Exosome secretion kinetics are controlled by temperature

Abstract

When multivesicular endosomes (MVEs) fuse with the plasma membrane, exosomes are released into the extracellular space where they can affect other cells. The ability of exosomes to regulate cells nearby or further away depends on whether they remain attached to the secreting cell membrane. The regulation and kinetics of exosome secretion are not well characterized, but probes for directly imaging single MVE fusion events have allowed for visualization of the fusion and release process. In particular, the design of an exosome marker with a pH-sensitive dye in the middle of the tetraspanin protein CD63 has facilitated studies of individual MVE fusion events. Using TIRF microscopy, single fusion events were measured in A549 cells held at 23–37°C and events were identified using an automated detection algorithm. Stable docking precedes fusion almost always and a decrease in temperature was accompanied by decrease in the rate of content loss and in the frequency of fusion events. The loss of CD63-pHluorin fluorescence was measured at fusion sites and fit with a single or double exponential decay, with most events requiring two components and a plateau because the loss of fluorescence was typically incomplete. To interpret the kinetics, fusion events were simulated as a localized release of tethered/untethered exosomes coupled with the membrane diffusion of CD63. The experimentally observed decay required three components in the simulation: 1) free exosomes, 2) CD63 membrane diffusion from the endosomal membrane into the plasma membrane, and 3) tethered exosomes. Modeling with slow diffusion of the tethered exosomes (0.0015–0.004 μm²/s) accurately fits the experimental data for all temperatures. However, simulating with immobile tethers or the absence of tethers fails to replicate the data. Our model suggests that exosome release from the fusion site is incomplete due to postfusion, membrane attachment.

Significance

Exosomes are nanoscale vesicles secreted from a wide variety of cells and they are involved with a number of disease states, from cancer to Alzheimer’s disease. For cells to secrete exosomes, multivesicular endosomes (MVEs) fuse with the plasma membrane in a constitutive fashion. Postfusion, exosomes circulate to affect cells both near and far. In this work, the fusion of MVEs was characterized by fluorescence microscopy and the rate of release was measured then simulated to develop a model for the fate of exosomes postfusion. Exosomes are secreted from MVEs more frequently at higher temperatures and modeling suggests that some exosomes remain attached to the cell surface for a period of time.

Introduction

Exosomes are a subset of small extracellular vesicles (sEVs) secreted from cells. They range from 30 to 100 nm in diameter and are formed from the inward budding of vesicles into the intraluminal space of late endosomes 1,2,3). Exosomes are secreted into the extracellular space when these multivesicular endosomes (MVEs) fuse with the plasma membrane (1,4,5,6). Once exosomes are secreted into the extracellular space, they can affect cells nearby and further away (7,8,9). Exosome secretion is used by healthy cells to maintain essential processes such as homeostasis (10) and cell motility (11,12); however, the release of exosomes can be exploited by unhealthy cells to aid in disease progression (13,14). Exosomes likely facilitate disease progression via the transfer of biomolecules from unhealthy to healthy cells (1,15,16,17,18). Specifically, exosomes and other sEVs have been shown to play a role in neurodegenerative diseases (19,20,21) and cancer (7). Many studies regarding disease states isolate sEVs, which are enriched in exosomes but not exclusively exosomes, therefore the term “sEV” is used. The discovery of exosomes and their involvement in disease states has motivated research into exosomes as markers for early detection and potential avenues for intervention.

Although MVE membrane fusion is likely an integral check point that can be harnessed to modulate exosome release, a mechanistic understanding of the fusion process is lacking. It is known that MVE fusion is a constitutive process but enhanced in the presence of Ca²⁺ (16,22). In a bulk sEV collection assay, more sEVs were collected from cells treated with Ca²⁺ ionophores (16). This was also observed in single fusion events, where ionophores increased the number of events (22); however, the magnitude of the effect of Ca²⁺ depended dramatically on the cell type (22). Basic information that can further our understanding of the fusion process, including whether MVEs are stably docked, identification of proteins that control fusion, and the dependence of MVE fusion on temperature, are unknown. This information can provide insights into similarities and differences between well-understood fusion mechanisms utilized by other vesicles, such as dense core vesicles, and potentially identify avenues for modulating exosome secretion.

Membrane fusion has been well studied regarding the regulated fusion of secretory vesicles, such as synaptic vesicles, dense core vesicles, and insulin granules. Analyzing the kinetics of content release during fusion between secretory vesicles and the plasma membrane has led to the identification of different fusion mechanisms, such as “kiss and run” (23) and a diversity of fusion modes that affect the release of content (24). As kinetic studies of different vesicles have provided highly important insights into stimulated exocytosis, similar studies regarding MVEs fusion will be essential to increase our understanding of exosome release.

To elucidate the kinetics of exosome secretion, a marker of intraluminal vesicles (ILVs), CD63, has been used to visualize single fusion events (25,26,27). CD63 is a tetraspanin protein present on the endosome and plasma membranes of cells and enriched on the membrane of ILVs that become exosomes upon MVE fusion (16). Several labs have tagged CD63 on the first extracellular loop with a pH-sensitive fluorescent protein (i.e., pHluorin or pHuji) such that the probe is quenched on the inside of the late endosome or when on the surface of an ILV (25,26,27). CD63-pHluorin is a pH-dependent probe, which remains quenched inside the acidic environment of an MVE and signals the onset of fusion by a sudden spike of fluorescence that gradually disperses radially.The rate at which fluorescence dissipates depends on how CD63-labeled exosomes leave the fusion site. Postfusion, exosomes have been observed to either diffuse far from the site of secretion (9) or remain close to the cell surface (28,29). Interestingly long-lasting fluorescence after fusion has been observed with CD63+ and CD81+ exosomes but not CD9+ exosomes (27). Tethering molecules, such as tetherin, may be involved with limiting the widespread release of a portion of exosomes (30). By observing the kinetics of single fusion events, we demonstrate that the release of exosomes from the fusion site is oftentimes noted to be incomplete and simulations of the process allow us to propose that this is due to the extent of exosome attachment to the cell surface.

In this work, single MVE fusion events were visualized using CD63 fluorescent probes and total internal reflection fluorescence (TIRF) microscopy. A non-small cell lung cancer cell line (A549) was used as a model system to investigate MVE fusion because they readily release exosomes, facilitating the imaging process, and exosomes have been shown to play an integral role in cancer progression (14,31,32,33). One challenge of measuring constitutive fusion events in a single vesicle fusion assay is the tedious process of manually finding and analyzing fusion events that occur at random points in time and in a relatively slow fashion (∼1–3 events per minute). To overcome this, an automated approach to detection and analysis capable of capturing both large and small fusion events that occur on top of a background of CD63-pHluorin on the plasma membrane was developed. Our results reveal that, before fusion, almost all MVEs are docked for at least 1 s, typically longer. The frequency of fusion increases and the kinetics of release is faster at higher temperatures. The kinetics of release, or loss of fluorescence postfusion, can relay information about the fate of exosomes. Through two types of analyses, 1) fitting the fluorescence loss and 2) simulations of the release event, we determine that a portion of exosomes are free to diffuse away from the fusion site, but some exosomes remain attached to the surface in many, but not all, fusion events. This attachment depends on temperature, with fewer exosomes remaining attached at higher temperatures.

Materials and methods

Cell culture

A549 cells were cultured in Dulbecco-modified Eagle’s minimum essential medium (DMEM) (Gibco 11965092). DMEM was supplemented with 10% fetal bovine serum (Sigma-Aldrich, St. Louis, MO). Cells were grown and maintained in a humidified 37°C, 5% CO₂ incubator. Cells used for microscopy were plated in LabTek 8-well dishes and were transiently transfected using Lipofectamine 3000 (Thermo Fisher Scientific, Waltham, MA) using 2.5 mg/mL of CD63-pHluorin plasmid DNA (gift from D.M. Pegtel (27)) according to the manufacturer’s protocol. For two-color MVE docking experiments, A549 cells were cotransfected with EGFP-CD63 and CD63-pHuji. CD63-pEGFP C2 was a gift from Paul Luzio (Addgene: http://n2t.net/addgene:62964; RRID:Addgene_62964) and CD63-pHuji was a gift from D.M. Pegtel (27). Cells were imaged between 24 and 48 h posttransfection. For Brefeldin A (BfA) treatment, BfA was purchased from Goldbio (St. Louis, MO, cat. no. B-930-5). A549 cells were transiently transfected with CD63-pHluorin as described above, then incubated in medium containing either 0.1% DMSO or 0.1% DMSO with 5 μg/mL BfA for 60 min at 37°C, then imaged. For Magic Red treatment, the Magic Red Cathepsin B Assay kit was purchased from Immunochemistry Technologies (Davis, CA, cat. no. 937) and used according to the manufacturer’s protocol.

sEV collection and slot blot

A549 cells were seeded 24–48 h before exosome collection in T75 cell culture flasks (CELLTREAT, Life Science Products, Frederick, CO) and allowed to reach 70–80% confluency. After reaching 70–80% confluency, cells were washed 3× in serum-free DMEM and incubated at 37°C, 5% CO₂ for 24–48 h in serum-free DMEM. The medium was collected and sEVs were isolated using ExoQuick-TC (Systems Biosciences, Palo Alto, CA, cat. no. EXOTC50A) following the manufacturer’s protocol. Cells that were incubated for 24 h transiently expressed CD63-pHluorin and cells that were incubated for 48 h were not transfected. Once sEVs were isolated, a slot blot (Bio-Rad, Hercules, CA, cat. no. 1706542) was used and CD63 was probed with the primary antibody anti-CD63 (Santa Cruz Biotechnologies, Dallas, TX, sc-5275, 1:1000) and a fluorescent, secondary anti-mouse antibody (Santa Cruz Biotechnologies, sc-516178, 1:1000), and then imaged on a Bio-Rad ChemiDoc MP Imaging System imager.

TIRF microscopy

The transiently transfected cells were imaged using an inverted Nikon microscope equipped with 491 and 561 nm lasers on an acousto-optic laser launch (Solamere Technology, Salt Lake City, UT) and an EMCCD camera (Andor iXon897). The laser power entering the back of the microscope was set to 30 mW for each laser and then reduced using ND filters in the laser path before excitation. To direct excitation light to the sample, a dual color, a TIRF-specific, dichroic beam splitter was used (Chroma Technologies, Bellows Falls, VT) and emission was split into red and green fluorescent channels using a Dual-View (Optical Insights). For pHluorin detection, a green laser (491 nm) was used for excitation and a 525/50 filter (Chroma Technologies) was used to detect emission. For pHuji detection, a yellow laser (561 nm) was used for excitation and a 605/75 nm filter (605/75) was used for detection. For magnification, a 60× 1.49 NA objective and an additional 2.5× lens were used such that one pixel is 109 nm. To improve the ability to focus on both colors simultaneously a long focal length (1000 mm) plano-convex lens was placed in the red emission channel of the dual view. Micromanager image acquisition software (34) was used to obtain data with a 50–100 ms exposure time continuously for 500–1100 frames. Before recording cells for two-color imaging, the TIRF field was aligned with 200 nm diameter, carboxylate-modified, yellow-green FluoSpheres (Thermo Fisher Scientific). Yellow-green FluoSpheres show up in both channels and are used for registration of green and red images in conjunction with a custom MATLAB code (The MathWorks, Natick, MA). The depth of field is also indirectly measured by the FluoSpheres and the sample temperature was maintained using an on-stage heater system (Bioscience Tools, San Diego, CA, TC-1-100S).

Two-color TIRF microscopy was performed followed by automated fusion detection using the pHuji channel. To count whether the MVE was “docked” before fusion, green fluorescence must be present before fusion and during fusion. If green was present only during fusion, the vesicle was counted as “not docked” before fusion. If green was not observed at all during the fusion event, the MVE likely did not contain EGFP-CD63 and we could not tell if the MVE was docked or not before fusion. Presence of EGFP-CD63 at the fusion site was confirmed both manually using EGFP-CD63 images and via the small increase in EGFP-CD63 brightness.

Image analysis

To identify fusion locations, image sequences with the pH-sensitive MVE fusion marker were subject to the following analysis using scripts designed in MATLAB (Fig. S1). The analysis code is available at GitHub:https://github.com/michelleknowles/membrane-fusion.

Step 1: Calculation of differential movies

Images were subtracted from one another to highlight the cellular locations where intensity has changed in time. An earlier time frame (t) was subtracted from a later time frame (t + 25 frames, t + 1.25 s). The maximum projection of the difference movie was used for the subsequent spot finding step. Bandpass filtering of the movie helped with finding spots (Fig. S2) but was not essential and not applied for the fusion events processed here.

Step 2: Fusion event localization

The average intensity of the maximum projection is used as the threshold to locate spots and a cell mask is applied to find the average cell intensity. Spots above the calculated threshold were located using a MATLAB tracking algorithm originally designed for use in IDL (35).

Step 3: Cropping raw data at fusion locations

A 25 × 25 pixel region with the potential fusion spot centered was cropped. If two color channels were measured, both were cropped.

Step 4: Output the intensity in time

The average intensity from a circle 7 pixels in diameter was measured. The average cell intensity within the cell boundary was defined as the background for fusion events and subtracted from fusion events. Intensity traces were normalized by first subtracting the minimum before fusion then dividing by the maximum intensity.

Step 5: Alignment of data in time

Since the fusion events are not synchronous, all events were aligned by setting the frame before fusion to be 0 s. A background average is calculated using first 35 data points of the fusion event intensity data and the onset of fusion is defined as the point where intensity spikes to 1.4 times the background average.

Step 6: Sort events

Events that are identified in the above steps are either fusing, moving, or docking vesicles. Initially, events were sorted by viewing the cropped movies prepared in step 3 or by the intensity traces in step 5. Fusion is noted by a sudden spike in intensity, which is followed by gradual loss of fluorescence over time. With a subset of vesicles, an automated protocol was developed to separate fusion from other events. Here, docking events were removed based on the slope of the decay after the maximum intensity. Slopes of 5% or greater were considered fusion or motion. Tracking was then performed on a subset of the potential fusion events and the rate of motion was used to filter out fusing vesicles from moving vesicles. Moving vesicles are removed based on the mean square displacement, and a diffusion coefficient greater than 0.010 μm²/s denoted a motile vesicle. These nonfusion events are shown in Fig. 2. In the code, the rates can be adjusted or analyzed by eye. This step is done separately from 1 to 5 and 7.

Figure 2.

Potential fusion events can be categorized as (A) fusion, (B) motion, or (C) docking. (A–C) Show single events, scale bar = 2 μm. (D) Docking (red) is noted by the lack of fluorescence loss in the intensity trace. The slope of a line fit from the maximum intensity to 1 s later is substantially lower than moving or fusing vesicles (n = 15 events each). (E) The moving vesicles can be separated from fusing vesicles based on the diffusion coefficient from tracking analyses (n = 15 vesicles each). (F) Fusion events should also show fluorescence expanding in time as CD63-pHluorin exosomes and CD63-pHluorin on the MVE membrane leave the fusion site. (G) The diameter as a function of time for fusion events (error bars are mean ± SE, n = 53). (H) The diameter of the fluorescence signal at the onset of fusion reflects the diameter of the MVE. The black dashed line is the diffraction limit of the microscope, measured using a yellow-green FluoSphere (d = 200 nm).

Step 7a: Calculate kinetic parameters

From the aligned intensity traces in time several pieces of information are noted in the results. The slope of the decay, the fraction lost at 1 and 5 s, was calculated. To determine the rate of decay, a double exponential decay shown in Eq. 1 was used to fit single events from the maximum intensity to the end of the trace.

$$F\left(t\right)\mathit{=}\mathrm{A}{e}^{\mathit{-}{k}_{\mathit{1}}t}\mathit{+}B{e}^{\mathit{-}{k}_{\mathit{2}}t}\mathit{+}\mathrm{C}$$ (1)

where A and B are the relative amounts of the fast and slow components that have corresponding rates k₁ and k₂. C is a plateau because some traces do not return to the background level over the time intensity was measured. The rates from fitting, k, were converted in t_1/2 where t_1/2 = ln(2)/k. All fitting was performed in GraphPad Prism.

Step 7b: Calculate size changes in time

The radial plot of the fusion event was determined as a function of time; fusion events expand in time (36). To determine the size of the fusion events, the cropped 25 × 25 pixel image at the moment of fusion (t = 0) was averaged with the frame before and the frame after (t = −0.05 to 0.05 s). This was done to improve the signal/noise ratio. For time progression of the diameter, three images were averaged for each time point measured for a subset of events (n = 53). If events proceeded quickly, the diameter could only be measured over a few frames before content was lost and a diameter could not be measured. To measure the diameter, a radial plot was calculated from the image such that the center pixel is position 0 and the equidistant pixels are averaged, as described in our previous work (37). The fullwidth at half-maximum was determined by linear interpolation between the two points surrounding the half-maximum intensity. The expansion in time was fit with a line.

Calculation of time to leave the fusion site for membrane diffusion of CD63-pHluorin

To measure the time, t_1/2, it would take for half the particles moving at the rate measured in fluorescence recovery after photobleaching (FRAP) to leave a 0.76 μm diameter circle we calculated the probability distribution of step sizes taken in space and time (38),

$$y\left(r,t\right)=r\left[A\exp \left(-\frac{r^2}{4Dt}\right)\right]$$ (2)

and solved for t_1/2, the time where half of the molecules are within a circle of radius, r = 0.38 μm, and half have traveled further. D is the diffusion coefficient (0.039 μm²/s) of CD63 on the plasma membrane yielding t_1/2 = 1.33 s. This was verified by simulations of a random walk, described below.

Modeling MVE fusion decays

MVE fusion events were modeled as a point source where 100 CD63-pHluorin molecules are deposited randomly within a 0.5 μm diameter region, approximately the size of the MVE diameter. Postdeposition, molecules can escape a circle of 0.76 μm diameter (7 pixels in the image analysis) around the fusion site in several ways: 1) as an untethered exosome moving at 6.5 μm²/s. This rate of diffusion is calculated using the Stokes-Einstein diffusion equation: D = k_BT/(6πηr), where k_B is the Boltzmann constant, T is 310 K, η is the viscosity of the aqueous buffer (0.69 cP), r is the radius of an exosome, which ranges from 15 to 60 nm, 50 nm was used. The rate a free exosome diffuses from the site of fusion varies by less than 10% over the range of temperatures measured in the experimental data. 2) As a molecule diffusing from the endosomal membrane into the plasma membrane at 0.039 μm²/s. This rate of diffusion comes from FRAP measurements of CD63-pHluorin diffusing on the plasma membrane of live A549 cells. 3) As tethered exosomes. Tethered exosomes were treated in two ways: 3A) the rate of motion of the tethered exosome is immobile unless the tether is broken, then the exosome diffuses at a rate of 6.5 μm²/s. In the simulation, the time constant at which half of the tethers are broken (t_1/2tether) was varied from 1 to 50 s to best fit the data and exosomes were attached with only one tether. 3B) Tethers were allowed to move and the diffusion coefficient, D_tether, was varied from 0 to 0.018 μm²/s to determine a best fit at 37°C. In both approaches to tethering, the fraction of CD63-pHluorin in the membrane, in tethered exosomes, and in free exosomes was varied to best match the data. The amount in the endosomal membrane was constrained to 0.3 ± 0.1 based on past work where the endosomal membrane contains 30–34% of the CD63 within an MVE (39). The simulation used 50 μs steps and allowed the molecules to move via the modes of motion described above. The location of the molecule was recorded every 50 or 100 ms, depending on the experimental data that the simulation was matching to; 23°C data were taken at a 100 ms/frame to obtain longer data sets due to the observed plateau in the events and 27, 32, and 37°C data were taken at a 50 ms/frame. The simulation lasted between 12.5 and 25 s and the number of molecules remaining in a 0.76 μm circle, the size over which the intensity was measured for the experimental data analysis, was quantified.

Simulations were run and a best match to the average decays was obtained for all temperatures and individual decays for 37°C data (n = 20 fusion events). For the average data, the ratio of free exosomes to endosomal CD63 diffusing on the membrane was determined by fitting the first 10 frames of the experimental data. The tethered exosomes do not contribute to intensity loss at this stage. Next, the D_tether was varied and the percent of tethered exosomes was determined to fit the long-time tail of the experimental data. At this point, the endosomal CD63 and the free exosomes do not contribute significantly to the experimental data. For fitting 8 simulations were performed and averaged. For individual traces, the initial 10 frames were fit as described for the average data. The D_tether or t_1/2tether was kept constant after being determined from fitting the average, but the fraction tethered was varied in 2% increments from 0 to 100%. The remainder of the CD63 was split between endosomal and free exosomes, as determined from fitting the averages. The absolute value of the differences between the simulation and the data was measured for each time point and the lowest sum of these differences was considered the best match for both average and individual traces.

Results

Automated detection and quantification of MVE fusion events

MVE fusion events were visualized using TIRF microscopy and quantified using CD63-pHluorin, a pH-sensitive fluorescent probe (27,40), where the pHluorin moiety is located between two transmembrane domains of CD63 such that it is localized to the interior of the MVE and on the exterior of the ILV (Fig. 1 A). While enclosed in the acidic environment of an MVE, pHluorin is quenched. Once the vesicle fuses with the plasma membrane, the pH change unquenches pHluorin leading to an abrupt fluorescence spike, marking the onset of MVE fusion (Fig. 1, B–D). The fluorescence rapidly appears at the fusion site and dissipates into the surrounding area (Fig. 1 D) as exosomes diffuse into the extracellular space and as CD63-pHluorin on the MVE membrane diffuses into the plasma membrane. The kinetics of MVE fusion can be measured from the intensity profile of single events in time (Fig. 1 C).

Figure 1.

MVE fusion events are visualized using CD63-pHluorin. (A) Diagram depicting the assay used to detect MVE fusion events. CD63-pHluorin is quenched when in an acidic vesicle. Once the MVE fuses, the pH increases and pHluorin emits green fluorescence. (B) Membrane fusion is observable in live A549 cells using TIRFM. A maximum projection of a difference movie with potential fusion events circled in green. The purple arrow marks the event in the example in (C–E). (C) The intensity profile of a single fusion event, where 0 s is defined as the onset of fusion. The intensity shown is the average intensity within a 0.76 μm diameter circle and normalized to the maximum intensity. (D) Montage of a single fusion event. (E) Montage of a difference movie of a single fusion event. Green box is 0 s. Scale bar, 2 μm (cell AM1587).

Although sites of fusion were visible by eye, manually locating and detecting each event is time consuming and subjective. To circumvent these issues, event detection was automated and the procedure is summarized in Fig. S1. Detection began by enhancing the signal of the fusion events relative to the background via the generation of a difference movie, like others have done for constitutive fusion measurements (41). The purpose of this step was to remove signal rendered by vesicles that are visible and stationary on the cell surface throughout the movie; fusing vesicles show a rapid fluorescence intensity spike due to pH change upon fusion and are generally not visible before the onset of fusion (Fig. 1 D). Signal enhancement obtained from the difference movie is apparent in Fig. 1 E, a single fusion event that has been bandpass filtered. Once located, potential fusion sites were cropped from the raw movie file into 25 × 25 pixel (2.7 × 2.7 μm) regions centered around the fusion spot and the fluorescence intensity within a 7 pixel (0.76 μm) circle, centered on the fusion spot, was quantified for each fusion event and output as a function of time (Fig. 1 C). Fluorescence intensity traces in time were used to identify fusion events and then further analyzed for kinetic information.

Upon vesicle fusion, the fluorescence intensity profile of a pH-dependent probe exhibits certain features, such as a rapid increase (1–2 frames) in fluorescence intensity at a localized spot followed by a cloud-like spread of intensity into a region around the initial spike location (Fig. 1 D) (27,36,41,42). Our approach to locating the fusion events works well for avoiding stationary, fluorescent vesicles; however, fluorescent vesicles that move or dock are detected. Fig. 2 shows the most common examples of nonfusion events detected. Among the potential fusion events identified using the automated approach, approximately 70% were fusion events (Fig. 2 A). The remaining 30% of nonfusion events were a combination of moving vesicles (Fig. 2 B) and docking vesicles (Fig. 2 C). While all three events show an increase in fluorescence in the intensity trace, the moving and docking vesicles are typically slower to reach the maximum intensity and the subsequent kinetics are different. Fusion events decay exponentially (Fig. 2 A). Moving vesicles typically have to remain stationary for a few frames to be detected as a spot, then move away causing a slower onset of fluorescence followed by a plateau or slow decay (Fig. 2 B). Docking events do not decay, resulting in a fluorescence plateau (Fig. 2 C). The slope from the maximum intensity to 1 s later was fit and used to bin the events into fusing, moving, and docking (Fig. 2 D). While slope is useful for binning docked vesicles from the others, overlap was observed for fusing and moving despite a significant difference in the data sets (unpaired t-test, p = 0.0307). Therefore, tracking was performed to determine a diffusion coefficient (Fig. 2 E); moving vesicles can be identified by a diffusion coefficient greater than 0.010 μm²/s.

After fusion occurs, CD63-pHluorin gradually moves away from the fusion site and a radial spread of the fluorescence is typically observed (Figs. 1 D and 2, F and G). This presumably occurs as exosomes diffuse from the fusion site and as CD63-pHluorin from the MVE surface laterally diffuses into the plasma membrane as depicted in Fig. 2 F. Single fusion events radially expand in time (Fig. S4) and the average of many fusion events also shows widening in time (Fig. 2 G). The width of the intensity at the fusion site increased in time due to the loss of CD63-pHluorin from the fusion site and into the surrounding area. At the moment of fusion, the radius relates the size of the MVE relates to the mean diameter in A549 cells, which is 453 nm (Fig. 2 H). Once the fusion events were isolated from other detected events, the intensity traces were further analyzed to characterize MVE size and MVE fusion kinetics.

Fusion events are MVEs, not lysosomal or trafficking vesicles

To verify if the fusion events observed were due to MVE fusion, lysosomal and trafficking vesicle fusion were examined. Cells were transfected with CD63-pHluorin and treated with Magic Red, a lysosome marker that becomes fluorescent when Cathepsin B, a lysosomal enzyme, cleaves the probe. Cells expressing CD63-pHluorin (Fig. 3 A) and labeled with Magic Red (Fig. 3 B) look very different when imaged using TIRFM. Magic Red puncta are further from the cell surface and appear larger. When CD63-pHluorin fusion events occur (Fig. 3 C), Magic Red (Fig. 3 D) is not lost at the same time, confirming that CD63-pHluorin fusion events are not lysosomal. This is also observed in the average intensity traces (Fig. 3, E and F), where Magic Red intensity does not decrease during fusion events. This agrees with recent work that showed a lack of correlation between the presence of Magic Red, TRPML1 (a lysosomal Ca²⁺ channel), EEA1 (an early endosome marker), and LC3B (a macroautophagy marker) with CD63 fusion events, suggesting that CD63 fusion events are a unique subtype of fusogenic endosomes (43). Next, trafficking from the ER to the Golgi was blocked with BfA to determine if trafficking vesicles are delivering CD63-pHluorin to the cell surface (Fig. 3 E). Acidic trafficking vesicles can deliver newly synthesized CD63-pHluorin to the plasma membrane (44). CD63-pHluorin fusion events were measured and the kinetics were not altered with the presence of BfA (Fig. 3 G). As a positive control, the stimulated secretion of VAMP2-pHmScarlet was measured with and without BfA treatment (Fig. S5) in PC12 cells. The loss of VAMP2 secretion demonstrates that BfA blocks trafficking at this concentration (5 μg/mL) and duration (60 m). Finally, sEVs carrying CD63 were secreted from cells over time. Small EVs were collected from A549 cells and measured for the presence of CD63 (Fig. 3 H). CD63 was present when collection took place over 24 h, when cells were transiently transfected with CD63-pHluorin and more was present in nontransfected cells after 48 h. These data suggest that MVE fusion is the primary source of fusion events.

Figure 3.

CD63-pHluorin fusion events are not lysosomes or trafficking vesicles. (A) CD63-pHluorin was expressed in A549 cells and imaged with TIRFM; an average of five frames (0.85 s) is shown starting with the first frame where fusion was observed within the pink circle. (B) Magic Red labels lysosomes, (C) a single fusion event CD63-pHluorin, and (D) Magic Red. The region is marked by a pink circle in (A). (E) CD63-pHluorin intensity during fusion (average of n = 17 events, 7 cells). (F) Magic Red intensity in time during CD63 fusion events (mean ± SE). The onset of CD63-pHluorin fusion is at 0 s. (G) Brefeldin A (BfA) (5 μg/mL for 60 min) was used to block ER to Golgi membrane trafficking. DMSO-treated cells (n = 25 events, 8 cells) and BfA-treated cells (n = 30 events, 9 cells) both had observable fusion events and the half time to reach the plateau was not significantly different (t-test, p = 0.44). (H) Small EVs were collected from A549 cells after 24 and 48 h incubation and precipitated with ExoQuick-TC and blotted on a slot blot for the presence of CD63.

MVE fusion leads to exosome release and deposition of CD63 on the plasma membrane

By observing individual fusion events, the rate of the loss of intensity in time was analyzed to determine the mechanism by which CD63-pHluorin leaves the fusion site. Heterogeneity in the rate of fluorescence decay was noted with two types of decay profiles (Fig. 4, A and B), where the decay profiles of some secretion events underwent a biphasic decay with an initial fast downward slope for approximately 3 s, followed by a gradual intensity loss phase (Fig. 4 A), whereas others showed a gradual loss of intensity (Fig. 4 B). The average of 110 events is shown in Fig. 4 C and many single fusion events taken from two cells are shown in Fig. S3 A to demonstrate the large heterogeneity in fusion events. A majority of the fusion events exhibited two-component decay profiles (Fig. 4 D) at 37°C; however, a few events were not able to be fit well if the movie ended before a long-time decay could be observed.

Figure 4.

Multiple kinetic modes are observed in MVE fusion events. (A) A single fusion event (black dots) fit to a one-component (red), or a two-component (blue) fit. (B) Sample fusion event that is fit well with a one-component decay. (C) Average time course of 120 fusion events. (D) Pie chart showing percent of one- vs. two- component decay profiles. (E) Average fast (blue) and slow (light blue) decay constants from a biexponential fit and a single exponential fit (red), where t_1/2 = ln(2)/k, was calculated using the rate constants from fits of individual fusion events. The mobility of CD63-pHluorin on the plasma membrane was measured by FRAP and an expected t_1/2 was calculated based on the diffusion coefficient and circle size (white) (mean ± SE). (F) Percent of the fast component in decay curves for fusion events at 37°C (mean ± SE). The gray band represents the portion of CD63 present on the endosomal membrane (30–34%) based on EM data (39). (G) Plateau for decay curves (median ± 95% CI). All data were taken at 37°C.

To determine the mechanism behind the biphasic decays, we considered the likely locations of the fluorescence probe. CD63-pHluorin, is present on the ILVs and the MVE surface. Recent EM data show that, on average, 30–34% of CD63 is retained on the endosomal limiting membrane, whereas the remainder is located on ILVs (39). When fusion occurs, it is expected that part of the decay is due to exosomes diffusing from the fusion site and the part of the decay is due to the CD63-pHluorin diffusing from the MVE membrane into the plasma membrane. Analysis of the rate of decay was performed to identify the amount of fluorescence loss due to each of these components. The traces that fit well to a two-component exponential function (n = 66 events) displayed fast and slow components with t_1/2 of 0.40 ± 0.04 s and 11.5 ± 2.4 s, respectively (Fig. 4 E, blue bars). The t_1/2 of the single-component fits (Fig. 4 E, red) fell between that of the two-component fit, suggesting that these are similar mechanisms. To determine which component of the kinetics was due to CD63-pHluorin diffusion in the plasma membrane, FRAP experiments were performed with CD63-pHluorin on the surface of A549 cells (Fig. S6). Recovery traces (Fig. S6 B) show that CD63-pHluorin is mobile on the plasma membrane and fitting of the data revealed that CD63-pHluorin diffused at a rate of 0.039 μm²/s (Fig. S6 D). Interestingly, there was no temperature dependence observed for the motion of CD63 on the plasma membrane (Fig. S6). To determine how long it takes for molecules to leave a region the size (d = 0.76 μm) over which intensity was measured for a fusion event, data were simulated for a 2D diffusion of particles moving at 0.039 μm²/s and deposited at the center of the circle. On average, half of the particles are lost from the fusion site within 1.3 ± 0.1 s (Fig. 4 E, white). The fraction of the intensity that moves at the faster rate is 40%, slightly higher than the expected amount to be present on the endosomal membrane (30–34% (39)) denoted by the gray bar. The plateau, although small at 37°C, was above 0 (Fig. 4 G) as fluorescence often remained postfusion.

MVE fusion is temperature dependent

MVE fusion kinetics were analyzed at four different temperatures: 23, 27, 32, and 37°C. The kinetics of MVE fusion are directly related to temperature; a decrease in temperature caused a decrease in the rate at which the CD63-pHluorin fluorescence signal was lost (Fig. 5, A–D). Although the lower temperatures reduced the rate of content loss, two phases were required when fitting the average decay. In Fig. 5 B, the decays are fit well with a biexponential function, but the single exponential fit in Fig. 5 C clearly misses the fast component for all temperatures. This suggests that the release event contains at least two mechanisms of leaving, likely the loss of CD63 on the endosomal membrane and loss of exosomes. Since the fast portion of the decay lasted about <1 s for 37°C (Fig. 4 E), we compared the amount of fluorescence remaining after 1 s for all of the different temperatures to isolate the fast component. The decays after 1 s trend as a function of temperature (Fig. 5 D); however, the mobility of CD63 on the membrane shows no temperature dependence in FRAP measurements (Fig. S6), suggesting that exosome secretion portion of the decay is temperature dependent, not the diffusion of CD63 through the membrane. Reduced temperatures also led to the incomplete loss of fluorescence from the fusion site. The average of the fusion events at 23°C retained approximately 50% of the maximum intensity at 10 s postfusion, whereas the average of the fusion events observed at 37°C only retained 50% of maximum intensity for approximately 1–2 s postfusion (Fig. 5 A). The amount of the decay that was in the fast component of the fit was not temperature dependent (Fig. 5 E); however, the remaining fluorescence in the plateau was more prominent at lower temperatures (Fig. 5 F). The presence of a plateau suggests that content is not fully released from the fusion site or remains attached to the cell surface. The rate of fusion events per minute (Fig. 5 G) also trended higher at higher temperatures. Overall, higher temperatures make fusion more frequent, faster to release content, and less fluorescence is retained at long times.

Figure 5.

Kinetics of MVE fusion events depend on temperature. (A) Average intensity traces in time of single MVE fusion events at 23°C (n = 86), 27°C (n = 83), 32°C (n = 77), and 37°C (n = 110). For fitting, the time alignment was done with respect to the maximum intensity (0 s) and individual traces were normalized by the maximum. Black lines are fits with a biexponential function. (B) Zoom of the biphasic fit and the (C) single exponential fit at short times. Error bars are mean ± SE. (D) The percent loss in intensity 1 s after the maximum. All are significantly different in t-tests (p < 0.05) from the nearest temperature (mean ± SE). (E) The portion of the decay that is fast for each temperature is not significantly different. The light gray bar indicates the amount of CD63 expected to be present on the MVE limiting membrane (mean ± SE). (F) The plateau from the biexponential fit relates the long time, remaining intensity (median ± 95%CI, only 23°C is significantly different from others in a t-test, p < 0.05). (G) Fusion events observed per minute of data acquisition at different temperatures (mean ± SE).

Exosome release, tethering, and CD63 membrane diffusion can account for the experimentally observed kinetics

Fitting data to an exponential function has limitations due to the signal/noise ratio and total duration of recordings available when measuring single fusion events. Therefore, we applied a simple, physical model to determine what types of CD63 motion were sufficient for explaining the release kinetics at different temperatures. In the experimental data, the loss of signal at the fusion site is due to CD63-pHluorin leaving. CD63-pHluorin can leave in several ways: 1) CD63-pHluorin is sorted onto ILVs that become exosomes and these exosomes can diffuse freely from the fusion site. 2) CD63-pHluorin exosomes can remain at the fusion site due to tethering (30). 3) CD63-pHluorin is on the endosomal membrane and can diffuse into the plasma membrane to leave the fusion site. To simulate the data, molecules were released from the fusion site, which was defined as a 0.76 μm diameter circle (Fig. 6 A). CD63-pHluorin molecules were allowed to start from anywhere within the MVE (∼500 nm diameter). The molecules move with one of the three motions described above. Free exosomes (Fig. 6, A and B, green) diffuse away almost instantly at a rate of 6.5 μm²/s, calculated for a 100 nm exosome moving in an aqueous solution. CD63-pHluorin located on the endosomal membrane (Fig. 6, A and B, blue) moves at a rate of 0.039 μm²/s as measured by FRAP of CD63-pHluorin on the plasma membrane (Fig. S6). These two components are not sufficient to recreate the long-time experimental data; both decay too quickly when compared with data. Attachment of exosomes to the cell surface has been observed in EM data (28,29), and tethering proteins have been identified (30). Therefore, a tethered exosome component was added to the simulation. Initially tethers were added and immobile; however, the slow decay of the long-time data could not be replicated (Fig. S7, A and B). We considered that the slow intensity loss could be due to photobleaching of CD63-pHluorin. Photobleaching was measured (Fig. 6 B, gray open squares) and the rate of photobleaching is extremely slow under the imaging conditions, much slower than the loss of fluorescence from the fusion site at long time (Fig. S7, C and D). Therefore, the tethered exosome component needed to vary in time to account for the long-time, slow decay observed during fusion events. Slowly moving tethered exosomes (Fig. 6, A and B, yellow) were included to account for the long-time decay observed. In this “diffusing, tethered exosome model,” tethers did not break, but diffused slowly and the diffusion coefficient was determined by the best fit to the experimental data (0.004 μm²/s at 37°C). The rate of motion got slower at colder temperatures (Table 1).

Figure 6.

Fusion events were simulated by allowing CD63 to leave the center of the fusion site as tethered exosomes, free exosomes, or membrane diffusion. (A) Depiction of the simulation in time. 100 particles are deposited at the center of the 0.76 μm circle and allowed to diffuse. A portion of the particles are freely diffusing as 100 nm diameter exosomes (green), diffusing on the plasma membrane (blue), or tethered (yellow). The larger green, blue, and yellow circles depict the motion of the bulk of particles and only a few individual particles are shown. (B) The number of particles remaining within the circle (d = 0.76 μm) is shown for each component. The loss of particles in time under the scenario where all particles move as untethered exosomes at a rate of 6.5 μm²/s (green), all particles move as CD63 does in the plasma membrane at a rate of 0.039 μm²/s (blue), or all are tethered and diffuse at 0.004 μm²/s (yellow). The data are shown as gray circles and the photobleaching rate as gray squares. (C) The average intensity loss observed in the experimental data for each temperature is shown with the best fit from simulations (n = 8). (D) The fraction of each component was varied to fit individual traces at 37°C.

Table 1.

Parameters for simulated fusion events best fits

	D (μm2/s)	23°C	27°C	32°C	37°C
No. of events	–	86	83	77	98
Dtether (μm2/s)	varies	0.0015	0.003	0.004	0.004
Fraction attached exosomes	varies	0.5	0.5	0.45	0.4
Fraction free exosomes	6.5	0.18	0.18	0.18	0.24
Fraction endosomal	0.039	0.32	0.32	0.37	0.36

Fusion events were simulated to best match the average intensity loss trace at each temperature. D_tether, the fraction attached, and the fraction free were unconstrained. The fraction on the endosomal membrane was constrained to 0.3 ± 0.1. See materials and methods for details.

At all temperatures, the loss of fluorescence from the fusion site could be modeled with three components. When comparing the simulation results with the average data at all temperatures, the variables in the simulation were: D_tether, the fraction of free exosomes, and the fraction of CD63 on the endosomal membrane, which was constrained to be 0.3 ± 0.1. The fraction of attached exosomes was set to be the remaining amount of CD63 that is not on the endosomal membrane or free exosomes. The best fit for the average data at each temperature is summarized in Table 1 and shown in Fig. 6 C and residuals are shown in Fig. S9. The amount of CD63 on the endosomal membrane was relatively constant for all temperatures and aligned well with past work that showed an endosomal fraction of 0.3–0.34 (39).

The individual fusion event traces at 37°C were also simulated to characterize the heterogeneity in MVE fusion that would be missed by looking at the average alone. All traces were modeled, both the biphasic and single-phase types. The D_tether was fixed at 0.004 μm²/s, as determined from the average; however, the fraction of CD63 on the endosomal membrane was varied as there is likely a distribution of CD63 localization from MVE to MVE. The fraction of CD63 on the endosomal membrane ranged from 0 to 0.6 (Fig. 6 D, blue) and the fractions of free and tethered exosomes (Fig. 6 D, green and yellow, respectively) also varied widely to accommodate the distribution of kinetics observed from single traces. Overall, the kinetics of individual MVE fusion events are highly variable but can be modeled with three types of motion for CD63-pHluorin loss from the fusion site.

MVEs are docked before fusion

The steps before membrane fusion were probed to determine whether fusion happens with newly arrived MVEs (Fig. 7 A, top) or MVEs that are docked before fusion (Fig. 7 A, bottom). Cells were transfected with both CD63-pHuji and EGFP-CD63. While CD63-pHuji is only visible upon the onset of fusion and serves as a fusion marker, EGFP-CD63 can be observed before fusion (22,27). By measuring MVEs that contain both CD63-pHuji and EGFP-CD63, MVEs would be visible before and during fusion (Fig. 7 B). In the CD63-pHuji channel (Fig. 7 C), fluorescence is not visible before fusion, but these events were used to locate fusion sites due to the unquenching of pHuji with a change of pH, similar to pHluorin. In the GFP-CD63 channel, green fluorescence is observed before fusion (Fig. 7 D). Visual assessment of vesicle docking revealed that 87 of 94 fusion events showed a visible vesicle docked at the site of fusion from the start of the movie, which is at least 1 s, but generally longer. Therefore, MVEs are typically docked to the membrane before fusion in A549 cells.

Figure 7.

MVEs dock before fusion. (A) Diagram illustration of crash fusion (top), and docking and fusion (below). (B) Depiction of assay used to detect MVE docking before fusion. In crash fusion, the eGFP-CD63 vesicle marker is not present before fusion; however, if vesicles dock, eGFP-CD63 is present before fusion. (C) Single images and quantified intensity plots for a fusion event in the CD63-pHuji channel. (D) Single images and quantified intensity plots of the same region during fusion of the eGFP-CD63 channel. Scale bar, 2 μm.

Discussion

In this work we describe an automated analysis for locating MVE fusion events from TIRF microscopy time series data and use it to characterize single MVE fusion events in A549 cells. To temporally characterize the fusion events, images were taken at a high frame rate (10–20 Hz) as events spontaneously occurred. Fusion events were located by a sharp, transient, increase in fluorescence that occurred upon a somewhat bright background since the CD63-pHluorin probe is not entirely contained within MVEs, but also appears on the cell membrane. To reduce this background, difference movies were calculated before plotting a maximum projection to locate fusion events (Figs. 1, S1, and S2). This highlighted fusion events as well as other transient events where nonacidified, fluorescent vesicles moved near the cell surface or docked (Fig. 2). The intensity trace of the events in time, measured from the raw data, allowed for differentiation between the types of events detected. Fusion was marked by a rapid increase in fluorescence and an exponential decay; however, diffusion and docking events were slower to increase, decayed differently, or had faster diffusion coefficients when tracked (Fig. 2). The fusion events observed were primarily MVE fusion, as opposed to lysosomes or trafficking vesicles (Fig. 3). Although this work focused on MVE fusion, the docked and moving events could be separately assessed in the future. This approach is capable of high-throughput, automated detection and characterization of membrane fusion events and adds to currently available protocols for fusion analysis (42,45,46).

The fusing MVEs ranged in diameter from diffraction limited to 800 nm in A549 cells. Others have reported that MVE size depends on the cells and organism. In HeLa cells, EM data show that MVEs range from 400 to 600 nm in diameter (27). Using super-resolution fluorescence methods, MVEs were measured to be 1 μm in diameter in MDA-MD-231 cells (16) and MVEs are 400 nm on average in C. elegans epithelial cells, based on EM analyses (47). In this work, MVEs secreted from A549 cells were on average 453 nm (Fig. 2). It should be noted, however, that the size is calculated at the first observed moment of fusion when the MVE membrane has fused with the plasma membrane. One limitation of this approach is that expansion and loss of fluorescent content could occur during the 50 ms exposure time; this would lead to overestimation of MVE size but our measurements largely agree with EM measurements (27), suggesting that not much fluorescence expansion occurs during the exposure time.

Before fusion, most MVEs dock at the plasma membrane (Fig. 7). In this work, only vesicles observed to undergo fusion based on the brightening of a red fluorescent pH-sensitive probe (CD63-pHuji) were measured. If docked, EGFP-CD63 within these vesicles is visible before fusion. Out of 94 observed fusion events, 87 were visibly docked previously and this largely agrees with past work where CHO cells and murine embryonic primary fibroblasts were treated in ways to increase the intracellular Ca²⁺ levels to stimulate fusion (22).

After docking, MVEs can fuse with the plasma membrane to release content and the release of content depends on temperature. Although A549 cells undergo MVE fusion at any temperature, higher temperatures increase the frequency of fusion events (Fig. 5 G) and the rate at which CD63 is lost from the fusion site (Fig. 5, A–D). In one respect, this is similar to stimulated fusion of synaptic vesicles, where temperature increases the number of stimulated fusion events in a variety of cell types (48,49). However, temperature has little effect on the rate at which soluble content is released and primarily affects fusion pore dilation, which occurs on a millisecond timescale in synaptic vesicle fusion (48). In MVE fusion, the mobility of exosomes attached to the cell membrane accounts for part of the differences observed at different temperatures (Table 1; Fig. 5 F); the initial loss of fluorescence, although significantly different between all temperatures, is small (Fig. 5 D) and likely due to free exosomes leaving the fusion site. The motion of CD63-pHluorin on the membrane was not dependent on temperature (Fig. S6). When modeling the data, the fluorescent decays at different temperatures could be recapitulated by changing the diffusion rate of attached exosomes (Table 1; Fig. 6 C).

Two approaches were taken to develop a model that can provide a mechanistic understanding of the postfusion kinetics; experimental data were both fit and simulated using a diffusion-based model. During MVE fusion, CD63-pHluorin is likely released in at least two different ways: as exosomes and by diffusion from the endosomal limiting membrane into the plasma membrane (Fig. 8). Based on the fitting and the simulations of the experimental data, we conclude that the fast (<2 s) component of the decay is a combination of free exosomes leaving and CD63 diffusing from the endosomal to the plasma membrane. The rationale is as follows:

• 1)Most MVE fusion events in A549 cells proceed via a biphasic decay. At 37°C, two-thirds of the traces show biphasic decays (Fig. 4 D) with the faster rate on the same order of magnitude as the rate of CD63-pHluorin diffusion in the plasma membrane (Fig. 4 E). It is worth noting that we assume that the rate of diffusion of CD63 into the plasma membrane from the endosomal membrane is similar to the rate of diffusion of CD63 on the plasma membrane, as measured by FRAP (Fig. S6). However, the rate of CD63 loss due to the fast component of fusion is approximately three times faster than expected based on CD63 plasma membrane diffusion (Fig. 4 E, white and dark blue). Here, the time it takes to leave the circle where fusion occurs, t_1/2, is 0.40 s for the fast component and the expected time of CD63 membrane diffusion, if CD63 moved at a rate of 0.039 μm²/s, is 1.33 s. Other transmembrane proteins have also been noted to diffuse faster during membrane fusion. In a study of constitutive fusion using TIRF microscopy, the transmembrane protein VSVG diffused from fusion sites at a rate of 0.11–0.13 μm²/s (41,50), whereas VSVG on the plasma membrane diffused at a rate of 0.038 μm²/s (51). The diffusion of VSVG from a fusion site is three times faster than diffusion of the protein already present on the plasma membrane. Therefore, it is possible that the rate of motion of CD63-pHluorin motion on the plasma membrane is slower than the rate newly delivered protein moves and this could account for the difference in the calculated rate at which protein leaves the fusion site (Fig. 4 E). If newly delivered protein moves faster, this would reduce the fraction of “free” exosomes in the model.
• 2)The fraction of the intensity that is lost at the faster rate (Fig. 4 F) also supports that this portion of the kinetics is primarily from CD63-pHluorin diffusing from the endosomal membrane into the plasma membrane. The fraction of the fast component is 0.42 ± 0.02 and the fraction of CD63 reported to be on the endosomal limiting membrane by EM data is 0.30–0.34 (39). The fraction fast and likely endosomal diffusion was also constant (0.40–0.46) and not significantly different over all temperatures measured (Fig. 5 E). In diffusion-based modeling of CD63 loss from the fusion site to simulate the decay curves, as opposed to fitting, approximately 35% of the CD63 needed to be on the endosomal membrane to recreate the experimental data (Table 1).
• 3)Fusion site diffusion simulations could not recreate the fast loss with only membrane diffusing CD63-pHluorin (Fig. 6 C). Therefore, a small portion of CD63 needed to reside on free exosomes (18–24%, Table 1) to model the experimental data at all temperatures. The contribution of free exosomes to the fast portion also supports why the rate of release measured from fitting is faster than expected from membrane diffusion (Fig. 4 E, dark blue and white). However, fitting data with more than two exponentials was not an approach to pursue; the isolation of these two kinetic components would be challenging with the noise in our experiments. Combined, this model and the data support that CD63 leaves fusion sites by diffusing into the plasma membrane and free exosomes contribute to the initial loss.

Figure 8.

Model of MVE fusion and how CD63 leaves the fusion site. (A) MVEs first dock then fuse. Postfusion, three outcomes can occur simultaneously in single fusion events: (B) exosomes quickly leave the fusion site (24%, <0.5 s), (C) CD63-pHluorin on the endosomal membrane diffuses into the plasma membrane (36%, 1–2 s). or (D) tethered exosomes remain attached to the membrane and slowly leave the fusion site (40%, 5–10 s). The percent of each component and approximate time to leave the fusion site were determined from a simulation of the average fusion decay measured at 37°C.

To better understand the mechanism behind the slow (>10 s) component of the fluorescence loss postfusion, several hypotheses were tested. First, the slow rate is almost 30 times slower than the fast rate at 37°C (Fig. 4 E), but it is not as slow as the photobleaching rate (Fig. S7, A and B), suggesting that CD63-pHluorin molecules are leaving the fusion site in a delayed fashion. Second, in past work on exosome secretion, others have also noted long-lived fluorescence in single fusion events and suggested that this was due to tethering of exosomes (27), although cells lacking one of the known tethers, tetherin (30) did not remove the long-lived fluorescence (27). Many molecules could potentially tether exosomes to the surface, such as tetraspanins, cell or matrix adhesion proteins, integrins, fibronectins, and other molecules (52,53,54); attachment of exosomes to the cell surface has been noted in EM data for many years (28,29). Exosomes themselves can act as attachment sites for migratory cells (25), demonstrating that they are capable of sticking to the cell surface. To better understand the delayed loss of fluorescence, exosome secretion was simulated as having attachments between the externalized exosome and the cell surface without knowing the identity of the attachment. To model the experimental data measured, membrane-attached exosomes needed to move slowly away from the fusion site (Fig. 6). They could not remain attached eternally (Fig. S7, C and D). By varying the diffusion rate as a function of temperature (0.0015–0.004 μm²/s), the model fit the experimental data well (Figs. 6 C and S8). Alternatively, tethers could dissociate over the course of many seconds (8–40 s, Fig. S8), where increased temperature decreased the time for the tether to break (Table S1). A model that has exosomes tethered but break in time, fits the slow fluorescence decay in the experimental data equally well (Fig. S8 C). Future experimental work in this field would help elucidate the molecules responsible for exosome attachment. Without yet knowing the details of what molecules are at work (i.e., tetherin, adhesion proteins, etc.), these approaches were able to model the long-time data well at all temperatures (Figs. 6 C and S8 C).

In neuroendocrine exocytosis, slow fusion has been attributed to: 1) the opening and closing of a fusion pore, commonly known as kiss and run (23), which significantly hinders content release. In dense core vesicle (DCV) fusion, small molecules are released first and larger peptide secretion occurs only after full fusion (55). It is currently not known if MVEs could undergo this type of fusion; however, larger vesicles, such as MVEs, typically create stable fusion pores that dilate (56). It is also unlikely that large molecules, such as CD63-pHluorin, or exosomes would escape without the formation of a large pore. However, it is worth noting that the events noted as “docking” in Fig. 2 could also be a form of transient fusion. 2) The incomplete flattening and quick endocytosis (“cavi-capture”) of the MVE postfusion could also give rise to slow secretion. In DCV fusion, incomplete flattening has been observed followed by endocytosis and reacidification on the order of 8 s in PC12 cells (57), but much longer in chromaffin and MIN6 cells (58). These occur at much slower timescales than those examined here and are unlikely to be responsible for the slow loss of fluorescence. However, more experiments are needed to characterize the shape of the MVE membrane postfusion and the rate of endocytosis and reacidification. 3) The content of DCVs has been noted to alter the release kinetics (59,60). Luminal proteins can affect fusion pore dilation and cause a plateau in fluorescence, as we observe in MVE fusion. 4) Finally, exosomes could temporarily stick to the glass surface used for imaging; however, different types of exosomes, for example, CD9+ instead of CD63+, display fast fusion kinetics when the two are compared side by side (27). Therefore, the data, model, and current literature support a slow, tethered exosome component (Fig. 8).

In summary, the kinetics of MVE fusion can relate the mechanism by which fluorescently labeled exosomes leave the fusion site and the release depends on temperature. In this work, the automation of the analysis and diffusion-based models have been developed to aid in the measurement of fusion kinetics. We conclude by proposing a model where exosomes are tethered in A549 cells; the tethers can move slowly in time and do so more readily under higher temperatures. Future studies of the role of kiss and run fusion, membrane shape changes during fusion, and exosome tethering in MVE membrane fusion will further expand our understanding of how exosomes are fully released from cells.

Author contributions

A.M. collected and analyzed data and contributed to writing. Z.O. developed and used the simulation model and collected data. A.W.W. collected and analyzed data. M.D.P. developed image analyses. M.T.N. and B.L.B. contributed to calculations and data collection. M.K.K. contributed to data analysis, writing, and model development.

Editor: Sarah Veatch.