Mutational analysis of the neurexin/neuroligin complex reveals essential and regulatory components

Abstract

Neurexins are cell-surface molecules that bind neuroligins to form a heterophilic, Ca²⁺-dependent complex at central synapses. This transsynaptic complex is required for efficient neurotransmission and is involved in the formation of synaptic contacts. In addition, both molecules have been identified as candidate genes for autism. Here we performed mutagenesis experiments to probe for essential components of the neurexin/neuroligin binding interface at the single-amino acid level. We found that in neurexins the contact area is sharply delineated and consists of hydrophobic residues of the LNS domain that surround a Ca²⁺ binding pocket. Point mutations that changed electrostatic and shape properties leave Ca²⁺ coordination intact but completely inhibit neuroligin binding, whereas alternative splicing in α- and β-neurexins and in neuroligins has a weaker effect on complex formation. In neuroligins, the contact area appears less distinct because exchange of a more distant aspartate completely abolished binding to neurexin but many mutations of predicted interface residues had no strong effect on binding. Together with calculations of energy terms for presumed interface hot spots that complement and extend our mutagenesis and recent crystal structure data, this study presents a comprehensive structural basis for the complex formation of neurexins and neuroligins and their transsynaptic signaling between neurons.

The heterophilic complex formed by cell adhesion molecules neurexin (Nrxn) and neuroligin (Nlgn) reflects the asymmetric nature of the synapse with presynaptic and postsynaptic specializations (1, 2). Nrxns and Nlgns are essential molecules because they perform important functions in synaptic transmission (3, 4) and differentiation of synaptic contacts (5, 6), and both molecules have been identified as candidate genes for autism (7, 8).

Nrxns form a family of transmembrane proteins with variable extracellular sequences. Nrxn genes (Nrxn1–3) give rise to α-neurexins and shorter β-neurexins that contain five (α-Nrxn) or two (β-Nrxn) splice sites (SS1–5) (9). Although they share most sequences, the essential role of α-Nrxn in neurotransmission cannot be replaced by β-Nrxn (10), and one ligand exists for α-Nrxn that does not bind to β-Nrxn (11). In contrast, Nlgn was discovered by its interaction with β-Nrxn (12). The cholinesterase-like adhesion molecule (CAM) domain of Nlgn interacts with the extracellular domain of β-Nrxn in a Ca²⁺-dependent manner, and binding is facilitated by the splice variation of β-Nrxn that lacks an insert in SS4 (13). Nlgn mRNA is also susceptible to splicing, at two positions referred to as A and B (12), including splice variants with no insert in B that bind to all β-Nrxns and presumably α-Nrxn (14). Therefore, the Nrxn/Nlgn complex involves a domain shared by α- and β-Nrxns, and any structural characterization needs to account for the Ca²⁺ dependence and regulation by alternative splicing (15–17).

Extracellularly, both Nrxns contain laminin-like (LG/LNS) domains that in α-Nrxn alternate with interspersed EGF-like sequences. α-Nrxn contains six (LNS1–6) and β-Nrxn only one LNS domain (9, 18), which is responsible for Nlgn binding (13) and contacts the Nlgn dimer from opposite sides (19). The crystal structure of the single LNS domain of Nrxn 1β demonstrated that it is related to lectin-like domains (20), and the structure of the second Nrxn 1αLNS domain revealed the presence of Ca²⁺ binding sites (21). Recently, three cocrystal studies of the Nrxn/Nlgn complex consistently identified the contact areas but differed in the actual residues of the interface and the identification of Ca²⁺ coordination sites (15–17). For example, the role of R109, N103, and D104 in β-Nrxn and of K306, E397, L399, and R597 in Nlgn1 remained ambiguous in these studies, raising the question of which amino acids are in fact essential for binding. Moreover, none of these crystal structures provided a rationale for binding of Nrxn to Nlgn2 (15–17), the major binding partner of α-Nrxn (22), raising the possibility of an alternative interface.

Here we reveal through mutagenesis experiments the essential residues of Nrxn that constitute the hydrophobic contact site surrounding the Ca²⁺ binding pocket, and we show that these residues are required for binding to all Nlgn isoforms. Similarly, we identified a single essential amino acid in Nlgn but conclude from numerous additional mutations and calculation of energy terms that the binding site on the surface of the Nlgn CAM domain is less distinct than on the Nrxn LNS domain.

Results

Residues Required for Ca²⁺ Coordination in the Nrxn/Nlgn Complex.

To better define the core structure of the LG/LNS family (9), we compared LNS folds of Nrxn to those of agrin, laminin, Gas6, SHBG, and pentraxin, employing pairwise structural alignments [supporting information (SI) Fig. S1A]. The core is composed of β-strands β2-β11 and is stabilized by hydrophobic residues that bind a fragment of the β10/β11 loop. The rigid core determines that residues of a Ca²⁺ coordination site are placed on top of three loops in agrin, laminin, and Nrxn (i.e., β5, turn β6/7, and loop β10/11), characterized by an rmsd of 0.63 Å [from Nrxn 1β; Protein Data Bank (PDB) ID code 1C4R]. To test our definition of a core structure, we calculated Cα-Cα difference distance plots and difference Ramachandran values for LNS pairs. Changes were mostly located at nonconserved loop regions, whereas no changes occurred at the Ca²⁺ sites (Fig. S1B). This rigidity is unique because Ca²⁺ binding is mostly mediated by flexible loops (23–25). In contrast, the β-turns are subject to alternative splicing in Nrxn and agrin and are termed “hypervariable regions” (20). Because sequence and length of loops differ between LNS domains, we next explored the possibility that the surface surrounding the Ca²⁺ binding site is responsible for their specific protein–protein binding properties. We mapped shape and electrostatic potentials, and we observed that, in most LNS domains, the Ca²⁺ site features as a negative pole surrounded by a ring of positively charged residues, e.g., in agrin (Fig. 1A and Figs. S2 and S3). However, the Nrxn1β LNS/Nrxn1α LNS6 domain differs from this “standard surface” because hydrophobic residues engulf the Ca²⁺ binding pocket (Fig. 1B), consistent with recent crystal structures of the Nrxn/Nlgn complex (15–17).

Fig. 1.

Ca²⁺ coordination in Nrxn. (A and B) Electrostatic surface analysis surrounding the Ca²⁺ binding site in agrin (A) and Nrxn1α LNS6 (B). Negative (red) and positive (blue) charges and long hydrophobic residues (arrowheads) are indicated. (C) Immunoblot showing pull-down experiments of Nlgn1 from brain using IgG fusion proteins of wild-type βLNS domain, various alanine mutations, and control vector. (D) Similar to Nlgn1 (C), Nlgn2 and Nlgn3 are precipitated by β-Nrxn with and without splice insert SS4 but failed to bind to Nrxn mutations βD137A and βN238A. Residue T235 in β-Nrxn does not inhibit interaction with Nlgn1–3s, in contrast to previous data (27). Original SDS gels and input control of fusion proteins for C are shown in Fig. S3.

To test which residues contribute to the Ca²⁺ dependency of the complex, we generated mutations at the Ca²⁺ site of LNS domains fused to human IgG-Fc. We mutated β-Nrxn D137 and N238 to alanine (short: βD137A and βN238A) and arginine (βD137R and βN238R), thereby disturbing side chains responsible for Ca²⁺ coordination in laminins (26). Because also α-Nrxn may bind to Nlgn (14), we simultaneously introduced the equivalent mutations into the isolated sixth LNS domain of α-Nrxn (Nrxn 1αLNS6; Fig. S2) to generate fusion proteins that lack the 38 N-terminal residues specific for Nrxn 1β (Fig. S4) (9). Both wild-type β-Nrxn (Fig. 1C, lane 3) and α-Nrxn (below, Fig. 2 and Fig. S5) LNS constructs precipitated Nlgns from mouse brain lysate, suggesting that the β-Nrxn-specific residues are dispensable for this interaction (but see refs. 13 and 14 for a different view). In contrast, the alanine mutations of the acidic Ca²⁺ binding residues in Nrxn 1αLNS6 and Nrxn 1βLNS failed to pull down Nlgn1 (Fig. 1C, lanes 4–7), as well as Nlgn2 and Nlgn3 (Fig. 1D, lanes 5 and 6). The complete loss of Nlgn binding for αD1183A, βD137A, αN1284A, and βN238A as well as for the respective arginine mutations (Table S1) identifies the Ca²⁺ binding site at the rim of the LNS fold as essential for binding to all Nlgns. These data are consistent with a cell culture study in which loss of synapse-inducing activity was observed for βD137A and βN238A (27). In that study, the Nrxn mutation βT235A also showed reduced clustering and preferentially labeled the surface of cells expressing Nlgn2. However, in our biochemical binding experiments βT235A and αT1281A pull down all Nlgn isoforms comparable to wild-type levels (Fig. 1 C and D, lane 3) and fail to discriminate between Nlgn1 (plus insert B) (Fig. 1C, lanes 8 and 9) and Nlgn2 (minus B) (Fig. 1D Upper, lane 7). In contrast, the presence of splice insert 4 in Nrxn (+SS4) shows reduced binding not only to Nlgn1 (Fig. S3), which confirms earlier results (12), but also to Nlgn2 and 3 (Fig. 1D, lane 4), shown here for the first time.

Fig. 2.

Low Ca²⁺ affinity is sufficient for Nrxn/Nlgn complex formation. (A) Electrostatic surface properties of Nrxn 1αLNS2 domain. (B) Binding of Nlgn to Nrxn 1αLNS6 (Upper) and binding of neurexophilin to Nrxn 1αLNS2 (Lower) are mutually exclusive. (C) Ca²⁺ coordination of LNS2 reconstructed in Nrxn 1βLNS by mutating βN238 to glycine. (D) The βN238G mutation binds recombinant Nlgn at a [Ca²⁺] of 2 mM but not at lower concentrations. The exposure time of lanes 2–4 in D was 10 min, and the exposure time was 1 min for all other lanes. Input controls of recombinant proteins are shown in Fig. S5.

Ca²⁺ Dependence of Nrxn/Nlgn Complex Formation.

To explore whether Nlgn binding of Nrxn is defined by calcium affinity, we made use of the LNS2 domain of Nrxn 1α (Fig. 2A), which coordinates Ca²⁺ at a position identical to that of LNS6 (21) and also contains a splice site (SS2). However, the two domains differ in that LNS2 is not able to bind Nlgn (Fig. 2B Upper, lane 4), and the Ca²⁺ site shows a variation: whereas αD329, αL346, and αM414 are conserved, the residue corresponding to αN1284 in LNS6 (or D3055 in laminin 2α; Fig. S2D) is replaced by a glycine (Fig. S2E). The alterability in Ca²⁺ affinity of isolated LNS2 by splice inserts (21) led to the suggestion that alternative splicing of αLNS6/βLNS may, in analogy, also reduce Ca²⁺ affinity of these domains below a threshold required for Nlgn binding (21, 28). To directly test the idea if the 10-fold reduction in complex formation caused by insert SS4 in LNS6 (Fig. S6B) depends on Ca²⁺ affinity, we mimicked the Ca²⁺ site of LNS2 in Nrxn 1αLNS6 and Nrxn 1βLNS domains by introducing mutations βN238G (Fig. 2C) and αN1284G (data not shown). In contrast to the alanine mutations of αN1284 and βN238 shown above, the glycine mutations surprisingly sustain Nlgn binding (Fig. 2D, lane 4, and data not shown). However, complex formation of βN238G/Nlgn1 requires a [Ca²⁺] of 2 mM, which is 200-fold higher compared with wild-type βLNS(-SS4)/Nlgn1 (Fig. 6A) and yet 20-fold higher than for βLNS(+SS4)/Nlgn1 (Fig. 6B). This is an important result because it reveals that the threshold for complex formation is far below normal [Ca²⁺] in the extracellular space and that factors other than Ca²⁺ affinity have to specify the exclusive binding of αLNS6/βLNS (but not LNS2) to Nlgn.

Fig. 6.

Hot spots of the Nrxn/Nlgn complex interface. (A) Summary of experimentally defined essential residues on Nrxn (Left) and Nlgn (Right) surfaces. Hydrophobic side chains of β-Nrxn G155, L234, I236, and R109 (magenta) surround the Ca²⁺ coordination site (yellow) and mediate binding to Nlgn. Other residues in this area have no effect (cyan). In Nlgn, D271 near splice inserts A and B and loop L34 (circles) is critical for Nrxn binding (red). A G500A mutation prevents complex formation at least in Nlgn1 (17), whereas L399 and N400 only show synergistic effects (16) (in pink), and many other residues do not affect binding (cyan). (B) Summary of calculated hot spots as determined by an alanine scan (see Tables S3 and S4 for details). Three Nrxn hot spots are identified, i.e., I236 (magenta), L234, and L135 (both orange). On Nlgn, only G500 (pink) and L399 (orange) are calculated as hot spots, whereas residues F499 and N400 (green) have an intermediate effect and G396 and E397 only marginally reduce complex stability (light green). Calculated ΔΔG values are rainbow color-coded from 0 (cyan) to 2 (magenta) kcal/mol. The Nrxn/Nlgn complex has been opened by 180° to expose the interface, and positions of Ca²⁺ ion spheres (1 and 2) are labeled.

Hydrophobic Residues Required for the Nrxn/Nlgn Complex.

Our analysis of surface charges indicated that the structure surrounding the Ca²⁺ site in αLNS6/βLNS (mostly hydrophobic) is distinct from LNS2 (mostly negatively charged), but these surface properties have not yet been probed experimentally. To distinguish the effect of candidate residues on shape or electrostatic surface properties at the contact area, we used site-directed mutagenesis to probe for Nlgn binding and a calcium-45 overlay assay to test integrity of Ca²⁺ coordination. First, we addressed the role of L1280 and I1282 because these residues differ in LNS domains of laminin or agrin. We inserted mutations αL1280S+I1282S+N1284D (Fig. 3A and green in C; and Fig. S7) and βL234S+I236S+N238D into αLNS6 and βLNS domains, respectively. These triple mutations mimic the Ca²⁺ site of laminin 2α LNS5 but remove the hydrophobic ring characteristic for Nrxn. We found that complex formation with Nlgn was completely abolished (Fig. 3A Upper, lanes 4 and 5), although Ca²⁺ coordination itself was preserved (Fig. 3A Lower). These data suggest that the hydrophobic residues mediate a direct interaction with Nlgn. To validate this hypothesis, we introduced a spacer into the interface by mutating residues αG1201 and βG155 to valine. These changes were introduced together with αT1202A and βT156A because the bulky valine may sterically interfere with the methyl group of threonine and alter the loop conformation (Fig. 3B and red in C). Both double mutations lacked the ability to bind Nlgn but were able to coordinate Ca²⁺ (Fig. 3B, lanes 4 and 5), suggesting that contact between Nrxn and Nlgn is very close in this area and that Ca²⁺ is enclosed within the interface. These mutagenesis data are consistent with recent cocrystals of the complex (15–17) and extend the structural analysis by demonstrating not only that they are part of the contact site but that their hydrophobicity is essential. Furthermore, we determined that the contact area on Nrxn essential for Nlgn1 binding is similarly required for the binding to Nlgn2 and Nlgn3 (Fig. 3D, lanes 4 and 5), suggesting that the same hydrophobic residues of Nrxn are required for all Nrxn/Nlgn complexes.

Fig. 3.

Nlgn binding requires hydrophobic residues in Nrxn. (A Upper) Nlgn from brain lysate is pulled down by wild-type Nrxn 1αLNS6 domain but not by mutations αL1280S+I1282S+N1284D or βL234S+I236S+N238D, respectively. (A Lower) Calcium-45 overlay dot blots demonstrate that Ca²⁺ binding itself is maintained in mutations. (B) An artificially extended surface blocks complex formation by mutating αG1201V+T1202A or βG155V+T156A (Upper) but leaves Ca²⁺ coordination unscathed (Lower). (C) A laminin Ca²⁺ binding site was mimicked in Nrxn by replacing hydrophobic residues L1280 and I1282 with serines and substituting N1284 by aspartate (green). Alternatively, the interface was extended by mutating αG1201V+T1202A (red). (D) As for Nlgn1 (A and B), mutations βG155V+T156A and βL234S+I236S+N238D completely block binding to Nlgn2 and Nlgn3 (lanes 4 and 5). Input control of recombinant proteins are shown in Fig. S7.

To control that the recombinant Nrxn LNS proteins used in our study were also able to bind to membrane-bound Nlgn, heterologous cells expressing full-length Nlgn with a GFP tag were incubated with fusion proteins. For example, wild-type and βT235A domains outlined membrane-bound Nlgn, whereas no surface labeling was observed with βG155V+T156A (Fig. S8).

Alternative Splicing in the Nrxn/Nlgn Complex.

Splicing at SS4 in Nrxn and site B in Nlgn was suggested as an important determinant in complex formation (12); however, previous studies have reported conflicting results because different assays and constructs were used. Because we wanted to solve the issue of how strongly splicing affects complex formation compared with the contact site mutations described above, we analyzed all relevant splice combinations under comparable conditions. We determined the binding of Nlgn containing the 9-aa-long B insert (Nlgn1+B) to βLNS without (βLNS-SS4) and with the 30-aa-long splice insert 4 (βLNS+SS4). Whereas Nlgn1+B binds to βLNS-SS4 at 10 μM [Ca²⁺] (Fig. S6A), βLNS+SS4 requires a 10-fold-higher concentration (Fig. S6B). In the reverse experiment, we observed that βLNS-SS4 and βLNS+SS4 precipitated Nlgn from brain lysate, but to a lesser extent if the +SS4 insert was present (Fig. 6C). To explore how splicing in Nlgn affects binding to α- and β-Nrxns, we expressed Nlgn1+B and Nlgn1-B (lacking the 9-aa insert) and compared binding to different Nrxn variants. Nrxn αLNS6+SS4 (Fig. S6D) and βLNS+SS4 (data not shown) were found to bind to both Nlgn1+B and Nlgn1-B, indicating that, in contrast to mutations of Ca²⁺ binding and contact residues (Figs. 1–3), no splice combination completely prevents complex formation. However, when we used Nlgn to pull down full-length α-Nrxn, we observed binding of Nlgn1-B to Nrxn 1α+SS4 (Fig. 4A, lane 4) but not of Nlgn1+B (Fig. 4A, lane 3), consistent with an earlier report (14). To define the additional factor in full-length α-Nrxn that prevents complex formation, we generated IgG fusion constructs of the last extracellular cassette of Nrxn 1α containing the splice insert (αLNS5_E_LNS6+SS4). Similar to full-length α-Nrxn, this construct showed no detectable binding to Nlgn1+B (Fig. 4B, lane 4), suggesting that the complete cassette determines a three-dimensional structure unfavorable of complex formation. Because EGF structures contain cysteine bonds (29), we tested whether reducing conditions could change the binding behavior of αLNS5_E_LNS6. In fact, binding to Nlgn1+B could be restored by addition of DTT (Fig. 4B, lanes 4 and 7), whereas binding of Nlgn1+B to an isolated αLNS6 domain as control was unaffected by DTT (Fig. 4B, lanes 3 and 5). Our results suggest that the preference of full-length Nrxn1α+SS4 for Nlgn1-B (14) may be determined by an EGF-tightened domain arrangement in the αLNS5_E_LNS6 cassette, allowing the first model prediction of their spatial arrangement (Fig. S9).

Fig. 4.

EGF3 domain of αNrxn+SS4 inhibits binding to Nlgn+B. (A) Full-length αNrxn+SS4 binds to IgG-Nlgn1-B but not to Nlgn1+B under identical conditions. (B) αLNS5-EGF3-LNS6 does not bind Nlgn1+B under nonreducing conditions (−DTT), whereas addition of DTT restores binding. As control, DTT does not change Nlgn binding of single αLNS6 domains. An extended set of pull-down experiments with more splice combinations is shown in Fig. S6.

Nlgn Interface Residues and Complex Formation.

Because mostly Ca²⁺ binding and hydrophobic residues in Nrxn are required for Nlgn binding, we aimed at identifying the relevant residues in Nlgn. First, we probed the degenerate EF hand (30) for its contribution to complex formation, but deletion (Nlgn1+BΔEF1) did not alter binding to βLNS (Fig. S10), thereby validating recent Nlgn crystal structures that did not detect Ca²⁺ bound to this motif (15–17, 31). Next, we selected four presumptive interface sites in Nlgn that match the criteria analyzed above, i.e., (i) the +SS4 and +B inserts may interfere to reduce (but not abolish) binding, (ii) the contact site is likely hydrophobic, and (iii) it should contain a residue capable of joining the Ca²⁺ coordination of Nrxn. We tested candidate sites by mutating key residues: First, A557 and V558 build a hydrophobic site oriented toward the synaptic cleft (19). However, aspartate and arginine mutations that mimic other cholinesterases did not change binding to Nrxn (Fig. S10). Second, a proline-rich sequence containing E99 that binds Zn in a derivatized acetylcholinesterase crystal (PDB ID code 1FSS) prompted us to mutate P100 to serine. The change should sterically block a close contact if present but failed to alter complex formation (Fig. S11A, lane 3). Third, close to splice site B, Nlgn contains acidic Q395 and E397 that lie next to a hydrophobic residue (L399). We mutated L399 to serine to test the importance of surface charge in that area and changed H294 to valine to probe for contribution of a nearby histidine as in Gas6 (32). None of these mutations, however, altered binding (Fig. 5A, lanes 6 and 9), a surprising finding because a recent cocrystal of the complex suggested that residues L399 and H294 may directly contact Nrxn (15), emphasizing the need for mutagenesis studies as performed here (Table S2). Finally, we focused on the area around D271 because it sits near the three major Nlgn loops and is embedded in a hydrophobic environment (P192, G266, and L273). Changing D271 to arginine resulted in a properly processed protein based on evidence that it reaches the surface of tsA201 cells (Fig. 5D) and an IgG fusion protein of D271R is produced properly by COS-7 (data not shown). Importantly, the Nlgn D271R mutation completely blocked binding to Nrxn (Fig. 5A, lane 3). When we tested the nearby residues with P192A, G266D, L273E, and double mutation G266D+L273D, however, none of these mutations abolished Nrxn/Nlgn complex formation (Fig. S11 A, lanes 6 and 9, B, lanes 3 and 9, and C, lane 3), raising the possibility that the interface on Nlgn may be less distinct as compared with Nrxn.

Fig. 5.

Residue D271 of Nlgn1 is essentially required for binding Nrxn. (A) Pull-down experiments of mutated recombinant Nlgn1 by βLNS-IgGFc fusion protein and IgGFc protein as negative control. (B–D) TsA201 cells expressing wild-type Nlgn1-B (B) or mutant G266D (C) and D271R (D) were surface-labeled with anti-Nlgn antibody 4C12 (45). (Scale bar: 5 μm.) Pull-down of additional mutations and input control of recombinant proteins are shown in Fig. S11.

Because of available data from cocrystal structures of the Nrxn/Nlgn complex (15–17, 31), the contribution of suspected residues to complex stability can be calculated and compared with mutagenesis analyses (Fig. 6). We estimated binding energies for wild-type and mutant complexes to assess the effect of hot spot residues (Tables S3 and S4). In Nrxn, the binding energies for mutants S107R, L135R, G155V+T156A, L234S+I236S+N238D, I236R, and N238R have positive ΔG values, indicating no complex formation, in agreement with experimental data (Fig. 6A). Interestingly, Nrxn mutations D137A and D137R do not destabilize the complex interface according to calculations, supporting its exclusive function in Ca²⁺ coordination, whereas N238A only slightly destabilized the complex. Calculated binding energies for all available alanine mutations (Table S4) are schematically summarized in Fig. 6B: only the Nrxn mutations of hydrophobic residues L135 (ΔΔG = 1 kcal/mol), L234 (ΔΔG = 1 kcal/mol), and I236 (ΔΔG = 2 kcal/mol) resulted in significant destabilization of the complex but no alanine mutation of polar residues such as S107 or R232. In Nlgn, none of the mutations gave positive ΔG values, including the quintuple mutation L399A+N400A+D402N+Q395A+E397A (16), and only the Nlgn residues L399 and G500 account as moderate hot spots according to energy terms (Fig. 6B and Table S4). However, our L399S mutation did bind Nrxn experimentally (Figs. 5C and 6A), and Nlgn2 (carrying a Q for the nonconserved G500) also formed a complex with Nrxn (Fig. 1F). Therefore, mutagenesis and calculated data together indicate that the interface in Nrxn is sharply delineated, whereas the binding site in Nlgn appears less well defined or even has alternative positions.

Discussion

Using mutagenesis, we identified single residues that determine the complex between Nrxn and Nlgn. All mutations tested thus far are summarized in Tables S1 and S2 and are displayed in Fig. 6A: essential residues of Nrxn (D137, G155, L234, I236, and N238 for β-Nrxn) and Nlgn (D271) showed an all-or-nothing effect in our binding assay (Figs. 1, 3, and 5). Analysis of Ca²⁺ binding (Fig. 3) and calculation of energy terms using available crystal data (Fig. 6B) revealed that essential residues contribute to Ca²⁺ coordination, to the hydrophobic interface, or to both.

Ca²⁺ Coordination.

Our comparison of LNS domains (21, 26, 32–37) revealed a conserved core structure (Fig. S1). Ca²⁺ binding in Nrxn (by D137, V154, I236 and N238 for β-Nrxn), agrin, and laminins represents a rigid “socket” with no flexible states as in EF hands (23), consistent with unchanged CD spectra of Nrxn and agrin (13, 38) and crystal data (15–17, 39). This rigidity allowed us to transfer epitopes from other LNS domains to Nrxn: βN238G mimics the Ca²⁺ coordination of Nrxn LNS2 but reduced Ca²⁺ affinity of the Nrxn/Nlgn complex ≈200-fold (Fig. 2). Because the complex still forms with this mutation at physiological [Ca²⁺], it appears unlikely that calcium affinity itself defines splice-code-specific pairings (21). Rather, crystal and NMR data revealed that insert SS4 exists in an equilibrium of two conformations (39, 40). The inactive form is stabilized by coordinating Ca²⁺ with high affinity (39), whereas binding follows after unfolding of SS4 (40). Therefore, the 50% reduction in complex formation when SS4 is present could sufficiently be explained by a reduced availability of the active form. In this interpretation, calcium may have a double function for Nrxn/Nlgn: first, it is required for complex formation, but, second, it slows down the process by keeping approximately half of the splice insert carrying variants, if present, in a reserve pool. Similarly, the additional presence of the third EGF domain as in full-length αNrxn+SS4 may lead to a spatial arrangement that makes binding to splice insert B-positive Nlgns even less likely (Fig. 4 and Figs. S6 and S9).

Binding Residues.

Crystal data revealed that the Nrxn/Nlgn1 interface may involve (i) four side-chain to main-chain H-bridges, (ii) two salt bridges, and (iii) 19 nonbinding interactions. Hydrogen bridges include one from βNrxn (βN238 to G396) and three from Nlgn (E397 to βT235 and βI236; N400 to βS107). In addition, βR232/D387 and βR109/E297 form salt bridges (15). Consistently, mutations of βS107A and βT235A do not alter its main-chain interaction and sustain complex formation (Fig. 1 and Table S4) (27). Similarly, no single mutation removing an H-bridge or salt bridge abolishes complex formation: Nlgn E297A, D387A, E397A, and N400A only slightly reduce binding (Table S3), and Nrxn R232A remained functional (27). Interestingly, residue βN238 prevents complex formation when changed to A (Fig. 1) but tolerates mutations to G (Fig. 2) or D (Fig. 3), indicating that it is essential for calcium coordination, but its side-chain interaction to G396 (15, 16) is dispensable. Another complex interaction profile is highlighted by βR109A, which showed no synaptogenic activity (27) but sustains complex stability based on calculated binding energies (Fig. 6B). Consistently, single mutations of Nlgn residues E297, H294, and L399 that are in binding distance to βR109 did not prevent complex formation (Fig. 5) (28), supporting a model of dynamic binding of Nrxn R109 with all three residues (15).

The hydrophobic ring of Nrxn (Fig. 1B) is expanded by Nlgn L399 and F499 to a hydrophobic shell that encloses the Ca²⁺ (15–17). The importance of this shell is highlighted by our triple mutation βL234S+I236S+N238D, which specifically reduces the hydrophobicity of the surface but does not introduce a steric hindrance (as imposed by βL135R, βL234R, βI236R, and N238R; Table S3 and Fig. S12) to prevent complex formation (Fig. 3) (15). Therefore, we propose that hot-spot residues on the Nrxn surface do coordinate Ca²⁺ (βD137 and βN238), make hydrophobic contacts (βG155, βL234, and βI236), or both (βI236)—while none of the H-bridges are essential. In addition, our results reveal that the hot-spot residues identified here are also required for binding to Nlgn2 and Nlgn3 (Figs. 1 and 3) and, presumably, for the structurally identical complex of Nrxn/Nlgn4 (17).

In Nlgns, mutagenesis studies are more difficult because several previous changes were retained in the ER (41–43); therefore, we tested each new mutation for its ability to reach the cell surface (Fig. 5, Fig. S9, and data not shown). Of all correctly delivered mutations, only D271R resulted in complete loss of Nrxn binding in our experiments (Figs. 5 and 6). Because D271 lies 12 Å apart from Nrxn D104 (15–17), the arginine introduced at position 271 may initiate a repelling reaction on R274 and R311, of which the latter is part of a presumed second Ca²⁺ site (16).

According to crystal data, Nlgn possesses up to four more candidate residues at the interface (G500, L399, N400, and E397) (15–17), but, based on calculated binding energies, only mutations of L399 and G500 could possibly destabilize the complex (Fig. 6B). Because L399S did not prevent Nrxn binding to Nlgn in our study (Fig. 5), G500 remains as a putatively important residue. However, Nlgn2 differs from Nlgn1 by a G500Q exchange calculated to sterically block Nrxn/Nlgn complex formation (Table S3), suggesting that G500 may not play a general role. In summary, we conclude that, whereas the binding interface on Nrxn is sharply delineated and conserved for the binding to presumably all Nlgn isoforms, the Nlgn interface tolerates more mutations and may even contain an alternative hot-spot area in Nlgn2.

Materials and Methods

Original cDNAs used for site-directed mutagenesis have been described (4, 11, 13, 44). Oligonucleotide primers and procedures for structural modeling, pull-down experiments, live cell labeling, and calcium-45 overlays are based on published protocols (13, 14) and are described in detail in SI Materials and Methods.