SALM4 suppresses excitatory synapse development by cis-inhibiting trans-synaptic SALM3–LAR adhesion

Synaptic adhesion molecules regulate various aspects of synapse development, function and plasticity. These functions mainly involve trans-synaptic interactions and positive regulations, whereas cis-interactions and negative regulation are less understood. Here we report that SALM4, a member of the SALM/Lrfn family of synaptic adhesion molecules, suppresses excitatory synapse development through cis inhibition of SALM3, another SALM family protein with synaptogenic activity. Salm4-mutant (Salm4−/−) mice show increased excitatory synapse numbers in the hippocampus. SALM4 cis-interacts with SALM3, inhibits trans-synaptic SALM3 interaction with presynaptic LAR family receptor tyrosine phosphatases and suppresses SALM3-dependent presynaptic differentiation. Importantly, deletion of Salm3 in Salm4−/− mice (Salm3−/−; Salm4−/−) normalizes the increased excitatory synapse number. These results suggest that SALM4 negatively regulates excitatory synapses via cis inhibition of the trans-synaptic SALM3–LAR adhesion.

S ynaptic adhesion molecules regulate synapse development, function and plasticity. Previous studies have identified a large number of synaptic adhesion molecules, including neuroligins and neurexins [1][2][3][4][5][6][7][8][9][10][11] . The known functions of these molecules include aspects of trans-synaptic interactions between pre-and postsynaptic adhesion molecules and positive synapse regulations (that is, promotion of synapse development, maturation and maintenance), although opposite aspects (cis-interactions between nearby synaptic adhesion molecules on pre-or postsynaptic membranes and negative regulations) are less well understood.
Functionally, SALMs regulate neurite outgrowth and branching 12,16,17 . In addition, they have been implicated in the regulation of synapse development and function. For instance, SALM1 interacts with the GluN1 subunit of NMDA receptors (NMDARs), and promotes dendritic clustering of NMDARs in cultured neurons 12 . SALM2 associates with both NMDARs and AMPA receptors (AMPARs), and promotes the development of excitatory (but not inhibitory) synapses 13 . SALM3 and SALM5 (but not the other SALMs) have synaptogenic activities, promoting both excitatory and inhibitory presynaptic differentiation in contacting axons 18 via trans-synaptic interactions with presynaptic LAR family receptor tyrosine phosphatases 19 . However, with the exception of SALM3, the functions of the SALMs have not been examined using in vivo approaches, such as mouse genetics.
Although individual SALM isoforms are structurally and functionally distinct, they have been shown to associate with one another and are likely to form larger complexes 20 . Specifically, SALM1-3 (but not SALM4 and -5) were found to form a co-immunoprecipitable complex in the rat forebrain. In heterologous cells, all of the SALMs can interact with one another, further extending the complexity of inter-SALM interactions 20 . In addition to these cis-interactions, SALMs can participate in trans-interactions, with SALM4 and -5 forming homophilic and trans-cellular adhesion complexes 20 . These results suggest that the individual SALM isoforms are unlikely to be physically or functionally isolated, but instead may participate in intricate interplays with other SALM isoforms. However, the details of such interactions remain largely unexplored.
In the present study, we generated and characterized Salm4-mutant (Salm4 À / À ) mice. To our surprise, we found that Salm4 À / À mice display increased excitatory and inhibitory synapse numbers in the hippocampus. Our data suggest that SALM4 cis-interacts with SALM3, inhibits the trans-synaptic interaction of SALM3 with presynaptic LAR and suppresses SALM3-dependent presynaptic differentiation at excitatory synapses.

Results
SALM4 expression pattern in the CNS. In order to understand the functions of SALM4, we first determined the expression patterns of Salm4 in the nervous system. SALM4 mRNAs were detected in the mouse brain and spinal cord at embryonic days 16 and 18, as determined by in situ hybridization (Fig. 1a). In postnatal brains, SALM4 mRNAs were detected in various brain regions, including the neocortex, striatum and hippocampus.
SALM4 proteins (B95 kDa) were mainly detected in the rat brain, as determined by immunoblot analysis using SALM4-specific antibodies and different tissue lysates (Fig. 1b,c and Supplementary Fig. 1a-c). SALM4 protein expression gradually increased during postnatal brain development ( Fig. 1d and Supplementary Fig. 1d). SALM4 proteins were detected in synaptic brain fractions, including crude synaptosomes, synaptic membranes and PSD fractions (Fig. 1e,f and Supplementary  Fig. 1e,f), consistent with the previously reported ultrastructural localization of SALM4 proteins around cell junctions, including neuronal synapses 20 .
Because SALM4 is known to regulate neurite outgrowth and branching in cultured hippocampal neurons 16,17 , we examined the morphology of Salm4 À / À CA1 pyramidal neurons in the hippocampus using biocytin dye injection and Sholl analysis. We found that Salm4 À / À neurons had normal dendritic complexities in both apical and basal dendrites (Fig. 2g).
Assessment of SALM4 protein expression by X-gal staining of Salm4 þ / À brain slices (5-week-old mice) revealed that SALM4 could be observed in various regions, including the neocortex, striatum, hippocampus and cerebellum (Fig. 2h). Notably, SALM4 X-gal signals in the hippocampus were stronger in the CA1 and CA3 regions than in the dentate gyrus. This contrasts with the similar levels of SALM4 mRNAs in the CA1/3 and dentate gyrus regions (Figs 1a and 2c), indicative of different expression levels of SALM4 mRNAs and proteins in different brain regions, similar to those observed in the cerebellum. SALM4 deletion had no effect on the levels of other SALMs or synaptic proteins, either in the whole brain or the hippocampus (Fig. 2i,j and Supplementary Fig. 1h).
Salm4 À / À mice display increased synapse numbers. Electron microscopic (EM) studies previously detected SALM4 signals in axons, dendrites and synapses 20 , suggesting that SALM4 may regulate synapse development and function. We thus measured spontaneous excitatory synaptic transmission in hippocampal CA1 pyramidal neurons. We found that the frequency (but not the amplitude) of miniature excitatory postsynaptic currents (mEPSCs) was significantly increased in Salm4 À / À neurons (Fig. 3a). A similar increase was also observed in the frequency (but not amplitude) of miniature inhibitory postsynaptic currents (mIPSCs; Fig. 3b). We could not detect any changes in the inputoutput relationship of evoked excitatory synaptic transmission and the paired pulse facilitation at Salm4 À / À Schaffer collateral-CA1 pyramidal synapses ( Supplementary Fig. 2), which is not in line with the increased frequency of mEPSCs (Fig. 3a). It is possible that mEPSCs in CA1 pyramidal neurons represent synaptic inputs other than those from the Schaffer collateral pathway (that is, the temporoammonic pathway), or that evoked EPSCs are less sensitive in detecting the changes in excitatory synapse number.
EM analysis of the Salm4 À / À hippocampus (CA1 stratum radiatum) revealed significant increases in the density and length of the PSD apposed to presynaptic vesicles, whereas there was no change in the thickness or perforation (a measure of maturation) of the PSD (Fig. 3c). In addition, the density of symmetrically electron-dense regions apposed to GABA (g-aminobutyric acid)positive inhibitory nerve terminals was increased, whereas there was no change in their length or thickness (Fig. 3d). These results suggest that SALM4 deletion increases the number of excitatory and inhibitory synapses in the hippocampus.
SALM4 re-expression rescues excitatory synapse numbers. To determine whether the increased synapse number observed in Salm4 À / À neurons was caused by SALM4 deficiency, we re-expressed SALM4 in Salm4 À / À CA1 pyramidal neurons by herpes simplex virus (HSV) infection on postnatal days (P) 15-18 ( Supplementary Fig. 3a). We found that SALM4 re-expression, confirmed by immunoblot analysis (Fig. 3e and Supplementary  Figs 1i and 3b), rescued the increased mEPSC frequency in Salm4 À / À neurons to a normal range without affecting the amplitude (Fig. 3f). In contrast, SALM4 overexpression in wildtype (WT) neurons had no effect on the frequency or amplitude of mEPSCs, suggesting that endogenous levels of SALM4 proteins are sufficient to have the full negative effect on mEPSC frequency.
In contrast, SALM4 expression in Salm4 À / À CA1 pyramidal neurons had no effect on the frequency or amplitude of mIPSCs ( Supplementary Fig. 3c). Moreover, Salm4 À / À neurons expressing enhanced green fluorescence protein (EGFP) showed mIPSC frequencies similar to those of WT neurons expressing EGFP, which is in contrast to the increased mIPSC frequency observed in uninfected Salm4 À / À neurons. This might be attributable to the viral infection, which may have adverse effects on inhibitory (but not excitatory) synaptic transmission under our experimental conditions. SALM4 cis-interacts with SALMs 2/3/5. To better understand the molecular mechanisms underlying the SALM4-dependent negative regulation of neuronal synapses, we performed an unbiased proteomic screen of SALM4-associated proteins in WT and Salm4 À / À (control) mouse forebrains. We first enriched mature membrane proteins, including SALM4, using a wheat germ agglutinin (WGA) column (Fig. 4a). When these preparations were immunoprecipitated with SALM4 antibodies and subjected to proteomic analyses, we identified SALM2 as a major protein associated with SALM4 in WT brains but absent in Salm4 À / À immunoprecipitates (of the total 20 peptides, 12 were SALM4 and 8 were SALM2; Fig. 4b,c).
This identification of an in vivo association between SALM4 and SALM2 contrasts with the previous reports that SALM1-3, but not SALM4 or -5, form complexes with one another in the rat brain 20 . This discrepancy might reflect that our antibodies are somehow more efficient in pulling down SALM4 proteins in complex with SALM2. Indeed, a previous study reported that SALM1-3 exhibited antibody-dependent differential coimmunoprecipitation, wherein SALM1 immunoprecipitates contained almost undetectable amounts of SALM2 and SALM3, but SALM2 and SALM3 immunoprecipitates contained significant amounts of SALM1 (ref. 20  To further characterize the interaction between SALM4 and SALM2, we performed co-immunoprecipitation experiments in heterologous cells. We found that SALM4 formed a complex with SALM2 ( Fig. 4d and Supplementary Fig. 4a). In addition, the extracellular (ecto) domain of SALM4 (but not the cytoplasmic domain; SALM4-Ecto) could associate with SALM2, suggesting that the ecto domains of SALM4 and SALM2 are involved.
SALM2 forms a complex with SALM3 in vitro and in vivo and with SALM5 in vitro 20 . In addition, SALM3 and SALM5 (but not the other SALMs) have synaptogenic activities 18 . We therefore hypothesized that SALM4 might suppress neuronal synapses through a cis-interaction with SALM3 and/or SALM5. To examine this hypothesis, we first tested whether SALM4 could associate with SALM3 and SALM5, in addition to SALM2.
We found that SALM4 forms a complex with SALM3 in heterologous cells, but neither the ecto nor cytoplasmic domain of SALM4 was sufficient to mediate the interaction ( Fig. 4e and Supplementary Fig. 4b)  (c) Lack of in situ hybridization signals in the Salm4 À / À brain (P56). (d) Salm4 À / À brain lacks detectable SALM4 proteins, as shown by two different SALM4 antibodies (1820 and 2026 guinea pig). (e) Normal gross morphology of Salm4 À / À brain regions (left, forebrain coronal section; right, cerebellar coronal section). Scale bar, 500 mm. (f) Normal numbers of neurons in hippocampal CA1 and CA3 regions. Scale bar, 20 mm. (g) Normal morphology of the apical and basal dendrites of Salm4 À / À hippocampal CA1 pyramidal neurons. Neuronal infusion of biocytin was followed by the analysis of apical (middle) and basal (right) dendritic complexity by Sholl analysis. Scale bar, 50 mm. (h) Distribution patterns of SALM4 proteins in the mouse brain, as revealed by X-gal staining of sagittal Salm4 þ / À brain slices at 5 weeks. Scale bars, 1 mm. (i) SALM4 deletion does not induce compensatory changes in other SALMs, as determined by immunoblot analysis of Salm4 À / À whole brain (WB) and hippocampal lysates (3 weeks). n ¼ 3 mice, ns, Student's t-test. (j) SALM4 deletion does not cause significant changes in the levels of synaptic proteins (scaffolds, receptors and signalling molecules). Salm4 À / À WB and hippocampal (HC) lysates (3 weeks) were immunoblotted by the indicated antibodies. n ¼ 3 mice, ns, Student's t-test.
important for the interaction with SALM2 and SALM5, while the transmembrane domain of SALM4 is important for the interaction with SALM3.
SALM4 inhibits synapse-promoting effects of SALM2/3/5. Because SALM4 associates with SALM2, SALM3 and SALM5, we tested whether SALM4 regulates the functional properties of SALM2/3/5. SALM2 has been shown to facilitate excitatory synapse differentiation, as shown by that cultured hippocampal neurons overexpressing SALM2 show increased excitatory synapse number 13 . When we coexpressed SALM2 and SALM4 in cultured hippocampal neurons, SALM4 reversed the SALM2dependent increase in excitatory synapse number, as indicated by the density of Shank1/2/3 ( Fig. 6a), an excitatory postsynaptic marker protein. In contrast, the size and intensity of Shank clusters were not changed by either the overexpression of SALM2 or SALM2 þ SALM4. This suggests that SALM4 suppresses excitatory synapse-promoting effect of SALM2. SALM3 and SALM5 have been shown to promote presynaptic differentiation in contacting axons 18 . We thus tested whether SALM4 could inhibit SALM3-or SALM5-dependent presynaptic differentiation, using neuron-fibroblast-mixed culture assays 21   heterologous cells markedly suppressed SALM3-dependent synapsin I clustering in contacting axons of cocultured hippocampal neurons (Fig. 6b). SALM4 had similar effects on SALM5-dependent synapsin I clustering. In control experiments, SALM4 coexpression had no effect on synapsin I clustering induced by LRRTM2, an LRR-containing adhesion molecule with synaptogenic activity 22,23 . SALM4 expression did not change the surface levels of coexpressed SALM3, SALM5 or LRRTM2, although there were decreasing tendencies, as determined by surface immunofluorescence staining (Fig. 6b) and surface biotinylation assays ( Fig. 6c and Supplementary Fig. 5a).

SALM4 inhibits the interaction between SALM3/5 and LAR.
Because SALM4 suppresses SALM3-and SALM5-dependent presynaptic differentiation, we next tested whether SALM4 could inhibit the interaction between SALM3 and its known presynaptic ligand, LAR 19 , or that between SALM5 and LAR 24 .
To this end, we tested whether the binding of LAR (a soluble form) to SALM3 displayed on heterologous cell surfaces could be inhibited by SALM4 coexpressed in the same cell. Indeed, the interaction of SALM3 with increasing concentrations of soluble LAR (extracellular Ig1-3 domains; LAR-Ig1-3-Fc) was significantly reduced by SALM4 coexpression (Fig. 7a). Similarly, LAR binding to SALM5 was significantly inhibited by SALM4 coexpression (Fig. 7b).
Double Salm3-Salm4 KO normalizes excitatory synapse numbers. The cis-regulation of SALM3/5 by SALM4 requires their expression in the same cells. In line with this, X-gal staining revealed that the expression patterns of SALM3 and SALM4 proteins significantly overlapped in mouse brain regions, including the neocortex, hippocampus and striatum ( Fig. 8a and Supplementary Fig. 6a), and also in hippocampal CA3 and CA1 regions. The potential overlap between SALM4 and SALM5 could not be tested because Salm5 À / À mice that allow X-gal staining were unavailable; however, we note that the known mRNA distribution pattern of SALM5 substantially overlap with those of SALM3 and SALM4 (refs 13,14).
We found that the increased mEPSC frequency observed in Salm4 À / À CA1 pyramidal neurons was normalized to near-WT levels in Salm3 À / À ; Salm4 À / À CA1 pyramidal neurons (Fig. 8b). This result also indicates that the reduced mEPSC frequency in Salm3 À / À neurons was normalized in the double KO mice, although the extent of normalization was weaker than that for Salm4 À / À neurons. The double KO had no effect on the amplitude of mEPSCs.
When mIPSCs were compared, we found that the double KO did not rescue the increased mIPSC frequency seen in Salm4 À / À mice (Fig. 8c), nor did it affect the mIPSC frequency in Salm3 À / À mice. Moreover, the observed mIPSC amplitudes were comparable across the four genotypes. Together, these results indicate that the double KO of SALM3 and SALM4 rescues the increased frequency of mEPSCs (but not mIPSCs) in Salm4 À / À mice.
We next performed EM analysis to see whether the double KO normalizes the increased excitatory and inhibitory synapse densities in Salm4 À / À mice. We found that the Salm3 À / À ; Salm4 À / À hippocampus (CA1 stratum radiatum) shows the density of the PSD significantly smaller than that of Salm4 À / À hippocampus but comparable to that of WT hippocampus (Fig. 8d). In contrast, the increased inhibitory synapse density in Salm4 À / À hippocampus was not normalized by the double KO (Fig. 8e). These results further suggest that SALM3 mediates the SALM4-dependent inhibition of excitatory synapse number.

Discussion
In the present study, we report that Salm4 À / À mice display increased excitatory and inhibitory synapse numbers in the hippocampus, and that SALM4 cis-interacts with SALM3, inhibits the SALM3-LAR interaction and suppresses SALM3-dependent presynaptic differentiation.
Our data indicate that a SALM4 mutant that lacks SALM3 binding (SALM4-Ecto), unlike full-length SALM4, loses the ability to inhibit LAR binding to SALM3 in HEK293T cells, and to rescue the increased mEPSC frequency in CA1 pyramidal cells (Fig. 7d-f). However, the extent of these reversals are partial, especially in the LAR-binding experiments in vitro. One possibility is that SALM4-Ecto, although it no longer associates with SALM3 in HEK293T cells under co-immunoprecipitation conditions (Fig. 4e), may still weakly associate with, or get sufficiently close to, SALM3 on the surface membrane through its intact extracellular domain and inhibit LAR binding to SALM3; the fact that SALM4-DLRR binds to SALM3 does not exclude the possibility that the LRR domain of SALM4 may still weakly bind to SALM3. This nonspecific inhibition, if present, might be stronger when SALM4-Ecto is highly expressed in HEK293T cells, whereas it is weaker when SALM4-Ecto is weakly expressed in neurons, or SALM4-Ecto cannot easily have an access to endogenous SALM3 at excitatory synaptic sites.
Our hypothesis (SALM3 inhibition by SALM4) is genetically supported by our observation that double KO of SALM3 and SALM4 normalizes mEPSC frequency and excitatory synapse number phenotypes. This suggests that SALM3, but not SALM5, may act as a major downstream effector of SALM4 for the inhibition of excitatory synapse number in the hippocampus. In line with this possibility, SALM3 and SALM4 show substantial overlaps in their protein distribution patterns in the brain (Fig. 8a). Moreover, a previous study showed that the temporal expression patterns of SALM3 and SALM4 mRNAs along the embryonic developmental stages are more similar than those of the other SALMs 14 . These results, however, do not exclude the possibility that SALM5 may closely interact with SALM4 in some other brain regions.
Our data, however, provide relatively few clues as to how SALM4 negatively regulates inhibitory synapses. The increased inhibitory synapse number in Salm4 À / À mice might merely represent a compensatory change induced by the increase in excitatory synapse number. Alternatively, it could be a more direct effect involving SALM4. In line with this, a previous study showed that SALM3 and SALM5, which associate with SALM4, can induce both excitatory and inhibitory presynaptic differentiation in cocultured neurons 18 . In addition, the LAR family proteins, which bind to SALM3 and SALM5, can induce both excitatory and inhibitory postsynaptic differentiations by interacting with NGL-3, TrkC, IL1RAPL1, IL1RAcP and Slitrks 2,6 .
It is likely that 'postsynaptic' SALM4 cis-interacts with SALM3 to suppress excitatory synapses. SALM4 is detected at both preand postsynaptic sites, and is thought to mediate homophilic and trans-synaptic adhesions 20 . This homophilic adhesion, however, is unlikely to contribute to the SALM4-dependent negative regulation of SALM3 because we herein showed that HSVmediated re-expression of SALM4 in Salm4 À / À CA1 pyramidal neurons (postsynaptic side) was sufficient to rescue the increased mEPSC frequency in a cell-autonomous manner.
Our study extends the reported complexity of interactions among the SALM family members. The proteomic data derived from WT and KO (control) brain samples suggest that there is a strong biochemical association between SALM4 and SALM2 in vivo. In addition, SALM4 associates with SALM3 and SALM5 in addition to SALM2 in heterologous cells. Given that SALMs 1-3 form a strong co-immunoprecipitable complex in the brain 20 , SALM4, together with SALMs 1-3 and SALM5, may form combinations of protein complexes larger than previously revealed in different brain regions. This complex might be further stabilized by interactions with PSD-95 through SALMs 1-3, and with LAR family proteins through SALMs 3 and 5 (refs 12-14,19). Therefore, the SALM4-dependent suppression of SALM2-dependent excitatory synapse development (Fig. 6a) could be mediated indirectly through SALM3. Conversely, the SALM4-dependent suppression of SALM3-dependent presynaptic differentiation could also be mediated or modulated by SALM2, a major binding partner of SALM4 in the brain (Fig. 4b,c).
Inhibition of neuronal synapses via cis-interactions of SALM4 is reminiscent of the previously reported cis-interaction between neuroligin-1 and neurexin-1b, which suppresses neuroligin-1dependent presynaptic differentiation in contacting axons 25 . In addition, postsynaptic MDGA1 was shown to cis-interact with neuroligin-2 and thereby suppress neuroligin-2-dependent inhibitory presynaptic differentiation 26,27 . These results suggest that postsynaptic adhesion molecules may cis-interact with one another to regulate their trans-interactions. In addition, we recently showed that a trans-interaction between postsynaptic NGL-1 and presynaptic netrin-G1 (a glycosylphosphatidylinisotol (GPI)-anchored synaptic adhesion molecule) induces the cisinteraction of netrin-G1 with LAR 28 . These results suggest that there may be functional interplays between cis-and transadhesions at the synapse.
SALMs have been implicated in autism spectrum disorders, intellectual disability and schizophrenia [29][30][31][32][33] . In addition, LAR family proteins are implicated in autism spectrum disorders, attention deficit/hyperactivity disorder, restless leg syndrome and schizophrenia 2,6 . Therefore, the mechanisms revealed in the present study might help us better understand these disorders.
In conclusion, our results suggest that postsynaptic SALM4 negatively regulates excitatory synapse numbers through a cisinteraction with SALM3 and the inhibition of both the SALM3-LAR interaction and SALM3-dependent presynaptic differentiation. Our study provides genetic support for the emerging notion that neighbouring adhesion molecules cis-interact with one another to regulate neuronal synapses.

Methods
Generation of Salm4 À / À mice. A mouse ES cell clone (15252A-A12; Lrfn3_AA12), derived from the C57BL/6N strain, was obtained from Velocigene (VG15252). The cassette (ZEN-Ub1) for an allele deletion contained LacZ and neomycin (b-geo) and parts of exons 2 and 3 of the Salm4/ Lrfn3 gene for homologous recombination. To generate male chimeric mice, cultured ES cells (C57BL/ 6N) were microinjected into the blastocyst of the C57BL/6J-Tyroc-2J4(albino B6). Chimeric mice were bred with albino B6 females (C57BL/6J-Tyroc-2J4) to generate germline-transmitted F0 mice (C57BL/6J-Tyroc-2J4 þ C57BL/6N strain). F0 mice were backcrossed to C57BL/6J for two to seven generations. The F2 mice were used for the analysis of brain morphology and synaptic protein levels. Electrophysiology and electron microscopy were performed using F3-7 generations. All mice were bred and maintained according to the KAIST Animal Research Requirements, and all procedures were approved by the Committees of Animal Research at KAIST. Mice were fed ad libitum by standard rodent chow and tap water, and housed under 12-h light/dark cycle (lights off at 19:00).
For field recordings, transverse hippocampal slices (400 mm) were prepared from 3-to 4-week-old male littermates. The brain was rapidly isolated and placed in ice-cold, oxygenated dissection buffer containing (in mM) 212.7 sucrose, 5 KCl, 1.23 NaH 2 PO 4 , 0.5 CaCl 2 , 10 MgCl 2 , 26 NaHCO 3 and 10 glucose. Hippocampal slices were prepared by cutting with a Leica VT1000P vibratome (Leica) and transferred for recovery to a holding chamber containing oxygenated ACSF consisting of (in mM) 124 NaCl, 5 KCl, 1.23 NaH 2 PO 4 , 2.5 CaCl 2 , 1.5 MgCl 2 , 26 NaHCO 3 and 10 glucose at 28-30°C for at least 1 h before recording. After recovery, slices were transferred to a recording chamber where they were perfused continuously with oxygenated ACSF (27-28°C) at a flow rate of 2 ml min À 1 . Hippocampal CA1 field excitatory postsynaptic potentials were evoked by Schaffer collateral stimulation (0.2 ms current pulses) with concentric bipolar electrode. Synaptic responses were recorded with ACSF-filled microelectrodes (1-2 MO) and were quantified as the initial slope of field excitatory postsynaptic potential). Data from slices with stable recordings (o5% change over the baseline period) were included in the analysis. All data are presented as mean ± s.e.m. normalized to the preconditioning baseline (at least 20 min of stable responses). The experiments were blind to mouse genotypes. Recordings were performed using an AM-1800 Microelectrode amplifier (A-M systems), and the IGOR software (WaveMetrics, Lake Oswego, OR, USA) was used for digitizing and analysing the responses.
Electron microscopy. WT, Salm4 À / À and Salm3 À / À ;Salm4 À / À mice were deeply anaesthetized with sodium pentobarbital (80 mg kg À 1 , intraperitoneally (i.p.)) and were intracardially perfused with 10 ml of heparinized normal saline, followed by 50 ml of a freshly prepared fixative of 2.5% glutaraldehyde and 1% paraformaldehyde in 0.1 m phosphate buffer (PB, pH 7.4). The hippocampus was removed from the whole brain, postfixed in the same fixative for 2 h and stored in PB (0.1 M, pH 7.4) overnight at 4°C. Sections were cut transversely on a Vibratome at 70 mm. The sections were osmicated with 0.5% osmium tetroxide (in 0.1 m PB) for 1 h, dehydrated in graded alcohols, flat embedded in Durcupan ACM (Fluka) and cured for 48 h at 60°C. Small pieces containing the stratum radiatum of hippocampal CA1 region were cut out of the wafers and glued on the plastic block by cyanoacrylate. Ultrathin sections were cut and mounted on Formvar-coated single-slot grids. For excitatory synapses, sections were stained with uranyl acetate and lead citrate, and examined with an electron microscope (Hitachi H-7500; Hitachi) at 80 kV accelerating voltage. For inhibitory synapses, sections were further immunogold-stained for GABA.
Postembedding immunogold staining for GABA. Sections were immunostained for GABA by the postembedding immunogold method, as previously described 42 , with some modifications. In brief, the grids were treated for 5 min in 1% periodic acid to etch the resin, and for 8 min in 9% sodium periodate to remove the osmium tetroxide, and then washed in distilled water, transferred to Tris-buffered saline containing 0.1% Triton X-100 (TBST; pH 7.4) for 10 min and incubated in 2% human serum albumin in TBST for 10 min. The grids were then incubated with rabbit antiserum against GABA (GABA 990, 1:10,000) in TBST containing 2% human serum albumin for 2 h at room temperature. The antiserum (a kind gift from Professor O.P. Ottersen at the Center for Molecular Biology and Neuroscience, University of Oslo) was raised against GABA conjugated to bovine serum albumin with glutaraldehyde and formaldehyde 43 , and characterized by spot testing 44 . To eliminate cross-reactivity, the diluted antiserum was pre-adsorbed overnight with glutaraldehyde (G)-conjugated glutamate (500 mM, prepared according to a previous report 45 ). After extensive rinsing in TBST, the grids were incubated for 3 h in goat anti-rabbit IgG coupled to 15 nm gold particles (1:25 in TBST containing 0.05% polyethylene glycol; BioCell Co., Cardiff, UK). After a rinse in distilled water, the grids were counterstained with uranyl acetate and lead citrate, and examined with an electron microscope (Hitachi H-7500; Hitachi) at 80 kV accelerating voltage. To assess the immunoreactivity for GABA, gold particle density (number of gold particles per mm 2 ) of each GABA þ terminal was compared with gold particle density of terminals, which contain round vesicles and make asymmetric synaptic contact with dendritic spines (background density). Terminals were considered GABA-immunopositive ( þ ) if the gold particle density over the vesicle-containing areas was at least five times higher than background density.
Quantitative analysis of excitatory and inhibitory synapses. For quantification of excitatory synapse, 24 electron micrographs representing 368.9 mm 2 neuropil regions in each mouse were taken at a Â 40,000. Number of spines (PSD density), proportion of perforated spines, PSD length and PSD thickness from each three WT, Salm4 À / À and Salm3 À / À ; Salm4 À / À mice were quantified by using the ImageJ software. For quantification of inhibitory synapse, 24 electron micrographs representing 655.5 mm 2 neuropil regions in each mouse were taken at a Â 30,000. Number of GABA þ terminals showing clear PSD (inhibitory synapse density), length and thickness of PSD contacting GABA þ terminals from each three WT, Salm4 À / À and Salm3 À / À ; Salm4 À / À mice were quantified by using the ImageJ software. The measurements were performed by an experimenter blind to the genotype. Digital images were captured with the GATAN DigitalMicrograph software driving a charge-coupled device camera (SC1000 Orius, Gatan) and saved as TIFF files. Brightness and contrast of the images were adjusted in Adobe Photoshop 7.0 (Adobe Systems).
Cultured neuron transfection and imaging. Cultured hippocampal neurons were prepared from E18 rat brains, as described previously 47 , on coverslips coated with poly-D-lysine and grown in Neurobasal medium supplemented with B-27 (Invitrogen), 0.5% fetal bovine serum, 0.5 mM Glutamax (Invitrogen) and sodium pyruvate (Invitrogen). For the overexpression of SALMs in cultured neurons, hippocampal neurons were transfected with pIRES-SALM2-EGFP, pGW1-Myc-SALM4 or EGFP (Control) using a CalPhos Kit (Clontech) at DIV10 and immunostained at DIV14. For the ICC, cultured neurons were fixed with 4% paraformaldehyde/4% sucrose, permeabilized with 0.2% Triton X-100 in PBS, immunostained with primary antibodies (against EGFP, Shank1 and SALM4 (to identify the triply transfected hippocampal neurons)), and Cy3-, Cy5-and fluorescein isothiocyanate-conjugated secondary antibodies (Jackson ImmunoResearch). The images were acquired using a confocal microscope (LSM710, Carl Zeiss) with a Â 63 objective lens. All image settings were kept constant. The Z-stacked images were converted to maximal projection and analysed to obtain the size, intensity and density of the puncta immunoreactivities derived from marker proteins. The quantification was performed in a blind manner using MetaMorph (Molecular Devices).
Mixed culture synapse-formation assays. Mixed culture assays were performed with HEK293T cells (American Type Culture Collection) as described 23 . Cultured hippocampal neurons were incubated until HEK293T cells transfected with EGFP (Control), or the indicated SALM3, SALM5 or LRRTM2 constructs, were added at 10 days in vitro (DIV 10) for further coculture experiments. HEK293T cells were transfected with FuGene (Roche, USA) as indicated. After 48 h, transfected HEK293T cells were trypsinized, seeded on the hippocampal neuron cultures at DIV 10, further cocultured for 48 h and double-immunostained with synapsin I, SALM4 and HA/GFP antibodies at DIV 12 as described previously 48 . All images were acquired using a confocal microscope. For quantifications, the contours of the transfected HEK293T cells were chosen as the region of interest. The fluorescence intensity of synapsin puncta normalized to each HEK293T cell area was quantified for red, green and blue channels with the MetaMorph Software (Molecular Device).
Recombinant protein expression and purification. The Ig1-3 domains (aa 30-318) of human LAR-RPTP was cloned into the BamHI and XbaI sites of the pAcGP67 vector (BD Bioscience) tagged with the Fc domain of human IgG gene. The High Five insect cells (Invitrogen) were transfected with corresponding P4 baculovirus for 3 days and harvested. The supernatants containing secreted proteins were loaded on protein A-Sepharose column (GE Healthcare Life Science) for the purification of Fc-tagged proteins. The affinity resin was washed with the buffer containing 50 mM Tris-HCl (pH 8.0) and 200 mM NaCl. Fc-tagged proteins immobilized in the protein A-Sepharose column were eluted with glycine buffer (100 mM glycine, pH 2.1).
Mass spectrometric screen of SALM4-associated proteins. Preparation of protein samples, affinity chromatography and proteomic analysis were performed as described previously 49 . Briefly, crude synaptosomes were prepared from mouse forebrains (15 WT and KO mice; 3 weeks), and solubilized in 1 Â PBS containing 0.5% Triton X-100, 0.1% SDS, 0.5 mM EDTA, 0.5 mM EGTA and protease inhibitor cocktail (Roche). From the solubilized proteins, mature membrane proteins were enriched using WGA column chromatography (WGA bead, Vector Laboratories) and eluted using N-acetyl-d-glucosamine (Sigma). The samples were incubated with SALM4 antibodies (2026, purified, Gp) prebound to protein A-Sepharose beads (GE Healthcare), followed by protein elution with 2.5 Â XT MOPS buffer (BIO-RAD). Boiled samples were resolved using 4B12% Criterion XT Precast Gels (BIO-RAD) followed by silver staining. From the stained gels, protein bands were excised and subjected to mass spectrometric analyses. Mass spectrometry was performed using LTQ-Orbitrap XL ETD (Thermo Scientific) at the Korea Basic Science Institute.
Statistics. The results of statistical analysis are indicated in Supplementary  Table 1.
Data availability. The authors confirm that all relevant data are available from the authors.