Smad2 and Smad3 Phosphorylated at Both Linker and COOH-Terminal Regions Transmit Malignant TGF-β Signal in Later Stages of Human Colorectal Cancer

Transforming growth factor (TGF)-β, a potent inhibitor of epithelial cell growth, acts as a tumor suppressor in early cancer. In later stages of cancer, however, TGF-β interacts with the Ras pathway to induce an epithelial-mesenchymal transition toward an invasive and metastatic tumor phenotype (1–3). Tumor-promoting effects of TGF-β have been shown by the ability of agents that block TGF-β signaling to inhibit invasiveness of cancer cell lines in vitro and their metastatic potential in vivo (4–6) and by data showing that TGF-β can directly stimulate growth and motility of cancer cells (7). Mechanisms by which TGF-β acts both to promote and inhibit tumor progression have been extensively studied but remain unclear.

Smads, central mediators converting signals from receptors for TGF-β superfamily members to the nucleus (8), are modular proteins with conserved Mad homology (MH) 1, intermediate linker, and MH2 domains (9). Catalytically active TGF-β type I receptor (TβRI) phosphorylates the COOH-terminal serine residues of receptor-activated Smads (10), which include Smad2 and the highly similar protein Smad3. The linker domain undergoes regulatory phosphorylation by Ras-associated kinases, including extracellular signal-regulated kinase (ERK), c-Jun NH₂-terminal kinase (JNK), and cyclin-dependent kinase (CDK) 4 (11–13). TβRI and Ras-associated kinases differentially phosphorylate Smad2 and Smad3 to create three phosphorylated forms: COOH-terminally phosphorylated Smad2/3 (pSmad2C and pSmad3C), linker phosphorylated Smad2/3 (pSmad2L and pSmad3L), and both linker and COOH-terminally phosphorylated Smad2/3 (pSmad2L/C and pSmad3L/C; refs. 14, 15). Phosphorylated Smad2 and Smad3 rapidly oligomerize with Smad4 and translocate to the nucleus, where they regulate transcription of target genes (16).

Reversible shifting of Smad3-mediated signaling between tumor suppression and oncogenesis in hyperactive Ras-expressing epithelial cells indicates that TβRI-dependent pSmad3C transmits a tumor-suppressive TGF-β signal, whereas oncogenic features such as cell growth and invasion are promoted through the JNK-dependent pSmad3L pathway (17, 18). Linker phosphorylation of Smad3 indirectly inhibits its COOH-terminal phosphorylation and subsequently suppresses tumor-suppressive pSmad3C signaling (Supplementary Fig. S1). TGF-β signaling confers a selective advantage on tumor cells by shifting from a tumor-suppressive TβRI/pSmad3C pathway to an oncogenic JNK/pSmad3L pathway during sporadic human colorectal carcinogenesis (19), an observation that has been extended to hepatic carcinogenesis (20–22). However, the biological significance of pSmad2L/C and pSmad3L/C remains unknown (15).

In this study, we investigated the role of pSmad2L/C and pSmad3L/C in cancer progression. Based on our findings, we conclude that perturbation of autocrine TGF-β signaling by nuclear CDK4 and cytoplasmic JNK accounts for the critical role of Smad in the malignant cell behavior characterizing advanced stages of human cancer.

Generation of mouse embryo–derived fibroblasts. Cells from several Smad3 wild-type (Smad3^+/+) and knockout (Smad3^Δex8/Δex8) littermate embryos (23) were pooled by genotype to generate fibroblasts, which were used for experiments as mouse embryo–derived fibroblasts (MEF) after three passages. Smad2^−/− MEFs were provided by Dr. A. Roberts (National Cancer Institute).

Cell culture. ACBRI 519 human intestinal epithelial cells and COLO320 human colon cancer cells were obtained from Applied Cell Biology Research Institute and Health Science Research Resources Bank, respectively. ACBRI 519 cells were sensitive to growth inhibition by TGF-β and COLO320 cells were resistant (24). Smad2^−/− and Smad3^−/− MEFs were used for infection experiments.

Domain-specific antibodies against phosphorylated Smad2 and Smad3. Nine polyclonal anti–phospho-Smad2 and anti–phospho-Smad3 sera were raised against the phosphorylated linker and COOH-terminal regions of Smad2 and Smad3 by immunization of rabbits with synthetic peptides (Supplementary Fig. S2; ref. 17). Relevant antisera were affinity purified using the phosphorylated peptides as described previously (25).

Constructs. Flag-tagged wild-type or CDK/JNK phosphorylation mutants of Smad2 and Smad3 were inserted into a retroviral vector (pQCXIN; Clontech) containing a puromycin resistance marker as described previously (17).

Retroviral infection and expression of Flag-tagged Smad2 and Smad3 mutants. Smad2^−/− or Smad3^−/− MEFs were infected with retroviral solutions carrying Flag-tagged Smad2 or Smad3 mutants lacking phosphorylation sites in the linker and/or COOH-terminal regions as described previously (17).

Immunoprecipitation and immunoblotting. Immunoprecipitation and immunoblotting were performed as described previously (12, 17). Where indicated, cells were pretreated with a JNK inhibitor SP600125 (Calbiochem), a TβRI inhibitor SB431542 (Calbiochem), or a CDK4 inhibitor (Calbiochem) for 2 h before adding TGF-β or platelet-derived growth factor (PDGF). To detect pSmad2L/C and pSmad3L/C, nuclear extracts were immunoprecipitated with anti-pSmad2/3C antibody and the immunoprecipitates were immunoblotted using anti-pSmad2/3L antibody. Immunoblots were also analyzed using 1 μg/mL anti–c-Myc antibody (Santa Cruz Biotechnology) and 0.1 μg/mL anti–matrix metalloproteinase-9 (anti–MMP-9) antibody (R&D Systems).

Nuclear/cytosolic extracts and in vitro kinase assay. Nuclear and cytosolic extracts were prepared according to the manufacturer's protocol (Qiagen). Activated JNK was immunoprecipitated from cytosolic extract with anti–phospho-JNK antibody (Promega). CDK4 was immunoprecipitated from nuclear extract with anti-CDK4 antibody (Millipore). The immune complexes were collected with protein G-Sepharose (GE Healthcare). Bacterial expression and purification of glutathione S-transferase (GST)-Smad2 and GST-Smad3 were carried out, and kinase reaction was performed as described previously (12).

Immunohistochemical studies. Immunohistochemical analyses of human liver specimens were performed as described previously (19).

[³H]thymidine incorporation. DNA synthesis was measured by incorporation of 1 μCi/mL [³H]thymidine (GE Healthcare) into 5% trichloroacetic acid–precipitable material after a 4-h pulse exposure as described previously (12).

Matrigel invasion assay. Membranes with 8-μm pores covered with Matrigel (BD Biosciences) on the upper surface were coated with type I collagen on the lower side. Where indicated, cells were pretreated with a MMP-2/MMP-9 inhibitor II (Calbiochem) for 1 h before adding TGF-β or PDGF. Infiltrating cells were counted in five randomly selected regions as described previously (17).

Tumor specimens. Pathology reports and histologic slides were reviewed independently by two pathologists at Kansai Medical University Hospital. Sites of cancers included transverse colon (n = 1), sigmoid colon (n = 10), and rectum (n = 9). Tumors were staged at the time of surgery by standard criteria using tumor-node-metastasis–based staging according to the unified international colorectal cancer staging classification (26).

We scored pSmad2 and pSmad3 positivity in cell nuclei as follows: 0, no positivity; 1, <25% of nuclei showing positivity; 2, from 25% to 50%; 3, from 50% to 75%; 4, >75% (20, 21). Written informed consent was obtained from each patient according to the Helsinki Declaration. We also obtained approval for this study from our institutional ethics committee.

Reverse transcription-PCR. Isolation of RNA, RT, and PCR of TGF-β type II receptor (27, 28); Smad2 (29); and Smad4 (30) was performed as described previously (19).

TGF-β signaling converts JNK-dependent pSmad2L into pSmad2L/C. To investigate the role of individual phosphorylation domains of Smad2 and Smad3, we generated a total of nine antibodies directed at each phosphorylation site in the linker and COOH-terminal regions (Fig. 1A; ref. 17). We initially studied the kinetics of Smad2 and Smad3 phosphorylation in MEFs after stimulation by PDGF. PDGF-dependent phosphorylation occurred on one threonine and three serine residues within the linker regions but was not observed at their COOH-terminal regions (Fig. 1B). The linker segments serve as substrates for JNK both in vivo and in vitro at 15 min after PDGF treatment (Fig. 1C). TGF-β signal in the presence of PDGF induced an increase in phosphorylation of Smad2L/C: pSmad2L/C appeared at 15 minutes, reached a maximum at 30 minutes, and returned to the basal level by 2 hours (Fig. 1D). TGF-β treatment induced only weak phosphorylation of Smad3L/C at Thr in comparison with pSmad2L/C at Thr. Because Ras-activated JNK signaling caused nuclear translocation of pSmad3L (Supplementary Fig. S3; ref. 17) and linker phosphorylation of Smad3 indirectly blocked its COOH-terminal phosphorylation (Supplementary Fig. S1), TGF-β signaling in the presence of activated Ras could induce COOH-terminal phosphorylation mainly of pSmad2L at Thr²²⁰ and Ser²⁴⁵.

pSmad2L/C signaling enhances cell invasion by up-regulation of MMP-9. We next assessed whether TGF-β and PDGF treatment induced infiltrative potency in Smad2^−/− MEFs expressing Smad2 mutants lacking individual phosphorylation sites (Fig. 2A). TGF-β and PDGF treatment additively increased the infiltrative potency of Smad2^+/+ MEFs (Fig. 2B). Forced expression of wild-type (WT*) Smad2 in Smad2^−/− MEFs led to recovery of infiltrative potency triggered by TGF-β and/or PDGF stimulation, indicating that Smad2 is essential for the ligand-dependent activities. As expected, Smad2^−/− MEFs carrying Smad2(3SA), which lacks COOH-terminal serine residues, showed no invasion in a chamber assay after either TGF-β or PDGF stimulation. Surprisingly, neither TGF-β nor PDGF treatment could induce any infiltrative potency in Smad2 mutants lacking individual phosphorylation sites at Thr²²⁰, Ser²⁴⁵, Ser²⁵⁰, or Ser²⁵⁵ in the linker region. Taken together, the results indicated that promotion of cell invasion by TGF-β required both complete linker and COOH-terminal phosphorylation of Smad2.

Because MMP-2 and MMP-9 are important in metastasis, we examined whether MMP-2 or MMP-9 is required for TGF-β– or PDGF-induced invasion by Smad2^+/+ MEFs and Smad2^−/− MEFs expressing Smad2WT*. Both basal and ligand-induced invasive activities in the MEFs were inhibited by a MMP-2/MMP-9 inhibitor (Fig. 2C) but were less inhibited by a MMP-13 inhibitor (data not shown). TGF-β and/or PDGF treatment to Smad2^+/+ MEFs did not show induction of MMP-2 expression (data not shown). In addition, because invasion and metastasis of cancer are closely associated with TGF-β–mediated up-regulation of MMP-9 (31), we investigated the effect of linker and/or COOH-terminal phosphorylation of Smad2 on induction of MMP-9 protein. Reflecting the invasion profile of the MEFs (Fig. 2B), introduction of Smad2WT* in Smad2^−/− MEFs led to recovery of ligand-dependent induction of MMP-9 protein. Combined treatment with TGF-β and PDGF of Smad2^−/− MEFs expressing Smad2 mutants lacking phosphorylation sites in either the linker region (Smad2EPSM) or the COOH-terminal region [Smad2(3SA)] could not induce MMP-9 protein (Fig. 2D). In support of this observation, we determined that TGF-β and PDGF-dependent MMP-9 activity each required both linker and COOH-terminal phosphorylation of Smad2 (data not shown). Collectively, the data showed that TGF-β together with PDGF induced MMP-9–mediated cell invasion through the pSmad2L/C pathway.

Linker phosphorylation of Smad2 requires TβRI-dependent COOH-terminal phosphorylation. We also investigated the kinetics of Smad2/3 phosphorylation in response to TGF-β stimulation. In contrast to the selective linker phosphorylation after PDGF treatment (Fig. 1B), TGF-β–dependent phosphorylation of Smad2 and Smad3 occurred not only at all linker sites but also at the COOH-terminal serine residues (Fig. 3A). Subcellular localization and linker/COOH-terminal phosphorylation of Smad2^−/− MEFs expressing Smad2 mutants lacking phosphorylation sites suggested that nuclear accumulation (Supplementary Fig. S4; ref. 32) and linker phosphorylation of Smad2 (Supplementary Fig. S5) required TGF-β–dependent COOH-terminal phosphorylation. In support of this conclusion, exposure to a TβRI inhibitor SB431542 caused loss of not only COOH-terminal phosphorylation but also linker phosphorylation (Fig. 3B).

Nuclear CDK4 converts pSmad2C and pSmad3C into pSmad2L/C and pSmad3L/C. TGF-β induced pSmad2L/C and pSmad3L/C after nuclear translocation of pSmad2C and pSmad3C (Fig. 3C; Supplementary Fig. S5). The results indicated that nuclear kinases could phosphorylate linker segments of pSmad2C and pSmad3C. A previous report suggested that CDK4 phosphorylated the linker region of Smad3 (13). Because CDK4 was localized in the nuclei irrespective of whether or not the Smad2^+/+ MEFs received TGF-β or PDGF treatment (Supplementary Fig. S6), and Thr in Smad2/3L, Ser²⁵⁵ in Smad2L, and Ser²¹³ in Smad3L served as substrates for CDK4 both in vivo and in vitro at 90 minutes after TGF-β and PDGF treatment (Fig. 3D), we concluded that CDK4 was involved in linker phosphorylation of pSmad2C and pSmad3C in the nuclei of the MEF.

Stimulation of cell growth by up-regulating c-Myc oncoprotein requires TGF-β–mediated linker phosphorylation. We further investigated whether linker phosphorylation affected growth of the MEF. As shown in Fig. 4A and B, expression of Smad2(3SA) in Smad2^−/− MEFs did not influence growth. However, the T220V, S245A, and S255A mutants showed suppression of growth in response to TGF-β. Smad2EPSM showed the most prominent TGF-β–mediated growth inhibition. These results indicated that Smad2 phosphorylation at Thr²²⁰, Ser²⁴⁵, and Ser²⁵⁵ collectively blocked the growth-inhibitory effect of TGF-β, although Smad2 mutants lacking individual phosphorylation sites in the linker region showed weaker suppression of growth in response to TGF-β than corresponding Smad3 mutants (Supplementary Fig. S7; ref. 13).

TGF-β inhibits cell growth through Smad-mediated transcriptional regulation of critical regulators of the cell cycle (9). The first direct transcriptional target of the TGF-β pathway that explains how this cytokine inhibits cell proliferation is c-Myc (33), whose expression in Smad2^+/+ MEFs and Smad2^−/− MEFs expressing Smad2WT* were initially repressed by TGF-β (Fig. 4C and D); later, however, TGF-β signaling underwent a complete change to induce c-Myc protein. In contrast, TGF-β persistently inhibited c-Myc expression and growth in Smad2^−/− MEFs expressing Smad2EPSM. Because expression of cyclin D1 in Smad2EPSM MEFs appeared at 1 h after addition of TGF-β, and remained high at 24 hours (Supplementary Fig. S8), the cyclin D1-CDK4 complex was active in the cells (34). Taken together, TGF-β inhibited cell growth by down-regulating c-Myc transcription through the pSmad2C and pSmad3C pathway; TGF-β signaling, in turn, enhanced cell growth by up-regulating c-Myc after the additional phosphorylation of pSmad2C and pSmad3C at their linker regions.

Malignant signaling by pSmad2L/C and pSmad3L/C in human metastatic colorectal cancer. We finally investigated whether perturbation of TGF-β signaling involved linker phosphorylation of Smad2 and Smad3 in 20 human metastatic colorectal tumors (Supplementary Table S1). Because perturbation of TGF-β signaling results from mutations that inactivate TGF-β receptors or Smad signal transducers in a portion of gastrointestinal cancers (9), we analyzed mutations of these genes in the human metastatic colorectal cancer samples. MH2 region sequencing in Smad4 gene revealed a G/A substitution at codon 361 in one cancer (Arg to His; patient 5 in Supplementary Table S1), but no mutations were found in TβRII or Smad2 genes from any colorectal tumor samples.

Figure 5A shows the distribution of TGF-β₁ and phosphorylated Smad2/3 in a metastatic colon tumor and nearby uninvolved normal liver tissue from patient 1 in Supplementary Table S1. Immunostaining for TGF-β₁ showed diffuse cytoplasmic positivity throughout the tumor specimen (Fig. 5,A, top left of α TGF-β₁ panel). The degree of Smad2C phosphorylation in the tumor was almost the same as that in normal liver, and pSmad2C was distributed evenly in the nuclei of tumor cells (Fig. 5,A, top left compared with bottom right of α pSmad2C panel). Although Smad3 was slightly phosphorylated at the COOH-terminal region in normal liver cells, Smad3C was highly phosphorylated in the nuclei of the tumor cells (Fig. 5,A, top left compared with bottom right of α pSmad3C panel). Both Smad2L and Smad3L showed little phosphorylation in normal liver (Fig. 5,A, bottom right of α pSmad2/3L Thr, αpSmad2L Ser²⁵⁵, and α pSmad3L Ser²¹³ panels). In contrast, Smad2 and Smad3 in the tumor were highly phosphorylated at the linker regions, and linker-phosphorylated Smad2 and Smad3 were localized in the nuclei of tumor cells (Fig. 5,A, top left of α pSmad2/3L Thr, αpSmad2L Ser²⁵⁵, and α pSmad3L Ser²¹³ panels). In particular, the invasion fronts of two metastatic colon tumors (patients 10 and 16 in Supplementary Table S1) showed strong nuclear signals for pSmad2/3L (Thr; Fig. 5B).

Domain-specific phospho-Smad2/3 immunoblotting of tissue extracts from the same patient showed that Smad2 and Smad3 in the tumor were highly phosphorylated at both linker and COOH-terminal regions (Fig. 5C,, top left). Phospho-Smad2/3L immunoblotting of pSmad2/3C immunoprecipitates from nuclear extracts showed that pSmad2L/C and pSmad3L/C were localized in cell nuclei of TGF-β–producing metastatic colon tumor (Fig. 5C,, top right). Taken together with in vitro experiments, aberrant phosphorylation of Smad2L/C and Smad3L/C could lead to uncontrollable tumor growth and invasion. In vitro kinase assay confirmed that cytosolic JNK and nuclear CDK4 obtained from the tumor tissue could phosphorylate Smad2 or Smad3 at their linker regions (Fig. 5C , bottom). These results pointed to additive requirement of JNK and CDK4 activities for the formation of pSmad2L/C and pSmad3L/C in advanced stages of human colorectal cancer.

Our results provide a new view of TβRI and Ras-associated kinases cooperating to promote cancer progression by up-regulating invasion- and growth-related proteins. Thus, pSmad2L/C and pSmad3L/C can mediate the malignant signaling that allows human metastatic cancer to adopt more invasive and proliferative properties required for progression (Fig. 5D).

In contrast to the finding of COOH-terminal phosphorylation of Smad2 and Smad3 in almost all cell types and tissues (10), timing, duration, extent, and functional implications of their linker phosphorylation depend on cell type and differ by the stage of cancer progression. Therefore, the role of linker phosphorylation in COOH-terminal phosphorylation has been controversial with varying data, suggesting that Ras-mediated linker phosphorylation either inhibits (11, 17, 19–22) or enhances (36–46) events downstream of TβRI. To explain the disparate findings, we analyzed phospho-Smad2/3 profiles in a single model of colon cancer using a pair of cell lines, human normal intestinal epithelial cells, and human colon cancer cells, to validate our hypothesis that linker phosphorylation changes during cancer progression (Fig. 6).

First, involvement of different kinases may explain for differing outcomes among various cell types and contexts. Normal epithelial cells generally show rapid phosphorylation at Smad2/3L in response to TGF-β (Fig. 6A; ref. 12). The kinases seem to act before Smad2 or Smad3 reaches the nucleus. Both JNK and ERK are localized in the cytoplasm and directly phosphorylate the linker regions, creating pSmad2L and pSmad3L (11, 12, 42). In contrast, TGF-β–induced linker phosphorylation in established mesenchymal cells such as MEFs is very slow. Mesenchymal cells show TGF-β–inducible linker phosphorylation after nuclear translocation of pSmad2/3C (Fig. 6B; present work). As epithelial cells are transformed into cancer cells, they come to exhibit strong constitutive linker phosphorylation (Fig. 6C; ref. 17). Nuclear CDK4, together with cytoplasmic JNK, converts a tumor-suppressive pSmad2/3C signal into malignant characteristics. Thus, TGF-β signaling confers a selective advantage on cancer cells by shifting from pSmad2C and pSmad3C pathway characteristic of mature epithelial cells to the pSmad2L/C and pSmad3L/C pathway, which is more characteristic of the state of flux shown by the activated mesenchymal cells. Loss of epithelial homeostasis and acquisition of a migratory, mesenchymal phenotype are essential for invasion in later stages of human cancer (47). In particular, tumors with features characteristic of epithelial-mesenchymal transition have been found at invasion fronts, where strong phosphorylation of Smad2L/C and Smad3L/C was observed (Fig. 5B; ref. 2).

The second possibility involves differential duration of linker phosphorylation among various cell types. Although linker phosphorylation is transient after TGF-β treatment of normal epithelial cells (Fig. 6A; ref. 12), TGF-β–inducible phosphorylation is generally persistent in various mesenchymal cells (Fig. 6B; present work). Moreover, constitutive linker phosphorylation is found in almost all types of cancer cells, including Ras-transformed cells and cancer tissues (Fig. 6C; refs. 17, 19–22). Phosphorylation of many transcription factors is controlled by a dynamic interplay between kinases and phosphatases. Several lines of evidences have identified small COOH-terminal domain phosphatase (SCP1-3) as Smad2/3 linker–specific phosphatases (15, 48). SCP1-3 dephosphorylate Ser²⁴⁵, Ser²⁵⁰, and Ser²⁵⁵ sites in Smad2L or Ser²⁰⁴, Ser²⁰⁸, and Ser²¹³ sites in Smad3L (48). Accordingly, SCP1-3 might dephosphorylate the Ser phosphorylation in normal epithelial cells. Linker phosphorylation in mesenchymal cells and cancer cells might show resistance to the phosphatases. Alternatively, these cells may not be able to induce or activate the relevant phosphatases. One should note that phosphorylation at Thr²²⁰ in Smad2L and at Thr¹⁷⁹ in Smad3L could not be dephosphorylated by SCP1-3, although several kinases (CDK4, JNK, and ERK) could phosphorylate the threonine residues (11, 13, 17, 48). In this regard, our current data showed that metastatic tumors were constitutively transmitting malignant signals through the phosphorylated threonine residues (Fig. 5).

The final possibility to consider involves the differences in effect of TGF-β on receptor tyrosine kinase (RTK)–mediated signaling between epithelial and mesenchymal cells. TGF-β and RTKs are known to exert mutually antagonistic effects on cell cycle control and apoptosis in normal epithelial cells (49, 50). Because linker phosphorylation of Smad3 indirectly inhibits its COOH-terminal phosphorylation (Supplementary Fig. S1), the proliferative effect mediated through the RTK-dependent pSmad3L pathway antagonizes TGF-β signaling through the cytostatic pSmad3C pathway in normal epithelial cells (14). In normal epithelial cells, pSmad2L/C and pSmad3L/C rarely exist either in vitro or in vivo (Fig. 6A; refs. 20–22). In contrast, TGF-β and RTKs synergistically promote growth and invasion of mesenchymal cells (1). Blocking either linker or COOH-terminal phosphorylation of Smad2 abrogated the synergistic responses of MEFs to TGF-β and PDGF (Figs. 2 and 4), indicating involvement of pSmad2L/C in this synergistic response of mesenchymal cells.

Thus far, we have investigated the phosphoisoforms of Smad2 and Smad3 in >2,000 human cancer tissue specimens, finding that their distribution correlates well with the pathologic features. In mature epithelial cells, pSmad2C and pSmad3C are predominantly located in the nuclei. In contrast, pSmad2L/C and pSmad3L/C show nuclear localization at invasion fronts of several types of advanced cancers with distant metastasis. Restoration in cancer of the lost tumor-suppressive function observed in normal epithelial cells, together with disruption of the fundamental signaling pathways that enable tumors to grow and invade, would represent an elegant therapeutic approach. Accordingly, we have reason to hope that pharmacologic inhibition of linker phosphorylation can suppress progression of human advanced cancer by causing a shift from malignant to tumor-suppressive TGF-β signaling (17, 22). Additionally, when evaluating effectiveness of targeted therapies for human cancer, domain-specific phosphorylation of Smad2 and Smad3 should be useful biomarkers likely to correlate with patient responses (22).

No potential conflicts of interest were disclosed.

Grant support: Grant-in-Aid for scientific research from the Ministry of Education, Science and Culture of Japan.

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 Dr. R. Derynck (University of California at San Francisco) for providing cDNAs encoding human Smad2 and Smad3; Drs. C. Stuelten and A. Roberts (National Cancer Institute) for providing Smad2^−/− MEFs; Drs. K. Kobayashi and K. Yokote (Chiba University) for providing Smad3^−/− mice; and N. Ohira for assistance with immunoblotting.