Nlgn3 gene was originally cloned from rat brains as a homolog of rat Nlgn1 (Ichtchenko et al., 1996). Nlgn3 is conserved from invertebrates to vertebrates including humans. For instance, human Nlgn3 shows a 92 or 98% sequence similarity at the transcriptional or translational levels, respectively, compared with the rat ortholog (Philibert et al., 2000). Drosophila Nlgn genes (Dnlg1-4) are independently diversified during evolution in invertebrates and vertebrates but have relatively high (approximately 20%) sequence homologies to human Nlgns (Knight et al., 2011). Human Nlgn3 is 32,272 bp in length on an X-chromosome region (Xq13.1) and composed of eight exons ranging from 60 to 1,864 bp (Philibert et al., 2000). The coding region spans from exon 2 to 8. Exons 7 and 8 are the largest exons, encoding about 65% of Nlgn3 protein (Figure 1). Nlgn3 promotor has a Wnt signal-responsive element, allowing for Wnt-mediated transcriptional control of Nlgn3 (Medina et al., 2018). Premature transcripts of Nlgn3 are processed by splicing out particular exons. Exons 3 and 4, encoding 20 amino acid sequences named A1 and A2, can be spliced out, thereby generating four theoretical splice isoforms from Nlgn3 (Nlgn3-A, +A1, +A2, and +A1A2). Exon 7 may also be alternatively spliced because some Nlgn3 mRNAs are truncated with a lack of exon 7 (Talebizadeh et al., 2006). The longest transcript of Nlgn3 is translated into Nlgn3, which is made up of 848 amino acids in humans and rats, partly through postsynaptic local translational machinery (Ichtchenko et al., 1996; Philibert et al., 2000; Cajigas et al., 2012).

Nlgn3 is a single membrane spanning protein with an N-terminal hydrophobic sequence with the characteristics of a cleaved signal peptide followed by a large extracellular domain, a highly conserved single transmembrane region, and a short cytoplasmic domain (Ichtchenko et al., 1996). This molecular structure is common to other Nlgn isoforms, as rat Nlgn1-3 share 52 and > 70% identical residues in the full length and extracellular domain, respectively (Ichtchenko et al., 1996; Fabrichny et al., 2007). The extracellular domain is composed of a large cholinesterase-like domain and O-glycosylated sequence (Figure 1). The cholinesterase-like domain has the α/β-hydrolase fold which is structurally shared with acetylcholinesterase and thyroglobulin (De Jaco et al., 2010). However, the α/β-hydrolase fold found in Nlgns is commonly non-catalytic due to a lack of an active site serine and alternatively responsible for binding to other synaptic molecules. The two alternatively spliced insertions A1 and A2 locate on the extracellular cholinesterase-like domain on Nlgn3. A1 is a sequence homologous to the splice insert A of Nlgn1 and contains two cysteines, suggesting a possible formation of a disulfide-bonded ring. In contrast, A2 is a sequence homologous to the splice insert A of Nlgn2 and has no cysteines (Hoffman et al., 2004). Importantly, the presence or absence of the splice insertions A1 and A2 play roles in synaptic function as described previously (Philibert et al., 2000; Oku et al., 2020; Uchigashima et al., 2020b). The extracellular cholinesterase-like domain also has the Nlgn-Nlgn dimerization interface which is highly conserved among different Nlgn isoforms (Dean et al., 2003; Arac et al., 2007; Yoshida et al., 2021).

Nlgn3 molecule forms homodimers and heterodimers with other Nlgn isoforms in the secretory pathway and translocate to the plasma membrane (Poulopoulos et al., 2012; Yoshida et al., 2021). Nlgn1 protein can participate in the heterodimerization with Nlgn3 (Poulopoulos et al., 2012). However, the heterodimer formation of Nlgn2 and Nlgn3 is controversial. Nlgn3 is anatomically co-localized and biochemically co-immunoprecipitated with Nlgn2 (Budreck and Scheiffele, 2007). Nlgn3 functionally requires Nlgn2 at hippocampal inhibitory synapses (Nguyen et al., 2016). These biochemical, anatomical, and functional pieces of evidence support the heterodimer formation of Nlgn2 and Nlgn3. However, an in situ chemical cross-linking study failed to detect heterodimers of Nlgn2 and Nlgn3 in hippocampal neurons (Poulopoulos et al., 2012). Dimerization of Nlgns, including Nlgn3, is essential for synapse specification and function (Dean et al., 2003; Ko et al., 2009; Shipman and Nicoll, 2012b). The significance of dimerization on Nlgn trafficking is controversial. Poulopoulos et al. (2012) reported that the lack of dimerization retains Nlgn in the secretory pathway although other groups reported that monomer mutants are able to translocate to the plasma membrane (Ko et al., 2009; Shipman and Nicoll, 2012b).

Nlgn3 protein can be cleaved on its juxtamembrane domain in an activity-dependent manner which is conserved from rodents to humans (Bemben et al., 2019). Although Nlgn1 can also be shed during elevated neuronal activity, Nlgn1 does not share the same molecular pathway for cleavage as Nlgn3. The shedding of Nlgn1 depends on NMDA receptor activation, Ca²⁺/calmodulin-dependent protein kinase, and proteolytic activity of a disintegrin and metalloproteinase domain-containing protein 10 (ADAM10) or matrix metalloproteinase 9 (MMP9) (Peixoto et al., 2012; Suzuki et al., 2012). In contrast, ADAM10 claves Nlgn3 under basal condition and activity-dependent cleavage of Nlgn3 is mediated by metabotropic glutamate receptor activation, protein kinase C (PKC) signaling, and MMPs, irrespective of any combination of Nlgn3 dimers (homomers or heteromers) (Kuhn et al., 2016; Venkatesh et al., 2017; Bemben et al., 2019). In addition, Nlgn3 confers sensitivity to PKC-dependent ectodomain shedding of Nlgn1 and Nlgn2 (Bemben et al., 2019). Interestingly, Nlgn3 extracellular ectodomain functions as a cortical neuronal activity-regulated glioma mitogen promoting high-grade glioma (HGG) proliferation and growth (Venkatesh et al., 2015, 2017) (see Section “Nlgn3 in tumors”). However, it remains elusive whether Nlgn3 ectodomain affects the translation of other proteins.

Nlgn3 gene is involved in a non-syndromic monogenic form of ASD. As of today, over ten missense mutations are identified on Nlgn3 locus in individuals with ASD (Figure 1; Jamain et al., 2003; Blasi et al., 2006; Jiang et al., 2013; Yu et al., 2013; Iossifov et al., 2014; Kenny et al., 2014; Redin et al., 2014; Xu et al., 2014; Deciphering Developmental Disorders, 2017; Yuen et al., 2017; Quartier et al., 2019). Among these mutations, R451C missense mutation has been most intensively studied (Jamain et al., 2003). The estimated occurrence of R451C mutation is < 3% among people with non-syndromic ASD (Quartier et al., 2019). R451C mutation site is located in the helix next to the Nlgn dimerization interface and is important for cell surface trafficking. It does not interfere with dimerization of Nlgn3 mutant protein (Poulopoulos et al., 2012), but shows retention of Nlgn3 mutant in the endoplasmic reticulum (ER), leading to a 90% loss of Nlgn3 from the cell surface (Chih et al., 2004; Comoletti et al., 2004). The mutant protein retained in the ER is preferentially degraded by the proteasome (Trobiani et al., 2018). Additionally, this retention can activate some pathways of the unfolded protein response (UPR) to induce the upregulation of the molecular chaperone BiP (immunoglobulin heavy-chain-binding protein) or ER stress-related transcription factor CHOP [C/EBP (CCAAT/enhancer-binding protein)-homologous protein], supporting a possible molecular mechanism that influences neuronal functions in individuals carrying the R451C mutation of Nlgn3 gene (Trobiani et al., 2018). Additional missense mutations of the R597W and P514S were reported in a few multiplex families with ASD and intellectual disability (Redin et al., 2014; Quartier et al., 2019). Importantly, all these missense mutations can cause severe loss of Nlgn3 expression at the postsynaptic membrane via abnormal membrane trafficking (Tabuchi et al., 2007; Quartier et al., 2019).

Deletions or frameshift mutations in the coding region of Nlgn3 were also reported in individuals with ASD (Levy et al., 2011; Sanders et al., 2011; Yuen et al., 2017). Furthermore, single nucleotide variants in the intron or 3′ untranslational region (3′UTR) of Nlgn3 were identified in a cohort of 144 males with ASD (Steinberg et al., 2012). The intronic variants locate within conserved transcription factor binding sites that could potentially affect gene regulation. The functional impact of the 3′UTR variant remains unclear because no effect on the expression of Nlgn3 was detected by luciferase assay. In addition, Nlgn3 might be associated with schizophrenia. A rare loss-of-function mutation of Nlgn3, which can cause a premature stop 42 codons downstream of the frameshift mutation in exon 2, was also found among a study of 273 people diagnosed with schizophrenia (Kenny et al., 2014).

Dysregulation of mTOR (mammalian target of rapamycin) and MAPK (mitogen-activated protein kinase) pathways are strongly implicated in ASD (Bramham et al., 2016; Winden et al., 2018). eIF4E (eukaryotic translation initiation factor 4E) is an mRNA cap-binding protein that has also been associated with ASD given that eIF4E-mediated translation is the final common process in mTOR, MAPK, and FMRP pathways (Amorim et al., 2018). eIF4E transgenic (TG) mice exhibit abnormal translation, ASD-like behaviors, and synaptic deficits (Santini et al., 2013). Importantly, eIF4E TG or 4E-BP2, an eIF4E repressor, KO mice display abnormal social behaviors and increased Nlgn translation (Gkogkas et al., 2013). Furthermore, Nlgn3 local translation is regulated by fragile X mental retardation protein encoded by fragile mental retardation 1 (Fmr1) gene, which is the most common monogenic cause of syndromic ASD (Chmielewska et al., 2019). Taken together, these findings indicate that Nlgns are downstream targets of ASD-associated signaling pathways.

Recently, Hornberg et al. (2020) reported a novel function of Nlgn3 protein to regulate mRNA translation. Shot-gun proteomics in the ventral tegmental area (VTA) of Nlgn3 KO mice identified altered expression of proteins associated with mRNA translation. Moreover, AHA (methionine analog azidohomoalanine) labeling elucidated the disruption of translation homeostasis in Nlgn3 KO mice. Treatment with ETC-168, an inhibitor of MAP kinase-interacting kinases (MNK) 1/2, ameliorated the dysregulation of mRNA translation and abnormal social behavior (Hornberg et al., 2020). These results clearly indicate that Nlgn3 is not just synaptic glue but acts as a synaptic regulator. Indeed, Nlgn3 regulates dendritic structure by modulating mTOR signaling (Xu et al., 2019) and Nlgn3 KO caused dysregulation of mGluR-dependent signaling (Baudouin et al., 2012; but see Rothwell et al., 2014; Zhang et al., 2015). These studies support the profound importance of Nlgn3 as an upstream regulator of translational pathways vulnerable in ASD.

Nlgn1-3 mRNA are highly expressed in the central nervous system (Ichtchenko et al., 1996; Gilbert et al., 2001). The expression levels in rat brains are low at birth and upregulated 2–3 times during postnatal development (Varoqueaux et al., 2006). A modest developmental increase of Nlgn3 mRNA has been reported based on quantitative RT-PCR of total mRNA levels in the mouse medial nucleus of the trapezoid body (MNTB) (Zhang et al., 2017). Nlgn3 is expressed in various types of neural cells: neurons, astrocytes, oligodendrocyte precursor cells (OPCs), newly formed oligodendrocytes, and myelinating oligodendrocytes (Zhang et al., 2014; Proctor et al., 2015; Stogsdill et al., 2017). Olfactory ensheathing glia are also known as Nlgn3-expressing cells (Gilbert et al., 2001). Nlgn3 expression levels vary depending on the cell type. In the ventral striatum (nucleus accumbens), there are two different types of projection type neurons called medium spiny neurons (MSNs). Nlgn3 mRNA levels, but not Nlgn1 and Nlgn2, are significantly higher in D1 dopamine receptor expressing-MSNs than in D2-MSNs (Rothwell et al., 2014). Interestingly, single cell RNAseq demonstrates that Nlgn3 levels are the highest in OPCs, which make bona fide glutamatergic and GABAergic synapses as postsynaptic cells (Bergles et al., 2000; Lin and Bergles, 2004; Ziskin et al., 2007). Nlgn3 contains four alternative splice variants (Nlgn3-A, +A1, +A2 and, +A1A2) with different expression levels. Nlgn3-A and Nlgn3+A2 splice isoforms are dominantly expressed in hippocampal CA1 pyramidal cells (Uchigashima et al., 2020b). Nlgn3 is also detected outside the central nervous system (CNS) (Philibert et al., 2000; Suckow et al., 2008). Rat Nlgn3 is expressed in pancreatic islet beta cells at lower levels than in the brain (Suckow et al., 2008).

Consistent with the expression pattern of Nlgn3 mRNA, Nlgn3 protein is richly expressed in the CNS. In the mouse brain, the expression of Nlgn3 begins as early as E12, peaks during postnatal weeks 2 and 3 when synaptogenesis is maximized, and lasts throughout the adult stage (Budreck and Scheiffele, 2007; Trobiani et al., 2018). Nlgn3 is detectable in both neuronal and non-neuronal cells (Varoqueaux et al., 2006; Proctor et al., 2015; Stogsdill et al., 2017). Some studies further demonstrate the in vitro and in vivo subcellular distribution of Nlgn3 by immunostaining with specific antibodies in the mouse CNS. Nlgn3 is expressed across the whole brain with variable intensities in distinct regions (Uchigashima et al., 2020b). The expression levels are high in the hippocampus, neocortex, striatum, and brain stem, and low in the thalamus and cerebellum. The hippocampus is the most extensively analyzed area. Immunopositive signals for Nlgn3 are associated with both excitatory and inhibitory synapses in hippocampal primary neurons (Budreck and Scheiffele, 2007) or the adult hippocampal CA1 region (Uchigashima et al., 2020b). Importantly, Nlgn3 is selectively expressed at inhibitory synapses expressing cannabinoid receptor CB1 or vesicular glutamate transporter type 3 (VGT3) [presumably co-expressing cholecystokinin (Cck)], but not at other major inhibitory synapses expressing parvalbumin (Pv) or somatostatin (Sst) in the hippocampal CA1 region (Uchigashima et al., 2020a). Input cell-dependent expression of Nlgn3 is also noted in the cerebellar cortex. Nlgn3 is localized at parallel fiber excitatory synapses, mossy fiber excitatory synapses, and a subset of inhibitory synapses including molecular layer interneuron-Purkinje cell synapses, but not at climbing fiber excitatory synapses (Baudouin et al., 2012; Lai et al., 2021). To date, no reports have addressed the distribution of endogenous Nlgn3 splice isoforms due to technical difficulty. However, Nlgn3 overexpression shows splice isoform-dependent targeting of Nlgn3 protein to distinct synapses as shown in a recent study (Uchigashima et al., 2020b).

The extracellular domain of Nlgn3 protein trans-synaptically binds to that of Nrxns in a Ca²⁺-dependent manner (Ichtchenko et al., 1996). Nrxns are presynaptic CAMs encoded in three genes (Nrxn1, Nrxn2, and Nrxn3) that are transcribed under different promoters as longer alpha (αNrxn1–3), shorter beta (βNrxn1–3), and Nrxn1-specific gamma (γNrxn1) isoforms (Figure 2; Tabuchi and Sudhof, 2002; Sterky et al., 2017). Each Nrxn gene has six alternative splice (AS) sites, named AS1–AS6, potentially generating thousands of Nrxn splice isoforms (Puschel and Betz, 1995; Ullrich et al., 1995; Missler and Sudhof, 1998; Gorecki et al., 1999; Schreiner et al., 2014; Treutlein et al., 2014). Notably, alternative splicing at AS4 is well characterized as a molecular switch that determines trans-synaptic CAM binding pairs. A surface plasmon resonance binding assay has demonstrated the comparative binding affinity of Nlgn proteins to βNrxn protein. Nlgn3 weakly binds to βNrxn compared with Nlgn1 and Nlgn2 with ∼10-fold lesser binding strength irrespective of the splice isoform of either binding partner (Koehnke et al., 2010). All Nlgn3 splice isoforms exhibit a higher binding affinity to βNrxn without the insertion of AS4 than those with AS4 (Ichtchenko et al., 1996; Koehnke et al., 2010). Structural analysis predicts the βNrxn1 binding site of Nlgn3 protein is composed of H279-E281, N362, and Q370-N377 (Yoshida et al., 2021). The ASD-associated R451C mutation on the extracellular domain can interfere with this binding (Comoletti et al., 2004; Yoshida et al., 2021). Nlgn3 can also bind to αNrxn1 protein regardless of the presence or absence of the splice insertion at AS4 (Boucard et al., 2005). Interestingly, a specific trans-synaptic interaction between Nlgn3-A and αNrxn1+AS4 controls inhibitory synaptic transmission in an input cell-dependent manner, as described in detail below (see section “Molecular Code” for an Input Cell-Dependent Function of Nlgn3 Protein at Hippocampal Inhibitory Synapses) (Uchigashima et al., 2020a). To the best of our knowledge, there are no comparative studies examining the binding affinity of Nlgn3 to αNrxn. Recently, heparin sulfate (HS) modification of Nrxn was reported to mediate trans-synaptic interactions with Nlgn1–4 proteins and regulate their synaptic functions (Zhang et al., 2018). The three lysine and arginine residues of human Nlgn3 (K609, R611, and R613), which are conserved across different Nlgn isoforms and animal species, are required for HS binding for full Nrxn binding, but not for the dimerization or membrane trafficking of Nlgn3 (Zhang et al., 2018).

Nrxn has been considered the sole presynaptic binding partner of Nlgn for decades, however, a recent study identified Nlgn3 as a selective ligand for presynaptically-expressed type IIA receptor protein tyrosine phosphatase delta (PTPδ) (Yoshida et al., 2021). Similar to Nrxn genes, PTPδ has multiple isoforms generated by alternative splicing of microexons (Takahashi and Craig, 2013). Yoshida et al. (2021) demonstrated that PTPδ lacking mini-exon B can trans-synaptically interact with Nlgn3 and influence presynaptic differentiation (Figure 2). Interestingly, a structural analysis predicted a direct interaction of G371, E372, L374, and N375 in the extracellular domain of Nlgn3 with the third immunoglobulin domain of PTPδ protein, suggesting that PTPδ and βNrxn1 potentially compete for binding to Nlgn3. Importantly, the Nlgn3 binding interface to PTPδ is not present in other Nlgns, further highlighting the versatile function of Nlgn3 compared with other Nlgns. This non-canonical Nlgn3-PTPδ signaling is hampered by the R451C mutation of Nlgn3 (Yoshida et al., 2021).

MAM domain-containing glycosylphosphatidylinositol anchor 1 (MDGA1) and 2 (MDGA2) proteins also compete for the Nrxn binding site on the extracellular domain of Nlgn3 via a cis interaction, blocking trans-synaptic interactions between Nlgn and Nrxn (Lee et al., 2013; Pettem et al., 2013; Connor et al., 2016; Elegheert et al., 2017; Gangwar et al., 2017; Figure 2). The binding affinity of each MDGA protein varies depending on the Nlgn isoform. Both MDGAs exhibit low binding affinity to Nlgn3 compared with Nlgn1 and Nlgn2 (Lee et al., 2013; Elegheert et al., 2017). Furthermore, the heterodimer of Nlgn3 with other Nlgn isoforms likely blocks MDGA-mediated blockage of Nrxn-Nlgn interactions because the immunoglobulin-tandem structure of MDGA is not able to bind to Nlgn heterodimers efficiently (Gangwar et al., 2017). Therefore, MDGA protein-mediated modulation would be less effective at Nlgn3-containing synapses, compared with Nlgn3-lacking synapses.

Secreted protein acidic and rich in cysteine-like 1 (SPARCL1 also known as Hevin) is an extracellular matrix released from astrocytes and important for excitatory synaptogenesis (Kucukdereli et al., 2011). Interestingly, SPARCL1 forms synaptic triads with αNrxn1 and Nlgn1, and Nlgn3 can be incorporated into this complex (Kucukdereli et al., 2011; Singh et al., 2016; Figure 2). However, the molecular mechanism underlying the synaptogenic effect of SPARCL1 remains controversial. Gan and Sudhof (2020) reported that SPARCL1 regulates excitatory synaptogenesis independent of Nlgns and Nrxns. Further studies are required to elucidate the physiological role of Nrxn-SPARCL1-Nlgn complexes on excitatory synapses.

The short intracellular domain of Nlgn3 contains the PDZ- (Irie et al., 1997) and gephyrin-binding motifs (Poulopoulos et al., 2009; Figure 2). Both PDZ- and gephyrin-binding motifs are conserved across all Nlgn isoforms, suggesting that all Nlgn isoforms can be potentially delivered into both excitatory and inhibitory synapses. Nlgn3 has some conserved phosphorylation sites in the intracellular domain. In the case of Nlgn1 and Nlgn2, the phosphorylation of the conserved tyrosine residue is functionally critical for excitatory and inhibitory synapses, respectively (Poulopoulos et al., 2009; Giannone et al., 2013). However, the functional roles of Nlgn3 phosphorylation sites remain unclear because no effects of the corresponding Nlgn3 phosphomutants have been detected on synaptic function.

One of the most important outcomes obtained by Nlgn3 mutant studies is the heterogeneity of mutation impact in the brain circuits. Three mutant mouse models, including knockout and missense ASD mutations, have revealed that different Nlgn3 mutations cause distinct abnormalities at synapses. This section highlights the differential impact of Nlgn3 mutations in excitatory and inhibitory synapses (Tables 1, 2).

A few studies have examined the impact of Nlgn3 protein on long-term synaptic plasticity. Parallel fiber-Purkinje cell synapses in the cerebellum can induce mGluR-mediated long-term depression (LTD) which underlies motor coordination. The effect of Nlgn3 KO on this form of LTD remains controversial. Baudouin et al. (2012) reported that constitutive Nlgn3 KO mice exhibited a loss of mGluR-mediated LTD. However, Zhang et al. (2015) showed no change in mGluR-mediated LTD in pan Nlgn-deficient Purkinje cells. Nlgn3 KD caused no significant change on Nlgn1-mediated long-term synaptic potentiation (LTP) at excitatory synapses in the hippocampal CA1 region (Shipman and Nicoll, 2012a).

In contrast to Nlgn3 KO or KD, Nlgn3-R451C mutation significantly affects long-term synaptic plasticity with variable phenotypes at distinct synapses. LTP is enhanced at hippocampal excitatory synapses (Etherton M. et al., 2011). On the other hand, LTD is impaired at corticostriatal excitatory synapses (Martella et al., 2018). Interestingly, the impairment in this form of LTD can be partly rescued by the activation of cannabinoid CB1 receptor, suggesting a dysfunction in endocannabinoid signaling in Nlgn3-R451C KO mice as demonstrated at hippocampal inhibitory synapses (Foldy et al., 2013; Martella et al., 2018).

A functional impact of Nlgn3 in remodeling of neuronal circuits has been reported. Nlgn3 can affect the turnover of spines that form synaptic structures with excitatory neurons. In layer 2/3 pyramidal neurons of the somatosensory cortex, Nlgn3-R451C KI mice show enhanced dynamics of PSD95+ spines, which may be subsequently associated with stable synaptic connectivity. Importantly, abnormal PSD95+ spine dynamics is shared with another mouse model of non-syndromic ASD in which the chromosomal region corresponding to human 15q11–13 is paternally duplicated (patDp/ + mice) (Isshiki et al., 2014).

Nlgn3 also contributes to the developmental elimination of climbing fiber-Purkinje cell synapses. Nlgn3-R451C KI mice impair elimination of redundant climbing fiber-Purkinje cell synapses from postnatal day 10–15 (Lai et al., 2021). Nlgn3 KO mice exhibit ectopic climbing fiber synapses through hetero-synaptic competition with parallel fiber synapses onto Purkinje cells (Baudouin et al., 2012). Therefore, the dysfunction of Nlgn3 likely causes abnormal remodeling of neuronal circuits which can be a major endophenotype of ASD.

One prevailing mechanism describing the pathophysiology of ASD is the imbalance between excitation and inhibition in neurons, which can be caused by circuit dysfunction described above. Supporting this notion, emerging evidence has revealed that functional impairment in Nlgn3 gene generates dysfunction in specific circuits. Based on the DSM-5, the diagnostic criteria to characterize ASD are (1) persistent deficits in social communication and interaction and (2) restricted, repetitive patterns of behavior, interests, or activities (American Psychiatric Association, 2013). Many behavioral studies investigating Nlgn3 mutant mouse and Drosophila (Yost et al., 2020) models corroborate the relevance of Nlgn3 dysfunction with ASD-like behavioral features and reveal roles for Nlgn3 in specific brain regions responsible for these behaviors (Table 3). The first study on Nlgn3-R451C, which mimics the human single nucleotide polymorphism (SNP) identified in individuals with ASD, opened the field for investigation of Nlgn3 mutants in ASD-associated phenotypes (Jamain et al., 2003; Tabuchi et al., 2007). This KI line displayed impaired sociability and learning and memory, which was further confirmed in follow-up studies (Etherton M. et al., 2011; Jaramillo et al., 2014). In addition, Nlgn3-R451C KI mice display increased repetitive behavior and aggressiveness, deficits in transitive inference, and abnormal wake and sleep EEG power spectral profiles (Burrows et al., 2015; Liu et al., 2017; Hosie et al., 2018; Norris et al., 2019). Nlgn3 KO mice also display behavioral abnormalities including hyperactivity, normal sociability but deficits in social recognition, reduced ultrasound vocalization, impaired learning and memory, and abnormal pheromone preference (Radyushkin et al., 2009; Dere et al., 2018). Of note, there are inconsistencies in behavioral phenotypes observed among Nlgn3 mutant lines. For example, different Nlgn3-R451C KI lines have distinct sociability and learning and memory phenotypes (Tabuchi et al., 2007; Chadman et al., 2008). Even Nlgn3 KO mouse lines displayed different behavioral phenotypes. These conflicts have been considered due to different genetic backgrounds (Jaramillo et al., 2014, 2018). However, other factors, like housing environments, may also contribute to the inconsistent results (Burrows et al., 2017, 2020; Kalbassi et al., 2017). For example, social environment can affect social submissive behaviors or hierarchy. Male Nlgn3 KO mice under mixed genotype housing with male wild type littermates are submissive to unfamiliar wild type mice, whereas those under single genotype housing are not (Kalbassi et al., 2017). This indicates that the patterns of animal housing could potentially change behavioral results.

Aberrant striatal development and connectivity are linked to repetitive behaviors and abnormal reward circuitry in ASD (Hollander et al., 2005; Langen et al., 2009; Di Martino et al., 2011; Langen et al., 2014; Abbott et al., 2018). Striatum is divided into dorsal and ventral regions which include the caudate nucleus and putamen, and nucleus accumbens, respectively. Nlgn genes are expressed in both regions, however, Nlgn3 expression levels vary depending on the cell type and striatal region. Nlgn3 mRNA levels, but not Nlgn1 and Nlgn2, are significantly higher in D1-MSNs than in D2-MSNs in the nucleus accumbens. In the dorsal striatum, there are no differences in the expression of Nlgn1-3 (Rothwell et al., 2014). Concomitant to this expression profile, Nlgn3 KO selectively reduces inhibitory synaptic transmission in ventral D1-MSNs but not D2-MSNs without changing excitatory basal synaptic transmission and mGluR-dependent plasticity. Nlgn3-R451C KI mutant mice also display reduced inhibition in D1-MSNs in the nucleus accumbens suggesting that Nlgn3 localizes primarily at inhibitory synapses and regulates D1-MSNs (Figure 3A).

Both Nlgn3 KO and Nlgn3-R451C KI mice display enhanced rotarod performance and stereotypic behaviors (Table 2; Rothwell et al., 2014). Purkinje neuron-specific Nlgn3 KO is sufficient to induce hyperactivity but does not affect motor learning. In contrast, elevated motor learning behavior is regulated by Nlgn3 expressed in D1-MSNs in the nucleus accumbens but not in the dorsal striatum (Rothwell et al., 2014), indicating that ASD-associated behavioral phenotypes are precisely regulated by Nlgn3 expressed in specific brain structures.

In VTA DA neurons, cocaine, social reward, and novelty stimuli induce AMPAR-mediated synaptic plasticity by regulating GluR2-lacking AMPAR trafficking (Luscher and Bellone, 2008; Bariselli et al., 2016, 2018). Nlgn3 KO and DA neuron-specific Nlgn3 KD caused the dysregulation of GluA2-lacking AMPAR insertion which correlates with abnormal social behavior (Bariselli et al., 2018; Figure 3B).

Recent studies have highlighted Nlgn3 function in VTA DA neurons on social behavior. Social reward, sociability, and social novelty were reduced by specific Nlgn3 KD in VTA DA neurons (Bariselli et al., 2018). Social novelty deficits in Nlgn3 KO mice were rescued by specific expression of Nlgn3 in VTA DA neurons (Hornberg et al., 2020). Furthermore, Nlgn3 KO dysregulates the translation of oxytocin signaling in VTA DA neurons through the MAP kinase cascade (Figure 3B). This suggests that Nlgn3 is not only a trans-synaptic regulator but also regulates translation and oxytocinergic signaling. It is particularly important to elucidate the localization of Nlgn3 in DA neurons. Where does Nlgn3 localize at excitatory, or inhibitory synapses? Further investigations are required to map the expression of Nlgn3 at distinct synapses.

Gamma band oscillations implicated in various phases of hippocampal- and cortical-dependent cognitive behaviors (Buzsaki and Wang, 2012; Adaikkan and Tsai, 2020) and dysfunction of gamma oscillations have been found in people with ASD (Rojas and Wilson, 2014). Gamma oscillations are elicited by the firing of fast-spiking inhibitory interneurons and Pv+ inhibitory interneurons, which are the dominant fast-spiking interneuron cell type, to generate gamma oscillations underlying various animal behaviors (Cardin et al., 2009; Buzsaki and Wang, 2012). Importantly, selective deletion of Nlgn3 gene in Pv+ interneurons and Nlgn3 KO mice display reduced power of gamma oscillations in the hippocampus (Polepalli et al., 2017; Modi et al., 2019; Figures 4A,B). Furthermore, Pv+ interneuron-specific Nlgn3 KO mice exhibit abnormal hippocampus-dependent learning and memory, including fear memory retention and extinction (Polepalli et al., 2017). Dysregulated gamma oscillation was also found in Nlgn3-R451C KI mice (Cao et al., 2018). Fast-spiking interneurons in the medial prefrontal cortex (mPFC) exhibited reduced excitability while optogenetic restoration of Pv+ neuron excitability rescued deficits in social novelty preference. Interestingly, the social submissive behavior observed in Nlgn3 KO mice was rescued by expression of Nlgn3 in Pv+ interneurons, suggesting that Pv+ interneurons may be a primary target causing abnormal phenotypes in Nlgn3 KO mice (Kalbassi et al., 2017). Taken together, these results strongly suggest that Nlgn3 in Pv+ interneurons regulate cognitive behaviors and dysregulation of excitatory and inhibitory balance is one of the endophenotypes of ASD, which offers a new therapeutic approach for social behavior challenges in individuals with ASD.

Numerous in vitro studies that cultured dissociated neurons with non-neuronal Nlgn-overexpressed cells have demonstrated that expression of Nlgns in non-neuronal cells is sufficient to induce presynaptic differentiation (Scheiffele et al., 2000; Graf et al., 2004; Chubykin et al., 2005; Dong et al., 2007; Ko et al., 2009; Etherton M. R. et al., 2011; Sun et al., 2011). These results raise an attractive hypothesis that postsynaptic Nlgns regulate presynaptic functions in a retrograde manner. This hypothesis was confirmed in both Nlgn1 and Nlgn3 (Futai et al., 2007; Wittenmayer et al., 2009; Stan et al., 2010; Shipman et al., 2011). Postsynaptic overexpression of Nlgn1 or Nlgn3 in hippocampal CA1 pyramidal neurons in organotypic slice cultures increased release probability. Our follow-up study suggests that this retrograde modulation is mediated by trans-synaptic interactions between Nlgn and Nrxn (Futai et al., 2013). Overexpression of Nlgn3 can increase the number of presynaptic terminals along the dendrites of hippocampal primary neurons, supporting a role for Nlgn3 protein in synapse formation (Chih et al., 2005). Furthermore, in vivo overexpression of Nlgn3+A2 in the neocortex selectively upregulated inhibitory but not excitatory presynaptic proteins such as VIAAT and glutamic acid decarboxylase (GAD) 65 on the overexpressing neurons without increasing postsynaptic gephyrin expression (Fekete et al., 2015). These studies support the retrograde modulation of synaptic function by postsynaptic Nlgn3. However, these studies were restricted to an overexpression approach and no change in presynaptic function has been reported in Nlgn3 KO, Nlgn3-R451C, Nlgn3 R704C, and Nlgn1/2/3 triple KO mice (Varoqueaux et al., 2006; Tabuchi et al., 2007; Etherton M. R. et al., 2011; Chanda et al., 2017).

Importantly, a recent report supports that endogenous Nlgn3 regulates excitatory presynaptic release probability in the hippocampus. Polepalli et al. (2017) found that Pv+ interneuron-specific Nlgn3 KO caused the loss of group III mGluR-mediated inhibition in excitatory presynaptic terminals and increased excitatory synaptic release probability on Pv+ neurons. These results suggest that postsynaptic Nlgn3 negatively regulates presynaptic release probability through presynaptic mGluR signaling (Figures 4A,B). It will be particularly intriguing to elucidate the molecular mechanism that bridges postsynaptic Nlgn3 and presynaptic mGluRs. Interestingly, Nlgn3-mediated retrograde modulation is also found in hippocampal Cck+/CB1+ inhibitory synapses. Cck+/CB1+ inhibitory synapses are persistently suppressed via the constitutive activation of presynaptic CB1R which is mediated by ambient eCB released from postsynaptic neurons (Katona and Freund, 2012). This mode of eCB-mediated suppression at Cck+/CB1+ inhibitory synapses is disrupted in both Nlgn3 KO and R451C KI mice (Foldy et al., 2013), suggesting a critical role of Nlgn3 in tonic eCB-mediated signaling at Cck+/CB1+ inhibitory synapses (Figures 4A,C).

The most fundamental and intriguing role of Nlgns is the impact of their trans-synaptic interactions on synaptic function. However, elucidating how trans-synaptic interactions shape functional synaptic properties has remained a challenge. To understand the physiological roles of trans-synaptic molecules one must be able to manipulate the expression of these molecules in pre- and post-synaptic neurons simultaneously and then elucidate the functional consequences of this manipulation. We have recently advanced methodology for performing non-overlapping gene transfections which allows us to express Nlgn and Nrxn isoforms in pre- and post-synaptic neurons simultaneously and decipher roles of specific Nlgn3 and Nrxn interactions on hippocampal inhibitory synaptic transmission (Keener et al., 2020b, a; Uchigashima et al., 2020a). Inhibitory synapses are characterized by distinct neurochemical markers specific to input cell types such as VGT3, Pv, and Sst (Pelkey et al., 2017). Nlgn3-A overexpression can selectively potentiate VGT3+ synapses but not Pv+ or Sst+ synapses. In contrast, Nlgn3+ A2 overexpression potentiates neither VGT3+ nor Pv+ synapses, although Sst+ synapses show different phenotypes depending on the context of experiments (timing and duration of overexpression, animal species, and methods to induce action potentials) (Horn and Nicoll, 2018; Uchigashima et al., 2020a).

What mechanism underlies the input cell- and splice isoform-dependent function of Nlgn3-A protein at inhibitory synapses on CA1 pyramidal cells? One promising mechanism is a trans-synaptic interaction of Nlgn3-A protein with a specific Nrxn isoform. Each Nrxn isoform has been hypothesized to regulate synaptic function via trans-synaptic interactions with postsynaptic binding partners in an isoform-dependent manner (Sudhof, 2017a). Interneuron type-dependent transcription patterns of Nrxns have been demonstrated within the hippocampal CA1 region (Fuccillo et al., 2015; Uchigashima et al., 2020a). Notably, αNrxn1+AS4 mRNA is expressed at higher levels in VGT3+ interneurons than in Pv+ or Sst+ interneurons. In hippocampal slice cultures, Nlgn3-A-mediated potentiation at VGT3+ synapses is abolished in VGT3+ interneuron-specific Nrxn1/2/3 triple KO mice. In contrast, this deficit can be rescued by selective expression of αNrxn1+AS4 protein but no other Nrxns including αNrxn1-AS4 and βNrxn3+AS4 protein (Uchigashima et al., 2020a). These results highlight the importance of specific trans-synaptic interactions between Nlgn3-A and αNrxn1+AS4 proteins which underlie Nlgn3-A-mediated potentiation at VGT3+ synapses in an input cell-dependent manner (Figures 4A,D). However, this specific interaction cannot be explained by biochemical studies in which αNrxn1 protein can bind to Nlgn3 protein regardless of the presence or absence of the splice insertion at AS4 (Boucard et al., 2005). βNrxn3+AS4 protein can bind to all four Nlgn3 splice isoforms (Koehnke et al., 2010). Thus, the underlying molecular mechanism to generate trans-synaptic interactions between Nlgn3-A and αNrxn1+AS4 proteins remains elusive. One possible explanation is a co-factor that can interact with Nlgn3-A and αNrxn1+AS4 proteins to regulate their trans-synaptic interaction. Indeed, the trans-synaptic interaction of Nlgn3 and Nrxn proteins can be modulated by PTPδ or MDGA protein (Lee et al., 2013; Pettem et al., 2013; Connor et al., 2016; Elegheert et al., 2017; Gangwar et al., 2017; Yoshida et al., 2021). This modulation might be structurally or functionally affected by the presence or absence of the short amino acid chain encoded by the alternatively spliced sequence. Another possible explanation is a contribution of other alternatively spliced sequences on Nrxn proteins, the function of which has not been well characterized yet. Further investigations are required in the future.

Although the function of Nlgns has been extensively examined in the context of neurons, recent evidence indicates that Nlgn proteins can regulate glial morphogenesis and differentiation. Astrocytic Nlgn1-3 proteins are required for astrocyte morphogenesis both in vivo and in vitro (Stogsdill et al., 2017). The role of Nlgn1-3 proteins in astrocyte morphogenesis differs in an isoform- and time-dependent manner (Stogsdill et al., 2017). Additionally, the Nlgn3-R451C mutation affects astrocytic morphology by reducing their branch point number, branch length, and territory in the dentate gyrus (Matta et al., 2020). Furthermore, oligodendrocytic Nlgn3 protein contributes to the differentiation of oligodendrocytes (Proctor et al., 2015) and the density of microglia is increased in the dentate gyrus of Nlgn3-R451C KI mice (Matta et al., 2020). Interestingly, glial progenitor cells from individuals with schizophrenia express significantly lower levels of Nlgn1, Nlgn2, and Nlgn3 mRNA compared with controls (Windrem et al., 2017), suggesting a role for glial Nlgn3 protein in pathophysiological conditions. In Drosophila, Dnlg3, a homolog of vertebrae Nlgn3 protein initially identified as gliotactin, is transiently expressed on peripheral gila and required for the formation of the peripheral blood-nerve barrier (Auld et al., 1995).

Nlgn3 protein can function as a cortical neuronal activity-regulated glioma mitogen (Venkatesh et al., 2015, 2017). Nlgn3 extracellular ectodomain is shed from neurons and OPCs likely through ADAM10 in a neuronal activity-dependent manner. The secreted ectodomain of Nlgn3 can act on glioma cells via unknown binding partners and recruits the phosphoinositide 3-kinase (PI3K)-mTOR pathway to promote glioma cell proliferation. Nlgn3 exposure also induces feedforward expression of Nlgn3 protein at the transcriptional and translational levels, resulting in further increases in Nlgn3 protein expression in tumor cells. Importantly, patient-derived pediatric glioblastoma xenografts to the frontal cortex demonstrate a notable inhibition of glioma growth in Nlgn3 KO mice for up to 6 months. Therefore, Nlgn3 mRNA expression levels are inversely correlated with overall survival, providing a strong prediction of survival in human HGG. Interestingly, there is no effect of Nlgn1, Nlgn4X, Nlgn4Y or Nlgn2 on glioma proliferation, and no significant association between Nlgn2 expression and patient survival in adult glioblastoma multiforme (GBM), suggesting a unique function of Nlgn3 in brain tumors.

A prevalent concern among people with ASD revolves around gastrointestinal (GI) symptoms such as constipation, diarrhea, and abdominal pain (Holingue et al., 2018). In support of this, non-syndromic ASD models with the deletion or mutation of Nlgn3 gene exhibit abnormal GI function. Nlgn3-R451C KI mice show a faster small intestine transit of carmine red dye administered by oral gavage and increased sensitivity to GABA_A receptor-mediated modulation (Hosie et al., 2019). Similarly, Nlgn3 KO mice exhibit faster colonic migrating motor complexes (Leembruggen et al., 2020). Morphologically, no differences in the number of many GI neuron subpopulations have been reported in Nlgn3 KO or R451C KI mice, except for an increase in small intestine myenteric neurons in Nlgn3-R451C KI mice (Hosie et al., 2019; Leembruggen et al., 2020; Sharna et al., 2020).