Transforming Growth Factor- b Receptor-associated Protein 1 Is a Smad4 Chaperone*

, Members of the transforming growth factor- b (TGF- b ) superfamily signal through unique cell membrane receptor serine-threonine kinases to activate downstream targets. TRAP1 is a previously described 96-kDa cytoplasmic protein shown to bind to TGF- b receptors and suggested to play a role in TGF- b signaling. We now fully characterize the binding properties of TRAP1, and show that it associates strongly with inactive heteromeric TGF- b and activin receptor complexes and is released upon activation of signaling. Moreover, we demonstrate that TRAP1 plays a role in the Smad-mediated signal transduction pathway, interacting with the common mediator, Smad4, in a ligand-dependent fashion. While TRAP1 has only a small stimulatory effect on TGF- b signaling in functional assays, deletion constructs of TRAP1 inhibit TGF- b signaling and diminish the interaction of Smad4 with Smad2. These are the first data to identify a specific molecular chaperone for Smad4, suggesting a model in which TRAP1 brings Smad4 into the vicinity of the receptor complex and facilitates its transfer to the receptor-activated Smad proteins.

In a recently described direct signaling pathway, the activated type-I receptor associates with and phosphorylates a family of receptor-activated Smad proteins, which then dissociate from the receptor, hetero-oligomerize with a common partner Smad4, and translocate to the nucleus, where they take part in transcriptional activation of target genes (4 -6). In addition, extensive cross-talk between the Smad pathway and other signal transduction cascades has been demonstrated, notably the JAK-STAT pathway (7), mitogen-activated protein kinase pathways (8 -10) as well as the vitamin D signaling pathway (11), the glucocorticoid receptor pathway (12), and the Wnt pathway (13).
Efforts to elucidate TGF-␤ signaling pathways have resulted in the identification of a number of other non-Smad proteins which interact directly with T␤RI and/or T␤RII, including SARA (14), STRAP (15), FKBP12 (16), the ␣ subunit of farnesyl transferase (17), TRIP-1 (18), and the B␣ subunit of phosphatase 2A (19). Functional roles in TGF-␤ signal transduction have been proposed for them but remain controversial (2, 19 -22). Of these, SARA, which has been shown to facilitate Smad2/3 recruitment to the activated T␤RI (14), is the only protein directly shown to modulate the Smad signaling pathway.
Recently, a novel protein called TRAP1 was identified in a yeast two-hybrid screen for a protein interacting with a mutationally activated T␤RI (23). Indeed, the C-terminal portion of TRAP1 (⌬TRAP1) was shown to interact specifically with mutationally or ligand-activated T␤RI but not with the quiescent receptor, suggesting that it might participate in signal transduction from the T␤RI. In functional assays, ⌬TRAP1 acted as an inhibitor of TGF-␤-mediated effects, suggesting that it might play a role in modulating this signal transduction pathway.
To better understand the biological role of TRAP1, we have now expressed the full-length molecule, and used biochemical and functional assays to show that TRAP1 differs in its binding characteristics from the previously published ⌬TRAP1 in that it predominantly associates with receptor complexes that are signaling deficient. Moreover, we show that TRAP1 specifically interacts with Smad4 and provide data suggesting that it functions as a chaperone for Smad4 in TGF-␤ signal transduction.

EXPERIMENTAL PROCEDURES
Cell Lines and Transfections-COS-1, HepG2, and NMuMg cells were maintained in high glucose-Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum, 100 units of penicillin/ml, and 100 g of streptomycin/ml. Cells were transfected with the constructs indicated using LipofectAMINE (Life Technologies), Superfect (Quiagen), or FUGENE (Roche Molecular Biochemicals) according to the manufacturer's instructions. Transfection protocols were generated using Cellputer Software. 2 Construction of Plasmids-Epitope-tagged TRAP1 and TRAP1 deletion constructs were polymerase chain reaction-amplified from pCDNA3-TRAP1 using the Roche Molecular Biochemical Expand TM High Fidelity PCR System and cloned into the mammalian expression vector pEFcx (gift from Dr. C. Hill), a derivative of pEF-BOS with a modified multiple cloning site, as well as into pEGFP (CLONTECH). For bacterial expression of GST-tagged fusion proteins, ⌬TRAP1 was subcloned into the pGEX4T3 vector. Smad4-MH2 was cloned into pCDNA4HisMax (InVitrogen) for in vitro translation. All constructs were sequence confirmed.
In Vitro Transcription/Translation-In vitro transcription/translation was carried out using commercially available reticulocyte lysates (InVitrogen) according to the manufacturer's instructions. In short, Smad4 constructs under the control of T7 promoters were incubated with the reticulocyte lysates in the presence of 20 mCi of [ 35 S]methionine (PerkinElmer Life Sciences) for 90 min at 30°C and subsequently incubated with GST-tagged ⌬TRAP1 bound to glutathione-agarose. Following washes in a buffer described for the immunoprecipitation studies, beads were boiled with sample buffer (NOVEX, containing 5% mercaptoethanol) and loaded onto precast SDS gels (NOVEX). Visualization was done by autoradiography.
Subcellular Localization by Immunofluorescent Confocal Microscopy-COS-1 cells were plated onto sterilized glass coverslips (Corning) on day 0, transiently transfected on day 1, serum-starved overnight on day 2, and processed on day 3. Following treatment, cells were fixed in 3.5% paraformaldehyde, permeabilized with 0.5% Triton X-100 in phosphate-buffered saline for 10 min, and incubated for 30 min at room temperature with either of the following primary antibodies: 12CA5 anti-HA mouse monoclonal antibody (1:1000, own production) or M2 anti-Flag mouse monoclonal antibody (1:1000, Sigma). Cells were washed 3 times with phosphate-buffered saline prior to incubation with another primary antibody and then incubated for 30 min with the following secondary antibodies: fluorescein isothiocyanate-conjugated goat anti-mouse antibody (Kirkegaard & Perry, 1:1000) and rhodamineconjugated goat anti-rabbit antibody (Jackson ImmonoResearch, 1:1000). Cells were mounted with medium containing 4,6-diamidino-2phenylindole and then visualized with a Zeiss confocal microscope.
Reporter Assays-HepG2 cells or COS-1 cells (3.5 ϫ 10 5 cells/well) were seeded into six-well tissue culture plates. Cells were transfected using Superfect (Quiagen) or Fugene (Roche Molecular Biochemicals), respectively, according to the manufacturer's instructions with the indicated amounts of DNA. Cells were co-transfected with either with a 3TP-Luciferase reporter (gift from J. Massaguè) or a SBE driven Luciferase reporter (gift from S. Kern). 24 h later, cells were shifted into low serum with or without TGF-␤ (5 ng/ml). Where indicated, luciferase activity was measured following 12-18 h of incubation using a commercially available kit (Pharmingen) and normalized to ␤-galactosidase expression.

TRAP1 Interacts Predominantly with Inactive T␤RI⅐T␤RII Complexes and Dissociates from Activated
Receptor Complexes-To investigate interactions of TRAP1 with TGF-␤ receptors, we have expressed the full-length molecule together with combinations of wild type and kinase-deficient T␤RI and T␤RII in vivo. COS-1 cells were transiently transfected with epitope-tagged receptor (HA) and TRAP1 (Flag) constructs, and cell lysates were analyzed by immunoprecipitation and Western blotting. In cells that overexpress only T␤RI, a weak interaction of full-length TRAP1 can be demonstrated with the kinase-deficient mutant but not the wild type receptor (Figs. 1A and 2A). In cells overexpressing only T␤RII, TRAP1 interacts with either wild type or kinase-deficient receptors, showing a stronger association with the kinase-deficient T␤RII (Figs. 1A and 2B). When both T␤RI and T␤RII are overexpressed simultaneously, in various combinations of wild type and kinase-deficient forms, TRAP1 binds most strongly to those complexes in which at least one of the receptors has been mutationally inactivated (Fig. 1B), making the complex incapable of signaling. A direct effect of TGF-␤ on association of TRAP1 with wild type receptor combinations could not be demonstrated (Fig. 1B), most likely due to a ligand-independent activation of the receptor complex in the context of overexpression of wild type receptors (20,24), a phenomenon which we have confirmed in functional assays (Fig. 3C, and data not shown).
We also examined the intracellular localization of TRAP1 and TGF-␤ receptors using epitope-tagged proteins and indirect immunofluorescence to confirm the above observations. In order to avoid ligand-independent activation of the signaling cascade, we again utilized mutationally activated and kinasedeficient forms of T␤RI to simulate the presence and absence of TGF-␤ signal, as described for the immunoprecipitation assays. As shown in Fig. 1C, TRAP1 co-localizes with TGF-␤ receptors when the receptors are mutationally inactivated, displaying the same patchy distribution pattern previously reported for the receptor complex (14,25,26). In the presence of constitutively activated receptors, however, the intracellular distribution of TRAP1 changes from the patchy pattern to a more diffuse pattern (Fig. 1D). This correlates with a reduction in the degree of TRAP1-receptor association consistent with that observed in immunoprecipitation/Western blotting. Taken together, these data suggest that TRAP1 associates with the TGF-␤ receptor complex and that the type II receptor is the primary binding partner. The fact that associations are strongest in the presence of kinase-deficient receptors suggests that activation of signaling through the TGF-␤ receptor complex results in dissociation of TRAP1 from this complex, in striking distinction to previously published data for the C-terminal fragment of TRAP1, ⌬TRAP1 (23).

TRAP1 Associates Only with Type II Receptors of the TGF-␤ and Activin Pathways, Whereas Its Pattern of Binding to Type I Receptors Is Less
Restricted-To determine whether the interactions of TRAP1 were specific to the TGF-␤ receptors T␤RI and T␤RII, we next examined the ability of TRAP1 to bind to other members of the TGF-␤ superfamily of receptors. As shown in Fig. 2A, TRAP1 associates also with several kinasedeficient type I receptors, specifically with ALK1, BMPR-IA (ALK3), ActRI (ALK4), and the T␤RI (ALK5), but not with the ALK2 receptor and only minimally with BMPR-IB (ALK6). With the exception of the binding of TRAP1 to ALK3, these data suggest that TRAP1 is likely to be involved in both TGF-␤ and activin but less likely in BMP signaling.
Since the primary binding partner for TRAP1 is the type II receptor, we tested the association of TRAP1 with various other type II receptors. Interestingly, only T␤RII and the activin type IIB receptor associate with TRAP1, whereas the BMP type II receptor does not (Fig. 2B). To explore the binding of TRAP1 to the activin receptors more fully, we examined the pattern of binding of TRAP1 to both wild type and kinase-deficient forms of ActRIIB. In contrast to the pattern observed with T␤RII, where the binding of TRAP1 was clearly stronger to the kinasedeficient form of the receptor ( Fig. 1A and Fig. 3A), we observed equal binding of TRAP1 to either active or inactive forms of ActRIIB (Fig. 3A). Similar to that observed with the TGF-␤ receptor complex, activation of the complex by overexpression of the constitutively activated form of ALK4 led to nearly complete dissociation of TRAP1 from the receptor complex. In an attempt to understand the basis of the difference in the comparative binding of TRAP1 to the wild type and kinasedeficient forms of T␤RII and ActRIIB, we examined the effect of overexpression of each of these receptor constructs on the activity of the TGF-␤/activin responsive 3TP-Lux reporter (28) (Fig. 3B). The ability of wild type T␤RII but not ActRIIB to activate signaling when overexpressed in this cellular context mirrors the apparent differential binding of TRAP1 to the wild type-and kinase-deficient forms of these two receptors and suggests that partial dissociation of TRAP1 from wild type T␤RII underlies its apparently stronger binding to inactive T␤RII. Whether the selective ability of T␤RII to activate signaling in COS-1 cells results from an autocrine loop or from differential expression of ALK5 compared with ALK4 is not known.

Full-length TRAP1 Contains Multiple Receptor-binding Domains, and Single Domains Show Reduced Binding Specificity-To establish the domains of TRAP1 involved in its binding
to T␤RII, C-and N-terminal deletion constructs and a middle region construct (Fig. 4A) were used in immunoprecipitation assays. As shown in Fig. 4B, full-length TRAP1, ⌬MC-TRAP1 (amino acids 1-215), ⌬NM-TRAP1 (amino acids 651-860), ⌬NC-TRAP1 (amino acids 238 -536), and ⌬TRAP1 (amino acids 474 -860) co-immunoprecipitate with T␤RII. It is noteworthy that binding of each of the deletion constructs is severalfold stronger than that of the wild type molecule, suggesting a complex conformational regulation of these binding domains. Even under stringent washing conditions (500 mM NaCl buffer), strong binding of all regions to T␤RII(KD) was detectable (data not shown). Since both the C-terminal deleted construct ⌬MC-TRAP1 and the N-terminal deleted construct ⌬NM-TRAP1 as well as the middle region ⌬NC-TRAP1 all bind strongly to T␤RII, there are at least three binding sites for interaction of TRAP1 with the receptor. HA-T␤RI (K232R, kinase deficient, KD) or HA-T␤RI (T204D, mutationally activated, a*), and/or T␤RII (wt) or T␤RII (K277R, KD). 30 h after transfection, cells were serum starved for 12 h in the absence or presence of 10 ng/ml TGF-␤1 as indicated (B only). Cell lysates were immunoprecipitated with anti-HA antibody and blotted with anti-Flag antibodies. C and D, TRAP1 co-localizes with TGF-␤ receptors in the absence of signaling. COS-1 cells were transiently transfected with both KD receptors (C) or with T␤RII and activated T␤RI (a*) (D). After a 24 -36-h incubation period, the cells were fixed, subjected to immunochemistry with primary antibodies against the Flag and HA epitopes, and secondary antibodies linked to fluorescein isothiocyanate or rhodamine dyes, and mounted with medium containing 4,6-diamidino-2-phenylindole (DAPI). Magnification: ϫ63. Since the association data for the full-length TRAP1 described in the legend to Fig. 1 differ from those previously described for the C-terminal ⌬TRAP1, we also investigated the interaction pattern of this truncated protein with T␤RI and T␤RII. In our hands, ⌬TRAP1 neither showed any preferential binding to activated, wild type, or inactivated TGF-␤ receptors nor did it discriminate between T␤RI and T␤RII, as shown in Fig. 4C, in striking distinction from the pattern shown for the full-length molecule. Since overexpression of either T␤RII or the activated form of T␤RI are sufficient to induce signaling, these data demonstrate that ⌬TRAP1, unlike full-length TRAP1, does not dissociate from the active signaling receptor complex.
Full-length TRAP1 Slightly Stimulates TGF-␤ Signaling-Initial functional studies of ⌬TRAP1 by Charng et al. (23) showed an inhibitory effect on TGF-␤ induced activity of the 3TP-Lux reporter, suggesting a functional role of the molecule in TGF-␤ signaling. Based on the different interaction patterns of full-length TRAP1 and ⌬TRAP1 molecules with the TGF-␤ receptors, we investigated whether their functional roles might also differ, and specifically whether ⌬TRAP1 might function as a dominant negative mutant of the full-length molecule. HepG2 cells were transiently transfected with the indicated TRAP1 constructs and the SBE luciferase reporter, consisting of 4 tandem repeats of the Smad-binding element CAGA (29) or the 3TP-Lux reporter (data not shown). Stimulation with TGF-␤ resulted in activation of luciferase activity with either promoter ( Fig. 5 and data not shown). Addition of TRAP1 slightly enhanced luciferase activity, whereas ⌬TRAP1 consistently inhibited luciferase activity, as shown previously (23). Similar results were observed in NMuMg cells with the 3TP-Lux reporter (data not shown).
These results distinguish functionally the full-length TRAP1 molecule and its N-terminal deletion mutant ⌬TRAP1. We propose that the relatively minor effect of the full-length molecule might be due to its abundance within the cell, making it a non-limiting component in the signal transduction pathway. In this respect, TRAP1 behaves like SARA, which was shown to have little effect on reporter gene activity (3TP-Lux and ARE-Lux) as a full-length molecule, but which was inhibitory in its truncated and functionally inactive forms (14).
Activation of the TGF-␤ Receptor Complex Leads to Dissoci-  3-7), and co-precipitating TRAP1 was probed with anti-GFP. The expression levels of TRAP1 and the receptor constructs are shown as loading controls. Note that in contrast to the apparently stronger binding of TRAP1 to the kinase-deficient form of T␤ RII, its binding to ActRIIB is independent of this inactivating mutation. B, T␤RII but not ActRIIB activates the transcription of the TGF-␤/activin-responsive 3TP-Lux reporter. The ability of transfected wild type and kinase-deficient forms of T␤RII and ActRIIB to trigger the corresponding signaling pathways was assessed by activation of the 3TP-Lux reporter, under the experimental conditions described in panel A. Note that T␤RII significantly activates transcriptional responses with the same order of potency of activated ALK4. In contrast ActRIIB elicits transcriptional responses only modestly higher than that of negative control and kinase-deficient ActRIIB and T␤RII. Results are expressed as ratio of luciferase activity (relative light units) to ␤-galactosidase activity and are representative of three independent experiments. Error bars represent standard deviation; where no error bars are visible, the error is too small to be shown.

ation of TRAP1 from the Receptor and Association of TRAP1
with Smad4 -Based on the fact that ⌬TRAP1 has an inhibitory effect on TGF-␤ signaling and fails to dissociate from active receptor complexes in contrast to full-length TRAP1, we hypothesized that TRAP1 might play a fundamental role in signaling from the TGF-␤ and activin receptors, and possibly other receptors of the TGF-␤ superfamily. Smad4 was a potential target for TRAP1 action, since it is an obligatory signaling intermediate in Smad-mediated signaling from all TGF-␤ superfamily receptors. We therefore investigated whether TRAP1 might interact with Smad4 in vivo, and whether such an interaction might be signal-dependent. COS-1 cells were transiently transfected with TRAP1 and Smad4 and with or without activated T␤RI. As shown in Fig. 6A, TRAP1 binds only weakly to Smad4 in the absence of a TGF-␤ signal. However, upon activation of the signaling cascade, a strong interaction can be demonstrated between TRAP1 and Smad4. This interaction most likely occurs off the receptor, as the activated receptor interacts only weakly with TRAP1, as shown above. These data are also consistent with the fact that a direct association of Smad4 with the activated receptor complex has not thus far been demonstrated (30,31). To address the question whether TRAP1 might also interact with other Smad proteins, we transfected Flag-and Myc-tagged Smad 1-8 constructs and utilized a GFP-tagged TRAP1 construct as a potential binding partner (Fig. 6B). The data show that Smad4 is the only Smad binding partner of TRAP1.
In order to determine whether the interaction of TRAP1 with Smad4 is direct or dependent on bridging molecules, we performed in vitro pull-down assays. Since full-length TRAP1 could not be expressed in our bacterial strains (DH5␣ and BL21, data not shown), we utilized bacterially expressed GST-⌬TRAP1 bound to glutathione-agarose and various in vitro transcribed/translated Smad4 constructs. The Smad4 constructs were labeled with [ 35 S]methionine and incubated with GST-agarose and GST-⌬TRAP1-agarose beads, followed by SDS-PAGE and autoradiography. As shown in Fig. 6C, ⌬TRAP1 pulls down Smad4 full-length protein weakly and almost no Linker ϩ MH2 and MH2 constructs. However, the MH1 domain, a MH1 ϩ Linker construct, as well as a SAD ϩ MH2 construct bind strongly to ⌬TRAP1. This suggests that there are at least two binding sites in Smad4 that facilitate a direct TRAP1-Smad4 interaction, and that these binding sites are conformationally restricted and regulated in the full-length Smad4 in vivo (as shown above). Together, these results demonstrate a signal dependent, direct TRAP1-Smad4 interaction.

TRAP1-Smad4 Association Is Transient and Is Disrupted by a Receptor-activated Smad-
The association of TRAP1 with Smad4 in the presence of an activated receptor led us to investigate whether TRAP1, Smad4, and Smad2 are part of a common complex and whether the observed association of Trap1 with Smad4 results from the artificial state generated by overexpression of Smad4 in the absence of an equivalent amount of a receptor-activated Smad. Based on the receptor binding pattern of TRAP1 to TGF-␤ and activin receptors, each of which activate Smad2 and Smad3, we examined the association of Smad4 with TRAP1 in COS-1 cells in the presence of increasing concentrations of Smad2 or the Smad2(3SA) mutant, which cannot be phosphorylated at its C-terminal end and thus cannot bind Smad4 following receptor activation.
Importantly, we show in Fig. 7 that the interaction of TRAP1  5) to a level comparable to that of the TRAP1-Smad4 association seen in the absence of an activated receptor (lane 2). These data are consistent with the expectation that, in the presence of high levels of receptor-activated Smad2, the major binding partner of Smad4 is Smad2 rather than TRAP1 (lane 5). The data demonstrate that TRAP1, Smad4, and Smad2 do not complex together, but rather suggest that the binding of Smad4 to TRAP1 or the activated Smad2 is mutually exclusive. We interpret these results to suggest that a putative endogenous Smad4-TRAP1 interaction is likely only very transient in vivo and is disrupted as soon as Smad4 binds to receptor-activated Smad2/3. Lane 6 provides further evidence for this model by showing that disruption of the TRAP1-Smad4 interaction is not merely a consequence of the presence of Smad2, but requires that Smad2 be activated by phosphorylation at its C terminus. This is demonstrated by comparing the ability of Smad2 and the inactive Smad2(3SA) mutant, which cannot be terminally phosphorylated, to disrupt the TRAP1-Smad4 interaction. In contrast to wild type Smad2, Smad2(3SA) slightly increases the association of TRAP1 with Smad4 (lane 6), possibly by blocking the activation of endogenous Smad2. Together, these data suggest that receptor activation leads to association of Smad4 and TRAP1, and that this likely transient state occurs prior to the association of Smad4 with a receptor-activated Smad.
Mutated Forms of TRAP1 Inhibit TGF-␤ Signaling by Interfering with Formation of the Smad2-Smad4 Complex-Since ⌬TRAP1 failed to dissociate from activated receptors (Fig. 4C) and inhibited TGF-␤ signaling (Fig. 5) (23), we hypothesized that the dissociation of TRAP1 from active receptors and its association with Smad4 might facilitate the interaction of Smad4 with Smad2/3. To test this model, we determined whether ⌬TRAP1 and other N-and C-terminal deletion mutants of TRAP1 would decrease the association of Smad4 with Smad2. As demonstrated in Fig. 8 4). We have occasionally seen slight inhibition by overexpressed TRAP1 in reporter gene assays as well, and suggest that its strong expression in the experiment shown might result in sequestration of Smad4 by TRAP1. In comparable immunoprecipitation experiments where TRAP1 expression levels were lower, we did not see this effect of full-length TRAP1 (data not shown).

DISCUSSION
The regulation of cellular signaling cascades involves a multitude of processes on various levels, including transcriptional regulation of signaling components, receptor trafficking, and phosphorylation events. In the case of signaling from TGF-␤ superfamily receptors, phosphorylation on serine and threonine residues of both the receptors themselves and of Smad proteins activates signal transduction. Other emerging mechanisms of signal regulation are the chaperone, anchoring, and scaffolding proteins that regulate the folding, the subcellular distribution, and the recruitment of signaling molecules (27). Such mechanisms have been identified in a variety of signaling pathways such as the steroid/glucocorticoid pathway, mitogenactivated protein kinase signaling cascades, protein kinase A and C signaling pathway, and others (32). Very recently, such mechanisms have been shown to be important in the TGF-␤ signaling pathway as well, in that a FYVE domain protein SARA has been shown to regulate the subcellular localization of two of the R-Smad proteins, Smad2 and Smad3 (14). However, despite evidence for such anchoring proteins and for identification of certain amino acids within the Smad and receptor proteins that confer specificity for the coordinated action of the pleiotropic receptors of the TGF-␤ superfamily (33), it cannot be ruled out that other, as yet unidentified mechanisms might not also contribute to the fine tuning of response in this cascade. It is therefore reasonable to assume that other anchoring and scaffolding proteins will be found which regulate and promote assembly of signaling intermediates in the TGF-␤ pathway. We now propose that, given its signal-dependent association with both T␤RII and ActRIIB and its interaction with Smad4, the previously identified TRAP1 protein plays such a chaperone role in signaling downstream of not only TGF-␤, but likely also activin. Our working hypothesis is that TRAP1 facilitates interaction of Smad4 with Smad2/3 proteins by binding to Smad4 in the vicinity of the activated receptor and mediating its transfer to the phosphorylated Smad2/3 (Fig. 9). In support of this model, we have shown that TRAP1 associates most strongly with receptors that are not actively signaling, as is the case either in the absence of ligand or, experimentally, by use of kinase-deficient receptors. Consistent with our data, we propose that activation of the receptor might then lead to a conformational change in TRAP1 such that it dissociates from the receptor and forms a transient complex with Smad4. Most importantly, the strong signal-dependent association of TRAP1 and Smad4 is seen only in the absence of co-expressed Smad2, suggesting that it might be a transient complex which can be visualized only in the experimental condition of overexpression of Smad4 in the absence of an acceptor R-Smad. TRAP1 and the activated, phosphorylated Smad2 bind Smad4 in a mutually exclusive fashion. The C-terminal phosphorylation deficient 3SA mutant of Smad2 does not interfere with Smad4-TRAP1 association, presumably because it does not leave the receptor and cannot function as an acceptor for TRAP1-activated Smad4. For the R-Smads, phosphorylation of the C-terminal SSXS motif has been proposed to relieve the autoinhibitory interaction of the MH1 and MH2 domains, freeing the MH2 domain to interact with the MH2 domains of other R-Smads or of Smad4 (30,34). At present, there is no known phosphorylationdependent mechanism for activation of Smad4 similar to that of the R-Smads. While the simpler explanation of our data is that Smad4 can bind either receptor-activated TRAP1 or Smad2/3, we propose instead that the interaction of receptoractivated TRAP1 with Smad4 might function to reduce the autoinhibitory MH1/MH2 domain interaction of Smad4 and thereby make it competent to interact with activated R-Smads. Whether there is a complementary TRAP1-related mechanism operative downstream of BMP receptors remains to be demonstrated.
The only other molecule described so far that serves as a chaperone in TGF-␤ signaling is SARA (14). Although there are no sequence homologies between TRAP1 and SARA, there are some strikingly similar characteristics. Both proteins associate with the TGF-␤ receptor complex, both are regulated in a ligand-dependent fashion, and both display mutually exclusive binding to their Smad protein partner with respect to the acceptor Smad. While TRAP1 binds Smad4 only in the absence of bound Smad2, SARA binds Smad2 only when it is not associated with Smad4. However, unlike domain binding data for SARA, we are unable to delineate any specific domain within TRAP1 necessary for the binding of the receptor and Smad4, but rather show that N-and C-terminal and middle regions all have binding activity. However, this is not without precedent in this field. Another chaperone/scaffolding protein, RAP, involved in folding and escorting certain low density lipoprotein receptor family proteins, has also been shown to bind multiple sites in the target receptor through multiple sites in the chaperone (35). Again with the TRIP-1 protein, shown previously to bind strongly to the type II TGF-␤ receptor, deletion studies showed that the receptor-binding domain could not be localized and that multiple regions of the molecule participated in the interaction (22). Moreover, the binding of multiple Smad4 domains to TRAP1 and vice versa supports our hypothesis that TRAP1 might serve as a scaffold that separates the Smad4-MH1 from the Smad4-MH2 domain in order to present Smad4-MH2 to the MH2 domain of Smad2. However, in the absence of three-dimensional structural information about TRAP1, we cannot, for the present, address the question of its binding properties in detail.
We observed both mild stimulatory and mild inhibitory effects of exogenous full-length TRAP1 on TGF-␤ signaling. Since TRAP1 is ubiquitously expressed (23), we propose that these variable effects might be dependent on the relative levels of endogenous TRAP1 expressed by a particular cell. In cells expressing low levels of TRAP1, exogenous TRAP1 might enhance TGF-␤ signaling, whereas in cells with higher TRAP1 levels, additional exogenous TRAP1 might sequester Smad4 from endogenous R-Smads and inhibit signaling. This behavior is similar to published data on the SARA protein and ARIP1, a molecule proposed to act as a scaffold in the activin pathway (36). Both proteins are thought to facilitate signaling, yet fulllength SARA stimulated only slightly or had no effect in functional assays (14), and full-length ARIP1 inhibited signaling (36). In this context, it is noteworthy that mutated forms of SARA or TRAP1 each inhibit signal transduction, suggesting that these molecules suppress the activity of the endogenous protein in a dominant-negative fashion. In the case of TRAP1, we have not yet determined whether this inhibitory activity might be due to the inability of these constructs to dissociate from the activated receptor complex, as shown for ⌬TRAP1, their inability to associate with Smad4, or possibly also their inability to release and/or activate Smad4 to associate with Smad2. Our data showing that expression of mutant forms of TRAP1 diminishes Smad2-Smad4 interaction do not distinguish between these possibilities. TRAP1 is a large protein (860 amino acids), and our data suggest that multiple subdomains contribute to its functional activity. Interestingly, we often observed two distinct TRAP1 bands. As the double band is observed only when utilizing a 3Ј-tagged full-length TRAP1 construct (such as Flag-and Myctagged TRAP1, e.g. Figs. 1, A and B, 2A, and 6A) but not a 5Ј-tagged construct (such as EGFP-TRAP1, Fig. 6B), the smaller molecule likely represents an N-terminally truncated TRAP1. Whether this is due to an alternative translation start site or proteolytic cleavage has yet to be determined.
During the course of this study, a report was published describing the interaction of TRAP1 with 5-lipoxgenase in a yeast two-hybrid system (37). However, no functional role was attributed to this binding activity, nor was the finding confirmed in mammalian cells. We have no data linking the effect of TRAP1 in TGF-␤ signaling with the function of 5-lipoxgenase; however, at this point, it should not be ruled out that TRAP1 could serve other, as yet unknown, functions in the TRAP1 Is a Smad4 Chaperone cells. In this regard, the human ortholog of the yeast vacuolar sorting protein, Vps39/Vam6p (38), previously published as a 3Ј cDNA named KIAA0770 (39) shows a 25% identity and 40% similarity to TRAP1. This new protein has now been shown to localize to the cytoplasmic face of lysosomes, suggesting that it, and by inference possibly also TRAP-1, may play a role in lysosome biogenesis. 3 In this regard, another vesicular trafficking protein, caveolin-1, has been shown to play a role in TGF-␤ signaling mediated by its direct interaction with T␤RI (40).
A recent data base search revealed that TRAP1 is localized on chromosome 2 and encoded by 11 exons (the terminal 34 base pairs were not part of the published contig). Given the functional importance of TRAP1 in TGF-␤ signaling, knowledge of its chromosomal localization will now enable investigation of whether this locus can be linked to any diseases in which TGF-␤ superfamily members are known to play a role. Studies addressing this question are under way in this laboratory.