Synaptic Cell Adhesion: A Structural Perspective

Abstract

Cell adhesion molecules (CAMs) play critical roles in mediating intercellular interactions in the context of the nervous system, such as guiding neuronal development, synapse formation and maturation, and synaptic plasticity. In addition to their extracellular adhesive roles, most CAMs induce intracellular signaling events and scaffold large protein complexes through intracellular domains. The molecular biology of how CAMs regulate synaptic development and function has been hugely advanced by decades of structural biology. These structures have illuminated multiple modes of CAM regulation, including how alternative splicing regulates CAM homotypic and heterotypic interactions. CAMs are diverse in size and contain a variety of adhesion domain classes such as immunoglobulin (Ig), leucine-rich repeats (LRR), and LamininG/Neurexin/Sex Hormone (LNS). In this chapter we detail structures of key synaptic adhesion complexes, including a mechanistic explanation of how these structures have informed functional work. Detailing the structural basis of synaptic adhesion provides a foundation for deciphering the complex interactions underlying neuronal connectivity and function in health and disease.

1. Introduction

Synapses represent the foundational units of communication within the human nervous system (NS), forming the labyrinthine network of connections between the eighty-six billion neurons which underlie all aspects of NS function (Scheiffele, 2003; Südhof and Malenka, 2008; Missler, Südhof and Biederer, 2012a; Südhof, 2018a, 2021a). Synapses allow neurons to transmit signals to one another, enabling the integration and processing of information and this occurs in a controlled and reproducible manner (Scheiffele, 2003; Südhof and Malenka, 2008; Missler, Südhof and Biederer, 2012a; Südhof, 2018a, 2021a). The dynamic properties of synapses, including their ability to undergo changes in strength and connectivity, underlie the brain's remarkable adaptability and capacity for learning and memory formation (Magee and Grienberger, 2020). Moreover, synaptic dysfunction is implicated in a wide range of neurological and psychiatric disorders, highlighting the pivotal importance of synapses in maintaining nervous system health and function (Südhof, 2021a). This chapter discusses the structural biology of cell adhesion molecules (CAMs) in synapse development and function. We detail representative structural studies of synaptic CAMs, and we review mechanistic details about synaptic adhesion and signaling events.

In the nervous system, CAMs play crucial roles in forming, organizing, maintaining, and pruning cell-cell interactions at various types of synapses (Yamagata, Sanes and Weiner, 2003; Missler, Südhof and Biederer, 2012a; Südhof, 2018a; Sanes and Zipursky, 2020). In response to adhesive events between axons and dendrites, intracellular signaling events are instigated in both pre- and post-synaptic cells which trigger the assembly of the synaptic machinery, specialization, and other events (Fig. 1). These signaling events include second messenger generation, scaffolding of elaborate protein networks, changes in the cytoskeleton, and potentially others (Fig. 1). Furthermore, since synapses can diverge into many specializations, this requires a multitude of CAMs on both sides of the developing synapse (Südhof, 2018b). Studies using proximity labeling experiments have identified the proteome at various synapse types (Ting et al., 2016; Cijsouw et al., 2018; van Oostrum et al., 2023). CAMs often function redundantly, which makes studying these systems difficult as perturbations can yield subtle phenotypes which are difficult to dissect (Südhof, 2018a; Sanes and Zipursky, 2020).

Figure 1: Synaptic CAM complexes are key for synapse development.

An excitatory synapse is drawn roughly to scale, with the pre-synaptic terminal on the left, and the post-synaptic terminal on the right. CAMs are shown in the synaptic cleft, and the events that they instigate are shown, such as pre-synpatic scaffolding (red and cyan scaffolding proteins) of synaptic vesicles (two concentric circles) and calcium channels (purple shapes). On the post-synaptic side, CAMs scaffold the tight clustering of neurotransmitter receptors such as NMDA or AMPA receptors and instigate intracellular signaling events such as G protein-signaling. The post-synaptic density is shown as a blue gradient to indicate that the region is densely packed with molecules. Puncta adherentia towards the edge of the synapse are shown, including the cadherin molecules that mediate symmetric adhesion and their links to the actin cytoskeleton. The image displays only the top half of the synapse for clarity.

The focus of this review is chemical rather than electrical synapses, and specifically excitatory synapses as they have been long studied as model systems (Rollenhagen and Lübke, 2006). Electron tomography analysis reveals that excitatory synapses typically have a synaptic cleft of 150-200 Å and can have a lateral size of ~500 nm (Fig. 1) (Burette et al., 2012; Tao et al., 2018; Cole and Reese, 2023). The synaptic cleft is densely packed with many electron-dense components in a highly organized fashion (Burette et al., 2012; Biederer, Kaeser and Blanpied, 2017; Tao et al., 2018; Cole and Reese, 2023); as early as 1959 a “band of extracellular material” was observed in the synaptic cleft (Gray, 1959) and this was later shown to have even higher electron density than the cytoplasm, suggesting extremely high local concentrations of biological macromolecules (Zuber et al., 2005). Observed density in the synaptic cleft bridges across the synapse and connects to intracellular components on both sides, consistent with the idea of CAMs locked in a transsynaptic adhesive embrace and coordinating intracellular events on either side of the synaptic cleft (Fig. 1) (Burette et al., 2012; Biederer, Kaeser and Blanpied, 2017; Tao et al., 2018; Cole and Reese, 2023). Studies using isolated synaptosomes, which can be isolated from brain tissue and form a minimal functional synapse, or neurons grown on grids, have achieved higher resolution in cryogenic electron microscopy (cryo-EM) strudies and may soon be able to resolve individual proteins at the synapse (Perez de Arce et al., 2015; Martinez-Sanchez et al., 2021; Sun et al., 2023; Held, Liang and Brunger, 2024). Overall, the synaptic cleft represents a crowded environment full of a multitude of interactions between different macromolecules.

Both homophilic and heterophilic CAM interactions are required for synapses and these coordinate to instigate a symphony of signaling events (Fig. 1) (Scheiffele, 2003; Südhof and Malenka, 2008; Togashi, Sakisaka and Takai, 2009a; Missler, Südhof and Biederer, 2012a; Südhof, 2018a, 2021a). For instance, homophilic interactions of classical cadherins are present during early synapse development and contribute to cytoskeletal coordination through catenin signaling (Seong, Yuan and Arikkath, 2015), and these interactions stabilize into puncta adherentia which are found adjacent to the synaptic active zone (AZ) and help to maintain structural integrity of the synapse (Togashi, Sakisaka and Takai, 2009b). The mature synapse is inherently asymmetric and requires heterophilic interactions between CAMs on axons and dendrites to develop (Südhof, 2021b). On the pre-synaptic side, molecules such as neurexins (Nrxs) and receptor protein tyrosine phosphatases (RPTPs) connect to adaptor molecules such as Calcium/Calmodulin activated serine kinase (CASK) (Hata, Butz and Südhof, 1996; Luo et al., 2020) and Liprinɑ (Marcó de la Cruz et al., 2024) to form the presynaptic AZ, which includes clustering of Ca²⁺ channels and formation of synaptic vesicles (SV) and the machinery for SV fusion. RPTP activity can also regulate synapse formation (Dunah et al., 2005) and is regulated in response to RPTP adhesion and clustering (Xie et al., 2020). On the post-synaptic side, molecules such as the neuroligins and latrophilins have conserved C-terminal PSD-95/Discs-large/ZO-1 (PDZ) ligands commonly interact with the PDZ domains of the membrane-associated guanylate kinase (MAGUK) or the SH3 and multiple ankyrin repeat (SHANK) families of scaffolding proteins (Irie et al., 1997; Tobaben, Sudhof and Stahl, 2000; Zhu, Shang and Zhang, 2016; Monteiro and Feng, 2017; Südhof, 2021b). MAGUK/SHANK proteins such as post-synaptic density 95 (PSD-95) can multimerize and scaffold large protein complexes which can induce the tight clustering of neurotransmitter receptors which is necessary for rapid response to neurotransmitter exocytosis (X. Chen et al., 2011; Chen et al., 2015; Zhu, Shang and Zhang, 2016). Recent work has begun to illustrate how MAGUKs and their scaffolded complexes can lead to phase-separated condensates (Zeng et al., 2019; Chen et al., 2020; Wu et al., 2021; Wang et al., 2024), consistent with the extremely high density of protein observed at the post-synaptic space in electron micrographs (Burette et al., 2012; Biederer, Kaeser and Blanpied, 2017; Tao et al., 2018; Cole and Reese, 2023). Furthermore, the post-synaptic adhesion G protein-coupled receptors (aGPCRs) including latrophilins (ADGRLs) and brain angiogenesis inhibitors (ADGRBs) are present on the post-synaptic side and can signal through heterotrimeric G protein pathways (Stephenson et al., 2013; Lala and Hall, 2022; Wang et al., 2024). ADGRBs can also coordinate cytoskeletal events directly through the RhoGEFs (Weng et al., 2019). Other signaling pathways such as protein kinase A and protein kinase B are important in synapse formation based on heterologous synapse formation assays (Jiang, Sando and Südhof, 2021). To summarize, a plethora of signaling events are unleashed following the heterophilic interaction and tight clustering of CAMs across the synapse.

Our understanding of cell adhesion is guided in part by the differential adhesion hypothesis (DAH) which suggests that cells organize themselves to minimize their adhesive free energy, considering the strengths of homophilic and heterophilic interactions that they engage in (Honig and Shapiro, 2020a). This is similar in principle to how oil and water will separate when mixed. However, the DAH does not consider several variables such as heterogeneity of protein expression across the cell surface, cis interactions, downstream signaling, and the plethora of adhesion molecules present on the cell surface of neurites (Honig and Shapiro, 2020a). Structural and molecular studies of the individual adhesion molecules that populate the neuron cell surface have helped further our understanding of how synaptic cell adhesion is guided (Fig. 2). Overall, CAM architecture manifests in a multitude of domain types and an extensive combination of these domains to make a variety of proteins (Honig and Shapiro, 2020b). This system is further diversified by spatiotemporal regulation of alternative splicing, protein expression, and post-translational modifications such as glycosylation (Missler, Südhof and Biederer, 2012b; Südhof, 2018b, 2021b; Honig and Shapiro, 2020b). These diversifications allow for the full complexity of the nervous system, including a full range of unique connections between neurons and specifications of a multitude of synapse types (Südhof, 2021b). Atomic models, primarily derived from X-ray crystallography and accompanying structural data, such as electron microscopy and small angle X-ray scattering have been instrumental in illuminating the mechanistic basis of CAM function at the synapse (Rudenko and Takahashi, 2019). These data help us to understand how CAMs operate within the tight distance constraints and high protein levels of the synaptic cleft, how they bind ligands, and how alternative splicing modulates CAM function.

Figure 2: The panoply of CAM interactions at the synapse lead to diverse outcomes.

The CAMs covered in this chapter are presented here to convey the complexity of events that can occur during synaptic development. Pre-synaptic molecules are presented on the left side, and post-synaptic molecules on the right side. Interactions validated by structural and mechanistic studies are shown using double-headed arrows.

Adhesion domains commonly present in CAMs

CAMs consist of a variety of extracellular domains which consist of domains composed mostly of beta-strands (Fig. 3). Features of CAM domains include disulfide bonds, Ca²⁺ binding sites, and glycosylation (Honig and Shapiro, 2020b). CAMs are often composed of multiple smaller domains in series as opposed to a single large fold (Honig and Shapiro, 2020b), potentially to allow for combinatorial shuffling of domains to increase protein diversity. Below, we outline several domain classes which are commonly found in CAMs and briefly discuss their overall features.

Figure 3: Common domain types observed in synaptic CAMs.

Common domains are shown using cartoon representations. Immunoglobulin domain: PDB 3M45, Lectin domain: PDB 6VHH, Laminin G domain: PDB 2R1D, Leucine-Rich Repeat domain: PDB 5Z8X.

Immunoglobulin domains form the largest superfamily of domains present in CAMs and consist of a beta-sandwich fold often stabilized by disulfide bonds (Fig. 3) (Honig and Shapiro, 2020b). Proteins such as RPTPs, MDGAs, SALMs, Slitrks, and more have Ig domains (Nam, Mah and Kim, 2011; Takahashi and Craig, 2013; Kim et al., 2017; Won, Lee and Kim, 2019). Proteins with Ig domains often have fibronectin type III (FNIII) domains which are also two sheet beta sandwich modules which are very similar in structure to Ig domains. Various surfaces of the Ig domain are used to interact with binding partners, although the longer lateral faces of the domain are typically used (Honig and Shapiro, 2020b).

The laminin/neurexin/sex hormone (LNS) domain fold consists of a beta-sandwich domain with a hypervariable surface on the shorter side face composed of loops (Rudenko et al., 1999). The hypervariable surface often coordinates Ca²⁺ and can be used to bind a variety of proteins (Rudenko et al., 1999). This domain type is most famously found in neurexins in the context of the synapse but is present in other proteins as well.

Leucine-rich repeat (LRR) domains consist of the conserved amino acid sequence, LxxLxLxxN/CxL, and are found as a tandem repeat of individual motifs, which results in an overall horseshoe-like shape. (Kobe and Kajava, 2001a; Roppongi, Karimi and Siddiqui, 2017). The LRR is capped on the N- and C-terminus by different capping regions (N-Cap and C-Cap) (Kobe and Kajava, 2001b; Roppongi, Karimi and Siddiqui, 2017). Various regions of the horseshoe can be used to interact with other proteins or in some cases proteoglycans. Synaptic proteins included in this class are the FLRT, LRRTM proteins as well as the SALM molecules (Nam, Mah and Kim, 2011; Lu et al., 2015; Roppongi, Karimi and Siddiqui, 2017).

Other proteins have unique folds adapted from other protein classes. Teneurins are extremely large synaptic molecules which contain several domains organized around a large beta-barrel and bear striking structural homology to bacterial Tc-toxins (Jackson et al., 2018; Li et al., 2018). Various surfaces of the beta barrel are used to interact with protein ligands. Neuroligin is another example, and it has an esterase domain similar to acetylcholinesterase which is an ɑ/β hydrolase fold (Araç et al., 2007); the neuroligin esterase domain is incompatible with catalysis. In the following sections we detail representative structural studies of important synaptic CAMs and how they inform mechanistic detail about synaptic adhesion and signaling events. Whenever possible, our structural renderings in the presented figures are to scale.

2. Neurexins and their interaction partners

The presynaptic Neurexins (Nrxs) and their post-synaptic binding partners, most notably the neuroligins (NLs), are broadly known as some of the most famous synaptic adhesion molecule (Südhof, 2017; Rudenko, 2019; Gomez, Traunmüller and Scheiffele, 2021). They serve as a model system for how we understand synaptic CAMs. Neurexins were identified due to their role as the Ca²⁺-dependent receptor for ɑ-latrotoxin (ɑLTx) (Petrenko et al., 1990). These proteins have been studied for over 25 years and are involved in regulating synaptic specificity and maturation (Südhof, 2017; Rudenko, 2019; Gomez, Traunmüller and Scheiffele, 2021). Nrxs and NLs are tightly associated with autism spectrum disorder and other neurological conditions (Südhof, 2017; Rudenko, 2019; Gomez, Traunmüller and Scheiffele, 2021). In mammals, a vast array of Nrx protein products is generated and presented on the pre-synaptic surface through a combination of three genes (Nrx1-3), each with three alternative promoters (ɑ, β, ɣ) and alternative splicing at six sites (SS1-6) (Schreiner et al., 2014; Treutlein et al., 2014; Fuccillo et al., 2015). Nrxs can interact with post-synaptic ligands depending on which version is produced and which ligands are present post-synaptically (Südhof, 2017; Rudenko, 2019). These ligands include various post-synaptic and soluble proteins which are highlighted in the following sections (Südhof, 2017; Rudenko, 2019).

The longest Nrx version, Nrxɑ, has 6 LNS domains (LNS1-LNS6) with single EGF repeats between LNS1 and LNS2 (EGFA), LNS3 and LNS4 (EGFB) and LNS5 and LNS6 (EGFC), a heparan sulfate (HS) modification between LNS6 and the membrane, and a small transmembrane region (TMR) (Fig. 4A) (Südhof, 2017; Gomez, Traunmüller and Scheiffele, 2021). Nrxβ contains only the LNS6 domain, a HS modification site, and the TMR (Südhof, 2017; Gomez, Traunmüller and Scheiffele, 2021). Nrxɣ is the shortest isoform and only has a short extracellular region which is HS modified and the TMR (Südhof, 2017; Gomez, Traunmüller and Scheiffele, 2021). At the far end of the intracellular C-terminus, Nrxs have a PDZ ligand, as many synaptic CAMs do (Fig. 4A) (Hata, Butz and Südhof, 1996; Rudenko, 2019; Gomez, Traunmüller and Scheiffele, 2021). Six alternatively spliced sites (SS1-6) generate further Nrx diversity. Sequencing studies detect thousands of neurexin isoforms within the mammalian brain (Schreiner et al., 2014; Treutlein et al., 2014; Fuccillo et al., 2015) and isolated neuron populations contain specific versions of neurexins rather than a broad array of neurexin isoforms (Fuccillo et al., 2015). Together, these data suggest that individual cells use a specific repertoire of neurexins to coordinate synaptic connections.

Figure 4: Features of the pre-synaptic hub molecule Neurexin.

A: Domain architectures of the neurexins. The sites of alternative splicing are shown as vertical dotted lines. SS: splice site. LNS: Laminin G, Neurexin, Sex-hormone. EGF: epidermal growth factor domain. B: The structure of Nrx1β is shown in cartoon representation with splice sites highlighted using grey spheres. C: The structure of Nrx1ɑ is shown in cartoon representation with D: an orthogonal view.

Nrx1β was the first to be structurally characterized (Fig. 4B) (Rudenko et al., 1999) and revealed the three-dimensional architecture of the LNS fold as two parallel beta sheets, each with seven strands. Importantly, this structure revealed that although SS2, SS3, and SS4 occur in different LNS domains, they all localize to the loop regions on the same surface on the LNS fold known as the hypervariable surface (Fig. 4B), suggesting that this surface forms a regulatory surface or a ligand binding surface.

Over a decade later, the Nrx1ɑ crystal structure was determined (F. Chen et al., 2011), which consists of an L-shaped molecule with LNS2-5 making the long leg of the L and the LNS6 protruding ~90° from LNS2-5 (Fig. 4C, D). In relation to this L shape, the LNS1 domain and EGFA are not ordered in crystal structures of Nrx1ɑ. Additional work confirmed that there are two major points of flexibility in Nrx1ɑ (Liu et al., 2018). First between LNS1 and LNS2, and second between LNS5 and LNS6. Based on their arrangement in Nrx1ɑ (F. Chen et al., 2011; Liu et al., 2018), the LNS domain hypervariable surfaces, some of which are ligand binding sites, present on the same side of the structure (Fig. 4C). This implies that multiple ligands could be recruited simultaneously if they bind to different LNS repeats (Liu et al., 2018). Furthermore, if protein partners need to interact with more than one LNS domain at the same time, it is possible in this arrangement (Liu et al., 2018). The orientation of the hypervariable surfaces also suggests the orientation of the Nrxɑ molecule relative to the membrane- the long leg of the L, including LNS1-5, likely sits parallel to the membrane surface (F. Chen et al., 2011; Liu et al., 2018). This orients the hypervariable surfaces of the LNS domains towards the post-synaptic membrane for post-synaptic adhesion molecules to interact with (Liu et al., 2018). This Nrx1ɑ structural data also revealed the three-dimensional positions of other splice inserts. SS1 is located between EGFA and LNS2, and SS6 is between LNS5 and EGFC (Fig. 4C), suggesting that these splice inserts could modulate the level of flexibility between domains and allow Nrx1ɑ to access different binding partners in the synaptic cleft (Liu et al., 2018). Finally, the SS5 insert contains a putative protease cleavage site which can be cleaved by metalloproteases which are present in the synaptic cleft (Liu et al., 2018). Thus, the SS5 insert may be a mechanism to generate secreted Nrx modules or to turnover Nrx-mediated adhesion sites.

Nrxs can engage in several heterophilic protein-protein interactions which leads to a multitude of adhesive outcomes (Südhof, 2017; Rudenko, 2019; Gomez, Traunmüller and Furthermore, these heterophilic interactions are regulated by alternative splicing (Südhof, 2017; Rudenko, 2019).

Neuroligins (NLs) are the best characterized Nrx binding partner and are membrane tethered proteins with large extracellular domains with homology to ɑ/β hydrolases such as acetylcholinesterases (Bemben et al., 2015). Following the extracellular domain (Fig. 5A), NLs have short TMRs, an intracellular region, and a PDZ ligand at the C-terminus (Fig. 5A) (Irie et al., 1997). There are four NLs in humans and NLs are highly found frequently mutated in autism spectrum disorder (ASD) (Bemben et al., 2015). Each NL is localized to different synaptic types, with NL1 and NL4 located at excitatory synapses, NL2 located at inhibitory synapses, and NL3 has been found at both types (Bemben et al., 2015; Marro et al., 2019). NLs are obligate dimers; mostly NLs are known to form homodimers, however there is evidence for NL heterodimers (Budreck and Scheiffele, 2007; Poulopoulos et al., 2012; Bemben et al., 2015). NLs are alternatively spliced at two sites known as splice site A and splice site B (SSA, SSB) (Fig. 5A) which regulate interactions with its binding partners (Chih, Gollan and Scheiffele, 2006; Koehnke et al., 2010; Lee, Dean and Isacoff, 2010).

Figure 5: Features of Nrx/NL interaction and modulation by MDGA.

A: Domain architectures and dimeric structure of the Neuroligins. B, C: Domain architectures and structures of the Nrx/NL interaction with orthogonal view shown in D. E: Domain layouts and structure of the MDGA/NL interaction with orthogonal view shown in F.

The first structures of NL were determined in 2007; these were released coincident with Nrxβ-NL complex structures (Araç et al., 2007; Fabrichny et al., 2007; Chen et al., 2008). These structures revealed that NLs do have ɑ/β hydrolase folds with a central twisted β-sheet that is surrounded by ɑ-helices (Fig. 5A). The obligate NL dimer is mediated by two helices from each monomer, with several hydrophobic residues involved. The esterase catalytic site residues are not compatible with catalysis, and the entrance to the active site is not solvent exposed as it would be in a catalytically competent esterase (Araç et al., 2007).

The Nrxβ-NL complex structures revealed the previously described Ca²⁺-dependent interaction (Ichtchenko, Nguyen and Südhof, 1996; Nguyen and Südhof, 1997) between the hypervariable surface of the LNS6 domain of Nrxβ (Fig. 5B) with the side face of the esterase domain of NL (Fig. 5 C, D) (Araç et al., 2007; Fabrichny et al., 2007; Chen et al., 2008). Strikingly, these structures revealed that the three-dimensional location of both the Nrx SS4 and NL SSB splice inserts are directly at the Nrxβ-NL binding interface, providing an atomic explanation for previous observations of splice-site specific regulation of Nrxβ-NL binding (Fig. 5 C, D) (Ichtchenko, Nguyen and Südhof, 1996; Boucard et al., 2005; Comoletti et al., 2006). Both the presence of Nrxβ SS4 and NL SSB reduce the affinity of the Nrxβ-NL interaction (Ichtchenko, Nguyen and Südhof, 1996; Boucard et al., 2005; Comoletti et al., 2006) which follows as these inserts will add extra residues directly at the Nrx-NL interface and could disrupt the tight network of interactions that form. Although there is no high-resolution structure of Nrxɑ in complex with any NL, biochemical studies have shown that these proteins do interact (Boucard et al., 2005).

The MAM-domain containing glycosylphosphatidylinositol anchor proteins (MDGA) form a regulatory axis for the Nrx/NL interaction (Pettem et al., 2013; Connor et al., 2016). MDGA proteins are found on the postsynaptic side and can inhibit Nrx/NL dependent synapse development of both excitatory and inhbitory synapses (Pettem et al., 2013; Connor et al., 2016). MDGA proteins contain ECRs with six Ig domains (Ig1–6), one FnIII domain, and a MAM domain (Fig. 5E). At the C-terminus these proteins are linked to the membrane by a glycosylphosphatidylinositol (GPI) anchor. Three groups determined crystal structures of the MDGA1 ECR in complex with NL (Elegheert et al., 2017; Gangwar et al., 2017; Kim et al., 2017). One structure used the full MDGA ECR, and this revealed a triangular architecture where the Ig1 domain contacts the FnIII domain (Fig. 5 E, F) (Elegheert et al., 2017). Ig1/2 form one face of the triangle, Ig3/4 the next, and Ig5/6 and FnIII the last face. Complexes of MDGA1 with NL1 revealed that the Ig1 domain forms extensive contacts with the side face of one NL protomer and Ig2 contacts the other NL protomer in a Ca²⁺-independent manner (Fig 5. E, F) (Elegheert et al., 2017; Gangwar et al., 2017; Kim et al., 2017). The binding surface of MDGA Ig1 to NL1 overlaps with LNS6 of Nrx binding to NL, which strongly suggests a mechanism for how MDGA proteins can regulate Nrx/NL interactions, through direct steric occlusion (Elegheert et al., 2017; Gangwar et al., 2017; Kim et al., 2017). The MDGA/NL interaction has a similar binding affinity as Nrx/NL, suggesting that these two interactions can compete depending on the expression levels of each protein present and which SS4 splice state Nrx is present in (Elegheert et al., 2017; Kim et al., 2017).

Nrxs also bind to another post-synaptic ligand, leucine-rich repeat transmembrane neuronal proteins (LRRTMs) (Ko et al., 2009; Siddiqui et al., 2010; Roppongi, Karimi and Siddiqui, 2017). LRRTMs are synaptogenic proteins implicated in neurological disorders such as schizophrenia, ASD, Alzheimer’s disease, and others (Roppongi, Karimi and Siddiqui, 2017). There are four LRRTM proteins in mammals, and the LRRTM domain architecture consists of an extracellular LRR region with N- and C-caps, followed by a single-pass transmembrane region and an intracellular region (Fig. 6A). Additionally, there is a C-terminal PDZ-binding motif which binds to MAGUKs, and this can induce clustering of excitatory neurotransmitter receptors (Roppongi, Karimi and Siddiqui, 2017). The structure of LRRTM2 in complex with Nrx1β revealed that LNS6 uses its hypervariable surface to interact with the C-cap of LRRTM2 in a Ca²⁺-dependent manner, and that this binding interaction directly conflicts with the Nrx/NL interaction (Fig 6B) (Yamagata et al., 2018). The binding affinity of Nrx/LRRTM2 is similar to Nrx/NL, but the binding affinity of Nrx to LRRTM2 is strikingly decreased in Nrx SS4+, whereas the shift due to SS4+ is more subtle for Nrx/NL (Yamagata et al., 2018). Thus, Nrx SS4+ will preferentially interact with NL over LRRTM.

Figure 6: Features of the Nrx/LRRTM and RPTP/NL interactions.

A: Domain architectures of neurexins and LRRTMs. B: Structure of the Nrx/LRRTM interaction C: Domain layouts of the RPTP and NLs. D: Structure of the RPTP/NL interaction with orthogonal view shown in E.

The type IIA receptor protein tyrosine phosphatases (RPTPs) form another family of pre-synaptic adhesion molecules which interact with NLs in a splice-isoform dependent manner (Takahashi and Craig, 2013). RPTPs contribute to synaptic adhesion and synapse organization, and they are associated with ASD and schizophrenia (Takahashi and Craig, 2013). RPTPs contain a series of three Ig domains (Ig1-3), four or eight FnIII domains depending on splicing (FnIII1–8), a short TMR, and two protein tyrosine phosphatase domains (Fig. 6C). RPTP function is modulated by four alternatively spliced inserts termed mini-exons A-D (meA-meD), each encoding short amino acid sequences with less than ten residues (Takahashi and Craig, 2013). meA inserts in Ig1, meB inserts between Ig2 and Ig3, meC in FnIII5, and meD in the region between the TMR and the first phosphatase domain. PTPδ is an RPTP which interacts with NL3, but only if the meB splice insert is absent (meB−) (Yoshida et al., 2021). Ig2 and Ig3 of PTPδ meB− interact with the esterase domain of NL3, where Ig3 sterically occludes LNS6 of Nrx from binding (Fig. 6D, E) (Yoshida et al., 2021). In contrast, PTPδ meB+ interacts with a separate set of ligands using Ig2 and Ig3; this is reviewed in a later section of this chapter (Takahashi and Craig, 2013; Yoshida et al., 2021).

Recent work has shown the importance of soluble molecules in the trans-synaptic space which can bridge pre- and post- synaptic CAMs together to mediate adhesion. Neurexophilins (Nxphs) are a family of secreted glycoproteins which are thought to play a role in GABAergic neurotransmission (Fig. 7A) (Born et al., 2014). Nxphs interact with the LNS2 domain of Nrx using a distinct surface, not the hypervariable surface employed for most Nrx interactions. This is done via a unique architecture where Nxph forms a contiguous β-sheet with Nrx LNS2 which extends the β-sandwich fold of the LNS2 domain (Fig. 7B, C) (Wilson et al., 2019). Nxph also provides some side chains which interact with LNS2 residues (Wilson et al., 2019). This Nrx/Nxph interaction is regulated by Nrx splice site SS2, where SS2+ increases Nxph affinity to Nrx (Wilson et al., 2019). Two versions of SS2+ both increase the affinity of Nxph for Nrx (Wilson et al., 2019).

Figure 7: Features of the Nrx/Nxph and Nrx/Clbn/GluD2 interactions.

A: Domain architectures of neurexins and neurexophilins. B: Structure of the Nrx/Nxph interaction with orthogonal view in C. D: Domain layouts of the Neurexins, Cerebellins, and Glutamate D2 receptors. E: Structure of the Nrx/Clbn/GluD2 interaction.

The cerebellin (Clbn) proteins are soluble molecules that link pre-synaptic Nrx to post-synaptic ionotropic Glutamate delta receptors (GluD1-2) to regulate levels of certain neurotransmitter receptors such as AMPA-type and NMDA-type receptors (Südhof, 2023). This is a phenomenon that depends on the adhesive properties of the GluD receptor rather than ion transduction; the transmembrane domain of GluD can be removed without defects in synapse formation whereas the intracellular portion of GluD is necessary for synapse formation (Dai et al., 2021) There are four Clbns in humans, and they consist of cysteine-rich amino terminal domains (NTDs) and C1q-like globular domains (Fig. 7D). Clbns form obligate trimers via their C1q domains, and these trimers can form dimers via its NTD to form a hexameric Clbn (Fig. 7D). Crystal structures and accompanying biophysical and biochemical data (Cheng et al., 2016; Elegheert et al., 2016) revealed that the amino-terminal domain (NTD) of GluD2 receptors can interact with a trimer of the Clbn1 C1q domain (Fig. 7E). Furthermore, a hexameric Clbn1 is necessary for interaction with Nrx1β, and the Clbn/Nrx1 interaction occurs only for Nrx1 SS4+ (Uemura et al., 2010; Elegheert et al., 2016). The interaction of Nrx1β with Clbn1 includes the hypervariable surface, but the high-resolution structure is not known (Elegheert et al., 2016).

For Nrxs, the SS4+ splice site serves as a master switch to regulate several interactions between LNS6 and post-synaptic adhesion partners. In the context of SS4+, interactions with NLs are slightly reduced in affinity, but the Nrx/NL pair still can interact (Ichtchenko, Nguyen and Südhof, 1996; Boucard et al., 2005; Comoletti et al., 2006). With Nrx SS4+, NL is more likely to interact with MDGA than Nrx as the complex affinities are similar (Elegheert et al., 2017). Next, interactions between Nrxs and LRRTMs are strongly reduced (Yamagata et al., 2018). Finally, interactions between Nrxs and Clbns are strongly enhanced, which form a trans-synaptic bridge to GluD2 (Elegheert et al., 2016).

How these proteins mediate their effects at the synapse are largely dependent on their intracellular PDZ ligands interacting with PDZ domains, most commonly of the MAGUK or SHANK families. Most of the above-mentioned proteins have PDZ ligands in their C-termini, and the structures of neurexin bound to the MAGUKs calcium/calmodulin dependent serine protein kinase CASK (PDB 6NID) (Hata, Butz and Südhof, 1996) and spinophilin (Muhammad et al., 2015) are known. Nrx interaction with CASK has been shown and CASK can phosphorylate the intracellular tail of Nrx (Mukherjee et al., 2008) and nucleate filamentous actin from Nrx (Biederer and Südhof, 2001). This Nrx/CASK interaction is crucial for organization of the pre-synaptic active zone through the Liprinɑ proteins (Marcó de la Cruz et al., 2024); the structure of CASK bound to Liprinɑ is also known, providing a pathway from Nrx to Liprinɑ (Wei et al., 2011). On the post-synaptic side, Neuroligin binds to the MAGUK PSD-95 (Irie et al., 1997) and several other PDZ domain proteins (Meyer et al., 2004) and the intracellular portion of GluD2 is necessary for its synaptic function through a PDZ interaction (Dai et al., 2021). NLs are shown to interact with a variety of MAGUKs (Irie et al., 1997; Meyer et al., 2004). MAGUKs have three PDZ domains as well as Guanylate kinase-like domains that bind to phosphorylated peptides, and thus they link huge assemblies of molecules (Zhu, Shang and Zhang, 2016) and likely play a master role in synapse organization.

Latrophilins and their binding partners

Another CAM interaction hub is formed by the presynaptic teneurins (TENs) and fibronectin leucine-rich repeat transmembrane proteins (FLRTs) interacting with the postsynaptic latrophilins (ADGRLs or LPHNs). ADGRLs were originally identified due to their role as the Ca²⁺-independent receptor of ɑLTx (Krasnoperov et al., 1997). ADGRLs are post-synaptic molecules that play key roles in synapse development through scaffolding of post-synaptic protein complexes as well as their ability to induce signaling through heterotrimeric G proteins (Sando and Südhof, 2021; Wang et al., 2024). These key receptors integrate signals from various sources and output through several mechanisms (Sando, Jiang and Südhof, 2019) making them kingpins of the synapse (Fig. 8). ADGRL function can now be modulated due to the recent development of pioneering synthetic antigen binders (sABs) targeting ADGRLs (Kordon et al., 2023). ADGRLs are some of the most highly relevant synaptic proteins (Südhof, 2018b) and through their interaction with the pre-synaptic TENs and FRLTs, the ADGRLs direct specific synapse development depending on the identity and splice isoform of each molecule present (Li et al., 2018; Sando, Jiang and Südhof, 2019). ADGRLs are associated with attention deficit hyperactivity disorder (ADHD), bipolar disorder, autism spectrum disorder (ASD), and cancers (Meza-Aguilar and Boucard, 2014). Shortly after their discovery, ADGRLs were bioinformatically identified as members of G protein-coupled receptor (GPCR) family (Lelianova et al., 1997; Sugita et al., 1998) and were later classified into the adhesion family of GPCRs (aGPCRs) (Fredriksson et al., 2003). There are three ADGRLs in vertebrates, and the domain architecture consists of a lectin (Lec) domain, an olfactomedin (Olf) domain, a hormone receptor (HormR) domain, and the GPCR autoproteolysis-inducing (GAIN) domain (Araç et al., 2012) (Fig. 9A). Following this is a GPCR seven-transmembrane domain (7TM) as well as a C-terminal intracellular region (Krasnoperov et al., 1997) with a PDZ ligand. ADGRLs are alternatively spliced in at least five different sites (termed herein using nomenclature for Mus musculus ADGRL3 SS1-SS5 from (Wang et al., 2024)). SS1 occurs at the far N-terminus of the receptor before the Lec domain, SS2 occurs between the Lec and Olf domains (this was previously referred to as ‘SSA’ (Boucard, Maxeiner and Südhof, 2014)), SS3 in the GAIN domain, SS4 in the 3^rd intracellular loop of the 7TM region, and SS5 at the far C-terminus (Wang et al., 2024). SS2 can regulate binding to teneurins (Boucard, Maxeiner and Südhof, 2014) and SS4/SS5 can modulate synaptic function of ADGRLs (Y. Wang et al., 2021; Wang et al., 2024).

Figure 8: The Latrophilins as a post-synaptic signal integrator.

The various interactions and features of latrophilins are shown as cartoons, including binding partners, modes of signaling, and scaffolding interactions.

Figure 9: The network of Latrophilin interactions revealed by structural biology.

A: domain architectures of the proteins involved in ADGRL interaction. B: Structural model of a super-complex between TEN/FLRT/ADGRL/UNC5D.

Biochemical studies in the early 2000s showed that ADGRLs consist of two non-covalently linked sequences, termed the N-terminal fragment (NTF) and C-terminal fragment (CTF). The NTF and CTF are generated by post-translational proteolysis at the GPCR proteolysis site (GPS) (Krasnoperov et al., 2002); the function of the GPS remained a mystery for over a decade. In 2012, the crystallographic structure of the ADGRL1 GAIN domain was revealed (Araç et al., 2012). This structure showed that the GAIN domain, the only ECR domain conserved throughout 32 of 33 aGPCRs (Prömel, Langenhan and Araç, 2013), houses its cleaved, yet associated, final β-strand which is the N-terminus of the CTF (Araç et al., 2012) and this constitutes the GPS. The relevance of this was still unclear until work from two groups (Liebscher et al., 2014; Stoveken et al., 2015) revealed that the final β-strand of the GAIN domain acts as the orthosteric agonist of the 7TM, leading to its designation as the tethered agonist (TA). The GAIN domain shields the TA inside of it until the NTF is shed from the CTF; this is hypothesized to occur via ligand binding and/or force application (Vizurraga et al., 2020; Lala and Hall, 2022). This TA-dependent mechanism for aGPCR activation now has been shown for many aGPCRs (Liebscher et al., 2014;Demberg et al., 2015, 2017; Schöneberg et al., 2015; Stoveken et al., 2015; Schöneberg, Kleinau and Brüser, 2016; Wilde et al., 2016; Perry-Hauser et al., 2022; Bernadyn et al., 2023) and the cryogenic electron microscopy (cryo-EM) structures of several aGPCRs (including ADGRL3) with the TA bound in the orthosteric site are available (Barros-Álvarez et al., 2022; Ping et al., 2022; Qu et al., 2022; Xiao et al., 2022). Furthermore, cryo-EM structures of ADGRL3 in complex with the four major classes of heterotrimeric G proteins are available (Qian et al., 2022).

However, efficient cleavage in the GAIN domain is not required for several functions of ADGRLs, including synapse formation (Prömel et al., 2012; Sando, Jiang and Südhof, 2019), suggesting that ADGRLs can operate independent of TA-mediated activation. There is work which supports the idea that allosteric communication between the ECR and 7TM can regulate signaling in several aGPCRs (Kishore et al., 2016; Salzman et al., 2017; Kordon et al., 2023, 2024; Bandekar et al., 2024). Furthermore, available evidence suggests that some aGPCRs are not cleaved in the consensus GPS within the GAIN domain (Bui et al., 2023), and some do not respond to TA stimulation in assays for G protein signaling or downstream pathways (Favara et al., 2021; Dates et al., 2024). Emerging work has shown the relative ECR/7TM orientation of ADGRL3 using cryo-EM and demonstrated that changes in ADGRL3 signaling are linked to changes in ECR/7TM orientation (Kordon et al., 2024). Taken together, this evidence suggests that the ECR of aGPCRs are not simply caps for the TA; they also communicate directly with the 7TM and thus merit intense study as the crown jewels of these enigmatic receptors.

The structure of the HormR domain is now known for several aGPCRs (Araç et al., 2012; Leon et al., 2020; Bandekar et al., 2024). Mysteriously, the function of the HormR domain remains unknown in any aGPCR although HormR domains are present in several aGPCR subfamilies (Vizurraga et al., 2020; Lala and Hall, 2022). Studies in the 2010s revealed the structural basis of the interaction of ADGRLs with their physiological ligands, the fibronectin leucine rich transmembrane proteins (FLRTs) and the teneurins (TENs) which are detailed below.

FLRTs were identified as candidate ADGRL ligands in an affinity chromatography-mass spectrometry screen against the ADGRL ECR (O’Sullivan et al., 2012) and they were shown to interact directly with ADGRLs and to form trans-synaptic complexes. FLRTs were shown to regulate excitatory synapse function (O’Sullivan et al., 2012).There are three FLRTs in mammals (FLRT1-3) (Lacy et al., 1999) and they have an N-terminal LRR region with N- and C- terminal caps, a FnIII domain, and a short transmembrane region (Fig. 9A) (Lu et al., 2015). FLRTs form a weak dimer species through their LRR which is thought to be a cis dimer (Seiradake et al., 2014; Lu et al., 2015).

Structural and biophysical studies performed by three competing groups revealed that the Olf domain of ADGRL interacts with the concave side of the FLRT LRR to make a tight complex (Jackson et al., 2015; Lu et al., 2015; Ranaivoson et al., 2015) (Fig. 9B) and that this interaction occurs in trans. Another post-synaptic FLRT ligand, uncoordinated 5 (UNC5), interacts with FLRT in a manner incompatible with FLRT dimerization (Fig. 9B) (Seiradake et al., 2014; Lu et al., 2015), although the relevance of this interaction to synaptic development is unclear.

The other major ligand of ADGRLs are the TENs. TENs are a family of type II transmembrane proteins that play essential roles in neuronal development, synaptogenesis, and circuit formation (Young et al., 2013; Berns et al., 2018). TENs regulate synapse formation (Leamey and Sawatari, 2014; Woelfle, D’Aquila and Lovejoy, 2016) and TEN perturbations impair synapse development (Silva et al., 2011; Hong, Mosca and Luo, 2012; Mosca et al., 2012; Boucard, Maxeiner and Südhof, 2014; Mosca, 2015). Mutations in TENs are linked to microphthalmia, congenital anosmia, essential tremors, and cancers (Aldahmesh et al., 2012; Ziegler et al., 2012; Hor et al., 2015; Alkelai et al., 2016; Chassaing et al., 2016; Graumann et al., 2017; Talamillo et al., 2017). There are four TENs in vertebrates (TEN1-TEN4), two in flies (TEN-m and TEN-a), and one in worms (Tucker et al., 2012; Tucker, 2018; Wides, 2019). TENs consist of an N-terminal intracellular region, a short TMR, and a large extracellular region with eight epidermal growth factor (EGF)-like repeats (EGF1-8) and a C-terminal region termed the TEN “superfold” which contains a cysteine-rich domain (CRD) which binds several Ca²⁺ ions, a transthyretin-like domain (TTR), two Ig-like domains (Ig), a β-propeller domain (β-prop), an extremely large β-barrel domain, and a toxin-like domain (Tox) (Fig. 9A) (Jackson et al., 2018; Li et al., 2018; Meijer et al., 2022). TENs form constitutive cis dimers linked by two intermolecular disulfide bonds in EGF2 and EGF5 (Li et al., 2018).

TENs were identified as binding partners of the ADGRLs using affinity chromatography and shown to interact trans-synapticallly (Silva et al., 2011). The minimal TEN binding region of ADGRLs was identified as the Lec domain (Boucard, Maxeiner and Südhof, 2014). Furthermore, the presence of SS2 in ADGRL1 was shown to decrease binding affinity to TENs without affecting the affinity of the ADGRL/FLRT interaction (Boucard, Maxeiner and Südhof, 2014). Although TENs have long been studied biochemically (Feng et al., 2002), structural work proved elusive as a large part of their sequences are not similar to any known eukaryotic protein. The first high-resolution structures of TENs were released in 2018 when two groups in parallel determined the structure of the monomeric TEN ECR (Jackson et al., 2018; Li et al., 2018). These structures revealed the striking architecture of the TEN ECR (Fig. 9B) termed the TEN “superfold” (TEN^SF)- a module of domains organized around an extremely large β-barrel in an architecture with strong structural resemblance to bacterial Tc toxins (Li et al., 2018). Within the TEN^SF, the Ig and β-propeller cap the “bottom” of the β-barrel and the Tox domain begins inside of the β-barrel and protrudes out of the side, and partially being present outside of the β-barrel on its side surface. The TTR and CRD domains interact with the Ig domain as part of the same TEN^SF module (Meijer et al., 2022). Later work from the same two groups showed that the ADGRL Lec domain interacts with the TEN β-barrel on the side of the β-barrel roughly opposite to Tox (del Toro et al., 2020; Li et al., 2020) and the Olf domain also interacts with the top corner of the barrel (Fig. 9B). This interaction is compatible with FLRT-Olf binding, in agreement with previous reports that binding of both FLRT and TEN to ADGRL is required for excitatory synapse specification (Sando, Jiang and Südhof, 2019) and the triple complex of ADGRL/TEN/FLRT was shown using soluble ECR constructs (Li et al., 2020). The ADGRL/TEN complex structures suggested a mechanism for how the presence of SS2 decreases ADGRL/TEN affinity as previously observed (Boucard, Maxeiner and Südhof, 2014). The SS2- structure (PDB 6SKA) (del Toro et al., 2020) (PDB 6SKA) has high-quality density for the Olf domain which also packs against TEN, whereas the SS2+ structure (PDB 6VHH) (Li et al., 2020) does not clearly resolve the Olf domain, suggesting the splice insert, which adds 5 residues in between Lec and Olf increases the spacing between these two domains in order to modulate binding of Olf to TEN.

Based on the known binding modes of the ADGRL ligands, a super-complex of TEN/FLRT/ADGRL/UNC5D can be modeled, suggesting that ADGRLs form the center of a nexus of signals from several inputs (Fig. 9B) (Lu et al., 2015; Jackson et al., 2016; Li et al., 2018, 2020; Araç and Li, 2019a, 2019b).

Similar to the Nrx paradigm, alternative splicing of pre-synaptic TENs regulates their interactions with post-synaptic ADGRLs (Li et al., 2020). TENs are alternatively spliced at two sites in their ECR; SSA is found in EGF8 and SSB in the β-propeller domain (Berns et al., 2018). The presence of the SSB site (SSB+) is thought to mediate homophilic interaction of TENs and this has been shown to disrupt the TEN/ADGRL interaction in an assay that used full-length proteins (Li et al., 2018, 2020). Structural work clarified that the SSB is present at a loop in the β-propeller domain and mediates contacts between two copies of the TEN^SF, suggesting it may mediate dimerization (Berns et al., 2018; Jackson et al., 2018). Mechanistic work agrees with the hypothesis that TEN dimerization via SSB disrupts the TEN/ADGRL interaction (Fig. 10) (Li et al., 2018, 2020). TEN -SSB has two freely rotating TEN^SF which can productively interact with post-synaptic ADGRLs (Fig. 10 A, B); TEN +SSB locks the two TEN^SF copies in a tight embrace via their β-propeller domains and limits the spatial extent of what they can explore (Fig. 10 C, D). Thus TEN +SSB is less likely to productively interact with ADGRLs. Furthermore, TEN2 SSB+ does not induce aggregation in trans with ADGRL3 (Li et al., 2018, 2020), and it shifts the synapse-forming properties of TEN2 from excitatory (SSB−) to inhibitory (SSB+) (Li et al., 2018; Sando, Jiang and Südhof, 2019). Other dimer species of TENs are reported, including the mutagenesis-validated asymmetric dimer of the D. melanogaster TEN-m (Li, Bandekar and Araç, 2023). The compact dimer species of mammalian TEN3 has been reported which is mediated by EGF8 and the Tox domains (Gogou et al., 2024) and this compact dimer only forms in TEN3 SSA+. This structure provides a hypothesis for how SSA could regulate access of TEN in the synaptic cleft, where SSA+ TEN3 would be locked in the compact dimer and sequestered from potential ligands. The compact TEN4 dimer is also available, which is mediated by the CRD and the Tox domains (Meijer et al., 2022).

Figure 10: Structural models for regulation of the TEN/ADGRL interaction by alternative splicing.

A, B: In the absence of the SSB insert, each TEN^SF can explore the synaptic cleft and find ADGRL molecules to interact with. C, D: In the presence of SSB insert, the two TEN^SF are locked together and cannot explore the synaptic space to bind to post-synaptic ADGRLs.

Downstream of receptor activation, ADGRLs stimulate the serum response element (SRE) pathway through the heterotrimeric G proteins Gɑ12/13. This phenomenon is robust and consistently observed across several independent groups at different universities (Nazarko et al., 2018; Mathiasen et al., 2020; Barros-Álvarez et al., 2022; Perry-Hauser et al., 2022; Qian et al., 2022; Bui et al., 2023; Kordon et al., 2023; Dates et al., 2024; Wang et al., 2024). Additionally, ADGRLs have been shown to regulate the cyclic adenosine monophosphate (cAMP) pathway (Nazarko et al., 2018; Röthe et al., 2019; Sando and Südhof, 2021; Qian et al., 2022; Kordon et al., 2023; Wang et al., 2024). Furthermore, the single splice variant of ADGRL3 which couples to Gɑ_s is critical to ADGRL synaptic function (Wang et al., 2024) and the cAMP signaling pathway is known to be important for synapse formation (Jiang, Sando and Südhof, 2021; Sando et al., 2022). Finally, ADGRLs have a PDZ ligand in their C-terminus which can interact with the Shank family of PDZ-domain containing proteins (Kreienkamp et al., 2000, 2002; Tobaben, Sudhof and Stahl, 2000). The structure of the C-terminal ADGRL PDZ ligand in complex with the Shank PDZ domain is available (Ponna et al., 2018) showing that ADGRL interacts with the PDZ domain in a typical manner. Also, the intracellular C-terminus of ADGRL is necessary for synapse formation (Sando and Südhof, 2021). Shanks can scaffold a large array of proteins including neurotransmitter receptors and the actin cytoskeleton, forming phase-separated macromolecular condensates which may be important for clustering neurotransmitter receptors (Monteiro and Feng, 2017; Wang et al., 2024). Thus, ADGRLs can output in two major modes: in post-synaptic signaling and through scaffolding of large protein complexes via the Shank proteins.

The interaction of the Reticulon-4 receptors (RTN4R/Nogo) with the brain angiogenesis inhibitors (ADGRBs or BAIs) form another interaction pair key for synapse development involving a post-synaptic aGPCR (J. Wang et al., 2021). RTN4Rs are presynaptic proteins with two close homologs, RTN4L1 and RTN4L2 and the three form a group of neuronal receptors with the ADGRBs as bona fide ligands (Chong et al., 2018). RTN4Rs have an N-terminal LRR region which is linked to the plasma membrane by a GPI anchor (Fig. 11A) (J. Wang et al., 2021). ADGRBs are post-synaptic proteins (Stephenson et al., 2013) expressed in the nervous system that are important for synapse formation (Kakegawa et al., 2015; Sigoillot et al., 2015) and they are associated with schizophrenia (DeRosse et al., 2008; Liao et al., 2012). There are three ADGRBs in mammals (ADGRB1–3) and they have ECRs consisting of an amino-terminal domain (NTD), 5 thrombospondin repeats (TSR1-5), a HormR, and a GAIN domain (Fig. 11A) (J. Wang et al., 2021). ADGRBs also have a GPCR 7TM as well as a C-terminal PDZ ligand (Stephenson et al., 2013).

Figure 11: Structural basis for the RTN4R/ADGRB interaction.

A: The domain layouts of RTN4R and ADGRBs. B: The structure of the RTN4R/ADGRB interaction.

The structural basis for the RTN4R/ADGRB1 interaction is known (Fig. 11B) and occurs through the interaction of the concave surface of the RTN4R LRR interacting with the TSR3 domain of ADGRB1 (J. Wang et al., 2021). Glycosylation on TSR3 plays a role in this interaction, with an O-linked glycosylation events playing a role in the interaction interface (J. Wang et al., 2021). Mutation of this interface was shown to affect several neuronal phenotypes (J. Wang et al., 2021).

ADGRBs can coordinate several events downstream of their activation, similar to ADGRLs. Signaling studies have provided evidence that ADGRBs signal to various pathways at least through G_12/13 and potentially through other means (Stephenson et al., 2013; Kishore et al., 2016) Additionally, the C-terminal ADGRB1 PDZ ligand can interact with a variety of PDZ domains, including PSD-95, in vitro (Stephenson et al., 2013) Other studies have shown that ADGRB1 can directly bind to the engulfment and cell motility protein (ELMO) via a short peptide sequence in the ICR. ELMO is an adaptor protein for the dedicator of cytokinesis protein (DOCK) (Weng et al., 2019), which is a Rho guanine nucleotide exchange factor (RhoGEF) that plays roles in coordinating cytoskeletal changes, and thus ADGRBs could regulate cytoskeletal activity through its ICR independently of G protein signaling.

Overall, the work described above strongly demonstrates how post-synaptic aGPCRs are integrators of adhesive signals and how they can disseminate these signals through multiple pathways, including G protein-signaling, PDZ-ligand interaction to scaffold protein complexes, and RhoGEFs to modulate the actin cytoskeleton (Lala and Hall, 2022). These events are all required for the proper development of the post-synaptic density (Südhof, 2021b).

Type IIa receptor protein tyrosine phosphatases and their interaction partners

The type IIa receptor protein tyrosine phosphatases (referred to as RPTPs herein), which consist of PTPδ, PTPσ, and leukocyte-common antigen related phosphatase (LAR), form another hub as presynaptic CAMs. They mediate synaptic adhesion and organization in a manner that is tightly regulated by alternative splicing (Takahashi and Craig, 2013). RPTPs are implicated in neuropsychiatric conditions such as ASD, bipolar disorder, restless leg syndrome, and schizophrenia and they have long been studied as molecules important for nervous system development (Takahashi and Craig, 2013).

RPTPs have large ECRs that begin with 3 Ig domains (Ig1-3) and continue with either 4 or 8 FnIII domains (FnIII1-8) depending on alternative splicing (Fig. 12A). A short TMR follows and finally the ICR has one functionally catalytic phosphatase domain (PTPA) and a second catalytically inert phosphatase domain (PTPB) (Takahashi and Craig, 2013). This domain architecture is studded with four small alternatively spliced inserts termed mini-exons (meA-meD). meA is in Ig2, meB between Ig2 and Ig3, meC in FnIII5, and meD in the ICR between the TMR and PTPA (Takahashi and Craig, 2013). As detailed in the previous Nrx hub section, RPTPs with meB− interact with NL3 (Yoshida et al., 2021). In the meB+ splice form, RPTPs have several ligands which are the focus of this section (Takahashi and Craig, 2013). RPTPs interact with proteoglycans including heparan sulfate proteoglycans (HSPGs) and chondroitin sulfate proteoglycans (CSPGs), with these molecules affecting the neuronal function of RPTPs (Coles et al., 2011). In a crystal structure of the LAR Ig1-Ig2 tandem, a highly conserved surface of positively charged residues on Ig1 binds to the small molecule mimic sucralose octasulfate (Coles et al., 2011). Heparan sulfate as a mimic of HSPG binding induced oligomerization of the PTPσ ECR but chondroitin sulfate did not, suggesting that these two glycans can differentially regulate RPTP clustering; the presence of HSPG may thus support the RPTP clustering needed to support molecular condensates and pre-synaptic specialization (Coles et al., 2011).

Figure 12: Structural basis for the RPTP/SALM and RPTP/Slitrk interactions.

A: The domain layouts of PTPδ and SALMs. B: The structure of the PTPδ/SALM5 interaction with orthogonal view on the right side. C: The domain layouts of PTPδ and Slitrks. D: The structure of the PTPδ/Slitrk2 interaction.

Several post-synaptic ligands of RPTPs have been confirmed by structural and functional work including synaptic adhesion-like molecules (SALM1-5) (Goto-Ito et al., 2018; Lin et al.,2018), Slit and Trk-like proteins (Slitrk 1–6) (Yamagata, Sato, et al., 2015), neurotrophin receptor tropomyosin-related kinase C (TrkC) (Coles et al., 2014), interleukin-1-receptor accessory protein-like 1 (IL1RAPL1) (Yamagata, Yoshida, et al., 2015), and interleukin-1 receptor accessory protein (IL1RAcP) (Yamagata, Yoshida, et al., 2015). This interaction network forms a strong parallel to the Nrx hub in its organization, where one pre-synaptic molecule set (RPTPs) interacts with various post-synaptic partners and the pre-synaptic molecule is regulated by alternative splicing (Takahashi and Craig, 2013).

The interaction of RPTPs with SALMs (alternate name: leucine rich fibronectin III-containing proteins - Lrfn) is key to synapse development and has been studied structurally (Nam, Mah and Kim, 2011). There are 5 SALMs in mammals (SALM1-5) and they are strongly associated with ASD and schizophrenia (Nam, Mah and Kim, 2011; Liu, 2019). SALMs are post-synaptic and are involved in the development of both excitatory and inhibitory synapses depending on the context (Nam, Mah and Kim, 2011; Liu, 2019). SALMs have LRR with N- and C-terminal cap regions, one Ig domain, and one FnIII domain, a TMR, and a short ICR with a PDZ ligand (Fig. 12A) (Liu, 2019). Crystallographic structures of SALM2 (Goto-Ito et al., 2018) and SALM5 (Goto-Ito et al., 2018; Lin et al., 2018) are available in complex with PTPδ are available. These structures revealed that PTPδ Ig2-Ig3 interacts with the SALM Ig domain and the C-cap of the SALM LRR using conserved residues (Fig. 12B). SALM dimerization occurs in an antiparallel manner and is critical for the synaptic functions of SALM as mutations in the dimerization interface reduce synaptic function (Goto-Ito et al., 2018; Lin et al., 2018). Additionally, PTPδ meB+ is present in all PTPδ-SALM structures, and although meB is close to the binding interface, it does not make any interactions with SALM (Fig. 12B). Interestingly, meB+ strongly enhances the binding affinity of the two proteins by >20-fold (Goto-Ito et al., 2018; Lin et al., 2018). One hypothesis to explain these results is that meB increases the spacing between Ig2 and Ig3 to allow Ig3 to productively bind to SALM (Goto-Ito et al., 2018).

Another key interaction for synapse development is that of the RPTPs with the Slitrks which is necessary for both excitatory and inhibitory synapse development, depending on the context (Won, Lee and Kim, 2019). Slitrks are named due to their homology to the axon guidance molecules Slit and the Trk receptors and they are highly expressed in the central nervous system. These key CAMs are linked to several neuropsychiatric disorders (Won, Lee and Kim, 2019). There are 6 Slitrks in mammals (Slitrk1-6) and they contain two LRR regions (LRR1 and LRR2), a short TMR, an ICR which is suggested to signal through phosphorylation of key tyrosine residues (Fig. 12C) (Won, Lee and Kim, 2019), and a C-terminal PDZ ligand. The interaction of PTPδ with Slitrk2 (and LAR with Slitrk1) involves Ig2-Ig3 sandwiching the Slitrk LRR1 and a loop of the LRR1 C-cap (Fig. 12D) (Um et al., 2014; Yamagata, Sato, et al., 2015). In this case, meB directly makes several interactions with the inside of the concave surface of the LRR and accordingly, meB− PTPδ does not exhibit any binding with Slitrk2 in surface plasmon resonance studies (Yamagata, Sato, et al., 2015). Furthermore, meB− LAR does not interact with Slitrk1 in cell-surface binding or cell aggregation experiments (Um et al., 2014). Interestingly, it was shown that the lateral interactions between different copies of the trans LAR/Slitrk1 heterodimer were important for heterologous synapse formation assays, suggesting that this species needs to cluster at high concentration to initiate synapse development (Um et al., 2014).

The RPTP-TrkC interaction is key for excitatory synapse formation (Takahashi et al., 2011). TrkCs have an LRR domain with N- and C-terminal cap regions, and two Ig domains in the ECR (Fig. 13A) (Coles et al., 2014). This is followed by a TMR, and a C-terminal PDZ ligand (Coles et al., 2014) and in some versions of TrkC there is a tyrosine kinase domain at the C-terminus (Takahashi et al., 2011). Trks are tyrosine kinase molecules that signal in response to binding of the small protein neurotrophin, but TrkC has adapted to bind PTPσ instead, as TrkC neurotrophin binding is not required for synapse formation whereas binding to PTPσ is (Takahashi et al., 2011). Furthermore, the variants of TrkC expressed in the synapse are versions lacking its kinase domain (Takahashi et al., 2011). The PTPσ/TrkC structure showed that Ig1/Ig2 of PTPσ interacts with the LRR and Ig1 of TrkC (Fig. 13B) (Coles et al., 2014). This sterically conflicts with the binding site for HSPG (Coles et al., 2011, 2014), and heparan sulfate can displace PTPσ from immobilized TrkC and disrupt artificial synapse formation through TrkC (Coles et al., 2014). Interestingly, this is the only RPTP ligand that binds to the Ig1/Ig2 tandem of any RPTP (Coles et al., 2014), in agreement with the hypothesis that TrkC competes with HSPG in a physiological context. All other characterized RPTP ligands bind to the Ig2/Ig3 tandem (Coles et al., 2014; Yamagata, Sato, et al., 2015; Yamagata, Yoshida, et al., 2015; Lin et al., 2018).

Figure 13: Structural basis for the RPTP/TrkC and RPTP/ILRAP interactions.

A: The domain layouts of PTPσ and TrkC. B: The structure of the PTPσ/TrkC interaction with orthogonal view on the right side. C: The domain layouts of PTPδ and ILRAPs. D: The structure of the PTPδ/ILRAP interaction.

IL-1RAcP and IL1RAPL1 (termed ILRAPs herein) are a group of post-synaptic molecules that partner with RPTPs for synapse formation (Yamagata, Yoshida, et al., 2015). ILRAPs are implicated in a range of disorders including ASD and X-linked intellectual disability (Pavlowsky et al., 2010). The two ILRAPs have three extracellular Ig domains (Ig1-Ig3), a short TMR, and an intracellular Toll/IL-1R homology domain (TIR) and C-terminal PDZ ligand (Pavlowsky et al., 2010) (Fig. 13C). The structures of both ILRAPs in complex with PTPδ are known (Yamagata, Yoshida, et al., 2015) and this work revealed that the ILRAP-Ig1 domain is cradled between PTPδ -Ig2 and PTPδ -Ig3, and furthermore ILRAP-Ig3 contacts PTPδ-Ig1 (Fig. 13D). Another structure of ILRAP in complex with meB− PTPδ was determined, and the ILRAP-Ig3/PTPδ-Ig1 were not observed, providing an explanation for meB+ regulation of the PTPδ/ILRAP ineraction (Yamagata, Yoshida, et al., 2015). Similar to that proposed for SALM, meB is thought to increase the distance between PTPδ-Ig2 and PTPδ-Ig3 such that the ILRAP-Ig3/PTPδ-Ig1 can productively occur (Yamagata, Yoshida, et al., 2015). This agrees with binding affinity measurements where meB− PTPδ has lower affinity for ILRAPs (Yamagata, Yoshida, et al., 2015). Additionally, these structures revealed that meA, which is located on a loop of PTPδ-Ig2, is directly located in the interaction surface with ILRAP-Ig1. This provides an explanation for affinity measurements that showed the meA splice insert is a very precise switch that distinguishes between the two homologs IL-1RAcP and IL1RAPL1, where meA− PTPδ retains interactions with IL-1RAcP but loses measurable interaction with IL1RAPL1 (Yamagata, Yoshida, et al., 2015).

In the RPTP hub, splicing plays a master regulatory role. meA+ is necessary for the RPTP/IL1RAPL1 and RPTP/NL interactions, but dispensable for RPTP/IL-1RAcP (Yamagata, Yoshida, et al., 2015), RPTP/SALM5 (Goto-Ito et al., 2018), and RPTP/Slitrk (Yamagata, Sato, et al., 2015). The presence of RPTP meB+ eliminates the RPTP/NL interaction (Yoshida et al., 2021) and decreases the affinity of the RPTP/TrkC interaction(Coles et al., 2014), but enhances the affinity for RPTP/ILRAPs (Yamagata, Yoshida, et al., 2015), the RPTP/SALM interaction (Lin et al., 2018), and the RPTP/Slitrk interaction (Yamagata, Sato, et al., 2015).

Downstream of RPTPs on the pre-synaptic side, the scaffolding molecule Liprinα binds to the PTPB domain of all RPTPs (Marcó de la Cruz et al., 2024), and other proteins such as the RhoGEF Trio (Debant et al., 1996) also interact with PTPB; these can organize and scaffold around the actin cytoskeleton. Liprinα proteins are master regulators of the presynaptic space; their deletion leads to empty boutons lacking presynaptic vesicles (Marcó de la Cruz et al., 2024). The structural basis for interaction of Liprinα with PTPδ (Wakita et al., 2020) and LAR (Xie et al., 2020) are known and involve the sterile alpha motif (SAM) domains of Liprinα interacting with the PTPB domain of each RPTP. On the post-synaptic side, the Slitrks have PDZ ligands which interact with PSD-95 and Shank proteins (Han et al., 2019), SALMs have a PDZ ligand and can interact with PSD-95 (Morimura et al., 2006; Wang et al., 2006), Slitrks interact with PSD-95 and Shanks (Han et al., 2019), and ILRAPL1 can interact with PSD-95 (Pavlowsky et al., 2010). Thus, the RPTPs and their post-synaptic ligands can robustly scaffold the components for productive synapse formation in a manner tightly regulated by alternative splicing.

Conclusions and Future Directions

This chapter summarizes several major points about the structural basis of synaptic adhesion, with a focus on the mechanisms of how alternative splicing modulates the function of synaptic CAMs. A variety of molecules are important for synaptic function, and they can be organized into several interaction hubs. The three most well-studied are presented here, the neurexin hub, the latrophilin hub, and the RPTP hub (Südhof, 2018b). Alternative splicing, primarily of pre-synaptic molecules, affects the structure of CAMs and this translates into functional effects (Südhof, 2018b). Several possibilities are apparent, such as inserting or removing specific residue side chains into interfaces (Elegheert et al., 2016), affecting the spacing between domains (Lin et al., 2018), and affecting the conformational flexibility of CAMs in the synaptic cleft (Li et al., 2020) and other possibilities are likely. In response to productive CAM adhesion, scaffolding of dense phase separated condensates occurs through the interaction of intracellular sequences with proteins such as the Liprinɑs and MAGUKs and these are likely key for organizing the pre-synaptic active zone and the post-synaptic density (Südhof, 2021b). How signaling events such as G protein and other pathways contribute to synapse formation are a hot topic (Jiang, Sando and Südhof, 2021; Wang et al., 2024). The future of this field likely includes detailed structural and functional studies of more synaptic CAMs. Structural studies thus far have focused on the isolated ECRs of synaptic CAMs and new exciting work including both the TM and ECRs of CAMs open new doors to understanding CAM function (Kordon et al., 2024). New trans-synaptic interactions necessary for synapse formation are continuing to be structurally characterized (Patil et al., 2023). These studies will be expanded on by the cutting-edge technique of cryogenic electron tomography (cryo-ET), which has the potential to resolve individual proteins present in the context of real, vitrified synapses (Tao et al., 2018; Sun et al., 2023; Held, Liang and Brunger, 2024). In conjunction with pioneering work on the molecular basis of CAM function (Marcó de la Cruz et al., 2024; Wang et al., 2024), future studies can characterize proteins at a level never seen before, and mechanistic work can use structural hypotheses to interrogate the details of how CAMs work and how mutations found in disease states can affect CAM functionality.

Ultimately, our work as a scientific community is oriented towards modulating these CAM interactions in disease states to serve human health needs, and the decades of structural and mechanistic work on CAMs has delivered many potential avenues for correcting altered synaptic function in disease states. The trans-synaptic adhesion of CAMs could be specifically blocked, or the lateral build-up of condensates could be blocked through interrupting intracellular scaffolding events, such as through MAGUKs or Liprinɑs. The PDZ domain/ligand interaction is common in the function of many CAMs, and this interaction is prime for targeting with small molecules (Dev, 2004; Christensen et al., 2019). Such discovery campaigns are underway for the treatment of neurological disorders (Dev, 2004; Vogrig et al., 2013; Christensen et al., 2019). CAMs with direct signaling activity such as aGPCRs can be targeted by modulating compounds to affect their signaling (Kordon et al., 2023). Downstream signaling events can be modulated using traditional kinase, phosphatase, and other inhibitors (Jiang, Sando and Südhof, 2021).

However, the most exciting possibilities illuminated by the studies outlined in this work include splice-isoform specific modulators. For example, molecules that bind to the sequence of Nrx SS4 could disrupt Nrx/Clbn/GluD2 complex formation without affecting Nrx/NL binding, and molecules which bind to the TEN SSB sequence could disrupt TEN SSB dimer formation and promote ADGRL/TEN interaction in TEN SSB+. Antibody-like molecules, including nanobodies (Nbs) and synthetic antigen binders (sABs) are prime candidates for splice-isoform selective binders as they can target epitopes with exquisite specificity (Paduch et al., 2013). Antibodies can be made which target specific pre-synaptic and post-synaptic molecules, to target alternatively spliced sequences directly, or to stabilize or block to the changed conformations that are directed by alternative splicing. Bi-specific antibodies which bind to the synaptic adhesion complex and promote adhesive events are also possible. Exciting studies have shown that sAB molecules can specifically disrupt the interaction of ADGRL with TEN, but not FLRT, and vice versa (Kordon et al., 2023), potentially providing a way to block excitatory synapse development. sABs are available that can modulate the signaling of ADGRL3 without affecting other ADGRLs (Kordon et al., 2023). The future of mechanistic molecular neurobiology is filled with challenges (Südhof, 2018b), and structural biology, when combined with mechanistic studies using in vitro and in vivo systems, has the unique ability to provide us atomic-level detail of the synapse and the broad, amazing array of proteins involved in its function. Protein structures have been a useful tool for decades and will continue to be extremely valuable in unraveling the mysteries of the human nervous system.

Methods

Structural analysis and figure preparation was performed using UCSF ChimeraX (Goddard et al., 2018; Pettersen et al., 2021; Meng et al., 2023). Sequence conservation analysis was performed using PROMALS3D (Pei, Kim and Grishin, 2008).