Neuroligins are postsynaptic adhesion proteins involved in the establishment of functional synapses in the central nervous system. In rodents, four genes give rise to neuroligins that function at distinct synapses, with corresponding neurotransmitter and subtype specificities. In the present study, we examined the interactions between the different neuroligins by isolating endogenous oligomeric complexes using in situ cross-linking on primary neurons. Examining hippocampal, striatal, cerebellar and spinal cord cultures, we found that neuroligins form constitutive dimers, including homomers and, most notably, neuroligin 1/3 heteromers. Additionally, we found that neuroligin monomers are specifically retained in the secretory pathway through a cellular quality control mechanism that involves the neuroligin transmembrane domain, ensuring that dimerization occurs prior to cell surface trafficking. Lastly, we identified differences in the dimerization capacity of autism-associated neuroligin mutants, and found that neuroligin 3 R471C mutants can form heterodimers with neuroligin 1. The pervasive nature of neuroligin dimerization indicates that the unit of neuroligin function is the dimer, and raises intriguing possibilities of distinct heterodimer functions, and of interactions between native and mutant neuroligins contributing to disease phenotypes.
The development of functional circuits in the central nervous system is based on the establishment of a variety of synapses between neurons. The formation of all fast transmitting synapses involves co-ordinated and specialized differentiation of pre- and post-synaptic membranes, a process driven by trans-synaptic adhesion. The synaptic adhesion system of presynaptic neurexins and postsynaptic neuroligins is one of very few systems known to function in the establishment of transmitting synapses with a wide range of specificities, encompassing glutamatergic [1,2], glycinergic [3,4], GABA (γ-aminobutyric acid)-ergic [4,5] and cholinergic  synapses.
Neuroligins derive from four genes in rodents, producing the homologous isoforms NL (neuroligin) 1–4 [7–9], which display synapse-specific functions. NL1 functions at excitatory synapses [1,2], whereas NL2 functions at inhibitory synapses [4,5,10]. NL4 is associated with glycinergic synapses in the spinal cord and retina , whereas NL3 localizes to as yet uncharacterized subsets of both excitatory and inhibitory synapses . These features have supported the notion of a direct correspondence of adhesion protein isoforms to synapse subtypes. According to the current paradigm, single neuroligin isoforms are targeted to and function at particular synapses, independently of other isoforms. However, in vitro data indicate that the extracellular domains of neuroligins interact to form oligomers. Recombinant soluble neuroligin fragments form homomeric tetramers , or homomeric dimers [13,14] in solution and in crystals [15–17]. These findings indicate the existence of in cis interactions driving neuroligins to form membrane oligomers in vivo. Although it is widely speculated that oligomerization might play a role in neuroligin function and autism pathology , endogenous oligomers have yet to be identified in neurons, leaving their native properties unaddressed.
In order to study neuroligin oligomerization, we isolated and characterized endogenous neuroligin oligomers from neurons. We examined their stoichiometry and composition in different neuronal populations, and identified endogenous isoform-specific neuroligin heteromers. We additionally identified elements of a trafficking mechanism that controls neuroligin oligomer assembly. Finally, we examined the effects of autism-associated neuroligin mutations on dimerization and trafficking. Our data provide evidence of pervasive interactions between neuroligins that can result in homomeric, mixed-isoform and mutant-containing complexes.
MATERIALS AND METHODS
Plasmids, cell culture and transfection
Myc–NL1, HA–NL2 (where HA is haemagglutinin) and HA–NL3 constructs were described previously . An NL1–GFP (green fluorescent protein) construct was produced by subcloning the rat NL1 coding sequence into pEGFP-N1. Human NL4–venus and NL4/R87W–venus constructs were generously provided by Dr Chen Zhang and Professor Thomas Südhof (Stanford University, Stanford, CA, U.S.A.) . Mutations of these constructs were generated using the QuikChange method (Stratagene). The open reading frames of all constructs were fully sequenced. Rat primary neurons were isolated and cultured from E (embryonic day) 18 hippocampus, E18 striatum and P6 cerebellum . Mouse primary neurons were isolated and cultured from E13 spinal cord and E16 hippocampus . Neurons were plated on plastic culture dishes in Neurobasal medium (Gibco) supplemented with Glutamax (Invitrogen), penicillin, streptomycin and B27 supplement (Gibco). Cultures were allowed to develop for 2 weeks in a 5% CO2 incubator at 37°C. COS7 and HEK (human embryonic kidney)-293FT cell lines were cultured in plastic culture dishes in DMEM (Dulbecco's modified Eagle's medium; Gibco) supplemented with 10% FBS (fetal bovine serum; Gibco) in a 5% CO2 incubator at 37°C. Cell lines were transiently transfected using standard lipofection procedures with FuGENE6 (Roche) or Lipofectamine 2000 (Invitrogen). Cell samples were prepared 16–24 h after transfection.
Cross-linking, surface biotinylation and deglycosylation
Cross-linking of cell-surface proteins was performed on primary neurons or transfected HEK-293FT cultures. Cells were treated with 0.1 mM of the membrane-impermeable cross-linker BS3 [bis(sulfosuccinimidyl) suberate; Pierce] in PBS for 10 min on ice. After treatment, the cross-linker was washed off and quenched using 100 mM ethanolamine. Cell lysates were prepared from cross-linker-treated cells using 1% SDS in a Ca2+-chelating Tris-buffered solution (5 mM EDTA, 50 mM Tris and 150 mM NaCl) containing protease inhibitors (1 mM leupeptin, 1 mg/ml aprotinin and 100 mM PMSF). Lysates were pressed through a syringe with a 27 gauge needle until DNA was sufficiently sheared, and subsequently centrifuged at 20000 g for 15 min. Supernatant SDS extracts were either prepared for SDS/PAGE or used as input for immunoprecipitation assays. Surface biotinylation was performed on transfected HEK-293FT cultures. Cells were treated with 0.5 mg/ml biotin (Pierce) in PBS for 30 min on ice and washed with 0.1 M glycine in PBS. Cell lysates were extracted in 1% Triton X-100 and 0.1% SDS in the above Tris-buffered solution, and centrifuged at 20000 g for 15 min. Cell extract supernatants were incubated with steptavidin-coated beads (Pierce) for over 6 h at 4°C. Beads were washed with lysis buffer and eluted directly in Laemmli buffer. The input, flow-through and bead-coupled fractions were separated on Tris-acetate 3–8% acrylamide gradient NuPAGE gels (Invitrogen).
Deglycosylation experiments were performed using the glycosidases EndoH (endoglycosidase H) and PNGase (peptide N-glycosidase) F (New England Biolabs) according to the manufacturer's protocols.
Immunoprecipitation and immunoblotting
For immunoprecipitations, SDS extracts were diluted 7-fold with 1% Triton-X 100-containing Ca2+-chelating Tris-buffered solution to allow for antigen recognition . Anti-GFP or anti-NL isoform-specific antibodies (see below) were then used to immunoprecipitate the respective NLs. Cross-linked and biotinylated samples were separated on Tris-acetate 3–8% acrylamide gradient NuPAGE gels (Invitrogen), whereas other samples alone were separated on 7.5% Tris-glycine gels. Separated proteins were electroblotted on to nitrocellulose membranes (Whatman).
The following antibodies were used for immunoprecipitations and Western blotting: rabbit anti-GluR2 (glutamate receptor 2; Western blot dilution 1:1000; Synaptic Systems), mouse anti-HA (clone 16B12) (Western blot dilution 1:3500; Covance), rabbit anti-GFP antibody Ab6556 for venus-fusion proteins (Western blot dilution 1:3500; Abcam) and mouse anti-GFP antibody clone 7.1 and 13.1 for GFP-fusion proteins (Western blot dilution 1:2000; Roche). Antibodies against specific NLs were described previously: mouse anti-NL1 (clone 4C12)  (Western blot dilution 1:10000; Synaptic Systems), rabbit NL2 antiserum  (Western blot dilution 1:3500), rabbit NL3 antiserum  (Western blot dilution 1:3500) and rabbit NL4 antiserum  (Western blot dilution 1:1000). HRP (horseradish peroxidase)-conjugated affinity purified secondary antibodies (dilution 1:10000; Molecular Probes) were used with enhanced chemiluminescence (GE Healthcare) for signal readout. Quantification of Western blotting band intensities was performed using ImageJ v1.43 software (http://rsbweb.nih.gov/ij/). The subtract background function was applied with a rolling ball radius of 50 pixels. Signal intensities were measured for each band, and averaged over three iterations.
Immunolabelling and imaging
HEK-293FT cells plated on gelatin-coated glass coverslips were transfected and fixed with 4% PFA (paraformaldehyde) 16–24 h after transfection. Cells were permeabilized with 0.1% Triton X-100 solution and blocked with 5% goat serum and 0.1% gelatin solutions. Cells were co-immunolabelled with rabbit anti-HA antibody SG77 (dilution 1:2000; Zymed) together with either mouse anti-KDEL receptor clone 10C3 (dilution 1:100; Stressgen) or mouse anti-GM130 (golgi matrix protein GM130) clone 35 (dilution 1:1500, BD Biosciences) and Alexa Fluor-conjugated secondary antibodies (dilution 1:1000; Molecular Probes). Single optical section imaging was performed on an inverse Leica DMIRE2 microscope connected to a Leica TCS SP2 AOBS confocal laser-scanning setup.
All of the data shown are representative of experiments performed at least in triplicate. Where applicable, statistical analyses were performed using Student's t test.
Neuroligins reside on the surface of neurons as dimers
To examine whether neuroligins natively form oligomers in neurons, we developed an in situ chemical cross-linking approach designed to preserve cell-surface in cis complexes in cultured primary neurons. Neuroligins are tightly intercalated at the synapse, engaging in multiple protein interactions, both intracellularly, and across the synaptic cleft. Our aim was to obtain stoichiometrically identifiable neuroligin oligomers by covalently cross-linking neuroligin complexes on the plasma membrane, while disrupting the interactions of neuroligins with postsynaptic scaffolds and presynaptic neurexins.
Towards this aim, neurons were bathed in chelating buffer to interfere with the Ca2+-dependent trans-synaptic interaction of neuroligins with neurexins . Subsequently, neurons were treated with BS3, a membrane-impermeable cross-linker with reactive groups spaced by an 11.4 Å (1 Å=0.1 nm) carbon arm, to covalently link proteins in close proximity. Under these conditions cross-linking is confined to protein interactions exposed to the extracellular space, thereby excluding cytosolic interactions of neuroligins with postsynaptic scaffolding proteins such as PSD-95 (postsynaptic density protein 95) , S-SCAM (synaptic scaffolding molecule)  and gephyrin . Subsequently, neurons were lysed in SDS-containing buffer to extract membrane proteins and dissociate protein complexes that were not covalently cross-linked. This procedure successfully resulted in a biochemical preparation where native plasma membrane complexes were preserved through covalent cross-linking, allowing stoichiometric and compositional analyses of neuroligin oligomers. It is noteworthy that this methodology is not limited to the study of neuroligins, as exemplified by the detection of cross-linked AMPA receptor complexes in these preparations (albeit with lower efficiency) (Figure 1B), and may have broader applicability to the study of native membrane protein complexes.
Neuroligins form dimers
The SDS/PAGE migration of native neuroligins corresponds to a molecular mass of about 120 kDa. After in situ cross-linking, the majority of neuroligin immunoreactivity appeared as a single band with electrophoretic mobility corresponding to approximately 270 kDa (Figure 1A). The high efficiency of cross-linking to a single band revealed that neuroligins on the surface of neurons are constituents of an in cis membrane complex. Although estimating the sizes of cross-linked adducts by comparing their migration on a gel to protein standards is confounded by the pronounced differences in the migration of linear compared with non-linear polypeptide chains, the approximate size of this complex is compatible with a neuroligin dimer, rather than higher-order oligomers. To examine whether the 270 kDa adducts represent neuroligin dimers, we applied the same cross-linking approach to COS7 cells transfected with neuroligin expression constructs. Cross-linking of HA-tagged NL2 (HA–NL2) yielded an adduct of similar size to that seen in neurons (Figure 1C). Furthermore, introduction of the double point mutation E603A/L604A, previously shown to disrupt recombinant neuroligin oligomerization in vitro , (HA–mono-NL2), resulted in the loss of the 270 kDa band (Figure 1C). Thus, as exogenous expression of NL2 alone is sufficient to form a 270 kDa adduct in non-neuronal cells, and as mutations known to interfere with oligomerization also interfere with adduct formation, the cross-linked complexes observed in neurons likely represent NL dimers.
Neuroligins can homodimerize and heterodimerize
The successful biochemical isolation of stoichiometric neuroligin oligomers from neurons and cell lines allowed us to address the long-standing question of whether neuroligins exist only as homomers, as observed in vitro, or whether they also occur as heteromers. In order to address the isoform composition of neuroligin dimers, we employed tagged expression constructs and isoform-specific neuroligin antibodies that we previously raised against NL1 , NL2 , NL3  and NL4 . The isoform-specificity of these antibodies was confirmed on brain lysates from neuroligin-deletion mutant mice that we previously generated  (Figure 1D).
To determine whether homomers and heteromers form, we co-expressed a fusion protein of NL1 and GFP (NL1–GFP) together with epitope-tagged Myc–NL1, HA–NL2 or HA–NL3 in HEK-293FT cells. After in situ cross-linking, we performed immunoprecipitations using an anti-GFP antibody, and compared cross-linked adducts in lysates and immunoprecipitates on Western blots with antibodies against endogenous NL1, GFP or the HA tag (Figure 1E). The fusion of GFP to NL1 led to a significant decrease in the electrophoretic mobility of cross-linked adducts containing NL1–GFP compared with epitope-tagged neuroligins, allowing us to distinguish neuroligin dimers that contained one, two or no NL1–GFP on the same Western blot by their distinct migration. In cells co-expressing NL1–GFP and Myc–NL1, we identified three cross-linked adduct bands immunoreactive for NL1 in equal stoichiometries (Figure 1E, upper-left-hand blot). The heavy and intermediate adducts were immunoprecipitated with anti-GFP antibody (Figure 1E, upper and lower left-hand blots) and were immunoreactive for GFP in Western blots, whereas the light adduct was not (Figure 1E, lower-left-hand blot). These profiles are consistent with the formation of NL1–GFP/NL1–GFP low-mobility dimers, NL1–GFP/Myc–NL1 intermediate-mobility dimers, and Myc–NL1/Myc–NL1 high-mobility dimers in equal proportions. These data confirm that NL1 can form homodimers on the plasma membrane. The heavy NL1–GFP homodimer displayed increased GFP immunoreactivity compared with the intermediate NL1–GFP/Myc–NL1 dimer (Figure 1E, lower-left-hand blot,) consistent with the increased avidity that the two GFPs afford. This effect of skewing band intensities was observed consistently on Western blots when comparing monomers to dimers containing two epitopes, and was taken into account when interpreting Western blot readouts.
In cells co-expressing NL1–GFP with either HA–NL2 or HA–NL3, we observed equivalent band patterns of heavy, intermediate and light dimer adducts. The immunoreactivities of these bands segregated in a manner consistent with the formation of both homodimers and heterodimers. Heavy and intermediate adducts displayed immunoreactivity for NL1 and were immunoprecipitated by anti-GFP antibody (Figure 1E, upper-centre blots). The intermediate and light adducts showed HA immunoreactivity. Importantly, HA immunoreactivity was also displayed by the intermediate adduct fraction that is immunoprecipitated by the anti-GFP antibody (Figure 1E, lower-centre blot). This pattern indicates that the heavy adducts correspond to NL1–GFP homomers, the light adducts correspond to HA–NL2 homomers, whereas the intermediate mobility adducts correspond to NL1–GFP/HA–NL2 heterodimers. Equivalent patterns were observed in cells co-expressing NL1–GFP and HA–NL3 (Figure 1E, upper- and lower-right-hand blots), together indicating that NL1, NL2 and NL3 homodimers, and NL1/NL2 and NL1/NL3 heterodimers form freely in non-neuronal cells.
Neuroligins form homodimers and selective heterodimers in neurons
Having observed that heterodimerization of neuroligins can occur, we proceeded to analyse the composition of endogenous neuroligin dimers in primary neuron cultures from hippocampus, striatum, cerebellum and spinal cord. We performed in situ cross-linking on these cultures and used the specific anti-neuroligin antibodies (Figure 1D) to individually immunopurify cross-linked adducts for each neuroligin. We subsequently probed these immunoprecipitates for immunoreactivity of other neuroligins by Western blotting. Immunoprecipitation of NL3 from cross-linked samples of hippocampal neurons in culture resulted in the enrichment of 120 kDa (monomeric) and 270 kDa NL3 immunoreactive bands. Probing this preparation with an anti-NL1 antibody revealed the presence of NL1 immunoreactivity selectively in the same 270 kDa NL3-containing cross-linked band (Figure 2A). These data demonstrate that the 270 kDa band contains cross-linked adducts composed of NL1 and NL3, thus positively identifying endogenous NL1/NL3 heterodimers from the surface of hippocampal neurons.
Endogenous neuroligins form select heterodimer species
The existence of NL1/NL3 heterodimers was confirmed by the reciprocal approach through the detection of NL3 immunoreactive 270 kDa adducts in NL1-immunopurified samples (Figure 2A). In these samples, faint NL2 immunoreactivity was also detected, indicating that a minor NL1/NL2 heterodimer population is also present in neurons. The existence of NL1/NL2 heterodimers and their low prevalence was also confirmed in NL2-immunopurified samples exhibiting faint NL1 immunoreactivity (Figure 2A). In contrast, NL3-immunopurified samples showed no NL2 immunoreactivity, and reciprocally NL2-immunopurified samples displayed no NL3 immunoreactivity (Figure 2A). These findings indicate that NL2/NL3 heterodimers do not occur on the surface of hippocampal neurons, despite the fact that these proteins have previously been reported to partially co-localize and co-precipitate .
Having identified NL1/NL3 and minor NL1/NL2 heterodimers in hippocampal neurons, we went on to perform the same analysis of dimer composition in cerebellar granule cell and striatal cultures, which contain largely homogeneous populations of glutamatergic and GABAergic neurons respectively. As in the hippocampus, these samples revealed abundant NL1/NL3, and minor NL1/NL2 heterodimers, whereas NL2 and NL3 were not detected together in dimers (Figure 2B).
To examine NL4-containing oligomers, we employed an NL4 antiserum to immunopurify and detect NL4 from neuron cultures. Like other neuroligins, NL4 cross-linked to dimer-sized bands in hippocampal cultures (Figure 2C and 2D). Owing to cross-linker-independent cross-reacting bands displayed by the anti-NL4 antibody close to the size of anticipated dimers, we confirmed that the putative dimer band was indeed NL4, as it was absent in neurons from NL4-deletion mutant mice (Supplementary Figure S1 at http://www.BiochemJ.org/bj/446/bj4460321add.htm). We went on to examine NL4 immunoprecipitates for immunoreactivites of other neuroligins, but found no evidence of NL4-containing heteromers (Figure 2C). Given the low abundance of NL4 in hippocampal cultures, even if considerable fractions of NL4 formed heterodimers, these could well fall below the detection threshold. We thus additionally cultured and analysed spinal cord neurons where NL4 is known to be abundant . Although we detected a robust NL4 dimer band from spinal cord neurons, no other neuroligin immunoreactivities were detected in the NL4-immunopurified dimers (Figure 2D), indicating that NL4 exclusively forms homodimers, and not heterodimers with other neuroligin isoforms.
Taken together, the above data show that neuroligins form both homodimers and selective heterodimers in neurons.
Neuroligin monomers are retained in the secretory pathway
The detection of neuroligin dimers in neurons raised the question of whether neuroligins reach the cell surface as preformed constitutive dimers, or whether dimerization occurs on the cell surface to accommodate a proposed role in adhesive signalling . To address this question, we examined the localization and trafficking of neuroligin mutants that do not dimerize . Expression of HA–mono-NL2 in hippocampal neurons led to accumulated HA immunoreactivity in somata and subsequent cell death, indicating that HA–mono-NL2 is toxic to neurons (results not shown). We thus resorted to cell lines, which tolerated the expression of monomeric neuroligin mutants, for further experiments.
We expressed HA–NL2 and HA–mono-NL2 in COS7 cells, and compared the efficiency of trafficking using glycosylation profile, surface biotinylation and immunocytochemistry readouts. Transfected neuroligins appear as a band doublet on Western blots. Treatment of lysates with EndoH, which cleaves immature ER (endoplasmic reticulum) N-linked glycans, or PNGaseN, which removes all of the N-linked glycans, showed that the lower band of the doublet corresponds to neuroligin that has yet to exit the ER. Comparison of the band intensity ratios of the doublet indicated the accumulation of immature HA–mono-NL2 in the ER as compared with the unmutated protein (Figure 3B).
Retention of NL monomers by a transmembrane asparagine residue
To directly address the efficiency of surface trafficking, we performed experiments in HEK-293FT cells where surface-exposed protein fractions were covalently labelled with biotin and subsequently purified and separated from intracellular fractions. These experiments confirmed that the maturely glycosylated bands correspond to surface exposed protein, and corroborated the loss of surface trafficking of the monomeric mutant (Figure 3C). To directly visualize the subcellular distributions of the proteins, we immunolabelled fixed cells for HA and secretory pathway proteins, KDEL receptor (ER marker) or GM130 (golgi marker). Although unmutated NL2 was abundant on the margins of the cell, monomeric NL2 was confined to intracellular secretory compartments (Figure 3F).
The above data collectively indicate that neuroligin monomers are retained in the secretory pathway and only traffic efficiently to the plasma membrane once dimerization has occurred.
A transmembrane asparagine residue mediates neuroligin monomer retention
As dimerization is a prerequisite for surface trafficking of neuroligins, we examined neuroligin sequences for putative motifs that could mediate the retention of monomers. We noted that the transmembrane domains of neuroligins are highly conserved and include an asparagine residue within the hydrophobic stretch (Figure 3A). Such polar transmembrane residues are known to be involved in the retention of dissociated subunits of membrane protein complexes in yeast  and the mammalian γ-secretase . We thus tested if this transmembrane domain asparagine residue is involved in the regulation of neuroligin trafficking.
Indeed, mutation of Asn691 to a leucine residue in monomeric NL2 (HA–mono-NL2/N691L) resulted in the rescue of the trafficking deficit caused by the monomeric mutation alone in all of the trafficking assays examined (Figures 3B, 3C and 3F). Similar observations were also made with NL1 and its homologous mutants (results not shown). Relative quantification of the ratio of mature to immature glycoprotein corroborated the significant facilitatory effect of the N691L mutation on the trafficking of the monomeric mutant (Figure 3D). The N691L mutation alone additionally facilitated trafficking of dimerization-competent NL2, though the effect was less pronounced.
Importantly, the N691L mutation rescued the trafficking phenotype of the monomeric mutant, without rescuing its dimerization capacity. After in situ cross-linking, the transmembrane mutation had no effect on the dimerization of NL2, whereas HA–mono-NL2/N691L displayed only background levels of dimer adducts, comparable with those displayed by the monomeric mutants alone (Figure 3E). Thus the transmembrane domain asparagine residue is specifically involved in the active retention of neuroligin monomers in the secretory pathway.
Trafficking and dimerization of neuroligin autism mutants
Strong evidence from both human and mouse genetics has implicated neuroligin mutations in autism-spectrum disorders [9,27,28]. These include the point mutations R471C in mouse NL3 (NL3/R471C) and R87W in human NL4 (NL4/R87W). Interestingly, both mutations affect protein trafficking and lead to retention of the mutant proteins in the secretory pathway [19,29]. Additionally, the R471C mutation has been predicted to disrupt oligomerization . We thus examined whether the transmembrane domain retention mechanism we report in the present study is also responsible for the retention of autism-associated neuroligin mutants, and whether these mutations interfere with dimerization.
Using HEK-293FT cells, we examined tagged HA–NL3 and NL4–venus, and their autism-associated mutants. The observed effect of the mutations confirmed the previously reported reduction of maturely glycosylated protein, and dramatic loss of total protein (Figure 4A). We introduced homologous transmembrane domain asparagine-to-leucine mutations in the autism-associated mutants, yielding HA–NL3/R471C/N722L and NL4/R87W/N689L–venus. Unlike in monomeric neuroligin mutants (Figure 3), mutation of the transmembrane asparagine residue did not restore normal expression of autism-associated NL mutants (Figure 4A). Thus the secretory pathway mechanisms that regulate neuroligin monomers and autism-associated mutants are distinct.
Trafficking and dimerization of autism-associated neuroligin mutants
We went on to examine whether autism mutants display dimers on the cell surface. In situ cross-linking of cells co-transfected with HA–NL3/R471C and NL1–GFP displayed both homodimer and heterodimer bands, similar to cells co-transfected with HA–NL3 and NL1–GFP (Figure 4B). These data show that the NL3 autism mutation does not interfere with dimerization, and that NL1/NL3/R471C mutant dimers can reach the cell surface. Importantly, these experiments demonstrate that the pathogenic R471C mutation in the NL3 gene may not only disrupt NL3 function, but may also exert an epistatic affect on NL1. Conversely, NL4/R87W–venus did not display dimers on the cell surface, despite NL4–venus showing robust homo- and hetero-dimers with Myc–NL1 in these cell line experiments (Figure 4C). This indicates that the NL4 autism mutant, unlike that of NL3, either inhibits dimerization, or causes complete sequestration of dimers from the cell surface. Whichever the case, the presence of surface dimers is a feature that potentially distinguishes the two autism-associated mutations in terms of their molecular characteristics.
Neuroligin biology has attracted considerable interest since neuroligins were first shown to be potent inducers of synaptic differentiation , and to display synapse-subtype specificities [1,3,5,11]. Furthermore, mutations in neuroligin genes were the first to be causally implicated in heritable forms of autism [9,27,28]. With neuroligins emerging as key regulators of synapse development, the present study addresses a critical gap in our understanding of their molecular diversity.
We developed a biochemical procedure that allowed us to examine the oligomerization state of native membrane proteins, and observed that neuroligins interact in cis to form stable dimers on the neuronal surface (Figure 2). Assembly of these dimers takes place in the early secretory pathway, and is a prerequisite for neuroligins to traffic to the cell surface (Figure 3). This is consistent with in situ cross-linking, showing that practically all of the surface-exposed neuroligins are in the dimeric state (Figures 1, 2 and 3). From these data it emerges that the molecular units of neuroligin that function at the postsynaptic membrane are the neuroligin dimers. Thus it is dimer species, not individual neuroligins, that should be considered when examining synaptic specificities.
Homodimerization between neuroligins occurs readily to produce the four homomeric species. NL4/NL4 dimers appear to be the exclusive form of NL4 on the plasma membrane, whereas NL2/NL2 dimers comprise the vast majority of NL2-containing dimers. The existence of NL1/NL1 and NL3/NL3 dimers was directly observed in vitro (Figure 1E), and can be inferred in native neuroligins by band intensity estimates (Figure 2). A novel neuroligin complex identified in the present study is the NL1/NL3 dimer, which was abundantly prevalent in hippocampal, striatal and cerebellar neurons. We also detected traces of a NL1/NL2 dimer species. These are the only detectable heteromeric combinations native to the neuron types we examined, and together with the four homodimers appear to represent the complement of neuroligin complexes on the neuronal plasma membrane.
Immunohistochemistry has shown NL1 and NL2 to be specifically associated with excitatory [1,2] and inhibitory synapses [4,5] respectively. NL4 is specific to glycinergic synapses , whereas NL3 immunoreactivity appears in subsets of both excitatory and inhibitory synapses . Revisiting these observations in light of the above findings, we propose that excitatory synapses can contain NL1 homodimers and NL1/NL3 heterodimers, the latter accounting for the NL3 fraction localized at subsets of excitatory synapses. Inhibitory synapses would invariably contain NL2/NL2 dimers, whereas a subset of inhibitory synapses would additionally accommodate NL3/NL3 dimers, corresponding to those showing NL3 immunoreactivity co-localizing with NL2. These two homodimers may both be part of higher order complexes at the same synapses, possibly representing distinct homodimer species of the same postsynaptic density, which is known to be insoluble in non-ionic detergents . This interpretation is consistent with both a lack of detectable NL2/NL3 heterodimers (Figure 2), and NL2/NL3 co-localization and co-precipitation from non-ionic detergent extracts . Lastly, the NL1/NL2 heterodimer, representing the least prevalent neuroligin complex, may be part of a small subset of synapses that could be either excitatory or inhibitory, since minor fractions of both NL1 and NL2 have been reported to rarely ‘mislocalize’ to inhibitory and excitatory synapses respectively .
The selective trafficking of neuroligin dimers, but not monomers, to the cell surface (Figure 3) is reminiscent of multi-subunit plasma membrane complexes, such as ion channels and transmitter receptors . The constituent subunits of such complexes are retained intracellularly if the complementary subunits are unavailable for complex assembly (e.g. GABAA receptor subunits  or γ-secretase subunits ). This phenomenon has been interpreted to represent a cellular quality control mechanism, which ensures that only fully assembled functional complexes reach their sites of action on the cell surface. An equivalent mechanism appears to operate on neuroligin complexes, where monomers can be considered subunits of the dimeric complex.
The identification of a conserved asparagine residue in the neuroligin transmembrane domain (Figure 3A) that is necessary for the retention of neuroligin monomers provides mechanistic insight into the quality control of neuroligin dimerization in the secretory pathway. Mutation of this asparagine residue to a hydrophobic leucine allowed monomers to aberrantly traverse the secretory pathway and reach the plasma membrane (Figure 3). Thus the polar side chain of the lipid-embedded asparagine residue allows the cell to recognize neuroligin monomers and retain them. Asparagine residues in lipid environments need to neutralize their side chain dipoles, and can do so by associating with one another [36,37]. In a neuroligin dimer, the close proximity of the two transmembrane domains would allow the side chains of the asparagine residue pair to interact and shield their dipoles from the hydrophobic environment. Our data are consistent with a model according to which a nascent neuroligin molecule prior to dimerization, or a dimerization-deficient neuroligin mutant, would stabilize its exposed transmembrane asparagine residue through a transmembrane domain interaction with an ER-resident protein. This association would lead to ER retention and prevent neuroligin monomers from further secretory pathway processing. Neuroligin dimerization as a competing interaction on the asparagine residue side-chain, or mutation of the asparagine to a hydrophobic residue, would disengage the neuroligin molecule from the ER-resident protein, releasing it to traffic towards the cell surface. The present example, together with that of the unrelated γ-secretase complex , substantiates the notion that transmembrane asparagine residues are crucial components of a broader mammalian cellular mechanism that ensures the proper function of protein complexes on the plasma membrane.
Neuroligin mutations associated with autism pathology have been proposed to impede neuroligin oligomerization  and trafficking [28,38]. Moreover, neuroligin dimerization was proposed to take place at the synapse to initiate signalling, and the disruption of this process was suggested to be involved in autism aetiopathology . Our data directly address these hypotheses. We find that the effects of the autism mutations on neuroligin trafficking are not mediated by the retention mechanism we identified, which appears to be dedicated to the control of dimerization. Given the drastic reduction in the amount of protein associated with autism mutations, a mechanism that involves protein degradation , such as ER-associated degradation, is more likely involved. Furthermore, as we find dimerization to be constitutive, it is likely not an initiating step in signalling pathways at the synapse. Examination of the NL3/R471C mutation by in situ cross-linking additionally revealed that these mutants retain their capacity to dimerize. Importantly, NL3/R471C mutants can also heterodimerize with NL1 (Figure 4B), indicating a possible epistatic effect of the NL3 mutation on NL1 function.
NL3/R471C knockin mice display behavioural hallmarks of autism, and have an imbalance in the ratio of synaptic excitation to inhibition. Strikingly, although the mutation leads to a 90% loss of protein, the resulting phenotype is not recapitulated in NL3 knockout animals, indicating that a loss-of-function effect of the mutation on NL3 is not solely responsible for the behavioural and synaptic abnormalities . Conversely, deletion of NL4 is sufficient to cause autism-associated behaviour . In the case of NL3, a putative NL1/NL3/R471C complex, as we observed in cell lines (Figure 4B), may represent a molecular difference between NL3 knockin and knockout animals, and thus a potential mediator of the autism phenotype associated with the NL3 gene. In the case of NL4, autism mutants showed a complete lack of surface dimers in cell lines (Figure 4C). This observation supports the loss-of-function effect of the mutation, and is in line with the recapitulation of the autism phenotype in NL4 knockout animals. The discrepancy between the effects of NL3 and NL4 autism mutations on surface dimers thus correlates with the genetic difference of the two loci in regards to the manifestation of autistic behaviour in mice.
Taken together, our data provide evidence that the functional neuroligin unit is the dimer. Distinct NL1/NL3 and NL1/NL2 heterodimers natively co-exist with neuroligin homodimers. The assembly and composition of these dimers is controlled by the postsynaptic neuron prior to their trafficking to the plasma membrane, where neuroligin dimers drive synaptic differentiation. Finally, interactions between neuroligins may contribute to the effects of neuroligin mutations observed in vitro and in animal models.
green fluorescent protein
glutamate receptor 2
golgi matrix protein GM130
human embryonic kidney
Alexandros Poulopoulos conceived the study. Alexandros Poulopoulos, Tolga Soykan, Liam P. Tuffy and Matthieu Hammer performed the experiments. All of the authors contributed to the writing of the paper.
We thank Dr Chen Zhang and Professor Thomas Südhof (Stanford University, Stanford, CA, U.S.A.) for generously providing reagents. We thank Klaus Hellmann (University of Göttingen, Göttingen, Germany) for excellent technical support, and Dr Sven Thoms (University of Göttingen, Göttingen, Germany) and Patrick F. Davis (Harvard, Cambridge, MA, U.S.A.) for careful feedback on the paper prior to submission.
This work was supported by the Max Planck Society, the German Research Foundation [grant numbers GRK 521 and FZT 103 (to F.V. and N.B.)], the European Commission (EUROSPIN, SynSys, EU-AIMS; to N.B.) and the Cure Autism Now Foundation (to F.V.). L.P.T. is a recipient of a Marie Curie IEF fellowship [grant number 274972].
These authors contributed equally to this work.
Present address: Department of Stem Cell and Regenerative Biology, Harvard University, Cambridge, MA 02138, U.S.A.