Rab2 regulates presynaptic precursor vesicle biogenesis at the trans-Golgi

Presynaptic precursors deliver synaptic proteins to nascent and growing synapses. The authors provide evidence that during an early precursor formation step, the small GTPase Rab2 regulates immature precursors formation at the trans-Golgi prior to subsequent Arl8-dependent maturation steps.

In the present study, we aim to identify novel regulators of precursor biosynthesis. We hypothesized that precursor biogenesis requires membrane remodeling enzymes, such as Rab (Ras-related in brain) proteins, small GTPases controlling vesicle budding, and fusion via effector protein recruitment (Stenmark, 2009;Zhen and Stenmark, 2015). Rab proteins function as molecular switches by changing between a GTP-bound active state to a GDP-bound inactive state (Stenmark, 2009;Zhen and Stenmark, 2015). We here show that presynaptic precursors originate from the trans-Golgi in an early step of the biosynthetic assembly pathway. This process requires the highly conserved, Golgi-related small GTPase Rab2 for precursor assembly and maturation. Rab2 is reported as a Golgi resident and acts bidirectionally in the ER to Golgi trafficking (Liu and Storrie, 2012;Saraste, 2016;Tisdale and Balch, 1996). In conjunction with the previously shown function of Rab2 and Rab2 effectors in the promotion of dense core vesicle (DCV) maturation in neurons (Ailion et al., 2014;Edwards et al., 2009;Hannemann et al., 2012;Sumakovic et al., 2009) and the association of rab2 mutations with neurodevelopmental defects, e.g., autism spectrum disorders, schizophrenia (Kiral et al., 2018;Takata et al., 2016), and memory and prefrontal morphology defects (Li et al., 2015), in humans, our data identify Rab2 as a crucial factor in early presynaptic precursor biogenesis.

Results
We performed a motoneuron-specific RNAi knockdown of neuronally expressed Rab proteins (Chan et al., 2011) analyzing AZ and SV protein distribution in motoneuron somata and terminals of Drosophila third instar larvae, identifying exclusively Rab2 as a potential regulator of precursor biogenesis.
Presynaptic proteins accumulate in the cell bodies of rab2 −/−deficient neurons Subsequently, we used a rab2 −/− mutant created by ORF excision (Chan et al., 2011) to establish a complete loss-of-function background. Mutant Rab2 protein level was evaluated by Western blot analysis ( Fig. S1 A), proving the absence of detectable Rab2 protein in the rab2 −/− null mutant. Mutant animals died at the early third larval stage, consistent with rab2 −/− null mutants previously described (Lőrincz et al., 2017). We performed immunofluorescence analysis in motoneuron cell bodies located in the cortex of larval ventral nerve cords (VNC) of rab2 −/− mutants. We detected a striking ectopic accumulation of the AZ scaffold protein Bruchpilot (BRP; Kittel et al., 2006;Wagh et al., 2006) with a strong increase of BRP protein level (Fig. 1, A and B). BRP did not accumulate diffusely in the cytoplasm, but in distinct, large aggregates not observed in controls (Fig. 1 A,zoom;and Fig. 1 C). We further analyzed the rab2 −/− mutants for a set of presynaptic proteins, including another AZ scaffold protein, RIM-binding protein (RIM-BP; Liu et al., 2011). RIM-BP also accumulated in rab2 −/− mutant somata (Fig. 1,D and E;and Fig. S1,B and G), and the large, ectopic RIM-BP accumulations overlapped with the BRP signal, while smaller RIM-BP dots also observed in the control were not BRP-positive. The same phenotype was observed for the release factor (m)UNC13A (Böhme et al., 2016;Fig. S2, A and B). Furthermore, SV proteins, here the vesicular glutamate transporter protein (VGlut; Santos et al., 2009), were also strongly accumulating in the rab2 −/− motoneuron somata (Fig. 1,F and G;and Fig. S1,C and H). However, in contrast to RIM-BP or UNC13A, VGlut aggregates were 2.5-fold larger than BRP aggregates (Fig. S1 D) and, interestingly, localized rather adjacent to the BRP accumulations. Other SV proteins, Synaptotagmin-1 (Syt-1; Zhou et al., 2017;Fig. S2, C and D) and Dap160/intersectin (Pechstein et al., 2010;Fig. S2, E and F), accumulated with a similar distribution. RNAi mediated knockdown of Rab2 verified by Western blot analysis (Fig. S3 A) consistently provoked significant ectopic accumulations of both AZ and SV proteins (BRP,VGlut, in motoneuronal cell bodies (Fig. S3, B-H). Unlike SV and AZ proteins, levels of the endogenous mitochondrial ATP synthase, which also traffics from the cell body to the neuronal terminal, were not increased in rab2 −/− mutant motoneuronal somata (Fig. S2,G and H), implying that loss of Rab2 could specifically impair presynaptic protein export, while other trafficking pathways are not affected.
We showed previously that axonally trafficking BRP-positive precursors are copositive for the lysosomal-associated membrane protein LAMP1 (Vukoja et al., 2018). We tested for LAMP1 protein levels in rab2 −/− mutants by expressing a GFP-LAMP1 construct (Pulipparacharuvil et al., 2005) in motoneurons and observed a strong increase of LAMP1 in rab2 −/− mutant cell bodies (Fig. 1,H and I;and Fig. S1,E and I). Similar to SV proteins, LAMP1-positive aggregates were larger and localized rather adjacent to ectopic BRP accumulations. Our findings agree with previous observations showing an increase of LAMP1 signal in larval brains of rab2 −/− mutant (Lund et al., 2018). Importantly, the ectopic presynaptic protein accumulations of rab2 −/− -deficient cells were not positive for standard markers of the lysosomal and autophagosomal system: Cathepsin L ( Finally, we could rescue the BRP accumulation phenotype back to wild-type BRP levels by reexpression of Rab2 at endogenous protein levels into rab2 −/− mutants (Fig. 1, L-N), proving the cell-autonomous nature of the rab2 −/− phenotype.
Two-electrode voltage clamp (TEVC) electrophysiological recordings showed that single action potential evoked response was reduced to half of the evoked excitatory junctional currents (eEJCs) in rab2 −/− mutants ( Thus, reduction of presynaptic material available at the synaptic terminal in the absence of Rab2 clearly restricts neurotransmitter release. However, those AZs still forming remain apparently functional, in accordance with the overall normal AZ architecture confirmed by microscopy. Rab2 is a component of anterogradely trafficking presynaptic precursors Could Rab2, as a membrane anchored GTPase, localize to in vivo trafficking precursors? By use of a genomic Rab2-GFP construct (genRab2-GFP), with the GFP inserted in the Rab2 ORF (Lund et al., 2018), we colabeled larval axons for BRP and observed distinct Rab2-positive puncta, partially copositive for BRP ( Fig. 3  A), also shown by line profile (Fig. 3 B) and Pearson's correlation coefficient (Fig. 3 C). Similarly, we detected puncta copositive for genRab2-GFP and VGlut (Fig. 3, D-F), Syt-1 (Fig. S5, A-C), or the lysosomal marker Spinster (Rong et al., 2011;Sweeney and Davis, 2002;Fig. 3, G-I). Rab2 in a GTP locked constitutively active form (Rab2 QL -YFP) expressed in motoneurons showed a similar colocalization profile (Fig. S5, D-L). Next, we directly monitored axonal trafficking in vivo in intact larvae (Andlauer and Sigrist, 2012a), therefore coexpressing strawberry-tagged "BRP short" (BRP D3 -straw; Fouquet et al., 2009) and Rab2-YFP or constitutively active Rab2 QL -YFP (Fig. 3,J and K;and Fig. S5,M and N). 50% of anterograde trafficking BRP-positive vesicles were copositive for Rab2 (wild-type) and 100% copositive for constitutively active Rab2 QL (Fig. 3 S5 T). As a membrane component of anterogradely trafficking precursor vesicles, we should expect Rab2 to reach the synaptic terminal. Indeed, we observed genRab2-GFP in a punctate pattern at the synaptic terminals, localizing mostly adjacent to the AZ (Fig. 3, R and S). Interestingly, Rab2 might here identify cargo-loaded precursors "waiting to be discharged" and/or vesicles destined for retrograde transport.
In summary, we provided evidence that Rab2 is associated to presynaptic precursors, supporting the hypothesis that ectopic presynaptic protein accumulations in rab2 −/− mutants consist of immature, trafficking-incompetent precursors.
Rab2 is described as a Golgi resident, and its depletion is reported to affect Golgi apparatus integrity (Aizawa and Fukuda, 2015;Maringer et al., 2016). Indeed, the number of Golgi complexes (measured by GM130) was increased by 20% ( Fig. 4 E), while Golgi area was decreased by 35% ( Fig. 4 F) in rab2 −/− mutants, pointing toward a Golgi fragmentation defect. GM130 protein level was not altered (Fig. 4 G). As rab2 −/− mutants showed an altered Golgi architecture, it is conceivable that generic protein biogenesis or general post-Golgi sorting defects could cause the ectopic presynaptic protein accumulation we observed at the trans-Golgi. To address this question, we first used RNAi lines targeting proteins of the coat protein II and I (COPII and COPI) complexes, implied in Golgi biogenesis, maintenance, and consequently function (Kondylis and Rabouille, 2009;Popoff et al., 2011;Ward et al., 2001), thus generically disrupting Golgi function. We restricted protein knockdown to the motoneurons to circumvent survival deficits. If generic Golgi disruption accounts for the rab2 −/− mutant phenotype, we should observe presynaptic protein aggregation upon Golgi disruption. Depletion of the COPII complex protein Secretory 23 (Sec23), a component of the COPII vesicle coating (Fromme et al., 2008), caused a much stronger Golgi fragmentation than Rab2 depletion ( Fig. 4 H), with an 87% increase in GM130 aggregate number Second, we asked if other than presynaptic proteins were also ectopically aggregating in Rab2-depleted somata and hence analyzed the transsynaptic membrane glycoprotein Neuroglian (Nrg) and the early endosomal marker Rbsn-5. Nrg protein remained bound to the plasma membrane in rab2 −/− mutant somata and did not aggregate into ectopic BRP accumulations, although the overall protein level was increased (Fig. 4, V and W). Nrg protein levels at the synaptic terminals were in fact increased, in contrast to the synaptic proteins being depleted in rab2 −/− mutants (Fig. 4, X-Z). In rab2 −/− somata labeled with Rbsn-5, uncommonly large Rbsn-5-positive organelles formed, which, however, were not copositive for BRP (Fig. S7, A-C). They likely represent enlarged early endosomes reflecting increased fluxes from the Golgi to early endosomes. Consistently, Rbsn-5 levels at Rab2-deficient terminals were not altered (Fig. S7,D and E). Thus, in the absence of Rab2, nonsynaptic proteins are still targeted to their membrane destination, and sorting from the Golgi is apparently not compromised, confirming that Rab2 specifically regulates presynaptic protein sorting from the trans-Golgi.
Thus, accumulating presynaptic proteins at the trans-Golgi of rab2 −/− mutant somata could represent immature precursors, interrupted in their biosynthetic maturation process due to loss of the membrane trafficking regulator Rab2.
Tubule-shaped vesicular membranes accumulate at the trans-Golgi in rab2 −/− mutants For a detailed evaluation of the presynaptic protein accumulations in rab2 −/− mutants, we performed ultrastructural analysis by EM. Electron micrographs showed a striking accumulation of circular-to oval-shaped vesicular membranes at the trans-Golgi (Fig. 6, A  , and the volume fraction of trans-Golgi networkassociated vesicles of rab2 −/− mutants compared with wild type was significantly increased (Fig. 6 C).
But how do these maturation-arrested, immature precursors compare with the PLVs we described previously to accumulate in the cell body of arl8 −/− mutants (Vukoja et al., 2018)? The direct comparison of arl8 −/− mutant and rab2 −/− mutant electron micrographs revealed that PLVs detected in the absence of Arl8 were, with a ∼60-nm short diameter, significantly larger than rab2 −/− mutant precursors at ∼40 nm, possessed a more electron dense core (Fig. 6, B and D), and had a clearly more uniform, circular shape. Long diameters of both mutants were similar at ∼75 nm (Fig. S8 C). Notably, the volume fraction taken by mitochondria (Fig. 6 E) was not altered in  in agreement with the confocal data (Fig. S1, Q and R), again showing that absence of Rab2 does not affect anterograde transport generically. Interestingly, we detected a threefold increase in the incidence of somatic structures resembling autophagocytic organelles in rab2 −/− mutants ( Fig. S8 D), potentially interesting in the context of the previously described Rab2 function in the autophagic degradative pathway (Ding et al., 2019;Fujita et al., 2017;Lőrincz et al., 2017;Lund et al., 2018).
We performed 3D electron tomography to evaluate more deeply the apparent shape differences of precursor vesicles in rab2 −/− and arl8 −/− mutants. Clearly, immature precursors in rab2 −/− mutants localized, in agreement with the confocal and STED light-microscopic data (Fig. 5, A-H), along the entire trans-Golgi, while the cis-Golgi facing the nucleus was devoid of such vesicular accumulations (Fig. 6 F). Immature precursors were detectable now in the 3D reconstruction as heterogeneous vesicular or short tubular structures (Fig. 6,G and H), with an average short/long diameter ratio of ∼1.8 (Fig. 6 I). By contrast, PLVs accumulating in arl8 −/− mutants were rather homogenously shaped, circular vesicles with a short/long diameter ratio close to 1 (Fig. 6, H and I). Consequently, immature precursors of rab2 −/− mutants were threefold smaller compared with those of arl8 −/− mutants (Fig. 6 J) and had a twofold smaller surface area ( Fig. 6 K).
In summary, we suggest that the 40-nm-size, tubule-shaped, clear-core vesicles forming at the trans-Golgi of rab2 −/− mutants represent immature and biosynthetically early arrested presynaptic precursors.

Discussion
Although of fundamental importance, our understanding of the cell biological origin and molecular mechanics of presynaptic precursor biogenesis remains fragmentary. We here provide evidence that early precursor biogenesis is regulated by the conserved, Golgi-related small GTPase Rab2 (Chan et al., 2011;Lardong et al., 2015;Liu and Storrie, 2012;Lőrincz et al., 2017;Saraste, 2016;Tisdale and Balch, 1996). We suggest that precursor formation takes place in two consecutive steps: (1) sorting of precursor cargo proteins from the trans-Golgi on independent exit routes for AZ and SV proteins, (2) followed by a subsequent maturation step merging both exit routes. Convergence could occur either by direct fusion of early precursors or through conventional sorting at endo-lysosomal compartments, from which mature precursors are retrieved at a later time point (Fig. 8 K).
Several independent lines of evidence, including confocal, super-resolution light, and EM, live imaging, and genetic assays, support this model. Upon depletion of Rab2, biosynthetic presynaptic material accumulates ectopically at the trans-Golgi of motoneuronal somata, overlapping with the trans-Golgi marker Syx16 (Figs. 5 and 6). Interestingly, these accumulations did not show a uniform presynaptic material distribution but a segregation into two adjacent but not overlapping fields, positive for either presynaptic proteins (BRP, RIM-BP, UNC13) or SV proteins (VGlut, Syt-1; Figs. 1 and 7). These aggregates apparently do not represent degradative compartments as they were devoid of lysosomal and autophagosomal machinery markers, i.e., Cathepsin L, p62, and ATG8a (Figs. 1 and S1). Ultrastructural analysis revealed that ectopic accumulations consist of tubule-shaped, clear-core vesicles with a short diameter of ∼40 nm volume (J; rab2 −/− : 0.56 ×10 6 nm 3 ± 21,040, n = 313; arl8 −/− : 1.44 ×10 6 nm 3 ± 48,906, n = 138); surface area (K; rab2 −/− : 33,563 nm 2 ± 794, n = 313; arl8 −/− : 60,795 nm 2 ± 1,508, n = 138). Scale bars: (A) 500 nm, (B) 100 nm, (F and G) 200 nm, and (H) 300 nm. All data provided as mean ± SEM. N represents single vesicles (D, I, J, and K) or sections (C and E) of one or two VNCs. ***, P < 0.001. wt, wild-type. ERGIC, ER-Golgi intermediate compartment. (Fig. 6). The elongated structures frequently appeared to consist of two to three attached spherical modules (Fig. 6 H), possibly consisting of immature precursors unable to proceed in a Rab2dependent fusion or fission process. By contrast, the previously identified PLVs observed in arl8 −/− mutants (Vukoja et al., 2018) were uniformly circular-shaped with ∼60-nm diameters, a consequently larger volume, and a dense core (Fig. 6), apparently representing more matured precursors. Moreover, the matured precursor field of arl8 −/− mutants showed no segregation of AZ and SV proteins and is thus likely the result of a subsequent fusion step of the post-Golgi early precursors, either directly by mutual fusion or via the endo-lysosomal compartments (Fig. 8 K). As this fusion step evidently does not occur in the absence of Rab2, immature precursors here have lost the ability to fuse to mature vesicles. These findings would place Arl8 function downstream of Rab2 in the precursor biogenesis pathway and, indeed, we could show by genetic epistasis experiments that Rab2 acts upstream of Arl8 (Fig. 8).
Rab2 as a membrane-associated protein could function either directly at the trans-Golgi or on the arrested immature precursors by recruitment of Rab2 interaction partners to the nascent precursor membranes executing the subsequent fusion process, for example, SNARE proteins or proteins of the homotypic fusion and protein sorting complex, such as VPS39, already shown to be a direct interaction partner of Rab2 (Gillingham et al., 2014).
Interestingly, both the SV protein-positive early (rab2 −/− mutants) and the matured (arl8 −/− mutants) precursors are positive for the lysosomal marker LAMP1 (Fig. 7). Thus, the immature precursors already express some lysosomal membrane identity. LMPs traffic on nonclathrin-coated vesicles from the trans-Golgi network to the late endosome (Pols et al., 2013). It is tempting to speculate that LAMP1-positive early precursors might be related to these organelles. In contrast, Arl8, the lysosomal kinesin adaptor, localized to both types of immature precursors in rab2 mutants (Fig. S5). Arl8, as a regulator of lysosome and lysosome-related organelle mobility (Khatter et al., 2015;Rosa-Ferreira and Munro, 2011), could confer motility to the mature precursors in the presynaptic precursor pathway by hooking them to the microtubule motor kinesin, as previously suggested in fly, mouse (Vukoja et al., 2018), and Caenorhabditis elegans (Klassen et al., 2010).
Downstream of somatic precursor biogenesis, Rab2 apparently remains associated, probably in its activated form, to the mature precursors as it trafficked anterogradely on BRP/Spinster-positive vesicles and effectively reached the synaptic terminal (Figs. 3 and S3). Retention of presynaptic proteins in the neuronal soma of rab2 −/− mutants consequentially caused a depletion of these proteins at the synaptic terminal and compromised presynapse formation and synaptic transmission (Fig. 2). The few synapses still forming remained roughly unaltered regarding architecture and SV number at the presynapse, suggesting a Rab2 function in the supply of presynaptic material rather than incorporation of precursor material into the assembling presynapse. However, as Rab2 is localized to the actively transported precursors (Fig. 3), it is conceivable that next to its upstream function in precursor biogenesis, Rab2 also directly contributes to the trafficking of presynaptic precursors.
Rab2 is described as a Golgi-related small GTPase implied in the bidirectional ER to Golgi trafficking (Liu and Storrie, 2012;Saraste, 2016;Tisdale and Balch, 1996), and indeed, we consistently observed subtle Golgi fragmentation defects in rab2 −/− mutants (Fig. 4, D-G), raising the possibility that generic Golgi sorting or protein biogenesis defects might be the primary cause of the rab2 −/− mutant phenotype. However, we could show that effective generic Golgi disruption did not phenocopy the rab2 −/− mutant phenotype, and other nonpresynaptic proteins were not accumulating at the trans-Golgi (Figs. 4 and S5), confirming a specific role for Rab2 in presynaptic precursor biogenesis.
The here-described function of Rab2 in precursor biogenesis could share mechanistic traits with the role of Rab2 and its effectors (RIC-19, RUND-1, CCCP-1, TBC-8) in DCV maturation in C. elegans (Ailion et al., 2014;Edwards et al., 2009;Hannemann et al., 2012;Sumakovic et al., 2009). DCVs store, transport, and release neuropeptides and neurotrophins through exocytosis. They form through secretion from the trans-Golgi network (Kim et al., 2006;Morvan and Tooze, 2008), where inadequate secretory proteins become incorporated into the immature DCVs, which are subsequently removed by clathrin-mediated membrane trafficking into the endo-lysosomal system (Ailion et al., 2014;Edwards et al., 2009;Kim et al., 2006;Morvan and Tooze, 2008;Sumakovic et al., 2009). Elegant studies in C. elegans showed that Rab2 and its effectors are required for the retention of neuropeptides in immature DCVs. Although we can only speculate, Rab2 function during the early phase of precursor biogenesis could comprise a similar mechanism by the retention or sorting of presynaptic and/or lysosomal proteins. Furthermore, future studies also need to evaluate a potential role of Rab2 in DCV maturation in Drosophila.
Our study contributes to a comprehensive model of the biosynthetic pathway underlying presynaptic precursor biogenesis and could support future research by connecting Rab2-related neurodevelopmental defects, e.g., memory defects in human (Li et al., 2015), autism spectrum disorders, or schizophrenia (Kiral et al., 2018;Takata et al., 2016), with the Golgi pathway-related neurodegenerative diseases (Rasika et al., 2018).

Rescue experiment
We used the genomic rab2 −/− mutant line (Chan et al., 2011), in which the Rab2 ORF was replaced by a Gal4, now transcriptionally regulated by the endogenous Rab2 transcription regulators, and thus expressed a UAS-Rab2-YFP at endogenous protein levels in the rab2 −/− mutant background (see also Tables 1 and 2).

Immunostainings of larval neuromuscular junctions (NMJs) and brains
Immunohistochemistry for confocal and STED microscopy For immunohistochemistry, dissections were performed in hemolymph-like solution 3 (HL3; Stewart et al., 1994;composition in millimolar: 70 NaCl, 5 KCl, 20 MgCl 2 , 10 NaHCO 3 , 5 trehalose, 115 sucrose, and 5 Hepes, pH adjusted to 7.2) by opening the third instar larvae dorsally along the midline and removing the entrails. Dissections were fixated with 4% paraformaldehyde in PBS (pH 7.2) for NMJ staining or Bouin's fixative for staining of brain samples for 10 min. After fixation, the filets were washed with PBS plus 0.05% Triton X-100 and blocked for 60 min in 5% normal goat serum (Sigma-Aldrich; S2007). For immunostainings, the larvae were incubated with primary antibodies at 4°C overnight and subsequently washed in a PBS plus 0.05% Triton X-100 solution for 2 h at RT. Larvae were then incubated for 2-3 h with secondary antibodies at RT. Washing procedures were repeated. Larvae were finally mounted in Vectashield (Vector Laboratories), Mowiol (Sigma-Aldrich), or ProLong Gold (Thermo Fisher Scientific). Primary and secondary antibodies used in these studies are shown in Table 3.

Image acquisition and analysis
Conventional confocal and STED images were acquired with Leica DMI 6000 (SP8) and TCS SP8 gSTED 3× microscopes (Leica Microsystems), respectively. For confocal scans, an HC PL APO CS2 63× /1.40 NA oil objective (Leica Microsystems) was used; for STED, an HC PL APO CS2 100×/1.40 NA oil objective (Leica Microsystems) was used. Images were acquired at ∼20°C, and fluorochromes are indicated in the antibody section described above. Imaging medium was immersion oil (Sigma-Aldrich; 55822). Mounting medium was . For signal detection, high-sensitive hybrid detectors with a 400-800-nm spectrum scanned for green and red channels and photomultiplier tubes with a 400-800-nm spectrum scanned for far red channels and were used for confocal scans. For STED, hybrid detector Sp GaAsP were used. Larval brain z-stacks had a step size of 0.3-0.5 µm between single optical slices. 40 z-slices of the central region along the dorsoventral axis were imaged. The NMJ z-stacks had a step size of 0.2-0.3 µm between single optical slices. All images were acquired using the LAS X software (Leica Microsystems). The ImageJ 1.52n software was used for the analyses of confocal and STED images. GraphPad Prism, v5.01, was used for statistical analyses. For STED microscopy, Huygens Deconvolution software was used applying a theoretical point spread function automatically computed based on pulsed-or continuous-wave STED-optimized function and the specific microscope parameters. Default deconvolution settings were applied. Images for figures were processed, if necessary, with ImageJ software to enhance brightness using the brightness/contrast function and smoothened (0.5-pixel Sigma radius) using the Gaussian blur function. Confocal stacks were processed with the image software Fiji (http://fiji.sc; Schindelin et al., 2012). Image analysis followed the standard protocol as described by Andlauer and Sigrist (2012b) and will be described in detail for the different analyses below. NMJ and VNC quantifications. Type 1b NMJs on muscle 4 were analyzed. Zoom areas in the VNC were acquired of motorneuronal somata either lateral or central (dorsal) of the neutrophil. The HRP-Cy5 antibody signal was used as the template for a mask, restricting the quantified area to the shape of the NMJ. For larval brain aggregate quantification, a region of interest (ROI) was drawn via the freehand selection tool including only the cortex region of the larval VNC (sparing neuropil region and background). The original confocal stacks were converted to maximal projections (for brain, taking only the central 20 z-slices of the VNC without adjacent axons or weakly stained tissue in deeper layers), and after background subtraction, a mask of the synaptic area/brain aggregates was created by applying a manual threshold to remove the irrelevant lowerintensity pixels. The segmentation of single spots was done semi-automatically via the command "Find Maxima" embedded in the Fiji software and by hand with the pencil tool and a line thickness of 1 pixel. To remove high-frequency noise, a Gaussian blur filter (0.5-pixel Sigma radius) was applied. The processed picture was then transformed into a binary mask using the same lower threshold value as in the first step. This binary mask was then projected onto the original unmodified image using the "min" operation from the ImageJ image calculator.
The number of AZ or VNC aggregates in the graphs reflects the density of particles in both tissues and was normalized to either the NMJ (HRP) area or the freehand selection ROI of the cortex area. The AZ area (nm 2 ) is the mean area value of all determined individual AZ area of one NMJ. For sum intensities "per NMJ area," every single pixel intensity within the single spots/particles of the corresponding channel was first added up (calculated as "integrated density" per AZ/aggregate in Fiji), and subsequently all spot values per NMJ/cortex were summed up in order to estimate total protein levels of all particles within the respective tissue. Finally, this total sum was divided by the respective NMJ or cortex area to normalize these estimates for variations in imaged tissue size. n represents the number of NMJs/larval brains. For sum intensities "of NMJ," protein mean pixel intensity within the HRP mask was calculated as "integrated density." For protein mean pixel intensity "of AZ," the mean pixel intensity of the individual AZs was determined and the mean value plotted. For protein mean intensity "of cortex area," an ROI around the whole cortex area excluding the neuropile was created using the HRP signal as a reference. Mean pixel intensity of protein within this ROI was determined and the mean value plotted. The same procedure was applied for NMJ ("protein mean pixel intensity of NMJ") using the HRP mask. All data were normalized to wild-type (100%). n for brain quantifications represents one VNC of one larva; four to eight larvae were analyzed. n for NMJs represents NMJs with one or two NMJs/animal and five or six animals/genotype.
Golgi (GM130) quantification of Golgi defects. Golgi quantifications were performed as described above regarding the basic image treatment. A stack of four to six slices from the VNC-zoom images was chosen and a maximum projection performed. Three to six clearly detectable cells were chosen for analysis and marked (ROI-area). Within this ROI, single GM130 spots were identified as described above, but without segmentation. The number of spots was divided by ROI-area and normalized to Rab2 knockdown (RNAi) and control Genomic Rab2-GFP Rab2 localization (Figs. 3, 4, and S5) Ok6-Gal4 × UAS-Rab2-YFP (wild-type) In vivo (Fig. 3) Rab2 localization (Fig. S5) Ok6-Gal4 × UAS-Rab2 Q65L -YFP In vivo (Fig. S4) Rab2 localization (Fig. S6) Ok6-Gal4, UAS-BRP-D3-Straw × UAS-Rab2-YFP (wild-type) In vivo (Fig. 3) Ok6-Gal4, UAS-Rab2-YFP (wild-type) × UAS-Spinster-RFP In vivo (Fig. 3) Ok6-Gal4, UAS-BRP-D3-Straw × UAS-Rab2 Q65L -YFP In vivo (Fig. S5) Ok6-Gal4, UAS-Rab2 Q65L -YFP × UAS-Spinster-RFP In vivo (Fig. S5) wild-type (100%). Area of the individual GM130 spot was determined, and the mean plotted, normalized to wild-type (100%). Mean pixel intensity of GM130 signal within the ROI-area was determined, and the mean plotted, normalized to wild-type (100%). n represents one VNC of one larva; four to eight larvae were analyzed. Pearson's correlation coefficient. For Pearson's correlation coefficient, an image stack was acquired, and one focal plane was chosen for analysis after ROI selection within a single cortex soma or axonal bundle, respectively. ROIs were selected around aggregates with potential signal overlap of one to three aggregates in a range of 1.5-2.5 µm 2 . ImageJ Coloc_2 was used to determine the Pearson's R value above threshold by testing the Rab2 channel against the other indicated second channel (BRP, VGlut, Spinster, etc.). As a control, the second channel was flipped horizontally ("inverted BRP, Spinster, VGlut, etc."), and subsequently the Pearson's R value above threshold was recalculated. This aims to show specific colocalization only in the original (not flipped/not inverted) overlapping channels and to exclude a general stochastic colocalization. Analyzed where overall 9-12 motoneuronal somata (one ROI per soma) of two or three VNCs, corresponding to two or three animals. For axonal colocalization, 10-15 axonal bundles (one ROI per axonal bundle) in three to five animals were selected.
BRP mean pixel intensity in boutons excluding AZs from STED analysis of NMJs. Bouton area was outlined manually (ROI-1), and outside bouton area was cleared. Gaussian blur with 2.0 was applied. A threshold of 30 was set and all AZs within this threshold marked (ROI-2). Subtraction of ROI-2 from ROI-1 (XOR function in Fiji ROI manager) and the new ROI was named ROI-3. In ROI-3 (bouton area without AZs), the mean pixel intensity was measured. Data were collected from two or three boutons of one NMJ per larva from three larvae. Determination of the number of AZs per bouton area and AZ ring diameter from STED analysis of NMJs. For BRP ring diameter (C-terminal antibody Nc82) from STED analysis of NMJs, Nc82positive AZs in a planar position with a minimum of three clusters/maxima were manually encircled, and ROIs were saved (1-14 AZs/bouton, three boutons/larvae, three larvae). ROI area was measured (Fiji) and diameter calculated assuming a circular ROI area according to the formula d 2 where d is the diameter and A is the area.
Area overlap measurement of BRP/VGlut and Arl8-GFP in percentage from neuronal somata. Single cells were manually chosen from three to five images/brain, three brains in total. Threshold values were set manually, and ROIs in both channels were outlined and the area measured. The AND function of the Fiji ROI manager was used to determine area overlap. Only the Arl8-positive area was determined by subtraction of the copositive area (AND; see above) from the total Arl8 area and percentages calculated.
Frequency determination in percentage of peak-to-peak distance of BRP-positive aggregates in rab2 −/− mutants and GM130marked Golgi. We place line through BRP aggregates and Golgi, saved as ROI, used ROI manager MultiPlot tool, and determined peak-to-peak distance. For random shifted data, the BRP channel was shifted by 80 pixels. A cutoff at 1.2-µm distance was used, binned for 0.06 µm. We measured 1-14 distances from two VNC images per larva, with a total of three larvae analyzed.
Center distance measurements of STED-imaged presynaptic protein aggregates in rab2 −/− mutant somata. Single-channel images were created for individual aggregates. By defining a suitable threshold for each aggregate separately, a binary mask was created. Then, outlines of the aggregate were detected by the magical wand tool and subsequently saved as ROI. Using the ImageJ built-in measurement function "center of mass," the respective x and y mass-central coordinates of each ROI (one per channel) were calculated. Distances of the centers of mass were calculated according to the function x/y are the x/y coordinates). Experimental n represents the number of single aggregates tested in VNC images of two or three animals.

In vivo live imaging and analysis
In vivo imaging of intact Drosophila larvae was performed as previously described (Andlauer and Sigrist, 2012a). Briefly, third instar larvae were put into a drop of Voltalef H10S oil (Arkema) within an airtight imaging chamber. The larvae were anesthetized before imaging with 10 short pulses of a desflurane (UAS) air mixture until the muscles relaxed. Axons were imaged using confocal microscopy immediately after exiting the VNC.
Kymographs were plotted using a custom-written Fiji script (based on tsp050706.txt TimeSpacePlot [Kymograph] from Jens Rietdorf, downloadable at http://www.embl.de/eamnet/html/ kymograph.html). In brief, temporal recordings of proximal axons (pixel size, 100 nm; t resolution, 0.8-1.7 s), expressing YFP/ mRFP/strawberry-labeled proteins of interest, were backgroundsubtracted, filtered via a Gausian blur (Sigma radius 0.5), and stabilized via the Fiji image registration plugin MultiStackReg. A time stack limited to the time frame of a single specific cotransport event was cropped from the original file. Sequentially, a segmented line ROI was drawn along the trajectory of the cargo (minimum of 30 s stable visibility). The final kymograph was (automatically) created by adding the line ROI (average width, 3-10 pixels) per time frame on top of the preceding line ROI in a concatenated fashion.
Quantification of (co)transport events in proximal axons was performed as follows. Per indicated coexpression genotype, two or three different axonal nerve bundles (5-min imaging time per axon) were imaged in five or six animals. Every file was screened for transport events by quantifying events of sufficient intensity and traceability (stable identification of the unidirectional cargo trajectory for at least 4 µm in one channel) via drawing a linear ROI starting from the point of first discovery to the end point in the last sufficiently recognizable section. The length of this ROI was divided by the time (number of traceable images of the single event × t resolution per file), and the resulting speed value (µm/s) was listed according to the cargo direction (anterograde/retrograde). Every ROI (created in channel one) was compared with the corresponding coexpression channel and only counted as cotransport if this second channel indicated a similar unipolar movement event in various images (at least three or four different temporal sections within the same time frame while indicating a sufficient signal overlap).
Electrophysiology TEVC recordings were performed as previously reported (Matkovic et al., 2013). In short, spontaneous recordings (miniature excitatory junction currents [mEJCs], 90 s), single-evoked recordings (eEJCs, 20 repetitions at 0.2 Hz), and high-frequency recordings (PP 10-ms or 30-ms interstimulus interval, PP10 or PP30, 10 repetitions at 0.2 Hz) were recorded. All experiments were performed on third instar larvae raised at 25°C on semidefined medium (Bloomington recipe). Dissection and recording medium was extracellular HL3 (Stewart et al., 1994;in millimolar: 70 NaCl, 5 KCl, 20 MgCl 2 , 10 NaHCO 3 , 5 trehalose, 115 sucrose, and 5 Hepes, pH adjusted to 7.2). Dissections were done in ice-cold Ca 2+ -free HL3 medium, while mEJC, eEJC, and highfrequency recordings were performed in 1.5 mM Ca 2+ HL3 at room temperature. For all physiological recordings, intracellular electrodes with a resistances of 15−25 MΩ (filled with 3 M KCl) were placed at muscle 6 of abdominal segment A2/A3. Acquired data were subjected to a low-pass filter at 1 kHz and sampled at 10 kHz. Traces of both miniature and evoked postsynaptic currents were recorded in TEVC mode (AxoClamp 2B; Axon Instruments). The command potential for mEJC recordings was −80 mV, and it was −60 mV for all other recordings. Only cells with an initial membrane potential between −50 and −70 mV and input resistances of ≥4 MΩ were used for further analysis. eEJC and PP traces were analyzed for standard parameters (amplitude, rise time, decay, charge flow, PP ratio) by using a semi-automatic custom-written MATLAB script (Mathworks; v. R2009a). Stimulation artifacts in eEJC recordings were removed for clarity. mEJC recordings were analyzed with pClamp 10 software (Molecular Devices).
Western blot analysis of larval brains Larval central nervous system (CNS) protein extractions were performed as follows: 10 CNSs were dissected from third instar larvae. The resulting tissues were sheared manually in 5 µl of 2% SDS aqueous solution using a micropistil secured into a 1.5-ml cup. Subsequently, 0.5 µl of a 10% Triton X-100 aqueous solution and 5 µl of 2× Laemmli sample buffer were added. Samples were then heated at 95°C for 10 min. Centrifugation was performed for 2 min at 16,000 g, in order to pellet debris. For larval CNS, 10 µl (equivalent to 10 larval CNS) was subjected to denaturing SDS-PAGE using a 6% Tris-HCl gel.

EM
Conventional embedding for transmission EM was performed as described previously (Matkovic et al., 2013). In brief, dissected third instar larvae were fixed with 4% PFA and 0.5% glutaraldehyde in 0.1 M dissolved in PBS for 10 min and subsequently in glutaraldehyde (2% glutaraldehyde in 0.1 M sodium cacodylate) for 60 min, then washed in sodium cacodylate buffer and postfixed with 1% osmium tetroxide and 0.8% KFeCn in 0.1 M sodium cacodylate buffer (1 h on ice). After washing with sodium cacodylate buffer and distilled water, the samples were stained with 1% uranyl acetate in distilled water. Samples were dehydrated, infiltrated in EPON resin, and subsequently muscles 6/7 of abdominal segment A2/3 were removed by dissection. Blocks, collected in an embedding mold, were polymerized and cut in 65-70-nm-thin serial sections. Sections were then postfixed and poststained with uranyl acetate/lead citrate. Micrographs were produced using a JEM 1011 microscope equipped with an Orius 1200A Gatan camera using the DigitalMicrograph software package (Gatan).
For quantification, the plasma membrane and the electrondense T-bar were detected by eye and labeled manually. T-bar roof size was measured using a straight line connecting the furthest distance of the upmost T-bar dense material (in relation to the plasma membrane). T-bar area was obtained by surrounding the dense material and measuring the area of the created ROI.
For 2D morphometric analysis, images of neuronal soma from VNCs were imaged using a Zeiss 900 transmission EM. For 3D analyses, 250-nm sections were cut and collected on coated slotted grids with 10-nm gold fiducials. Once the Golgi was located, a series of images from +60°to −60°were taken with a 1°step using a Tecnai G20 microscope. Etomo/IMOD (Kremer et al., 1996; https://bio3d.colorado.edu/imod/) and Microscopy Image Browser (Belevich et al., 2016; http://mib.helsinki.fi/ index.html) were used to work with 3D volumes and render 3D models of Golgi and Golgi-associated structures.
Quantification and statistical analysis Data were analyzed using GraphPad Prism v5.01. Normality was tested with the D'Agostino and Pearson omnibus normality test. If the sample size was too low for D'Agostino and Pearson omnibus normality test (n > 7), Gaussian distribution was assumed. If data were normally distributed (or assumed to be normally distributed for n < 7) and two groups were compared, an unpaired, two-tailed t test was used ( Fig. 1, B . If data were normally distributed (or assumed to be normally distributed for n < 7) and more than two groups were compared, a one-way ANOVA with Tukey's post hoc test was used (Fig. 1, M and N; and Fig. 8, B-E). If data were not normally distributed and two groups were compared, a Mann-Whitney U test was used (Fig. 2, J, L, N, P, and R; Fig. 3, C, F, and I; Fig. 4, C, E, S, U, and Z; Fig. 5, K, M, and N; Fig. 6, C, D, and I-K; and Fig. 7 H). If data were not normally distributed and more than two groups were compared, the Kruskal-Wallis test with Dunn's post hoc test was used (Fig. 8 J). For supplemental figures, see supplemental figure legends. Significance is noted by *, P < 0.05; **, P < 0.01; ***, P < 0.001. No statistical methods were used to predetermine sample sizes, but sample sizes in these experiments were similar to those generally employed in the field. Data collection and analyses were not performed blind to the conditions of the experiments, nor was data collection randomly. For electrophysiological recordings, genotypes were measured in an alternating fashion on the same day and strictly analyzed in an unbiased manner.