Neurexins play a crucial role in cerebellar granule cell survival by organizing autocrine machinery for neurotrophins

Neurexins (NRXNs) are among the key presynaptic cell adhesion molecules that regulate synapse function and formation via trans-synaptic interaction with postsynaptic ligands. Here, we generated cerebellar granule cell (CGC)-specific Nrxn triple-knockout (TKO) mice to allow the deletion of all NRXNs. Unexpectedly, most CGCs died in these mice. The requirement of NRNXs for cell survival was reproduced in cultured CGCs. We showed that the axons of cultured Nrxn TKO CGCs that were not in contact with the postsynaptic structure had defects in the formation of presynaptic protein cluster and action potential-induced Ca2+ influx. Additionally, these cells were impaired in the secretion from axons of depolarization-induced fluorescence-tagged brain-derived neurotrophic factor (BDNF), and the cell-survival defect was rescued by the application of BDNF. Our results suggest that CGC survival is maintained by autocrine neurotrophic factors, and that NRXNs organize the presynaptic protein clusters and the autocrine neurotrophic factor secretory machinery independent of contact with postsynaptic ligands.


INTRODUCTION
Synapse formation is a multistep process that is triggered after the initial contact of an axon with a postsynaptic cell; however, this process is thought to vary according to the type of synapse 1 . During this process, synaptic cell adhesion molecules called synapse organizers play crucial roles in triggering the formation, organization, and functional specialization of synapses. Neurexins (NRXNs) are among the major presynaptic organizers and interact with various types of postsynaptic cell adhesion molecules including neuroligins (NLGNs), leucine-rich repeat transmembrane neuronal proteins (LRRTMs), and the cerebellin precursor protein 1 (Cbln1)-glutamate receptor δ2 (GluD2) complex 2 . In mammals, 3 NRXNs are encoded by three genes (Nrxn1, Nrxn2, and Nrxn3). Two alternative promoters in each gene produce a long α-NRXN and a short β-NRXN, and each transcript undergoes alternative splicing at six sites (splice sites [1][2][3][4][5][6] and at two sites (splice sites 4 and 5), respectively 2 . Although Nrxn mRNAs are widely expressed in the brain, the expression patterns of multiple Nrxn variants differ according to cell type 3 . The functions of NRXNs in various types of synapse are supported by several physiological evidences. Genetic manipulations of Nrxn genes in mice have shown that α-NRNXs and β-NRXNs play different roles in synaptic functions in different types of synapses [4][5][6][7] . Triple deletion of Nrxn1, Nrxn2, and Nrxn3 in four different types of synapses, i.e., climbing fiber (CF)-Purkinje cell (PC) synapses, synapses formed by parvalbumin-or somatostatin-positive interneurons on pyramidal layer 5 neurons of the medial prefrontal cortex, and the calyx of Held synapse, yields severe but distinct synaptic phenotypes, ranging from impairments in their distribution and functions to decreases in synapse number 8,9 .
The process of synapse formation between parallel fibers (PFs), which are the axons of cerebellar granule cells (CGCs), and cerebellar PCs has unique characteristics compared with that of other types of synapses, such as those in neocortical and hippocampal pyramidal neurons; i.e., the dendritic spines of cerebellar PCs are formed by cell-intrinsic factors without presynaptic contact 1,10 . Additionally, presynaptic elements of the PFs with vesicular profiles similar to those of synaptic vesicles are also intrinsically formed in the absence of their specific postsynaptic compartments 10,11 . However, the physiological role of these presynaptic-like structures remains unknown, and the cell-intrinsic factors that are necessary for the formation of the presynaptic-like structure have not been identified.
Conversely, the synapse organizers that mediate the formation of PF-PC synapses have been identified.
Previously, we found that the PF-PC synapses are formed and maintained by a trans-synaptic interaction between postsynaptic GluD2 and presynaptic NRXNs through the secreted Cbln1 12 . The functional importance of this ternary complex at the PF-PC synapse has been confirmed by both conventional and conditional GluD2 or Cbln1 knockout mice, in which cerebellar long-term depression (LTD) is impaired and approximately half of spines lack synaptic contact with the PF terminal 12-16 . However, in contrast with GluD2 and Cbln1, the physiological roles of the presynaptic-organizing NRXNs in CGCs have not been fully clarified. 4 In the present study, we generated CGC-specific Nrxn1, Nrxn2, and Nrxn3 triple-knockout (Nrxn TKO) mice to clarify the physiological role of NRXNs in CGCs. Contrary to the initial expectation, we found that almost all CGCs died in these mutant mice. This phenotype was entirely different from those of both conventional and conditional GluD2 or Cbln1 knockout mice [12][13][14][15][16] . We examined the effect of conditional Nrxn TKO in cultured CGCs, and found that NRXNs were required for their survival. The cultured Nrxn TKO CGCs had a morphological and functional defect of presynaptic-like structure which is formed independently of postsynaptic contact. Notably, in these cells, the activity-induced axonal neurotrophic factor secretory machinery was severely impaired, and the cell-survival defect was rescued by application of brain-derived neurotrophic factor (BDNF). Thus, our results suggest that NRXNs regulate the functional organization of the presynaptic-like structure and that axonal NRXNs are essential for the self-regulatory machinery of neurotrophic-factor-mediated CGC survival.
The size of the cerebellum of CGC-specific Nrxn TKO mice was smaller than that of control mice ( Fig. 1d), and their total amount of cerebellar protein was reduced to approximately 50% of that of wildtype or control floxed Nrxn (fNrxn) mice (Fig. 1e). There was no substantial difference in that of forebrain proteins among genotypes. To examine the deletion of NRXN proteins in CGC-specific Nrxn TKO mice, we performed a western blot analysis using a pan anti-NRXN antibody. Protein bands of approximately 160-200 kDa, corresponding to the size of α-NRXNs 18 , were detected in cerebellar 5 homogenates prepared from control fNrxn and wild-type mice. Conversely, the 90-100 kDa bands corresponding to the size of β-NRXNs 18 were not clearly detected in these mice. In CGC-specific Nrxn TKO mice, these signals were reduced to 38% of those detected in control mice (Fig. 1f,g). There was no substantial difference in the NRXN signals in forebrain homogenates among genotypes. We estimated an approximately 80% reduction of cerebellar NRXN proteins in CGC-specific Nrxn TKO mice (Fig. 1e,g).

Loss of CGCs in CGC-specific Nrxn TKO mice
CGC-specific Nrxn TKO mice showed severe ataxic gait and could not walk along a straight line using regular steps as control mice did. These mutant mice showed poor performance and no improvement in the constant-speed rotarod test (Supplementary Figure 2). To examine the formation of the cerebellar circuit, we performed an immunohistochemical analysis using antibodies against the vesicular glutamate transporter 1 (VGluT1), VGluT2, and calbindin. In the cerebellum, VGluT1 and VGluT2 are predominantly expressed in PF and CF terminals, respectively 19 . The parasagittal cerebellar sections prepared from 8-week-old CGC-specific Nrxn TKO mice were much smaller than those from control mice (Fig. 2a). The punctate staining signals for VGluT2 in the molecular layer were significantly increased in CGC-specific Nrxn TKO mice compared with control mice. In contrast, the staining signals for VGluT1 were decreased in the mutant mice compared with control mice (Fig. 2b,c). The VGluT1 signals in the mutant mice were different in each lobule, with the most prominent reduction detected in lobule IX (Fig. 2a,b). Unexpectedly, the DAPI staining signals were dramatically decreased in the GC layer of CGC-specific Nrxn TKO mice. The layered GC structure seemed to have disappeared in lobule IX of the mutant mice (Fig. 2b). Prominent reduction of VGluT1 signals in the molecular layer and of DAPI signals in the GC layer suggested the loss of CGCs in the mutant mice.

NRXNs are essential for the survival of CGCs
In the mouse, the precursors of CGCs proliferate at the surface of the developing cerebellum and migrate inwardly to form the internal granule layer. This process ends by postnatal day 20 (P20), when cerebellar circuit formation is completed 20 . To examine whether the loss of CGCs in CGC-specific Nrxn TKO 6 mice was caused by developmental defects, we analyzed the changes of the cerebellar structures with age. In the CGC-specific Nrxn TKO mice, the size of the cerebellum decreased gradually from 3 to 11 weeks of age (Fig. 3a). In lobules IV/V of 3-week-old mutant mice, the density of CGCs was comparable to that observed in control mice, but gradually decreased in as the animals aged. The well-defined layered structure of CGCs had almost completely disappeared at 11 weeks of age (Fig. 3b,c). Moreover, in lobule IX, the density of CGCs was decreased to 22% of that of control mice at 3 weeks of age, and the well-defined layered structure of CGCs had almost completely disappeared at 8 weeks of age (Fig.   3b,c). Conversely, the densities of PCs were increased in the mutant mice (Fig. 3b,d). An electron microscopic analysis revealed that the densities of PF-PC synapses were unchanged and decreased in lobules IV/V and IX of 4-week-old mutant mice, respectively. Consistent with the time course of CGC densities, the densities of PF-PC synapses were decreased from 4 weeks to 8 weeks in both lobules of the mutant mice (Supplementary Figure 3). These results suggest that CGCs were decreased in young mutant mice, whereas surviving CGCs formed synapses with PCs and died at a later stage. The differences in the time course of the density of CGCs among lobules were well matched with the developmental time course of Cre activity in GluN2C +/iCre mice, in which Cre-mediated recombination was first observed in lobules VIII and IX at P7 and gradually expanded to other lobules as the cerebellum developed (Supplementary Figure 4). Thus, the differences in the time course of the decrease in the density of CGCs is probably attributed to the difference in the time course of Nrxn deletion in each lobule. To examine whether the loss of CGCs in the CGC-specific Nrxn TKO mice is caused by apoptotic cell death, we performed TUNEL staining. The number of TUNEL-positive cells in the CGC layer was significantly increased in 3-week-old CGC-specific Nrxn TKO mice compared with control mice (Fig. 3e,f), suggesting the apoptotic death of CGCs. Collectively, these results indicate that NRXNs are essential for the survival of CGCs.
Next, we examined whether NRXNs have functional redundancy regarding CGC survival. We generated CGC-specific Nrxn1/2 double-knockout (DKO), Nrxn2/3 DKO, and Nrxn1/3 DKO mice (Supplementary Figure 5a). In contrast with CGC-specific Nrxn TKO mice, these mutant mice grew normally with no obvious ataxic gait, and showed apparently normal cerebellar structures (Supplementary Figure 5b). The DAPI signals in the GC layer and staining signals for VGluT1, VGluT2, 7 and calbindin were indistinguishable from those detected in control mice (Supplementary Figure 5c).
These results suggest that NRXN1, NRXN2, and NRXN3 are all expressed in CGCs and have functional redundancy, at least regarding CGCs survival. Finally, deletion of all NRNXs results in CGC death.

PF-PC synapse-independent regulation of CGC survival by NRXNs
Next, we investigated whether the loss of CGCs in CGC-specific Nrxn TKO mice was caused by dysfunction and/or loss of PF-PC synapses. For this, we used cultured CGCs, as they form no synapses among them and the majority of presynaptic-like structures with presynaptic vesicular profiles are not opposed to definite postsynaptic structures [21][22][23] , whereas functional synapses may be formed only in a small portion of CGCs, as synaptic transmission has been observed in CGCs 24 . The cultured CGCs that were prepared from fNrxn pups (termed cultured fNrxn CGCs) were used to analyze the effects of Nrxn There were no substantial differences in cell survival among NRXN1β(+S4), NRXN1α(+S4), and NRXN1α(-S4) co-infected cultures, suggesting that NRXN1β and NRXN1α have the same function, at least regarding cultured CGC survival. The extracellular regions are not identical among the NRXN1 variants, but share some sequences. Conversely, the transmembrane and cytoplasmic regions are identical among them. Thus, we speculated that the transmembrane and/or cytoplasmic region of NRXN1 are responsible for cultured CGC survival. To examine this hypothesis, we constructed three NRXN1β(+S4) mutants: (1) the cytoplasmic region of NRXN1β(+S4) was replaced with that of Ncadherin (NRXN1β(+S4)/N-cad), (2) the extracellular domain of NRXN1β(+S4) was fused to the transmembrane and cytoplasmic regions of interleukin-2 receptor-alpha (IL-2Ra) (NXN1β(+S4)/IL-2Ra); and (3) the extracellular region of NRXN1β(+S4) was fused to GPI (NRXN1β(+S4)-GPI) (Fig.   4d). The reduction in the number of CGCs in iCre-infected cultures was not rescued by these mutants (Fig. 4e,f). These results suggest that NRXNs regulate the viability of cultured CGCs through their Cterminal region.

Non-cell-autonomous factors are sufficient for the survival of cultured Nrxn TKO CGCs
To confirm further the role of NRXNs in cultured CGC survival, the cultured fNrxn CGCs were sparsely transfected with the expression vector for EGFP with or without that for iCre. In control EGFPtransfected fNrxn CGCs, numerous punctate staining signals for NRXNs were detected along the axons (Supplementary Figure 7a). These signals could not be detected in EGFP and iCre-transfected fNrxn CGCs, suggesting knockout of all NRXNs. Unexpectedly, the number of EGFP-positive CGCs was comparable between control and iCre-transfected fNrxn CGCs (Supplementary Figure 7b,c). This result was entirely different from the results obtained for the lentiviral-mediated Nrxn TKO (Fig. 4b-f). As one of the major differences between these two experiments was the population of Nrxn TKO cells used in the cultures (Supplementary Figures 6 and 7), we hypothesized that a non-cell-autonomous factor affects CGC survival. To examine whether cultured CGC survival is regulated non-cell-autonomously, we prepared a mixed culture of wild-type and Nrxn TKO CGCs (Fig. 5a). The co-culture of EGFPlabeled Nrxn TKO CGCs with non-EGFP-labeled Nrxn TKO CGCs led to a significant decrease in the numbers of both EGFP labeled and non-labeled cells compared with those in control cultures. Co-culture with non-EGFP-labeled wild-type CGCs restored the number of EGFP-labeled Nrxn TKO CGCs to the same level as that detected in the control cultures (Fig. 5b,c). These results suggest that non-cellautonomous factors and/or factors from wild-type CGCs are sufficient for the survival of Nrxn TKO CGCs.
Given that cultured CGCs do not form synapses among them, the paracrine action of neurotrophic factors would be one of the candidates for supporting Nrxn TKO CGC survival non-cell-autonomously in the mixed culture. Thus, to examine the effects of neurotrophic factors on the survival of cultured Nrxn TKO CGCs, we applied BDNF, neurotrophin (NT3), insulin-like growth factor-1 (IGF-1), fibroblast growth factor-1 (FGF-1), or FGF-2 extracellularly. The application of BDNF and IGF-1 significantly increased the cell viability of Nrxn TKO CGCs (Fig. 5d,e). In contrast, NT3, FGF-1, and FGF-2 had no positive effect on the survival of these cells. Among the neurotrophic factors tested here, BDNF was the most effective regarding the survival of cultured Nrxn TKO CGCs. These results raise the possibility that neurotrophic factor secretion is impaired in these cells.

Cultured Nrxn TKO CGCs have an impaired neurotrophic factor secretory machinery in axons
To examine whether the neurotrophic factor secretory machinery is impaired in Nrxn TKO CGCs, we used a super-ecliptic pHluorin-tagged BDNF (BDNF-pHluorin). The fNrxn CGCs were sparsely transfected with BDNF-pHluorin with or without iCre, and the BDNF-pHluorin secretion was quantified by monitoring its fluorescence increase upon exocytic secretion to the extracellular space. In control cultures, the number of BDNF-pHluorin puncta along the axons was significantly increased after the application of 50 mM KCl, suggesting the BDNF-pHluorin secretion from axons. In contrast, that in Nrxn TKO CGCs was not increased by the application of KCl at high concentration ( Fig. 6a,b).
Conversely, the expression of the chimeric mutant NRXN1β(+S4)/N-cad failed to rescue the fluorescence increase (Fig. 6a,b). These results suggest that the machinery of axonal neurotrophic factor secretion is impaired in Nrxn TKO CGCs and that it requires the C-terminal region of NRXNs.

Cultured Nrxn TKO CGCs exhibit impaired action potential-induced Ca 2+ influx into presynaptic-like structures
Neurotrophic factors are stored in dense-core vesicles, and exocytotic vesicular secretion is triggered by membrane depolarization-induced Ca 2+ influx 25 . In cultured CGCs, activity-dependent Ca 2+ influx and exocytosis occur at the site of axonal presynaptic-like structures with synaptic vesicle-like assemblies 21,22 . Thus, to examine the membrane-depolarization-induced Ca 2+ influx in the presynapticlike structure of CGCs, we constructed vesicle-associated membrane protein-2 (VAMP2) fused with the fluorescent calcium sensor GCaMP6s 26 at its N terminus (GCaMP6s-VAMP2). A previous study showed that GCaMP5G-synaptobrevin-2/VAMP2 is targeted to the presynaptic terminal and can be used for the analysis of presynaptic Ca 2+ transients 4 . The cultured CGCs were sparsely transfected with GCaMP6s-VAMP2 and TagRFP, and stained with antibodies against the active zone protein Bassoon and GCaMP6s. The majority of GCaMP6s-VAMP2 signals were well merged with Bassoon signals  Hz (2-80 stimuli), which were saturated after 25 stimuli (Fig. 7e,f). Fluorescence transients were blocked by tetrodotoxin, suggesting that the Ca 2+ influx was induced by action potential stimulation. In Nrxn TKO CGCs, the action potential-induced fluorescence transients were decreased, suggesting that action potential-induced Ca 2+ influx into presynaptic like structure is impaired. Decreased fluorescence transients were restored by co-transfection with NRXN1β(+S4), but not the chimeric mutant NRXN1β(+S4)/N-cad (Fig. 7e,f).

NRXNs are essential for the formation of presynaptic-like structures in cultured CGCs axons
The presynaptic-like structures in the axons of CGCs have vesicular profiles that are similar to those of synaptic vesicle structures, in which presynaptic proteins, such as the active zone protein Bassoon, VGluT1, and synapsin, are accumulated [21][22][23] . Thus, we examined the formation of presynaptic-like structures in Nrxn TKO CGCs. fNrxn CGCs were sparsely transfected with EGFP with or without iCre, and immunostained with antibodies against Bassoon, synapsin, and VGluT1, respectively. Nrxn TKO CGCs showed a significant decrease in the densities of punctate staining signals for Bassoon, synapsin, and VGluT1 along the axons compared with those detected in control CGCs ( Fig. 8a-f). Furthermore, the intensities of those signals were significantly decreased in Nrxn TKO CGCs (Fig. 8a-c and 8g-i).
Co-expression of NRXN1β(+S4) restored both the densities and intensities of Bassoon, synapsin, and VGluT1 puncta. Conversely, the chimeric mutant NRXN1β(+S4)/N-cad failed to restore these defects ( Fig. 8a-i). These results suggest that NRXNs regulate the formation of axonal presynaptic-like structures having presynaptic proteins through their C-terminal region, independent of PC contacts.

DISCUSSION
In the cerebellum, the GluD2-Cbln1-NRXN ternary complex regulates PF-PC synapse formation 12 . The analysis of both conventional and conditional KO mice provided evidence that GluD2 and Cbln1 play an essential role in the formation and maintenance of PF-PC synapses and in cerebellar LTD [12][13][14][15][16] .
However, the physiological role of NRXNs in CGCs has not been fully clarified. Here, we generated CGC-specific Nrxn KO mice and found that NRXNs organize the axonal neurotrophic factor secretory machinery and are thus essential for CGCs survival. We also discovered that NRXNs regulate the formation of presynaptic-like structures independent of postsynaptic ligands.
CGC-specific Nrxn TKO mice showed severe loss of CGCs (Fig. 2). A developmental analysis of CGC-specific Nrxn TKO mice cerebella suggested that NRXNs play a crucial role in the survival of CGCs (Fig. 3). These phenotypes were entirely different from those of conventional and conditional GluD2 or Cbln1 KO mice [12][13][14][15][16] . Recent studies of four different types of synapses in conditional Nrxn TKO mice, i.e., CF-PC synapse, synapses formed by parvalbumin-or somatostatin-positive interneurons on pyramidal layer 5 neurons in the medial prefrontal cortex, and the calyx of Held synapse, reported defects in synapse-specific functions, ranging from a decrease in synaptic density to impairments of synaptic transmission and action-potential-induced Ca 2+ influx 8,9 . Intriguingly, the effect of the deletion of all NRXNs in CGCs was strikingly different from those effects. During development, defects in synaptic connection or functions often induce apoptotic neuronal cell death in the brain 27 . In fact, a reduction of the CGC layer (14% reduction) was observed in GluD2-deficient mice, in which PF-PC synapse number was decreased to approximately half of that observed in wild-type mice 14 .
Several mutant mice, such as PC degeneration mice, cysteine protease cathepsin B and C double KO, and Zürich I prion protein (PrP) KO mice expressing a truncated PrP, exhibited an almost complete loss of cerebellar PC and a decreased number of CGCs; nevertheless, CGCs were still present in the inner GC layer 28-30 . These phenotypes were entirely different from our observation of almost complete loss of CGCs in CGC-specific Nrxn TKO mice (Fig. 3). Thus, complete loss of CGCs in the mutant mice cannot be ascribed to PF-PC synapse loss or dysfunction. To address this, we used cultured CGCs and showed the requirement of NRXNs for the survival of cultured CGCs (Fig. 4). Importantly, the cultured CGCs do not form synapses among them, and the axonal presynaptic-like structures containing presynaptic proteins lack contact with postsynaptic structures [21][22][23] . Thus, our results suggest that NRXNs regulate CGC survival in a non-synaptic manner. Consistent with this notion, functional and morphological defects were observed in cultured Nrxn TKO CGC axons (Figures 6-8).
It has been reported that neurotrophic factors support neuronal survival 25 . In the cultured Nrxn TKO CGCs, the defect of CGC survival was rescued by co-culture with wild-type CGCs or the application of the neurotrophic factors BDNF or (partially) IGF-1 (Fig. 5). In cultured Nrxn TKO CGCs, activitydependent axonal BDNF-pHluorin secretion was impaired (Fig. 6). These results suggest that neurotrophic factors act in a autocrine and/or paracrine manner for the survival of cultured CGCs, and that one of the causes of Nrxn TKO CGC death should be the impairment of neurotrophic factor secretion from the axons of CGCs. Consistent with this notion, it is suggested that the depolarizationinduced Ca 2+ influx that is critical for CGC survival occurs in the axonal varicosity compartment, rather than the dendritic compartment 22 . Unlike axonal varicosities in the cultured CGCs, axonal varicosities of PF that make synapses in contact with the PCs are covered with glial processes in vivo. Conversely, PFs form bundles with honeycomb structure without glial processes 11 . Thus, in addition to autocrine action at PF-PC synapses, autocrine and/or paracrine action at axons may work in vivo to support CGC survival. Autocrine regulation of neuronal survival by neurotrophic factors was also reported in cultured dorsal root ganglion and hippocampal neurons 31,32 . In our experimental condition, BDNF was most 13 effective on the survival of Nrxn TKO CGCs (Fig. 4e,f). BDNF and its receptor, TrkB, are reportedly expressed in cultured CGCs 33 . Previous studies showed that BDNF, NT-3, FGF-2, and IGF-1 enhance the survival of cultured CGCs 33-36 , albeit with variable effects, presumably because of the developmental stage of the CGCs. In the rodent cerebellum, the BDNF and NT-3 mRNAs are expressed in CGCs, which also express TrkB and TrkC 37-39 . Several studies have suggested that BDNF and NT-3 are secreted from PF terminals and act in an anterograde or autocrine fashion 40-43 . Conversely, IGF-1 is predominantly expressed in cerebellar PCs, while its receptor is expressed in both cerebellar PCs and CGCs 44-46 . It is thought that IGF-1 secreted by PC is taken up by CGCs 34,44 . Therefore, it is possible that multiple factors both from PFs and PCs sustain CGC survival in vivo.
To date, several molecules that regulate neurotrophic factor release have been identified. BDNF, which is one of the most studied neurotrophic factors, is stored in dense-core vesicles and its release is normally triggered by membrane depolarization induced Ca 2+ influx through voltage-gated Ca 2+ channels 25 . However, there are several unsolved questions about how its release site is constructed and functioned. In the cultured CGCs, we showed that action-potential-induced Ca 2+ influx occurred at the axonal presynaptic-like structure (Fig. 7). Consistently, previous studies showed that high-KCl-induced Ca 2+ influx and exo-endocytosis non-synaptically occurs in the axonal presynaptic-like structure of cultured CGCs, in which presynaptic proteins are accumulated 21,22 . In the cultured Nrxn TKO CGCs, the action potential-induced Ca 2+ influx into the axon was impaired (Fig. 7). Therefore, it is likely that the defect of axonal BDNF-pHluorin release in Nrxn TKO CGCs was caused by the defect of actionpotential-induced Ca 2+ influx into the axonal presynaptic-like structure (Fig. 7). Thus, these results suggest that NRXNs organize the Ca 2+ influx machinery that is essential for the release of neurotrophic factors. Our results also provide a molecular insight into the organization and function of the neurotrophic factor release site. Previous studies of α-NRXN TKO, conditional NRXN TKO, or β-NRXN TKO mice showed that both α-NRXNs and β-NRXNs play a role in controlling presynaptic Ca 2+ channels in several types of synapses via different mechanisms 4,7,8 ,47 . Here, the action-potential-induced Ca 2+ influx into the axonal presynaptic-like structure was impaired in Nrxn TKO CGCs. Furthermore, this defect was rescued by the expression of NRXN1β(+S4) (Fig. 7). Conversely, β-NRXNs reportedly function at the synapse, where they regulate Ca 2+ influx by controlling the tonic postsynaptic endocannabinoid signaling mediated by 2-AG 4 . Thus, the regulatory mechanism of axonal Ca 2+ influx mediated by NRXNs in cultured CGCs appears to be distinct from the synaptic functions of NRXNs.
The synapse formation process has been studied well in the neuromuscular junction. During development, prior to innervation, acetylcholine receptors are clustered in the prospective synaptic region of muscle fibers, which is called muscle pre-patterning. In mammals, pre-patterning requires the muscle-specific receptor tyrosine kinase MuSK and the low-density lipoprotein receptor-related protein 4 (Lrp4). Interestingly, muscle Lrp4 also binds to motor axons and sends a retrograde signal to the motor nerve for presynaptic differentiation 48 . In the cerebellum, presynaptic elements of the PF that exhibit vesicular profiles are intrinsically formed in the absence of specific postsynaptic compartments 10,11 . This is reproducible in in vitro CGCs cultures, in which axonal presynaptic-like structures containing synaptic proteins are intrinsically formed without contacting postsynaptic structures [21][22][23] . A previous study suggested that the presynaptic-like structures containing synaptic vesicle clusters observed in the cultured CGCs may represent the early stages of the presynaptic structure, which provides local platforms for synapse formation, as the axonal structural changes induced by GluD2 preferentially occurred in these regions 49 . In cultured Nrxn TKO CGCs, the density and intensity of Bassoon, synapsin, and VGluT1 puncta along the axons were severely decreased, suggesting a crucial role for NRXNs in the formation of the presynaptic-like structure (Fig. 8). These defects, as well as other defects, were rescued by NRXN1β(+S4), but not by its C-terminal mutant (Figs 6-8). Thus, it is reasonable to assume that the primary function of NRXNs in CGCs is the regulation of the organization of the axonal presynaptic-like structure through their C-terminal region. NRXNs interact with CASK and Mint through their C-terminal region 2 . However, the signal transduction pathway underlying this phenomenon has not been elucidated. Therefore, further studies are required to address this issue. We previously found that the number and intensity of Bassoon and VGluT1 in the axonal varicosities are increased after the application of the recombinant tetrameric GluD2 N-terminal domain in wild-type CGCs, but not in Cbln1 KO CGCs, suggesting that presynaptic differentiation is induced by the NRXN-Cbln1-GluD2 interaction 23 . It was also suggested that presynaptic differentiation is induced by the multimerization of NRXNs via their postsynaptic ligands, such as GluD2-Cbln1 or NLGN1 23,50 . In contrast, multimerization of NRXNs does not seem to be necessary for the formation of the presynapticlike structure, as NRXNs in the axons of CGCs lack contact with the postsynaptic ligands that are required for NRXN multimerization.
In conclusion, by analyzing conditional Nrxn TKO CGCs both in vivo and in vitro, we found novel function of NRXNs in CGCs survival. Our results suggest that, in CGCs, NRXNs play essential roles in the formation and function of presynaptic-like structures containing presynaptic proteins, independent of binding to their postsynaptic ligands. Moreover, our results suggest that the survival of CGCs is maintained by autocrine neurotrophic factors with a secretory machinery that is organized by NRXNs.

Western blot analysis
25 Cerebellar and forebrain homogenates were prepared from 8-week-old mice. The brains were homogenized with homogenate buffer containing 0.32 M sucrose, 1 mM NaHCO3, 1 mM MgCl2, 0.5 mM CaCl2, and protease inhibitors using a Teflon-glass homogenizer, followed by centrifugation at 710 × g for 10 min at 4°C, as described previously 54 . Proteins were quantified using the Quick Start Bradford Protein Assay (Bio-Rad). Equal amounts of protein were separated by SDS-PAGE, transferred to a PVDF membrane, and incubated with rabbit anti-pan NRXN (0.42 μg/ml) 55

Antibody production
The rabbit anti-synapsin I polyclonal antibody was produced by the SIGMA custom antibody production services. The sequence of the antigenic polypeptide was NYLRRRLSDSNFMANLPNGYMTDLQRPQP, corresponding to the amino-terminus of synapsin I 60 .
The specificity of the antibody was confirmed by western blotting using a mouse brain homogenate (Supplementary Figure 8).

Construction of expression vectors
The coding sequence of the double human influenza hemagglutinin (2×HA) epitope tag High-magnification images were deconvolved using Huygens Essential version (Scientific Volume Imaging).

Behavioral test
A rotarod apparatus consisting of a rod with a diameter of 3. than that of background signals on the same axon. Contiguous puncta were separated from each other using the "segmented particles" tool. Average fluorescence events before (2 min recording, 4 events) and after (12 min recording, 24 events) stimulation were quantified.

Ca 2+ imaging
For Ca 2+ imaging experiments, the CGCs were plated on a glass-bottom culture dish as described above. The

Statistical analysis
The results of at least two independent experiments were subjected to statistical analyses. No statistical method was used to determine sample size. Statistical significance was evaluated using the Kruskal-Wallis test followed by the post-hoc Steel-Dwass test, two-way or one-way ANOVA followed by Tukey's or Dunnett's post hoc test, or Student's t-test using the R software (R Core Team, 2017).
Statistical significance was assumed when P < 0.05.

Data availability
The data generated in this experiment are available from the corresponding author upon reasonable request.