Identification of a Surface for Binding to the GDNF-GFRα1 Complex in the First Cadherin-like Domain of RET*

The RET receptor tyrosine kinase is activated by binding to a ligand complex formed by a member of the glial cell line-derived neurotrophic factor (GDNF) family of neurotrophic factors bound to its cognate GDNF-family receptor-α (GFRα) glycosylphosphatidylinositol-linked co-receptor. Molecular modeling studies of the extracellular domain of RET (RETECD) have revealed the existence of four cadherin-like domains (CLD1–4) followed by a cysteine-rich domain. Cross-linking experiments have indicated that the RETECD makes direct contacts with both the GDNF ligand and GFRα1 molecule in the complex, although it has low or no detectable affinity for either component alone. We have exploited sequence and functional divergences between the ectodomains of mammalian and amphibian RET molecules to map binding determinants in the human RETECD responsible for its interaction with the GDNF-GFRα1 complex by homologue-scanning mutagenesis. We found that Xenopus RETECD was unable to bind to GDNF-GFRα-1 or neurturin (NTN)-GFRα-2 complexes of mammalian origin. However, a chimeric molecule containing CLD1, -2, and -3 from human RETECD, but neither domain alone, had similar binding activity as compared with wild type human RETECD, suggesting the existence of an extended ligand binding surface within the three N-terminal cadherin-like domains of human RETECD. Subsequent loss-of-function experiments at higher resolution identified three small subsets of residues, mapping on the same face of the molecular model of RET CLD1, that were required for the interaction of human RETECD with the GDNF-GFRα1 complex. Additional experiments demonstrated that N-linked glycosylation of human RETECD was not required for ligand binding. Based on these observations, we propose a model for the assembly and architecture of the GDNF-GFRα1-RET complex.

Glial cell line-derived neurotrophic factor (GDNF) 1 was initially identified as a potent survival factor of ventral midbrain dopaminergic neurons (1) and has since its discovery been intensely studied due to its potential utility as a therapeutic agent for the treatment of neurodegenerative diseases, such as Parkinson's disease (2). GDNF is a disulfide-linked homodimer consisting of two polypeptide chains of about 110 residues each. The overall disulfide arrangement of GDNF conforms to the structural cystine knot motif (3). Sequence and structural similarities have indicated that GDNF is a distant member of the transforming growth factor-␤ superfamily of ligands. However, unlike typical members of this family that signal through receptor serine-threonine kinases, GDNF signals through a receptor complex formed by the receptor tyrosine kinase RET and a glycosylphosphatidylinositol-anchored, ligand binding moiety, the GDNF family receptor ␣1 (GFR␣1). Four structurally related but distinct ligands, namely GDNF, neurturin (NTN), persephin, and artemin, utilize RET as signaling receptor with the aid of four different GFR␣ receptors (GFR␣1-4), which confer ligand specificity (4,5). In addition, an alternative receptor complex for GDNF family ligands, involving the neural cell adhesion molecule in collaboration with GFR␣ proteins, has recently been identified (6).
Both gain-and loss-of-function mutations in the RET gene have been identified in human diseases. Mutations inducing constitutive dimerization or activation of the RET tyrosine kinase lead to familial and sporadic cancers in neuroendocrine organs, including multiple endocrine neuroplasias type 2A and 2B and familial medullary thyroid carcinoma (7,8). Loss-offunction mutations in RET cause a dominant genetic disorder of neural crest development known as Hirschsprung disease, which results in a lack of neurons in distal segments of the enteric nervous systems and colon aganglionosis (9). Although RET has no detectable affinity for any of the GDNF family ligands in the absence of GFR␣ receptors, chemical cross-linking and co-immunoprecipitation experiments have indicated that RET can still make direct contacts with both the GDNF and GFR␣1 molecules in the complex (10 -12). A structural and functional understanding of the protein-protein interactions at play in the GDNF-GFR␣1-RET ternary complex is still lacking and will be required for the rational design of small molecules capable of mimicking the effects of GDNF.
The binding determinants that mediate the interaction between GDNF family ligands and GFR␣ molecules have been investigated using different approaches, including alanine and homologue-scanning mutagenesis (12)(13)(14). Key residues involved in the interaction with GFR␣ receptors were found in the two ␤Ϫhairpin "fingers" of the GDNF molecule (13,14). Intriguingly, residues at analogous positions have been shown to participate in the interaction of classical transforming * This work was supported by grants from the Swedish Foundation for Strategic Research, the Swedish Medical Research Council (K99-33X-10908-06C), the Vth Framework Program of the European Union (QLRT-1999-00099), Centrala Försökdjursnä mnden, the Swedish Fund for Research without Laboratory Animals, and the Karolinska Institute. 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. growth factor-␤ proteins with their cognate type II receptors (15). The major determinant of ligand binding in GFR␣ molecules has been localized to the most conserved region of the molecule, a central domain predicted to contain four conserved ␣-helices and two short ␤-strands (12). Distinct hydrophobic and positively charged residues in this central region were required for the binding of GFR␣1 to GDNF (12). Because the RET ECD is unable to bind members of the GDNF family directly, and only weakly to GFR␣ molecules (16), this receptor is likely to interact with a composite surface formed by residues from both GDNF and GFR␣ molecules. However, the regions and residues in RET ECD that participate in these interactions have remained unknown.
The overall molecular architecture of the RET ECD was recently elucidated using a bioinformatics approach (17). In that study, it was found that the RET ECD comprises four N-terminal domains with similarity to classical cadherin molecules, socalled cadherin-like domains or CLDs, followed by a C-terminal cysteine-rich domain. Multiple alignments indicate that the RET ECD from a number of different species, including human, mouse, chick, frog, fish, and fly, appear to conform to this organization (17). The highest degree of sequence similarity between the RET ECD and cadherins is found in and around a highly conserved calcium binding site present between CLD2 and CLD3 but, unlike classical cadherins, absent between all other RET ECD subdomains (17).
In the present study, we have investigated the location and biochemical characteristics of ligand binding determinants in the human RET ECD . For this purpose, we have employed homologue-scanning mutagenesis, taking advantage of the inability of the Xenopus RET ECD to interact with complexes between GDNF family ligands and GFR␣ molecules of mammalian origin, despite its overall structural similarity to the human RET ECD .

MATERIALS AND METHODS
DNA Constructs-All expression constructs were generated in the pSecTag2AHA system (18). The cDNA encoding the mature part of the RET ectodomain was amplified by PCR and cloned into the SfiI and NotI sites of the pSecTag2AHA vector. The chimeric constructs were generated by splicing by overlap extension (19). The integrity of the cloning junctions of all constructs were confirmed by automated DNA sequencing. The regions targeted for mutation by en-bloc mutagenesis were identified using the GETAREA 1.1 software (20) (www.scsb. utmb.edu/getarea/area_man.html) with the coordinates from the modeled CLD (1-3) as input (17).
Transfection and Selection of Stable CHO Cell Lines-Chinese hamster ovary (CHO) cells were maintained in a humid atmosphere of 5% CO 2 in Dulbecco's modified Eagle's medium (Invitrogen) supplemented with 2 mM L-glutamine and 60 g/ml gentamycin and 10% fetal bovine serum. Freshly split CHO cells were transfected using FuGENE 6 with the different chimeric constructs. Forty-eight h after transfection, 800 g/ml of hygromycin B (Invitrogen) was applied in complete Dulbecco's modified Eagle's medium. Colonies were picked within 1-2 weeks of selection and expanded in 96-well plates. The expression of RET protein was determined by Western blot using the anti-HA antibody B16.12 (Covance, Biosite) as primary antibody. Single clones were expanded for expression experiments.
Expression of Soluble Chimeric RET ECD Proteins-Stably transfected CHO cells were expanded into 10-cm plates and allowed to reach 80% confluence. Subsequently, the serum-containing medium was removed, and the cells were rinsed twice with PBS. Serum-free medium consisting of Dulbecco's modified Eagle's medium supplemented with 2 mM L-glutamine and 1 mM Hepes and 60 g/ml gentamycin was added to the cells. To enhance the yield, prevent CHO cell proliferation, and alleviate potential misfolding routes during protein maturation, the CHO cells were expanded at 37°C but shifted to 30°C for the period of protein expression in serum-free medium as described previously (18). The incubation and protein production were continued for 3-4 days at ϩ30°C. The serum-free medium was concentrated using Centriprep-10 (Amicon). The concentration of the protein was determined either by Western blotting with anti-HA antibodies or by sandwich enzyme-linked immunosorbent assay using anti-HA antibodies as capture tools and a polyclonal rabbit anti-RET ECD antibody (18) as detection reagent.
Deglycosylation Assays-To verify the structural integrity (and consequent ability of the molecules to pass the "quality control checkpoint" of the endoplasmic reticulum) of the RET domain chimeras and finetuned mutants, the chimeric RET ECD mutants were subjected to deglycosylation assays using endoglycosidase H (Endo H) and peptide:Nglycosidase F (PNGase F) deglycosidases. Typically, 50 l of conditioned medium was either treated with deglycosidase or left untreated. The deglycosylation was carried out according to instructions of the manufacturer (New England Biolabs).
Binding Assays-The binding experiments were performed essentially as described (18). Briefly, 50 ng of human GDNF (PeproTech) was mixed with 250 ng of a fusion protein formed by rat GFR␣1 and the Fc domain of human IgG (GFR␣1-Fc, R&D Systems) in PBS. In some experiments, the related ligand NTN (human) and the Fc fusion of its cognate GFR␣2 receptor (from rat) were also used. After 5 min of incubation at room temperature, the protein solution was added to the wells of an enzyme-linked immunosorbent assay plate (MaxiSorp, Nunc). The following day, the wells were rinsed by PBS and blocked with 2% skimmed milk powder in TBS (2% MTBS) for 1 h at room temperature. Equal amounts of RET ECD mutants were added to the GDNF-GFR␣-1-Fc coated wells and to wells only coated with 2% MTBS. The binding was allowed to proceed for 1 h at room temperature. The washing was performed with TBS with 0.1% Tween 20 three times and TBS three times. Following washing, bound RET ECD molecules were detected by a monoclonal anti-HA antibody at a 1:2000 dilution in 2% MTBS. Finally, the monoclonal anti-HA antibody was detected with an anti-mouse horseradish peroxidase-conjugated antibody (DAKO) at a 1:2000 dilution in MTBS. The reaction was developed by addition of 3,3Ј,5,5Ј-tetramethyl benzidine (TMB) substrate according to the instructions of the manufacturer (Pierce).
Native Deglycosylation of Human RET ECD and Binding Analysis-To examine the influence of the N-linked carbohydrates attached to the ectodomain of RET, the carbohydrates were enzymatically removed under native conditions. Briefly, 1 g of semipurified HA-tagged RET ECD was deglycosylated by incubation with 20 units of PNGase F overnight at 4°C. Half of the reaction was applied for binding experiments as described above. The remaining portion was taken for Western blot analysis to verify the removal of N-linked carbohydrates.

Preferential Interaction of the Mammalian GDNF-GFR␣-1
Ligand Complex with Human, but Not Xenopus, RET ECD -To evaluate the functional capabilities of wild type and mutant RET ECD molecules, we have developed a solid-phase binding assay using immobilized human GDNF, rat GFR␣1, or the GDNF-GFR␣1 complex as target ligands (18). Recombinant RET ECD molecules were produced as epitope-tagged, soluble proteins in serum-free supernatants of stable transfected CHO cell lines as described previously (18). As shown in Fig. 1A, human RET ECD was able to detect the GDNF-GFR␣1 complex with an EC 50 of ϳ0.2 nM, a value comparable with the binding affinity reported previously using cell-based cross-linking binding assays (10). The human RET ECD did not interact with either GDNF or GFR␣1 alone (Fig. 1A), in agreement with previous observations. Similar to cell-based binding assays (17,21,22), the interaction between the human RET ECD and the GDNF-GFR␣1 complex in our solid-phase binding assay was dependent on Ca 2ϩ , as it was totally abolished in the presence of 1 mM of the Ca 2ϩ -specific chelator EGTA (Fig. 1B). Based on its structural role in the cadherin molecule, the binding of Ca 2ϩ ions to the CLD2/CLD3 interface of the RET ECD is thought to rigidify the relative orientations of these two cadherin-like domains in RET ECD (23,24). The requirement of Ca 2ϩ for the ability of the RET ECD to interact with the GDNF-GFR␣1 complex suggests that residues located on both sides of the Ca 2ϩ binding site between CLD2 and CLD3 may contribute to ligand binding. Despite its overall structural similarity to human RET ECD (17), the Xenopus RET ECD has only 45% amino acid identity to its human counterpart, indicating a significant level of sequence divergence between the two species. In contrast, the intracellular domains of human and Xenopus RET display greater than 85% sequence identity. When compared with human RET ECD , Xenopus RET ECD showed negligible binding to the mammalian GDNF-GFR␣1 complex in the concentration range tested (Fig. 1C), indicating that divergent regions between human and Xenopus RET ECD may represent specific ligand binding determinants.
A Strategy for Homologue-scanning Mutagenesis of the Human RET ECD -The inability of Xenopus RET ECD to interact with the mammalian GDNF-GFR␣1 complex despite its overall structural similarity to the human RET ECD allowed us to use a homologue-scanning mutagenesis approach to study structurefunction relationships in the RET ECD . Based on the subdomain boundaries defined in a previous bioinformatics study on the RET ECD (17), a series of chimeric molecules was constructed by swapping different subdomains between human and Xenopus RET ECD (Fig. 2A). Stable CHO cell lines secreting different epitope-tagged chimeric Xenopus/human RET ECD molecules were generated as described previously (18).
In a previous study, we found that certain subdomains of the RET ECD , mainly CLD1, -2, and -3, have an intrinsic susceptibility to misfolding that makes them particularly vulnerable to inactivating mutations such as those found in patients with Hirschsprung disease (18). Misfolded RET ECD molecules are retained intracellularly in the endoplasmic reticulum and eventually ubiquitinated and degraded, although a fraction may also get access to the extracellular space by direct leakage from the endoplasmic reticulum, particularly after overexpression (18). Folded and misfolded RET ECD molecules can be distinguished by the sensitivity of the latter to Endo H (18). Upon exit from the endoplasmic reticulum, correctly folded glycoproteins lose sensitivity to Endo H as carbohydrates of higher complexity are added in the Golgi complex. The structural integrity of chimeric RET ECD molecules was examined by subjecting the secreted proteins to Endo H treatment as described previously (18). As a control, RET ECD proteins were treated with PNGase F, which removes sugars from both folded and misfolded proteins. As shown in Fig. 2A, all chimeric RET ECD proteins were resistant to Endo H treatment (solid arrowheads) but were still sensitive to PNGase F as expected (open arrowheads). These results indicated that this set of Xenopus/ human chimeric RET ECD molecules was folded correctly.  2. CLD1, -2, and -3 of the human RET ECD are required for binding to the mammalian GDNF-GFR␣1 ligand complex. In A, chimeric RET ECD constructs produced in supernatants of stably transfected CHO cells grown at 30°C were subjected to deglycosylation as indicated. RET ECD proteins were detected with an anti-HA antibody. Deglycosylation-resistant (solid arrowheads) and -sensitive (empty arrowheads) protein species are indicated. All constructs were sensitive to PNGase F digestion, as expected, but were largely resistant to Endo H. The diagram summarizes the chimeric molecules generated and their corresponding nomenclature. h, human; x, Xenopus. B, anti-HA tag blot of CHO cell supernatants showing that chimeric molecules were produced at comparable levels. IB, immunoblot. C, solid-phase binding assay of chimeric RET ECD molecules. Wells were coated with GDNF-GFR␣1-Fc complex (solid bars) or PBS (white bars) and subsequently blocked with low-fat milk. Results were normalized to the binding of wild type human RET ECD . Control denotes supernatant from mocktransfected CHO cells. Shown are means Ϯ S.D. of triplicate observations. D, solid-phase binding assay of chimeric RET ECD molecules. Wells were coated with NTN-GFR␣2-Fc complex (solid bars) or PBS (white bars) and subsequently blocked with low-fat milk. Shown are means Ϯ S.D. of triplicate observations.

CLD1, -2, and -3 of the Human RET ECD Are Required for Binding to the Mammalian GDNF-GFR␣1 Ligand Complex-
The ligand binding activity of chimeric RET ECD proteins was evaluated using a solid-phase binding assay with the GDNF-GFR␣1 complex immobilized to 96-well plates. Chimeric molecules were also tested against a complex formed by human NTN and rat GFR␣2. Equivalent levels of wild type and chimeric proteins were present in these experiments as demonstrated by Western blotting using an antibody against the HA epitope tag present in all the constructs (Fig. 2B). Xenopus RET ECD was not able to interact with the NTN-GFR␣2 complex (data not shown). As shown in Fig. 2, C and D, binding activity to either ligand complex required all three N-terminal cadherin-like domains of RET ECD (i.e. CLD1, -2, and -3) to be of human origin. In contrast, the species of origin of CLD4 and the cysteine-rich domain had no impact on the ligand binding activity of chimeric RET ECD molecules (Fig. 2, C and D). Similar structure-activity profiles were observed toward both GDNF-GFR␣1 and NTN-GFR␣2 complexes, indicating that the RET-ECD interacts with different members of the GDNF and GFR␣ families in a similar fashion. The fact that residues on both sides of the Ca 2ϩ coordination site were required for ligand binding is in agreement with the recognized importance of Ca 2ϩ for stabilizing the RET ECD in a conformation competent for ligand binding and receptor activation (Fig. 1B) (17, 22).

Distinct Clusters of Exposed Residues in CLD1 Are Required for Binding of Human RET ECD to the GDNF-GFR␣1 Complex-
Having established the existence of ligand binding determinants within CLD1, -2, and -3 of the human RET ECD , we set out to define more precisely the location and identity of functionally important residues in these three domains. The GETAREA 1.1 program (www.scsb.utmb.edu/getarea/area_man.html) was used to generate surface accessibility plots of CLD1, -2, and -3 of the human RET ECD using the coordinates of their modeled structures (17) (see the supplemental figure). Segments displaying more than 50% surface accessibility were examined for their degree of sequence similarity to analogous segments in the Xenopus RET ECD . Exposed segments of 6 -12 residues in CLD1, -2, and -3 of the human RET ECD displaying more than 50% divergence from the equivalent Xenopus sequences were targeted for a second round of homologue-scanning mutagenesis (Fig. 3A). A total of 10 chimeric constructs (designated with the Roman numerals I-X) were generated and produced in the supernatants of stable CHO cell transfectants as above (Fig.  3B). With the exception of chimeric protein X, all other chimeras were resistant to Endo H digestion and therefore considered correctly folded (Fig. 3B). Equivalent amounts of wild type human RET ECD and chimeric proteins I-IX (Fig. 3C) were compared for their ability to bind the mammalian GDNF-GFR␣1 complex in a solid-phase based binding assay as above.
FIG. 3. Distinct clusters of exposed residues in CLD1 are required for binding of human RET ECD to the GDNF-GFR␣1 complex. A, alignment of the human and Xenopus RET ECD with conserved residues boxed in black, and surface-exposed, variable regions used to produce chimeric molecules boxed in gray. hRet, human RET; xRet, Xenopus RET. In B, chimeric RET ECD constructs produced in supernatants of stably transfected CHO cells grown at 30°C were subjected to deglycosylation as indicated. RET ECD proteins were detected with an anti-HA antibody. Deglycosylation-resistant (solid arrowheads) and -sensitive (empty arrowheads) species are indicated. All constructs were sensitive to PNGase F digestion, as expected. Constructs I-IX were largely resistant to Endo H, whereas construct X was sensitive, indicating a misfolded protein. The diagram summarizes the chimeric molecules generated and their corresponding nomenclature. C, anti-HA tag blot of CHO cell supernatants showing that wild type hRET ECD and chimeric molecules I-IX were produced at comparable levels. IB, immunoblot. D, solid-phase binding assay of chimeric RET ECD molecules. Wells were coated with GDNF-GFR␣1-Fc complex (solid bars) or PBS (white bars) and subsequently blocked with low-fat milk. Results were normalized to the binding of wild type human RET ECD . Control denotes supernatant from mock-transfected CHO cells. Shown are means Ϯ S.D. of triplicate observations. As shown in Fig. 3D, chimeras I, III, IV, V, and VI all displayed a reduction in binding as compared with the wild type human RET ECD .
Chimeric protein I carries 4 amino acid replacements as compared with wild type human RET ECD , namely S32L, D34K, A35D, and W37Y, suggesting that one or more of those positions are crucial for the interaction of the RET ECD with the GDNF-GFR␣1 complex. Based on the modeled structure of CLD1 (17), residues in region III are predicted to form part of a loop between the C and D ␤-strands of this domain. This exposed segment is shorter and highly divergent in the Xenopus RET ECD , indicating that residues in this loop region may also contribute to the binding of the human RET ECD to the GDNF-GFR␣1 complex.
In contrast to chimeras I, III, and V, chimeric protein IV retained ϳ20% binding to the GDNF-GFR␣1 complex (Fig. 3D). Of the 8 amino acid exchanges in this region, three involve the replacement of positively charged residues by uncharged residues (Fig. 3A), indicating that one or more of those charges are important for formation of the GDNF-GFR␣1-RET complex. The complete absence of detectable binding in chimeric protein V indicates that one or more of the residues exchanged in this region are important for ligand binding. Of the 9 amino acid differences between the human and Xenopus sequences in this region, R133L and W139N may be the most significant ones as they involve residues that are enriched in known protein-protein interfaces (25). Finally, a 60% reduction in ligand binding was observed in chimeric protein VI (Fig. 3D), which involves 9 amino acid replacements in CLD2 (Fig. 3A). Of note, this was the only set of mutations outside CLD1 that affected the interaction of the RET ECD with the GDNF-GFR␣1 complex. Interestingly, three of the exchanges in this region involve the replacement of two polar and one uncharged residue by positively charged residues (Fig. 3A). Taken together, they indicate that although CLD1, -2, and -3 of human RET ECD are all required for ligand binding, the most important determinants appear to be concentrated in CLD1, the most N-terminal subdomain of the human RET ECD .
N-linked Carbohydrates in Human RET ECD Are Dispensable for Ligand Binding-Protein glycosylation can have a modulatory effect on protein-protein interactions, and introduction of N-linked glycosylation sites has been used as a mutagenesis strategy (26). The human RET ECD is abundantly N-glycosylated, but the role of this post-translational modification in ligand binding is unknown. Interestingly, predicted N-glycosylation sites in the human RET ECD are not evenly distributed, but the majority (9 of 12) of them appear downstream of the Ca 2ϩ coordination site, in accord with the location of ligand binding determinants in the CLD1. We treated the human RET ECD with PNGase F under native conditions and examined its ability to bind to the GDNF-GFR␣1 complex. Treatment with PNGase F resulted in complete deglycosylation of the native protein, comparable with that obtained after prior denaturation (Fig. 4A). As shown in Fig. 4B, RET ECD incubated in deglycosylation buffer with or without PNGase F without prior denaturation showed no loss of binding as compared with a non-treated control. Denaturation prior to enzymatic deglycosylation resulted in complete loss of binding, as expected (Fig.  4B). Thus, N-linked carbohydrates are unlikely to play a role in the assembly of the RET-GDNF-GFR␣1 complex and may instead be of importance for the folding and maturation of the RET ECD in the secretory pathway, as suggested previously (18). DISCUSSION The receptor tyrosine kinase RET has remained in the limelight ever since its discovery as a transforming protein in 1987 (27) and the subsequent elucidation of its participation in the receptor complex for GDNF family ligands (10,28,29). Despite having an extracellular domain of more than 600 residues, RET cannot engage any of these ligands directly but requires the auxiliary GFR␣ receptors for activation (29,30). Although the RET ECD was known to make direct contacts with both GDNF ligands and GFR␣ receptors in the complex, binding determinants in the RET ECD molecule and the overall architecture of the complex remained to be characterized.
In this study, we set out to identify functional determinants in the RET ECD responsible for its association with the GDNF-GFR␣1 complex using a homologue-scanning mutagenesis approach based on the differential abilities of human and Xenopus RET ECD to interact with ligand complexes of mammalian origin. Using this approach, binding determinants were found to be concentrated in the N-terminal CLD1 of the human RET ECD . Within this region, three discrete segments, ranging from 6 to 12 residues (i.e. I, III, and V in Fig. 3A), could not be replaced by equivalent sequences from Xenopus RET ECD without complete loss of activity, indicating that these segments are required for the binding of the human RET ECD to the mammalian GDNF-GFR␣1 complex. Importantly, these replacements had no detectable effects on protein production, stability, secretion, or folding, at least at 30°C. Because of their relatively high solvent accessibility, these epitopes are likely to be directly involved in the interaction of RET with its ligands. When visualized on the modeled three-dimensional structure of the CLD1 (17), regions I, III, and V are all localized on the same face of the model (Fig. 5A), delineating a probable surface for interaction with the GDNF-GFR␣1 complex. In support of this notion, the only site of N-glycosylation known in the human CLD1, namely Asn-98, 2 is located in the opposite side of the 2 S Kjaer, unpublished observations. domain, a position that is sterically compatible with the proposed location of the ligand binding interface.
Bioinformatics analysis of the RET ECD has recently indicated a structural organization resembling that of classical cadherins with four cadherin-like domains followed by a Cterminal cysteine-rich domain not related to cadherin sequences (17). The x-ray crystal structure of the complete extracellular domain of C-cadherin has recently been solved, revealing an elongated rod-shaped structure (24). Two types of interactions between different cadherin molecules could be defined in these crystals: cis interactions were formed laterally between adjacent cadherin molecules, whereas trans interactions linked cadherin molecules in opposite orientations, presumably representing the kind of interactions responsible for cell-cell contact. The trans interface was defined by a conserved tryptophan side chain (Trp-2) at the N-terminal end of the cadherin molecule from one cell, which was shown to insert into a hydrophobic pocket in the cadherin molecule from the opposing cell (24). The importance of Trp-2 for cadherin-mediated cell adhesion had been inferred independently from structurefunction analyses (31). However, Trp-2 is not conserved in RET ECD sequences from different organisms, and no functionally analogous residues can be identified in the modeled structure of the RET ECD (17). Moreover, no adhesive function has to date been attributed to RET molecules (32).
Given the pivotal role played by the CLD1 in the interaction of RET with its ligands, a straight rod-like organization similar to that of classical cadherins would place the major ligand binding site in the RET ECD away from the plasma membrane, where the membrane-anchored GDNF-GFR␣1 complex is likely to be. In fact, structure-function studies of the GFR␣1 molecule have indicated that the N-terminal domain of this receptor is dispensable for ligand binding and have instead localized the binding determinants toward the middle and Cterminal portions of the molecule (12), in agreement with a membrane-proximal site of complex assembly. How could these apparently contradictory observations be reconciled?
An analogous topological problem has been posed by the cytokine receptor complex formed by gp130, IL-6R␣, and IL-6. Like RET, gp130 has a large multidomain extracellular region, and ligand binding determinants have been localized to the three most N-terminal domains (33). On the other hand, IL-6R␣, like GFR␣1, has ligand binding determinants in its membrane-proximal domains (34). A recent crystallographic analy-sis of the assembly of this complex revealed a bent structure for gp130 and IL-6R with their ligand binding domains forming a "table" that rests on "legs" composed by the more C-terminal domains of the molecules (35). In an analogous fashion, we hypothesize that a bent arrangement of the RET ECD would allow its ligand binding domain to reach the GDNF-GFR␣1 complex located closer to the membrane (Fig. 5B). It is worth noting that this arrangement would not be possible had the RET ECD retained all the interdomain Ca 2ϩ binding sites that characterize classical cadherins. As mentioned previously, the single Ca 2ϩ binding site in the RET ECD is located at the interface of CLD2 and CLD3, suggesting a straight and rigid conformation for these two domains. On the other hand, the lack of Ca 2ϩ binding sites in all of the other interdomain regions suggest a greater degree of flexibility as compared with classical cadherins that may allow the RET ECD to bend toward the cell membrane. Clearly, validation of this hypothesis awaits the structural determination of the RET ECD in complex with GFR␣1 and GDNF.