Cerebellin–neurexin complexes instructing synapse properties

Cerebellins (Cbln1-4) are secreted adaptor proteins that connect presynaptic neurexins (Nrxn1-3) to postsynaptic ligands (GluD1/2 for Cbln1-3 vs. DCC and Neogenin-1 for Cbln4). Classical studies demonstrated that neurexin-Cbln1-GluD2 complexes organize cerebellar parallel-fiber synapses, but the role of cerebellins outside of the cerebellum has only recently been clarified. In synapses of the hippocampal subiculum and prefrontal cortex, Nrxn1-Cbln2-GluD1 complexes strikingly upregulate postsynaptic NMDA-receptors, whereas Nrxn3-Cbln2-GluD1 complexes conversely downregulate postsynaptic AMPA-receptors. At perforant-path synapses in the dentate gyrus, in contrast, neurexin/Cbln4/Neogenin-1 complexes are essential for LTP without affecting basal synaptic transmission or NMDA- or AMPA-receptors. None of these signaling pathways are required for synapse formation. Thus, outside of the cerebellum neurexin/cerebellin complexes regulate synapse properties by activating specific downstream receptors.


Introduction
Cerebellins are secreted proteins that are encoded by four genes (Cbln1-4). The domain structure of Cerebellins resembles that of complement factor C1q (C1qA-C), C1ql proteins (C1ql1-4, for 'C1q like'), and adiponectin ( Figure 1a) [1]. These proteins share a similar Cterminal C1q-like domain but include different N-terminal multimerization motifs. Here, Cerebellins contain a short cysteine-rich sequence (w30 residues), adiponectin and complement C1q proteins a collagen-like domain (w50 residues), and C1ql proteins both cysteine-rich and collagen-like domains (Figure 1a). The various C1q-domain proteins assemble into homo-and heteromultimers, with cerebellins assembling into trimers via their C-terminal C1q-domains and into dimers of trimers via their N-terminal cysteine-rich domain, thus creating a two-headed hexamer (Figure 1b) [2e6]. Cbln1, Cbln2, and Cbln4 proteins readily form homohexamers but Cbln3 is unable to assemble into multimers on its own. Instead, Cbln3 is secreted as a heteromultimer with Cbln1 or Cbln2 [7]. It is unclear if Cbln1, Cbln2, and Cbln4 heteromultimerize with each other.
Presynaptic neurexin-Cbln1 and -Cbln2 complexes bind to postsynaptic GluD1 and GluD2 (gene symbols Grid1 and Grid2) via the cerebellin C1q-domains (Figure 1b) [13]. GluDs exhibit the domain structure of ionotropic glutamate receptors and are evolutionarily related to AMPA-(AMPARs) and NMDA-receptors (NMDARs) [4,14]. Despite their ionotropic receptor structure, GluDs were classically thought not to function as ligandgated ion channels [14], but recent data suggest that they might, after all, also gate ion fluxes [15e17]. Independent of whether or not GluDs are in fact ionotropic receptors, GluDs clearly transduce presynaptic Cbln1/2-neurexin signals into a postsynaptic response at subsets of synapses [14]. It is likely that different Cbln1, Cbln2, and Cbln3 homo-and heteromultimers have distinct affinities for GluDs, and that GluD1 and GluD2 exhibit distinct affinities for various cerebellins, but these affinities have not been comprehensively examined. Thus, Cbln1 and Cbln2, alone or in a complex with Cbln3, are secreted adaptors that connect presynaptic neurexins to postsynaptic GluDs (Figure 1b). Presynaptic neurexins form trans-synaptic complexes with postsynaptic GluD1/2 via binding to Cbln1/2/3 or with postsynaptic deleted-incolon-cancer (DCC) and Neogenin-1 via binding to Cbln4, with Cbln1-4 acting as secreted adaptor molecules in these complexes. a, Schematic of cerebellin domain structures in comparison to those of the related C1ql's, adiponectin, and complement factors C1qA, B, and C (SP = signal peptide; C = cysteine-rich domain; GxxG = collagen-like domain). Note that the cysteine-rich and collagen-like domains are dimerizing and trimerizing sequences, In contrast to Cbln1-3, Cbln4 binds only poorly -if at allto GluD1 and GluD2 [3,6]. Instead, Cbln4 tightly binds to the axon-guidance Netrin receptor DCC (for 'deleted in colon cancer') and its close homolog Neogenin-1 [18,19]. Cbln4 competes with netrins for DCC and Neogenin-1 binding, suggesting that it binds to the same site as Netrins on the membrane-proximal FNIII domains. The binding of Cbln4 to DCC and Neogenin-1 was a huge surprise because DCC and Neogenin-1 have no sequence similarity to GluD1 or GluD2 and Cbln4 has no similarity to Netrins, and because Cbln4deficient mice exhibit little evidence of axonal pathfinding deficits [19,20].
Besides binding to Cerebellins, neurexin SS4þ splice variants bind to neuroligins, albeit with a lower affinity than Cerebellins [9,10]. In addition, neurexin SS4splice variants bind also to neuroligins as well as to other ligands, most prominently LRRTM1 and LRRTM2.
Cerebellins (except for the cerebellum-specific Cbln3) are expressed throughout the brain, albeit in distinct selective patterns [21e23] (Figure 1c). While Cbln1 is abundantly produced in cerebellar granule cells but only sparsely in other brain regions, Cbln2 is robustly expressed, among others, in the cortex, subiculum, and lateral and medial habenula ( Figure 1c). Moreover, cerebellar granule cells express Cbln2 instead of Cbln1 during early postnatal development [21]. Cbln4, conversely, is highly expressed in the medial habenula in neurons, the entorhinal cortex, and cortical and hippocampal somatostatin-positive interneurons [21,24e26]. The overall picture that emerges from these studies is that a limited subset of neurons in brain abundantly produce cerebellins. Moreover, expression of cerebellins may be activity dependent. Retinal light exposure upregulates expression of Cbln1 and Cbln4 in the visual cortex [27], but puzzlingly neuronal activity downregulates Cbln1 expression in cerebellar granule cells [28], indicating that further clarifications are needed.

Function of neurexin-Cbln1-GluD2 complexes in the cerebellum
Initially unconnected work revealed that GluD2 and Cbln1 knockout (KO) mice exhibit a strikingly similar cerebellar phenotype (reviewed in Ref. [14]). Brilliant experiments from the Yuzaki laboratory subsequently showed that Cbln1 binds to GluD2 as a postsynaptic receptor [13], while parallel pioneering studies from the Mishina laboratory identified neurexins as presynaptic Cbln1 receptors [8].
GluD1 and Cbln1 KOs cause a complete loss of parallelfiber synapse LTD, a partial decrease (w40e50%) of parallel-fiber synapse numbers, and an emergence of 'naked' spines but no changes in the Purkinje cell dendritic arbor [29,30]. The Cbln1 and GluD2 KO phenotypes are commonly interpreted as reflecting an impairment in synapse formation, with 'naked spines' suggesting that spines can form without synaptic inputs [1]. An alternative hypothesis, however, is that the loss of trans-synaptic neurexin-Cbln1-GluD2 signaling destabilizes parallel-fiber synapses, leading to their elimination [10]. The second hypothesis would explain why only a subset of parallel-fiber synapses are lost in the Cbln1 and GluD2 KO mice even though LTD is completely abolished and would interpret 'naked spines' as remnants of eliminated synapses. Moreover, the second hypothesis is supported by the finding that in GluD1 KO mice, parallel-fiber synapse formation apparently proceeds normally initially [31] and contain more postsynaptic AMPARs consistent with a signaling impairment [32].
How can we differentiate between the hypotheses that Cbln1 and GluD2 mediate formation of parallel-fiber synapses or that they organize synapse properties required for synapse maintenance? An ingenuous experiment provides a clue: Yuzaki and colleagues demonstrated that recombinant Cbln1 introduced into the cerebellar cortex either in an acute slice or in vivo restores synapse numbers in Cbln1 KO mice within hours [33], suggesting that exogenous Cbln1 protein can reactivate a pre-existing neurexin and GluD2 receptor machinery to re-establish synapses that were lost in the KO. This intriguing finding supports the notion that synapse loss in Cbln1 KO cerebellum is due to a disengagement of a signaling complex that causes synapse destabilization but that can be quickly reversed by addition of recombinant protein.
Function of neurexin-Cbln1/2-GluD1/2 complexes outside of the cerebellum Neurexins are extensively alternatively spliced at six canonical sites, of which splice site 4 (SS4) has been studied in greatest detail because it regulates the interactions of neurexins with at least three families of trans-synaptic ligands, neuroligins, cerebellins, and respectively, while the C1q-domain forms trimers. Thus, cerebellins are dimers of trimers (hexamers), adiponectin and complement factors C1qA-C trimers of timers (nonamers), and C1q-like proteins form dimers of timers of trimers (octadecamers). b, Diagram of cerebellin-based trans-synaptic neurexin complexes involved in synapse organization in comparison to other neurexin complexes. Cbln1-3 complexes bind to GluD1/2 whereas Cbln4 complexes bind to DCC and neogenin-1. c, Expression of Cbln1, Cbln2, and Cbln4 in the hippocampal formation and surrounding areas visualized by single-molecule in situ hybridization illustrates restricted and selective synthesis of these neurexin-adaptor molecules in a few types of neurons. Cbln1 (yellow) is abundantly expressed primarily in the cerebellum, Cbln2 (light blue) in the habenula and in subsets of subiculum and cortical neurons, and Cbln4 in the entorhinal cortex, medial habenula, and hippocampal interneurons (data courtesy of Dr. Kif Liakath-Ali). Distinct functions of Nrxn1-Cbln1/2 and Nrxn3-Cbln1/2 complexes with GluD1/2 in regulating synaptic NMDA-receptors (NMDARs) and AMPAreceptors (AMPARs), respectively (adapted from Ref. [37]). Left, schematic of the mouse brain; right, illustrations of trans-synaptic neurexin-Cbln1/2 complexes regulating NMDARs and AMPARs in the mPFC (1, top), subiculum (2, middle), and cerebellum (3, bottom). Nrxn1 with an insert in alternatively spliced sequence 4 (SS4) upregulates postsynaptic NMDARs by binding to Cbln1 or Cbln2, whereas Nrxn1 lacking an insert in SS4 has no effect. Nrxn3 with an insert in alternatively spliced sequence 4 (SS4), conversely, downregulates postsynaptic AMPARs also by binding to Cbln1 or Cbln2, whereas Nrxn3 lacking an insert in SS4 again has no effect [35][36][37]. All effects are mediated by postsynaptic GluD1 or GluD2 (gene symbols Grid1 and Grid2, respectively) that regulate NMDARs and AMPARs via distinct cytoplasmic sequences. The Nrxn1 signaling pathway is observed in the mPFC and in CA1/subiculum synapses of the hippocampus but not in parallel-fiber synapses of the cerebellum that lack NMDARs, whereas the Nrxn3 signaling pathway is found in CA1/subiculum synapses and in parallel-fiber synapses of the cerebellum but not in the mPFC.
The regulation of NMDARs and AMPARs by Nrxn1 SS4þ and Nrxn3 SS4þ , respectively, was mediated by Cbln2 and GluD1 [36], suggesting that presynaptic Nrxn1 and Nrxn3 exert differential postsynaptic effects by binding to the same Cbln2-GluD1 receptor complexes [36,37] ( Figure 2). This differential regulation depended on the short SS4 insert sequences (30 residues) that differ between Nrxn1 and Nrxn3, such that swapping SS4 inserts between Nrxn1 and Nrxn3 was sufficient to switch their functions [36].
How do postsynaptic GluD1/2 differentially transduce presynaptic Nrxn1 SS4þ and Nrxn3 SS4þ signals? Strikingly, distinct cytoplasmic sequences mediated the Nrxn1-and Nrxn3-dependent regulation of NMDARs and AMPARs, suggesting that GluD1/2 act as differential signal transducers (Figure 3a) [36]. Replacing the entire ionotropic transmembrane architecture of GluD1 or GluD2 with a single transmembrane region from CD4 did not block their function, suggesting that the activation of GluD1/2 signaling by binding of Nrxn1 SS4þ -Cbln2 and Nrxn3 SS4þ -Cbln2 complexes does not operate by a transmembrane conformational change analogous to that of ionotropic glutamate receptors [36] (Figure 3a). Moreover, short peptide sequence motifs that are required in full-length GluD1 for regulation of NMDARs and AMPARs (Figure 3b), when inserted into an otherwise unrelated cytoplasmic sequence in the context of the minimal GluD1-CD4 construct, were sufficient to mediate regulation of NMDARs and AMPARs by Nrxn1 SS4þ -Cbln2 and Nrxn3 SS4þ -Cbln2 complexes, respectively (Figure 3cef) [36]. Thus, the functional architecture of GluD1 is surprisingly simple in that an extracellular Cbln2-binding domain connected to an intracellular peptide motif is sufficient to mediate the regulation of NMDARs and AMPARs by Nrxn1 SS4þ -Cbln2 and Nrxn3 SS4þ -Cbln2 complexes (Figure 3).
To generalize these findings beyond CA1/subiculum synapses, two other brain regions expressing Cbln1 or Cbln2 were recently tested [37]. In the medial prefrontal cortex (mPFC), SS4-alternative splicing of Nrxn3 had no effect on AMPARs, suggesting that this pathway is inactive. However, SS4-alternative splicing of Nrxn1 profoundly altered NMDAR responses (Figure 2). Consistent with this finding, the Cbln2 deletion suppressed NMDARs but had no effect on AMPARs or on synapse or spine numbers [37]. These results differ from those reported in a study using a human BAC transgenic mouse that increases Cbln2 expression, which concluded that modest upregulation of Cbln2 expression dramatically increases spine numbers in the mPFC, although synapses were not examined [38]. Future experiments will have to address this discrepancy, but the lack in that study [38] of functional synapse studies, of a genetic Cbln2 deletion, and of a more robust overexpression of Cbln2 raises questions. Given that the conclusion that Cbln2 expression levels regulate spine numbers was based on the introduction of a human enhancer into the mouse Cbln2 gene and not on direct manipulations of Cbln2 expression, one wonders whether other actions of the human enhancer independent of Cbln2 might not be responsible for the observed increase in spine numbers.
In contrast to the mPFC, expression of Nrxn1 SS4þ or conditional deletion of Cbln1 had no effect on NMDARs in cerebellar parallel-fiber synapses, as expected given that parallel-fiber synapses lack NMDARs. Constitutive expression of endogenous Nrxn3 SS4þ , however, suppressed parallel-fiber synapse AMPARs, whereas deletion of Cbln1 enhanced AMPARs ( Figure 2). Thus, the Nrxn3-Cbln1-GluD2 signaling pathway regulating AMPARs likely also operates in cerebellar parallel-fiber synapses. The results in the cerebellum do not necessarily mean that neurexin-Cbln2 signaling has no role in parallel-fiber synapse formation, as both effects could operate in parallel.

Cerebellin-mediated block of self-inhibition in adjacent synapses
Reciprocal dendro-dendritic synapses are formed between dendrites of excitatory and inhibitory neurons in several brain regions, in particular between excitatory mitral and inhibitory granule cells in the olfactory bulb [39e46]. At dendro-dendric synapses, presynaptic and postsynaptic specializations are in close vicinity, creating a microcircuit consisting of only two adjacent reciprocal synapses in neighboring dendrites (Figure 4) [39e41]. However, this synapse design has an inherent liability: How are cis-interactions between trans-synaptic adhesion molecules in adjacent pre-and postsynaptic specializations prevented? Such cis-interactions would be detrimental since they would inactivate signaling by trans-synaptic adhesion molecules. A similar liability exists for axo-axonic synapses in which presynaptic terminals receive a synaptic input while also forming a synaptic output.
Recent experiments on neurexins and cerebellins in the olfactory bulb revealed an ingenious self-avoidance mechanism that counteracts this liability [47]. Specifically, local expression of Cbln2 that binds to neurexins in mitral cells in the olfactory bulb blocks cis-interactions between neurexins and neuroligin-1 that would otherwise inactivate synaptic inputs from granule cells, thereby freeing neuroligin-1 for trans-interactions with granule cell neurexins. This block is possible because Cbln2 has a higher affinity for SS4þ neurexins than neuroligin-1 [47]; trans-synaptic neuroligin-1 interactions are presumably not blocked because the granule cell neurexins are partially expressed as SS4splice variants and/or because granule cell neurexins mediate their functions, at least in part, by binding to mitral cell dystroglycan [48,49] (Figure 1b). In this manner, Cbln2 ensures functionality of dendrodendritic synapses.

Function of Cbln4-neurexin complexes
Relatively little is known about Cbln4 functions. When Cbln4 was identified as a DCC and Neogenin-1 ligand [18,19], extensive searches for an axonal guidance phenotype in constitutive Cbln4 KO mice failed to detect major changes. Only a modest and transient increase in the number of wandering axons in the brachial plexus was observed, consistent with the predominantly postnatal expression of Cbln4 [18,19]. Moreover, extensive analyses of single and triple KO mice constitutively lacking Cbln4 either alone or in combination with Cbln1 and Cbln2 also failed to detect major changes in brain architecture, suggesting that the major function of Cbln4, like that of Cbln1 and Clbn2, is not in axon guidance [20]. Furthermore, these studies did not observe a large amount of synapse loss in any brain region, suggesting that Cbln4 is not a major contributor to the formation or maintenance of synapses. Cbln4 KO mice did, however, exhibit major behavioral changes in a number of tasks, suggesting an important function in brain [20,24]. These results indicate a postdevelopmental role of Cbln4 similar to, but different from, Cbln1 and Cbln2.
Four subsequent studies examined Cbln4 function in specific brain regions. Studies on the medial habenula revealed that different types of medial habenula neurons that project to distinct areas of the interpeduncular nucleus express high levels of Cbln2 or Cbln4 [24]. Strikingly, conditional deletion of Cbln4 from the medial habenula elicited a strong behavioral anxiety phenotype but produced no impairments in synapse numbers or basal synaptic transmission as monitored in the target interpeduncular nucleus [24]. A potential explanation for this phenotype was suggested by a later finding in the entorhinal cortex, another brain region that expresses high levels of Cbln4 (Figure 1c). Conditional deletion of Cbln4 from the entorhinal cortex that innervates the dentate gyrus via the perforant path [50e52] caused no changes in synaptic connectivity or basal transmission strength in perforant path synapses [25], which is similar to the lack of an effect of the deletion of Cbln4 on medial habenula/interpeduncular nucleus synaptic connectivity [24]. The conditional deletion of Cbln4 from the entorhinal cortex, however, caused a remarkable and complete loss of LTP [25]. Dentate gyrus neurons robustly express Neogenin-1 but not DCC as a potential ligand for presynaptic neurexin-Cbln4 complexes [25]. Conditional deletion of Neogenin-1 in the dentate gyrus replicated the phenotype of the Cbln4 deletion in the entorhinal cortex. Again, normal basal synaptic transmission was observed with a complete absence of LTP in perforant-path synapses [25]. These studies were the first to reveal a physiological function of Cbln4-binding to its Neogenin-1/DCC receptors and uncovered a selective role of Cbln4 in synaptic plasticity. They were also instrumental in identifying a molecular mechanism for dentate gyrus LTP, which was the first type of LTP discovered [53] but whose molecular characteristics remained uncharacterized.
A completely different view of Cbln4 function emerged from two recent studies of Cbln4 in the mouse cortex [26,54]. The first of these studies [26] is easily overlooked because it doesn't mention Cbln4 in the title or abstract but beautifully shows that Cbln4 is highly expressed in somatostatin-positive (SSTþ) interneurons in the cortex and that an shRNA to Cbln4 decreases, whereas overexpression of Cbln4 increases, synapse numbers formed by SST þ interneurons on cortical pyramidal neurons as analyzed by fluorescence labeling. The shRNA data were replicated by the second study that puzzlingly identified GluD1 as the Cbln4 receptor [54] even though Cbln4 is not known to bind to GluD1. The large synapse number increase induced by Cbln4 overexpression in cortex suggests that Cbln4 may have a role in cortical inhibitory synapses that differs from its function in other brain regions and that could plausibly be mediated by DCC and/or Neogenin-1.
GluDs transduce presynaptic Nrxn1-Cbln2 and Nrxn3-Cbln2 signals that control NMDARs and AMPARs, respectively, by a simple sequencedependent transduction mechanism (data replotted from Ref. [36]) a, Schematic of the constructs used to examine the GluD signal transduction mechanism. GluDs' exhibit the same domain architecture as AMPARs and NMDARs to which they are homologous (GluD1, left). A simple GluD derivative containing only the N-terminal extracellular cerebellin-binding domain (ATD) and the C-terminal cytoplasmic sequence connected via the CD4 transmembrane region (GluD1-CD4 TMR and GluD2-CD4 TMR ) is fully competent to transduce presynaptic Nrxn1-Cbln2 and Nrxn3-Cbln2 signals into postsynaptic regulation of AMPARs and NMDARs, respectively (center). Moreover, GluD1-CD4 TMR constructs with cytoplasmic sequences containing only the sequence motifs that are required for the AMPAR-and NMDAR-regulation (GluD1-CD4 TMR Motifs) is also fully enabled to transduce presynaptic Nrxn1-Cbln2 and Nrxn3-Cbln2 signals (right). b, Cytoplasmic tail sequences of the 'GluD1-CD4 TMR Motif3 and Motif4 constructs. Sequences derived from GluD1 are colored in red and green. c-f, Rescue experiments in GluD1 CRISPR-KO neurons illustrating the ability of wild-type but not mutant GluD1-CD4 TMR Motifs constructs to rescue selected GluD1 KO phenotypes, such that the Motif4a construct reverses the increase in AMPAR-EPSC amplitude produced by the GluD1 KO whereas the Motif4 construct restores the decreased NMDAR-EPSC amplitude (C, sample traces; D-F, bar graphs of the NMDA/AMPA ratio (D), the AMPAR-EPSC amplitudes (E), and the NMDAR-EPSC amplitudes (F)).
Clarification of this astounding effect and its mechanism will provide major insights into this intriguing question.

Outlook
Current data define multiple cerebellin-dependent signaling pathways that organize synapses. The use of distinct neurexin output signaling pathways that are mediated by neurexin-Cbln1/2-GluD1/2 and neurexin-Cbln4-DCC-Neogenin-1 interactions and that serve diverse synaptic roles enables neurexin complexes to control an array of synaptic functions. Of these functions, the regulation of postsynaptic NMDARs and AMPARs is likely crucially involved in tuning neural circuits because NMDAR and AMPAR responses determine the nature of synaptic signals. The cerebellin-dependent changes in synapse numbers e the loss of cerebellar parallel-fiber and cortical inhibitory synapses observed upon downregulation of Cbln1 and Cbln4 expression, respectively e A non-canonical self-avoidance function for Cbln1 and Cbln2 is required for synaptic transmission at dendro-dendritic synapses. At dendrodendritic synapses in the olfactory bulb, granule cell dendrites (left) form inhibitory synapses on mitral/tufted cell dendrites (right), while mitral/tufted cell dendrites form excitatory synapses on granule cell dendrites, placing reciprocal inhibitory (top) and excitatory synapses (bottom) into close proximity. Binding of presynaptic granule-cell neurexins to postsynaptic mitral-cell Nlgn2 and/or Nlgn3 is essential for granule cell/mitral cell synaptic transmission, but this trans-interaction is suppressed when mitral-cell neurexins in adjacent mitral cell/granule cell synapses form cis-complexes with mitral-cell Nlgn2/3. Cbln1 and Cbln2 block such cis-interactions by binding to neurexins with a higher affinity than neuroligins [47]. This self-avoidance mechanism requires predominant expression of SS4+ splice variants of neurexins in mitral cells since SS4-splice variants do not bind to Cbln1 and Cbln2. Note that in addition to these trans-synaptic interactions, granule cell/mitral cell synapses are controlled by the interaction of presynaptic granule-cell neurexin-3 with postsynaptic mitral-cell dystroglycan [49].
are more difficult to understand since it is unclear whether the synapse losses observed are due to changes in the formation, maintenance, or elimination of synapses that normally turn over continuously. Clearly cerebellins do not obligatorily mediate synapse formation or suppress synapse elimination since in at least some synapses, deletion of cerebellins and their pre-and postsynaptic receptors have dramatic effects on synapse properties but not on synapse numbers. The major problem here is that in general, synapse formation, maintenancy, and elimination are poorly defined processes. Developing and applying assays for that purpose is a high priority goal.
Cerebellins mediate just one of multiple neurexin output pathways that in many synapses likely compete with each other, as shown most clearly for reciprocal dendro-dendritic synapses [47] (Figure 4). Other competitive pathways will likely emerge since diverse neurexin ligands often bind to the same neurexin sequences, generally in a manner regulated by alternative splicing [10]. Thus, the action of neurexin-cerebellin complexes cannot be examined in isolation but has to be considered in the context of an overall larger transsynaptic interaction network (Figure 1b). Building models of this network, however, is a daunting challenge given that the relative affinities of various neurexin interactions and the precise levels of various isoforms and their splice variants at synapses are unknown.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability
This is a review article and does not report unpublished data.