Structural basis of SALM3 dimerization and synaptic adhesion complex formation with PTPσ

Synaptic adhesion molecules play an important role in the formation, maintenance and refinement of neuronal connectivity. Recently, several leucine rich repeat (LRR) domain containing neuronal adhesion molecules have been characterized including netrin G-ligands, SLITRKs and the synaptic adhesion-like molecules (SALMs). Dysregulation of these adhesion molecules have been genetically and functionally linked to various neurological disorders. Here we investigated the molecular structure and mechanism of ligand interactions for the postsynaptic SALM3 adhesion protein with its presynaptic ligand, receptor protein tyrosine phosphatase σ (PTPσ). We solved the crystal structure of the dimerized LRR domain of SALM3, revealing the conserved structural features and mechanism of dimerization. Furthermore, we determined the complex structure of SALM3 with PTPσ using small angle X-ray scattering, revealing a 2:2 complex similar to that observed for SALM5. Solution studies unraveled additional flexibility for the complex structure, but validated the uniform mode of action for SALM3 and SALM5 to promote synapse formation. The relevance of the key interface residues was further confirmed by mutational analysis with cellular binding assays and artificial synapse formation assays. Collectively, our results suggest that SALM3 dimerization is a pre-requisite for the SALM3-PTPσ complex to exert synaptogenic activity.

. This study revealed that LRR domains of SALM5 are involved in dimerization, and the different structures indicated that overall structural features are conserved between SALM2, SALM3 and SALM5 20 . However, structural details of SALM family beyond SALM2 and SALM5 remain to be determined including SALM3 and the molecular details of its interaction with LAR-RPTPs. SALM3 is expressed early during embryonic stage in the brain, whereas the other SALMs are expressed from the embryonic stage, and the expression is maintained or increased after birth 21 . In vivo studies using SALM3 knockout (SALM3 −/− ) mice displayed markedly reduced excitatory synapse number and locomotor activity, and exhibit behavioral hypoactivity, which underlines the importance of SALM3 in excitatory synapse development 14 . SALM3 interacts with all three known members of LAR-RPTPs (LAR, PTPσ and PTPδ) to induce presynaptic differentiation 14 . The interaction between SALM3 and LAR-RPTPs is markedly enhanced in the presence of the LAR-RPTP mini-exon B (meB) 14,20 . Besides SALMs, LAR-RPTPs form trans-synaptic complexes with different families of postsynaptic adhesion proteins, which include NGL-3, IL1RAcP, neurotrophin receptor TrkC and the Slitrks 22,23 . The extracellular region of LAR-RPTPs contain three Ig domains, four to eight FnIII domains and multiple splicing variants at several sites: mini-exon A (meA) in the Ig2 domain, mini-exon B (meB) between Ig2 and Ig3 domain and mini-exon C (meC) located in the FnIII domain (Fig. 1a). In the cytoplasmic region, LAR-RPTPs contain two tandem tyrosine phosphatase domains 6,22 . The Ig2 and Ig3 domains of PTPδ and PTPσ are involved in interaction with the LRR and Ig domains of SALM5 17,18 .
In this study, we have solved the crystal structure of mouse SALM3 LRR domain and show that it forms a dimer that is close to identical to that of SALM5, and the residues that are involved in dimerization and their interactions are largely conserved, and our complementary biophysical studies support this. We also solved the small angle X-ray scattering (SAXS) structure of SALM3-PTPσ complex that verifies the formation of the 2:2 trans-heterotetrametric complex between the proteins. Mutational studies with a cell-based binding assay and a synapse formation assay confirm the binding mechanism of SALM3 to PTPσ and the identity of the key interface residues that are largely the same as for the SALM5-PTPδ interaction. Therefore, we conclude that SALM3 and SALM5 interact in similar manner with their LAR-RPTP ligands to induce presynaptic differentiation. Solution The overall SALM3 LRR domain structure. The missing parts for the LRR domain shown as dashed lines, LRRNT and LRRCT-capping domains are colored blue (LRRNT) and red (LRRCT) respectively. The Nand C-termini of the protein are marked with 'Nt' and 'Ct' and the residue number, disulphides are indicated in yellow as sticks. (c) Dimeric assembly observed in the crystal structure with one of the subunits shown with transparent surface, and N-glycosylation present in the LRRNT Asn25 shown as sticks. On the left structure is viewed from the "top" towards the concave side of the LRR dimer, orthogonal to the interface, and on the right from the side, glycosylated Asn25 marked. Molecular structure figures were prepared with PyMol, version 1.5.0.5 (www.pymol .org). www.nature.com/scientificreports/ scattering studies suggest that the dimeric complex has significant conformational flexibility. Our results also confirm that dimerization is absolute requirement for induction of synapse formation by the SALM3, similar to SALM5, and thus appears to be common for the SALM family proteins involved in synapse formation.

Results and discussion
The structure of the SALM3 LRR domain. The structure of mouse SALM3 LRR domain (Fig. 1b, c) was solved at 2.8 Å resolution in space group P2 1 ( Table 1). The crystals exhibited high degree of twinning, making the structure solution demanding (see methods). However, it was possible to obtain a good quality refined model once the correct space group was resolved. Four molecules, or two dimers, were present in the asymmetric unit of the crystal, one monomer had an N-glycan partly visible in electron density at Asn25 at the N-terminal end of the domain (Fig. 1c). Overall two N-glycosylation sites are predicted to be present in SALM3 LRR domain (Asn25 and Asn70), with most of the glycans disordered in the structure, however. The LRR domain shows a typical extracellular LRR domain structure, similar to the known SALM5 LRR structure (Supplementary Fig. S1) 20 , with an extend β-sheet on the concave side of the structure formed by the seven LRRs (L1-L7) (Fig. 1b) of variable length (22-24 residues) with consensus motif of "L 1 xxL 4 xL 6 xL 8 xxN 11 xL 13 " with aliphatic Leu or Ile residues mostly occupying positions 1, 4, 5, 6 and 13 and forming the hydrophobic core of each repeat, and the absolutely conserved Asn11 hydrogen bonding to the previous repeat. The LRR domain further contains an N-terminal LRRNT capping subdomain (residues 17-48) and an α-helical C-terminal LRRCT capping domain (residues 228-278) (Fig. 1b,c), which are conserved in the vertebrate extracellular LRR domain family 24 . Both the LRRNT and LRRCT subdomains of SALM3 are stabilized by two disulphide bridges; in the LRRNT between Cys17-Cys27 and Cys21-Cys34, and in the LRRCT between Cys238-Cys257 and Cys240-Cys278. Cys278 also is the last residue of the whole LRR domain. A flexible loop (residues 216-227) is present between the 7th LRR and the LRRCT, not visible in the SALM3 crystal structure (Fig. 1b, c). The possible functional significance of this loop remains unknown, as its not involved in LAR-RPTPs recognition at least in SALM5-PTPδ complex structures 17,18 , while such an extension might be expected to have some functional significance, as e.g. in the case of platelet glycoprotein IB alpha LRR domain 25 .
Dimerization through the LRR domain is a defining feature of the SALM family. The SALM3 LRR domain forms a dimer similar to those observed for SALM5 and SALM2 crystal structures 17,18,20 (Fig. 1). It appears that the SALMs form a highly conserved dimeric structure based on the known structures of SALM2, SALM3 and SALM5 and our data on SALM1 oligomerization 20 , and the overall sequence conservation of the interface residues in the protein family ( Supplementary Fig. S2, Table S1). The SALM3 LRR structure and dimer interface are close to identical to that of SALM5 (Fig. 2, Supplementary Fig. S1, Table S1) 20 . The hydrophobic interactions and hydrogen bonding networks are conserved at the interface with Gln131 and Arg107 forming www.nature.com/scientificreports/ hydrogen bonds across the dimer interface to Asn156 of the other monomer and vice versa (Fig. 2, Table S1). The SALM3 LRR domain, like SALM5 and SALM2 forms an antiparallel side-by-side dimer, with a rather small interface with buried surface area of 1,001 Å 2 . The interface consists of the "one-layer" central hydrogen bonding network (Fig. 2) described above, and hydrophobic interactions formed by the N-and C-terminal regions (Fig. 2). The dimer species is however very stable in solution for both SALM3 and SALM5, and also SALM1 based on our observations 20 . Monomeric species have not been observed in vitro. Here, we analysed the oligomeric state in solution for both the LRR domain and LRR-Ig construct for SALM3 by size exclusion chromatography-coupled multiangle static laser light scattering (SEC-MALLS) (Fig. 3a, b). Both appear as dimers based on observed molecular weights of ca. 60 and 80 kDa (Fig. 3a, b); twice the size of molecules weights calculated from sequence (30,9 and 40 kDa Fig. S1). Thus, we conclude that the dimerization mechanism through the LRR domains is conserved in the family, generating a unique dimeric post-synaptic LAR-RPTP ligand, with possible functional implications. Interestingly all the other characterized LAR-RPTP ligands (Slitrks, IL1RAPL1 and TrkC and NGL-3) 26-29 are monomeric. Dimerization of SALMs might be a way of regulation of the function and induction of presynaptic signaling by LAR-RPTPs, as in the case of SALM3 and SALM5 biological function depends on the oligomerization 17,18 (see below), and SALM4 regulates the function of SALM3 by cis-inhibition, most likely through heterodimer formation 16 . Function of SALM2 remains unclear, but it appears to be able to bind the LAR-RPTP receptors 18 , whereas the function of SALM1 appears quite different, with reported involvement in actin mediated post-synaptic neurexin clustering, and no synapse formation activity reported 30 .
The solution structure of SALM3-PTPσ complex. In order to understand the binding mode of SALM3 to its PTPσ ligand, we have characterized the solution structure of the SALM3 LRR-Ig -PTPσ Ig1-Ig3 complex using size exclusion chromatography coupled with SAXS (SEC-SAXS). The SEC-profile of SALM3-PTPσ complex shows a single major peak, and the frames (350-365) with similar Rg-values, as calculated from the X-ray scattering intensities were averaged for further analysis (Fig. 3c). The SALM3 LRR-Ig and PTPσ Ig1-Ig3 structures were modelled based on the known crystal structures, as described in the methods. Rigid body fitting of the complex and ab initio envelope modelling were done against the obtained solution scattering data. The complex structure was modelled by rigid body fitting, with the LRR dimer structure as one rigid body and each SALM5 Ig-domain with its respective bound PTPσ ligand Ig1-Ig3 structure as the two other rigid bodies (Figs. 1, 3), The starting model was based on the SALM5-PTPδ complex crystal structure (PDB 5XNP) 17 and the final rigid body refinement results against the SAXS data demonstrate that a 2:2 complex between SALM3 and PTPσ is formed similar to SALM5. Both 2:2 complex models with P1 and P2 symmetry fit the data well with X 2 -values of 1.7 and 2.5 respectively (Fig. 3d-f and Supplementary Fig. S3). This is further supported by the ab initio envelope calculations that fit well with the rigid body models (Fig. 3g, h). Both methods give a result that fit a 2:2 complex best. We further confirmed that 1:1 and 2:1 complexes do not fit to the data ( Supplementary Fig. S3).
It should be noted that while it is possible to obtain a good fit to the model with P2 symmetry constraint, we stress that due to the limited resolution of SAXS data, the model is likely representing an average snapshot of a number of conformations present in solution. This is further confirmed by the fact that the fit can be slightly improved if no symmetry constraints are used, indicating certain degree of flexibility in the modelled regions.

◂
We have also previously shown that the position of the Ig-domain is flexible relative to the LRR domain dimer for both SALM3 and SALM5 20 . Taken together, our results suggest 2:2 stoichiometry for the SALM3 LRR-Ig:PTPσ Ig1-Ig3 complex in solution, and we propose a schematic model for the orientation of the complex organization in the synaptic cleft (Fig. 3i). Thus, we here show that SALM3 forms a complex with the presynaptic PTPσ receptor with a similar mechanism to SALM5 binding to PTPδ, and that we observe that the structure exhibits certain amount of flexibility that may be requirement for the molecular environment in the synaptic cleft between the two cellular membranes. Based on the SAXS data, we further validated the interaction sites in the SALM3-PTPσ complex by site-directed mutagenesis and cell-based binding and heterologous synapse formation assays.
Mutational analysis of SALM3-PTPσ interface. The recent SALM5-PTPδ complex crystal structures at 3.7 Å and 4.2 Å resolution showed that the LRRCT-region and Ig-domain of SALM5 interact with the Ig2 and Ig3 domains of PTPδ, and the presence of meB sequence enhanced the interaction 17,18 . We also reported in our earlier study 20 using surface plasmon resonance (SPR) that the interaction of SALM3 and PTPσ is enhanced by ten-fold in presence of meB. The K d for SALM3 interaction with PTPσ-meB + by SPR was measured to be 2.2 µM 20 .
Multiple sequence alignment of SALM3 and SALM5 sequences ( Supplementary Fig. S2), and the amino acid conservation plotted on the SALM3-PTPσ complex model, and comparison of aligned SALM2, SALM3 and SALM5 complex structures (Fig. 4a-c, Supplementary Fig. S4) show that the PTPσ interacting residues of SALM3 are mostly conserved in SALM3 and SALM5 in the vertebrates ( Fig. 4a; Supplementary Fig. S2, Table S2). In addition, in the LRR domain dimer structure of SALM3, the concave surface shows large number of conserved residues between SALM3 and SALM5 based on sequence analysis (Fig. 4a) 20 . This might indicate a potential binding site, the ligand for which remains to be identified.
Based on the sequence analysis, and in particular the SAXS structure of SALM3-PTPσ complex, we suggest that the binding interface of SALM3-PTPσ is similar to SALM5-PTPδ, with the observed dependence on the presence of meB 14,20 , structural alignment of SALM3 model to SALM2, and SALM5 complexes indicate the interaction sites are also structurally highly conserved ( Supplementary Fig. S4). In order to verify this, we generated several deletion and point mutations to SALM3 extracellular domain at the SALM3-PTPσ ligand interface modelled based on the binding interface interactions of SALM5-PTPδ structure. The wild-type (WT) and mutant SALM3 proteins were expressed in HEK293T cells, and tested for interaction with PTPσ-Fc fusion protein using an in-cell western binding assay (Fig. 4d, e). Initially, we measured the apparent binding affinity of soluble dimeric PTPσ-Fc to SALM3 on the cell surface to be 22.32 ± 5 nM (Fig. 4d). It should be noted that this is an overestimate of the 1:1 affinity, due to the avidity effects arising from to the use of a dimeric Fc-fusion construct. The 1:1 binding affinity for the interaction is 2.2 µM, as noted above 20 . The negative controls showed no detectable signal (Fig. 4f), which confirmed the specificity of the assay. To assess the effect of the point mutations, we used 150 nM PTPσ-Fc.
Firstly, SALM3 ΔLRR, SALM3 ΔLRR + ΔFn and SALM3 ΔIg-Fn domain deletion mutants were expressed on the surface of HEK293T cells, but showed no interaction with PTPσ (Fig. 4e). The cell-based binding studies on SALM3 deletion mutants verified that both the LRR and Ig-domain of SALM3 are involved and necessary for the PTPσ interaction, and neither alone is enough for binding to the PTPσ. Point mutations on the SALM3 LRR and Ig-domains were analysed for PTPσ interaction to probe and verify more closely the importance of binding interfaces of SALM3-PTPσ complex. The surface expression and secretion was verified by immunofluorescence and western blot detection of the variants as described in the Methods and Supplementary Fig. S5 and S6.
The binding interface in SALM3-PTPσ complex is formed by three interaction sites (Fig. 4b, Supplementary  Table S2), similar to SALM5-PTPδ (PDB 5XNP, 5XWT) 17,18 . In site I the SALM3 LRRCT forms an interface with the Ig2-domain of PTPσ, in site II SALM5 Ig-domain interacts with Ig3-domain of PTPσ and in site III the SALM3 Ig-domain binds to Ig2-domain of PTPσ 17 . We generated mutations in SALM3 on these three sites (Fig. 4b, c) in order to validate the SALM3 complex structure.
In the case of SALM3-PTPσ complex, in site I, Arg247 (SALM5 Arg253) forms hydrogen bond with the PTPσ Gln202 and back bone carbonyl of Val225 (Fig. 4c) www.nature.com/scientificreports/ shows no interaction with PTPσ (Fig. 4e). This confirms its critical importance for the interaction for SALM3 with PTPσ. In an earlier study 18 , it was reported that SALM5 Arg253Ala decreased the binding affinity with PTPδ eight-fold compare to the wild-type (WT). The site consists of additional hydrophobic interactions by residues Pro212, Cys 246, Leu249 and Trp250 in SALM5 with PTPδ, corresponding to SALM3 Pro210, Cys240, Leu243 and Trp244 residues, in equivalent positions (Supplementary Table S2). Sequence analysis further confirms that the residues are highly conserved in SALM3 and SALM5 (Supplementary Fig. S2 and Table S2). Essentially all the PTPσ residues involved are conserved on all three interfaces ( Supplementary Fig. S7).
In site II, SALM3 Arg285 (SALM5 Arg291) forms an ion pair with PTPσ Asp275, and hydrogen bonds to the main chain carbonyl group on Thr277 of PTPσ (Fig. 4c). These interactions are conserved between the two complexes. However, not all interface residues in the SALM Ig-domains are conserved. SALM3 Leu305 (Arg311 in SALM5) binds to a hydrophobic pocket surrounded by Met282, Tyr266 and Glu279 side chains (Fig. 4c). This www.nature.com/scientificreports/ is different in SALM5 where the Arg311 hydrogen bonds to adjacent PTPδ Tyr257 and backbone carbonyl groups at Glu270-Asp271. These amino acid differences at the binding site might lead to slight changes in specificity or affinity in vivo.
In addition, at site II in the SALM3 complex Leu282 (SALM5 Leu288) forms hydrophobic interactions with PTPσ Met305 and the peptide main chain at the turn residues 306-307 of the PTPσ Ig3-domain (Fig. 4c). The cell-based binding measurements of SALM3 Arg285Ala-Leu282Arg and Arg285Ala-Leu305Ala double mutants with PTPσ showed no signal for interaction (Fig. 4e), and thus verify the importance of site II and Arg285, Leu282 and Leu305 for PTPσ binding. In previous studies SALM5 Leu288Ala mutation was reported to decrease affinity by 10-fold 18 , and SALM5 Lys309 and Arg311 when mutated to Asn and Thr to introduce an N-glycosylation site in place of Lys309 also abolished binding 17 .
In site III of the model, Glu359 (SALM5 Glu365) forms hydrogen bond with the PTPσ Thr153, and a salt bridge to PTPσ Lys144. SALM3 Leu321 and His313 (SALM5 Leu327 and His319) form hydrophobic interactions with Asp138, Met139 and Ala157 of PTPσ (Fig. 4c). SALM3 Thr354 is also part of the interface and packs against Leu155 of PTPσ. An earlier study 17 reported that based on Ala-scanning mutagenesis that SALM5 Ser360Ala-Se329Ala double mutant (Ser360Ala equal to SALM3 Thr354) significantly reduced the interaction with LAR 18 . Our cell-based binding study showed that SALM3 Leu321Ala-His313Ala double mutation abolishes the interaction with PTPσ, and SALM3 Thr354Arg-Glu359Ala double mutant decreases binding towards PTPσ by ~ 50% but did not abolish the interaction (Fig. 4e). Thus, in site III the hydrophobic interactions by Leu321 and His313 appear to be most critical for the affinity in SALM3 of the residues at this interface (Fig. 4c, e and Supplementary  Table S2). Overall, all three sites observed as binding interfaces for SALM5 appear to be functionally equivalent in SALM3, and our mutational data on the interfaces nicely complements the earlier limited mutational studies on SALM5 17,18 .
Finally, for SALM5, it was reported that the Asn158Ala and Gln134Asn SALM5 mutations (equivalent to SALM3 residues Asn156 and Gln131) that interfere with the LRR dimerization hardly affect the binding to PTPδ 18 . Our cell-based binding studies show that the SALM3 Arg107Ser-Gln131Ala double mutation at the dimer interface interferes directly with PTPσ binding (Fig. 4e), which is in contrast to SALM5 results. This suggests dimerization on cell surface could be direct pre-requirement for activity, at least in the case of SALM3, and potentially based on the highly similar interaction also in SALM5. We also verified that the purified Arg107Ser-Gln131Ala mutant protein actually is monomeric in solution, while two other double mutants Leu282Arg-Arg285Ala and Thr354Arg-Glu359Ala where dimeric and expressed and behaved as the wild type SALM3 (Supplementary Fig. S8).
The effect of SALM3 variants on presynaptic differentiation. Since SALM3 functions as a synapse organizer inducing presynaptic differentiation through binding to presynaptic LAR-RPTPs, we tested the effect of the SALM3 mutations on presynaptic differentiation activity using heterologous synapse-formation assays. SALM3 mutants expressed in HEK293T cells were cultured with hippocampal neurons (DIV9), and the presynaptic differentiation was detected by immunostaining against the presynaptic marker synapsin I (Fig. 5). The synaptogenic activity of SALM3 was clearly detected with the WT SALM3, but the interface mutants targeting the SALM3-PTPσ interaction abolished synapsin I clustering (Fig. 5). In addition, Thr354Arg-Glu359Ala double mutant at site III showed significantly decreased activity, despite exhibiting 50% residual binding at saturating concentration towards the PTPσ-Fc (Fig. 4e). Also the Leu282Arg mutation alone at site II was sufficient to compromise the SALM3 activity in inducing presynaptic differentiation (Fig. 5). In case of SALM5, it was previously reported 18 that SALM5 Arg253Ala (Site I of SALM5-PTPδ complex), SALM5 Leu288Ala (Site II) and SALM5 Ile358Ala (Site III) mutants severely reduced the synaptogenic activity. In addition, the Arg107Ser-Gln131Ala SALM3 mutant that interferes with the SALM3 LRR dimerization also abolished the synaptogenic activity (Fig. 5). These results suggest that SALM3 mediated dimerization is critical for interaction with PTPσ and the formation of SALM3-PTPσ trans-synaptic signaling complex to induce presynaptic differentiation. Overall, the binding assays results are well correlated with data obtained from heterologous synapse formation analyses. This further suggests that all interactions that are important for the complex formation with PTPσ are also required for the synaptogenic activity. Thus the results verify the biological significance of all the three interaction sites, and the fact that the described SALM3-PTPσ interactions are necessary and sufficient for presynaptic differentiation by SALM3. Recently Xie et al. 31 reported that LAR D1 phosphatase domains can dimerize to promote clustering and activation of synaptic differentiation. We suggest that the dimeric 2:2 complex formation by SALMs on binding to LAR-RPTPs in turn may promote the intracellular homophilic interactions of LAR-RPTPs and further activation of presynaptic differentiation.

conclusions
SALM3 was the first protein in the SALM-family to be characterized as a ligand for presynaptic LAR-RPTPs 14 .
We have now resolved its molecular mechanism of action and mode of oligomerization, which we conclude are similar with SALM5. Taken together, we conclude that dimerization might be a universal mechanism that is conserved across all SALM family proteins and that LAR-RPTP recognition occurs via the same binding interfaces. We hypothesize that SALM3 dimerization promotes presynaptic differentiation by bringing the LAR-RPTP receptors together and enhancing their clustering. Further studies are warranted to systematically test how SALM dimerization is involved in mediation of various synaptic functions and LAR-RPTP activation.

Materials and methods
Cloning, protein expression and purification. Mouse SALM3 LRR (residues 17-284) was cloned into Drosophila pRMHA3 expression vector 32 with CD33 signal sequence at the N-terminus and a C-terminal Fctag preceded with a Prescission protease cleavage site. The cDNA for SALM3 was obtained from ImaGenes GmbH. The cloning and protein purification of SALM3 LRR-Ig (residues 17-367) and mouse PTPσ Ig1-3-meB + (resdiues 33-331) have been described in earlier 20 . For cell-based binding assays, the mouse SALM3 LRR-Ig-Fn (residues 17-510) and the deletion mutants SALM3 LRR (residues 17-284) "ΔIg-Fn", SALM3 Ig (residues 280-367) "ΔLRR + Fn" and SALM3 Ig-Fn (residues 280-502) "ΔLRR" were cloned in the pDisplay plasmid vector. The constructs contained an N-terminal HA-tag followed by the cloned SALM3 sequence, and at the C-terminus Myc-tag followed by transmembrane domain. For mutational studies, we used site directed mutagenesis 31 to mutate the plasmid constructs of SALM3 LRR-Ig-Fn. The oligonucleotide pairs used for the cloning and mutagenesis are listed in Supplementary Table S4. For SALM3 LRR, the protein expression was verified by transient transfection and western blot detection with goat anti-human polyclonal horse radish peroxidase (HRP) conjugated antibody (Abcam ab98567). A stable Drosophila S2 cell line for SALM3 LRR expression was then generated for the protein purification, and protein was expressed in standard manner as described previously 20 . The protein with the C-terminal Fc-fusion was affinity purified with protein-A sepharose (Invitrogen). Samples were eluted with 0.1 M glycine pH 3.0 directly to a neutralizing buffer, with final composition of 60 mM Tris pH 7.4, 300 mM NaCl. The tagged proteins were incubated with Precission protease for 16 h at 4 °C to remove the Fc tag. Precission protease was produced as a GST fusion in Escherichia coli BL21 (DE3) from pGEX-6P-1 vector (Addgene). The cleaved Fc-fusion protein were passed through a protein-A column, and flow-through containing the cleaved SALM3-LRR was collected and gel filtered with Superdex 75 10/300 (GE Healthcare) in 60 mM Tris pH 7.5, 300 mM NaCl and concentrated and stored at − 80 °C for further use. Crystallization and structure determination and refinement. The SALM3 LRR construct was concentrated to 7.2 mg/ml and exchanged to 20 mM Tris pH 7.4, 100 mM NaCl for crystallization. Thin plate-like crystals appeared from 0.17 M ammonium sulfate pH 7.5, 25% w/v PEG 4,000, 10% glycerol at + 20 °C. Crystals were harvested directly from the drop and flash frozen for data collection. The crystals diffracted to 2.8 Å and crystallized in space group P2 1 , and were twinned (pseudo-orthorhombic) due to the β-angle being almost equal to 90° (Table 1) with high twin-fraction of ca. 34% with the twinning operator h, − k, − l and with two dimers in the asymmetric unit ( Table 1). The twinning affected the quality of maps and model unavoidably to some extend and complicated the refinement process, however final electron density was of satisfactory quality (Supplementary Fig. S9).
The SALM3 crystal structure was solved by molecular replacement with the CCP4 package and PHASER 33 . An initial solution for the SALM3 LRR was found using the LRR domain of SALM5 (PDB 6F20) 20 as a template. After molecular replacement and initial refinement with REFMAC 34 , the R-factors for the initial LRR domain model were R work /R free = 41.2%/46.9%. The model was further refined with PHENIX 35 and BUSTER 36 and rebuild with COOT 37 . The final model refined with BUSTER had R-factors of 26.91/28.50% (Table 1).
Small-angle X-ray scattering data collection and analysis. SAXS data on the SALM3 LRR-Ig and PTPσ complex was measured by the SEC-SAXS method at Diamond Lightsource on the B21 beamline. For SALM3 LRR-Ig and PTPσ complex preparation, SALM3 LRR-Ig and PTPσ were mixed in the molar ratio of 1:1.2 of SALM3 LRR-Ig (60 µM) and PTPσ (80 µM) in total volume of 50 µl, followed by incubation at 4 °C for 1 h. Sample compartment and exposure cell were cooled to 4 °C. Superdex 200 PC 3.2/30 (GE Healthcare) column was used with flow rate of 0.5 ml/min in 30 mM Tris HCl pH7.5, 150 mM NaCl. Data processing was performed using ScÅtter 38 and DATASW 39 . Particle shapes at low resolution were reconstructed ab initio by DAMMIN 40 in P1 with P(r) function generated from the data truncated to 0.25 Å −1 . A total of 10 independent reconstructions were performed, and the models were averaged with the program DAMAVER 41 . Rigid body modeling was performed using CORAL 42 , where available partial crystal structures were treated as rigid bodies while missing residues were modelled ab initio as a Cα-chain by the software. The final model was converted to a full all-atom model using FULCHER tool (Shkumatov A. et al., unpublished). The modeling was done with either P1 or P2 symmetry. Porod and dimensionless Kratky plots were calculated for estimation of D max and the overall compactness of the models (Supplementary Fig. S3).
Homology modeling and conservation analysis. Homology models of SALM3 LRR-Ig and PTPσ Ig1-Ig3 structures for SAXS studies and interface analysis were generated based on the known SALM5 (PDB 6F2O) and PTPδ crystal structures (PDB 5XNP) using SWISS-MODEL (https ://swiss model .expas y.org/) 43 . Sequence conservation was analysed and plotted on the SALM3 structure using CONSURF 44  cell-based binding assays. We employed the in-cell western blot method 47 to detect the SALM3 and PTPσ interaction on cell surface with the Odyssey Infrared Imaging System (LI-COR Biosciences) following manufacturer's recommendations. In this assay, SALM3 constructs were expressed in HEK293T cells and PTPσ-Fc was added on the cells in a 96-well plate and the binding was detected with anti-human Fc antibody and signal per well was read with the Odyssey Imaging system. For transfection, HEK293T cells were grown to 90% confluency on a T75 flask using DMEM media with 10% FBS (Thermo Fisher Scientific) at 37 °C and 5% CO 2 . Cells were seeded into a poly-L lysine-coated 96 well plate at approximately 20,000 cells/well and allowed to grow 24 h at 37 °C until the confluency was 90%. After 24 h, the cells were transfected with SALM3-pDisplay plasmids (400 µg/well) and polyethylenimine (1.2 µg/well). The transfected cells were incubated at 37 °C for 48 h. The SALM3 fusion protein expression in transfected cell was detected with Floid™ Cell Imaging system (Thermo Fisher Scientific) using immunofluorescence.
For immunofluorescence detection, transfected cells were fixed under non-permeabilizing conditions with 4% paraformaldehyde (PFA) for 20 min and blocked at room temperature with 1xPBS buffer containing 1% BSA for 60 min, followed by addition of 1:300 dilution primary antibody (rabbit anti-HA, Santa Cruz) and incubation for 30 min with show shaking at room temperature. The cells were stained with 1:500 dilution secondary antibody (Alexa 488 anti-rabbit, Molecular probes), and the expression was detected with Floid™ Cell Imaging system to verify the surface expression of proteins ( Supplementary Fig. S5).
For in-cell western blot experiment, SALM3 transfected HEK293T cells were blocked with EGB buffer 48 (10 mM HEPES pH 7.2, 168 mM NaCl, 2.6 mM KCl, 2 mM CaCl 2 , 2 mM MgCl 2 , 10 mM D-glucose, and 5% BSA) for 2 h at room temperature with slow shaking. Blocking buffer was removed and cells were covered in 50 µl of PTPσ-Fc diluted in EGB buffer with 1% BSA. The cells were incubated overnight at 4 °C with gentle shaking. Cells were further treated with 1:200 dilution of secondary antibody (goat anti-human IgG IRDye 800CW), and the SALM3 and PTPσ binding intensity was detected using LI-COR Odyssey (169 µm resolution, medium quality) with channel (800 nm) intensity of 8. For the binding assay a concentration series of PTPσ-Fc (1-250 nM) was used, and 150 nM PTPσ-Fc (close to saturation) was used for mutational studies.
Heterologous synapse-formation assays. HEK293T cells were transfected with the indicated mutants and the WT SALM3 cloned in pDisplay vector (Thermo Fisher Scientific; listed in Table S4). After 48 h, transfected HEK293T cells were trypsinized, seeded onto cultured hippocampal neurons at DIV9, and then coimmunostained with antibodies against synapsin 49 at DIV11. Images were acquired by confocal microscopy (LSM700, Carl Zeiss). For quantifications, the contours of transfected HEK293T cells were chosen as the ROI. The fluorescence intensities of synaptic marker puncta, normalized with respect to the area of each HEK293T cell were quantified for both red and green channels using MetaMorph Software (Molecular Devices). Surface expression was verified as explained in the Supplementary Fig. S10. All data are expressed as means ± SEM. All experiments were repeated using at least three independent cultures and data were evaluated statistically using a nonparametric Kruskal-Wallis test, followed by Dunn's multiple-comparison test for post hoc groups comparisons (n values used are indicated in the figure legends). Significance was indicated with an asterisk (compared with a value from control group) or hashtag (compared with a value from SALM3 WT-group).