Endocytic BDNF secretion regulated by Vamp3 in astrocytes

Brain-derived neurotrophic factor (BDNF) regulates diverse brain functions via TrkB receptor signaling. Due to the expression of TrkB receptors, astrocytes can internalize extracellular BDNF proteins via receptor-mediated endocytosis. Endocytosed BDNF can be re-secreted upon stimulation, but the molecular mechanism underlying this phenomenon remains unrecognized. Our study reveals that vesicle-associated membrane protein 3 (Vamp3) selectively regulates the release of endocytic BDNF from astrocytes. By using quantum dot (QD)-conjugated mature BDNF (QD-BDNF) as a proxy for the extracellular BDNF protein, we monitored the uptake, transport, and secretion of BDNF from cultured cortical astrocytes. Our data showed that endocytic QD-BDNF particles were enriched in Vamp3-containing vesicles in astrocytes and that ATP treatment sufficiently triggered either the antero- or retrograde transport and exocytosis of QD-BDNF-containing vesicles. Downregulation of Vamp3 expression disrupted endocytic BDNF secretion from astrocytes but did not affect uptake or transport. Collectively, these results provide evidence of the selective ability of astrocytic Vamp3 to control endocytic BDNF secretion during BDNF recycling.


Methods
Detailed information of the materials and resources is included in Table 1. All animal experimental procedures were approved by the Institutional Animal Care and Use Committee of the Korea Brain Research Institute (IACUC-2017-0047). All experiments were carried out in accordance with the approved guidelines and regulations.
Primary astrocyte culture. We utilized an AWESAM astrocyte culture protocol as reported previously 19 with minor modifications to acquire cultured astrocytes that had an in vivo-like morphology. Cortical astrocytes were prepared from embryos from wild-type C57BL/6 mice on days E17-18. Cortices were dissected in dissection medium (10 mM HEPES in HBSS) at 4 °C and then incubated in 0.25% trypsin-EDTA in a 37 °C water bath for 20 min with gentle inversion every 5 min. After trypsinization, the tissue was washed in dissection medium at 4 °C five times and then triturated with 1 ml of NB+ medium (2% B-27 supplement, 2 mM GlutaMax, 5000 U/ml penicillin and 5000 µg/ml streptomycin in neurobasal medium). Dissociated cells were filtered through a cell strainer and plated on 0.04% polyethylenimine (PEI)-coated cell culture dishes (4 × 10 6 cells/60 mm dish) in culture media (10% FBS, 5000 U/ml penicillin and 5000 µg/ml streptomycin in DMEM). Seven days after plating the dissociated cells, the dishes were shaken at 110 rpm for 6 h. The cells were then washed with 1× PBS three times, treated with 0.25% trypsin, and plated on 0.04% PEI-coated glass-bottom dishes (3 × 10 4 cells/dish) or 18 mm coverslips in a 12-well plate (1 × 10 4 cells/well) in NB+ medium containing HBEGF (50 µg/ml).
To screen Vamp3 siRNA, C8-D1A (mouse type 1 astrocyte cell line) cells were cultured in DMEM supplemented with 10% FBS at 37 °C under 5% CO 2 . Each siRNA (100 nM) was transfected into C8-D1A cells using RNAi Max according to the manufacturer's protocol. Two days after transfection, samples were analyzed by western blotting with an anti-Vamp3 primary antibody or β-actin-HRP and HRP-conjugated anti-rabbit secondary antibody. The screening of Vamp3 siRNAs revealed that siVamp3 #1 effectively diminished the level of endogenous Vamp3 (Fig. S2). Therefore, only siVamp3 #1 was employed in the experiments.
TrkB-targeting shRNA lentiviral particles were purchased from Sigma (shRNA-pLKO.1-hPGK-puro-CMV-tGFP). The shRNA target sequences are described in the Key Resources Table. To assess the knockdown efficiency of TrkB shRNA, cortical neurons from E17-18 C57BL/6 mouse embryos were cultured. Each Lenti-shTrkB particle was transduced into cortical neurons at 5 DIV. Three days after transduction, total RNA was extracted using TRIzol reagent. Each RNA sample (0.3 µg) was reverse transcribed into cDNA by using SuperScript III reverse transcriptase. To determine the reduction in TrkB RNA levels, PCR was performed using TrkB and β-actin primers. Because shTrkB #1 reduced the level of endogenous TrkB more effectively than shTrkB #2 (Fig. S1), only shTrkB #1 was used in the experiments.

Immunocytochemistry.
To determine the localization of QD-BDNF, cultured astrocytes were incubated with 2 nM QD-BDNF for 20 min and then fixed with 4% paraformaldehyde (PFA). For immunostaining, the cells were permeabilized with 0.1% Triton X-100 for 10 min and then blocked with 5% normal goat serum for 1 h at room temperature. After blocking, the cells were incubated with anti-Rab5, anti-Rab7, anti-Rab11, anti-Lamp1, anti-Vamp3, or anti-chromograninB for 1 h and then incubated with an anti-Alexa 488 secondary antibody for 1 h at room temperature. QD imaging. For monitoring endocytic BDNF, 50 nM biotinylated mature BDNF (bt-BDNF) or 50 nM biotinylated bovine serum albumin (bt-BSA) was incubated with 50 nM streptavidin-conjugated quantum dot 655 (st-QD655) at 4 °C overnight at a ratio of 2:1. QD-BDNF or QD-BSA was then filtered with a 100 kDa Amicon filter to remove unconjugated mBDNF, BSA, or QDs, and 1% BSA containing PBS was added to the filtrates. Astrocytes were incubated with QD-BDNF or QD-BSA on 12-13 DIV, and the medium was then replaced with an extracellular solution (in mM; 119 NaCl, 2.5 KCl, 20 HEPES, 2 CaCl 2 , 30 glucose, and 2 MgCl 2 , pH 7.4) containing 4 µM QSY21. Time-lapse images were taken by using a confocal laser scanning microscope (TCS SP8, Leica) at a 1 Hz rate using a 63× oil objective. ATP (100 µM) or ionomycin (1 µM)-containing extracellular solution was added to stimulate the astrocytes. QD655 fluorescence was excited with a 561 nm laser and assessed with a HyD (hybrid) detector in the range of 650-695 nm.
Image and statistical analyses. Image processing and analysis were performed using ImageJ/FIJI software (NIH, USA). To analyze the kinetics or secretion of BDNF particles, regions of interest (ROIs) of astrocytic processes were manually selected and linearized. The linearized time-lapse images were transformed into kymographs using the KymographBuilder plugin in ImageJ/FIJI. After extracting the X and Y coordinate data for each particle from the kymograph, the direction, distance, and velocity were determined. Immobile or QD-BDNF particles in the stationary mode were defined when the particles showed the travel distance less than the diameter of a single QD-BDNF particle (~ 0.6 µm). Trafficking of QD-BDNF particles over 0.6 µm were categorized as the anterograde or retrograde transport, depending on the direction of the particle transports. The complete disappearance of QD-BDNF fluorescence was defined as QD-BDNF exocytosis. The percentage of QD-BDNF secretion was determined by dividing the number of secreted QD-BDNF particles with the total number of QD- www.nature.com/scientificreports/ BDNF particles on the kymograph (number secreted QD-BDNF particles/all QD-BDNF particles × 100 (%)).
To calculate colocalization ratios of QD-BDNF particles, images with QD-BDNF particle were segmented and transformed to the binary images to identify the region of interests (ROIs) of all observed QD-BDNF particles. A total number of QD-BDNF particles was derived from the total number of ROIs in these images. Next, colocalized QD-BDNF particles with vesicle markers were determined when more than 80 % area of the ROI was occupied by the fluorescence signal of vesicle markers. This 80% threshold was based on our confocal imaging conditions as follows: using the oil-immersed 63× lens (numerical aperture (NA) = 1.4) and the 561 nm excitation laser, the approximate lateral resolution of our confocal imaging was about 150 nm (d = 0.37λ/NA, according to the Abbe diffraction limit; λ = 561 nm, NA = 1.4). Because this lateral resolution of our confocal imaging was about 25% of the diameter of single QD particles (the diameter of the single QD particle = ~ 600 nm; Fig. 1), QD-BDNF particles showing 75% or more overlap of their area with fluorescence signals of vesicle markers may be considered as 'colocalized' with the tested marker. In Fig. 3B,D, we counted the number of QD-BDNF particles showing 80-100% overlap of their areas with vesicle markers as the number of colocalized QD-BDNF. In Fig. 3D above, colocalized Vamp3-EGFP signals were defined as ones showing 100% overlap of their areas with tested vesicle markers. The colocalization ratio was calculated by the following equation: To analyze the structural complexity of astrocytes induced by BDNF, 2 nM of QD-BDNF was treated for 20 min. The morphological complexity of astrocytes was defined by the shape index (SI; cell perimeter 2 /area -4π). It is known that greater SI values well correspond to increased complexity of cell morphology, but perfect circles show SI = 0 12,20 .
Statistical analyses were performed using Prism 8.0 software (GraphPad). Statistically significant differences between two groups were determined using Student's unpaired t-test, and three or more groups were compared using one-way ANOVA with Dunnett's multiple comparisons test. The Kolmogorov-Smirnov test was used to examine the statistical significance of the percentages of cumulative distribution between the two groups. All data were from three independent batches of cultured astrocytes and are indicated as the mean ± standard error of the mean (SEM).

Results
Monitoring endocytic BDNF in cultured astrocytes using QD-BDNF. To directly monitor endocytic BDNF in astrocytes, we utilized biotinylated recombinant mature BDNF directly associated with streptavidin-QDs as described previously (ref. 7 ; see Methods for detailed information). With this method, the fluorescence of the extracellular QD-conjugated mature BDNF complex (QD-BDNF; Fig. 1A) could be cancelled by a hydrophilic fluorescence quencher, QSY21 (4 µM), in the extracellular media, but QD-BDNF fluorescence Colocalization ratio = (number of colocalized particles with vesicular markers)/(total number of particles). www.nature.com/scientificreports/ was recovered after endocytosis (Fig. 1A,C). Under our imaging conditions, the smallest and most observable two-dimensional size of purified QD-BDNF was approximately 0.3 µm 2 , indicating a single QD-BDNF particle (Fig. 1B). The intracellular uptake of QD-BDNF particles into astrocytes was mediated by receptor-mediated endocytosis, as (1) QD-BSA treatment resulted in no intracellular QD particles (Fig. 1C), and (2) the number of intracellular QD-BDNF particles (Fig. 1D) from astrocytes was significantly reduced by shRNA-mediated genetic knockdown (KD) of TrkB expression (Fig. S1). Moreover, our QD-BDNF particles were bioactive, because cultured astrocytes showed more complex morphology after QD-BDNF treatment (Fig. 1E), consistent with a previous report 12 . Since astrocytic TrkB.T1-dependent structural complexity is important for the structural and functional maturation of astrocytes 12 , QD-BDNF uptake under our conditions appeared to be mediated by TrkB.T1. We next explored the ideal concentration and incubation time for the QD-BDNF treatment of cultured astrocytes to track single QD particles. QD-BDNF (0.5-5 nM) was applied to cultured astrocytes for 5 minutes (min) up to 4 h. Treatment with 2 nM QD-BDNF for 20 min resulted in most density and fraction of intracellular single QD-BDNF particles (Fig. 1F-I), and all QD-BDNF tracking and secretion experiments were therefore carried out under this condition.
ATP triggers the transport and secretion of endocytic BDNF in astrocytes. We next monitored intracellular QD-BDNF particles in astrocytes to investigate the transport and secretion of endocytic mBDNF. Since astrocytes can be stimulated by extracellular ATP due to the expression of diverse P2 receptors 24 , 100 µM ATP was added to QD-BDNF-containing astrocytes expressing EGFP ( Fig. 2A,B) to induce the transport and secretion of QD-BDNF. Most QD-BDNF particles remained immobile (stationary mode) before ATP treatment (Fig. 2C). However, ATP stimulation triggered either the anterograde or retrograde transport of QD-BDNF (Fig. 2C), leading to an increase in the distance of QD-BDNF trafficking (Fig. 2D). No ATP-induced changes in speeds of QD-BDNF transport were detected (Fig. 2E). These results suggest that the transport of endocytic BDNF is dependent on ATP-induced intracellular signaling.
We next assessed whether ATP stimulation evokes endocytic BDNF release in astrocytes. The exocytosis of endocytic QD-BDNF could be detected by the disappearance of QD-BDNF fluorescence due to the exposure of QD-BDNF to the QSY21 quencher via opened vesicle pores 7 . Despite a few spontaneous QD-BDNF exocytosis events (5.28 ± 1.76%), QD-BDNF exocytosis was significantly increased (19.37 ± 4.75%; Fig. 2F) after the ATP treatment, consistent with another study 25 . This ATP-induced QD-BDNF secretion was abolished by the expression of the tetanus toxin light chain (TLC) in astrocytes (Fig. 2F), supporting the idea that endocytic BDNF release is SNARE-dependent. Ca 2+ signaling seems to play a limited role in endocytic BDNF secretion, because ATP-induced QD-BDNF secretion was partially reduced by the chelation of intracellular Ca 2+ , but a direct Ca 2+ elevation by the ionomycin treatment failed to trigger QD-BDNF secretion (Fig. 2F). These results suggest that cooperative actions of other mechanisms with Ca 2+ signaling are required for the full exocytosis of endocytic BDNF-containing vesicles. Finally, as reported in neurons 7 , BDNF secretion events were frequently observed in immobile vesicles before ATP treatment (Fig. 2G), suggesting that the arrival of endocytic BDNF vesicles at secretion sites is a prerequisite for exocytosis events.
QD-BDNF particles were widely detected in all the tested vesicular fractions (Fig. 3A,B). Of note, the colocalization ratio of QD-BDNF with Vamp3 was highest among that with other vesicular markers (Fig. 3B), suggesting that a large portion of internalized BDNF molecules was sorted into Vamp3-positive vesicles. To further characterize the Vamp3-positive QD-BDNF-containing vesicles, additional immunocytochemistry analyses of astrocytes with both QD-BDNF particles and Vamp3-EGFP were performed with vesicular marker antibodies (Fig. 3C). Regardless of whether QD-BDNF particles were detected, Vamp3-positive vesicles were enriched in vesicles containing Rab5, Rab7, or ChgB (Fig. 3C,D). However, Vamp3-positive vesicles with QD-BDNF were more colocalized with Rab5-, Lamp1-, or ChgB-positive vesicles than Vamp3-negative ones (Fig. 3C,D). Given that astrocytic Vamp3-containing vesicles are implicated in the exo-and endocytotic cycling of endosomes 16 , our results suggest that Vamp3 participates in endocytic BDNF recycling in astrocytes.
Vamp3 is required for ATP-induced endocytic BDNF secretion from astrocytes. Since our results showed that endocytic BDNFs were enriched in Vamp3-containing astrocytic vesicles (Fig. 3), ATPinduced BDNF secretion may frequently occur at Vamp3-positive vesicles. We thus compared the fraction of QD-BDNF particles displaying the exocytosis event from Vamp3 ( +) vesicles to that from Vamp3-negative (−) vesicles (Fig. 4A). Few very spontaneous QD-BDNF secretion events were observed regardless of the presence of Vamp3 in QD-BDNF-containing vesicles (Fig. 4B), indicating that spontaneous endocytic BDNF release does not involve Vamp3. ATP-induced QD-BDNF secretion events was also observed from both Vamp3-positive and Vamp3-negative vesicles (Fig. 4B), but QD-BDNFs in Vamp3-positive vesicles were secreted more frequently than those in Vamp3-negative vesicles (Fig. 4C). Despite the possible effect of Vamp3-EGFP overexpression on distribution of endocytic QD-BDNF, these results propose the involvement of Vamp3-positive vesicles in ATPinduced endocytic BDNF secretion. www.nature.com/scientificreports/ Next, we tested whether Vamp3 directly participates in endocytic BDNF exocytosis by using the siRNA mediated KD method (Fig. S2C). We first assessed whether the endocytosis or transport of QD-BDNF was affected by Vamp3 KD (Fig. 5A-E). Vamp3 KD failed to alter the endocytosis (Fig. 5B) or ATP-induced antero-or retrograde transport of QD-BDNF (Fig. 5C-E). By contrast, astrocytes with Vamp3 KD showed significantly reduced ATPtriggered QD-BDNF secretion to ~ 76% (% QD-BDNF secretion: siSCR = 29.61 ± 2.84 vs. siVamp3 = 6.92 ± 1.36; Fig. 5F). This reduced QD-BDNF exocytosis was successfully restored by the delivery of the siRNA-insensitive Vamp3 construct together with Vamp3 siRNAs (Fig. 5F). Together, these results indicate that Vamp3 selectively controls endocytic BDNF exocytosis in astrocytes.

Discussion
In this work, we showed the direct uptake and recycling of mBDNF in astrocytes by utilizing QD-BDNF as a proxy for the extracellular BDNF protein. After secreted from source cells, neurotrophin proteins seem to be internalized by binding to corresponding Trk receptors on nearby target cells, but direct monitoring of endogenous neurotrophin has been hampered due to their relatively low concentration in live cells. Because QD is a fluorescent nanoparticle with an excellent photostability and could stably tracked in live cells with a high signal-to-noise ratio, the QD-linked neurotrophin sensor has been widely used to examine the transport and activity-dependent secretion of neurotrophin-containing endosomes in live cells 7,17,18 . Using QD-linked mBDNF, a previous study founds TrkB-dependent mBDNF internalization, as well as complexin 1/2 (Cpx1/2)/ synaptotagmin 6 (Syt6)-dependent re-secretion of endocytic mBDNF 7 . However, it has not examined whether mBDNF is directly internalized and recycled in astrocytes and what molecular mechanisms handle endocytic mBDNF secretion from astrocytes, although astrocytic p75NTR-dependent endocytosis of neuronal proBDNF and its re-secretion were reported 10 .
When treated with purified QD-BDNF particles, there was an increase in the complexity of astrocytic morphology (Fig. 1E), as found from other studies showing TrkB.T1-dependent structural complexity and maturation of astrocytes 11,12 . Given that TrkB-shRNA expression diminished QD-BDNF internalization (Fig. 1D), QD-linked mBDNF endocytosis and morphological changes seem to be mediated by TrkB.T1. Because ATP stimulation of astrocytes was sufficient for triggering SNARE-dependent release of endocytic QD-BDNF (Fig. 2F), our study proposes that neuronal mBDNF directly takes part in the process of astrocytic modulation of extracellular BDNF concentration, in addition to TrkB.T1-dependent regulation of astrocyte functions.
We revealed that Vamp3 is one of important regulators of ATP-triggered endocytic BDNF secretion. Among all tested vesicular pools, Vamp3-positive vesicles in the fraction of early endosomes, lysosome, or secretory granule contained most endocytic QD-BDNFs (Fig. 3B). However, other vesicular fractions such as Rab7 or Rab11-positive endosomes or other SNARE-containing vesicles may contain a portion of endocytic BDNF, because we found significant colocalization of QD-BDNF in both Vamp3-positive and -negative vesicles with corresponding vesicular markers but no significant colocalization with MitoTrackers (Fig. 3D). Because Vamp3 is an enriched vSNARE in astrocytes 29 and involved in endosome recycling 16 , it is possible that recycling of endocytic BDNF-containing vesicles in astrocytes requires the role of Vamp3. Indeed, our findings support this notion; we observed the secretion of QD-BDNF by ATP stimulation frequently from Vamp3-EGFP-containing vesicles (Fig. 4). Vamp3 KD was successful in diminishing ~ 76% of ATP-induced QD-BDNF exocytosis (Fig. 5), supporting the idea that Vamp3 is one of the main regulators for endocytic BDNF secretion. However, neither endocytosis nor transports of QD-BDNF requires Vamp3, as shown by no changes in QD-BDNF uptake and transports by astrocytic Vamp3 KD (Fig. 5). These results indicate a selective role of Vamp3 in endocytic BDNF release. It is unclear how ATP stimulation of astrocytes caused increased the antero-or retrograde transport of  Our work also uncovered the complex molecular nature underlying endocytic BDNF secretion from astrocytes. We discovered that chelation of ATP-induced Ca 2+ elevation partially reduces QD-BDNF exocytosis, whereas a direct increase in intracellular Ca 2+ concentration cannot evoke QD-BDNF exocytosis (Fig. 2F). These findings imply the requirement of additional signaling pathway for full exocytosis of endocytic BDNF. For example, modification of cAMP concentration through P2 receptor activation 30,31 or A2 receptors 32 , may influence endocytic BDNF release by activating cAMP-dependent signaling pathways important for vesicle docking or exocytosis 29,33 . Moreover, Vamp3-independent mechanisms may also be implicated in regulating endocytic www.nature.com/scientificreports/ BDNF release, because we observed a significant number of QD-BDNFs localized in the Vamp3 negative vesicles (Fig. 3) and ATP-triggered QD-BDNF release events from Vamp3 (-) vesicles (Fig. 4B). These findings support the notion that astrocytic mBDNF recycling involves multiple but differential signaling pathways. Additional studies will further explore the other aspects of molecular events regulating BDNF recycling in astrocytes and their physiological functions in synaptic plasticity and cognitive functions.

Data availability
All data generated or analyzed during this study are included in this published article (and its supplementary information files).