Phosphatidylinositol 3,5-bisphosphate facilitates axonal vesicle transport and presynapse assembly

Abstract

Neurons relay information via specialized presynaptic compartments for neurotransmission. Unlike conventional organelles, the specialized apparatus characterizing the neuronal presynapse must form de novo. How the components for presynaptic neurotransmission are transported and assembled is poorly understood. Our results show that the rare late endosomal signaling lipid phosphatidylinositol 3,5-bisphosphate [PI(3,5)P₂] directs the axonal co-transport of synaptic vesicle (SV) and active zone proteins in precursor vesicles (PVs) in human neurons. PVs are distinct from conventional secretory organelles, endosomes, and degradative lysosomes and are transported by coincident detection of PI(3,5)P₂ and active ARL8 via kinesin KIF1A to the presynaptic compartment. Our findings identify a crucial mechanism mediating the delivery of SV and active zone proteins to developing synapses.

One-Sentence Summary:

A PI(3,5)P₂-regulated precursor organelle mediates the delivery of presynaptic building blocks for neurotransmission to developing synapses in human neurons.

Membrane-bounded compartments are a hallmark of eukaryotic cells (1, 2). In contrast to cell division that allows for organelles to be inherited from the mother cell (1, 2), during the differentiation of stem cells into neurons the specialized secretory apparatus for neurotransmission must form de novo. How synaptic vesicles (SVs) (3) and the presynaptic active zone (4, 5) scaffolds are formed, transported, and assembled into a functional presynaptic compartment for neurotransmission in developing mammalian central nervous system neurons is unknown (6, 7). Most SV proteins are transmembrane proteins that are somatically synthesized in the endoplasmic reticulum (ER) from where they are exported to the trans-Golgi-network (8), a compartment that is largely absent from axons and presynaptic nerve terminals (9). As a consequence, presynaptic biogenesis requires the formation of SV protein-containing precursor vesicles (PVs) that are axonally transported to the nascent presynapse (6, 10) via the kinesin KIF1A (called UNC104 in C. elegans and UNC104/ IMAC in D. melanogaster) (7, 11-16). Early studies of mouse peripheral axons in situ and cultured rat hippocampal neurons revealed the presence of heterogenously sized tubulovesicular structures that may represent presynaptic precursor organelles (17-19). Later works suggested the existence of specialized axonal transport vesicles for active zone proteins (20, 21). Whether the alleged carriers for SV and active zone proteins are common (22) or distinct organelles, what their cell biological identity is, and how this identity relates to the machinery for axonal transport remains to be fully elucidated.

Axonally transported PVs represent a distinct type of organelle.

We adopted a previously described protocol for the differentiation of human induced pluripotent stem cells (iPS) into glutamatergic neurons (iNs) (23) (Fig. S1A). Co-culture of these developing iNs with mouse astrocytes enabled the formation of functional pre- and postsynaptic compartments (Fig. S1B-E). Optical imaging of mature iNs (days in vitro [DIV] 28-30) expressing a chimera between the SV protein Synaptophysin and pH-sensitive pHluorin-GFP and stimulated with varying numbers of action potentials revealed potent exo-endocytic responses with kinetic parameters (Fig. S1F,G) similar to the kinetics of exo-endocytosis measured in mouse hippocampal neurons (24). Human iNs were capable of synaptic transmission and displayed short-term depression and asynchronous release in response to high-frequency stimulation (Fig. S1H). To follow the development of functional presynaptic compartments we monitored the axonal transport of presynaptic components in iNs using multicolor spinning disk confocal imaging paired with automated dual-channel tracking and kymographic analyses (Fig. S1A,I). The SV protein Synaptophysin-mRFP was present in diffraction-limited motile puncta, the majority of which underwent anterograde axonal transport (see Fig. 1B,C for endogenous Synaptophysin-eGFP). Anterogradely moving Synaptophysin puncta contained other SV proteins such as VAMP/ Synaptobrevin 2 (Fig. S1J,L) or the vesicular glutamate transporter 1 (VGLUT1) (Fig. S1L), suggesting that SV proteins might be axonally transported as a single entity. We found that further key components of the presynapse such as the active zone proteins Bassoon and MUNC13-1, or the presynaptic adhesion molecule Neurexin 1β also underwent anterograde axonal co-transport with SV proteins (Figs. S1K,L). Synaptophysin-mRFP was similarly co-transported with vGLUT1, the active zone proteins Bassoon and MUNC13-1, and Neurexin1β, but not with mitochondria in mouse hippocampal neurons (Fig. 1F). To better understand these findings, we generated genome-engineered iNs expressing Synaptophysin-eGFP from its endogenous locus (Fig. S1M). Endogenous Synaptophysin-eGFP was largely contained in diffraction-limited anterogradely moving puncta (Fig. 1B-E) together with other SV and active zone proteins (Fig. 1B-D; Movies S1,S2). Inhibition of protein synthesis depleted anterogradely moving Synaptophysin vesicles (Fig. 1E).

Figure 1 ∣. Nascent SV and active zone proteins are cotransported in iN axons.

(A) Scheme of the presynaptic compartment. (B,C) Kymographs showing the trafficking of endogenous SYP^endo-eGFP with VAMP2-SNAP (B) or MUNC13-1-SNAP (C). X-axis=30 μm, Y-axis=1 min. Vertical lines=static foci. Scale bar, 10 μm. (D) Fraction of anterogradely moving SYP^endo-eGFP vesicles co-trafficking with SYP-SNAP (87.3 ± 3.2%), VAMP2-SNAP (87.1 ± 4.3%), MUNC13-1-SNAP (88.4 ± 7.0%). n=4 experiments (≥ 50 vesicles each). (E) Number of anterogradely trafficking SYP^endo-eGFP vesicles (per min/ 30 μm length) in proximal axons of day 12-13 iNs following 24h incubation with DMSO, CHX or Anisomycin: DMSO, 11.5 ± 1.3; CHX 2.2 ± 0.2; Anisomycin, 2.1 ± 0.1. n = 3 experiments (≥ 50 vesicles each). One-way ANOVA followed by Dunnett’s post-test; p<0.0001 ****; p<0.01 **; p>0.05 ns. (F) Fraction of anterogradely moving vesicles co-labeled for SYP-mRFP and SYP-eGFP (92.8 ± 2.6%), vGlut1-venus (89.6 ± 3.3%), eGFP-Bassoon (79.2 ± 7.5%), MUNC13-1-eGFP (82.7 ± 5.9%), Neurexin1β-eGFP (83.4 ± 3.4%), or Mito-Halo (6.5 ± 1.8%) in proximal axons of mouse hippocampal neurons. n=3 experiments (≥ 50 vesicles each). (G) Kymographs illustrating co-trafficking of endogenous SYP^endo-eGFP with ARL8B-iRFP in iN. X-axis=30 μm. Y-axis=1 min. Vertical lines=static foci. Scale bars, 10 μm. (H) Fraction of anterogradely moving SYP^endo-eGFP vesicles co-labeled for LAMP1-mRFP (61.0 ± 10.4%) or ARL8B-iRFP (56.3 ± 6.3%). n=5 experiments (≥ 50 vesicles each). (I) Fraction of anterogradely moving LAMP1^endo-eGFP vesicles co-labeled for SYP-mRFP (44.0 ± 6.4%) or Syt1-mRFP (62.9 ± 8.9%). n=4 experiments (≥ 50 vesicles each). (J) 3D reconstruction of PVs (green) recruited to mitochondrion (red). Scale bar, 500 nm. (K) Ultrastructure of PVs recruited to mitochondria analyzed by FIB-SEM. * marks vesicles. Scale bars, 100 nm. (L) Cumulative plot of diameters of PVs and SVs. (M) Violin plot of the distribution of PV sizes. Whiskers show the min/max, box borders indicate 1^st and 3^rd quartiles, line indicates the median. Data are mean ± SEM.

Next, we tested whether axonally transported PVs contain elements of the secretory pathway. Markers of the ER such as SEC61β or KDEL, ARF1, a marker for Golgi-derived transport carriers, the cis-Golgi protein Giantin, trans-Golgi network protein 2, or RAB6, a marker for post-Golgi vesicles, were absent from axonally transported PVs (Fig. S2A,B). Anterogradely trafficked PVs were also segregated from axonally transported mitochondria (Fig. 1F; S2A,B). Axonally transported PVs displayed moderate co-transport with RAB5, a protein present on early endosomes and on SVs (3, 25), whereas the late endosomal marker RAB7 was largely absent from PVs. Axonally trafficked PVs were poorly accessible to recycling endosomal markers such as internalized transferrin (Fig. S2A,B). In contrast, anterogradely moving PVs in about 50-60% of cases contained late endosomal/ lysosomal membrane proteins such as LAMP1, CD63, and the lysosome-associated small GTPase ARL8B (Fig. S2A,B). Co-transport of PVs with lysosomal membrane markers such as ARL8B (Fig. S2B; Movie S3) and LAMP1 (Fig. S2B) was confirmed in iN endogenously expressing Synaptophysin-eGFP (Fig. 1G,H) and, conversely, in LAMP1-eGFP knock-in (KI) iN (Figs. 1I, S1M). As axonal PVs are involved in the delivery of newly synthesized presynaptic proteins, we expected them to be distinct from degradative lysosomes. Consistently, we found that anterogradely transported PVs lack lysosomal cathepsin activity (Fig. S2C) and are non-acidic (Fig. S2F). Lysotracker-positive degradative lysosomes containing endogenous LAMP1-eGFP were mostly found in neuronal somata, although motile acidic lysosomes were detectable in axons (Fig. S2D,E) [consistent with (26)].We conclude that axonally transported PVs are distinct from conventional secretory organelles, recycling endosomes, and mature lysosomes. Instead, they may represent a neuron-specific biogenesis organelle, which derives from a pathway that sorts lysosomal membrane proteins [see (26-28)].

To characterize PVs ultrastructurally, we devised a chemical genetic strategy to enrich these rare and transient organelles at a defined axonal location, thereby enabling their ultrastructural characterization by correlative light and electron microscopy (CLEM). We co-expressed the FRB domain of mTOR kinase targeted to the outer membrane of mitochondria (Mito-FRB) with a chimera between Synaptophysin and FK506-binding protein 12 (SYP-FKBP). In the absence of rapamycin SYP-FKBP localized to axonal PVs that displayed no colocalization with Mito-FRB-positive mitochondria (Fig. S2I,J). Rapamycin induced FRB/ FKBP heterodimerization and caused PVs marked by SYP-FKBP to be recruited to the mitochondrial surface. Sequestration of SYP-FKBP-containing PVs at mitochondria occurred in tandem with the mitochondrial accumulation of endogenous Piccolo, a major active zone protein, and the SV calcium sensor Synaptotagmin 1 (Fig. S2I,J). We then determined the ultrastructure of mitochondrially sequestered presynaptic PVs in proximal axons by correlative spinning disk confocal light and focused ion beam milling scanning electron microscopy (FIB-SEM) (Fig. S2G). FIB-SEM and 3-dimensional reconstruction (Fig. 1J) revealed the presence of a population of vesicles and tubules in direct contact to the mitochondrial surface (Fig. 1K, I-VI). Quantitative morphometric analysis showed that > 90% of these had diameters of 50-400 nm (Fig. 1L) with a mean size of 166 ± 14 nm and a median diameter of 87 nm (Fig. 1M), a size distribution distinct from that of mature SVs (Fig. 1L). Minor fractions of vesicles contained electron-dense material (5%) (Fig. 1K, III) or comprised large multivesicular bodies (6%) (Fig. 1K, VI). The latter may originate from the degradative pathway and are unlikely to relate to presynaptic biogenesis. No vesicle or tubule accumulations were found in the absence of rapamycin (Fig. S2H). These data (Fig. 1J-M) show that presynaptic biogenesis in mammalian CNS neurons involves vesicular and tubular compartments of 67 nm (Q1-lower quartile) to 220 nm (Q3 -upper quartile), consistent with the reported ultrastructure of alleged presynaptic transport packets in mouse saphenous nerves in situ (17) and in rat hippocampal axons (19).

Anterograde axonal transport and presynaptic delivery of PVs depend on ARL8A/B and KIF1A.

Prior studies in invertebrate models (29-31) have implicated the small GTPase ARL8 (compare Fig. 1), a regulator of anterograde lysosome motility (32, 33), in presynaptic biogenesis. If the organelles identified ultrastructurally (Fig. 1J-M) represent PVs, their axonal transport should require ARL8. We generated knockout (KO) iPS lines lacking either one or both human isoforms of ARL8 (termed ARL8A and ARL8B) (Fig. S3A). Complete loss of ARL8A and B in three independent double KO lines strongly impaired anterograde axonal transport of Synaptophysin-positive PVs (Figs. 2A,B; S3A,B; Movie S4), a phenotype restored by re-expression of ARL8A- or ARL8B-c-myc (Fig. 2A,B; Fig. S3D). Single loss of ARL8B alone produced a mild phenotype (Fig. S3C). We conducted an unbiased proteomic screen for ARL8 interacting factors. This analysis identified SV cargos and KIF1A (Fig. 2C,D), a kinesin implicated in axonal transport of SV proteins (12-16, 34). KIF1A-eGFP underwent efficient co-transport with SV and active zone proteins and with ARL8B (Fig. 2E,F). CRISPR/Cas9-mediated KO (Fig. S3E) or shRNA-mediated knockdown (Fig. S3F,L) of KIF1A severely compromised anterograde transport of PVs and this defect was restored by re-expression of wild-type KIF1A but not by a disease-associated KIF1A rigor mutant (32, 35) (Fig. 2G,H; Movie S5). A similar blockade in anterograde axonal transport of PVs was observed in primary mouse hippocampal neurons expressing the KIF1A rigor mutant (Fig. 2I,J). Anterograde axonal PV transport proceeded nearly unperturbed in the near absence of classical kinesin KIF5B or the kinesin-adaptors DENN/MADD (Fig. S3G-I,L) and was moderately reduced upon depletion of FYCO1 (Fig. S3J,M). To probe the interplay between ARL8A/B and KIF1A, we monitored the localization and axonal transport dynamics of the C-terminal membrane-binding (36) tail domain of KIF1A (KIF1A^tail). KIF1A^tail-eGFP colocalized with Synaptophysin-mRFP-containing PVs in wild-type neurons, but was largely diffusely localized in ARL8A/B double KO iN. Re-expression of hyperactive ARL8B•GTP (QL) but not inactive ARL8B•GDP (TN) rescued defective KIF1A^tail-eGFP localization to anterogradely moving PVs (Fig. S3N,O). Loss of KIF1A had no substantial impact on the recruitment of ARL8B to PVs expressing Synaptophysin-eGFP (Fig. S3P,Q).

Figure 2 ∣. Axonal transport of PVs is controlled by ARL8A/B and KIF1A.

(A) Kymographs of trajectories of SYP-eGFP vesicles in WT or ARL8A/B double KO iN. X-axis=30 μm. Y-axis=1 min. Scale bar, 10 μm. (B) Number of anterogradely moving SYP-eGFP vesicles (per min/ 30 μm length): WT, 9.2 ± 0.6; WT + ARL8B-Myc, 8.7 ± 1.4; ARL8A/B double KO, 2.6 ± 0.9; ARL8A/B double KO + ARL8B-Myc, 7.9 ± 1.7. n = 3 experiments (≥ 50 vesicles each). One-way ANOVA followed by Dunnett’s post-test; p<0.0001 ****; p<0.01 **; p>0.05 ns. (C) Scatter plot of ARL8B-mCherry interacting proteins in iN. Blue, lysosomal proteins; green, kinesins; orange, SV proteins. (D) Co-immunoprecipitation of ARL8B-mCherry with KIF1A-eGFP in developing iNs. Bound proteins were detected by immunoblotting. Input, 5% of total lysate. (E) Kymographs illustrating co-trafficking of KIF1A-eGFP with SYP-mRFP (top) or ARL8B-mCherry (bottom) in iN. X-axis=30 μm. Y-axis=1 min. Scale bars, 10 μm. (F) Fraction of anterogradely moving SYP-mRFP (76.9 ± 6.5%), MUNC13-1-SNAP (79.7 ± 5.9%), or ARL8B-mCherry (64.1 ± 10.1%) puncta co-trafficking with KIF1A-eGFP. n = 5 experiments (≥ 50 vesicles each). (G) Kymographs of trajectories of SYP-eGFP vesicles in iNs that are WT, KIF1A KO, or KO expressing full-length KIF1A or Rigor mutant KIF1A. X-axis=30 μm. Y-axis=1 min. Scale bar, 10 μm. (H) Number of anterogradely moving, retrogradely moving, or stationary SYP-eGFP vesicles (per min/ 30 μm length): WT, 9.7 ± 1.7; KIF1A KO, 2.4 ± 0.5; KO + KIF1A-FL, 8.7 ± 0.3; KO + KIF1A-rigor, 2.4 ± 0.3. n = 3 experiments (≥ 50 vesicles each). One-way ANOVA followed by Dunnett’s post-test; p<0.0001 ****; p<0.01 **; p>0.05 ns. (I) Kymographs showing trajectories of SYP-mRFP vesicles in primary mouse hippocampal neurons expressing full length KIF1A or Rigor mutant KIF1A. Scale bar, 10 μm. (J) Number of anterogradely moving SYP-mRFP vesicles (per min/ 30 μm length) in mouse hippocampal neurons: KIF1A-FL, 8.1 ± 1.1; KIF1A-Rigor, 2.5 ± 0.8. n = 3 experiments (≥50 synaptic puncta each). t test; p<0.0001 ****; p<0.01 **; p>0.05 ns. Data are mean ± SEM.

If anterogradely transported PVs represented precursor organelles for presynapse assembly, stalling their axonal transport should impair synapse assembly. Indeed, ARL8A/B double KO iNs suffered from a severe reduction in the number of Bassoon puncta per dendrite length, while the density of postsynaptic Homer1 puncta was unaffected (Fig. 3A,B). Moreover, in comparison to synapses from WT neurons, ARL8A/B double KO iNs displayed a profound overall reduction in the amounts of SV and active zone proteins (Fig. 3C-F; S4). ARL8A/B loss did not affect postsynaptic glutamate receptors (Fig. S4B,C,I) or Homer1 (Fig. 3C,D,H, S4). ARL8A/B loss also did not impact the delivery of presynaptic voltage-gated P/Q-type calcium channels (Fig. 3G; S4D), suggesting that P/Q-type calcium channels may be transported to synapses via a distinct mechanism. Lentiviral re-expression of ARL8B rescued defective presynaptic delivery of SV and active zone proteins in ARL8A/B double KO iN (Fig. 3A-F; S4). Loss of KIF1A phenocopied ARL8A/B double KO with respect to reduced density of presynaptic Bassoon puncta, whereas postsynaptic Homer1 puncta density was unchanged (Fig. 3I; S5A). Synapses identified by Homer1, PSD-95 or GLUN1 further showed a severe reduction in the amounts of SV and active zone proteins (Fig. 3J,K;S5B,C,E,G), whereas presynaptic calcium channels (Fig. 3L; S5D), Homer1 (Fig. S5B,E,F) or postsynaptic NMDA receptors (Fig. 3M; S5C) were unaltered. Defective presynaptic delivery of SV and active zone proteins was restored by re-expression of KIF1A (Fig. 3J-K; S5B,C,E,G). Similar defects in presynaptic biogenesis determined by VGLUT1 immnuostaining were observed in D. melanogaster larval neuromuscular junction (NMJ) synapses from unc104 (the KIF1A ortholog) hypomorphic animals (Fig. 3N; S5H,I). Defective presynaptic biogenesis in KIF1A KO and ARL8A/B double KO iNs were paralleled by severely reduced stimulus-evoked postsynaptic calcium responses measured by Xph20-GCaMP7f (Fig. 3O,P), likely as a consequence of impaired presynaptic glutamate release. Consistently, overexpression of the transport-defective KIF1A rigor mutant impaired SV exocytosis in primary mouse hippocampal neurons (Fig. S5K). In contrast, presynaptic calcium influx was unaltered in KIF1A KO or ARL8A/B double KO iN (Fig. S5J), in agreement with the lack of effect of ARL8A/B or KIF1A loss on the amounts of presynaptic voltage-gated calcium channels (Figs. 3G,L, S4D,S5D).

Figure 3 ∣. Loss of ARL8A/B or KIF1A/ Unc104 impairs presynaptic biogenesis and function.

(A) Representative confocal images of WT iNs, ARL8A/B double KO iNs or double KO iNs expressing ARL8B-c-myc at day 30 immunostained for presynaptic Bassoon, postsynaptic Homer1 and MAP2. Scale bar, 5 μm. (B) Number of synapses and of pre- and postsynaptic puncta per 20 μm dendrite length. n = 3 experiments ( ≥50 puncta each). (C,D) Representative confocal images of WT iN, ARL8A/B double KO iN or double KO iN expressing ARL8B-c-myc at day 30 immunostained for Synaptophysin (C), Bassoon (D) and Homer1 (C,D). Scale bars, 10 μm. (E-H) Quantified mean fluorescence intensity of pre- and postsynaptic proteins in WT iN, ARL8A/B double KO iN or double KO iN expressing ARL8B-c-myc (day 30). n=3 experiments (≥100 synapses each). SYP, Synaptophysin (E), Bassoon (F), Cav2.1 (G), Homer1 (H). (I) Number of synapses and of pre- and postsynaptic puncta per 20 μm dendrite length in WT iNs, KIF1A KO iNs or KO iNs expressing KIF1A (day 30). n = 3 experiments (≥50 puncta each). (J-M) Quantified mean fluorescence intensity of pre- and postsynaptic proteins in WT iNs, KIF1A KO iNs or KO iNs expressing KIF1A (day 30). See (E-H) for abbreviations. n = 3 experiments (≥100 synaptic puncta each). (N) Defective presynaptic biogenesis in absence of Unc104-mediated PV delivery in D. melanogaster. Reduced amounts of VGLUT at Unc104 mutant NMJs. WT, 100 ± 11; Unc104, 6.3 ± 0.8. n = 10 (WT) and 10 (Unc104) NMJs. t test; *P < 0.05, *P < 0.01, ****P < 0.0001. (O) Schematic representation of pre- and postsynaptic calcium sensors SYP-CGaMP6f and Xph20-CGaMP7f. (P) Mean ΔF/F0 response of Xph20-CGaMP7f to 100 action potentials (APs) (10 Hz, 10 s) stimulation in WT, ARL8A/B double KO or KIF1A KO iNs. Data were normalized to pre-stimulation. n=3 independent experiments. (B, E-M) One-way ANOVA followed by Dunnett’s post-test; p<0.0001 ****; p<0.01 **; p>0.05 ns. All data represent mean ± SEM.

Presynaptic biogenesis in human neurons is thus mediated by ARL8A/ KIF1A-dependent axonal transport and delivery of PVs to nascent synapses.

Anterograde axonal transport of PVs is controlled by the rare signaling lipid PI(3,5)P₂.

Conflicting data regarding the role of the BORC complex, a postulated upstream activator of ARL8 (32, 37), in axonal transport of SV proteins have been reported (29, 30, 38, 39). We found that lentiviral knockdown of the BORC subunits Myrlysin or Diaskedin impaired anterograde axonal transport of LAMP1-positive lysosomes (Fig. S6A-D), but had little effect on anterograde motility of PVs expressing endogenous Synaptophysin-eGFP (Fig. S6E,F). We generated BORCS5 RNAi lines in D. melanogaster. In contrast to loss of ARL8 (30), depletion of BORCS5 had no major effect on the presynaptic staining intensity of the SV protein VGLUT or on NMJ area (Fig. S6I-K), although BORCS5 loss impaired autophagy/ lysosome-mediated turnover of p62 (Fig. S6L-N). Surprisingly, depletion of the BORC-specific subunits Myrlysin or Diaskedin rescued defective anterograde transport of PVs in ARL8A/B double KO neurons (Figs. 4A, S6H), suggesting that loss of Myrlysin or Diaskedin causes the upregulation of an endogenous mechanism that renders PV transport ARL8A/B-independent. Recent data from fibroblasts show that loss of BORC boosts the amounts of the rare late endosomal/ lysosomal signaling lipid PI(3,5)P₂ (40-42) via a poorly understood mechanism (37). We speculated that elevation of PI(3,5)P₂ in absence of BORC might promote anterograde PV transport. Depletion of Myrlysin in iPS cells increased PI(3,5)P₂ monitored by ratiometric imaging via a fluorophore-conjugated engineered p85α-cSH2 domain from class I PI 3-kinase (ep85α-cSH2) as a sensor (43) (Figs. 4B, S7A). To directly probe the function of PI(3,5)P₂ in anterograde axonal PV transport, we analyzed the effects of pharmacological inhibitors of PI(3,5)P₂ or PI(3)P synthesis. Inhibition of PI(3,5)P₂ synthesis by the PI(3)P 5-kinase PIKFYVE inhibitor Apilimod (Fig. 4C) or of its precursor PI(3)P (see Fig. S7D) by VPS34-IN1, SAR405, or Compound 19 (Figs. 4C, S7B,C) potently reduced the fraction of anterogradely moving PVs. Moreover, Apilimod caused a small reduction in transport velocity of the remaining motile PVs (mean±SEM [μm/s]: DMSO ctrl: 1.825 ± 0.1339 vs. Apilimod: 1.595 ± 0.1306). In contrast, anterograde axonal PV transport proceeded unperturbed upon inhibition of PI(4)P synthesis by PI4KIIIβ-IN10 or blockade of type II phosphatidylinositol 5-phosphate 4-kinases (Fig. S7B,C). Anterograde transport of Synaptophysin-eGFP was also inhibited upon genetic inhibition of PI(3,5)P₂ synthesis in PIKFYVE KO iNs (Figs. 4D; S7E,F), depletion of PIKFYVE by lentiviral shRNA (Figs. 4A, S6G,H; Movie S6), or hydrolysis of PI(3)P/ PI(3,5)P₂ on PVs by Rapalog-induced local recruitment of the lipid phosphatase MTMR2 (Fig. S7G,H). PIKFYVE depletion abrogated anterograde axonal transport of PVs in Myrlysin- or Diaskedin-depleted ARL8A/B double KO iN (Fig. 4A). Next, we tested whether PIKFYVE acts locally on PVs to promote axonal transport. Endogenous PIKFYVE co-precipiated with Syp^endo-eGFP from iN lysates (Fig. S8A) and endogenous mCherry-PIKFYVE^endo (Fig. S8B) colocalized with PVs expressing VAMP/ Synaptobrevin 2 (Pearson 0.51±0.03) or Synaptotagmin1 (Pearson 0.44±0.02) in PIKFYVE KI iNs (Fig. 4F). Moreover, axonal PVs co-localized with the PI(3,5)P₂-sensing engineered p85α-cSH2 domain in live iNs (Fig. 4G) and co-stained with recombinant SNXA, a selective PI(3,5)P₂-binding protein (44) in fixed iNs (Fig. S8C). Colocalization of PVs with either of these probes was abrogated by inhibition of PI(3,5)P₂ synthesis (Figs. 4G, S8C).

Figure 4 ∣. PI(3,5)P₂ synthesis regulates presynaptic biogenesis.

(A) Motility of SYP-mRFP vesicles in WT and ARL8A/B double KO iNs transduced with control lentivirus (shCTRL) or following lentiviral knockdown of Myrlysin, Diaskedin, or PIKFYVE. n = 3 experiments (≥ 50 vesicles each). (B) PI(3,5)P₂ measured by quantitative ratiometric imaging in situ: WT, 0.05 ± 0.004; shMyrlysin, 0.15 ± 0.01. t test; *P < 0.05, **P < 0.01, ****P < 0.0001. (C) Depletion of PI(3)P or PI(3,5)P₂ inhibits anterograde transport of SYP^endo-eGFP vesicles in iNs. Number of SYP^endo-eGFP vesicles (per min/ 30 μm length) moving anterogradely, retrogradely, or remaining stationary per minute in iNs treated (30 min) with DMSO, VPS34-IN1 or Apilimod. DMSO, 11.1 ± 2.2; VPS34-IN1, 1.7 ± 0.9; Apilimod, 3.6 ± 1.1. n = 5 experiments (≥ 50 vesicles each). (D) Number of anterogradely-moving SYP-eGFP vesicles in day 12-14 iNs (per min/ 30 μm length): WT, 8.5 ± 0.9; PIKFYVE KO #1, 3.8 ± 0.05; PIKFYVE KO #2, 2.5 ± 0.4; PIKFYVE KO #3, 4.3 ± 0.2. n = 3 experiments (≥ 50 vesicles each). (E) 2xPH-(KIF1A) associates with PI(3,5)P₂ liposomes. Bound fraction (% of total): No liposomes, 1.7 ± 0.6%; no PI, 5.5 ± 2.0%; PI(3,5)P₂, 60.9 ± 4.0%; PI(4,5)P₂, 12.9 ± 5.5%; PI(3)P, 24.0 ± 8.1%. n = 4 experiments. (A,C,D,E) One-way ANOVA followed by Dunnett’s post-test; p<0.0001 ****; p<0.01 **; p>0.05 ns. (F) Colocalization of endogenous mCherry-PIKFYVE with VAMP2 or SYT1 (green) in iN axons. Scale bar, 10 μm. Bottom, line intensity scans from 10 μm segments. (G) Representative confocal images of iN axons expressing SYP-mRFP and microinjected with the PI(3,5)P₂-sensing fluorophore-labeled ep85α-cSH2 domain. Scale bar, 10 μm. Bottom, line intensity scans from 10 μm segments. (H) Confocal images of WT and Fab1 mutant NMJs stained for VGLUT (green) and HRP as axonal membrane marker (magenta). Top: overview. Scale bars, 5 μm. Insets, zooms. Scale bars, 2 μm. (I) Quantification of representative data shown in (H). WT, 100 ± 11.7; Fab1-RNAi, 51.2 ± 5.5. n=10 NMJs each. t test; *P < 0.05, **P < 0.01, ****P < 0.0001. Data are mean ± SEM.

KIF1A harbors a phosphoinositide-binding pleckstrin-homology (PH) domain (45). We found that a KIF1A mutant lacking its PH domain fails to rescue defective axonal transport of PVs in KIF1A KO iN (Fig. S9A,B; Movie S5). GST-fused recombinant PH domain of KIF1A directly bound to PI(3,5)P₂ and, to a lesser extent, to PI(3)P in vitro (Fig. 4E; S9C). When expressed in human iNs, KIF1A^tail-GFP (Fig. S9D,E) or its PI(3,5)P₂-binding PH domain (Fig. S9H,I) were recruited to anterogradely moving axonal PVs. Inhibition of PI(3,5)P₂ synthesis or genetic depletion of PIKFYVE caused the partial dissociation of KIF1A^tail-GFP (Fig. S9D-G) or mCherry-PH-(KIF1A) (Fig. S9H,I) from motile PVs. A PH domain mutant of KIF1A lacking the ability to bind PI(3,5)P₂ (Fig. S9J,K) displayed a greatly reduced ability to associate with axonally transported PVs (Fig. S9H,I). Rapalog-induced hydrolysis of PI(3)P/ PI(3,5)P₂ on PVs by MTMR2 displaced full-length KIF1A-eGFP from PVs (Fig. S9L). These data indicate that PI(3,5)P₂ among possible other functions might contribute to recruiting KIF1A to axonal PVs.

Lastly, we analyzed the long-term consequences of reduced PI(3,5)P₂ synthesis for presynaptic biogenesis. iNs depleted of PIKFYVE displayed reduced synaptic amounts of the SV marker Synaptophysin and the active zone protein Bassoon (Fig. S10A-D), whereas postsynaptic Homer1 remained unchanged (Fig. S10E). Genetic depletion of the D. melanogaster PIKFYVE ortholog Fab1 partially phenocopied loss of KIF1A/UNC104 (compare Fig. 3N), showing a reduction in SV proteins such as VGLUT (Fig. 4H,I) and a, possibly compensatory, increase in NMJ length (Fig. S10F-I).

Discussion

The biology behind the origin of transport organelle(s) responsible for the assembly of the presynaptic compartment for neurotransmission in the mammalian CNS and in humans is a long-standing matter of investigation that remains to be fully elucidated. Our findings demonstrate that mammalian presynaptic biogenesis occurs by axonal transport of PVs marked by the endolysosomal signaling lipid PI(3,5)P₂ (Fig. S11). These PVs carry not only endogenous SV cargos but also active zone proteins as well as β-neurexin to nascent presynapses, consistent with data in D. melanogaster (30), in C. elegans (31, 46), and with earlier studies in mammalian CNS neurons (19, 22). PVs comprise surprisingly heterogenous vesicular and tubular carriers that resemble the organelles accumulated in rat proximal axons in response to anterograde transport blockade observed decades ago (17), yet, are clearly distinct in morphology, shape, and size from mature SVs. PVs are devoid of markers of the secretory pathway and distinct from recycling endosomes and from mature degradative lysosomes. Instead, PVs represent a neuron-specific organelle, which may derive from a pathway that sorts lysosomal membrane proteins (28, 47, 48).

We further show that axonal transport of presynaptic PVs depends on the poorly understood rare signaling lipid PI(3,5)P₂. P(3,5)P₂ has been found to affect multiple organelles within the endolysosomal system including late endosomes and lysosomes, macropinosomes (40-42) and recycling endosomes (49). Moreover, reduced PIKFYVE function was found to impair synapse formation (50). The ARL8-KIF1A-PI(3,5)P₂ pathway for presynaptic biogenesis may thus reflect the relationship of SVs and related secretory organelles to the endolysosomal system (51) as underscored by the presence of proteins implicated in lysosome function on SVs including ARL8A/B, LAMP5, LAMP1, LAMP2, the neuronal AP-3 complex, ATG9A, RAB26, and VPS34 (52).

Mutations in KIF1A are linked to epilepsy, hereditary spastic paraplegia, i.e. a rare gait disorder caused by axonal defects/ axonal degeneration in the spinal cord (53-55) and to the peripheral neuropathy Charcot-Marie-Tooth Disease (56). We thus speculate that defects in PI(3,5)P₂-regulated axonal PV transport may be linked to some of these rare inherited neurological disorders and peripheral neuropathies.