Runt-Related Transcription Factor RUNX3 Is a Target of MDM2-Mediated Ubiquitination

Deregulation of proliferation in mammalian cancer cells involves the synergistic induction of promitogenic regulatory pathways and inactivation of antimitogenic factors. The regulatory functions of the two classic tumor suppressors, p53 and retinoblastoma protein (pRB), are well understood. Accumulating evidence indicates that the cell fate–determining Runt-related (RUNX) transcription factors play complementary biological roles to the p53 and pRB tumor suppressors in both hematopoietic and nonhematopoietic cancers, including cancers of the gastrointestinal tract as well as breast and prostate (1). Although tumor-related modifications in the p53-pRB network are evident in the same spectrum of tumors, a fundamental question is whether and how RUNX-responsive regulatory pathways are mechanistically linked to p53-pRB–dependent cellular surveillance mechanisms.

Several oncogenic and DNA damage–related pathways converge on p53, which is the key effector of a critical checkpoint that arrests the cell cycle (2). The p53 protein is normally labile and maintained at very low levels. Mouse double minute 2 (MDM2; HDM2 in human, called here MDM2), a RING finger E3 ubiquitin ligase, plays an essential role in p53 degradation by promoting ubiquitination of p53, which induces nuclear export and proteasomal degradation of p53 (3). The stabilization of p53 and its activation is mediated by the p14^ARF (human)-p19 ^ARF (mouse)/MDM2 pathway in response to oncogenic stimulation or by DNA damage–dependent activation through the ATM/ATR kinase pathway (4). p14^ARF inhibits the ubiquitin ligase activity of MDM2 by binding to a central region of MDM2. MDM2 can also inhibit p53-mediated transcriptional activation in a ubiquitination-independent manner (5). MDM2 also has p53-independent functions as reflected by elevation of MDM2 activity in patients with p53 mutations (6, 7) and MDM2-mediated growth advantage in the absence of p53 (8, 9).

RUNX transcription factors play pivotal roles in normal development and neoplasia (10). Deregulation of the biological functions of the three human RUNX family genes, RUNX1/AML1, RUNX2/CBFA1, and RUNX3/PEBP2αC (11), contributes to cancer. RUNX1 is required for hematopoiesis and is genetically altered in leukemia (12–14). RUNX2 is linked to osteogenesis (15, 16) and alterations in human RUNX2 levels are associated with cleidocranial dysplasia (17, 18) and osteosarcoma (19). RUNX3 is required for the development of CD8-lineage T cells (20, 21) and TrkC-dependent dorsal root ganglion neurons (22, 23). RUNX3 is the smallest member of the RUNX family and is prototypical for the tumor-suppressive potential of these proteins (24). For example, lower levels of RUNX3 have been shown to be causally associated with human cancers for stomach (25), bladder (26), and colon (27). Recently, RUNX3 was also identified as one of the five most informative genes for the CpG island methylator phenotype of colorectal cancer (28).

Because p53 and RUNX3 both control cell cycle progression and apoptotic processes (25, 29, 30), we postulate that these two proteins may be controlled by the same inhibitory pathway. Our results show that RUNX3 is stabilized by oncogenic Ras-dependent induction of the p14^ARF-MDM2 pathway. MDM2 interacts with RUNX3 and suppresses the transcriptional activity of RUNX3 activity by blocking its transactivation potential as well as by facilitating MDM2-mediated ubiquitination and nuclear exclusion of RUNX3. Our data reveal that RUNX3 and p53 are both connected to the MDM2 pathway. This key finding indicates that the MDM2 pathway simultaneously controls two major tumor suppressor pathways with ubiquitous (p53) and lineage-specific (RUNX3) functions.

Plasmids and antibodies. Full-length cDNA as well as serial deletion and point mutants of RUNX3 (NM_004350), MDM2 (NM_002392), and p14^ARF (NM_000077), respectively, were amplified by PCR and subcloned into pCS4-3Myc or pCS4-3HA. Anti-RUNX3 (5G4) and anti-MDM2 (SMP14, D-12) antibodies were purchased from Abcam and Santa Cruz Biotechnology, respectively.

Cell culture and transfection. Human embryonic kidney cells (HEK-293) and HeLa cells were maintained in DMEM and MKN45 was maintained in RPMI (Life Technologies) supplemented with 10% fetal bovine serum (Life Technologies) and 100 units/mL penicillin-streptomycin (Life Technologies) at 37°C in a humidified atmosphere with 5% CO₂. Cell lines were obtained from Korea Research Institute of Bioscience and Biotechnology. Transient transfection was carried out using Lipofectamine Plus reagent (Invitrogen) according to the manufacturer's instruction. The small interfering RNA (siRNA) for MDM2 (si-MDM2: 5′-UUACAGCACCAUCAGUAGGUACAGA-3′; Invitrogen) and RUNX3 (si-RX3-1: 5′-AACCUGAUGCCAUAGACUC-3′ and si-RX3-2: 5′-UGUUCUCAAACCAUCUCUG-3′; Bioneer) were used for knockdown MDM2 or RUNX3, respectively. Reporter assays, immunoprecipitation, and immunoblotting analysis were done as described previously (29).

In vitro ubiquitination assay. GST-MDM2 and His-tagged Runt domain were overexpressed in Escherichia coli BL21 (DE3) and purified according to standard procedures. Ubiquitination assays were carried out by adding 20 ng each of human recombinant ubiquitin-activating enzyme (E1), ubiquitin-conjugating enzyme GST-tagged UbcH5b (E2), and HA-tagged ubiquitin, 300 ng purified GST-MDM2, and 100 to 400 ng His-Runt domain in ubiquitination reaction buffer [50 mmol/L HEPES (pH 7.4), 2.5 mmol/L precoupled Mg-ATP solution, 0.5 mmol/L phenylmethylsulfonyl fluoride]. Reactions were carried out for 2 h at 37°C.

Apoptosis assay by Annexin V staining. HEK-293 cells were treated with RUNX3-specific siRNA (si-RX3-1 or si-RX3-2) or nonspecific siRNA (si-con) at 50 nmol/L with Lipofectamine RNAiMAX (Invitrogen). Twenty-four hours after siRNA treatment, the cells were transfected with RUNX3 and/or K-Ras-V12 expression plasmid and incubated in serum-free DMEM for 4 h. The medium was changed with 0.5% fetal bovine serum/DMEM and further cultured 44 h. The cells were harvested and treated with Annexin V-FITC and propidium iodide according to the manufacturer's instruction (FITC-Annexin V apoptosis detection Kit1; BD Biosciences, Pharmingen) and analyzed by flow cytometry using BD FACSCalibur (BD Biosciences, Pharmingen).

Ras activation elevates RUNX3 protein levels. Activating K-Ras mutations are known to modulate the p14^ARF-MDM2 pathway. We examined whether an activated Ras mutant controls RUNX3 protein expression. Human embryonic kidney (HEK-293) cells were transfected with a K-Ras mutant that is constitutively activated by a Val¹² mutation or an inactivating Asn¹⁷ mutation. Increased expression of K-Ras-V12 results in a dose-dependent increase in RUNX3 protein levels, but no increase in RUNX3 protein is observed with the Ras-17N mutant (Fig. 1A). Dose-dependent increases in the levels of endogenous RUNX3 protein and known transcriptional targets of RUNX3, p21^WAF/CIP and Bim, were also observed on exogenous expression of K-Ras-V12 (Fig. 1B). The induction of p21 and Bim by K-Ras-V12 did not occur when RUNX3 was depleted by RUNX3 siRNA knockdown, suggesting an essential role for RUNX3 in the K-Ras-V12–dependent induction of p21 and Bim (Supplementary Fig. S1). The increase of RUNX3 protein level appears to be independent from RUNX3 transcription, because RUNX3 mRNA levels are not affected by the expression of K-Ras-V12 (Fig. 1B).

Because all three Runx proteins induce a senescence-like growth arrest in primary mouse embryonic fibroblasts in response to activated Ras (31), we examined whether the levels of RUNX1 and RUNX2 are also increased by K-Ras-V12. Interestingly, however, the levels of the RUNX3 homologues, RUNX1 and RUNX2, were not increased by K-Ras-V12 expression (Supplementary Fig. S2) in HEK-293 cells. This result suggests that RUNX3 is a specific target of K-Ras-V12 at least in HEK-293 cells, and the molecular basis for this selectivity remains to be resolved. We next examined the involvement of p14^ARF in the K-Ras-V12–induced increase of RUNX3 levels. Forced expression of K-Ras-V12 or p14^ARF in A549 lung cancer cells, which do not express p14^ARF, increases RUNX3 only modestly and expression of both genes further increased the levels of RUNX3 protein (Fig. 1C).

Annexin V-FITC and propidium iodide staining followed by flow cytometric analysis revealed that transient expression of K-Ras-V12 or RUNX3 alone in HEK-293 cells increased the number of apoptotic cells to twice that of mock-treated cells (Fig. 1D). Coexpression of K-Ras-V12 and RUNX3 further increased the number of apoptotic cells by ∼2-fold. Importantly, treatment of RUNX3 siRNA abolished the increase of apoptosis by K-Ras-V12 (Fig. 1D). Knockdown of RUNX3 by specific siRNAs is shown in Supplementary Fig. S1. These data together suggest that RUNX3 is a component of an oncogene surveillance pathway that prevents pathologic consequences of abnormal oncogene activation.

MDM2 interacts with RUNX3. Because the predicted effect of elevating p14^ARF is inhibition of MDM2, the E3 ubiquitin ligase that is the principal attenuator of p53 protein levels (32), we postulated that MDM2 may also directly minimize protein expression of RUNX3 through physical interactions between RUNX3 and MDM2. We coexpressed epitope-tagged HA-RUNX3 or HA-p53 and Flag-MDM2 proteins in HEK-293 followed by coimmunoprecipitation and immunoblotting (immunoprecipitation-Western blot analysis). The results show that MDM2 is present in immunoprecipitates obtained with anti-HA (RUNX3) antibody (Fig. 2A). Comparable amounts of MDM2 are detected in anti-HA (p53) precipitates, suggesting that MDM2 physically interacts with both RUNX3 and p53 (Fig. 2A). The interactions between endogenously expressed RUNX3 and MDM2 were examined in HEK-293 cells that produce two main MDM2 isoforms (90 and 75 kDa). Lysates were analyzed by immunoprecipitation with a specific RUNX3 monoclonal antibody (5G4) followed by immunoblotting with anti-MDM2 or anti-RUNX3 antibody. The results establish that both MDM2 isoforms coimmunoprecipitate with RUNX3 (Fig. 2B), indicating that endogenous RUNX3 interacts with endogenous MDM2.

MDM2 downregulates RUNX3 protein levels. We investigated whether MDM2 levels are inversely correlated not only with the levels of p53 but also with that of RUNX3. Epitope-tagged HA-RUNX3 or HA-p53 were each coexpressed with increasing amounts of Flag-MDM2 in HEK-293 cells and the levels of the expressed proteins were monitored by immunoblotting. Strikingly, expression of MDM2 protein decreases the levels of both p53 and RUNX3 (Fig. 2C). The complementary experiment in which MDM2 is depleted by siRNA administration reveals that a reduction in MDM2 increases endogenous RUNX3 protein levels in both MKN45 gastric carcinoma cells and HEK-293 cells (Fig. 2D).

Acidic domain of MDM2 mediates direct interactions with the Runt-homology domain of RUNX3. To delineate the domains of RUNX3 and MDM2 that are required for the interaction between the two proteins, serial deletion mutants of RUNX3 and MDM2 were examined by immunoprecipitation-Western blot analysis (Supplementary Fig. S3A and C). Deletion of the RUNX3 COOH-terminal region beyond amino acid 234 or the NH₂-terminal region before amino acid 110 does not diminish binding to MDM2 (Supplementary Fig. S3B). However, removal of COOH-terminal region beyond amino acid 187 increased MDM2 binding. Deletion of the highly conserved Runt-homology domain between amino acids 54 and 187 abolishes interactions with MDM2. Taken together, these results indicate that the distal region of the Runt-homology domain (amino acids 110-187) is essential for the RUNX3/MDM2 interaction and amino acids 187 to 234 inhibit the binding.

Complementary immunoprecipitation-Western blot experiments were done with full-length RUNX3 and MDM2 protein deletion mutants to define the domain of MDM2 that interacts with RUNX3. Deletion mutants encompassing the acidic domain of MDM2 were capable of binding full-length RUNX3, but deletion mutants lacking this domain were not (Supplementary Fig. S3C and D). This result suggests that the acidic domain of MDM2 is essential for binding to RUNX3. Notably, deletion of NH₂-terminal region, which interacts with p53, is dispensable for the MDM2/RUNX3 interactions. Thus, MDM2 interacts with p53 and RUNX3 through separate regions.

Because CBFβ, a heterodimeric partner of all three RUNX proteins, recognizes the conserved Runt domain, we examined whether CBFβ interferes with the interaction between RUNX3 and MDM2. Expression of fixed amounts of RUNX3 and MDM2 and progressively increasing amounts of CBFβ2 followed by immunoprecipitation-Western blot analysis revealed that CBFβ2 does not interfere with the RUNX3-MDM2 interaction (Supplementary Fig. S4).

MDM2 ubiquitinates RUNX3. The RING finger domain of MDM2 represents the active site for its E3 ubiquitin ligase activity and mediates the ubiquitination of MDM2 itself and the pRB and p53 tumor suppressors (33). To determine whether RUNX3 is also a target of the E3 ubiquitin ligase activity of MDM2, we performed in vivo ubiquitination assays. A mutant MDM2 protein (C438A) that lacks ubiquitin ligase activity was used as a negative control. Flag-tagged wild-type and C438A mutant MDM2 proteins were coexpressed with Myc-RUNX3 and HA-Ub. Immunoprecipitation-Western blot analysis results reveal that wild-type MDM2 efficiently ubiquitinates RUNX3, but ubiquitination of RUNX3 by the MDM2(C438A) mutant is severely reduced (Fig. 3A). The ubiquitination activity of MDM2 on RUNX3 was further examined under physiologic conditions by a loss-of-function approach. MKN45 cells were treated with siRNA for MDM2 or a nonsilencing RNA as control. Ubiquitination of endogenous RUNX3 is evident under control conditions but decreased on siRNA-mediated knockdown of MDM2 protein (Fig. 3B). Because p14^ARF is a key negative regulator of MDM2, we tested whether p14^ARF inhibits MDM2-mediated RUNX3 ubiquitination. The result revealed that MDM2-mediated ubiquitination of RUNX3 is significantly reduced on coexpression of p14^ARF (Fig. 3C). These results suggest that RUNX3 is a direct target of E3-ubiquitin ligase activity of MDM2 and p14^ARF negatively regulates MDM2-mediated RUNX3 ubiquitination.

RUNX3 protein is ubiquitinated by MDM2 at Lys⁹⁴ and Lys¹⁴⁸. MDM2 conjugates ubiquitin on lysine residues within its target proteins. To define the location of key lysine residues that modulate RUNX3 stability, we examined which RUNX3 deletion mutants are degradable by MDM2. RUNX3 protein deletion mutants were expressed with or without MDM2 and steady-state protein levels were measured by immunoblotting (Fig. 4A). Full-length RUNX3 as well as all other deletion mutants including a mutant protein containing only the NH₂-terminal domain from amino acids 1 to 187 was still downregulated by coexpression of MDM2 protein. Although RUNX3(1-187) interacts with MDM2 more strongly than wild-type RUNX3 or several deletion mutants, this increased binding does not correlate with increased MDM2-mediated degradation. This result suggests that the COOH-terminal region or RUNX3 beyond amino acid 187 may be required for efficient MDM2-mediated RUNX3 degradation.

We deduced that lysines required for MDM2-mediated ubiquitination are located within the Runt-homology domain. To verify this assumption, we examined whether the Runt-homology domain is ubiquitinated in vitro. Ubiquitination analysis using a purified recombinant His-tagged protein spanning the Runt-homology domain of RUNX3 reveals that the Runt-homology domain is ubiquitinated by MDM2 as reflected by a distinct mobility shift (Fig. 4B). Unlike in vivo assays that exhibit polyubiquitination, in vitro assays only permit monoubiquitination on each lysine residue (34). Calculation of the relative molecular mass of the Runt-homology domain (∼17 kDa) and the corresponding ubiquitinated form (∼39 kDa) reveals that the difference in the two bands is ∼22 kDa. Because ubiquitin has a molecular mass of ∼8.5 kDa, the size difference suggests that ubiquitination of RUNX3 occurs at either two or three sites.

To pinpoint lysine residues responsible for ubiquitination of RUNX3, we substituted the five lysine residues located within the Runt-homology domain to arginines. At least three of these residues (K94, K129, and K148) are located on the surface of the protein based on the three-dimensional structure of the Runt domain (35). Full-length wild-type and lysine mutant RUNX3 were each coexpressed together with MDM2 in HEK-293 cells and sensitivity to MDM2-mediated degradation was examined. The results show that the single lysine mutant K94R is more resistant to MDM2-mediated degradation than lysine mutant K148R, whereas the K129R mutant is as sensitive as the wild-type RUNX3 protein (Fig. 4C). Combination of the K94R and K148R mutations (double-mutant K94R/K148R) generates a stable protein that is more resistant than the single-mutant K94R. Importantly, triple mutation of the three remaining lysine residues within the Runt-homology domain (K129R/K186R/K192R) results in a protein that is as sensitive to MDM2-dependent destabilization as the wild-type RUNX3 protein. Taken together, we conclude that K94 and K148 are key target residues for ubiquitination and K94 may represent the primary site.

We validated the importance of K94 and K148 in regulation of RUNX3 protein levels by in vivo ubiquitination analysis in HEK-293 cells. Mutation of K94 (K94R mutant) virtually abolishes ubiquitination by MDM2, whereas a single mutation of K148 (K148R mutant) merely reduces ubiquitination and mutation of K129 (K129R mutant) has no discernible effect (Fig. 4D). These results suggest that ubiquitination occurs primarily on K94 and secondarily on K148 of RUNX3. It is possible that the two modification events are linked and that ubiquitination of K94 may be required for the second ubiquitination at K148.

MDM2 suppresses the transcriptional activity of RUNX3. MDM2 inhibits p53-mediated transcriptional activity by direct binding to the DNA-binding motif of p53 and by ubiquitination of p53, which results in nuclear export to the cytoplasm. We therefore investigated whether MDM2 downregulates RUNX3-mediated transcriptional activity using the RUNX3-responsive p21^CIP/WAF1 promoter fused to the luciferase reporter gene (29). We find that the transactivation activity of RUNX3 is suppressed by MDM2 in a dose-dependent manner (Fig. 5A). Interestingly, the suppression of RUNX3 activity by MDM2 is released by the coexpression of p14^ARF (Fig. 5B). Reporter gene assays reveal that mutant MDM2(C438A), which lacks ubiquitin ligase activity, also inhibits RUNX3 stimulation of p21 promoter activity although less effectively than wild-type MDM2 (Fig. 5C). This result suggests that the transcriptional activity of RUNX3 is controlled by MDM2 through an ubiquitination-independent mechanism. Consistent with this interpretation, the transactivating activity of the ubiquitination-defective mutant RUNX3-K94R is also suppressed by MDM2 although not to the same extent as wild-type RUNX3 (Fig. 5D). Collectively, these results suggest that, similar to p53, RUNX3 is inactivated by MDM2 by both ubiquitination-dependent destabilization and ubiquitination-independent interference with transcriptional activation.

MDM2 facilitates nuclear export of RUNX3. MDM2 induces nuclear export of p53, which subsequently leads to the degradation of the protein. To investigate the possibility that MDM2 also facilitates nuclear exclusion of RUNX3, we transiently transfected Myc-RUNX3 and Flag-MDM2 proteins in HeLa cells and analyzed the subcellular localization of RUNX3 by immunofluorescence microscopy. RUNX3 alone and MDM2 alone are each exclusively localized in the nucleus (Fig. 6A), corroborating previous studies (36). However, when both genes are expressed within the same cell, RUNX3 becomes localized in the cytoplasm, indicating export from the nucleus (Fig. 6B). However, an ubiquitination-defective MDM2 mutant (MDM2-339; see Supplementary Fig. S3), which lacks the RING finger domain, fails to translocate RUNX3 to cytoplasm (Fig. 6C). Hence, the E3-ubiquitin ligase activity of MDM2 is required for the cytoplasmic sequestration of RUNX3. Consistent with this observation, wild-type MDM2 fails to translocate the ubiquitination-defective mutant RUNX3-K94R to the cytoplasm (Fig. 6D). Fractionated nuclear and cytoplasmic proteins form transfected cells expressing RUNX3 and MDM2 or MDM2-339 was examined by immunoblotting analysis. The results confirmed that MDM2 but not MDM2-339 facilitates cytoplasmic localization of RUNX3 (data not shown). Taken together, these results indicate that MDM2 induces nuclear export of RUNX3 through ubiquitination. We conclude that MDM2 controls RUNX3 through the same mechanisms as documented for p53.

RUNX3 is a tumor suppressor and inactivation of its nuclear function is prevalent in various tumors. Here, we provide insight into a key mechanism by which RUNX3 exerts its biological functions. Our data indicate that RUNX3 is a principal lineage-specific component of the p14^ARF-MDM2 oncogenic surveillance pathway, thus clarifying how deregulation of RUNX3 activity is linked to cancer susceptibility.

Unlike the ubiquitously expressed p53 protein, RUNX3 is expressed in distinct tissues and regulate cell growth and differentiation in a lineage-specific manner. However, RUNX3 has also many similarities with p53. For example, RUNX3 and p53 are very frequently inactivated in various kinds of tumors and the frequencies are ∼50% (25, 37). Both RUNX3 and p53 are DNA-binding transcription factors that directly control the cell cycle and apoptosis (25). Both exert their growth-suppressive effects by regulating analogous molecular pathways; RUNX3 is associated with its ability to induce cell cycle arrest and programmed cell death by transcriptional upregulation of the CDK inhibitor p21^CIP/WAF1 (29) and the apoptotic regulator Bim (Bcl2L11; ref. 30), respectively, and p53 also controls p21^CIP/WAF1 and apoptotic mediators.

Although the p53 pathway has a major role in the cellular protection mechanisms that prevent pathologic consequences of abnormal oncogenic stimulation (4), this pathway often fails to respond to oncogene activation during cancer development. For example, K-Ras codon 12 mutation is frequently found at preneoplastic stages of lung adenocarcinoma, but p53-null mutations are found at even later stages (38, 39). This result suggests that there are p53-independent surveillance mechanisms that guard against unscheduled mitogenic cues. Interestingly, RUNX3 is frequently downregulated in preneoplastic human lung adenocarcinomas (40). This result could possibly suggest that inactivation of RUNX3 may be involved in p53-independent oncogene surveillance mechanisms.

In this study, we investigated how and when RUNX3 is activated and found that RUNX3 can be activated through 14^ARF-MDM2 pathway when Ras is abnormally activated. In normal conditions, MDM2 inactivates the transactivation activity of RUNX3 through protein/protein interactions as well as mediates ubiquitination of RUNX3 that stimulates nuclear export and subsequent degradation of RUNX3. Proteasomal degradation of p53 is a staged event in which MDM2 and other E3 ubiquitin ligases (e.g., COP1 and Pirh2) ubiquitinate p53 sequentially (41, 42). MDM2 stimulates nuclear export of p53 mainly through monoubiquitination or biubiquitination (43, 44). Our results show that MDM2 promotes nuclear export of RUNX3. Similar to p53, MDM2 may support monoubiquitination of Runx3 in the nucleus, thus promoting nuclear export and cytoplasmic polyubiquitination by other ubiquitin E3 ligases before proteasomal delivery.

The NH₂-terminal region of MDM2 directly binds to p53 and inhibits transactivation by p53. This activity is independent of the ability of MDM2 to reduce p53 levels through its COOH-terminal enzymatic domain (5). We found that MDM2 also inhibits RUNX3-mediated transcriptional activity in a concentration-dependent manner even if the ubiquitin-conjugating activity of MDM2 is compromised (C438A mutant). Thus, analogous to p53, MDM2 can inhibit RUNX3 activity through two separate functions: it promotes degradation of RUNX3 and can block transcriptional activation. Our data show that one key difference between p53 and RUNX3 is that the latter binds to the acidic domain of MDM2, a region that is adjacent to the NH₂-terminal domain of MDM2 that interacts with p53 and pRB (45, 46).

The p14^ARF-MDM2-p53 pathway is subject to p53-dependent feedback regulation and resembles a molecular oscillator. MDM2 gene transcription is activated by p53 as a normal part of the mechanisms that resolves a p53-dependent growth arrest in response to DNA damage or oncogenic stimuli. The linkage of this pathway to RUNX3 indicates that p53 can indirectly control RUNX3 levels. In conclusion, our results suggest involvement of RUNX3 in oncogene surveillance and a remarkable dual role for MDM2 in attenuating the levels of p53 and RUNX3 as principal tumor suppressors.

No potential conflicts of interest were disclosed.

Grant support: Creative Research Grant R16-2003-002-01001-02006 from the Korea Science and Engineering Foundation (S.-C. Bae) and NIH grants R01-AR49069 (A.J. van Wijnen) and P01-CA082834 (G.S. Stein).

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.

We thank Jane Lian and Janet Stein for stimulating discussions.