Article

Confirmation of the functional unit of the NL1–NX1 complex

To reconstitute the NL1–NX1 complex, we expressed the NX1 LNS domain and the NL1 CLD using baculovirus. Crystals of the complex were obtained with NL1 and NX1 combined in the presence of Ca²⁺. The structure was determined by molecular replacement. The asymmetric unit contained two NX1 and two NL1 molecules. Because each NL1 was involved in crystal packing with four neighboring NX1 molecules, it was unclear which pairs of NL1 and NX1 are the functional units. The interaction between neuroligins and neurexins has not been systematically mapped, despite existing mutagenesis data on both NL1 and NX1^{20, 25}. To prevent the misidentification of packing artifacts as functional interfaces, we studied all four NL1-NX1 crystal contacts, which bury surface areas of 1,160 Ų, 1,030 Ų, 560 Ų and 140 Ų, respectively (Fig. 1 and Supplementary Fig. 1 online). We tested the two larger interfaces (1 and 2) by designing five NX1 mutants (S107R, L135R, D137S, I236R and N238R) within interface 1 and two NX1 mutants (R112E and T179A) within interface 2, and we compared the NL1-binding capacities of the mutants and wild-type NX1. His-tag pull-down experiments (Fig. 1a) indicated that only the interface-1 mutations reduced NL1 binding. To further evaluate the importance of interface 1, we tested mutations (E397A and N400A) of two NL1 residues central to interface 1 and involved in many NL1-NX1 interactions. These NL1 mutants did not show detectable binding to wild-type NX1 in calorimetric experiments (Supplementary Fig. 2 online). The mutagenesis data therefore indicated that interface 1 (see the buried interfaces in Fig. 1) was physiologically relevant.

Figure 1: Structure of the NL1–NX1 complex.

(a) His-tag pull-down assay showing effects of mutagenesis of NX1 surface residues on NL1 binding. Left, Coomassie-stained SDS-PAGE gel, with upper bands being His-tagged NL1 attached to Ni-NTA resin, and lower bands being non-His-tagged NX1 mutants retained by NL1. Lane assignments are indicated in the corresponding table (right), along with the NL1-binding capacities of the mutants compared with the wild type. (b) The 2:2 NL1–NX1 complex in a ribbon model, viewed from the NX1 membrane. NL1, orange and pink; NX1, blue and cyan; Ca²⁺ ions, red balls. (c) The complex viewed from the side.

Full size image (96 KB)

Structure of the NL1–NX1 complex

The NL1–NX1 complex contained a central NL1 dimer and two NX1 monomers, one attached to each side (Fig. 1b,c). NL1 dimerization ensures NX1 clustering. The NL1 monomer contained a 550–amino acid, globularly shaped CLD, with a central 11-strand -sheet flanked by -helices. The NL1 dimer in the complex, as expected from its 30% sequence identity to AChE²⁸, could be superimposed upon the mouse AChE dimer (r.m.s. deviation 1.4 Å for C atoms) (Supplementary Fig. 3 online). The dimer interface of NL1, as in AChE, was formed by four -helices near the C termini, with two helices (residues 450–460 and 620–635) from each monomer. The dimer interface was large, burying 1,590 Ų of surface area, and was primarily hydrophobic. With the Met459-Met459 and Leu633-Leu633 contacts from both chains as the dividing line, the interface appeared as a duplex of two equivalent subinterfaces. Each subinterface was marked by a central Phe458 surrounded by Trp463 and Met459 from its own monomer and Met459, Leu625 and Leu629 from the opposing monomer (Supplementary Fig. 3). The NX1 LNS domain was -sandwich-like, as described¹⁸. Each NX1 used a loop-rich surface lying on the side of the -sandwich, previously designated as the 'hypervariable surface'¹⁸, to contact NL1, whereas NL1 used the region around the exposed side of the central -sheet to contact NX1. With respect to the orientation of NX1 relative to NL1, the -sheets of NX1 and the long axis of NL1 dimer were vertical, but not, as suggested in previous small-angle scattering studies²¹, parallel to each other.

The Ca²⁺ binding site is in NX1 but also relies on NL1

After refining the protein part of the structure, we searched the asymmetric unit for Ca²⁺. Initially, our effort was compromised by the requirement for sodium citrate, a known chelating reagent, at high concentration (1 M) for crystallization. Nevertheless, the SIGMA-weighted composite omit 2F_o – F_c map and the protein-only F_o – F_c electron density map (Supplementary Fig. 4 online) showed a unique site in each pair of NL1 and NX1, surrounded by four NX1 oxygen atoms. The electron density for the site was modest. We speculated that it might be a Ca²⁺ ion present at partial occupancy. Although the total Ca²⁺ concentration was 10 mM during crystallization, free Ca²⁺ was estimated to be only 10^{- 2} mM because of citrate chelation. We therefore speculated that citrate chelation competed with the Ca²⁺ binding site, resulting in the low occupancy. To verify our speculations, we replaced citrate with acetate in the mother liquor before measuring X-ray diffraction data and obtained a data set at a resolution of 3.5 Å. After simple rigid-body refinement to accommodate cell-dimension differences, we calculated electron density maps using the data from the acetate-adapted crystal. Strong electron density at this position unambiguously indicated that it was a Ca²⁺ binding site (Fig. 2a and Supplementary Fig. 4).

Figure 2: The Ca²⁺ binding site in the NL1-NX1 complex.

(a) Coordination of the Ca²⁺ and its effect on NL1-NX1 interaction. NX1, cyan; NL1, pink. Ca²⁺ and water molecules, large and small red balls, respectively. Hydrogen bonds and Ca²⁺-oxygen interactions, dashed red lines. F_o – F_c electron density calculated with the Ca²⁺ and water molecules omitted from the model is shown contoured at 4.5. The Ca²⁺ ion is octahedrally coordinated by four protein oxygen atoms from NX1 and two water molecules. WAT1 is bonded to both the Ca²⁺ and two NL1 oxygen atoms. The NX1 Asn238 side chain is bound to both the Ca²⁺ and the NL1 main chain. (b) Overlaying the structures of the Ca²⁺ bound to NX1 (Ca²⁺, red ball; NX1, cyan sticks) with the Ca²⁺ bound to NX1-LNS2 (both Ca²⁺ and NX1 colored orange) shows that the Ca²⁺ binding sites are in similar positions with respect to the LNS domain. The main difference is that the positioning of WAT1 in NX1 requires NL1, but the equivalent water in the NX1-LNS2 structure can be positioned by the side chain of Tyr412.

Full size image (62 KB)

The bound Ca²⁺ had direct contacts only with NX1, and not with NL1 as suggested elsewhere²². The Ca²⁺ binding site in NX1 was located at the hypervariable surface. Coordination of the bound Ca²⁺ was consistent with an octahedral geometry (n = 6) (Fig. 2a), with two main chain carbonyl oxygen atoms (Val154 and Ile236), two side chain oxygen atoms (Asp137 and Asn238) from NX1, and two water molecules ligated to the Ca²⁺. Although NX1 provided all the non-solvent oxygen atoms for the bound Ca²⁺, we did not detect binding between Ca²⁺ and NX1 in the absence of NL1 using calorimetry (Supplementary Fig. 5 online), consistent with previous observations that Ca²⁺ incubation does not induce structural change of NX1²². Thus, Ca²⁺ binding seems to require the presence of NL1. In the NL1–NX1 structure, one of the water molecules in Ca²⁺ coordination, WAT1, was hydrogen bonded to the main chain carbonyl of Gln395 and the side chain carboxyl of Glu397 from NL1, suggesting that NL1 acts in Ca²⁺ binding by coordinating this water molecule for the completion of the Ca²⁺ binding site.

Superimposition of the structure of NX1 in our complex with the structure of NX1-LNS2 (ref. 29) showed that their Ca²⁺ ions were bound at similar positions (Fig. 2b). Most structural elements for Ca²⁺ coordination were conserved, except that one water molecule in NX1-LNS2 was replaced with the amide carbonyl from NX1 Asn238. Because the NX1-LNS2 Ca²⁺ binding site includes more water molecules, we find it surprising that NX1-LNS2 alone binds Ca²⁺ at measurable affinity (400 M)²⁹, but NX1 alone does not show detectable Ca²⁺ binding. One explanation may be the differences in the positionings of these water molecules, especially the one at the NX1 WAT1 position (Fig. 2b).

The interface between NL1 and NX1

The NL1-NX1 interface contained a protruding surface patch at the edge of central -sheet of NL1 and part of the hypervariable surface of NX1 (Fig. 3). The majority of the interface on the NL1 side emanated from loop 395–402, a smaller fraction from the short, flanking loop 499–502 and a minor fraction from side chains extending from loop 292–310 and loop 380–388. The NX1 side of the interface consisted of two long, parallel loops (101–110 and 231–241) and a short loop (131–137). The 1,160-Ų total buried solvent-accessible surface area between each pair of NX1 and NL1 molecules was smaller than the average of 1,600 400 Ų expected for protein-protein interfaces³⁰. The interface was primarily composed of hydrophilic residues. The only interactions between hydrophobic residues were between NX1 Leu135 and NL1 Phe499 and between NX1 Ile236 and NL1 Leu399. A total of six direct hydrogen bonds were formed between NL1 and NX1, mostly contributed by only two residues, Glu397 and Asn400, from NL1 (Fig. 3). Mutation of either residue to alanine (E397A or N400A) abolished NL1-NX1 binding, confirming their essential roles (Supplementary Fig. 2). At the edge of the interface, two salt bridges were formed, between NL1 Asp387 and NX1 Arg232 and between NL1 Glu297 and NX1 Arg109. Notably, these two salt bridges were not grouped with other interactions and therefore are not reinforced by the adjacent structural elements. Water molecules filled in the unoccupied space between the opposing surfaces of NX1 and NL1, with some forming hydrogen bonds with both molecules. The hydrophilic and water-supplemented nature of the NL1-NX1 interface is consistent with previous studies showing that NL1-NX1 binding affinity is inversely related to the ionic strength of the solution¹⁹.

Figure 3: The NL1-NX1 interface.

(a) Side view. NL1 and NX1 backbones are shown as coils. The side chains of the important residues in the NL1-NX1 interaction are shown as sticks. Water molecules and Ca²⁺, red balls. NX1 splice site 4 (cyan square) lies near the salt bridge between NL1 Asp387 and NX1 Arg232. (b) View from the NL1 direction. NL1 splice site B, (pink square) lies near the salt bridge between NL1 Glu297 and NX1 Arg109.

Full size image (58 KB)

Specificity of neuroligin-neurexin interactions

The ranking order of neuroligins toward binding NX1 is NL1 > NL4 NL3 > NL2 (ref. 20). Sequence alignment of neuroligins showed that their central neurexin-binding loops (NL1 395–402) were identical (Supplementary Fig. 6 online), and the surrounding small loops bore differences. NL1 and NL4 differ only slightly (NL1 Tyr499 versus NL4 Phe463), explaining their close neurexin-binding affinities. The weaker NX1-binding affinity of NL3 can be attributed to the Tyr His substitution at the NL1 Tyr295 position, where a hydrophobic side chain is required for positioning the 294–297 loop for contacting NX1. NL2 differs markedly from other neuroligins by a glycine glutamine substitution (NL1 Gly500 versus NL2 Gln475). In the NL1–NX1 complex, NL1 Gly500 faced the NX1 surface closely and could not be substituted by any residue with a side chain, as the side chain would conflict with NX1 Ser239. Hence, structural rearrangement would be required in NL2 to accommodate Gln475 when binding neurexins. This would be unfavorable, because the flanking residues interacted extensively with NX1 in the complex. On the basis of our structure, therefore, NL2 is not an optimal ligand for neurexins.

Effects of splice sites on NL1-NX1 interaction

Of the three alternative spice sites harbored in the complex, the site A and site B inserts of NL1 reduce NL1-NX1 binding by two- to five-fold²⁰, whereas the SS 4 insert in neurexins nearly blocks the neuroligin-neurexin association^{26, 27}. Our structure indicated that site A of NL1 was more than 25 Å from the edge of the NL1-NX1 interface, and it is therefore unlikely to interfere with the neuroligin-neurexin association (Fig. 4a). In comparison, site B of NL1 was only 10 Å from the edge of the interface and so it is more likely to interfere with NL1-NX1 binding.

Figure 4: Alternative splice sites in the NL1–NX1 complex.

(a) The locations of the NL1 sites, A and B, and NX1 SS 4 on the surface of the complex. Each residue immediately adjacent to the potential insert is colored green and shown in space-filling representation. (b) Calorimetric titration between NX1 and NL1 without the site B insert. (c) Calorimetric titration between NX1 and NL1 with the site B insert. The NL1 protein does not have the site A insert, and the NX1 protein does not have the SS 4 insert.

Full size image (84 KB)

To determine how the site-B insert of NL1 produces its inhibitory effect, we used isothermal titration calorimetry to compare the thermodynamic profiles of NL1, with and without the site-B insert (NL1-A versus NL1-AB), in binding to NX1 (Fig. 4b,c). The measured affinities for NL1-A/NX1 and NL1-AB/NX1 were 430 nM and 93 nM, respectively. Both binding events were enthalpy-driven and were entropically unfavorable. The H value of approximately - 20 kcal mol^{- 1} agreed with the number of salt bridges and hydrogen bonds in the complex. Comparing the two thermodynamic profiles showed that the site-B insert produced an enthalpic loss (H difference 1.4 kcal mol^{- 1}) but an entropic gain (TS difference 0.5 kcal mol^{- 1}). There are two possible ways that the site-B insert, which was near the NL1-NX1 interface, could exert its negative influence on binding. One possibility is steric hindrance: bringing the two proteins together could limit the flexibility of the insert, resulting an entropic loss. Such a possibility does not agree with our thermodynamic data, which showed that the site-B insert brought an entropic gain. The other possibility is that the site B insert disturbs or weakens an existing hydrogen bond or salt bridge. This possibility is consistent with the thermodynamic data: the enthalpic loss would be due to bond breaking, and the associated entropic gain would be due to increased conformational flexibility.

Consistent with the second possibility, analyzing the NL1-NX1 structure showed that the NL1 site-B insert could interfere with an adjacent, dynamic salt bridge between NL1 Glu297 and NX1 Arg109. In the structure of the complex without the NL1 site-B insert, the NX1 Arg109 side chain showed no electron density beyond C, indicating that it was flexible (Supplementary Fig. 7 online), but in its low-energy, extended conformation, both the N1 and N atoms of NX1 Arg109 were at appropriate distances from the carboxyl group of NL1 Glu297 to form a salt bridge. Neuroligin-neurexin interactions have been shown to be electrostatically driven¹⁹. Mutation of E297A (combined with a remote K306A mutation that is not likely to be relevant) reduces NL1 binding to NX1 by five-fold²⁰, supporting the existence of this salt bridge. Given the flexibility of NX1 Arg109, the correct conformation and the availability of NL1 Glu297 become important for the energetics of this salt bridge. In NL1 without the site-B insert, Glu297 is held inflexible in a very tight turn; in NL1 with the site B insert (298–306), the insert, which is next to Glu297, would probably release this tight-turn conformation and result in increased flexibility of Glu297, thus destabilizing the salt bridge with NX1 Arg109. The effect of replacing Glu297 with alanine in NL1 without insert B (ref. 20) is similar to that of adding insert B (Fig. 4); each reduces binding by about five-fold, further supporting the concept that insert B acts though modulating the NL1 Glu297–NX1 salt bridge. Notably, the inhibitory effect of the NL1 site-B insert seems to be augmented by the presence of N-linked glycosylation in the middle of the insert residues²³. This is perhaps because the N-linked glycan, with its large size, can restrain the conformation of the insert and adjacent residues, including Glu297, decreasing the possibility that NL1 Glu297 will achieve the conformation that allows salt bridge formation with NX1 Arg109.

Analysis of the structural elements around NX1 SS 4 indicated that the SS 4 insert may use a similar mechanism to that of the NL1 site-B insert to modulate NL1-neurexin recognition. This insert site is distant from the main NL1-NX1 binding interface, and is unlikely to act through steric hindrance. However, it is immediately adjacent to NX1 Arg232, which forms a salt bridge with NL1 Asp387 in the complex (Fig. 3). An insertion of 30 amino acids between Ala200 and Gly231 is likely to alter the conformation of NX1 Arg232, resulting in the loss or weakening of this salt bridge. Because incorporating the SS-4 insert reduces NL1-NX1 binding to an undetectable value, direct thermodynamic evidence is unavailable for this supposition.

Effects of the autism-related neuroligin mutations

Aberrant neuroligin genes have been linked to autism^{11, 15, 16, 31}, but it remains unclear how the point mutations of neuroligins are related to the genesis of the disease. An NL3 mutation (R451C) and three NL4 mutations (G99S, K378R and V403M), located in the CLD domain, have been found in autistic subjects^{11, 15}. These residues, corresponding to Arg473, Asp106, Lys414 and Val439 of NL1, respectively, are highly conserved, except for NL1 Asp106 (Supplementary Fig. 6). An inspection of the locations of these residues in the NL1–NX1 complex indicated that they were distant from the neuroligin-neurexin interface and the neuroligin dimer interface, and so are unlikely to affect neuroligin-neurexin binding or neuroligin dimerization (Fig. 5). All but Asp106 were largely buried beneath the protein surface in NL1. This was unexpected, in that most of these residues—in particular, Lys414 and Arg473—are large hydrophilic residues. Their buried nature indicates that most of the autism mutations in neuroligins compromise structural integrity. The mutations most likely result in insufficient folding, secretion or viability of the specific types of neuroligins, which would alter their functions at the synapse.

Figure 5: Autism-related neuroligin point mutations.

The mutated residues, NL1 Asp106 (corresponding to the NL4 G99S mutation), NL1 Lys414 (corresponding to the NL4 K378R mutation), NL1 Arg473 (corresponding to the NL3 R451C mutation) and NL1 Val439 (corresponding to the NL4 V403M mutation), are colored green and shown in space-filling representation, and are mapped on the NL1–NX1 complex (ribbons). These residues are away from the binding and dimerization interfaces, and are mostly buried.

Full size image (71 KB)

Constructs and mutagenesis.

The coding sequence of the NL1 extracellular CLD domain (residues 26–638) was cloned adjacent to an N-terminal 7-His-tag in pAcGP67A (PharMingen). The coding sequence for the NX1 LNS domain (residues 82–292), devoid of splice insert (residues 201–230), was cloned adjacent to a C-terminal 7-His-tag, also in pAcGP67A. A cleavable glutathione S-transferase (GST) fusion NX1 construct with no His-tag was prepared using pET-49 (Novagen). We used overlap extension PCR, based on pAcGP67A-NL1 and pET-49-NX1 constructs, to generate mutation in NL1 and NX1, respectively. We verified all constructs and mutations by DNA sequencing.

Protein preparation.

We prepared His-tagged NL1, its mutants and NX1 from baculovirus-infected insect cells. Recombinant baculovirus was produced by transecting sf9 cells with the pAcGP67A constructs together with linearized baculovirus DNA (BacVector-3000, EMD). After amplification, we used the viruses to infect 1–6 liters of Hi5 cells at a density of 1.8 10⁶ cells ml^{- 1} in HyQ SFX medium (HyClone). Cell cultures were grown for 66 h and were then collected by centrifugation. The proteins in the supernatant were captured by Ni-NTA resin and eluted with 200 mM imidazole pH 7.5. They were further purified with size exclusion columns equilibrated with HBS buffer (10 mM HEPES, 150 mM NaCl, pH 7.5).

Non-His-tagged NX1 and its mutants were expressed as intracellular soluble GST fusion proteins from rosetta-gami2 Escherichia coli cells (Novagen) induced at 25 °C. We purified proteins with a glutathione-agarose column and gel filtration, and cleaved them with HRV-3C protease overnight. The cleaved NX1 proteins were passed through a second glutathione-agarose column to separate uncleaved proteins, and further purified by a second gel filtration column.

Crystallization and X-ray data collection.

NL1 and NX1 proteins were concentrated to 10 mg ml^{- 1} in HBS with 5 mM CaCl₂ and combined at a 3:1 ratio (v/v) for crystallization. We obtained crystals through sitting-drop vapor diffusion. The drops contained equal volumes of the protein and the reservoir solution (1.0 M sodium citrate, 0.1 M sodium cacodylate, pH 6.5, 0.2 M NaCl, 10 mM CaCl₂). X-ray diffraction was measured at 0.9797 Å wavelength at beamlines 22-BM-C and 21-ID-D at the Advanced Photon Source, Argonne, Illinois, USA. Crystals were cryoprotected with 10% (v/v) glycerol in the mother liquor before flash-cooling to 100 K. The data were indexed, integrated and scaled with HKL2000 (ref. 37). To prevent Ca²⁺ from being chelated by citrate, we used a separate cryoprotecting solution with acetate replacing citrate and with a higher Ca²⁺ concentration (3 M sodium acetate, 0.1 M sodium cacodylate, pH 6.5, 0.1 M CaCl₂, 10% (v/v) glycerol) to gradually adapt the crystals before flash-cooling for X-ray diffraction.

Structure determination and refinement.

We determined the structure using the molecular replacement program PHASER³⁸, with mouse AChE (PDB 1N5M) and free NX1 (PDB 1C4R) as search models. We rebuilt the model of the complex using O (ref. 39) and refined it using CNS⁴⁰. In the final model, each NL1 contained two N-linked glycans (Asn109 and Asn547) with well-defined electron density; a third potential CLD glycosylation site, Asn343, was mutated to glutamine to facilitate crystallization. In the Ramachandran plot, 85.0%, 14.4%, 0.6% and 0% of the residues fell in favored, allowed, generally allowed and disallowed regions, respectively. The statistics of data collection and refinement are summarized in Table 1.

Table 1: Data collection and refinement statistics

Full table

His-tag pull-down assays.

Small columns loaded with 0.5 ml Ni-NTA agarose beads (bed volume 0.25 ml) were washed with binding buffer (20 mM HEPES, pH 7.5, 10 mM CaCl₂). We then loaded an equal amount of N-terminally His-tagged NL1 proteins (0.2 ml at 0.8 mg ml^{- 1}) onto each column and washed the columns with 2 ml binding buffer. The His-tag-free NX1 and its mutants, cleaved from GST fusion protein, were buffer-exchanged into the binding buffer with gel filtration, adjusted to equal concentrations (0.5 mg ml^{- 1}) and loaded in equal amounts (0.5 ml) onto the NL1-bound columns. Binding buffer (2 ml) was applied to wash off unbound NX1 proteins. The retained NX1 proteins, together with NL1 proteins, were then eluted off the Ni-NTA columns with 0.5 ml 0.2 M imidazole, pH 7.5 in HBS, and analyzed with SDS-PAGE.

Isothermal titration calorimetry.

We performed calorimetric titrations on a VP-ITC calorimeter (MicroCal) at 30 °C and processed the data with MicroCal Origin 5.0 software. The protein concentration was determined by the bicinchoninic acid assay with BSA as the standard (Pierce). For measuring NL1-NX1 interactions, the proteins purified from insect cells were buffer-exchanged into an identical lot of binding buffer (20 mM HEPES, pH 7.5, 10 mM CaCl₂) using gel filtration to control buffer heat dilution effects. Before each titration, protein samples were degassed for 10 min. We then added the NX1 proteins to NL1 over the course of many injections until the NL1 was fully saturated. For measuring binding interactions between NX1 and Ca²⁺, we added native NX1 to a chelating solution with a final concentration of 50 mM EDTA, pH 8.0, to remove residual Ca²⁺ potentially derived from cell culture. The protein was then buffer-exchanged to 20 mM HEPES, pH 7.5, using gel filtration. We prepared a CaCl₂ solution by dissolving CaCl₂ directly into the same 20 mM HEPES buffer and then added it to NX1 during the course of titration.

Accession codes.

Protein Data Bank: Coordinates and structural factors have been deposited with accession code 3B3Q.

Note: Supplementary information is available on the Nature Structural & Molecular Biology website.

Author Contributions

X.C., H.L., A.H.R.S. and P.J.F. carried out experiments; X.H. supervised the research; X.C. and X.H. wrote the article.