Structural Analysis of an Epidermal Growth Factor/Transforming Growth Factor-α Chimera with Unique ErbB Binding Specificity*

Various chimeras of the ErbB1-specific ligands epidermal growth factor (EGF) and transforming growth factor-α (TGFα) display an enlarged repertoire as activators of ErbB2·ErbB3 heterodimers. Mutational analysis indicated that particularly residues in the N terminus and B-loop region of these ligands are involved in the broadened receptor specificity. In order to understand the receptor specificity of T1E, a chimeric ligand constructed by the introduction of the linear N-terminal region of TGFα into EGF, we determined in this study the solution structure and dynamics of T1E by multidimensional NMR analysis. Subsequently, we studied the structural characteristics of T1E binding to both ErbB1 and ErbB3 by superposition modeling of its structure on the known crystal structures of ErbB3 and liganded ErbB1 complexes. The results show that the overall structure of T1E in solution is very similar to that of native EGF and TGFα but that its N terminus shows an extended structure that is appropriately positioned to form a triple β-sheet with the large antiparallel β-sheet in the B-loop region. This conformational effect of the N terminus together with the large overall flexibility of T1E, as determined by 15N NMR relaxation analysis, may be a facilitative property for its broad receptor specificity. The structural superposition models indicate that hydrophobic and electrostatic interactions of the N terminus and B-loop of T1E are particularly important for its binding to ErbB3.

The ErbB signaling network is composed of the ErbB1 (or EGFR), ErbB2 (HER2/Neu), ErbB3 (HER3), and ErbB4 (HER4) tyrosine kinase receptors. Upon ligand binding, these receptors dimerize into a variety of homodimeric and heterodimeric receptor complexes whereby the intrinsic kinases become activated, which results in a cascade of second messengers and a diversity of subsequent downstream signaling (1,2). ErbB receptors play an important role in growth and differentiation of cells, whereas overexpression of both receptors and ligands has been found in several human cancers (3). As a consequence, the ErbB signaling network is increasingly used as a therapeutic target for the development of anti-tumor drugs (4,5). Recent crystallographic studies have boosted this field of research by providing a wealth of information on the structure of ErbB-ligand complexes (6,7).
A critical step in ErbB receptor signaling is the binding of EGF-like ligands to the extracellular domain of the receptor. More than a dozen soluble ligands have been identified that can be categorized into three distinct groups. A first group, composed of epidermal growth factor (EGF), 1 transforming growth factor-␣ (TGF␣), and amphiregulin, binds specifically to ErbB1. A second group consists of neuregulin (NRG) with its multiple isoforms, which have specific affinity for ErbB3 and ErbB4. A third group binds to both ErbB1 and ErbB4 and is composed of ␤-cellulin, heparin-binding EGF, and epiregulin (8,9). For ErbB2, no soluble ligand has been identified, but it forms the preferred dimerization partner for all other members of the ErbB family.
Despite having distinct receptor binding specificity, all ErbB growth factors have an EGF-like domain as a common motif, which is defined by three disulfide bridges that generate three looped regions, designated the A-, B-, and C-loop, in addition to linear N and C termini. A single hinge residue between the fourth and fifth cysteine divides the EGF-like ligand into an Nand C-terminal half, each with a two-stranded antiparallel ␤-sheet.
For several ligands, including EGF, TGF␣, NRG-1␣, and betacellulin, solution structures have been determined by NMR analysis under a variety of experimental conditions (10 -15). Recently, also crystal structures have been reported of EGF (16), ErbB3 in unliganded form (17), and ErbB1 in complex with EGF (7) or TGF␣ (6). The last two structures established the relative orientation of the four domains in the extracellular part of the receptor and confirmed earlier observations that both EGF and TGF␣ bind a single ErbB1 receptor molecule, which subsequently dimerizes to form a 2:2 complex. Moreover, these studies indicated that specific residues in the B-loop region of EGF and TGF␣ are involved in binding to the first extracellular domain (domain I) of the ErbB1 receptor, whereas several residues in the A-loop, C-loop, and C terminus are in close contact with the third extracellular domain (domain III). Unlike EGF, also residues in the N terminus of TGF␣ directly contact the receptor in domain I.
Site-directed mutagenesis and phage display studies have * 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.
The atomic coordinates and structure factors ( identified amino acids that have only limited freedom of mutation and are thought to be involved directly in receptor binding (18). For EGF and TGF␣ binding to ErbB1, these highly conserved residues are located around the second and sixth cysteine and in the C terminus (19,20). In contrast, for NRG binding to ErbB3, hydrophobic and charged residues located in the N terminus and B-loop appear of primary importance (21)(22)(23). Supposing that all EGF-like ligands bind their respective ErbB receptor in a similar orientation, it appears that domain I of ErbB3 contributes mostly to ligand specificity, whereas for ErbB1 this is located in domain III (24,25). In order to study ligand binding specificity, we have previously used a domain exchange strategy between EGF and TGF␣ and shown that replacing the N-terminal linear region of EGF with that from TGF␣ results in a chimera, designated here as T1E, with unique ErbB binding specificity. T1E not only maintains high affinity for ErbB1, the natural receptor for both EGF and TGF␣, but has in addition gained the ability to bind to ErbB2⅐ErbB3 heterodimers with nearly identical affinity as NRG-1 (26). Subsequent mutation analysis showed that the TGF␣ residues His 4 and Phe 5 in the N terminus and the EGF residue Leu 28 at the tip of the B-loop are essential for the high affinity of T1E for ErbB2⅐ErbB3 heterodimers. Replacement of Leu 28 for the Glu residue present at the equivalent position in TGF␣ strongly impaired binding to ErbB2⅐ErbB3 heterodimers without affecting binding to ErbB1 (26). These results indicated that the combination of residues in the N terminus of TGF␣ and in the B-loop region of EGF mediate the enhanced binding affinity for ErbB2⅐ErbB3 heterodimers.
High affinity binding requires a proper three-dimensional structure of the ligand in combination with specific residues in the receptor binding domain. In order to understand the altered receptor specificity of T1E compared with EGF and TGF␣, we have studied structure and dynamics of T1E by twoand three-dimensional heteronuclear NMR techniques. Subsequently, we modeled the interaction of T1E with both the ErbB1 and ErbB3 receptors, on the basis of the recently published crystal structures of the extracellular regions of ErbB3 (17) and the EGF⅐ErbB1 and TGF␣⅐ErbB1 complexes (6,7). Using these models, the specific roles of residues in the N terminus and B-loop of T1E in ErbB2⅐ErbB3 interaction were analyzed.

EXPERIMENTAL PROCEDURES
Construction of T1E Expression Vector-To produce sufficient amounts of T1E for NMR analysis, we used the Pichia pastoris protein expression system (Invitrogen). Thereto, T1E was cloned into the P. pastoris expression vector pPICZ␣A. A gene construct encoding the human TGF␣/EGF chimera T1E, previously described by Stortelers et al. (26), was used as a template for the amplification of the T1E fragment by PCR. The primers used for PCR introduced an XhoI and a SalI restriction site permitting directional cloning of the amplified DNA in frame with the ␣-factor leader sequence in the pPICZ␣A expression vector. A stop codon preceding the SalI site prevented expression of the His tag. The amplified fragment was first subcloned into pCR2.1 TOPO® cloning vector (Invitrogen). The recombinant TOPO vector was digested with XhoI-SalI to generate a 189-bp fragment, which was subsequently introduced into the pPICZ␣A expression vector using its XhoI-SalI sites. The correctness of the constructs was confirmed by DNA sequencing. Approximately 7 g of the DNA construct were linearized with BstXI prior to transformation of P. pastoris X33 (Mutϩ) and KM71H (MutS) cells. The electroporation method of the EasyS-elect™ Pichia expression kit (version F; Invitrogen) was used for transformation, and the transformants were plated on MD and MM agar plates to screen for methanol utilization (Mut) phenotype. Several Mutϩ and MutS clones were put on plates with very high zeocin concentrations (up to 1 mg/ml) to select for the so-called "jackpot" clones with multiple integrations. Southern blotting was used to confirm the presence of multiple integrations in the genome of selected high zeocinresistant colonies. In small scale expression screens, the expression levels showed to be proportional to the number of integrations. A Mutϩ clone with multiple integrations was selected for large scale expression and 15 N labeling.
Expression of Recombinant 15 N-labeled T1E in P. pastoris-To produce 15 N-labeled T1E, we slightly modified the protocol of Wiles et al. (27). In brief, 25 ml of BMG culture medium (0.1 M potassium phosphate (pH 6.0), 0.34% (w/v) yeast nitrogen base without amino acids or ammonium sulfate, 1% (v/v) glycerol, 0.00004% (w/v) biotin, and 1% (w/v) ammonium sulfate) were inoculated with a fresh colony of the selected Mutϩ clone and grown overnight while shaking at 30°C to midlogarithmic phase. The culture was scaled up to 500-ml batches of BMG medium without ammonium sulfate but containing 0.1% (w/v) 15 Nlabeled ammonium chloride as its sole nitrogen source (99% 15 N; Campro Scientific, Veenendaal, The Netherlands) in 2-liter baffled flasks. After overnight culture, the cells were transferred to 500 ml of BMM induction medium (0.1 M potassium phosphate (pH 6.0), 0.34% (w/v) yeast nitrogen base without amino acids or ammonium sulfate, 0.5% (v/v) methanol, 0.00004% (w/v) biotin, and 1% (w/v) 15 N-labeled ammonium chloride) and grown for 9 days with additional supplies of methanol every 24 h. The minimal amount of nitrogen source needed for optimal protein production was determined to be 0.1% in the growth medium and 1% in the induction medium.
Protein Purification-The P. pastoris culture supernatant was subjected to cation exchange chromatography (SP-Toyopearl 550C, Tosoh Corp., Tokyo, Japan) and eluted with a linear salt gradient (0.2-1.0 M NaCl in 0.05 M NaAc). The fractions were tested for binding affinity to ErbB1 by an 125 I-mouse EGF binding competition assay as previously described by van de Poll et al. (28) and analyzed in parallel by nonreducing SDS-PAGE and Western blot analysis using a polyclonal antibody (Ab-3) raised against recombinant wild type human EGF (Oncogene Science Inc., Cambridge, MA). Specific antibody binding was detected by a goat anti-rabbit antibody linked to horseradish peroxidase and visualized by enhanced chemiluminescence. Fractions positive for ErbB1 binding and displaying only a single band on Western blot were collected, dialyzed against 0.2 M HAc, and lyophilized. After resolubilization in 0.5 M HAc, the protein was finally purified by reverse phase high pressure liquid chromatography on a Deltapak C 18 column (Waters Associates, Milford, CT). Elution was carried out using a linear gradient of 20 -40% acetonitrile in 0.1% trifluoroacetate. Using matrixassisted laser desorption-ionization time-of-flight mass spectrometry, the purified protein was identified as 15 N-labeled T1E-(1-54) lacking the C-terminal arginine. The fractions corresponding to the main peak were collected, and the volume was reduced using speedvac centrifugation and subsequently dialyzed against double distilled water to remove remaining traces of organic solvents and finally against 50 mM phosphate buffer, pH 6.3. We thus obtained ϳ13 mg of uniformly 15 N-labeled T1E/liter of culture medium. The labeled material was biologically indistinguishable from the E. coli-derived T1E described previously (26).
Data Analysis and Structure Determination-The NMR sample was prepared to contain 0.8 mM protein in 95% H 2 O, 5% D 2 O (v/v), in 50 mM sodium phosphate buffer (pH 6.3) with 20 g/ml Pefabloc (Roche Applied Science). NMR spectra were recorded at 298 K on Varian Unity Inova spectrometers operating at 800-and 600-MHz 1 H resonance frequencies. The following multidimensional experiments were recorded and analyzed: 15 N HSQC, HNHA, HNHB, two-dimensional TOCSY, two-dimensional NOESY, 15 N TOCSY, 15 N NOESY, and 15 Nfiltered TOCSY (for both the aromatic and the aliphatic region). All spectra were processed on a Silicon Graphics work station using NMRPipe software (29). Backbone and side chain resonances were assigned using XEASY (30). Nuclear Overhauser effect (NOE) peaks from the two-and three-dimensional NOESY spectra were classified either as very weak, weak, medium, or strong with an upper distance restraint of 7.0, 5.0, 3.5, and 2.8 Å, respectively. Structure calculations were performed using the program XPLOR 3.851 (31). Dihedral angle restraints were derived from 3 J(H N H)-coupling constants, which were measured by a three-dimensional HNHA experiment (32). Subsequently, additional and angle restraints were predicted using TA-LOS (33), resulting in a total of 98 dihedral angle restraints. To improve local geometry and electrostatics, the lowest energy structures were refined in water using a restrained molecular dynamics protocol under a CHARMM22 force field (34). gray and blue (␤-sheet)-colored residues are EGF-based. b, sequence alignment of T1E with TGF␣, EGF, and NRG-1␤. The arrows depict residues that mediate enhanced binding affinity for ErbB2⅐ErbB3 heterodimers. For clarity, the -loop of NRG-1␤, which is known to be dispensable for EGF⅐ErbB1 and TGF␣⅐ErbB1 complexes were taken from Protein Data Bank entries 1M6B (17), 1IVO (7), and 1MOX (6), respectively.
Relaxation Measurements-15 N longitudinal relaxation rate R 1 , the 15 N transverse relaxation rate R 1 , and the steady-state { 1 H}-15 N NOE were measured at 298 K at 14.1 tesla field strength. R 1 relaxation rates were determined from a series of six interleaved spectra recorded with relaxation delays of 0.016, 0.128, 0.256, 0.512, 0.768, and 1.024 s. For the determination of R 1 relaxation rates, six spectra were recorded in an interleaved manner with relaxation delays of 0.016, 0.032, 0.048, 0.064, 0.112, and 0.128 s. An overall rotational correlation time ( c ) was first estimated from the mean ratio R 1 /R 1 and optimized to 5.4 ns using a grid search routine of the Modelfree program. This value is higher than expected for a monomeric 54-amino acid protein at 298 K and probably results from (partial) aggregation, since at concentrations above 0.8 mM the sample started to precipitate. Using the optimized c value, the backbone dynamics were further analyzed on a per residue basis by Lipari-Szabo analysis (38,39) using the program Modelfree 4.01 (40). We analyzed the generalized order parameter (S 2 ), which is correlated with backbone rigidity, the internal correlation time ( e ), which is the time scale of internal motion (picosecond to nanosecond), and the chemical exchange contribution R ex describing motions on millisecond to microsecond time scale.
Homology Modeling and Structural Superposition Modeling-To compare binding of T1E to the extracellular domains of ErbB1 and ErbB3, we built structural superposition models of T1E⅐ErbB1 domain I and T1E⅐ErbB3 domain I. Structures of T1E and ErbB1 domain I were available (1P9J (this paper) and 1MOX), but the crystal structure of ErbB3 (1M6B) published by Cho and Leahy (17) lacks the N-terminal 35 residues. To compensate for this lack of structural information, we first built a homology model of this region (residues 5-35) of ErbB3 by assuming that it has a similar structure as the corresponding region in ErbB1. In order to do so, sequences of ErbB1 domain I and ErbB3 domain I were aligned using ClustalW (41), which showed that this region of these two receptors has a sequence identity of 29.0% and a sequence similarity of 67.7%. The crystal structure of domain I of ErbB1 (1MOX) was subsequently used as a template structure for homology modeling with the program WHAT-IF (37).

RESULTS
Structure Calculation and Quality-Using a series of homonuclear and heteronuclear NMR experiments, 1 H and 15 N resonance assignments were obtained for T1E using the Wü thrich sequential assignment protocol (42). The proton resonance assignments were mainly derived from two-dimensional TOCSY and two-dimensional NOESY spectra, whereas a three-dimensional 15 N NOESY spectrum was used to identify the sequential 15 N resonances of backbone amide groups and to verify the obtained proton resonance assignments.
From the NOE spectra, a set of 660 NOEs was collected, composed of 309 intraresidual, 194 sequential, 67 medium range, and 90 long range NOEs. This resulted in an average of 12.2 distance restraints per residue. Additional constraints included nine hydrogen bonding restraints, three constraints from the disulfide bridges, and 98 and angle restraints. We calculated an ensemble of 36 high resolution NMR structures of T1E. A superposition of this ensemble in two different orientations is shown in stereo view in Fig. 1a. The structural statistics are listed in Table I. All accepted structures contained neither distance violations greater than 0.5 Å nor angle violations greater than 5°from experimental data. Analysis of the Ramachandran plot showed that for the 36-structure ensemble 82.9% of the residues were found in the most favored regions, 16.4% in the additionally allowed regions, and 0.7% in generously allowed regions. No residues were found in the disallowed regions of the Ramachandran plot.
The data presented in Fig. 1a show that the protein contains two major stretches of secondary structure, consistent with the EGF fold. A large anti-parallel ␤-sheet is formed by residues Val 21 -Ile 25 (␤ 1 ) and Lys 30 -Asn 34 (␤ 2 ), and a smaller anti-par-ErbB3 binding, is not shown. Color coding is as in a. c, T1E structure ensemble superimposed for B-loop residues (pairwise r.m.s. deviation 1.27 Ϯ 0.4 Å) and C-loop residues (pairwise r.m.s. deviation 0.91 Ϯ 0.29 Å), respectively. d, ribbon presentation of the T1E structure color-coded according to residue-specific S 2 values. Residues with S 2 ϭ 0 -0.7 are displayed in pink, S 2 ϭ 0.7-0.8 in violet, and S 2 ϭ 0.8 -1.0 in light blue. No data are available for residues colored gray. The A-, B-, and C-loop as well as N and C termini are indicated, and the cysteine bridges are depicted by gray lines. e, ribbon presentation of the T1E structure color-coded according to residue-specific R ex values. Residues requiring a chemical exchange contribution are shown in green. f, superposition of the lowest energy structure of T1E onto EGF and TGF␣ in native and complexed forms. T1E is shown in green, native EGF in light blue (molecule A from 1JL9), native TGF␣ in orange (1YUG, mean structure), EGF complexed to ErbB1 in blue (from 1IVO.pdb), and TGF␣ complexed to ErbB1 in red (from 1MOX.pdb). Notably, the N terminus adopts an extended conformation that approximately follows the ␤ 1 -strand, although it is much less ordered than the large ␤-sheet in the B-loop itself (Fig. 1a). Its conformation is determined, however, by several unambiguous, long range NOEs between Val 1 and Leu 28 , Val 2 and Leu 28 , Ser 3 and Leu 28 , Phe 5 and Met 23 , and Asn 6 and Tyr 24 . In 4 of 36 structures in the ensemble, this resulted in a triple ␤-sheet formation of the N terminus with the antiparallel ␤-sheet in the B-loop, based upon criteria used by PROCHECK-NMR.
Relaxation and Dynamics Parameters-Despite the high degree of structural similarity, most members of the EGF family of polypeptide growth factors show marked differences in receptor binding specificities. Since ligand binding may be influenced by intrinsic dynamical properties of the protein, we characterized the backbone dynamics of T1E.  The nice dispersion in the 15 N HSQC spectra allowed the measurement of { 1 H}-15 N NOE values for all but one nonproline residue. Data for His 12 did not yield reliable fits in the relaxation analysis. As shown in Fig. 2 To further analyze the backbone internal motions, the Modelfree approach was used (38,39). Backbone amide order parameters S 2 were determined for 46 residues (Fig. 2). These S 2 values were grouped into three classes, which are visualized in Fig. 1d using a color-coding scheme. The local flexibility of the N-and C-terminal ends is reflected by S 2 values of 0.2-0.7 and e values of Ͼ500 ps. Interestingly, the residues in the N and C termini, as well as His 18 , Gly 20 , Ala 27 , Leu 28 , Gly 41 , and Trp 51 , could only be fitted using a model incorporating an intermediate time scale motional parameter, although for the nonterminal residues the fit yielded relatively large error bars on the accompanying e values. In Fig. 1e, a ribbon presentation of the lowest energy T1E structure is shown with residues that required a contribution of the transverse relaxation rate R ex in green. Non-zero values of R ex were found for several residues mainly in the A-loop (residues Ser 11 -Cys 16 ) and to a lesser extent also in the B-loop and C-loop region, which is indicative for chemical exchange processes within the millisecond to microsecond time range (Fig. 1e). Overall, the structure of T1E is relatively flexible throughout the entire amino acid sequence, with enhanced backbone flexibility and internal motions in the N and C termini and chemical exchange contributions in A-, B-, and C-loop regions. Structural Superposition of T1E with Related Structures-Since the chimera T1E is composed of parts of the EGF and TGF␣ sequences, we compared the T1E structure with the parental molecules by superposition. We superimposed the lowest energy T1E structure onto native EGF (molecule A from crystal dimer 1JL9) and native TGF␣ (mean NMR structure 1YUG) as well as onto crystal structures of EGF and TGF␣ in the receptor-bound state, as obtained from Protein Data Bank entries 1IVO and 1MOX, respectively (Fig. 1f) A Comparison of T1E Binding to ErbB1 Domain I and ErbB3 Domain I-T1E is unique in its properties since it can bind both to ErbB1 and ErbB3. Since the amino acid requirements for ligand binding to ErbB1 and ErbB3 are very much different, the three-dimensional structure of T1E provides the opportunity to compare the structural requirements for binding of a single ligand to two distinct receptor molecules. Domain exchange and phage display studies of T1E (26,43) indicated that residues His 4 and Phe 5 in the N terminus and Leu 28 in the B-loop region of T1E determine its affinity for ErbB2⅐ErbB3 binding, strongly suggesting a direct role for the involvement of receptor domain I. This is further substantiated by the observation that these three residues (see the arrows in Fig. 1b) are functionally conserved in NRG-1␤, the natural ErbB2⅐ErbB3 agonist. Table II lists the amino acids in EGF and TGF␣ that directly interact in the crystal structures with specific residues in domain I of ErbB1 (6,7) and also surveys the corresponding residues in domain I of ErbB3. Most residues in ErbB1 involved in ligand binding are conserved if not identical in ErbB3. Examples are Tyr 101 , which is known to interact with His 4 in the free N terminus of TGF␣, and Glu 90 which is responsible for the formation of a salt bridge with Lys 28 in the ␤ 2 -strand of the B-loop of EGF (Lys 29 in TGF␣). Furthermore, Tyr 45 and three Leu residues in ErbB1 that interact with B-loop residues of EGF have a conserved hydrophobic character. In contrast, Asn 128 and Arg 125 , two residues in ErbB1 that interact with residues at the tip of the B-loop of TGF␣, are clearly different in ErbB3, comprising residues Lys 132 and Tyr 129 , respectively.
To visualize possible critical interactions in ligand-receptor binding, we generated structural models by superimposing the lowest energy T1E structure onto the domain I regions of both ErbB1 and ErbB3 receptor. Focus was laid on the ligandbinding interface of domain I, since this domain is most probably involved in binding the N-terminal and B-loop region of T1E that confer specificity toward ErbB2⅐ErbB3 complexes (26). The structural superposition modeling was composed of a three-step procedure. We first superimposed T1E onto complexed TGF␣. The structure of complexed TGF␣ was preferred over that of complexed EGF, because only the N terminus of TGF␣ is known to be involved in receptor binding (6). In the second step, the complete domain I of ErbB3 (with the homology-modeled N-terminal region; see "Experimental Procedures") was superimposed on ErbB1 domain I obtained from the crystal structure of TGF␣⅐ErbB1. Finally, an overlay of these two sets was made based on the orientation of complexed TGF␣ in both superpositions, thereby assuming that T1E binds ErbB1 in a similar manner as TGF␣ does and that binding to ErbB3 domain I is similar to binding to ErbB1 domain I. Fig. 3a compares the hydrophobic character of the receptor surface residues in these structural superposition models of T1E⅐ErbB1 domain I and T1E⅐ErbB3 domain I. Of particular interest are the residues juxtaposed to the crucial N-terminal T1E residues His 4 and Phe 5 . In the T1E⅐ErbB1 complex, His 4 is in close proximity to Ser 99 and Tyr 101 , allowing for stacking interaction with the latter. This interaction appears to be preserved in the T1E⅐ErbB3 complex, where the homologous Tyr 104 replaces Tyr 101 , whereas Leu 102 occupies the position of Ser 99 . Similarly as in TGF␣, T1E-Phe 5 is oriented toward Leu 69 and Tyr 45 in the T1E⅐ErbB1 complex. This residue faces a much more hydrophobic environment in the T1E⅐ErbB3 complex, formed by Met 72 and Leu 48 . Furthermore, ErbB1 contains a large Phe residue at position 20. In ErbB3, two small residues, Gly 50 and Ala 23 , are located in the corresponding region, making the hydrophobic cavity in ErbB3 larger and better available for large hydrophobic N-terminal ligand residues. These differences in hydrophobicity between ErbB1 domain I and ErbB3 domain I strongly suggest a role for specific hydrophobic contacts in ligand-ErbB receptor binding selectivity.
In addition to specific hydrophobic contacts, we compared the contribution of electrostatic interactions in receptor binding specificity. Fig. 3b displays the electrostatic potentials for the ErbB1 domain I and ErbB3 domain I complexed with T1E. The ligand binding interface of the ErbB3 domain I shows a larger surface with electronegative potential than the corresponding surface of ErbB1 domain I. Residues jointly responsible for this electronegative patch on ErbB3 domain I are Glu 45 , Asp 93 , and Glu 131 . Furthermore, the positive electrostatic potential resulting from Arg 125 and Lys 13 in ErbB1 are absent in ErbB3. In Fig. 3c, the receptor binding interface of the ligands is shown. T1E, EGF, and TGF␣ are displayed as a surface representation with electrostatic potentials colored as in Fig. 3b, together with the backbone trace of ErbB1 domain I. The models in Fig. 3c  FIG. 3. Comparison of the T1E⅐ErbB1 domain I and T1E⅐ErbB3 domain I complexes obtained by structural superposition. a, hydrophobic interactions of T1E N terminus with domain I of ErbB1 and ErbB3. The backbone trace of T1E is colored green with selected side chains His 4 and Phe 5 shown in ball-and-stick representation (blue). Domain I of both receptors is shown in a space-filling representation and colored according to hydrophobicity (52), ranging from yellow (less hydrophobic) to red (more hydrophobic). b, electrostatic potential plots of ligand-binding interfaces of ErbB1 domain I and ErbB3 domain I in surface representation. Relative orientation of T1E is shown by the green backbone trace. Backbones of residues His 4 , Phe 5 , and Leu 28 are shown in yellow. Backbone traces of acidic and basic residues of T1E are colored are rotated 180°around the vertical axis with respect to Fig. 3b. In the TGF␣⅐ErbB1 complex, Glu 27 at the B-loop tip of TGF␣ interacts with Arg 125 of ErbB1 (6). In comparison, this negatively charged Glu 27 of TGF␣ would probably be located in an unfavorable negatively charged environment on ErbB3, in particular close to Glu 131 . In line with these considerations, previous work showed that in EGF/TGF␣ chimeras, Glu 27 prevents interaction with ErbB2⅐ErbB3 complexes, whereas the corresponding Leu in EGF is favorable to this interaction (26). Furthermore, a comparison of the electrostatic potentials of the three ligands (Fig. 3c) reveals that Arg 22 in TGF␣ is responsible for a large positively charged patch, whereas EGF is more negatively charged on this side of the molecule as a result of the presence of Glu 5 and Glu 24 in this region. T1E lacks Glu 5 , but Glu 26 at the end of the ␤ 1 -strand is also responsible for a small negative patch on T1E. Together, these structural superposition models show large differences in electrostatic potentials of both ErbB receptors and EGF-like ligands, indicating that electrostatic attraction and repulsion play a major role in ErbB receptor specificity of ligand binding. DISCUSSION EGF, TGF␣, and the chimera T1E all bind to the ErbB1 receptor with high affinity. Of these three ligands, T1E is unique because it is also able to interact with ErbB2⅐ErbB3 heterodimers (26). Ligand binding results from a proper conformation in combination with the presence of specific residues at positions that directly interact with the receptor. In the present study, we have determined the solution structure and quantified the backbone dynamics of T1E. To explain the binding selectivity of T1E, we have constructed models of T1E⅐ErbB1 domain I and T1E⅐ErbB3 domain I, using structural superposition. The present data indicate that the additional affinity of T1E for ErbB2⅐ErbB3 heterodimers most probably stems from the interplay of three different factors: (i) the conformational effect of the N terminus, (ii) the dynamic properties of T1E, and (iii) the presence of specific residues that form part of the binding epitope for domain I of ErbB3. The structural superposition models of T1E⅐ErbB1 domain I and T1E⅐ErbB3 domain I strongly suggest that both hydrophobic interactions and electrostatic attraction/repulsion may be prominent aspects in determining ErbB ligand binding selectivity. In addition, our data provide structural support for the observation that His 4 and Phe 5 in the N terminus and Leu 28 in the B-loop region are critical residues in mediating ErbB3 binding affinity.
The overall structural fold of T1E in solution is very similar to the solution structures of other members of the EGF family, such as EGF and TGF␣ (10,14): an N-terminal half with a flexible loop region (A-loop) and a large antiparallel ␤-sheet (forming the B-loop) and a C-terminal half that contains a minor antiparallel ␤-sheet (forming the C-loop). The N-and the C-terminal halves of EGF-like ligands are connected by a hinge residue, Asn 34 in the case of T1E. In comparison with EGF and TGF␣, the N terminus of T1E displays ,-angles consistent with an extended conformation. In 4 of 36 structures in the ensemble, the N terminus explicitly forms a triple ␤-sheet with the B-loop. The observation that NRG-1␣, which is a natural ligand for the ErbB2⅐ErbB3 heterodimer, also contains a triple ␤-sheet in this structurally homogeneous region (13) suggests that stabilization of the N-terminal region into an extended structure could be a requirement for EGF-like growth factors to bind ErbB3. Although the N terminus shows enhanced backbone flexibility, receptor binding of suitably folded T1E may drive the equilibrium of T1E into the receptor-bound, triple ␤-sheet-containing structure.
Whereas the secondary structure and global fold of EGF-like structures are quite similar, the overall r.m.s. deviations of the superpositions are relatively high (Fig. 1f). This may represent real differences between these structures or, alternatively, suggest that some of the structures are less well defined, either too loose or too tight. In our study, we used a CHARMM22 water refinement protocol (34) to improve the quality of the structure ensemble. This resulted in reliable Z-scores for the local geometry (Table I). Furthermore, the Ramachandran Z-score, which is a good indicator of the overall quality of the ensembles, yielded highly acceptable values for NMR structure ensembles. Hence, we consider our structure to be reliable in terms of a representation of our experimental data.
Although uncertainty in NMR data does not necessarily indicate motion between the N-and C-terminal halves, the presence of hinge-bending motions has previously been suggested for ErbB ligands (14, 44 -47). The present observation that the B-and C-loop regions of the T1E ensemble superimpose nicely (see Table I) suggests that such motions could indeed be at play. In addition, many residues in T1E (Asp 13 , Gly 14 , Tyr 15 , Cys 16 , Val 21 , Glu 42 , Arg 43 , Gln 45 , and Tyr 46 ) at the interface between the N-and C-terminal halves exhibit nitrogen-15 exchange line broadening (Fig. 1e), similarly as observed previously for TGF␣, which is consistent with the existence of hingebending motions. Furthermore, superposition of TGF␣ structures in bound and native state revealed a change in relative orientation of the N-and C-terminal halves, which could indicate that bending of the hinge is a prerequisite for proper receptor binding (6).
Dynamics may play a significant role in broad receptor binding specificity of T1E. From the backbone relaxation data of T1E and the subsequent Modelfree analysis, it appears that T1E is a relatively flexible molecule. The protein exhibits significant backbone motions in the picosecond to nanosecond time scale, and ϳ40% of the residues require a chemical exchange contribution. The residues with the largest R ex values are located in the A-loop (Figs. 1e and 2). Furthermore, the data show R ex values for groups of residues that are separated in sequence but close together in space (Asp 7 and Cys 8 (in the N terminus); Met 23 , Tyr 24 , and Ile 25 (in the ␤ 1 -strand); and Tyr 31 and Ala 32 (in the ␤ 2 -strand)) that may be indicative of synchronous motion of these domains. It has been proposed that flexibility may correlate with broadened receptor-ligand binding specificity (48), and the observed flexibility of T1E may thus be a facilitative property for its broad receptor specificity. T1E can adopt multiple conformations, thus facilitating binding to domains I and III of both ErbB1 and ErbB3 receptors by an induced fit mechanism. In this view, the conformation of T1E in which the N terminus and the B-loop form a triple ␤-sheet could particularly be favorable for binding to domain I of ErbB3. If only 10% of the total available T1E is in the correct conformation to bind ErbB3, as indicated by the percentage of structures showing a triple ␤-sheet, this could explain the red and blue, respectively. Residues potentially involved in specific ligand-receptor interactions are indicated by number (T1E residues in green). Dark red indicates electronegative potential, dark blue indicates positive electrostatic potential, and white indicates a potential of zero. Electrostatic potentials were calculated using the program MOLMOL (35). c, electrostatic potential plots of receptor-binding interfaces of T1E, EGF, and TGF␣ in surface representation. Relative orientation of the ErbB1 domain I is shown as a light blue backbone trace. For the T1E⅐ErbB1 complex, the structural superposition model is shown, whereas for the EGF⅐ErbB1 and TGF␣⅐ErbB1 complex, the crystal structures with Protein Data Bank entry 1IVO (7) and 1MOX (6) were used, respectively. The orientation of the models is rotated 180°around the vertical axis with respect to b. Electrostatic potential color coding is as in b.
relatively low affinity of T1E for ErbB3 receptors alone and the need for stabilization by ErbB2 (26). A role for dynamics in the ligand-receptor recognition process has previously also been suggested for NRG-1␣ (49), where flexibility was observed in regions that are clearly important for receptor binding, such as both the N-and C-terminal regions (23). Our relaxation data, together with those described for other ErbB ligands, indeed suggest that dynamic flexibility contributes to broad specificity of the ErbB receptor recognition process.
To elucidate the role of specific residues in N terminus and B-loop of T1E for ErbB3 receptor binding, we made structural superposition models of T1E⅐ErbB1 domain I and T1E⅐ErbB3 domain I. From these models, we propose that the N terminus of T1E is involved in binding to ErbB3 by hydrophobic packing. His 4 and Phe 5 , two residues in the N terminus of T1E that were earlier found to be important for ErbB2⅐ErbB3 binding specificity (26), are in proximity to a hydrophobic pocket in ErbB3 (Fig. 3a). The observation that this pocket is more hydrophobic and larger than the corresponding region in ErbB1 suggests that it favors binding of large hydrophobic residues such as His 4 and Phe 5 . Moreover, the preformed extended conformation of the N terminus in T1E could facilitate surface presentation of His 4 and Phe 5 , enhancing the formation of the required intermolecular contacts. In addition, these large hydrophobic residues in the N terminus of T1E could also be involved in compensating for less favorable acidic residues in the B-loop region. Unfavorable electrostatic interactions between Glu 26 in T1E and Glu 74 in ErbB3 could potentially be masked by His 4 and Phe 5 in the N terminus. This hypothesis is supported by phage display studies, which indicated that for optimal ErbB2⅐ErbB3 binding EGF should preferentially contain an aromatic, a hydrophobic, and a positively charged amino acid in its linear N terminus (43). The observation that a positively charged residue is favored in one of the positions in the N terminus suggests that masking the acidic Glu residue present at the end of the ␤ 1 -strand (corresponding to Glu 26 of T1E) is required for proper binding to the ErbB3 domain I.
Several studies with mutant growth factors have suggested that acidic residues in the B-loop region are involved in ErbB binding selectivity. Chimeras that contain the B-loop region of TGF␣ are unable to activate ErbB2⅐ErbB3 heterodimers (26,50), particularly because the acidic Glu 27 at the tip of the TGF␣ B-loop is a negative determinant for interaction with ErbB2⅐ErbB3 heterodimers (26). In the present T1E⅐ErbB3 domain I model, Leu 28 is positioned at the tip of the B-loop of T1E pointing toward a large electronegative patch on the ErbB3 ligand binding site, whereas the corresponding region in ErbB1 is predominantly positively charged. In the TGF␣⅐ErbB1 receptor complex, the corresponding Glu 27 is involved in an electrostatic interaction with Arg 125 , thereby stabilizing the ligand-receptor complex (6). In contrast, the equivalent region in ErbB3 does not contain positively charged residues that could stabilize the interaction, as shown in Fig.  3b. We therefore argue that the large electronegative patch, in particular Glu 131 in domain I of ErbB3, is responsible for electrostatic repulsion of Glu 27 at the tip of the B-loop of TGF␣, thereby inhibiting this interaction.
In the present study, we determined the solution structure and dynamics of T1E and analyzed its unique ErbB receptor binding specificity by using structural superposition models of T1E⅐ErbB1 domain I and T1E⅐ErbB3 domain I. Our results provide new insights into the subtle mechanism of ligand binding specificity of the ErbB receptor family. An attractive goal is to combine residues specific for ErbB1 and ErbB3 binding into one chimeric EGF-like molecule. Recent studies from our lab-oratory have shown that the affinity of T1E for ErbB3 binding can be enhanced significantly by mutations in both the N-and C-terminal regions (43,51), without loss of ErbB1-specific affinity. By using a chimera with a combination of these mutations, a direct comparison of ErbB1 and ErbB3 binding mechanisms can be studied on the basis of a single ligand.