Hot Spots in Tcf4 for the Interaction with β-Catenin*

Cloning, Expression, and Purification of Proteins—The armadillo repeat region (amino acid residues 134–671) in β-catenin was cloned and purified as described previously (). Tcf4-CBD (residues 1–56) was cloned into the NdeI and XhoI restriction sites of pET20b. To quantitate the recombinant Tcf4 protein spectroscopically, a tryptophane residue was introduced between the CBD and the C-terminal His₆ tag. Sitedirected mutagenesis of Tcf4 was performed using the QuikChange kit (Stratagene) and oligonucleotides that were complementary to 15 bp upstream and downstream of the mutation site on both strands. Tcf4-CBD mutants were transformed into Escherichia coli (BL21-DE3). Transformed bacteria were grown to mid-log phase at 37 °Cin2× yeast tryptone medium, and protein expression was induced by addition of 1 mm isopropyl-1-thio-β-d-galactopyranoside. Induced cells were then grown for 3 h and harvested by centrifugation. Cell pellets were resuspended and lysed in PBS (Sigma) using a French press. The recombinant proteins were purified using pre-charged Ni-Sepharose beads (Pharmacia) and a standard protocol (Novagen). The proteins were judged to be pure after affinity chromatography as assessed by SDS-PAGE and matrix-assisted laser desorption ionization time-of-flight mass spectrometry. Tcf4 wild type and mutants were quantitated using an absorbance coefficient at 280 nm of 5.69 × 10³ mol/cm.

Isothermal Titration Calorimetry—Calorimetric measurements were carried out at constant temperatures using a VP-isothermal titration calorimetry titration calorimeter (MicroCal). Samples were extensively dialyzed against PBS, 1 mm DTT. Heats of dilution were measured in blank titrations by injecting the protein into the buffer used in the particular experiment and subtracted from the binding heats. Thermodynamic parameters were determined by non-linear least squares methods using routines included in the Origin software package (MicroCal). Due to the large binding enthalpy of the interaction, a protein concentration in the calorimeter of 2 μm was sufficient to obtain a good signal-to-noise ratio and to reduce “c-values” (c = K_B [P] N; [P] = protein concentration; N = stoichiometry) below 500. Measured thermodynamic values are usually a function of solvent conditions. To indicate this, observed binding enthalpy and entropy changes are abbreviated as ΔHobs and ΔSobs, respectively throughout this publication. The program RIBBON has been used to generate the Figs. 2, 3, 4 ().

In this study we mutated all Tcf4-CBD residues that make significant contacts with β-catenin to alanine and studied the binding of these mutants using isothermal titration calorimetry. The results were analyzed on the basis of the published structures of Tcf-β-catenin complexes. The studied protein domain of Tcf4 contained residues 1–56, which have been shown to be sufficient for high affinity binding to β-catenin (, ). To determine the protein concentration spectroscopically and facilitate purification, we introduced a tryptophane residue and a His₆ tag at the C terminus of Tcf-CBD. These additional C-terminal residues did not alter binding affinities for β-catenin (Table I). In this study we noted that the slightly different solvent conditions used (PBS, 1 mm DTT) result in somewhat smaller binding enthalpies than reported in an earlier study in which a buffer of 50 mm sodium phosphate, pH 7.4, 100 mm NaCl, and 2 mm DTT was used. A representative set of experimental data is shown in Fig. 1.

Interaction of the Tcf4 β-Hairpin and Its Flanking Residues (Residues 2–15)—Most of the β-hairpin formed by the xTcf3 residues 2–15 has been found to be disordered in human Tcf4 structures. The region that was well defined started with Tcf4 residue Leu¹² in the structure determined by Graham et al. (Protein Data Bank accession number 1jdh, referred to henceforth as Tcf4^1jdh) () and residue Gly¹³ in the structure determined by Poy et al. (Protein Data Bank accession number 1jpw, referred to henceforth as Tcf4^1jpw) (). This finding is surprising, considering the high sequence similarity between xTcf3 and Tcf4 (Fig. 2). In xTcf3, the main side chain interactions are formed by Asp¹⁰, which forms a salt bridge with the β-catenin residue Arg⁴⁷⁴, and Glu¹¹, which interacts with His⁵⁷⁸, Arg⁵⁸², Arg⁶¹², and Tyr⁶⁵⁴ on β-catenin (Fig. 2). In the two Tcf4 structures reported, the location of the N terminus is quite different. Whereas in Tcf4^1jdh residues 12–16 superimpose well with the corresponding residues in xTcf3, the N terminus in Tcf4^1jpw does not interact with β-catenin and is oriented toward the solvent. The first 12 N-terminal residues in Tcf4 have been shown to be dispensable in solid phase competition assays (enzyme-linked immunosorbent assay) as well as in reporter gene assays (, ). Indeed, deletion of the first 9 (ND9) or 12 (ND12) residues of Tcf4 had only a marginal effect on binding constants (Table I), whereas binding enthalpies were found to be reduced by about 5 kcal/mol. This suggests that the Tcf4 N terminus interacts with β-catenin but that it does not contribute to binding free energy change due to enthalpy-entropy compensation mechanisms. A similar effect was observed for the mutation D10A, which bound with a ΔΔHobs of +6.5 kcal/mol but with a binding constant similar to that of wild type Tcf4. Interestingly, the mutation D11A also reduced the K_B of the interaction by about 5-fold. Thus, the mutation D11A has a larger effect on the binding of Tcf4 than deletions of the entire N terminus. It has been observed previously that effects of single mutations may be much stronger than the contribution of the entire site as a whole (). Cooperative effects are usually invoked to explain such observations. However, the observed data might also be the result of local changes in protein structure, differences in the dynamics of both proteins, or changes in surface hydration. To confirm this result, we also mutated the residues on β-catenin that interact with Glu¹¹ in xTcf3. The mutations R612A and Y654A did not significantly affect binding of wild type Tcf4 (Table II). Surprisingly, we observed an ∼30-fold decrease in binding constant for wild type Tcf4 for the β-catenin mutations H578A and R582A, suggesting that wild type Tcf4 interacts with this binding site on β-catenin. The Tcf4 residue Asp¹¹ corresponds to a glutamic acid in xTcf3. Possibly, the elongation of the side chain by one CH₂ group at this critical position renders this interaction more favorable. Due to the lack of structural information on the Tcf4 N terminus, we can only speculate about the molecular mechanisms that give rise to this binding behavior. However, the measured data demonstrated that under the chosen experimental conditions, the N-terminal residues of Tcf4 do interact in solution with β-catenin and that this interaction involves the β-catenin residues that are contacted by xTcf3.

Interaction of the Tcf4 Extended Region—This region is structurally very conserved among the three reported Tcf structures and superimposes very well with residues 671–686 of the β-catenin ligand E-cadherin as well as the Tcf antagonist ICAT (, , ). The Tcf extended region binds along a shallow groove with highly positively charged surface patches (charge buttons). Consequently, the electrostatic interactions observed in the β-catenin-Tcf complexes have been discussed as the main interaction sites that drive the binding of Tcf (). Mutagenesis data on β-catenin showed that mutation of the residues Lys⁴³⁵, Asn⁴²⁶, His⁴⁷⁰, Arg⁴⁶⁹, and Lys⁵⁰⁸ to alanine strongly reduces binding of Lef1 and Tcf4 in co-immunoprecipitation experiments (). These residues form polar contacts with Tcf4 residues 16–19. Asp¹⁶ has been identified previously as a critical residue for Tcf binding (, ). The Tcf4 residue Asp¹⁶ forms hydrogen bonds with the β-catenin residues Lys⁴³⁵ and His⁴⁷⁰ as well as several water-mediated hydrogen bond networks (Fig. 2). We find in this study that the residue Asp¹⁶ defines the most important interaction hot spot on Tcf4. The mutation D16A resulted in a ΔΔG of +2.4 kcal/mol. The neighboring residue Glu¹⁷ forms a salt bridge with Lys⁵⁰⁸. However, upon mutation of Glu¹⁷ to alanine, the binding constant decreased only ∼3-fold ().

The interaction with the first charge button is followed in Tcf4 by the hydrophobic residues Leu¹⁸, Ile¹⁹, and Phe²¹, which interact with hydrophobic groups of the β-catenin residues Thr³⁹³, Cys⁴²⁹, Cys⁴⁶⁶, Ala⁴⁶⁷, Pro⁴⁶³, and Arg³⁸⁶ (Fig. 3). Single alanine mutants of Leu¹⁸, Ile¹⁹, and Phe²¹ did not significantly affect Tcf4 binding in isothermal titration calorimetry experiments. For the related transcription factor Lef1, an important contribution of residue Phe²⁴ for β-catenin binding (Phe²¹ in Tcf4) has been discussed using a yeast two-hybrid system. This residue has also been shown to be important for the translocation of endogenous β-catenin into the nucleus (). In addition, the Tcf4 double mutant I19A,F21A has been shown to reduce reporter gene activity by ∼40% in transient transfection experiments (). In light of our binding data, it is likely that mechanisms other than β-catenin binding are responsible for the observed effect of these mutants in these cell-based experiments.

The cluster of hydrophobic Tcf4 residues is followed by negatively charged residues, which form the interaction with the second charge button defined by the β-catenin residue Lys³¹². The three-dimensional structure of this region differs significantly in the two Tcf4 crystal structures published. As in xTcf3, in Tcf4^1jpw the interaction with the second charge button is made by Glu²⁴, whereas in Tcf4^1jdh, a similar interaction is formed by Glu²⁸. In this structure, Glu²⁴ makes no interaction with β-catenin but points to the solvent space. For the mutant E24A, we found a significant difference in binding enthalpy (ΔΔHobs, +7.4 kcal/mol), which is, however, largely offset by favorable entropic contributions. This mutation resulted in a ΔΔGobs of only 0.9 kcal/mol and an ∼4-fold decreased binding constant.

The mutant E28A resulted in only moderate reduction of ΔHobs and in similar binding constants with respect to the wild-type protein. In addition to Glu²⁸, Glu²⁹ also makes polar contacts to the second charge button in Tcf4^1jdh. The observed difference in ΔHobs of 5.8 kcal/mol for mutant E29A supports such an interaction, even though binding constants are not altered. This highly acidic region compensates for single mutations as indicated by the structural plasticity shown by the two structures of Tcf4-β-catenin complexes. This hypothesis is supported by co-immunoprecipitation experiments, demonstrating that the mutation of all 4 glutamate residues in that region (Glu²⁴, Glu²⁶, Glu²⁸, and Glu²⁹) abolishes binding to β-catenin, whereas single mutations had either no effect or only a small effect on binding ().

Mutation N34A only moderately changed ΔHobs and had a negligible effect on the binding constants. This residue is disordered in Tcf4^1jpw and forms a hydrogen bond with Arg³⁴² in Tcf4^1jdh.

Binding of the entire extended region of Tcf4 that includes residues 9–28 was probed using a synthetic peptide (Tcf^9–28). The peptide bound with an affinity of 2.4 ± 0.3 × 10⁶ mol^-1,an enthalpy change of -9.0 ± 0.2 kcal/mol, and an entropy change close to 0 (Table I). These binding data demonstrate that a large portion of the binding free energy change (about 75%) is due to interactions formed by the extended region of Tcf4.

Interaction of the Tcf4 Helical Region—The amphipathic α-helix described for all three Tcf structures packs against a hydrophobic surface area on β-catenin defined by the residues Phe²⁵³, Phe²⁹³, and Ile²⁹⁶ and the aliphatic portions of His²¹⁹, Lys²⁹², and Lys³³⁵ (Fig. 4). The Tcf4 residues Leu⁴¹, Val⁴⁴, and Leu⁴⁸ form the main hydrophobic contacts in this interface. In general, mutation of the hydrophobic residues significantly reduces binding affinities of Tcf4. We observed an ∼15-fold reduction for the mutations L41A and V44A as well as a 5-fold reduction for the mutant L48A (Fig. 5) ().

The Tcf4 residues Asp⁴⁰ and Lys⁴⁵ form polar and electrostatic interactions along the edge of the β-catenin groove. Asp⁴⁰ is positioned between Arg³⁷⁶ and Lys³³⁵ in β-catenin, and the Tcf4 residue Lys⁴⁵ forms a hydrogen bond with the β-catenin residues His²⁶⁰ and Asn²⁶¹. These interactions are also preserved in the xTcf3 structure. The mutation D40A resulted in an ∼3-fold reduction of the Tcf4 binding constant. The thermodynamic reasons for the reduced binding constant were mainly due to a reduction of favorable binding enthalpies. Mutation of the Tcf4 residue Lys⁴⁵ to alanine resulted in slightly larger ΔHobs and in similar binding constants (Fig. 6).

Heat Capacity Changes upon Binding of Tcf4 —The temperature dependence of the binding enthalpy changes (ΔHobs) for the Tcf4 CBD resulted in the determination of a heat capacity change upon binding (ΔC_p) of -1.45 kcal/(mol K), a value very close to the one that has been published using slightly different buffer conditions (Fig. 7) (). In contrast to that very large heat capacity change, a linear fit of binding enthalpy data measured on Tcf4^9–28 resulted in a Δ C_p of only -0.18 kcal/(mol K). Thus, part of the large negative heat capacity change observed upon binding of Tcf4-CBD to β-catenin is a consequence of the formation of the C-terminal helix observed in all Tcf-β-catenin crystal structures.

Binding data of a number of Tcf4 and β-catenin mutants have been reported in the literature using several diverse in vivo and in vitro assay techniques, which makes it difficult to compare the reported experimental data. We used titration calorimetry to generate a comparable set of Tcf4 mutant binding data that describes the effect of single alanine mutations for all Tcf4 residues in the β-catenin-Tcf interface. The single mutations show that Tcf4 binds in a bi-dentate mode: the largest effects of single alanine mutants were observed for residue Asp¹⁶ and for hydrophobic contacts of the C-terminal helix of the CBD (Fig. 5). The strength of a thermodynamic analysis of a binding event is that calorimetric data give a deeper insight into the nature of the interaction than binding constants alone. It has been observed that the altered interaction made by a mutant often does not result in a change in binding constant, even though large differences in binding energetics have been observed. In this study, most mutations of polar residues in the Tcf4-β-catenin interface resulted in a significant change in binding enthalpy, which was compensated for by more favorable entropic contributions to the binding. As a consequence, only minor changes in the free energy of binding and in the resulting binding constants were observed (Fig. 6). This phenomenon of enthalpy-entropy compensation is a general hallmark of the binding energetics of mutated protein-protein interfaces and indicates a large degree of cooperativity in these interactions (). On a molecular level, enthalpy-entropy compensation can be explained by the large energetic cost of side chain dehydration and by the unfavorable entropic terms that result from the reduction of side chain mobility upon formation of a polar interaction. It has recently been shown in a detailed study () on the contribution of electrostatic interactions of single residues to the binding energy in protein-protein interfaces that electrostatics strongly opposed binding in two of the four cases studied. One case had a net effect close to zero, and another case provided a significant contribution to the binding free energy (). Structural features that explained favorable electrostatic interactions in that study were more favorable desolvation terms due to already partially buried charged groups in the free monomer or the formation of partially exposed ion pairs in the complex as well as the formation of salt bridge networks. Because Tcf4 is unstructured in solution, a mechanism that invokes already buried charged groups is not likely. However, the fact that a large number of residues interact in an extended conformation with an interface that consists mainly of charged and polar interactions will favor the formation of networks and partially exposed ion pairs. Indeed, residue Asp¹⁶ forms multiple polar contacts in a network of direct and water-bridged hydrogen bonds.

Upon mutation of a polar side chain to alanine, one binding partner remains hydrated and is not restrained when the complex forms. The thermodynamic consequences of this are that binding enthalpies are reduced because polar interactions have been broken, but favorable entropic terms arise from the gained flexibility of the remaining binding partners. In general, mutations of most polar or charged residues in an interface of two proteins result in complex destabilization. This is the case even if the polar interaction itself is destabilizing. Consequently, mutagenesis data of charged residues reflect primarily the effect of the removal of a charge but do not indicate whether these polar interactions per se stabilize or destabilize protein-protein interactions ().

An interesting aspect of the thermodynamic data generated in this study is that residues that are disordered in crystal structures may have a significant influence on binding thermodynamics. In particular, such contributions were seen for mutations of the N-terminal region that corresponds to a β-hairpin loop in Tcf3 and is disordered in both crystal structures of the Tcf4-β-catenin complex and in the region between the Tcf4 residues Glu²⁴ and Glu²⁹ that is disordered from residue 26 on in Tcf4^1jpw and forms an α-helix in Tcf4^1jdh. Cooperative contributions have been observed for polar residues that are separated by less than 7 Å. Consequently, at least for the Tcf N terminus, such relatively short-range cooperative effects can be ruled out. This indicates that the N terminus of Tcf4 physically interacts with β-catenin and in particular with the residues Arg⁶¹², Tyr⁶⁵⁴, His⁵⁷⁸, and Arg⁵⁸² that were mutated in this study.

Binding of Tcf does not significantly alter the overall conformation of β-catenin compared with its unbound state (, ). Main chain root mean square deviation between β-catenin without ligand and the two published Tcf4-β-catenin complexes were only 0.7 Å, confirming the role of the armadillo repeat region as a rigid structural platform. However, binding to β-catenin induces structural changes in the Tcf ligand, which is mainly unstructured in solution (). Thus, folding events in Tcf4 upon binding to the β-catenin surface additionally modulate the binding energetics of Tcf4 β-catenin complex formation. This study showed that one effect of such folding events is the determined large negative heat capacity increment of -1.45 kcal/(mol K). This effect can be largely assigned to folding events because the extended region of Tcf4 alone (residues 9–28) bound with a heat capacity change of only -0.18 kcal/(mol K).

A large number of thermodynamic studies showed that binding energetics can be calculated on the basis of polar and apolar surface area changes upon complex formation (, ). However, surface area changes calculated on the basis of the xTcf3-β-catenin complex predicted a heat capacity increment of -0.067 kcal/(mol K) (), and the new structural data on Tcf4-β-catenin complexes did not significantly change these calculated values. Large deviations of calculated values and experimental heat capacity changes have been described for many other systems, and possible mechanisms for such discrepancies have been discussed by us and by others (, , , , ). The comparison between experimental heat capacity increments of Tcf4-CBD and Tcf ^9–28 suggests that folding events upon binding are largely responsible for the large negative heat capacity increment.

Implications for Drug Development—Because of the central role of the Tcf/Lef β-catenin interaction in Wnt signaling, the development of a low molecular weight antagonist is an interesting strategy for anti-cancer therapy. In contrast to many active sites in proteins that can be efficiently inhibited by small inhibitors, protein interaction sites usually span much larger surface areas. However, a number of studies have shown that these large interfaces often contain interaction hot spots that may be used as binding sites for much smaller antagonists (, ). In this study, we showed that, at least in vitro and on a single amino acid residue basis, interference with the contact site of the Tcf residue Asp¹⁶ would be the most promising strategy. A hydrophobic binding pocket on β-catenin in the vicinity of Asp¹⁶ that could harbor a small molecule has been described (). However, such an inhibitor would also compete with the β-catenin ligand cadherin, which would lead to the retention of the cadherin in the endoplasmic reticulum and degradation of the protein (). A recent structural study on the Tcf antagonist ICAT showed that even though ICAT, cadherin, and Tcf have partially overlapping binding sites, ICAT competed with Tcf binding but not with cadherin binding in vivo (, ). Possible explanations for this observation could be that cadherin binds much more strongly to β-catenin than Tcf and ICAT or that the binding is regulated in vivo by local concentration differences and/or by post-translational modifications. Recently, it has been reported that cadherin binding is modulated by phosphorylation, which increases its binding constant to β-catenin by 3 orders of magnitude (, ). A large body of structural information on β-catenin complexes has been reported recently (, , , , ). This knowledge will certainly aid the structure-based design of Tcf antagonists. However, due to the large binding constants of Tcf4, the observed structural plasticity, the high degree of cooperativity, and the electrostatic nature of the interaction, the development of efficient antagonists will be challenging.

We thank Jan Malyszko for the synthesis of oligonucleotides and DNA sequencing, Dr. M. Sigvardsson (Lund University) for the His₆-β-catenin expression system, and Dr. A. Schnuchel (Pharmacia Corp.) for providing us with the Tcf4-W-His₆ wild type expression system.