Positive and negative regulation of TGF-β signaling

Transforming growth factor (TGF)-β and related factors are multifunctional cytokines that regulate growth, differentiation, adhesion and apoptosis of various cell types (Roberts and Sporn, 1990). More than 30 proteins have been identified as members of the TGF-β superfamily, which includes TGF-βs, activins and bone morphogenetic proteins (BMPs). Activins and BMPs play important roles in early embryogenesis (Kingsley, 1994; Hogan, 1996; Harland and Gerhart, 1997); activins induce dorsal mesoderm in Xenopusembryos, whereas BMPs induce ventral mesoderm. BMPs also play critical roles in morphogenesis of various tissues. TGF-β plays an important role in early embryonic development (Goumans et al., 1999), but might play more crucial roles at relatively late stages of development and in adult tissues. TGF-β acts as a potent growth inhibitor for most types of cells, including epithelial cells, endothelial cells, hematopoietic cells and lymphocytes (Roberts and Sporn, 1990; Miyazono et al., 1994). In addition, TGF-β functions as a fibrogenic factor and is responsible for tissue sclerosis of liver, kidney, lung, skin and other tissues.

TGF-β and related factors are produced as dimeric precursors, in which the C-terminal portions form active ligands following proteolytic processing (Miyazono et al., 1993; Kingsley, 1994). The secreted TGF-β-like factors bind to two different types of serine/threonine kinase receptor: type I and type II (Fig. 1; Heldin et al., 1997; Massague, 1998; Zhang and Derynck, 1999). The type II receptor kinases are constitutively active; upon ligand binding, the type II receptors activate the type I receptor kinases through phosphorylation of the juxtamembrane domain (mainly the glycine-serine-rich domain or GS domain) of type I receptors. The type I receptor kinases then activate intracellular substrates, the central signal messengers being Smad proteins (Miyazono et al., 2000). Smads include three subclasses: receptor-regulated Smads (R-Smads), common-partner Smads (Co-Smads) and inhibitory Smads (I-Smads). R-Smads are anchored to the cell membrane through membrane-bound proteins, including Smad-anchor for receptor activation (SARA; Tsukazaki et al., 1998). R-Smads directly interact with and become phosphorylated by type I receptors. R-Smads then form complexes with Co-Smads and migrate into the nucleus, where they regulate transcription of target genes (Fig. 1). In mammals, Smad2 and Smad3 are TGF-β/activin-specific R-Smads, whereas Smad1, Smad5 and, presumably, Smad8 are BMP-specific R-Smads. Smad4 is the only Co-Smad in mammals, but two Co-Smads, Smad4α and Smad4β, have been identified in Xenopus(Howell et al., 1999; Masuyama et al., 1999). Smad6 and Smad7 act as I-Smads.

Signaling by TGF-β-like factors is regulated in both positive and negative fashions, and is tightly controlled temporally and spatially through multiple mechanisms at the extracellular, membrane, cytoplasmic and nuclear levels. Positive regulation could be critical for amplification of signaling by TGF-β-like factors. Negative regulation plays an important role in restriction and termination of signaling, and often occurs through a negative feedback loop (Fig. 2). Negative regulation is also crucial in early embryonic development and morphogenetic processes, limiting the range of signaling by TGF-β-like factors and forming a gradient of ligand activity. Signaling by TGF-β-like factors is also regulated through cross-talk with other signal transduction pathways, including MAP kinase pathways and JAK/STAT pathways. Perturbation of the negative regulation of TGF-β signaling might be linked to the pathogenesis of various clinical disorders, especially progression of tumors.

Positive regulation of TGF-β and BMP signaling, especially the induction of ligands and their signaling components, often occurs through the action of TGF-β-like factors themselves. For example, three mammalian isoforms of TGF-β (TGF-β1, TGF-β2, and TGF-β3) are auto- and cross-induced by different TGF-β isoforms (Bascom et al., 1989; Kim et al., 1990; O’Reilly et al., 1992). Nodal and its related proteins, which play an important role in early embryogenesis and act through activin receptors and Smad2 (Nomura and Li, 1998), are also induced by nodal signaling (Meno et al., 1999). In certain types of cell, TGF-β receptors might be induced by ligand stimulation (Bloom et al., 1996).

Transcription factors that function as targets of TGF-β-like factors are also induced by ligand stimulation. TGF-β induces production of a transcription factor, Runx3 (formerly termed PEBP2αC/Cbfa3/AML2), in B lymphocytes (Shi and Stavnezer, 1998). Newly synthesized Runx3 in turn forms a complex with Smad3 activated by TGF-β, and cooperatively induces IgA class switching in B lymphocytes (Hanai et al., 1999). c-Jun is induced by TGF-β (Wong et al., 1999) and regulates the transcription of target genes in concert with Smads (Zhang et al., 1998; Liberati et al., 1999).

Smad signaling is also positively modulated through cross-talk with other signaling pathways. Smads might be activated by tyrosine kinase receptor signals under certain conditions (de Caestecker et al., 1998). TGF-β activates non-Smad pathways, including the three distinct MAP kinase pathways: the Erk, c-Jun N-terminal kinase (JNK) and p38 MAP kinase pathways (Hartsough and Mulder, 1995; Atfi et al., 1997; Adachi-Yamada et al., 1999; Hocevar et al., 1999; Zhou et al., 1999). JNK and p38 MAP kinase phosphorylate c-Jun and ATF-2 (also called CRE-BP-1), respectively. c-Jun is a component of the AP-1 transcription factor, whereas ATF-2 acts as a homodimer as well as as a heterodimer with c-Jun. Smad3 physically interacts with phosphorylated c-Jun and ATF-2, although phosphorylation of c-Jun and ATF-2 appears not to be required for interaction with Smad3. Smad3 and Smad4 then act in concert with c-Jun and ATF-2 in transcriptional activation of target genes (Zhang et al., 1998; Liberati et al., 1999; Sano et al., 1999).

TGF-β is secreted as a latent complex, which must be activated to exhibit its biological effects (Saharinen et al., 1999). TGF-βs are synthesized as precursor forms; the N-terminal portions of the TGF-β precursors are cleaved off, but remain bound to the C-terminal active dimers and maintain them in inactive forms (Miyazono et al., 1993). In contrast, activins and BMPs do not form such TGF-β-like latent complexes and therefore do not require prior activation to exert biological effects; however, their activities are tightly regulated by specific antagonists. Two different types of antagonist have been identified: those that directly bind ligands, and those that belong to the TGF-β superfamily and interfere with binding of ligands to specific receptors.

Various antagonists that directly bind BMPs have been identified. These include noggin, chordin, cerberus and its related proteins, and follistatin. Cerberus, gremlin, caronte, DAN and other structurally related proteins are collectively termed the DAN family (Hsu et al., 1998). Proteins of the DAN family have a conserved cystine-knot motif, which is also found in other growth factors, including TGF-β-like factors (Pearce et al., 1999; Rodriguez Esteban et al., 1999). However, other BMP antagonists lack sequence similarity with each other.

Why are there so many antagonists of BMPs? One important reason may be that these antagonists have distinct expression profiles and regulate different biological responses in vivo. Noggin and chordin are secreted by Spemann’s organizer, and induce neural tissue from ectoderm and dorsalize ventral mesoderm (Piccolo et al., 1996; Zimmerman et al., 1996). Cerberus plays an essential role in formation of head-like structure. A cerberus-like protein, caronte, plays a critical role in the establishment of left-right asymmetry (Rodriguez Esteban et al., 1999; Yokouchi et al., 1999). Limb development is controlled by various BMP antagonists, including noggin, chordin, follistatin and gremlin, which have distinct roles in limb morphogenesis (McMahon et al., 1998; Capdevila et al., 1999; Merino et al., 1999). Noggin is also involved in hair-follicle induction (Botchkarev et al., 1999).

Another important reason may be that these antagonists have different affinities for various BMP isoforms (some of them are termed growth/differentiation factors or GDFs) as well as other factors. Both noggin and chordin directly and specifically bind BMPs with high affinity, and abolish the activity of BMPs. Noggin binds to BMP-2, BMP-4 and GDF-6 with high affinity, but to BMP-7 with low affinity (Zimmerman et al., 1996; Chang and Hemmati-Brivanlou, 1999). Follistatin was originally identified as an antagonist of activins, but it has also been shown to bind BMPs (Yamashita et al., 1995; Iemura et al., 1998). Caronte binds to BMP-4, BMP-7 and nodal, but not to activin A (Rodriguez Esteban et al., 1999; Yokouchi et al., 1999). Cerberus binds to multiple growth factors, including BMPs, nodal-like proteins and a non-TGF-β-superfamily protein, Wnt, but not to TGF-β1 (Hsu et al., 1998; Piccolo et al., 1999). Interestingly, the binding of cerberus to these growth factors occurs through independent binding sites in the cerberus molecule (Piccolo et al., 1999). In contrast, various BMP antagonists might interact with similar domains in the BMP molecules, given that noggin competes with cerberus, gremlin and DAN for binding to BMP-2 (Hsu et al., 1998).

A third reason may be that these antagonists diffuse in tissues at different rates. Chordin is ~120 kDa, which is much larger than other BMP antagonists, and might thus diffuse less efficiently in tissues (Piccolo et al., 1996). Formation of morphogen gradients is critical for pattern formation in early development (Shimizu and Gurdon, 1999): BMPs and activins differentially activate target genes, depending on their concentrations. A gradient of BMP-4 activity is produced by the interaction between BMP-4 and noggin (or chordin; Holley et al., 1996). High concentrations of BMP-4 induce blood formation in Xenopusembryos, whereas complete loss of BMP activity caused by chordin results in neurogenesis (Fig. 2B). Importantly, low concentrations of BMP-4, which are partially antagonized by chordin, lead to formation of muscle.

These BMP antagonists can also exert effects though a negative feedback loop (Fig. 2A); expression of noggin is induced by BMP-2, BMP-4 and BMP-6 in osteoprogenitor cells, and noggin in turn abolishes the bioactivity of BMPs (Gazzero et al., 1998). Interestingly, TGF-β1 also transiently induces the expression of noggin (Gazzero et al., 1998), which suggests that BMP signaling is cross-regulated by TGF-β1 through noggin.

Most of the BMP antagonists act by binding ligands directly, whereas certain activin antagonists bind to receptors, preventing ligand binding. Mammalian lefty-1, lefty-2 and zebrafish antivin form a subfamily of the TGF-β superfamily (Meno et al., 1998, 1999; Thisse and Thisse, 1999). Lefty/antivin plays an essential role in the formation of left-right asymmetry. Lefty and antivin might function as monomers, since these molecules lack both the long α helix loop (α3) required for dimerization and the cysteine residue involved in intermolecular disulfide bridging. Lefty/antivin is an inhibitor of nodal and its related proteins. Because the biological effects of lefty/antivin are overridden by overexpression of the extracellular domains of activin type II receptors, lefty/antivin blocks nodal signaling through competition for binding to type II receptors, and thus functions as a pseudoligand (Meno et al., 1999; Thisse and Thisse 1999). BMPs also bind to activin type II receptors (Yamashita et al., 1995), but it is currently unknown whether lefty/antivin blocks BMP signaling. Nodal upregulates the expression of nodal itself, but at the same time induces the expression of lefty/antivin and regulates its own bioactivity by a negative feedback mechanism (Meno et al., 1999; Saijoh et al., 2000).

An important question remaining to be answered is whether the inhibin α chain acts as a pseudoligand, or whether inhibin exerts activity through binding to specific receptors. Activins are disulfide-linked dimers composed of β chains; in mammals, four β chain isoforms (βA, βB, βC and βE) have been identified. Inhibins are heterodimers composed of α and β chains. Both α and β chains are members of the TGF-β superfamily, but the α chain is distantly related to other members of this superfamily. Inhibin might antagonize activins in certain types of cell. In human hepatoma cells and erythroleukemia cells, inhibin was shown to bind to activin type II receptors through β subunits and to compete with activins for receptor binding. Upon binding of inhibin, activin type II receptors were unable to recruit type I receptors, which are essential for signaling (Xu et al., 1995; Lebrun and Vale, 1997). However, certain bioactivities of inhibins (e.g. inhibition of FSH release) might not be elicited by the antagonistic effects against activins (Lebrun and Vale, 1997); in fact, specific inhibin-receptor-like proteins have been isolated in gonadal tumors obtained from mice lacking the inhibin α subunit (Draper et al., 1998), which suggests that inhibin is not a simple pseudoligand for activin receptors but instead exerts effects through specific cell surface receptors in certain types of cell.

Signaling is regulated at the cell membrane and in the cytoplasm through mechanisms common to both the TGF-β/activin and BMP signaling pathways. BAMBI is a pseudoreceptor for serine/threonine kinase receptors found in Xenopusembryos (Onichtchouk et al., 1999) and exhibits a high degree of sequence similarity to the human nmagene product (Degen et al., 1996). nmais downregulated in metastatic melanoma cell lines (Degen et al., 1996). BAMBI is structurally similar to type I serine/threonine kinase receptors but lacks an intracellular domain. The expression profile of BAMBI is similar to that of BMP-4 in Xenopusembryos, and BAMBI requires BMP signaling for expression. BAMBI stably interacts with various type I serine/threonine kinase receptors, as well as type II receptors, and abolishes signaling by BMPs, activins and TGF-βs. BAMBI might thus be induced by BMPs and auto-regulate BMP signaling; in addition, it might cross-regulate signaling by other members of the TGF-β superfamily.

The FK506-binding immunophilin, FKBP12, interacts with type I receptors for TGF-β family proteins in yeast two-hybrid screens (Wang et al., 1994). FKBP12 is an abundant protein, and binds to a Leu-Pro sequence in the GS domain of type I receptors. Loss of binding of FKBP12 to type I receptors leads to spontaneous receptor activation in the absence of ligand stimulation (Wang et al., 1996; Chen et al., 1997). FKBP12 might therefore safeguard TGF-β-like-factor signaling through protection against ligand-independent activation of type I receptors by type II receptors. However, analysis by gene targeting failed to demonstrate a role of FKBP12 in signaling by TGF-β-superfamily members (Shou et al., 1998). Further studies may be needed to elucidate the functional importance of FKBP12 in the TGF-β signal transduction.

I-Smads belong to the Smad protein family and function as antagonists of R-Smad and Co-Smad signaling. Smads have conserved N- and C-terminal domains termed Mad homology (MH) 1 and MH2 domains, respectively, which are linked by nonconserved linker regions. I-Smads have MH2 domains, but their N-terminal sequences are highly divergent from the MH1 domains of other Smads. In a similar way to R-Smads, I-Smads interact with type I receptors activated by type II receptors; whereas R-Smads rapidly dissociate from activated type I receptors, I-Smads stably interact with them (Fig. 1; Hayashi et al., 1997; Imamura et al., 1997; Nakao et al., 1997; Inoue et al., 1998). I-Smads have also been reported to compete with Co-Smad for formation of complexes with R-Smads (Hata et al., 1998).

Expression of I-Smads is induced by various extracellular stimuli, including peptide growth factors and mechanical stress (Topper et al., 1997; Afrakhte et al., 1998). Cells treated with interferon (IFN)-γ become resistant to the effects of TGF-β (Ulloa et al., 1999). IFN-γ activates the JAK/STAT pathway; STAT1 activated by IFN-γ induces the expression of Smad7, which in turn blocks TGF-β signaling. TGF-β signaling is thus regulated through cross-talk with the IFN-γ/STAT pathway via production of Smad7 (Fig. 2C).

Importantly, I-Smad expression is upregulated by TGF-β/activins and BMPs (Tsuneizumi et al., 1997; Afrakhte et al., 1998; Ishisaki et al., 1998; Takase et al., 1998). Smad7 is a potent inhibitor of signaling by both TGF-β/activins and BMPs, and expression of Smad7 is induced by direct effects of Smad3 and Smad4 on the Smad7promoter (Nagarajan et al., 1999). Smad6 preferentially inhibits BMP signaling rather than TGF-β signaling, and Smad6 expression is induced by direct effects of BMP-activated Smads on the Smad6promoter (Ishida et al., 2000). R-Smads/Co-Smads and I-Smads thus form a negative feedback loop in the signaling pathways activated by the TGF-β superfamily (Fig. 2A). Moreover, Smad7 is an inhibitor of both TGF-β/activins and BMPs; therefore, Smad7 induced by TGF-β cross-regulates the activity of BMPs. Smad6 and Smad7 are highly expressed in pancreatic cancers (Kleeff et al., 1999a,b), which results in resistance of cells to the growth inhibitory effects of TGF-β, and possibly tumor progression.

The Erk MAP kinase pathway is activated by peptide growth factors, including epidermal growth factor (EGF) and hepatocyte growth factor (HGF). Erk phosphorylates serine or threonine residues in the PX(S/T)P or (S/T)P motif in the linker regions of Smads. Although R-Smads phosphorylated by Erk can form complexes with Smad4, they do not migrate into the nucleus, and thereby Erk inhibits the signaling by TGF-β and BMPs (Kretzschmar et al., 1997, 1999). Ras is involved in the activation of the Erk MAP kinase pathway by tyrosine kinase receptors in peptide growth factor signaling. Cells transformed by oncogenic activated Ras become resistant to the growth inhibitory effect of TGF-β; this might be due to the inhibition of nuclear translocation of TGF-β-specific R-Smads (Kretzschmar et al., 1999).

After translocation into the nucleus, Smads regulate transcription of target genes by binding to their consensus DNA sequences, interacting with other transcription factors and recruiting transcriptional coactivators or co-repressors (Fig. 1; Derynck et al., 1998; Miyazono et al., 2000). For example, Smads activated by TGF-β directly bind to the Smad-binding elements (SBE) in the promoter of the gene that encodes plasminogen activator inhibitor 1 (PAI-1; Dennler et al., 1998; Song et al., 1998; Stroschein et al., 1999a). Smads also physically interact with a transcription factor, TFE3, that binds to an E-box sequence located adjacent to SBE in the PAI-1promoter (Hua et al., 1998, 1999). TFE3 thus functionally synergizes with Smads in activation of the PAI-1promoter. Smads also recruit the transcriptional coactivator p300/CBP (Feng et al., 1998; Janknecht et al., 1998; Nishihara et al., 1998; Pouponnot et al., 1998; Shen et al., 1998; Topper et al., 1998), which has histone acetyltransferase activity (Fig. 3). Through acetylation of core histones and possibly other proteins, p300/CBP loosens the structure of chromatin and increases accessibility of transcriptional complexes to the basal transcriptional machinery (Travers, 1999).

Several transcriptional co-repressors interact with Smads. Interestingly, these co-repressors preferentially modulate signaling by TGF-β, but not that by BMPs. A homeodomain protein of the TALE class, TGIF, is the first transcriptional co-repressor shown to interact with Smads (Wotton et al., 1999). c-Ski and its related protein SnoN (Ski-related novel gene) are also Smad-binding transcriptional co-repressors; c-Ski binds to Smad2, Smad3 and Smad4, but not to BMP-activated Smads (Akiyoshi et al., 1999; Sun et al., 1999a; Luo et al., 1999). skiwas discovered as an oncogene in the avian Sloan Kettering retroviruses, and its overexpression leads to transformation of avian fibroblasts (Colmenares and Stavnezer, 1989). TGIF and c-Ski compete with p300/CBP for interaction with TGF-β-specific R-Smads, and they repress the transcription of target genes induced by TGF-β (Fig. 3). Both TGIF and c-Ski recruit histone deacetylases (HDACs) to Smad complexes, which leads to the transcriptional repression of target genes (Akiyoshi et al., 1999; Wotton et al., 1999). Depending on their levels of expression, TGIF and c-Ski could limit the magnitude of TGF-β signaling. Expression of c-Ski is modulated under various conditions; for example, it is induced during differentiation of erythroleukemia cells into megakaryocytic cells (Namciu et al., 1994). When the expression of c-Ski is upregulated during oncogenesis (Nomura et al., 1989), cells become resistant to the growth inhibitory effect of TGF-β and, therefore, transformed (Luo et al., 1999; Sun et al., 1999a).

Since SnoN is structurally related to c-Ski and recruits HDAC1 (Nomura et al., 1999), it might repress the transcription of TGF-β-responsive genes through a mechanism similar to that of c-Ski. However, SnoN is a more effective repressor of transcription induced by Smad2 than of that induced by Smad3 (Stroschein et al., 1999b), and regulates TGF-β activity through a complex mechanism. In the absence of ligand stimulation, SnoN represses the spontaneous activation of TGF-β-responsive genes. Upon TGF-β stimulation and nuclear accumulation of Smad3, SnoN is rapidly degraded by cellular proteasomes (Stroschein et al., 1999b; Sun et al., 1999b). c-Ski is also degraded upon TGF-β stimulation, although the half-life of c-Ski is longer than that of SnoN (Sun et al., 1999b). SnoN might thus transiently disappear from cells after activation of Smad3 by TGF-β and thereby make TGF-β activity more overt. After a time lag, however, the expression of SnoN is induced by TGF-β signaling, and SnoN in turn terminates TGF-β signaling through negative feedback regulation.

Smads are degraded by the ubiquitin-proteasome pathway in both ligand-dependent and ligand-independent fashions. Ubiquitination of proteins occurs by E1 ubiquitin-activating enzymes, E2 ubiquitin-conjugating enzymes and E3 ubiquitin-ligases. Although ubiquitination can occur in the absence of E3 ligases on certain proteins, E3 ligases are important for specific recognition of target proteins and, ultimately, degradation by the 26S proteasomes (Laney and Hochstrasser, 1999). Smurf1 is an E3 ubiquitin-ligase of the Hect family that specifically interacts with Smad1 and Smad5 specific for BMP signaling pathways (Zhu et al., 1999). Smurf1 has two copies of the WW motif, which is responsible for binding to a PY motif (PPXY sequence) in the linker regions of Smad1 and Smad5. Smurf1 degrades Smad1 and Smad5 ligand-independently, limiting the intracellular amounts of Smad1 and Smad5. Since the intracellular pool of Smad4 is limited in cells, and since Smad4 is a component shared by the TGF-β/activin and BMP pathways (Candia et al., 1997), these two pathways might compete with each other for formation of complexes with Smad4 in signal transduction. Ectopic expression of Smurf1 in Xenopusembryos results in the repression of the ability of cells to respond to BMPs and, at the same time, enhances their responsiveness to activins (Zhu et al., 1999). Smurf1 might thus influence the balance of BMP and activin signaling in Xenopusembryos. Whether Smurf1-like molecules for the TGF-β-specific R-Smads or other Smads exist remains to be determined.

The half-life of Smad2 activated by TGF-β is shorter than that of non-activated Smad2. Since treatment by proteasome inhibitors prolongs the half-life of activated Smad2, degradation of activated Smad2 might occur through the ubiquitin-proteasome pathway (Lo and Massague, 1999; Fig. 1). Phosphorylation of Smad2 is not required for the degradation of Smad2; instead, nuclear localization of Smad2 appears to be most important for its ubiquitination (Lo and Massague, 1999). Smad2 activated by TGF-β migrates into the nucleus, where it regulates the transcription of target genes. At the same time, however, proteasomes might begin to degrade nuclear Smads, causing the irreversible termination of TGF-β signaling.

The activities of TGF-β, activins and BMPs are regulated through multiple mechanisms. Regulation by direct binding of ligands by extracellular antagonists appears to be very important for the regulation of BMP activities. There are nearly 20 BMP isoforms in mammals (Kawabata et al., 1998; Kawabata and Miyazono, 2000); these isoforms have different profiles of expression, different affinities for receptors and therefore unique biological activities in vivo. It is thus reasonable to propose that a wide variety of extracellular antagonists that have different specificities regulate the actions of BMPs during early development and morphogenetic processes. Activins and nodal-like proteins exert effects through activin receptors, and they are also regulated by extracellular antagonists. In addition to proteins that directly bind ligands, several activin/nodal antagonists are pseudoligand-type molecules that compete with ligands for receptor binding.

Regulation of TGF-β and BMP signaling at cell membrane and cytoplasmic levels is less specific to each signaling pathway. BAMBI and FKBP12 modulate the effects of both TGF-β/activin and BMP signaling. Smad6 activity is specific to BMP signaling but Smad7 blocks both TGF-β/activin and BMP signaling. Moreover, regulation by Erk occurs in both the TGF-β and BMP signaling pathways. Although a Hect family E3 ligase has been identified only for BMP-specific Smads, other E3 ligases that act on TGF-β-specific R-Smads or other Smads probably exist. Negative modulation in the nucleus by transcriptional corepressors appears to be unique to TGF-β signaling; thus far, negative regulation of BMP signaling in the nucleus has not been demonstrated. Both TGIF and c-Ski specifically bind to Smad2 and Smad3, but not efficiently to BMP-specific R-Smads. Perturbation of TGF-β signaling is one of the most crucial events in tumor progression. Mutations of the TGF-β type II receptor and Smad4 have been identified in various cancers and frequently occur in certain colorectal cancers and pancreatic cancers (Markowitz et al., 1995; Hahn et al., 1996; Miyaki et al., 1999). Current findings have shown that overexpression or constitutive activation of negative regulators of TGF-β signaling (i.e. I-Smads, Ras and c-Ski) might also lead to resistance to the growth-inhibitory effect of TGF-β and transformation of cells. It will be interesting to determine how these regulatory molecules are expressed, activated and orchestrated in the TGF-β and BMP signaling pathways under various physiological and pathological conditions.

I thank the colleagues in my laboratory for valuable discussion.