Specific Neuroligin3–αNeurexin1 signaling regulates GABAergic synaptic function in mouse hippocampus

Synapse formation and regulation require signaling interactions between pre- and postsynaptic proteins, notably cell adhesion molecules (CAMs). It has been proposed that the functions of neuroligins (Nlgns), postsynaptic CAMs, rely on the formation of trans-synaptic complexes with neurexins (Nrxns), presynaptic CAMs. Nlgn3 is a unique Nlgn isoform that localizes at both excitatory and inhibitory synapses. However, Nlgn3 function mediated via Nrxn interactions is unknown. Here we demonstrate that Nlgn3 localizes at postsynaptic sites apposing vesicular glutamate transporter 3-expressing (VGT3+) inhibitory terminals and regulates VGT3+ inhibitory interneuron-mediated synaptic transmission in mouse organotypic slice cultures. Gene expression analysis of interneurons revealed that the αNrxn1+AS4 splice isoform is highly expressed in VGT3+ interneurons as compared with other interneurons. Most importantly, postsynaptic Nlgn3 requires presynaptic αNrxn1+AS4 expressed in VGT3+ interneurons to regulate inhibitory synaptic transmission. Our results indicate that specific Nlgn–Nrxn signaling generates distinct functional properties at synapses.


Introduction
In central synapses, cell adhesion molecules (CAMs) are major players in trans-synaptic interactions (de Wit and Ghosh, 2016) that serve a primary role in initiating synapse formation by directing contact between axonal and dendritic membranes. Emerging evidence suggests that trans-synaptic interactions are also important for synapse identity, function, plasticity, and maintenance (Biederer et al., 2017;Campbell and Tyagarajan, 2019;Südhof, 2017). Numerous CAM variants exist due to large gene families and alternative splicing, generating a vast array of possible combinations of pre-and postsynaptic CAMs. Although some specific trans-synaptic interactions of CAMs have been reported to underlie distinct synaptic properties (Chih et al., 2006;Fossati et al., 2019;Futai et al., 2013), elucidating synaptic CAM complexes that dictate synapse identity and function remains a major challenge.
Nrxn-Nlgn interactions depend on Nrxn protein length (long form [a] vs short form [b]), splice insertions at AS4 of Nrxns, and splice insertions of Nlgns. For example, Nlgn1 splice variants that have splice insertions at site B have higher binding affinities for bNrxn1-AS4 (bNrxn1 lacking alternative splice insertion at AS4) than for bNrxn1+AS4 (containing an alternative splice insertion at AS4) (Boucard et al., 2005;Koehnke et al., 2010;Reissner et al., 2008). However, it is largely unknown which Nrxn-Nlgn combination defines specific synapse functionality. We recently found that Nlgn3D, which lacks both of the A1 and A2 alternative splice insertions, is the major Nlgn3 splice isoform expressed in hippocampal CA1 pyramidal neurons and regulates both excitatory and inhibitory synaptic transmission (Uchigashima et al., 2020). However, to the best of our knowledge, the synapses at which Nlgn3D interacts with presynaptic Nrxn isoform(s) have not been identified.
In the present study, we show that Nlgn3 is selectively enriched at vesicular glutamate transporter 3-expressing (VGT3+) Cck+ inhibitory terminals in the hippocampal CA1 area. Gain-of-function and loss-of-function studies revealed that Nlgn3 regulates VGT3+ interneuron-mediated inhibitory synaptic transmission. Importantly, the effect of Nlgn3 on VGT3+ synapses was hampered by the deletion of all Nrxn genes in VGT3+ interneurons and rescued by the selective expression of aNrxn1+AS4 in VGT3+ interneurons. These results suggest that the trans-synaptic interaction between aNrxn1+AS4 and Nlgn3 underlies the input cell-dependent control of VGT3+ GABAergic synapses in the hippocampus.

Results
Nlgn3 is enriched at VGT3+ GABAergic synapses in the hippocampal CA1 region In a recent study, we demonstrated that Nlgn3 localizes at and regulates both inhibitory and excitatory synapses in the hippocampal CA1 area (Uchigashima et al., 2020). However, the distribution of Nlgn3 at different types of inhibitory synapses has not yet been addressed. Therefore, we first examined which GABAergic inhibitory synapses express Nlgn3 in the CA1 area by immunohistochemistry. Cck+, Pv+, and Sst+ interneurons are the primary inhibitory neurons in the hippocampus. Moreover, the cell bodies and dendritic shafts of CA1 pyramidal cells are targeted by Cck+ and Pv+, and Sst+ interneurons, respectively (Pelkey et al., 2017). Our Nlgn3 antibody with specific immunoreactivity was validated in Nlgn3 KO brain (Figure 1-figure supplement 1A and B) and displayed a typical membrane protein distribution pattern in the hippocampus, as we recently reported ( Figure 1A; Uchigashima et al., 2020). Inhibitory synapses expressing Nlgn3 were identified by colocalization of Nlgn3 signals with vesicular inhibitory amino acid transporter (VIAAT) signals. Four different inhibitory axons/terminals were visualized by anti-VGT3 and -CB1 (markers for Cck+ interneurons) (Früh et al., 2016), -Pv, and -Sst antibodies. We found that signal intensities for Nlgn3 were considerably high at GABAergic synapses co-labeled with VGT3 or CB1 ( Figure 1B,C, and F) and low or at approximately noise levels at those co-labeled with Pv ( Figure 1D and F) or Sst ( Figure 1E and F) in the CA1 region. Noise levels were obtained from images that lacked true close apposition of signals for Nlgn3 and synaptic markers observed by rotating the Nlgn3 channel 90( Figure 1F and Figure 1-figure supplement 1G). Moreover, co-localization of Nlgn3 signals with these markers demonstrated similar findings ( Figure 1G). Therefore, these data strongly suggest that Nlgn3 is preferentially recruited to Cck+ GABAergic synapses, but not to Pv+ or Sst+ inhibitory synapses.

Nlgn3 regulates inhibitory synaptic transmission at VGT3+ GABAergic synapses
To determine whether Nlgn3 has specific roles at VGT3+ inhibitory synapses, we assessed the effect of overexpressing Nlgn3D, the major Nlgn3 splice isoform expressed in CA1 pyramidal neurons (Uchigashima et al., 2020), on input-specific inhibitory transmission. To distinguish a subset of GABAergic synapses and evoke cell-specific synaptic transmission, we generated three cell type-specific fluorescent lines by crossing VGT3-Cre, Sst-Cre, or Pv-Cre with a TdTomato (RFP) reporter line, producing respectively, VGT3/RFP, Sst/RFP, and Pv/RFP mouse lines. TdTomato-expressing cells in each of the three fluorescent mouse lines were distributed in the CA1 in a layer-dependent manner (Figure 2-figure supplement 1). We evaluated the effect of Nlgn3D overexpression (OE) on unitary inhibitory postsynaptic currents (uIPSCs) by triple whole-cell recordings using organotypic slice cultures from each mouse line. Two to three days after transfection of Nlgn3D or enhanced green fluorescent protein (EGFP) control by biolistic gene gun, current-and voltage-clamp recordings were conducted from a presynaptic RFP+ interneuron and postsynaptic EGFP or EGFP/Nlgn3D-positive and -negative postsynaptic pyramidal neurons, respectively ( Figure 2A). RFP interneurons expressing VGT3 in the pyramidal cell layer and stratum (st.) oriens and radiatum, Pv in the st. pyramidale and Sst in the st. oriens, were chosen (Figure 2-figure supplement 1). uIPSCs were evoked by inducing APs in RFP+ neurons. The amplitude and paired-pulse ratio (PPR), monitoring release probability, of uIPSCs and synaptic connectivity were compared between Nlgn3D-transfected anduntransfected neurons ( Figure 2B-E). Importantly, VGT3+ inhibitory interneurons displayed clear potentiation of uIPSCs onto CA1 pyramidal neurons overexpressing Nlgn3D ( Figure 2C). Paired APs of VGT3+ neurons with short intervals (50 ms) induced paired-pulse depression (PPD) of uIPSCs. Nlgn3D displayed reduced PPD compared with untransfected neurons, consistent with previous work (Futai et al., 2007;Shipman et al., 2011;Uchigashima et al., 2020; Figure 2D). As PPR inversely correlates with presynaptic release probability, these results suggest that Nlgn3D OE can facilitate presynaptic GABA release. In contrast, Nlgn3D OE reduced uIPSCs in Pv+ inhibitory synaptic transmission, but had no effect on PPR as reported previously (Figure 3A-D; Horn and Nicoll, 2018). Lastly, Nlgn3D OE did not alter uIPSCs or PPR mediated by Sst+ interneurons ( Figure 3E-H). No effect of biolistic transfection with EGFP alone was found on uIPSC amplitude, PPR, or , and Sst (E). The boxed area in low magnification images is enlarged in lower panels. Arrowheads indicate Nlgn3 immunofluorescent puncta associated with (yellow) or distant from (white) interneuron markers. (F and G) Summary of the relative intensity (F) and co-localization frequency (G) of Nlgn3 immunofluorescent signals at different inhibitory synapses. Plots are obtained from each synapse for intensities (F) or image for co-localizations (G). Noise levels for the intensity and colocalization were obtained from images with the Nlgn3 channel rotated 90˚(90˚rotation). ***p<0.001; n.s. not significant; One-way ANOVA with Sidak's post hoc test. Bars on each column represent mean ± SEM. Scale bars, 100 mm (A) and 2 mm (B-E). The online version of this article includes the following figure supplement(s) for figure 1: connection probability at Pv+ and Sst+ inhibitory synapses. The above results strongly suggest that Nlgn3 modifies inhibitory synaptic function depending on the type of presynaptic interneuron with which it interacts. Interestingly, Nlgn3D OE did not increase synaptic connectivity ( Figure 2E), suggesting that (i) Nlgn3 regulates pre-existing inhibitory inputs on postsynaptic neurons and/or (ii) postsynaptic Nlgn3D OE is not sufficient to induce new synapse formation.
Overexpression of postsynaptic Nlgn3A2, a Nlgn3 splice isoform including the A2 cassette, in CA1 pyramidal neurons has been reported to differentially regulate Pv+ and Sst+ inhibitory synapses (Horn and Nicoll, 2018). Horn and Nicoll reported that human Nlgn3A2 OE reduces Pv+ and increases Sst+ inhibitory synaptic transmission, which is inconsistent with our findings in Sst+ synapses ( Figure 3F). This suggests that the signaling interaction between the input neuron and different Nlgn3 splice isoforms may generate distinct inhibitory regulatory mechanisms. In addition, we also   reported that Nlgn3A2 OE increases evoked inhibitory synaptic transmission in the CA1 region (Uchigashima et al., 2020). Therefore, we tested the effect of mouse Nlgn3A2 OE on inhibitory synaptic transmission at Pv+, Sst+, and VGT3+ synapses (Figure 3-figure supplement 1). To our surprise, mouse Nlgn3A2 did not potentiate VGT3+ and Sst+ inhibitory synapses ( Figure 3-figure supplement 1A-H). Nlgn3A2 OE reduced uIPSCs at Pv+ synapses, like Nlgn3D ( Figure 3I-L). These findings suggest that Nlgn3A2 regulates inhibitory synapses through different interneuron type(s).

Nlgn3 knockdown reduces VGT3+ inhibitory synaptic transmission in the hippocampal CA1 region
We next tested the impact of acute Nlgn3 knockdown (KD) on inhibitory synaptic transmission. Organotypic slice cultures prepared from C57BL/6J mice were biolistically transfected with shRNA against Nlgn3 (shNlgn3#1), which exhibits over 90% KD efficiency specific to Nlgn3 isoforms ( Figure 4C,G, and K) and connection probability ( Figure 4D,H, and L) between transfected and untransfected neurons. Taken together, these data suggest that Nlgn3 is required for synaptic transmission specifically at VGT3+ GABAergic synapses in an input cell-dependent manner.

Lack of Nrxn genes in presynaptic VGT3+ neurons abolishes the effect of postsynaptic Nlgn3D OE
Our results suggest that Nlgn3 is preferentially located at VGT3+ inhibitory synapses and regulates inhibitory synaptic transmission. Postsynaptic Nlgns couple with presynaptic Nrxns to form trans-synaptic protein complexes that regulate synapse formation and function (Südhof, 2017). Does inputspecific Nlgn3D ( Figure 2C) require presynaptic Nrxn proteins? To address this question, we generated VGT3 neuron-specific Nrxn triple knockout (TKO) with TdTomato reporter gene mouse line (Nrxn1/2/3 f/f /VGT3-Cre/TdTomato: NrxnTKO/VGT3/RFP). This mouse line is fertile with KO of Nrxn1, 2, and 3 specifically in TdTomato-positive VGT3+ neurons ( Figure 5A). First, we assessed the impact of Nrxn TKO on VGT3+ synaptic transmission. Pre-and postsynaptic dual whole-cell recordings were performed between RFP-positive VGT3+ interneurons in CA1 st. pyramidale and nearby CA1 pyramidal neurons ( Figure 5B-E). NrxnTKO/VGT3/RFP mice displayed reduced uIPSC, PPR at 25 ms inter-pulse interval, and connectivity compared with wild-type (WT) VGT3/RFP mice ( Figure 5C-E). Intrinsic excitability was comparable between WT and NrxnTKO VGT3+ neurons (Figure 5-figure supplement 1). These results suggest that Nrxns in VGT3+ interneurons regulate synaptic transmission without changing intrinsic membrane properties. Next, we transfected Nlgn3D in NrxnTKO/VGT3/RFP slice cultures and performed triple whole-cell recordings as described above using VGT3/RFP slice cultures ( Figure 2B-E). VGT3+ interneurons lacking Nrxns induced synaptic release regardless of Nlgn3D gene transfection, indicating that presynaptic Nrxns in VGT3+ interneurons are not essential for synaptogenesis. Importantly, we observed no enhancement of uIPSC amplitude in Nlgn3-overexpressed neurons compared with untransfected pyramidal neurons ( Figure 5F-I). These results strongly suggest that presynaptic Nrxn proteins are necessary for regulating inhibitory synaptic transmission through postsynaptic Nlgn3.
aNrxn1 and bNrxn3 mRNAs are highly expressed in VGT3+ interneurons Our results above clearly suggest that presynaptic Nrxn proteins are important for the function of postsynaptic Nlgn3D. We therefore hypothesized that Nrxn isoforms highly expressed in VGT3+ interneurons functionally couple with postsynaptic Nlgn3D. To address this hypothesis, we examined the mRNA expression patterns of a and b isoforms of Nrxn1-3 in hippocampal CA1 interneurons by    (FISH). The specificities of cRNA probes for 6 Nrxn isoforms were validated recently (Uchigashima et al., 2019). mRNAs encoding all Nrxn isoforms, except bNrxn1, were detected not only in the st. pyramidale but also in scattered cells across all layers ( Figure 6figure supplement 1A-F). Importantly, aNrxn1 and bNrxn3 mRNAs appeared to be enriched in scattered cells within the st. radiatum or pyramidale (arrows in Figure 6A-C and G-I), where VGT3+ interneurons are dominantly distributed compared with Pv+ and Sst+ interneurons (Pelkey et al., 2017). Double FISH signals were twice as strong for aNrxn1 and bNrxn3 mRNAs in VGT3+ ( Figure 6A,G, and J) interneurons than in Pv+ ( Figure 6B,H, and J) and Sst+ ( Figure 6C,I, and J) interneurons. In contrast, there were no differences in the signal intensities for the remaining Nrxn isoforms between VGT3+ and other interneurons ( Figure 6D-F and J and Figure 6-figure supplement 1G-P). These findings suggest that aNrxn1 and bNrxn3 mRNAs are highly expressed in VGT3 + interneurons compared with Pv+ or Sst+ interneurons.
Our rescue approach suggests that aNrxn+AS4 regulates inhibitory synaptic transmission with Nlgn3D. The structures of a and bNrxns are similar; therefore, different aNrxn+AS4 isoform(s) may    be able to functionally substitute aNrxn1+AS4. Although our FISH results demonstrated comparable levels of aNrxn3 among VGT3+, Pv+, and Sst+ interneurons ( Figure 6D-F and J), our single-cell RNA sequencing data indicate that aNrxn3+AS4 is another dominant Nrxn splice isoform in VGT3+ interneurons ( Figure 7E). Thus, we tested whether aNrxn3+/-AS4 functionally couples with Nlgn3D in NrxnTKO/VGT3/RFP slice cultures ( Figure 9E-H). To our surprise, neither aNrxn3+AS4 nor aNrxn3-AS4 pairing with postsynaptic Nlgn3D had any effect on inhibitory synaptic transmission. These results strongly suggest that aNrxn1+AS4, but not aNrxn3+AS4, has a unique signal in the extracellular domain important for synaptic function with Nlgn3D.

Discussion
Synaptic protein-protein interactions are critical for the development, maturation, and survival of neurons. However, it is technically challenging to physiologically characterize trans-synaptic CAM protein interactions in two different neurons due to the difficulty in identifying  synaptically connected neuronal pairs in the brain. Co-culture approaches consisting of non-neuronal cells transfected with different Nrxn splice isoforms and dissociated neurons expressing endogenous Nlgns (or expressing Nrxn-binding proteins in non-neuronal cells and observing their interactions with endogenous Nrxns) have begun to elucidate the roles of trans-synaptic Nrxn/Nlgn isoforms on the clustering of pre-/postsynaptic molecules (Chih et al., 2006;Kang et al., 2008;Ko et al., 2009;Nam and Chen, 2005;Scheiffele et al., 2000). However, this approach is limited to primary cultures and cannot address whether these trans-synaptic interactions are sufficient to induce functional synapse diversification. To fully understand the physiological roles of trans-synaptic molecules, one must be able to manipulate the expression of these molecules in pre-and postsynaptic neurons simultaneously followed by determination of the functional consequences of such manipulation. Using our newly developed gene electroporation method that enables us to transfect genes in minor cell types such as specific inhibitory interneurons (Keener et al., 2020a;Keener et al., 2020b), we demonstrated for the first time that aNrxn1+AS4 and Nlgn3D, which are endogenously expressed in VGT3+ inhibitory and CA1 pyramidal neurons ( Figure 7D; Futai et al., 2007;Shipman et al., 2011;Uchigashima et al., 2020), respectively, form a specific signal that dictates inhibitory synaptic transmission. It has been reported that Nlgn proteins regulate inhibitory synaptic transmission in an input cellspecific manner. For example, a Nlgn2 KO mouse line displays deficits in fast-spiking but not Sst+ interneuron-mediated inhibitory synaptic transmission in the somatosensory cortex (Gibson et al., 2009). Furthermore, Nlgn3 KO mice and the Nlgn3 R451C knock-in mutant line, which mimics a human autism mutation, showed Pv+ or Cck+ input-specific abnormal inhibitory synaptic transmission in the hippocampal CA1 region (Fö ldy et al., 2013). Therefore, the function of postsynaptic Nlgns is determined by the type of presynaptic inputs it receives, supporting the intriguing hypothesis that specific Nrxn-Nlgn binding regulates synaptic function. Our gene expression and functional assays highlight that aNrxn1 is abundantly expressed in VGT3+ interneurons compared with other interneurons ( Figure 6J) and aNrxn1+AS4 is the dominant aNrxn1 splice isoform expressed in VGT3  H). Numbers of cell pairs: bNrxn3+AS4, bNrxn3-AS4, aNrxn3+AS4, and aNrxn3-AS4 at VGT3+ synapses (eight pairs/three mice, 9/3, 5/3, and 11/3). The number of tested slice cultures is the same as that of cell pairs. Mann-Whitney U-test. + interneurons ( Figure 7C) which regulates inhibitory synaptic transmission with Nlgn3D ( Figure 8C). Our FISH experiments, which did not distinguish the AS4 insertion, detected aNrxn1 at low levels in Pv+ interneurons ( Figure 6J). The expression of aNrxn1+AS4 was previously confirmed in Pv+ interneurons (Fuccillo et al., 2015). Considering no effects of Nlgn3D OEs on Pv+ inhibitory synapses (Figure 2), other postsynaptic mechanism(s) might exist to regulate inhibitory synaptic transmission with aNrxn1+AS4 at Pv+ inhibitory synapses.
Biochemical studies have demonstrated that any Nrxn can bind to any Nlgn with different affinities, with the exception of aNrxn1 and Nlgn1, which do not interact (Boucard et al., 2005;Reissner et al., 2008). The AS4 insertion has a critical role in changing Nrxns' affinity to postsynaptic binding partners, including Nlgn (Südhof, 2017). A splice insertion at AS4 in bNrxn1 can weaken its interaction with Nlgns (Koehnke et al., 2010). Indeed, we have reported that bNrxn1-AS4 but not bNrxn1+AS4 increases synaptic transmission through its interaction with Nlgn1 (Futai et al., 2013). Similarly, aNrxn1-AS4 specifically interacts with Nlgn2 and enhances inhibitory synaptic transmission (Futai et al., 2013). In contrast, we found that an insertion at AS4 in aNrxn1 can increase inhibitory synaptic transmission at VGT3+ synapses through a trans-synaptic interaction with Nlgn3D. It is particularly interesting that only aNrxn1+AS4, but not aNrxn3+AS4, encoded uIPSC enhancement with postsynaptic Nlgn3D (Figures 8C and 9F). These results suggest that structural differences beyond the AS4 site exist between these two aNrxns. Since the structures of the major Nrxn domains, such as LNS and EGFA-EGFC, are similar between aNrxns, differential alternative splicing events that occur at other AS sites may regulate binding with Nlgn3D. Further structural and functional analyses targeting these AS events could reveal a novel AS structure that modulates Nrxn-Nlgn signaling on synaptic function. In contrast to biochemical analyses using purified Nrxn and Nlgn proteins, our results are based on intact synapses composed of a number of molecules including Nrxn and Nlgn proteins. Additional synaptic molecules could be involved in inhibitory synaptic functions mediated by trans-synaptic interactions between aNrxn1+AS4 and Nlgn3D at VGT3+ synapses. In particular, it will be important to address whether aNrxn1+AS1, aNrxn3+AS4, and bNrxn3-AS4, which are highly expressed in VGT3+ neurons (Figure 7), can encode specific synaptic functions when interacting with other postsynaptic Nrxn-binding partners such as Nlgn2, CST-3, and IgSF21 (Pettem et al., 2013;Tanabe et al., 2017;Um et al., 2014).
Nlgns function as homomeric or heteromeric dimers that bind to monomeric Nrxns (Budreck and Scheiffele, 2007;Poulopoulos et al., 2012). The function of human Nlgn3A2 at inhibitory synapses requires the presence of Nlgn2 in hippocampal neurons (Nguyen et al., 2016), suggesting that heterodimers of Nlgn3A2 and Nlgn2 are formed at inhibitory synapses. However, our finding of Nlgn3mediated synaptic potentiation at VGT3+ synapses was based on Nlgn3D OE without the replacement of other Nlgns. Further studies are necessary to determine whether the function of Nlgn3D requires the formation of heterodimers with Nlgn2 at VGT3+ synapses.
Deletion of Nrxns or Nlgns has been reported to have little effect on synapse formation (Chanda et al., 2017;Chen et al., 2017;Varoqueaux et al., 2006). Our manipulation of the expression levels of either presynaptic Nrxns or postsynaptic Nlgn3 consistently did not affect synaptic connectivity, a measurement of the number of active synapses (Figures 2, 3, and 6). However, Nlgn3D formed new synapses only when aNrxn1+AS4 was simultaneously expressed in VGT3+ interneurons ( Figure 8E). This suggests that trans-synaptic interactions of Nrxns and Nlgns may control not only synapse function but also synapse number, while the lack of Nrxns or Nlgns on either side of the synapse can be compensated by other CAM interactions. In this context, it is important to address whether the subcellular localization of Nrxn and Nlgn proteins is regulated by synaptic activity, such as homeostatic synaptic plasticity (Mao et al., 2018).
Note that our results, demonstrating the role of postsynaptic Nlgn3D and A2 isoforms in CA1 pyramidal neurons on inhibitory synaptic transmission, have some inconsistencies with a previously published study, which found that Nlgn3A2 differentially regulates Pv+ and Sst+ inhibitory synapses (Horn and Nicoll, 2018). Horn and Nicoll reported that OE of human Nlgn3A2 reduced Pv+ and increased Sst+ inhibitory synaptic transmission. The former finding is consistent with our results in Figure 3B and Figure 3-figure supplement 1, displaying that Nlgn3D or A2 OE reduce Pv+ uIPSC amplitude. This may suggest that Pv+ neurons have a unique trans-synaptic regulatory mechanism compared with other interneurons and that Nlgn3D or A2 OE may disrupt endogenous GABA A R complexes at Pv+ inhibitory synapses. In contrast, Nlgn3D or A2 OE did not increase Sst+ uIPSCs ( Figure 3F and Figure 3-figure supplement 1), as observed in Horn and Nicoll, 2018. This difference might be due to variations in experimental approaches including the Nlgn3 clone used (human Nlgn3A2 versus mouse Nlgn3D splice isoform) and the duration of transgene or shRNA expression in hippocampal CA1 pyramidal neurons (2-3 weeks of OE versus 2-3 days of OE). Additionally, Horn and Nicoll crossed Pv-Cre and Sst-Cre, the same cre lines we tested, with Ai32 mice, a cre-dependent channelrhodopsin line (Madisen et al., 2012), to evoke Pv+ or Sst+ neuron-mediated synaptic transmission, respectively (Horn and Nicoll, 2018). However, our current and previous studies for mouse line validation indicate that while Pv-Cre and Sst-Cre lines exhibit highly specific cre expression in these cell types, these lines have leaky cre expression in other cell type(s) (Figure 2figure supplement 1, yellow arrow heads) (Mao et al., 2015). Therefore, light-evoked activation of nonspecific cell types may contribute to the inconsistent results in synaptic transmission. Surprisingly, Nlgn3A2 did not increase VGT3+ inhibitory synaptic transmission. One possible explanation is that Nlgn3A2 couples with Cck+ neurons that do not express VGT3. Cck-and vasoactive intestinal polypeptide (VIP)-positive but VGT3-negative inhibitory interneurons have been identified in the hippocampal CA1 region (Somogyi et al., 2004). Furthermore, most of the Cck+ neurons in this area express CB1 transcripts (Katona et al., 1999). Therefore, CB1 signals in Figure 1C can originate from Cck+/VGT3+ and Cck+/VIP+ interneurons. Further studies are required to identify other types of interneurons that are capable of coupling with Nlgn3A2 by using other cell type-specific cre lines (e.g. VIP-cre).

Single-and double-labeled FISH
Single/double FISH was performed using our recently established protocol (Uchigashima et al., 2019). VGT3+, Sst+, and Pv+ interneurons were identified with cRNA probes characterized previously (Omiya et al., 2015;Song et al., 2014;Yamasaki et al., 2016). All procedures were performed at room temperature unless otherwise noted. Briefly, fresh frozen sections were fixed with 4% paraformaldehyde, 0.1M PB for 30 min, acetylated with 0.25% acetic anhydride in 0.1M triethanolamine-HCl (pH 8.0) for 10 min, and prehybridized with hybridization buffer for 30 min. Hybridization was performed with a mixture of fluorescein-(1:1000) or DIG-(1:10,000) labeled cRNA probes in hybridization buffer overnight followed by post-hybridization washing using saline-sodium citrate buffers at 75˚C. Signals were visualized using a two-step detection method. Sections were pretreated with DIG blocking solution for 30 min and 0.5% tryamide signal amplification (TSA) blocking reagent in Tris-NaCl-Tween 20 (TNT) buffer for 30 min before the first and second steps. During the first step, sections were incubated with peroxidase-conjugated anti-fluorescein antibody (1:500, Roche Diagnostics) and TSA Plus Fluorescein amplification kit (PerkinElmer) for 1 hr and 10 min, respectively. In the second step, sections were incubated with peroxidase-conjugated anti-DIG antibody (1:500, Roche Diagnostics) and TSA Plus Cy3 amplification kit (second step) with the same incubation times. Between the first and second steps, residual peroxidase activity was inactivated with 3% H 2 O 2 in TNT buffer for 30 min. Sections were incubated with DAPI for 10 min for nuclear counterstaining (1:5000, Sigma-Aldrich).

Triple staining of Nlgn3 and synaptic markers
Mice were fixed by transcardial perfusion with 3% glyoxal fixative (Richter et al., 2018). Brains were cryoprotected with 30% sucrose in 0.1 M PB to prepare 50-mm-thick cryosections on a cryostat (CM1900; Leica Microsystems). All immunohistochemical incubations were performed at room temperature. Sections were incubated with 10% normal donkey serum for 20 min, a mixture of primary antibodies overnight (1 mg/ml), and a mixture of Alexa 488-, Cy3-, or Alexa 647-labeled species-specific secondary antibodies for 2 hr at a dilution of 1:200 (Invitrogen; Jackson ImmunoResearch, West Grove, PA). Images were taken with a confocal laser scanning microscope equipped with 473, 559, and 647 nm diode laser lines, and UPlanSApo (10Â/0.40), UPlanSApo (20Â/0.75), and PlanApoN (60Â/1.4, oil immersion) objective lenses (FV1200; Olympus, Tokyo, Japan). To avoid crosstalk between multiple fluorophores, Alexa488, Cy3, and Alexa647 fluorescent signals were acquired sequentially using the 488 nm, 543 nm, and 633 nm excitation laser lines. All images show single optical sections (800 Â 800 pixels). The analysis was performed using ImageJ software (https:// imagej.nih.gov/ij/). Briefly, the signal intensity and co-localization frequency of Nlgn3 puncta in the hippocampus CA1 were measured in the region of interest (ROI) selected from inhibitory synapses co-labeled for VIAAT and interneuron markers: VGT3, CB1, Pv, and Sst. To assess the noise levels for intensities and co-localizations, we analyzed images with the Nlgn3 channel rotated 90˚to identify true close appositions of Nlgn3 signals and synaptic markers (Singh et al., 2016;Stogsdill et al., 2017). The noise levels for the signal intensity and co-localization frequency of Nlgn3 signals were comparable among the four distinct synapses ( Figure 1G), suggesting that the distribution pattern or ROI of Nlgn3 signals and synaptic markers were unlikely to be biased. All the data for each group were obtained from two mice and pooled together.

Single-cell sequencing and analysis
Single-cell RNA extraction The cytosol of four VGT3-positive neurons in CA1 origins was harvested using the whole-cell patchclamp technique described previously (Futai et al., 2013;Uchigashima et al., 2020). The cDNA libraries were prepared using a SMART-Seq HT Kit (TAKARA Bio) and a Nextera XT DNA Library Prep Kit (Illumina) as per the manufacturers' instructions. The final product was assessed for its size distribution and concentration using a BioAnalyzer High Sensitivity DNA Kit (Agilent Technologies) and loaded onto an S1 flow cell on an Illumina NovaSeq 6000 (Illumina) and run for 2 Â 50 cycles according to the manufacturer's instructions. De-multiplexed and filtered reads were aligned to the mouse reference genome (GRCm38) using HISAT2 (version 2.1.0) applying -no-mixed and -nodiscordant options. Read counts were calculated using HTSeq by supplementing Ensembl gene annotation (GRCm38.78). Gene expression values were calculated as transcripts per million (TPM) using custom R scripts (Source code 1). Genes with no detected TPM in all samples were filtered out. Our data set was then combined with the 'Mouse Whole Cortex and Hippocampus SMART-seq' data portal from the Allen Institute for Brain Science where a complementary set of 76,533 total cells were primarily collected from >20 areas of mouse cortex and hippocampus of~8-week-old pan-GABAergic, pan-glutamatergic, and pan-neuronal transgenic lines . The gene expression data matrix (matrix.csv) which stores raw read counts for every cell in the data set and cell metadata (metadata.csv) containing information such as sample names, brain regions of origin, cell type designations (e.g. 'GABAergic', 'Non-neuronal', and 'Glutamatergic') and cell type subclass designations (e.g. 'SST', 'L6 CT', and 'Astrocyte') was downloaded from the portal. Neuronal cells only from the hippocampus (2367 cells) were merged with our dataset (total 2382 cells). In order to minimize batch effect between our data and the Allen Brain Atlas, systematic differences in sequencing coverage across batches were removed by rescaling the size factors using the multiBatchNorm function from the batchelor R package (Haghverdi et al., 2018), and then a batch effect correction based on linear regression model was applied using the rescaleBatches function from the batchelor package. A tSNE plot was then generated using Rtsne R package (van der Maaten and Hinton, 2008). Eighty-five randomly selected hippocampal GABAergic neurons from the Allen Brain Atlas dataset and 15 of our cells were selected, and the batch-corrected expression levels of Nrxn genes were visualized in a heatmap using ComplexHeatmap R package (Gu et al., 2016). For splice isoform quantification, kallisto (Bray et al., 2016) was used by supplementing the transcript fasta file (Mus_musculus.GRCm38.cdna.all.fa). Each isoform was summarized manually to account for inclusion or exclusion of the AS4 exon in the a or b isoforms. The manually curated transcript IDs are provided in Table 1.

Single-cell RT-qPCR
Isolation of single-cell cytosol and preparation of single-cell cDNA libraries were performed by the same method described in single-cell sequencing and analysis. For validation of the Nrxn KO mouse line, the following TaqMan gene expression assays (Applied Biosystems) were used: Nrxn1 (Mm03808857_m1), Nrxn2 (Mm01236856_m1), Nrxn3 (Mm00553213_m1), and Gapdh (Mm99999915_g1). The relative expression of Nrxns was calculated as follows: Relative expression = 2 Ct,Gapdh /2 Ct,Nrxns ; Ct, threshold cycle for target gene amplification.

Single-cell electroporation
A detailed protocol is described in our recent publication (Keener et al., 2020a;Keener et al., 2020b). Briefly, the slice cultures were perfused with filter-sterilized aCSF consisting of (in mM): 119 NaCl, 2.5 KCl, 0.5 CaCl 2 , 5 MgCl 2 , 26 NaHCO 3 , 1 NaH 2 PO 4 , 11 glucose, and 0.001 mM tetrodotoxin (TTX, Hello Bio Inc), gassed with 5% CO 2 /95% O 2 , pH 7.4. Patch pipettes (4.5-8.0 MW) were each filled with plasmids containing either tag-BFP and Nrxn or EGFP and Nlgn3D (0.05 mg/ml for each plasmid) and respectively electroporated in TdTomato-positive VGT3+ interneurons and CA1 pyramidal neurons. The same internal solution for single-cell sequencing was used. A single electrical pulse train (amplitude: À5 V, square pulse, train: 500 ms, frequency: 50 Hz, pulse width: 500 ms) was applied to the target neurons. After electroporation, the electrode was gently retracted from the cell and the slices were transferred to a culture insert (Millipore) with slice culture medium in a petri dish and incubated in a 5% CO 2 incubator at 35˚C for 3 days.

Electrophysiology
Whole-cell voltage-and current-clamp recordings were performed on postsynaptic and presynaptic neurons, respectively. Nlgn3D and Nlgn3A2 constructs or shRNAs were transfected at DIV6-9 or DIV2 and subjected to recordings at 2-3 or 5-12 days after transfection, respectively. DIV10-14 organotypic slice cultures prepared from WT (VGT3/RFP) and KO (NrxnTKO/VGT3/RFP) mice were evaluated for KO of Nrxns at VGT3+ synapses. The extracellular solution for recording consisted of (in mM): 119 NaCl, 2.5 KCl, 4 CaCl 2 , 4 MgCl 2 , 26 NaHCO 3 , 1 NaH 2 PO 4 , 11 glucose, and 1 kynurenic acid (Sigma), gassed with 5% CO 2 and 95% O 2 , pH 7.4. Thick-walled borosilicate glass pipettes were pulled to a resistance of 2.5-4.5 MW. Whole-cell voltage clamp recordings were performed with internal solution containing (in mM): 115 cesium methanesulfonate, 20 CsCl, 10 HEPES, 2.5 MgCl 2 , 4 ATP disodium salt, 0.4 guanosine triphosphate trisodium salt, 10 sodium phosphocreatine, and 0.6 EGTA, adjusted to pH 7.25 with CsOH. For current-clamp recordings, cesium in the internal solution was substituted with potassium and the pH was adjusted with KOH. GABA A receptor-mediated inhibitory postsynaptic currents (IPSCs) were measured at Vhold ± 0 or À70 mV. Thirty to forty consecutive stable postsynaptic currents were evoked at 0.1 Hz by injecting current (1 nA) in presynaptic interneurons. Synaptic connectivity was tested by applying 25 consecutively paired (at 50 ms intervals) stimulations; responses larger than 5 pA observed within 5 ms after the onset of either of the pulses were counted as evoked unitary GABA A R-IPSC. Recordings were performed using a Multi-Clamp 700B amplifier and Digidata 1440, digitized at 10 kHz and filtered at 4 kHz with a low-pass filter. Data were acquired and analyzed using pClamp (Molecular Devices).

Statistical analyses
Results are reported as mean ± SEM. Statistical significance, set at p<0.05, was evaluated by one-or two-way ANOVA with Sidak's post hoc test for multiple comparison, Mann-Whitney U-test, and Student's t-test for two-group comparison.
. Transparent reporting form

Data availability
Sequencing data have been deposited in GEO under accession code GSE150989.
The following dataset was generated: