Because commercial TIRF microscopes do not allow polarization control, we built a TIRF microscope that allows continuous variation of the excitation polarization (Supplementary Figure S1 and Materials and Methods). The light exiting each laser is polarized, with a polarization ratio >1:100 (Materials and Methods). The outputs from individual lasers are coupled into polarization maintaining fibers, which are then combined into a single fiber using a wavelength division multiplexer, preserving polarization. This fiber is mounted into an optomechanical cage system with a rotation stage (Supplementary Figure S1). The fiber can be rotated continuously manually to vary the polarization of the excitation beam. The beam is expanded and deflected by a motorized mirror that also sets the position of the beam focused at the back focal plane, thereby controlling the incidence angle. The setup is controlled through the open-source software micro-manager (Edelstein et al., 2014) (see Materials and Methods). Supported bilayers were generated using the vesicle fusion method in microfluidic channels, as described in Karatekin and Rothman (2012). Rotating the excitation polarization resulted in a sinusoidal variation of the mean fluorescence intensity of LR-PE doped SBLs as expected (Supplementary Figure S2). A Glan-Taylor prism can be inserted into the rotation mount before the beam expansion optics to improve the polarization ratio, but this makes it more challenging to maintain the laser beam’s position at the back focal plane of the objective (hence the evanescent depth) fixed as the rotation mount is rotated.

For SUV-SBL fusion experiments, the SBL is reconstituted with neuronal/exocytotic t-SNARE proteins Syntaxin-1 and SNAP25. After incubation of the coverslip with SUVs for at least 30–60 min, the microfluidic chamber is rinsed. The formation of a homogenous and continuous SBL is verified using the 1,2-dioleoyl-sn-glycero-3-phosphoethanolamine-N-(7-nitro-2-1,3-benzoxadiazol-4-yl) (NBD-PE) label in the SBL. Fluorescence recovery after photobleaching of NBD-PE was used to ensure the fluidity of the SBLs (Materials and Methods). If a SBL passed these quality checks, SUVs reconstituted with the cognate neuronal/exocytic v-SNARE vesicle-associated membrane protein 2 (VAMP2, also known as synaptobrevin-2) were introduced at a continuous flow rate of 2 μL/min (with a flow cell cross-section of 300 μm by 75 μm, the mean linear flow rate was ∼1.5 mm/s). The SUVs were labeled with LR-PE, excited at 561 nm. We monitored vSUV docking and fusion events continuously using TIRF microscopy, and recorded stream acquisitions at 56 frames/s. As control experiments, we incubated the tSBLs with a solution containing the soluble cytoplasmic domain of VAMP2 (CDV) that competes with the full-length VAMP2 on SUVs for binding the t-SNAREs on the SBL. The rate of fusion events (normalized to SUV lipid concentration and detection area) was 5-fold smaller for the control, consistent with previous reports (Karatekin et al., 2010; Nikolaus and Karatekin, 2016; Stratton et al., 2016).

We first reproduced previous results (Stratton et al., 2016) obtained with a commercial TIRF microscope (Nikon Eclipse Ti), in which the excitation polarization was fixed to s-pol (perpendicular to the plane of incidence, Figure 2C, corresponding to θ = 0 ° on our rotation stage). With this polarization, the total intensity in a 22 μm by 22 μm box (82 by 82 pixels) around the docking/fusion site first increases suddenly upon docking of a SUV to a value I d o c k (Figure 1B). After a delay τ d o c k , when the intensity has decreased slightly to a value I f u s due to photobleaching in the SUV, membrane fusion results in a second increase in the total intensity over timescale τ r e l e a s e as the fluorophores are transferred from the SUV into the SBL where they are brighter. This increase tends toward a maximum value I S B L but reaches a value I m a x < I S B L due to photobleaching. The total intensity is lower by a factor λ T I R F = I d o c k / I S B L in the SUV compared to the SBL. After reaching I m a x , the intensity decays with characteristic timescale τ b l e a c h due to photobleaching which is stronger for fluorophores in the SBL (the box size is chosen large enough that no labels leave the box during this time).

If the fusion pore did not hinder release, then release kinetics would be limited by how rapidly the fluorophores diffuse around the SUV and occur on a time scale τ v e s = A v e s / D l i p , where A v e s is the SUV area and D l i p lipid diffusivity. By contrast, if the fusion pore slowed release, τ r e l e a s e would be significantly longer than τ v e s . Stratton et al. (2016) defined a pore openness, P o , to quantify the degree by which the fusion pore impedes release kinetics: P o = g τ v e s τ r e l e a s e ,   τ v e s = A v e s / D l i p (1) where g = b / 2   π   r p is a factor that reflects the role of the pore geometry on lipid release rate. With typical values for the height of the pore (Breckenridge and Almers, 1987) b = 15  nm and the fully open pore radius r p = 3  nm, g is of order unity (Stratton et al., 2016). Note that r p includes half the bilayer thickness (2 nm), i.e., the radius of the aqueous lumen of the pore is 1 nm. For a two-state (open-closed) pore, P o is the fraction of the time the pore is in the open state. For a flickering pore with a continuously varying size in time Eq. 1 is equally valid, with P o the time-averaged pore radius relative to the fully open pore radius (Stratton et al., 2016).

Thus, if we knew τ v es , we could deduce whether the fusion pore significantly impedes release. We can in fact estimate τ v es for every fusion event, from combining the intensity of the docked vesicle, I d o c k , single labeled-lipid intensity in the SBL, I l i p , the intensity reduction factor λ T I R F , the known labeling density ρ l i p , and the lipid diffusivity, D l i p . As the lipid labels diffuse away from the fusion site, individual labeled lipids become discernible and thus single-lipid intensity in the SBL, I l i p (Supplementary Figure S2), and lipid diffusivity, D l i p , can be measured from single particle tracking. This intensity is reduced by a factor λ T I R F in the SUV, i.e., the intensity of a single lipid label in the SUV, averaged over all locations in the vesicle, is λ T I R F   I l i p . Thus, A v e s = I d o c k / ( λ T I R F   I l i p   2 ρ l i p ) . (2)

Estimation of λ T I R F is not trivial, because of photobleaching. For a good estimate, fitting of the total intensity time profile to a model is needed. However, hindering of release by the fusion pore can qualitatively modify the kinetics of release. Stratton et al. (2016) considered two limiting cases for release kinetics 1) diffusion-limited release (lipids are released as rapidly as they can diffuse through the pore’s neck), 2) pore-limited release (the pore slows release by flickering). These two limiting cases produce qualitatively different release kinetics (Stratton et al., 2016). In the former case, the fraction of labeled lipids remaining in the vesicle a time t after the pore first opens, ϕ v e s , decays with an inverse time dependence, ϕ v e s = τ v e s / t , whereas the latter produces an exponential decay, ϕ v e s = e − t / τ r e l e a s e . Since a priori it is not known which limiting case describes release better, a procedure was adopted by Stratton et al. (2016) whereby it is assumed release is pore-limited. The total intensity profile I t o t ( t ) was then fitted to the kinetics expected for this case to extract τ r e l e a s e and λ T I R F . In addition, lipid diffusivity, D l i p , was measured by tracking of individual lipid dyes as they became discernible in the SBL (Stratton et al., 2016). Combining these parameters allowed estimation A v e s , τ v e s , and P o (Eqs. 1, 2). Only small values of P o are consistent with the pore-limited release assumption. If this procedure produced a P o value nominally ≥ 1 , the event was flagged as a diffusion-limited release. It was then verified that the intensity profiles for such events are better described by the diffusion-limited release case.

No significant differences were found between results obtained using the custom-built pTIRF set to s-pol excitation or the Nikon Eclipse Ti TIRF microscope (Stratton et al., 2016), validating measurements with the new instrument.

One of the major bottlenecks with the procedure above is estimating lipid diffusivity by tracking of individual lipids. In addition, with polarizations that produced lower fluorescence intensities in the SBL that made single-particle tracking even more challenging, the procedure described above was not practical here. We therefore adopted a modified procedure in which the total fluorescence intensities in five concentric circles of increasing size were fitted simultaneously to a model function which captures both the radial shape of the point-spread function and the expected radial spreading of the released dye due to diffusion (Supplementary Figure S3 and Materials and Methods). This procedure provided a more constrained fit and allowed us to estimate the lipid diffusion coefficient, D l i p , in addition to the bleaching time, τ b l e a c h , the lipid release time, τ r e l e a s e , and the intensity reduction factor λ T I R F as independent parameters in the fit. The vesicle intensity just after docking, I d o c k was extracted from the intensity profile with the largest circle radius (11 pixels). The bleaching rate was assumed to be proportional to excitation efficiency (i.e., bleaching in the SUV was assumed to be λ T I R F times the rate in the SBL). To estimate the single lipid intensity I l i p ( θ ) , we tracked single labeled lipids in the SBL for θ = 0 ° (s-pol) for which single-lipids were brightest (Supplementary Figure S2). For other polarizations, tracking was not feasible or reliable, so we measured the variation of the average SBL intensity, labeled with 0.5% LR-PE, as a function of polarization angle and used this information as correction factor (Supplementary Figure S2). We then used Eq. 2 to calculate the vesicle area A v e s , and Eq. 1 to obtain the pore openness. A flowchart summarizing the procedure is shown in Supplementary Figure S4. P o quantifies how much lipid release is slowed compared to free diffusion through an un-restricted fusion pore. For our system, we have previously shown that pore flickering is the main mechanism of release-slowing (Stratton et al., 2016), enabling P o to be interpreted as a duty-cycle, but it can equally be used to empirically quantify release-slowing in systems where the mechanism has not been established.

Using this procedure, and for θ = 0 ° , we found that for most events, lipid release was slower than expected for lipid diffusion, i.e., τ r e l e a s e ≫ τ v e s . Pore openness values were 0.0003 < P o < 0.90 for most fusion events (78%), with mean = 0.39. For 19% of the pores, the procedure returned a nominal P o value > 1 , indicating a “permanently” open pore, P o = 1 . Note that “permanently” here means that the pore was open long enough that all lipid labels were released during a single flicker. For the largest vesicles studied ( R v e s ≈ 80   nm ) , τ v e s ≈ 70   ms (with D l i p ≈ 1.1   μ m 2 / s ), so “permanently open” pores could in fact be flickering at frequencies <14 Hz. Overall, these results are consistent with those reported previously (Stratton et al., 2016), validating the new procedure.

What is the contribution of the excitation polarization to the intensity change λ T I R F as a lipid-dye is transferred from the SUV to the SBL membrane? Here we use an extrapolation procedure to estimate this quantity. We start with the relationship between the vesicle area (hence vesicle radius R v e s = A v e s / ( 4 π ) ), the intensity reduction factor, λ T I R F , and the normalized docked intensity, I d o c k / I l i p , through Eq. 2 (note that the labeling density ρ l i p is fixed). Thus, the vesicle radius is a function of λ T I R F and I d o c k / I l i p . The parameters λ T I R F and I d o c k / I l i p are actually not independent from one another, but the theoretical relationship between the two is complex (Stratton et al., 2016). Thus, we determined an empirical relationship between these two quantities by plotting λ T I R F as a function of I d o c k / I l i p in Figure 3A for θ = 0 ° (s-pol). Using this relationship, one can obtain vesicle size either as a function of the normalized docked intensity, R v e s ( I d o c k / I l i p ) , or as a function of the intensity reduction factor, R v e s ( λ T I R F ) . Both representations are useful. The former is plotted in Figure 3B, which shows that the normalized docked intensity uniquely determines R v e s . That this should be case is not entirely obvious, because the polarization contribution to I d o c k / I l i p varies as a function of distance from the interface in a non-monotonic manner (Stratton et al., 2016). The latter representation is inverted, then the value of λ T I R F as R v e s → 0 is extrapolated to estimate the pure polarization contribution, as the evanescent field decay contribution vanishes at this limit. It was shown by Stratton et al. (2016) that the slope at the origin is λ T I R F 0 / δ T I R F M , where λ T I R F 0 is the value of the intensity reduction factor extrapolated to zero vesicle radius and δ T I R F M is the characteristic decay length of the evanescent field. For the incidence angle used, we independently estimated δ T I R F M ≈ 78   nm here (Materials and Methods). To estimate the slope at origin, we fitted a line to R v e s ≤ 35   nm values, while constraining the x-intercept to be equal to δ T I R F M ≈ 78   nm , obtaining, slope = − 0.0105 ± 0.0004   nm , and λ T I R F 0 = 0.82   (95% confidence interval, CI = 0.79 − 0.85 ).

We repeated the procedure described above at additional excitation polarizations (Supplementary Figure S5), finding λ T I R F M 0 = 0.83   ( C I = 0.78 − 0.88 ) , 1.51   ( 1.43 − 1.60 ) , and 1.97   ( 1.85 − 2.01 ) for θ = 30 , 60 , and 90 ° , respectively (Figure 3D). Thus, for θ = 60° and 90° (p-pol) cases, the pure polarization contribution is a decrease in total intensity upon lipid transfer from the SUV to the SBL, with λ T I R F M 0 > 1 . At finite values of R v es , the polarization effect competes with the intensity enhancement from the evanescent field decay effect (Figure 2). The two contributions are equal at R v e s ∗ = 22.6   nm   ( CI = 20.0 − 25.8   nm )  and  36.8   nm  ( 30.4 − 47.9   nm ) for θ = 60 ° and 90 ° , respectively (Supplementary Figure S5). For these polarizations, 59, and 80% of all events, respectively, had λ T I R F > 1 . By contrast, only 4% of all events for θ = 0 ° or 30 ° had λ T I R F > 1 , due to noise. For vesicle sizes close to R v e s ∗ , the net change in the total intensity upon fusion is near zero, such as the case depicted in Figure 2D, middle. For such cases, analysis and extraction of a P o value is particularly challenging.

Overall, these results show that for polarizations that contribute a decrease to the change in total intensity upon SUV-SBL fusion ( λ T I R F M 0 > 1 ) , there is a polarization-dependent vesicle size for which the net intensity change is nil, making analysis of lipid mixing kinetics very challenging. For the commonly used lipid dye LR-PE, pure polarization effects can vary by a factor of ∼2.4; importantly, a polarization with a net positive contribution to the signal change can be chosen to ensure all events can be analyzed for lipid mixing kinetics.

Two of the main bottlenecks in the analysis pipeline are visual identification of fusion events and single-molecule tracking of lipids for estimation of D l i p (Supplementary Figure S4). When using non-optimal excitation polarizations, the most obvious effect is that it becomes much harder to visually identify fusion events and distinguish them from undocking events. Thus, considerably more time is spent on identification of events. Although we did not use an automated event-detection algorithm, it is likely that event detection would be more challenging for such an algorithm for non-optimally excited samples.

Once fusion events are correctly identified, with non-optimal excitation polarizations the most important challenge for analysis is a loss of sensitivity to single lipids diffusing in the supported bilayer after vesicle fusion. This made single-molecule tracking based estimation of lipid diffusion more challenging in all but the s-polarized case. We were able, however, to estimate diffusion coefficients based on the lateral spread of the released dye, with these estimates being broadly compatible across polarizations. The distribution of other fitted parameters was also broadly similar across all excitation polarizations (e.g., see distributions for τ d o c k ; Supplementary Figure S6). We nonetheless observed subtle differences in the observed distribution of pore openness ( P o ) , with an apparent reduction in the number of slow-release events (small P o values) for p-polarization (Supplementary Figure S6). A potential explanation for this is that the relative fluorescence enhancement/reduction is much less pronounced than for p polarized case, making it significantly harder to unambiguously estimate the time at which the fusion pore opens. This is particularly likely for the slower release events where, in the absence of an enhancement on fusion, the docked vesicle signal will continue to dominate until well after the fusion pore opens.

Together these results let us conclude that choosing an optimal excitation polarization (e.g., s-polarized for LR-PE) is desirable, but that with knowledge of the actual polarization state and a suitable numeric model, useful measurements can be obtained on systems (such as many off-the shelf commercial TIRF microscopes) where polarization may not be controllable nor optimal.

As an illustrative application of pTIRF microscopy to the study of fusion pores, we explored the relationship between fusion pore openness and SNARE copy numbers and membrane curvature. How membrane fusion depends on SNARE copy numbers has been studied both using bulk and single-event assays. In single-event SUV-SBL fusion assays, it was found that 5–10 trans-SNARE complexes (SNAREpins) are required for rapid membrane fusion (Domanska et al., 2010; Karatekin et al., 2010). Bulk SUV-SUV fusion studies reported as few as a single SNAREpin could mediate lipid mixing (van den Bogaart et al., 2010), but content release required more (Shi et al., 2012), suggesting more SNAREpins drive larger fusion pores. Consistent with this idea, monitoring release of differently sized cargo in a nanodisc-SUV bulk fusion assay, Bello et al. showed that release of larger cargo required more SNAREpins (Bello et al., 2016). In nanodisc-based assays where single-pore conductance reflects pore size, it was shown that the mean fusion pore size increases with increasing SNARE copy numbers (Wu et al., 2017; Bao et al., 2018), an effect attributed to entropic repulsion among the SNARE complexes lining the pore’s waist (Wu et al., 2017). These results in reconstituted assays are consistent with observations in live cells that cargo release is faster when more SNAREs are available (Zhao et al., 2013; Acuna et al., 2014; Bao et al., 2018).

To test if pore openness, P o , depended on the number of v-SNAREs per liposome, N S N A R E , we plotted the fraction of open pores ( P o ≥ 0.9 ) as a function of N S N A R E (Figure 4A). Because our P o estimates are most reliable for θ = 0 ° , we used the corresponding data set for this analysis. The fraction of open pores f P o ≥ 0.90 increased with N S N A R E up to N S N A R E ≈ 30 , above which it plateaued around 0.5. We explored this relationship further, by plotting P o values for flickering pores ( P o < 0.9 ) as a function of N S N A R E . There was a weak trend for P o to increase with increasing N S N A R E , up to N S N A R E ≈ 30 (Figure 4B). Note that because the v-SNAREs are incorporated with a random orientation into the SUVs (Karatekin et al., 2010), the effective number of v-SNAREs facing outside the SUV on these plots, N S N A R E o u t , is ∼1/2 the plotted (total) value.

Membrane curvature could also contribute to fusion pore dynamics. Smaller liposomes fuse faster both in bulk experiments (Malinin et al., 2002; Hernandez et al., 2012; Yang et al., 2021) and computer simulations (Gao et al., 2008), although the effects on fusion pore dynamics are not clear. In PC12 and chromaffin cells, larger vesicles have more stable initial fusion pores (Zhang and Jackson, 2010), suggesting high membrane curvature may contribute to fusion pore expansion. To test for membrane curvature effects, we first plotted the fraction f P o ≥ 0.90 of open pores ( P o ≥ 0.9 ) as a function of vesicle curvature, C = 2 / R v e s , but no clear correlation emerged (Figure 4C). However, a weak trend emerged for P o values for flickering pores ( P o < 0.9 ) to decrease as a function of increasing curvature, C = 2 / R v e s , at high curvatures corresponding to the range of liposome sizes for which N S N A R E o u t ≲ 15 (Figure 4D).

Overall, these observations suggest that pore openness increases with vesicle area, likely with contributions from both SNARE copy numbers and membrane curvature.

1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC), 1,2-dioleoyl-sn-glycero-3-phospho-L-serine (sodium salt) (DOPS), 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine (POPE), 1-stearoyl-2-arachidonoyl-sn-glycero-3-phosphoethanolamine (SAPE), 1,2-dioleoyl-sn-glycero-3-phosphoethanolamine-N-(lissamine rhodamine B sulfonyl) (LR-PE), 1,2-dioleoyl-sn-glycero-3-phosphoethanolamine-N-(7-nitro-2-1,3-benzoxadiazol-4-yl) (NBD-PE), and 1,2-dioleoyl-sn-glycero-3-phosphoethanolamine-N-[methoxy(polyethylene glycol)-2000] (PEG2K-PE) were purchased from Avanti Polar Lipids (Alabaster, AL). NBD-PE (0.5 mol%) was included only in the t-SNARE SBL to test SBL fluidity using fluorescence recovery after photobleaching (Karatekin and Rothman, 2012; Nikolaus and Karatekin, 2016).

Recombinant proteins vesicle-associated membrane protein 2 (VAMP2, also known as synaptobrevin-2), syntaxin-1, and synaptosomal-associated protein 25 (SNAP25) were expressed, purified, and reconstituted into SUVs as described in detail previously (Karatekin et al., 2010; Karatekin and Rothman, 2012). Plasmids were a generous gift from J. E. Rothman (Yale University). A lipid-to-protein ratio (L:P) of 200 was used for vSUVs and 5,000 or 10,000 for tSBLs.

For reconstitution of protein, we used the method of refs. Karatekin et al. (2010); Karatekin and Rothman (2012). We used the following molar ratios for the vSUVs: POPC/DOPS/(SAPE or POPE)/PEG2KPE/LR = 54/20/20/5/1. The t-SBLs harbored one t-SNARE complex for every 5,000 or 10,000 lipids, and the lipid composition was (in molar ratios) POPC/DOPS/(SAPE or POPE)/PEG2KPE/NBD-PE = 69.5/10/15/5/0.5. As negative controls, we included the soluble cytoplasmic domain of the v-SNARE VAMP2 (CDV, residues 1–92, 10 μM) which associates with the t-SNAREs on the tSBL and inhibits docking and fusion of vSUVs. SUV diameters were 20–200 nm, estimated from dynamic light scattering (DynaPro NanoStar, Wyatt Technology, Santa Barbara, CA, United Sates).

For every batch of vSUVs, we calculated the actual lipid-to-protein ratio (LP) following ref. Karatekin et al. (2010). The lipid concentration was estimated using LR-PE fluorescence which was independently calibrated using standard solutions. The protein concentration was estimated from SDS-PAGE gels stained with Sypro Orange (MilliporeSigma, St. Louis, MO) running against a known concentration. Assuming an area per lipid (Hung et al., 2007) a l i p = 0.70   n m 2 we estimated the SNARE density Γ S N A R E = 1 / ( L P × a l i p ) for each batch. For most batches, the resulting snare density was 3600   c o p i e s / μ m 2 , with actual L P ≈ 400 . For every event that could be analyzed, the vesicle area A v e s was calculated using Eq. 2, which allowed us to estimate the number of v-SNAREs per SUV, N S N A R E = A v e s × Γ S N A R E .

Outputs from a LuxX 488 nm or 638 nm continuous-wave diode-pumped laser (200 or 150 mW maximum power, respectively, Omicron, Rodgau-Dudenhofen, Germany) were each coupled to a polarization maintaining fiber. The output of a 561 nm laser (150 mW maximum power, Cobolt 04-01 Series, Jive, Solna, Sweden) was modulated by an acousto-optical modulator (PCAOM V-50, Crystal Technology, Inc., Palo Alto, CA) before coupling into another polarization maintaining fiber. The fibers were combined into a single fiber using a polarization maintaining wavelength division multiplexer (OZ Optics, Ottawa, Canada). The fiber carrying the combined wavelengths was mounted onto a manual rotation mount (Thorlabs, Newton, NJ). We set the desired excitation field polarization by rotating this mount. If desired, the polarization ratio can be improved by inserting a Glan-Taylor prism, but in practice this complicates alignment as polarization is rotated and was not used. The beam was then expanded and passed through an adjustable diaphragm before being reflected by a mirror whose position was controlled by a motorized actuator (CONEX-TRB12CC DC servo actuator, Newport, Irvine, CA). The beam then went through a tube lens, an excitation filter (ZET488/10x, ZET561/10x, or ZET640/20x, Chroma), and a dichroic mirror (ZT488rdc, ZT640rdc, Chroma) before focusing onto the back focal plane of an Olympus PlanApo 60x/1.45 Oil TIRF objective, mounted on an inverted microscope (IX81, Olympus, Tokyo, Japan). Fluorescence was collected through the same objective, passed through a HHQ500LP and ET525/50m, HHQ575LP, and ET610/60M, or HQ660LP and ET700/75 (Chroma), and detected using an EM-CCD camera (Ixon-ultra-897, Andor, Belfast, United Kingdom). One pixel corresponded to 265 nm in the sample plane. We stream-recorded 60 s movies (3,300 frames) with exposure time 17.8 ms (duty cycle 18.3 ms). The microscope, including the mirror position for setting the evanescence depth was controlled by micro-manager (Edelstein et al., 2014) (the configuration file is available upon request). All experiments were carried at 32°C, using a heated stage insert (Thermo Plate, Tokai Hit, Shizuoka-ken, Japan).

The evanescent field depth was estimated by measuring the angle of incidence, θ , of the excitation beam with respect to the normal of the imaging plane, and using (Axelrod, 2008) δ T I R F   =     λ 0 / 4 π ( n g 2   s i n 2   θ − n w 2 ) − 1 / 2 , where λ 0 = 561   nm is the laser excitation wavelength, and n g = 1.52 and n w = 1.33 are the refractive indices for glass and water, respectively. An N-BK7 right-angle prism (20 mm per side, PS908, Thorlabs) was coupled to the TIRF objective using a cover slip and oil matching the refractive index of the glass. Adjusted to the same mirror position used in the SUV-SBL fusion experiments, the laser beam passed from the objective into the prism undeflected but was refracted at the glass-air interface as it emerged from the prism. The beam was projected onto a wall and the simple geometry was used to calculate the angle of incidence, θ = 71.8 − 72.6 ° , corresponding to δ T I R F = 77 − 79   nm . The highest intensity of the projected spot on the wall was found at 72.2 ° , δ T I R F = 78   nm .

We followed ref. Karatekin and Rothman (2012). Briefly, microfluidic channels were made by bonding a block of poly(dimethyl siloxane) (PDMS) replica of a microfabricated structure onto a glass coverslip. Prior to bonding, holes were punched into the PDMS block using a hole puncher (Schmidt Manual Press, Schmidt Technology, Cranberry Twp., PA) to connect tubing for introducing solutions. Coverslips (24 mm by 60 mm) were treated with air plasma for 10 min in a plasma cleaner (LTD Model SP100 Plasma system, Anatech, United Sates, Sparks, NV) before bonding to the PDMS. The PDMS was not plasma treated but was placed under vacuum for at least 20 min to avoid bubble formation during experiments. After assembly of microfluidic channels, a diluted and degassed tSUV or pfSUV suspension was introduced into the channels and incubated for at least 30 min. Unbound SUVs were rinsed away. SBL fluidity was tested using fluorescence recovery after photobleaching (FRAP) using the 488 nm laser to excite NBD-PE (Karatekin and Rothman, 2012; Nikolaus and Karatekin, 2016). Occasionally, SUVs adhered to the glass coverslip but did not burst and form a continuous fluid bilayer. In such cases, the coverslips were additionally incubated with reconstitution buffer (25 mM HEPES-KOH, 140 mM KCl, 100 µM EGTA, and 1 mM DTT, pH 7.4) with 10 mM Mg²⁺ for at least 30 min and then thoroughly rinsed with Mg²⁺ free buffer.

Prior to flowing SUVs into a microfluidic channel, the SBL in the viewfield was continuously bleached by 561 nm excitation to reduce background fluorescence. When the first v-SUVs reached the viewfield, image acquisition was initiated to record a 1-min movie consisting of 3,300 frames. Data was recorded for four different excitation polarizations, 0° (s-pol), 30°, 60°, and 90° (p-pol).

Analysis of vesicle fusion was done offline and began with visual identification of fusion events which were then tracked using the SpeckleTrackerJ plugin (Smith et al., 2011) of ImageJ (Schneider et al., 2012) with subpixel resolution. Tracks started the first frame in which a SUV docked onto the SBL until the frame in which it fused with the SBL, as evidenced by the onset of a sudden change in fluorescence intensity, accompanied by the spread of the signal. The track length defined the docking-to-fusion delay τ d o c k . Fusion events were further analyzed using PYME (www.python-microscopy.org) and the Python Anaconda platform. The pixel intensities surrounding a particle’s centroid position were summed within concentric circular regions of interest (ROI) with five different radii (3, 5, 7, 9, and 11 pixels radius). The analysis extended from 10 frames prior to docking until 50 frames after fusion. For each fusion event, the total intensity for all radii were simultaneously fitted to a model function which captures both the radial shape of the point-spread function and the expected radial spreading of the released dye due to diffusion. The use of a model encoding this radial information and fitted to multiple different sized ROIs allowed us to estimate the lipid diffusion coefficient, D l i p , the bleaching time, τ b l e a c h , and the lipid release time, τ r e l e a s e , as independent parameters in the fit. Our specific model function describing the total intensity within a radius R was as follows: I ( R , t ) =   I d o c k e d ( R ,   t ) +   I r e l e a s e ( R ,   t ) (3) where: I d o c k e d ( R , t ) = e − ( t − t d o c k ) τ b l e a c h ∫ 0 R P S F ( r )   d r { 0   t < t d o c k 1   t d o c k ≤ t < t f u s i o n e − ( t − t f u s i o n ) τ r e l e a s e t f u s i o n ≤ t I r e l e a s e ( R , t ) =   G e − G ( t − t f u s i o n ) τ b l e a c h ∫ 0 R [ P S F ⊗ H d i f f u s i o n ] ( r )   d r { 0 t < t f u s i o n [ 1 − e − ( t − t f u s i o n ) τ r e l e a s e ] t f u s i o n ≤ t (4) and P S F is an approximation to the microscope point spread function, H d i f f u s i o n = e − r 2 4 π D l i p t is the 2D diffusion Green’s function, ⊗ represents convolution and G = 1 / λ TIRF is the gain in intensity that a dye molecule experiences transiting from the vesicle into the bilayer. The bleaching rate is assumed to be proportional to excitation efficiency. The traces were background subtracted and normalized to the total intensity of the docked vesicle before fitted using a weighted least squares fit. Free parameters were G , D l i p , τ r e l e a s e , τ b l e a c h . Fit quality was evaluated visually for every fit and poor fits were excluded from further analysis. Fusion events with docking-to-fusion delays t d o c k < 4 frames were excluded, as shorter docking times did not result in good fits.