Rapid capture of vesicles at the mitochondria suggests INVs are highly mobile

Previously, we showed that it is possible to capture INVs at the surface of the mitochondria using an inducible heterodimerization system (Fig 1A). INVs have a small marker protein, tumor protein D54 (TPD54/TPD52L2), which binds tightly to the vesicle outer membrane via C-terminal amphipathic helices (Larocque et al, 2021; Reynaud et al, 2022). Relocalization of mCherry-FKBP-TPD54 to MitoTrap at the mitochondria causes the capture of INVs from the cytoplasm to the mitochondria surface (Larocque et al, 2020, 2021). Upon induction, the relocalization of fluorescence to the mitochondria is rapid (τ = 38.6 s) (Fig 1B and C) (Video 1 and Video 2). To compare the relocation of vesicles with the relocalization of a protein that has similar properties, we used a TPD54 mutant (R159E) that cannot bind INVs (Larocque et al, 2021). This construct, mCherry-FKBP-TPD54(R159E), was relocalized an order of magnitude more quickly (τ = 3.8 s) consistent with a freely diffusing protein (dimeric mass 128 kD). Induced relocalization is not an active process. That is, mitochondria do not “attract” FKBP-tagged proteins from far away; instead, relocalization occurs through chance collisions between the vesicles and the mitochondria Therefore, these experiments suggest that INVs have a high mobility inside cells, but that their mobility is slower than a freely diffusing protein. The mobility is such that all INVs in the cell can be captured at the mitochondria in just over a minute. Besides vesicle mobility, the other factor in relocalization efficiency is the abundance and spatial distribution of the MitoTrap. To quantify the median distance that an INV may need to traverse to collide with a mitochondrial surface, we generated a 3D distance map from MitoTrap-positive structures in a cell. This analysis revealed a distance of 1.1 ± 0.2 μm (mean ± sd, Fig 1D and E). This means that when we add rapamycin, a vesicle is on average ∼1.1 μm away from a mitochondrial surface and is captured with a T_1/2 = 26.6 s. Using these measurements, we can estimate an apparent diffusion coefficient of ∼0.3 μm² s⁻¹ for INVs. This rough calculation suggests that INVs collectively have high motility, regardless of their mode of transport.

Direct imaging of INVs reveals heterogeneous, but mainly diffusive, movement

To begin to understand the mobility of INVs, we used fast capture imaging of GFP-TPD54 knock-in cells (see the Materials and Methods section). By eye, three states of motion were evident. First, the majority of vesicles underwent a random, Brownian motion (Fig 2A and B) (Video 3). Second, there were a small number of immobile vesicles, likely attached to a larger subcellular structure. Third, another minority of vesicles underwent directed movement. Vesicles could switch between these states over time and were otherwise indistinguishable in terms of size or brightness. Simultaneous imaging of microtubules and INVs revealed that the directed motions likely correspond to motor-based transport along microtubules (Fig 2C and Video 4). Using single particle tracking, we could identify directed motions and found that they constituted 5–11% of tracks (Fig 2D and E). These directed motions could be reduced by depolymerizing microtubules using nocodazole (20 μM) but not by actin depolymerization using latrunculin B (1 μM), confirming that they represent microtubule-dependent transport (Fig 2D and E and S1A–C and Video 5). However, the size of this microtubule-dependent fraction of INVs motions is very small, and we questioned whether it could account for transport of INV cargo.

Diffusion of INVs dominates recovery after photobleaching

To further investigate INV mobility, we used FRAP. In GFP-TPD54 knock-in cells, almost all of the TPD54 is associated with INVs and the association is tight (Larocque et al, 2020). Therefore, the bleach of GFP-TPD54 and subsequent recovery of signal in the bleached region is an indirect readout of INV mobility (Fig 3A). The recovery of GFP-TPD54 was with τ = 12.7 s, whereas recovery of GFP or GFP-TPD54 R159E mutant that is not associated with INVs was τ = 0.5 s or 1.0 s, respectively (Figs 3B and S3). The order of magnitude difference in recovery mirrors the observations made by relocalization experiments (Fig 1), and, together with the observation of high mobility (Fig 2) suggests that recovery reflects INV movement rather than exchange of protein on the surface of stationary INVs.

To estimate the contribution of diffusion to INV mobility, we used osmotic compression of cells (Guo et al, 2017). Addition of PEG1450 at up to 25% is a reversible manipulation to decrease cellular volume by water exchange and thereby limit diffusion because of a crowding effect (Fig S2A–D). Compression of cells using 6% or 12% PEG1450 resulted in a slowing of GFP-TPD54 FRAP to τ = 23.8 s or 24.1 s and an increase in the immobile fraction from ∼20% to ∼45–60% (Fig 3C and D). GFP-TPD54 R159E diffusion was also slowed to τ = 2.0 s or 4.9 s (Figs 3C and D and S3). This manipulation was reversible (Fig S2D) and supports the idea that mobility is mainly via diffusion.

To assess the contribution of active transport to INV mobility, we performed two manipulations. First, nocodazole (20 μM for > 1 h) was used to depolymerize the microtubule network. FRAP of GFP-TPD54 in these cells was indistinguishable from control conditions (τ = 9.4 s control, 9.0 s nocodazole; Figs 3E and F and S3). Second, ATP depletion (see the Materials and Methods section) was used to inhibit motor activity (Video 5). Again, FRAP of GFP-TPD54 was indistinguishable from control τ = 9.2 s control, 9.0 s ATP depletion; Figs 3G and H and S3) The lack of effect of either manipulation suggests that diffusion of INVs dominates the recovery of GFP-TPD54 signal after photobleaching and points to diffusion rather than active transport as being the main mode of INV mobility.

Diffusion coefficient of INVs

Given the mobility of INVs is mainly diffusive, can we determine the diffusion coefficient for this vesicle class? The FRAP experiments allow us to infer that GFP-TPD54 associated with INVs has a diffusion coefficient of 0.220 ± 0.096 μm² s⁻¹ (Fig 4A). By contrast, the R159E mutant of TPD54 which does not bind INVs has an inferred diffusion coefficient of 2.20 ± 0.62 μm² s⁻¹. These values correspond reasonably well to our initial estimate based on the induced relocalization experiments (diffusion coefficients are summarized in Fig 4B). To verify these measurements, we used two further approaches.

First, fluorescence loss in photobleaching (FLIP). Whereas FRAP assesses the diffusion coefficient of local INVs, FLIP experiments allow us to evaluate the mobility of the whole cellular vesicle content. Here, GFP-TPD54 knock-in HeLa cells grown on micropatterns to standardize the procedure were subjected to repeated photobleaching and the loss of cellular fluorescence was measured (Fig 4C and Video 6). Using a small target region of the cell (3.2% of the total area), the total cellular signal was bleached in ∼15 min using pulses every 3 s (τ = 430 s, Fig 4C and D). This result underscores the high mobility of INVs, because it suggests that every INV in the cell had passed through this small bleaching area in ∼15 min. To derive a diffusion coefficient from these data, we used a computer simulation and asked what coefficient could describe these observations. Using pulses of 60 ms, with the same interval of repeated photobleaching (see the Materials and Methods section, Fig 4E), the simulations suggested D ≈ 0.32 μm² s⁻¹.

Second, we returned to single particle tracking to directly analyze the movement of individual vesicles. This is a significant technical challenge. Acquisition must be fast enough to capture the motions accurately in GFP-TPD54 knock-in cells, and the signal-to-noise was very low at high frame rates. A more significant issue was the sheer density of INVs (estimated 20,000 per cell) which reduces the accuracy of tracking, especially if they move at high speed and in a nondirected manner (Fig S4A). To track these vesicles accurately, we bleached StayGold-TPD54 in a subregion of the cell and then tracked INVs entering this blank region (Fig S4B–E and Video 7). Again, the tracked vesicles mainly moved by random diffusion (Fig 4F). The diffusion coefficient derived from the mean squared displacement of all tracks was 0.104 μm² s⁻¹ (Fig 4G). Single-track analysis shows that the population follows normal diffusion where the median MSD exponent α = 0.97 (Fig 4H). The diffusion coefficients measured or inferred in all experiments are summarized in Fig 4B and suggest that INV motion is mainly diffusive with a D of 0.1–0.3 μm² s⁻¹.

Small trafficking vesicles move by diffusion

To characterize the movement of INVs and compare this with the mobility of other vesicle types, we again used a fast capture imaging and single-particle tracking approach. Tracking was done with cells expressing markers of different small vesicles (GFP-SCAMP3, ATG9A-GFP, GFP-Rab30, StayGold-TPD54, mCherry-Rab11, GFP-2xML1N, GFP-SCAMP1, GFP-Clathin light chain a, GFP-Rab35), markers of endosomes and lysosomes (GFP-Rab5, LAMP1-mCherry-FRB) or a microtubule plus-end protein as a control for directed movement (mNeonGreen-EB3) (Figs 5 and 6 and Video 8). The density of small vesicles was much higher (1.9–3.2 vesicles μm⁻²) than that of endosomes, lysosomes, and MT plus-ends (Figs 5A and 6A). INVs marked by TPD54 were the most numerous of all those we imaged (5.4 vesicles μm⁻² after photobleaching to reduce the density). The motion of all vesicle types was heterogeneous, with three states of movement and switching between these states, as described above for INVs (Fig 2). However, the proportions of tracks in each of these three states differed by eye. For example, LAMP1 had a large fraction of tracks moving in a directed manner, whereas clathrin-coated vesicles had a smaller fraction, and a considerable fraction of CCVs moved by passive diffusion (Fig 5B). We therefore analyzed single-particle tracking data using a custom-written R package, TrackMateR (Royle, 2022). Using the MSD exponent (α) to characterize each track allowed us to visualize the motions for each movie (Fig 5B) and for the entire dataset (Fig 5C), and confirmed that LAMP1 was mainly directed, whereas smaller vesicles mainly had diffusive mobility.

To examine in more detail, TrackMateR allowed us to measure many parameters that collectively describe the motion of each vesicle track during the imaging period (Figs 6B–F and S5A–C). We used a dimension-reduction method, principal component analysis, to compute and visualize a mobility profile for each vesicle type, using eight of these parameters for each track (total, 96,496 tracks; Fig S5A–C). Plotting all track data on the two principal component axes allowed us to see the shape of the data for each marker (Fig 6E). The similarity of these data shapes could be assessed using a distance-based method (see the Materials and Methods section). Subsequent hierarchical clustering revealed that the markers could be grouped into three classes based upon their motion and density (Fig 6F). The purely directed motion of EB3—corresponding to the growth rate of microtubules—placed it clearly in a third category. The endosomal and lysosomal markers Rab5 and LAMP1 were most similar to EB3, having a mainly directed motion. The remaining vesicle markers were in one robust class containing Rab35, Clathrin, SCAMP1, TPD54, SCAMP3, Rab30, ATG9A, Rab11, and ML1N (Fig 6F), and it was not possible to further differentiate this class. The motion of these vesicles was mainly diffusive, with an MSD exponent α ≈ 1 (Fig 6C and D). This class was distinct and robust when the clustering was repeated with subsampling of the data. To summarize, small trafficking vesicles (which include INVs and clathrin-coated vesicles) are numerous and mainly move by normal diffusion. Conversely, larger objects which are lower abundance have a higher fraction of directed motion which is consistent with microtubule-based transport.

Some INVs are constitutively exocytosed

To understand how diffusion of INVs is related to vesicle transport, we need to be able to assay the delivery of INV cargo at its target compartment. Because INVs are involved in anterograde transport and receptor recycling (Larocque et al, 2020, 2021), we reasoned that we should be able to assay the delivery of INV cargo at the plasma membrane. We used synaptophysin–pHluorin (sypHy) as a cargo molecule and assayed delivery by direct visualization of fusion using either TIRF microscopy of spinning disk confocal microscopy (see the Materials and Methods section). Bright flashes corresponding to exocytic events could be imaged, occurring with a frequency of ∼0.5 μm⁻² min⁻¹ (Fig 7A). We confirmed that sypHy was trafficked in INVs by performing a co-relocation experiment. Here, INVs were captured on the mitochondria by the induced relocalization of mCherry-FKBP-TPD54, and the co-relocation of sypHy was confirmed in the green channel (Fig 7B). The co-relocation was even more obvious when the pHluorin was unquenched with 50 mM NH₄Cl, which suggests that the INVs carrying sypHy are acidified. No co-relocation of sypHy was observed when the mCherry-FKBP-TPD54 R159E mutant was relocalized to the mitochondria, confirming that the co-relocation was specific to INVs (Fig 7B).

To confirm that the exocytic carriers that deliver sypHy to the plasma membrane are indeed INVs, we measured the exocytic frequency of sypHy before and after the capture of INVs are the mitochondria (Fig 7C). To do this, we co-expressed sypHy with blue MitoTrap and either mCherry-FKBP-TPD54 WT or mCherry-FKBP as a control. We measured a ∼50% drop in the exocytic rate, 3 min after rapamycin addition in cells co-expressing mCherry-FKBP-TPD54, whereas there was no change in cells co-expressing mCherry-FKBP which had no capture of INVs at the mitochondria (Fig 7C).

Diffusion is important for constitutive exocytosis of INVs

Having confirmed that we could assay the delivery of INVs at a target compartment, we wanted to test if diffusion of INVs was important for delivery. We initially used osmotic compression induced by incubating cells in a solution containing 12% PEG1450. Under these conditions, we saw a dramatic reduction in the rate of sypHy exocytosis compared with control conditions, suggesting that diffusion is important for vesicles to deliver their cargo (Fig 8A). However, it is likely that osmotic compression affects other forms of transport besides diffusive movement. To specifically slow the diffusion of INVs, we used a large (estimated 14 nm) protein “dongle” (Küey et al, 2019) to increase the effective size of INVs (Fig 8B). We reasoned that in GFP-TPD54 knock-in cells, expression of a GFP nanobody coupled to three copies of FKBP would bind GFP-TPD54, decorating the surface of INVs to generate “furry” vesicles. Because diffusion is related to the size of the particle, an artificial increase in size will slow diffusion. We calculated that an INV with an effective diameter of 36 nm has a theoretical diffusion coefficient of 0.29 μm s⁻², whereas an INV decorated with dongles is predicted to have a 64 nm diameter, resulting in a 0.16 μm s⁻² diffusion coefficient (see the Materials and Methods section, Fig 8C).

We first confirmed that diffusion of furry vesicles was slowed compared with regular INVs by performing FRAP measurements on endogenous GFP-TPD54. The measured diffusion coefficient reduced from 0.24 μm s⁻² in control to 0.18 μm s⁻² in cells where the dongle was expressed (Fig 8D). Using the curve in Fig 8C, these measured diffusion coefficients correspond to diameters of 44.0 nm and 56.4 nm for the INV and the INV plus dongle, respectively. Under these conditions, we next analyzed the rate of sypHy exocytosis and observed an ∼50% decrease in cells expressing the dongle compared with the control (Fig 8E and F), suggesting that restricted diffusion reduces vesicle delivery to the target site. However, it is possible that the dongle impairs INV fusion and that affects the measured rate of exocytosis. To investigate this possibility, we inspected the movies to determine if we could detect any vesicles that approached the membrane but did not fuse. Out of 3,767 approaches (2,307, control; 1,460, dongle-expressing), all 3,767 fused immediately upon approach to the membrane with no vesicles approaching that did not result in a fusion signature. We therefore examined the events themselves to see if we could find any evidence of impaired fusion (Fig 8G–I). The peak amplitude of fusions of a random sample of events was similar between control and dongle-expressing cells (Fig 8G and H). Moreover, the apparent rise time of a time-resolved average of events in dongle-expressing cells was indistinguishable from that of control cells (Fig 8I). These findings argue against the idea that the measured decrease in exocytic rate in dongle-expressing cells was because direct interference in fusion itself. In summary, these experiments strongly suggest that diffusion is an important mode of transport for delivery of cargo at the target compartment.

Molecular biology

The following plasmids were available from previous work: GFP-FKBP-TPD54 WT and R159E mutant (Larocque et al, 2021), synaptophysin-pHluorin/sypHy (Granseth et al, 2006), mCherry-MitoTrap (Cheeseman et al, 2013), LAMP1-mCherry-FRB (Küey et al, 2022), GFP-Rab35, GFP-Rab30, mCherry-FKBP-Rab11a (Larocque et al, 2020), and GFP-clathrin light chain A (Royle et al, 2005). StayGold-TPD54 was made by DNA synthesis (Twist Bioscience) using StayGold N1/C4 as the tag (Hirano et al, 2022). GFP-SCAMP1 and GFP-SCAMP3 were cloned by PCR and insertion into BglII and SalI sites in pEGFP-C1. ATG9A-GFP was cloned by amplifying human ATG9A (gift from Sharon Tooze) by PCR to SalI and EcoRI sites, remove the stop codon and insert into pFKBP-EGFP-N1 before moving the insert to pEGFP-N1 using the same sites. Plasmids mCherry-Rab5a (#2769), pEBFP2-N1 (#54595), GFP-2xML1N (#67797) or mNeonGreen-EB3-7 were from Addgene or Allele Biotech. Blue MitoTrap (pMito-eBFP2-FRB), was cloned by cutting eBFP2 from pEBFP2-N1 and inserting in place of mCherry in pMito-mCherry-FRB using AgeI and BsrGI sites.

Cell biology

Wild-type HeLa cells (HPA/ECACC 93021013) or GFP-TPD54 knock-in (clone 35) HeLa cells (Larocque et al, 2020) were maintained in DMEM with GlutaMAX (Thermo Fisher Scientific) supplemented with 10% FBS and 100 U ml⁻¹ penicillin/streptomycin. RPE1 cells stably expressing EB3-tdTomato (kind gift from Anne Straube, Warwick), were cultured in DMEM/Nutrient Mixture F-12 Ham (Sigma-Aldrich) supplemented with 10% FBS, and 100 U ml⁻¹ penicillin/streptomycin, 2 mM L-glutamine, and 0.25% sodium bicarbonate (Sigma-Aldrich); supplemented with 500 μg ml⁻¹ G418. All cells were kept in a humidified incubator at 37°C and 5% CO₂ and were routinely tested for mycoplasma contamination by a PCR-based method. Cells were transfected with GeneJuice (Merck) according to the manufacturer’s instructions.

Relocalization of FKBP-tagged TPD54 to MitoTrap was induced by application of rapamycin (Alfa Aesar) at a final concentration of 200 nM, as described previously (Larocque et al, 2020). For microtubule depolymerization, cells were treated with 20 μM nocodazole for at least 1 h. For actin depolymerization, cells were treated with 1 μM latrunculin B (Merck) for 25 min.

For ATP-depletion experiments, cells in full media were washed twice with PBS, then cultured in DMEM glucose-free (11966-025; Gibco) supplemented with 10% FBS, and 100 U ml⁻¹ penicillin/streptomycin, 6 mM of 2-deoxyglucose (Apexbio), and 10 mM of sodium azide (G-Biosciences) for 1 h. The efficiency of depletion was assayed by measuring intracellular ATP levels using Luminescent ATP Detection Assay Kit (ab113849; Abcam) according to the manufacturer’s instructions.

Osmotic compression was applied by exchanging full culture medium with full media supplemented with various concentrations of PEG1450 (P7306; Sigma-Aldrich). Washout of PEG1450 was done by progressive and successive dilutions until the medium was completely changed to avoid osmotic shock by immediate replacement.

Where cells were grown on micropatterns, a total of 27,000 Hela cells were seeded onto micropatterned chips (CYTOO chips Standard; CYTOO Inc.) and incubated for 1 h. As cells began to attach to the micropatterns, the medium was changed to remove all nonattached cells and then returned to the incubator to complete attachment. Later the same day, cells were either imaged live or fixed for immunofluorescence; only micropatterns containing a single cell were analyzed. Here, cells were fixed at room temperature with PFA solution (3% formaldehyde and 4% sucrose in PBS) for 15 min and permeabilized at room temperature in 0.5% Triton X-100 in PBS (or 0.1% saponin in PBS) for 10 min, followed by blocking with 3% BSA in PBS for 1 h. To amplify GFP signal, cells were incubated with GFP-booster (Alexa Fluor 488, 1:200; Invitrogen). After three PBS washes, coverslips were mounted with Mowiol solution DAPI.

To visualize the cytoskeleton, microtubules were stained in live cells using 0.5 μM of SiR-tubulin (Spirochrome).

Microscopy

For live cell imaging, cells were plated onto fluorodishes (WPI) and imaged in complete media in an incubated chamber at 37°C and 5% CO₂. All live cell imaging was done using a Nikon CSU-W1 spinning disc confocal system with SoRa upgrade (Yokogawa) with a Nikon 100 × 1.49 NA oil CFI SR HP Apo TIRF objective with optional 2.8× intermediate magnification and 95B Prime camera (Photometrics). This system has a CSU-W1 (Yokogawa) spinning disk unit with 50 μm and SoRa disks (SoRa disk used), Nikon Perfect Focus autofocus, Okolab microscope incubator, Nikon motorized xy stage, and Nikon 200 μm z-piezo. Excitation was via 405 nm, 488 nm, 561 nm and 638 nm lasers with 405/488/561/640 nm dichroic and Blue, 446/60; Green, 525/50; Red, 600/52; FRED, 708/75 emission filters. Acquisition and image capture were via NiS Elements (Nikon).

For fast imaging to analyze vesicle movement, images were typically acquired every 60 ms. Exposure time was limited so a high laser power was used (∼80%) and movies were limited by photobleaching. StayGold imaging allowed a lower laser power ∼40% and longer movie acquisition. Although the vesicles are subresolution, the fluorescent spots captured for analysis were from a single intensity population, indicating single vesicle resolution (Fig S6). Exceptions were Rab30 and Rab35, where a subpopulation of brighter spots indicates larger vesicles. To image exocytosis, early experiments used a Nikon TIRF microscope system, although we found that imaging the ventral plane of the cell with the spinning disc system gave similar results. Images were acquired every 200 ms, which was sufficient to see events and spread of sypHy in the plasma membrane. For experiments where exocytosis was imaged in clone 35 cells, we found that the increase in fluorescence from sypHy fusion and dequenching was easily distinguishable from GFP-TPD54 in the background. For experiments where pHluorin was fully dequenched, a solution containing 50 mM NH₄Cl was added to the cells.

FRAP and FLIP experiments were controlled by NiS Elements. Photobleaching was performed via C-TIRF 405 C200812 Chroma filter. A circular region of interest (ROI) of 4 or 6 μm was defined for FRAP or FLIP respectively, and 100% laser power, 40 µs dwell time were empirically determined to bleach the GFP- or StayGold-tagged constructs. The effective size of the bleached region was confirmed by bleaching a fixed cell and measuring the bleached area. For FRAP experiments, images were taken with low laser power to limit photobleaching, within the same focal plane at regular intervals (between 80 ms and 1 s) to monitor fluorescence recovery in the ROI. For FLIP experiments on GFP-TPD54, regular (every 3 s) sequential bleaching was done with continual image acquisition.

Data analysis

Generally, images were analyzed in Fiji (Schindelin et al, 2012) using custom scripts and the outputs were processed and plotted using R. For quantification of TPD54 relocalization, we used a custom-written ImageJ script. Briefly, mitochondria were segmented using LabKit and used as a mask to quantify the mean pixel density in the TPD54 channel in each frame after background subtraction. Data were averaged and fitted with a single exponential function as described below. Equivalent results were obtained with GFP-FKBP tags rather than mCherry-FKBP, and with a C-terminally tagged WT rather than an N-terminally tagged construct.

Mean pixel densities of the FRAP/FLIP region were measured for each cell, background subtracted, and then normalized to the pre-bleach point. In R, the average relative intensity was used for fitting using the equation,y0+A expx−x0τwith the reciprocal of the SD of each time point used for weighting (FRAP only). The diffusion coefficients from FRAP experiments were calculated using the corrected the Soumpasis equation (Kang et al, 2012),D=re2+rn28τ1/2where r_e is the effective post-bleach radius and r_n is the user-defined radius, both in μm; τ_1/2 is the half time determined on the recovery curves in seconds.

To calculate mitochondrial proximity, cells and mitochondria were each segmented by two LabKit classifiers using the TPD54 and MitoTrap channels, respectively. A distance map from mitochondria (within the cell of interest) was generated using the 3D suite ImageJ (Ollion et al, 2013). Voxel 3D distances for the entire cell volume (excluding mitochondria) were imported into R and the median distance was calculated for each cell.

All single particle tracking was done using TrackMate v7.10.2 (Tinevez et al, 2017). The outputs were analyzed using the R package TrackMateR, which was written specifically for this project (Royle, 2022). To identify directed transport of vesicles, TraJClassifier was used (Wagner et al, 2017). Tracks classified by TraJ as “directed” contained significant diffusive motion, so we pooled all tracks not classified as directed and measured their efficiency (related to linearity of track). The upper limit for the experiment was then used to identify tracks that were moving in a truly directed manner. Track parameters assessed by TrackMateR were used in the principal component analysis. The individual points on PC2 versus PC1 were used to generate a contour map which is a 3D shape that describes the population of tracks for each marker in principal component space, with density as a third dimension. For hierarchical clustering, a dissimilarity matrix was assembled from pairwise comparisons (total distance) of these 3D shapes in IGOR Pro. Similar results were achieved using other methods including Hausdorff distance, and with other dimension reduction techniques. We were unable to further distinguish Rab35, clathrin, SCAMP1, TPD54, SCAMP3, Rab30, ATG9A, Rab11, and ML1N into subcategories of motion.

To quantify exocytosis, we constructed a simple automated detection procedure. All movie frames were projected onto a single image and Find Maxima in Fiji was used to center ROIs to measure intensity over time in the movie. These outputs were read into R, normalized, and then a peakfinding procedure was used to find peaks (exocytic events) and thereby calculate the number of events per unit area per unit time. A manual method was also used following an automated blinding procedure. Here, exocytic events were checked manually to ensure that they represented true exocytic events.

Theory and simulations

To estimate the diffusion of an INV or an INV decorated with dongles, we used the Stokes–Einstein equation,D=kBT6π η rwhere k_B is the Boltzmann constant, T is temperature, η is the cytoplasmic viscosity, taken as 4.4 × 10⁻² Pa s (Kalwarczyk et al, 2011), and r is the radius of a spherical particle.

We note that measurements of diffusion of 40 nm diameter genetically encoded multimeric nanoparticles in HEK293 cells estimate D_eff = 0.5 μm² s⁻¹ (Delarue et al, 2018), which is higher than this prediction. This may reflect a lower cytoplasmic viscosity of HEK293 cells compared with HeLa, or alternatively, diffusion of INVs may be slowed by transient interactions with cellular components that inert genetically encoded multimeric nanoparticles are not subject to.

To approximate the diffusion coefficient from FLIP measurements, a custom-written routine for IGOR Pro 9 was used. Briefly, a circular cell area of 15 μm radius was used and the bleach area set as a variable between 2.6–2.8 μm radius in five steps (to simulate small differences in the bleach area and cell footprint). 100 vesicles were simulated to diffuse with their positions captured at 60 ms intervals, and a bleach pattern of 60 ms bleach every 3 s used to match the experiments. Any vesicles that fell inside the bleach region while the laser was simulated as on were inactivated. The half-time of inactivation was found by averaging 10 simulations at defined D and bleach radius. This simple simulation assumes that the number of vesicles in the cell is constant (no new vesicles are made and no vesicles fuse during the simulation period).

To assess the ability of TrackMate to accurately track high-density, high-mobility vesicles at different frame rates, a simulation was used in Fiji where vesicles were “imaged” in a simulated confocal plane as they diffused through a 14 × 14 × 14 μm volume. The resulting images from these simulations were automatically tracked by TrackMate using identical settings to detect the spots and the outputs analyzed by TrackMateR. The ground truth locations of spots outputted by Fiji were used to verify the mobility of simulated vesicles.