Calsyntenin-3 directly interacts with neurexins to orchestrate excitatory synapse development in the hippocampus

Calsyntenin-3 (Clstn3) is a postsynaptic adhesion molecule that induces presynaptic differentiation via presynaptic neurexins (Nrxns), but whether Nrxns directly bind to Clstn3 has been a matter of debate. Here, we show that β-Nrxns directly interact via their LNS domain with Clstn3 and Clstn3 cadherin domains. Expression of splice site 4 (SS4) insert-positive β-Nrxn variants, but not insert-negative variants, reversed the impaired Clstn3 synaptogenic activity observed in Nrxn-deficient neurons. Consistently, Clstn3 selectively formed complexes with SS4-positive Nrxns in vivo. Neuron-specific Clstn3 deletion caused significant reductions in number of excitatory synaptic inputs, and moderate impairment of light-induced anxiety-like behaviors in mice. Moreover, expression of Clstn3 cadherin domains in CA1 neurons of Clstn3 conditional knockout mice rescued structural deficits in excitatory synapses, especially within the stratum radiatum layer. Collectively, our results suggest that Clstn3 links to SS4-positive Nrxns to induce presynaptic differentiation and orchestrate excitatory synapse development in specific hippocampal neural circuits.


Introduction
Synaptogenic adhesion molecules, a class of synaptic transmembrane proteins that induce synaptic differentiation in vitro (Missler et al., 2012;Südhof, 2017;Um and Ko, 2013), are central to various aspects of synapse development, but their precise roles in synapse assembly, validation, and/or plasticity in vivo are only beginning to be revealed (Ko et al., 2015;Missler et al., 2012;Südhof, 2018). Presynaptic neurexins (Nrxns) and leukocyte common antigen-related receptor protein tyrosine phosphatases (LAR-RPTPs) are among the synaptogenic adhesion molecules that have emerged as key platforms that facilitate convergence of diverse signals from multifarious postsynaptogenic ligands at mammalian synapses Südhof, 2018). Of particular note is the fact that, although Nrxns and LAR-RPTPs are evolutionarily conserved, only a subset of their ligands in mammals has homologs in invertebrate species that also play crucial roles in various aspects of central nervous system development, hinting at the possibility that Nrxns and LAR-RPTPs serve fundamental functions through these selective adhesion pathways.
Calsyntenins (Clstns) are evolutionarily conserved synaptogenic adhesion proteins of the cadherin superfamily that are expressed most highly in the brain (Sotomayor et al., 2014). Synaptic functions of the three vertebrate Clstn family members have recently been reported. For example, juvenile Clstn1-deficient mice exhibit compromised excitatory synaptic transmission, possibly owing to disrupted targeting of N-methyl-D-aspartate (NMDA) receptor subunits (Ster et al., 2014). They also show increased synaptic levels of GluN2B subunit-containing NMDA receptors, enhanced longterm potentiation (LTP) and greater filopodia-like dendritic protrusions in the hippocampus, but decreased dendritic arborization, suggesting that Clstn1 mediates dendritic transport of NMDA receptor subunits and regulates spine maturation during early development (Ster et al., 2014). In addition, Clstn1 regulates guidance receptor trafficking by shuttling Rab11-positive vesicles, leading to switching of commissural axon responsiveness (Alther et al., 2016). Clstn1 also contributes to peripheral sensory axon arborization, branching, endosomal dynamics, and microtubule polarity in zebrafish (Lee et al., 2017;Ponomareva et al., 2014). Clstn2, on the other hand, plays a non-redundant role in inhibitory synapse development and influences a subset of cognitive abilities (Lipina et al., 2016;Ranneva et al., 2017). Clstn3 was identified as a postsynaptogenic adhesion molecule that acts through presynaptic Nrxns (Pettem et al., 2013;Um et al., 2014). However, in contrast to the report of Pettem et al. (2013), we did not detect direct interactions between Clstn3 and α-Nrxns.
In the present study, we revisited these issues. Strikingly, utilizing newly engineered Nrxn1 expression vectors to increase Nrxn1β expression levels, we found that recombinant Clstn3 bound both Nrxn1β and Nrxn1α, with a slight preference for splice site 4 (SS4) insert-positive variants, requiring an Nrxn splice variant containing an insert at SS4 as a functional receptor for its presynaptic differentiation-inducing activity. Conditional deletion of Clstn3 in neurons led to drastic reductions in excitatory synapse structures (but not basal excitatory synaptic transmission), and mild impairments in a subset of hippocampus-dependent mouse behaviors. Finally, adeno-associated virus-mediated expression of Nrxn-binding CST-3 cadherin domains was sufficient to rescue decreased excitatory synapse puncta density in CA1 stratum radiatum layers of Clstn3-deficient mice. Viewed together, our results revise our previous molecular model, showing that Clstn3 directly interacts with SS4-positive Nrxn splice variants to induce presynaptic differentiation, and suggesting synapse-organizing functions of Clstn3 that may control specific synaptic inputs from Schaffer-collateral afferents in the hippocampus. expression was not likely attributable to differences in the position of the FLAG epitope between rNrxn1β (N-terminus) and our original mNrxn1β construct, as evidenced by the comparably robust binding of a newly generated N-terminally FLAG-tagged mNrxn1β +SS4 (FLAG-mNrxn1β +SS4 ) and our original C-terminally FLAG-tagged mNrxn1β +SS4 (mNrxn1β +SS4 -FLAG) to Ig-Clstn3 in cell surfacebinding assays (Figure 1-figure supplement 1B and 1C).
Cadherin domains of Clstn3 mediate direct binding to β-Nrxns. In addition to cell surface-binding assays, in vitro pull-down assays clearly showed binding of Clstn3 to both Ig-mNrxn1β +SS4 and Ig-mNrxn1α +SS4 . To identify a minimal Clstn3 domain involved in Nrxn binding, we used three different FLAG-tagged Clstn3 constructs: Full (full-length Clstn3), Cad (containing tandem cadherin domains, the transmembrane segment and intracellular residues of Clstn3), and ΔCad (full-length Clstn3 lacking the tandem cadherin domains) (see Figure 2A for schematic diagram of Clstn3 constructs).
Both Ig-mNrxn1α +SS4 and Ig-mNrxn1β +SS4 pulled down Clstn3 Cad, but not Clstn3 ΔCad (Figures 2B   and 2C), suggesting that cadherin domains of Clstn3 mediate binding to Nrxns. The results of these and aforementioned cell-based surface-binding assays do not exclude the possibility that intermediate(s) expressed in HEK293T cells may bridge indirect associations of Nrxns with Clstn3.
Clstn3 binds to an LNS domain of mNrxn1β in a Ca 2+ -dependent manner. We next investigated which Nrxn sequences mediate Clstn3 binding. To this end, we generated the following FLAG-tagged mNrxn1β-deletion constructs: mNrxn1β ΔHRD, which lacks a β-Nrxn-unique histidine-rich sequence; mNrxn1β ΔLNS, which lacks a β-Nrxn LNS domain; and ΔStalk1, which lacks the entire stalk region ( Figure 2G). We found that Ig-Clstn3 bound comparably to HEK293T cells expressing mNrxn1β ΔHRD or mNrxn1β wild-type (WT), but did not bind HEK293T cells expressing mNrxn1β ΔLNS ( Figures 2H and 2I), indicating that the Nrxn1β LNS domain is necessary for Clstn3 binding.
Interestingly, deletion of the entire stalk region (mNrxn1β ΔStalk1) diminished Clstn3 binding as well as NL-2 binding (Figures 2H and 2I). The stalk region contains ~40 residues with several putative Olinked glycosylated sites and a short cysteine-loop sequence composed of two conserved cysteines flanking an 8-residue acidic sequence (Sterky et al., 2017). We thus hypothesized that O-linked glycosylation of mNrxnβ regulates Clstn3 binding and that mNrxn1β ΔStalk1 displayed weak Clstn3 binding because it lacks O-linked glycosylation. To test this, we generated mNrxn1β constructs containing point mutations or deletions in the conserved stalk region. Clstn3 binding was retained in mNrxn1β constructs in which putative O-glycosylated threonines or serines were replaced with glycines (mNrxn1β ΔCHO), the cysteine-loop sequence was deleted (mNrxn1β ΔCysL), the stalk region was partially deleted (mNrxn1β ΔStalk2), or the conserved serine residue for attaching heparan sulfate chains was replaced with alanine (mNrxn1β ΔHS) (Figure 2-figure supplement 1A-D).
However, Clstn3 did not bind Nrxn1γ, a newly identified Nrxn1 isoform (Sterky et al., 2017), or other heparan sulfate proteogylcans, such as glypicans or syndecans ( Figure 2H . These findings suggest that the Nrxn1β LNS domain is a Clstn3-binding site. We further found that treatment with the Ca 2+ chelator EGTA prevented these interactions (Figure 2-figure supplement 3).

Generation and Characterization of Clstn3 Conditional Knockout (cKO) Mice.
To elucidate the physiological significance of Clstn3-Nrxn interactions in vivo, we used Clstn3 tm1a(EUCOMM)Hmgu mice in which a targeting cassette harboring FRT, lacZ, and loxP sites was inserted between exon 7 and exon 8, resulting in a 'knockout-first' lacZ-reporter-tagged Clstn3 tm1a insertion allele with conditional potential (Figure 4-figure supplement 1A) (Skarnes et al., 2011). The Clstn3 tm1a allele was confirmed by genomic polymerase chain reaction (PCR) (Figure 4-figure supplement 1B), and loss of Clstn3 protein in Clstn3 tm1a/tm1a mice was verified by the absence of detectable Clstn3 immunoreactivity to two Clstn3-specific antibodies (JK001 and JK091; Figure 4-figure supplement 1C and 1D). Deletion of Clstn3 in Clstn3 tm1a/tm1a mice did not affect the expression levels of diverse synaptic proteins, including Clstn1 and Clstn2 (Figure 4-figure supplement 1E and 1F). In addition, the gross morphology of Clstn3 tm1a/tm1a mice was normal and their neuron numbers were comparable to those in WT mice, as assessed by Nissl and NeuN (neuronal marker) staining (Figure 4-figure supplement 1G-I). We then crossed Clstn3 tm1a/tm1a mice with an FLPe knock-in strain to remove contaminating transgenes and the neomycin resistance cassette to generate Clstn3 tm1c(EUCOMM)Hmgu mice.
Before deciding which Cre driver lines to cross for the current study, we analyzed Clstn3 expression patterns in the mouse brain. For this, we performed RNAscope-based in situ hybridization analysis using a Clstn3-specific probe. This analysis showed that Clstn3 mRNA was strongly expressed in most pyramidal cell layers and interneurons in the mouse hippocampus (Figure 4-figure   supplement 2A). Notably, Clstn3 expression in pyramidal cell layers in the CA3 region was particularly prominent (Figure 4-figure supplement 2A and 2C). Clstn3 mRNA was also expressed in parvalbumin (PV)-and somatostatin (SST)-positive GABAergic interneurons in the hippocampus, as previously reported (Pettem et al., 2013) (Figure 4-figure supplement 2B). Moreover, immunofluorescence analyses using a Clstn3-specific antibody (JK091) showed that the expression pattern of Clstn3 protein was similar to that of Clstn3 mRNA in the hippocampus (Figure 4-figure   supplement 2D). Based on these results, the Nestin-Cre (Nestin-Clstn3) driver line was chosen to further cross with Clstn3 tm1c(EUCOMM)Hmgu mice to generate Clstn3-cKO mice (Figures 4A and 4B).
These mice were largely indistinguishable from control littermates (Clstn3 f/f ; Ctrl) in terms of birth rate, although Nestin-Clstn3 mice weighted marginally (but significantly) less at both postnatal day 30 (P30) and P54, which is a generalized metabolic phenotype of the Nestin-Cre driver line (Galichet et al., 2010) ( Figure 4C; Table S2). In addition, gross morphology (as assessed by Nissl staining) and neuron numbers (as assessed by NeuN staining) in Nestin-Clstn3 mice were comparable to those of control littermates (Figure 4D-F). Semi-quantitative immunoblot analyses showed that the relative expression levels of various synaptic proteins in the hippocampus ( Figure 4G) and cortex ( Figure   4H) of Nestin-Clstn3 mice were unchanged compared with littermate controls (Figures 4G-I).
Collectively, these data suggest that Clstn3 is not essential for mouse survival or breeding, and does not affect the expression levels of synaptic proteins.
Excitatory synapse development is impaired in CA1 hippocampal pyramidal neurons of Clstn3-cKO mice. We next performed immunohistochemistry to quantify the intensity of excitatory and inhibitory synapses, identified by labeling with antibodies to VGLUT1 (vesicular glutamate transporter 1) and GAD67 (glutamic acid decarboxylase 67), respectively. To this end, we analyzed the effect of Clstn3 deletion on the intensity of excitatory and inhibitory synaptic puncta in various layers of the hippocampal CA1 region (Figure 5) using both Nestin-Clstn3 ( Figure 5) and Clstn3 tm1a/tm1a mice ( Figure 5-figure supplement 1). We found a significant decrease in the intensity of VGLUT1 puncta in most layers of hippocampal CA1 regions (stratum oriens [SO] and stratum radiatum [SR]), but not in the stratum lacunosum moleculare (SLM) layer, in both Clstn3-cKO ( Figures 5A and 5C) and Clstn3 tm1a/tm1a mice (Figure 5-figure supplement 1). In contrast, the intensity of GAD67 puncta was unchanged in all examined layers of hippocampal CA1 regions (SO, SR, stratum pyramidale [SP] and SLM) (Figures 5B, 5D and Figure 5-figure supplement 1). To complement these anatomical analyses, we performed whole-cell electrophysiological recordings of miniature excitatory and inhibitory postsynaptic currents (mEPSCs and mIPSCs) in brain slices from Nestin-Clstn3 and littermate WT mice ( Figure 5-figure supplement 2). Surprisingly, no differences in the amplitude or frequency of mEPSCs and mIPSCs were detected between Nestin-Clstn3 and Ctrl mice ( Figure 5-figure supplement 2). Thus, our data suggest that Clstn3 is specifically required for excitatory synapse structures, but not basal excitatory synaptic transmission, in CA1 hippocampal pyramidal neurons.

Clstn3 KO mice display increased anxiety-like behavior.
To explore the behavioral consequences of Clsnt3 loss in mice, we tested Clstn3-cKO mice for changes in select forms of learning and memory using 8-10-wk-old male Nestin-Clstn3 mice, together with control male littermates (Ctrl; Figure 6). We initially tested anxiety-like behaviors using the open-field (OF) test, elevated-plus maze (EPM) test, and light/dark t ransition (LDT) test. For the OF test, both Ctrl and Nestin-Clstn3 mice were allowed to freely explore an open field for 30 min, and then general activity was measured by calculating the velocity and total distance traveled. Anxiety-like behavior was assessed by calculating the frequency of entries into the center of the field and time spent in the center of the field.
We found no differences in these parameters in Nestin-Clstn3 mice compared with control mice (Figures 6A-E). There were also no differences between Nestin-Clstn3 mice and control mice in either the number of entries into open arms or the total time spent in open arms in the EPM test ( Figures 6F-H). However, in the LDT test, which measures the level of anxiety in response to bright light (unlike the EPM test, which measures the level of anxiety in response to exposure to an open space), there was a significant increase in anxiety-like behavior in Nestin-Clstn3 mice compared with control mice (both Nestin-Cre and Clstn3 tm1a/tm1a mice; data not shown), measured as the frequency of entering the brightly lit chamber and the total time spent in the lighted chamber (Figures 6I and 6J).
Nestin-Clstn3 mice displayed normal spatial working memory (measured by the Y-maze test) ( Figures 6K-M) and short-term memory (assessed by the novel object-recognition (NOR) test) ( Figures 6N-Q). Overall, these data suggest that Nestin-Clstn3 mice display moderately increased  (Figure 7-figure supplement 2). Taken together, these results suggest that Clstn3 organizes the development of hippocampal CA1 excitatory synapses, and likely acts through cadherin domains-mediated interactions with presynaptic Nrxns to control the properties of specific excitatory synaptic projections involving the SR layer.

Discussion
The present study was initiated with the goal of reconciling discrepancies surrounding molecular mechanisms of Clstn3-mediated synapse development. For this, we revisited the most critical (and controversial) issues using newly engineered Nrxns expression vectors and recombinant proteins, and a newly developed Clstn3-cKO mice. Taken together with our previous findings, the current study's three principal observations provide plausible explanations for discrepancies surrounding the molecular mechanisms of Clstn3-mediated synapse development.
First, we found that Clstn3 binds directly to β-Nrxns (Figures 1 and 2). This interaction was only detected using newly engineered Nrxn1β constructs that were expressed at ~100-fold higher levels Instead, we conclude that this sequence may influence the orientation and/or conformation of the LNS domain in a manner that is critical for its interaction with Clstn3. Structural studies have suggested that a β-Nrxn LNS domain containing an SS4 insertion exists as a dynamic equilibrium between two conformational states: one in which the splice insert forms a protruded αhelix that supports binding to cerebellins (Cblns), and one in which the additional residues adopt a βsheet conformation and restore binding to NLs and LRRTMs (Reissner et al., 2013). Because Clstn3 exhibits a slight preference for binding SS4-positive Nrxns, it is tempting to speculate that restoration of downstream residues promoted a conformational change in the SS4 insert, leading to Clstn3 binding. As noted above, the current study, unlike our previous study and that of Pettem et al., employed newly engineered Nrxn recombinant proteins to provide the first demonstration of direct interactions with recombinant Clsn3 proteins (Figure 2). Moreover, in keeping with results from the current study, it was previously reported that β1-Nrxn binds immobilized Clstn3 with a complex binding mode (Lu et al., 2014). Although it was claimed that Clstn3 interacts with α-Nrxns (but not β-Nrxns), and we previously failed to detect the interaction of Nrxns with Clstn3 (Pettem et al., 2013;Um et al., 2014), we would argue that conclusions based on binding experiments employing a limited set of methodologies could unintentionally be misleading, as was the case for our previous report (Um et al., 2014).
Second, re-expressing select Nrxn1 variants in Nrxn-deficient neurons rescued impaired Clstn3 synaptogenic activity (Figure 3). Previously, we reported that presynaptic Nrxns serve as functional receptors for postsynaptic Clstn3, but do not directly interact with them (Um et al., 2014). Our new data indicate that our previous model should incorporate direct binding of β-Nrxns to Clstn3 (Figures   1-3). In support of this interpretation, BAM-2, a Nrxn-related C. elegans homolog, was recently reported to bind CASY-1 and mediate neural circuit wiring of male-specific hook-sensory HOA neurons in C. elegans (Kim and Emmons, 2017). Intriguingly, CASY-1 does not directly interact with NRX-1, a canonical C. elegans Nrxn ortholog (Kim and Emmons, 2017). Although BAM-2 exhibits considerable sequence homology/similarity with α-Nrxns (Colavita and Tessier-Lavigne, 2003), whether these vertebrate and worm genes are functionally homologous remains to be determined.
Nrxns interact with various postsynaptogenic proteins, including NLs, LRRTMs, Cblns, and latrophilins (Südhof, 2017). These interactions are dynamically modulated by the alternative splicing status of Nrxns, mainly at the canonical SS4 splice site. Strikingly, the synapse-promoting activity of Clstn3 required Nrxn splice variants containing the SS4 insert (Figure 3), as has been shown for other Nrxn ligands (i.e., Cblns) (Südhof, 2017). These observations are consistent with slightly stronger binding of Nrxn1β +SS4 to Clstn3, which contrasts with the exclusive binding of Nrxn1β +SS4 to Cblns (Figures 1G and 1H). Pull-down experiments in mouse brains also showed that Clstn3 is primarily associated with Nrxn-SS4-positive variants in vivo and that its synaptogenic activity requires the binding to Nrxn-SS4-positive variants (Figures 3C and 3D) (Um et al., 2014). Additionally, the shed ectodomain of Clstn3 suppresses the synaptogenic activity of NL-2 and LRRTM2, suggesting that Clstn3 competes with these Nrxn ligands (Pettem et al., 2013). However, LRRTM2 interacts only with Nrxn-SS4-negative splice variants (Ko et al., 2009a), whereas Clstn3 activity requires Nrxn-SS4-positive splice variants; moreover, a β-Nrxn LNS domain (identical to the sixth LNS domainssixth LNS domain in α-Nrxns) is sufficient for Clstn3 binding. Although we still failed to detect clear interactions of α-Nrxns with Clstn3 in cell surface-binding assays (data not shown), we observed robust Nrxn1α-Clstn3 interactions in brain pull-down and direct binding assays (Figures 2E, 3C and   3D). It is possible that the surface-binding assays employed here were too stringent to observe this interaction, or the cell surface presentation of α-Nrxns was not optimal for binding. Thus, we conclude that Clstn3 binds to both αand β-Nrxns through a common LNS domain, but a subset of Clstn3 synaptic functions might require other unidentified ligands in vivo. Moreover, given the activity-dependent regulation of Nrxn alternative splicing at the SS4 site (Iijima et al., 2011), future studies should investigate whether Clstn3-Nrxn interactions could be fine-tuned depending on distinct activity patterns in vivo. Furthermore, given a recent report that different SS4-positive splice variants of two different Nrxns (Nrxn1 vs. Nrxn3) differentially control different postsynaptic responses (Dai et al., 2019), it is possible that Clsnt3 preferentially partners with a specific Nrxn in vivo.
Third, re-expressing Clstn3 cadherin domains was sufficient to completely rescue the impaired excitatory synapse structures in hippocampal CA1 neurons from both Clstn3 tm1a/tm1a and Nestin-Clstn3 mice (Figure 7 and Figure 5-figure supplement 1), but only in the SR layer (but see Figure 7figure supplement 2). In contrast, the structural deficits in other hippocampal CA1 layers were rescued by expression of full-length Clstn3 protein, but not by expression of partial Clstn3 proteins (Figure 7 and Figure 5-figure supplement 1). Although it is possible that other unidentified proteins could also bind to Clstn3 cadherin domains, it is likely that presynaptic neurexins and postsynaptic Clstn3 control the properties of excitatory synapse development in specific Schaffercollateral projections within the SR layer. A recent paper highlighted the fact that the distinctive features of two different Schaffer-collateral projections differentially regulate mushroom spine density and high-magnitude LTP in the SO layer, organized by heterophilic type II cadherins (Basu et al., 2017). Moreover, Nrxn genes show differential, but overlapping, isoform-and region-dependent expression in different classes of neurons, and undergo highly distinctive, cell type-specific alternative splicing (Nguyen et al., 2016;Ullrich et al., 1995). One plausible scenario would be that CA3 neurons projecting to CA1 neurons in the SR layer, but not the SO layer, express higher levels of SS4-positive Nrxns at their nerve terminals. A recent extensive chromogenic and fluorescent in situ hybridization analysis showed that the levels of Nrxn mRNAs in hippocampal subfields are significantly lower in GABAergic neurons than excitatory neurons (Uchigashima et al., 2019), which might account for preferential deficits on excitatory synapses in CA1 hippocampal regions of Clstn3 KO mice. Another possible scenario is that major glutamatergic axon fibers of Schaffer-collateral (SC) pathways are targeted to CA1 pyramidal neurons in the SR layer, whereas a subpopulation of axon fibers of SC pathways innervate both bistratified GABAergic interneurons in the SO layer and CA1 pyramidal neurons. Additional studies will also be required to determine the identity of neurons in vivo that are responsible for Nrxn-Clstn3 interactions in other neural circuits inside and outside of the hippocampus. Notably, in stark contrast to the previous report that decreased inhibitory synapse structure and transmission in Clstn3-KO mice (Pettem et al., 2013), we found that inhibitory synapses in CA1 hippocampal layers and the cortex were morphologically normal in our Clstn3-KO mice ( Figure 5 and Figure 5-figure supplement 1). Moreover, to our surprise, Clstn3-KO mice did not exhibit any alterations in basal synaptic transmission at either excitatory or inhibitory synapses ( Figure 5-figure supplement 2), suggesting that Clstn3 is required for structural integrity, but not basal synaptic transmission, in hippocampal CA1 neurons. The reason for this discrepancy is currently unclear, but it is likely that Clstn family proteins are functionally redundant in the maintenance of basal synaptic transmission; alternatively, Clstn3 may be specifically required for certain forms of synaptic plasticity. A notable difference between the study of Pettem et al. and the current study is that the former targeted exons 2 and 3 of the mouse Clstn3 gene, whereas our study targeted exon 8 ( Figure 4A). In addition, Pettem et al employed an EIIa-Cre driver line that permits germ line deletion in both excitatory and inhibitory neurons by driving expression of Cre recombinase in the early mouse embryo, whereas we used a pan-neuronal Nestin-Cre driver line that is presumed to be similar to EIIa-Cre. Because Clstn3 is expressed in both excitatory and inhibitory neurons (Figure   4-figure supplement 2), it is possible that specifically deleting Clstn3 in GABAergic inhibitory neurons may produce marked deficits in GABAergic synapse development, as previously documented (Pettem et al., 2013), but this did not clearly manifest in our Clstn3-cKO mice. A more systematic, follow-up investigation to probe cell type-specific contributions of Clstn3 to distinct anatomical and electrophysiological phenotypes should address these important, but puzzling, observations. Future studies should also probe how Nrxn-Clstn3 complexes mediate synaptic specificity involving the SR layer of the hippocampal CA1 region. Furthermore, rigorous analyses are warranted to address whether and how this circuit specificity is related to the altered anxiety-like behavior observed in Clstn3-cKO mice (Figure 6). A complete understanding of how Clstn3 functions will also require investigation of the potential contributions of other Clstn3 domains (e.g., LNS domain, intracellular region) to the functions of Clstn3, as was similarly shown in C. elegans (Ikeda et al., 2008;Kim and Emmons, 2017;Thapliyal et al., 2018).
In summary, the present study unambiguously establishes that Clstn3 plays a role in specifying the properties of a specific hippocampal CA1 neural circuit, in part by regulating excitatory synapse development through formation of physical complexes with specific Nrxn splice variants.

Materials and methods
Plasmids. Nrxn rescue vectors were generated by PCR-amplification of full-length sequences of mouse Nrxn1β -SS4 , Nrxn1β +SS4 , rat Nrxn1β -SS4 , rat Nrxn1β +SS4 , bovine Nrxn1α -SS4 and bovine Nrxn1α +SS4 , followed by digestion with NheI and BsrGI and cloning into the TKD vector [L-313 vector]. Three nucleotides (underlined) in the GTGCCTTCCTCTATGACAACT sequence were then mutated to render it shRNA-resistant. pGW1-FLAG-mNrxn1β -SS4 and pGW1-FLAG-mNrxn1β +SS4 were generated by PCR-amplification of full-length mNrxn1β -SS4 and mNrxn1β +SS4 , respectively, digestion with KpnI and EcoRI, and cloning into a modified pGW1 vector containing a mouse Clstn1 signal peptide and FLAG epitope (pGW1 vector; British Biotechnology, Oxford, UK). pCMV-IgC-mNrxn1β -SS4 and pCMV-IgC-mNrxn1β +SS4 were generated by PCR-amplification of the indicated extracellular regions of mNrxn1β -SS4 (aa 1-359) and mNrxn1β +SS4 (aa 1-389), respectively, followed by digestion with EcoRI and SalI and cloning into a pCMV-IgC vector. pGW1-FLAG-Clstn3 and pGW1-FLAG-Clstn3-ΔCAD were generated by PCR-amplification of full length mouse Clstn3 and a Clstn3 fragment (aa 258-956), respectively, followed by digestion with KpnI and EcoRI and cloning into the pGW1-FLAG vector. pGW1-FLAG-Clstn3-Cad used in Figure 2A was generated by PCR amplification of two Clstn3 fragments (aa 29-252 and aa 848-956), followed by digestion with KpnI and EcoRI for the first fragment and EcoRI only for the second fragment, and subsequent cloning into the pGW1-FLAG vector. The following deletion constructs of mNrxn1β were generated using pGW1-FLAG-mNrxn1β +SS4 : pGW1-FLAG-mNrxn1β-ΔLNS6 was generated by PCR amplification of mNrxn1β (aa 293-468), followed by digestion with KpnI and EcoRI and cloning into the pGW1-FLAG vector. pGW1-FLAG-mNrxn1β-ΔStalk1 was generated by PCR amplification of two mNrxn1β fragments (aa 1-315 and aa 347-468), followed by digestion with KpnI and EcoRI for the first fragment and EcoRI for the second fragment, and subsequent cloning into the pGW1-FLAG vector.

Animals. Clstn3 tm1a(EUCOMM)Hmgu mice, in which a targeting cassette harboring FRT, lacZ and loxP
sites is inserted between exon 7 and exon 8, resulting in a 'knockout-first' lacZ reporter-tagged

Production of recombinant lentiviruses and adeno-associated viruses (rAAVs). 1. Lentivirus
production Recombinant lentiviruses were produced as previously described (Lee et al., 2013). In brief, HEK293T cells were transfected with three plasmids-Lentivirus vectors, psPAX2, and pMD2G-at a 2:2:1 ratio using Lipofectamine 2000 (Thermo-Fisher Scientific) according to the manufacturer's protocol. After 72 h, lentiviruses were harvested by collecting the medium from transfected HEK293T cells and briefly centrifuging at 1,000 × g to remove cellular debris. Filtered media containing 5% sucrose were centrifuged at 117,969 × g for 2 h; supernatants were then removed and the virus pellet was washed with ice-cold phosphate-buffered saline (PBS) and resuspended in 80 μl PBS.

AAV production HEK293T cells were co-transfected with the indicated AAV vectors and pHelper
and pRC1-DJ vectors. Seventy-two hours later, transfected HEK293T cells were collected, lysed, and mixed with 40% polyethylene glycol and 2.5 M NaCl, and centrifuged at 2000 × g for 30 min. The cell pellets were resuspended in HEPES buffer (20 mM HEPES; 115 mM NaCl, 1.2 mM CaCl2, 1.2 mM MgCl2, 2.4 mM KH2PO4) and an equal volume of chloroform was added. The mixture was centrifuged at 400 × g for 5 min, and concentrated three times with a Centriprep centrifugal filter (Millipore) at 1,220 × g for 5 min each and with an Amicon Ultra centrifugal filter (Millipore) at 16,000 × g for 10 min. Before titering AAVs, contaminating plasmid DNA was eliminated by treating 1 μl of concentrated, sterile-filtered AAVs with 1 μl of DNase I (Sigma-Aldrich) for 30 min at 37°C.
After treatment with 1 μl of stop solution (50 mM ethylenediaminetetraacetic acid) for 10 min at 65°C, 10 μg of protease K (Sigma-Aldrich) was added and AAVs were incubated for 1 h at 50°C. Reactions were inactivated by incubating samples for 20 min at 95°C. The final virus titer was quantified by qRT-PCR detection of EGFP sequences and subsequent reference to a standard curve generated using the pAAV-U6-EGFP plasmid. All plasmids were purified using a Plasmid Maxi Kit (Qiagen GmbH).

LC-MS/MS protein analysis.
Peptides were analyzed using a nanoflow LC-MS/MS system consisting of an Easy nLC 1000 system (Thermo-Fisher) and an LTQ Orbitrap Elite mass spectrometer (Thermo-Fisher) equipped with a nano-electrospray source. Peptide solutions (5-µl aliquots) were loaded onto a C18 trap column (20 × 75 µm, 3 µm particle size; Thermo-Fisher) using an autosampler. Peptides were desalted and concentrated on the column at a flow rate of 5 µl/min. Trapped peptides were then separated on a 150-mm custom-built microcapillary column consisting of C18 (particle size, 3 µm; Aqua Science, Yokohama, Japan) packed in 100-µm silica tubing with a 6µm inner diameter orifice. The mobile phases A and B were composed of 0% and 100% acetonitrile, respectively, and each contained 0.1% formic acid. The LC gradient began with 5% B for 5 min and was increased to 15% B over 5 min, 50% B over 55 min, and 95% B over 5 min, then remained at 95% B over 5 min, followed by 5% B for an additional 5 min. The column was re-equilibrated with 5% B for 15 min between runs. A voltage of 2.2 kV was applied to produce the electrospray. In each mass analysis duty cycle, one high-mass resolution (60,000) MS spectrum was acquired using the orbitrap analyzer, followed by 10 data-dependent MS/MS scans using the linear ion trap analyzer. For MS/MS analysis, a normalized collision energy (35%) was used throughout the collision-induced dissociation phase. All MS and MS/MS spectra were acquired using the following parameters: no sheath and auxiliary gas flow; ion-transfer tube temperature, 200°C; activation Q, 0.25; and activation time, 20 ms. Dynamic exclusion was employed with a repeat count of 1, a repeat duration of 30 s, an exclusion list size of 500, an exclusion duration of 60 s, and an exclusion mass width of ± 1.5 m/z.

MS data analysis.
MS/MS spectra were analyzed using the following analysis protocol, referencing the UniProt mouse database (09-15-2015 release). Briefly, each protein's reversed sequence was appended onto the database to calculate the false discovery rate. Peptides were identified using ProLuCID (Xu et al., 2015) in Integrated Proteomics Pipeline software, IP2 (http://www.integratedproteomics.com), with a precursor mass error of 25 ppm and a fragment ion mass error of 600 ppm. Trypsin was used as the protease, and two potential missed cleavages were allowed. Carbamidomethylation at cysteine was chosen as a static modification, and methionine oxidation was chosen as a variable modification. Protein lists consisting of two or more peptide assignments for protein identification (false-positive rate < 0.01) were prepared by filtering and sorting output data files.

Primary neuron culture, immunocytochemistry, image acquisition, and quantitative analyses.
The indicated analyses were performed using cultured, E18-derived, rat hippocampal neurons and confocal microscopy, as previously described Um et al., 2014a;Um et al., 2014b).
Rat hippocampal neurons were prepared from E18 rat brains and cultured on coverslips coated with poly-D-lysine in Neurobasal media supplemented with B-27 (Thermo-Fisher), 0.5% fetal bovine serum, 0.5 mM GlutaMax (Thermo-Fisher), and sodium pyruvate (Thermo-Fisher). For immunocytochemistry, cultured neurons were fixed with 4% formaldehyde/4% sucrose, permeabilized with 0.2% Triton X-100 in PBS, and immunostained with primary antibodies and Cy3-or fluorescein isothiocyanate (FITC)-conjugated secondary antibodies (Jackson ImmunoResearch, West Grove, PA, USA). Images were acquired using a confocal microscope (LSM700; Carl Zeiss) equipped with a 63× objective lens. All image settings were kept constant. Z-stack images were converted to maximal projection. All images were quantitatively analyzed in a blinded manner using MetaMorph software (Molecular Devices).
Cell-surface-binding assays. Ig-fusion proteins of Clstn3, Ig-Nrxn1β splice variants, and IgC alone (Control) were produced in HEK293T cells. Soluble Ig-fused proteins were purified using protein A-Sepharose beads (GE Healthcare). Bound proteins were eluted with 0.1 M glycine (pH 2.5) and immediately neutralized with 1 M Tris-HCl (pH 8.0). Transfected HEK293T cells expressing the indicated plasmids were incubated with 10 μg/ml Ig-fused proteins for 2 h at 37°C. Images were acquired using a confocal microscope (LSM700; Carl Zeiss).
Heterologous synapse-formation assays. Heterologous synapse-formation assays were performed using recombinant Clstn3 fusion proteins as previously described (Ko et al., 2009). Briefly, HEK293T cells were transfected with EGFP (negative control) or the indicated Clstn3 constructs using Lipofectamine 2000 (Thermo-Fisher). After 48 h, the transfected HEK293T cells were trypsinized, seeded onto in vitro day10 (DIV10) hippocampal neurons, co-cultured for an additional 72 h, and double-immunostained on DIV13 with antibodies against EGFP, HA, and the indicated synaptic markers (synapsin, VGLUT1, or GAD67). All images were acquired using a confocal microscope (LSM700; Zeiss). For quantification, the contours of transfected HEK293T cells were chosen as the region of interest (ROI). Fluorescence intensities of synaptic markers in each ROIs were quantified for both red and green channels using MetaMorph software (Molecular Devices). Normalized synapse density on transfected HEK293T cells was expressed as the ratio of red to green fluorescence.

Semi-quantitative immunoblot analysis.
For semiquantitative immunoblot analyses, brains from P42 WT, Clstn3 tm1a/ tm1a , Clstn3 fl/fl , or Clstn3 fl/fl ::Nestin-Cre (Nestin-Clstn3) mice were homogenized in 0.32 M sucrose/1 mM MgCl2 containing a protease inhibitor cocktail (Thermo-Fisher Scientific) using a Precellys Evolution tissue homogenizer (Bertin Co.). After centrifuging homogenates at 1,000 × g for 10 min, the supernatant was transferred to a fresh microcentrifuge tube and centrifuged at 15,000 × g for 30 min. The resulting synaptosome-enriched pellet (P2) was resuspended in lysis buffer and centrifuged at 20,800 × g, after which the supernatant was analyzed by Western blotting. Quantitation was performed using ImageJ software (National Institutes of Health). Nissl staining. WT and Clstn3 tm1a/tm1a mice were perfused first with PBS and then with 4% paraformaldehyde by cardiac injection. Fixed brain tissue was isolated, post-fixed for 12 hours at 4°C, and dehydrated in 30% sucrose for 6-8 days. Thereafter, brain tissue was embedded in OCT (Optimal Cutting Temperature) compound and stored at -80°C. The frozen tissue was mounted and cut at a thickness of 20 μm using a cryostat (Leica CM5120). Slices were mounted on glass slides, washed three times with PBS (15 min each), and permeabilized with 0.1% Triton X-100 in PBS for 10 min.  Slides were imaged with an LSM700 microscope (Zeiss) and analyzed using MetaMorph software (Molecular Devices).

Immunohistochemistry. WT and
Stereotaxic injection of rAAVs. Adult (~8-wk-old) male Clstn3 −/− or WT littermate mice (C57BL/6 strain) were anesthetized with Avertin (400 mg/kg body weight) by intraperitoneal injection. rAAV solutions (titers ≥ 1 × 10 11 viral genomes/ml) were injected with a NanoFil syringe (World Precision Instruments) at a flow rate of 0.1 μl/min. The coordinates used for the CA1 region of the dorsal hippocampus were AP -2.5 mm, ML ± 1.5 mm, DV +1.5 mm (from the dura). The site at DV +1.5 mm received a 1-μl injection. Injected mice were allowed to recover for at least 14 d following surgery prior to use in experiments.
Electrophysiology. Electrophysiological recordings were performed in acute hippocampal CA1 slices.

Mouse behaviors.
Behavioral experiments were performed in the following order: open-field test, Ymaze test, novel object-recognition test, light/dark transition test, and elevated plus-maze test. All behavioral analyses were performed using male mice. 1. Y-maze test. A Y-shaped white acrylic maze with three 40-cm-long arms at a 120° angle from each other was used. Mice were introduced into the center of the maze and allowed to explore freely for 8 min. An entry was counted when all four limbs of a mouse were within the arm. The movement of mice was recorded by a top-view infrared camera, and analyzed using EthoVision XT 10 software. 2. Open field test. Mice were placed into a white acrylic open-field box (40 × 40 × 40 cm), and allowed to freely explore the environment for 30 min under low-light conditions (~30 lux). The traveled distance moved and time spent in the center zone by freely moving mice were recorded by a top-view infrared camera, and analyzed using EthoVision XT 10 software (Noldus). 3. Novel object-recognition test. An open-field maze was used in this test.
Mice were habituated to the maze for 10 min. For training sessions, two identical objects were placed in the center of the maze at regular intervals, and mice were allowed to explore the objects for 5 min.
After the training session, mice were returned to their home cage for 24 h. For novel objectrecognition tests, one of the two objects was exchanged for a new object, placed in the same position of the maze. Mice were returned to the maze and allowed to explore freely for 5 min. The movement of mice was recorded by infrared camera, and the number and duration of contacts were analyzed using EthoVision XT 10 (Noldus). 4. Elevated plus-maze test. The elevated plus-maze is a plusshaped (+) white acrylic maze with two open arms (30  5  0.5 cm) and two closed arms (30  5  30 cm) positioned at a height of 75 cm from the floor. Light conditions around open and closed arms were ~300 and ~30 lux, respectively. For the test, mice were introduced into the center zone of the elevated plus-maze and allowed to move freely for 10 min. All behaviors were recorded by a top-view infrared camera, and the time spent in each arm and the number of arm entries were measured and analyzed using EthoVision XT 10 software (Noldus). 5. Light/Dark transition test. The light/dark box consists of an open-roof, white (light) chamber conjoined to a closed black (dark) chamber with a small entrance to allow free movement between the two chambers. The light chamber was illuminated at 350 lux. The time spent in each chamber and the number of entries were measured and analyzed using EthoVision XT 10 software (Noldus).
Statistics. All data are expressed as means ± standard errors of the mean (SEM), and significance is indicated with asterisk (compared with a value from control group) or hashtag (compared with a value from experimental group). All experiments were performed using at least three independent cultures and the normality of data distributions was evaluated using the Shapiro-Wilk test, followed by non-parametric Kruskal-Wallis test with Dunn's multiple-comparison test for post hoc group comparisons, using cell numbers or the number of experiments as the basis for 'n'.   mNrxn1α +SS4 and mNrxn1α -SS4 encode full-length mouse Nrxn1α with and without an SS4 insert, respectively. bNrxn1α +SS4 and bNrxn1α -SS4 encode full-length bovine Nrxn1α with or without an SS4 insert, respectively. mNrxn1β +SS4 and mNrxn1β -SS4 similarly encode full-length mouse Nrxn1β with or without an SS4 insert, respectively. rNrxn1β +SS4 and rNrxn1β -SS4 encode full-length rat Nrxn1β with or without an SS4 insert, respectively. mNrxn1α and mNrxn1β constructs were cloned into the pCAGGS vector, whereas bNrxn1α and rNrxn1β constructs, which have long been used to characterize Nrxnbinding ligands, were generated in the Südhof laboratory. Equal amounts of expression plasmids were transfected into HEK293T cells. Serial dilutions of lysates of mNrxn1-expressing cells were required to directly compare expression levels of the indicated Nrxn expression constructs. Note that expression levels of pCAGG mouse Nrxn1 expression plasmids were ~100-fold higher than those used previously.