ebpα-negative MSCs (OS; C/ebpα fl/fl MSCs) (Fig. 6A) in a dose-dependent manner (Fig.  6B); however, its effect on Myc downregulation was smaller in p53negative but C/ebpα-positive MSCs (OS MSCs) (Fig. 6A). These results demonstrate that induction of Myc by Runx3 is dependent on both p53 and C/ebpα deficiency. Correspondingly, administration of AI-10-104 had an obvious therapeutic effect on OS; C/ebpα fl/fl mice that developed lower extremity OS (Fig. 6C). The clear pharmacological effect of AI-10-104 may be attributed to its pan-reactivity, simultaneously inhibiting Runx2 which exerts oncogenicity in support of Runx3 in p53-deficient OS development [3]. In fact, AI-10-104 also showed an inhibitory effect on the interaction between Runx2 and Cbfβ in OS; C/ebpα fl/fl MSCs ( Supplementary Fig. 6).
Taken together, the present results indicate that the oncogenic potential of Runx3, which aberrantly upregulates Myc, is suppressed by both tumor suppressors, p53 and C/ebpα, in normal MSCs. The p53-negative MSCs (OS MSCs), in which only C/ ebpα attenuates Runx3 and Runx3 becomes active, are considered to be in a pre-OS state. Once C/ebpα is downregulated, fully unleashed Runx3 abnormally upregulates Myc and promotes osteosarcomagenesis (Fig. 6D).

DISCUSSION
In mice, C/ebpα deletion promoted p53-deficient osteosarcomagenesis and conferred tumorigenicity on p53-deficient MSCs. In humans, C/EBPα was defined as a favorable prognostic factor for OS. The present data indicate that C/ebpα plays a tumor suppressor role in the development of OS, although genetic alterations in the C/ebpα gene have not yet been reported in human OS. To the best of our knowledge, this study is the first to report Osx-positive MSC/osteoprogenitor-specific KO of C/ebpα, an adipo-TF, in mice. In the presence of p53, loss of C/ebpα modestly enhanced bone formation in adult mice without an oncogenic phenotype, suggesting that inhibition of adipogenesis by its deficiency resulted in a relative enhancement of osteogenesis, consistent with the phenotype of systemic C/ebpα-deficient mice [27]. Although the tumor suppressive function of C/ebpα is not obvious in response to its mere deletion, it appears to function as an attenuator against a powerful oncogenic force, i.e., Myc dysregulation by Runx3 in the absence of p53, as if it were a substitute for p53.
Along with TP53 mutations, genetic inactivation of RB-1 plays a key role in human OS development [1]. In mice in which p53 and/ or Rb are deleted by Osx-Cre, OS development is dependent on loss of p53 but not Rb, whereas it is potentiated by loss of Rb [5,6,28], suggesting that E2Fs can promote p53-deficient osteosarcomagenesis. C/ebpα decreases the activity of E2Fs [13,29]. Moreover, C/ebpα inhibits Rb phosphorylation through direct interaction with Cdk2 and Cdk4, leading to Cdk/cyclin kinase inactivation [20,21]. Thus, the possible impact of loss of C/ ebpα on Rb inactivation and/or E2F activation leading to progressive tumorigenicity in OS mice and OS MSCs needs to be examined in more detail. However, in this study, Myc upregulation caused by C/ebpα deficiency was not primarily dependent on Rb phosphorylation or E2F activity.
During normal differentiation, TFs of one lineage (osteogenesis) repress TFs of the other lineage (adipogenesis), thereby completing and maintaining the differentiation state [11]. During osteogenesis, adipo-TFs are generally downregulated; however, in abnormal osteogenesis, namely, osteosarcomagenesis, the tight regulation is disturbed. Among adipo-TFs, the downregulation of C/ebpα suggests that other oncogenic factors involved in OS development repress C/ ebpα. Downregulation of C/ebpα by multiple mechanisms has been recognized in a variety of tumors: by microRNA (miR)-182 in acute myeloid leukemia development [30], by miR-101 in tumor-associated macrophages [31], epigenetically by hyper-methylation of the C/ebpα promoter in head and neck squamous cell carcinoma [32], by HIF-1 in breast cancer hypoxia [33], and by TGFβ in epithelial-tomesenchymal transition mediating breast cancer metastasis [34]. Myc and C/ebpα may also form a negative regulatory loop, as suggested by a previous report that Myc represses C/ebpα expression in hibernoma cells [35]. In any case, oncogenic signaling pathways in the microenvironment of p53-deficient OS undoubtedly play an important role in C/ebpα downregulation.
In addition to C/EBPα, PPARγ, a master regulator of adipogenesis, was identified as the most favorable prognostic adipo-TF in the TARGET cohort analysis (Fig. 1C and Supplementary Fig. 1B), and its tumor suppressive roles in OS have been suggested [15,36]. The inhibitory regulation of Runx2, a master regulator of osteogenesis, by PPARγ has been documented in the context of adipogenic/osteogenic regulation [37]. Furthermore, the oncogenic role of Runx2 in the genesis of p53-deficient OS was previously suggested [3,38,39]. Therefore, although not examined in this study, PPARγ likely functions in a repressive manner against Runx2 and possibly Runx3 in the p53-deficient OS development.
The cooperative actions of C/ebps with Runx1 or Runx2 in hematopoiesis [23] or osteogenesis [40], respectively, have been reported. RUNX1 and C/EBPα act synergistically to increase M-CSF receptor expression via their respective binding sites, which are located adjacent to each other in the promoter of the M-CSF receptor [23]. Runx2 and C/ebpβ directly interact and synergistically increase osteocalcin expression via a C/ebp consensus site in the osteocalcin promoter [40]. In addition, Runxs and C/ebpα compete with each other for binding to an overlapping regulatory element in the promoter region of CD11a integrin in myeloid cells [41]. In contrast to these reports, this study demonstrated the inhibition of Runx3 DNA binding to mR1, an indispensable element (a Runx consensus site) in the Myc promoter for aberrant upregulation of Myc, by C/ebpα through direct protein-protein interaction. This is a previously unreported inhibitory interaction mode between Runxs and C/ebps as oncogenes and anti-oncogenes, similar to the aforementioned inhibitory interaction of E2Fs and C/ebpα, in undifferentiated proliferating cells such as MSCs or in tumorigenic cells lacking p53. In p53-deficient thymic lymphomagenesis, we recently reported that Runx1 shows oncogenic transactivation to upregulate Myc via mR1 [42]. In p53-deficient T-cell lymphomas, the presence or absence of inhibition of Runx1 DNA binding by C/ebpα is also an interesting issue.
This manuscript emphasizes the essentiality of the oncogenic Runx3-Myc axis in p53-deficient osteosarcomagenesis by identifying a novel tumor suppressive role for C/ebpα. C/ebpα functions as an auxiliary brake against oncogenic Runx3 in case p53 inhibition is no longer effective as shown in Fig. 6D. The loss of C/ ebpα causes the pre-OS to develop into OS. A Runx inhibitor was effective even in the absence of both tumor suppressors, p53 and C/ebpα. Runx is a very effective therapeutic target for OS.

MATERIALS AND METHODS Mouse lines
Floxed mouse lines of p53 [43], Runx3 [44], and Myc [45] and an mR1mutated mouse line (mR1 m/m ) [3] were described previously. Sp7/Osx-Cre (no.006361) and a floxed mouse line of C/ebpa (no.006230) lines were purchased from Jackson Laboratory. The line no. 006230, which originally possessed the Mx1-Cre gene, was backcrossed to remove it before crossing with the Sp7/Osx-Cre lines. All mouse studies were performed in the C57BL/6 background using approximately equal numbers of males and females. The details of all animal experiments, including the number of mice (sample size) to be used, were reviewed and approved by the Animal Care and Use Committee of Nagasaki University Graduate School of Biomedical Sciences (no. 2104011709-2). Four mice were housed in each cage. Mice were reared in a pathogen-free environment on a 12 h light cycle at 22 ± 2°C.

BM-MSCs and mOS and human OS cells
All cells used in this study were confirmed to be free of mycoplasma infection and maintained in F12/DMEM supplemented with 10% fetal bovine serum. For the generation of BM-MSCs, BM cells were flushed from the femur of mice with F12/DMEM. Cd11b-and Cd45-negative adherent BM cells, which were negatively selected using a magnetic cell sorting system (MACS; Miltenyi Biotec) consisting of CD11b (no. 130-049-601) and CD45 (no. 130-052-301) MicroBeads and MS Columns (no. 130-042-201), were used as BM-MSCs. Similarly, for the generation of mOS cells, adherent cells obtained from randomly selected mouse OS tissues that were minced and collagenase I-digested were negatively selected using a MACS. Cd11band Cd45-negative OS cells were used as mOS cells. ST2/HS-Os-1 and MG-63 were purchased from RIKEN and JCRB, respectively.

Stable and inducible expression of exogenous genes
MSCs and mOS cells were retrovirally transfected with pMSCV-vector (Clontech/Takara) or pMSCV-mouse-C/ebpα, and selected using puromycin. Resistant cells were used without cloning. Expression of C/ebpα was induced in ST2 cells or p53-deleted (Δp53) ST2 cells using the Retro-X Tet-One Inducible Expression System (Clontech/Takara). Cells were retrovirally transfected with pRetroX-teton-mouse C/ebpα and selected with puromycin. Resistant cells were treated with 100 ng/ml doxycycline (dox) for the indicated times.

RNA-seq
RNA was extracted from MSCs and mOS cells isolated from three individual OS mice using the NucleoSpin RNA kit. RNA quality was assessed on a Bioanalyzer (Agilent). Libraries were prepared using the TruSeq stranded mRNA Library Prep Kit (Illumina). Next-generation sequencing (NGS) libraries were sequenced as 100 bp paired end reads using MGI's DNBSEQ-G400RS, with an average of 40 million reads and an average mapping rate of 83%. RNA counts were quantified using Kallisto3 by pseudo-aligning FASTQ reads to the mouse genome (mm10). RNA-seq data generated and submitted to DDJB sequence read archive with the accession number DRA011168 in the last study [3] were used for OS tissues of OS mice.
In each of the three groups (MSCs, mOS cells, and OS tissues), genes were excluded from subsequent analyses if their average expression counts were zero. Principal component analysis was performed using the prcomp function of R. Differential expression analysis of OS tissues vs. MSCs was performed using DEseq2, and the results were used to draw both MA plots and heatmaps. Of 48 genes with overlapping Gene Ontology terms GO:0003700 (DNA binding TFs) and GO:0045444 (Fat Cell Differentiation), 45 orthologs between humans and mice were considered adipo-TFs in this study.
Gene expression microarray data and the corresponding clinical information from 89 primary OS patients were obtained from the TARGET project (https://ocg.cancer.gov/programs/target) under accession number phs000468 on NCBI dbGaP. Correlation analysis was performed using Spearman's correlation. For survival analysis, 86 datasets possessing both sequencing and clinical information were used. Hazard ratios were calculated with the Cox proportional hazard model after the patients were stratified into low-and high-expressing groups based on the median value for each gene.