Deorphanizing FAM19A proteins as pan-neurexin ligands with an unusual biosynthetic binding mechanism

Synapses are the basic units of information processing in the brain. Although a fundamental understanding of how synapses transfer and process information has emerged recently, how synapses are formed, maintained, and continuously restructured throughout life remains largely unknown. Synapses are likely controlled at every phase of their life cycle by trans-synaptic adhesion signals, starting with their initial formation and the specification of their properties, and continuing throughout life with their restructuring during synaptic plasticity, but how precisely synaptic adhesion molecules shape synaptic properties remains unclear. Many candidate synaptic adhesion molecules have been described (Jang et al., 2017; Krueger-Burg et al., 2017; Ribic and Biederer, 2019; Südhof, 2018). Among these, neurexins are arguably the best studied (reviewed in Han et al., 2019; Kasem et al., 2018; Reissner et al., 2013; Rudenko, 2019; Südhof, 2017). Despite much work, however, the functions and mechanisms of action of neurexins remain enigmatic.

Accumulating evidence suggests that neurexins play distinct roles at different types of synapses rather than performing a single canonical function at all synapses (e.g., see Anderson et al., 2015; Aoto et al., 2015; Chen et al., 2017). Neurexins are highly conserved type I membrane proteins predominantly localized to presynaptic terminals (Missler et al., 2003; Trotter et al., 2019; Ushkaryov et al., 1992), although astrocytes also express neurexins (Gokce and Südhof, 2013; Tasic et al., 2018; Zeisel et al., 2018). Each of the three vertebrate neurexin genes (Nrxn1-3) produces two isoforms (α- and β-neurexins) via different promoters (Ushkaryov et al., 1992, 1994). In addition, the neurexin-1 gene contains a third promoter that produces Nrxn1γ (Sterky et al., 2017). α-Neurexins contain six laminin-neurexin-sex hormone-binding globulin (LNS) domains with three interspersed epidermal growth factor (EGF) domains, a juxtamembranous stalk region that includes a cysteine-loop (CysL) domain, a transmembrane region, and a short cytosolic tail containing a PSD-95/Discs-large/ZO-I (PDZ)-binding domain (Gokce and Südhof, 2013; Ushkaryov and Südhof, 1993; Ushkaryov et al., 1992). β-Neurexins possess a unique N-terminal sequence and splice into α-neurexins just N-terminal to the sixth (last) LNS domain (Ushkaryov et al., 1992, 1994). Nrxn1γ also contains a unique N-terminal sequence that is followed by part of the stalk region (Sterky et al., 2017). Due to the usage of up to six canonical sites of alternative splicing and three different promoters, thousands of neurexin variants are expressed (Schreiner et al., 2014; Treutlein et al., 2014). Adding to this complexity, the neurexin stalk region is modified by heparan sulfate (HS) in a subset of neurexins (Zhang et al., 2018).

Neurexins organize both the pre- and postsynaptic machinery and determine specific synaptic properties by forming trans-synaptic complexes with a multitude of diverse ligands that include intracellular, transmembrane, and secreted proteins (Südhof, 2017). However, known neurexin ligands seem to account for only a fraction of neurexin interactions, and identifying and validating novel neurexin ligands remain a challenge.

FAM19As (FAM19A1-A5, aka TAFA1-5), constitute a family of brain-enriched proteins that are differentially expressed across brain regions (Tom Tang et al., 2004; Yong et al., 2020; Fig. S1). FAM19As resemble cytokines and are evolutionarily highly conserved in vertebrates, but their functions remain enigmatic. FAM19As contain a signal peptide followed by a conserved region with 10 cysteine residues that likely form intramolecular disulfide bonds and include a CC-chemokine motif with distant homology to macrophage inflammatory protein 1α (Tom Tang et al., 2004).

Fam19a1 knockout (KO) mice have decreased body weights, are hyperactive, and exhibit reduced anxiety-like behaviors as well as impaired fear memory acquisition and recall (Lei et al., 2019; Yong et al., 2020). In contrast, Fam19a2 and Fam19a3 KO mice exhibit increased anxiety-like behaviors (Choi et al., 2018; Kim et al., 2017), while Fam19a4 KO mice show increased injury-induced pain hypersensitivity and abnormal neuronal excitability (Delfini et al., 2013). These findings broadly suggest that FAM19As are important for normal nervous system function, but a mechanistic understanding of their roles is lacking. A number of potential receptors for FAM19As have been reported, including G-protein–coupled receptor 1 for FAM19A1 (Zheng et al., 2018), formyl-peptide receptor (FPR) 1 for FAM19A4 (Wang et al., 2014), and FPR2 and sphingosine-1-phosphate receptor 2 for FAM19A5 (Park et al., 2017; Wang et al., 2018). However, stoichiometric interactions were not documented for these putative receptors, and the FAM19A binding sites remain uncharacterized.

In the present study, we have identified FAM19A1-A4 proteins as stoichiometric subunits of neurexin complexes. We first detected FAM19A1 and FAM19A2 in a proteomic screen of endogenous neurexin complexes immuno-isolated from mouse brain. FAM19A1-A4 bind to all major neurexin isoforms by forming intermolecular disulfide bridges via the neurexin CysL-domain and via binding to an adjacent sequence of neurexins. Neurexins, in turn, are required for the stable expression and surface transport of FAM19A1. Exogenous FAM19A1 impairs neurexin O-glycosylation and HS modification, and lowers the levels of total, but not surface-localized, α-neurexins in mixed neuron/glia cultures obtained from mouse hippocampus. Thus, our findings deorphanize a family of brain-enriched putative cytokines as neurexin ligands and demonstrate that FAM19As regulate neurexin modifications, thereby expanding our understanding of the neurexin interactome to include regulatory subunits.

To systematically identify neurexin ligands, we performed affinity immuno-purifications of endogenous neurexin-1 from adult mouse brain. Using quantitative liquid chromatography-tandem mass spectrometry (Sterky et al., 2017), we found that FAM19A1 and FAM19A2 were coimmunoprecipitated with neurexins (Fig. 1 A). FAM19A1 and FAM19A2 (aka TAFA1 and TAFA2; Tom Tang et al., 2004) belong to a family of five cysteine-rich, vertebrate-specific, and evolutionarily conserved secreted brain proteins (Fig. 1 A and Fig. S1). Consistent with a sequence resemblance to cytokines, FAM19As were shown to bind to several G-protein–coupled receptors (GPCRs; Park et al., 2017; Wang et al., 2014, 2018; Zheng et al., 2018). Given these previous findings, the purification of endogenous neurexin/FAM19A complexes from brain was surprising. However, we isolated FAM19A1 and FAM19A2 in a complex with neurexins from brain homogenates in independent experiments using multiple neurexin antibodies, suggesting that FAM19As are robust constituents of neurexin complexes in brain.

FAM19A1-A5 are highly homologous (Fig. 1 A), suggesting that FAM19A3, FAM19A4, and FAM19A5 may also bind to neurexins. To test this and to validate the interactions of FAM19A1 and FAM19A2 with neurexins, we performed coimmunoprecipitation (coIP) experiments with recombinant proteins. We coexpressed in transfected HEK293T cells each of the five FAM19A isoforms (tagged with a C-terminal V5 epitope) and secreted, truncated neurexin-1β (Nrxn1β-ECD). Nrxn1β-ECD was tagged with a C-terminal Myc epitope and contained all Nrxn1β extracellular domains, but lacked its transmembrane region and cytoplasmic tail. We immunoprecipitated FAM19A1-A5 from the transfected HEK293T cell medium using V5 antibodies and analyzed the immunoprecipitates by immunoblotting for Nrxn1β and FAM19As (Fig. 1 B). We also examined a Nrxn1β mutant in which the CysL-domain of neurexins was deleted (referred to as Nrxn1β-ECD-ΔCysL) because we observed possible disulfide bonding of FAM19As with Nrxn1β (see below). The immunoprecipitations of secreted neurexin/FAM19A complexes revealed that FAM19A1-A4, but not FAM19A5, bound to the Nrxn1β-ECD (Fig. 1 B). Strikingly, deletion of the CysL-domain of Nrxn1β abolished FAM19A-binding.

To independently confirm binding of FAM19As to Nrxn1β, we analyzed the medium of HEK293T cells coexpressing Nrxn1β-ECD or Nrxn1β-ECD-ΔCysL with FAM19A1-A5 by non-reducing or reducing SDS-PAGE and immunoblotting (Fig. 1 C). Under non-reducing conditions, Nrxn1β migrated as a covalent complex with FAM19A1-A4, but not with FAM19A5. The Nrxn1β/FAM19A complexes were absent after deletion of the CysL-domain and were dissociated by reducing agents (Fig. 1 C). Subtype-specific peptides for FAM19A3 and FAM19A4 were not detected by mass spectrometry in our original experiments, probably because FAM19A3 and FAM19A4 expression are very low in brain (Fig. S1; Saunders et al., 2018; Tasic et al., 2018; Tom Tang et al., 2004; Zeisel et al., 2018).

Do FAM19A1-A4 proteins bind to all neurexin isoforms? To address this question, we measured the surface exposure of neurexin/FAM19A complexes on HEK293T cells expressing full-length neurexins. In this assay, we coexpressed various EGFP-tagged full-length neurexin isoforms or control proteins with V5-tagged FAM19A1-A5 proteins and imaged cell-surface levels of FAM19A1-A5 by immunocytochemistry (Fig. 2 and Fig. S2). As a positive control, we used cerebellin-1 (Cbln1), a secreted neurexin ligand (Matsuda and Yuzaki, 2011; Uemura et al., 2010). As expected, Cbln1 was displayed at the cell surface when Nrxn1β containing splice site 4 (SS4) was coexpressed, but not when Nrxn1β lacking SS4 was coexpressed with Cbln1 (Fig. 2, A and B). As a negative control, we used leucine-rich repeat transmembrane protein 2 (LRRTM2) because FAM19As are not expected to bind to LRRTM2 (Fig. 2 C).

When we analyzed various splice variants of α- and β-neurexins and all five FAM19As (FAM19A1-A5), we found that all neurexin splice variants tested formed surface-exposed complexes with coexpressed FAM19A1-A4, but not with FAM19A5 (Fig. 2, D–P; and Fig. S2). Thus, consistent with the immunoprecipitation experiments, FAM19A1-A4, but not FAM19A5, bind to α- and β-neurexins independent of alternative splicing.

To obtain direct evidence for a stable complex of neurexins and FAM19A1-A4, we purified recombinant Nrxn1β-ECD, Nrxn1β-ECD/FAM19A1 complexes, and FAM19A1 from the medium of transfected HEK293S cells (which lack N-acetylglucosaminyltransferase I, GnTI⁻). Nrxn1β-ECD was Myc- and 6xHis-tagged, and FAM19A1 was V5- or Twin-Strep-tagged (Fig. 3 and Fig. S3, H–P). An O-linked glycosylation-deficient Nrxn1β-ECD mutant in which all serines and threonines in the O-glycosylated stalk region were changed to glycines was used to reduce the heterogeneity of recombinant Nrxn1β-ECD (note that neurexin is poorly HS-modified in HEK293 cells; Zhang et al., 2018). Whereas FAM19A1 was efficiently secreted from HEK293S cells in a complex with Nrxn1β-ECD, low levels of secreted FAM19A1 were detected when FAM19A1 was expressed alone (Fig. S3, H–J), suggesting that FAM19A1 is poorly secreted in the absence of neurexins. Moreover, recombinant FAM19A1, when expressed without neurexins, partly formed disulfide-bonded dimers (Fig. S3, B and L). To enable the purification of sufficient quantities of recombinant FAM19A1 for biophysical studies, we replaced the endogenous signal peptide with that of Igκ and tagged FAM19A1 with a Twin-Strep moiety that allows efficient affinity purification but confers microheterogeneity to FAM19A1 (Fig. 3 F), likely because the bacterial Twin-Strep tag is O-glycosylated in eukaryotic cells.

We analyzed purified recombinant Nrxn1β-ECD, FAM19A1, and Nrxn1β-ECD/FAM19A1 complexes by size-exclusion chromatography coupled with multi-angle light scattering (SEC-MALS) to determine their absolute molar masses (Fig. 3, G and H). These experiments revealed that the Nrxn1β-ECD is a monomer and FAM19A1 is a mixture of monomers and dimers in solution, while Nrxn1β-ECD and FAM19A1 form a stoichiometric heterodimer (Fig. 3 I). Thus, Nrxn1β-ECD and FAM19A1 form a stable 1:1 complex.

We observed a robust FAM19A cell-surface signal in the surface-exposure assay when FAM19As were coexpressed with Nrxn1β (Fig. 2, Fig. S2, and Fig. S3, E and F). Upon adding recombinant FAM19A1 to cells expressing Nrxn1β, however, we detected only weak FAM19A1 binding that was independent of the Nrxn1β CysL-domain, suggesting it is non-specific (Fig. S3, E and F). Moreover, whereas coexpressed FAM19A1 was disulfide-bonded to Nrxn1β (Fig. 1 C and Fig. 5 E), mixing of recombinant FAM19A1 and Nrxn1β-ECD purified separately failed to produce disulfide-bonded Nrxn1β/FAM19A1 complexes (Fig. S3 G). Together, these results suggest that FAM19As bind to neurexins by forming intermolecular disulfide bridges during transport through the secretory pathway.

To test whether recombinant Nrxn1β/FAM19A1 complexes are also disulfide bonded, we analyzed purified recombinant Nrxn1β-ECD/FAM19A1 complexes by SDS-PAGE in the absence and presence of reducing agents (Fig. 4 A). The complex was stable in SDS-sample buffer in the absence of reducing agents despite a high concentration of SDS, but was dissociated by disulfide bond reduction. Furthermore, on native PAGE (in which the migration of proteins is not strictly dependent on molar mass), the Nrxn1β-ECD/FAM19A1 complex was maintained in the absence of reducing agents, but disrupted by reducing agents (Fig. 4 B).

These experiments show that in the complex, Nrxn1β and FAM19A1 are disulfide-bonded to each other. Are the disulfide bonds between Nrxn1β and FAM19A1 produced physiologically within the secretory pathway, or do they form artifactually in the oxidizing environment of the medium after secretion and/or cell lysis (Johnston and Südhof, 1990)? The observations that separately purified FAM19A1 and Nrxn1β-ECD do not spontaneously form covalent disulfide-bonded complexes (Fig. S3 G) and that recombinant FAM19A1 fails to bind to surface-displayed Nrxn1β (Fig. S3, E and F) indicate that the disulfide-bonded Nrxn1β/FAM19A1 complex results from chaperoned interactions in the secretory pathway. To independently test this hypothesis, we expressed Nrxn1β-ECD and FAM19A1 either alone or in combination in HEK293T cells, collected the cells and the medium directly in SDS-sample buffer, and analyzed the samples without or with the addition of reducing agents by SDS-PAGE and immunoblotting (Fig. 4, C and D). The Nrxn1β-ECD/FAM19A1 complex was readily detectable as a single band on SDS-PAGE under non-reducing conditions both in the medium and the cells when the two proteins were coexpressed. Thus, the complex likely forms in cells before secretion and exposure to the extracellular milieu, suggesting it is assembled in the secretory pathway.

We next investigated which neurexin sequences mediate the formation of the neurexin/FAM19A complex. Since all neurexin isoforms and splice variants tested form this complex (Fig. 2 and Fig. S2), we analyzed the FAM19A-binding sequences of Nrxn1β-ECD as the smallest neurexin variant that still binds to FAM19As. We systematically tested deletions of various Nrxn1β-ECD domains in Nrxn1β/FAM19A1 complex formation using coIPs (Fig. 5, A–D). Only the deletion of the CysL-domain of Nrxn1β blocked assembly of Nrxn1β/FAM19A1 complexes, although some Nrxn1β constructs were expressed poorly, likely due to misfolding (Fig. 5 B). In these experiments, we also examined whether O-glycosylation of the stalk domain of Nrxn1β is required for Nrxn1β/FAM19A1 complex formation, but observed no change in complex formation as a function of glycosylation (Fig. 5 B and Fig. S4 A).

The results from the deletion mutants agree with the original validation experiments of the neurexin/FAM19A complex, in which deletion of the neurexin CysL-domain also blocked complex formation (Fig. 1, B and C). Since these findings suggest that the CysL-domain is essential for FAM19A binding, we analyzed its sequence in detail. The CysL-domain consists of a nine-residue loop with two flanking prolines and five negatively charged residues (CPSDDEDIDPC). Strikingly, no mutation of the loop sequence, including a substitution of positively charged arginines for the negatively charged residues, affected Nrxn1β/FAM19A1 complex formation (Fig. 5, C and D; and Fig. S4 A). Thus, despite remarkable sequence conservation, the actual sequence of the CysL-domain is not important for FAM19A1-A4 binding. Moreover, neither expanding the size of the loop by inserting two glycine residues nor replacing the entire nine-residue Nrxn1β CysL-domain with the non-homologous 13-residue cysteine-loop from Type A γ-aminobutyric acid (GABA)_A-β3 receptors (Miller and Aricescu, 2014) disrupted binding (Fig. 5, C–E). Mutation of the two cysteine residues forming the disulfide-bonded loop (Sterky et al., 2017), however, completely abolished formation of Nrxn1β/FAM19A1 complexes (Fig. 5, C–E). Thus, the only feature of the Nrxn1β CysL-domain that appears to enable FAM19A1 binding is the cysteine-loop itself, not its size or sequence.

These observations raised the question of whether FAM19A1 generally binds to all cysteine-loop domains. However, the extracellular domain of neuroligin-1 containing an insert in splice site A1 that includes disulfide-bonded loops (Araç et al., 2007; Hoffman et al., 2004; Ichtchenko et al., 1996) failed to form a complex with FAM19A1 (Fig. 5 D and Fig. S4 A), suggesting specificity. To test this further, we asked whether FAM19A1 could form a complex with the solvent-exposed cysteine-loops of GABA_A-β3 receptors (Miller and Aricescu, 2014). We coexpressed FAM19A1-V5 with HA-tagged full-length Nrxn1β (positive control), mutant full-length Nrxn1β lacking the CysL-domain (negative control), and GABA_A-β3 receptors. We then measured binding of FAM19A1 to these proteins using coIPs with antibodies against HA or V5 (Fig. 5 F). FAM19A1 did not bind to GABA_A-β3 receptors, suggesting that the cysteine-loop domain alone cannot form neurexin/FAM19A complexes. Consistent with this conclusion, FAM19A1 did not bind to Nrxn1γ, which contains the CysL-domain of Nrxn1β but differs in its upstream sequences and which does bind to CA10, another recently described cis-ligand of neurexins (Fig. S4, B and C; Sterky et al., 2017).

We noted that whenever we coexpressed FAM19A1 with a protein that does not bind to FAM19A1, such as Nrxn1β lacking a CysL-domain or GABA_A-β3 receptors (Fig. 5 F), FAM19A1 levels appeared to be lower than when FAM19A1 was coexpressed with a neurexin that binds to FAM19A1, despite the fact that the expression conditions were the same. Quantifications confirmed that the secretion of FAM19A1 was ∼5- and ∼30-fold higher, respectively, when FAM19A1 was coexpressed with WT Nrxn1β than when it was coexpressed with mutant Nrxn1β or with GABA_A-β3 receptors (Fig. 5 G). These observations, consistent with the finding that FAM19A1 is poorly transported through the secretory pathway in the absence of neurexin (Fig. S3, H–J), support the conclusion that neurexin binding to FAM19A1 in the secretory pathway stabilizes FAM19A1.

Since neurexins shape synapse properties (Südhof, 2017), does FAM19A1 binding to neurexins affect synapse formation and/or synaptic transmission? To address this question, we prepared mixed cultures of neurons and glia from the hippocampus of newborn mice that are triple homozygotes for the conditional knockout (cKO) of neurexins-1, -2, and -3 (referred to as Nrxn123 cKO mice; Chen et al., 2017). We infected the cultures at day in vitro (DIV) 4 with lentiviruses encoding mutant inactive (ΔCre; control; Kaeser et al., 2009) or active Cre-recombinase (Cre), and at DIV6 with lentiviruses encoding FAM19A1-V5 or with control lentiviruses. We analyzed the cultures by immunocytochemistry, biochemistry, and electrophysiology at DIV14–16. With this experimental design, we compared cultures expressing or lacking α- and β-neurexins, without or with coexpression of FAM19A1 (see Fig. S5, E and F, demonstrating that the Nrxn123 cKO ablates neurexin expression).

Immunocytochemical labeling for vesicular glutamate transporter 1 (vGluT1) and vesicular GABA transporter (vGAT), markers for excitatory and inhibitory synapses, respectively, showed that neither the neurexin deletion nor the expression of exogenous FAM19A1 had a significant effect on synaptic puncta density or size, quantified at secondary or tertiary dendrites visualized by MAP2 staining (Fig. 6, A–C and G–I). These results agree with previous studies demonstrating that neurexins are not essential for synapse formation (Chen et al., 2017; Luo et al., 2020; Missler et al., 2003) and suggest that FAM19A1 as a neurexin ligand is not required for synapse formation.

Next, we asked whether FAM19A1 regulates synaptic transmission even if it does not affect synapse numbers. To gain initial insight into this question, we measured miniature excitatory and inhibitory postsynaptic currents (mEPSCs and mIPSCs, respectively) as a function of exogenous FAM19A1 expression (Fig. 6, D–F and J–L). We observed only a modest effect of FAM19A1 on mEPSC amplitudes and frequencies (Fig. 6, D–F). However, FAM19A1 caused a significant decrease (∼20%) in mIPSC frequency (Fig. 6, J–L), suggesting that FAM19A1 binding to neurexins impacts synaptic function.

In the experiments of Fig. 6, we also analyzed surface-displayed FAM19A1 as a function of neurexin expression. FAM19A1 was abundantly present on the neuronal cell surface in a punctate pattern when endogenous neurexins were present (Fig. 6, A and G; and Fig. S5, A and B). Deletion of neurexins, however, decreased surface-displayed FAM19A1 levels dramatically (∼75–90%; Fig. 6 A, C, G, and I; and Fig. S5, A and B). FAM19A1 “puncta” were more abundant than synaptic puncta as identified by immunocytochemistry for vGluT1 and vGAT (Fig. 6), and were primarily localized at dendrites (Fig. S5, A and B). These results suggest that neurexins are the major binding partners of FAM19A1 because their deletion greatly decreases the levels of exogenously expressed surface-displayed FAM19A1.

Given that neurexins are essential for surface transport of FAM19A1, does FAM19A1 regulate the intracellular transport and post-translational modification of neurexins? To address this question, we generated dissociated hippocampal cultures containing or lacking endogenous α- and β-neurexins without or with expression of exogenous FAM19A1 as described in Fig. 6. We then biotinylated the cell-surface proteins of these cultures at DIV16, purified the biotinylated cell-surface proteins using streptavidin, and analyzed the total cellular proteins (“input”), “flowthrough” proteins (non-biotinylated, predominantly intracellular proteins), and “biotinylated” purified cell-surface proteins (Fig. 7 A).

We first reexamined the effects of endogenous neurexins on exogenous FAM19A1. Deletion of neurexins suppressed total (∼45% decrease) and surface-exposed FAM19A1 levels (∼60% decrease, Fig. 7 B), confirming the conclusion of Fig. 6 that neurexins are the major physiological binding partners of FAM19A1 and suggesting that neurexin sequences within or surrounding the CysL-domain are not saturated by endogenous ligands.

Next, we examined the effects of exogenous FAM19A1 on neurexins. Surprisingly, exogenous FAM19A1 expression decreased (∼40%) the total levels of α-neurexins without changing their surface levels (Fig. 7 C). Furthermore, FAM19A1 caused a large shift in the apparent mobility of α-neurexins on reducing SDS-gels (Fig. 7 D). As shown below, this mobility shift was due to a FAM19A1-induced decrease in the O-glycosylation and HS modification of α-neurexins. Under control conditions, ∼70% of surface α-neurexins were extensively O-glycosylated and/or HS-modified. Upon coexpression of FAM19A1, only ∼30% of surface α-neurexins were extensively O-glycosylated and/or HS-modified (Fig. 7 D). Additionally, exogenous FAM19A1 increased the surface levels of non-HS–modified β-neurexins, although monitoring β-neurexins is difficult due to their low abundance and extensive glycosylation (Fig. 7, E and F).

To further characterize the effect of FAM19A1 on neurexins, we examined their O-glycosylation and HS modification in detail. We purified the biotinylated cell-surface proteins of cultured hippocampal cells without and with exogenous FAM19A1 expression as described above, and treated the surface proteins with a mix of glycosidases to remove N- and most O-linked sugars, or with heparinases I, II, and III to remove HS (Fig. 8 and Fig. S5 H). Afterward, we analyzed the proteins by reducing SDS-PAGE and immunoblotting.

As observed above (Fig. 7), exogenous FAM19A1 decreased the apparent molecular weight of most α- and β-neurexin bands by at least 20 kD (Fig. 8, B and E). Strikingly, N- and O-linked deglycosylation of neurexins caused a similar, but not identical, shift in apparent molecular weight of α-neurexins as exogenous FAM19A1 expression. The combination of FAM19A1 expression and deglycosylation, conversely, produced a slightly bigger shift in molecular weight than deglycosylation or FAM19A1 expression alone (Fig. 8, A–C). Whereas exogenous FAM19A1 had no detectable effect on HS-modified β-neurexins, it slightly decreased the apparent molecular weight of non-HS–modified β-neurexins. Viewed together, these results indicate that FAM19A1 significantly down-regulates neurexin O-glycosylation.

Heparinases also caused major shifts in the apparent molecular weights of neurexins, as shown previously (Zhang et al., 2018). These shifts differed from those observed for N- and O-deglycosylation, although for α-neurexins they were again broadly similar to shifts induced by exogenous FAM19A1 (Fig. 8, D–F). The combination of heparinases and FAM19A1 expression produced only a slightly bigger shift in the molecular weight of α-neurexins than either condition alone, suggesting that FAM19A1 suppresses the HS modification of neurexins (Fig. 8, E and F). Although heparinase treatment led to a similar de-enrichment of HS-modified β-neurexins with and without exogenous FAM19A1, a slight enrichment in lower molecular weight species was maintained with exogenous FAM19A1, which may be partly due to differences in O-glycosylation. Moreover, an analysis of surface proteins modified by HS using an antibody to the “HS stub,” which remains on proteins after HS removal by heparinases, revealed that α-neurexins represent a minor component of the HS-modified proteome (Fig. S5 H).

Viewed together, these experiments suggest that subsets of α- and β-neurexins are modified by O-linked sugars and/or by HS. FAM19A1 appears to regulate both modifications. Given that in neurexins, both the serine that is modified by HS (Ser³¹⁶ in Fig. 5 C; Zhang et al., 2018) and the serines and threonines that are O-glycosylated (Ushkaryov et al., 1992, 1994) are proximal to the FAM19A1 binding site and that FAM19A1 binding occurs within the secretory pathway, the regulation of these modifications by FAM19A1 is plausible.

The regulation of the O-glycosylation and HS modification of neurexins by FAM19As raises the question of whether such regulation is, in turn, regulated. As an initial approach to this question, we analyzed the expression of FAM19As and neurexins as a function of activity in hippocampal cultures (Fig. 9). Strikingly, neuronal depolarization decreased Fam19a1 and Fam19a2 expression by ∼60% and ∼20%, respectively, over a 48-h time course (Fig. 9, C and D). In contrast, Car10 and Nrxn3 were up-regulated ∼30–40%, whereas Car11 and Nrxn2 were down-regulated ∼25–30% (Fig. 9, F, G, I, and J). Since Fam19a1, Fam19a2, Car10, and Car11 are differentially expressed in hippocampal neurons (Fig. S1, E–H), these data suggest that the expression of these genes is subject to divergent activity-dependent regulation.

Neurexins are presynaptic adhesion molecules that function as central regulators of synapse properties (reviewed in Kasem et al., 2018; Südhof, 2017). Despite much work, however, the functions of neurexins and their mechanisms of action are incompletely understood. A major emerging theme for understanding neurexins is their vast number of physiological ligands, which at present encompass at least eight gene families (cerebellins, α-dystroglycan, neurexophilins, neuroligins, LRRTMs, calsyntenins, C1qls, and CA10/11) with more than 25 members. All of these ligands, except for CA10/11 (Sterky et al., 2017) and C1ql2/3 (Matsuda et al., 2016), bind to either the LNS2 or LNS6 domain, and for many domains of neurexins, no ligands have yet been described. Here, we identified FAM19A1-A4 as novel interactors of α- and β-neurexins that exhibit remarkable properties unlike other previously studied ligands. FAM19A1-A4 were presumed to be secreted cytokines that bind to GPCRs in a signaling function. However, we demonstrate that they are stoichiometric subunits of neurexin complexes. FAM19As are poorly secreted on their own and require neurexins for stable transport through the secretory pathway. All previously described ligands interact with neurexins in a reversible fashion, whereas the neurexin-binding mode of FAM19A1-A4 is unusual in that it requires transit through the secretory pathway and involves the formation of intermolecular disulfide bonds mediated by the neurexin CysL-domain. To the best of our knowledge, this is a structurally unprecedented interaction mode that differs from that of other neurexin ligands.

Seven lines of evidence demonstrate that the neurexin/FAM19A complex represents a physiological interaction, that neurexins are the predominant binding partners for FAM19A1-A4, and that their interaction serves to regulate post-translational processing of neurexins. First, well-controlled immunoprecipitations of endogenous neurexins from mouse brain showed that endogenous FAM19A1 and FAM19A2 are present in a complex with neurexins in brain (Fig. 1 A). Given that FAM19A isoforms are selectively expressed in small subsets of neurons throughout the brain (Fig. S1) and thus are not abundant proteins, and given the cis requirement for neurexin/FAM19A binding (Fig. S3, E–G), the robust detection of FAM19A1 and FAM19A2 in these immunoprecipitations suggests that neurexin/FAM19A complexes are prevalent in neurons expressing FAM19A1-A4 isoforms. Second, recombinant neurexins bind to FAM19A1-A4, but not FAM19A5, when produced in transfected HEK293T cells, as documented by coIPs, migration on SDS-PAGE under non-reducing but not reducing conditions, and surface exposure assays of FAM19A1-A4 and neurexins (Fig. 1, Fig. 2, and Fig. S2). Third, biophysical measurements, in particular SEC-MALS, demonstrated that Nrxn1β forms a stoichiometric, heterodimeric, covalent complex with FAM19A1 (Fig. 3 and Fig. 4). Fourth, mutagenesis showed that the two cysteine residues of the Nrxn1β CysL-domain were the only absolute requirement for assembly of the covalent Nrxn1β/FAM19A1 complex (Fig. 5). These two cysteine residues presumably bind to cysteine residues in FAM19A1-A4, which would explain why FAM19A5 does not bind because it lacks two conserved cysteine residues (Fig. 1 A). Additionally, the stoichiometry of the Nrxn1β/FAM19A1 complex suggests that it involves two intermolecular disulfide bonds. FAM19A1 binding to Nrxn1β was independent of the neurexin CysL sequence, but was specific overall since the cysteine-loop sequence of GABA_A-β3 receptors bound to FAM19A1 when it was transferred into Nrxn1β, but not when it was retained within GABA_A-β3 receptors. Moreover, Nrxn1γ, which contains the neurexin-1 CysL-domain and binds to CA10/11 but lacks other N-terminal neurexin-1 sequences (Sterky et al., 2017), did not bind to FAM19A1 (Fig. S4, B and C). Thus, additional N-terminal Nrxn1β sequences likely facilitate FAM19A binding even though no single specific sequence we tested was absolutely required. These data on the structure of the neurexin/FAM19A complex reveal a remarkable mechanism, namely the formation of a covalent complex in the secretory pathway. Fifth, FAM19A1 expression levels were significantly lower when FAM19A1 was expressed alone or coexpressed with non-binding proteins compared with when FAM19A1 was coexpressed with neurexins (Fig. 5, D, F, and G; and Fig. S4 A). Similarly, FAM19A1 expression levels were dramatically lowered when endogenous neurexins were deleted (Fig. 7 B), demonstrating that FAM19A1 requires neurexin binding for stability. Sixth, exogenously expressed FAM19A1 was displayed on the neuronal cell surface in a punctate pattern when endogenous neurexins were present, but was largely absent when endogenous neurexins were deleted (Fig. 6, C and I; and Fig. S5, A and B). Thus, neurexins are the major receptors for FAM19A1. Seventh and finally, FAM19A1 expression altered neurexin O-glycosylation and HS modification (Fig. 7, Fig. 8, and Fig. S5 H). Thus, consistent with a disulfide-mediated interaction that occurs within the secretory pathway, FAM19A1 regulates the post-translational modification of neurexins.

Viewed together, our results demonstrate that FAM19A1-A4 are physiological subunits of neurexin-ligand complexes. How can we reconcile our observations with studies suggesting that FAM19A1 and FAM19A4 are secreted ligands for G-protein–coupled receptor 1 (Zheng et al., 2018) and FPR1 (Wang et al., 2014), respectively? Our results do not rule out the possibility that FAM19As perform multiple functions and act as cytokines in non-neuronal cells and as subunits of neurexin complexes in neurons. However, all studies identifying putative receptors for FAM19As as candidate cytokines were performed by measuring surface binding of recombinant proteins to transfected cells. Although informative, such surface binding can be misleading given the small amounts of proteins involved and the indirect nature of the assay. No direct interaction studies with recombinant proteins, as shown in Fig. 3 here, were performed. Clarification of the role of the GPCRs implicated as FAM19A receptors will require such studies. It is remarkable that all GPCRs proposed as FAM19A receptors are also known to function as receptors for other unrelated ligands, and that different receptors were suggested for different FAM19A isoforms despite their homology.

What is the physiological role of the neurexin/FAM19A complex? Despite more than 1,000 papers, the functions of neurexins are only partly understood. A synaptic role is overwhelmingly demonstrated, but owing to the expression of thousands of splice variants, such a role appears to be multifaceted and diverse (Südhof, 2017). In cultured neurons, neurexins have modest effects on synapses (see, for example, the lack of an effect of pan-neurexin deletions on synapse numbers in Fig. 6). However, in vivo neurexins control fundamental features of synapses, such as the release probability (Chen et al., 2017), Ca²⁺-channels (Missler et al., 2003; Brockhaus et al., 2018; Luo et al., 2020), and post-synaptic receptor composition (Aoto et al., 2013; Dai et al., 2019). At synapses, neurexins form nanoclusters (Trotter et al., 2019), and the neurexin modifications that are regulated by FAM19As may affect such nanoclusters. Fam19a mRNAs are expressed in specific subsets of neurons throughout the brain (Saunders et al., 2018; Tasic et al., 2018; Zeisel et al., 2018; Fig. S1, E and F), which may give rise to the different phenotypes of Fam19a1-a4 KO mice and add to the diversity of neurexin functions by conferring specific properties onto neurexins in neurons coexpressing FAM19As with neurexins. Consistent with this possibility, FAM19A1 regulates the diverse O-glycosylation and HS modification of neurexins, suggesting that FAM19As may shape the functional diversity of neurexins by controlling their post-translational modifications (Fig. 10). These modifications appear to affect the steady-state levels of neurexins, as shown by the decreased levels of total α-neurexins upon exogenous expression of FAM19A1 (Fig. 7 C), but more importantly likely alter ligand interactions of neurexins (Roppongi et al., 2020; Zhang et al., 2018). Consistent with a role for FAM19As in regulating neurexins, FAM19A1 is itself regulated by activity, as shown in a subset of excitatory neurons in vivo (Hrvatin et al., 2018) and as we observed in cultured hippocampal cells (Fig. 9). Such activity-dependent regulation of FAM19A expression in turn could contribute to the control of neurexin O-glycosylation and HS modification. Thus, we propose that FAM19As are selectively expressed in an activity-dependent manner in a subset of neurons to form stoichiometric cis-complexes with neurexins, thereby regulating their post-translational modifications and activities.

Mice were group-housed with littermates of the same sex and maintained on a 12-h light/dark cycle, with access to food and water ad libitum. All procedures were performed in adherence with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and were approved by the Stanford University Administrative Panel on Laboratory Animal Care.

The proteomic screen for endogenous neurexin ligands was previously described in Sterky et al. (2017). Briefly, brains from adult Nrxn1^HA/HA (Trotter et al., 2019) or WT (negative control) mice were extracted and flash-frozen in the Südhof laboratory at Stanford University and sent to the Fakler laboratory at the University of Freiburg, where plasma membrane–enriched fractions were prepared. After solubilization with CL-91 buffer (Logopharm), membranes were subjected to affinity purification (AP; detailed in Schwenk et al., 2014) using the following antibodies: (1) anti-HA (Roche, 11867423001, lot 10145700), (2) anti-HA (Thermo Fisher Scientific, 26183, lot OJ187900), (3) anti-Nrxn1 (Synaptic Systems, 175103), or (4) anti-Nrxn2 (Südhof laboratory). Eluates of APs were trypsinized and analyzed by HPLC-coupled mass spectrometry (MS) using an Orbitrap Elite mass spectrometer (Thermo Fisher Scientific; as described in Schwenk et al., 2014). Protein amounts were quantified by a label-free procedure using peptide peak signal intensities (peak volumes, PV), and molecular abundances were estimated using the abundance_{norm, spec} score, defined as the sum of all assigned and protein isoform-specific PVs divided by the number of MS-accessible protein isoform-specific amino acids (Bildl et al., 2012). The specificity of proteins copurified in APs was evaluated by calculating protein ratios (rPVs) from the respective peptide PVs (TopCorr method; Bildl et al., 2012). Specificity thresholds of APs were determined empirically from rPV histograms of all proteins detected in the respective AP versus control (samples from WT mice for anti-HA APs and from preimmunization IgGs for anti-Nrxn APs). All proteins with rPVs above threshold in both anti-HA APs and at least one anti-Nrxn AP were considered specifically copurified.

Unless otherwise noted, mouse protein coding sequences were used to generate all constructs. For transfections of HEK293T cells, constructs encoding FAM19A1-A5 or CA10 (Sterky et al., 2017) contained the endogenous signal peptide and were cloned into pEB-Multi-Neo (Fig. 1, B and C; Fig. 2; Fig. S2; Fig. S3, A–G; Fig. 4, A, C, and D; and Fig. S4; and Fig. 5). For transduction using BacMam viruses, the endogenous signal peptide was used for FAM19A1-V5 and FAM19A1-V5-6xHis (Fig. 3, A, B, G, and I; Fig. 4 B; and Fig. S3, H–J). For the BacMam virus encoding FAM19A1-Twin-Strep, the signal peptide was replaced with that of Igκ (Fig. 3, E, F, H, and I; and Fig. S3, K–O). For transfections of HEK293T cells, constructs encoding the WT or mutant Nrxn1β^SS4-SS5--ECD contained the Igκ signal peptide and were cloned into pEB-Multi-Neo (Fig. 1, B and C; Fig. 4, A, C, and D; Fig. 5, A–E; and Fig. S4 A). Similarly, BacMam virus encoding Igκ-Nrxn1β^SS4-SS5--ECD-ΔCHO-Myc-6xHis was used (Fig. 3, A–D, G, and I; Fig. 4 B; and Fig. S3 G; Sterky et al., 2017). For constructs encoding full-length neurexins with C-terminal EGFP tags, the endogenous signal peptides were used (Sterky et al., 2017; Fig. 2 and Fig. S2). For constructs in the HIV-1 flap, ubiquitin promoter, woodchuck hepatitis virus post-transcriptional regulatory element (FUW) vector encoding full-length WT or mutant Nrxn1β or Nrxn1γ with an HA tag downstream of the signal peptide, the endogenous signal peptide was used (Fig. 5, F and G; Fig. S3, E–F; and Fig. S4, B and C). GABA_AR-β3 was cloned into pCMV-Sport6 and contained a C-terminal HA tag, and the endogenous signal peptide was replaced with that of chicken LAR-type receptor phosphotyrosine-phosphatase σ (PTPRS; Miller and Aricescu, 2014; Fig. 5, F and G).

Hippocampal cultures containing neurons and glia (i.e., mixed) were generated from post-natal day (P) 0 Nrxn123 cKO mice (Fig. 6, Fig. 7, and Fig. 8; Chen et al., 2017) or CD-1 mice (Fig. 9). The hippocampi of pups were dissected and mixed, regardless of gender. Hippocampi were dissected in ice-cold HBSS and kept on ice until digestion with 0.2 µm-filtered 10 U/ml papain (Worthington Biochemical Corporation) in HBSS in a 37°C water bath for 20 min. Hippocampi were washed twice with HBSS and once with plating medium (MEM [Gibco] supplemented with 2 mM L-glutamine [Gibco], 0.4% glucose [Sigma-Aldrich], 2% Gem21 NeuroPlex Serum-Free Supplement [Gemini], and 5% FBS [Atlanta]). Cells were dissociated in plating media and plated onto autoclaved glass coverslips coated with Matrigel Membrane Matrix (Corning) in 24-well plates (one pup’s hippocampi/12 wells). This day was considered DIV0. On DIV1, 80% of the plating medium was replaced with fresh, prewarmed growth medium (Neurobasal-A [Gibco] supplemented with 2% Gem21 NeuroPlex Serum-Free Supplement and 2 mM L-glutamine [Sigma-Aldrich]). At DIV3–4, when glia were nearly confluent, 50% of the medium was replaced with fresh, prewarmed plating medium and a final concentration of 2 µM cytosine arabinoside. At DIV8, 50% of the medium was replaced with fresh, prewarmed plating medium. Analyses were performed at DIV14–16.

At DIV4, hippocampal cultures were infected with lentiviruses encoding Cre or ΔCre. At DIV6, hippocampal cultures were infected with control lentiviruses or lentiviruses encoding mouse FAM19A1-V5 (containing the endogenous FAM19A1 signal peptide).

At DIV14, CD-1 hippocampal cultures underwent a 50% conditioned media swap yielding a final concentration of 100 µM picrotoxin or vehicle (0.1% DMSO). RNA was collected 2, 6, 24, and 48 h after treatment with picrotoxin or vehicle. Prior to lysis for RNA extraction, cells were washed once using ice-cold Dulbecco's phosphate-buffered saline (DPBS; Gibco).

HEK293T cells (ATCC CRL11268) were grown in complete DMEM (cDMEM), which consisted of DMEM (Gibco), 5% FBS (Sigma-Aldrich), penicillin, and streptomycin. All transfections were performed using calcium phosphate, either with a CalPhos Mammalian Transfection Kit (Takara) or reagents generated in-house. For coIP experiments, HEK293T cells were plated onto 21-cm² dishes. For immunocytochemistry and cell surface binding experiments, HEK293T cells were plated onto poly-L-lysine (150–300 kD; Sigma-Aldrich)–coated glass coverslips in 24-well plates. Medium was replaced with fresh, prewarmed cDMEM 1 h before transfection. Unless otherwise noted, cells were washed once with prewarmed DPBS, and medium was replaced with fresh, prewarmed cDMEM ∼12 h after transfection. Samples were collected ∼48 h after transfection.

For qRT-PCR experiments, RNA concentration was measured using a NanoDrop 1000 Spectrophotometer (Thermo Fisher Scientific), and 8 ng total RNA was used per reaction. Transcripts were probed using PrimeTime qPCR Probe Assays (Integrated DNA Technologies), which consist of two primers and one FAM-labeled, ZEN/IBFQ-quenched 5′ nuclease probe. The following predesigned assays were used (gene, assay ID): Actb (Mm.PT.51.14022423), Arc (Mm.PT.58.5865502.g), Car10 (Mm.PT.58.11765793), Car11 (Mm.PT.58.32895602), Fam19a1 (Mm.PT.56a.6079538), Fam19a2 (Mm.PT.58.7298614), Fam19a4 (Mm.PT.56a.9330679), Fam19a5 (Mm.PT.56a.8916996), Fos (Mm.PT.58.29977214), and Gapdh (4352932E, Applied Biosystems). Additionally, the following assays were used (gene, primer 1, primer 2, probe): Nrxn1αβγ, 5′-CGA​TGT​CAT​CTG​TCC​CAA​CA-3′, 5′-GCC​ATC​GGA​TTT​AGC​ACT​GTC-3′, 5′-TGG​AGC​TGC​ACA​TAC​ACC​AAG​GAA-3′; Nrxn2αβ, 5′-TGA​TAT​TTG​CCG​TCG​CTC​AC-3′, 5′-GGT​GCG​AGT​GGA​CAG​TG-3′, 5′-TCA​ATT​GTA​ATG​TCG​TCC​GTG​CCC​A-3′; and Nrxn3αβ, 5′-CAC​TGA​TAA​TGA​ACG​CCT​CCA-3′, 5′-CCT​TTG​TCC​TTT​CCT​CCG​ATG-3′, 5′-CCT​TTT​TCC​TGC​AGC​CAC​TCC​TCT-3′. Assays generating threshold cycle (C_t) values >35 were omitted. C_t values for technical replicates differed by less than 0.5. C_t values were averaged for technical replicates.

For qRT-PCR from brain tissue, dissections and RNA extraction were performed as described in Seigneur and Südhof (2017). VeriQuest Probe One-Step qRT-PCR Master Mix with ROX (Affymetrix) and an ABI 7900HT Fast RT-PCR instrument (Applied Biosystems) were used. Samples were run in technical triplicates with Gapdh and Actb as loading controls. Data were normalized to the arithmetic mean of Gapdh and Actb, and mean 2^−ΔCt values are plotted.

For qRT-PCR from hippocampal cultures, total RNA was isolated using a RNeasy Mini Kit (Qiagen) per the manufacturer’s instructions. TaqMan Fast Virus 1-Step Master Mix (Thermo Fisher Scientific) and a QuantStudio 3 RT-PCR System (Thermo Fisher Scientific) were used, and samples were run in technical duplicates with Gapdh as a loading control. Additionally, data were normalized to the control group within each experimental replicate (vehicle 2 h post-treatment). Mean 2^−ΔΔCt values are plotted.

All lentiviral protein expression was driven by the human synapsin I promoter (Südhof, 1990). LV-SynI-Cre-EGFP and LV-SynI-ΔCre-EGFP were generated by the Janelia Viral Tools facility and added to Nrxn123 cKO hippocampal cultures at DIV4 (Kaeser et al., 2009). Control and mouse FAM19A1-V5 lentiviruses were generated as described in Trotter et al. (2019). Briefly, the medium of HEK293T cells was replaced with fresh cDMEM containing 25 µM chloroquine diphosphate (Sigma-Aldrich) 1 h before transfection. Cells were transfected with lentiviral long terminal repeat (LTR)-containing vectors encoding FAM19A1-V5 or control (empty vector) with helper plasmids pRSV-REV, pMDLg/gRRE, and pVSVG. 16–20 h later, cells were washed with DPBS, and the medium was replaced with neural culture growth media. 48 h after transfection, viral supernatant was collected, cleared by centrifugation at 1,500 g for 10 min, aliquoted, and stored at −80°C. Viral supernatant (25 µl/well) containing control or FAM19A1-V5 lentiviruses was added to Nrxn123 cKO hippocampal cultures on DIV6.

Experimenters were blind to the treatment groups during recordings and analysis.

For electrophysiological recordings of hippocampal cultures at DIV14–16, a coverslip was placed in a recording chamber mounted onto an Axioskop FS-2 upright microscope (Zeiss) equipped with differential interference contrast (DIC) and fluorescence filters, as well as a light-emitting diode (LED) source. Coverslips were maintained at room temperature (RT). Recordings were performed with a two-channel Axoclamp 700B amplifier (Axon Instruments) and digitalized with a Digidata 1440 digitizer (Molecular Devices) controlled by Clampex 10.7 (Molecular Devices). Patch pipettes (resistance of 2.5–3.5 MΩ) were pulled using borosilicate glass (WPI) on a two-stage vertical puller (Narishige). Neurons were voltage-clamped at −70 mV. The recording chamber was continuously perfused with oxygenated (95% O₂, 5% CO₂) standard artificial cerebrospinal fluid containing (in mM) 119 NaCl, 2.5 KCl, 1.3 MgCl₂, 2.5 CaCl₂, 11 glucose, 1 NaH₂PO₄, and 26 NaHCO₃. For mEPSC recordings, the following internal solution was used (in mM): 135 cesium methanesulfonate, 8 NaCl, 10 Hepes, 2 ATP-Mg, 0.3 GTP-Na, 0.1 Spermine, 7 phosphocreatine, and 0.3 EGTA, 300 mOsm/l, pH 7.3, adjusted with CsOH). The external solution contained 1 µM tetrodotoxin (TTX; American Radiolabeled Chemicals) and 100 µM picrotoxin (Tocris Bioscience). For mIPSC recordings, the internal solution consisted of (in mM) 135 CsCl, 10 Hepes, 1 EGTA, 4 ATP-Na, and 0.4 GTP-Na, 300 mOsm/l, pH 7.3, adjusted with CsOH). The external solution contained 1 µM TTX (American Radiolabeled Chemicals), 10 µM 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX; Tocris Bioscience), and 50 µM AP5 (Tocris).

Miniature currents were detected and analyzed offline in Clampfit 11.0.3 (Molecular Devices) using template matching. A lowpass Gaussian 1 kHz filter was used. The first minute of the recording was excluded from analysis. Events <8 pA were excluded. Cells for which the axis resistance >15 MΩ, or for which the mean amplitude and/or frequency were statistical outliers (mean ± 2× standard deviation), were removed. Both cumulative distributions and averages are shown. Specifically, the graphs in the inset of Fig. 6, E, F, K, and L, show the average of the amplitude and frequency of each sample and have been statistically compared using two-tailed unpaired t tests, as indicated in the figure legend. For mEPSC recordings, data in these graphs are means ± SEM of 33 and 37 cells recorded across six independent cultures; for mIPSC, data in these graphs are means ± SEM of 25 cells recorded across five independent cultures. For cumulative distributions, we plotted the amplitude and inter-event interval of single events across all recorded cells (180 events for all cells except for one cell, for which only 133 events could be scored). We used the standard approach of analyzing the same number of events for each cell in order to avoid biasing the analysis toward outliers. The cumulative distributions have been compared using the Kolmogorov–Smirnov test.

All samples were kept on ice until the addition of Laemmli sample buffer. Proteins were eluted from beads using 2× Laemmli sample buffer containing either 200 mM dithiothreitol (DTT) or 20 mM tris(2-carboxyethyl)phosphine hydrochloride (TCEP) and heating at 65°C for 10 min. Samples were analyzed by SDS-PAGE.

Approximately 6 h after transfection, cells were washed once with prewarmed DPBS, and the medium was replaced with 3 ml of prewarmed Freestyle 293 Expression Medium (Gibco). The medium was collected ∼48 h after transfection. Hepes (10 mM final concentration), EDTA (5 mM final concentration), and cOmplete, EDTA-free protease inhibitor (Roche) were added to each sample at collection. Samples were centrifuged at 1,500 g for 15 min at 4°C in a swingout, tabletop centrifuge to pellet floating cells, and the medium was transferred to a fresh tube. 200 µl of medium (input) was removed and stored at −80°C. 40 µl of 1:1 slurry of washed mouse anti-V5 agarose beads (Thermo Fisher Scientific, A7345) was added to each sample, and samples were rotated overnight at 4°C. Samples were centrifuged at 1,500 g for 7 min at 4°C, and the medium (flow through) was discarded. Beads were washed three or four times with ice-cold DPBS before elution.

Prior to collection, cells were washed once using ice-cold DPBS and collected in 500 µl of ice-cold DPBS containing cOmplete, EDTA-free protease inhibitors. Samples were centrifuged at 3,000 rpm for 5 min at 4°C. The DPBS was removed, and cell pellets were resuspended in 200 µl coIP buffer (100 mM NaCl, 4 mM KCl, 2 mM Ca₂Cl, 2 mM MgCl₂, and 20 mM Tris-HCl, pH 7.4) containing 1% Triton X-100 (Sigma-Aldrich) and cOmplete, EDTA-free protease inhibitors. Samples were incubated on ice for 1 h and gently vortexed every 15 min. Samples were then centrifuged at 14,000 rpm for 15 min at 4°C, and the clarified lysate was collected. A fraction (input) was saved and stored at −80°C. For each sample, 150 µl cleared lysate was mixed with 450 µl of coIP buffer and 30 µl of 1:1 slurry of washed mouse anti-V5 or anti-HA agarose beads (A2095, Millipore). Samples were rotated overnight at 4°C. Beads were washed five times with coIP buffer containing 0.1% Triton X-100 before elution.

Plates were placed on ice, and cells were washed once with ice-cold DPBS. Cells were fixed using 4% paraformaldehyde in DPBS for 20 min at RT, washed three times with DPBS, blocked (5% normal goat serum [Sigma-Aldrich] in DPBS) for 1 h at RT, and incubated with primary antibodies in blocking solution overnight at 4°C. Coverslips were washed three times with DPBS, incubated with secondary antibodies in blocking solution for 1 h at RT, and washed three times with DPBS. HEK293T cell–containing coverslips were mounted onto positively charged glass slides using Fluoromount G without DAPI (Southern Biotech). Hippocampal cultures underwent total labeling (see below).

Following surface labeling, hippocampal cultures were permeabilized in permeabilization/blocking buffer (0.2% Triton X-100 in blocking solution) for 1 h at RT, incubated with primary antibodies for total labeling in permeabilization/blocking buffer overnight at 4°C, washed three times with DPBS, incubated with secondary antibodies for total labeling in permeabilization/blocking buffer for 1 h at RT, washed three times with DPBS, and mounted onto positively charged glass slides using Fluoromount G without DAPI. The following primary antibodies were used: mouse anti-V5 (1:1,000; Invitrogen, R960-25, RRID: AB_2556564), chicken anti-microtubule-associated protein 2 (MAP2; 1:1,000; Aves Labs, MAP, RRID: AB_2313549), rabbit anti-vGluT1 (1:1,000; homemade, YZ6089), rabbit anti-vGAT (1:1,000; Millipore, AB5062P), and rabbit anti-HA (C29F4; 1:1,000; Cell Signaling, rabbit mAb 3724). The following Alexa Fluor Highly Cross-Adsorbed Secondary Antibodies (Invitrogen; raised in goat, IgG [Heavy + Light]) were used: anti-mouse 405 (1:500 or 1:1,000), anti-rabbit 405 (1:500 or 1:1,000), anti-mouse 546 (1:1,000), anti-rabbit 546 (1:1,000), and anti-chicken 647 (1:500 or 1:1,000).

Approximately 48 h after transfection, cells were washed three times with ice-cold Hepes bath buffer containing the following (in mM): 140–150 NaCl, 4–5 KCl, 2 CaCl₂, 1 MgCl₂, 10 glucose, and 10 Hepes, with the pH adjusted to 7.4 using NaOH [300 mOsm/l]). Next, cells were incubated in Hepes bath buffer with or without 100 nM FAM19A1-V5-His for 4 h at 4°C with gentle shaking. Cells were washed three times at 4°C with Hepes bath buffer, fixed, and surface labeled as described above.

All steps were performed on ice or at 4°C. Cells were washed twice with ice-cold DPBS containing 1 mM MgCl₂ and 0.1 mM CaCl₂ (DPBS+MC), then incubated with 1 mg/ml EZLink-Sulfo-NHS-SS-Biotin (in DPBS+MC; Thermo Fisher Scientific) at 4°C for 30 min while gently shaking and protected from light. Cells were washed twice with quenching solution (100 mM glycine in DPBS+MC) for 10 min per wash with gentle shaking, then twice with PBS+MC. Cells were lysed in RIPA buffer (150 mM NaCl, 5 mM EDTA, 1% Triton X-100, 0.1% SDS, 25 mM Tris-HCl, pH 7.6) containing cOmplete, EDTA-free protease inhibitors (cRIPA; 50 µl/well). Lysate was centrifuged at 14,000 rpm for 15 min at 4°C. Cleared lysate was collected and stored at −80°C. 100–110 µl cleared lysate was used for experiments shown in Fig. 7. The total volume for the pull-down was brought up to 200 µl using cRIPA, and 35 µl of prewashed Dynabeads MyOne Streptavidin C1 (Thermo Fisher Scientific) was added to each sample. Samples were rotated at 4°C for 3 h. Flowthrough was saved, and beads were washed three times with cRIPA, then eluted using 2X Laemmli sample buffer (containing 200 mM DTT or 10 mM TCEP) at 65°C for 10 min. All samples were analyzed by SDS-PAGE.

Cell surface proteins were purified as described above, except that equal amounts of total proteins were loaded across all groups within an experiment, the total volume for the pull-down was brought up to 300 µl, and 40 µl of prewashed Dynabeads MyOne Streptavidin C1 was added to each sample. Protein levels were measured using a BCA Protein Assay Kit (Pierce) per the manufacturer’s instructions.

After incubation of lysate with Dynabeads MyOne Streptavidin C1 beads, beads were washed twice with cRIPA, once with complete protein deglycosylation mix (cPDM) buffer (50 mM NaCl, 20 mM Tris-HCl, pH 7.5, and cOmplete, EDTA-free protease inhibitors) containing 1% Triton X-100, and once with cPDM buffer containing 0.2% Triton X-100. Deglycosylation Mix Buffer 2 (NEB), 10 mM TCEP, and cOmplete, EDTA-free protease inhibitors were added to the beads, and the samples were placed at 75°C for 10 min. Samples were briefly placed on ice. Protein Deglycosylation Mix II (NEB) was added, and samples were incubated at RT for 30 min. Next, samples were incubated at 37°C for 1 h with gentle shaking. Finally, Laemmli buffer (containing 200 mM DTT or 10 mM TCEP) was added, and samples were placed at 65°C for 10 min. The eluate was collected and stored at −80°C until evaluation by SDS-PAGE.

After incubation of lysate with Dynabeads MyOne Streptavidin C1 beads, beads were washed twice with cRIPA, once with Heparinase Reaction Buffer (NEB) containing 1% Triton X-100 and EDTA-free cOmplete protease inhibitors, and once with Heparinase Reaction Buffer containing 0.2% Triton X-100 and EDTA-free cOmplete protease inhibitors. Bacteroides eggerthii heparinases I, II, and III (1 µl of each, NEB) were added, in Heparinase Reaction Buffer containing 10 mM TCEP, 0.2% Triton X-100, and cOmplete, EDTA-free protease inhibitors, to each sample. Samples were incubated at 30°C for 2 h, with gentle shaking. Laemmli buffer (containing 200 mM DTT or 10 mM TCEP) was added to the samples, and they were placed at 65°C for 10 min. The eluate was collected and stored at −80°C until evaluation by SDS-PAGE.

4–20% Criterion TGX Precast Midi Protein gels (BioRad; 18-well, 30 µl) were used for all experiments except those shown in Fig. 8. For experiments shown in Fig. 8, 4–15% Criterion TGX Precast Midi Protein gels (BioRad; 12+2 well, 45 µl) were used. Proteins were transferred to a 0.2-µm-pore nitrocellulose membrane using a Criterion Blotter (BioRad) with plate electrodes, with transfer settings of 400 mV for 1.5 h at 4°C. The transfer buffer was composed of ice-cold Tris-glycine with 20% methanol. Membranes were blocked with 5% nonfat milk (Carnation) in Tris-buffered saline with 0.1% Tween-20 (TBST) for 1 h at RT, then incubated in primary antibody in blocking solution overnight (unless otherwise noted). Membranes were washed three times with TBST and incubated with secondary antibodies in blocking solution for 1 h at RT. Membranes were washed three times and scanned using an Odyssey CLx Imager (Li-COR) in automatic mode for dynamic range detection. When multiple antibodies were used, staining and scanning were done sequentially. Immunoblots were analyzed using free Image Studio Lite Version 5.2 (Li-COR). Channels were pseudo-colored using Adobe Photoshop CS (64-bit). Normalization was performed as described in the figure legends.

The following primary antibodies were used for immunoblotting: chicken anti-c-myc (1:500; Aves Labs, ET-MY100, RRID: AB_2313514), mouse anti-V5 (1:1,000; Invitrogen, R960-25, RRID: AB_2556564), rabbit anti-V5 (1:1,000; Millipore, AB3792), rabbit anti-Strep-tag II (1:1,000; Abcam, ab76949), rabbit anti-myc (clone 71D10; 1:1,000; Cell Signaling, rabbit mAb 2278), mouse anti-HA.11 (clone 16B12; 1:1,000; BioLegend, MMS-101R, RRID: AB_256533), mouse anti-GAPDH (clone 6C5; 1 h incubation at RT; 1:5,000; Millipore, MAB374), rabbit anti-pan-Nrxn (1–2 overnight incubation at 4°C; 1:1,000; Millipore, ABN161-I), mouse anti-GABA_AR β2/3 (clone 62-3G1; 1:500; NeuroMab, 75–363, RRID: AB_2315837), mouse anti-Δ-HS (3G10 epitope, clone F69-3G10; 1:1,000; Amsbio, 370260-S), and rabbit anti-synapsin (1:1,000; homemade, YZ6079). The following secondary antibodies (Li-COR) were used (1:10,000 in blocking solution): IRDye 800CW donkey anti-mouse (926–32212; RRID: AB_2716622), IRDye 680LT donkey anti-mouse (926–68022; RRID: AB_2814906), IRDye 800CW donkey anti-rabbit (926–32213; RRID: AB_621848), and IRDye 680LT donkey anti-rabbit (926–68023; RRID: AB_10706167).

Specific bands in pan-neurexin immunoblots were identified by their disappearance after neurexin deletion in Nrxn123 cKO hippocampal cultures infected with lentiviral Cre. Non-specific bands were identified based on their persistence, without a qualitative change in band intensity, after neurexin deletion in Nrxn123 cKO hippocampal cultures infected with lentiviral Cre.

400 ml of Expi293F cells (Gibco) at a density of 3 × 10⁶ viable cells/ml were transfected using a ExpiFectamine 293 Transfection Kit (Gibco) per the Expi293 Expression System user guide instructions. 6 d later, one cOmplete, EDTA-free protease inhibitor tablet was added to the culture, and it was harvested. The cell culture suspension was centrifuged at 4,000 g for 30 min to pellet down cells. The medium was collected and the pH was adjusted to 8.0 using 20 mM (final concentration) Tris-HCl (pH 8.0). Imidazole was added to a final concentration of 10 mM. 5 ml of prewashed 1:1 HisPur Ni NTA resin (Thermo Fisher Scientific) in equilibration buffer consisting of 20 mM Tris-HCl, 300 mM NaCl (20/300 TBS), and 20 mM imidazole was added to the media. Batch binding was allowed to occur for 2 h at RT, with gentle stirring. The resin was collected and washed with ∼25 ml of 20/300 TBS containing 25 mM imidazole. Recombinant protein was eluted in 20 ml of 20/300 TBS containing 200 mM imidazole. The eluate was concentrated to 1 ml using a prewet centrifugal filtration unit containing a 3-kD molecular weight cut-off membrane (Amplicon). The sample was injected onto a Superdex 75 Increase 10/300 GL column (GE) connected to an NGC Quest 10 Chromatography System (BioRad) and eluted over 1 column volume (CV) using 20/300 TBS. Peak fractions were analyzed by Coomassie staining of non-reducing and reducing SDS-gels. The peak fraction corresponding to the FAM19A1-V5-His monomer was used for cell surface binding.

For recombinant protein expression for SEC-MALS, the BacMam system (Dukkipati et al., 2008; Goehring et al., 2014; Wilson et al., 2019) was used. Constructs were cloned into the pEG BacMam vector and transformed into MAX Efficiency DH10Bac Competent Escherichia coli (Invitrogen) to generate recombinant bacmid DNA. SF9 cells were transfected to generate BacMam viruses, which were then amplified. BacMam viruses were used to transduce HEK293S GnTI⁻ cells (ATCC CRL-3022) or, in the case of FAM19A1-V5-6xHis, FreeStyle 293-F cells (Gibco; grown per the manufacturer’s instructions) were used for protein expression. HEK293S GnTI⁻ cells were grown in suspension in FreeStyle 293 Expression Medium (Gibco) containing 2% FBS at 37°C and with 8% CO₂. When the cell density reached 1–2 × 10⁶ cells/ml, 1–2 liters of cell cultures were (1) cotransduced with BacMam viruses encoding Igκ signal peptide-Nrxn1β^SS4-SS5--ECD-Myc-6xHis and FAM19A1-V5 containing the endogenous signal peptide, or transduced with BacMam viruses encoding (2) Igκ signal peptide-Nrxn1β^SS4-SS5--ECD-Myc-6xHis, (3) Igκ signal peptide-FAM19A1-Twin-Strep, or (4) FAM19A1-V5-His containing the endogenous signal peptide. For the purification of the Nrxn1β-ECD-Myc-6xHis/FAM19A1-V5 complex, 1:2 volumes of supernatants containing the BacMam viruses were used for transduction. 10 mM sodium butyrate was added to the cultures 8–24 h after transduction, and the incubation temperature was lowered to 30°C. Approximately 72–96 h after transduction, cOmplete, EDTA-free protease inhibitors were added to the cell culture, cells were pelleted by centrifugation, and the medium was collected. At RT, medium was concentrated, and the buffer was exchanged into 20/150 TBS using a 10-kD membrane (Omega Centramate Pall T-series). For the purification of 6xHis-tagged proteins or protein complexes, this buffer also contained 10 mM imidazole. The resulting protein-containing solution was loaded onto 5 ml His Trap HP(GE) columns using an ÄKTA Start (GE) at 4°C. Columns were washed with a solution of 20/300 TBS containing 25 mM imidazole. 6xHis-tagged proteins/protein complexes were eluted using a 50–500-mM imidazole gradient (in 20/300 TBS) over 10 CV. Peak fractions were evaluated by Coomassie staining of reducing and non-reducing SDS-gels, and fractions of interest were combined and concentrated using a 3- or 10-kD-molecular-weight cut-off centrifugal filtration unit. Further purification steps were performed as described below, with columns attached to an ÄKTA pure purification system (GE). For the purification of Nrxn1β-ECD-Myc-6xHis, the immobilized metal affinity chromatography (IMAC)-purified protein solution was loaded onto a HiLoad 16/600 Superdex 200 PG column and eluted over 1 CV using 20/150 TBS. For the purification of the Nrxn1β-ECD-Myc-6xHis/FAM19A1-V5 complex, fractions of interest from IMAC were combined and dialyzed in 20/50 TBS. The resulting sample was loaded onto a Mono Q 4.6/100 OE column (GE), and the proteins of interest were eluted using a 30 CV gradient of 20/50 to 20/500 TBS. Fractions of interest were pooled and loaded onto a Superdex 200 10/300 GL column and eluted using 20/150 TBS. Fractions of interest were pooled, the sample was diluted such that the concentration of NaCl was 50 mM, and the sample was again loaded onto a Mono Q 4.6/100 OE column (GE). Proteins were eluted with a 40 CV gradient of 20/50 to 20/500 TBS. For the purification of FAM19A1-V5-6xHis, fractions of interest were pooled, loaded onto a Superdex 75 10/300 GL column, and eluted over 1.25 CV using 20/150 TBS. For the purification of FAM19A1-Twin-Strep, the concentrated, buffer-exchanged protein-containing solution was loaded onto StrepTrap HP columns (GE), washed with 20/150 TBS, and batch-eluted using 2.5 mM desthiobiotin (Sigma-Aldrich) in 20/150 TBS. The eluate was concentrated and loaded onto either a Superdex 75 10/300 GL column (GE) or a HiLoad Superdex 75 PG column for small- and large-scale purification, respectively. Protein concentration measurements were performed using a NanoDrop 1000 spectrophotometer (Thermo Fisher Scientific), with the expected molecular weights and extinction coefficients taken into account.

For experiments shown in Fig. 4 A, Igκ-Nrxn1β-ECD-Myc-6xHis and FAM19A1-V5 were coexpressed in FreeStyle 293-F cells transfected using polyethylenimine. The recombinant Nrxn1β-ECD/FAM19A1 complex was purified from the medium by batch elution of bound Nrxn1β-ECD from a TALON column, followed by SEC on a Superdex 200 column.

200 μl of ∼1 mg/ml purified recombinant protein was used for SEC-MALS. For the Nrxn1β-ECD/FAM19A1 complex and Nrxn1β-ECD, a Superdex 200 10/300 GL column (GE) was used. For FAM19A1-Twin-Strep, a Superdex 75 10/300 GL column (GE) was used. The flow rate was 0.5 ml/min in PBS, pH 7.4 (Sigma-Aldrich). Using Dawn Heleos-II and Optilab rEX instruments (Wyatt Technology), UV absorption at 280 nm, light scattering, and differential refractometry data were recorded for protein elution profiles. A bovine serum albumin reference was used to correct the baseline. Data analyses were performed using ASTRA 7.1.2 (Wyatt Technology). A differential refractive index increment (dn/dc) value of 0.185 was used for all calculations.

Absolute mass determination was performed by the Stanford University Mass Spectrometry core. Samples were analyzed, as provided, by electrospray ionization mass spectrometry on the Agilent 1260 HPLC and Bruker MicroTOF-Q II. Solvent A consisted of 0.05% trifluoroacetic acid (TFA) and 0.09% formic acid in water, and Solvent B consisted of 0.1% formic acid in acetonitrile. The gradient was as follows (time in minutes, Solvent A, Solvent B): 0–1.5, 95%, 5%; 2, 65%, 35%; 10, 54%, 46%; 11–12, 5%, 95%; 13, 95%, 5%. The column was an Agilent Pursuit 5 diphenyl 150 × 2.1 mm, the temperature was 80°C, and the flow rate was 0.3 ml/min. The volume injected was 5 µl. Data were collected in full-scan MS mode with a mass range of 400–4,000 D. The collision radio frequency (RF) setting was 800 Vpp.

The experimenter was blind to the treatment groups during image acquisition and analysis.

Confocal images were acquired at RT using an inverted Nikon A1Rsi Eclipse Ti confocal microscope (Nikon) operated by NIS-Elements AR acquisition software (Nikon). All images were 1,024 × 1,024 pixels. When multiple channels were used, images in each channel were acquired sequentially to avoid bleed-through. The same imaging settings were used throughout for all groups within an experimental replicate and were optimized to prevent pixel saturation. All images for a given experimental replicate were acquired on the same day with the same acquisition settings. Positive and negative control groups were included within each experimental replicate.

Confocal images of HEK293T cells were acquired using a 20× objective, 3.46× digital zoom, and a z-stack distance of 2 µm. Fields of view were chosen using the EGFP signal only.

Confocal images of hippocampal cultures were acquired using a 60× oil objective (Apo, NA 1.4), 1.82× digital zoom, and a z-stack distance of 0.5 µm. Neurons were selected based on health (i.e., MAP2 integrity) based on the MAP2 signal alone.

Confocal image analysis was performed using NIS-Elements Advanced Research software (Nikon). Maximum intensity projections were generated and analyzed, and are depicted in representative images. All images acquired within the same day underwent equal background subtraction before analysis. For synapse analysis, 20–40 µm lengths of secondary or tertiary dendrites (no intersecting dendrites) were quantified per neuron, with 9–13 neurons analyzed per experiment (n = 4 independent cultures). These regions of interest were drawn solely based on MAP2 staining. Excitatory and inhibitory synapses were defined as vGluT1- or vGAT-positive puncta associated with MAP2, respectively, using the General Analysis function. Puncta density, area, and mean intensity were averaged per independent culture before statistical analysis. Mean intensity values were normalized to the control group within each independent culture.

Statistical analyses were performed using GraphPad Prism 8 software as described in the figure legends.

Fig. S1 shows expression of Fam19a mRNAs in different brain regions as a function of development and cell type in mouse brain. Fig. S2 shows that FAM19A1-A4, but not FAM19A5, binds to all neurexin isoforms and splice variants tested as assessed by the neurexin-mediated exposure of FAM19A proteins on the surface of HEK293T cells coexpressing the various FAM19A proteins and neurexins. Fig. S3 shows that exogenously added FAM19A1 protein binds weakly to surface β-neurexin, in contrast to coexpressed FAM19A1 protein, which binds stoichiometrically, and purified recombinant FAM19A1 exists as a monomer as well as disulfide-mediated dimers. Fig. S4 shows that FAM19A1 binding to Nrxn1β is dependent on the cysteine residues in the neurexin Cys-loop domain, but the Cys-loop domain is insufficient for FAM19A1 binding since FAM19A1 does not bind to Nrxn1γ, which contains the Cys-loop domain, whereas CA10 does bind to Nrxn1γ. Fig. S5 shows that the deletion of neurexins decreases levels of surface-exposed FAM19A1 localized adjacent to dendrites but has no effect on the intrinsic electrical properties of neurons. It shows quantification of Nrxn123 deletion in Nrxn123 cKO mouse hippocampal cultures used in cell surface biotinylation experiments and assessment of the HS modification of neurexins using HS removal by treatment with heparinases.

We thank S.C. Wilson, R.A. Pfuetzner, J.H. Trotter, E.M. Seigneur, Z. Liu, K.J. Gan, and K.A. Raju (Stanford University School of Medicine, Stanford, CA) for advice and reagents.

This study was supported by grants from the National Institutes of Health (MH052804 to T.C. Südhof and R37MH63105 to A.T. Brunger) and the Deutsche Forschungsgemeinschaft (EXC 2189, 390939984 to B. Fakler). A.J. Khalaj was supported by a National Science Foundation Graduate Research Fellowship (DGE-114747).

The authors declare no competing financial interests.

Author contributions: A.J. Khalaj, F.H. Sterky, B. Fakler, and T.C. Südhof designed the experiments. A.J. Khalaj performed all experiments except for the proteomic screen, which was performed by F.H. Sterky and J. Schwenk, and the electrophysiological recordings, which were performed by A. Sclip. All authors analyzed the data, and A.J. Khalaj and T.C. Südhof wrote the paper with input from all authors.