Calsyntenin-3 Molecular Architecture and Interaction with Neurexin 1α*

Background: Calsyntenin-3 (Cstn3) promotes synapse development, controversially interacting with neurexin 1α (n1α). Results: Cstn3 binds n1α directly, and its structure adopts multiple forms. Conclusion: Cstn3 interacts with n1α via a novel mechanism and can produce distinct trans-synaptic bridges with n1α. Significance: A complex portfolio of molecular interactions between proteins implicated in autism spectrum disorder and schizophrenia guide synapse development. Calsyntenin 3 (Cstn3 or Clstn3), a recently identified synaptic organizer, promotes the development of synapses. Cstn3 localizes to the postsynaptic membrane and triggers presynaptic differentiation. Calsyntenin members play an evolutionarily conserved role in memory and learning. Cstn3 was recently shown in cell-based assays to interact with neurexin 1α (n1α), a synaptic organizer that is implicated in neuropsychiatric disease. Interaction would permit Cstn3 and n1α to form a trans-synaptic complex and promote synaptic differentiation. However, it is contentious whether Cstn3 binds n1α directly. To understand the structure and function of Cstn3, we determined its architecture by electron microscopy and delineated the interaction between Cstn3 and n1α biochemically and biophysically. We show that Cstn3 ectodomains form monomers as well as tetramers that are stabilized by disulfide bonds and Ca2+, and both are probably flexible in solution. We show further that the extracellular domains of Cstn3 and n1α interact directly and that both Cstn3 monomers and tetramers bind n1α with nanomolar affinity. The interaction is promoted by Ca2+ and requires minimally the LNS domain of Cstn3. Furthermore, Cstn3 uses a fundamentally different mechanism to bind n1α compared with other neurexin partners, such as the synaptic organizer neuroligin 2, because Cstn3 does not strictly require the sixth LNS domain of n1α. Our structural data suggest how Cstn3 as a synaptic organizer on the postsynaptic membrane, particularly in tetrameric form, may assemble radially symmetric trans-synaptic bridges with the presynaptic synaptic organizer n1α to recruit and spatially organize proteins into networks essential for synaptic function.

A growing number of synaptic organizers form heterophilic trans-synaptic molecular bridges to mediate synaptic differentiation (the assembly and maintenance of presynaptic and postsynaptic machineries capable of synaptic transmission) (1,2). Many of these synaptic organizers are now implicated in neuropsychiatric disorders, underscoring their importance in regulating functional neuronal circuits (2). Calsyntenin-3 (Cstn3 4 or Clstn3) was recently identified as a synapse-organizing protein (3), and its role was confirmed by multiple groups (4). Cstn3 localizes in part to the postsynaptic membrane (5). On the cell surface, Cstn3 triggers inhibitory and excitatory presynaptic differentiation in contacting axons (3). Cstn3 knock-out mice display both decreased inhibitory and excitatory synaptic densities and deficits in synaptic transmission, suggesting that Cstn3 is needed for the development of both GABAergic and glutamatergic synapses (3). Members of the calsyntenin family play a role in learning and memory (6 -8). Calsyntenins have also been linked to Alzheimer disease and appear to shield amyloid-␤ precursor protein from the proteolytic production of amyloidogenic A␤ peptide (9 -11). The extracellular domain of calsyntenins is composed of two cadherin domains, an LNS (laminin, neurexin, sex hormone-binding globulin) domain and an ␣-helix/␤-strand-containing domain (␣/␤ domain). Calsyntenins are proteolytically cleaved (9), and their shed ectodo-main can oppose the function of the full-length calsyntenin (3). The Cstn3 ectodomain was shown to bind the extracellular domain of neurexin 1␣ (n1␣), a synaptic organizer found on the presynaptic membrane, in synaptosomal pull-down assays and cell-based binding and recruitment assays (3) as well as via independent proteomic approaches in our laboratory. However, Um et al. (4) were not able to reproduce the interaction between Cstn3 and n1␣ in similar cell surface binding assays, raising a question of whether Cstn3 binds n1␣ directly.
To gain insight into the structure and function of Cstn3, we used biochemical, biophysical, and structural methods to interrogate Cstn3 and its interaction with n1␣. We show that the extracellular domain of Cstn3 forms both functional monomers and tetramers, and we reveal their architectures and demonstrate direct interaction between Cstn3 and n1␣. Our results suggest that Cstn3 works in concert with n1␣ by forming a trans-synaptic bridge to promote synapse development.

Protein Expression and Purification
The human Csnt3 ectodomain or fragments (accession number BC104767), followed by a C-terminal tag ASTSHHHHHH, was produced using baculovirus-mediated overexpression in HighFive cells with Insect-XPRESS ϩ L-glutamine medium (Lonza). Briefly, medium containing the secreted proteins was concentrated after protease inhibitors were added, dialyzed overnight (25 mM sodium phosphate, pH 8.0, 250 mM NaCl), and purified with a nickel-nitrilotriacetic acid column (Invitrogen; 25 mM sodium phosphate, pH 8, 500 mM NaCl, eluted with an imidazole gradient). Subsequently, the protein was dialyzed into 25 mM Tris, pH 8, 100 mM NaCl, 0.5 mM PMSF; incubated with 5 mM CaCl 2 for 0.5 h; applied to a MonoQ column (GE Healthcare) equilibrated with 25 mM Tris, pH 8, 50 mM NaCl; and subsequently eluted with an NaCl gradient. Cstn3-LMW and Cstn3-HMW eluted separately on the MonoQ column. Last, proteins were applied to a HiLoad Superdex-200 16/60 size exclusion column (GE Healthcare; 25 mM Tris, pH 8, 100 mM NaCl). Purified proteins were stored in 25 mM Tris, pH 8, 100 mM NaCl in flash-frozen aliquots. Purified hexahistidinetagged bovine neurexin 1␣ and its fragments as well as hexahistidine-tagged rat neuroligin 2 (NL2) containing insert ϩA2 were produced in a manner similar to that described previously (16). Recombinant rat neurexin 1␤ was produced as described (26). For analytical size exclusion chromatography, proteins were loaded on a Superdex 200 PC 3.2/30 column in 25 mM Tris, pH 8.0, 120 mM NaCl, 5 mM CaCl 2 and run at 0.08 ml/min. Protein standards (Sigma; 200, 66, 29, and 12.4 kDa) and blue dextran (2000 kDa) were used to calibrate the column loaded in a 50-l sample volume and run at 0.08 ml/min.
EM Data Acquisition and Image Preprocessing-The negative stain micrographs were acquired at room temperature on a Gatan UltraScan 4Kϫ4K CCD by a Zeiss Libra 120 transmission EM (Carl Zeiss NTS) operating at 120 kV at ϫ125,000 magnification (0.94 Å/pixel) under low defocus (Ϫ0.1 to ϳϪ0.6 m) by following a strategy developed for imaging small and asymmetric proteins (28). The micrograph defocus and astigmatism were examined by ctffind3 (FREALIGN software package) (29). Micrographs with distinguishable drift effects were excluded, and the rest were corrected for the contrast transfer function with the SPIDER software (30). Only isolated particles were initially selected and windowed by the EMAN software (31) and then manually adjusted. A total of 11,186 Cstn3-LMW particles and 6699 Cstn3-HMW particles were windowed and selected from 1063 micrographs. These particles were then aligned and classified by reference-free class averaging using refine2d.py (EMAN).
Electron Tomography (ET) Data Acquisition and Image Preprocessing-ET data of Cstn3-LMW and Cstn3-HMW were acquired with the equipment and conditions described above. The specimens on a high tilt room temperature holder were tilted at angles ranging from Ϫ66 to 66°in steps of 1.5°. The tilt series were acquired under a total electron dose of ϳ200 e Ϫ /Å 2 by the Gatan tomography software. The tilt series were initially aligned together with the IMOD software package (32), and the contrast transfer function was determined by ctffind3 (FREALIGN software) and corrected by TOMOCTF (33). The tilt series for a singled out ("targeted") Cstn3 particle was tracked and selected at 150 ϫ 150-pixel size.
Individual Particle Electron Tomography (IPET) Three-dimensional Reconstruction-Ab initio three-dimensional reconstructions were conducted using the IPET reconstruction method (34). In brief, the tilt series of a targeted particle was directly back-projected into a three-dimensional map to generate an "initial model." The projections of the initial model were then used as the references for tilting image alignment. During this process, a set of automatically generated Gaussian low pass filters and automatically generated masks were sequentially applied to both the references and tilt images. The three-dimensional map from the previous iteration was used as the new initial model for the next iteration until the changes in translational parameters were less than 1 pixel in total.
IPET Fourier Shell Correlation (FSC) Analysis-The resolutions of the IPET three-dimensional reconstructions were determined by FSC analysis by splitting the center refined raw ET images into two groups (odd-or even-numbered indices according to the order of tilting angles). Each group was used independently to generate a three-dimensional reconstruction by IPET; the two IPET three-dimensional reconstructions were then used to compute the FSC curve over their corresponding spatial frequency shells in Fourier space (using the "RF 3" command in SPIDER) (30). The frequency at which the FSC curve fell to a value of 0.5 was used to assess the resolution of the final IPET three-dimensional density map.
Single-particle Three-dimensional Reconstruction-Two IPET three-dimensional density maps of Cstn3-LMW (monomer) and two IPET maps of Cstn3-HMW (tetramer) were low passfiltered to 26 and 30 Å, respectively. The maps were then used as ab initio initial models for their corresponding single-particle multireference refinement (multirefine in EMAN) (31). The final maps refined from the two Cstn3-LMW (monomer) particles showed a resolution of 15.0 and 15.9 Å, respectively (based on the 0.5 FSC criterion (31)), whereas the final maps refined from the two Cstn3-HMW (tetramer) particles showed a resolution of 16.0 and 16.5 Å, respectively. For the Cstn3-HMW (tetramer) refinement, C 4 symmetry was enforced. All maps were then low pass-filtered to 16 Å for structural manipulation. Domains were roughly assigned and colored using Color Zone in Chimera (35) by fitting homology models of the Cad1-Cad2 tandem and the LNS domain, whereas the remaining molecular volume was assigned to the ␣/␤ domain. As a basis for Cstn3 Cad1-Cad2, Cad2-Cad3 (also known as EC2-EC3) from mouse cadherin-8 (Protein Data Bank entry 2A62 (36)) was used and has 25.5% sequence identity to the human Cstn3 counterpart. For the Cstn3 LNS domain, the LNS domain L2 from the rat n1␣ (Protein Data Bank 2H0B (20)) was used and has 23.4% sequence identity to the human Cstn3 counterpart.
To compare the two final Cstn3-HMW tetramer maps, they were aligned by proc3d and align3d (EMAN) before FSC computation. The subunits within each tetramer density map were extracted using the Volume Eraser option and then aligned to the Cstn3 monomer (monomer reconstruction 2) by optimizing the maximal cross-correlation of density maps with a contour level of ϳ25 kDa using Chimera. The aligned density maps were further aligned prior to FSC calculation. The rotational autocorrelation was computed using the measure rotation and measure corr commands in Chimera.
Statistical Analyses of the Particle Size-A total of 1776 Cstn3-LMW particles, 1033 Cstn3-HMW particles in the presence of 3 mM CaCl 2 , and 1017 Cstn3-HMW particles without additional Ca 2ϩ were selected from a total of 722 micrographs. Only "top view" Cstn3-HMW particles were used to measure the diameter in two orthogonal directions. The longest dimension of the Cstn3 monomer particle was used to represent its size. The geometric mean (the square root of the product) of two perpendicular diameters was used to represent the diameter of the Cstn3 tetramer. Histograms for the dimensions of the particles were generated and fitted with a Gaussian function in Origin version 7.5 (sampling step of 5.0 Å). Four density maps have been deposited to the Electron Microscopy Data Bank (EMD-6009, EMD-6010, EMD-6011, and EMD-6012).

Solid Phase Binding Assays
Proteins were biotinylated at room temperature by dialyzing them into PBS (100 mM sodium phosphate, pH 7.2, 150 mM NaCl), subsequently incubating them with a 5-fold molar excess of EZ-Link NHS PEG4-Biotin (Pierce) for 30 min and then dialyzing them into 25 mM Tris, pH 8, 100 mM NaCl. The labeling efficiency was typically 4 -8 biotins/molecule as determined by the Pierce biotin quantitation kit. Solid phase binding assays were carried out at room temperature. For assays with immobilized neurexins, 200 ngr neurexin in binding buffer/ Ca 2ϩ (20 mM Tris, pH 8.0, 100 mM NaCl, 5 mM CaCl 2 ) was coated in 96-well plates (Corning Costar 9017) for 2 h at 150 rpm. As a background control, a series of wells was incubated with buffer but no neurexin. The wells were subsequently emptied; washed three times with 300 l of binding buffer/Ca 2ϩ for 30 s at 400 rpm or, for Ca 2ϩ -free conditions, with binding buffer/ EDTA (20 mM Tris, pH 8.0, 100 mM NaCl, 20 mM EDTA); and finally blocked with blocking buffer (2% (w/v) gelatin (Sigma G7663) in binding buffer/Ca 2ϩ or binding buffer/ EDTA) for 2 h. Wells were then incubated with increasing concentrations of biotinylated Cstn3-LMW*, Cstn3-HMW*, or NL2* in binding buffer/Ca 2ϩ with 0.25% BSA (in triplicate) or binding buffer/EDTA (20 mM Tris, pH 8.0, 100 mM NaCl, 20 mM EDTA) with 0.25% BSA (in duplicate) for 1 h, emptied, and washed again three times with binding buffer/Ca 2ϩ or binding buffer/EDTA. For assays with immobilized Cstn3, 200 ngr Cstn3 (or fragments) in binding buffer/Ca 2ϩ was coated for 1 h at 150 rpm, and the wells were treated as described above except that they were blocked with blocking buffer containing 3% gelatin. Wells were then incubated with increasing concentrations of biotinylated neurexin 1␣* in binding buffer/Ca 2ϩ (in triplicate) for 1 h. To develop the signal, wells were incubated with anti-streptavidin HRP conjugate (Sigma S2438; diluted 1:5000 in blocking buffer) for 45 min, washed three times with binding buffer/Ca 2ϩ or binding buffer/EDTA, and then incubated with the substrate o-phenylenediamine (Calbiochem) for 10 min. The reaction was stopped by adding 50 l/well 0.5 M H 2 SO 4 , and the absorbance was read at 490 nm. Data were analyzed with Prism (GraphPad). Specific binding was expressed as total binding in the presence of Ca 2ϩ minus bind-

Calsyntenin-3 Architecture and Interaction with Neurexin 1␣
ing in the absence of immobilized bait. The K D was calculated by non-linear regression using a model for "one-site specific binding" and further visualized with Scatchard plots. Error bars show the S.E.

Co-immunoprecipitation and Cell Surface Binding Assays
HEK cells were transfected with the indicated expression vectors (3) using TransIT-LT1 transfection reagent (Mirus) and grown for 48 h. Cell lysates were extracted with 1% Triton X-100 in TBS with Complete protease inhibitor (Roche Applied Science), incubated with 1 g of anti-HA antibody (Roche Applied Science) overnight at 4°C and then with Protein G beads for 1 h 4°C. The beads were then washed with 0.1% Triton X-100 three times and eluted with SDS-sample buffer. Samples were analyzed by Western blot with anti-GFP antibody (Invitrogen). For the cell-based binding assay, COS7 cells were transfected with the indicated expression vectors (37) using TransIT-LT1 transfection reagent (Mirus) and grown for 24 h. Cstn3-Fc protein was generated as described previously (3).

Cstn3
Monomers and Tetramers-To study the structure of Cstn3, we expressed the Cstn3 ectodomain in insect cells and found that it formed two distinct molecular species, a low molecular weight (LMW) and a high molecule weight (HMW) multimeric form (Fig. 1, a-c). Under reducing conditions, both Cstn3-LMW and Cstn3-HMW migrated as ϳ100 kDa bands by SDS-PAGE, consistent with a monomer (Fig. 1d). However, under non-reducing conditions, Cstn3-HMW migrated as multimers held together with intersubunit disulfide bonds unlike Cstn3-LMW, which was clearly still monomeric (Fig.  1d). To test whether Cstn3 multimerizes with itself or other calsyntenin family members in a cellular environment, we coexpressed HA-tagged Cstn3 with CFP-tagged Cstn1, Cstn2, Cstn3, or the unrelated N-cadherin. We then pulled down HA-Cstn3 and probed if a CFP-tagged partner co-immunoprecipitated. HA-Cstn3 co-immunoprecipitated Cstn3-CFP, suggesting that Cstn3 forms multimers with itself, a specific association, as indicated by lack of co-precipitation of Cstn1-CFP or N-cadherin-CFP (Fig. 1e).
In EM micrographs, Cstn3-LMW corresponded to monomers that are flexible in structure (Fig. 2) and span a diameter of 85-100 Å (Fig. 3). Single-particle three-dimensional reconstruction was performed by multireference refinement methods ( Fig. 2 and supplemental Fig. S1). Two separate single-particle reconstructions were carried out, each using a different ab initio density map obtained from IPET (34) as an initial model, and revealed conformational variability (monomer reconstruction 1 in Fig. 2, d and e, and monomer reconstruction 2 in Fig. 2,  f and g). To relate the size of the individual domains to the complete Cstn3 ectodomain, homology models of the Cstn3 Cad1-Cad2 tandem and the LNS domain were docked in the molecular envelopes. The remaining density was assigned to the ␣/␤ domain, which lacks a clear structural homologue. Our reconstructions suggest that the Cstn3 monomer can adopt more opened or more closed conformations (Fig. 2, compare e and g).
In EM micrographs, Cstn3-HMW corresponded to multimers ( Fig. 4 and supplemental Fig. S2). The multimers contained an internal 4-fold axis, suggestive of a tetramer, based on self-rotation and cross-correlation analysis of electron tomographic density maps from two independent particles (supplemental Fig. S2). The four monomers in the Cstn3 tetramer are related by a 90°rotation around a common 4-fold axis as  ). b, 12 representative raw images of Cstn3 monomers. c, four representative reference-free class averages. d, single-particle three-dimensional reconstruction using the first of two ab initio density maps obtained from IPET as an initial model (monomer reconstruction 1). Three orthogonal views are displayed using two isosurface contour levels (corresponding to volumes of ϳ50 and 25 kDa). e, high contour isosurface shown as in d was colored according to the molecular volumes of the four domains comprising the Cstn3 extracellular region (two cadherin domains in orange and yellow, LNS domain in green, and the ␣/␤ domain in blue). f, single-particle three-dimensional reconstruction using the second of two ab initio density maps obtained from IPET as an initial model (monomer reconstruction 2). Three orthogonal views are displayed using the same isosurface contour levels as above. g, same high contour isosurface shown as in f was  The histogram is displayed in cyan bars, and the distribution of the particle diameter is fitted by a Gaussian curve. The largest population (12.6%) has a diameter of 95.4 Ϯ 2.5 Å.

Calsyntenin-3 Architecture and Interaction with Neurexin 1␣
opposed to a dimer of dimers. As in the case of the Cstn3 monomer, we performed single-particle three-dimensional reconstruction of the Cstn3 tetramer with refinement methods (31) using two different ab initio IPET reconstructions as initial models and applying C 4 symmetry ( Fig. 4 and supplemental Fig. S2).
The Cstn3 tetramer resembles an unopened flower. Each monomer forms one of four petals. The N-terminal cadherin domains each form the tip of a petal, and the C-terminal portions form the base of the flower. The molecular symmetry of the Cstn3 tetramer is compatible with ectodomains tethered to the postsynaptic membrane by single C-terminal trans-membrane segments (5). Whereas the monomers flex somewhat in the tetramer (mimicking the slight opening of a flower), overall the conformation is similar between tetramer particles (Fig. 5), although only to a resolution of ϳ28 Å (i.e. significantly lower than the ϳ16 Å resolution of the monomer and tetramer reconstructions estimated with the FSC ϭ 0.5 criteria) (supplemental Figs. S1h and S2j). The addition of Ca 2ϩ significantly increased  ). b, 12 representative raw images of Cstn3 tetramers. c, four representative reference-free class averages. d, single-particle three-dimensional reconstruction using an ab initio density map obtained from IPET as an initial model and refined with C 4 symmetry. Three orthogonal views are displayed using isosurface contour levels (corresponding to volumes of ϳ100 and 200 kDa). e, high contour isosurface shown as in d but color-coded according to domains as described in the legend for   a and b, three orthogonal views of the two different Cstn3 tetramer three-dimensional single-particle reconstructions (tetramer reconstruction 1 and tetramer reconstruction 2), colored according to their domains (see the legend for Fig. 2). The density maps displayed as two isosurface contour levels were aligned with each other to facilitate structural comparison through FSC computation. c, the FSC curves of these two tetramers cross the 50% threshold at 28.0 Å. d and e, a monomer subunit extracted from each of the Cstn3 tetramer density maps. f, for comparison purposes, the Cstn3 monomer from Cstn3-LMW particles in the same view as d and e. g, FSC analyses on each subunit of the Cstn3 tetramers compared with the free Cstn3 monomer show that the FSC curves cross the 50% threshold at 22.2 and 28.5 Å, respectively. affinity (K D ϭ 34 Ϯ 3 nM), and binding was promoted by Ca 2ϩ (Fig. 7, a and b). Cstn3 tetramers interacted with n1␣ similarly (K D ϭ 38 Ϯ 6 nM) (Fig. 7, c and d). For comparison, the ectodomain of the synaptic organizer NL2, another well validated Ca 2ϩ -dependent binding partner, also bound n1␣ with high affinity (K D ϭ 3 Ϯ 1 nM) (Fig. 7, e and f). To determine the minimal fragment of Cstn3 necessary to bind n1␣, we tested the binding of n1␣ to immobilized Cstn3 and a series of Cstn3 fragments (Fig. 8, a and b). In this flipped assay, immobilized Cstn3 interacted with soluble n1␣ with even higher affinity (K D ϭ 5 Ϯ 1 nM) (Fig. 8c). A Cstn3 fragment containing just the LNS and ␣/␤ domain bound n1␣ as well as full-length Cstn3, suggesting that the cadherin tandem is not required for high affinity binding (Fig. 8d). Indeed, the Cstn3 LNS domain alone was sufficient to bind n1␣ similarly to the full-length Cstn3 ectodomain (Fig. 8e), whereas the isolated ␣/␤ domain was not (Fig. 8b).
Cstn3 Uses a Distinct Mechanism to Interact with n1␣-Because well known partners of n1␣, such as neuroligins, LRRTMs, and latrophilin, strictly require the presence of the neurexin L6 domain for binding (15,22,38,39), we tested this dependence for Cstn3 as well. Strikingly, the extracellular domain of n1␣ lacking EGF-C and L6 (L1L5) bound Cstn3 as well as the full-length n1␣ ectodomain containing all nine domains (Fig. 9, a and b). In contrast, interaction between NL2 and n1␣ was completely abolished when these two domains were absent (Fig. 9c). This suggests that Cstn3 and NL2 use different mechanisms to interact with n1␣.
We confirmed the interaction between Cstn3 and n1␣ by surface plasmon resonance (Fig. 9, d-f). Full-length n1␣ (L1L6) bound Cstn3 similarly as the truncated ectodomain (L1L5) (i.e. K D ϳ33 Ϯ 8 nM and K D ϳ45 Ϯ 5 nM, respectively). The n1␣ fragment L5L6 also bound Cstn3 although more weakly (K D ϳ105 Ϯ 10 nM) (Fig. 9, d and f), whereas the presence of splice insert SS#4 did not significantly affect the interaction (L5L6 SS#4) K D ϳ55 Ϯ 10 nM (data not shown). In contrast, L1L6 and L5L6 bound NL2, but L1L5 (lacking the essential L6 domain) did not (Fig. 9, e and f). Furthermore, although L5L6 binds Cstn3 and NL2 with similar affinity (K D ϳ105 Ϯ 10 nM versus 83 Ϯ 2 nM), different molecular mechanisms underlie these two interactions, because L5L6 associates with and dissociates from Cstn3 more slowly compared with its interaction with NL2, which quickly reaches equilibrium but also quickly falls apart (k a ϳ17,370 Ϯ 54 M Ϫ1 s Ϫ1 and k d ϳ0.001823 Ϯ 0.000006 s Ϫ1 for Cstn3 versus k a ϳ67,030 Ϯ 1000 M Ϫ1 s Ϫ1 and k d ϳ0.00556 Ϯ 0.000057 s Ϫ1 for NL2). The single neurexin L2 domain did not bind Cstn3 appreciably, indicating that very specific regions of the n1␣ ectodomain interact with Cstn3, whereas surprisingly, n1␤ also bound immobilized Cstn3, although the data did not fit a simple 1:1 stoichiometric model, clearly indicating a more complex binding mode and complicating quantitative assessment (data not shown). Thus, based on our solid phase binding and surface plasmon resonance assays, we conclude that the ectodomains of Cstn3 and n1␣ interact with each other directly. We confirmed the domains mediating interaction between Cstn3 and n1␣ in cell-based binding assays (Fig. 10). Using a panel of n1␣ mutants lacking various domains expressed on the cell surface, we monitored the ability of these variants to bind soluble Cstn3-Fc. Consistent with the biochemical results, Cstn3-Fc bound cells expressing n1␣ L1L6, n1␣ L3L6, and n1␣ L5L6 similarly. Furthermore, the neurexin 1␣ mutation D1176A, which abolishes the Ca 2ϩ -binding site in the L6 domain and prevents cell surface-expressed neurexin 1␣ from recruiting NL2 on the surface of dendrites in neuronal co-culture assays (37), had no effect on the interaction between Cstn3-Fc and n1␣, substantiating that Cstn3 utilizes a different mechanism to bind neurexin 1␣ than NL2. In cell-based assays, we do not detect interaction between Cstn3 and n1␤, although their soluble ectodomains interact in biochemical assays. Possibly, differences in glycosylation (by expressing proteins in COS cells, insect cells, and Escherichia coli) and/or the presentation of the proteins on the cell surface compared with our biochemical assays impacted binding.

DISCUSSION
We present here the first molecular insight into Cstn3. We used a unique combination of electron microscopy and tomography (34) to obtain unprecedented structural information on a multidomain and apparently flexible protein, intractable by other methods. The structures of Cstn3 were studied by optimized negative stain EM, a validated method that has proven successful for relatively small proteins (28), rather than conventional negative stain EM or cryo-EM. As with any negative stain EM technique, we cannot exclude potential artifacts as a result of its chemical characteristics, which might impact the resolution limit, the shape, and/or the conformation of the observed macromolecules. Therefore, structural interpretations must be made with due caution. Nevertheless, our reconstructions provide important biological insight in conjunction with our biochemical, biophysical, and cell-based assays. We show that the Cstn3 ectodomain forms monomers and tetramers that are stabilized by disulfide bonds and Ca 2ϩ ions (Figs. 1, 2, 4, and 6) and that Cstn3 also forms multimers with itself in cell-based assays (Fig. 1e). The symmetry of the Cstn3 tetramer is compatible with four subunits tethered to a synaptic membrane. Importantly, the molecular 4-fold symmetry places putative protein binding sites similarly with respect to the synaptic membranes, facilitating the recruitment of partners from the presynaptic (trans-interaction) or the postsynaptic (cis-interaction) side.

Calsyntenin-3 Architecture and Interaction with Neurexin 1␣
the ratio of Cstn3 monomers and tetramers in the brain is not known (nor whether it changes or yet other multimers exist), we show that Cstn3 monomers and tetramers bind n1␣ with similar affinity (Fig. 7; see below). According to the neuroliginbased model, Cstn3 tetramers would cluster neurexins, but Cstn3 monomers would not, raising the question of whether monomeric Cstn3 has a different or even opposing function in the synaptic cleft. Remarkably, more than 50% of Cstn3 is present in the brain as a soluble ectodomain shed through proteolysis (3). It will be important to determine whether Cstn3 monomers and tetramers are equally susceptible to ectodomain shedding. Strikingly, intersubunit disulfide bonds interconnect and stabilize the Cstn3 tetramer. Intersubunit disulfide bonds in the extracellular space are also known to stabilize the cisdimer of E-cadherin, promoting homophilic adhesion (44), and cis-tetramers of ␥-protocaderins (45). Calsyntenins play a role in cognition, and an increasing body of work suggests a link between oxidative status and neurodegenerative disorders (46), so it is tantalizing to speculate that oxidative stress in the brain might shift the balance between Cstn3 monomers and tetramers in the synaptic cleft by promoting intersubunit disulfide bond formation, thereby altering Cstn3 function. Further studies are needed to determine the multimerization state of Cstn3 in the synaptic cleft and the factors that regulate the balance between monomers and tetramers.
We show that the ectodomains of Cstn3 and n1␣ interact with each other directly with nanomolar affinity promoted by Ca 2ϩ (Figs. 7-9). These data agree with cell-based assays, which observed similar Ca 2ϩ -dependent interaction between a soluble Fc domain-fused Cstn3 ectodomain and cell surface expressed n1␣ (3) (Fig. 10). Interaction between Cstn3 and n1␣ involves the LNS domain of Cstn3 and L5-EGFC-L6 of n1␣ (Figs. 8 -10). Whereas Cstn3 and NL2 both bind n1␣ with nanomolar affinity, promoted by Ca 2ϩ , Cstn3 uses a different molecular mechanism to interact with n1␣ compared with NL2, one that does not rely on L6 (Figs. 9 and 10). Nevertheless, the Cstn3 binding site on n1␣ may be spatially close to L6 because the Cstn3 ectodomain can suppress the synapse-promoting activity of NL2 and LRRTM2 (3), possibly by competing for neurexin interaction. Presentation may also influence binding, because the soluble n1␤ ectodomain readily interacts with Cstn3 biochemically but in cell-based binding or recruitment assays does not (3) (Fig. 10). Likewise, the isolated Cstn3 LNS domain binds n1␣ in solid phase assays but when tethered to the cell surface does not, although this domain is certainly required for binding because its deletion or mutation disrupts  DECEMBER 12, 2014 • VOLUME 289 • NUMBER 50 neurexin interaction and disrupts the synapse-promoting activity of Cstn3 (3). Although it is possible that technical aspects account for some of these minor differences, overall, the data presented here consistently and strongly indicate that Cstn3 interacts directly with n1␣.

Calsyntenin-3 Architecture and Interaction with Neurexin 1␣
Our structural studies provide insight into the function of monomeric and tetrameric Cstn3 alone and in the context of a trans-synaptic bridge with n1␣ (Fig. 11). Although domains cannot be unambiguously assigned in our molecular envelopes of Cstn3, given their low resolution, our tentative assignment of the Cstn3 LNS domains places them in a ring on the outside of the Cstn3 tetramer where they could recruit potentially up to four n1␣ molecules. Other as yet unidentified partners could also bind Cstn3 to form mixed macromolecular assemblies. Interestingly, in the cerebellum, tetramers of the postsynaptic GluR␦2 receptor assemble four neurexins, which triggers synapse formation, although binding is indirect and requires a tripartite complex with cerebellins (47). The cadherin domains in Cstn3 appear to fulfill a role different from those seen in classical cadherins (48), because the N-terminal Cstn3 Cad1 domain does not mediate side-by-side dimers, and tryptophan residues underlying this interaction are not present (Fig. 12). Furthermore, the Cstn3 Cad1-Cad2 tandem aligns best with the internal EC2-EC3 in the five-domain mouse cadherin 8 (Fig. 12). However, similarly to cadherins, the interface between Cad1 and Cad2 in Cstn3 is probably rigidified by multiple Ca 2ϩ ions binding between the adjacent domains because numerous calcium binding residues are conserved (Fig. 12). FIGURE 11. Model of the interaction between Cstn3 and neurexin 1␣ at a synapse. Monomeric Cstn3 would recruit a single neurexin 1␣ monomer (left), whereas Cstn3 tetramers would be able to recruit multiple neurexin 1␣ monomers (right), clustering the presynaptic and postsynaptic network more extensively. Neurexin 1␣ L2-L6 is shown (Protein Data Bank entry 3QCW) (16) with domains L1-EGF-A modeled in translucently to reflect their unknown location. FIGURE 12. Sequence comparison between Cstn3 Cad1-Cad2 and cadherin EC domains. Shown is sequence alignment of human Cstn3 Cad1 and Cad2 with the EC1, EC2, and EC3 domains of chick cadherin 5 (Q8AYD0), mouse cadherin 6 (P97326), mouse cadherin 8 (P97291), mouse cadherin 11 (P55288), mouse cadherin 2 (P15116), frog cadherin (P33148), mouse cadherin 1 (P09803), and human cadherin 1 (P12830) using Clustal Omega. The human Cstn3 Cad1 and Cad2 domains align well with the EC2 and EC3 domains of cadherins. Cstn3 shares sequence identity with many of the residues that form Ca 2ϩ binding sites in cadherin EC2, suggesting that the linker region between Cstn3 Cad1 and Cad2 is stabilized by Ca 2ϩ ions as well. However, Cstn3 Cad2 lacks a characteristic critical stretch of five residues that forms part of the Ca 2ϩ -binding sites that stabilize the linker region between cadherin EC3 and EC4. Tryptophan residues that underlie the strand swap dimerization mechanism of EC1 domains in cadherins are boxed in red. Amino acids aligning with the Ca 2ϩ -binding residues found in mouse cadherin 8 EC1-EC2-EC3 are highlighted in cyan. Blue and magenta spheres indicate the identities of the Ca 2ϩ ions bound in cadherins.
Our results support the idea that neurexins in combination with different interacting partners form very distinct heterophilic trans-synaptic bridges. We suggest that these trans-synaptic bridges are not intended to just align the intracellular presynaptic and postsynaptic machineries (that are attached to the respective cytoplasmic tails) but also to orient macromolecular assemblies in the synaptic cleft itself, altogether, recruiting and spatially organizing proteins into networks essential for synaptic function.