SorCS1-mediated sorting in dendrites maintains neurexin axonal surface polarization required for synaptic function

The pre- and postsynaptic membranes comprising the synaptic junction differ in protein composition. The membrane trafficking mechanisms by which neurons control surface polarization of synaptic receptors remain poorly understood. The sorting receptor Sortilin-related CNS expressed 1 (SorCS1) is a critical regulator of trafficking of neuronal receptors, including the presynaptic adhesion molecule neurexin (Nrxn), an essential synaptic organizer. Here, we show that SorCS1 maintains a balance between axonal and dendritic Nrxn surface levels in the same neuron. Newly synthesized Nrxn1α traffics to the dendritic surface, where it is endocytosed. Endosomal SorCS1 interacts with the Rab11 GTPase effector Rab11 family-interacting protein 5 (Rab11FIP5)/Rab11 interacting protein (Rip11) to facilitate the transition of internalized Nrxn1α from early to recycling endosomes and bias Nrxn1α surface polarization towards the axon. In the absence of SorCS1, Nrxn1α accumulates in early endosomes and mispolarizes to the dendritic surface, impairing presynaptic differentiation and function. Thus, SorCS1-mediated sorting in dendritic endosomes controls Nrxn axonal surface polarization required for proper synapse development and function.


Introduction
Neurons are highly compartmentalized cells that need to maintain distinct membrane identities. This is especially clear at the synapse, where pre-and postsynaptic membranes differ dramatically in their protein composition [1]. Membrane trafficking mechanisms that organize and maintain the polarized distribution of receptor proteins during the formation, maturation, and plasticity of synapses are of key importance for the proper function of neural circuits [2,3].
Neurexins (Nrxns), expressed from 3 genes as α-, β-, and γ-Nrxns, are essential presynaptic adhesion molecules that engage in a network of interactions with multiple pre-and postsynaptic extracellular ligands [4] and act in a cell-type-specific manner to regulate synapse number, function, and plasticity [5][6][7][8][9]. Mutations  neuropsychiatric disorders, especially schizophrenia and autism [10][11][12], underscoring the importance of Nrxns for circuit function. Despite their essential role, the molecular mechanisms that control presynaptic polarization and abundance of Nrxns are largely unknown [13]. Previous studies have shown that surface trafficking of Nrxns requires the C-terminal PSD-95/discs large/zona occludens-1 (PDZ)-binding motif [14,15] and that surface mobility depends on synaptic activity and interaction with extracellular ligands [16,17]. The vertebrate Sortilin-related CNS expressed 1-3 (SorCS1-3) proteins are sorting receptors belonging to the vacuolar protein sorting 10 (VPS10P) family receptors that have emerged as major regulators of intracellular protein trafficking [18,19]. SorCS proteins are required for the surface expression of glutamate receptors, neurotrophin receptors, transporters, and adhesion molecules [20][21][22][23][24]. These sorting receptors are thus essential for synaptic transmission and plasticity, but the mechanisms by which SorCS proteins sort their cargo are poorly understood. Using quantitative proteomics analysis, we recently established a critical role for SorCS1 in Nrxn surface trafficking [20]. Although our previous observations suggested that SorCS1 might regulate endosomal trafficking of Nrxn, the intracellular pathway via which Nrxn is trafficked to synapses, the regulation thereof by SorCS1, and the relevance of SorCS1-mediated sorting of Nrxn and other axonal cargo proteins for synaptic development and function remained elusive.
Here, we use epitope-tagging to elucidate the subcellular localization of endogenous SorCS1 and Nrxn1α. We show that SorCS1 acts in dendritic endosomes to control a balance between the axonal and dendritic surface distribution of Nrxn1α in the same neuron. Although predominantly considered to be a presynaptic molecule, we find that newly synthesized Nrxn1α first traffics to dendrites, where it is endocytosed, followed by transcytosis to the axon. SorCS1 controls Nrxn axonal surface polarization by facilitating the transition of endocytosed Nrxn from early to recycling endosomes and does so by interacting with Rab11 family-interacting protein 5 (Rab11FIP5)/Rab11 interacting protein (Rip11), a novel SorCS1 cytoplasmic binding partner we identify. In the absence of SorCS1, Nrxn1α accumulates in early endosomes and mispolarizes to the dendritic surface, impairing presynaptic differentiation induced by the postsynaptic Nrxn ligand Neuroligin (Nlgn). This defect can be rescued by a Nrxn1α mutant that bypasses SorCS1-mediated sorting and transcytosis to polarize to the axonal surface, but not by wild type (WT) Nrxn1α. Finally, we show that SorCS1 is required for presynaptic function. Together, our observations indicate that SorCS1-mediated sorting in dendritic endosomes controls Nrxn axonal surface polarization required for proper synapse development and function.

SorCS1 controls an axonal-dendritic balance in Nrxn1α surface polarization
We first determined the subcellular distribution of endogenous SorCS1, which has been difficult to assess because of a lack of suitable antibodies. We generated a SorCS1 knock-in (KI) mouse (Sorcs1 HA ), using CRISPR/Cas9 to insert a hemagglutinin (HA) epitope tag in the Sorcs1 locus after the propeptide domain (S1A Fig). Correct insertion of the HA tag was confirmed by sequencing and western blot analysis (S1B Fig). HA immunodetection in cultured Sorcs1 HA cortical neurons revealed prominent punctate endogenous SorCS1 immunoreactivity in soma and dendrites (Fig 1A), which was mimicked by exogenously expressed HA-SorCS1 in hippocampal neurons (Fig 1B and 1C). These results are in agreement with a previous immunohistochemical study showing a somatodendritic distribution of SorCS1 [25] and suggest that SorCS1-mediated sorting of cargo proteins, including Nrxn, occurs in this cellular compartment.
To begin to elucidate how SorCS1 regulates membrane trafficking of its cargo Nrxn, we electroporated cortical Sorcs1 flox/flox mouse neurons with Cre recombinase (Cre) to remove SorCS1 (Sorcs1 knock out [KO]). We transfected neurons with HA-Nrxn1α, one of the most abundant Nrxn isoforms in the brain [26], and live-labeled them with an HA antibody to detect surface Nrxn1α, followed by Ankyrin-G (AnkG) and microtubule-associated protein 2 (MAP2) immunodetection to label the axonal and dendritic compartment, respectively. Loss of SorCS1 decreased Nrxn1α axonal surface levels and concomitantly increased dendritic surface levels in the same cell, changing Nrxn1α surface polarization from axonal to dendritic (Fig 1D and 1E). Re-expression of WT SorCS1 (SorCS1 WT ) in Cre-positive Sorcs1 flox/flox neurons restored Nrxn1α axonal surface polarization (Fig 1D and 1E), indicating that the defect occurred cell autonomously. In contrast, rescue with an endocytosis-defective mutant [27] of SorCS1 (SorCS1 Y1132A ), which remains on the plasma membrane (S1C Fig and S1D Fig), did not rescue Nrxn1α surface polarization (Fig 1D and 1E). To determine whether SorCS1 controls the subcellular distribution of endogenous Nrxn, we immunostained Sorcs1 flox/flox neurons electroporated with Cre or enhanced green fluorescent protein (EGFP) with a pan-Nrxn antibody directed against the conserved cytoplasmic tail [28] (S1E Fig and S1F Fig), which specifically labels endogenous Nrxn (S1G Fig and S1H Fig). Similar to the observations with transfected HA-Nrxn1α in Sorcs1 KO neurons, loss of SorCS1 reduced axonal intensity of endogenous Nrxn and decreased the axonal-dendritic ratio compared with control neurons (S1E Fig and S1F Fig). Conversely, overexpression of SorCS1 WT in hippocampal neurons, which express very low levels of Sorcs1 [29], increased Nrxn1α axonal surface polarization and concurrently decreased dendritic surface levels, whereas overexpression of SorCS1 Y1132A did not alter Nrxn1α surface polarization (S1I Fig and S1J Fig). Together, these results indicate that somatodendritic SorCS1 controls an axonal-dendritic surface balance of Nrxn1α. Our findings suggest that SorCS1 is either required for endocytosis of Nrxn1α or acts in endosomes to bias Nrxn1α surface polarization toward the axon.

Indirect axonal surface trafficking of Nrxn1α in mature neurons
Nrxns are considered to be predominantly presynaptic proteins, although several independent studies have reported the presence of a dendritic pool of Nrxns [20,[30][31][32][33]. To unequivocally determine the subcellular localization of endogenous Nrxn, we generated a Nrxn1α KI mouse (Nrxn1α HA ), inserting an HA epitope tag in the Nrxn1α locus after the signal peptide (S2A Fig). Correct insertion of the HA tag was confirmed by restriction digest, sequencing, and western blot analysis (S2B Fig). We cultured Nrxn1α HA cortical neurons together with WT cortical neurons to reliably detect endogenous HA-Nrxn1α in the somatodendritic and axonal compartment ( S2C Fig and S2D Fig). At 3 days in vitro (DIV), HA-Nrxn1α surface distribution was polarized towards the axon (Fig 2A and 2B). As neurons matured, surface HA-Nrxn1α axonal surface levels decreased, whereas dendritic surface levels increased (Fig 2A  and 2B). Surface HA-Nrxn1α remained polarized towards the axon at all developmental time points analyzed (Fig 2A and 2B). A similarly polarized distribution of Nrxn was observed in WT neurons labeled with the pan-Nrxn antibody ( S2E Fig and S2F Fig). Together, these results indicate that endogenous Nrxn1α is an axonally polarized surface protein that accumulates in dendrites as neurons mature.
To determine how Nrxn1α is trafficked in cortical neurons, we employed the retention using selective hooks (RUSH) system [34] to retain Nrxn1α in the endoplasmic reticulum (ER) and induced synchronous release and transport through the secretory pathway by Red arrowheads indicate the axon, and the blue asterisk marks the cell body. (B) Quantification of panel A: surface HA-Nrxn1α fluorescence intensity in axon and dendrites normalized to DIV3 neurons and the ratio of axonal-dendritic surface HA-Nrxn1α intensity. DIV3 (n = 30 neurons); DIV7 (n = 29); DIV11 (n = 24). �� P < 0.01; ��� P < 0.001 (Kruskal-Wallis test followed by Dunn's multiple comparisons test, 3 independent cultures). (C) Schematic representation of streptavidin-KDEL (ER hook) and reporters (Nrxn1α and TfR) used in RUSH experiments. Addition of biotin dissociates the reporters from the ER hook, inducing synchronous release from the ER and transport through the secretory pathway. (D) DIV9 WT mouse cortical neuron co-expressing SBP-EGFP-Nrxn1α and streptavidin-KDEL immunostained for MAP2 (blue), Ankyrin-G (red), and EGFP-Nrxn1α (grayscale and green) at t = 0 before adding biotin. (E) Live-cell imaging in DIV8 to DIV10 WT rat cortical neurons co-expressing SBP-EGFP-Nrxn1α and ER hook. After 24 to 31 h of expression, neurons were imaged every 5 min for 2.5 h. Biotin was added 10 min after the beginning of the imaging session. Shown are representative images of SBP-EGFP-Nrxn1α fluorescence in dendrites and axon before (t0) and 30, 60, 90 and 120 min after adding biotin. Red arrowheads indicate axon and black arrows indicate SBP-EGFP-Nrxn1α-positive puncta. See also S1 Movie. (F-I) Live-cell imaging in DIV8 to DIV10 WT rat cortical neurons co-expressing SBP-EGFP-Nrxn1α and ER hook. After 21 to 30 against the axon initial segment (AIS) protein Neurofascin to visualize the axonal compartment (S3A Fig). At t = 0, we observed diffuse SBP-EGFP-Nrxn1α fluorescence in soma and dendrites, in a pattern resembling the ER (Fig 2D). Following application of biotin, Nrxn1α fluorescence coalesced into punctate structures that started trafficking abundantly in the somatodendritic compartment, followed by Nrxn1α trafficking in the axon after a marked delay We performed several controls to verify that the delayed axonal trafficking of Nrxn1α is not an artefact of prolonged retention in the ER and synchronized release, which might overwhelm the trafficking machinery. Trafficking of the somatodendritic cargo protein transferrin receptor (TfR) [35] was detected only in the somatodendritic compartment ( S3E Fig and S3F Fig, S4 Movie). In the absence of biotin, dendritic SBP-EGFP-Nrxn1α-positive structures were not motile (S5 Movie), indicating that the delayed trafficking of Nrxn1α to axons is not simply caused by prolonged ER retention. Moreover, immature DIV3 neurons displayed Nrxn1α trafficking to the axonal compartment shortly after biotin application (S3G Fig and S3H Fig, S6 Movie), further indicating that the delayed axonal trafficking of Nrxn1α in mature neurons does not result from prolonged retention in the ER but correlates with the maturation state of the neuron. Taken together, these results show that following transport through the secretory pathway, Nrxn1α is first trafficked to the somatodendritic compartment and appears in axons with a marked delay.

Nrxn1α is transcytosed from the dendritic surface to axons
Our observations suggest a dendrite-to-axon transcytotic pathway for axonal surface polarization of Nrxn1α. The transcytotic model predicts that Nrxn1α is first trafficked to the somatodendritic plasma membrane, where it is internalized into endosomes and trafficked to the axonal compartment via endosome-derived carriers [36]. To determine whether Nrxn1α is internalized from the dendritic plasma membrane, we followed the fate of Nrxn1α throughout the endosomal pathway. Internalised Nrxn1α was largely restricted to the dendritic compartment ( S4A Fig and S4B Fig)  Nrxn1α axonal surface levels and a concomitant increase in dendritic surface levels. Expression of GTP binding-deficient mutants of Rab5 (S34N) and Rab11 (S25N) to interfere with transport to EEs and REs, respectively, mimicked the effect of blocking endocytosis (S5E- S5H  Fig). However, expression of Rab7 (T22N) to interfere with transport to late endosomes did not affect the axonal surface polarization of Nrxn1α ( S5I Fig and S5J Fig). Thus, endocytosis, EE and RE transport, but not late endosomal transport, are required for accumulation of Nrxn1α on the axonal plasma membrane, indicating transcytotic trafficking of Nrxn1α.

A SorCS1-Rab11FIP5 interaction controls Nrxn1α transition from early to REs
The transcytotic trafficking route of Nrxn1α and somatodendritic distribution of SorCS1 suggest that sorting of Nrxn1α occurs in dendrites. To determine how SorCS1 controls transcytotic trafficking of Nrxn1α, we followed the fate of internalized HA-Nrxn1α in Sorcs1 KO neurons. Endocytosis of HA-Nrxn1α in dendrites was not affected in Sorcs1 KO neurons compared with control cells (S6A Fig and S6B Fig). We next analyzed colocalization of internalized Nrxn1α with EEs (EAA1), fast REs (Rab4), REs (Rab11) in dendrites, and REs (Rab11) in the axon, respectively, of Sorcs1 KO neurons. Quantification revealed an increase in the colocalization of internalized Nrxn1α with EAA1-positive EEs and Rab4-positive fast REs in dendrites of Sorcs1 KO neurons (Fig 3A-3D). Colocalization of internalized Nrxn1α with Rab11-positive REs in Sorcs1 KO neurons was decreased in dendrites (Fig 3E and 3F) and even more strongly reduced in the axon (Fig 3G and 3H). Thus, endocytosed Nrxn1α accumulates in EEs and is missorted to Rab4-fast REs in the absence of SorCS1-mediated sorting. These observations are consistent with our finding that dendritic surface levels of Nrxn1α are increased in Sorcs1 KO neurons (Fig 1D and 1E), which is likely due to increased recycling of Nrxn1α via Rab4-fast REs back to the dendritic plasma membrane. Similarly, the reduced axonal surface levels of Nrxn1α in Sorcs1 KO neurons (Fig 1D and 1E) likely result from decreased sorting of Nrxn1α to Rab11-REs. Together, these results demonstrate that SorCS1 controls axonal surface polarization of Nrxn1α by facilitating the transition from EEs to Rab11-REs.
We reasoned that SorCS1 might interact with additional proteins to facilitate Nrxn1α sorting from early to REs. Rab11FIP5 or Rip11 is prominently present in the raw mass spectrometry (MS) data set we obtained after affinity purification (AP) of SorCS1 complexes from rat brain extracts with 2 independent antibodies [20] (AP-MS data available online). Rip11, which belongs to the family of Rab11-interacting proteins [37], localizes to REs and regulates transcytosis of proteins from the basolateral to the apical plasma membrane in polarized epithelial cells [38,39]. Western blot analysis of immunoprecipitated HA-SorCS1 from postnatal Sorcs1 HA KI cortical extracts showed a robust Rip11 band coprecipitating with SorCS1 but not with control mouse immunoglobulin G (IgG) (Fig 3I). Rip11 displayed a punctate distribution in dendrites and colocalized with SorCS1 (Fig 3J and 3K). Expression of a DN form of Rip11 that inhibits the transport from early to REs [40] reduced axonal surface levels of HA-Nrxn1α and increased dendritic surface levels, shifting Nrxn1α surface polarization from axonal to dendritic (Fig 3L  and 3M), similar to loss of SorCS1. Expression of WT Rip11, on the other hand, increased axonal surface polarization of HA-Nrxn1α (Fig 3L and 3M). Together, these results demonstrate that a SorCS1-Rip11 interaction facilitates Nrxn1α sorting from EEs to REs, preventing missorting to fast REs and late endosomes and biasing trafficking to the axonal surface.

Selective missorting of transcytotic cargo in the absence of SorCS1
We next asked whether the surface polarization of other axonal membrane proteins was affected in the absence of SorCS1. The cell adhesion molecules neuron-glia cell adhesion molecule L1 (L1/NgCAM) and contactin-associated protein-like 2 (Caspr2) are both targeted to the axonal surface but via distinct trafficking routes. L1/NgCAM is transcytosed from dendrites to axon [41,42], whereas Caspr2 undergoes nonpolarized delivery followed by selective endocytosis from the somatodendritic surface resulting in axonal polarization [43]. L1/ NgCAM levels are down-regulated in Sorcs1 KO synaptosomes [20]. Sorcs1 flox/flox cortical neurons electroporated with Cre or EGFP were transfected with myc-L1 or HA-Caspr2 and immunostained for surface myc or HA. Consistent with a role for SorCS1 in regulating transcytosed axonal cargo, surface polarization of L1/NgCAM, but not of Caspr2, was perturbed in

Loss of SorCS1 impairs Nrxn-mediated presynaptic differentiation
We next sought to determine motifs in the Nrxn1α protein sequence that contain signals necessary for its membrane trafficking, with the goal of identifying a Nrxn1α mutant that can bypass SorCS1-mediated sorting and transcytotic trafficking. We systematically tested the effect of a series of Nrxn1α cytoplasmic deletions on surface polarization (Fig 4A and 4B, S8A-S8C Fig) and found that removing the 4.1-binding motif in the cytoplasmic domain (Nrxn1α Δ4.1) was the only deletion that dramatically increased Nrxn1α axonal surface polarization ( Fig 4A and 4B, S8A-S8C Fig). Remarkably, axonal surface polarization of Nrxn1α Δ4.1 was unaffected by loss of SorCS1 (Fig 4C and 4D). Moreover, Dynasore treatment to block endocytosis did not impair axonal surface polarization of Nrxn1α Δ4.1 (Fig 4E and 4F), indicating that the Nrxn1α Δ4.1 mutant bypasses dendritic endocytosis and sorting by SorCS1 to polarize to the axonal surface.

SorCS1-mediated sorting is required for presynaptic function
Missorting of Nrxns, which regulate neurotransmitter release [5][6][7], would also be expected to impair presynaptic function. To assess the consequences of loss of SorCS1 on synaptic function, we recorded spontaneous miniature excitatory postsynaptic currents (mEPSCs) from Sorcs1 flox/flox autaptic cortical cultures electroporated with Cre or EGFP. SorCS1 loss strongly decreased mEPSC frequency (Fig 5A and 5B). Decay kinetics and amplitude of mEPSCs were not altered by loss of SorCS1 (Fig 5A and 5B), suggesting that the amount of neurotransmitter release per vesicle and postsynaptic receptor properties and/or receptor density were not affected. The amplitude and total charge transfer of single evoked EPSCs (eEPSCs) were reduced in Sorcs1 KO neurons (Fig 5C and 5D). The mEPSC frequency and eEPSC amplitude defects can be attributed to a decrease in the readily releasable pool (RRP) size, in the vesicular release probability (Pves), or in active synapse number. Indeed, we previously found an increase in the number of silent excitatory synapses after loss of SorCS1 [20]. To assess RRP size, we applied a hyperosmotic sucrose stimulus. The amplitude and total charge transfer of Dendritic sorting of presynaptic organizer Nrxn regulates synapse development  (Fig 5E and 5F) and RRP size ( Fig 5G) induced by single application of 0.5 M sucrose were strongly decreased in Sorcs1 KO neurons, in line with a reduction in active synapse number. Because eEPSC amplitude and RRP size were proportionally reduced in Sorcs1 KO cells, the Pves (eEPSC charge/initial sucrose charge) was unchanged ( Fig 5H). To further examine potential presynaptic release defects in Sorcs1 KO neurons, we performed train stimulations at different frequencies to probe for short-term plasticity defects that cannot be detected by induction of single EPSC and single sucrose application. Repeated stimulation at 10 Hz produced a more pronounced rundown of normalized evoked responses (synaptic depression) in Sorcs1 KO neurons compared with control cells (Fig 5I and 5J), suggesting a presynaptic defect during sustained periods of activity. Other stimulation frequencies and paired-pulse stimulations also showed increased synaptic depression in SorCS1 KO neurons ( Fig 5K). Calculating the RRP size by back-extrapolation of cumulative synchronous charge during 10 Hz train stimulation confirmed a reduction in RRP size in Sorcs1 KO neurons ( Fig  5L), indicating a presynaptic defect independent of the silent synapses phenotype that we described previously [20]. Together, these observations indicate that loss of SorCS1 impairs neurotransmitter release, reminiscent of Nrxn loss-of-function [5][6][7], indicating that SorCS1mediated sorting of Nrxns in dendrites is required for normal presynaptic function.

Discussion
The membrane trafficking mechanisms by which neurons control the polarized distribution and abundance of synaptic receptors are poorly understood. Here, we show that SorCS1-mediated sorting in dendritic endosomes controls a balance between axonal and dendritic surface levels of Nrxn and maintains Nrxn axonal surface polarization required for presynaptic differentiation and function.

A SorCS1-Rab11FIP5/Rip11 interaction sorts axonal cargo from EEs to REs
The mechanism by which SorCS1 regulates trafficking of neuronal receptors has remained unclear. We find that SorCS1 controls a balance between axonal and dendritic surface polarization of its cargo Nrxn. In mature neurons, newly synthesized Nrxn1α traffics to the somatodendritic surface, followed by endocytosis and transcytosis to the axon. Accordingly, blocking endocytosis, EE-, and RE-mediated transport all result in accumulation of Nrxn1α on the dendritic surface. These effects are mimicked by Sorcs1 KO. We find that endosomal SorCS1 controls the transition of endocytosed Nrxn1α from EEs to REs and identify the Rab11-interacting protein Rab11FIP5/Rip11 as a novel SorCS1 cytoplasmic interactor. Interference with Rip11 function alters the axonal-dendritic surface balance of Nrxn1α in the same way that Sorcs1 KO does. Rab11FIPs function as linkers between Rab11 and motor proteins to promote sorting of cargo from EEs to REs [37,47] and transcytosis in polarized epithelial cells [38,39], but their function in neuronal transcytosis has not been explored. Our data suggest that SorCS1/Rip11 form a protein complex that localizes to dendritic endosomes and sorts internalized Nrxn1α from early to recycling endosomes, thereby biasing the trafficking of Nrxn1α-depression in Sorcs1 KO neurons throughout different interstimulation intervals. Control (n = 18 neurons); Cre (n = 14). � P < 0.05 (two-tailed t test, 3 independent experiments). (L) Estimation of the RRP size used during neuronal activity by back-extrapolating a linear fit of the steady state current towards the ordinate axis intercept, which represents the initial RRP size before train stimulations (10 Hz). RRP size of active synapses was reduced in DIV14 to DIV16 Sorcs1 KO neurons. Control (n = 18 neurons); Cre (n = 13). �� P < 0.01 (Mann-Whitney test, 3 independent experiments). Underlying numerical values can be found in S1 Data. Graphs show mean ± SEM. Cre, Cre recombinase; DIV, days in vitro; eEPSC, evoked excitatory postsynaptic current; EGFP, enhanced green fluorescent protein; KO, knock out; mEPSC, miniature excitatory postsynaptic current; RRP, readily releasable pool; SorCS1, Sortilin-related CNS expressed 1. https://doi.org/10.1371/journal.pbio.3000466.g005 Dendritic sorting of presynaptic organizer Nrxn regulates synapse development containing REs to the axon while preventing Nrxn1α missorting to the dendritic surface and lysosomes.
SorCS1-mediated sorting of Nrxn1α in dendrites can be circumvented by deleting the 4.1-binding motif in the cytoplasmic tail of Nrxn1α, which results in strong polarization of Nrxn1α to the axonal surface. Surface polarization of Nrxn1α Δ4.1 is unaffected by blocking endocytosis or removing Sorcs1. It remains to be determined whether Nrxn1α's 4.1-binding motif, which lacks canonical YXXØ and (D/E)XXXL(L/I) dendritic sorting motifs [48] but contains several tyrosine residues, is required for initial sorting to dendrites or for insertion in the dendritic membrane. Binding of the α-amino-3-hydroxy-5-methyl-4-isoxazole propionictype receptor (AMPAR) subunit GluA1 to the 4.1N protein is required for GluA1 membrane insertion [49]. We find that the extracellular domain of Nrxn1α is required for somatodendritic sorting (S8D Fig and S8E Fig), suggesting that deletion of the 4.1-binding motif in Nrxn1α impairs dendritic insertion but not initial sorting to this compartment.
Our results show that loss of SorCS1 impairs transcytotic trafficking of Nrxn and L1/ NgCAM, but not of Caspr2, which traffics via the selective endocytosis/retention pathway, supporting a role for SorCS1 in regulating transcytotic trafficking of axonal cargo proteins. It is unlikely that all membrane proteins that have been identified using surface proteome analysis to depend on SorCS1-3 for their surface trafficking [20,23,24] undergo transcytosis. The SorCS1 cargo amyloid precursor protein (APP) [50][51][52], however, is transcytosed in human neurons [53]. SorCS1 likely interacts with additional proteins, such as the retromer complex [50], to regulate the surface trafficking of a diverse set of cargoes.

SorCS1-mediated surface polarization of Nrxn is required for presynaptic differentiation and function
The selective impairment in Nlgn1-induced presynaptic differentiation in Sorcs1 KO neurons and rescue thereof by Nrxn1α Δ4.1, but not by WT Nrxn1α, indicates that this defect is due to a decrease in the abundance of axonal surface Nrxns. Nrxns play key roles in neurotransmitter release [5][6][7]. Alpha-Nrxn KO in neocortical cultured slices reduces mEPSC/miniature inhibitory postsynaptic current (mIPSC) frequency and evoked inhibitory postsynaptic current (eIPSC) amplitude [7]. β-Nrxn KO in cortical neurons decreases mEPSC frequency and eEPSC amplitude [5]. Our results show that loss of SorCS1 in autaptic cortical neurons decreases mEPSC frequency, eEPSC amplitude, and sucrose-evoked responses, in line with the reported decreased fraction of active synapses in Sorcs1 KO neurons [20]. Whether this decrease in the fraction of active synapses is also happening in autaptic cultured neurons requires further experimental testing. Furthermore, we found an increase in synaptic depression of normalized eEPCSs during stimulus trains, similar to α-Nrxn KO neurons [7]. Together, these results suggest that mistrafficking of Nrxn is a major contributor to the defects in presynaptic function in Sorcs1 KO neurons. These defects are likely attributed to a decrease in active synapse number and a presynaptic defect in neurotransmitter release, rather than a decrease in excitatory synapse density, which is unaffected in cultured Sorcs1 KO cortical neurons [20] and in Nrxn TKD hippocampal neurons [46].
The axonal-dendritic balance of Nrxn1α is developmentally regulated. Early in neuronal development, endogenous Nrxn1α is enriched on the axonal surface and traffics directly to the axon after trans-Golgi network (TGN) exit. As neurons mature, Nrxn1α follows an indirect trafficking pathway after TGN exit via SorCS1-mediated sorting in dendritic endosomes and transcytosis to the axonal surface. It will be interesting to determine whether expression of Sorcs1 and Rip11 is developmentally regulated in these neurons. The biological function of this developmental regulation and circuitous trafficking route is not clear. Transcytosis could provide an efficient way of delivering receptors from a reservoir of readily synthesized proteins. Similar to other transcytosed axonal cargo, a dendritic pool of Nrxn may supply the demand for receptors along the axonal surface, either constitutively [54,55] or in response to ligand-receptor interactions and signalling [54,56].
Dendritically localized Nrxns might have a function in this compartment. Postsynaptically expressed Nrxn1 decreases Nlgn1's synaptogenic effect in hippocampal neurons, likely by cisinhibition of Nlgn1 [32]. In retinal ganglion cells, a shift in Nrxn localization away from dendrites has been proposed to allow dendritic innervation [30]. In cortical neurons, however, we observe the opposite: Nrxn surface expression in dendrites increases with maturation. Dendritic Nrxns might modulate the function of postsynaptic Nrxn ligands, such as Nlgns, which control postsynaptic neurotransmitter receptor function [57], or leucine-rich repeat transmembrane neuronal protein 1 (LRRTM1), which shapes presynaptic properties [58]. At the Caenorhabditis elegans neuromuscular junction, the ectodomain of postsynaptic Nrxn is proteolytically cleaved and binds to the presynaptic α2δ calcium channel subunit to inhibit presynaptic release [59].
In conclusion, impaired trafficking of critical cargo proteins in the absence of SorCS1 and resulting defects in synaptic function might contribute to the pathophysiology of the neurodevelopmental [60][61][62][63] and neurodegenerative disorders [52,64] with which SORCS1 has been associated. These observations underscore the notion that intracellular sorting is a key contributor to the proper maintenance of synaptic protein composition and function.

Ethics statement
All animal procedures were approved by the Institutional Animal Care and Research Advisory Committee of the KU Leuven (ECD P015/2013 and ECD P214/2017) and were performed in accordance with the Animal Welfare Committee guidelines of the KU Leuven, Belgium. The health and welfare of the animals were supervised by a designated veterinarian. The KU Leuven animal facilities comply with all appropriate standards (cages, space per animal, temperature, light, humidity, food, water), and all cages are enriched with materials that allow the animals to exert their natural behaviour. Both males and females were used for all experiments. To the best of our knowledge, we are not aware of an influence of sex on the parameters analyzed in this study.

Neuronal cultures
Primary neuronal cultures for imaging. Neurons were cultured from E18 to E19 Wistar rat embryos (cortical), from E18 to E19 WT C57BL/6J mice embryos (cortical and hippocampal), from E18 to E19 Sorcs1 flox/flox mice (cortical), from E18 to E19 Sorcs1 HA mice (cortical), and E18 to E19 Nrxn1α HA mice (cortical), as previously described by Savas and colleagues [20]. Cultures were maintained in a humidified incubator of 5% (vol/vol) CO 2 /95% (vol/vol) air at 37˚C, feeding the cells once per week by replacing one-third of the medium in the dish. Cortical neurons from WT and Sorcs1 flox/flox mice were electroporated with DNA just before plating using an AMAXA Nucleofector kit (Lonza, Basel, Switzerland, Cat #VPG-1001). Cortical neurons from Sorcs1 flox/flox mice were transfected at DIV7 using Effectene (Qiagen, Hilden, Germany, Cat #301425). Rat cortical neurons and neurons from WT (cortical and hippocampal) and Sorcs1 flox/flox (cortical) mice were transfected at DIV6 to DIV8 using calcium phosphate. Mouse cortical neurons from Nrxn1α HA and Sorcs1 HA were mixed with WT cortical neurons which allowed for reliable immunodetection of endogenous HA-Nrxn1α and HA-SorCS1.

Neuron transfection
Rat cortical neurons (live-cell imaging experiments) and neurons from WT (cortical and hippocampal) and Sorcs1 flox/flox (cortical) mice were transfected at DIV6 to DIV8 using calcium phosphate, with the exception of the DIV3 live-cell imaging experiment, in which rat cortical neurons were transfected at DIV2. A total of 2 μg of DNA or 1 μg of DNA from each DNA construct (double co-transfections) were used per coverslip, with the exception of co-transfections of EGFP-tagged Rab proteins or Rip11 (WT and dominant negatives) with extracellular HA-tagged Nrxn1α, in which 0.75 μg and 1.25 μg of DNA was used for EGFP-tagged and HA-Nrxn1α DNA constructs, respectively. Briefly, DNA plasmids were diluted in Tris-EDTA buffer (10 mM Tris-HCl and 2.5 mM EDTA [pH 7.3]), followed by dropwise addition of CaCl 2 solution (2.5 M CaCl 2 in 10 mM HEPES [pH 7.2]) to the plasmid DNA-containing solution to give a final concentration of 250 mM CaCl 2 . This solution was subsequently added to an equal volume of HEPES-buffered solution (274 mM NaCl, 10 mM KCl, 1.4 mM Na 2 HPO 4 , 42 mM HEPES [pH 7.2]) and vortexed gently for 3 s. This mixture, containing precipitated DNA, was then added dropwise to the coverslips in a 12-well plate, containing 250 μL of conditioned neuronal culture medium with kynurenic acid (2 mM), followed by 2 h incubation in a 37˚C, 5% (vol/vol) CO 2 /95% (vol/vol) air incubator. After 2 h, the transfection solution was removed, after which 1 mL of conditioned neuronal culture medium with kynurenic acid (2 mM) slightly acidified with HCl (approximately 5 mM final concentration) was added to each coverslip, and the plate was returned to a 37˚C, 5% (vol/vol) CO 2 /95% (vol/vol) air incubator for 20 min. Finally, coverslips were then transferred to the original dish containing the conditioned culture medium in the 37˚C, 5% (vol/vol) CO 2 /95% (vol/vol) air incubator to allow expression of the transfected constructs. Protein expression was typically for 24 h (live-cell imaging) or 48 h (immunocytochemistry).

Immunocytochemistry
Cells were fixed for 10 to 15 min in 4% (wt/vol) sucrose and 4% (wt/vol) paraformaldehyde in PBS (PBS: 137 mM NaCl, 2.7 mM KCl, 1.8 mM KH 2 PO 4 and 10 mM Na 2 HPO 4 [pH 7.4]) at room temperature and permeabilized with PBS + 0.25% (vol/vol) Triton X-100 for 5 min at 4˚C. Neurons were then incubated in 10% (wt/vol) bovine serum albumin (BSA) in PBS for 1 h at room temperature to block nonspecific staining and incubated in appropriate primary antibodies diluted in 3% (wt/vol) BSA in PBS (overnight, 4˚C). After washing 3 times in PBS, cells were incubated with the secondary antibodies diluted in 3% (wt/vol) BSA in PBS (1 h, room temperature). The coverslips were mounted using Prolong Gold Antifade mounting medium (Thermo Fisher Scientific, Waltham, MA, Cat #P36930).
In all immunocytochemistry experiments performed to determine the subcellular distribution of proteins of interest (Nrxns, SorCS1, Caspr2, L1) the axonal compartment was labeled using antibodies against Ankyrin-G, a marker of the AIS (NeuroMab, Davis, CA, Cat #73-146 or Santa Cruz Biotechnology, Dallas, TX, Cat #sc-31778), and the somatodendritic compartment was labeled using an anti-MAP2 antibody (Abcam, Cambridge, England, Cat #ab5392). See S2 Table for a  Neurons were then fixed in 4% (wt/vol) sucrose and 4% (wt/vol) paraformaldehyde in PBS for 10 min at room temperature, followed by several washes in PBS and blocking in 10% (wt/vol) BSA in PBS for 1 h at room temperature. Neurons were then incubated with anti-rabbit secondary antibody diluted in 3% (wt/vol) BSA in PBS (1 h, room temperature). Following permeabilization, neurons were processed for immunocytochemistry as described above.
Surface immunostaining of extracellular myc-tagged L1. To allow surface immunostaining of myc-L1 (live-labeling proved to be impossible), neurons were fixed first, followed by several washes and blocking, and then incubated with mouse anti-myc (1:1,000 dilution; Santa Cruz Biotechnology, Dallas, TX, Cat #sc-40) diluted in 3% (wt/vol) BSA in PBS overnight at 4˚C. Subsequently, neurons were incubated with the respective secondary antibody for 1 h at room temperature and processed for immunocytochemistry as described above.
Antibody pulse-chase experiments. Cultured living neurons were incubated at room temperature for 10 min in the presence of a high concentration (1:250) of mouse anti-HA antibody (Covance, Princeton, NJ, Cat #MMS-101P) against extracellular HA-tagged Nrxn1α and SorCS1, diluted in conditioned medium. Neurons were then washed with prewarmed PBS at 37˚C to remove the unbound antibody and were further incubated in antibody free conditioned medium in a 37˚C, 5% (vol/vol) CO 2 /95% (vol/vol) air incubator (for different periods) to allow the internalization of antibody-bound receptors. After this incubation, neurons were fixed in 4% (wt/vol) sucrose and 4% (wt/vol) paraformaldehyde in PBS for 10 min at room temperature. Next, neurons were either exposed to a supersaturating concentration (1:300) of the first of 2 secondary antibodies, to label the primary antibody-bound surface pool of protein, and/or incubated overnight with anti-mouse Fab fragments (0.25 mg/mL; Santa Cruz Biotechnology, Dallas, TX, Cat #715-007-003) in 5% (wt/vol) BSA in PBS to block all primary antibody-bound receptors that were not internalized and/or not labeled by the first secondary antibody. After permeabilization, cells were processed for immunocytochemistry as described above, and the pool of internalized receptors was labeled by incubation with the second secondary antibody (1:1,000) for 1 h at room temperature. This strategy allows differential labeling of cell surface and internalized pools of protein.
Dynasore treatment. Mouse cortical neurons expressing extracellular HA-tagged Nrxn1α for 30 h were either treated with 20 μM Dynasore (Sigma-Aldrich, St. Louis, MO, Cat #D7693) or DMSO (vehicle) and added to the coverslips in a 12-well plate containing conditioned neuronal culture medium for 18 h. Afterwards, surface HA-Nrxn1α was labeled in live neurons as described above.

Image analysis and quantification
All images obtained from immunocytochemistry experiments in fixed cells were captured on a Leica SP8 laser-scanning confocal microscope (Leica Micro-systems, Wetzlar, Germany). The same confocal acquisition settings were applied to all images taken from a single experiment. Parameters were adjusted so that the pixel intensities were below saturation. Fiji analysis software was used for quantitative imaging analysis. Z-stacked images were converted to maximal intensity projections and thresholded using constant settings per experiment.
Quantification of axonal and dendritic immunofluorescence intensity and ratio of axonal-dendritic immunofluorescence intensity. Fluorescence intensity was measured as the sum of integrated intensity in representative portions of axons and dendrites using Ankyrin-G and MAP2 as guides, respectively. Axonal and dendritic intensities were divided by neuritic length and total intensity (axonal + dendritic) ('Relative Axonal Intensity' and 'Relative Dendritic Intensity') to adjust measurements across cells with varying expression levels (exogenous expression). The A-D ratio was then calculated by dividing the values of axonal and dendritic intensities obtained for every cell ('Axonal/Dendritic Intensity'). A uniformly distributed protein yields an A-D ratio of around 1. A preferentially dendritically localized protein yields an A-D ratio < 1, whereas a preferentially axonally localized protein yields an A-D ratio > 1. To quantify the fluorescence intensity of endogenous Nrxns, axonal and dendritic intensities were not normalized to the total intensity ('Axonal Intensity' and 'Dendritic Intensity').
Polarity index. In some cases, polarization of cargo was determined by using the classical polarity index originally described by Sampo and colleagues and Wisco and colleagues [68,69]. This index does not provide a direct measurement of the axonal and dendritic fluorescent intensities, but it allows a quick estimation of the compartmentalized polarization of proteins of interest. One-pixel-wide lines were traced along 3 dendrites and representative portions of the axon, using MAP2 and Ankyrin-G as guides, in nonthresholded images. The mean intensities (dendritic was averaged from 3 dendrites) were used to calculate the dendrite:axon (D:A) 'Polarity Index'. D:A = 1, uniform staining; D:A < 1, preferential axonal staining; D:A > 1, preferential dendritic staining.
Long-term live-cell imaging. Fluorescence intensity was measured as the sum of integrated intensity in representative portions of axons and dendrites (using Ankyrin-G and MAP2 as guides) and cell bodies. Axonal and dendritic intensities were divided by neuritic length and somatic intensities by somatic area ('Nrxn Intensity' or 'TfR Intensity').

Antibody pulse-chase experiments.
Fluorescence intensity was measured as the sum of integrated intensity in representative portions of axons and dendrites using Ankyrin-G and MAP2 as guides, respectively. Intensities of internalized SorCS1 and Nrxn were divided by neuritic length and by the total intensity (surface + internal) ('Relative Intensity of Internalized Nrxn' or 'SorCS') or by the total internalized intensity (axonal + dendritic) ('Relative Intensity of Internalized Nrxn'). Intensities of surface SorCS1 were also divided by neuritic length and by the total intensity (surface + internal) ('Relative Intensity of Surface SorCS1').
Manders coefficient. Manders coefficient was measured by using the Fiji plugin JACoP [70]. Manders coefficient measures the proportion of the signal from channel 'a' that coincides with the signal in channel 'b' over the total intensity of 'a'-M1 coefficient [71].
Colocalization of internalized Nrxn with endosomal markers in Sorcs1 KO cells. The colocalization of internalized Nrxn with endosomal markers was evaluated by measuring the density of double-positive puncta for internalized Nrxn and endosomes and by measuring the intensity of Nrxn present in these puncta. Fluorescence intensity was measured as the sum of mean intensity of internalized Nrxn puncta (defined as 0.02 μm 2 -infinite) in representative portions of axons and dendrites (using Ankyrin-G and MAP2 as guides), colocalizing with endosomal markers. Intensities of internalized Nrxn were divided by neuritic length ('Nrxn Intensity in Double Positive Puncta'). Density of internalized Nrxn puncta (defined as 0.02 μm 2 -infinite), colocalizing with endosomal markers, was also divided by neuritic length ('Double Positive Puncta Density').

Live-cell imaging
For live-cell imaging of fluorescently tagged Nrxn1α-and TfR-positive vesicles, neurons were transfected with a bicistronic expression plasmid encoding Streptavidin-KDEL and SBP-EGFP-Nrxn1α or Streptavidin-KDEL and TfR-SBP-EGFP, respectively, using the RUSH system [34]. After transfection, neurons were maintained in Neurobasal medium without B27 supplement, because the presence of D-biotin in B27 interferes with the RUSH system, or in Neurobasal medium supplemented with N-2 supplement (Thermo Fisher Scientific, Waltham, MA, Cat #P36930), which lacks D-biotin. Live-cell imaging was performed at room temperature (approximately 20˚C) in imaging medium (119 mM NaCl, 5 mM KCl, 2 mM CaCl2, 2 mM MgCl2, 30 mM glucose, and 10 mM Hepes [pH 7.4]). For long-term live-cell imaging experiments DIV8 to DIV10 or DIV3 rat cortical neurons, after 24 to 31 h or 25 to 29 h of protein expression, respectively, were imaged on a Nikon Eclipse Ti A1R confocal microscope (Minato, Japan). Images were collected with a 40× oil objective (1.3 NA; Nikon, Minato, Japan). Focal drift during the experiment was avoided by using the perfect focus system feature of the Nikon system. Laser intensities were kept as low as possible. Time lapses lasted for 2.5 h and frames were acquired as z-stacks every 5 min. Synchronous release of Nrxn1α or TfR was induced by application of 40 μM D-biotin (Sigma-Aldrich, St. Louis, MO, Cat #B4501) 10 min after the beginning of the imaging session. Protein synthesis was inhibited by including 20 μg/ mL cycloheximide (Sigma-Aldrich, St. Louis, MO, Cat #C4859) in the imaging medium. For short-term live-cell imaging experiments, DIV8 to DIV10 rat cortical neurons expressing Streptavidin-KDEL and SBP-EGFP-Nrxn1α, 21 to 30 h after protein expression, were imaged on a Nikon spinning-disk confocal microscope equipped with a 60× oil objective (1.4 NA; Nikon). Time-lapses were performed by sequential capture of 200 ms images of SBP-EGFP-Nrxn1α every second for 60 to 120 s. Focal drift during the experiment was corrected automatically using the autofocus feature of the Nikon system. Neurons were imaged either for 20 to 34 min or 1 to 2 h after exposure to 40 μM D-biotin. In both experiments, a single image of live-stained pan-Neurofascin was taken at the beginning of every imaging session.

Live-labeling of the AIS.
Before every live-cell imaging experiment, primary rat cortical cultured neurons were live-labeled with an anti-pan-Neurofascin antibody (NeuroMab, Davis, CA, Cat #75-172) to distinguish the axon from the dendrites. Live-cell imaging experiments were performed with rat cortical neurons because live-labeling of the AIS with the anti-pan-Neurofascin antibody did not work in mouse cultured neurons in our hands. Briefly, coverslips with neurons were quickly rinsed in prewarmed neuronal culture medium. Neurons were incubated with anti-pan-Neurofascin antibody (1/500) diluted in conditioned neuronal culture medium for 10 min in a 37˚C, 5% (vol/vol) CO 2 /95% (vol/vol) air incubator. Coverslips were then quickly washed twice with prewarmed neuronal culture medium. Finally, neurons were incubated with an Alexa-555 anti-mouse (1/400; Thermo Fisher Scientific, Waltham, MA, Cat #A31570) secondary antibody diluted in conditioned neuronal culture medium and incubated for 10 min in a 37˚C, 5% (vol/vol) CO 2 /95% (vol/vol) air incubator. Neurons were quickly rinsed twice with prewarmed neuronal culture medium and used for live-cell imaging experiments.
Analysis of Nrxn1α-and TfR-positive vesicle transport. Dendrites and axon were imaged from the same neuron using spinning-disk confocal microscopy. Time lapses were performed by sequential capture of 200 ms images every second for 60 to 120 s. Acquisitions were analyzed using an Igor-based software developed by Pieter Vanden Berghe and Valérie Van Steenbergen (Lab. for Enteric NeuroScience and Cell Imaging Core, KU Leuven). Kymographs were created using a segmented line along the axon/dendrite (1 dendrite was analyzed per cell) from the soma towards the neurite tip, so that anterograde movement occurred from left to right. On the kymographs, single vesicle movement episodes were distinguished as tilted straight lines. Single vesicle pauses were distinguished as straight lines. The number of moving anterograde/retrograde particles in the dendrite and axon was determined manually by drawing lines on top of the trajectories of single particles obtained from the kymographs. Vesicles were defined as motile when showing net displacement of � 2 μm.
Immunoprecipitation of HA-tagged endogenous SorCS1. Cortices from adult Sorcs1 HA KI mouse brains were dissected and homogenized in homogenization buffer (50 mM HEPES [pH 7.4], 100 mM NaCl, 2 mM CaCl 2 , 2.5 mM MgCl 2 ) supplemented with protease inhibitors (Roche, Basel, Switzerland, Cat #11697498001) using a Dounce homogenizer. Cortical homogenates were extracted with 1% CHAPSO (VWR, Radnor, PA, Cat #A1100.0005) in homogenization buffer while rotating end-over-end for 2 h at 4˚C and centrifuged at 100,000g for 1 h at 4˚C to pellet insoluble material. Supernatants were precleared by adding 100 μL of Protein-G agarose beads (Thermo Fisher Scientific, Waltham, MA, Cat #22852BR) and rotating for 1 h at 4˚C, then incubated with 5 μg of mouse IgG or 50 μL of anti-HA magnetic beads (Thermo Fisher Scientific, Waltham, MA, Cat #88836) and rotated end-over-end overnight at 4˚C. A total of 60 μL of Protein-G agarose beads were added to the IgG control sample and rotated for 1 h at 4˚C, followed by 3 washes in cold extraction buffer and once in PBS. Anti-HA magnetic beads were washed according to the manufacturer's protocol using a magnetic stand. Protein-G agarose beads and anti-HA magnetic beads were heated for 15 min at 40˚C in 50 μL 2X sample buffer and analyzed by western blotting.

LV production
Second generation VSV.G pseudotyped LVs were produced as described by Dittgen and colleagues and Kutner and colleagues [72,73]. HEK293T cells were transfected with control (mCherry) or Cre-T2A-mCherry-containing pFUGW vector plasmids and helper plasmids PAX2 and VSVG using Fugene6 . Currents were low-pass filtered at 3 kHz and stored at 20 kHz. Patch pipettes were pulled from borosilicate glass using a multistep puller (P-1000; Sutter Instruments, Novato, CA). Pipette resistance ranged from 3 to 5 MΩ. The series resistance was compensated to approximately 75%. Only cells with series resistances below 15 MΩ were included for analysis. All recordings were made at room temperature. Spontaneous glutamatergic release (sEPSC) was recorded at −70 mV. Evoked release was induced using brief depolarization of the cell soma (from 70 to 0 mV for 1 ms) to initiate action potential-dependent glutamatergic release (eEPSCs). A fast local multibarrel perfusion system (Warner SF-77B, Warner Instruments, Hamden, CT) was used to determine the RRP size using external recording solution containing 500 mM sucrose. A custom analysis procedure in Igor Pro (Wavemetrics Inc., Lake Oswego, OR) was used for offline analysis of evoked and sucrose responses. Spontaneous events were detected using Mini Analysis program (Synaptosoft, Fort Lee, NJ).