High Resolution Mapping of the Binding Site of TrkA for Nerve Growth Factor and TrkC for Neurotrophin-3 on the Second Immunoglobulin-like Domain of the Trk Receptors*

Neurotrophic factors are important for survival and maintenance of neurons during developmental and adult stages of the vertebrate nervous system. The neurotrophins mediate their signal into the cell by specific interaction with tyrosine kinase receptors of the Trk family. The extracellular immunoglobulin-like domain of the Trk receptors adjacent to the membrane has previously been shown to be the dominant element for specific neurotrophin binding. Using computer graphics models of the human TrkA and TrkC immunoglobulin-like domains as a guide, the residues involved in binding to their respective neurotrophins were mapped by mutational analysis. TrkC primarily utilizes loop EF, between β-strands E and F, for binding. In contrast, TrkA utilizes the EF loop as well as additional residues, the latter being prime candidates for determining the specificity of TrkAversus TrkC. When selected TrkC and TrkA mutants with reduced binding were expressed on NIH3T3 cells, neurotrophin-induced autophosphorylation was strongly reduced or absent.

The domain organization of the extracellular portion of the Trk receptors has been proposed based on sequence information (13). According to this proposal, the extracellular portion of the Trk receptors is composed of a cysteine-cluster, a leucinerich motif, and a second cysteine-cluster, followed by two immunoglobulin-like domains. Analysis of a series of Trk receptor domain deletions and chimeras between the different members of the Trk family has shown that the second immunoglobulinlike domain of TrkC and TrkB confers specificity and affinity to their respective neurotrophins (14) and that both the first and second immunoglobulin-like domains are important for TrkA binding and specificity (14,15). The leucine-rich motif has been implicated to bind neurotrophins (16,17), although no recruitment of BDNF binding was observed when this motif was transferred into a TrkC background (14). In contrast, exchanging the second immunoglobulin-like domain from TrkB into TrkC resulted in high affinity BDNF binding (14). In addition, a fragment of TrkA comprising the two immunoglobulin-like domains has been shown to bind NGF and inhibit neurite outgrowth (18,19).
In this study, the neurotrophin binding sites on TrkA and TrkC have been determined using molecular modeling and mutagenesis of the second immunoglobulin-like domain. Most of the NT-3 binding site on the TrkC receptor is contributed by residues in a loop connecting ␤-strands E and F at the membrane proximal portion of the binding domain. In contrast, while this same loop is important for TrkA binding of NGF, numerous residues outside of this loop also make major contributions to binding, suggesting a mechanism of specificity discrimination.

Molecular Modeling of the TrkA and TrkC Binding Domains-
The models of human TrkA domain 5 (TrkAd5; second immunoglobulin-like domain) and TrkC domain 5 (TrkCd5) were based on the crystal structure of the first domain of human vascular adhesion molecule-1 (VCAM) (20). The sequences of human TrkAd5 and TrkCd5 were compared with the sequences of immunoglobulin-like domains for which crystallographic or NMR coordinates are available, e.g. CD4, CD2, telokin, titin, etc. None of the available structures exceeded 10% identity with either TrkAd5 or TrkCd5 (TrkAd5 and TrkCd5 have 36% identity). Since no single immunoglobulin-like domain stood out as a template, the VCAM structure was chosen as the template, since it contains individual immunoglobulin-like domains, as opposed to antibody crystal structures in which individual domains form an interface to another immunoglobulin domain. Although the percentage of identity of TrkAd5 and TrkCd5 to any specific immunoglobulin-like domain is low, the wealth of crystal and NMR structures of immunoglobulin-like domains increases the reliability of models of these domains. Difficulties concerning specific loops or amino acids are addressed below.
When the sequence of TrkCd5 was aligned with the sequence of human VCAM (Fig. 1), it became apparent that the buried disulfide bond, which is a hallmark of immunoglobulin domains, was absent but that the hallmark tryptophan residue was conserved (residue 304). Cys 289 is in ␤-strand B, as in VCAM, but the two residues at positions ϩ2 and Ϫ2 from Cys 289 are glutamic acid (Fig. 1). Since these two glutamic acids are unlikely to be buried, Cys 289 must be on the exterior of the ␤-strand. Likewise, Cys 331 has a glutamic acid at position Ϫ2 from it and is also likely to be exposed; in addition, Cys 331 is in ␤-strand E, not in ␤-strand F as in VCAM (Fig. 1). Upon building the model, Cys 289 and Cys 331 were found to be near one another on the two adjacent ␤-strands B and E, again suggesting that the disulfide in TrkCd5 is exterior, not interior. In domain 5 of both TrkA and TrkC, the sequence of loop EF can accommodate the conformation found in VCAM domain 1 (20) or can accommodate an ␣-helix as found in the CH2 and CH3 domains of IgG (21).
The VCAM domain 1 was transformed into TrkCd5 in three steps. First, all residues in the ␤-strands were changed to the TrkCd5 sequence using the INSIGHT-II program (Biosym Technologies, San Diego, CA). If possible, conformations of TrkCd5 side chains were kept similar to those of VCAM; otherwise, they were based on rotamer libraries (22), packing, and hydrogen bond considerations. Second, loop structures connecting the ␤-strands were gleaned from a search of loops of similar size from the Protein Data Bank (23) using the loop search algorithm in the INSIGHT-II program. Finally, the relatively large segment connecting ␤-strands C and E (24 residues long; His 306 -Gly 330 ) was modeled in two conformations: one similar to the conformation found in telokin (24) and titin (25) and the other similar to the conformation found in IgG Fc domains CH2 and CH3 (21) (shown in Fig. 2). It is not possible to discern which conformation will be adopted. The TrkAd5 model was then generated from the TrkCd5 model using similar methods.
The TrkAd5 and TrkCd5 models were subjected to energy minimization using the DISCOVER program (Biosym Technologies, San Diego, CA). The all-atom AMBER forcefield (26) as provided in the DISCOVER program was used for all calculations, employing a 14-Å cut-off for nonbonded interactions, a linear dielectric (⑀ ϭ 4.0 ϫ r), and 1-4 atom-atom interactions (i.e. atoms connected by two intervening covalently bonded atoms) scaled by a 0.5 factor. Prior to minimization, hydrogen atoms were added to the structure using INSIGHT-II, and positions of hydrogens on Ser, Thr, and Tyr side chains were checked visually for proper alignment in hydrogen bonds, if present. Energy minimization was performed in three stages. In stage one, 500 cycles of steepest descents minimization was employed with 47 nitrogen donor: oxygen acceptor atom pairs constrained to be within hydrogen bonding distance (2.90 Å; force constant of 50 kcal/Å). The latter were included to retain the ␤-strand nature of the immunoglobulin-like domain. In stage two, conjugate gradient minimization was employed for 2000 iterations, and the hydrogen bond constraints were retained. In stage three, another 2500 cycles of conjugate gradient minimization were employed with the hydrogen bond constraints released.
Mutagenesis and Recombinant DNA Manipulations-An immunoadhesin (27) of the extracellular portion of human TrkC, in which the extracellular portion is fused to the Fc of a human antibody, was previously constructed, sequenced, and subcloned into a vector that allows for production of double-and single-stranded DNA in Escherichia coli as well as expression of mature receptors in a mammalian system under control of the cytomegalovirus promoter (28). A similar immunoadhesin was constructed using human TrkA. Oligonucleotide site-directed mutagenesis (29) on these vectors generated the specific mutants. After transformation into the E. coli strain XL1-Blue (Stratagene, San Diego, CA), colonies were screened for the presence of the desired mutation by sequencing single-stranded DNA using the Sequenase version 2.0 kit (U.S. Biochemical Corp.). For each selected clone, the entire DNA sequence coding for the receptor immunoadhesin was verified. Double-stranded DNA used for transfection of human kidney 293 cells was isolated from XL 1-Blue with the QIAGEN DNA purification kit (Qiagen Inc., Chatsworth CA).
Design, Construction, Expression, and Purification of Receptor Mutants-All receptor mutants were constructed in the context of the full-length extracellular portion and expressed as immunoadhesins. The extracellular portions of TrkA and TrkC expressed in this fashion have been shown to display the expected specificities (28) and to interact with high affinity with their respective neurotrophin ligands (30). Furthermore, a panel of NT-3 mutants displayed a similar profile of affinities to TrkC whether the latter was expressed as immunoadhesin or on NIH3T3 cells (30), indicating that the binding domains of TrkC have a similar conformation in both systems. Plasmid DNA coding for the TrkA or TrkC immunoadhesin mutants were introduced into the human fetal kidney cell line 293 by calcium phosphate precipitation (31). The proteins were expressed and purified as described previously (14) and were more than 95% pure as judged from SDS-polyacrylamide gel electrophoresis (data not shown).
Iodination of NT-3-NT-3 was labeled with 125 I using lactoperoxidase (Calbiochem). Briefly, 8 g of purified NT-3 in 20 l of 10 mM sodium acetate, 140 mM NaCl, pH 5.5, was mixed with 37.5 l of 0.4 M sodium acetate, pH 5.6, 0.015 IU of lactoperoxidase, and 2 mCi of 125 I. The reaction was started by adding 15 l of diluted hydroxyperoxide (1:174,000 of 30% stock) (Mallinckrodt, Phillpsburg, NJ) and was kept at room temperature for 5 min. An additional 15 l of diluted hydroxyperoxide was added, and the reaction was kept at room temperature for another 5 min. The reaction was stopped by the addition of 15 l of 20 mM N-acetyl-L-tyrosine (Sigma) in water. The labeled NT-3 was isolated from free iodine by use of a PD-10 desalting column (Pierce). The trichloric acetic acid precipitability of the fractions containing the labeled NT-3 was usually higher than 90%, and the specific activity was between 2000 and 3000 Ci/mmol.
Binding TrkC Mutants to NT-3-Competition binding assays were performed using a 96-well plate format and purified receptor and ligand preparations. The binding constants (K d ) of receptor mutants are related to inhibition constants (IC 50 ) of competition experiments using labeled and unlabeled neurotrophins by the Cheng-Prusoff equation (K d ϭ IC 50 /(1 ϩ L*/K dL* )), where L* and K dL* are the concentration of the labeled ligand and its affinity to the receptor, respectively (32). Prior to the competition binding experiment, the concentration of each receptor immunoadhesin was adjusted to bind 10 -15% of the total labeled ligand in the absence of competitor to satisfy true competition requirements (e.g. K R t Ͻ 0.1) and to prevent tracer ligand depletion due to increased receptor concentrations (33). Native TrkC bound 10% of the total labeled NT-3 at a concentration of 7.1 Ϯ 1.7 ng/ml (five independent expressions). Some of the mutants (e.g. N335A/T338A/N342A) did not bind any labeled NT-3 above background when assayed at 2000 ng/ml concentration; therefore, no competition experiment could be performed. Hence, the affinities of these mutants were estimated to be reduced at least 200-fold and were labeled as NB (no binding) in Table I.
After coating each well with 100 l of 5 g/ml goat F(abЈ) 2 antihuman Fc IgG (Organon Technika, Westchester, PA) in 0.1 M Tris, pH 9.5, for 15 h at 4 -8°C, the wells were aspirated, washed three times with PBS, incubated for 2 h with 100 l of a solution of the receptor immunoadhesin in binding buffer, and washed with PBS. Binding buffer consists of Leibovitz's L-15 medium supplemented with 5 mg/ml bovine serum albumin (Intergen, Purchase, PA), 0.1 mg/ml horse heart cytochrome c (Sigma), and 20 mM HEPES, pH 7.2. In competition experiments, 50 l of binding buffer was immediately added to the wells to prevent drying. Purified, unlabeled NT-3 was serially diluted in binding buffer to a concentration range of 4096 -0.125 pM. 25 l of serial dilution was added per well followed by 25 l of labeled NT-3. The final concentration of labeled NT-3 in each well was approximately 30 pM. After 3 h of incubation at room temperature, the wells were washed with PBS, 0.5% Tween 20 (Sigma), and the bound radioactivity was counted. Competition data were fit to a four-parameter equation using the Kaleidagraph software (Abelbeck Software). All competition binding experiments were repeated at least three times.
Binding of TrkA and TrkC Mutants to Monoclonal Antibodies-The monoclonal antibodies against the extracellular domains of TrkA and TrkC were developed using an analogous protocol to a described procedure (34). Microtiter plates (Nunc, Naperville, IL) were coated with 100 l/well of 1 g/ml goat F(abЈ)2 anti-human IgG Fc-specific antibody (Cappel-Organon Teknika, Durham, NC) in 0.05 M carbonate buffer, pH 9.6, overnight at 4°C. The plates were washed with PBS, 0.05% Tween 20, and the excess binding sites were blocked with 200 l/well of PBS, 0.05% Tween 20, 0.5% bovine serum albumin (Calbiochem) (buffer A) for 1-2 h at ambient temperature. The plates were washed again, and 100 l of 1 g/ml Trk or Trk mutant immunoadhesins diluted in buffer A were added to the appropriate wells. After incubation for 1-2 h at ambient temperature, plates were washed and 100 l of human TrkAor TrkC-specific mouse mAbs (1 g/ml) were added, incubated for 1 h at ambient temperature, and washed. Horseradish peroxidase-conjugated goat anti-mouse Fc IgG antibody (Cappel-Organon Teknika), 1:5000 in buffer A, was added to the plate, incubated for 1 h at room temperature, and washed. The color was developed for 10 -20 min with 100 l of o-phenylenediamine (Sigma) substrate solution (5 mg of o-phenylenediamine in 12.5 ml of PBS containing 4 mM H 2 O 2 ), and the reaction was stopped with 100 l of 2.5 N H 2 SO 4 . Absorbances were recorded at 490 nm with a 405-nm reference filter (UVMax kinetic microplate reader, Molecular Devices, Palo Alto, CA).
Binding of TrkA Mutants to NGF-An NGF/TrkA assay similar to the one described for NT-3/TrkC was evaluated but was found to lack the sensitivity required to delineate differences in binding, possibly due to the covalently bound iodine on the labeled NGF interfering with binding. Instead, the mAb binding assay described above was modified to evaluate the binding of TrkA mutants to NGF. The anti-NGF and anti-TrkA mAbs used in the assay were developed using an analogous protocol to one previously described (34). Plates were coated with 100 l/well anti-NGF mAb 911 (1 g/ml) overnight at 4°C. The plates were washed with PBS, 0.05% Tween 20, and the excess binding sites were blocked with 200 l/well buffer A for 1-2 h at ambient temperature. NGF (100 l of a 1 g/ml solution in buffer A) was then added and incubated for 1 h. Serial dilutions of TrkA were added in duplicate (100 l of solutions starting at 1 or 16 g/ml; 8 dilutions total), and the bound receptor was detected with biotinylated anti-TrkA specific mAb 1487 (100 l of a 1 g/ml solution in buffer A) (see Table II). Horseradish peroxidase-conjugated streptavidin (Sigma), 1:1000 in buffer A, was added, and the absorbances were recorded at 490 nm with a 405-nm reference filter (UVMax kinetic microplate reader, Molecular Devices). A similar enzyme-linked immunosorbent assay-based assay was also evaluated for NT-3/TrkC to determine if this assay format was more sensitive than the 125 I-labeled NT-3 format. A selected panel of TrkC mutants showed the same pattern of binding as in the labeled NT-3 format. The enzyme-linked immunosorbent assay format was not more sensitive than the labeled NT-3 format, and the rest of the TrkC mutants were not repeated using the enzyme-linked immunosorbent assay format.
Construction of Stable Transfected NIH3T3 Cells-The cDNA coding for the extracellular domain of TrkC and the mutants T338A and N335A/T338A/N342A were amplified using specific PCR primers, which incorporated flanking restriction sites as described previously (14). Similarly, a fragment coding for the transmembrane and the intracellular domain of TrkC was amplified using PCR. These fragments were ligated into the mammalian expression vector pMEXneo (5). NIH3T3 cells were transfected with these constructs using lipofection (DOTAP reagent; Boehringer Mannheim). At least 10 single colonies for each of the constructs were selected for G418 resistance, expanded, and assayed for receptor expression by binding to wheat germ agglutinin or immunoprecipitation with the pan-Trk antibody 443 as described previously (35). For all constructs, at least 90% of the selected colonies expressed the desired receptor.
The cDNA coding for the extracellular domain of TrkA and the mutants E301A, T292A, and T319A/H320A/N323A were amplified using specific PCR primers, which incorporated flanking restriction sites as described previously (14). Similarly, a fragment coding for the transmembrane and the intracellular domain of TrkC was amplified using PCR. These fragments were ligated into the mammalian expression vector pMEXneo (5). NIH3T3 cells were transfected with these constructs using lipofection (DOTAP reagent; Boehringer Mannheim). At least 10 single colonies for each of the constructs were selected for G418 resistance, expanded, and assayed for receptor expression by binding to wheat germ agglutinin or immunoprecipitation with the pan-Trk antibody 45 (a gift of Dr. Barbara Hempstead) as described previously (35).
Stimulation of TrkC Receptor Autophosphorylation on NIH3T3 Cell Lines by NT-3-Approximately 1 ϫ 10 7 cells were treated at 37°C for 5 min with the appropriate neurotrophin. With NT-3, they were treated at 100 ng/ml or as indicated in Fig. 3; with NGF, they were treated at the different concentrations shown in Figs. 6 and 7. Nonidet P-40 plate lysis and immunoprecipitation with pan-Trk antiserum 443 (NT-3/ TrkC) or pan-Trk antibody 45 (NGF/TrkA) was performed as described previously (35). The phosphotyrosine content was analyzed by Western blot using monoclonal antibody 4G10 (Upstate Biotechnology, Inc., Lake Placid, NY) as described previously (35,36). For each of the constructs, two different cell lines were selected and assayed for neurotrophin-induced autophosphorylation in triplicate. (35) were plated onto 35-mm collagen-coated tissue culture dishes and grown overnight. The TrkC immunoadhesin or TrkC mutant immunoadhesins were mixed with NT-3 at the indicated concentrations in 2 ml of medium, incubated for 30 min at room temperature, and then added to the cells. The proportion of neurite-bearing cells was determined by counting the number of cells containing processes at least twice the length of the cell body after 3 days. The photographs in Fig. 4 were taken after 3 days. Inhibition experiments were performed at least three times using soluble receptor from two expressions.

The Three-dimensional Model of the TrkC Binding Domain
Identifies Surface-exposed Residues-To guide mutagenesis and aid in interpreting the results, computer graphics models of the second immunoglobulin-like domains of TrkA and TrkC were generated based on the crystal structure of human vascular cell adhesion molecule-1 domain 1 (20). The model predicts seven ␤-strands ( Fig. 1) in a c-type arrangement, with four strands (A, B, D, and E) forming sheet I and three strands (C, F, and G) forming sheet II of the ␤-sandwich (37). Loops will be denoted by the two ␤-strands that the loop connects, e.g. loop AB connects ␤-strands A and B.
One of the original hallmarks of an immunoglobulin-like fold is a conserved disulfide bond between strands B and F connecting the two ␤-sheets in the ␤-sandwich structure (38). However, many examples of immunoglobulin-like domains show that the disulfide bond can be moved within strands B, C, E, and F or may be absent (37). As described under "Experimental Procedures," the disulfide bond in domain 5 of the Trks is probably on the exterior of the domain. Surface-exposed residues were identified by inspection of the models. For the majority of residues, the contribution to binding was probed by mutation to alanine (39). Some residues, including native alanine residues, were changed to amino acids other than alanine, e.g. the analogous amino acid of a different Trk receptor. Selected residues that showed a pronounced effect when mutated to alanine were subsequently mutated to other amino acids to determine how a particular side chain contributes to affinity.
The Binding Site on TrkC for Its Ligand NT-3-The most important binding determinants in TrkC were discovered by analysis of the single alanine mutants E287A, R295D, N335A, T338A, and N342A (mutants are designated by the one-letter code for the native amino acid, the residue number, and the substituted amino acid; multiple mutants are separated by slashes; numbering is that of mature human TrkA or TrkC). Residue Glu 287 is located in loop AB ( Fig. 2A), and the affinities of E287A, E287K, and E287H to NT-3 were reduced 7-, 27-, and 3-fold, respectively, when compared with TrkC (Table I). That the most pronounced effect on the affinity resulted from charge reversal of the side chain suggests that Glu 287 may be inter-

FIG. 1. Alignment of human TrkA (trkAd5) and TrkC (trkCd5) domain 5 sequence with human vascular adhesion molecule-1 domain 1 (VCAMd1).
Buried residues in VCAM domain 1 are denoted by a filled circle above the residue. ␤-Strands in the VCAM crystal structure (20) and the TrkC model are underlined and labeled. The lowercase letters at the C terminus of TrkC and TrkA were not included in the immunoadhesin constructs. TrkC and TrkA residue numbers are under the amino acid sequence.
acting with a basic amino acid on NT-3. Residue Arg 295 is in loop BC ( Fig. 2A), and the analogous residues in TrkA and TrkB are aspartic acid and lysine, respectively. R295A had no effect on binding, but R295D reduced binding by an order of magnitude (Table I) and suggests that Arg 295 in TrkC might interact with an acidic residue in NT-3.
Residues Asn 335 , Thr 338 , and Asn 342 are located in loop EF ( Fig. 2A), and the affinities to NT-3 of N335A, T338A, and N342A were reduced by 38-, 30-, and 24-fold, respectively (Table I). Thr 338 and Asn 342 are conserved among the three Trk receptors, while position 335 is aspartic acid in TrkB and asparagine in TrkA. Two other residues in loop EF at positions 336 and 340 vary among the three Trk receptors; however, alteration of these two residues did not affect binding (K336A and K336A/Y340A, Table I). At position Asn 335 , the affinities of N335Q, N335D, and N335S were reduced by 21-, 0.8-, and 1.2-fold, respectively, compared with native TrkC (Table I).
Replacing Asn 335 with a longer side chain with the same functional group in mutant N335Q resulted in a pronounced decrease of affinity, which indicates that the space in the binding pocket at this position may be very limited and that no additional atoms can be accommodated. Introduction of a charged but isosteric amino acid in N335D as well as exchanging asparagine for serine did not decrease affinity to NT-3. These results indicate that Asn 335 probably acts as a hydrogen bond acceptor in the complex.
The affinity of T338S for NT-3 was equivalent to native TrkC, while T338A resulted in a reduction of 30-fold. When Thr 338 was changed to valine, which has a similar size but no hydrogen bonding potential, the affinity decreased by 12-fold (Table I). This suggests that in the complex with NT-3, the major contribution of Thr 338 to the affinity is by forming a hydrogen bond involving its hydroxyl oxygen, although some steric component is evident when T338A and T338V are compared. A larger side chain cannot be accommodated at position 338, since T338N had its affinity reduced 23-fold when compared with native TrkC.
Residue Asn 342 was probed by mutations to aspartic acid and serine with reduction of affinities of 20-and 21-fold, respectively, while mutation to glutamine resulted in no detectable binding (Table I). As at position 335, replacing Asn 342 with a larger side chain reduced binding. In contrast to position 335, however, making Asn 342 smaller (N342A, N342S) also reduced binding, as did removal of the side chain NH 2 group (N342D). These data suggest that the Asn 342 side chain acts as a hydrogen bond donor and that the steric requirements for hydrogen bond(s) and/or packing are quite rigid.
In addition to the four major contributors, two other residues showed minor effects; C331A in ␤-strand E and H363A in ␤-strand G had their affinities reduced 2.9-and 2.0-fold, respectively. All other residues that were tested displayed affinities that were unchanged compared with native TrkC (Table  I). Notably, alteration of the cysteine residues to alanine either had a minor effect (C331A) or no effect (C289A). This supports the hypothesis that the disulfide bond of TrkC is not involved in maintaining the structure of the domain, unlike the conserved disulfide bond of canonical immunoglobulin-like domains (38). The second immunoglobulin-like domain of TrkC contains two possible N-linked glycosylation sites at positions Asn 344 and Asn 357 (Fig. 1). Changing these asparagines to alanine had no effect on binding (Table I).
In addition to the two immunoglobulin-like domains, the leucine-rich motif has been previously implicated in binding neurotrophins (16,17). However, no recruitment of BDNF binding was observed when the TrkB leucine-rich motif was transferred into a TrkC background (14). In this study, a chimeric protein in which the leucine-rich motif of TrkA was substituted into TrkC (mutant A2C) showed no binding to NGF but bound NT-3 as well as native TrkC (Table I).
The Binding Site on TrkA for Its Ligand NGF-As for TrkC, loop EF in TrkA contains residues that play a major role in binding NGF. T319A, H320A, and N323A (Fig. 2B) all exhibited greater than 100-fold reduction in binding. More modest reductions were evident for other residues in loop EF, V321A (24-fold) and Q317A (6-fold) ( Table I)   a TrkA results are expressed as the mean of two independent experiments Ϯ S.D.; mutants with effects Ͼ10-fold were assayed at least four times, and results are expressed as the mean Ϯ S.D. Mutants exhibiting a reduction in binding between 5-and 30-fold are in italic type, and those greater than 30-fold are in boldface type. NB means that no binding could be detected under the conditions described under "Experimental Procedures." Possible N-linked glycosylation sites are underlined.
b TrkC results are expressed as the mean of two independent experiments Ϯ S.D. Mutants exhibiting a reduction in binding greater than 5-fold (boldface type) were assayed at least four times, and results are expressed as the mean Ϯ S.D. NB means that no binding could be detected under the conditions described under "Experimental Procedures." Possible N-linked glycosylation sites are underlined. c A2C binding to TrkA is the mean of six assays. d The IC 50 (mutant)/IC 50 (TrkA) of these mutants may be slightly too large due to the use of mAb A1487 as the detection agent in the NGF/TrkA assay (see "Results").   (Table I).
It should be noted that the NGF/TrkA assay used may have overestimated the reduction in binding of a few mutants because the anti-TrkA mAb used as the detecting agent in the assay, A1487 (Table II), also showed slightly reduced binding to several mutants compared with native TrkA (Table III). The mAb A1487 was the only mAb that could be used in the assay, since the other four mAbs either mapped to domain 5 (A1486 and A1502) or were blocking mAbs (A1488 and A1503) ( Table  II). For mutants C267A, P269E, C312A, and T327A/L329A, mAb A1487 showed 11-27% reduction in binding compared with TrkA; therefore, the IC 50 mutant/IC 50 TrkA values in Tables I and III are probably inflated by these same percentages. Regardless, P269E remains a major determinant in TrkA binding to NGF, while the other three, especially C267A and C312A, are of lesser importance.
Several of the TrkA mutants deserve closer analysis. Compared with TrkC, where altering the cysteine residues to alanine resulted in little or no effect on binding, in TrkA altering the cysteines to alanine did effect reduction in binding. However, the reduction in binding (C267A, Ͻ14-fold; C312A, Ͻ28fold) was less dramatic than for several other TrkA mutants, e.g. P269E at Ͼ100-fold, H310A at 79-fold and T319A at Ͼ100fold (Table I). The second immunoglobulin-like domain of TrkA contains four possible N-linked glycosylation sites at positions Asn 285 , Asn 290 , Asn 305 , and Asn 325 (Fig. 1). Changing these asparagines to alanine had little or no effect on binding to NGF ( Table I), suggesting that if carbohydrate is present it does not play a role in binding. At three of these sites, altering the threonine side chain 2 positions downstream from the asparagine (part of the consensus Asn-X-(Thr/Ser) glycosylation se-quence) did effect a relatively large reduction in binding, 20 -24-fold at each site (Table I). This implies that these three threonine residues interact directly with NGF; if lack of carbohydrate were the cause for these mutants not binding well, then the same should have been exhibited by changing the coupled asparagines to alanine. Finally, as with TrkC, residues in TrkA loop EF, which individually abrogated binding when changed to alanine were combined in double and triple mutants, and these showed no detectable binding ( Table I).
One of the crystal structures of murine NGF showed the presence of a zinc-binding site in the central ␤-strand bundle involving His 84 and Asp 105 as zinc ligands (40). The model of TrkA (Fig. 2B) shows that in ␤-strand A and loop AB, four residues (His 258 , Glu 262 , His 264 , and His 265 ) could provide one or more possible zinc-binding ligands to complement the zincbinding site in NGF. Alternatively, His 310 in loop EF (important for the NGF/TrkA interaction but not the NT-3/TrkC interaction), could also be a possible zinc-binding ligand. When the TrkA binding assay was performed in the presence of 2 mM and 10 mM EDTA, no abrogation of binding was apparent (Table I). Hence, though zinc was bound to NGF in the crystal structure, it apparently is not involved in the NGF/TrkA interaction.

Binding of a Panel of Monoclonal Antibodies Confirms the Structural Integrity of TrkA and TrkC Mutants-A loss in
activity of a protein upon mutation of a side chain can be attributed either to the importance of this side chain to the function of the protein or to an involvement in maintaining the structural integrity of the protein. A method to verify the structural integrity of mutants with reduced activity is to determine their reactivities against a panel of monoclonal antibodies (41).
A panel of five mAbs reactive against the extracellular domain of human TrkA was generated. To determine the domain(s) on the extracellular part of TrkA that interacts with each of these mAbs, their affinities to previously described TrkA mutants (14) were determined. All of the mAbs were specific for TrkA and did not bind to TrkB or TrkC (Table II). The domain mapping of the mAbs is provided in Table II. The blocking mAb A1502 only bound well to constructs in which domain 5 was present (Table II); the epitope of mAb A1502 was further resolved to include the TrkA segment connecting do-   (Table III).
Of the two mAbs that map to domains 1-4, i.e. A1503 and A1487, one blocks NGF binding, while the other does not. Of the two mAbs that map to domain 5, i.e. A1486 and A1502, one blocks, and the other does not. Two of the three blocking mAbs include domains 2 and 4 in their epitope (A1488 and A1503), but the third blocking mAb maps only to domain 5 (A1502). Hence, the pattern of blocking versus nonblocking mAbs does not unambiguously resolve which domains need to be occluded to block NGF binding by TrkA.
The panel of anti-human TrkA mAbs was used to test the structural integrity of the TrkA mutants that exhibited reduction in NGF binding. Most of the TrkA mutants bound to the five mAbs as well as native TrkA; those that showed reduced binding (i.e. Ͻ95% compared with native TrkA) to at least one of the mAbs are listed in Table III. Four mutants showed reduced binding only to mAb A1502, and these are probably part of the epitope of this mAb. Two mutants, T307A and N332A, showed reduced binding only to mAb A1486 and may comprise part of the epitope of this mAb. Three mutants (P269E, C312A, and T327A/L329A) showed reduced binding to mAbs A1488, A1487, and A1486, although binding was still 51-83% compared with native TrkA. These residues may interact directly with an mAb (e.g. A1486, which maps to domain 5), may play a role in interdomain interaction, or may perturb the local structure of domain 5. That these three mutants bound well to the other two mAbs, however, suggests that any structural deformation is not global.
A panel of four mAbs reactive against the extracellular domain of human TrkC was also generated. To determine the domain(s) on the extracellular part of TrkC that interacts with each of these antibodies, their affinities to previously described TrkC mutants (14) were determined. All of the mAbs were specific for TrkC and did not bind to TrkB, although mAb C1435 did show some residual binding to TrkA (Table IV). The domain mapping of the mAbs is provided in Table IV.
The panel of anti-human TrkC mAbs was used to test the structural integrity of the TrkC mutants E287A, N335A, T338A, N342A, and N335A/T338A/N342A, all of which showed reduced binding to NT-3. These mutants bound all four mAbs with affinities similar to TrkC, i.e. between 84 and 100% (Table  V). Hence, the observation that all four mutants bind the mAbs C1435, C1437, and C1438 demonstrates that the first three domains of the TrkC mutants are structurally intact and that these mutations do not lead to global unfolding of the protein.
Most importantly, these mutants also bound to mAb C1436, which was mapped to the two immunoglobulin-like domains, confirming that the mutations did not lead to unfolding of the binding domain.
In addition to binding by the panel of mAbs, it should be noted that the TrkA and TrkC mutants were expressed as immunoadhesins (27), i.e. with an immunoglobulin Fc fused to the C terminus of the Trk extracellular portion. Since these proteins were purified by interaction of the Fc with Staphylococcus aureus protein A and then analyzed for proper molecular size by SDS-polyacrylamide gel electrophoresis (data not shown), the entire Trk extracellular portion was necessarily expressed and folded such that it was not degraded during expression.
Neurotrophin-induced Autophosphorylation of TrkC Receptor Mutants-To test the ligand interactions of the receptor mutants in a cellular environment, stable NIH3T3 cell lines expressing TrkC and the mutants T338A and N335A/T338A/ N342A were constructed. The expression of the receptors was demonstrated by immunoprecipitation with the pan-Trk antibody 443 (5) (Fig. 3A). As reported for the rodent Trk receptors (6,35,42), the apparent molecular weights of the full-length human TrkC and TrkC mutants indicate the presence of a significant amount of glycosylation (calculated molecular mass of full-length, unglycosylated Trk receptors is about 95 kDa). For each of the constructs, two cell lines with different levels of expression were chosen for NT-3-induced autophosphorylation: C-18 and C-3 for TrkC; 48-2 and 48-4 for T338A; and 92-3 and 92-5 for N335A/T338A/N342A. The addition of NT-3 at 100 ng/ml resulted in rapid autophosphorylation of TrkC (C-18/ϩ and C-3/ϩ; Fig. 3B), while no signal was detectable in the absence of NT-3 (C-18/Ϫ and C-3/Ϫ; Fig. 3B). The cell lines expressing T338A at low level  and high level (48-4) did show neurotrophin-dependent activation at 100 ng/ml of NT-3 (48 -2/ϩ and 48 -4/ϩ; Fig. 3B), but the intensity of the signal was reduced when compared with the TrkC cell lines. Note that the expression levels of C-3 and 48-4 are similar (Fig. 3A), but the response to 100 ng/ml NT-3 was strongly reduced in 48-4 when compared with C-3 (Fig. 3B). Finally, no NT-3-dependent activation of the N335A/T338A/N342A mutant receptor was observed when the cell lines 92-3 and 92-5 were tested at 100 ng/ml NT-3 (92-3/ϩ and 92-5/ϩ, Fig. 3B). Dose-response autophosphorylation experiments showed that the TrkC receptor reaches saturation at 50 ng/ml (Fig. 3C). In agreement with the reduced binding affinity of the T338A mutant for NT-3, the neurotrophin-induced autophosphorylation was significantly less efficient and reaches saturation at approximately 500 a Only mutants that showed reduced binding (Ͻ95%) to at least one mAb are included in Table III. All mutants in Table I in boldface or italics were tested against the entire panel of mAbs.
b IC 50 (mutant)/IC 50 (TrkA) from Table I. c TrkA or TrkA mutants were incubated at a concentration of 1 g/ml with immobilized monoclonal antibodies. The amounts of bound receptor mutants are expressed as a percentage of the TrkA value (set at 100%). The results are the average of 2-4 independent binding experiments Ϯ S.D. ng/ml (Fig. 3D). These results establish that in a cellular environment a TrkC receptor mutant with the point mutation T338A does interact with NT-3 less efficiently than native TrkC and that the triple mutation N335A/T338A/N342A resulted in a TrkC receptor mutant that does not respond to NT-3 stimulus by autophosphorylation.

Soluble TrkC Mutants Do Not Inhibit NT-3-induced Neurite Extension on PC12/TrkC Cells-PC12
cells that were engineered to constitutively express rat TrkC respond to NT-3 by forming neurite extensions (35). This terminal differentiation event is dose-dependent (35), and consequently PC12/TrkC cells have become a useful tool for the analysis of TrkC receptor agonists (30). In this study, soluble immunoadhesins of TrkC and mutants of TrkC were used to compete for NT-3 and test for inhibition of NT-3 induction of neurites on PC-12/TrkC cells in analogy to experiments where the soluble Trk receptor immunoadhesins were used to inhibit neurotrophin-induced survival of peripheral neurons (28). In this assay, the level of inhibition of neurite outgrowth by a particular receptor mutant is inversely proportional to its relative affinity to soluble NT-3; reduction in a mutant's ability to compete for and prevent NT-3 binding to TrkC on the cell results in increased neurite outgrowth (compared with soluble native TrkC).
When NT-3 was added to the PC12/TrkC cells at concentra-tions of 1 ng/ml or 250 pg/ml, 67 Ϯ 6 and 59 Ϯ 3% of the cells differentiated and extended neurites (Fig. 4, A and B). When NT-3 was absent, no differentiation of neurons occurred (Fig.  4C). The addition of 40 or 100 ng/ml soluble TrkC to 1 ng/ml NT-3 completely blocked the neurite extension response of PC12/TrkC cells (Fig. 4, D and E; Table VI). The same was observed when the ratio of TrkC to NT-3 was even higher (400 ng/ml TrkC, 250 pg/ml NT-3; Fig. 4F). A concentration of 40 ng/ml soluble TrkC was the minimal concentration at which the saturating amount of 1 ng/ml NT-3 was inhibited completely (data not shown). Similarly, a series of TrkC mutants were tested for their ability to block NT-3 action on PC12/TrkC cells. The triple mutant N335A/T338A/N342A did not inhibit NT-3 at any of the three receptor/NT-3 ratios (Fig. 4, G-I; Table  VI), and the PC12/TrkC cells differentiated similarly as when NT-3 was added to the cells in the absence of receptor. This supports the results from the binding studies where it was shown that this TrkC mutant was unable to interact with NT-3. Even the single alanine mutants of these residues (N335A, T338A, and N324A) as well as N335Q, T338V, N342D, N342Q, and E287K completely lost the ability to inhibit NT-3 at all three receptor/NT-3 ratios (Table VI) and resulted in differentiated PC12/TrkC cells equivalent to Fig. 4, G-I. In addition to mutations of the residues Asn 335 , Thr 338 , Asn 342 , and Glu 287 , we tested mutants that may be close in the threedimensional structure to the main binding determinants. While all of these mutants inhibited NT-3 action at high receptor/NT-3 ratios (similar to native TrkC), the mutants E283A, R285A, L333A, N341A, and N307A were less potent than TrkC at lower ratios (Table VI). The observation that N307A and N341A did not inhibit neurite outgrowth at the lowest ratio tested (1 ng of NT-3/40 ng of receptor) suggests that this assay is more sensitive than the plate-based competition assay in which N307A and N341A showed binding equivalent to native TrkC (Table I).

NGF-induced Autophosphorylation of TrkA Receptor
Mutants-To test the ligand interactions of the receptor mutants in a cellular environment, stable NIH3T3 cell lines expressing TrkA and the mutants T292A, E301A, and T319A/H320A/ N323A were constructed. The expression of the receptors was No Yes a TrkC mutants previously described (14) were used to evaluate binding to monoclonal antibodies. Mutants s1B-s5B are TrkC receptors in which each of the five domains was individually replaced by the homologous TrkB sequence. Mutants ⌬6C and ⌬7C have the first four and three domains of TrkC, respectively, deleted. The mutant A2C, described in this study, is a TrkC in which domain 2 has been substituted with domain 2 from TrkA.
b The origin of each of the five domains of the extracellular portion of the Trk mutant is denoted by A for TrkA, B for TrkB, and C for TrkC. Deleted domains are denoted by a dash.
c TrkC or trkC mutants were incubated at a concentration of 1 g/ml with immobilized monoclonal antibodies. The amounts of bound receptor mutants are expressed as a percentage of the TrkC value (set at 100%). The results are the average of four (eight for ⌬7C) independent binding experiments Ϯ S.D. NB means that less than 1% binding was observed.
d TrkC domains comprising the epitope of each mAb. e mAb blocks NT-3 binding to TrkC.  Table I. b TrkC or TrkC mutants were incubated at a concentration of 1 g/ml with immobilized monoclonal antibodies. The amounts of bound mutants are expressed as a percentage of the TrkC value (set at 100%). The results are the average of four independent binding experiments Ϯ S.D. NB means no binding detectable. demonstrated by immunoprecipitation with the pan-Trk antibody 45 (Fig. 5). As with the TrkC receptors, the apparent molecular weights of the full-length human TrkA and TrkA mutants indicate the presence of a significant amount of glycosylation. For each of the constructs, two cell lines with different levels of expression were chosen for NGF-induced autophosphorylation: A-3 and A-5 for TrkA; 10-6 and 10-9 for E301A; 40-2 and 40-5 for T292A; and 51-4 and 51-7 for T319A/ H320A/N323A. The addition of NGF at 100 ng/ml (ϩ in Fig. 6) resulted in autophosphorylation of TrkA and all TrkA mutants except for the triple mutant (51-4 and 51-7; Fig. 6). Doseresponse autophosphorylation experiments showed that the TrkA receptor reaches saturation at 25 ng/ml (Fig. 7, top). The E301A mutant, with NGF binding reduced by 3-fold, also showed saturation at 25 ng/ml (Fig. 7, middle), while the T292A mutant, with NGF binding reduced by 20-fold, showed saturation at 50 ng/ml (Fig. 7, bottom). Hence, as also found with the TrkC mutants, reduced binding of NGF by TrkA mutants correlated with reduced autophosphorylation in a cellular environment. DISCUSSION Neurotrophins mediate their signal into the cell through specific interactions with tyrosine kinase receptors of the Trk family (reviewed in Refs. 11 and 43). Mutational analyses of human NT-3 (30) and mouse and human NGF (44,45) have defined the binding sites on these neurotrophins for TrkC and TrkA, respectively. In addition, the regions controlling specificity of the neurotrophins have been localized to the central ␤-strand bundle (46) and at least one loop connecting the ␤-strands (47). Notably, the residues that are involved in binding are not identical to those controlling specificity (46).  1-4), the human TrkC mutant T335A at low  and high  levels at 0 and 100 ng/ml NT-3 (lanes 5-8), and the triple mutant N332A/T335A/ N342A at high (92-3) and low (92-5) levels at 0 and 100 ng/ml NT-3 (lanes 9 -12). The receptors were immunoprecipitated with the pan-Trk antiserum 443, and bands were stained with the anti-phosphotyrosine antibody 4G10 after Western transfer as described (35,36). C, NT-3induced dose response of TrkC (cell line C-3) autophosphorylation. Concentrations of NT-3 are given in ng/ml. The gel was exposed to film for 10 s. D, NT-3-induced dose response of autophosphorylation of the human TrkC mutant T335A (cell line . Concentrations of NT-3 are given in ng/ml. The gel was exposed to film for 2 min. While the binding sites on NGF and NT-3 for their cognate Trk receptors have been determined by mutational analysis, the details of the binding site on the Trk receptors have remained unexplored. It has been shown previously that the second immunoglobulin-like domain of the extracellular portion of TrkC is the main determinant for interaction with NT-3 (14) and that both the first and second immunoglobulin-like domains are important for TrkA binding of NGF (14,15). The present study involved a more detailed analysis of the second immunoglobulin-like domain of TrkA and TrkC using molecular modeling and surface scanning mutagenesis.
Loop EF Plays a Major Role in Binding Neurotrophins-TrkC is the most stringent member of the Trk family in terms of specificity; it binds NT-3 with high affinity but not the other neurotrophins (7). TrkA preferably binds NGF, and TrkB preferably binds BDNF and NT-4/5, but at least in vitro, they can bind NT-3 with lower affinity (36, 48 -53). This study has shown that both TrkA and TrkC utilize loop EF as a dominant component of their binding site (Table I). All three Trk receptors possess a conserved sequence in loop EF: abPTHcNNGd, where a is N or D; b is Q, N, or K; c is V, M, or Y; and d is N or D (Fig. 1). While the threonine and second conserved asparagine residues are important for both TrkA and TrkC (Table I), the two receptors differ in their utilization of the other residues in loop EF. In TrkA, Gln 317 , His 320 , and Val 321 are involved in binding, whereas in TrkC the analogous Lys 336 , His 339 , and Tyr 340 are uninvolved; in TrkA, Asn 316 is unimportant, while the analogous TrkC residue Asn 335 is important (Table I). This difference in utilization of residues in this loop suggests that loop EF may play a role in determining the specificity, as well as binding, of the neurotrophins.
Inspection of the three-dimensional structures of immunoglobulins and immunoglobulin-like domains determined by x- FIG. 6. NGF-induced autophosphorylation of receptor mutants. NIH3T3 cell lines expressing human TrkA or TrkA mutants were exposed for 5 min to 0 (Ϫ) or 100 (ϩ) ng/ml NGF. After cell lysis, the receptors were immunoprecipitated with the pan-Trk antibody 45, and bands were stained with the anti-phosphotyrosine antibody 4G10 after Western transfer as described (35,36). , or T292A TrkA mutant (bottom) were exposed for 5 min to 0, 10, 25, or 50 ng/ml NGF. After cell lysis, the receptors were immunoprecipitated with the pan-Trk antibody 45, and bands were stained with the anti-phosphotyrosine antibody 4G10 after Western transfer as described (35,36). The gels were exposed to film for 1 min. ray crystallography reveals that loop EF may have one of two basic conformations: ␣-helical as in immunoglobulin CH2 and CH3 domains (21) or with a side chain:main chain hydrogen bond as in VCAM domain 1 (20), CD4 domain 2 (54,55), and telokin (54). The amino acid sequence of loop EF in the Trks could accommodate either conformation. In all three Trk receptors, loop EF has a hydrophobic residue (Val 321 in TrkA; Tyr 340 in TrkC), which could anchor a putative ␣-helix to the domain, similar to the function of the loop EF tryptophan residues in immunoglobulin CH2 and CH3 domains (21). In the alternative conformation, the first conserved asparagine residue (Asn 322 in TrkA; Asn 341 in TrkC) could provide the side chain:main chain hydrogen bond as found in VCAM domain 1, for example (20). However, changing this asparagine residue to alanine in both TrkA and TrkC had little or no effect on binding (Table I); this would argue against this conformation, since alteration of the analogous EF loop asparagine in VCAM domain 1 affected both binding to receptor and anti-VCAM mAbs (56). If loop EF adopts this alternative conformation, then the loop EF hydrophobic residue would be solvent-exposed and therefore could be important for determination of specificity of the Trk receptors by direct interaction with neurotrophins. The results presented here do not allow a definitive choice between the two possibilities.
Mutations to amino acids other than alanine at positions within loop EF in TrkC point to rather stringent requirements for binding. TrkC residue Asn 342 was the most sensitive to mutation, since all replacements at this position resulted in a reduction of TrkC receptor function (Table I). Mutations at Asn 335 and Thr 338 were tolerated if they involved amino acids that were isosteric or similar in chemical character, but exchanges to larger side chains reduced binding significantly, indicating that the spatial requirements at this part of the binding site may be very restricted.
Loop EF and loop AB, which play a role in the interaction of TrkA and TrkC with their ligands, have previously been shown to be important for molecular recognition in immunoglobulins and immunoglobulin-like domains with their receptors. Mutational analysis of the interaction of IgE with its high affinity receptor demonstrated that, among others, loop EF residues in the IgE C⑀3 domain are major contributors to affinity (57). The three-dimensional structure of the complex between S. aureus protein A and IgG Fc (domains CH2 and CH3) revealed that major contacts are formed by residues in loop EF (Gln 311 and Leu 314 ) and loop AB (Met 252 , Ile 253 , and Ser 254 ) of the CH2 domain (21); these same two loops are likewise utilized when IgG binds to the neonatal Fc receptor (58). In VCAM, residue Asn 65 in loop EF is important for binding to the human receptor but not the murine receptor (56). Although other immunoglobulin-like domains may utilize loops other than AB and EF, e.g. mucosal addressin cellular adhesion molecule (59) and intercellular adhesion molecule (60), the common usage of loops EF and AB by functionally quite different members of the immunoglobulin superfamily suggests that this arrangement of functional side chains provides an example of a structural motif optimal for protein-protein interaction.
The three-dimensional relationship of the five domains in the Trk receptors is unknown. The complexity of the epitopes for the anti-TrkA and anti-TrkC antibodies in conjunction with blocking versus nonblocking character (Tables II and IV) suggests that two or more of the domains may interact with one another. For example, it has been shown previously that TrkC domain 5 alone can bind NT-3 and undergo autophosphorylation (14). However, the anti-TrkC mAb C1435, which binds to domains 1-3, blocks NT-3 binding to full-length TrkC (Table  IV). Similarly, some TrkA mutants in domain 5 affect binding to mAb A1488, which does not map to TrkA domain 5 (Tables  II and III), Since both domain 4 and domain 5 are important for NGF binding in TrkA (14,15) the TrkA domain 5 mutants P269E and T327A/L329A (Tables I and III) could affect binding by disrupting the domain 4-domain 5 interface (rather than by direct interaction with NGF). The short link between domains 4 and 5 also suggests that domain 5 loops BC, DE, and/or FG could form interdomain contacts with domain 4 (similar to the loop interdomain contacts in VCAM (20)); mutations in these loops might also disrupt the domain-domain interface.
Specificity Determinants of Neurotrophin/Receptor Interactions-While both TrkA and TrkC utilize loop EF as a major component of the binding site, the two receptors differ in their interaction with the amino acids in loop EF, as noted above, as well as in their utilization of the rest of the second immunoglobulin-like domain. TrkC is the most stringent among the Trk receptors in its binding of neurotrophins, yet only a few residues outside of loop EF seem to be involved in binding: Glu 287 , Cys 331 , and His 363 (Table I, Fig. 2A). In contrast, in TrkA numerous residues outside of loop EF were major (Pro 269 , His 310 , Ser 340 /Met 342 ) or minor (e.g. Val 261 , His 264 , Thr 292 , Thr 307 , Thr 327 /Leu 329 , Asn 332 , Glu 355 /Asp 356 ) contributors to binding (Table I, Fig. 2B). Two modes of neurotrophin/Trk interaction, which are not mutually exclusive, are possible. First, it has been shown previously that both NGF and NT-3 use residues in their central ␤-strand bundle for binding (30,44). However, NGF also utilizes residues at its N terminus (44,45,61), whereas NT-3 does not (30). This suggests that the central ␤-strand residues in the neurotrophins might interact with loop EF in the Trk receptors, while the NGF N-terminal residues interact with other portions of the TrkA second immunoglobulin-like domain (interactions not present in the NT-3/TrkC interaction). One candidate for control of binding and specificity is ␤-strand B/loop AB. In TrkC, one residue in the N-terminal portion of ␤-strand B (Glu 287 ) is important for binding NT-3, while in TrkA, several residues in loop AB (Val 261 , Met 263 , and His 264 ) are important. Second, while portions of the TrkC second immunoglobulin-like domain outside of loop EF may not play a role in binding, they may play an important role in specificity by providing negative interaction with the noncognate neurotrophin ligands. Hence, NGF may be prevented from binding to TrkC not by residues in loop EF (most of which are conserved) but by residues external to loop EF; similarly, NT-3 may be prevented from binding to TrkA by negative interaction with residues outside of loop EF. This possibility is supported by the difference in binding affinity exhibited by the TrkC R295A and R295D mutants. The TrkC R295A mutant showed no reduction in binding, but a charge reversal, TrkC R295D (TrkA has an aspartic acid at this position), effected a 12-fold reduction in binding (Table I). Further support comes from previous findings that residues in the neurotrophins involved in binding may be segregated from those involved in specificity (46) and that changing specific residues in NT-3 altered binding to TrkA and TrkB but not TrkC (62).
TrkA and TrkC also differ in the nature of residues in the segment connecting the second immunoglobulin-like domain to the transmembrane sequence (Fig. 1). Several residues in the TrkA segment affected NGF binding when mutated (Table I), and these may provide additional binding/specificity determinants not present in TrkC. A naturally occurring TrkA receptor isoform carries an insert of six amino acids between the second immunoglobulin-like domain and the transmembrane sequence (63), and a similar isoform has also been detected in human TrkC (28). When tested in fibroblasts, the presence or absence of the insert did not affect the ligand specificity of the receptor (63). However, when the TrkA isoform containing the six-amino acid insert was expressed in PC12nnr5 cells and analyzed for NT-3-induced biological responses, significantly more neurite outgrowth was observed than in the cell line expressing a variant lacking the insert (64). This suggests that the presence of this insert might modulate the specificity of TrkA in a PC12 cell environment. Although the mechanism of this effect is unknown, it is conceivable that the insert may influence the orientation of the nearby binding domain, or, alternatively, amino acids in the insert may interact directly with neurotrophins.
While this study has provided a high resolution map of the residues in the second immunoglobulin-like domain of TrkA and TrkC involved in neurotrophin binding, it has only provided a glimpse of possible modes of specificity determination for these receptors. The data presented here show that TrkC may have fewer positive interactions with its cognate ligand NT-3 than does TrkA with its cognate ligand NGF and suggest that negative interactions may dominate the specificity. To more fully understand the mechanism of specificity, mutants of the Trks that can recruit binding of the noncognate neurotrophin ligands need to be evaluated, similar to studies on the specificity of neurotrophins (30,46,47).