Research ArticleStructural Biology

Hydrophobic patches on SMAD2 and SMAD3 determine selective binding to cofactors

See allHide authors and affiliations

Sci. Signal.  27 Mar 2018:
Vol. 11, Issue 523, eaao7227
DOI: 10.1126/scisignal.aao7227

The SMAD cofactor code

SMADs are transcription factors that execute the transcriptional outputs of transforming growth factor–β (TGF-β) signaling. SMAD2 and SMAD3 form complexes with SMAD4 that translocate into the nucleus. Whether these complexes stimulate or repress the expression of target genes depends on cofactors that interact with the SMADs; therefore, the underlying mechanism is critical for the appropriate response to TGF-β signaling. Miyazono et al. determined the crystal structures of SMAD2 in complex with the transcriptional corepressor SKI and SMAD3 in complex with the transcription factor FOXH1. The authors identified multiple hydrophobic patches on the surfaces of SMAD2 and SMAD3 that mediated the interactions with these cofactors. The cofactors interacted with the SMADs through different subsets of these patches, suggesting that cooperation and competition between cofactors determine which cofactors bind to the SMADs in vivo.

Abstract

The transforming growth factor–β (TGF-β) superfamily of cytokines regulates various biological processes, including cell proliferation, immune responses, autophagy, and senescence. Dysregulation of TGF-β signaling causes various diseases, such as cancer and fibrosis. SMAD2 and SMAD3 are core transcription factors involved in TGF-β signaling, and they form heterotrimeric complexes with SMAD4 (SMAD2-SMAD2-SMAD4, SMAD3-SMAD3-SMAD4, and SMAD2-SMAD3-SMAD4) in response to TGF-β signaling. These heterotrimeric complexes interact with cofactors to control the expression of TGF-β–dependent genes. SMAD2 and SMAD3 may promote or repress target genes depending on whether they form complexes with other transcription factors, coactivators, or corepressors; therefore, the selection of specific cofactors is critical for the appropriate activity of these transcription factors. To reveal the structural basis by which SMAD2 and SMAD3 select cofactors, we determined the crystal structures of SMAD3 in complex with the transcription factor FOXH1 and SMAD2 in complex with the transcriptional corepressor SKI. The structures of the complexes show that the MAD homology 2 (MH2) domains of SMAD2 and SMAD3 have multiple hydrophobic patches on their surfaces. The cofactors tether to various subsets of these patches to interact with SMAD2 and SMAD3 in a cooperative or competitive manner to control the output of TGF-β signaling.

INTRODUCTION

Members of the transforming growth factor–β (TGF-β) superfamily are multifunctional cytokines that control various processes in vertebrate cells (1). During embryogenesis, TGF-β superfamily signaling controls self-renewal and differentiation of stem cells to determine cell fate (2, 3). In adult tissues, TGF-β signaling regulates biological processes such as cell proliferation, apoptosis, immune responses, autophagy, cell migration, angiogenesis, extracellular matrix production, and senescence (410). The dysregulation of TGF-β signaling is associated with various diseases and medical conditions, such as cancer, fibrosis, inflammation, hereditary hemorrhagic telangiectasia, Marfan syndrome, and Shprintzen-Goldberg syndrome (5, 1116). TGF-β signals are transduced into the cell through two types of serine-threonine kinase receptors: TGF-β type I and type II receptors (TβRI and TβRII, respectively). Once TGF-β proteins are recognized by TβRI and TβRII, TβRI phosphorylates the C-terminal SXS (Ser-X-Ser) motif of receptor-regulated SMAD (R-SMAD) proteins (R-SMADs: SMAD1, SMAD2, SMAD3, SMAD5, and SMAD8). The R-SMAD proteins are transcription factors that consist of an N-terminal MAD homology 1 (MH1) domain, which binds to DNA, and a C-terminal MH2 domain, which mediates protein-protein interactions (fig. S1, A and B). Two phosphorylated R-SMAD proteins use their MH2 domains to form an active heterotrimeric complex with the common-mediator SMAD (Co-SMAD) SMAD4 (fig. S1C), and the heterotrimeric complex translocates into the nucleus to control the transcription of target genes (17). For example, SMAD2 and SMAD3 that are activated by TGF-β signaling form SMAD2-SMAD2-SMAD4, SMAD3-SMAD3-SMAD4, and SMAD2-SMAD3-SMAD4 complexes, and these complexes regulate TGF-β–dependent gene expression (18). The activated receptors also associate with other signaling components to regulate various TGF-β–dependent biological processes through mechanisms that do not depend on SMADs (19).

SMAD proteins form complexes with distinct partners to affect varied cellular processes. The proteins with which they interact do not share any conserved motifs. For example, the MH2 domains of SMAD2 and SMAD3, which share 97% amino acid sequence identity, form complexes with the membrane-associated Smad anchor for receptor activation (SARA; also known as zinc finger FYVE-type containing 9 or ZFYVE9) (20, 21); transcription factors like Forkhead box protein H1 (FOXH1) and Smad interacting protein 1 (SIP1) (22, 23); transcriptional corepressors like the proto-oncoprotein SKI, the SKI-related protein (SNON; also called SKI-like proto-oncogene or SKIL) (SKI homolog), and TGF-β–induced factor homeobox (TGIF) (2426); transcriptional coactivators such as CREB (cAMP response element–binding protein)–binding protein (CBP) and its homolog p300 (27); and proteins that terminate TGF-β signals such as MAN antigen 1 (MAN1; also known as LEM domain-containing protein 3 or LEMD3) and transmembrane prostate androgen-induced protein (TMEPAI) (28, 29). Because the DNA-binding specificity of the SMAD3 MH1 domain is weak and SMAD2 does not directly bind to DNA due to a loop insertion in the MH1 domain, cooperation of SMAD2 and SMAD3 with other cofactors is critical for the TGF-β–dependent regulation of many genes (30). The structures of several dimeric SMAD complexes have been solved by x-ray crystallography: SMAD2-SARA (20), SMAD3-SARA (21), SMAD4-SKI (31), and SMAD4-SNON (32). Structures have also been described for heterotrimeric SMAD2-SMAD2-SMAD4 and SMAD3-SMAD3-SMAD4 complexes (33). However, the structural basis underlying the selection of cofactors by SMAD proteins is not understood.

FOXH1 is one of the forkhead/winged-helix transcription factors that regulate the differentiation of stem cells. In mammals, the SMAD2-FOXH1 complex recognizes the goosecoid (Gsc) promoter to stimulate Gsc expression, which results in specification of the anterior primitive streak, whereas the SMAD3-FOXH1 complex suppresses activation of the Gsc promoter (22, 34). There are two SMAD2- and SMAD3-binding motifs, a FAST/FOXH1 motif (FM), and a SMAD interaction motif (SIM) in the SMAD interaction domain of FOXH1 (Fig. 1A). Although the FM is specific for FOXH1, the Pro-Pro-Asn-Lys-Ser sequence of the SIM is also present in the Xenopus laevis homeodomain transcription factor Mixer and in TMEPAI, which sequesters SMAD2 and SMAD3 from TβRI kinase activation (29, 3537). SKI is a transcription corepressor for SMAD proteins and acts as a negative regulator of TGF-β signaling (24, 38). High abundance of SKI is observed in many human cancers, and reduction of SKI abundance in cancer cells inhibits tumor growth (11, 39). SKI has an R-SMAD–binding domain and a SMAD4-binding domain (Fig. 1B) that mediate its interactions with R-SMADs and Co-SMAD, respectively. Mutations in the R-SMAD–binding domain cause Shprintzen-Goldberg syndrome (15, 16). Although the structure of the SMAD4-SKI complex shows that binding to SKI disrupts the interaction between the R-SMAD and the Co-SMAD (31), the R-SMAD–binding domain of SKI interacts with the heterotrimeric form of SMAD2 and SMAD3 (21, 40). Because the SMAD-binding domains of both FOXH1 and SKI contain proline residues, FOXH1 and SKI are predicted to bind to SMAD2 and SMAD3 in a manner similar to SARA, which uses a Pro-rich sequence to interact with the hydrophobic patches of the MH2 domains of SMAD2 and SMAD3; SKI and the SIM of FOXH1 compete with SARA for binding to SMAD2 and SMAD3 (21, 37); however, SKI does not affect the binding of FOXH1 to SMAD2 and SMAD3 (41). This implies that FOXH1 and SKI bind to the different surfaces of SMAD2 and SMAD3.

Fig. 1 Structure determination of the SMAD3-FOXH1 and SMAD2-SKI complexes.

(A) Domain structure of human Forkhead box protein H1 (FOXH1) and amino acid sequence alignment of vertebrate FOXH1 homologs. FOXH1 contains the forkhead (FH) DNA-binding domain and the SMAD interaction domain (SID). The SID contains both the FAST/FOXH1 motif (FM) and SMAD interaction motif (SIM). Schematic representations of the fragments used in this study are shown below the sequence alignment. (B) Domain structure of human SKI and amino acid sequence alignment of SKI homologs. SKI contains the receptor-regulated SMAD (R-SMAD)–binding domain (SBD), the Dachshund homology domain (DHD), and the SMAD4-binding domain (SMAD4-BD). Residues that are indispensable for SMAD2 and SMAD3 binding (21, 40) are marked with Xs. Residues that cause Shprintzen-Goldberg syndrome when mutated (15, 16) are indicated with asterisks. (C and D) Interaction of thioredoxin (Trx)–tagged SMAD2C-2E and Trx–SMAD3C-2E with the indicated histidine (His)– and Trx-tagged fragments of FOXH1 (C) and SKI (D) analyzed by a His-tag pull-down assay. The gel images are representative of triplicate experiments. (E) Monomeric structure of SMAD3 (green) in complex with FOXH1 (orange). (F) Trimeric structure of SMAD3 in complex with FOXH1. (G) Monomeric structure of SMAD2 (cyan) in complex with SKI (red). (H) Trimeric structure of SMAD2 in complex with SKI.

To clarify the cofactor selection mechanism of SMAD2 and SMAD3, we determined the structures of the SMAD3-FOXH1 and SMAD2-SKI complexes by x-ray crystallography. The structural analyses of the complexes showed that FOXH1 and SKI bind to SMAD2 and SMAD3 using mechanisms that are different from those by which SARA binds to SMAD2 and SMAD3. There are multiple hydrophobic patches on the MH2 domains of SMAD2 and SMAD3, and different cofactors tether to distinct subsets of these patches. Our results reveal the mechanism by which SMAD2 and SMAD3 bind various cofactors with high selectivity.

RESULTS

Structure determination of SMAD3-FOXH1 and SMAD2-SKI complexes

Before our cocrystallization experiments, we determined the SMAD-binding regions of FOXH1 and SKI based on sequence conservation. The amino acid sequence alignment of the FOXH1 SIM (36) shows that the Asp-Leu-Asp and Pro-Pro-Asn-Lys-Ser motifs are highly conserved among orthologs (Fig. 1A). The amino acid sequence alignment of the R-SMAD–binding domain of SKI and that of the closely related SNON shows that residues 17 to 41 of SKI are conserved among the homologs (Fig. 1B). A pull-down assay using thioredoxin (Trx)–tagged SMAD2, SMAD3, FOXH1, and SKI (fig. S1, A, B, D, and E) showed that residues 322 to 345 of FOXH1 and residues 16 to 40 of SKI were sufficient for binding to SMAD2 and SMAD3 (Fig. 1, C and D). Cocrystals of the SMAD3-FOXH1 complex were obtained using an MH2 domain of SMAD3 (residues 220 to 416) in which the C-terminal phosphorylation region was truncated (SMAD3C-dC) and residues 322 to 345 of FOXH1 that were tagged at the N terminus with a 6× histidine (His) tag and a Trx tag (His-Trx-FOXH1) (fig. S1, A and D). Cocrystals of the SMAD2-SKI complex were obtained using the phosphorylated-state mimics (S465E and S467E) of the MH2 domain of SMAD2 (residues 262 to 467, SMAD2C-2E), and residues 16 to 40 of SKI fused to acidic residues (Ser-Asp-Glu-Asp) at the C terminus (fig. S1, B and E).

The structures of the SMAD3-FOXH1 and SMAD2-SKI complexes were determined by x-ray crystallography at resolutions of 2.40 and 1.85 Å, respectively. The structures of the complexes show that both SMAD2 and SMAD3 contain a three-helix bundle region, a β-sandwich region, and a loop-helix region, similar to previously reported SMAD2 and SMAD3 structures (Fig. 1, E to H, and figs. S2, A to C, and S3, A to C) (33). Although the cofactors of SMAD2 and SMAD3 were predicted to bind to SMAD2 and SMAD3 using binding mechanisms similar to those of SARA (20, 21), the structures of the SMAD3-FOXH1 and SMAD2-SKI complexes show that FOXH1 and SKI bind to SMAD2 and SMAD3 using different mechanisms distinct from one another and from SARA.

Structure of the SMAD3-FOXH1 complex

In the SMAD3-FOXH1 complex, FOXH1 adopts an extended structure that contains an N-terminal amphipathic helix (helix 1), a β strand (strand 1), and a C-terminal hydrophobic helix (helix 2) that contact SMAD3 (Fig. 2 and fig. S2, D and E). Although we obtained crystals of the SMAD3-FOXH1 complex using a form of SMAD3 lacking the C-terminal phosphorylation region that mediates trimerization of SMAD3, the SMAD3 in the SMAD3-FOXH1 complex forms a trimeric structure in the crystal similarly to other SMAD structures (Fig. 1F) (33, 42, 43). The interfacial area between SMAD3 and FOXH1 is approximately 1112 Å2; thus, 10.5% of the total surface area of SMAD3 is involved in binding FOXH1. The SMAD3 structure in the SMAD3-FOXH1 complex is similar to the structure of SMAD3 in the SMAD3-SMAD4 heterotrimeric complex [Protein Data Bank (PDB) ID: 1U7F] (33). The root mean square deviation (RMSD) between the structure of SMAD3 in the SMAD3-FOXH1 complex and the chain A structure of SMAD3 in the SMAD3-SMAD4 complex, which interacts with SMAD4 at the three-helix bundle region, is 0.57 Å for 186 superposed Cα atoms. The RMSD between the structure of SMAD3 in the SMAD3-FOXH1 complex and the chain C structure of SMAD3 in the SMAD3-SMAD4 complex, which interacts with SMAD4 at the loop-helix region, is 0.56 Å for 184 superposed Cα atoms (fig. S4A).

Fig. 2 Structure of the SMAD3-FOXH1 complex.

(A) The interaction between SMAD3 (green) and FOXH1 (orange) at the three-helix bundle region of SMAD3. Hydrogen bonds are shown as cyan dotted lines. (B) The interaction between SMAD3 and FOXH1 at the β-sandwich region of SMAD3. (C) SDS–polyacrylamide gel electrophoresis (PAGE) of proteins after His-tag pull-down assay of SMAD3 (Trx–SMAD3C-2E) and the indicated forms of FOXH1 [His-Trx-FOXH1(322–345)]. The image is a composite from two separate gel images. The image is representative of triplicate experiments. (D) Relative affinities of the indicated His-Trx-FOXH1(322–345) mutants for binding to Trx–SMAD3C-2E. Data are mean ± SEM from triplicate experiments. *P < 0.05 compared to wild type (WT) by Student’s t test. **P < 0.01 compared to WT by Student’s t test.

In the three-helix bundle region of SMAD3, the hydrophobic surface of the H2-H3-H4 helices interacts with the hydrophobic surface of the amphipathic helix 1 of FOXH1 (Fig. 2A). The hydrophobic surface containing Pro318, Pro328, Leu351, Leu366, and Met369 of SMAD3 forms hydrophobic contacts with the surface containing Leu323, Leu324, Leu327, and Phe331 of FOXH1. In this region, SMAD3 and FOXH1 form four intermolecular hydrogen bonds to stabilize the complex (table S1). In the β-sandwich region, the β strand of FOXH1 forms a β sheet with one of the β sheets of SMAD3 (β1-β11-β10-β7-β8-FOXH1 strand 1), and the C-terminal hydrophobic helix of FOXH1 forms contacts with the hydrophobic surface surrounded by helix H2 and strands β8 and β9. These contacts are stabilized by 10 hydrogen bonds (Fig. 2B and table S1). In this region, the highly conserved Pro-Pro-Asn-Lys-Ser motif of FOXH1 is accommodated by the cleft between the H2 helix and the L3 loop of SMAD3, and the residues Asn337, Lys338, and Ser339 of the FOXH1 sequence form the β strand that is parallel to the β8 strand of SMAD3. The side-chain amide group of Asn337 in FOXH1 is inserted into the H2-L3 cleft of SMAD3 and forms four hydrogen bonds with Val331, Trp380, Ala382, and Glu383 of SMAD3 to stabilize the SMAD3-FOXH1 interaction. In addition, the side chains of Lys338, Ile340, and Val343 of FOXH1 form hydrophobic contacts with the hydrophobic surface of SMAD3 that contains Tyr324, Trp326, Cys332, Ile334, and Leu340.

To analyze the importance of the interacting residues of SMAD3 and FOXH1, we created alanine mutants at positions Asp326, Leu327, Asp328, Leu330, Phe331, Pro335, Pro336, Asn337, Lys338, Ser339, Ile340, and Val343 in FOXH1 to evaluate the importance of their side chains and analyzed the SMAD3-binding ability of these mutants by pull-down assays (Fig. 2, C and D). The L327A and F331A mutants, which affect the N terminus of the SMAD-binding region of FOXH1, reduced SMAD3-binding ability compared to the wild-type protein. In contrast, mutations of the highly conserved acidic residues Asp326 and Asp328 did not reduce the ability of FoxH1 to bind SMAD3; however, the side chain of Asp328 forms a hydrogen bond with SMAD3, suggesting that the SMAD3-FOXH1 interactions at the three-helix bundle region are mostly hydrophobic. Individual mutations of the Pro336, Asn337, Lys338, and Ser339 residues at the C terminus of the SMAD-binding region of FOXH1 to alanine (P336A, N337A, K338A, and S339A) greatly reduced the SMAD3-binding ability of FOXH1 (Fig. 2, C and D). This sequence is conserved in other SMAD2- and SMAD3-binding proteins such as Mixer and TMEPAI and is necessary for SMAD2 and SMAD3 binding by these proteins; mutations in this sequence result in loss of function of these proteins in cells (29, 37). Although mutations in the C-terminal helix 2 of FOXH1 (I340A and V343A) had little effect on SMAD3 binding, the pull-down assay of a deletion construct of FOXH1 showed that residues 341 to 345 of FOXH1 are required for SMAD2 and SMAD3 binding (Fig. 1C). The C-terminal helix 2 of FOXH1 is also important for SMAD3 binding.

Structure of the SMAD2-SKI complex

In the SMAD2-SKI complex, the R-SMAD–binding domain of SKI forms an N-terminal amphipathic helix and a C-terminal tail structure and interacts with the H3-H5 face of the three-helix bundle region of SMAD2 (Fig. 3 and fig. S3, D and E). Similar to SMAD3 in the SMAD3-FOXH1 complex, the SMAD2 of the SMAD2-SKI complex forms a trimeric structure in the crystal (Fig. 1H). The interfacial area between SMAD2 and SKI is approximately 920 Å2, so 7.7% of the total surface area of SMAD2 is involved in SKI binding. The structure of SMAD2 in the SMAD2-SKI complex is similar to the structure of SMAD2 in the SMAD2-SMAD4 heterotrimeric complex (PDB ID: 1U7V) (33). The RMSD between the structure of SMAD2 in the SMAD2-SKI complex and the chain A structure of SMAD2 in the SMAD2-SMAD4 complex, which interacts with SMAD4 at the three-helix bundle region, is 0.69 Å for 183 superposed Cα atoms. The RMSD between the structure of SMAD2 in the SMAD2-SKI complex and the chain C structure of SMAD2 in the SMAD2-SMAD4 complex, which interacts with SMAD4 at the loop-helix region, is 0.66 Å for 181 superposed Cα atoms (fig. S4B).

Fig. 3 Structure of the SMAD2-SKI complex.

(A) The interaction between SMAD2 (cyan) and SKI (red) at the three-helix bundle region of SMAD2. Hydrogen bonds are shown as cyan dotted lines. (B) SKI Leu32 binding pocket of SMAD2. The surface of SMAD2 is colored gray. (C) SDS-PAGE of proteins after His-tag pull-down assay of SMAD2 (Trx–SMAD2C-2E) and forms of SKI [His-Trx-SKI (16–40)] bearing the indicated amino acid substitutions that cause Shprintzen-Goldberg syndrome. The gel image is representative of triplicate experiments.

At the H3-H5 face of the three-helix bundle region of SMAD2, the hydrophobic surface of the amphipathic helix of SKI (Leu17, Leu21, and Phe24) forms hydrophobic contacts with the hydrophobic surface of SMAD2 (Phe390, Ala391, Leu394, Ala395, Val398, Pro445, Trp448, and Val452), and the complex is stabilized by five hydrogen bonds (Fig. 3A and table S2). Among the interacting residues, the side chains of Phe24 and Pro35 in SKI form stacking interactions with the side chain of Trp448 in SMAD2. In addition to the intermolecular interactions at the H3-H5 face of SMAD2, the side chain of Leu32 in SKI is inserted into the hydrophobic pocket between the three-helix bundle region and the β-sandwich region of SMAD2, which contains Trp274, Tyr340, Gly342, Val345, Phe385, Asn387, Leu442, and Pro445, to stabilize the SMAD2-SKI interaction (Fig. 3B). In this region, Ser31 and Leu32 in SKI form hydrogen bonds with Gly342 and Asn387 in SMAD2 to further stabilize the SMAD2-SKI complex (table S2).

Previous studies have shown that mutations to Leu17, Leu21, and Phe24 cause a loss in the SMAD3-binding ability of SKI (Fig. 1B) (21, 40). The structure of the SMAD2-SKI complex shows that these residues form hydrophobic contacts with the H3-H5 face of SMAD2. Mutations in the R-SMAD–binding domain of SKI cause Shprintzen-Goldberg syndrome (Fig. 1B) (15, 16). The SMAD2-SKI structure shows that these mutations affect residues that are important for the interaction between SMAD2 and SKI. Introducing the mutations that cause Shprintzen-Goldberg syndrome into SKI (for example, S28T, S31L, L32V, G34S, and P35S) abrogated the ability of SKI to bind SMAD2 (Fig. 3C). We predict that the loss of the interaction between SMAD and SKI at the H3-H5 face of SMAD2 and SMAD3 is responsible for the increased TGF-β signaling that causes Shprintzen-Goldberg syndrome (15).

Comparison of SMAD-cofactor complexes

To analyze the cofactor-binding regions of SMAD2 and SMAD3, we compared the structures of the SMAD3-FOXH1 and SMAD2-SKI complexes with the preexisting structures of SMAD-cofactor complexes (Fig. 4, A and B). Although FOXH1 and SKI were predicted to bind to SMAD2 and SMAD3 in a manner similar to SARA (21, 37), the structures of the complexes show that FOXH1 and SKI interact with different regions of SMAD2 and SMAD3. The FOXH1 and SKI constructs used for structure determination do not compete with one another for SMAD2 and SMAD3 binding (fig. S5, A and B). The structures of the SMAD2-SARA and SMAD3-SARA complexes (20, 21) show that the SMAD-binding domain of SARA can be divided into three substructures: (i) the rigid coil that interacts with the hydrophobic surface of helix H2 and strands β8 and β9 of SMAD2 and SMAD3, (ii) the amphipathic α helix that interacts with the hydrophobic surface of strands β5 and β6 of SMAD2 and SMAD3, and (iii) the β strand that interacts with the hydrophobic surface of helices H3 to H5 of SMAD2 and SMAD3. Here, we designate the hydrophobic patches on the surfaces of SMAD2 and SMAD3 that interact with the rigid coil, the amphipathic α helix, and the β strand of SARA as patches B2, B1, and A1, respectively (Fig. 4C). SARA tethers to these patches to interact with SMAD2 and SMAD3; these three substructures of SARA are indispensable for the binding of SARA to the MH2 domains of SMAD2 and SMAD3 (20, 44).

Fig. 4 Structural comparison of the SMAD-cofactor complexes.

(A) The structures of the SMAD-cofactor complexes. The SMAD2, SMAD3, and SMAD4 structures are shown as gray surface models. The bound FOXH1 SIM, SKI R-SMAD–binding domain, Smad anchor for receptor activation (SARA) Smad–binding domain, and SKI SMAD4–binding domain are shown as orange, red, black, and yellow cartoon models, respectively. (B) The FOXH1 SIM, SKI R-SMAD–binding domain, SARA Smad–binding domain, and SKI SMAD4–binding domain binding surfaces of SMAD proteins are colored orange, red, black, and yellow, respectively. (C) Schematic diagram of the hydrophobic patches of the MAD homology 2 (MH2) domains of SMAD2 and SMAD3.

Our structural analyses of the SMAD3-FOXH1 and SMAD2-SKI complexes show that FOXH1 and SKI interact with SMAD2 and SMAD3 using different combinations of patches. The combinations of patches used for FOXH1 and SKI bindings are also different from that used for SARA binding. FOXH1 interacts with SMAD3 by tethering to the hydrophobic surface of the H2-H3-H4 face of the three-helix bundle region (patch A2), the cleft between the H2 helix and the L3 loop (patch B3), and patch B2. The cofactor-binding patches B2 and B3 correspond to the SKI-binding surface of SMAD4 (Fig. 4, A and B) (31). The structure of the SMAD2-SKI complex shows that SKI mainly interacts with SMAD2 at patch A1, similar to SARA. However, SKI interacts through its amphipathic α helix, and the direction of the peptide chain of SKI is opposite to that of SARA. In contrast, SARA interacts with patch A1 of SMAD2 through a β strand (20, 21). In addition, SKI interacts with SMAD2 using an additional small patch on SMAD2 that is not used for binding to SARA (patch A3, the small pocket of SMAD2 between the three-helix bundle region and the β-sandwich region that interacts with Leu32 of SKI).

The structures of the SMAD-cofactor complexes reveal that the MH2 domains of SMAD2 and SMAD3 have multiple cofactor-binding patches on their surfaces and that cofactors tether to distinct combinations of the patches to interact with SMAD2 and SMAD3. To confirm this cofactor-binding model of SMAD2 and SMAD3, we generated chimeric proteins of the cofactors and analyzed the SMAD2- and SMAD3-binding abilities of these chimeric proteins using a surface plasmon resonance (SPR) assay (Fig. 5 and figs. S6, A to C, and S7, A to D). Trx-SARA(697–715)-FOXH1 contains residues 697 to 715 of SARA fused to Trx at the N terminus and to FoxH1 (residues 316 to 355) at the C terminus; Trx-SARA(682–715)-FOXH1 contains residues 682 to 715 of SARA fused to Trx at the N terminus and to FoxH1 (residues 316 to 355) at the C terminus; Trx-FOXH1-SKI contains residues 316 to 355 of FOXH1 fused to Trx at the N terminus and to SKI (residues 16 to 45) at the C terminus (fig. S1F). If the cofactor-binding patches of SMAD2 and SMAD3 function independently of one other, the chimeric protein that tethers to more patches should interact with SMAD2 and SMAD3 more strongly than the individual proteins from which the chimera is constructed. These chimeric proteins did not interact with the SMAD4 MH2 domain (fig. S8). The SPR assay using the phosphorylated-state mimics of SMAD2 and SMAD3 (His–Trx–SMAD2C-2E and His–Trx–SMAD3C-2E) and FOXH1 (Trx-FOXH1), which tethers to patches A2-B3-B2, showed that FOXH1 bound to SMAD2 and SMAD3 with dissociation constants of 4.82 and 8.00 μM, respectively. In contrast, the SARA-FOXH1 chimeric protein that tethers to patches A1-A2-B3-B2 [Trx-SARA(697–715)-FOXH1] bound to His–Trx–SMAD2C-2E and His–Trx–SMAD3C-2E with dissociation constants of 0.84 and 0.75 μM, respectively. The SARA-FOXH1 chimeric protein that tethers to patches B1-A1-A2-B3-B2 [Trx-SARA(682–715)-FOXH1] bound to His–Trx–SMAD2C-2E and His–Trx–SMAD3C-2E with dissociation constants of 0.48 and 0.54 μM, respectively (Fig. 5, A to C, and fig. S1, A, B, D, and F). As expected, the chimeric proteins that tether to more patches interacted with SMAD2 and SMAD3 the most strongly (Fig. 5, B and C, and fig. S7, A and B). Because FOXH1 preferentially binds to phosphorylated SMAD2 (22), Trx-FOXH1 bound only weakly to the forms of SMAD2 and SMAD3 (Trx–His–SMAD2C-dC and Trx–His–SMAD3C-dC; fig. S1, A and B) lacking the phosphorylation site (>100 μM; figs. S6, A and B, and S7, C and D). In contrast, Trx-SARA(682–715)-FOXH1 bound to Trx–His–SMAD2C-dC and Trx–His–SMAD3C-dC with approximately the same dissociation constants as His–Trx–SMAD2C-2E and His–Trx–SMAD3C-2E, respectively. Because SARA binds to the monomeric state of SMAD2 and SMAD3 (20, 21), we predicted that Trx-SARA(682–715)-FOXH1 would also bind to the SMAD2 and SMAD3 in the monomeric state and disrupt the trimerization of SMAD2 and SMAD3. Gel filtration analysis showed that Trx-SARA(682–715)-FOXH1 indeed disrupted the trimerization of SMAD3 (Fig. 5D). We also generated the FOXH1-SKI chimeric protein (Trx-FOXH1-SKI), which should be able to bind to two SMAD2 or two SMAD3 proteins simultaneously (Fig. 5A and fig. S1F), and analyzed the SMAD2- and SMAD3-binding ability of this chimeric protein. The SPR analyses showed that Trx-FOXH1-SKI bound tightly to His–Trx–SMAD2C-2E and His–Trx–SMAD3C-2E; the dissociation constants for His–Trx–SMAD2C-2E and His–Trx–SMAD3C-2E were 28 and 40 nM, respectively (Fig. 5E). Gel filtration analysis showed that Trx-FOXH1-SKI enhanced the trimerization of SMAD2 and SMAD3, as expected (Fig. 5D). We also analyzed the SMAD2- and SMAD3-binding abilities of the chimeric proteins through a His-tag pull-down competition assay (figs. S1F and S9, A and B). The presence of Trx-SARA(697–715)-FOXH1, Trx-SARA(682–715)-FOXH1, or Trx-FOXH1-SKI reduced the ability of His-Trx-FOXH1(322–345) to bind to Trx–SMAD2C-2E and Trx–SMAD3C-2E. In contrast, Trx-FOXH1-SKI inhibited the ability of His-Trx-SARA(697–715)-FOXH1 and His-Trx-SARA(682–715)-FOXH1 to bind to Trx–SMAD2C-2E and Trx–SMAD3C-2E, but Trx-FOXH1 did not. The results of the binding assays and the gel filtration assays suggest that molecules that tether to hydrophobic patches have the potential to modulate characteristics of SMAD2 and SMAD3, such as multimerization and cofactor binding.

Fig. 5 SMAD2- and SMAD3-binding abilities of the chimeric proteins.

(A) Map of the contacts made between the indicated chimeric proteins and the MH2 domain (green) of SMAD2 or SMAD3. A1, hydrophobic surface of H3-H5; A2, hydrophobic surface of H2-H3-H4; A3, small pocket between the three-helix bundle region and the β-sandwich region; B1, hydrophobic surface of strands β5 and β6; B2, hydrophobic surface of helix H2 and strands β8 and β9; B3, cleft between the H2 helix and the L3 loop. SARA, FOXH1, and SKI in chimeric proteins are colored black, orange, and red, respectively. (B and C) Surface plasmon resonance (SPR) diagrams showing the binding of the chimeric proteins Trx-FOXH1, Trx-SARA(697–715)-FOXH1, and Trx-SARA(682–715)-FOXH1 to His–Trx–SMAD2C-2E (B) and His–Trx–SMAD3C-2E (C). The indicated concentrations of chimeric proteins were injected through flow cells. The dissociation constants (KD) were obtained by curve fitting based on a steady-state affinity model. Data are mean ± SEM of triplicate experiments. RU, response units. (D) Gel filtration analysis of Trx-SARA(682–715)-FOXH1 (Mr = 19566) and Trx-FOXH1-SKI (Mr = 20529) mixed with His–Trx–SMAD2C-2E (Mr = 36946) or His–Trx–SMAD3C-2E (Mr = 36930) as indicated. The peak positions of the marker proteins are indicated by black triangles at the top of the chromatogram. Each protein peak (red line) was analyzed by SDS-PAGE and shown at the right. Data are representative of three experiments. AU, absorbance units. (E) SPR diagrams showing the binding of Trx-FOXH1-SKI to His–Trx–SMAD2C-2E and His–Trx–SMAD3C-2E. The KD values and kinetic parameters were obtained from curve fittings based on a two-state reaction model. Data are representative of triplicate experiments.

To further confirm the cofactor-binding model of SMAD2 and SMAD3, we performed a coimmunoprecipitation assay in cultured cells. We transiently transfected human embryonic kidney (HEK) 293 cells with Myc-SMAD2 and HA-FOXH1 with or without FOXH1(322–345), SARA(697–715)-FOXH1, or SARA(682–715)-FOXH1, and immunoprecipitated the SMAD2-FOXH1 complexes using an antibody recognizing Myc (fig. S10). Coexpression of the FOXH1 fragment or SARA-FOXH1 chimeric proteins with Myc-SMAD2 and HA-FOXH1 partially inhibited activin-induced association of SMAD2 and FOXH1 (fig. S10, A and B). Luciferase assays using SMAD-binding elements also confirmed that increasing the expression of SARA-FOXH1 chimeric proteins significantly inhibited activin- and TGF-β–dependent transcriptional activation through the SMAD signaling pathway (fig. S10, C and D). This inhibition was not observed when bone morphogenetic protein (BMP)–dependent transcriptional activation was examined with a BMP-SMAD–specific luciferase reporter (fig. S10E), indicating the specificity of inhibition with SARA-FOXH1 chimeric proteins. These results revealed that SMAD proteins use multiple hydrophobic patches to associate with cofactors with high specificity in cells.

DISCUSSION

SMAD proteins are the core transcription factors involved in TGF-β signaling and regulate the expression of many genes by forming various SMAD-cofactor complexes, some of which are stimulatory and others repressive. Although the cofactor selectivity of SMAD proteins is a key factor for understanding TGF-β–dependent gene regulation, the mechanism by which SMAD proteins select cofactors was previously unclear. Here, we analyzed the cofactor selection mechanism of the MH2 domains of SMAD2 and SMAD3. The structures of the SMAD3-FOXH1 and the SMAD2-SKI complexes show that each cofactor interacts with the MH2 domains of the SMAD by binding to a combination of hydrophobic patches.

We determined the structure of the SMAD3-FOXH1 complex using the FOXH1 SIM. FOXH1 has an additional SMAD2-binding motif (FM) that is located N-terminal to the SIM (35). Because FOXH1 preferentially binds to phosphorylated SMAD2 (21), the FM and SIM of FOXH1 are predicted to bind to heterotrimeric complexes composed of SMAD4 and two molecules of either SMAD2 or SMAD3. The similarity between the SMAD3 structure in complex with FOXH1 and that in complex with SMAD4 (fig. S4A) suggests that the FOXH1 SIM binds the SMAD3-SMAD4 heterotrimeric complex similarly to how it binds to SMAD3 in the SMAD3-FOXH1 complex. Because the heterotrimeric complexes include two SMAD2 or two SMAD3 proteins, cofactors of SMAD2 and SMAD3 are predicted to be able to bind to the heterotrimeric complex not only by using a combination of the hydrophobic patches of one SMAD2 or one SMAD3 protein but also by using a combination of the hydrophobic patches of two SMAD2 or two SMAD3 proteins (Fig. 6A). A previous study has shown that FOXH1 forms a complex composed of one FOXH1 molecule, two SMAD2 molecules, and one SMAD4 molecule on DNA after TGF-β stimulation (45). The structure of the SMAD3-FOXH1 complex shows that the FOXH1 SIM interacts with one SMAD3 using the patches A2, B2, and B3. The stoichiometry of the SMAD2-SMAD4-FOXH1 complex suggests that the SIM interacts with one SMAD2 and that the FM interacts with the other SMAD2 in the SMAD2-SMAD4 heterotrimeric complex (Fig. 6B).

Fig. 6 Proposed cofactor-binding mechanism of the SMAD2-SMAD4 and SMAD3-SMAD4 heterotrimers.

(A) SMAD heterotrimers are composed of two SMAD2 or SMAD3 (SMAD2/3) molecules (green) and one SMAD4 molecule (blue). The diagram represents the surface of each SMAD protein’s MH2 domain. Each SMAD2 (or SMAD3) protomer in the heterotrimer has hydrophobic patches (A1 to A3 and B1 to B3) on its MH2 domain with which it interacts with SMAD cofactors. (B) A cofactor that makes contacts with the hydrophobic patches of two SMAD2/3 protomers (magenta) would generate a 1:2:1 cofactor A:SMAD2/3:SMAD4 complex. (C) A cofactor that tethers to hydrophobic patches of one SMAD2/3 protomer (pink) would create a 2:2:1 cofactor B:SMAD2/3:SMAD4 complex. (D) A cofactor that interacts with SMAD2/3 and SMAD4 (brown) would form a 1:2:1 cofactor C:SMAD2/3:SMAD4 complex. (E) Cofactors that do not share hydrophobic patches would cooperatively bind to SMAD heterotrimers to form transcription factor complexes such as a 1:2:2:1 cofactor A:cofactor B:SMAD2/3:SMAD4 complex, 2:2:2:1 cofactor B:cofactor D:SMAD2/3:SMAD4 complex, or a 1:1:2:1 cofactor A:cofactor C:SMAD2/3:SMAD4 complex.

The structure of the SMAD2-SKI complex shows that the R-SMAD–binding domain of SKI interacts with the trimeric SMAD2 structure (Fig. 1H). SKI has an R-SMAD–binding domain, which interacts with trimeric forms of SMAD proteins (21, 40), and a SMAD4-binding domain, which disrupts SMAD trimerization (31). The similarity between the structure of SMAD2 in complex with SKI and that of SMAD2 in complex with SMAD4 (fig. S4B) suggests that the R-SMAD–binding domain of SKI is able to bind to the SMAD2-SMAD4 heterotrimeric complex by itself and forms a complex consisting of two SKI molecules, two SMAD2 molecules, and one SMAD4 molecule (Fig. 6C). The R-SMAD–binding domain of SKI is predicted to repress TGF-β signaling through competition with the transcription coactivators CBP and p300, which bind to the heterotrimeric SMAD complex (40). In addition, because the SMAD4-binding domain of SKI disrupts the heterotrimeric SMAD complex (31), recruitment of the SMAD4-binding domain of SKI to the heterotrimeric SMAD complex by the interaction between the R-SMAD–binding domain of SKI and the R-SMAD would also be predicted to disrupt the heterotrimeric SMAD complex and repress TGF-β signaling. Meanwhile, SNON, which shares 37% amino acid sequence identity with SKI, stabilizes the SMAD3-SMAD4 heterotrimeric complex (32). The structure of the SMAD4-SNON complex shows that the SMAD4-binding domain of SNON should not interfere with SMAD trimerization. SNON is predicted to interact with the SMAD3-SMAD4 heterotrimeric complex using the interaction between the R-SMAD–binding domain of SNON and one SMAD3 and the interaction between the SMAD4-binding domain of SNON and SMAD4 (Fig. 6D); a connection between one SMAD3 and SMAD4 in the heterotrimeric complex would stabilize the SMAD3-SMAD4 heterotrimeric complex. Stabilization of the SMAD complex was also observed in the gel filtration assay using the FOXH1-SKI chimeric protein (Fig. 5D).

The structures of the SMAD3-FOXH1 and SMAD2-SKI complexes show that the MH2 domains of SMAD2 and SMAD3 have multiple hydrophobic patches on their surfaces (patches A1 to A3 and B1 to B3; Fig. 4C). Other SMAD cofactors, the structures of which remain unknown, are also predicted to bind to these hydrophobic patches (table S3). For example, because CBP competes with SKI for binding to SMAD2 and SMAD3 (40), CBP and its homolog p300 are predicted to bind to patches A1 or A3, or both. The transcriptional corepressor TGIF, which competes with p300 (25), is also predicted to bind to the same site. The R-SMAD–binding domain of the SKI homolog SNON (Fig. 1B) is also predicted to bind to patches A1 and A3. Because the transcription factor Mixer and TMEPAI, which terminates TGF-β signals, contain conserved sequences that are similar to the Pro-Pro-Asn-Lys-Ser sequence of the FOXH1 SIM, Mixer and TMEPAI are predicted to bind to patches B2 and B3. Although MAN1, which terminates TGF-β signals, does not contain sequences that are similar to SIM, MAN1 competes with the FOXH1 SIM peptide for binding to SMAD2 (28). MAN1 is also predicted to bind to patches B2 and B3. These predictions suggest that patches A1 and A3 are used for the binding of transcriptional coactivators and corepressors and that patches B2 and B3 are the surfaces used for the binding of transcription factors and for TGF-β signal termination.

However, these cofactors may also use additional patches for SMAD2 and SMAD3 binding, and determination of the precise mechanisms by which each cofactor binds to SMAD2 and SMAD3 requires structural studies of the relevant complexes. Although SMAD cofactors that use common hydrophobic patches for SMAD2 and SMAD3 binding compete with one another for binding, SMAD cofactors that use different hydrophobic patches could bind to SMAD2 and SMAD3 cooperatively (Fig. 6E), provided the full-length proteins do not sterically hinder the formation of such complexes. For example, the transcriptional coactivators or corepressors that bind to patches A1 and A3 are predicted to bind to SMAD2 and SMAD3 cooperatively with transcription factors that bind to patches B2 and B3. In fact, TGIF binds to the SMAD2-SMAD4-FOXH1 complex (25). In addition, our pull-down assay showed that the R-SMAD–binding domain of SKI and the FOXH1 SIM cooperatively bind to SMAD2 and SMAD3 (fig. S5, A and B). The functions of SMAD2 and SMAD3 are regulated by various SMAD cofactors; therefore, elucidation of the cooperative and competitive SMAD2- and SMAD3-binding mechanisms of SMAD cofactors is necessary to understand TGF-β signal transduction in cells.

Because the dysregulation of TGF-β signaling is associated with various diseases, such as cancer and fibrosis (11, 12), the TGF-β signaling pathway has been considered a good drug target. Several drugs that target TGF-β and TGF-β receptors are in clinical trials (6). However, the multifunctionality of TGF-β makes it difficult to develop new drugs that do not have the potential to cause various side effects. Targeting TGF-β signaling therapeutically, in addition to affecting the intended process, may also affect other biological processes, such as cell proliferation, apoptosis, immune response, autophagy, cell migration, angiogenesis, extracellular matrix production, and senescence (410). To precisely regulate TGF-β signaling, it is important to not target elements of the signaling machinery that are common to all or many TGF-β responses, such as the TGF-βs themselves and TGF-β receptors, and to instead target proteins further downstream in the pathway, such as cofactors that interact with SMADs. A similar strategy has achieved positive results in other signaling pathways. For example, the direct inhibition of the NOTCH transcription factor complex using a stapled peptide whose structure is constrained by a covalent linkage between side chains of two amino acids antagonizes NOTCH signaling (46). In addition, a peptide corresponding to a region of the transcription factor FOXO4 selectively disrupts FOXO4-p53 interaction, induces apoptosis in senescent cells, and restores tissue homeostasis in response to chemotoxicity and aging (47). However, the poor understanding of the cofactor selection mechanism of SMAD2 and SMAD3 has hindered drug design strategies that target SMAD2-cofactor and SMAD3-cofactor interactions; to inhibit specific cofactor interactions with SMADs, it is necessary to understand the cofactor selection mechanism of SMAD2 and SMAD3. Our structural study identified the six hydrophobic patches that are used for cofactor binding of SMAD2 and SMAD3 (patches A1 to A3 and B1 to B3; Fig. 4C). Cofactors can bind to the MH2 domains of SMAD2 and SMAD3 in a cooperative or competitive manner based on the combinations of the hydrophobic patches that the cofactors require for binding (Fig. 6).

Small molecules that bind to one or more hydrophobic patches or chimeric proteins such as those constructed in this study could inhibit specific cofactor binding to SMADs and, thus, regulate TGF-β signaling more precisely than existing drugs that target TGF-β signaling through different mechanisms. For example, inhibiting protein-protein interactions at patch A1 or A3, or both, inhibits the binding of transcriptional coactivators (CBP and p300) and corepressors (SKI, SNON, and TGIF) to SMAD2 and SMAD3 without inhibiting the binding of cofactors that interact with other patches (for example, FOXH1) and without inhibiting non-SMAD TGF-β signaling. Because the abnormal activation and repression of TGF-β signals by the binding of p300 and SKI sometimes cause tumor heterogeneity, tumor growth, and fibrosis (11, 39, 48, 49), a molecule that inhibits protein-protein interactions at patch A1 or A3, or both, may become an effective drug that precisely regulates TGF-β signaling in specific contexts. Our results show the potential for developing new anticancer and antifibrotic drugs that modulate TGF-β signaling with fewer side effects.

MATERIALS AND METHODS

Expression and purification of recombinant proteins

Gene fragments of human SMAD2 (NM_005901), SMAD3 (NM_005902), SMAD4 (NM_005359), FOXH1 (NM_003923), and SKI (NM_003036) were amplified by polymerase chain reaction from complementary DNA and cloned into the pET-48b (+) plasmid (Novagen). The protein constructs used in this study are summarized in fig. S1. Each plasmid was modified using the PrimeSTAR Mutagenesis Basal Kit (TaKaRa) and the In-Fusion HD Cloning Kit (Clontech). For the copurification and cocrystallization of FOXH1 and SKI, gene fragments of SMAD3C-dC and SMAD2C-2E were cloned into pET-48b (+), and the His-tag region of pET-48b (+) was removed. For copurification and cocrystallization with SMAD3C-dC, the pET-48b (+)–FOXH1 plasmid was modified to express N-His-tag-Trx-tag-Gly-Ser-FOXH1(322–345) (His-Trx-FOXH1). For copurification and cocrystallization with SMAD2C-2E, the pET-48b (+)–SKI plasmid was modified to express N-His-tag-Trx-tag-Gly-Ser-HRV3C protease site-SKI (16–40)-Ser-Asp-Glu-Asp-C [His-Trx-HRV3C-SKI (residues 16 to 40)]. For the pull-down assay and the SPR assay, the expression plasmids of SMAD2, SMAD3, FOXH1, and SKI were modified to express N-His-tag-HRV3C protease site-Trx-tag-Gly-Ser-target protein-C.

For SMAD2, SMAD3, and SMAD4 expression, the constructed plasmids were transformed into Escherichia coli BL21(DE3) cells (Novagen) harboring the pG-KJE8 plasmid (TaKaRa). The recombinant E. coli cells were cultivated at 37°C in LB medium containing kanamycin (20 μg/ml), chloramphenicol (50 μg/ml), arabinose (0.5 mg/ml), and tetracycline (5 ng/ml) under aerobic conditions until the optical density at 600 nm reached 0.6. The expression of SMAD2, SMAD3, and SMAD4 was induced by the addition of isopropyl-β-d-thiogalactopyranoside (IPTG) at a final concentration of 0.1 mM. After cultivation at 25°C overnight, the cells were harvested by centrifugation at 5000g for 10 min. For FOXH1 and SKI expression, the constructed plasmids were transformed into E. coli Rosetta(DE3) cells (Novagen). The recombinant E. coli cells were cultivated at 37°C in LB medium containing kanamycin (20 μg/ml) and chloramphenicol (50 μg/ml) under aerobic conditions until the optical density at 600 nm reached 0.6. Protein expression was induced by the addition of 0.1 or 1 mM IPTG for FOXH1 and SKI, respectively. After cultivation at 18°C (FOXH1) or 25°C (SKI) overnight, the cells were harvested by centrifugation at 5000g for 10 min.

For the cocrystallization of SMAD3-FOXH1, E. coli cells overexpressing Trx-tagged SMAD3C-dC without the His tag and His-Trx-FOXH1 were mixed and resuspended in 50 mM tris-HCl (pH 9.0), 200 mM NaCl, 10 mM imidazole, 10% glycerol, and 1 mM tris(2-carboxyethyl)phosphine (TCEP), and the cells were lysed by sonication. After centrifugation at 40,000g for 30 min, the supernatant was purified using Ni-NTA Superflow resin (Qiagen). The SMAD3-FOXH1 complex was eluted with a buffer solution containing 50 mM tris-HCl (pH 9.0), 200 mM NaCl, 200 mM imidazole, 10% glycerol, and 1 mM TCEP. The fused Trx tag of SMAD3 was removed by treatment with HRV3C protease (4°C, overnight). The SMAD3-FOXH1 complex was purified using a Mono Q HR 10/10 column (GE Healthcare) pre-equilibrated with 10 mM tris-HCl (pH 9.0), 10% glycerol, and 1 mM TCEP, and the protein complex was eluted with a linear gradient of 0–1 M NaCl. The SMAD3-FOXH1 complex was further purified using a Superdex 200 HR 10/30 column (GE Healthcare) pre-equilibrated with 10 mM tris-HCl (pH 9.0), 10% glycerol, and 1 mM TCEP. The purified complex was concentrated to 16 mg/ml for crystallization.

For the cocrystallization of SMAD2-SKI, E. coli cells overexpressing the Trx-tagged SMAD2C-2E without the His tag and His-Trx-HRV3C-SKI (16–40) were mixed and resuspended in 50 mM tris-HCl (pH 9.0), 200 mM NaCl, 10 mM imidazole, 10% glycerol, and 1 mM TCEP and lysed by sonication. After centrifugation at 40,000g for 30 min, the supernatant was purified using Ni-NTA Superflow resin and was treated with HRV3C protease using the same method as described for the SMAD3-FOXH1 purification. The SMAD2-SKI complex was further purified using a Mono Q HR 10/10 column as described above. The buffer of the SMAD2-SKI solution was exchanged with 50 mM tris-HCl (pH 9.0), 200 mM NaCl, 200 mM imidazole, 10% glycerol, and 1 mM TCEP, and the protein solution was concentrated to 7.3 mg/ml for crystallization.

For binding assays, His–Trx–SMAD2C-2E, His–Trx–SMAD3C-2E, Trx–His–SMAD2C-dC, Trx–His–SMAD3C-dC, and Trx-His-SMAD4C were purified using Ni-NTA and Mono Q HR 10/10 columns as described above. His-Trx-FOXH1, His-Trx-SKI, and their mutants (except the L330A and F331A mutants of FOXH1) were purified using the same method as the purification of the His–Trx–SMAD2C-2E protein, with the exception that the purification buffers did not contain TCEP. The L330A and F331A mutants of FOXH1 were purified using a similar method as the purification of the His–Trx–SMAD2C-2E protein. Briefly, the protein solution obtained after Ni-NTA purification was purified using an XtalSpeed DA01 column (Mitsubishi Chemical). Trx–SMAD2C-2E and Trx–SMAD3C-2E were purified using a similar method as the purification of the His–Trx–SMAD2C-2E protein. Briefly, protein solutions obtained after Ni-NTA purification were treated with HRV3C protease at 4°C overnight to cleave the N-terminal His tag. The solutions were further purified using a Mono Q HR 10/10 column. Trx-FOXH1 and Trx-SKI proteins were purified using a similar method as the purification of the Trx–SMAD2C-2E protein, with the exception that the purification buffers did not contain TCEP. Purified proteins were stored at −80°C until further use.

Expression and purification of chimeric proteins

Gene fragments of SARA(697–715)-FOXH1, SARA(682–715)-FOXH1, and FOXH1-SKI (fig. S1F) were cloned into the pET-48b (+) plasmid and were modified to express N-His-tag-HRV3C protease site-Trx-tag Gly-Ser-target protein-C as described above. Each protein was overexpressed using the same method as FOXH1 overexpression. The Trx-SARA(697–715)-FOXH1 and Trx-SARA(682–715)-FOXH1 proteins were purified using a similar method as the purification of the FOXH1 L330A protein, with the exception that the N-terminal His tags of these proteins were removed by treatment with HRV3C protease after Ni-NTA purification. The Trx-FOXH1-SKI protein was purified using the same method as the Trx-FOXH1 purification. His-Trx-SARA(697–715)-FOXH1 and His-Trx-SARA(682–715)-FOXH1 were purified using the same method as the His-Trx-FOXH1 protein purification. The purified proteins were stored at −80°C until further use.

Pull-down assay

For the pull-down assay of FOXH1 with SMAD2 and SMAD3, Trx–SMAD2C-2E or Trx–SMAD3C-2E (1.25 μM), His-Trx-FOXH1 (1.25 μM), and the Ni-NTA Superflow resin were mixed in 1 ml of 50 mM tris-HCl (pH 9.0), 200 mM NaCl, 10 mM imidazole, 10% glycerol, and 1 mM TCEP. For the pull-down assay of SKI with SMAD2 and SMAD3, Trx–SMAD2C-2E or Trx–SMAD3C-2E (7.5 μM), His-Trx-SKI (7.5 μM), and Ni-NTA Superflow resin were mixed in 500 μl of 50 mM tris-HCl (pH 9.0), 200 mM NaCl, 10 mM imidazole, 10% glycerol, and 1 mM TCEP. For the cooperative binding assay of FOXH1 and SKI, Trx–SMAD2C-2E or Trx–SMAD3C-2E (7.5 μM), Trx-FOXH1 (7.5 μM), His-Trx-SKI (16–40) (7.5 μM), and Ni-NTA Superflow resin were mixed in 500 μl of 50 mM tris-HCl (pH 9.0), 200 mM NaCl, 10 mM imidazole, 10% glycerol, and 1 mM TCEP. For the pull-down assay of SMAD4 with the chimeric proteins [Trx-FOXH1, Trx-SKI, Trx-SARA(697–715)-FOXH1, Trx-SARA(682–715)-FOXH1, and Trx-FOXH1-SKI], Trx-His-SMAD4 (1.25 μM), the chimeric protein (1.25 μM), and the Ni-NTA Superflow resin were mixed in 1 ml of 50 mM tris-HCl (pH 9.0), 200 mM NaCl, 10 mM imidazole, 10% glycerol, and 1 mM TCEP. For the pull-down competition assay of the chimeric proteins, Trx–SMAD2C-2E or Trx–SMAD3C-2E (1.25 μM), the His-tagged chimeric protein [His-Trx-FOXH1(322–345), His-Trx-SARA(697–715)-FOXH1, or His-Trx-SARA(682–715)-FOXH1] (1.25 μM), the competitor [Trx-FOXH1, SARA(697–715)-FOXH1, Trx-SARA(682–715)-FOXH1, or Trx-FOXH1-SKI] (1.25 μM), and the Ni-NTA Superflow resin were mixed in 1 ml of 50 mM tris-HCl (pH 9.0), 200 mM NaCl, 10 mM imidazole, 10% glycerol, and 1 mM TCEP. The proteins remaining on the resin were then washed with 50 mM tris-HCl (pH 9.0), 200 mM NaCl, 20 mM imidazole, 10% glycerol, and 1 mM TCEP. The proteins eluted with 50 mM tris-HCl (pH 9.0), 200 mM NaCl, 200 mM imidazole, 10% glycerol, and 1 mM TCEP were separated by SDS–polyacrylamide gel electrophoresis (PAGE) and visualized by Coomassie staining. In the mutation analysis of FOXH1, the densities of the protein bands were quantified using ImageJ software (National Institutes of Health), and the concentrations of the Trx–SMAD3C-2E bands were normalized to those of the FOXH1 bands in the same lane of the SDS-PAGE gel.

Crystallization and structure determination

The concentrated SMAD3-FOXH1 and SMAD2-SKI complexes were crystallized by the sitting-drop vapor-diffusion method at 20°C. Crystals of the SMAD3-FOXH1 complex were obtained in a reservoir solution containing 0.1 M citrate (pH 5.4), 0.8% ethylene imine polymer, and 0.5 M NaCl. Crystals of the SMAD2-SKI complex were obtained in a reservoir solution containing 0.1 M sodium acetate trihydrate (pH 4.6) and 1.85 M sodium formate. The x-ray diffraction data sets of the crystals of the SMAD3-FOXH1 and SMAD2-SKI complexes were collected at beamline AR-NE3A of the Photon Factory (Tsukuba, Japan) under cryogenic conditions (95 K). For cryoprotection, the SMAD3-FOXH1 crystal was soaked in reservoir solution supplemented with 40% glycerol for a few seconds. The x-ray diffraction data were indexed and integrated with XDS (50) and scaled with Scala in the CCP4 suite (51). The crystal of the SMAD3-FOXH1 complex diffracted x-rays to a resolution of 2.40 Å. The crystal of the SMAD3-FOXH1 complex belonged to the space group P6322 with the unit cell parameters a = b = 131.87 Å and c = 91.60 Å. The crystal of the SMAD2-SKI complex diffracted x-rays to a resolution of 1.85 Å. The crystal of the SMAD2-SKI complex belonged to the space group I23 with the unit cell parameters a = b = c = 110.84 Å. The initial models of the SMAD3-FOXH1 complex and the SMAD2-SKI complex were determined by the molecular replacement method using the program MOLREP (52) with the coordinates of the SMAD3 MH2 domain (PDB code: 1MJS) (21). The initial models were refined and rebuilt using the programs Phenix.refine (53) and Coot (54). The geometries of the final structures were evaluated using the program MolProbity (55). The data collection and refinement statistics of the structures of the SMAD3-FOXH1 and SMAD2-SKI complexes are summarized in table S4.

SPR analysis

The SPR analyses between SMAD2 and the chimeric proteins and between SMAD3 and the chimeric proteins were performed using the Biacore T200 System (GE Healthcare). All SPR measurements were performed at 25°C in 10 mM tris-HCl (pH 9.0), 150 mM NaCl, 50 μM EDTA, 0.05% Tween 20, and 1 mM TCEP. The His–Trx–SMAD2C-2E, His–Trx–SMAD3C-2E, Trx–His–SMAD2C-dC, and Trx–His–SMAD3C-dC proteins (15 nM) were immobilized onto a Series S Sensor Chip NTA (GE Healthcare) (approximately 175 response units). Various concentrations of Trx-FOXH1, Trx-SARA(697–715)-FOXH1, Trx-SARA(682–715)-FOXH1, and Trx-FOXH1-SKI were loaded on the chip. The data were analyzed with the Biacore T200 Evaluation Software (GE Healthcare). To correct baseline drift, we subtracted the signal obtained from the buffer injection (0 concentration of analyte) from each response curve. When we used Trx-SARA(697–715)-FOXH1 and Trx-SARA(682–715)-FOXH1 as analytes, the response curves drifted below the baseline during the dissociation phase. These data suggest that the immobilized His–Trx–SMAD2C-2E, His–Trx–SMAD3C-2E, and Trx–His–SMAD2C-dC were dissociated from the sensor chip by the SMAD-chimera interactions or that the conformations of SMAD2 and SMAD3 were modified by the binding of Trx-SARA(697–715)-FOXH1 and Trx-SARA(682–715)-FOXH1. The dissociation constants (KD) and the corresponding standard errors for Trx-FOXH1, Trx-SARA(697–715)-FOXH1, and Trx-SARA(682–715)-FOXH1 were obtained by curve fitting based on a steady-state affinity model. To reduce the effect of baseline drift, we calculated the responses at each concentration at the position 20 s after starting the injection with a window of 5 s (the binding of the chimeric proteins to His–Trx–SMAD2C-2E and His–Trx–SMAD3C-2E), at the position 10 s after starting the injection with a window of 5 s (the binding of Trx-FOXH1 to Trx–His–SMAD2C-dC and Trx–His–SMAD3C-dC), at the position 50 s after starting the injection with a window of 5 s [the binding of Trx-SARA(697–715)-FOXH1 to Trx–His–SMAD2C-dC], or at the position 4 s before the end of injection with a window of 5 s [the binding of Trx-SARA(697–715)-FOXH1 to Trx–His–SMAD3C-dC, and that of Trx-SARA(682–715)-FOXH1 to Trx–His–SMAD2C-dC and Trx–His–SMAD3C-dC]. The KD values and kinetic parameters (kd1, kd2, ka1, and ka2) for Trx-FOXH1-SKI were obtained by curve fitting based on a two-state reaction model. The constants ka1 and kd1 indicate the association rate constant for analyte binding (M−1 s−1) and the dissociation rate constant for the dissociation of the analyte from the complex (s−1), respectively. The parameters ka2 and kd2 indicate the forward rate constant for the conformational change of the complexes (s−1) and the reverse rate constant for the conformational change of the complexes (s−1), respectively. KD values were calculated as (kd1/ka1) × [(kd2/(kd2 + ka2)].

Oligomeric state analysis by gel filtration chromatography

First, 10 μM His–Trx–SMAD2C-2E (or His–Trx–SMAD3C-2E) and 10 μM chimeric protein [Trx-SARA(682–715)-FOXH1 or Trx-FOXH1-SKI] were mixed in 10 mM tris-HCl (pH 9.0), 150 mM NaCl, 50 μM EDTA, 0.05% Tween 20, and 1 mM TCEP. Next, the samples were loaded onto a Superdex 200 HR 10/30 column (GE Healthcare) and eluted with buffer containing 10 mM tris-HCl (pH 9.0), 150 mM NaCl, 50 μM EDTA, 0.05% Tween 20, and 1 mM TCEP. To estimate the oligomeric state of the complexes, we used the following standard proteins: thyroglobulin (Mr = 669,000), ferritin (Mr = 440,000), aldolase (Mr = 158,000), conalbumin (Mr = 75,000), ovalbumin (Mr = 44,000), and ribonuclease A (Mr = 13,700).

Coimmunoprecipitation assay in vivo

Gene fragments of FOXH1(322–345), SARA(697–715)-FOXH1, and SARA(682–715)-FOXH1 were cloned into the pCAG-IP-FLAG plasmid. HEK293 cells were cultured in Dulbecco’s modified Eagle’s medium (Wako) supplemented with 10% fetal bovine serum (FBS) (Thermo Fisher Scientific) and penicillin/streptomycin (Wako) at 37°C in a humidified atmosphere of 5% CO2. The cells were transiently transfected with indicated combinations of plasmids using PEI-MAX (Polysciences). The cells were lysed in lysis buffer [20 mM tris-HCl (pH 7.4), 150 mM NaCl, 0.5 mM EDTA, 1% Triton X-100, 1 mM Na3VO4, 25 mM β-glycerophospate, 25 mM NaF, and cOmplete (Roche)] 72 hours after transfection and rotated at 4°C for 1 hour. After centrifugation, the lysate was rotated with anti-Myc 9E10 antibody (Santa Cruz Biotechnology) bound to Protein G Sepharose 4 Fast Flow (GE Healthcare) overnight. The beads were washed with lysis buffer three times. The bound proteins were eluted by boiling with sample buffer, resolved by SDS-PAGE, and immunoblotted with the antibodies indicated. Antibodies were used at the following concentrations: Myc 9E10 (1:1000) and HA 3F10–horseradish peroxidase (1:500) (Roche). The membrane was incubated with the antibodies at room temperature for 1 hour. The protein bands were visualized by enhanced chemiluminescence using the SuperSignal West Femto Maximum Sensitivity Substrate (Thermo Fisher Scientific), and the images were captured by a LAS 4000 (GE Healthcare). Densitometric analysis of the blots was performed with ImageJ software.

Luciferase assay

HepG2 cells and C2C12 cells were cultured and transfected as described above using luciferase reporter plasmids, CAGAx12-luc (56) and BRE-luc (57), respectively. The medium was changed to 0.2% FBS–containing medium supplemented with activin (20 ng/ml), TGF-β (1 ng/ml), or BMP4 (20 ng/ml) 24 hours after transfection, and the cells were cultured for 48 hours. These ligands were purchased from R&D Systems. Luciferase assay was performed using the Dual-Luciferase Reporter Assay System (Promega) according to the manufacturer’s instructions. The activities of firefly and renilla luciferases were measured with the GloMax Luminometer (Promega).

Computational analysis

The structures of the SMAD3-FOXH1 and SMAD2-SKI complexes were analyzed using a set of computer programs as follows: PISA (58), for the analysis of the protein interface, surface, and assemblies; Clustal Omega (59), for the amino acid sequence alignment; ESPript (60), for the preparation of alignment figures; LigPlot+ (61), for the depiction of schematic diagrams of intermolecular interactions; and PyMOL (http://pymol.org), for the depiction of structures.

SUPPLEMENTARY MATERIALS

www.sciencesignaling.org/cgi/content/full/11/523/eaao7227/DC1

Fig. S1. Protein constructs.

Fig. S2. Structure of the SMAD3-FOXH1 complex.

Fig. S3. Structure of the SMAD2-SKI complex.

Fig. S4. Superposition of SMAD structures.

Fig. S5. Cooperative binding of FOXH1 and SKI to SMAD2 and SMAD3.

Fig. S6. SMAD2- and SMAD3-binding ability of the chimeric proteins.

Fig. S7. Curve fitting of the SPR data in Fig. 5 and fig. S6.

Fig. S8. His-tag pull-down assay of the chimeric proteins with SMAD4.

Fig. S9. His-tag pull-down competition assay.

Fig. S10. Functional assay in cells.

Table S1. Hydrogen bonds between FOXH1 and SMAD3.

Table S2. Hydrogen bonds between SKI and SMAD2.

Table S3. Putative functions of the hydrophobic patches of SMAD2 and SMAD3.

Table S4. Summary of data collection and refinement statistics of the SMAD3-FOXH1 and SMAD2-SKI complexes.

REFERENCES AND NOTES

Acknowledgments: The synchrotron radiation experiments were performed at beamline AR-NE3A in the Photon Factory (Tsukuba, Japan) (2015G033 and 2016G650). Funding: This work was supported by the Platform for Drug Discovery, Informatics, and Structural Life Science (PDIS) from the Ministry of Education, Culture, Sports, Science and Technology, Japan and by the Japan Society for the Promotion of Science (JSPS) KAKENHI grant numbers 15K14708, 17K19581, and 23228003. Author contributions: M.T. and M.A. conceived and designed the project. K.M. performed the protein expression, purification, crystallization, and structure determination of the SMAD3-FOXH1 complex as well as the pull-down, SPR, and gel filtration assays. S.M. performed the protein expression, purification, crystallization, and structure determination of the SMAD2-SKI complex. T.I. assisted in the protein expression, purification, and crystallization, as well as the pull-down, SPR, and gel filtration assays. A.K. performed the coimmunoprecipitation assay and luciferase reporter assay and supported in protein expression. K.M. and M.T. wrote the manuscript, and M.T. edited the manuscript. Competing interests: The authors declare that they have no competing interests. Data and materials availability: Coordinates and structural factors are deposited in the PDB under the accession numbers 5XOC (SMAD3-FOXH1 complex) and 5XOD (SMAD2-SKI complex).
View Abstract

Navigate This Article