Role of neurexin heparan sulfate in the molecular assembly of synapses – expanding the neurexin code?

Synapses are the minimal information processing units of the brain and come in many flavors across distinct circuits. The shape and properties of a synapse depend on its molecular organisation, which is thought to largely depend on interactions between cell adhesion molecules across the synaptic cleft. An established example is that of presynaptic neurexins and their interactions with structurally diverse postsynaptic ligands: the diversity of neurexin isoforms that arise from alternative promoters and alternative splicing specify synaptic properties by dictating ligand preference. The recent finding that a majority of neurexin isoforms exist as proteoglycans with a single heparan sulfate (HS) polysaccharide adds to this complexity. Sequence motifs within the HS polysaccharide may differ between neuronal cell types to contribute specificity to its interactions, thereby expanding the coding capacity of neurexin diversity. However, an expanding number of HS‐binding proteins have been found capable to recruit neurexins via the HS chain, challenging the concept of a code provided by neurexin splice isoforms. Here we discuss the possible roles of the neurexin HS in light of what is known from other HS‐protein interactions, and propose a model for how the neurexin HS polysaccharide may contribute to synaptic assembly. We also discuss how the neurexin HS may be regulated by co‐secreted carbonic anhydrase‐related and FAM19A proteins, and highlight some key issues that should be resolved to advance the field.


Introduction
The brain exerts its functions by the transmission and processing of signals across synaptic connections. A heterogeneity of synaptic shapes and properties ensures functional specialisation in different circuits, but all synapses share a set of defining features: a presynaptic terminal filled with synaptic vesicles that upon calcium influx release their neurotransmitter content into the synaptic cleft to activate receptors on the postsynaptic membrane. The presynaptic and postsynaptic membranes also contain cell adhesion molecules that form physical connections across the synaptic cleft. Such interactions are thought to facilitate not only the formation but also subsequent specification, plasticity and maintenance of a synapse by coordinating its molecular architecture on either side of the cleft [1-3].
Neurexins constitute a well-known example of presynaptic cell adhesion molecules, which have been extensively studied for two principal reasons: First, binding of neurexins to postsynaptic partners is a strong inducer of synapse formation in vitro. For example, nonneuronal cells with ectopic expression of neurexin ligands induce the formation of presynaptic specialisation in contacting axons of co-cultured neurons (socalled 'artificial synapses') [1,4]. Conversely, expression of neurexins on non-neuronal cells leads to the assembly of postsynaptic specialisations on contacting neurites [5]. Second, variants in neurexin genes have repeatedly been linked to human neurodevelopmental and psychiatric diseases, including schizophrenia, intellectual disability, autism and Tourette syndrome (reviewed in [6][7][8]).
Neurexins interact with diverse postsynaptic cell adhesion molecules and secreted synaptic organisers [8,9]. These ligands in turn recruit additional synaptic components, such as intracellular scaffold proteins and postsynaptic receptors, that collectively contribute to the molecular assembly of contacting presynaptic and postsynaptic compartments. Extensive alternative splicing gives rise to an array of neurexin isoforms, which show preference for distinct ligands and thereby generate a 'splice-code' thought to play an instructive role in synapse assembly [8,10]. Of note, neurexins are not required for synapse formation per se, but are instead thought to specify the molecular architecture and properties of specific synapses after initial assembly. The recent finding by Ann-Marie Craig's lab that a substantial proportion of neurexin molecules carry a single heparan sulfate (HS) polysaccharide chain not only expands neurexin heterogeneity but also adds an additional layer of complexity to the 'neurexin code' by increasing the number of possible neurexin ligands [11]. Here, we discuss possible structure-function relationships of the neurexin polysaccharide and its possible regulation.

Neurexin protein structure and binding partners
Neurexins are in mammals encoded by three genes (NRXN1-3 in humans) with broad expression throughout the central nervous system [12,13]. Neurexins are predominantly expressed in neurons and concentrate at presynaptic termini [13,14], but some isoforms are also expressed abundantly in astrocytes, perhaps highlighting a broader role. Through the use of alternative promoters, each gene gives rise to both longer a-neurexin or shorter b-neurexin transcripts. In addition, the NRXN1 gene encodes an even shorter c-neurexin isoform [15,16]. The different isoforms differ in the composition of their extracellular domains (Fig. 1A); a-neurexins contain six LNS (laminin-neurexin-sex hormone binding globulin) domains that are interspersed by three EGF (epidermal growth factor-like) domains. Together they form an L-shaped array in the synaptic cleft with their ligand-binding surfaces aligned towards the postsynaptic membrane [17][18][19]. The b-neurexins contain only the most membrane-proximal LNS domain. The extracellular domains of neurexins are further diversified by alternative splicing at six conserved sites (SS1-6) [12,20]. This results in > 10 000 possible unique transcripts, of   which at least several hundred can be detected in mouse brains by long-read sequencing [20,21]. Although it remains unclear if all transcripts exist in quantitatively meaningful amounts, their relative expression differs between cell types and brain regions and are thought to contribute to functional specialisation of synapses in distinct circuits [21][22][23].
All neurexin isoforms, including the short cneurexin isoform, share a 'stalk' region that contains a conserved cysteine-loop (Cys-loop), a transmembrane domain and a relatively short intracellular sequence ending with a PDZ-binding motif [15,39,40]. The stalk region is rich in serine and threonine residues and is known since the initial cloning of neurexins to be extensively glycosylated by mucin-type O-GalNAc glycans [13]. These glycosylations may explain why it was only recently found that a specific serine residue within this sequence serves as an attachments site for a HS chain [11]. The HS-modified serine is conserved throughout vertebrate evolution and is present in Drosophila and possibly also in C. elegans. It is now evident that the HS chain plays an important role for at least some neurexin functions, although its exact role is not yet well understood.
HS-a rare glycan modification that expands neurexin structural complexity HS are linear polysaccharides ubiquitously found on cell surfaces and in the extracellular matrix. HS are invariably synthesised onto core proteins to form HS proteoglycans (HSPGs). Compared to the large number of glycoproteins carrying N-or mucin-type Oglycosylations, the number of HS-containing proteins is limited to only~20 vertebrate HSPGs identified to date, half of which belong to only two protein families: the transmembrane syndecans (4 members in humans) and the GPI-anchored glypicans (6 members) [41,42]. Nevertheless, HS are involved a wide range of physiological processes including cell migration, axon guidance, synaptogenesis, angiogenesis and lipid metabolism, and often show a high degree of functional redundancy [43,44]. These functions are commonly mediated by the HS chain(s) through the engagement of HS-binding proteins such as growth factors, cytokines and membrane receptors.
HS biosynthesis occurs in a step-wise manner beginning in the late ER/cis-Golgi with the transfer of a xylose (Xyl)-residue to specific serine residues in the core protein by xylosyltransferases [45]. The assembly continues with the enzymatic addition of two galactoses (Gal) and one glucuronic acid (GlcA), which completes the formation of the tetrasaccharide linkage region (Fig. 2) [46]. A fifth enzyme adds a Nacetylated glucosamine (GlcNAc) to this sequence, followed by the extensive addition of alternating units of GlcA and GlcNAc by the HS polymerases (EXT1/2) ( Fig. 2; for a more detailed description of HS biosynthesis, please see [47,48]).
The growing precursor polysaccharide is extensively modified by N-deacetylase/N-sulfotransferases (NDSTs), which substitute the acetyl group in GlcNAc residues with an N-linked sulfate group (to generate GlcNS). This process is generally incomplete, such that the NDST enzymes only modify certain regions along the chain, which are referred to as NS domains [49]. The HS polymer undergoes further modification, mainly in such NS domains ( Fig. 2): epimerisation at the C5 position convert GlcA residues to iduronic acid (IdoA), and most IdoA residues become sulfated at their C2 position. Moreover, GlcNS residues may become sulfated at C6 and C3 positions, respectively, by different O-sulfotransferases [50]. The polymer thus obtains a domain-type arrangement with sulfated regions (NS domain) interspaced with regions of nonsulfated (NAc) and mixed (NAc/NS) domains (Fig. 2) [51]. The modification patterns of mature HS chains can be further trimmed by extracellular 6-O-sulfatases [52]. Glycan-binding proteins preferentially bind NS domains with different modifications (exemplified below). A single HS polysaccharide contains several NS-domains along its linear chain. As each NSdomain may display a certain degree of structural In contrast to the synthesis of nucleic acids and proteins, the enzymes that produce HS polymers do so without a deterministic template. Yet, studies of HS biosynthesis point to a regulated process, resulting in polymers with cell-and tissue-specific sulfation patterns [47,53]. The composition of HS modifications in mice, for example, shows distinct tissue-specific patterns [54]. However, this and similar studies rely on the analysis of chemically or enzymatically released di-, tetra-and hexasaccharides, and their results may represent an oversimplification of cell-type specific structural heterogeneity of HS polymers. Due to technical limitations, complete structural information of full-length HS polymers is still lacking. An alternative strategy for HS structural characterisation is to utilise antibodies to detect specific HS epitopes defined by their fine structure. In a C. elegans study, such HS epitopes were found to be spatially restricted to specific cells and subcellular compartments in a pattern that was conserved between nematode species [55].
HS-modifying enzymes are differentially expressed across neurons in mouse brains [56], and it thus seems likely that also the pattern of HS modifications will be cell-type dependent also in the mammalian brain. The functional importance of HS structures for normal brain development and function have been exemplified both by studies of knockout models and patients with rare genetic diseases. For example, mutations in the  HS polymerase EXT1 have been associated with intellectual disability and autism [57]. Autism-like behaviours were also observed in mice knockout for Ext1 in excitatory forebrain neurons [58]. Moreover, numerous knockout studies in mouse, C. elegans and Drosophila have demonstrated the importance of specific HS-modifications such as 2-O-and 6-O-sulfation in cell migration, axon pathfinding and synapse formation (reviewed in [44]). Variants in the HS Ndeacetylase/N-sulfotransferase gene NDST1 have also been found to cause recessive intellectual disability [59], and a genome-wide association study has linked NDST3 to schizophrenia [60]. These examples point to essential roles of HS structures in physiological settings, but their in vivo structure-function effects remain mostly unknown. The large number of possible chain modifications allows the HS polysaccharide to accommodate an extremely high number of sequence possibilities [47]theoretically more than a million for a single octasaccharide chain. The number of actual combinations is however significantly restricted by the expression, regulation and specificity of the involved biosynthetic enzymes [61], but the resulting complexity may nevertheless hold significant information capacity. This idea has led to the hypothesis of an 'HS code' whereby specific HS sequences specify neurodevelopmental processes, including synaptogenesis [62]. The concept of a code, although intellectually appealing, may however be somewhat misleading. First, it is difficult to envision how a strict code may emerge from the nontemplate synthesis. Moreover, many HS-binding proteins show a sliding scale of affinities to different polysaccharide sequences, thereby questioning the degree of specificity ('coding') that can be achieved by the system. Nevertheless, certain protein-glycan interactions demonstrate remarkable specificity to a specific HS sequence.

Specificity and selectivity of glycanprotein interactions
Interactions between HS and HS-binding proteins rely mainly on ionic interactions between basic amino acid side chains and negatively charged sulfate and carboxyl groups of the polysaccharide [41,63]. More than a hundred different HS-binding proteins have been identified [41,64]. Early attempts to characterize HSbinding sites in proteins identified so-called Cardin-Weintraub sequences [65], composed of basic amino acids in linear peptides (-X-B-B-X-B-X-) and (-X-B-B-B-X-X-B-X), in which B denotes basic and X hydropathic residues, respectively [66]. However, most HS-binding proteins lack such sequences and no single consensus protein sequence exists. The ability of diverse proteins to bind HS instead appears to have emerged through convergent evolution, whereby basic residues arranged on the folded protein surface create charged patches that favour HS binding [41,66]. The HS-binding neurexin ligands neuroligin-1, LRRTM2 and LRRTM4 illustrate this concept (Fig. 3, further discussed below).
Most HS-binding proteins also bind heparin, a widely used anticoagulant, which is often substituted for HS due to its wide availability. However, heparin differ from HS by being both more extensively and more uniformly modified. The heparin molecule has a very high frequency (> 80%) of IdoA and GlcNS residues with one or more O-sulfate groups per disaccharide [67], resulting in a highly negatively charged chain with little domain organisation. As a consequence, protein binding to heparin is largely determined by the overall ionic interactions between the protein and the polysaccharide, rather than certain structural features along the chain. In contrast, HS has a much lower content of GlcNS residues (~40-50%). It is also less O-sulfated, containing an average number of 0.20-0.75 O-sulfate groups per disaccharide units, depending on the cellular source [54,67]. The lower degree of sulfation together with its domain-type arrangement (Fig. 2) provides HS polysaccharides with considerable structural variability. Due to this so-called fine structure, HS-protein interactions display a higher degree of specificity compared to heparin-protein interactions: some protein-HS interactions require a strict sulfation sequence, while others are less stringent and rather depend on the general degree of sulfation [68]. For example, relatively non-specific charge interactions appear to dominate the interactions between HS and members of the fibroblast growth factor (FGF) family [69,70]. In a library of HS oligosaccharides, different FGFs were found to bind identical HS structures with low specificity but similar affinities, which correlated with the overall degree of saccharide sulfation [70]. Hepatocyte growth factor can also bind HS with low specificity, without the need for a particular HS sequence, and may even bind other GAGs provided they have sufficient charge density [71]. Nevertheless, many HS-binding epitopes are not indifferent to the HS fine structure but bind with higher affinities to chains with sulfate groups at specific positionsachieving what has been conceptualised as 'intermediate specificity' [66]. In contrast, the extensively studied binding of antithrombin to heparin, which confers anticoagulant activity, shows remarkable specificity to a pentasaccharide sequence with a specific sulfation  [73]. The 3-O-sulfate is a key determinant for the antithrombin-pentasaccharide interaction and alone accounts for 60% of the binding energy [74]. Several other HS-binding proteins bind with high specificity to 3-O-sulfated HS and/or other rare structural variants including N-unsubstituted glucosamine units [75,76]. For example, neuropilin-1, a protein implicated in angiogenesis and axon guidance, exerts its effect through an interaction with 3-Osulfated HS [75]. The extent of 3-O-sulfation on natural HS is largely unknown, but available studies suggest that only 0-10% of disaccharide units, depending on the HS source, carry this modification [77]. Despite being relatively rare, mice and humans have seven different 3-Osulfotransferases (HS3STs) with partially different substrate specificities [78] and which are regulated in a spatiotemporal manner during mouse brain development [56,79]. Genetic and biochemical results point towards an importance of this modification for synapse development and function. Nematodes knockout for 3-O-sulfotransferases show impaired synapse formation [80], and a peptide blocking 3-O-sulfated HS interferes with synapse formation and function in cultured neurons from mouse hippocampi [81]. A single nucleotide polymorphism in HS3ST5 has also been linked to risk for autism in genome-wide association studies [82]. If and to what extent the neurexin HS contains 3-O-sulfate modifications, or other rare modifications, along its chain remain unknown.

Neurexin glycan-protein interactions
Initial experiments have highlighted the importance of the HS polysaccharide for normal neurexin functions. Zhang et al. generated knock-in mice with a mutation in the HS-modified serine of Nrxn1 [11]. These mice showed reduced postnatal survival, analogous to other constitutive Nrxn knockouts [83]. A detailed assessment of mossy fiber-CA3 synapses in the hippocampus revealed structural and functional phenotypes consistent with both presynaptic and postsynaptic defects at this synapse. A role for the neurexin HS in vivo was additionally demonstrated in Drosophila [11]. Fly knockouts for the single neurexin orthologue, which are viable but show reduced larval mobility and number of neuromuscular boutons, could be rescued by neuron-specific expression of Drosophila neurexin but only partially so by a mutant neurexin lacking HS [11]. Similar results were obtained using cultured mouse hippocampal neurons with knockdown of a combination of neurexin isoforms: knockdown Neurol l l l l l l l l l l l li i i i i ig i i in -1 1 1 1 1 1 1 1 1 1 1 1 1 β-Ne 157 R K158 Fig. 3. HS-binding surfaces of neurexin ligands. (A) Structure of the neuroligin-1 dimer, with one monomer represented as a surface colored according to its electrostatic potential. The charged canyon facing the opposite neuroligin-1 monomer (shown in pale green cartoon) mediates HS-binding. Some conserved basic residues in this region (labelled in insert) have been shown to confer heparin-binding and effects in assays for synapse formation and function [11]. The LNS-domain of b-neurexin, in complex with the surface-represented neuroligin-1 monomer, is shown in blue cartoon representation (based on PDB 3BIW [85]). (B) Structure of LRRTM2 (PDB 5Z8Y [84]), represented as a surface colored according to its electrostatic potential. Basic residues shown to confer heparin-binding and functional effects in synapse formation assays [11] are highlighted (insert). (C) Model of mouse Lrrtm4 as predicted by Alphafold (Q80XG9-F1 [111]) and represented as a surface colored according to its electrostatic potential. Basic residues shown to confer heparin-binding and functional effects in synapse formation assays [88] are highlighted (insert). Electrostatic surfaces (AE 5 K b T/e c , lower left) were rendered using the Adaptive Poisson-Boltzmann Solver (APBS) plugin of PyMOL (v. 2.5.2). neurons showed reduced numbers of excitatory miniature currents and were rescued by re-expression of shRNA-resistant neurexin but not with neurexin in which the HS-modified serine had been mutated [11].
Mechanistically, the neurexin HS has been shown to cooperate with the established (and crystallized) protein-protein interactions between the neurexin LNS6 domain and its canonical ligands neuroligins and LRRTM2 [84,85]. Both neuroligins (Fig. 3A) and LRRTM2 (Fig. 3B) contain conserved arginine and lysine residues with sidechains juxtaposed on the proteins' surfaces to form HS-binding motifs [11]. Mutating these residues reduced both proteins' ability to be retained on heparin columns as well as their ability to induce artificial synapses, suggesting that they mediate functionally important heparin/HS binding sites. The observation that both neuroligins and LRRTM2 independently have evolved capacity for HS-binding further strengthens the notion that these interactions are of functional importance. Nevertheless, exactly how these proteins bind to HS and whether they preferentially or exclusively engage specifically modified saccharide units remain unknown.
In addition to the cooperative protein/glycan interactions of neuroligin and LRRTM2, a number of synaptic HS-binding proteins have been found to interact with the neurexin HS chain alone. For example, postsynaptic LRRTM4-previously shown to bind the HS of synaptic glypican-4 [86,87]-was found capable to induce artificial synapses through interaction with the neurexin HS chain alone [88]. The short c-neurexin isoform is sufficient to support this function, suggesting that the extracellular HS chain with the intracellular neurexin sequence alone are sufficient for a minimal neurexin function. However, this artificial synapse formation function can be accomplished by any in vitro manipulation that results in the clustering of neurexin via its HS chain, including the exposure to beads coated with the HS-binding protein pleiotrophin or simply to beads made positively charged by coating with poly-lysine [11]. These findings may to some extent reflect the permissive nature of artificial synapse formation assays, but at the same time suggest that hundreds of different HS-binding proteins could serve as neurexin ligands. How such glycan-protein interactions relate to the apparent specificity of the neurexin protein-protein interactions and the concept of a 'splice code' is not known. A better understanding of the fine structure of the HS chain and its affinities towards specific ligands may be necessary to comprehend the function of neurexin complexes at synapses.

Possible roles of the neurexin HS
Similar to the neurexin proteins themselves [89], neurexin HS chains are likely to exert multiple functions in distinct synapses and circuits. Two conceptually different functions of neurexin HS can be envisioned, which importantly are not mutually exclusive (Fig. 4). First, as exemplified for other interactions above, the neurexin HS may contain rare modifications that confer distinct functions, possibly spaced as separate domains along the polysaccharide (Fig. 4A). Such functional domains may be regulated in a cell-type specific  [112]. HS length in astrocytic neurexin-1a appears to be significantly longer [95].

258
The To what extent the fine structure of the neurexin HS contribute to its functions, for example to stabilize neurexinneuroligin and neurexin-LRRTM2 complexes [11] or to bind specific postsynaptic ligands such as LRRTM4 [88], remains unknown. However, in vivo results demonstrate instructive roles for alternative splicing of neurexin-1 and À3 at SS4 [37,38] suggesting that glycan-protein interactions are insufficient to alone maintain synaptic properties, at least at the synapses studied. It thus appears likely that the protein-protein interactions between neurexin and its ligands form more stable complexes than that of glycan-protein interactions alone. Nevertheless, the HS chain may contribute to these interactions and differences of the polysaccharide fine structures, possibly regulated in a cell-type specific manner, serve to fine-tune the net affinities to the ligands at specific synapses [90]. The ligands may themself contribute to such fine-tuning of affinities. For example, neuroligins are also subject to alternative splicing at two conserved sites that regulate their affinity to neurexins [24,36]. One of the splice insertions in NLGN1 has been found to regulate its affinity to HS [91], thereby extending the concept of trans-synaptic splice codes to include the HS of neurexins.
Alternatively, the HS chain may also promote neurexin functions through permissive roles. Here, interactions of low-intermediate affinities and presumably lower specificities support the assembly of neurexin complexes. The synaptic cleft is a molecularly crowded yet highly organised structure [3] with neurexin complexes concentrated in nanoclusters [14]. The neurexin HS chain may extend significantly in the lateral direction as its predicted length is three times the width of the synaptic cleft or more. In this environment, the HS chain may serve to 'capture' ligands and guide their assembly into the complex by directing their diffusion in the membrane (Fig. 4B). On/off glycan binding will allow ligands such as neuroligins and LRRTM2 to slide along the polysaccharide until their proteinprotein interactions with the neurexin LNS6-domain result in a stable complex. Similar restricted diffusion along HS chains have been described for FGF2 [92] and may explain how secreted proteins can form morphogen gradients during development [93]. The observation that both neuroligins and LRRTM2 with mutated HS-binding motifs were unable to recruit endogenous neurexins in co-culture assays may supports this hypothesis [11]. Both neurexins and neuroligins are subject to continuous turnover by metalloproteases at the cell surface [14,94], and such a mechanism may aid the replacement of neurexin complexes at stable synapses. Ligand recruitment may also explain how astrocytic neurexin-1a contributes to synapse assembly at tripartite synapses (synapses with an astrocytic process invading the cleft). Interestingly, the HS chains of astrocyte neurexin appears to be longer than that of neurexin on neurons [11,95], in agreement with this hypothesis. Another permissive function of the neurexin HS chain could be the local recruitment and enrichment of soluble factors, for example FGF family known to promote the assembly of presynapses [96,97].
The neurexin HS may also serve to recruit receptors in cis. For example, the presynaptic LAR-type tyrosine phosphatase receptors PTPRS (a.k.a. RPTPr), which is known to bind HS and play a role in synapse assembly [98]. PTPRS has recently been found to physically and functionally interact with neurexins via the neurexin polysaccharide chain [88,99]. Specific HS domains may promote these functions, for example by positioning PTPRS as a co-receptor at defined length from the neurexin complex. How many ligand-binding sites the neurexin HS chain may accommodate is not known, but co-immunoprecipitation experiments suggest that the same chain can simultaneously can bind to both postsynaptic LRRTM4 in trans and presynaptic PTPRS and cis [88].

Regulation of HS-containing neurexin
Analysis of the migration of neurexins on protein gels have suggested that only a fraction (~70%) of neurexin molecules carry the HS chain [11,100], suggesting that neurexins serve as 'part-time' HSPGs with varying degree of HS occupancy [101,102]. The fraction differs between brain regions [100] and cell types [95], suggesting that HS substitution is regulated in a cell-type specific manner analogous to the diversification by means of different splice isoforms. However, as all neurexin isoforms contain the HS-modified serine, the question arises of how HS biosynthesis onto the neurexins core protein may be regulated. Cell-type specific differences in the relative levels of necessary biosynthetic enzymes, as well as their organisation within the Golgi, provides a possible explanation. However, this explanation is difficult to reconcile with the notion that some cells predominantly express non-HS neurexin but still express other HSPGs, such as syndecans and glypicans.
Another possibility is that mechanisms specific to neurexins regulate HS-chain initiation. The Cys-loop sequence located just C-terminal of the HS-modified serine, which is conserved among neurexins but absent in other HSPGs, may play a role. Two recently identified neurexin ligands, the carbonic anhydrase-related protein CA10 [15] and members of the FAM19A [34] protein family, bind to this sequence. Both CA10 and FAM19A1-4 are relatively small (35 and 12 kDa, respectively) secreted proteins that bind to robustly neurexins only when expressed in the same cell (i.e., in cis). By engaging the neurexin Cys-loop, an intermolecular disulfide bond is formed to one of its cysteine residues, generating covalent CA10-neurexin or FAM19A-neurexin complexes, respectively. While specific cysteines in each protein are required for these covalent complexes to form, other neurexin residues are additionally required: Two hydrophobic residues N-terminal of the Cys-loop are required for CA10 to bind to neurexin [15]. The same residues do not affect binding of FAM19A1, which instead appears to depend on sequences further N-terminal; in contrast to CA10, FAM19A1 does not interact with NRXN1c [34]. Binding of either CA10 or FAM19A1 to neurexin leads to expression of neurexin lacking HS [34,100], but FAM19A1 also prevents O-GalNAc glycosylation [34], further highlighting their different modes of interaction. As CA10 prevents addition of the priming xylose [100], the complex presumably forms prior to Golgi entry and blocks access of xylosyltransferases to serine by steric hindrance. Interestingly, the expression of both CA10 and FAM19A1 are differentially regulated by synaptic activity in cultured hippocampal neurons [34]. Increased neuronal firing caused elevated expression of CA10 and reduced expression of FAM19A1, suggesting a possible mechanism for cotranslational regulation of neurexin HS in an activitydependent manner. However, this hypothesis is currently speculative as it remains to be shown that glycosylation patterns of neurexin changes upon neuronal activity or absence of CA10 and/or FAM19A1-4.

The synaptic HS glycoproteome
An unanswered question is how the neurexin HS chain relates to those of other HSPGs, such as glypicans and syndecans, at specific synapses. Glypican-4, for example, concentrate at presynaptic terminals [86] and promote synapse formation in different systems, also when secreted from astrocytes [90]. Previous work has shown that the HS of glypican-4, similar to that of neurexins, can bind to both LRRTM4 and PTPRS [86][87][88]. Co-immunoprecipitation experiments have shown that LRRTM4 cannot simultaneously bind to both glypicans and neurexins [86], suggesting that LRRTM4 contain a single binding site for HS that competes for binding to HS of glypicans and neurexins, respectively. Which HSPG chain that constitutes its physiological ligand, and whether they have overlapping or complimentary functions, remain unknown. Glypican-4 was also recently found to bind, via its HS chain, to the orphan G-protein coupled receptor GPR158 [103]. Interestingly, mice knockout for Gpr158 showed defects in mossy fibre-CA3 synapses resembling those of Nrxn1DHS mice [11,103]. Clearly, both glycan-protein and protein-protein interactions, and their affinities for different ligands, need to be taken into account in models of the molecular architecture of specific synapses.
Finally, it cannot be excluded that additional synaptic HSPGs remain to be discovered. A glycoproteomic method to isolate and identify chondroitin sulfate and HS glycopeptides by nano-liquid chromatographytandem mass spectrometry (nLC-MS/MS) has been developed [104,105]. This methodology enabled the identification of two novel HSPGs in murine insulinsecreting cells [42], together with numerous novel CSPGs in humans and in various animal models [104,106,107]. A similar approach could be used to characterize synaptic HSPGs.

Discussion and outlook
A comprehensive understanding of neurexin functions is still lacking, despite being extensively studied. One reason for this is the extensive complexity that arise from its numerous isoforms, glycans modifications and ligands. The discovery of neurexins as HSPGs have explained some previous observations [11] but at the same time shown that the system may be even more complex than previously thought. We have argued that a better understanding of the fine structure of the neurexin HS chain as well as its affinities for specific ligands will be required to fully comprehend its function. Addressing these issues, however, faces technical challenges as no currently available method permits sequencing of the entire polysaccharide, and attempts to do so are complicated by the large heterogeneities in HS length and fine structure. Nevertheless, improved methods will advance the field. For example, an optimised mass spectrometry-based method has been employed to determine the predominant sequence of the single chondroitin sulfate chain of bikunin [108]. In addition, methods based on engineered glycosyltransferases that allow chemo-genetic labelling of HS could be used to reveal cell-type specific heterogeneities in HS chain composition [109]. Meanwhile, carefully controlled indirect approaches will have to be taken. In summary, we find it unlikely that the neurexin HS chain would contain strict codes for ligand binding. Instead, variations in its fine structure may fine tune affinities to various ligands to illustrate a common theme in biology: it is not all-or-nothing but everything in between. Whether the neurexin HS chain and its fine structure are dynamically regulated across the lifetime of specific synapses, and thereby may contribute to synaptic plasticity, should be one of several interesting topics for future studies.