Transforming Growth Factor-β Induces Formation of a Dithiothreitol-resistant Type I/Type II Receptor Complex in Live Cells*

Transforming growth factor-β (TGF-β) binds to and signals via two serine-threonine kinase receptors, the type I (TβRI) and type II (TβRII) receptors. We have used different and complementary techniques to study the physical nature and ligand dependence of the complex formed by TβRI and TβRII. Velocity centrifugation of endogenous receptors suggests that ligand-bound TβRI and TβRII form a heteromeric complex that is most likely a heterotetramer. Antibody-mediated immunofluorescence co-patching of epitope-tagged receptors provides the first evidence in live cells that TβRI·TβRII complex formation occurs at a low but measurable degree in the absence of ligand, increasing significantly after TGF-β binding. In addition, we demonstrate that pretreatment of cells with dithiothreitol, which inhibits the binding of TGF-β to TβRI, does not prevent formation of the TβRI·TβRII complex, but increases its sensitivity to detergent and prevents TGF-β-activated TβRI from phosphorylating Smad3 in vitro. This indicates that either a specific conformation of the TβRI·TβRII complex, disrupted by dithiothreitol, or direct binding of TGF-β to TβRI is required for signaling.

activins and bone morphogenetic proteins from mammals. The TGF-␤s are important modulators of development, the extracellular matrix, and the immune response; they are potent growth inhibitors in many cell types, and their receptors and some downstream signaling elements are tumor suppressors (1)(2)(3)(4)(5).
TGF-␤ signals through the sequential activation of two serine-threonine kinase cell surface receptors (6 -10), termed type I and type II (T␤RI and T␤RII). These two receptors physically associate to form a stable complex (8,11). Several chimeric receptor systems have established that this complex is required for signaling (12)(13)(14)(15)(16). Whether it is preformed or ligand-induced is controversial. Isolation of a T␤RI⅐T␤RII complex from detergent lysates was possible only after pretreatment with TGF-␤ (9). Ligand-independent interactions between receptor cytoplasmic domains, however, have been detected in transfected COS cells and in the yeast two-hybrid system, 2 and work with TGF-␤2 (which requires both receptors to bind) suggests that at least a small percentage of the cell surface receptor population is in preformed complexes (9,(17)(18)(19). Data from experiments on the effect of DTT also raise questions about the role of ligand in the complex. Treatment of cells with DTT prevents TGF-␤ binding to T␤RI but not to T␤RII (8,20) and, in vitro, prevents formation of the T␤RI⅐T␤RII complex. 3 The stoichiometry of the signaling complex is not known. T␤RI and T␤RII form ligand-independent homodimers in the endoplasmic reticulum and on the cell surface (21,22). The simplest model is that two homodimers form a heterotetrameric signaling complex induced or activated by TGF-␤. This view is supported by studies based on functionally complementary type I receptor mutants and on chimeric TGF-␤/erythropoietin receptors demonstrating that type I dimers are required for signaling (13,23). Yamashita et al. (24) used nonreducing/reducing two-dimensional SDS-PAGE to isolate ligand-bound T␤RI and T␤RII homo-and heterodimers. They speculate that these are derived from heterotetramers, although their data are also consistent with smaller complexes.
We report here studies on the physical nature of the T␤RI⅐T␤RII complex. We demonstrate that the complex is most likely a stable heterotetramer. Our studies show conclusively that some T␤RI⅐T␤RII complexes exist at the surface of live cells in the absence of ligand, and that TGF-␤ significantly enhances heterocomplex formation. DTT treatment, which prevents TGF-␤ binding to T␤RI, does not inhibit complex forma-* This work was supported by National Institutes of Health Grants CA63260 (to H. F. L.) and DK02290 (to R. G. W.) and by grants from the Israel Science Foundation administered by the Israel Academy of Arts and Sciences and from the Israel Cancer Research Fund (to Y. I. H.). 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. c These authors contributed equally to this work.  The abbreviations used are: TGF-␤, transforming growth factor-␤; T␤RI,  T␤RII, T␤RIII, types I, II, and III TGF-␤ receptors; octyl-POE, n-octylpolyoxyethylene; DTT, dithiothreitol; PAGE, polyacrylamide gel electrophoresis; PBS, phosphate-buffered saline; MES, 4-morpholineethanesulfonic acid; GST, glutathione S-transferase; HA, hemagglutinin. tion but does result in the failure of T␤RI from DTT-treated cells to phosphorylate Smad3 in vitro. We suggest that ligand binding to the two receptors may have different functions, and that complex formation itself is not sufficient for signal initiation.
Velocity Centrifugation on Sucrose Gradients-The velocity centrifugation technique has been described elsewhere (22). Briefly, cleared lysates (150 l in MNT lysis buffer) were mixed with 50 l size markers (29 -669 kDa; Sigma) and layered over 7.5-30% sucrose gradients in MNT/1% octyl-POE. Centrifugation was for 8 h, 60,000 rpm, in an SW60 rotor (Beckman) at 4°C. 250 l fractions were removed sequentially from the top of each gradient. After removal of 25 l for SDS-PAGE and Coomassie staining (to analyze migration of markers for each individual gradient), fractions were immunoprecipitated with ␣-IIC or VPN. Samples were analyzed by 7.5% SDS-PAGE. Autoradiographs were quantified with a LaCie Silverscanner II and MacBAS (Fuji) software.

RESULTS
T␤RI and T␤RII Form Stable, Heterotrimeric, or Heterotetrameric Complexes-We studied the size of the cell surface ligand-bound and cross-linked T␤RI⅐T␤RII complex using sucrose gradient velocity centrifugation, a technique we used previously to show that both receptors form homodimer-sized complexes in the endoplasmic reticulum (22). Cells were lysed with octyl-POE, a nonionic detergent with a density of 1 and a high CMC that enables analysis of the migration of detergentsolubilized proteins rather than of micelles (31). 125 I-TGF-␤1bound and cross-linked T␤RII from COS7 cells transfected with T␤RII alone migrated with a peak centered at fraction 7, the position of a 150-kDa marker protein (Fig. 1A, top and bottom panels). Although absolute size determinations can be inaccurate for detergent-solubilized membrane proteins, this is consistent with a homodimer bound to one or more molecules of TGF-␤1, as expected from previous gradient analysis and immunofluorescence (21,22). T␤RII from COS7 cells cotransfected with both T␤RI and T␤RII (Fig. 1A, middle and bottom panels) and immunoprecipitated with an antibody against T␤RI, thus in a T␤RI⅐T␤RII complex, migrated more quickly with a peak centered around fraction 8, indicative of a heterotrimeric or tetrameric complex.
L6 rat myoblasts, TGF-␤-responsive cells which lack the type III TGF-␤ receptor, show similar receptor complexes. TGF-␤1bound type I and II receptors immunoprecipitated with an antibody against T␤RII both migrated with a peak centered at fraction 10, correlating with markers between 230 and 265 kDa (Fig. 1B, top and bottom panels), and consistent with a heterotetrameric or trimeric complex. Nearly identical results were obtained with TGF-␤2 and with other cell lines, including Mv1Lu, HYB2, THP1, and GH3 (data not shown). Immunoprecipitation with an antibody against T␤RI (data not shown) gave similar results, indicating the absence of a significant population of ligand-bound T␤RII, which is not in a complex with T␤RI. Pretreatment of cells with DTT (Fig. 1B, middle and bottom panels) prevented ligand binding and cross-linking to T␤RI and slowed the migration of T␤RII to a peak centered around 155 kDa. The DTT-treated complex is likely a dimer (for comparison, see singly transfected COS cells in Fig. 1A), indicating disruption of the T␤RI⅐T␤RII complex under these conditions. DTT does not appear to affect T␤RII homodimers, a finding consistent with previous results (21). The DTT effect appears to be on the cross-linking of ligand to T␤RI rather than on cross-linking between T␤RI and T␤RII, because immunoprecipitation of T␤RI from DTT-treated cells following ligand binding and cross-linking showed the absence of ligand-labeled T␤RI (Fig. 3). Analogously, ligand binding and cross-linking after DTT treatment either to endogenous TGF-␤ receptors (20) or to COS cells transiently expressing T␤RI and T␤RII (not shown) failed to reveal T␤RI labeling.
T␤RI and T␤RII Form a Ligand-dependent Complex in Live Cells-To study the T␤RI⅐T␤RII complex in live cells, we used T␤RI and T␤RII carrying HA or Myc epitope tags at their extracellular termini for immunofluorescence co-patching, a technique that we developed and have described previously (21,22). Briefly, a tagged receptor at the surface of live, unfixed cells (in the cold to avoid internalization) is forced into patches by a double layer of bivalent IgGs where the secondary antibody is coupled to one fluorophore (e.g. fluorescein, which emits green fluorescence). A second receptor, containing a different tag, is labeled by antibodies coupled to a second fluorophore (e.g. Cy3, red fluorescence). The cells are examined by fluorescence microscopy to determine whether the two receptors are swept into mutual (yellow) or separate (red or green) micropatches. We have employed this method successfully to demonstrate that all three TGF-␤ receptors form ligand-independent homodimers (21,22). Fig. 2 shows the results of co-patching experiments performed on COS7 cells co-transfected with T␤RI-Myc and T␤RII-HA. The labeling specificity is high, as shown in a control experiment on cells transfected with T␤RI-Myc alone ( Fig.  2A, inset); these cells show only Cy3 labeling. In the absence of ligand ( Fig. 2A), 15-20% of the patches were mutual (yellow), although the majority were separate (either green or red). In the presence of 250 pM TGF-␤1, the percentage of mutual patches markedly increased to 40 -50% (Fig. 2B), demonstrating that T␤RI and T␤RII at the surface of live cells have an inherently low probability of forming heterocomplexes that is significantly enhanced by ligand binding. The fraction of a given receptor type in heterocomplexes is proportional to the FIG. 1. Type I and II receptors form stable heterotetrameric or trimeric complexes in COS7 and L6 cells. Cells were cross-linked to 125 I-TGF-␤1, lysed, and subjected to velocity centrifugation over sucrose gradients. A, COS7 cells were transfected with T␤RII alone (top panel) or together with T␤RI (middle panel). Fractions from the sucrose gradients were immunoprecipitated with anti-T␤RII (top) or anti-T␤RI (middle) and analyzed by SDS-PAGE. Migration of the type II receptor on each gradient was quantified and normalized to the peak value and is shown in the bottom panel. Size estimates (in kDa) are from size markers included in each gradient tube. Receptors from doubly transfected cells (middle) were immunoprecipitated with antibody against T␤RI, ensuring that all T␤RII measured is part of the complex and not from singly transfected cells. Quantification of the immunoprecipitated ligand-labeled type I receptor results in a nearly identical curve (data not shown). The broad nature of the peak for cells transfected with T␤RII alone may reflect some association of T␤RII with the small number of endogenous T␤RI found in COS7 cells. B, L6 cells were treated as in A, except that cells in the middle panel were treated with 2 mM DTT before ligand binding. Fractions were immunoprecipitated with anti-T␤RII. Migration of T␤RII (middle panel) or both receptors (top) on the gradients was quantified (bottom panel). Note that fraction numbers are comparable only within a given experiment, and that migration of markers for part A and part B was slightly different. Control experiments demonstrate that only minimal dissociation of receptor-bound but noncross-linked ligand occurred in detergent lysates over the 8 h required for centrifugation (data not shown).
number of yellow patches divided by the sum of yellow and red (for the red-labeled receptor type) or yellow and green (for green-labeled receptors). These fractions are similar for T␤RI and T␤RII (23-29% and 57-63% in the absence and presence of ligand, respectively), in accord with a 1:1 stoichiometric ratio in the heterocomplex.
Ligand binding and cross-linking to T␤RI co-expressed with T␤RII is abrogated by DTT pretreatment of cells (Ref. 20 and Fig. 1B, middle panel). DTT treatment, however, did not dissociate the ligand-independent T␤RI⅐T␤RII complexes (around 20% co-patching), and did not affect the enhancement of heterocomplex formation by TGF-␤1 (around 50% co-patching) (Fig. 2, C and D). The fraction of each receptor type in yellow patches also remained similar.
DTT Increases the Detergent Sensitivity of the Complex-In contrast to the persistence of T␤RI⅐T␤RII heterocomplexes in DTT-treated live cells (Fig. 2, C and D), velocity sedimentation experiments showed that receptor heterocomplexes were disrupted by DTT (Fig. 1B). We suspected that this disparity resulted from destabilizing effects of the detergent used to solubilize the receptors for velocity sedimentation. We therefore examined the effect of various detergents on T␤RI⅐T␤RII co-immunoprecipitation from DTT-pretreated Mv1Lu cells (Fig. 3). Immunoprecipitation with an antibody against T␤RI (Fig. 3, right panel) was used to assess the integrity of the receptor complex in DTT-treated cells, because T␤RII residing in heterocomplexes would still be labeled by ligand and would co-precipitate with T␤RI even if the latter was unlabeled. Under the most stringent lysis conditions used (buffers 1-3), there was no co-immunoprecipitation, indicating the absence of in-tact heteromeric complexes after detergent solubilization. These results hold in different cell lines and with antibodies raised against different type I receptor epitopes (not shown). The same was also true with buffer 4, which was used for the velocity centrifugation experiment in Fig. 1B. With a fifth lysis buffer, however, we detected a small amount of intact T␤RI⅐T␤RII complex in the presence of DTT (Fig. 3, right panel, buffer 5), suggesting that buffer and detergent conditions determine the integrity of the complex in DTT-treated cells. Together with the demonstration of T␤RI⅐T␤RII heterocomplex formation in live cells pretreated with DTT (Fig. 2), these results suggest that DTT treatment alters the conformation of the heterocomplex formed; the altered complex is less stable (as reflected in its increased detergent sensitivity) and most likely has a different conformation, resulting in the failure of ligand binding and cross-linking to T␤RI.

DTT Pretreatment Prevents TGF-␤-induced Activation of the Type I Receptor-In epithelial cells, an essential step in TGF-
␤-mediated growth inhibition and PAI-1 promoter activation is activation of the ability of T␤RI to rapidly phosphorylate Smad3 at its C-terminal SSVS motif (28). Because DTT treatment did not block the ligand-mediated association of T␤RI with T␤RII, we examined the signaling capability of these complexes as reflected by their ability to phosphorylate Smad3 in vitro. Mv1Lu cells (untreated or pretreated with DTT) were exposed to TGF-␤1 to allow heterocomplex formation and activation of T␤RI. In one study, lysates from these cells were incubated with recombinant GST-Smad3, and the isolated complex was subjected to an in vitro kinase reaction. In untreated cells (Fig. 4A, lanes 1 and 2), TGF-␤ mediated a greater than 20-fold increase in Smad3 phosphorylation; two-dimensional tryptic mapping indicated that this in vitro phophorylation occurred at the same site as in vivo (data not shown). DTTpretreated cells, however, showed no ligand-induced phosphorylation (Fig. 4A, lanes 3 and 4). Similar results (Fig. 4B) were obtained by first immunoprecipitating lysates with antibodies against T␤RI, then incubating the complexes with a GST-Smad3 fusion protein in an in vitro kinase reaction. Although there is background phosphorylation of the GST-Smad3 construct in this assay, peptide competition during the immunoprecipitation (Fig. 4B, lanes 5-8) eliminates all of the TGF-␤inducible phosphorylation, indicating that T␤RI is responsible. Thus, the T␤RI⅐T␤RII complex formed after DTT treatment is inactive and cannot mediate the earliest step in TGF-␤ downstream signaling phosphorylation of Smad3.

DISCUSSION
Our major findings are: 1) T␤RI and T␤RII form a stable, heteromeric complex, most likely a heterotetramer; 2) in live cells, the two receptors have an intrinsic affinity for each other that is markedly increased by TGF-␤ exposure; and 3) DTT does not prevent the formation of this complex but increases its detergent sensitivity and blocks TGF-␤-induced activation of the type I receptor, as measured by its ability to phosphorylate Smad3.
We previously used velocity centrifugation to demonstrate that T␤RI and T␤RII each form homodimer-sized complexes in the endoplasmic reticulum (22). We use similar technology here to show that ligand-bound and cross-linked receptors in transfected COS7 cells and in L6 cells expressing native receptors form detergent-stable complexes whose size is consistent with heterotrimers or heterotetramers. In L6 cells (Fig. 1B), the migration of the T␤RI⅐T␤RII complex is consistent with a heterotetramer or heterotrimer bound to one TGF-␤ molecule. In COS7 cells, the complex migrates significantly faster than a homodimeric complex (Fig. 1A, top panel) (22); the wide nature of the peak, and the shoulder at higher fractions, may be because of dissociation of larger complexes during centrifugation. Furthermore, co-patching results obtained in live cells (Fig. 2) are most consistent with a heterotetramer, the fractions of T␤RI and T␤RII in mutual patches are similar, as expected for a stoichiometric ratio of 1:1. It should be stressed, however, that a definite determination of the stoichiometric ratio depends on an accurate measurement of the surface levels of both receptors on the cells scored, which is not feasible by the current methods.
The tetrameric nature of the complex is also supported by data indicating that a dimer of T␤RII and more than one T␤RI reside in the heterocomplex. Evidence that there are two T␤RII in the complex comes from the demonstration by velocity centrifugation that treatment of cells with DTT results in a homodimer-sized complex of T␤RII. This suggests that there were two type II receptors in the original, heteromeric complex; otherwise, one must assume that the normally dimeric T␤RII (21,22) is monomeric in the heterocomplex and reassociates to form dimers after DTT treatment and detergent solubilization. Type I receptors, which are homodimers when expressed alone and remain dimeric after DTT treatment, (22) are likely to be multimeric in the active complex, as indicated by experiments demonstrating the existence of functionally complementary T␤RI mutants (23). Taken together, the sedimentation velocity, co-patching, and functional complementation studies imply that the signaling TGF-␤ receptor complex contains two T␤RI and two T␤RII polypeptides. We provide the first evidence in live cells that the formation of the T␤RI⅐T␤RII complex is TGF-␤-dependent (Fig. 2, A and  B). Previous studies have demonstrated the TGF-␤ dependence of the T␤RI⅐T␤RII heterocomplex but used receptors in detergent lysates, leaving open the possibility that TGF-␤ stabilized the complex in detergent but did not induce its formation (8,9,11). Co-patching studies examining heterocomplex formation in the intact plasma membrane show a marked increase in mutual aggregates in the presence of TGF-␤1. There is, however, an intrinsic affinity between the two receptors, as demonstrated by the 15-20% that reside in mutual aggregates in the absence of TGF-␤ ( Fig. 2A). Although these experiments used cells overexpressing the transfected receptors, as required for visualization by immunofluorescence, there are several indications that the heterocomplexes observed are not the result of high expression levels. The co-patching experiments examine single cells under the microscope, and the use of double labeling by IgG or biotin/streptavidin enhancement enabled us to analyze cells expressing as few as 15,000 surface receptors (evaluated by photomultiplier-based measurement of the fluorescence intensity on cells expressing known receptor levels, as described by Henis et al. (21)). These cells, with receptor levels higher but of the same order of magnitude as untransfected cells, yielded the same co-patching results as cells expressing receptor levels 10-fold higher. In addition, in all cases where the level and percentage of higher complexes formed were high enough to allow detection by ultracentrifugation in untransfected cells expressing the receptors (including ligand-labeled T␤RI⅐T␤RII heterocomplexes in the current work, and T␤RI or T␤RII homodimers in earlier studies) (22), a good agreement was obtained between the results on these cells and on transiently expressing COS cells. Furthermore, the results obtained in co-patching experiments depended on the receptor types examined, with low co-patching for type II/type III TGF-␤ receptor heterodimers versus high co-patching levels for T␤RI and T␤RII homodimers (21,22). This emphasizes the specificity of the interactions measured. We note, however, that COS cells overexpressing T␤RI and T␤RII exhibit some ligand-independent receptor phosphorylation (18), raising the possibility that the number of T␤RI⅐T␤RII complexes is larger in this system than in cells expressing the receptors endogenously.
The ligand-independent association of a fraction of T␤RI and T␤RII is consistent with our previous finding that TGF-␤2, unlike TGF-␤1, binds to a preformed complex of T␤RI and T␤RII (19). Such preformed complexes do not appear to mediate TGF-␤-independent signal transduction, 4 raising the question whether the role of ligand is to increase the number of complexes or to effect a necessary and stabilizing conformational change. A recent model of the T␤RI ectodomain (based on certain cysteine motifs shared with protectin (CD59)) proposes that its surface has two distinct binding sites, one each for ligand and T␤RII (32). This model is in agreement with the finding shown in Fig. 2 that there is an intrinsic affinity between T␤RI and T␤RII, which is increased by TGF-␤.
To further investigate the relationship between TGF-␤-mediated heterocomplex formation and signaling, we studied the effects of DTT treatment on complex stability and its ability to phosphorylate Smad3 in vitro. Pretreatment with DTT prevents TGF-␤ binding and cross-linking to the type I receptor (8,20). We demonstrate that it also increases the detergent susceptibility of the complex (Figs. 1B and 3), and blocks TGF-␤induced activation of the ability of T␤RI to phosphorylate Smad3 in vitro (Fig. 4) but does not change the percentage of T␤RI and T␤RII in mutual complexes with or without TGF-␤1 (Fig. 2). The observation that multiple cross-linkers, including difluorodinitrobenzene (with a spacer length of 3Å) can crosslink TGF-␤ to T␤RI suggests that cross-linking results reflect direct binding of ligand to T␤RI rather than their fortuitous proximity within a complex. 4 Our DTT data therefore suggest that either a specific T␤RI⅐T␤RII complex conformation, destroyed by DTT, or direct ligand binding to T␤RI are required for signal transduction. It is unclear whether ligand binding to T␤RI stabilizes the complex with T␤RII or serves a different function altogether.
Our data agree only in part with a previously published report that DTT pretreatment does not affect type I receptor phosphorylation and does not measurably alter the amount of ligand-bound and cross-linked T␤RI⅐T␤RII complex in detergent lysates (33). Although we found that TGF-␤1 can mediate heterocomplex formation after DTT treatment, especially in intact cells (Fig. 2), we have been able to demonstrate at most a small percentage of T␤RI⅐T␤RII complex using the same detergent lysis conditions as these investigators (see Fig. 3, lysis buffer 5), in accord with a second report (6). In some of the experiments by Vivien and Wrana (33), the antibody used was against a C-terminal epitope of T␤RI that has significant sequence identity to other type I receptors (34); however, antibody cross-reactivity could not explain all the differences between their results and ours.
These results raise interesting questions about the stoichiometry of TGF-␤ receptors in the signaling complex. Although our data are most consistent with a heterotetrameric (T␤RI) 2 (T␤RII) 2 receptor complex, and we have obtained similar results with several cell lines, there are suggestions in the literature that the stoichiometry of the complex may vary. For example, the ratio of T␤RI to T␤RII in microvascular endothelial cells was significantly higher in three-dimensional versus two-dimensional cultures, corresponding to increasing resistance to the anti-proliferative but not the matrix-inducing effects of TGF-␤ (35). In other cell types, including bone, the ratio of type I to type II receptors may also be important, particularly in differentiating growth inhibition from other effects of TGF-␤ (36 -38). The migration over sucrose gradients of the T␤RI⅐T␤RII complex from nonepithelial cell systems remains to be determined.
The  3 and 4) or without (lanes 1 and 2) 2 mM DTT (5 min, 37°C). After extensive washing, they were incubated with (lanes 2 and 4) or without (lanes 1 and 3) 100 pM TGF-␤1 (10 min, 37°C). Cells were rapidly chilled and lysed. Cleared lysates were incubated with a GST-Smad3 fusion protein for 3 h. After retrieval of the fusion protein by glutathione beads, it was subjected to an in vitro kinase reaction then analyzed by SDS-PAGE. Control cells were allowed to bind 125 I-TGF-␤1 under the same conditions and were then cross-linked, lysed, and analyzed by SDS-PAGE; there was no binding of ligand to T␤RI in DTT-pretreated cells, confirming that the 10-min incubation period at 37°C was not sufficient for arrival of significant amounts of non-DTT-exposed T␤RI at the cell surface (data not shown). B, Mv1Lu-Flag-N-Smad3 cells were treated as above, except lysates were immunoprecipitated with an anti-type I antibody (6) with (lanes 5-8) or without (lanes 1-4) 3 g of competing peptide before incubation with the GST-Smad3 fusion protein.
subunit bound to a type I and a type II receptor, or two ligand homodimers, each subunit bound to one of the four receptors in a presumed heterotetramer. The sucrose gradient data presented here show a surprisingly small increase in complex size when T␤RI is added to TGF-␤-bound T␤RII, suggesting that the addition of T␤RI to T␤RII is not accompanied by the recruitment of additional ligand molecules.