Identification of a Linear Epitope in Sortilin That Partakes in Pro-neurotrophin Binding*

Sortilin acts as a cell surface receptor for pro-neurotrophins (pro-NT) that upon complex formation with the p75 neurotrophin receptor (p75NTR) is able to signal neuronal cell death. Here we screened a sortilin peptide library comprising 16-mer overlapping sequences for binding of the pro-domains of nerve growth factor and brain-derived neurotrophic factor. We find that a linear surface-exposed sequence, 163RIFRSSDFAKNF174, constitutes an important pro-NT binding epitope in sortilin. Systematic mutational analysis revealed residues Arg163, Phe165, Arg166, and Phe170 to be critical for the interaction. Expression of a sortilin mutant in which these four amino acids were substituted by alanines disrupted pro-NT binding without affecting receptor heterodimerization with p75NTR or binding of ligands that selectively engages the centrally located tunnel in the β-propeller of sortilin. We furthermore demonstrate that a peptide comprising the ligand-binding epitope can prevent pro-NT-induced apoptosis in RN22 schwannoma cells.

neurotrophins (pro-NTs), that can be processed to their mature form in the secretory pathway as well as outside the cell (12)(13)(14). It is well appreciated that pro-NTs may also have functions in their own right, because they elicit activities opposite to those of their mature counterparts (15)(16)(17)(18)(19)(20). Although mature NTs can stimulate neuronal survival and long term potentiation and synaptic strengthening, pro-NTs may induce apoptosis and long term depression (21)(22)(23). Sortilin is required for the death signaling pathway, because it forms a high affinity binding site together with the p75 NTR neurotrophin receptor (24). Inhibiting binding of the growth factor pro-domains to sortilin prevents apoptosis by pro-NTs from occurring in cultured cells as well as in vivo, e.g. following spinal cord injury (25)(26)(27)(28)(29)(30)(31)(32).
Recently, the structure of sortilin, the archetype receptor of the VPS10p-domain family, was solved at the atomic level in complex with the 13-amino acid-long neuropeptide, neurotensin (33). The crystal structure revealed a unique 10-bladed ␤-propeller structure containing a centrally located tunnel with a diameter exceeding 25 Å that harbors the binding site for neurotensin. Although pro-NT binding to sortilin is partly inhibited by neurotensin (25) the size of the pro-NT pro-domain (ϳ100 -110 amino acids) likely precludes entry of the entire domain into the tunnel. We therefore set out to study if a linear surface exposed amino acid sequence in sortilin might assist in pro-NT binding.

EXPERIMENTAL PROCEDURES
Materials and Proteins-To obtain tagged forms of the neurotrophin pro-domains readily for detection in SPOT analysis, we prepared constructs for each protein allowing for addition of N-terminal S-peptide and polyhistidine tags. Template cDNA for human NGF and BDNF consisted of ATCC clones used for generation of fragments spanning residues Glu 1 -Arg 102 of NGF and Ala 1 -Arg 110 of BDNF using the primer pairs 5Ј-GGTATTGAGGGTCGCGAACCACACTCAGAGAGC-AATGTCCC-3Ј and 3Ј-GGGGGAAGTTGTCCTGAGT-GTCCTCGTTCGCCACTCCGAGATTGAGAGGAGA-5Ј, and 5Ј-GGTATTGAGGGTCGCGCCCCCATGAAAGAA-GCAAACATCCGAGG-3Ј and 3Ј-CACGTTTGTACAGG-TACTCCCAGGCCGCGACTCCGAGATTGAGAGGAGA-5Ј, with compatible overhangs for ligation-independent cloning into the pET-30 Xa/LIC vector from Novagen (catalog no. 70073-3) and amplification using Phusion DNA polymerase and following the protocol as provided by the manufacturer. Proteins were expressed in the BL21/DE3 strain of Escherichia coli, efficiently extracted from bacterial inclusion bodies using Bugbuster reagent from Novagen (catalog no. 70921) with added benzonase (Novagen, catalog no. 70750), and purified by standard nickel-nitrilotriacetic acid affinity chromatography in 500 mM NaCl, 5 mM Imidazole, and 20 mM Tris-HCl, pH 8.0. Protein elution was done in buffer supplemented with 20 mM EDTA. Verification of the intact tagged versions of HisS-NGFpro and HisS-BDN-Fpro was carried out by SDS-PAGE analysis followed by Coomassie staining or Western blotting using either antibody against the histidine tag from Sigma (H-1029) and secondary HRP-conjugated anti-mouse antibody from Calbiochem (catalog no. 401207), or alternatively by direct binding of HRP-conjugated S-protein from Novagen (catalog no. 69047-3).
For the production of the sortilin ectodomain, a construct encompassing the entire coding region of the N-terminal part of human sortilin, including the endogenous signal peptide and followed by a C-terminal polyhistidine tag inserted in the pCEP-Pu vector was kindly provided by D. Militz, Max-Delbrueck-Center for Molecular Medicine, Berlin. The DNA was transferred into EBNA 293 cells that were selected by G418 (Invitrogen catalog no. 10131-027, 300 g/ml) and Puromycin from Sigma (catalog no. P8833, 1 g/ml) before proteins were collected from medium conditioned for 48 h, and used for purification by applying to nickel-nitrilotriacetic acid-Sepharose. The secreted recombinant sortilin polypeptide chain spanning the intact extracellular domain of human sortilin thus ended at Ser 725 (ϩAMIEGRGVGHHHHHH containing the fXa site and polyhistidine tag). The quality of the protein was tested by silver staining of SDS-PAGE analysis.
Peptides for binding and competition studies were synthesized in-house at the Charite (Berlin) or purchased from Eurogentec. GST-NGFpro was prepared as previously described and unprocessed precursor pro-NGF was purchased from Millipore (catalog no. GF210).
Sortilin Mutants-Single amino acid substitutions of ligand binding residues within the extracellular domain of sortilin were introduced using the QuikChange II XL site-directed mutagenesis kit from Stratagene (catalog no. 200521) and the wild-type sortilin pCEP-Pu expression construct as template. Mutagenesis primers were from BioTez (Berlin). Multiple rounds of mutagenesis allowed for generation of double, triple, and quadruple mutant constructs knocking out several ligand binding residues in the sortilin domain.
To generate the quadruple mutations in a sortilin construct encoding the entire sortilin receptor, including the transmembrane and cytoplasmic domains, we took advantage of internal HpaI and BsrGI (New England Biolabs catalog nos. R0105S and R0575S, respectively) restriction recognition sites in the sortilin sequence. Using both enzymes we cut out the sortilin fragment containing the 4A mutations from the pCEP-Pu expression vector for the soluble sortilin and exchanged that for the corresponding fragment from a full-length sortilin expression vector in pcDNA3.1/zeo(ϩ) thereby generating a mutant full-length (fl)-sortilin-4A construct. In parallel, the same fragment was also introduced into an endocytosis-deficient mutant of sortilin (34) to generate a version of full-length sortilin-4A that resides at the cell surface (fl-sortilin-4A ⌬endo ).
CD Spectroscopy-The purity of sortilin proteins was assessed by SDS-PAGE, and the purified proteins were dialyzed against 20 mM NaH 2 PO 4 , pH 7.4. Protein concentrations were determined using the absorbance at 280 nm and an estimated extinction coefficient of 133,240 M Ϫ1 cm Ϫ1 (from the ExPaSy proteomics server tools). Ten CD spectra were recorded at 25°C for each protein on a Jasco J-810 spectropolarimeter (Jasco Spectroscopic, Japan) using a polypeptide concentration of ϳ0.1 mg/ml and a cuvette of 2-mm path length. CD data were obtained in the range from 260 to 200 nm at a resolution of 1 nm using a bandwidth of 2.0 nm. The scan speed was 100 nm/min, and the response time was 1 s.
Surface Plasmon Resonance Analysis-Determination of direct binding of ligand to immobilized protein was performed on a biacore2000 instrument (Biacore, Sweden) using CaHBS as standard running buffer (10 mM HEPES, pH 7.4, 140 mM NaCl, 2 mM CaCl 2 , 1 mM EGTA, and 0.005% Tween 20). A biosensor chip from Biacore (CM5, catalog no. BR-1000-14) was activated using the NHS/EDC method as described by the supplier followed by coating with sortilin to a protein density of 79 fmol/ mm 2 and used for affinity measurements of the recombinant pro-domains of NGF and BDNF. Preparation of a biosensor surface with pro-sortilin followed an equal procedure. Regeneration of the flow cell after each cycle of ligand binding experiment was done by two 10-l pulses of regeneration buffer (10 mM glycine-HCl, pH 4.0, 500 mM NaCl, 20 mM EDTA, and 0.005% Tween 20) and a single injection of 0.001% SDS. Fitting of sensorgrams for affinity estimations was done using BIAevaluation version 3.1.
Following similar protocols, immobilization of HisS-NGFpro or HisS-BDNFpro was also done on a CM5 biosensor chip using the NHS/EDC coupling kit according to the manufacturer's instructions (Biacore, Sweden), giving similar surface densities of immobilized protein (ϳ300 fmol/mm 2 ). Purified peptides were applied to the chip at increasing concentrations to verify the direct binding of pro-neurotrophic domains to linear sortilin peptides. This chip was subsequent used to examine the binding of 390 nM wild-type sortilin domain in CaHBS buffer at a flow of 5 l/min, in the absence or presence of competing sortilin peptide. In another competition assay, we applied the unprocessed versions of pro-NGF (50 nM) and pro-BDNF (50 nM) as well as receptor-associated protein (RAP, 90 nM) to immobilized sortilin and measured the binding in absence or presence of 200 M competing peptide.
Cellulose Membrane Preparation-We generated peptide libraries for all members of the VPS10p-domain receptor gene family or specific peptide variations in terms of substitution or length analyses of identified ligand binding peptides. A total of 2181 peptides was used for representation of the sortilin gene family, corresponding to 273 peptides for sortilin (accession code: CAA66904), 734 peptides for sorLA (accession code: NP_003096), 389 peptides for sorCS1 (accession code: NP_001013049), 382 peptides for sorCS2 (accession code: Q96PQ0), and 403 peptides for sorCS3 (accession code: CAI64579), with a 13-amino acid overlap between 16-mer peptides (35).
A cellulose support was prepared as an N-modified celluloseaminohydroxylpropyl ether membrane, and all rounds of synthesis started with spot definition by 9-fluorenylmethoxycarbonyl-␤-alanine-pentafluoophenyl ester that created an alanine linker between peptide and membrane. Then followed an automated linear synthesis of stepwise addition of the different amino acids protected at their N-terminal by 9-fluorenylmethoxycarbonyl and appropriate side-chain protection for the growing peptide chain. The pattern of de-protection, activation, and coupling continued until 16-mer peptides were produced, resulting in an equally distributed array of covalently anchored peptides to the cellulose support at their C-terminal ends with N-terminal free ends (details are given in Ref. 36). Removal of the side protection groups was done in two steps: First the membrane was treated with 90% trifluoroacetic acid (in dichlormethane, containing 3% triisobutylsilane and 2% H 2 O) and secondly with 60% trifluoroacetic acid (in dichlormethane, containing 3% triisobutylsilane and 2% H 2 O). To remove trifluoroacetic acid salts the membrane was washed several times with H 2 O, ethanol, Tris-buffered saline, and ethanol again, and finally dried. Finally, the membrane was blocked in blocking buffer from Sigma (catalog no. B6429) dilated in Tris-buffered saline (pH 8.0) and supplemented with 5% saccharose (Merck, catalog no. K32055087-422) for 2 h before the predefined peptide library is ready for ligand binding analysis.
Binding Studies of Cellulose-bound Peptides-The membrane-bound libraries were incubated with the combined S-peptide and polyhistidine-tagged pro-domains (10 g/ml) in blocking buffer overnight at 4°C, followed by a second incubation with 1 g/ml of HRP-conjugated S-protein from Novagen (catalog no. 69047-3) also in blocking buffer but for 3 h at room temperature. Subsequently, the membrane was washed three times for 10 min with Tris-buffered saline before quantitative characterization of bound ligand was carried out using the UptiLight chemiluminescence substrate from Uptima (catalog no. UP99619A) and the LumiImager instrument from Roche Diagnostics, providing the spot signal intensities in Boehringer light units. Alternatively, detection of bound ligand was performed by an immunochemical assay with an antibody against the histidine tag from Sigma (catalog no. H-1029) and a secondary HRP-conjugated anti-mouse antibody from Calbiochem (catalog no. 401207). Incubations followed standard Western blotting procedures and spot detection as above.
The method of substitution analysis and length analysis to identify unique single amino acid residues and to determine the minimal peptide sequence, respectively, for binding HisS pro-domains to the sortilin peptide, followed similar protocols as for the initial testing of ligand binding to the spotted membrane.
Pulldown Assay-The expressed extracellular domains of sortilin-WT or sortilin-4A was incubated with either GSTtagged NGFpro or the propeptide of sortilin and precipitated using 100 l of glutathione (GSH)-Sepharose beads (Amersham Biosciences, catalog no. 17-0756-01). The amount of applied receptor domains was determined by precipitation using Talon beads as control. Bound protein was separated by SDS-PAGE analysis and visualized using anti-histidine antibody by standard Western blotting.
Ligand Internalization by HEK 293 Cells-To study the effect of disrupting the identified binding site in full-length sortilin, HEK 293 cells were stably transfected with plasmids encoding either fl-sortilin-wt or fl-sortilin-4A. Receptor expression was confirmed by Western blotting using a sortilin-specific monoclonal antibody and by immunofluorescence. For immunofluorescence, cells were fixed in 4% paraformaldehyde, permeabilized with 0.1% Triton X-100, and subsequently incubated with rabbit anti-sortilin polyclonal antibodies followed by donkey anti-rabbit Alexa-488-conjugated secondary antibodies (1:300). Internalization of bound ligands was studied by the incubation of cells for 30 min at 37°C with anti-sortilin polyclonal antibodies (10 g/ml), GST-NGFpro (1 g/ml), or 100 nM BDNF pro-domain (Alomone) diluted in culture medium. Internalized GST-NGFpro was visualized by the incubation with goat anti-GST (Amersham Biosciences, 1:300) followed by donkey anti-goat Alexa-488-conjugated secondary antibodies. Internalized BDNF pro-domain was visualized with rabbit polyclonal antibodies raised against the BDNF pro-domain (Alomone, 1:200).
p75 NTR :Sortilin Co-immunoprecipitation-HEK 293 cells stably transfected with constructs encoding fl-sortilin-WT ⌬endo or fl-sortilin-4A ⌬endo were incubated in phosphate-buffered saline (with 1 mM CaCl 2 and MgCl 2 ) for 90 min at room temperature and then treated with 5 nM reducible protein crosslinker dithiobis(succinimidylpropionate) (Pierce) according to the manufacturer's instructions. After wash, cells were lysed on ice for 10 min in TNE buffer (20 mM Tris-HCl, pH 8.0, 1% Nonidet P-40, 10 mM EDTA) supplemented with complete protease inhibitor mixture. Samples were immunoprecipitated overnight at 4°C by use of Gammabind G-Sepharose beads (Amersham Biosciences) coupled with anti-p75 NTR (10494) from Abcam. Unspecific binding was removed by washing five times in Tris-buffered saline containing 0.05% Tween 20, and proteins were eluted by boiling samples in reducing sample buffer (20 mM dithioerythritol, 2.5% SDS). Protein samples were subjected to SDS-PAGE and Western blotted using anti-sortilin from BD Transduction Laboratories (catalog no. 612100) and anti-p75 NTR .
Apoptosis Assay-The effect of sortilin-derived peptides on pro-NGF-induced cell death was studied in RN22 schwannoma cells. Cells were cultured in Dulbecco's modified Eagle's medium added 10% fetal bovine serum, and seeded in 96-well plates at a density of 10,000 cells/well. After 24 h, the cells were washed twice in serum-free medium and incubated for 72 h in the presence of increasing concentrations of the sortilin-derived sort166 -181 peptide diluted in Dulbecco's modified Eagle's medium without serum and phenol red, containing 30 nM selenium, 5 g/ml insulin, 5 g/ml transferrin, 30 nM triiodothyronine, and 10 nM pro-NGF (Chemicon). Dead and live cells were subsequently quantified using the MultiTox-Fluor Multiplex cytotoxicity assay (Promega, Madison, WI).

Identification of a Linear Pro-NT Binding Sequence in Sortilin-
We first produced the pro-domains of NGF and BDNF as recombinant fusion proteins containing two N-terminal tags, a polyhistidine and an S-peptide tag, HisS-NGFpro and HisS-BDNFpro, respectively, to facilitate protein purification and detection of ligand binding to immobilized sortilin peptides (supplemental Fig. S1A). The S-tag system is based on the interaction of the 15-amino acid S-tag peptide with a 104-amino acid-long S-protein derived from pancreatic ribonuclease A (37). To confirm the receptor-binding properties of the recombinant pro-domains, we tested their binding to immobilized sortilin using surface plasmon resonance (SPR) analysis. The calculated affinities of HisS-NGFpro (K D ϳ 9 nM) and HisS-BDNFpro (K D ϳ 1.6 nM) were in accordance with past findings (supplemental Fig. S1B) (25,27).
To screen for potential linear ligand binding sites in the VPS10p-domain receptors, we synthesized a peptide library in which each of the receptors were dissected into consecutive 16-mer peptides overlapping by 13 amino acids and subsequently spotted onto filters. Binding of HisS-NGFpro and HisS-BDNFpro was then tested for their abilities to interact with the receptor peptide libraries by a so-called SPOT binding analysis (38 -40). A specific sequential triplet of signals was observed for peptides derived from sortilin, suggesting that this particular VPS10p-domain contains a linear binding epitope for the neurotrophin pro-domains. Interestingly, the HisS-NGFpro and HisS-BDNFpro gave virtually identically results, suggesting that they may share a common (linear) binding site (Fig. 1).
We focused the subsequent studies on sortilin and prepared a second peptide array comprising 273 immobilized sortilinderived peptides. In this set of experiments we took advantage of both tags present in the HisS-BDNFpro ligand as we used either a peroxidase-conjugate of the S-protein or an immunoassay with an antibody against the polyhistidine tag for detection. The analysis of the two assays gave slightly different, but overlapping, results. Whereas detection with S-protein resulted in strong positive signals for peptides 67-69, spanning sortilin residues 166 -187 (sort166 -187) (Fig. 2, A and C), the immunoassay identified peptides 64 -66, comprising residues 157-178, as the ligand binding site (Fig. 2, B and C). The shift in signal between the two detection methods likely represents different sterical hindrance from the S-protein and antipolyhistidine antibody. We considered the signal corresponding to peptides 22-26 in the anti-histidine blot unspecific, because a similar spot pattern was observed on membranes incubated with the secondary antibody alone (data not shown). Collectively, these experiments suggest that amino acids 163-174 in sortilin, corresponding to the overlapping residues in the peptide sequences, constitute a binding epitope for pro-NT (Fig. 2C).
To confirm binding of HisS-NGFpro and HisS-BDNFpro to sort163-174, we synthesized a peptide comprising these 12 amino acid residues. The peptide was subsequently subjected to SPR analysis (BIAcore) using a Sensorchip coupled with HisS-NGFpro and HisS-BDNFpro, thereby reversing the system from immobilized peptides to peptides in solution. We found that the higher affinity of the BDNF pro-domain as compared with the NGF pro-domain for sortilin (supplemental Fig. S1B and Refs. 25, 27) could be replicated when using the sortilin peptide. This suggests that the linear ligand recognition sequence sort163-174 may account for the difference in binding properties between pro-BDNF and pro-NGF (Fig. 2D).
Identification of Key Residues in Sortilin for Pro-domain Binding-Using substitutional analysis we next set out to map key residues in the linear sortilin sequence that contribute to pro-neurotrophin binding. A variant of the SPOT synthesis method was applied to generate a new library in which every residue in the sequence 163 RIFRSSDFAKNFVQTD 178 had been replaced by all 20 naturally occurring L-amino acid residues. The SPOT filters were then incubated with ligand, as exemplified by HisS-BDNFpro (Fig. 3A), and the intensity of the binding signals were quantified and used to calculate the replacement variability for each amino acid. Incubation with HisS-BDNFpro and HisS-NGFpro showed that both ligands were critically dependent on amino acids Arg 163 , Phe 165 , and Arg 166 inasmuch as the variance was low, i.e. within 10 -35% (Fig. 3, B and C). Interestingly, while the variance of residues Phe 170 and Phe 174 amounted to 60 -75% for HisS-NGFpro, indicating little impact on binding, the same amino acids were noticeable more important for the interaction between HisS-BDNFpro and the sortilin peptide as reflected by a variance between 30 and 35%.
To determine the minimal linear sortilin sequence capable of binding the pro-NT pro-domains we carried out a so-called length analysis. In practice, we synthesized the peptide sort163-178 and removed consecutive single residues from the N-and/or C-terminal ends of the peptide (Table 1). Using this assay we narrowed down the length of the binding epitope to a central region of the sequence corresponding to the 12-mer fragment 163 RIFRSSDFAKNF 174 . This result is in accordance with the substitution analysis, which also showed that residues 175 VQTD 178 do not significantly influence ligand binding (Fig.  3B). Interestingly, by further trimming the peptide we found that both halves of this fragment corresponding to the tetramer 163 RIFR 166 and the pentamer 170 FAKNF 174 , respectively, were able to associate independently with HisS-BDNFpro (Table 1) and HisS-NGFpro (data not shown).
This observation prompted us to repeat the substitution analysis, but this time with peptide libraries of 163 RIFRSSDF 170 and 170 FAKN-FVQTD 178 , because each of these fragments contains one of the two putative binding epitopes. Using this approach we confirmed that an exchange of Arg 163 , Phe 165 , or Arg 166 within the N-terminal half and Phe 170 , Lys 172 , or Phe 174 within the C-terminal half significantly decreased binding of the HisS-BDNFpro ligand (Fig. 4A). Similar results were obtained for HisS-NGFpro (data not shown), further corroborating the presence of a shared binding site in sortilin for both pro-NTs. In further support of this notion, binding of HisS-BDNFpro to immobilized sortilin, as determined by SPR analysis, was completely prevented by preincubation with saturating concentrations of HisS-NGFpro (supplemental Fig.  S2). Collectively, the above data suggest the existence of a common binding epitope comprising two interaction sites in which three residues in the second epitope may account for the stronger binding of HisS-BDNFpro as compared with HisS-NGFpro (Fig. 4B).
Ligand Binding to Sortilin-4A-Sortilin is produced with a propeptide that needs to be cleaved off to allow entrance of ligands into the tunnel (41). We therefore expressed a noncleavable soluble sortilin variant as described (41), immobilized it on a biosensor chip, and compared binding of unprocessed pro-BDNF and neurotensin to the mutant with that of the fully processed wild-type receptor. We found that both receptors bound pro-BDNF (Fig. 5A), in agreement with the existence of a binding site located away from the blocked central cavity. For comparison, binding of neurotensin was completely abolished in the non-processed sortilin variant (Fig. 5B), demonstrating that binding of this ligand entirely depends on access to the tunnel epitope.
To determine the impact of the linear pro-NT binding epitope in the full-length VPS10p-domain, we expressed in EBNA 293 cells a set of single residue alanine substitution mutants (R163A, F165A, R166A, F170A, K172A, and F174A) of the sortilin ectodomain. As a control, we included the mutation R160A, a residue outside the ligand-binding peptide sequence. All mutants were effectively secreted from the cells in similar quantities and showed a migration pattern in non-reducing gel analysis identical to that of the non-mutated sortilin domain, suggesting correct disulfide bridging of the mutants (supplemental Fig. S3). To further demonstrate the intact structure of the VPS10p ␤-propeller, we determined the CD spectra for each variant and found that none of the mutations resulted in aberrant folding of sortilin (supplemental Fig. S3). Binding of GST-tagged NGFpro was subsequently tested using SPR analysis with each receptor mutant immobilized on individual flow cells at similar protein densities, thus allowing for direct comparison of bound ligand. However, none of the single residue mutants displayed any substantial reduction in ligand affinity (maximally 2-to 3-fold reduced binding) as deduced by fitting a series of ligand concentration sensorgrams (supplemental Fig. S3B and Table 2).
Based on the existence of two independent motives within the linear binding site (cf. Table 1) we speculated that failure of the single residue mutants to show reduced GST-NGFpro affinity was accounted for by binding to the non-mutated residues in the second site. We therefore hypothesized that disrupting an entire motif in combination with substitution of a single critical residue from the adjacent motif might be required to efficiently abolish ligand binding. To address this premise, we mutated the four most important residues within the 163 RIFRSSDFAKNF 174 motif (cf. Fig. 4) thereby generating a construct with a R163A,F165A,R166A,F170A quadruple mutation, denoted sortilin-4A. This receptor mutant was also efficiently expressed by EBNA 293 cells, able to be purified using Ni 2ϩ -affinity chromatography, and folded properly as demonstrated by CD analysis (supplemental Fig. S3C).
The interaction of sortilin-4A with NT pro-domains was studied by SPR analysis and using a ligand precipitation assay. HisS-BDNFpro and GST-NGFpro were immobilized on the sensor chip, and equal amounts (100 nM) of wild-type sortilin (sortilin-WT) and sortilin-4A were applied. We found that HisS-BDNFpro and GST-NGFpro exhibited considerable stronger binding to sortilin-WT as compared with sortilin-4A (Fig. 6, A and B). In another set of SPR experiments, where sortilin-WT and sortilin-4A were the immobilized parts, we found that GST-NGFpro bound to the mutant receptor with a Ͼ10-fold lower affinity, i.e. K D ϳ26 nM versus ϳ2 nM, corroborating an important role of the linear binding site also in the VPS10p holo-domain (Fig. 7, A and B). A similar decrease in GST-NGFpro binding to sortilin-4A was revealed when comparing the ability of GST-NGFpro to precipitate wild-type and mutant sortilin in solution. In practice, the two VPS10p-domains were incubated with GST-NGFpro, allowed to form complexes, and then precipitated by glutathione beads. The precipitates were resolved by SDS-PAGE analysis, and bound sortilin was subsequently detected by Western blotting (Fig.  7C). In line with the SPR analysis the mutated sortilin domain exhibited considerably less interaction with GST-NGFpro. To assure that binding of ligands that engage the tunnel of the ␤-propeller far away from the site of the quadruple mutation was unperturbed, we repeated the experiment using the GSTtagged sortilin propeptide as a ligand (33). Importantly, this ligand efficiently precipitated sortilin-4A, confirming that the mutated domain is capable of interacting with other ligands independent of the linear binding site. To provide evidence for a functional role for the linear pro-NT binding site in the full-length and membrane-associated receptor, we expressed the entire wild-type and mutated sortilin receptors, designated fl-sortilin-WT and fl-sortilin-4A, respectively, in HEK 293 cells. The mutated receptor was mainly localized in perinuclear vesicles in a pattern indistinguishable from that of wild-type sortilin (Fig. 8A). Identical expression levels of fl-sortilin-WT and -4A were documented The quantification of a representative experiment testing the influence of the length of the sort163-178 peptide binding to HisS-BDNFpro in a modified SPOT analysis. Peptides with a high ligand affinity are shown on a grey background. For this presentation, an intensity threshold for the signal intensity of 30,000 BLU have been chosen as indication for high affinity. The presence of the two identified minor binding sites 163 RIFR 166 and 170 FAKNF 174 is presented on a black background.
by Western blot analysis of cell extracts (Fig. 8A). Normal internalization of an antibody against the sortilin extracellular domain confirmed that the mutant was also capable of endocytic uptake. Notably, the GST-propeptide of sortilin was equally well taken up by cells expressing wild-type and mutant receptor (Fig. 8B). However, whereas fl-sortilin-WT efficiently internalized the pro-domains of both NGF and BDNF, cells expressing fl-sortilin-4A did not (Fig. 8C). This demonstrates that the linear and surface-exposed pro-NT binding site is also functional in the full-length and membrane-anchored receptor.
Because sortilin and p75 NTR can physically associate in the plasma membrane (11,25), likely by a mechanism that is independent of the ␤-propeller tunnel, we next examined if fl-sortilin-4A were impaired in p75 NTR binding. To increase cell surface expression and the co-immunoprecipitation signal, we deleted a previously well described endocytosis signal (34) in the sortilin cytoplasmic tail of both receptors generating fl-sortilin-WT ⌬endo and fl-sortilin-4A ⌬endo , respectively. Interest-

TABLE 2 Binding kinetics of GST-NGFpro binding to sortilin mutants
Concentration series of 10, 20, 30, 40, and 50 nM of GST-NGFpro was applied to biosensor chips with immobilized sortilin single residue mutants. Affinities and rate constants were estimated using the BIAevaluation software, using the 1:1 Langmuir binding isotherm.  APRIL 16, 2010 • VOLUME 285 • NUMBER 16

Pro-BDNF/Pro-NGF Binding Site in Sortilin
ingly, the two receptor variants precipitated p75 NTR equally well, suggesting that hetero-oligomerization between sortilin and p75 NTR is independent of the surface-exposed pro-NT binding epitope (Fig. 8D).
The Linear Sortilin Epitope Prevents Pro-NT Binding and Death Induction-The importance of the linear epitope for pro-NT binding was further substantiated in a competition assay where the pro-domains of BDNF and NGF were immobilized on a biosensor chip, and the interaction with saturating concentrations of 200 nM soluble sortilin-WT was measured in the absence or presence of a competing peptide spanning the ligand-binding sequence. A clear decrease in receptor-ligand complex formation was observed in the presence of peptide. Thus, the response for sortilin-WT binding to HisS-BDNFpro was lowered from 300 RU in the absence of competitor to 100 RU in the presence of 200 M sort166 -181 (Fig. 9A). Quantification of the equilibrium response at increasing amounts of peptide showed a concentration-dependent competition profile with the inhibition being stronger for the interaction of sortilin with HisS-BDNFpro than for HisS-NGFpro, in line with a higher affinity of the former pro-domain (Fig. 9B). Moreover, the peptide also blocked the interaction of pro-BDNF and pro-NGF with sortilin immobilized on the chip surface (Fig. 10, A  and B). As a control for the specificity of the inhibitor, binding of sortilin to an unrelated ligand, the RAP (42, 43), was not affected by the sortilin peptide (Fig. 10C).
We finally asked if the sortilin-derived peptide was capable of blocking pro-NGF-induced apoptosis by the sortilin:p75 NTR complex. To this end we incubated the schwannoma cell line RN22 with 10 nM pro-NGF in the absence or presence of the peptide sort166 -181 and scored apoptosis after 72 h. Indeed, the peptide inhibited pro-NGF-induced cell death in a concentration-dependent manner with Ͼ60% reduction in the   (10,20,30,40, and 50 nM) tested for binding to immobilized extracellular domains of sortilin-4A (A) and sortilin-WT (B), demonstrating a strong decrease in the binding capacity for the NGF pro-domain upon the quadruple mutation in sortilin-4A. A higher than 10-fold decrease in affinity (WT: K D ϳ2 nM versus 4A: K D ϳ26 nM) was estimated using the BIAevaluation software. C, medium from EBNA 293 cells producing either the soluble ectodomain of sortilin-WT or sortilin-4A were incubated with a GST-tagged variant of the receptor propeptide (GSTpropept) or GST-NGFpro and precipitated by glutathione (GSH) beads. The amount of total secreted sortilin-WT and sortilin-4A (i.e. input) was determined by precipitation using Talon beads binding to the histidine tag within the sortilin domains. The precipitated proteins were subjected to SDS-PAGE analysis and visualized by Western blot analysis for sortilin. dead/live cell ratio at the highest peptide concentration used (Fig. 11A). Because reduced killing could also arise from a deficient physical interaction between sortilin and p75 NTR , we also examined if peptide sort166 -181 would interfere with receptor hetero-oligomerization. We observed no effect on the ability of p75 NTR to precipitate sortilin in the presence of 100 M of the peptide inhibitor (Fig. 11B). These data show that blocking the interaction with the linear epitope in sortilin is sufficient to specifically prevent binding of pro-NTs to sortilin, leaving ligand binding inside the tunnel as well as receptor hetero-oligomerization intact.
In conclusion, we find that sortilin harbors a linear binding epitope that selectively engages pro-NTs. We propose that this sequence may be used to derive a peptide antagonist or small organic molecule that can prevent apoptosis.

DISCUSSION
Pro-neurotrophins were initially considered inert precursors that merely serve as a reservoir for their mature counterparts, but it is now well established that they also exhibit activities on their own (16,23). Although mature NTs stimulate neuronal survival and differentiation by engaging their specific tyrosinereceptor kinases (Trks) in conjunction with p75 NTR , pro-NTs can elicit apoptosis in cells expressing sortilin and p75 NTR (1,24,44,45). Although the signaling pathway has not been fully elucidated, it has been shown that the simultaneous binding of the pro-NT pro-domain to sortilin and the mature part to p75 NTR induces formation of a tripartite death-inducing complex capable of activating the Jun-kinase signaling pathway (46 -49). We here identified a linear and surface-exposed amino acid sequence in sortilin that partakes in pro-NT binding and plays a pivotal role in death induction by pro-NTs.
The molecular structure of the sortilin ectodomain in complex with the neuropeptide neurotensin was recently solved at atomic resolution. The structure revealed an unprecedented 10-bladed ␤-propeller encircling a large tunnel that encompasses the binding site for neurotensin, the sortilin propeptide, and RAP (33,41,50). Intriguingly, we found that the surface of the receptor may engage in binding of pro-NTs. First, the soluble sortilin peptide sort163-174 bound to immobilized NT pro-domains and vice versa. Second, this interaction critically depended on the surface-exposed amino acids Arg 163 , Phe 165 , and Arg 166 for both NGFpro and BDNFpro, whereas Phe 170 and Phe 174 also contributed to the binding of the BDNF pro-domain, an observation that may explain the higher affinity of sortilin for pro-BDNF than pro-NGF ( Fig.  2D and supplemental Fig. S1) (25,27). Third, mutated sortilin harboring alanine substitutions in the surface-exposed residues Arg 163 , Phe 165 , Arg 166 , and Phe 170 showed only little interaction with pro-NTs as determined by SPR analysis, coimmunoprecipitation, and cellular uptake studies, whereas binding of the sortilin propeptide was unperturbed. Fourth, the peptide sort166 -181 competitively inhibited sortilin binding of pro-NTs, but not of p75 NTR , which likely engages an alternative surface epitope on the ␤-propeller, nor RAP or the sortilin  APRIL 16, 2010 • VOLUME 285 • NUMBER 16

Pro-BDNF/Pro-NGF Binding Site in Sortilin
propeptide that selectively target the tunnel. In agreement, the sortilin peptide was a potent inhibitor of pro-NT induced apoptosis in RN22 cells.
In the SPOT analysis, application of the two detection methods gave similar, though not identical, results confirming the specificity of the binding between the receptor peptides and the NT pro-domains. Although HisS-NGFpro and HisS-BDNFpro binding to the sortilin peptides gave rise to the same spots when both were assayed with the S-protein (peptides 67-69) (Fig. 1), we saw a consistent shift toward earlier sequences (peptides 64 -66) when developed with the anti-histidine antibody (Fig.  2). We speculate this difference could be accounted for by sterical hindrance of the detection system from the solid support. Thus, the S-protein is only 104 residues, whereas the anti-histidine immunoassay requires binding of two large antibodies, each roughly of 1500 amino acids.
Interestingly, the identification of two positively (Arg 163 and Arg 166 ) but no negatively charged residues in the sortilin epitope suggests that electrostatic interactions may contribute to the binding between the receptor and pro-NTs. In agreement with this model, the BDNF pro-domain contains 17 acidic residues many of which are located in a region suggested to be vital for the interaction with sortilin (51).
Sequence analysis of sortilin predicted the presence of several Asp-box motives throughout the VPS10p-domain (52). Indeed, x-ray analysis of sortilin revealed that Asp-boxes constitute a well defined structure that folds into a surfaceexposed hairpin-loop bridging strands 3 and 4 in each of the 10 ␤-blades of the propeller (33). The Asp-box consensus sequence can be expressed as X-X-Ser-X-Asp-X-Gly-X-Thr-Trp/Phe-X, where X represents any amino acid (53). Interestingly, this motif, 163 RIFRSSDFAKNF 174 , is part of the identified pro-NT binding in blade 2 of the receptor. Aspboxes normally serve a structural function in ␤-propellers, where the conserved residues participate in folding and ensure the stability of the domain (54). The variable residues are free to carry out other functions, e.g. ligand binding as observed for nucleotide binding to microbial ribonucleases (53), in line with our finding that these amino acids are involved in binding of the NT pro-domain. It is noteworthy that, although most ␤-propellers bind their ligands at the top face, as is also the case for neurotensin and the sortilin propeptide (33), binding to the bottom face or side domain has also been reported (55-57).  The amount of pro-NGF-induced cell death after 72 h were quantified using a fluorescence-based assay, and plotted against competitor concentrations as the ratio between dead and live cells in percentage. B, cells co-transfected with constructs for a sortilin variant devoid of endocytosis (fl-sortilin-WT ⌬endo ) together with p75 NTR were incubated in the absence or presence of 100 M sort166 -181 peptide, and immunoprecipitated using an antibody against p75 NTR . Lysates and precipitated proteins were subjected to SDS-PAGE and analyzed by Western blotting for either p75 NTR or sortilin as indicated.
Our data do not exclude that NGFpro and BDNFpro may also target the epitope in the tunnel of the ␤-propeller. Although the residues identified in the current study are exposed on the rim and outer surface of the barrel, it is possible that the Ͼ100-amino acid-long NT pro-domains simultaneously can associate with additional binding epitopes, including the tunnel. This would be in accordance with the suggested highly flexible and extended conformation of the pro-NGF prodomain (58,59) and in line with the recently reported crystal structure of the pro-NGF:p75 NTR complex that shows the prodomain to accommodate an exposed and disordered conformation free to interact with sortilin (49).
By SPOT peptide mapping we only identified one linear binding site in sortilin comprising amino acids 163-174. We believe this fact is due to the tunnel surface binding site for neurotensin being comprised of residues located far away from each other in the primary sequence, e.g. Lys 227 , Ser 283 , Arg 292 , and Tyr 318 (33). These residues are separated by up to 90 amino acids within the primary structure thereby representing a discontinuous binding epitope not suitable for detection by a peptide mapping approach.
There are experimental data to support as well to refute an additional pro-NT binding epitope in the tunnel. We found that, following disruption of the four most critical amino acids (Arg 163 , Phe 165 , Arg 166 , and Phe 170 ), for the interaction with pro-NT there was an ϳ25% residual binding still present, arguing for one or more additional interaction sites (Fig. 6). On the contrary, mutations in residue Ser 283 , which is indispensable for binding of the sortilin propeptide inside the tunnel, completely abrogated binding of this peptide while pro-NGF binding was unperturbed (33). Thus the experimental evidence in support of a composite pro-NT binding site, that overlaps with that of the sortilin propeptide and neurotensin in the tunnel, is contradictory. However, it is possible that pro-NT might engage the tunnel via an epitope in the vicinity of but independent of amino acid Ser 283 . This is supported by the observation that binding of the pro-NGF pro-domain is significantly better inhibited by the 13-amino acid-long full-length neurotensin than by residues 11-13, which constitute the sortilin binding site in the neuropeptide (33). These data may suggest that neurotensin competes with pro-NGF for binding due to steric hindrance in the tunnel. Collectively, the above data underscore the requirement of elucidating the atomic structure of the sortilin:pro-NT complex to precisely and unanimously identify the number and location of all binding epitopes.
Recent evidence has suggested that pro-NT-induced cell deaths play an important role in conditions characterized by apoptosis such as aging, and in pathological conditions, including seizures, spinal cord injury, retinal dystrophy, and Alzheimer disease (15, 18, 24, 28, 29, 60 -65). In accordance, recent reports have described changes in the levels of pro-NTs in many of these conditions, including early stages of Alzheimer disease (60, 66 -68). Of particular interest, increased glycation and lipoxidation of pro-NGF in the hippocampus and entorhinal cortex of Alzheimer disease patients makes it less susceptible to processing into mature NGF. Furthermore, in these patients TrkA expression is decreased, whereas the expression of p75 NTR and sortilin are unchanged (69,70). 4 Possibly, an imbalance between the relationship of receptor expression and proversus mature NGF changes the equilibrium between survival and apoptosis signaling in favor of the latter (71,72). Prompted by the biological significance of pro-NTs in cell death, we tested if the ligand-binding peptide of sortilin might by capable of preventing apoptosis. Indeed, we found that the peptide offered efficient protection against pro-NGF-induced apoptosis in cultured RN22 schwannoma cells, a finding that holds promise for the idea that an antagonist that can prevent pro-NGF binding to sortilin might be an efficient target to prevent pro-apoptotic conditions.
The peptide used in the present work is unlikely to be suitable for therapeutic use due to its modest affinity and accordingly high concentrations required to inhibit pro-NT binding. Yet, the peptide may serve as a lead for rational drug design aimed at increasing the binding affinity for NT pro-domains as well as to enhance the pharmacological properties in terms of solubility, stability, and efficacy. Ideally, such a compound should specifically prevent pro-NT-induced apoptosis, leaving the trophic actions of mature neurotrophins unchanged.