Dystroglycan Binding to α-Neurexin Competes with Neurexophilin-1 and Neuroligin in the Brain

Background: Extracellular matrix dystroglycan has essential functions at the neuromuscular junction and at inhibitory synapses in the brain. Results: Brain dystroglycan competes with neurexophilin-1 and neuroligins for binding to presynaptic α-neurexins. Conclusion: Competition between α-neurexin ligands in combination with alternative splicing determines formation of important trans-synaptic complexes. Significance: This is the first analysis of binding interference in α-neurexin multiplexes.

␣-Neurexins (␣-Nrxn) are mostly presynaptic cell surface molecules essential for neurotransmission that are linked to neuro-developmental disorders as autism or schizophrenia. Several interaction partners of ␣-Nrxn are identified that depend on alternative splicing, including neuroligins (Nlgn) and dystroglycan (␣DAG). The trans-synaptic complex with Nlgn1 was extensively characterized and shown to partially mediate ␣-Nrxn function. However, the interactions of ␣-Nrxn with ␣DAG, neurexophilins (Nxph1) and Nlgn2, ligands that occur specifically at inhibitory synapses, are incompletely understood. Using site-directed mutagenesis, we demonstrate the exact binding epitopes of ␣DAG and Nxph1 on Nrxn1␣ and show that their binding is mutually exclusive. Identification of an unusual cysteine bridge pattern and complex type glycans in Nxph1 ensure binding to the second laminin/neurexin/sex hormone binding (LNS2) domain of Nrxn1␣, but this association does not interfere with Nlgn binding at LNS6. ␣DAG, in contrast, interacts with both LNS2 and LNS6 domains without inserts in splice sites SS#2 or SS#4 mostly via LARGE (like-acetylglucosaminyltransferase)-dependent glycans attached to the mucin region. Unexpectedly, binding of ␣DAG at LNS2 prevents interaction of Nlgn at LNS6 with or without splice insert in SS#4, presumably by sterically hindering each other in the u-form conformation of ␣-Nrxn. Thus, expression of ␣DAG and Nxph1 together with alternative splicing in Nrxn1␣ may prevent or facilitate formation of distinct trans-synaptic Nrxn⅐Nlgn complexes, revealing an unanticipated way to contribute to the identity of synaptic subpopulations.
Unlike Nrxn variants that are expressed in most excitatory and inhibitory neurons (43), the ␣-Nrxn-specific ligand Nxph1 is restricted to inhibitory interneurons (36,44), similar to ␣DAG, which also prefers subsets of inhibitory synapses where it may co-localize with Nlgn2 (45)(46)(47)(48). Nxphs comprise a family of glycoproteins (Nxph1-4) that exhibit characteristics of secreted, preproprotein-derived molecules (35,36), but the structural determinants of their interaction with LNS2 (34) are still unclear. DAG in turn is produced from an evolutionarily conserved single gene (49) and proteolytically cleaved into extracellular ␣DAG and transmembrane ␤-DAG that remain non-covalently attached (50). Although specificity of ␣DAG binding to matrix proteins such as laminin comes from glycosylation of distinct residues in the mucin-rich regions (51,52), the glycan moiety required for association of Nrxn (37,53) is undetermined.
Here, we present the distinct interaction sites of Nxph1 and ␣DAG at the LNS2 and study their cross-talk with ligands of the LNS6 domain of ␣-Nrxn. Surprisingly, we observed that binding of ␣DAG and Nxph1 is mutually exclusive and that association of ␣DAG at LNS2 prevents formation of the trans-synaptic complex with Nlgn at LNS6. These are important results because impairments in ␣-Nrxn⅐Nlgn complexes are linked to neurodevelopmental disorders (54 -56), and there is symptomatic overlap with cognitive defects observed in DAG-associated muscular dystrophy syndromes (53,57,58).
All enzymes for restriction sites, dephosphorylation, ligation, and appropriate buffers were purchased from New England Biolabs (Ipswich, MA). Custom-made primers were made by Sigma. PCR was carried out with iProof TM high fidelity PCR (Bio-Rad), and DNA fragments were isolated using phenol-chloroform extraction or QiaEx (Qiagen, Hilden). All resulting intermediaries and final constructs were confirmed by DNA sequencing (GATC Biotech, Konstanz, Germany).
For co-sedimentation assays, secreted Fc-tagged proteins were bound to Protein A-conjugated Sepharose beads overnight, washed three times, and either analyzed directly or used in pulldown experiments essentially as described (20). In short, mouse brains were disrupted with a Polytron followed by Dounce homogenization in buffer H (100 mM NaCl, 5 mM CaCl 2 , 50 mM Tris, pH 7.5). Triton X-100 was added to a final concentration of 1% (w/v) for 3 h at 4°C followed by centrifugation at 220,000 ϫ g for 30 min. COS-7 or N2a cell lysates were obtained from scraped cells with 1% Triton X-100 in buffer H for 30 min at 4°C and centrifugation (15,000 ϫ g, 1 min). Aliquots of lysate were added to purified Fc fusion proteins in buffer H containing 0.1% Triton X-100 for binding at 4°C overnight. After washing, bound proteins were analyzed by SDS/ PAGE, Coomassie staining, and/or immunoblotting (Bio-Rad).
To obtain a pure Fc-Nxph1⅐LNS2-HA complex for mass spectrometry, we eluted free and LNS2-HA-bound Fc-Nxph1 from protein A beads with glycine buffer (50 mM glycine, pH 1.8) for 30 min, neutralized the eluate to pH 7 with 1 M Tris/HCl pH 8, separated the protein mixture on a Superdex 200 gel filtration column XK 16, and collected the fraction containing only Fc-Nxph1⅐LNS2-HA complex using an Ä kta TM prime system (GE Healthcare). The protein solution was concentrated to 500 l by Amicon ultracel filters (3K) and dialyzed against 10 mM ammonium bicarbohydrate in Slide-A-Lyzer TM chambers (Thermo Scientific) before mass spectrometry analysis.
Surface Plasmon Resonance (SPR) Analysis-Fc-Nxph1⅐LNS2-HA eluted from protein A beads as described above was bound covalently on a CMD 200m chip using 5 mM sodium acetate, pH 5, and 1-ethyl-3-(3-dimethyl-aminopropyl)carbodiimide, N-hydroxysuccinimide (EDC-NHS) following the manufacturer's protocol (Reichert Technologies, New York). Binding occurred randomly in a flow of 50 l/min, and stochastically either Nxph1, LNS2, or both proteins were irreversibly immobilized on the chip. 50 mM Tris, pH 7.4, served as the running buffer, and several solutions were tested to release either Nxph or LNS2 from their complexes. Complexes that were covalently linked to the chip via both proteins remained inert to elution buffers used, leading to a systematic underestimation of the eluted fraction under all conditions that did not bias our rela-tive comparison between WT and point-mutated complexes. Measurements were done with a two-channel Reichert SR7000DC SPR System (Reichert Technologies).
Mass Spectrometry-Nxph1 samples were digested in ammonium bicarbonate buffer (10 mM) overnight with trypsin, chymotrypsin, or a 1:1 mixture of trypsin and chymotrypsin at 37°C or with thermolysin at 65°C. Digest mixtures were dried, redissolved in water, and dried again. For separation of N-glycopeptides, ZIC-HILIC ProteaTips were used as described previously (61). Products were analyzed by nanoESI quadrupole time-of-flight mass spectrometer (Micromass, Manchester, UK) equipped with a Z-spray source in positive ion mode. Spectra were acquired at source temperature of 80°C, a desolvation gas (N 2 ) flow rate of 75 liters/h, a capillary voltage of 1.1 kV, and a cone voltage of 30 -40 V. For low energy collision-induced dissociation experiments, the (glyco)peptide precursor ions were selected in the quadrupole analyzer and fragmented in the collision cell using a collision gas (Ar) pressure of 3.0 ϫ 10 Ϫ3 Pa and collision energies of 20 -40 eV (Elab). The (glyco)peptide structures were deduced from the resulting fragment ion spectra (Table 1).
Structural Modeling-SwissProt entries NRX1A_RAT (Q63372-7), NXPH1_RAT (Q63366), and DAG1_Mouse (Q62165) were used to generate models (Figs. 3A, 4E, 6, 7C, and 8A). Gas6⅐Axl (PDB code 2C5D) served as template for the LNS2⅐Nxph1 complex, in which LNS2 (PDB code 2H0B) replaces Gas6. Axl residues Thr-204 to Lys-208 served as the backbone for the ␤-␤ sheet interaction coordinates, and the Nxph1 sequence was passed in single amino acids steps through this backbone structure to generate all possible LNS2⅐Nxph1 complexes. Each peptide complex was scored by calculating stability with FoldX. The complete C-terminal domain of Nxph1 was then homology-modeled using PAF (PDB code 2KCN) that contains an identical cysteine pattern, whereas the N-terminal domain was modeled by threading with PHYRE2. Both domains were manually connected and glycosylated using GLYCAM carbohydrate builder. An alternative model of the Nxph1 C-terminal domain was generated using coordinates of snake neurotoxin (PDB code 1NTN). For pictograms of ␣DAG, the mucin region was modeled using PHYRE2 (80) and a distorted structure as template (PDB code 4A54). The C-terminal domain as described (63) was modeled using the N-terminal domain structure (PDB code 1U2C), and parts were manually assembled and glycosylated with GLYCAM. For complex presentation, ␣DAG was manually docked to the Nrxn1␣ structure (PDB code 3R05), whereas Nlgn1 and Nlgn2 dimers were placed according to co-crystals (PDB code 3B3Q). All structures were rendered and visualized using PyMOL.

RESULTS
␣DAG, Nxph1, and ␣-Nrxn Are Expressed at Inhibitory Synapses-Although localization of ␣DAG at GABAergic terminals could be demonstrated by immunocytochemistry (45,48), the hypothesized presence of Nxph1 relied on indirect evidence from in situ hybridization data (36) and functional deficits observed in electrophysiological recordings from knockout (KO) neurons (64).
Here, we used immunogold electron microscopy to probe the ultrastructural localization of endogenous Nxph1 in cortical tissue of adult mice. To distinguish between actual localization and residual background from a polyclonal antiserum raised against the loop region (see "Experimental Procedures"), we compared labeling patterns in wild-type samples with KO and performed control labeling without first antibody using Lowicryl-embedded brain tissue. Although negative controls showed essentially no labeling (data not shown), Nxph1 normally localizes specifically to membranes of symmetric, inhibitory synapses in the neocortex (Fig. 1A). Asymmetric contacts, corresponding to excitatory synapses, were not labeled (arrow in Fig. 1B), whereas Nxph1 concentrated at the synaptic cleft of symmetric profiles (arrowhead in Fig. 1B) or in compartments of the secretory pathway such as rough endoplasmatic reticulum or Golgi cisternae (Fig. 1C), expected for a neuropeptidelike protein (35). To validate these observations, we also tested the ultrastructural distribution of its cognate receptor Nrxn and the trans-synaptic interaction partners of Nrxn, Nlgn1, and Nlgn2 in the same samples. We observed Nrxn at the synaptic cleft of both symmetric (Fig. 1D) and asymmetric (Fig. 1E) contacts in addition to localization in the secretory pathway ( Fig.   1F), consistent with their widespread expression and function in inhibitory and excitatory synapses (4,43). Demonstrating the reliability of our protocol, we could confirm the subtype-specific distribution of Nlgn2 at inhibitory (Fig. 1G) and Nlgn1 ( Fig. 1H) at excitatory synapses as reported (47,65). To finally brace against artifacts from our post-embedding procedure that may bias localization toward plasma membranes, we applied a pan-synapsin antibody but observed the expected different pattern over synaptic vesicles in presynaptic profiles (Fig.  1I). We conclude from our current and published results that Nxph1 is actually present at inhibitory synapses along with ␣-Nrxn, Nlgn2, and ␣DAG, providing a rationale for biochemical investigations of ␣-Nrxn/Nxph1-based multiplexes that might play a specific role at the GABAergic synaptic subpopulation.
␣DAG Binding to Nrxn-Binding of brain ␣DAG to ␣-Nrxn was reported (37), but its structural determinants and consequences for other interaction partners of Nrxn remained open. An obstacle had been the lack of information on ␣-Nrxn conformation; however, recent crystal data of extracellular sequences (33,66,67) allowed us to study the effects of ␣DAG binding based on structural predictions. The structure of ␣-Nrxn consists of six LNS domains ( Fig. 2A, green) intercepted by three EGF-like domains ( Fig. 2A, yellow) that assemble into a rigid core of LNS2-to-LNS5 (33,66). EGF2 and EGF3 show a typical ababcc cysteine knot pattern that tightly joins their adjacent LNS domains, whereas we determined here an aabbcc connectivity of EGF1 by mass spectrometry ( Fig. 2A, LDEX n GVC, m/z exp 1173.81; m/z calc 1173. 44) that can open a gap between LNS1 and LNS2 by more than 11 Å. Our observation explains the highly variable linkage of LNS1 (66,68) and makes LNS2 accessible. The u-form conformation of ␣-Nrxn ( Fig. 2A) opens the possibility that binding partners of the backfolded LNS2 domain interfere with ligands at the LNS6 domain. To address this important possibility, we first determined the exact binding epitopes of ␣DAG and Nxph1 using a combination of co-precipitation assays and site-directed mutagenesis as previously established (20).
Building on the sole previous study on ␣DAG-Nrxn interaction (37), we confirmed the binding of DAG to Nrxn1␣ and Nrxn1␤ and then tested all LNS domains individually (Fig. 2B). We found that ␣DAG from mouse brain interacts with both LNS2 and LNS6 (Fig. 2B, upper panel). This is an interesting result because only these domains, but none of the other four LNS, were shown to mediate all ligand binding (69), emphasizing the need to explore potentially competing complexes. Testing pulldown of Nlgn1 in the same experiments validated the known binding site at LNS6 (Fig. 2B, lane 14, middle panel) and Nrxn1␤ (lane 3) (15-17, 20, 70). In line with earlier studies which discovered that interaction of Nlgn1 and Nrxn depends on alternative splicing (25,26,71), Nlgn1 could not be pulled-down from brain lysates by full-length extracellular Nrxn1␣(ϩSS#4) (Fig. 2B, lane 4, middle panel) or by LNS5-EGF3-LNS6 cassette with insert (Fig. 2B, lane 8, middle panel). ␣DAG also prefers splice insert-free LNS domains (37) and thus binds to LNS2, LNS6, and Nrxn1␤ without insert in SS#2 or SS#4 in our co-sedimentation assay (Fig. 2B, lanes 3, 5-6, 10,  and 14, upper panel). Because ␣DAG is able to interact with two LNS domains, it can interact with Nrxn1␣(ϩSS#4, ϪSS#2) that has a blocked LNS6 but an insert-free LNS2 domain (lane 4). This is an interesting aspect as Nrxn in adult brains mostly contain (ϩSS#4) mRNA variants (22,72), suggesting that alternative splicing in Nrxn may affect several ligands simultaneously. Although some knowledge on competitive interaction of Nlgns and LRRTM with Nrxn is available (42), it is unclear if ␣DAG also competes for the same epitope at LNS6 and how its binding to LNS2 is affected by Nxph (34).
To determine the site of Nxph1 binding, we had to develop a modified binding assay because normal pull-down failed (Fig. 2C). Instead, we co-expressed recombinant Nxph1 with Fc-tagged Nrxn1 constructs in HEK293 cells, precipitated the pre-formed Nxph1⅐Nrxn-IgGFc complexes secreted into culture media with protein A beads, and tested binding by immunoblotting (see "Experimental Procedures" for details). This approach also allowed the reverse experiment with mutated Nxph1 residues, which were virtually impossible to accomplish by adenovirus-mediated transfer used previously to generate sufficient amounts of Nxph (34). Using the co-expression assay, we confirmed the binding of Nxph1 to isolated LNS2 (Fig. 2D, lane 6) and to the LNS1-EGF1-LNS2 cassette (lane 5), whereas other Nrxn1␣ domains (lanes 4, and 7-12) or Nrxn1␤ (lane 13) do not bind.
Interaction Sites of ␣DAG and Nxph1 on LNS2-To study the characteristics of ␣DAG and Nxph1 binding epitopes on LNS2, wild-type and mutated Fc-tagged LNS2 domains were immobilized on beads and tested for their ability to precipitate endogenous ␣DAG from brain lysate (Fig. 2E). As predicted from its calcium dependence (37)  Immunoelectron microscopy of Lowicryl-embedded neocortical tissue from murine brain was used to determine the exact localization of Nxph1 (A-C) and its cognate receptor Nrxn (D-F). sv, synaptic vesicle. Post-embedding with 10-nm gold-labeled secondary antibodies reveal Nxph1 only at symmetric, type 2 terminals (A and B, arrowheads), whereas asymmetric, type 1 contacts (B, arrows) are devoid of gold particles (circled in red in all panels). Nrxn is seen at both type 2 (D, arrowhead) and type 1 (E, arrow) synapses, representing inhibitory and excitatory terminals, respectively. C and F, labeling of Nxph1 and Nrxn in Golgi cisternae (Go) demonstrate their passage through the secretory pathway. G-I, control experiments showing the predicted differential distribution of the trans-synaptic Nrxn ligand Nlgn2 at symmetric (G) and Nlgn1 at asymmetric (H) synapses. I, a different labeling pattern is observed with anti-synapsin antibodies, confirming its association with synaptic vesicles in the presynaptic terminal of a type 1 spinous contact (sp). Scale bars, 200 nm, except in C and F, 300 nm.
␣-Neurexin Multiplexes panel), these results suggest that hydrophobicity of these Nrxn residues is required for ␣DAG binding, in contrast to a basic epitope in laminin interacting with ␣DAG (73,74). Although glycosylation of ␣DAG is essential for Nrxn binding (37,53), deletion of the LNS2 loop ␤11-␤12 (residues 428 -438), recently proposed to contain a carbohydrate binding site (33), demonstrates that it is not involved in ␣DAG binding (Fig. 2E,  lane 17, upper panel). Similarly, many additional residues were tested by site-directed mutagenesis but do not influence ␣DAG association (lanes 6 -10, 13, 14, and 16, upper panel). We mapped these data to the surface of the LNS2 structure, resulting in delineation of the ␣DAG epitope (magenta in Fig. 2, H-J).
Similar to ␣DAG, little is known about the Nxph1 binding epitope; therefore, we tested the entire surface of LNS2 to determine residues required for the Nxph1⅐LNS2 interface. For example, we deleted distinct loops like residues 428 -436 (Fig.  2G, lane 17) and changed hydrophobic to charged residues at strategic positions. We observed that the interface requires hydrophobicity of residue Ile-401 ( Because Ile-401 is part of the ␤10 strand in LNS2, we asked if Nxph1 engages in side chain-independent ␤-␤ interactions. We mutated the central residues of ␤10 to prolines and observed loss of Nxph1 complex formation (Fig. 2G, lanes 9 -10), whereas binding to ␣DAG persisted (Fig. 2E, lanes 9 -10). These data explain why binding between Nxph1 and ␣-Nrxn occurs calcium-and splice site-independently (34); the binding epitope at LNS2 (blue in Fig. 2, H and J) is distant to these positions and also non-overlapping with the ␣DAG binding site (magenta in Fig. 2, H-J).
In a first attempt to assemble the Nxph1⅐LNS2 complex by bioinformatics, we identified the crystal structure of Gas6 LNS1⅐Axl as a structural template. The interaction of ␤-strand residues 204 -208 of Axl with the correspondent ␤10 of Gas6 LNS1 (Fig. 2K, middle panel) is not their only contact interface (75) but best suits our purpose of modeling a Nxph1⅐LNS2 peptide complex (left and right panels). We generated models of all 150 possible LNS2⅐Nxph1 peptide combinations and calculated the relative change in free binding energy (⌬⌬G, Fig. 2L, right). These results show that any sequence of 5 residues of Nxph1 will bind to ␤10 of LNS2 with the only exception that a tyrosine is not allowed at position 2. This restriction only limits the number of potential complexes to 140, indicating that a mutagenesis study of single positions in Nxph1 has to await more structural information.
Nxph1 Prevents Simultaneous Binding of ␣DAG at LNS2-Our identification of separate binding epitopes for ␣DAG and Nxph1 suggested that simultaneous binding of both LNS2 ligands should be possible. We obtained complexes of Fc-tagged Nxph1 with a soluble extracellular domain of Nrxn1␣(ϩSS#4) or with LNS2-HA by co-expression that were purified and used to pull down ␣DAG from neuron-like N2a cells, a rich source of endogenous ␣DAG. Surprisingly, the Nxph1⅐Nrxn1␣ complexes could not interact with ␣DAG, and no triple complex was formed (Fig. 3A, lanes 7 and 8), whereas control pull-down with Nxph1-free Fc-Nrxn1␣(ϩSS#4) (Fig. 3A, lane 5) or Fc-LNS2 (lane 4) reliably bound ␣DAG. These results indicate that the presence of Nxph1 may sterically constrain ␣DAG binding to ␣-Nrxn, prompting us to examine key aspects of the Nxph1 structure in ␣-Nrxn binding.
Based on sequence analysis (35,36), Nxph1 was identified as a preproprotein with putatively secreted mature protein consisting of glycosylated N-terminal and cysteine-rich C-terminal domains (Fig. 3B). To determine the contribution of Nxph1 domains to complex formation with ␣-Nrxn, we probed if N-terminal or C-terminal sequences are involved and observed that both are required (Fig. 3C). In addition, we found that secretion of the C-terminal domain is reduced, possibly pointing to a role of the six cysteines in fold stabilization.
The highly conserved cysteines in the C-terminal domain should help to classify its structural fold (76). However, bioinformatic prediction programs like Raptor (77), I-Tasser (78), Rossetta (79), or Phyre (80) failed to predict the cysteine connectivity or the fold. We, therefore, purified recombinant Fctagged mature Nxph1 in complex with LNS2-HA and analyzed

␣-Neurexin Multiplexes
the structure by mass spectrometry methods (81). Our co-expression system produced two secreted protein fractions, free Nxph1_Fc and Nxph1_Fc, bound to LNS2-HA in a ratio of ϳ2:1 that we separated by gel filtration. Surprisingly, mass spectrometry revealed disulfide bonds in an abbacc pattern (Fig. 4A, left  and right panels) and not a more frequent cysteine knot, which is present, for example, in Nrxn1␣ EGF2 and EGF3 ( Fig. 2A) (33,66). This connectivity was the same in free and LNS2bound Nxph1. Because this abbacc pattern is rare but might be fold-stabilizing (82, 83), we successively opened all bridges by cysteine to serine mutations (CS, a to a-b-c) and found that any two bridges can be opened at the same time without an effect (Fig. 4B, lanes 4 -6, 10, 11, 14, and 15, upper panel). However, asymmetric triple mutations containing C239S strongly reduce Nxph1 secretion (lanes 12 and 13, lower panel). Similarly, opening of all three cysteine bridges reduced both binding capabilities (lane 7, upper panel) and secretion (lower panel), which is likely explained by a destabilized fold (84,85). The strong effect by opening the third bridge (lanes 7 and 10 -15) highlights Cys-239 as a key residue in fold stabilization. Although these results demonstrate that Nxph1 contains a rare fold with unusual cysteine pattern required for its own secretion bound to ␣-Nrxn, they do not solve the question of why additional binding of ␣DAG to preformed Nxph1⅐␣-Nrxn is blocked.
Because the N-terminal domain of Nxph1 is also involved in complex formation (Fig. 3C, lane 5), we evaluated the contribution of N-glycosylation, its distinctive feature. N-Glycosylation is not a prerequisite for complex formation as shown by tunicamycin treatment (Fig. 4C, upper panel) and point mutations of all three relevant residues including triple mutation N146D/ N156D/N162D (lower panel). Although this observation is consistent with earlier data (36), we now found in SPR experiments with preloaded Fc-Nxph1⅐LNS2-HA that the glycosylation is critical for complex stability (Fig. 4D); wild-type complex bound to SPR chips resisted stringent elution (lanes 1-8, upper  panel), and only near-denaturing conditions (6 M urea, lane 9; 7 M guanidinium chloride, lane 10) dissolved the complex. However, the complex with non-glycosylated triple Asn to Asp mutations (3xND) already started to fall apart at 1 M NaCl (Fig.  4D, ND, lower panel), pointing to an unexpected role of Nxph1 glycans in strengthening the interaction with ␣-Nrxn. We, therefore, analyzed the glycosylation pattern by mass spectrometry and observed two sites occupied by complex type glycans and one by high mannose-type oligosaccharides ( Fig. 4E and Table 1). Although complex type glycans with terminal sialic acids and core fucose as seen here on Asn-146 and Asn-162 of Nxph1 are unusual for such proteins (86), we observed the same N-glycans on its cognate receptor Nrxn1␣ (Table 1). The glycans identified add ϳ4 kDa to the N-terminal domain, leading to similar molecular weights for both domains (Fig. 3C). To visualize the complete, glycosylated mature Nxph1, we generated a model structure with glycans (Fig. 4F). The N-glycosylated N-terminal domain is shown as a single ␤-turn (Fig. 4F, ␤1/␤2, light blue ribbon) that is flexibly linked (gray helical linker and antibody epitope) to the C-terminal domain. Because only NMR data of the antifungal protein PAF (83) described an abbacc cysteine fold, we used these coordinates to model the C-terminal domain, which constitutes a three-leafed sevenstranded ␤-fold (␤3-␤9, dark blue ribbon) stabilized by three cysteines (yellow sticks). In contrast to PAF, we have determined constant connectivity, but cysteine isomerization can explain the stabilizing effect of asymmetric triple-mutation CS-aϩC256S (Fig. 4B, lane 14), whereas CS-aϩC239S appears unstable (lane 12). Assuming the same cysteine cluster as in the protein PAF, the free Cys-239 in CS-aC256S could form an alternative cysteine bond to Cys-194 of a, whereas vice versa a free C256 in CS-aϩC239S was not in reach of a. From this model it is likely that the C-terminal domain will bind to ␤10 of ␣-Neurexin Multiplexes LNS2 (Fig. 2J). In addition, the high mannose-type glycan on Asn-156 is likely to be buried in the interface with ␣-Nrxn to protect Nxph1 from ubiquitination and endoplasmic reticulum-associated degradation (87).
Determinants of ␣DAG Binding to ␣-Nrxn-Although our experiments above were performed with endogenous, glycosylated dystroglycan from brain or N2a cells, purified recombinant ␣DAG variants are necessary for mutagenesis. Such experiments were difficult because even large amounts of Fctagged ␣DAG secreted from HEK293 cells were hardly detectable by standard antibodies VIA4 -1 or IIH6C4, suggesting insufficient or inappropriate glycosylation (data not shown). This situation changed with the identification of LARGE that successively adds disaccharides of xylose-glucosamine terminal to complex O-mannosyl glycans of the mucin region (Fig. 5A), which are required for ␣DAG binding to laminin and agrin (88,89). To test the role of LARGE for Nrxn binding, we co-transfected HEK293 cells with Fc-tagged ␣DAG and LARGE and found that glycosylation of ␣DAG by LARGE is sufficient for binding to endogenous and recombinant Nrxn1␣ (Fig. 5B, lane 3, first and second panel). More importantly, using LARGEmodified recombinant ␣DAG, we were able to pull down ␣-Nrxn from mouse brain lysates (lane 3, first panel), an experiment not even reported for the intensely investigated Nlgn. Although dependence of Nrxn binding on LARGE is consistent with binding to laminin, the sites for the LARGE-mediated glycosylation appear different; we investigated by mutagenesis if Nrxn binds to the ␣DAG region including Thr-315 and Thr-317 that mediate laminin binding (90) but noticed that the ␣DAG⅐Nrxn1␣ complex formation is not reduced if this region is mutated (Fig. 5C, lane 7, first and second panel).
To determine the glycosylated region of ␣DAG that is responsible for Nrxn binding, we tested the complete and either half of the mucin region (Mucin, muc1, and muc2 in Fig. 5A). We observed that either half of the mucin region is sufficient to precipitate brain or recombinant ␣-Nrxn (Fig. 5C, lanes 4 and  5). In contrast, deletion of the complete mucin region abolished Nrxn binding (lane 6). A control mutation of the single N-glycosylation site Asn-139 in the N-terminal domain of ␣DAG had  , right panel). B, successive Cys-to-Ser mutations (lanes 4 -7), analysis of single mutations (lanes 10 -11), and their combinations (lanes [12][13][14][15] identify Cys-239 as most sensitive for Nxph1 binding to LNS2 (lanes 12 and 13). C239S reduces secretion of Nxph1 when combined with cysteine bridge a (lanes 12-13); similar combinations with C256S have no effect (lanes 14 -15). C, N-glycosylation of Nxph1 is not required for complex formation with LNS2. Binding of recombinant mature Nxph1 to Fc-tagged LNS2 co-expressed in COS7 cells without (lane 3) and with (lane 4) the addition of tunicamycin (upper panel). Successive Asn-to-Asp mutations of all N-glycosylation sites (Asn-146, Asn-156, and Asn-162) did not prevent complex formation (lanes 3-6, lower panel). D, N-glycosylation stabilizes the Nxph1⅐LNS2 complex. In a reversed surface plasmon resonance experiment, purified Nxph1-Fc⅐LNS2-HA complex covalently linked to CMD chip was tested to dissolve by serial injection of 100 mM NaOAc pH4 (1), 50 mM Tris, pH 8.9 (2), 250 mM NaCl (3), 500 mM NaCl (4), 1 M NaCl (5), 2 M NaCl (6), 4 M NaCl (7), 5 mM EDTA (8), 6 M urea (9), and 7 M guanidinium chloride (10). The wild-type (WT) complex was not affected by most conditions (the base line not changed, 100% binding) but can be disassembled by denaturing agents (9 and 10), serving as the reference for releasable protein (0% binding). In contrast to WT (square, upper panel; black trace, lower panel), deglycosylation by the triple Asn to Asp mutation (3ϫ ND) released 9% of the complex (red trace, lower panel) after injection of 1 M NaCl. E, glycosylation pattern of the Nxph1 N-terminal domain as determined by mass spectrometry (for details, see Table 1). Residues Asn-146 and Asn-162 were bound to complex type glycans containing fucose and sialic acids, and Asn-156 was linked to high mannose glycans. F, molecular model structure of Nxph1 with glycans attached. OCTOBER 3, 2014 • VOLUME 289 • NUMBER 40

TABLE 1 N-Glycan Structures of Nxph1 and Nrxn1␣
Results are from nanoESI quadrupole time of flight mass spectrometry. Glycans of a complex type were found at Asn-146 and Asn-162 of Nxph1 and at Asn-125, Asn-190 and Asn-797 of Nrxn1␣. The high mannose type N-glycan exclusively attached to Asn-156 was shown by analyzing wild-type and mutant proteolytic glycopeptides where Asn-162 is inactivated by mutation to Asp. A ϭ Gal, AN ϭ GalNAc, GN ϭ GlcNAc, F ϭ Fuc, M ϭ Man, NA ϭ NeuAc.

␣-Neurexin Multiplexes
27594 JOURNAL OF BIOLOGICAL CHEMISTRY VOLUME 289 • NUMBER 40 • OCTOBER 3, 2014 no effect (Fig. 5D, lane 4). These data indicate that binding of Nrxn to ␣DAG is locally less restricted compared with laminin, which binds mainly to a distinct N-terminal part of the mucin region (52,90,91).
Multiplexes of Nrxn1␣-The mutually exclusive binding of Nxph1 and ␣DAG to LNS2 of ␣-Nrxn shown here indicates that multiple ligand interactions have to be considered to understand the behavior of Nrxn-based molecular complexes. Based on the u-form conformation of ␣-Nrxn recently identified ( Fig. 2A), we asked if ligands of the LNS2 and LNS6 domains may influence each other. We found that Nxph1 in complex with full-length Nrxn1␣ at LNS2 prevents successive binding of ␣DAG because binding to its second site on the LNS6 is blocked by the presence of an insert at SS#4 (Fig. 6A,  lane 4). A triple complex of Nxph1⅐Nrxn1␣⅐␣DAG is possible, however, when the insert is missing (lane 5). Interestingly, binding of ␣DAG to LNS2 of Nxph1-free Nrxn1␣(ϩSS#4) (lane 3) or to LNS6 of Nxph1⅐Nrxn1␣(ϪSS#4) (lane 5) is similarly efficient, indicating an undisturbed binding of ␣DAG to LNS6 when Nxph1 is associated with LNS2. To brace against artifacts from a mixture of ␣-Nrxn with and without Nxph1 in these pulldown assays, we used Fc-tagged Nxph1 for a 1:1 stoichiometry with complexed ␣-Nrxn. After purification, Nxph1 is present as shown by anti- Nxph1 (lanes 4 and 5), and the complex was used to test the binding to the third ligand ␣DAG (Fig. 6A, pictograms in the right panel).
To address a second multiplex, we probed the binding of Nlgn1 to a preformed Nxph1⅐Nrxn1␣ complex (Fig. 6B). We observed that Nlgn1 binds normally to its sole interaction site at LNS6 independent of the presence of Nxph1 on LNS2 (Fig. 6B,  lanes 3 and 4), resulting into a Nxph1⅐Nrxn1␣⅐Nlgn1 triple complex (Fig. 6B, right panel). To analyze the influence of Nxph1 even on residual low affinity binding of Nlgn1 to LNS6, we also included the interaction of Nlgn1(ϩSSB) to Nrxn1␣(ϩSS#4), a pairing of variants that can barely be detected after long incubation times (lane 6) and is even ineffective in synapse formation assays (25,31). Despite this weak interaction, the Nxph1⅐Nrxn1␣⅐Nlgn1 triple complex could form in the presence of both splice inserts (lane 5), suggesting that interactions of Nxph1 and Nlgn1 with ␣-Nrxn occur independently of each other, in contrast to ␣DAG.
Finally, we examined simultaneous binding of ␣-Nrxn to Nlgns and ␣DAG. This is an important triple complex because ␣DAG was reported to mainly localize to inhibitory synapses (45), similar to the Nlgn2 variant (47) that can form a physiologically relevant trans-synaptic complex with ␣-Nrxn at this synapse (10,92,93). We, therefore, included Nlgn2 along with Nlgn1(ϪB) in our experiments and noticed that both bind equally to Nrxn1␣(ϩSS#4) (Fig. 6C, lane 4), consistent with a recent SPR study using an isolated LNS6 domain (71). Surprisingly, a preformed complex of Fc-tagged ␣DAG with Nrxn1␣(ϩSS#4) in which ␣DAG can only be bound to LNS2 prevented any detectable binding of Nlgn1 or Nlgn2 (Fig. 6C,  lane 3). In support, the reverse experiment with Fc-tagged Nlgn(ϪB) bound to Nrxn1␣(ϩSS#4) also failed to precipitate ␣DAG (data not shown), suggesting that a triple complex of ␣-Nrxn⅐Nlgns/␣DAG is unlikely to occur in brain. This is an unexpected result because ␣DAG and Nlgn compete in binding to ␣-Nrxn at different domains. However, it might be explained by the u-form of ␣-Nrxn that brings binding sites on LNS2 and LNS6 very close to each other ( Fig. 2A). Thus, the relatively large and similar size of ␣DAG and Nlgn dimers (Fig. 6C, molecules shown in the pictograms are to scale) might sterically hinder each other when bound to LNS2 and LNS6, respectively.
Binding Site of ␣DAG on LNS6 -Because the u-form conformation of ␣-Nrxn allows ␣DAG to bind simultaneously to LNS2 and LNS6, we finally analyzed the binding epitope of ␣DAG on the LNS6 domain. To determine the ␣DAG⅐LNS6 interface, we first confirmed the dependence of ␣DAG on alternative splicing in SS#4 as suggested by Sugita 7A, lanes 3 and 5, first panel), and inclusion (ϩSS#4) completely blocks this interaction (Fig. 7A, lanes 2 and 4, first panel). Because the binding is also calcium-dependent (37), we could successfully abolish the interaction by alanine mutations of the calcium-coordinating residues Asp-1183 (lane 9, first panel) and Gly-1201 (lane 10). As Leu-1280 and Ile-1282 were identified as hot spot residues at the interface of the Nlgn1⅐LNS6 complex (15-17, 20, 94), we introduced a triple mutation L1280S/I1282S/N1284D that removes hydrophobicity from the surface surrounding the calcium coordination site and observed that it prevents both Nlgn1 and ␣DAG binding to LNS6 (lane 11). This result surprisingly indicates that hydrophobic residues are essential for ␣DAG. More importantly, we identified an arginine mutation of Ile-1282 that is able to discriminate between Nlgn and ␣DAG binding to LNS6 by blocking ␣DAG and leaving Nlgn1 and Nlgn2 unscathed (lane 13). These findings suggest that the more limited hydrophobicity of arginine side chains is sufficient for Nlgn association but abolishes ␣DAG binding. In addition to these overlapping residues, we also discovered that residue Thr-1281 is an exclusive hot spot for ␣DAG binding not shared by Nlgn (lane 12). Together, our results reveal that the binding epitope for ␣DAG on LNS6 completely circles the calcium binding site (Fig. 7A, right panel) but also raises the question of how the different epitopes of ␣DAG on LNS2, LNS6, and on laminin LNS domains relate to each other.
To directly compare binding preferences of ␣DAG at the two ␣-Nrxn and the laminin LNS domains (51,73), we took advantage of their conserved rigid fold (20). We generated hybrid constructs of LNS2, LNS6, and LAM␣2LNS5 that express swapped calcium coordination sites and found that the calcium coordination of LNS6 could be transferred to LNS2 with intact ␣DAG binding (Fig. 7B, lane 4) but not vice versa (lane 7). The laminin calcium coordination contains two serines (PDB code 1QU0) but cannot support ␣DAG binding when transferred to LNS2 or LNS6 (lanes 5 and 8) even with conserved calcium binding to such a hybrid (20). These data suggest that binding of ␣DAG to LNS domains of Nrxn1␣ is structurally different to interaction with laminin. This conclusion is supported by the fact that the entire surface of laminin LNS domains is positively charged except for the calcium binding groove (Fig. 7B, right model) and requires basic residues (51,73,74,95). In contrast, the rim surfaces of LNS2 (Fig. 7B, left model) and LNS6 (not shown) are mostly negatively charged and display hydrophobic residues near the calcium binding groove. This unusual hydrophobic property of the Nrxn LNS domains was recognized to serve as the LNS6⅐Nlgn1 interface (15-17, 20, 70) but also mediates the important interaction of ␣DAG to ␣-Nrxn as shown above. Comparison of the two ␣DAG sites on LNS2 and LNS6 revealed that binding of Nxph1 might sterically hinder the approximation of ␣DAG to the calcium binding site (Fig.  7C, left model). Interestingly, the insert in SS#4 in switched conformation (71) of the LNS6 (middle model) is located at the same side where Nxph1 associates with LNS2, possibly mim-icking a similar steric block. In both cases, the binding obstacle may hinder attachment of ␣DAG to the side near ␤10-strand of the respective LNS domain (right model).

DISCUSSION
This work presents the first biochemical and structural analysis of binding interference in ␣-Nrxn-based complexes and identifies important determinants for competition in multiple interactions with Nlgn, ␣DAG, and Nxph.
Technical Considerations-To study the ␣-Nrxn multiplexes, we improved methods to isolate and purify ␣DAG and Nxph1. First, DAG is expressed in most tissues including brain (96), but only ␣DAG glycosylated by LARGE is able to bind to laminins (97) and Nrxn1␣ (Fig. 5B). We modified the method of Sugita et al. (37) by omitting preselection with wheat germ agglutinin because this yielded more Nrxn1␣ binding ␣DAG. Although wheat germ agglutinin has been successfully applied to characterize ␣DAG binding to laminin (51,52,73,90,97), laminin also precipitates more ␣DAG without wheat germ agglutinin (98). Second, we observed that neuroblastoma N2a cells are a rich source of Nrxn binding ␣DAG that facilitated a simple lysis procedures with Triton X-100 as detergent. Third, recombinant Nxph1 has previously been generated by adenovirus-mediated transfection of PC12 cell cultures (35). Here, we developed a less cumbersome alternative strategy by co-expressing Nxph1 with Nrxn1␣/LNS2 in HEK293 cells. This procedure yields a high amount of Nxph1⅐LNS2 complex sufficient even for mass spectrometric analysis of glycan moieties and cysteine connectivity. With these improved tools, we studied Nrxn1␣ forming binary and tertiary complexes with Fc-tagged ␣DAG, Nxph1, and Nlgns using a strategy that we successfully applied to determine hot spot residues at the Nrxn⅐Nlgn interface (20). The results from that previous biochemical investigation were entirely consistent with crystallographic studies (15)(16)(17) attesting the reliability of the current approach.
Promiscuity of LNS Domains-The calcium coordination site is described as the major binding region in LNS domains for proteins and steroid hormones (19). Our study extends its versatility to binding of glycans as determined here for ␣DAG⅐LNS2 and ␣DAG⅐LNS6. In addition, our identification of the Nxph1 binding epitope highlights ␤10 as a second versatile region because the receptor-tyrosine kinase Axl also binds to a corresponding region at LNS1 of Gas6 (75). In support, ␣DAG likely covers ␤10 when competing with Nxph1 for binding to LNS2 (Fig. 7C), and the insert in SS#4 can replace ␤10 of LNS6 (71).
It is remarkable that all known, structurally diverse Nrxn ligands bind solely to the LNS2 and/or LNS6 domains. Even more astonishing is the observation that Nlgn, LRRTM, and ␣DAG may compete for the same epitope at LNS6 (Fig. 7C). The reason may reside in an unusual calcium-induced hydrophobic and water-layered interface as shown for Nlgn1⅐LNS6 (15)(16)(17)20). Such hydrophobic environments are actually predicted to make stable connections to structurally diverse ligands through dynamic variations of hydrophobic contact points (99). Consistently, the conserved interfaces of Nlgn1⅐LNS6 (PDB codes 3WKF and 3B3Q) and Nlgn4⅐LNS6 (PDB code 2XB6) differ in their hydrophobic contact points. Moreover, we show ␣-Neurexin Multiplexes  (green) forming binary and triple complexes with Nrxn1␣ (white) as determined experimentally in this study. All of these molecules are present or even enriched at inhibitory terminals (Refs. 10, 36, 45 and47 and Fig. 1). We used available crystal structures of Nrxn1␣ (PDB codes 3POY and 3R05) and Nlgn2 (PDB code 3BL8) and generated model structures of ␣DAG without the N-terminal domain (97) and of Nxph1 to create complexes. The Nlgn dimer and ␣DAG are of comparable size (radius of gyration (Rg) of 40 and 35 Å, respectively), and both are as long as the rigid core unit LNS2-to-LNS5 of ␣-Nrxn (Rg ϭ 37 Å). Both ␣DAG and Nlgn2 cover the LNS5 that may cause steric hindrance and explain the mutually exclusive binding of ␣DAG and Nlgn2 to Nrxn1␣. B, probability of Nrxn1␣ in complex with ␣DAG (magenta), Nxph1 (blue), and/or Nlgn2 (green), calculated from the total number of possible complexes that are limited by competitive binding (left panel). ␣DAG complexes appear with higher probability when Nrxn1␣ is splice insert-free, whereas Nlgn2 complexes increase with inclusion of inserts at SS#2 and/or SS#4. If non-competitive binding was assumed, the splice-insert dependence of Nlgn2 and the advantage of ␣DAG over Nlgn2 complexes would be diminished (no competition, right panel). The size of molecules in A are to scale, and the modeled complexes were generated using two criteria, (i) coverage of hot spots and (ii) maximal surface area buried (see "Experimental Procedures"). Note that in addition to the complexes shown, other conformations are not excluded. The model of Nlgn⅐Nrxn1␣ modified from Refs. 33 and 66. OCTOBER 3, 2014 • VOLUME 289 • NUMBER 40 here that ␣DAG also requires hydrophobic residues in the vicinity of the calcium binding sites of LNS2 and LNS6 (Figs. 2E and 7A).

␣-Neurexin Multiplexes
Because Nrxn binds to complex O-mannose-type glycans that completely cover the mucin region of ␣DAG (Fig. 5C) (100 -103), it is likely that LNS domains bind only to the accessible sugars but not to protein residues of ␣DAG. Accordingly, the DAG sequence is highly conserved, and ligand specificity derives from post-translational glycosylation (49). We report here that Nrxn binding depends on LARGE that adds multiple xylose-glucosamine saccharides to glycans attached to ␣DAG. It is still unknown how exactly sugars bind to LNS domains, but hydrophobic residues are commonly found to participate in binding oligosaccharides (104,105), and their contribution for stability of a protein⅐glycan complex can be considerable (106). This idea is supported by crystal structures in which mannose (PDB code 1KZA) or galactose (PDB code 1TLG) coordinate a calcium ion and alkyl or aromatic side chains, performing a hydrophobic stacking with the hydrophobic "bottom" of a sugar pyranose ring (104). Consistently, we determined Tyr-1281 at LNS2 (Fig. 2E) and Leu-1280 and Ile-1282 at LNS6 (Fig.  7A) as glycan interacting residues.
LNS2 and LNS6 are the only LNS of ␣-Nrxn with a hydrophobic calcium coordination site. In analogy, ␣DAG binds solely to the second LNS of pikachurin that has a phenylalanine at the position corresponding to Tyr-412 of LNS2 in Nrxn1␣ (107,108). In contrast to these hydrophobic LNS domains, the binding of ␣DAG to laminin is mediated by basic residues on LAMA2 LG4 -5 and LAMA1 LG4 (51,73,74,95). This positive surface fits to the finding that negatively charged sulfated-glycans like heparan sulfate bind to laminin LNS (51,73,74,95). Furthermore, a crystal structure of LAM␣2LNS5 (PDB code 1QU0) revealed that a sulfate ion bound to calcium and the ␣DAG N-terminal mucin region contains sulfated (91) in addition to phosphorylated glycans (52,109), which explains why laminin binding is restricted to this region (52,90,91). In contrast, Nrxn1␣ also interacts with the C-terminal mucin region (Fig. 5C) that is free of sulfated glycans (91). Together, these data allow the definition of two classes of ␣DAG binding epitopes in the vicinity to calcium coordination sites: (i) a basic surface binding to sulfated or phosphorylated glycans and (ii) a hydrophobic surface binding directly to the pyranose ring of sugars. Although laminin and agrin belong to the first class, pikachurin and Nrxn fall into the second class that is related to C-type carbohydrate sites of lectins, including the ␣DAG binding concanavalin A (110).
Multiplexes of ␣-Nrxn-We tested biochemically the capability of ␣-Nrxn to form simultaneously complexes ("multiplex") with its ligands Nxph1, ␣DAG, and Nlgn. For the purpose of a comprehensive discussion, all possible and "forbidden" multiplexes of ␣-Nrxns are schematically displayed in Fig. 8A. The summary shows, for example, that there are (i) no complexes with both ␣DAG and Nlgn and (ii) no complexes with ␣DAG when SS#2 and SS#4 contain inserts, and (iii) all triple complexes include Nxph1. Importantly, all Nrxn1␣ interactions with ␣DAG, Nlgn, or Nxph1 investigated here represent irreversible complexes under physiological conditions. Nrxn1␣ bound to ␣DAG or Nlgn can only be disassembled by removal of calcium with EDTA (26,37), whereas Nxph1 can only be dissociated from Nrxn1␣ by near-denaturing conditions (Fig. 4D) (34,36). As a consequence, disassembly of these complexes needs stringent measures, for example, extracellular metalloproteases that have been shown to cleave ␤DAG (111), Nlgn1 (112), and Nrxn1␣ (113) from their membrane-bound C terminus.
Under the simplistic assumption that Nrxn1␣ ligands are abundantly available and binding occurs randomly, the probability of complex formation is determined by the number of particular ligand complexes divided by all possible complexes. We have analyzed this probability for Nrxn1␣ complexes with ␣DAG (Fig. 8B, magenta), Nxph1 (blue), and Nlgn2 (green) at a "virtual" inhibitory terminal because these proteins are present at those synapses (45,47) (Fig. 1). Because the probability depends on the number of relevant splice sites with insert (0, 1, or 2; at SS#2 and SS#4 together), the analysis reveals that ␣DAG in complex with splice insert-free ␣-Nrxn forms with a probability of nearly 0.6 ( Fig. 8B, magenta, left panel). Nlgn2 complexes (green) reach the same value when ␣-Nrxn contains both inserts, preventing ␣DAG binding. These individual probabilities of ␣-Nrxn complexes are a result of the competitive binding that we have biochemically determined for ␣DAG, Nxph1, and Nlgn. If non-competitive binding was assumed, the probability of an ␣DAG⅐␣-Nrxn complex without splice inserts would be increased to ϳ0.8 (Fig. 8B, magenta, right panel), whereas Nlgn2⅐␣-Nrxn complexes would not change much (green, right panel). Consequently, our analysis shows that the presence of ␣DAG can change the probability of Nlgn2⅐␣-Nrxn complexes.
␣DAG is expressed much earlier in development than Nrxn1␣ (114,115), suggesting that ␣DAG is an early binding partner. Our hypothesis that the probability of an ␣DAG⅐Nrxn1␣ complex is higher when Nrxn1␣ carries no splice inserts (Fig. 8B) may have important functional implications because some studies indicate that juvenile neurons express mostly insert-negative Nrxn(ϪSS#4) variants (116). 3 Moreover, the number of insert-positive variants appears to increase with synapse maturation (116), and ϩSS#4 expression is reduced after applying a learning and memory paradigm (117). In contrast to ␣DAG, Nlgn has a higher probability of interacting with Nrxn1␣-containing inserts (Fig. 8B). Consistent with a role in later developmental stages, there is evidence that synaptic activity and maturation of synapses can increase the insert-positive variants via the calcium/ calmodulin-dependent kinase pathway that involves RNA-binding protein SAM68 (24,118). Thus, developmental and/or activity-regulated control of alternative splicing in Nrxn could modify the composition of multiplexes with ␣DAG, Nxph1, and Nlgn at inhibitory synapses, adding an exciting and unanticipated layer of complexity to the regulation of these essential molecules.