Decorin Binds to a Narrow Region of the Epidermal Growth Factor (EGF) Receptor, Partially Overlapping but Distinct from the EGF-binding Epitope*

Decorin, a small leucine-rich proteoglycan, is a key regulator of tumor growth by acting as an antagonist of the epidermal growth factor receptor (EGFR) tyrosine kinase. To search for cell surface receptors interacting with decorin, we generated a decorin/alkaline phosphatase chimeric protein and used it to screen a cDNA library by expression cloning. We identified two strongly reactive clones that encoded either the full-length EGFR or its ectodomain. A physiologically relevant interaction between decorin and EGFR was confirmed in the yeast two-hybrid system and further validated by experiments using EGF/EGFR interaction and transient cell transfection assays. Using a panel of deletion mutants, decorin binding was mapped to a narrow region of the EGFR within its ligand-binding L2 domain. Moreover, the central leucine-rich repeat 6 of decorin was required for interaction with the EGFR. Site-directed mutagenesis of the EGFR L2 domain showed that a cluster of residues, His394-Ile402, was essential for both decorin and EGF binding. In contrast, K465, previously shown to be cross-linked to epidermal growth factor (EGF), was required for EGF but not for decorin binding. Thus, decorin binds to a discrete region of the EGFR, partially overlapping with but distinct from the EGF-binding domain. These findings could lead to the generation of protein mimetics capable of suppressing EGFR function.

Decorin, a prototype member of an expanding family of small leucine-rich proteoglycans (1), plays pivotal roles in modulating matrix assembly (2)(3)(4)(5) and cell proliferation (6 -8). Most of the biological functions of decorin are mediated by the protein core's unique organization of 10 tandem leucine-rich repeats (LRR) 1 which fold into an arch-shaped structure (9) whose concave surface is well suited to bind both globular and nonglobular proteins (2,10,11) as well as metal ions (12). Decorin expression is markedly suppressed in most transformed cells derived from primary malignant tumors (13)(14)(15) or in cells transformed by oncogenes such as vSrc (16), vJun (17), and ATF3 (18). On the contrary, decorin expression is markedly up-regulated during quiescence (19,20), and its levels can reach ϳ40-fold in post-confluent fibroblasts (21). We have previously shown that decorin expression is enhanced around invasive carcinomas (22) and have proposed that decorin might represent a natural antagonist to the growing cancer cells (1). This working hypothesis is corroborated by the established effects of decorin on growth factor-mediated tumor progression (23,4) and by its profound cytostatic effects on a wide variety of tumor cell lines (13,14,24,25). Lack of decorin is permissive for tumor development insofar as bitransgenic mice, lacking both decorin and the tumor suppressor p53, develop an accelerated lymphoma tumorigenesis (26). These earlier reports have been supported by the recent observation that decorin gene expression is differentially down-regulated in hepatocellular (27) and ovarian (28) carcinomas vis á vis their normal counterparts.
The decorin-induced effects are mediated, at least in part, by a specific interaction between decorin protein core and the epidermal growth factor receptor (EGFR) (29,30). This interaction triggers a signal cascade leading to activation of mitogen activated protein kinases (29), mobilization of intracellular calcium (31), up-regulation of p21 WAF1/CIP1 (p21) (25,32), and ultimately to growth suppression (13,14,25). However, following a transient stimulation of the EGFR kinase, decorin causes a sustained down-regulation of the EGFR (33) and other ErbB members of receptor tyrosine kinase (34), an action that would negatively affect tumor growth. Additional decorin-mediated effects are observed in normal cells such as macrophages (35), endothelial (36), and mesangial cells (37). Interestingly, EGF suppresses decorin expression in normal fibroblasts, suggesting a negative feedback loop between growth-promoting and growth-suppressing factors (38).
In this study we utilized an expression cloning strategy and discovered two transmembrane polypeptides that bound a soluble decorin/AP chimera. The two proteins encompassed either the full-length EGFR or a truncated version lacking the intracellular domain. The specific binding of decorin to the EGFR (K D ϳ27 nM) could be demonstrated in six different cell lines following transient transfection with the EGFR ectodomain. Moreover, we demonstrated a physiologically relevant interaction between decorin and EGFR using the yeast two-hybrid system, also confirmed by EGF/EGFR experiments. Using a panel of deletion mutants, decorin binding was mapped to a narrow region of the EGFR ectodomain within the ligandbinding L2 domain. Deletion mutants of decorin further revealed that the central leucine-rich repeat 6 (LRR 6 ) of decorin was required for proper interaction with the EGFR. Site-di-rected mutagenesis of the EGFR L2 domain showed that a cluster of residues, His 394 -Ile 402 , was essential for both decorin and EGF binding; however, there was differential binding following multiple amino acid substitutions. These in vivo data confirm the specific binding of decorin to the EGFR and show that the decorin protein core binds to a discrete region of the EGFR, partially overlapping with but distinct from the EGF-binding epitope. Our results could potentially lead to the generation of protein mimetics that could antagonize EGFR activity in a variety of tumors in which EGFR is overexpressed.

EXPERIMENTAL PROCEDURES
Expression of a Decorin/AP Chimeric Proteoglycan-The cDNA encoding the heat-stable placental AP was amplified from the APtag vector (39) by using two oligonucleotide primers with two different restriction sites, KpnI and EcoRI, and ligated into the expression vector pcDNA3 (Invitrogen). Similarly, a PCR-generated full-length human decorin cDNA was digested with EcoRI and XbaI and ligated at the 5Ј end of the AP/pcDNA3. Correct ligation and open reading frames were determined by automatic sequencing of the entire construct. Human squamous carcinoma A431 cells were stably transfected with this vector. The clones were selected on the basis of Northern and Western blot analyses as described before (34). Two independent clones, 9 and 10, with high expression of decorin/AP chimeric protein were expanded, and conditioned media were collected for interaction study. The Great EscAPe™ SEAP system (CLONTECH) was used for AP assays. Briefly, conditioned media from clone 10 from untransfected cells (negative control) and from cells stably transfected with and secreting an AP construct (positive control) were collected. Aliquots were incubated at 65°C for 30 min to inactivate endogenous phosphatases, cooled on ice, and then mixed with CSPD substrate/chemiluminescent enhancer for 10 min followed by exposure to x-ray film for 30 s. The standard AP curve was generated with conditioned media from a stable AP construct. Plates were treated as discussed before and were quantified on a Spectramax plate reader (Molecular Devices). Multiple runs of the standard were grouped for generation of the curve.
Expression Cloning, Transient Cell Transfection, and Binding Studies-The cDNA was made from mRNA extracted from A-431 cells, sequentially digested with EcoRI and SalI, and ligated in the proper orientation into the pcDNA3 expression vector according to manufacturer's instructions (Stratagene). High efficiency ultra-competent Epicurean cells (Stratagene) were transformed with 1 l of this ligated material. The bacteria were divided into ϳ100 pools. Plasmid DNA was prepared from each pool and transiently transfected into subconfluent COS-1 cells, essentially as described before (39). In initial screenings, we found that no decorin/AP-or AP-containing media bound to the surface of COS-1 cells, in agreement with previous results (40,41). After 2 days, the transfected COS-1 cells were incubated for 90 min with media from the decorin/AP-secreting cells, which were concentrated 5-fold using a Centricon apparatus (50-kDa cutoff). The cells were washed seven times with phosphate-buffered saline, fixed for 30 s in 60% acetone, 3% formaldehyde, 20 mM HEPES (pH 7.5), and washed twice for 5 min each in 150 mM NaCl, 20 mM HEPES (pH 7.5). The plates were then floated on a 65°C water bath for 30 min to inactivate endogenous cellular phosphatases. After rinsing with 100 mM Tris-HCl (pH 9.5), 100 mM NaCl, 5 mM MgCl 2 , the cells were stained for 72 h in the same buffer containing 10 mM L-homoarginine, 0.17 mg/ml 5-bromo-4-chloro-3-indolyl-phosphate (BCIP) and 0.33 mg/ml nitro blue tetrazolium (NBT) as color substrate (Promega). Several positive clones were identified as purple-stained cells representing decorin interacting proteins. In initial screening, AP-containing media did not react with COS-1 cells. Two clones (17-10 and 17-14) were subjected to two successive rounds of screening by diluting the original positive clones 1:48. Finally, single cDNA pools from tertiary screening were diluted into 12 pools for the last screening. When all the clones were positive, plasmids were subjected to automatic sequencing.
To determine whether the interaction observed in the yeast could also be obtained in mammalian cells, we subcloned the EGFR into the eukaryotic expression vector pcDNA3. Various cell lines were transiently transfected with 2 g of EGFR/pcDNA3 using the polyamine reagent TransIT ® (Mirus Corp.). After 48 h, the cells were incubated with decorin/AP-containing media, washed several times, and processed for histochemical detection of AP. For binding studies, COS-1 cells were transfected in the same manner and incubated at 4°C for 2 h with 125 I-labeled decorin protein core, a human recombinant protein core containing O-linked oligosaccharides but lacking the glycosaminoglycan side chain (42). The decorin core was labeled to high specific activity (ϳ10 18 cpm/mol) using Iodogen essentially as described before (34). Estimates of receptor affinity and number of binding sites were achieved using the Wizard program of the Sigma Plot 2001 package.
Yeast Two-hybrid System, Deletion Mutants, and Site-directed Mutagenesis-The ectodomain of the EGFR was generated using primers containing SalI and BglII restriction sites and was ligated into pGAD424 (CLONTECH) vector as prey in the same reading frame as the GAL4 activating domain. In a similar way, decorin cDNA was ligated in pGBT9 (Matchmaker 2, CLONTECH) vector as bait. The different deletion constructs of the ectodomain of EGFR in pGAD424 vector were made by PCR using the same restriction sites as described above. As a positive control, we synthesized human EGF by using two 159-bp oligonucleotides of sense and antisense EGF containing two additional restriction sites for EcoRI at the 5Ј end and SalI1 at the 3Ј end. Yeast reporter strain SFY526 was grown on YPD plates. The growth media were inoculated with a single colony 2-3 mm in diameter and incubated at 30°C with shaking (250 rpm) for 16 -18 h. The cells were grown at 1:30 dilution for 3 h to reach ϳ0.2 A, pelleted by centrifugation (1000 ϫ g for 5 min twice), and re-suspended in sterile Tris EDTA/LiAc solution. These competent cells were transformed with 1 g of EGFR-pGAD424 and decorin-pGBT9 vector constructs with PEG/LiAc solution and plated on S.D. plates lacking Leu and His, or on S.D. plates lacking Leu, His, and Trp. The transformants were grown for 2-4 days at 30°C. The colonies were lifted with 3MM paper and completely submerged in liquid nitrogen for 10 s. The filters were allowed to thaw at room temperature and placed on another paper pre-soaked in Z buffer/X-␣-gal solution. The ␤-galactosidaseexpressing colonies turned blue in 1-2 h. Various deletion mutants of the EGFR or decorin were generated by PCR, including suitable 5Ј restriction endonuclease sites to allow unidirectional cloning into the various vectors. Site-directed mutagenesis was performed by PCR using 40 -50 nucleotide oligomers containing the suitable restriction site at their 5Ј ends and internal substitutions to allow one or multiple amino acid mutations. All of the constructs were fully sequenced before transfection. Additional experimental details are provided in the legends to the figures and the text.

Secretion of Functional Decorin/AP Chimeric Proteoglycan-
The first step for performing expression cloning was to generate a soluble decorin proteoglycan with an easily detectable marker. To this end, we fused the coding region of human decorin to that of the heat-stable human placental AP, a readily detectable histochemical reporter (39), and expressed the resulting chimeric proteoglycan in A431 squamous carcinoma cells using pcDNA3 vector. Among more than 40 clones resistant to G418, we selected two (clones 9 and 10), which expressed a message that hybridized with decorin cDNA (Fig. 1A, top panel). As negative controls, we also generated stable transfectant clones expressing AP alone. Of interest, the expression of decorin/AP caused a concurrent induction of p21 (Fig. 1A, middle panel), a potent cyclin-dependent kinase inhibitor (43), which has been shown to be a downstream target of decorin (29,32). Both clones expressed a single broad band centering at ϳ150 kDa, consistent with the combined sizes of decorin (ϳ100 kDa) and AP (ϳ50 kDa). The chimeric proteoglycans were sensitive to chondroitinase ABC digestion (Fig. 1B), further establishing their proteoglycan nature. Clone 10 was assessed for AP activity using either a standard colorimetric reaction or chemiluminescence (Fig. 1C) and found to contain high levels of expression as compared with standard AP. Quantitative analysis, using a standard curve based on recombinant AP activity (Fig. 1D), revealed that clone 10 produced ϳ10 g/ml of decorin/AP chimeric proteoglycan/10 7 cells/24 h. Collectively, these results indicate that decorin/AP is properly folded because of the following: (a) it has full biological activity since it is capable of inducing p21 and causing growth retardation of the two clones (not shown), in agreement with our previous data (32); (b) it is a fully glycosylated proteoglycan as shown by its migration on SDS-PAGE and its sensitivity to chondroitinase ABC (not shown); and (c) it shows strong AP activity, utilizing either a classical colorimetric reaction or chemiluminescence.
Expression Cloning of the EGFR-To identify decorin-interacting proteins we generated a cDNA library from mRNA of A431 squamous carcinoma cells using random priming and oligo-deoxynucleotidyltransferase. The resulting cDNAs were subcloned into the expression vector pcDNA3 and divided into 100 pools, each containing ϳ5,000 individual cDNAs. Each pool was transiently transfected into green monkey COS-1 cells and screened for cells that bound to soluble decorin/AP (39). The decorin/AP-containing media were concentrated 5-fold by ultrafiltration (M r ϭ 50 kDa cutoff), thereby further purifying the medium and removing degradation products and smaller proteins that could potentially interfere with the assay. Using this approach, we isolated two cDNA pools that, when transfected, induced the surface binding of decorin/AP chimeric proteoglycan ( Fig. 2A). Each pool was subjected to two additional screenings by ϳ50-fold dilutions. After three rounds of screening, we identified and characterized two cDNAs that evoked a robust in situ binding when transiently expressed by COS-1 cells. Clone 17-10 contained an insert of ϳ6 kb (Fig. 2B), whereas clone 17-14 contained an insert of ϳ2.4 kb (Fig. 2C). Interestingly, the nucleotide sequence of the entire 17-10 cDNA revealed a single open reading frame that matched identically with the full-length EGFR (GenBank accession number NM_005228). Clone 17-14 encoded a C-terminal truncated form of the EGFR encompassing the ectodomain and the transmembrane region, but lacking the intracellular domain (Fig. 2D). Collectively, these findings provide independent confirmation that decorin is a biological ligand for the EGFR.
Specific Interaction Between Decorin/AP Proteoglycan and EGFR-To determine whether EGFR was indeed interacting with decorin/AP chimeric proteoglycan, we transiently transfected five different human cell lines of diverse histogenetic origin with the EGFR ectodomain (clone 17-14) expression vector. In all cases there was a pronounced binding of decorin/AP (Fig. 3A). Note that the histochemical detection of AP was done for only 1 h, because prolonged incubation revealed the endogenous EGFR that is highly expressed in A431 and moderately expressed in the other tumor cell lines. Thus, EGFR is a transmembrane protein that selectively interacts with decorin. Because the EGFR ectodomain was capable of interacting with soluble decorin/AP chimera, we conclude that the intracellular domain of the EGFR is not essential for proper decorin binding.
Next, to establish the affinity of decorin for EGFR and to exclude the possibility that the interaction may be mediated by the AP moiety, we utilized a radio-ligand binding assay. To this end, COS-1 cells, which express very low levels of EGFR, 2 were transiently transfected with either EGFR ectodomain-containing pcDNA3 or empty vector and then incubated at 4°C with increasing concentrations of 125 I-labeled decorin protein core (42). The binding was saturable in the range of 25-50 nM, in contrast to the mock-transfected cells that showed much lower levels of binding and did not display saturation (Fig.  3B, top panel). The binding data yielded a straight line in a Scatchard plot (Fig. 3B, lower panel), with a K D ϭ 27.5 Ϯ 4 nM and a number of ligand molecules of ϳ1.8 ϫ 10 5 per cell. These values are comparable to those obtained with mammary carcinoma cells (K D ϭ 71 Ϯ 7 nM) (34) and further indicate that the glycosaminoglycan chains are not directly involved in this interaction since we utilized a radio-labeled protein core.
EGFR/Decorin Interactions Using the Yeast Two-hybrid System-To further prove a physical interaction between the decorin protein core and the EGFR, we utilized the yeast twohybrid system. We first cloned the EGFR ectodomain into the pGAD vector containing the activating domain (prey), whereas the full-length human decorin protein core was cloned into the pGB vector containing the binding domain (bait). As positive and negative controls, we utilized either the pGB-53 (harboring 2 M. Santra, C. C. Reed, and R. V. Iozzo, unpublished observation. FIG. 1. Generation of a cellular system secreting a functional decorin/AP chimeric proteoglycan. A, Northern blotting of three clones resistant to G418 for at least 6 weeks in culture. Notice that the expression of decorin in clones 9 and 10 (top panel) correlates with an induction of endogenous p21 (middle panel). The exposure of the blot in the top panel was for 1 h. The blot was stripped and rehybridized with a full-length p21 cDNA. The middle panel was exposed for 4 h. B, Western immunoblotting of serum-free media conditioned by various clones as indicated following detection with an anti-decorin antibody. Notice the presence of a broad band with a typical migration on SDS-PAGE for a proteoglycan, centering at ϳ150 kDa (left panel). Both bands were sensitive to chondroitinase ABC (20 mU for 1 h) digestion (right panel), further establishing the proteoglycan nature of the decorin/AP chimera. C, alkaline phosphatase activity from clone 10, control AP, and untransfected A431 media using scalar dilutions. The AP activity was detected by chemiluminescence using the SEAP detection kit (CLONTECH). D, standard curve using recombinant AP and chemiluminescence (q). The values were pooled from four independent experiments. Clone 10 secreted 10 -12 ϫ 10 6 relative light units/15 l of media (f) following conditioning for 24 h in serum free medium. This corresponds to ϳ10 g/ml of decorin/AP chimeric proteoglycan/10 7 cells/24 h. the p53 gene) or the pGB-Lam5Ј (harboring the lamin 5 gene) combined with pGAD-T (containing the SV40 T antigen), respectively. In addition, as a positive control for EGFR we synthesized a cDNA encompassing the human EGF sequence known to have high affinity for the EGFR (44) and cloned it into the pGB vector. Before proceeding with the actual screening, the bait plasmid harboring the EGFR ectodomain was tested alone and found not to induce yeast growth in triple minus (Trp Ϫ , His Ϫ , Leu Ϫ ) media. In addition, all the constructs were assayed for growth in double minus (Trp Ϫ , Leu Ϫ ) media as a control for transfection efficiency (45). In the co-transformant reactions we found a robust growth in triple minus media for both constructs (Fig. 3C). Thus, the decorin protein core interacts with the EGFR ectodomain in vivo to an extent comparable to the EGF/EGFR interaction. These data also represent the first demonstration of a specific EGF/EGFR interaction using the yeast two-hybrid system. Moreover, these findings strengthen the notion that it is a real protein/protein interaction unaffected by, and independent of, the presence of glycosaminoglycan side chains or AP.
Decorin Specifically Interacts with the Ligand-binding (L2) Domain of the EGFR-Next, we sought to establish the precise location of the decorin-binding site within the ectodomain of the EGFR by generating 10 deletion mutants of the EGFR, ⌬13⌬10 (Fig. 4A). Growth was observed in cells co-transformed with either the full-length decorin or EGF and the first nine deletion constructs, but it was absent using ⌬10 (Fig. 4A). In addition to growth in triple minus, transcription of LacZ containing the upstream binding sites of GAL4 and the subsequent ability of co-transformant yeast strains to express functional ␤-galactosidase is further proof of a true protein/ protein interaction (46,47). All the deletion mutants, with the exception of ⌬10, showed a robust production of blue colonies (Fig. 4B). Identical ␤-galactosidase assays were obtained by co-transfection of EGF and EGFR (not shown). Thus, the 50 amino acid residues between 365 and 414 are likely to be either directly involved in binding to, or providing direct structural support to, the binding regions of both decorin and EGF. This region corresponds to the ligandbinding domain of EGFR (48), also known as the L2 domain (see under "Discussion").
The Central LRR 6 of Decorin Is Required for Interaction with the EGFR-Next, we investigated the structural requirements for decorin/EGFR interaction by generating seven deletion mutants of decorin in the yeast pGB vector. A three-dimensional model of decorin (9) and the seven deletions, which encompass various LRRs, are illustrated in Fig. 5. Interestingly, growth in triple minus media was obtained with the intact decorin (Fig.  5A) and the first three deletion mutants ⌬1-⌬3 (Fig. 5, B-D), the latter containing LRR 6 -10 . No growth was obtained if additional LRRs were lost in deletion mutants ⌬4 -⌬6 (Fig. 5,  E-G). However, ⌬7, a C-terminal deletion of the last two LRRs, was capable of supporting growth in triple minus media (Fig.  5H). Finally, an additional deletion mutant ⌬8, which contained LRR 1-6 (Met 1 -Ser 195 ), exhibited significant growth and ␤-galactosidase activity (not shown). Thus, we conclude that The ectodomain is divided into four subdomains designated by L1, S1, L2, and S2, respectively. The two inserts were sequenced in their entirety, and the protein sequence for the 5Ј and 3Ј ends of the two interacting clones is annotated in the bottom two schemes. the central LRR 6 is required for proper interaction with the EGFR. This is notable because a similar region has been involved in the binding to collagen (49 -51) and transforming growth factor-␤ (52), further stressing that this region, which lines the apex of the arch-shaped decorin (9), is directly responsible for the active binding of various ligands with no obvious structural similarities. Indeed, this region is enriched in charged amino acid residues (Fig. 5), and this may favor a ligand-binding property.
Decorin and EGF Bind to a Narrow Region within the L2 Domain of the EGFR-To further narrow down the exact binding region of decorin and EGF within L2, we generated four additional deletion mutants of the EGFR ectodomain, ⌬9 1 3⌬9 4 , spaced by ϳ10 amino acid residues (Fig. 6A). The C ␣ trace representation of the L2 module and various deletion mutants are shown in Fig. 6B. This three-dimensional model (53), based on the crystal structure of the insulin-like growth factor-1 receptor (54), predicts the formation of a pocket lined by L1, S1, and L2 domains. The results showed that all the deletions with the exception of ⌬9 4 formed prominent colonies on selective media (Trp Ϫ , His Ϫ , Leu Ϫ ) and generated robust ␤-galactosidase activity for both EGF and decorin (Fig. 6C). Interestingly, the first solvent-exposed face of the L2 domain (residues 310 -481) contains a number of highly conserved hydrophobic residues that likely interact with the EGF (see below). In addition, this face contains Lys 465 (Fig. 6B), an amino acid that can be cross-linked to EGF (55). Although in the deletion mutant ⌬9 the top third of L2, which is closely associated with the hydrophobic patch of amino acids, was missing, there was still full EGF/EGFR and decorin/ EGFR interaction. Even in the ⌬9 3 deletion mutant, which lacked nearly half of the L2 domain (Fig. 6B), there was still full interaction (Fig. 6C). However, the ⌬9 4 mutant abolished both EGF and decorin binding. Therefore, the ⌬9 3 region (residues His 394 -Ile 402 ) is either directly involved in binding or plays a Following careful washing, the cells were lightly fixed, heat inactivated, fixed again, and processed for AP detection. Note that the colorimetric reaction was performed for only 1 h, because prolonged incubation revealed the endogenous EGFR, which is highly expressed in A431 and moderately expressed in the other tumor cell lines. B, binding isotherm (top) and Scatchard analysis (bottom) for the binding of the 125 I-labeled decorin protein core to COS-1 cells following transient transfection with either EGFR-containing pcDNA3 (q) or empty vector (E). The values represent the mean Ϯ S.E. (n ϭ 3). An estimate of receptor affinity (K D ) and number of binding sites was obtained using the Wizard program of the Sigma Plot 2001 computer package. C, interaction of decorin or EGF with the EGFR using the yeast two-hybrid system. Representative experiments are shown in which yeast cells were transfected with the plasmids as indicated followed by a 4-day incubation at 30°C in media lacking Leu, His, and Trp. As positive and negative controls, either the pGB-53 (harboring the p53 gene) or the pGB-Lam5Ј (harboring the lamin 5 gene) combined with pGAD-T (containing the SV40 T antigen) was utilized, respectively (CLONTECH). No growth was triggered when yeast cells were transfected with the EGFR-pGAD alone (not shown), further indicating the specificity of the reaction. Individual colonies are shown in two independent experiments. significant structural role in supporting the binding region for both EGF and decorin.
Mutational Analysis of ⌬9 3 Reveals Subtle Requirements for EGF and Decorin Binding-To further investigate the precise binding site of decorin and EGF on the L2 domain, we generated single or multiple amino acid substitutions of ⌬9 3 (Fig. 7,  A and B). Interestingly, conserved amino acid substitution such as I401V in ⌬9 3a or double substitutions such as I401V and I402V in ⌬9 3b did not cause any change in either growth on selective media (Fig. 7B) or ␤-galactosidase activity (not shown) for yeast cells co-transfected with either EGF or decorin. However, the simultaneous mutation of Leu 399 , Ile 401 , and Ile 402 to Val (EGFR ⌬9 3c ) abolished the interaction with decorin without affecting EGF interaction. In addition we generated four single amino acid substitutions between Arg 403 and Lys 407 . Even non-conserved amino acid substitutions such as those present in ⌬9 3d -⌬9 3g did not alter the binding of either ligand (Fig. 7B), thus further confirming the specific interac-tion discussed previously in the narrow region of the ⌬9 3 deletion mutant. Finally, we wanted to test whether Lys 465 , a lysine residue that can be affinity cross-linked to EGF (55), could interfere with EGF or decorin binding. K465E caused a complete block of the EGF/EGFR interaction without affecting decorin/EGFR interaction (Fig. 7B). Collectively, these results indicate that decorin and EGF have partially overlapping, but yet distinct, binding epitopes on the EGFR ectodomain. This differential binding might explain the dissimilar effects of EGF and decorin on the EGFR activity. DISCUSSION The EGFR transduces signals from the extracellular environment that trigger diverse cellular functions and affect growth. Most of the known ligands for the EGFR are peptide growth factors derived from partial proteolysis of larger membrane-associated proteins that can engage the EGFR in both membrane-intercalated and soluble forms, as well as when tethered to a suitable matrix (56,57). Decorin, a ubiquitous leucine-rich proteoglycan, is a biological ligand for the EGFR (30) despite the fact that it has no homology to EGF or related peptides. Decorin interacts with the EGFR in a protracted way, leading to a sustained down-regulation of EGFR kinase activity and thus rendering tumor cells less activated and reducing their growth potential both in vitro and in vivo (33). In addition, we have recently discovered that transient transgene expression of replication-deficient adenovirus-containing decorin caused a significant growth inhibition of colon carcinoma and squamous carcinoma tumor xenografts (58). The cytostatic effects of decorin correlated with a markedly reduced proliferative index and an attenuation of the EGFR kinase activity both in vitro and in vivo, thus further stressing the important physiological role of decorin in EGFR function. Our findings provide an independent confirmation of previous studies that have used chemical (59,60), immunological (61), and genetic approaches (62) and further refine the mapping of EGF as well as decorin to a discrete region within the L2 domain of the EGFR. Our data raise the intriguing possibility that solidstate signals in the extracellular environment may cooperate or compete with traditional soluble ligands, such as EGF or TGF-␣, to determine the signaling properties of the EGFR in vivo as recently proposed for the EGF-like repeats of tenascin-C (63).
Expression Cloning of the EGFR or Its Ectodomain Using Decorin/AP Chimeric Proteoglycan-Although decorin interacts with the EGFR (29 -31), there are at least two lines of evidence suggesting the presence of additional cell surface receptors for decorin. First, decorin affects a variety of tumor cell lines with diverse histogenetic backgrounds, some of which could conceivably not express the EGFR (14). Second, decorin inhibits the growth of thymic lymphoma cells (26) and normal macrophages (35), two types of cells that either lack or express very low levels of EGFR. Thus, we used expression cloning utilizing a soluble proteoglycan/AP chimera to screen a cDNA library. This screening, though very laborious, has several advantages over conventional screening. All the experiments are performed with living eukaryotic cells; therefore, any biological interaction occurs in a physiological environment and takes place at the cell surface. In fact, because the method relies on an interaction between a tagged soluble product and a living cell surface, all the intracellular proteins are excluded from the screening (39). Among 0.5 ϫ 10 6 clones, two strongly interacting clones were identified that encoded either the full-length EGFR or its ectodomain. We have previously shown that decorin interacts with the soluble form of EGFR (30), a truncated 105-kDa species that lacks the transmembrane and in- FIG. 4. Decorin and EGF interact specifically with the ligandbinding domain (L2) of the EGFR. A, schematic representation of the EGFR and various deletion mutants, ⌬1-⌬10, which were cloned into the prey pGAD vector. The numbers below each deletion fragment designate the starting residue according to the mature protein, i.e. starting from residue 25 after removing the 24 amino acid signal peptide. Growth in Trp Ϫ /Leu Ϫ /His Ϫ media is indicated for either EGF or decorin (as bait) by semi-quantitative assessment, with maximal growth at ϩϩϩϩ. B, representative ␤-galactosidase assays of all the clones containing the various deletion mutants and decorin as bait. Similar results were obtained with the EGF as bait (not shown). tracellular domains and is the product of an aberrant 2.8-kb mRNA synthesized by A431 cells (64). Interestingly, the 17-14 clone we isolated was not this secreted species, because it contained the transmembrane domain. These findings argue against a bias of our screening for detecting mRNA species with high copy number, such as the amplified EGFR in A431 cells  ) and various deletion mutants. The model predicts the formation of 10 LRRs, with two less-conserved LRRs at the N and C terminus of the decorin protein core (9). Basic and acidic residues are shown in blue and red, respectively. The N-terminal Ser 7 , the site of the single glycosaminoglycan attachment in mammalian cells, is shown in violet. The deletion mutants were generated by PCR using appropriate restriction sites at their 5Ј end to allow unidirectional cloning into the pGB vector. The starting of each deletion mutant and the number of LRRs are also shown as well as the results of growth assays of the yeast two-hybrid system in triple minus media. Amino acid numbering follows the mature protein core (9). Each growth value was estimated in two independent experiments, with 5-10 colonies per group. ␤-galactosidase assays of all the growing colonies were also performed to further confirm the growth in triple minus media (not shown). An additional deletion mutant, ⌬8, which contained LRR 1-6 (Met 1 -Ser 195 ) exhibited significant growth and ␤-galactosidase activity (not shown). 6. Decorin and EGF bind to a finite region within the ligand-binding domain (L2) of the EGFR. A, deletion mutants of L2. These mutants were generated by PCR using appropriate restriction sites at their 5Ј end to allow unidirectional cloning into the pGAD vector. The numbers on the left of each deletion mutant correspond to the initial amino acid of the mature EGFR. B, C ␣ trace representation of the EGFR L2 model (53) based on the crystal structure of the insulin-like growth factor-1 receptor (54). The various deleted regions of L2 are in purple. Lys 465 , which has been cross-linked to EGF (55), is shown in red. C, representative ␤-galactosidase assays for decorin/EGFR and EGF/EGFR interacting clones. Three independent clones were selected for each interaction. The clones in the middle and right panels represent additional individual isolates, with the labels corresponding to those in the left panel. (64). These results indicate that decorin and EGFR have high affinity binding sites for each other. This was confirmed by radio-ligand binding assays using transient cell transfection with the EGFR and radioiodinated decorin protein core. The results showed a single straight line in a Scatchard plot yielding a K D ϭ 27.5 nM, similar to that previously obtained before with untransfected tumor cells (34).

FIG.
Decorin and EGF Bind to a Narrow Sequence within the L2 Domain of the EGFR-To prove a physical interaction between decorin protein core and the EGFR, we utilized the yeast twohybrid system. As a positive control, we utilized a 159-bp cDNA encoding the 53-residue human EGF. In all cases there was a robust growth in selective media indicating that the decorin protein core interacts with the EGFR ectodomain in vivo to an extent comparable to the EGF/EGFR interaction. Our findings represent the first demonstration of a specific EGF/EGFR interaction in the yeast and strengthen the notion that decorin/ EGFR is a bona fide protein/protein interaction, independent of the glycosaminoglycan side chains.
To establish the precise location of the decorin-binding site within the ectodomain of the EGFR, we generated 10 deletion mutants, ⌬13⌬10. Growth, as well as robust ␤-galactosidase activity, was observed in numerous individual isolates cotransformed with either the full-length decorin or EGF and the first nine deletion constructs; however, it was absent using ⌬10. Thus, the region that binds both EGF and decorin, residues 365-413, corresponds to the ligand-binding L2 domain of the EGFR (48). To further narrow down this region, we generated four additional deletion mutants, ⌬9 1 3⌬9 4 , spaced by ϳ10 amino acid residues. The results showed that all the deletions, with the exception of ⌬9 4 , formed prominent colonies on selective media and generated robust ␤-galactosidase activity. Thus, a relatively narrow region, ⌬9 3 (His 394 -Ile 402 ), near the C terminus of L2 is essential for the binding of both EGF and decorin. We should point out, however, that deletion analyses suffer from a "one way" approach when a three-dimensional binding region made up of sequentially distant but spatially close residues is dissected by consecutive subtractions. Once serial deletions abolish binding, no further information can be gleaned, especially regarding the importance of residues further downstream. Ablation of binding can be caused by direct deletion of the binding site or by disruption of the local protein structure without the actual binding site ever being overlapped. The failure of both decorin and EGF to bind to ⌬10 provides evidence that the ⌬9 3 region (His 394 -Ile 402 ) is necessary for binding but cannot give further insights without additional experimental data. The EGFR model (53), however, provides additional insight into the ⌬9 3 region. According to the model structure (see Fig. 7A), His 394 -Ile 402 occur on the opposite side of the L2 domain putatively involved in EGF binding as evidenced by cross-linking experiments (55). This additional evidence suggests that His 394 -Ile 402 may not play a direct interacting role with EGF but are nevertheless likely to be important for the overall local conformation of L2.
The 621 amino acid residues of the extracellular domain of the human EGFR can be subdivided into four subdomains (L1, S1, L2, and S2) where L and S stand for large and small Residues deleted in construct EGFR ⌬9 3 , which interacts with both decorin and EGF (see Fig. 6), are shown as a transparent dot surface. Cyan residues indicate hydrophobic amino acid residues located on face 2 of the L2 domain, which are conserved in many ErbB family members and homologous receptors. The triple mutant EGFR ⌬9 3c , containing mutation of Leu 399 , Ile 401 , and Ile 402 (shown in green) into valine, disrupts decorin interaction with EGFR but has no effect on EGF binding. In contrast, mutation of Lys 465 (shown in red) into glutamic acid, has the opposite effect; that is, it abolishes EGF but not decorin binding. Blue residues, indicating mutated residues Arg 403 , Arg 405 , Thr 406 , and Lys 407 in EGFR mutant constructs ⌬9 3d , ⌬9 3e , ⌬9 3f , and ⌬9 3g , respectively, have no effect on either decorin or EGF binding. B, chart indicating EGFR constructs, deletions, and mutations used in yeast two-hybrid analysis. Color of mutations corresponds to A. The growth of colonies co-transfected with either EGFR or decorin (Dcn) and the EGFR deletion mutants with various amino acid substitutions are depicted on the right, with ϩϩϩ as maximal growth. The growth was systematically confirmed by ␤-galactosidase assays using appropriate positive and negative controls (not shown). Each value was derived from 4 -5 individual yeast colonies, and each experiment was repeated 3-5 times with comparable results. domains, respectively (65,66). Several lines of evidence indicate that L2 plays a major role in EGF binding. First, epitopes for three monoclonal antibodies that competitively block EGF binding were mapped to Ala 351 -Asp 364 in L2 (61). Second, crosslinking experiments mapped the binding of EGF to a CNBr fragment encompassing L2 (59) and further showed that the NH 2 group of EGF residue Asn 1 was linked to Lys 336 in L2 (60). Third, EGF and TGF-␣ bind to a proteolytic fragment of the EGFR, which consists of a 40-kDa peptide (residues 302-503) encompassing the whole L2 domain and a small part of the S1 and S2 domains (67). Fourth, deletion studies of the extracellular portion of the EGFR have shown that L2, flanked by the two cysteine-rich S1 and S2 domains, contributes most of the binding forces for the interaction between EGF and EGFR (68). Fifth, EGF and TGF-␣ bind to a soluble L2 domain with affinities (K D ϭ 400 nM) indistinguishable from those of the soluble EGFR ectodomain (69). Finally, domain-swapping experiments between human and chicken EGFR, implicated both the N-and C-terminal halves of L2 as strong interacting sites (62,70). Thus, L2 contains at least two contiguous regions that together provide most of the interactions that define specificity of EGF to the receptor. Our results confirm these data and further demonstrate that, at least in the yeast two-hybrid system, only the C-terminal part of L2 is essential for binding to either decorin or EGF.
Decorin LRR 6 Is Required for Proper Interaction with the EGFR-The decorin protein core consists of about 10 relatively conserved LRRs contained between two Cys-rich regions (7). Interestingly, we found that the central LRR 6 , which lines the apex of the arch-shaped decorin (9), is required for proper interaction with the L2 domain of the EGFR. A significant advance in understanding the structural basis of LRR-protein interactions derives from crystallographic studies of ribonuclease inhibitor (71). Co-crystallization of ribonuclease inhibitor and either human placental ribonuclease A (72) or angiogenin, another ribonuclease, (73) has been achieved. Both structures encompass a horseshoe-shaped protein composed of repeated ␤-strand/␤-turn structures alternating with ␣-helices. In each LRR module the leucine residues form a hydrophobic core, whereas the side chains of the amino acid residues adjacent or flanking the leucine residues are solvent exposed and interact with the ligand (74). It is remarkable that 26 of 28 contact points for the ribonuclease inhibitor/ribonuclease A complex, and 25 of 26 contact points for the ribonuclease inhibitor/ angiogenin complex are located in the ␤-strands and ␤-turns. The majority of hydrogen bonds and van der Waals contacts in the two complexes are quite distinct, suggesting a high degree of specificity (73). Thus, it is not surprising that decorin also shows a significant degree of specificity for its interaction with the EGFR. The three-dimensional model of decorin, based on the crystallographic structure of the ribonuclease inhibitor (9), predicts that the solvent exposed concave surface of the decorin molecule would have a relatively large number of contact points with suitable ligands (see Fig. 5). This is notable because a similar region has been involved in the binding to collagen type I (49 -51) and transforming growth factor-␤ (52), suggesting that the LRR 6 (9) is highly active in binding various proteins with no obvious structural similarities.
Mutational Analysis Reveals Subtle Requirements for EGF and Decorin Binding-There are several residues in the modeled EGFR L2 domain where the side chains are hydrophobic, solvent-exposed, and conserved in two or more of the human EGFR homologs (53). The stretch of 20 amino acids between residues 394 and 414 contains numerous hydrophobic and solvent-exposed amino acids that could interact with both ligands. Interestingly, conserved amino acid substitution such as I401V in ⌬9 3a or double substitutions such as I401V and I402V in ⌬9 3b did not cause any change in either growth on selective media or ␤-galactosidase activity. However, a triple substitution in ⌬9 3c blocked decorin/EGFR interaction, but not EGF/EGFR interaction. In contrast, K465E caused a complete block of EGF/EGFR interaction, but had no effect on the decorin/EGFR interaction, thereby indicating a strict specificity for the binding of either ligand onto the L2 domain. FIG. 8. Changes in electrostatic potential of the EGFR L2 domain may affect EGF binding. A, molecular surface of the EGFR model (Pdb entry 1DNR) (53) implicated in EGF binding, colored by electrostatic potential, and contoured at Ϯ3.5 kT. Positively charged groups are blue, acidic groups are red, and uncharged regions are white. Lys 465 (arrow) is seen below a hydrophobic patch in the central portion of the model. B, same region as in panel A after mutating Lys 465 to glutamic acid (K465E). As shown, this single amino acid mutation can significantly change the charge of the L2 face to a more negative surface. C, molecular surface of one monomer from the crystallographic dimer of the human EGF (Pdb entry 1JL9) (76), which binds to wild type receptor but not to the K465E mutant receptor. Note that human EGF presents a negative charge over much of its surface, consistent with reduced binding to K465E. Panels were produced and mutations performed with Swiss-Pdb Viewer (75).
To investigate this interaction in more depth, we performed an investigation of the electrostatic potential of both wild type and mutant EGFR L2 face using the Swiss-Pdb Viewer (75). The face of EGFR implicated in EGF binding presents a central uncharged face containing several conserved hydrophobic residues ringed by several positively charged groups (Fig. 8A). The molecular surface of the crystallized human EGF (76) shows many acidic surface groups and some neutral regions (Fig. 8C) and is consistent with interaction on the putative wild type EGFR surface. Mutation of Lys 465 to glutamic acid (Fig. 8B) significantly alters the local surface charge and, interestingly, abolishes EGF binding in the yeast two-hybrid system. Collectively, these results indicate that decorin and EGF have partially overlapping, but yet distinct, binding epitopes on the EGFR ectodomain. This differential binding might explain the dissimilar effects of EGF and decorin on the EGFR activity.
It has been shown that the amino acid residues in the EGFR that are essential for binding the LA22 monoclonal antibody (Ala 351 -Asp 364 ) are not critical for the binding of EGF or TGF-␣ because the epitope was eliminated, whereas ligand binding was unaffected (77). Thus, because the monoclonal antibody is bulky it is likely that the EGF binding site is located in the near vicinity and that the competition with EGF occurs by steric hindrance. These data are in full agreement with our present findings that show a binding of the EGF at ϳ30 residues downstream of the monoclonal antibody LA22 binding. Moreover, because the distance between L1 and L2 that borders the ligand-binding cavity in the EGFR is ϳ3 nm (53), and because the overall dimensions of the decorin arch-shaped structure are 6.5 ϫ 4.5 ϫ 3 nm (distance between the two arms ϫ the distance between the base of the arch and the apex ϫ the overall thickness) (9), there is sufficient space in the cavity to accommodate a single decorin molecule. It is likely that decorin may wrap around one single EGFR, thereby interacting with the solvent-exposed face of L2 and some posterior structures as predicted from the mutational analysis.