Abstract

   Extracellular vesicles (EVs) are endogenous membrane‐derived vesicles
   that shuttle bioactive molecules between glia and neurons, thereby
   promoting neuronal survival and plasticity in the central nervous
   system (CNS) and contributing to neurodegenerative conditions. Although
   EVs hold great potential as CNS theranostic nanocarriers, the specific
   molecular factors that regulate neuronal EV uptake and release are
   currently unknown. A combination of patch‐clamp electrophysiology and
   pH‐sensitive dye imaging is used to examine stimulus‐evoked EV release
   in individual neurons in real time. Whereas spontaneous electrical
   activity and the application of a high‐frequency stimulus induce a slow
   and prolonged fusion of multivesicular bodies (MVBs) with the plasma
   membrane (PM) in a subset of cells, the neurotrophic factor basic
   fibroblast growth factor (bFGF) greatly increases the rate of
   stimulus‐evoked MVB‐PM fusion events and, consequently, the abundance
   of EVs in the culture medium. Proteomic analysis of neuronal EVs
   demonstrates bFGF increases the abundance of the v‐SNARE
   vesicle‐associated membrane protein 3 (VAMP3, cellubrevin) on EVs.
   Conversely, knocking‐down VAMP3 in cultured neurons attenuates the
   effect of bFGF on EV release. The results determine the temporal
   characteristics of MVB‐PM fusion in hippocampal neurons and reveal a
   new function for bFGF signaling in controlling neuronal EV release.

   Keywords: CD63, exosome release, fibroblast growth factor,
   multivesicular bodies, pHluorin, VAMP3
     __________________________________________________________________

   Extracellular vesicles (EVs) are endogenous membrane‐derived vesicles
   that shuttle biomolecules from cell to cell. The molecular mechanisms
   that modulate EV release in the central nervous system are poorly
   understood. Here, it is demonstrated that fibroblast growth factor 2
   increases the expression of VAMP3/cellubrevin to enhance the release of
   EVs from hippocampal neurons.

   graphic file with name ADVS-7-1902372-g007.jpg

1. Introduction

   Extracellular vesicles (EVs) are small membrane‐derived vesicles that
   shuttle bioactive macromolecules such as cytosolic and membrane
   proteins, RNA, or DNA between cells.[46]1 EVs travel wide distances in
   bodily fluids and are well suited to act as long‐range messengers that
   deliver state‐dependent molecular information within (and across)
   organs. Moreover, EVs‐based approaches hold great potential for their
   use as diagnostic or therapeutic nanocarriers (for review see refs.
   [47]2, [48]3). As determined by their biogenesis, at least three main
   subgroups of EVs have been defined: ectosomes (also termed
   microvesicles), exosomes, and apoptotic bodies. These entities
   originate from different cellular sites: microvesicles (50–1000 nm) are
   budding directly from the plasma membrane (PM), whereas exosomes
   (40–150 nm) are derived from multi‐vesicular bodies (MVBs) of the late
   endo‐lysosomal pathway, where exosomes are formed by inward budding of
   the limiting membrane into the MVB lumen. The biogenesis of EVs is
   either assisted by the endosomal sorting complex required for transport
   (ESCRT), or alternatively, is driven by cone‐shaped lipids, such as
   ceramide.[49]4 The same cell can state‐dependently release different
   subpopulations of exosomes and exosome‐like EVs with diverse cargo
   compositions to induce various effects on target cells.[50]5, [51]6,
   [52]7

   Before exosomes can be released into the extracellular space, MVBs must
   fuse with the PM and a number of molecular mechanisms facilitate the
   otherwise energetically unfavorable MVB‐PM fusion and exosome release.
   The family of Rab GTPases was shown to control MVB inward budding,
   transport of MVBs, and MVB‐PM docking in a cell‐type specific
   manner.[53]8, [54]9, [55]10, [56]11, [57]12, [58]13 In addition, MVB‐PM
   fusion may be facilitated by different soluble NSF‐attachment receptor
   (SNARE) protein family members in different model systems.[59]12,
   [60]14, [61]15, [62]16, [63]17, [64]18, [65]19, [66]20 Similar to
   canonical neurotransmission, EV‐release appears to be
   calcium‐responsive in most cases,[67]12, [68]21, [69]22 although MVB‐PM
   fusion may also be triggered through alternative calcium‐independent
   mechanisms.[70]20

   In the central nervous system (CNS), EVs released by neurons and glial
   cells contribute to the complex networks of cell‐to‐cell signals that
   underlie CNS physiology and pathology.[71]23, [72]24, [73]25 Emerging
   evidence indicates that EV‐mediated signaling regulates neuronal
   firing,[74]26, [75]27, [76]28, [77]29 synaptic plasticity,[78]17,
   [79]30, [80]31, [81]32, [82]33 and myelin formation.[83]34, [84]35 The
   presence of EVs in cerebrospinal fluid (CSF) underscores their
   relevance in neural tissue.[85]36 In the injured CNS, EVs were recently
   found to cross the blood–brain barrier and promote
   neuroinflammation.[86]37, [87]38 Furthermore, certain types of EVs
   appear to support neuronal survival under conditions of ischemic
   stress.[88]9, [89]26, [90]39 In neurodegenerative diseases, EVs
   contribute to the seeding and spreading of toxic protein aggregates,
   and influence the aggregation process and clearance of these
   aggregates.[91]40

   In EV‐mediated neuron–glia communication, a general theme is that the
   secretion of EVs appears to be coupled to neuronal activity. Neuronal
   depolarization and glutamate release has been suggested to stimulate
   the secretion of exosomes from neurons[92]41, [93]42, [94]43, [95]44,
   [96]45 and oligodendrocytes[97]9 while serotonin is reported to
   stimulate microglial secretion of EVs.[98]46 The EVs may in turn
   transfer bioactive RNAs, proteins, and lipids between cells, thereby
   regulating neuronal activity.[99]17, [100]25, [101]30, [102]31,
   [103]33, [104]41 Similar to neurotransmitter vesicles fusing with the
   PM, exosome release can be evoked by electrical activity in conjunction
   with calcium influx and depends on SNARE proteins. However, the
   temporal characteristics of stimulus‐evoked EV release and
   cellular‐molecular mechanisms that segregate it from neurotransmitter
   release are currently unknown.

   Fibroblast growth factors (FGFs) are a family of pleiotropic growth and
   differentiation factors that regulate CNS homeostasis in health and
   disease. Although FGFs are best known for their roles in the early
   steps of patterning of the neural primordium and proliferation of
   neural progenitors, they have equally important roles in the adult
   brain; where they regulate neuronal calcium homeostasis and plasticity
   and promote neuroprotection and repair in response to neural tissue
   damage (reviewed in ref. [105]47). In addition to these physiological
   roles, bFGF has been implicated in responses to neuronal injury[106]48,
   [107]49, [108]50 or psychiatric conditions (reviewed in refs. [109]51,
   [110]52). For instance, a large body of evidence implicated FGF2 and
   the FGFR in mood and anxiety disorders (reviewed in refs. [111]51,
   [112]52) where FGF2 has been consistently found to be decreased in
   post‐mortem brain tissue from depressive subjects and in the brain of
   animal models including the hippocampus,[113]53, [114]54, [115]55,
   [116]56, [117]57 which is a highly relevant region for mood
   disorders.[118]58 Single nucleotide polymorphism (SNP) in the FGF2 gene
   was found to be associated with side effects and responsiveness to
   antidepressant treatment.[119]59 Conversely, chronic antidepressant or
   anxiolytic treatment resulted in an increase in FGF2[120]60 and
   administration of FGF2 exerted antidepressant properties.[121]61,
   [122]62 FGFs are expressed throughout the developing and adult CNS,
   with FGF‐1 and FGF‐2 (basic FGF or bFGF) being the most abundant FGF
   family member in the hippocampus.[123]63, [124]64, [125]65, [126]66
   bFGF is released from glia and—to a lesser degree—from hippocampal
   neurons[127]67 and exhibits the highest affinity for the FGF receptor 1
   (FGFR1).[128]68 Conversely, FGFR1 is the most abundant FGF receptor
   subtype in the hippocampus, where it is predominantly expressed in
   CA1/CA3 pyramidal neurons[129]69, [130]70, [131]71, [132]72 (reviewed
   in ref. [133]65). Despite their important role in hippocampal
   homeostasis, the contribution of FGFs to regulating neuronal EV release
   has not yet been addressed.

   Here, we used a combination of patch‐clamp electrophysiology and
   pH‐sensitive dye imaging to examine stimulus‐evoked EV release in
   individual neurons in real time. We found that cultured hippocampal
   neurons have a low consecutive rate of spontaneous EV release events.
   Different from previous reports,[134]41, [135]42, [136]43, [137]44,
   [138]45 we found that high‐frequency electrical stimulation (HFS) does
   not reliably induce EV release in cultured hippocampal neurons and that
   only a small subset of cells responds to HFS with a burst‐like increase
   of MVB‐PM fusion events. Unlike synaptic vesicles, which fuse with the
   PM within milliseconds in response to PM depolarization, the
   stimulus‐evoked fusion of MVBs occurred only after tens of seconds and
   exhibited a significant time lag between the initial stimulus‐evoked
   calcium signal and the eventual MVB‐PM fusion event. Treatment with
   basic fibroblast growth factor (bFGF) greatly increased the rate of
   stimulus‐evoked MVB‐PM fusion events and consequently the abundance of
   EVs in the culture medium. Further proteomic analysis of neuronal EVs
   demonstrated bFGF to increase the abundance of the v‐SNARE
   vesicle‐associated membrane protein 3 (VAMP3, cellubrevin) on EVs.
   Conversely, knocking‐down VAMP3 in cultured neurons attenuated the
   effect of bFGF on EV release. In summary, our results thus describe for
   the first time the specific temporal characteristics of MVB‐PM fusion
   in hippocampal neurons in response to electrical stimulation,
   uncovering a significant temporal separation of neuronal calcium influx
   and MVB‐PM fusion. In addition, our data reveal a new function of bFGF
   signaling in controlling neuronal EV release and thus support the
   investigation of growth factor‐mediated signal transduction via EVs in
   the healthy and diseased CNS.

2. Results

2.1. Stimulus‐Induced MVB‐PM Fusion Has an Abate Success Rate and Slow
Temporal Kinetic

   A number of reports connected EVs release to neuronal activity in
   conjunction with calcium influx.[139]41, [140]42, [141]43, [142]44,
   [143]45 However, because these studies examined bulk release in
   EV‐enriched medium pellets, the specific spatial and temporal kinetics
   of neuronal MVBs fusion in response to a depolarizing stimulus are
   currently unknown. In order to visualize MVB‐PM fusion in individual
   cells, we transduced primary hippocampal neurons after 3 days in vitro
   (DIV) with a lentivirus to express pCD63‐pHluorin[144]20, [145]22,
   [146]73, [147]74 (Figure S1a, Supporting Information). Compared to the
   neutral pH value of the extracellular compartment (7.4), the endosomal
   and lysosomal lumen pH is maintained in a range of 4.5–6.5, due to the
   activity of ATP‐dependent proton pumps in the membrane of
   endosomes.[148]75 Because of the topological orientation of the tag,
   with pHluorin facing the acidic intraluminal side of the MVB, the
   fluorescence emitted from pHluorin is quenched. Upon fusion of the MVB
   with the PM, the low luminal pH is immediately neutralized, resulting
   in an increase in fluorescence intensity. Different from examining bulk
   release in EV‐enriched medium pellets, this approach therefore allows
   for the visualization of MVB‐PM fusion events in single cells.

   At day 12 post‐transduction with pCD63‐pHluorin, CD63‐pHluorin
   fluorescence co‐localized with Rab7‐positive late endosomal vesicles
   (Figure S1b,c, Supporting Information) and strongly increased upon
   perfusion with 50 mm NH[4]Cl (Figure S1d,e, Supporting Information),
   thus confirming a pH‐dependent quenching of fluorescence in acidic cell
   organelles. We examined the cell culture medium from
   CD63‐pHluorin‐transduced neurons and found GFP to be present in
   EV‐enriched medium pellets together with the EV proteins CD81 and
   Alix/AIP1 (Figure S1f, Supporting Information). When loaded on
   iodixanol gradients, CD81 and Alix/AIP1 segregated to low‐density
   fractions, consistent with the typical density of exosomes[149]5
   (Figure S1f, Supporting Information). Compared to CD81 and Alix/AIP1,
   the distribution of GFP on the gradient was found to be somewhat
   broader, indicating that CD63‐pHluorin labels a more heterogeneous
   population of EVs (Figure S1f–i, Supporting Information). Based on
   these results we concluded that CD63‐pHluorin is successfully
   distributed to MVBs/late endosomes and to exosome‐like EVs in primary
   hippocampal neurons. Live cell imaging of CD63‐pHluorin‐expressing
   neurons (Figure [150]1a,[151]b) over a period of 5 min revealed sudden
   increases in fluorescence, suggestive of MVB‐fusion events in a small
   number of cells (n = 4/34 cells, 11.7%) (Movie S1, Supporting
   Information). Each event consists of a punctate, burst‐like increase in
   fluorescence with a comparably long signal duration and slow decay rate
   (Figure [152]1c) and occurred over a period of several minutes (avg.
   event rate: n = 2 ± 0.65 events per cell per minute). MVB‐PM fusion
   events were predominantly located to the soma, whereas axo‐dendritic EV
   release was extremely scarce in cultured neurons. We concluded that
   cultured hippocampal neurons have a low consecutive MVB‐PM rate under
   unstimulated conditions.

Figure 1.

   Figure 1
   [153]Open in a new tab

   High‐frequency stimulation (HFS) evokes MVB‐PM fusion in a subset of
   neurons. a) Photomicrograph illustrating a cultured rat hippocampal
   neuron (DIV 12) transduced with pCD63‐pHlurin under differential
   interference contrast (DIC) (left). Photomicrograph of the same cell
   with the focus plane adjusted to the the plasma membrane and under
   fluorescence microscopy (excitation at 488 nm) in b/w (middle) and in
   pseudo color (right). b) Representative photomicrographs from time
   lapse fluorescence live‐cell imaging (excitation wave length at 488 nm)
   from a single cell at t [1] = 0 s and t [2] = 125 s in pseudo color
   (first and second from left). Subtracting both images from each other
   (t [2] − t [1]) represents the sum of changes in fluorescence over
   time. Changes in fluorescence in the last as compared to the first
   frame are indicated in purple (third from left). Projection image with
   the images from (d) and (f) merged (fourth from left). c)
   Representative fluorescence traces (Δf/f) illustrating the rise in
   fluorescence in the ROIs indicated in (b). The respective ROIs are
   color‐coded in red, purple, green, and orange. High magnification
   microscopic image illustrating a burst‐like increase in fluorescence in
   the first ROI (orange). d) Schematic illustrating the high‐frequency
   stimulus (HFS) protocol applied in cultured hippocampal neurons. A HFS
   has been given for 1 s at 100 Hz. e) Representative images in pseudo
   color from live‐cell imaging of patch‐clamped CD63‐pHlurin‐transduced
   hippocampal neurons during and after HFS. The CD63‐pHlurin fluorescence
   signal remained stable as long as the holding membrane potential was
   maintained at V [Hold] (−70 mV) prior to the HFS. A burst increase in
   fluorescence is elicited in response to a HFS. f) Representative
   fluorescence trace illustrating a burst‐like increase in CD63‐pHlurin
   fluorescence following HFS in the cell indicated in (d). g)
   Photomicrograph illustrating a hippocampal neuron filled with Oregon
   Green BAPTA 488 through the patch pipette. h) Graph illustrating a rise
   in cytosolic calcium in response to HFS from the ROI in red in (g).

   When whole‐cell patch‐clamped, a subset of CD63‐pHluorin‐transduced
   neurons (n = 5/34 neurons, 14.7%) likewise exhibited a series of rapid
   fluorescence bursts in response to a brief high‐frequency stimulus
   (HFS) of 100 Hz over 1 s (Figure [154]1d; Movie S2a,b, Supporting
   Information). In HFS‐responsive cells, the burst‐like increase in
   pHluorin‐fluorescence occurred on average after an interval of 38.51 s
   (38.51 ± 12.54 s) between the HFS and the first burst with a relatively
   slow temporal rate ranging from seconds to minutes (t [1/2] = 86.83 ±
   0.6149 s; n = 5) and no immediate gross decay or signal termination was
   observed during recording (Figure [155]1e,[156]f). Conversely, the same
   HFS evoked a fast and immediate increase in intracellular calcium as
   measured by the calcium indicator Oregon Green 488 BAPTA‐1 (Figure
   [157]1g,[158]h). These results thus demonstrate that stimulus‐evoked
   MVB‐PM fusion has an abated success rate in cultured neurons and a
   comparably long time lag between the stimulus and MVB‐PM fusion. In
   order to further investigate the role of intracellular calcium stores
   for EV release, we perfused cultured neurons with thapsigargin (10 µm)
   while imaging CD63‐phluorin fluorescence (Figure S2, Supporting
   Information). Thapsigargin is a non‐competitive inhibitor of the
   sarco/endoplasmic reticulum Ca^2+ ATPase (SERCA) that raises cytosolic
   (intracellular) calcium concentrations by blocking the ability of the
   cell to pump calcium into the sarcoplasmic and endoplasmic
   reticula.[159]76 Whereas we detected EV release events in response to
   HFS, perfusion of cultured neurons with thapsigargin had no effect (n =
   60 neurons) (Figure S2c, Supporting Information).

2.2. Basic Fibroblast Growth Factor Increases Stimulus‐Evoked MVB‐PM Fusion

   We next aimed to identify candidate molecules that regulate MVB‐PM
   fusion in hippocampal neurons. Among other candidates, we investigated
   the effect of bFGF on MVB‐PM fusion. When we treated cultured
   hippocampal neurons after 9–12 days in vitro (DIV) for 24 h with bFGF
   (50 ng mL^−1), we found the proportion of neurons exhibiting a
   HFS‐evoked increase in CD63‐pHluorin fluorescence to be greatly
   enhanced (Figure [160]2a; Movie S3a–d, Supporting Information). In
   bFGF‐treated cells, CD63‐pHluorin fluorescence increased over a period
   of several minutes (t [1/2] = 117.8 ± 2.67 s; n = 17) following a HFS
   (Figure [161]2b,[162]c). Overall, the proportion of neurons that
   exhibited an increase in fluorescence increased to ≈60% in bFGF‐treated
   and HFS‐stimulated neurons (Figure [163]2d).

Figure 2.

   Figure 2
   [164]Open in a new tab

   bFGF increases stimulus‐induced CD63‐pHluorin fluorescence indicative
   of MVB‐PM fusion. a) Images in pseudo color illustrating an increased
   fluorescence signal in bFGF‐treated neurons in response to a HFS. b)
   Representative fluorescence traces illustrating an increase in
   CD63‐pHluorin fluorescence following HFS in VEH‐ and bFGF‐treated
   neurons and in unstimulated cells (V [Hold]). c) Bar graph illustrating
   the average maximal fluorescence in response to a HFS (number of
   cells/condition: bFGF + HFS vs bFGF + V [Hold]; n = 17 and 10; p =
   0.0007). d) Bar graph demonstrating the relative proportion of neurons
   that exhibited an increase or decrease of fluorescence or no change in
   response to a HFS in the same groups (number of cells/condition (bFGF +
   HFS; n (increase) = 31, n (decrease) = 9, n (no change) = 9 (n total =
   49); bFGF + V [Hold]; n = n (increase) = 2, n (decrease) = 1, n (no
   change) = 7 (n total = 10); VEH + HFS; n = n (increase) = 1, n
   (decrease) = 1, n (no change) = 9 (n total = 11), respectively). Data
   are shown as mean ± s.e.m. For comparison, a two‐tailed unpaired t‐test
   was used in (c). ***p < 0.001.

   In principle, an increase in fluorescence is due to MVB‐PM fusion and
   exposure of the pHluorin tag to the extracellular environment.
   Alternatively, it may be a consequence of an increased pH‐value inside
   the MVB or an increase in the expression of CD63‐pHluorin. To address
   the effect of bFGF on the pH‐value inside MVBs, we treated cultured
   neurons with bFGF and imaged CD63‐pHluorin fluorescence under steady
   state and after treatment with 50 mm NH[4]Cl (Figure [165]3a,[166]b).
   Whereas the absolute fluorescence value was similar in either condition
   (Figure [167]3c), we found that bFGF‐treatment (50 ng mL^−1 for 24 h)
   led to a significant decrease in fluorescence relative to the maximal
   NH[4]Cl‐evoked fluorescence signal (Figure [168]3d,[169]e). These
   results indicate a reduced pH value inside MVBs in bFGF‐treated
   neurons. In line with previous reports,[170]77, [171]78, [172]79 bFGF
   treatment also led to an increase in the intracellular calcium
   concentration as measured by Fura‐2 calcium imaging (Figure [173]3f).
   bFGF had no statistical significant effect on the expression of
   pHluorin over a period of 48 h, although there was a trend toward a
   reduced pHluorin expression in bFGF‐treated neurons (Figure S3a–c,
   Supporting Information). Based on our findings, we concluded that the
   HFS‐evoked increase in CD63‐pHluorin fluorescence in bFGF‐treated
   neurons represents MVB‐PM fusion rather than an increased pH‐value
   inside the MVB or the overall expression of CD63‐pHluorin.

Figure 3.

   Figure 3
   [174]Open in a new tab

   bFGF treatment results in acidification of CD63‐positive multivesicular
   endosomes. a,b) Photomicrographs illustrating CD63‐pHluorin
   fluorescence in unstimulated bFGF‐ and VEH‐treated neurons before and
   after application of NH[4]Cl. c) The raw fluorescence values are
   similar in bFGF‐ and VEH‐treated neurons (number of cells/condition n =
   510 and 505; p = 0.1488). d,e) Relative to the maximal fluorescence
   evoked by NH[4]Cl, the CD63‐pHluorin signal in bFGF‐treated neurons (50
   ng mL^−1, 24 h) is slightly decreased, indicating a decreased pH‐value
   in MVBs of bFGF‐treated neurons (p < 0.0001). f) Treatment with bFGF
   leads to a small increase of the intracellular calcium concentration as
   measured by Fura 2 (number of cells/condition n = 312; p = 0.0003).
   Data are shown as mean ± s.e.m. For comparison, a two‐tailed unpaired
   t‐test was used. ***p < 0.001.

2.3. bFGF Increases EV Release in Cultured Hippocampal Neurons

   In order to further examine the effect of bFGF on neuronal EVs, we
   treated cultured rat hippocampal neurons at DIV 9–12 with bFGF or
   vehicle (VEH). We quantified the number of EVs in the culture medium of
   neurons treated with VEH or bFGF by nanoparticle tracking analysis
   (NTA) and found that bFGF (50 ng mL^−1 for 24 h) increased the number
   of EVs without having a significant effect on EV size (Figure
   [175]4a–[176]c). Medium EVs had an average size of 130.2 ± 3.15 nm,
   consistent with the size of exosomes.[177]1 Treatment with bFGF (50 ng
   mL^−1 for 24 or 48 h) had no statistical significant effect on the
   release of LDH (Figure S4a, Supporting Information) or the amount of
   apoptotic neuronal nuclei (Figure S4b,c, Supporting Information), thus
   demonstrating that the increased number of EVs is not a consequence of
   neuronal cell death or compromised membrane integrity in bFGF‐treated
   neurons. FGFs act through FGF receptors that are classified as
   high‐affinity tyrosine kinase receptors.[178]80 In accord, the effect
   of bFGF on EV number was mediated through activation of tyrosine
   kinase‐dependent receptors, because co‐application of genistein (50
   µm), an inhibitor of (receptor) tyrosine kinase activity, decreased the
   effect of bFGF to sub‐normal levels (Figure [179]4d–[180]f). In line
   with a calcium‐dependent mechanism of bFGF‐induced exosome release,
   co‐application of the cell‐permeant calcium chelator BAPTA‐AM (1 µm)
   normalized the effect of long‐term bFGF treatment on exosome release
   (Figure [181]4g–[182]i). To further examine the effect of bFGF on
   exosome release, we quantified the abundance of the exosome‐associated
   proteins CD81, Alix/AIP1, and GFP (pHluorin) in EV‐enriched medium
   pellets from bFGF and VEH‐treated neurons. Consistent with our NTA
   data, we found that treatment with bFGF increased the abundance of
   these proteins, whereas co‐application of BAPTA‐AM normalized the
   effect of bFGF on CD81, Alix/AIP1 and GFP (Figure [183]4k–[184]l).
   Different from these compounds, transduction of neurons with
   CD63‐pHluorin alone had no effect on the abundance of EVs (Figure
   S3d,e, Supporting Information). In summary, these results demonstrate
   that prolonged (24 h) treatment with bFGF increases neuronal EVs in a
   calcium‐dependent manner. On the other hand, bFGF had no effect on the
   size or number of MVBs as measured by immunocytochemistry (ICC) and
   electron microscopy (EM) (Figure S5, Supporting Information).

Figure 4.

   Figure 4
   [185]Open in a new tab

   bFGF increases the release of extracellular vesicles (EVs) from
   cultured hippocampal neurons. a) Averaged curves illustrating the
   number of particles/size in the cell culture medium of bFGF‐ or vehicle
   (VEH)‐treated hippocampal neurons as measured by nanoparticle tracking
   analysis (NTA). b) Bar graphs demonstrating an increased particle
   number in bFGF‐treated neurons as compared to VEH‐treated cells as
   measured by NTA (n = 11 and 15 samples/condition; p < 0.001). c)
   Treatment with bFGF has no statistical significant effect on the
   average particle size (p = 0.2687). d) Results from NTA illustrating
   the number of particles/size in medium from bFGF or VEH‐treated
   hippocampal neurons with or without co‐application of 50 µm genistein
   (GEN). e,f) Genistein reduces the bFGF‐induced increase in particle
   number but had no effect on the size of particles (n = 5
   samples/condition; p < 0.001). g) Results from NTA illustrating the
   number of particles/size in medium from bFGF or VEH‐treated hippocampal
   neurons with or without co‐application of 1 µm BAPTA‐AM. h,i) BAPTA‐AM
   normalized the bFGF‐induced increase in particle number but had no
   effect on the size of particles (n = 3 samples/condition; p < 0.001).
   k) Western blot illustrating the abundance of CD81, Alix/AIP1, and GFP
   in the EV‐enriched pellet of cell culture medium from bFGF and
   VEH‐treated cells with or without co‐application of 1 µm BABPTA‐AM. l)
   Bar graphs illustrating an increase of CD81, Alix/AIP1, and GFP in the
   EV‐enriched pellet of bFGF‐treated neurons (VEH vs bFGF: n = 5; p =
   0.0006 for CD81, p = 0.0013 for Alix/AIP1, p = 0.0028 for GFP).
   Co‐application of BABPTA‐AM normalized the bFGF‐induced increase in
   these proteins in the EV‐enriched pellet. Data are shown as mean ±
   s.e.m. For comparison, a two‐tailed paired t‐test was used in (b) and
   an unpaired t‐test in (e,h,l) ***p < 0.001, **p < 0.01, *p < 0.05.

2.4. bFGF Modulates the Abundance of EV‐Enriched SNARE Proteins

   We instead hypothesized that bFGF affects exosome release through
   stimulating MVB‐PM fusion. Previous research implicated a number of
   molecules in mediating MVB‐PM fusion, such as Rab GTPases[186]8,
   [187]9, [188]10, [189]11, [190]12, [191]13 and soluble NSF‐attachment
   receptor (SNARE) protein family members.[192]12, [193]14, [194]15,
   [195]16, [196]17, [197]18, [198]19, [199]20 The abundance of these
   molecules on the MVB limiting membrane is likely mirrored by their
   abundance on exosomes as a result of their mechanism of generation with
   inward budding into the lumen of MVBs. Investigating exosomal proteins
   will thus support the understanding of molecular changes of the MVB
   limiting membrane. Therefore, instead of investigating cell lysates, we
   performed label‐free quantification (LFQ) of proteins using
   high‐resolution mass spectrometry (MS) on isolated EV‐enriched pellets
   from bFGF and VEH‐treated neurons (n = 6 per condition) (Figure
   [200]5a). Principal component analysis (PCA) revealed a clear
   discrimination between the treatment conditions (Figure [201]5b),
   further confirmed by unsupervised hierarchical clustering of the
   differentially expressed proteins (Figure [202]5c). The Pearson's
   correlation (PC) analysis of log[2] transformed protein LFQ intensities
   revealed a clear separation between the two groups and an excellent
   reproducibility of the replicates within the two groups (Figure S6,
   Supporting Information). In total 2258 distinctly expressed proteins
   were identified (Table S1, Supporting Information). In total, 705
   proteins showed a significantly changed abundance in EVs (FDR < 0.05)
   in response to treatment with bFGF. Out of these differentially
   abundant proteins, 441 proteins had a fold change of 1.5 or above with
   a p‐value < 0.05. Among those 441 proteins, 235 proteins had an
   increased and 206 proteins had a decreased abundance, respectively
   (Figure [203]5d). Based on functional pathway enrichment analysis, we
   found strong enrichment in the categories “vesicle mediated transport,”
   “regulation of vesicle mediated transport,” and “import to the cell”
   (Figure [204]5e). We next developed an interaction network and ran a
   centrality analysis to determine critical signaling hub proteins. We
   computed four network parameters: i) betweenness, ii) closeness, iii)
   stress, and iv) degree. Based on this analysis, the centrality ranks of
   53 proteins were calculated (Figure [205]5f). Among these proteins, the
   v‐SNARE proteins vesicle‐associated membrane protein 2 and 3 (VAMP2 and
   VAMP3) had the highest centrality measures (Figure [206]5g).
   Remarkably, bFGF decreased the abundance of VAMP2 (0.69‐fold) whereas
   it increased the abundance of VAMP3 (1.71‐fold) in EV‐enriched medium
   pellets. Visualization of their functional interaction partners
   demonstrated their functional relationship with a variety of proteins
   implicated in membrane fusion (Figure [207]5h). In summary, our
   proteomic data analysis thus demonstrates bFGF to affect key intrinsic
   molecular mechanisms responsible for vesicle membrane fusion in
   hippocampal neurons.

Figure 5.

   Figure 5
   [208]Open in a new tab

   Treatment with bFGF modulates the abundance of SNARE proteins on EVs.
   a) Schematic illustrating the workflow used to analyse the proteomic
   data from MS analysis. b) PCA plot illustrating a segregation of the
   two treatment condition (bFGF vs VEH). c) Hierarchical clustering and
   heat map showing the relative expression value (z‐score and
   log[2]‐transformed LFQ protein intensities) of 705 differentially
   abundant proteins (FDR < 0.05) in EVs from bFGF‐ and VEH‐treated
   neurons. Each column represents a replicate of either treatment
   condition and each row corresponds to a differentially abundant
   protein. d) Volcano plot of statistical significance (−log[10]
   transformed p‐value, −log[10](0.05) = 1.30) against log[2] transformed
   protein LFQ ratios between the two treatment conditions. Blue
   (decreased) and red (increased) dots indicate proteins with
   significantly changed abundance. e) Bar graph illustrating the results
   of a functional pathway analysis based on Gene Ontology, Reactome, and
   KEGG resources. f) Network analysis illustrating candidate protein
   interactions. g) Table illustrating the top 53 hub proteins against
   four centrality calculations in a network topology. h) Protein–protein
   interaction network illustrating functional interaction partners of
   VAMP2 and VAMP3. Source to target topology is considered for network
   depiction.

2.5. bFGF Enhances Neuronal EV Release in a VAMP3 Dependent Mechanism

   Based on our results we hypothesized bFGF to affect MVB‐PM fusion and
   exosome release by modulating the expression of v‐SNARE proteins. In
   order to test this hypothesis, we first examined the abundance of VAMP2
   and 3 in neuronal cell lysates and EV‐enriched pellets from bFGF‐ or
   VEH‐treated hippocampal neurons by Western blot. In accord with our MS
   data, we found the abundance of VAMP2 to be decreased in cell lysates
   from cultured neurons in response to treatment with bFGF, whereas the
   abundance of VAMP3 remained unaffected (Figure [209]6a–[210]c). In
   contrast to cell lysates, the abundance of VAMP3 was increased in
   EV‐enriched medium pellets from bFGF‐treated neurons, where VAMP2 was
   not detectable (Figure [211]6d,[212]e). Since exosomes are generated
   from inward budding of MVB membrane patches,[213]1 a high level of
   VAMP3 on exosomes may be reflected by an increased abundance of VAMP3
   on the cytoplasmic MVB membrane and thus help to fuse MVBs to the PM.
   In order to test this hypothesis, we used a siRNAs specific for Vamp3
   to knock‐down VAMP3 in cultured neurons (Figure [214]6f,[215]g; Figure
   S7, Supporting Information). Consistent with a role of VAMP3 for EV
   release, we found that siRNA‐mediated knock‐down of VAMP3 attenuated
   the number of EVs for both VEH‐ and bFGF‐treated neurons as measured by
   NTA (Figure [216]6h,[217]i). In accord with a specific effect of VAMP3
   on MVB‐PM fusion, the EV size remained unaffected with Vamp3 siRNA
   treatment (Figure [218]6k). Our results therefore suggest that bFGF
   increases EV release in part through a VAMP3‐dependent mechanism. Our
   results suggest a non‐redundant functional role of the homologs VAMP2
   and VAMP3 for EV release which can serve as a molecular basis for the
   segregation of MVB and neurotransmitter carrying synaptic vesicle
   fusion events, requiring the presence of VAMP2.[219]81

Figure 6.

   Figure 6
   [220]Open in a new tab

   The bFGF‐induced increase in EV release is partly mediated through
   VAMP3. a) Western blot illustrating the abundance of VAMP2 and VAMP3 in
   cell lysates from bFGF‐ and VEH‐treated neurons. b,c) The abundance of
   VAMP2 is decreased in neuronal cell lysates in response to
   bFGF‐treatment, whereas the abundance of VAMP3 remains unaffected (n =
   26/condition; VAMP3: p = 0.6541; VAMP2: p = 0.0252). d) Western blot
   illustrating the abundance of VAMP2 and VAMP3 in EV‐enriched pellets
   from bFGF‐ and VEH‐treated neurons. e) Treatment with bFGF increases
   the abundance of VAMP3 on EVs released from cultured neurons, whereas
   VAMP2 is undetectable on neuronal EVs (n = 12/condition; p = 0.0027
   f,g) Western blot and bar graph illustrating the knock‐down of VAMP3 in
   response to treatment with Vamp3 siRNAs (n = 12/condition; p < 0.001).
   h) Averaged curves from NTA illustrating the the number of
   particles/size in the cell culture medium of bFGF or VEH‐treated
   hippocampal neurons with our without co‐application of Vamp3 siRNAs.
   i,k) Bar graph illustrating a reduced number of EVs in neurons
   pre‐treated with Vamp3 siRNA (10 nm for 48 h). Co‐application of Vamp3
   siRNAs partially prevents the bFGF‐mediated increase in EVs (VEH vs
   bFGF vs Vamp3 siRNA vs bFGF + Vamp3 siRNA: number of samples/condition
   n = 24 vs 12 vs 24 vs 12; p < 0.0001). Treatment with bFGF or Vamp3
   siRNAs has no effect on the size of EVs as measured by NTA. Data are
   shown as mean ± s.e.m. For comparison, a two‐tailed unpaired t‐test was
   used in (c), (e), and (g) and a paired t‐test was used (i) ***p <
   0.001, **p < 0.01, *p < 0.05.

3. Discussion

   In our study, we examined the effect of electrical stimulation on
   neuronal EV release in single cells and in real time. Although cultured
   hippocampal neurons exhibit a constitutive release of EVs, we found
   MVB‐PM fusion to be a relatively rare event in cultured neurons under
   unstimulated conditions (Figure [221]1). Because the vast majority of
   our cultured neurons at DIV 9–12 had spontaneous miniature excitatory
   postsynaptic currents (mEPSCs), spontaneous action potentials (APs),
   and spontaneous calcium oscillations (data not included), these results
   suggest that basal neuronal activity is uncoupled from neuronal MVB‐PM
   fusion and exosome release. Our data therefore implies that neuronal
   activity and exosome release are segregated in neuronal cells under
   physiological conditions, presumably as a consequence of the different
   effector molecules required for synaptic vesicle versus MVB‐PM fusion
   and their differential sensitivity to cellular stimuli. Consistent with
   this view, the application of a HFS induced MVB‐PM fusion events in a
   small population of neurons (≈10%) (Figure [222]1d–[223]f). Whereas
   MVB‐PM fusion followed the HFS with a time lag of several tens of
   seconds, the same HFS evoked a fast and immediate increase in
   intracellular calcium (Figure [224]1g,[225]h). These results thus
   demonstrate that different from neurotransmitter vesicle fusion, MVB‐PM
   fusion is not directly coupled to a rise in intracellular calcium and
   likely involves a complex network events downstream the initiating
   calcium signal. The results from investigating the effect of
   store‐derived calcium signals (Figure S2, Supporting Information)
   further suggests that the calcium signal required for EV release is
   likely to be more complex than a simple rise in cytosolic calcium. Our
   results put previous reports, which directly connect MVB‐PM fusion and
   neuronal exosome release to neuronal activity, in a different
   perspective.[226]41, [227]42, [228]43, [229]44, [230]45 The results
   from these previous studies were derived from quantifying EVs in the
   cell culture medium in response to long‐lasting neuronal membrane
   depolarization without examining the effect of a brief and well‐defined
   electrical stimulus on MVB‐PM fusion. Thus, these results may be a
   consequence of using a non‐physiological stimulus. At present, it is
   unclear, why the effect of HFS is restricted to such a small neuronal
   population. Akin to presynaptic terminals, where a readily releasable
   pool (RRP) of transmitter vesicles is required for successful
   neurotransmitter release,[231]82 PM‐docked and therefore releasable
   MVBs may only be present in a small subset of neurons. In B‐lymphocytes
   two pools of MVBs can be identified based on their membrane cholesterol
   content, where only MVBs with high membrane cholesterol levels are able
   to fuse with the plasma membrane and release exosomes.[232]83
   Alternatively, stimulus‐responsive MVB‐PM fusion may require the
   expression of cell‐type specific SNARE‐associated molecules. Because of
   their particular role for EV biology, we therefore suggest the novel
   term e‐SNARE for SNARE proteins implicated in the release of exosomes.
   Our results imply that growth factor stimulate EV release by affecting
   the molecular machinery required for vesicle fusion, which include
   SNAREs such as VAMP3 (Figure [233]5 & [234]6). The fact that blocking
   receptor tyrosine kinases by genistein decreases the number of EVs to
   sub‐normal levels (Figure [235]4d,[236]e) suggests a baseline
   activation in cultured neurons by endogenous pleiotropic growth factor
   support. Because of its unspecific effect on receptor tyrosine kinases,
   genistein likely affects a number of growth factor signaling networks
   other than downstream from bFGF receptors, including the classical
   neurotrophins nerve growth factor (NGF) and brain derived neurotrophic
   factor (BDNF).[237]84 Interestingly, we found these latter molecules to
   mimic the effect of bFGF on EV release (Figure S8, Supporting
   Information). These results point toward a broader regulatory role of
   neurotrophic factors for neuronal EV release and future studies should
   examine the specific effects of distinct neuronal growth factors on EV
   release. On the other hand, although growth factors enhance
   stimulus‐responsive EV release, their role is additive to other
   factors, since genistein did not entirely block EV release in these
   cells. In addition, the effect of genistein on EV release exceeded the
   effect of BAPTA treatment, likely because of its additional off‐target
   effects.[238]85 Future research will need to examine the specific
   cellular and molecular factors that determine the availability of MVBs
   for stimulus‐evoked PM fusion and the different populations of
   releasable MVBs in neuronal cells. In support of such studies, emerging
   evidence indicates that the coupling of neuronal EV release to neuronal
   activity could be functionally relevant for plasticity‐associated
   processes in the CNS.[239]17, [240]30, [241]31, [242]41, [243]86,
   [244]87, [245]88

   In the present study, we examined the effect of bFGF on EV release in
   cultured hippocampal neurons. Our results demonstrate that bFGF
   stimulates EV release from cultured hippocampal neurons (Figure
   [246]4). Our proteomic analysis and molecular biology experiments
   further suggest bFGF to affect key cellular protein circuits that
   control vesicle membrane fusion (Figure [247]5). In summary, our data
   thus reveal a new and previously unrecognized function of bFGF in
   controlling neuronal EV release, and implicate the SNARE protein VAMP3
   in the bFGF‐mediated effect on EV release. Since MVB‐PM fusion and
   neurotransmission are both mediated by SNARE proteins and because both
   are regulated by similar stimuli, the question raises how neurons
   segregate neurotransmitter release from exosome release. The SNAREs
   that are used for synaptic vesicle exocytosis are VAMP2 (also called
   synaptobrevin 2), syntaxin 1a, and SNAP‐25.[248]89 In accord, mice
   lacking VAMP2 have a significantly reduced synaptic transmission
   (around 10% of control),[249]81 emphasizing the role of VAMP2 for
   synaptic vesicle fusion, where VAMP3 is not required. Our biochemical
   results (Figure [250]6) demonstrate a strong signal for VAMP3 in EVs,
   whereas VAMP2 was not detectable by Western blot, suggesting low
   abundance. Our results suggest a non‐redundant functional role of VAMP2
   and VAMP3 for EV release which may serve as a molecular basis for the
   segregation of neurotransmitter and stimulus‐evoked EV release in
   neurons. Although an attractive model, the role of VAMP 3 for EV
   release is likely to be more complex, because we found that treatment
   with Vamp3 siRNA had no consistent effect on the abundance of the EV
   signature proteins CD81, CD9, Alix or Flotilin‐1 in EV‐enriched cell
   culture medium pellets (Figure S9, Supporting Information). A number of
   reasons may account for such inconsistencies between the two different
   assays. First, bFGF has a profound effect on the abundance of
   EV‐enriched proteins, as demonstrated by the MS‐based quantification of
   the ExoCarta top 100 EV‐enriched proteins in bFGF and VEH‐treated
   neurons (Figure S10 and Table S2, Supporting Information), that is,
   bFGF affects the abundance of these proteins in each vesicle and
   independent from its effect on the number of vesicles. Second, recent
   evidence suggests that various EV subpopulations may exist within a
   given biological specimen, each of them with a unique protein
   composition, internal content and function.[251]90 Both of these
   factors fairly contribute to EV complexity in biological fluids and
   future research thus needs to address the specific effect of bFGF and
   VAMP3 on distinct EV subpopulations and their protein composition.
   Moreover, VAMP2 and 3 interact with numerous other syntaxins (Figure
   [252]5h), thus adding another layer of complexity.

   Our biochemical (Figure S1f, Supporting Information) and NTA data
   (Figure [253]4a) suggest that bFGF predominantly affects the abundance
   of exosomes over other types of EVs. To confirm this hypothesis, we
   applied pH‐sensitive dye imaging where pHluorin is tagged to the
   EV‐enriched tetraspanin CD63. In line with our biochemical data, we
   found that treatment with bFGF increased the stimulus‐evoked increase
   in pHluorin fluorescence (Figure [254]2), thus further demonstrating
   that bFGF stimulates the release of exosomes over other types of EVs.
   The latter interpretation is however mitigated by restriction of the
   imaging method we applied: since we used patch‐clamp electrophysiology
   in conjunction with conventional epifluorescence microscopy for imaging
   CD63‐pHluorin instead of using total internal reflection microscopy
   (TIRF), an increase in fluorescence could be a result of MVB‐PM fusion
   and exosome release, or alternatively, be due to an abrupt,
   stimulus‐induced increase in the pH‐value inside CD63‐tagged MVB in the
   absence of MVB‐PM fusion. We are aware of these concerns but favor the
   former explanation because i) bFGF treatment clearly results in an
   increased abundance of EVs in the cell culture medium (Figure
   [255]4a,[256]b) and ii) because treatment with bFGF rather decreases
   the pH‐value inside CD63‐tagged MVBs under unstimulated conditions
   (Figure [257]3a–[258]e). In fact, it would be even tempting to
   speculate about an increased EV release as a result of a decreased
   endo‐lysosomal pH‐value, similar to what has been shown for synaptic
   vesicles,[259]74 although this hypothesis clearly deserves further
   investigation. Although our data favor a loose coupling between calcium
   and EV release, we found bFGF‐stimulated exosome release to be
   calcium‐dependent after prolonged (24 h) treatment, as BAPTA‐AM
   normalized the effect of bFGF (Figure [260]4g–[261]i) and on the other
   hand bFGF increased cytoplasmatic calcium concentrations (Figure
   [262]3f), presumably as a result of an activation of
   voltage‐independent calcium stores.[263]77, [264]78, [265]79 Moreover,
   epifluorescence imaging allowed us to capture a wider area including
   the entire soma and dendritic arbor, where TIRF is restricted to a
   small area due to the curved geometry of neuronal cells.

   Although calcium has become a common theme for exosome release, other
   studies found MVB‐PM fusion to be independent from intracellular
   calcium in HeLa cells.[266]20 Future studies will need to examine the
   specific role of the intracellular neuronal calcium domains for the
   effect of bFGF on MVB‐PM fusion. In principal, increase in exosomes is
   due to an enhanced MVB‐PM fusion, or be the consequence of an increased
   MVB size or number. Our data point toward an effect of bFGF on MVB
   availability or fusion rate because we found no difference in the size
   or number of MVBs, although there was a trend towards more and bigger
   MVBs in our EM analysis (Figure S5, Supporting Information). This is in
   contrast to previous reports, demonstrating the volume of MVBs to
   increase in response to treatment with the neurotrophin
   prosaposin[267]91 or when the neurotrophin receptor TrkB is
   overexpressed.[268]92

   Although providing new insights into the molecular mechanism of
   neuronal exosome release, the translational potential of our results is
   limited by the lack of in vivo data. Therefore, future studies will
   have to confirm our results in vivo, for example, by using suitable
   reporter animal models that have been published recently.[269]25 It
   would be interesting to determine, if EV release exhibits short‐ or
   long‐term changes through electrical stimulation in vivo, like this is
   known to be the case for hippocampal synaptic transmission (i.e.,
   long‐term potentiation). Such in vivo or ex vivo experiments will
   provide further insight into EV‐mediated cell‐to‐cell communication in
   the intact CNS, although adding another layer of complexity that comes
   with the presence of different cell types in the CNS and thus making
   these experiments technically more challenging.

4. Conclusion

   In summary, our results thus reveal a new function for neurotrophic
   factor signaling in controlling activity‐dependent neuronal EV release
   and support the investigation of growth factor‐mediated signal
   transduction via EVs. Our results aid the examination of growth
   factor‐mediated EV release in pathological conditions, where EVs
   contribute to the progression of neurodegenerative diseases, such as
   Parkinson's disease (PD) or Alzheimer's disease (AD).[270]93, [271]94,
   [272]95, [273]96, [274]97 Furthermore, our results will promote the
   investigation of EV‐associated changes in responses to neuronal
   injury[275]48, [276]49, [277]50 or psychiatric conditions where bFGF
   has been shown to be strongly implicated (reviewed in refs. [278]51,
   [279]52).

5. Experimental Section

Cell Culture

   Animals were handled in accordance with the regulations of local
   authorities and the animal welfare committee of the Ludwig Maximilian
   University Munich, Germany. Rat hippocampal neurons were prepared from
   embryonic day 18 CD rats (Charles River) and cultivated in Neurobasal
   media containing 2% B27, 0.25% glutamine, and 0.125% glutamate
   (Invitrogen) as described before.[280]98 Unless otherwise indicated,
   neurons were treated with 50 ng mL^−1 basic fibroblast growth factor
   (Peprotech) diluted in PBS.

Lentiviral Production and Virus Transduction

   CD63‐pHluorin, where pHluorin is cloned into the second luminal loop of
   mouse CD63, was expressed from a modified FUGW lentiviral vector under
   the control of the neuron‐specific human synapsin promoter
   (pCD63‐pHluorin). The generation of the plasmid is described in detail
   elsewhere.[281]74 pCD63‐pHluorin was obtained from Dr. Benjamin Rost
   (German Centre for Neurodegenerative Research, Berlin, Germany). Viral
   particles were produced as described before.[282]99 Unless described
   otherwise, rat hippocampal neurons were transduced with CD63‐pHluorin
   lentiviral particles at day 3 in vitro (DIV). After 24 h, the medium
   was removed and the cells washed three times with PBS and fresh medium
   was replenished.

Small Interfering RNA Transfection

   Small interfering RNAs (siRNAs) (Silencer Select siRNAs, Thermo Fisher
   Scientific) targeting VAMP3 (#4 390 771) or control siRNAs (#4 390 843)
   were mixed with Lipofectamine RNAi Max according to the manufacturer's
   recommendations. The siRNAs were applied at DIV7 at a final
   concentration of 10 nm and the cells were incubated with the siRNAs for
   the following 2 days until DIV 9, the cells were harvested for Western
   blot and the medium collected for NTA.

Live Cell Microscopy

   All live cell imaging experiments were performed on an inverted Leica
   DMI6000 B microscope controlled via Leica Application Suite X (LAS X)
   software. For CD63‐pHluorin imaging, hippocampal neurons, cultured on
   glass bottom cell culture dishes (Ibidi), were transferred to the
   microscope stage at DIV 12–14 and perfused with ACSF containing the
   following: 140 mm NaCl, 2.4 mm KCl, 10 mm Hepes, 10 mm glucose, 4 mm
   CaCl[2], and 4 mm MgCl[2], 320 mOsmol/L, pH 7.4. To visualize pHluorin
   fluorescence, the cells excited with a UV light source equipped with a
   488 nm excitation filter and imaged at 1 frame per second by a CD
   camera (Hamamatsu). Ammonium chloride (NH[4]Cl) was applied at a
   concentration of 50 mm diluted in ACSF through a ValveLink8.2 perfusion
   system (AutoMate). Fura‐2 AM calcium imaging was performed as described
   before.[283]100 Leica Application Suite X (LAS X) software Graph Pad
   Prism 7 were used for data analysis.

Preparation of EV‐Enriched Pellets from Neuronal Cell Culture Medium

   To prepare EV‐enriched pellets, cell culture medium was changed at DIV
   11 and collected after 24 h (5 mL medium/500 000 cells). EVs were
   isolated by differential ultracentrifugation as previously
   described[284]5 with minor modifications. In brief, the conditioned
   medium was centrifuged at 300 × g for 10 min at 4 °C to pellet cells.
   The supernatant was centrifuged at 1 000 × g for 20 min at 4 °C,
   transferred to new tubes, and centrifuged in a MLA‐80 rotor (Beckman)
   for 90 min at 100 000 × g to obtain a 100 K pellet. All pellets were
   washed in 1 mL of PBS and re‐centrifuged at the same speed in a TLA‐55
   rotor (Beckman) before being re‐suspended in 35 µL of sterile PBS or
   otherwise in 35 µL RIPA buffer. The resulting pellet was defined as
   EV‐enriched medium pellet.

Iodixanol Gradient Separation

   Iodixanol gradients were prepared as described.[285]5 In brief,
   EV‐enriched medium pellets (20 mL medium/1‐2 mio. cells) were washed
   and resuspended in 1.5 mL buffer containing: 0.25 m sucrose, 10 mm Tris
   pH 8.0, 1 mm EDTA (pH 7.4), and mixed 1:1 with 60% (w/v) stock solution
   of iodixanol/Optiprep. A 40% iodixanol working solution was prepared
   and used to prepare 20% and 10% (w/v) iodixanol solutions. Next, 1.3 mL
   20% (w/v) iodixanol and 1.2 mL 10% iodixanol were layered on top of the
   EV suspension and tubes were centrifuged for 1 h at 4 °C at 350 000 × g
   in a SW55Ti rotor (Beckman). Ten fractions of 490 µL were collected
   from the top of the tube. The density was assessed with a refractometer
   and the fractions were diluted with 2.5 mL PBS and centrifuged for 30
   min at 100 000 × g in a TLA 110 rotor (Beckman). These concentrated
   fractions were resuspended in 30 µL of RIPA and separated on gels.

Western Blot

   Cultured neurons or EV‐enriched pellets were lyzed in RIPA buffer incl.
   a protease and phosphatase inhibitor cocktail (Complete Protease
   Inhibitor Cocktail, PhosStop Phosphatase Inhibitor Cocktail, both
   Roche). The lysates were incubated for 30 min on ice and centrifuged at
   5000 × g for 15 min at 4 °C, and the supernatant retrieved. Protein
   concentrations were determined using the BCA method (Thermo Fisher
   Scientific) and a spectrophotometer (NanoDrop, Thermo Fisher
   Scientific). After adjusting protein concentrations, protein lysates
   were heated to 75 °C for 15 min in Laemmli sample buffer containing 10%
   beta‐mercaptoethanol. SDS‐PAGE was performed by using Mini‐Protean TGX
   Gels (Bio‐Rad) and a Tris‐glycine‐based running buffer. The protein was
   blotted onto polyvinylidene difluoride (PVDF) membranes (Bio‐Rad) at
   150 mA for 60–120 min on ice. The membranes were blocked with 5% dry
   milk in TBST wash buffer (tris‐buffered saline containing 0.05% tween)
   for 1 h and incubated at 4 °C overnight under gentle shaking with the
   primary antibody in TBST/5% BSA (Cell Signaling Technology). The
   membranes were washed and incubated with the respective HRP‐conjugated
   secondary antibody (Vector Labs) in TBST/5% milk for 1 h, followed by
   further washing and exposure to Clarity Western blot ECL Substrate
   (Bio‐Rad) or ECL Prime (GE Healthcare). Chemiluminescence was detected
   with LI‐COR Odyssey Fc Imaging system and analyzed by Image Studio
   software (Licor). Western blot images were further processed by ImageJ
   and the optical densities of target proteins were scaled to the
   respective loading controls for statistical analysis. Therefore, a
   normalized experimental signal (measured experimental signal/lane
   normalization factor) was used, where the lane normalization factor was
   calculated from the ratio between the measured signal of a housekeeping
   protein for each lane divided by the highest measured signal of
   housekeeping protein on the blot. For quantifying exosomal proteins by
   Western blot, the signal was normalized to the protein concentration of
   the cell lysate.

Reagents and Antibodies

   The following primary antibodies were used for Western blot and ICC:
   anti‐GFP (GTX113617, GeneTex), anti‐Alix/AIP1 (ABC40, Merck), anti‐CD81
   (sc‐166029, Santa Cruz Biotechnology), rabbit anti‐rat Rab5 (C8R1, Cell
   Signaling Technology), rabbit anti‐rat Rab7 (D95F2, Cell Signaling
   Technology), anti‐Rab11 (D4F5, Cell Signaling Technology), rabbit
   anti‐rat EEA1 (C45B10, Cell Signaling Technology), anti‐LBPA (Z‐PLBPA,
   Echelon Biosciences), anti‐CHC (D3C6, Cell Signaling Technology). The
   following chemical compounds and growth factors were used: BAPTA‐AM
   (Tocris), Genistein (Tocris), EGTA (Tocris), bFGF (Peprotech),
   Poly‐l‐Lysine (Sigma), Tetrodotoxin (TTX) (Abcam), Tetraethylammonium
   chloride (TEA) (Abcam).

Immunocytochemistry and Confocal Microscopy

   For ICC, cultured neurons were fixed for 15 min at room temperature in
   4% paraformaldehyde, washed once for 10 min in PBS, and permealized for
   1 min with 0.125% Triton X‐100 in PBS. Unspecific antibody binding was
   blocked by 1% bovine serum albumin in PBS for 1 h followed by
   incubation with the respective primary antibody at 4 °C overnight.
   Unbound antibodies were removed by washing with PBS and the cells were
   incubated with Alexa Fluor‐conjugated secondary antibodies for 1 h at
   room temperature. Unbound secondary antibodies were removed by washing
   in PBS and the cells were air‐dried, covered with fluorescence mounting
   medium (Dako), and sealed under cover glass. Images were captured using
   a Leica SP5 confocal microscope and analyzed using Fiji software
   ([286]https://fiji.sc/Fiji).

Electrophysiology and Imaging

   All electrophysiological experiments were performed on rat hippocampal
   neurons at DIV 12–14 grown on glass coverslips (German Glass)
   pre‐treated with 65% nitric acid and coated with 1 mg mL^−1
   Poly‐d‐Lysin (Sigma). Experiments were performed on an upright Olympus
   BX51 microscope equipped with a Heka EPC 10 USB amplifier (Heka) under
   the control of Patchmaster 10 software (Heka). Intracellular (pipette)
   solution contained the following: 136 mm KCl, 17.8 mm Hepes, 1 mm EGTA,
   0.6 mm MgCl[2], 4 mm NaATP, 0.3 mm Na[2]GTP, 15 mm creatine phosphate,
   and 5 U mL^−1 phosphocreatine kinase, 315–320 mOsmol L^−1, pH 7.4.
   Extracellular solution contained the following: 140 mm NaCl, 2.4 mm
   KCl, 10 mm Hepes, 10 mm glucose, 4 mm CaCl[2], and 4 mm MgCl[2], 320
   mOsmol L^−1, pH 7.4; patch‐pipette solution 136 mm KCl, 17.8 mm Hepes,
   1 mm EGTA, 0.6 mm MgCl[2], 4 mm NaATP, 0.3 mm Na[2]GTP, 15 mm creatine
   phosphate, and 5 U mL^−1 phosphocreatine kinase, 315–320 mOsmol L^−1,
   pH 7.4. Electrophysiological data were analyzed with Axograph X version
   1.6.5. (Axon Instruments). Live‐cell imaging videos in conjunction with
   patch‐clamp experiments were captured time locked to an electric
   stimulus by using a mounted CD camera (ExiBlue, QImaging) controlled by
   µManager software at a frame rate of 10 frame per second.

Nano‐Particle Tracking Analysis

   For NTA experiments, the cell culture medium was changed at DIV 11 and
   collected after 24 h (1.5 mL medium/150 000 cells). A NanoSight LM10
   NTA apparatus (Malvern) was used for all NTA measurements. For each
   sample, at least three videos of 60 s with more than 200 detected
   tracks per video, and in at least one dilution, were taken and analyzed
   using the NanoSight LM10 NTA software v3.00 and Graph Pad Prism 7.

Single Cell Imaging

   For local (single cell) calcium measurements, neurons were filled with
   0.05 mm Oregon Green 488 BAPTA‐1 AM (Life Technology) in the
   patch‐pipette solution over a period of 2–2.5 min. Captured videos were
   edited and analyzed using Fiji (ImageJ) and Graph Pad Prism 7.

Quantification of Neuronal Cell Death

   Cell death in cultured neurons was quantified as described
   before.[287]101 In brief, lactate dehydrogenase (LDH) released into the
   culture medium was measured using the CytotoxOne Membrane Integrity
   Assay (Promega) according to the manufacturer's instructions with a
   Fluostar Omega fluorescence reader (BMG Labtech).

Electron Microscopy

   For EM, rat hippocampal neurons at DIV 12–14 were grown on glass
   coverslips (German Glass), pre‐treated with 65% nitric acid, and coated
   with 1 mg mL Poly‐d‐Lysin (Sigma). The medium was removed, the cells
   washed three times with PBS, and pre‐fixed for 5 min with 4% PFA. After
   this, they were washed three times with 75 mm cacodylate buffer
   including 150 mm NaCl and 2 mm MgCl[2] (fixation buffer) and post‐fixed
   with 0.2% osmium tetroxide. The cells were again washed once with
   buffer and three times with H[2]O[bidest]. Dehydration was carried out
   in a graded acetone series. The 20% acetone step additionally included
   1% uranyl acetate. Finally, the cells were embedded in the epoxy resin
   Epon 812. In general, embedding of the cells, sectioning, and
   post‐staining were carried out as described.[288]102 Transmission
   electron microscopy was performed on a Zeiss EM 912 equipped with an
   integrated OMEGA energy filter and operated at 80 kV in the zero‐loss
   mode and the images analyzed by Fiji (ImageJ) software.

LC‐MS/MS Analysis

   The EV‐enriched pellets were lyzed in 80 µL of a modified RIPA lysis
   buffer (50 mm TrisHCl pH 8, 150 mm NaCl, 5 mm EDTA, 1% v/v Triton
   X‐100, 0.5% w/v sodium deoxycholate, 0.1% w/v) with protease inhibitors
   (Sigma‐Aldrich, USA) on ice with intermediate vortexing. 20 µL H[2]O,
   10 µL 100 mm MgCl[2], and 25 units Benzonase (Sigma‐Aldrich, USA) were
   added followed by an incubation for 30 min at 37 °C at 1400 rpm in a
   Thermomixer (Eppendorf). Undissolved material was removed by
   centrifugation for 5 min at 20 000 g and 4 °C. The supernatants were
   transferred to 1.5 mL protein Lobind Tubes. Proteins were reduced by
   addition of 9 µL of 200 mm dithiothreitol (Biozol) in 50 mm ammonium
   bicarbonate and incubation for 30 min at 37 °C. Cysteine residues were
   alkylated by addition of 18 µL 400 mm iodoacetamide (Sigma) and
   incubation for 30 min at room temperature in the dark. Afterward, the
   reaction was quenched by adding another 9 µL of 200 mm dithiothreitol.
   Proteolytic digestion was performed using a modified protocol for
   single‐pot solid‐phase enhanced sample preparation (SP3).[289]103
   Briefly, after binding of proteins to 40 µg of a 1:1 mixture of
   hydrophilic and hydrophobic magnetic Sera‐Mag SpeedBeads (GE
   Healthcare) with a final concentration of 70% acetonitrile for 30 min
   at room temperature, beads were washed twice with 200 µL 70% ethanol
   and twice with 180 µL acetonitrile. For proteolytic digestion, 125 ng
   LysC and 125 ng trypsin (Pomega) were added in 20 µL 50 mm ammonium
   bicarbonate followed by an incubation for 16 h at room temperature. The
   supernatants were transferred to fresh 0.5 mL protein lobind tubes
   (Eppendorf). For improved peptide recovery, 20 µL 0.1% formic acid were
   added to the magnetic beads followed by sonication for 30 s in a
   sonication bath (Hielscher Ultrasonics GmbH). The supernatants were
   combined and dried by vacuum centrifugation. Peptides were re‐dissolved
   in 20 µL 0.1% formic acid. The peptide concentration was estimated
   using Nanodrop at 280 nm (Thermo).

   The peptides were analyzed on an Easy nLC 1000 nanoHPLC (Thermo
   Scientific) which was coupled online via a Nanospray Flex Ion Source
   (Thermo Sientific, USA) equipped with a PRSO‐V1 column oven (Sonation)
   to a Q‐Exactive mass spectrometer (Thermo Scientific). An amount of 1
   µg of peptides per sample was separated on an in‐house packed C18
   column (30 cm × 75 µm ID, ReproSil‐Pur 120 C18‐AQ, 1.9 µm, Dr. Maisch
   GmbH) using a binary gradient of water i) and acetonitrile ii)
   supplemented with 0.1% formic acid (0 min, 2% B; 3:30 min, 5% B; 137:30
   min, 25% B; 168:30 min, 35% B; 182:30 min, 60% B) at 50 °C column
   temperature. A data‐dependent acquisition method was used. Full MS
   scans were acquired at a resolution of 70 000 (m/z range: 300–1400, AGC
   target: 3E+6). The ten most intense peptide ions per full MS scan were
   selected for peptide fragmentation (resolution: 17 000, isolation
   width: 2 m/z, AGC target: 1E+5, NCE: 25%). A dynamic exclusion of 120 s
   was used for peptide fragmentation. The raw data were analyzed with the
   software Maxquant (maxquant.org, Max‐Planck Institute Munich) version
   1.6.1.0.[290]104

   The MS data were searched against a canonical fasta database of Rattus
   norvegicus from UniProt (download: March 5, 2018, 29 975 entries).
   Trypsin was defined as protease. Two missed cleavages were allowed for
   the database search. The option first search was used to recalibrate
   the peptide masses within a window of 20 ppm. For the main search
   peptide and peptide fragment mass tolerances were set to 4.5 and 20
   ppm, respectively. Carbamidomethylation of cysteine was defined as
   static modification. Acetylation of the protein N‐term as well as
   oxidation of methionine was set as variable modifications. The false
   discovery rate for both peptides and proteins was adjusted to less than
   1%. Label free quantification (LFQ) of proteins required at least two
   ratio counts of razor peptides. Only razor and unique peptides were
   used for quantification. The protein LFQ intensities were log[2]
   transformed and a two‐sided Student's t‐test was applied to evaluate
   the significance of proteins with changed abundance. Additionally, a
   permutation based false discovery rate estimation was used.[291]105

Network Architecture and Centrality Analysis

   Raw MS data were analyzed using R statistical environment version i386
   3.3.3 (http://www.r‐project.org./). A hierarchical cluster analysis was
   performed using Euclidean similarity metric in R and clustered
   heat‐maps were visualized using the Heatmap3 R package. In order to
   check for the reproducibility of the array replicates Pearson's
   correlation analysis was performed using R and plotted as heat‐maps. In
   order to overview the array replicates distribution and segregation PCA
   was performed using “procomp” function of the stat package in the R
   statistical environment.

   Accession IDs of 2258 protein hits from (MS) data were used to obtain
   protein information and gene IDs using Uniprot database[292]106 and
   were selected for further pathway analysis. Both up‐regulated and
   down‐regulated peptide hits with ±1.5‐fold enrichment in MS data were
   further considered for pathway enrichment analysis. METASCAPE[293]107
   ([294]https://metascape.org) was used in order to perform pathway
   enrichment analysis; which contains Kyoto Encyclopedia of genes and
   genomes (KEGG). Pathways were considered statistically significant
   using a p‐value ≤ 0.05. Consideration for further analysis was given to
   the candidates contributing to enrichment of the pathways reported in
   vesicle mediated transport, regulation of vesicle mediated transport,
   and import to the cell.

   Protein–protein interactions (PPINs) were retrieved by using STRING
   database.[295]108 A parameter of high confidence (0.7) was set for
   functional interaction among the candidates of interest and an
   intereactome among candidate proteins was obtained. The intereactome
   maps obtained from STRING database were visualized by using Cytoscape
   v3.6.1 ([296]https://www.cytoscape.org/)[297]109 and hub proteins were
   obtained through centrality analysis.

Conflict of Interest

   The authors declare no conflict of interest.

Supporting information

   Supporting Information
   [298]Click here for additional data file.^ (1.2MB, pdf)

   Supplemental Table S1
   [299]Click here for additional data file.^ (284.1KB, xlsx)

   Supplemental Table S2
   [300]Click here for additional data file.^ (20.6KB, xlsx)

   Supplemental Movie 1
   [301]Click here for additional data file.^ (6.5MB, avi)

   Supplemental Movie 2a
   [302]Click here for additional data file.^ (1.1MB, avi)

   Supplemental Movie 2b
   [303]Click here for additional data file.^ (4.5MB, avi)

   Supplemental Movie 3a
   [304]Click here for additional data file.^ (18.3MB, mov)

   Supplemental Movie 3b
   [305]Click here for additional data file.^ (18.1MB, mov)

   Supplemental Movie 3c
   [306]Click here for additional data file.^ (18.8MB, mov)

   Supplemental Movie 3d
   [307]Click here for additional data file.^ (17.1MB, avi)

Acknowledgements