The significance of valine 33 as a ligand-specific epitope of transforming growth factor alpha.

Although binding of epidermal growth factor (EGF) and transforming growth factor alpha (TGFalpha) to the EGF receptor (EGFR) is mutually competitive, their binding is not identical, and their biological activities are not always equivalent. To probe for ligand-specific interactions, we have synthesized analogues of TGFalpha with modifications to the residue lying between the fourth and fifth cysteines (the "hinge"). Although this residue lies in a structurally conserved region of the protein, it is not conserved within the EGFR ligand family. Our results show that in TGFalpha there is a preference for a bulky hydrophobic hinge residue; this contrasts with EGF, for which a hydrogen bond donor functionality is preferred. Sequence analysis of the human EGFR ligands revealed that the nature of the hinge residue correlated with the sequence in the B-loop beta-sheet. As this region is an important determinant in recognition of TGFalpha by the chicken EGFR, we assessed the mitogenicity of the TGFalpha hinge mutants, as well as the other EGFR ligands, using chicken embryo fibroblasts. The preference of the chicken EGFR for TGFalpha hinge mutants with hydrophobic side chains paralleled that of the human EGFR. Betacellulin and heparin-binding EGF-like growth factor also possess an hydrophobic hinge; both were at least as potent as TGFalpha for chicken embryo fibroblasts. EGF and amphiregulin, both with hydrogen bond donor functionalities at their hinge, displayed markedly decreased in potency by comparison with TGFalpha. We propose that EGFR ligands can be subclassified into TGFalpha-like and EGF-like and that this is of functional significance, identifying a potential mechanism whereby EGFR can discriminate between its ligands.

Although binding of epidermal growth factor (EGF) and transforming growth factor ␣ (TGF␣) to the EGF receptor (EGFR) is mutually competitive, their binding is not identical, and their biological activities are not always equivalent. To probe for ligand-specific interactions, we have synthesized analogues of TGF␣ with modifications to the residue lying between the fourth and fifth cysteines (the "hinge"). Although this residue lies in a structurally conserved region of the protein, it is not conserved within the EGFR ligand family. Our results show that in TGF␣ there is a preference for a bulky hydrophobic hinge residue; this contrasts with EGF, for which a hydrogen bond donor functionality is preferred. Sequence analysis of the human EGFR ligands revealed that the nature of the hinge residue correlated with the sequence in the B-loop ␤-sheet. As this region is an important determinant in recognition of TGF␣ by the chicken EGFR, we assessed the mitogenicity of the TGF␣ hinge mutants, as well as the other EGFR ligands, using chicken embryo fibroblasts. The preference of the chicken EGFR for TGF␣ hinge mutants with hydrophobic side chains paralleled that of the human EGFR. Betacellulin and heparin-binding EGF-like growth factor also possess an hydrophobic hinge; both were at least as potent as TGF␣ for chicken embryo fibroblasts. EGF and amphiregulin, both with hydrogen bond donor functionalities at their hinge, displayed markedly decreased in potency by comparison with TGF␣. We propose that EGFR ligands can be subclassified into TGF␣like and EGF-like and that this is of functional significance, identifying a potential mechanism whereby EGFR can discriminate between its ligands.
The epidermal growth factor receptor (EGFR) 1 ligands, transforming growth factor ␣ (TGF␣) (1), epidermal growth factor (EGF) (2), amphiregulin (AR) (3), heparin-binding EGF (HB-EGF) (4), and betacellulin (BTC) (5) are characterized by a three looped EGF-motif, formed from a single polypeptide chain constrained by three intramolecular disulfide bridges. The structure can be conveniently divided into an N-domain, which possesses a prominent ␤-sheet, and a C-domain (6). These are connected by a single residue, which lies at end of the ␤-sheet and is sometimes referred to as the hinge (see Fig. 1, a and b).
EGF and TGF␣ have been the subjects of many studies that have attempted to understand the details of the EGFR-ligand interaction (reviewed in Ref. 7). Most of these studies have concentrated upon highly conserved residues in an effort to elucidate the common features responsible for the binding of all EGFR ligands. Yet despite their similar three dimensional structure (8) and their mutual inhibition, evidence exists for EGF and TGF␣ that their binding may not be identical (9). Furthermore, although these two growth factors have similar activities in some assays (10), in other cases they exert differential effects. For example, EGF and TGF␣ exhibit distinct potencies in a variety of contractile smooth muscle systems (11), TGF␣ is a more potent angiogenic mediator (10), and it is more effective than EGF at inducing Ca 2ϩ release from bones in culture (12). TGF␣ is also more vasoactive than EGF, but unlike EGF it does not induce a refractory period following treatment (13). Although some some of these differences may be related to differences in processing or handling of the ligand (14), the fact that the chicken EGFR exhibits binding affinities for mEGF and hTGF␣ that differ by 2 orders of magnitude (15) suggests that the ligands interact with the receptor in nonidentical ways. Using mEGF/TGF␣ chimeras, we have shown that the B-loop ␤-sheet and hinge residue, which vary markedly between EGF and TGF␣, are determinants in recognition of mEGF and hTGF␣ by the chicken EGFR (16).
All known EGF-like motifs have a strikingly consistent structural core defined by their arrangement of the cysteines. The importance of disulfide constraints in this core region is immediately apparent in a superimposition of NMR-derived conformations of mEGF and hTGF␣ (Fig. 1b). In spite of differing hinge residues (and overall homologies of about 30%), in most cases, the RMS superimposition of cysteine ␣ carbons is less than 0.6 Å, and the largest deviations are between alternative conformations of the same growth factor. The hinge residue (indicated by the arrow in Fig. 1b) is part of this well preserved structural core. It is therefore significant that in EGF, point mutations of this receptor contact residue have established that a hydrogen bond donor residue is preferred (17), whereas this residue is valine in TGF␣. This led us to explore whether this residue might represent a ligand-specific site of interaction with the EGFR. Production of TGF␣ Yeast Expression Vector-TGF␣ was expressed in Saccharomyces cerevisiae using the yeast expression vector pWYG9/ EGF (20) in which a synthetic TGF␣ mini-gene obtained from clone BB23 replaced that of mEGF. The TGF␣ expression vector was constructed using the following steps: (i) the 179-base pair HindIII-EcoRI fragment containing the hTGF␣ mini-gene was cloned into the HindIII and EcoRI sites of M13mp18 (21); (ii) site-directed mutagenesis (22) was used to remove the Met codon from the 5Ј end of TGF␣ and to insert a XhoI restriction site and codons for Lys and Arg (generation of a Kex 2 proteolytic cleavage site) adjacent to the first valine of hTGF␣ forming M13/TGF/2; (iii) the 715-base pair AatII-BclI fragment from pWYG9/ EGF containing the yeast gal7 promoter and ␣-factor prepro leader sequence fused to mEGF was inserted into the AatII and BamHI sites of pUC18 to form pUC/EGF/1; (iv) the 173-base pair XhoI-EcoRI fragment from M13/TGF/2 was inserted into the XhoI-EcoRI sites of pUC/ EGF/1 to replace the mEGF coding sequence and form pUC/TGF/2; (v) the 715-base pair AatII-BamHI fragment from pUC/TGF/2 was inserted into the AatII-BclI sites of pWYG9/EGF to form pWYG9/TGF/2. As the TGF␣ mini-gene contained a KpnI site not present in mEGF, this was used to ensure correct replacement of mEGF by hTGF␣.

Materials-All
Production of Hinge Mutants-Mutation of Val-33 in TGF␣ was achieved using either site-directed mutagenesis of M13/TGF/2 or by polymerase chain reaction-directed mutagenesis of the HindIII-EcoRI fragment of BB23 using a two-step amplification procedure (23) and Vent polymerase. For the polymerase chain reaction-directed mutagenesis, the antisense primers across the mutation site (TAACCAGAGTG-GCATACGCATGCCGGTTTGT) incorporated base changes at Val-33 (ATA) for Ile (GAT), Gly (GCC), and Asn (GTT); the primers used at the ends of the TGF␣ DNA incorporated BamHI restriction sites were: 5' sense, ccggatccATGGTTGTATCCCAC, and 3Јantisense, tcggatccTT-ATTAAGCTAGCA.
Expression and Purification of Recombinant TGF␣s and mEGF-TGF␣, mEGF, and the mutant growth factors fused to the ␣-factor secretion signal of vector pWYG9 were expressed in S. cerevisiae S150 -2B 2 0 (␣Leu2 His3 Ura3 Trpl). The recombinant yeast cells were initially grown in yeast peptone (YP) medium before harvesting and concentrating 10 times by resuspension in 100 ml of synthetic complete medium (24), which lacked dextrose but contained twice as much yeast nitrogen base as well as 2% (w/v) galactose for induction of growth factor expression. After 24 h, the cell-free supernatant was harvested by centrifugation and stored frozen at Ϫ20°C prior to purification.
The cell free yeast supernatant was adjusted to 40 mM trifluoroacetic acid and loaded onto a C 18 Sep-Pak extraction cartridge equilibrated with 40 mM trifluoroacetic acid. After washing with 140 ml of 10% acetonitrile in 40 mM trifluoroacetic acid, the growth factors were eluted using 50% acetonitrile in 40 mM trifluoroacetic acid. After lyophilization, the growth factors were further purified by Mono Q anion exchange chromatography (16) and C 18 reversed phase chromatography using a Pep RPC C 18 HR5/5 (Pharmacia Biotech Inc.) with a linear gradient of 0 -42% acetonitrile in 40 mM trifluoroacetic acid/H 2 O over 12 ml. Samples were lyophilized, and protein concentrations were determined either by ELISA (for TGF␣ mutants) as described below or by absorbance at 280 nm (mEGF) as described previously (16).
Laser Desorption Mass Spectroscopy-Molecular masses were assessed using a laser desorption mass spectrometer as described previously (16).
Competitive ELISA for TGF␣-Immunoreactive TGF␣ was determined in competitive ELISA using a monospecific antibody raised against the C-terminal 17 residues of TGF␣ (epitope 44 -50). Standard TGF␣ (12.5 ng/ml) was immobilized in the wells of an ELISA plate using 35 mM NaHCO 3 /15 mM Na 2 CO 3 buffer containing 10 mM dithiothreitol and 0.02% NaN 3 (w/v), pH 9.6, for 18 h at 4°C. A second ELISA tray was prepared containing dilutions of the recombinant TGF␣ samples and a TGF␣ standard in 50 mM Tris-HCl buffer containing 0.25% (w/v) BSA, 145 mM NaCl, 0.05% (v/v) Tween 20 (TB/BSA buffer) to which was added an equal volume of sheep anti-TGF␣ antibody (10 g/ml in TB/BSA containing 20 mM dithiothreitol). After incubation overnight at 4°C, the first plate containing the immobilized TGF␣ was blocked with TB/BSA buffer for 1 h at 37°C before addition of 100 l of the TGF␣/antibody mixture from the second tray. Following another overnight incubation at 4°C, antibody bound to the immobilized TGF␣ was detected using an anti-sheep immunoglobulins antibody conjugated to horseradish peroxidase, followed by development using 0.01% hydrogen peroxide in 0.1 M sodium acetate buffer with 0.15 M 3, 3Ј,5,5Јtetramethylbenzidine dihydrochloride as chromogen. The reaction was halted by the addition of 2 M H 2 SO 4 (50 l/well), and absorbance was read at 450 nm. The TGF␣ in the samples was estimated by reference to a TGF␣ standard curve in the range of 64 -500 ng/ml.
Cell Culture-Cells were routinely cultured in Dulbecco's modified Eagle's medium, containing 10% (v/v) heat inactivated fetal bovine serum, 50 IU/ml penicillin, 50 IU/ml streptomycin, 1 mM L-glutamine, 1 mM sodium pyruvate, and 1 ϫ nonessential amino acids. The NR6/HER medium was supplemented with 100 g/ml of geneticin sulfate to maintain the selection of the EGFR encoding plasmid.
Mitogenesis Assays-Mitogenic activity of recombinant growth factors was determined by measuring the stimulation of incorporation of [ 125 I]iododeoxyuridine into DNA of NR6/HER cells or primary cultures of chicken embryo fibroblasts (CEFs) as described previously (16). In the case of the CEFs, responsiveness to growth factor stimulation tended to vary between individual primary cultures (identified as CEF1, etc) and declined en passage; assays were routinely performed using primary cultures between passages 3 and 10. EC 50 values were determined by nonlinear curve fitting to the logistic function f(x) ϭ (a Ϫ d)/(1 ϩ (x/c) d ) ϩ d. 125 I-Labeled mEGF Competitive Binding Assays-Relative binding affinities were determined by measuring the ability of recombinant growth factors to displace [ 125 I]mEGF binding to EGFR on HN5 cells (25). IC 50 values were determined by nonlinear curve fitting as above.
Receptor Tyrosine Phosphorylation-EGF receptor phosphorylation was determined using HN5 cells in 96-well plates at 50% confluence. Cells were incubated with 50 l of sample diluted in Dulbecco's modified Eagle's medium for 10 min at 22°C. The samples were then removed, and cells were washed twice in phosphate-buffered saline before solubilization in 50 l of 1 ϫ sample buffer (62.5 mM Tris-HCl, pH 6.8, 2% (w/v) SDS, 5% (v/v) glycerol, 0.002% (w/v) bromphenol blue, 5% (v/v) mercaptoethanol) containing phosphatase and protease inhibitors (1 mM NaF, 1 mM Na 3 VO 4 , 70 M phenylmethylsulfonyl fluoride) and heating at 95°C for 3 min. Samples were analyzed by SDS-polyacrylamide gel electrophoresis and Western blotting using a biotinylated antiphosphotyrosine antibody (PY20) as described previously (26). The phosphorylated band at 170 kDa was confirmed as EGFR using a polyclonal anti-EGFR antibody.

RESULTS
As Val-33 lies in a structurally conserved region of TGF␣, the first series of Val-33 substitutions were chosen because they were known to be functional in other members of the growth factor family (i.e. Val 3 Ile, Val 3 Asn, and Val 3 Lys). These TGF␣s were all efficiently expressed by the recombinant yeast cells, and each was readily purified by ion exchange and reversed phase chromatography. The latter procedure ensured that any incorrectly folded growth factor was separated from the purified product (27); in each case, purification yielded a single major peak of protein (e.g. Fig. 2a), and molecular masses where confirmed by laser desorption mass analysis (e.g. Fig. 2b).
Conservative substitution of Val-33 3 Ile was found to have no effect on the ability of the recombinant TGF␣ to compete in receptor binding assays (Fig. 3a and Table I)  enhanced its ability to stimulate DNA synthesis ( Fig. 3b and Table I). However, substitution of Val-33 with a hydrogen bond donor functionality caused a parallel reduction in EGFR binding and induction of DNA synthesis; in particular, the Lys-33 mutant showed a two log reduction in activity. The initial series of mutations indicated that there was a preference for a hydrophobic hinge residue in TGF␣. This was confirmed by comparing the activities of two other hydrophobic substitutions of the hinge residue (Val 3 Leu and Val 3 Met), both of which exhibited activities similar to that of wild type TGF␣. A glycine hinge mutant was also examined as an example of the absence of functionality at this position. This growth factor was expressed poorly by the recombinant yeast cells, and although we had insufficient material to determine the binding affinity of the Val-33 3 Gly mutant, in the more sensitive mitogenesis assay, its activity was found to lie between those of the Asn-33 and Lys-33 mutants.
Ligand binding to EGFR leads to activation of its intrinsic protein tyrosine kinase and initiation of intracellular signals. The ability of the TGF␣ hinge mutants to activate EGFR autophosphorylation in HN5 squamous carcinoma cells closely paralleled their observed receptor binding and mitogenic activities (Fig. 4). All TGF␣s with hydrophobic substitutions of the hinge residue stimulated EGFR autophosphorylation; consistent with its slightly enhanced ability to stimulate DNA synthesis, the Ile-33 mutant also showed slight enhancement of EGFR phosphorylation relative to wild type TGF␣. In the case of the Lys-33 or Asn-33 mutants, we estimate that EGFR phosphorylation decreased to 5 or 25% of wild type, respectively.
The chicken EGFR displays markedly different affinities for mEGF and TGF␣ (15), and we have previously shown that this difference in binding is due, in part, to differences in the B-loop ␤-sheet and hinge residue (16). When tested in mitogenesis assays using chicken embryo fibroblasts (CEFs) a preference for a hydrophobic hinge in TGF␣ is clearly evident (Fig. 5). Both TGF␣ and the hydrophobic Ile-33 substitution were potent mitogens (EC 50 values, 14 and 4.2 pM, respectively), whereas the TGF␣ Asn-33 and Lys-33 mutants (EC 50 values, 100 and 342 pM, respectively) were intermediate in potency between TGF␣ (EC 50 , 14 pM) and mEGF (EC 50 , 4000 pM).
BTC and HB-EGF also have an hydrophobic hinge residue. When we assessed their mitogenic activities using CEFs, we found that both were at least as effective as TGF␣ (Fig. 6). In three independent assays, BTC was equipotent with TGF␣, whereas HB-EGF was a superagonist; the enhanced activity of HB-EGF was variable and probably due to differential expression of heparan sulfate proteoglycans. Like mEGF, AR has a hydrogen bond donor hinge residue, but it was even less effective than mEGF as a mitogen for CEFs. This can be explained by the truncation of the C-terminal tail of AR, which reduces mitogenic potency by 1-2 orders of magnitude (28). DISCUSSION The EGF structural motif is found in a wide variety of extracellular proteins, and almost every type of functional group  is tolerated at the hinge. Thus, although this residue lies within the core of the protein, there is no intrinsic structural requirement for any particular side chain. This is confirmed by NMR-derived structures of many EGF modules (29) and in a study of several EGF hinge mutants (17), where only proline substitution was shown by NMR spectroscopy to affect the ligand structure. Therefore, changes in activity arising from substitutions of the hinge residue in TGF␣ or EGF cannot readily be dismissed as the result of conformational perturbations.
We show that substitution of Val-33 with other bulky hydrophobic residues did not significantly alter the ability of TGF␣ to bind or activate either the human or the chicken EGFR, whereas substitution with Asn, as found in EGF, or Lys, as found in AR, produced ligands with reduced affinity. Our observations contrast with similar work by Niyogi and co-workers (17) or Koide et al. (30), which showed that in EGF, hydrogen bond donor residues were preferred over hydrophobic residues. We also demonstrate that in most cases, receptor binding affinity and mitogenicity are similarly affected. For the Lys-33 mutant, however, ligand affinity was very poor compared with mitogenic potency, but this is in agreement with other mutagenesis studies. For example, in a study of Tyr-37 mutations in EGF, the mutants exhibited differing receptor-binding affinities and abilities to activate the EGFR tyrosine kinase, but most mutants stimulated thymidine incorporation similarily to wild type (31). Likewise, mutations of Leu-47 dramatically reduced receptor binding affinity but had a more modest influence on mitogenic activity (32). This must in part reflect the complexity of the bioassays where the measurements are dependent on numerous kinetically controlled cellular events. Such complex factors may also underlie the slightly superagonistic activity of the Ile-33 mutant, which exhibited equivalent affinity to wild type TGF␣ but which showed elevated tyrosine kinase activity.
Examination of the sequences of known human EGFR ligands reveals that they can be divided into two families, EGFlike or TGF␣-like, based upon the nature of their hinge residue. This classification also correlates with a particular pattern of residues in the ␤-sheet, being characterized most conveniently by the presence or the absence of a proline; TGF␣, HB-EGF, and BTC are TGF␣-like ligands, whereas EGF and AR are EGF-like. This evidence implicates the B-loop ␤-sheet as a recognition element in EGFR ligands, a role supported by recent work with EGF/TGF␣ B-loop chimeras (16). It is also consistent with our observation that the EGFR ligands fall into two groups when tested on chicken EGFR: the high affinity TGF␣-like ligands and the two low affinity EGF-like ligands. Examination of known EGFR ligands from different species (Table II) indicates that with the exception of rat and mouse HB-EGF, they can all be grouped according to the same classification. Presumably, the difference in the rodent HB-EGFs reflects some variation within the EGFR from this species; it would be of interest to determine whether rodent HB-EGF has a lower affinity for the human EGFR, as does mEGF for the chicken EGFR (15). Differences in behavior of rodent HB-EGF have already been reported in studies of cellular sensitivities to diphtheria toxin. Whereas human or monkey cells are extremely sensitive to diphtheria toxin, rodent cells are not, even though the HB-EGF precursor (the diphtheria toxin receptor) shows about 80% sequence identity between human and mouse. Recent studies with chimeric HB-EGFs have indicated that the most critical residues for toxin binding lie between residues 122-148 (33, 34); significantly the hinge residue lies within this region (residue 133).
It is well known that EGF and TGF␣ compete for binding to EGFR, suggesting a common ligand binding site. However, it need not be the case that the recognition of both EGF and TGF␣-like ligands is through homologous residues or that EGF and TGF␣ bind in identical conformations. For example, identical interactions for EGF and TGF␣ could be achieved by different residues providing the same critical functional groups in a common receptor binding pocket. However, we have been unable to identify any spacially related functionality in the region of the ␤-sheet and the hinge residue, so it is unlikely that this can account for the differing hinge preferences we have observed. We therefore conclude that the B-loop ␤-sheet region provides a mechanism for discrimination between EGF and TGF␣ and that this part of the growth factor recognizes different receptor residues depending on its composition. When combined with other information, our evidence suggests a model for the binding of ligands to the EGFR. The picture that is emerging is of both common and specific points of interaction between the EGFR and its ligands. In common among EGFR ligands are residues such as Arg-42 and Leu-48, which contribute a large proportion of the binding free energy, as well as Phe-15, Phe-17, and His-18, which are semi-conserved and together may form a receptor binding patch (8). It is noteworthy that this pattern of conservation is largely preserved in the heregulins and viral growth factors and thus this epitope might be a common feature of all c-erbB ligands. Ligand-specific interactions involving the B-loop ␤-sheet may provide a mechanism for discrimination between individual growth factors. In this ligand-binding model, the N-domain of EGF and TGF␣ may be accommodated in exclusive subsites, or they may occupy a common subsite utilizing distinct or overlapping functional epitopes, as is the case for the human growth hormone receptor (35). Either mechanism would be consistent with the differential inhibition of binding of EGF and TGF␣ by the 13A9 monoclonal antibody (9). Ligand-specific interactions may then result in the transduction of different signals or signals of differing intensity or duration and consequent differences in cellular response.