Functional mapping of receptor specificity domains of glial cell line-derived neurotrophic factor (GDNF) family ligands and production of GFRalpha1 RET-specific agonists.

The glial cell line-derived neurotrophic factor (GDNF) family ligands (GFLs) (GDNF, neurturin, artemin, and persephin) are critical regulators of neurodevelopment and support the survival of midbrain dopaminergic and spinal motor neurons in vitro and in animal disease models making them attractive therapeutic candidates for treatment of neurodegenerative diseases. The GFLs signal through a multicomponent receptor complex comprised of a high affinity binding component (GDNF-family receptor alpha-component (GFRalpha1-GFRalpha4)) and the receptor tyrosine kinase RET. To begin characterization of GFL receptor specificity at the molecular level, we performed comprehensive homologue-scanning mutagenesis of GDNF, the prototypical member of the GFLs. Replacing short segments of GDNF with the homologous segments from persephin (PSPN) (which cannot bind or activate GFRalpha1.RET or GFRalpha2.RET) identified sites along the second finger of GDNF critical for activating the GFRalpha1.RET and GFRalpha2.RET receptor complexes. Furthermore, introduction of these regions from GDNF, neurturin, or artemin into PSPN demonstrated that they are sufficient for activating GFRalpha1. RET, but additional determinants are required for interaction with the other GFRalphas. This difference in the molecular basis of GFL-GFRalpha specificity allowed the production of GFRalpha1. RET-specific agonists and provides a foundation for understanding of GFL-GFRalpha.RET signaling at the molecular level.

The glial cell line-derived neurotrophic factor (GDNF) 1 family ligands (GFLs) are distant members of the TGF-␤ superfamily that are potent neurotrophic factors in vitro and are critical for the development of distinct neuronal populations in vivo (1). There are currently four known GFLs, GDNF (2), neurturin (NRTN (3)), artemin (ARTN (4)), and persephin (PSPN (5)). Each GFL is a demonstrated survival factor for dopaminergic ventral midbrain neurons cultured from the embryo (2, 4 -6). Furthermore, GDNF, NRTN, and PSPN have also been shown to support the survival of motor neurons in culture (5,(7)(8)(9). Therefore there is considerable interest in the therapeutic potential of these molecules for use in treating neurodegenerative diseases such as Parkinson's disease and amyotrophic lateral sclerosis, which is underscored by the success of GDNF in relieving disease symptoms in several animal models of Parkinson's disease (10). However the difficulty of delivering these large polypeptides to their intended site of action in the central nervous system makes it desirable to understand the molecular basis of their action to aid in attempts to produce minimal agonists of the GFL receptor system.
The GFLs signal through a unique multicomponent receptor complex, consisting of a high affinity glycosylphosphatidylinositol-anchored binding component (GFR␣1-GFR␣4) and the receptor tyrosine kinase RET. Extensive in vitro and in vivo experimentation has supported the hypothesis that for each GFL there is a preferred GFR␣ receptor, to which the GFL binds with highest affinity and most potently activates RET. These preferred interactions are GDNF-GFR␣1, NRTN-GFR␣2, and ARTN-GFR␣3 (4,9,11,12). PSPN does not bind or activate any of the mammalian GFR␣s, but does bind to a protein currently only identified in chicken called GFR␣4 (5,13,14). However, despite these preferential interactions, there is also clear cross-talk between the different ligand-receptor pairs. The known alternative interactions are NRTN-GFR␣1, ARTN-GFR␣1, and GDNF-GFR␣2 (4,11,15,16). As GFR␣1 and GFR␣2 are often expressed by the same or closely neighboring neurons (6,17), it is still unclear through which receptor exogenously administered GDNF and NRTN mediate their neurotrophic actions on various neuronal populations.
Whereas the in vitro interactions between the different GFLs and GFR␣s are now relatively well understood, the molecular basis of this specificity and cross-talk is unknown (18). Furthermore, the precise details of receptor complex formation and stoichiometry are also poorly characterized. The crystal structure of GDNF indicates that it is a disulfide-bonded dimer that is significantly similar to the structure of TGF-␤2, as predicted by the cysteine spacing of its primary sequence (19 -21). However, the structure itself yields only speculative information regarding receptor binding surfaces. Furthermore, analogy to other TGF-␤ superfamily members regarding receptor binding surfaces would likely be unfounded as the receptors used by GDNF and the TGF-␤s are drastically different and likely hold no structural similarity.
To characterize the structural requirements of GDNF for interacting with its receptors, we performed homologue-scanning mutagenesis, replacing short blocks (ϳ8 amino acids) of GDNF sequence with the homologous blocks from PSPN, which cannot bind or activate GFR␣1⅐RET or GFR␣2⅐RET. This screen identified two critical regions for the GDNF-GFR␣1 interaction and at least one additional region critical for the alternate GDNF-GFR␣2 interaction. These blocks are discontinuous along the primary sequence of GDNF but are directly adjacent along the second finger of GDNF when mapped to the crystal structure. We further demonstrate that these two regions from any of the GFR␣1 agonists (GDNF, NRTN, or ARTN) are sufficient to activate GFR␣1⅐RET in the context of the PSPN molecule and that these chimeric mutants function as GFR␣1⅐RET-specific agonists in vitro. Finally, we identify an additional region that is critical for the NRTN-GFR␣2 and ARTN-GFR␣3 interactions. These data identify a putative receptor interaction site shared by the GFLs to activate GFR␣1⅐RET and identify additional determinants required for GDNF and NRTN to activate GFR␣2⅐RET and for ARTN to activate GFR␣3⅐RET.

EXPERIMENTAL PROCEDURES
Mutagenesis, Synthesis, and Purification of GFL Mutants-All mutants were produced by fusion polymerase chain reaction mutagenesis. The amino acid sequence of rat GDNF, mouse NRTN and PSPN, and human ARTN was used for all constructs. Polymerase chain reaction products were directly cloned into the MluI and XbaI sites of plasmid pCB6 (22), and the inserts were sequenced entirely. For mutant G-hf-GDNF the 6His-FLAG tag was inserted between the third and fourth residue of mature GDNF, making the sequence (N-. . . RLKR SPD-HHHHHHDYKDDDD-KQAL . . . -C). For mutant N-hf-GDNF, residues 1-38 were truncated from mature GDNF and attached to the preproregion of NRTN. The resulting sequence was (N-. . . RRAR PGA-HH-HHHHDYKDDDDK-RGCV . . . -C). For N-hf-ARTN, residues 1-13 were truncated from mature human ARTN and attached to the preproregion of NRTN, with the resulting sequence (N-. . . RRAR PGA-HH-HHHHDYKDDDDK-RGCR . . . -C). For mutant P-hf-PSPN, the tag was inserted between the third and fourth residue of mature PSPN, resulting in N-. . . RLPR ALA-HHHHHHDYKDDDDK-GSCR . . . -C. Expression plasmids for homologue-scanning mutants were produced as above by fusion polymerase chain reaction, and all were based on the N-hf-GDNF construct.
For mutant protein production in COS cells, expression plasmids were transfected using the DEAE-dextran/Chloroquine method (23). COS cells were plated onto 10-or 15-cm dishes, transfected, and switched to Dulbecco's modified Eagle's medium containing 1% fetal calf serum. After 4 -5 days, conditioned medium was collected, cleared, and either concentrated using Centriprep-10 concentrators (Amicon) or purified by nickel chromatography (Qiagen). Proteins were visualized by immunoblotting with anti-FLAG M2 monoclonal antibody (Sigma). Relative quantities of FLAG-tagged proteins were determined by an enzyme-linked immunosorbent assay also using the anti-FLAG M2 antibody. Briefly, purified proteins or conditioned medium were coated to Nunc-Immuno MaxiSorp microtiter plates overnight at 4°C. Plates were washed (3ϫ in Tris-buffered saline, 0.03% Tween 20), blocked (blocking solution: Tris-buffered saline, 1% bovine serum albumin) for 1 h at 25°C, washed again (5ϫ), and then incubated with a 1:500 dilution of anti-FLAG M2 antibody in blocking solution for 1.5 h at 25°C. After another wash (5ϫ), plates were incubated with horseradish peroxidase-conjugated anti-mouse antibodies diluted 1:5000 in blocking solution. Finally, wells were washed 5ϫ, and the presence of horseradish peroxidase was assayed by the addition of the chromogenic substrate 3,3Ј,5,5Ј-tetramethylbenzidine for 5 min. The reaction was stopped by adding an equal volume of 0.5 M H 2 SO 4 , and color was measured in a plate reader at 450 nm. The absolute concentration of GDNF in purified and conditioned medium samples was determined for the G-hf-GDNF construct using the GDNF E-MAX enzyme-linked immunosorbent assay kit following the manufacturer's instructions (Promega).
Receptor Activation, Complex Formation, Cell Proliferation, and Differentiation Assays-Receptor activation assays were performed as described previously (4). RET-3T3 cells were plated at 85,000 cells/well in 12-well plates and transfected using Superfect (Qiagen) with the reporter plasmids (250 ng/well Gal4-Luc, 50 ng/well Gal4-Elk), CMV-lacZ (50 ng/well) for transfection normalization, a CMV-GFR␣ (250 ng/well) expression plasmid, a mutant GFL construct (250 ng/well), and 650 ng/well pBluescript as a carrier for a total of 1.5 g of DNA/well. Cells were switched to 0.5% serum-containing medium the morning after transfection and harvested 36 h later. The average luciferase activity of duplicate or triplicate samples was normalized to ␤-galactosidase ac-tivity of the cotransfected lacZ reporter. For the receptor complex formation assay, Neuro2a cells were transiently transfected using Superfect with an expression plasmid for rat GFR␣1 with a FLAG tag inserted by polymerase chain reaction mutagenesis after the signal sequence. Cells were treated with 25 ng/mL of the indicated factor for 6 h; the cells were washed once with cold phosphate-buffered saline, lysed, and immunoprecipitated using an anti-FLAG M2 monoclonal antibody conjugated to agarose (Sigma). After washing, the immunoprecipitated samples were visualized by immunoblotting with an anti-RET antibody (C-9, Santa Cruz Biotechnology). The NBL-S proliferation assay and the SH-SY5Y differentiation assay were performed as described (4).
Primary Cultured Neuron Survival Assays-Rat cerebellar granule cell dissection and culture was performed as described previously (24). Timed pregnant Harlan Sprague-Dawley rats were purchased from Harlan Sprague Dawley (Indianapolis, IN). At postnatal day 7, cerebella were dissected, cut into 1-mm pieces, and incubated for 15 min in 0.3 g/ml trypsin (Worthington) at 37°C. The pieces were triturated with a fire-polished Pasteur pipette in the presence of trypsin, and the resulting cell suspension was passed through a Nitex filter (size 3-20/ 14; Tetko, Elmsford, NY). Cells were plated at a density of 2.3 ϫ 10 5 cells/cm 2 in 4-well dishes (Nunc) coated with 0.1 mg/ml poly-L-lysine. Plating medium (K25 ϩS) consisted of basal medium Eagle (Life Technologies, Inc.) containing 10% dialyzed fetal bovine serum, 20 mM KCl, 100 unit/ml penicillin and 100 g/ml streptomycin. To reduce the number of non-neuronal cells, 3.3 g/ml aphidocholine was added to the medium 24 h after plating.
Granule cell transfections were performed using a modified calcium phosphate protocol as described previously (25). Expression plasmids for GFR␣1, GFR␣2, and RET are described elsewhere (11). At 5 days in vitro, medium was replaced with Dulbecco's modified Eagle's medium (Life Technologies, Inc.) for 1 h. During this time, an equal volume of solution containing 0.25 M CaCl and 67 g/ml DNA was added to a 2ϫ HEPES-buffered saline (274 mM NaCl, 10 mM KCl, 1.4 mM Na 2 HPO 4 -7H 2 O (Fisher), 15 mM dextrose, and 42 mM HEPES (free acid), pH 7.07) and incubated in the dark at room temperature for 25 min. 30 L of the precipitate (1 g of DNA) were added to each 4-well dish and incubated at 37°C for 1 h. Cells were washed twice with Dulbecco's modified Eagle's medium and then returned to plating medium. Transfection efficiency was approximately 0.2-0.5%.
To quantify transfection results, the number of initial EGFP-positive cells in a defined area (minimum of 150 cells) of two to four wells (of a 4-well dish) per condition were counted 24 h after transfection, then rinsed the cultures twice in Dulbecco's modified Eagle's medium and switched them to high potassium/serum (K25ϩS), low potassium/no serum (K5-S) medium alone, or K5-S supplemented with the indicated factors for 48 h. After this period, the number of enhanced green fluorescent protein-positive cells remaining in the originally defined area of the well was counted again to obtain the percentage neuronal survival.
Sequence Analysis and Molecular Visualization-Amino acid sequence comparison and alignment was performed using the MegAlign program of the DNAstar software package. Molecular modeling and analysis was done using the Swiss PDB viewer (26), and models were rendered using Mole software (Applied Thermodynamics). The GDNF structure was modeled after entry 1AGQ (21) of the Protein Data Bank. The "C/D" dimer from 1AGQ is shown, with the missing segment of monomer "C" built as determined for monomer "D."

RESULTS
The N-terminal Extension of GDNF Is Not Required for Activity-An alignment of the mature rat GDNF with mouse NRTN, ARTN, and PSPN is shown in Fig. 1. Rat GDNF was the basis for all mutagenesis in this study because its structure has been determined (21). The largest notable difference between the GFLs is the N-terminal extension before the first structural cysteine, which varies from 40 amino acids in GDNF to only 5 amino acids in NRTN and PSPN, and shows little similarity between the different family members. Furthermore GDNF, NRTN, and ARTN have multiple RXXR consensus subtilisin-like proprotein convertase cleavage sites (27), and therefore multiple isoforms of these extensions are possible. The ones shown are consistent with N-terminal sequencing of GDNF and NRTN (2, 3), the single cleavage site in PSPN (5), and a cleavage site conserved between mouse and human ARTN (4).
To determine if the large N-terminal extension of GDNF is required for its activity, we took advantage of the differential processing of GDNF and NRTN by mammalian cells. Chimeric constructs with the prepro-region from NRTN attached to GDNF were generated, and we assessed both their processing and their ability to activate the GFR␣1⅐RET receptor complex (Fig. 2). Tandem 6X histidine and FLAG tags were inserted after the RXXR cleavage site of GDNF or NRTN to allow purification and tracking of the proteins. When expressed in COS cells, similar quantities of tagged GDNF with its own pro-region (G-hf-GDNF) or the NRTN pro-region (N-hf-GDNF) could be purified from conditioned medium using nickel chromatography (Fig. 2B). Only species corresponding to the expected processing events were observed in the medium, indicating that the pro-domains of NRTN and GDNF are sufficient to direct proper processing of their different N-terminal extensions.
To assess the ability of these proteins to activate the high affinity GDNF receptor, GFR␣1⅐RET, we utilized the Gal4-Elk1/Gal4-luciferase reporter system as described previously (4). This system monitors the level of RET activation of the mitogen-activated protein kinase pathway, which induces the Gal4-Elk1 fusion protein to activate the Gal4-luciferase reporter, giving a facile measure of receptor activation (28,29). The reporter plasmids, together with expression plasmids for GFR␣1 and the indicated GDNF construct, were cotransfected into fibroblasts that stably express RET (Fig. 2C). Subsequently, the steady-state level of GFR␣1⅐RET activation was measured after two days of autocrine ligand stimulation by the cells. Wild-type GDNF, G-hf-GDNF, and N-hf-GDNF all showed comparable activation of GFR␣1⅐RET in this assay, whereas tagged PSPN constructs with either the PSPN or NRTN pro-region did not activate the receptor. These results are consistent with previous receptor activation and binding experiments indicating that GDNF, but not PSPN, can bind and activate RET through GFR␣1 (4,5). Furthermore, these data indicate that the N-terminal extension of GDNF is not required for its ability to activate the GFR␣1⅐RET receptor complex, and the location of the epitope tag does not interfere with the function of GDNF.
Homologue-scanning Mutagenesis Identifies Regions of GDNF Critical for Interacting with GFR␣1 and GFR␣2-Although the crystal structure of GDNF is known, the molecular determinants for GDNF-GFR␣ receptor binding and specificity are currently undefined. To identify regions of GDNF that are required for its ability to activate the GFR␣1⅐RET receptor complex, we utilized homologue-scanning mutagenesis together with the receptor activation assay described above. Ten GDNF mutants (termed "GPG" mutants) were generated by replacing blocks of nonconserved sequence from PSPN into GDNF (Fig. 3, see Fig. 1 for sequence changes). This method allows comprehensive mutagenesis of sites responsible for the difference in the ability of GDNF and PSPN to activate GFR␣1⅐RET and is highly likely to maintain structural integrity of the mutants because the replacements are from a homologous (and likely structurally similar) protein. All mutants were produced at similar quantities when transfected into COS cells, and the processing of each appeared normal (Fig. 3B).
Comparison of the GPG mutants with GDNF revealed that although many mutants showed slightly decreased activity, only one (GPG-F2a) completely lost the ability to activate the GFR␣1⅐RET receptor (Fig. 4A). This region maps to the bottom of the GDNF molecule along the second finger (Fig. 4B) (21). The only other notable mutant that consistently showed activity 60% or lower than full-length GDNF was GPG-F2c, which FIG. 1. Primary sequence alignment of the mature GFLs. A, rat GDNF is shown with mouse NRTN, ARTN, and PSPN. Secondary structure elements ("␣" for ␣-helix, "␤" for ␤-strand, and 3 for 3 10 -helix) as determined for GDNF (21) are indicated above the alignment. The regions delineated for homologue-scanning mutagenesis are in colored blocks, with the name of the region ("F1a-b" for finger 1, "Ha-d" for heel, and "F2a-d" for finger 2) below the given block. The color of each block corresponds to the color scheme shown in B. B, representation of backbone of the GDNF dimer showing the first (blue) and second (red) fingers and the heel region (green) of the molecule. maps to the strand adjacent to the region F2a along the bottom of the molecule.
It is well established that GDNF is also capable of binding and activating the GFR␣2⅐RET receptor, albeit with a lower affinity than NRTN (11,15,16), whereas PSPN cannot (5,13). Therefore, to compare the molecular determinants of GDNF required for GFR␣1 versus GFR␣2 interaction, we tested the ability of this same panel of mutants to activate the GFR␣2⅐RET receptor complex. Several of the GPG mutants were significantly more attenuated in their ability to activate RET through GFR␣2 than they were through GFR␣1 (Fig. 4C). As before, mutant GPG-F2a was unable to activate the receptor. Interestingly, mutant GPG-F2c, which was the second most affected mutant in activating GFR␣1⅐RET, showed essentially no activation of GFR␣2⅐RET. Finally, the mutant that showed the most contrast in its ability to activate GFR␣1⅐RET versus GFR␣2⅐RET was mutant GPG-Ha, which was severely affected only in its ability to activate GFR␣2⅐RET. Interestingly, mapping these three regions to the GDNF crystal struc-ture reveals that they form a continuous surface, with residues from F2a and F2c from one monomer directly adjacent to region Ha from the heel of the other monomer, suggesting a possible receptor-ligand interaction surface (Fig. 4D). In summary, the above mutagenesis data suggest that two adjacent regions of GDNF located along the second finger are necessary for interaction with both GFR␣1 and GFR␣2 (regions F2a and F2c) and that additional regions appear to be required for GDNF interaction with GFR␣2.
Identification of Regions F2a and F2c from GDNF, NRTN, or ARTN as Sufficient for Activating the GFR␣1⅐RET Receptor Complex-The homologue-scanning screen above suggests several regions that are necessary for the full ability of GDNF to activate the GFR␣1⅐RET receptor. However, using only loss of function analysis it is difficult to assess whether critical receptor-contact residues are being altered directly or if the mutations induce structural changes that alter receptor contact surfaces elsewhere on the molecule. To identify if any of the critical regions identified above (either alone or in combination) are sufficient to allow binding and activation of the GFR␣1⅐RET receptor complex, we generated PSPN mutants with the putative critical regions from GDNF replacing the corresponding regions of PSPN. We then determined if these "PGP" mutants had gained the ability to activate the GFR␣1⅐RET receptor complex (Fig. 5).
We focused initially on two candidate regions, F2a and F2c . GDNF with a 6-histidine and FLAG tag inserted two residues after the RXXR cleavage site is designated "G-hf-GDNF." To generate GDNF lacking the N-terminal extension, a chimera was produced with the NRTN prepro-region attached to GDNF truncated two residues upstream of the first structural cysteine. B, production of tagged GDNF proteins in COS cells. The indicated construct was transiently transfected into COS cells; recombinant protein was purified from conditioned medium using nickel chromatography and visualized on an anti-FLAG immunoblot. Monomeric G-hf-GDNF ran at ϳ25 kDa, and N-hf-GDNF at ϳ21.5 kDa, consistent with processing at the predicted RXXR cleavage sites from GDNF and NRTN. Therefore N-hf-GDNF produces GDNF with the 40-amino acid N-terminal extension replaced by two amino acids and the tandem His-FLAG tags. C, comparison of GFR␣1⅐RET receptor activation by wild-type and tagged GDNF constructs. RET-expressing fibroblasts were cotransfected with GFR␣1, the reporter plasmids (Gal4-Elk1 and Gal4-luciferase), and the indicated GDNF expression construct. Reporter activation was measured after 36 h of autocrine ligand stimulation by the cells and reported as fold activation over the control condition. Mean and S.D. are shown for duplicates from a representative experiment. Tagged full-length and truncated GDNF showed similar receptor activation to wild-type GDNF, whereas tagged PSPN with its own (P-hf-PSPN) or NRTNЈs pro-region (N-hf-PSPN) does not activate GFR␣1⅐RET.

FIG. 3. Homologue-scanning mutagenesis of GDNF.
A, schematic representation of mature region from GDNF, PSPN, and the homologue-scanning mutants. Mutant GPG-Hc is a deletion of amino acids 88 -92 of mature GDNF to correspond with the lack of these residues in PSPN. The sequence replacements in these mutants are shown in Fig. 1. B, anti-FLAG immunoblot of total cell lysates from COS cells transfected with the different homologue scanning mutant constructs. Mutants were produced in similar quantities. The two major bands indicated by arrows represent the unprocessed and mature forms of the mutants. The lower band corresponds in size to the secreted product (see Fig. 2B). The smaller size of PSPN than N-hf-GDNF reflects the fact that GDNF is glycosylated (2), whereas PSPN is not. along the second finger of GDNF, because mutant GPG-F2a was entirely inactive, and mutant GPG-F2c was the only other mutant that consistently showed decreased activation of GFR␣1⅐RET. Replacement of either region F2a or F2c alone from GDNF into PSPN was not sufficient to allow the resulting mutants (PGP-F2a or PGP-F2c) to activate GFR␣1⅐RET. However, when both regions were placed into PSPN the resulting mutant (PGP-F2ac) gained the ability to activate the GFR␣1⅐RET receptor at a level comparable to full-length GDNF (Fig. 5B). Furthermore, treatment of GFR␣1-transfected Neuro2a cells (which endogenously express Ret) with either GDNF or mutant PGP-F2ac induced comparable GFR␣1⅐RET complex formation, demonstrating that PGP-F2ac had gained the ability to bind and induce receptor complex formation (Fig. 5C). To determine if mutant PGP-F2ac was also able to elicit biological responses comparable to GDNF, we tested its ability to induce proliferation in the NBL-S neuroblastoma cell line, which expresses GFR␣1 and RET and responds to GFR␣1⅐RET agonists (GDNF, NRTN, ARTN) by proliferation (Fig. 5D) (4). Mutant PGP-F2ac purified from transiently transfected COS cell-conditioned medium induced robust proliferation of NBL-S cells, similar to GDNF at all doses, whereas PSPN did not. Furthermore, PGP-F2ac was also able to induce neurite outgrowth in the GFR␣1⅐RET-expressing neuroblastoma line SH-SY5Y similar to GDNF (Fig. 5E) (4). NRTN and ARTN are also able to activate GFR␣1⅐RET (4,30), and therefore we examined if the same regions (F2a and F2c) from these two additional GFR␣1 agonists are also sufficient to activate the GFR␣1⅐RET receptor (Fig. 6). Interestingly, the corresponding PSPN mutants with regions F2a and F2c from NRTN and ARTN were also capable of activating the GFR␣1⅐RET receptor complex, at levels comparable to GDNF and PGP-F2ac (Fig. 6B). This indicates that elements of regions F2a and F2c from all the known GFR␣1 agonists are sufficient to activate GFR␣1⅐RET when placed in the context of PSPN. Furthermore, these regions appear to be sufficient only for GFR␣1⅐RET activation, as the same mutants did not activate GFR␣2⅐RET comparably to GDNF (Fig. 6C). Mutants PNP-F2ac and PGP-F2ac did elicit low level activation of GFR␣2⅐RET, whereas PAP-F2ac did not. This is consistent with the ability of GDNF and NRTN, but not ARTN, to activate GFR␣2⅐RET (4) and indicates that regions F2a and F2c are also FIG. 4. Receptor activation analysis of homologue-scanning mutants. A, GFR␣1⅐RET receptor activation by homologue-scanning GDNF mutants. Values are represented as a percentage of reporter activation by GDNF. The mean and standard deviation of triplicate measurements from a representative experiment are shown. Similar results were obtained from three other experiments. Mutant GPG-F2a was severely attenuated in its ability to activate GFR␣1⅐RET. All other mutants showed slightly decreased activity at 70 -80% of GDNF. The only other mutant that consistently showed receptor activation below 60% of GDNF was GPG-F2c. Arbitrary boundary lines at 33 and 66% of wild-type GDNF activation are designated in color to simplify visualization in B. All mutants in the yellow and red regions were significantly less active than full-length GDNF (p Ͻ 0.05, Student's paired t test). B, location of putative critical GFR␣1⅐RET interaction domains identified in A on a space-filling model of the GDNF crystal structure (21). Regions are colored green above 66%, yellow below 66%, and red below 33% of GDNF activation. The upper representation is a side view of the GDNF dimer, and the lower representation is from a viewpoint below the dimer. C, GFR␣2⅐RET receptor activation by homologue-scanning mutants. The mean and standard deviation of duplicate measurements from a representative experiment are shown. Similar results were obtained from three other experiments. As in A, all mutants in the yellow and red regions were significantly less active than full-length GDNF (p Ͻ 0.05, Student's paired t test). Mutants GPG-F2a and GPG-F2c were both severely attenuated in GFR␣2⅐RET activation, similar to control and PSPN. Three additional mutants were more attenuated in their ability to activate GFR␣2⅐RET than GFR␣1⅐RET (F1b, Hb, Hc). Finally, mutant GPG-Ha was significantly less effective at activating GFR␣2⅐RET (ϳ27%) than GFR␣1⅐RET (ϳ80%). D, location of putative critical GFR␣2⅐RET interaction domains identified in C. Color scheme and viewpoints of structure are same as for B.
involved in binding and activating GFR␣2⅐RET but that additional regions are required for full activation.
Additional Determinants Are Critical for NRTN to Activate GFR␣2⅐RET and for ARTN to Activate GFR␣3⅐RET-The observation that mutants PGP-F2ac and PNP-F2ac cannot fully activate GFR␣2⅐RET is consistent with the homologue-scanning mutagenesis above (Fig. 4), which suggested that regions in addition to F2a and F2c are required for GDNF to activate GFR␣2⅐RET. In particular, mutant GPG-Ha was significantly attenuated in its ability to activate GFR␣2⅐RET (ϳ27%), whereas it showed a minor loss in the ability to activate GFR␣1⅐RET (ϳ80%, Fig. 4). To address whether region Ha contains additional molecular determinants required for activating GFR␣2⅐RET, we generated a PSPN mutant with regions F2a, F2c, and Ha replaced by the corresponding regions from NRTN, the highest affinity GFR␣2 agonist (PNP -Ha/F2ac, Fig.  7A). Whereas mutant PNP-F2ac elicited only a minor increase in GFR␣2⅐RET activation as before, mutant PNP-Ha/F2ac was significantly better at activating GFR␣2⅐RET; however, it did not restore full activity (usually 70 -80% of GDNF, Fig. 7B). Therefore, consistent with the homologue-scanning mutagenesis above, these data indicate that regions F2a and F2c of GDNF and NRTN are only sufficient to activate GFR␣1⅐RET, and additional regions are required for full activation of GFR␣2⅐RET (region Ha and perhaps others).
ARTN is a recently identified member of the GFLs that can activate GFR␣1⅐RET, and the only member of the family that can activate GFR␣3⅐RET (4). As shown above, regions F2a and F2c from ARTN are sufficient to activate GFR␣1⅐RET. In an initial attempt to characterize the molecular determinants of the ARTN-GFR␣3 interaction, we examined the ability of PSPN-ARTN chimeras to activate the GFR␣3⅐RET receptor complex (Fig. 7, C and D). As expected, N-hf-ARTN gave robust activation of GFR␣3⅐RET, whereas N-hf-GDNF showed no activity. Mutant PAP-F2ac, which is capable of activating GFR␣1⅐RET (see Fig. 6), was also unable to activate FIG. 5. GDNF/PSPN chimeras identify regions F2a and F2c from GDNF as sufficient for GFR␣1⅐RET activation and bioactivity on GFR␣1⅐RET-expressing cells. A, schematic diagram of GDNF/PSPN replacement chimeras (PGP) for gain-of-function analysis. The PSPN amino acid sequence in parentheses below the schematics was replaced by a GDNF sequence, shown above. B, both regions F2a and F2c from GDNF are required for activating GFR␣1⅐RET in the context of PSPN, as measured by the luciferase-based reporter assay. C, mutant PGP-F2ac induces GFR␣1⅐RET receptor complex formation. Neuro2a cells either untransfected (RET) or transfected with FLAG-tagged GFR␣1 (RET⅐GFR␣1) were treated with the indicated factor at 25 ng/mL. Cells were lysed, immunoprecipitated with an anti-FLAG antibody, and immunoblotted using an anti-RET antibody. A fraction of the total lysate shows the RET doublet, with the upper band corresponding to the mature cell-surface form. As expected, treatment of nontransfected Neuro2a cells (that express only RET) with GDNF does not lead to complex formation. However, GFR␣1transfected Neuro2a cells treated with GDNF or PGP-F2ac, but not PSPN, results in RET coimmunoprecipitating with GFR␣1. IP, immunoprecipitate. D, histogram of bromodeoxyuridine (BrdU) incorporation by NBL-S neuroblastoma cells in the presence of the indicated factors purified from conditioned medium of transiently transfected COS cells. Mutant PGP-F2ac stimulates proliferation of NBL-S cells at a similar level to GDNF at all doses tested. E, SH-SY5Y neuroblastoma cells were cultured in the presence of no factor (CTRL), retinoic acid (RA, 10 M), or conditioned medium from COS cells expressing the indicated construct. GDNF and PGP-F2ac stimulated neurite outgrowth in these cells, whereas PSPN did not.
PSPN/GDNF Chimeras Function as GFR␣1-specific Agonists in Vitro-The homologue-scanning mutagenesis and gain of function experiments above demonstrate that for the GFR␣1⅐RET agonists (GDNF, NRTN, and ARTN), regions F2a and F2c are sufficient for interaction with GFR␣1 but that additional regions are required for interaction with GFR␣2 and GFR␣3 (Figs. 6 and 7). This suggests that the PSPN mutants containing only regions F2a and F2c from GDNF, NRTN, or ARTN may function as GFR␣1⅐RET-specific agonists. To test this idea, we developed a cell survival assay using rat cerebellar granule cells in culture. These cells do not express endogenous RET or GFR␣ coreceptors and can be transfected efficiently, therefore providing a model system of cell survival with defined receptor components. Cerebellar granule cells transfected with RET or GFR␣1 alone do not survive in the absence of high potassium or in the presence of GDNF; however, cells cotransfected with GFR␣1 and RET survive in the presence of GDNF at levels comparable to the control condition of high potassium plus serum (Fig. 8A) (24). Consistent with other assays, cerebellar granule cells transfected with GFR␣1⅐RET or GFR␣2⅐RET survive in the presence of either GDNF or NRTN, but not PSPN (Fig. 8B). However, mutant PGP-F2ac was only capable of supporting the survival of GFR␣1⅐RET-transfected, but not GFR␣2⅐RET-transfected cells. This is consistent with its minimal ability to activate the GFR␣2⅐RET receptor in fibroblasts (Fig. 6C) and indicates that unlike GDNF and NRTN, mutant PGP-F2ac functions as a GFR␣1⅐RET-specific agonist in this in vitro cell survival paradigm. DISCUSSION We report a comprehensive mutagenesis analysis of GDNF, the prototypical member of the GDNF family of neurotrophic , and Ha together from NRTN into PSPN restores activity to ϳ70% (PNP-Ha/F2ac), significantly greater than regions F2a and F2c alone. This is consistent with the homologue-scanning mutagenesis above (Fig. 4), which suggested that region Ha is also required for GDNF interaction with GFR␣2. As expected, region Ha alone from NRTN was not sufficient to activate GFR␣2⅐RET (PNP-Ha). C, schematic of ARTN/PSPN chimeras used to delineate requirements for GFR␣3 activation. D, as reported previously, ARTN activates the GFR␣3⅐RET receptor, whereas GDNF cannot. Mutant PAP-F2ac was also not capable of activating GFR␣3⅐RET, indicating that like the NRTN-GFR␣2 interaction, the ARTN-GFR␣3 interaction requires regions in addition to those sufficient to activate GFR␣1⅐RET (F2a and F2c). However, further addition of region Ha (mutant PAP-Ha/F2ac) from ARTN restored significant activity (ϳ70%), indicating region Ha is also critical for the ARTN-GFR␣3 interaction. factors, in an attempt to characterize at the molecular level the interaction between GDNF and its receptors GFR␣1 and GFR␣2. This analysis revealed similar regions are sufficient for GFR␣1 interaction by all known GFR␣1 ligands; GDNF, NRTN, and ARTN. Furthermore we found that additional regions are required for activating GFR␣2⅐RET and GFR␣3⅐RET, which allowed the production of GFR␣1⅐RET-specific agonists as tested in receptor activation and cell survival assays.
The N-terminal Extension of GDNF Is Not Required for Activity-The N-terminal extension before the first structural cystine-knot cysteine is highly variable among the TGF-␤ superfamily members. In the case of TGF-␤2, it exists as a short ␣-helix and is stabilized by an additional disulfide bond relative to other members of the superfamily (19,20). However, in GDNF and OP-1, the only other members of the TGF-␤ superfamily that have been crystallized, the 35-37 residue N-terminal extension was unresolved and therefore does not adopt a consistent conformation in solution (21,31). We found that truncating the N-terminal extension did not affect the ability of GDNF to activate its receptors. This is consistent with experiments characterizing monoclonal antibodies against GDNF that also suggested the N-terminal extension is not required for receptor binding or bioactivity (32). The role of the N-terminal extension of GDNF remains unclear; however, because it does not appear to be involved in receptor interaction, possibilities such as its importance for stability and/or binding additional proteins or to extracellular matrix elements warrant consideration.
Characteristics of the Putative GFR␣ Interaction Site of GDNF-The mutagenesis reported here indicates that critical residues for GFR␣ receptor interaction and specificity are located in the second finger region of the GDNF molecule. However, whereas the use of homologue-scanning mutagenesis is ideally suited for identifying sites involved in differential receptor specificity, it cannot delineate all residues involved in receptor binding, as some of these may be identical in GDNF and PSPN. The fact that residues from regions F2a and F2c are sufficient when placed in the context of PSPN to activate GFR␣1⅐RET indicates that residues in these regions are likely to be directly involved in binding to GFR␣s. Although it is possible that the GFLs also contact RET in the active receptor complex, the fact that all members of the GFLs, including PSPN, appear to signal through RET makes it doubtful that the regions identified by our mutagenesis scheme are involved in RET interaction directly (4,14). Because we focused on functional activation of the GFR␣⅐RET receptor complex rather than receptor binding assays, we cannot exclude the possibility that some of the nonfunctional mutants produced here may still be capable of binding to the GFR␣ coreceptors. Such mutants could potentially function as receptor antagonists, and we are currently investigating this possibility.
Current evidence suggests that the stoichiometry of the active receptor complex for GDNF is (GDNF) 1 (GFR␣) 2 (RET) 2 (33). Residues from regions F2a and F2c project essentially from the bottom the GDNF structure and form symmetric sites on both monomers of the molecule (see Fig. 4B). Interestingly, residues from region Ha (which is critical for GDNF-GFR␣2 and ARTN-GFR␣3 interaction) lie directly adjacent to regions F2a and F2c from the finger 2 region of the opposing monomer and form continuous surfaces along the side and bottom of the molecule (see Fig. 4C). Because these regions are all critical for GFR␣ specificity, we suggest that the molecular surface formed by the heel of one monomer and the second finger of the adjacent monomer form GFR␣ binding surfaces on GDNF.
Comparison of structure-function studies from other TGF-␤ and cystine-knot superfamily members reveals common themes in the location of receptor binding surfaces. The best characterized of these is the binding of vascular endothelial growth factor with its receptors KDR and Flt-1, where extensive mutagenesis and co-crystallization have been performed (34,35). Mutagenesis of vascular endothelial growth factor identified a cluster of residues critical for KDR binding, several of which were along the adjacent ␤-strands of the second finger, similar to regions F2a and F2c identified here for GDNF. Furthermore, even though the orientation of the monomers in the vascular endothelial growth factor dimer is different from GDNF, the receptor binding site on vascular endothelial growth factor involves residues from finger 2 and the analogous heel region of the opposing monomer (34). Finally, mutagenesis of TGF-␤ indicates that the binding affinity of the different TGF-␤ isoforms for the TGF-␤ type II receptor is also determined by residues along the second finger of the molecule, analogous in location to region F2c (36,37). Therefore it is possible that although the receptor systems and dimerization modes are strikingly different for these different cystine-knot superfamily proteins, the location of their receptor interaction surfaces may be quite similar.
Differences in GFR␣1-GFR␣2-GFR␣3 Interaction-In our analysis of the structural determinants of GDNF required for activating its two functional receptors, GFR␣1⅐RET and GFR␣2⅐RET, we observed that only regions F2a and F2c were required for activating GFR␣1⅐RET but additional regions (including region Ha) were required for activating GFR␣2⅐RET. GFR␣1⅐RET is the most promiscuous of the GFL receptors and is able to interact with three of the four known ligands. This is FIG. 8. Mutant PGP-F2ac functions as a GFR␣1⅐RET-specific agonist in a neuronal survival assay. A, cerebellar granule cells dissected from 5-day-old rats were transfected with the indicated receptor components and cultured for 3 days in the presence of high potassium plus serum (K25ϩS), low potassium (K5), or low potassium plus GDNF (GDNF). Only neurons transfected with GFR␣1 and RET together survived in the presence of GDNF. B, cerebellar granule cells transfected with GFR␣1⅐RET or GFR␣2⅐RET were cultured in the presence of the indicated factors. NRTN was bacterially produced and added at 50 ng/mL. All others were produced in COS cells and added as concentrated conditioned medium. Whereas GDNF and NRTN support the survival of neurons expressing GFR␣1⅐RET or GFR␣2⅐RET similarly, mutant PGP-F2ac only supports the survival of GFR␣1⅐RET expressing neurons. consistent with GFR␣1⅐RET having the most minimal requirements for being activated (regions F2a and F2c) and additional regions being required for activating GFR␣2⅐RET and GFR␣3⅐RET. It is of note that regions F2a and F2c from GDNF, NRTN, or ARTN conferred similar activity to PSPN in the receptor activation assay. It will be interesting to determine if these mutants confer similar GFR␣1-binding characteristics to their parent constructs (PGP-F2ac with higher affinity than PNP-F2ac or PAP-F2ac) or if they all bind GFR␣1 with similar affinity, suggesting that regions of NRTN and ARTN outside F2a and F2c may actually reduce their affinity for GFR␣1.
Although region Ha is critical for the NRTN-GFR␣2 and ARTN-GFR␣3 interactions, PSPN mutants containing regions Ha, F2a, and F2c were not fully active, suggesting that either additional regions are necessary or the binding determinants from NRTN and ARTN are not presented properly in the context of the PSPN molecule. A better understanding of receptor complex formation and stoichiometry and direct co-crystallization will be required to confirm and refine our understanding of the receptor binding surfaces of these molecules.
Utility of GFL Chimeras as GFR␣1-specific Agonists-The GFLs are promising candidates as therapeutic agents for the treatment of neurodegenerative diseases, including Parkinson's disease. GDNF, NRTN, and ARTN all support the survival of cultured dopaminergic neurons of the embryonic ventral midbrain (2,4,6), and GDNF and NRTN have both shown to be effective in supporting the survival of midbrain dopaminergic neurons in animal injury models (6,10). Several lines of evidence argue that the effects of GDNF, NRTN, and ARTN on dopaminergic ventral midbrain neurons are mediated through the GFR␣1⅐RET receptor system. First, because GFR␣3 is not expressed in the ventral midbrain and ARTN cannot utilize GFR␣2, survival promotion of these neurons by ARTN is likely through its ability to activate GFR␣1⅐RET (4). Second, GFR␣2 expression is diffuse and weak in the pars compacta region of the substantia nigra and does not colocalize with tyrosine hydroxylase staining neurons, in contrast to the significantly stronger expression of GFR␣1, which does colocalize with tyrosine hydroxylase staining neurons (6). Finally, the ability of both GDNF and NRTN to support the survival of dopaminergic ventral midbrain neurons is lost in GFR␣1 knockout mice, indicating that at least in the embryo the survival promotion of dopaminergic ventral midbrain neurons is only through GFR␣1⅐RET signaling (38). Because there are several central and peripheral sites of GFR␣2⅐RET expression that could lead to side effects in the treatment of patients with GDNF or NRTN, the GFL chimeras that function as GFR␣1⅐RET-specific agonists described here (PGP-F2ac and particularly PAP-F2ac) could be therapeutically useful as they would minimize potential GFR␣2-related side effects, while maintaining the ability to support survival and growth of the desired GFR␣1⅐RET expressing target neuron populations.