Individual Vesicle Fusion Events Mediated by Lipid-Anchored DNA

Abstract

Membrane fusion consists of a complex rearrangement of lipids and proteins that results in the merger of two lipid bilayers. We have developed a model system that employs synthetic DNA-lipid conjugates as a surrogate for the membrane proteins involved in the biological fusion reaction. We previously showed that complementary DNA-lipids, inserted into small unilamellar vesicles, can mediate membrane fusion in bulk. Here, we use a model membrane architecture developed in our lab to directly observe single-vesicle fusion events using fluorescence microscopy. In this system, a planar tethered membrane patch serves as the target membrane for incoming vesicles. This allows us to quantify the kinetics and characteristics of individual fusion events from the perspective of the lipids or the DNA-lipids involved in the process. We find that the fusion pathways are heterogeneous, with an arrested hemi-fusion state predominating, and we quantitate the outcome and rate of fusion events to construct a mechanistic model of DNA-mediated vesicle fusion. The waiting times between docking and fusion are distributed exponentially, suggesting that fusion occurs in a single step. Our analysis indicates that when two lipid bilayers are brought into close proximity, fusion occurs spontaneously, with little or no dependence on the number of DNA hybrids formed.

Introduction

Membrane fusion is central to many biological processes, including endo- and exocytosis and the transfer of membrane proteins between cellular compartments. The process of vesicle docking and fusion is mediated by formation of the SNARE protein complex, made up of recognition partners on the vesicle and target membranes, with many other accessory proteins assisting or regulating the process (1–3). Although extensively studied, essential questions about vesicle fusion, including the number of components involved and the precise physical mechanism, are not well understood and seemingly small differences in procedures and components among labs lead to different conclusions. Due to the complexity of the fusion reaction and the proteins involved, reductionist model systems can complement in vivo data to yield a better understanding of this biological process. Many such systems have been described that use the SNARE proteins (4–19) or synthetic surrogates for them (20–27), and these are providing valuable insight, as well as stimulating the development of increasingly realistic assays.

We have developed a model system (25–27) that employs synthetic DNA-lipid conjugates as surrogates for the SNARE machinery. This model system affords easy control over DNA sequence, binding geometry, and length—factors less easily probed in SNARE-mediated fusion—and it allows us to examine how fusion proceeds once the vesicle and target membrane are brought close together in the absence of accessory factors. The binding specificity of these DNA-lipid conjugates avoids having to deal with dead-end complexes due to promiscuous binding of incorrect partners, and the conjugates spontaneously insert into lipid membranes without requiring detergent dialysis—two issues that can be troublesome in reconstituted SNARE systems (1,2,28). Hybridization of complementary DNA pairs on different membranes, when anchored in the correct orientation, enables fusion between small vesicles in bulk (25,26). Both lipid and content mixing are observed. Previously, the mechanism of this multistep reaction could not be addressed as the kinetics of the docking and fusion reactions are convoluted in such ensemble measurements.

We recently developed a model membrane architecture—a DNA-tethered bilayer patch—that allows direct observation of individual vesicle-to-planar bilayer fusion events to better investigate the mechanism of DNA-mediated vesicle fusion ((27), Fig. 1 A). In this system, we use DNA in two independent roles. The first is used to construct the target tethered patch with the DNA linked to the lipid anchor at its 5′ end on both surfaces—DNA hybridization in this configuration is referred to as the tethering orientation. A second, orthogonal DNA sequence is used to initiate vesicle docking and fusion, with the DNA anchored at its 3′ on one surface and its 5′ end on the other—hybridization between these strands is referred to as the zippering orientation and brings the membranes into close apposition to facilitate fusion. This is directly analogous to the proposed geometry of the docked SNARE complex at the presynaptic membrane (27,29), an improvement over many model systems that study fusion only between vesicles.

Figure 1.

(A) Vesicle-to-tethered membrane patch assay. The target membrane patch, tens of μm in diameter, is tethered by 5′anchored 24mer DNA (magenta strand, 8 nm in height) to a modified glass coverslip, and also displays 5′anchored zippering DNA (blue) (27,30). DNA strands (magenta and blue) are on both leaflets of the tethered patch—not shown in the schematic to avoid confusion. A vesicle ∼50 nm in diameter, with 2% Texas Red-labeled lipids (red dots, both leaflets) and displaying 3′coupled zippering DNA (blue, outer leaflet only) docks via DNA hybridization and then fuses to the tethered membrane patch, monitored by fluorescence microscopy. Nothing is drawn to scale. (B) Example traces of different fusion behaviors, showing the integrated fluorescence intensity in a ROI around each vesicle over time: (i) docking-only, (ii) hemi-fusion-only, (iii) hemi-then-full-fusion, (iv) full-fusion-only. (C) Proportions of fusion outcomes observed (65 DNA/vesicle, 3′PolyA). Error bars = STD of three experiments. To minimize sampling bias, only events that docked >60 s before the end of the 100 s video stream were included (N > 400 for all experiments). (D) The distribution of % intensity decrease upon hemi-fusion. (E) The distribution of % of dye-labeled-lipid in the outer leaflet of vesicles, as determined by dithionite quenching of NBD-PE fluorophores in the outer leaflet of DNA-tethered vesicles.

Herein, we use the information gathered from this system to construct a mechanistic model of the DNA-mediated fusion process and thereby provide insight into several important questions for the biological fusion reaction: Is fusion a linear pathway or is it heterogeneous? What is the role of hemi-fusion? How many binding partners are needed for fusion to proceed? Does DNA hybridization (or SNARE partnering) actively transduce zippering into membrane fusion? We observe that heterogeneous fusion behavior, with an arrested hemi-fusion state being the predominant outcome, emerges when the two membranes are brought into close apposition via DNA hybridization. The kinetics and fusion outcomes of the system are not substantially altered by the DNA sequence or number density on the vesicle and there is a gap of seconds between completion of DNA hybridization and a fusion transition. Furthermore, we observe that very few hybrids, possibly just one, are sufficient to allow this fusion behavior to proceed, and that the kinetics of the system are not greatly limited or enhanced by the formation of additional hybrids.

Materials and Methods

Reagents

1,2-Dioleoyl-sn-glycero-3-phosphatidylcholine (DOPC), 1,2-dioleoylphosphatidyl-sn-glycero-3–ethanolamine (DOPE), and cholesterol (Ch) were purchased from Avanti Polar Lipids. Texas Red-DHPE and Oregon Green-DHPE were purchased from Invitrogen (Grand Island, NY). Ethynyl phosphonic acid and triethyl 2,2′,2″'-(4,4′,4″-nitrilotris(methylene)tris(1H-1,2,3-triazole-4,1-diyl))triacetate (TTMA) was generously provided by the Chidsey lab at Stanford. DNA oligonucleotides were synthesized at the Protein and Nucleic Acid facility at Stanford.

Preparation of vesicles displaying DNA

DNA-lipids were prepared as in (25) by covalent attachment of a synthetic lipid-phosphoramidite as the final base to either the 3′ or 5′ end of an oligo on resin (sequences in Table S1), and subsequent purification on HPLC. Fluorescent dyes were conjugated to lipid-DNA postdeprotection using NHS chemistry (see the Supporting Material Section 1.1).

Vesicles were prepared by extrusion. A mixture of 2:1:1 DOPC/DOPE/Ch was dried from chloroform under a stream of nitrogen, then under vacuum for 3 h. For lipid mixing experiments, 2% Texas Red-DHPE (TR-DHPE) or 2% Oregon Green-DHPE (OG-DHPE) was also added. The lipid film was rehydrated to 0.4 mg/mL in 10 mM sodium phosphate, 240 mM sodium chloride buffer, pH 7.4 and extruded 29 times through a 30 nm polycarbonate membrane (Avanti Polar Lipids, Alabaster, AL). The resulting vesicle diameters were 48 ± 12 nm, measured by dynamic light scattering (Fig. S1). DNA-lipids (at tens of μM) were added to 5 μL of vesicles at 0.4 mg lipids/mL to yield the desired DNA density and are only displayed on the outer vesicle leaflet.

Giant unilamellar vesicles (GUVs) destined for tethered bilayers were made by gentle hydration (27). Their lipid composition was 2:1:1 DOPC/DOPE/Ch with 0.01% OG-DHPE added to locate the tethered membrane. They contained two different DNA-lipids (tethering strand: 0.5% Sequence 3, and zippering strand: sequence and concentration as specified). DNA-lipids are in both leaflets. For experiments using dye-labeled DNA, the OG-DHPE was left out. Instead, 0.01% dye-labeled DNA-lipid was included as the zippering strand.

Fusion experiments

A DNA-alkyne (Sequence 2-alkyne) was covalently attached to an alkyl azide-functionalized glass coverslip using click chemistry. GUVs with complementary tethering DNA (Sequence 1) and the specified zippering DNA were incubated with the substrate until the GUVs deformed and ruptured to form tethered bilayer patches, ∼20–30 min (30). Membrane patches were thoroughly rinsed to remove any lipid debris and were characterized by uniformity of fluorescence and absence of lipid structures such as tubules. Once a suitable patch was located, vesicles (∼10 μL at 1 μg lipids/mL) with the complementary zippering strand were added to the solution above the membrane patch. In the lipid mixing experiments (e.g., Fig. 1 A), docking was observed by appearance of a fluorescent, diffraction-limited vesicle on the patch. Hemi-fusion was observed by a sudden decrease of fluorescence within the vesicle region of interest (ROI) as some portion of its lipid dye was transferred to the target patch and rapidly diluted via diffusion. In full fusion, the fluorescence intensity completely disappeared (operationally defined as losing >90% of original intensity). Analysis was performed in a homemade MATLAB program (The MathWorks, Natick, MA). For all experiments, we only analyzed events >5 μm away from the edge of the tethered patch to avoid edge effects.

In the dye-DNA experiments (see Fig. 3 A), the tethered patches displayed 0.01% 5′PolyT-Alexa546 DNA-lipid and the vesicles displayed 3′PolyA DNA-lipid (and no lipid dye). Analysis and calculation of number of DNA hybrids formed is given in the Supporting Material Section 1.4.

Figure 3.

(A) Schematic for DNA hybrid formation experiment. The tethered bilayer contains a low percent (0.01% ∼100 DNA-lipids/μm²) of Alexa546-5′PolyT (green stars), and the vesicles contain 3′PolyA. Docking is observed by the appearance of a spot as DNA hybrids form between tethered patch and vesicle, indicated by a (B) fast <1 s or (C) fast then slow (∼10 s) rise in the intensity of the ROI against a dim background of laterally mobile-labeled DNA on the patch. Hemi/full fusion is detected by a sharp decrease in the intensity to baseline levels, as the labeled DNA-lipid hybrids diffuse into the much larger membrane patch area. DNA is displayed only on the outside of the vesicle, so hemi/full fusion is indistinguishable. (D) Distribution of the number of DNA hybrids formed between vesicle and tethered patch before undergoing hemi- or full fusion. Calculation of number of hybrids is given in the Supporting Material Section 1.4.

NBD-dithionite quenching

Vesicles were prepared as previously mentioned, but containing 5% NBD-PE (headgroup labeled) instead of 2% TR-DHPE and displaying many 5′PolyT DNA lipids (∼195 DNA/vesicle). These vesicles were tethered to an Egg PC glass supported lipid bilayer (SLB) displaying the complementary 5′PolyA DNA-lipids. This DNA-lipid pair will hybridize in the tethering orientation (cf. Fig. S7) and can form many hybrids, holding the vesicle immobilized and apart from the SLB membrane. After an initial fluorescence micrograph was taken of many tethered vesicles, 8 μL of 100 mM sodium dithionite was added to the ∼40 μL solution above the vesicles. Following 2 min incubation, the chamber was rinsed with buffer and then a final micrograph was taken. A comparison of the initial and final fluorescence intensities of many vesicles yielded a distribution of percentage of NBD fluorophores in the outer leaflet (see Fig. 1 E).

Microscopy

All fusion experiments were performed on a Nikon Ti-U microscope with a 100× oil immersion objective (Nikon Instruments, Melville, NY; NA = 1.49). The excitation source was a Nikon Intensilight, which illuminated the sample uniformly using a liquid light guide. Images were recorded using an Andor iXon 897 (Andor Technology, Belfast, United Kingdom), and were processed with Metamorph software (Molecular Devices, Sunnyvale, CA). See the Supporting Material Section 1.3 for details on image acquisition and processing.

Results

In a typical experiment, a dilute suspension of vesicles displaying DNA was manually pipetted into the solution above a tethered membrane patch displaying the complementary DNA sequence. Individual docking and fusion events were monitored via fluorescence microscopy by observing a lipid dye or a dye-labeled DNA-lipid. We previously observed that DNA-mediated fusion resulted in content transfer across the tethered bilayer and was distinguishable from vesicle rupture above the target bilayer (27), a common outcome observed by groups studying vesicle fusion to SLBs (15). These content transfer events were rare (∼10%), therefore, in this report we focus on understanding the DNA-mediated fusion process from the perspective of the lipids and DNA-lipids. Using these fusion markers, we examined the fusion outcomes and kinetics, dependence on DNA sequence and number density, and behavior of DNA-lipids during fusion, each discussed separately below.

Fusion outcomes and kinetics

We first investigated the various fusion outcomes in lipid mixing experiments, in which we monitored the transfer of a lipid dye from vesicle to target patch (Fig. 1 A). The incoming vesicle displayed a moderately high density of DNA-lipids (65 DNA/vesicle, comparable to the reported number of ∼70 synaptobrevin proteins on an average synaptic vesicle (31), as did the target bilayer (0.5 mol % or ∼2500 DNA/μm² on the top-facing leaflet). The high density on the target ensures that the DNA-lipids will not become depleted during an experiment, in which hundreds of fusion events may occur on an individual patch. We analyzed the docking to fusion waiting times and behavior for hundreds of vesicles and observed four different outcomes following docking, classified as docking-only, hemi-fusion-only, hemi-then-full-fusion, and full-fusion-only—each discussed below. In control experiments, where we used noncomplementary sequences in the two membranes, or where the incoming vesicle contained no DNA-lipids, neither docking nor fusion was observed to occur (data not shown). Furthermore, if the tethering rather than zippering orientation of DNA was used to dock vesicles to the target bilayer, fusion did not occur (see the Supporting Material Section 7).

Docking-only

A substantial proportion of vesicles (∼25–35%) were arrested following docking. These vesicles never transferred any of their lipid dye to the tethered patch during the experiment (Fig. 1 B(i)) and were classified as docking-only events. At least some docking-only events could be observed for hours without change, suggesting a rather permanently arrested docked state of the vesicle (see Fig. S2, b and d).

Hemi-fusion–only

The predominant behavior (∼60–80%) of vesicles following docking was to transfer some, but not all, of their lipid dye to the tethered patch after some waiting time (t_wait, e.g., Fig. 1 B(ii)), and then to retain the remaining dye for the rest of the experiment. These events are classified as hemi-fusion-only. These hemi-fused vesicles appeared to be quite stable and could be observed for minutes and even hours without undergoing any change in fluorescence intensity beyond photobleaching (Fig. S2, a and c).

To understand the docking to hemi-fusion transition, we performed a kinetic analysis of these events. Fig. 2 A shows the docking to hemi-fusion waiting times as a cumulative distribution function (CDF). This CDF could be fit to an exponential function using maximum likelihood estimation (mean waiting time τ_wait = 11.3 s, 75 DNA/vesicle), suggesting that the transition from docking to hemi-fusion occurs in a single step.

Figure 2.

(A) Cumulative distribution function (CDF) of docking to hemi-fusion waiting times (blue) in vesicle-to-tethered membrane lipid mixing experiments (cf. Fig. 1A, 0.3 mol % 5′PolyT DNA-lipids in target membrane), with single exponential model (black line) fit by maximum likelihood estimation (see the Supporting Material Section 6). (B) CDFs across various average numbers of DNA/vesicle (1 to 65 DNA/vesicle, 3′PolyA). Complementary 5′PolyT DNA-lipids in the target membrane were kept constant at 0.5 mol %. See Table 1 for statistical data and Fig. S4 for actual DNA/vesicle distributions.

To further characterize the hemi-fusion-only events, we quantified the percent decrease in fluorescence intensity upon hemi-fusion for hundreds of events (Fig. 1 D). Assuming that the lipid dye is distributed randomly between the inner and outer leaflets of the vesicle, we can estimate the expected mean percent intensity decrease upon hemi-fusion as the ratio of the outer to the total surface area of the vesicle. Using the average diameter of our vesicles (48 ± 12 nm, measured by dynamic light scattering) and an estimated bilayer thickness of 5 nm, we calculated an expected mean percent intensity decrease of 61%.

Our observed mean (∼68%) was near the expected mean, however the distribution was much wider than expected (∼40–90%). Such wide-ranging values should not be possible for vesicles of reasonable size (vesicle diameter would need to be <20 nm to obtain a value of >80%, and values <50% should never be possible for a spherical vesicle). Although it is plausible that the wide distribution is due to nonrandom distribution of the lipid dye in the inner and outer leaflets, NBD-quenching of the lipid dye in the outer leaflet of tethered vesicles shows a narrow distribution of the percent dye in the outer leaflet (Fig. 1 E). Rather, it is possible that some apparent hemi-fusion events are in fact a transient merging of both leaflets, leading to a wider distribution of lipid dye transfer percentages than expected for pure hemi-fusion. Although this is a tentative interpretation, it is also consistent with our previous observation in content transfer experiments that some vesicles release only a portion of their contents during fusion events (27).

Hemi-then-full-fusion

Consistent with a canonical picture of vesicle fusion as proceeding from docking through a hemi-fused intermediate to full fusion (content transfer), we observed events that transferred a portion of their lipid dye to the tethered patch following docking, and then, after a further waiting period, transferred their remaining dye to the patch (Fig. 1 B(iii) and Movie S1). These events are classified as hemi-then-full-fusion events. Interestingly, these hemi-then-full-fusion accounted for <2% of events, even though this might be expected to be the canonical pathway. Because of their rarity, we performed no further analysis of these events.

Full-fusion–only

A small percentage (<5%) of vesicles transferred all their fluorescent dye to the target tethered membrane in one step, after a short waiting period following docking (Fig. 1 B(iv) and Movie S2). These are classified as full-fusion-only events because they appeared to transition directly to full fusion without passing through an observable hemi-fused intermediate.

Both the full-fusion-only and hemi-then-full-fusion events should result in content transfer across the tethered membrane. Together they account for ∼5% of all events observed, consistent with our previous report (27) that DNA-mediated content transfer across a tethered patch was very rare—only ∼10%. That report also showed that lipid-mixing and content transfer occurred simultaneously for our system (see Fig. 3 in (27)).

As with the hemi-fusion-only events, we performed a kinetic analysis of full-fusion-only events (N = 68), although their rarity means that this provides at best only an estimate of the full fusion kinetics. The CDF of the docking to full fusion wait times was also exponential (see Fig. S3), with an average waiting time of 15 s.

Dependence on number of DNA-lipids/vesicle

To examine the dependence of fusion on the number of available binding partners, we varied the number of DNA-lipids on the incoming vesicles from 65 DNA/vesicle to 1 DNA/vesicle (corresponding to 0.25 to 0.004 mol % DNA) while keeping the number density of DNA-lipids on the target tethered patch constant, and characterized the fusion behavior of vesicles at each number density.

Consistent with earlier work that quantified the rate of docking between mobile DNA-tethered vesicles with strands in the tethering orientation (32), we qualitatively observed that docking to the tethered patch (hybridizing in the zippering orientation) became rarer as the number of DNA/vesicle was reduced.

Surprisingly, decreasing the number of DNA/vesicle did not substantially alter the probability of achieving a certain fusion outcome, even at 1 DNA/vesicle. For all numbers of DNA/vesicle, hemi-fusion-only events predominated overwhelmingly, although interestingly we observed that docking-only events decreased to <10% for 1, 5, and 10 DNA/vesicle (data not shown) with a concomitant increase in hemi-fusion events. Even more surprisingly, and in contrast to conclusions inferred from bulk experiments (25,26), decreasing the number of DNA/vesicle did not dramatically alter the kinetics of the docking to hemi-fusion transition (Table 1 and Fig. 2 B). Down to 5 DNA/vesicle, the CDFs showed quite similar kinetic behavior with averages ∼5–7 s. At 1 DNA/vesicle, the CDF was consistently slower, but the increase in average waiting time was not dramatic, increasing only to ∼9 s.

Table 1.

Docking to hemi-fusion wait time statistics

DNA Sequence	Number density	Mean twaita	Nb
Vesicle: 3′Poly A;	1 DNA/vesicle	9.3 ± 1.3 sec	369
	5 DNA/vesicle	5.3 ± 0.7 sec	410
	10 DNA/vesicle	4.8 ± 0.9 sec	248
Target: Poly T	25 DNA/vesicle	6.7 ± 1.1 sec	513
	65 DNA/vesicle	7.2 ± 1.2 sec	531
DNA Sequence	Number Density	Mean twaita	N
Vesicle: 3′ Seq 2;	5 DNA/vesicle	11.5 ± 2.0 sec	159
	10 DNA/vesicle	9.5 ± 2.1 sec	148
Target: Seq 1	25 DNA/vesicle	12.7 ± 2.6 sec	115
	65 DNA/vesicle	12.3 ± 2.1 sec	230

^aErrors for the mean waiting time are 95% CI determined from bootstrap resampling of the docking to hemi-fusion wait times (N_bootstraps = 10,000). The bootstrap distribution of mean waiting times was symmetrical on either side of the mean to within 0.2 s and the higher error estimate was chosen.

^bN is the number of docking to hemi-fusion events in the data set.

In contrast to the fusion kinetics and outcome probabilities, the lateral mobility of the docked vesicles was significantly dependent on the DNA/vesicle number density (see Movie S3). At 65 DNA/vesicle, docked vesicles were mostly immobile; at 10 DNA/vesicle, some docked vesicles were slightly mobile; and at 1 DNA/vesicle, most docked vesicles were highly mobile on the surface of the tethered membrane. Presumably, mobility reflects the number of tethers formed between the vesicle and the tethered patch. However, the increased mobility of vesicles at lower numbers of DNA/vesicle did not hinder the transition to hemi-fusion. After diffusing for some time, mobile vesicles underwent hemi-fusion to the tethered patch and then remained fixed thereafter, presumably because the outer leaflet of the vesicle had merged with the tethered membrane, disallowing further diffusion.

Across several experiments, we observed that the slight pattern in the CDFs in Fig. 2 B for the number densities ranging from 65 to 5 DNA/vesicle appeared to be consistent and not entirely the result of experimental noise. A close inspection of these CDFs reveals that a slow population grows in at 25 and 65 DNA/vesicle, at which number densities the vesicles are immobile. This might indicate that having too many available tethers can actually inhibit the transition to hemi-fusion. A discussion of this, including kinetic models and fits to the data are outlined in the Supporting Material Section 5.

We note that the DNA/vesicle number densities in these experiments are the expected averages and are calculated using the average vesicle diameter (48 ± 12 nm). We confirmed that the actual average number density was near the expected average by using a dye-labeled DNA-lipid, but the width of a typical DNA/vesicle distribution is broad (Fig. S4). These distributions raise a potential issue for the limiting case of 1 DNA/vesicle, as the data may be biased toward the tail of the distribution via selection during docking. In that case, having 1 DNA/vesicle may not be enough to mediate fusion, or may only mediate it very slowly, but that would not be reflected in the data. As an approximate measure, the fusion of vesicles containing 0.25 DNA/vesicle on average, where by Poissonian statistics it would be rare for a vesicle to have >1 DNA, was measured (data not shown). For these vesicles, the docking rate is so low that it became impractical to collect statistical information, but we observed that docked vesicles could undergo hemi-fusion on timescales similar to the nominally 1 DNA/vesicle experiment. This issue is likely to complicate other reports of number dependence, and the best solution would be to work with populations of vesicles purified by the number of DNA (or SNARE)/vesicle (in progress).

Effect of DNA sequence

Previously, we observed that a PolyA/T repeating sequence mediated both faster vesicle-vesicle docking in individual vesicle docking measurements (32) and faster vesicle-vesicle fusion in bulk experiments (25) than a fully overlapping sequence (Sequences 1 and 2 in Table S1). However, the docking study could not measure fusion, and the bulk experiments convolved docking and fusion, therefore it was not possible to discern the effect of DNA sequence on the fusion reaction. To determine how DNA sequence (and consequently the energetics of DNA hybridization) influences fusion behavior following docking, we performed lipid-mixing experiments as previously mentioned, but using Sequence 1 and 3′Sequence 2 (T_m = 68.3°C) as the zippering DNA strands (blue strands in Fig. 1 A) instead of the PolyA/T DNA sequences (Tm=55.3C). These TM values are calculated values for DNA oligomers (not DNA-lipids) in solution using the OligoAnalyzer tool on the www.idtdna.com website ([Na+] = 250 mM, [DNA] = 0.25 M (arbitrarily chosen)). The values are meant only to be a comparison demonstrating a difference in energetics of hybridization and cannot be directly extrapolated to DNA-lipid hybridization where the lipid anchors constrain the DNA to apposing membrane surfaces and the relevant concentration is the number density on each membrane. However, note that if the energy of DNA hybridization influences fusion behavior, DNA duplexes of the same length but with different T_m should exhibit different behavior.

Across four DNA/vesicle number densities (65, 25, 10, and 5 DNA/vesicle), only very moderate changes in kinetics or fusion outcomes were observed between the different DNA sequences. For the fully overlapping sequence, hemi-fusion-only was still the predominant outcome and the docking to hemi-fusion transition still followed exponential kinetics (Fig. S6), with an average wait time of ∼10–12 s for all number densities (Table 1). This indicates that DNA sequence (and the energetics of DNA hybridization) does not strongly influence the fusion behavior of vesicles following docking; implying that the increased rate of fusion observed in bulk experiments (25) reflected a higher rate of docking, consistent with previous measurements of the sequence dependence of docking between individual tethered vesicles (32).

Also consistent with that implication, we observed that docking to the tethered patch was qualitatively much slower for the fully overlapping DNA sequence compared to the PolyA/T pair, presumably the result of increased geometrical constraints placed upon the fully overlapping sequences to initiate contact at the correct sequence location to successfully hybridize. This dramatically decreased docking rate made it impractical to collect data on vesicles containing 1 DNA/vesicle for the fully overlapping sequence, where docking became so rare that meaningful statistics could not be gathered.

Behavior of DNA-lipids during fusion

To gain insight into fusion from the perspective of the DNA-lipids, we investigated DNA hybrid formation subsequent to docking to determine how quickly and how many hybrids form between the vesicle and target membrane. To accomplish this, DNA-lipids labeled at the membrane distal end with a fluorescent dye (typically Alexa546) were incorporated into the target tethered bilayer and the number of dye-labeled DNA-lipids that formed hybrids with a docked vesicle as illustrated in Fig. 3 A was quantified. In these experiments, the DNA-lipid density in the tethered patch is ∼50 DNA-lipids per μm² on the top leaflet. To ensure that this lower density did not cause DNA-lipid diffusion to limit the observed rate of hybrid formation, we performed a simple calculation. The diffusion coefficient of lipids in the tethered membrane is ∼5–6 μm²/s (30). Assuming a similar diffusion coefficient for DNA-lipids in the target membrane, the encounter time of a DNA-lipid diffusing randomly on the patch with a docked vesicle is ∼0.3 ms, and should not be rate-limiting. Likewise, a DNA-lipid diffusing randomly on a 50 nm vesicle will have a mean encounter time of <0.5 ms to reach any given target 5 nm in radius on the vesicle surface (33).

Before vesicle addition, the tethered bilayer has a dim, uniform background due to the rapidly diffusing dye-labeled DNA-lipids in the target patch. Upon vesicle docking (defined as the video frame in which a bright spot is first detected), dye-labeled DNA-lipids gather at the site of the docked vesicle as DNA-lipids on the vesicle hybridize to dye-labeled DNA on the target patch. This produces a local increase in the density of dye-labeled DNA at the site of the docked vesicle, resulting in a bright spot above the background (Fig. 3 B). The background-subtracted intensity of this spot is proportional to the number of DNA hybrids formed, which can be calculated using a calibration curve (see the Supporting Material Section 1.4). Due to the background created by the dye-DNA-lipids in the tethered patch (translated into DNA hybrid effective units, the noise is ∼15 DNA hybrids for a vesicle-sized ROI), this experiment must be performed at high—65 and 42—DNA/vesicle densities. Upon hemi-fusion or full fusion (indistinguishable here), the spot rapidly disappears as the hybridized DNA duplexes diffuse into the membrane patch.

Two control experiments were performed. To verify the simultaneity of lipid transfer and DNA duplex release, we performed an experiment in which we monitored lipid mixing and DNA hybridization together, and observed that the release of the hybridized DNA duplexes into the membrane patch always coincided with a lipid transfer event (Fig. S9). To more directly measure DNA hybridization, we also performed a fluorescence resonance energy transfer experiment using a dye-labeled DNA on the vesicle and on the patch. As expected, the fluorescence resonance energy transfer ratio within a vesicle-sized ROI increased rapidly upon vesicle docking and diminished to background upon fusion (data not shown). However, we opted to use the one color hybridization experiment shown in Fig. 3 A for our analysis because of higher signal to noise.

When measuring how quickly hybrids form between vesicle and target membrane, we observed two behaviors (Figs. 3 B-C). For many vesicles, the intensity of the spot reached its maximum value in the initial frames upon docking, and then remained flat until fusion occurred. This indicates that most DNA hybrids formed rapidly upon docking (<1s), and that the number of hybrids remained constant during t_wait (Fig. 3 B). Other vesicles underwent a rapid rise in intensity during the first frames upon docking, but then the intensity slowly continued to rise for some longer time periods (often 10 s or more after docking, see Fig. 3 C). This suggests that many DNA hybrids are formed upon docking but then additional hybrids form more slowly over the next few seconds. The maximum number of hybrids formed did not seem to be correlated with the time it took to form them, suggesting that this slower accumulation of hybrids was not due to having more available tethers on the vesicle.

We next looked at the maximum number of hybrids formed by many docked vesicles and compared that to the number of DNA-lipids added, on average, to the vesicles (Fig. 3 D and Fig. S11). As expected, the distributions were wide (see the Supporting Material Section 4), but were surprisingly centered near the number added to the vesicles, suggesting that all DNA on the vesicles could hybridize with DNA on the tethered bilayer—as many as 60–70 duplexes on average in Fig. 3 D, with even higher numbers at the tail of the distribution. The minimum contact area between the vesicle and tethered bilayer that would be required to accommodate this many DNA hybrids was estimated by assuming that the duplexes hybridize completely and that they are equally spaced around the perimeter of the contact area. Then, if the width per duplex is 2 nm, a circular contact area of diameter >140 nm/π = 44 nm would minimally be required to accommodate 70 hybrids. Because the average diameter of our vesicles is ∼50 nm, this suggests that the vesicles are greatly deformed and/or that incomplete hybridization (i.e., hybridization only at the membrane distal end) allows a staggered configuration of duplexes that would not require so large a contact area. Atomic force microscopy studies of vesicles adsorbed onto a glass surface have indicated that vesicles can deform quite dramatically upon adhesion, flattening to a width/height ratio of approximately five (34). Furthermore, recent cryo-electron microscopy images of tightly docked vesicles via SNARE complex formation suggest that such deformation may be possible for docked vesicles (Fig. 2 D and Fig. 3 C in (35)). It is also possible that the target membrane deforms locally to increase the contact area with the incoming vesicle, however similar results were obtained with SLBs as the target, a membrane that should be less deformable (data not shown).

Discussion

In this study, we examined vesicle-to-tethered bilayer fusion events mediated by DNA-lipids (Fig. 1 A) as a model system for SNARE-mediated vesicle fusion. In our system, the incoming vesicles dock and fuse to a target membrane (tens of microns in diameter) that is held 8 nm from the surface by 24mer DNA tethers to minimize bilayer-surface interactions. This geometry mimics the curvatures of bilayers in the synapse, where synaptic vesicles (∼40 nm in diameter, (31)) fuse to the locally planar plasma membrane. Using DNA-cholesterol conjugates, others have studied lipid-mixing events between vesicles and planar SLBs (36); using our DNA-lipid conjugates, we also observe lipid-mixing, but not content transfer between vesicles and SLBs (data not shown). Because content transfer is not observed (presumably due to bilayer-surface interactions, see also (15)), we prefer tethered bilayers to SLBs as planar target membranes.

Fig. 4 contains a graphical summary model of DNA-mediated vesicle fusion to a tethered bilayer based on the results described herein. Each state is discussed in detail below.

Figure 4.

Graphical summary model of DNA-mediated fusion to a tethered patch. The locations of dye-labeled lipid molecules and DNA-lipids at each step reflect the understanding derived from the current study. The thickness of the arrows at each step represents the proportion of vesicles observed to undergo the various transitions. The dotted arrows to and from partial fusion represent our incomplete understanding of this transition—we have observed evidence of these transitions, but have not characterized them as a percentage of total events.

DNA hybrid formation and vesicle mobility upon docking

By assaying DNA hybrid formation during vesicle docking and fusion (Fig. 3), we observed that all available DNA-lipids on the vesicle (at least ∼70/vesicle on average) can form hybrids with the target tethered patch. The number of DNA/vesicle (and consequently the number of hybrids formed upon docking) appears to govern the lateral mobility of the vesicle once docked to the tethered patch by DNA hybridization (Movie S3). As the number of DNA/vesicle decreases, vesicle mobility on the surface increases.

Vesicles arrested after docking

A significant proportion of vesicles with 65 DNA/vesicle (∼25–35%, see Fig. 1 C) were arrested after docking. Vesicles arrested after docking have been observed in a variety of SNARE-based fusion model systems, with proportions varying widely from 0% to 90% (6,9,13,16–18). What causes these vesicles to become arrested remains unknown. One possibility is that fusion components (DNA-lipid, synaptobrevin, etc.) may become trapped between vesicle and target membranes, disallowing fusion by keeping the membranes apart. If this were the case, one would expect in our system that lower DNA/vesicle densities should produce fewer vesicles arrested after docking. At these lower densities, the vesicles are highly mobile and any trapped DNA-lipid could presumably easily escape or, in the case of a single hybrid, there would be no obvious mechanism to trap anything. Consistent with that expectation, we observed fewer docking-only vesicles (<10%) at low numbers of DNA/vesicle (data not shown). Lipid composition and vesicle curvature may also have a role in arresting vesicles after docking—these are discussed further in sections below.

Multiple fusion outcomes following docking

Several fusion outcomes were observed and are proportionately represented by the thickness of the arrows in Fig. 4. Most vesicles (thick black arrow) underwent hemi-fusion, where they were able to remain stable for minutes to hours (Fig. S2). <2% of hemi-fused vesicles (thinnest black arrow) later transitioned to full fusion and <5% (medium black arrow) of vesicles transitioned directly from docking to full fusion without passing through a measurable hemi-fusion intermediate. Based on the range of values observed for the percent intensity decrease upon lipid-mixing (Fig. 1 D), we inferred that there are also some partial fusion events in which both leaflets exchange momentarily with the target membrane—consistent with reports of transient fusion pore opening in SNARE-based systems (9,37,38) as well as partial content transfer events we observed previously (27). The exact proportion of such events, presumed to be small, is unknown (dotted lines). Surprisingly, the division among the various fusion pathways was not dramatically influenced by the DNA/vesicle number density or by the DNA sequence. Once the two membranes are brought close enough together, the various fusion behaviors emerge spontaneously, with hemi-fusion predominating in all cases. A very rough estimate of that distance, based on the width of a DNA duplex and linkage to the lipid anchor, is ∼2–3 nm, although this distance could be even smaller should the duplex be pulled out of the way.

Other researchers have suggested that vesicle curvature may play an important role in determining fusion intermediates, kinetics, and outcomes (e.g. (39)). However, we did not observe a correlation between the initial intensities of the vesicles and the division among the fusion pathways or kinetics (data not shown), suggesting that the role of vesicle size and consequently curvature is not the major deciding factor in our data set, which only examined vesicles from a limited size distribution (Fig. S1). This is consistent with our observation that DNA-mediated fusion between mobile vesicles tethered to an SLB occurred on a similar timescale to fusion in our vesicle-to-tethered patch system (40).

Hemi-fusion kinetics and mechanistic implications

As the predominant outcome in our system, the hemi-fused state is of particular interest. Based on experiments monitoring lipid mixing (Fig. 1 D) and those monitoring dye-labeled DNA-lipids (Fig. 3 and Section 9 in the Supporting Material), we infer that this intermediate has fully exchanged its outer leaflet, including any DNA-lipids or hybrids, with the target membrane, as drawn in Fig. 4.

We observed that the CDFs for hemi-fusion wait times were exponential (Fig. 2), with k on the order of 0.1 s⁻¹. The CDF can reveal information on the number of steps before fusion. If there are one or more steps before fusion (and if they occur on dissimilar timescales), the waiting times will follow a gamma distribution and the CDF will have an initial lag before rising. This has been observed in both SNARE-mediated and virus-mediated fusion of vesicles to planar bilayers, where the rate-limiting steps were attributed to the formation of more fusion pairs during the waiting time (6,41). A gamma distribution did not fit our CDFs at all, indicating that hybridization of DNA-lipids may be occurring much faster than protein binding in SNARE- or virus-based systems. In our system, the waiting time CDFs were always exponential, implying a stochastic process from docking to fusion.

Consistent with the implication that forming a certain number of DNA hybrids is not a rate-limiting step, the hemi-fusion rate was very similar across a wide range of DNA/vesicle and for different DNA sequences (Table 1). Vesicles with >1 DNA/vesicle underwent slightly faster hemi-fusion, with a possible minimum near 10 DNA/vesicle, but the difference was not dramatic, especially considering that docked vesicles with low DNA/vesicle densities were quite mobile—suggesting they were not held as closely to the target membrane as those with higher DNA/vesicle densities. Furthermore, the observation that many DNA hybrids could form shortly after vesicle docking and then remain constant in number until fusion occurred (Fig. 3 B) also suggests that DNA hybrid formation is not directly coupled to the fusion event. The lack of strong dependence of the rate of hemi-fusion on the sequence or number of DNA hybrids, in combination with the seconds timescale between docking due to hybridization and (hemi)fusion, suggests that the energy of DNA hybridization is not actively transduced into membrane fusion (hemi- or full) as any energy gained in the binding event is dissipated prior to the fusion event.

What, then, is the nature of the waiting time between docking and fusion? To explain slow SNARE-mediated vesicle-to-SLB fusion events on the order of seconds, others have suggested that local lipid tail fluctuations could initiate fusion between apposing membranes (12). In that case, the waiting time between docking and hemi/full fusion would be the time until a local lipid fluctuation occurred to initiate fusion. This mechanism, where lipid tails infrequently sample the polar interface leading to nucleation of a fusion pore, has been demonstrated in simulations, although only for highly constrained, curved vesicles on timescales much shorter than our experiments (42,43). This stochastic process should produce an exponential distribution of waiting times, consistent with our observations, but is difficult to verify experimentally.

Vesicles arrested after hemi-fusion

Vesicles arrested at hemi-fusion have been observed in other model systems, although typically at lower percentages than observed here (9,13). One explanation for why hemi-fusion is the predominant endpoint of our system is that the lipid anchor of our DNA-lipids spans only one leaflet of the vesicle bilayer. The importance of the SNARE transmembrane domains for promoting full fusion has been suggested by a number of reports (10,14,44–47), although one study showed fusion could be achieved with the addition of several accessory proteins even if the SNARE transmembrane domain was replaced with an anchor spanning only one leaflet (44). Because our DNA-lipids only span half the bilayer, all hybridized pairs can diffuse into the target bilayer upon hemi-fusion, leaving the hemi-fused vesicle without any DNA-lipids as depicted in Fig. 4. This could explain why the hemi-fused state is the predominant endpoint in our system.

Lipid composition may also have a role to play in arresting the vesicles at hemi-fusion (or at docking). The particular lipid composition used here (2:1:1 DOPC/DOPE/Chol) was chosen based on other model membrane fusion systems (21) and because we observed the greatest extent of DNA-mediated lipid and content mixing in bulk experiments, while ensuring no leakage of vesicle contents or nonspecific fusion (25). However, this composition may stabilize the hemi-fused intermediate so that transition to full fusion becomes difficult. Indeed, other researchers have demonstrated that lipids with negative curvature (such as PE) can stabilize the hemi-fused intermediate (9,13,48), and removal of negatively curved lipids can in some cases lead to an increased propensity for full fusion events (13). PE is, however, a significant component of the synaptic vesicle lipid composition (31) suggesting that the biological fusion machinery must be able to overcome or avoid any hemi-fusion intermediates stabilized by lipid composition.

Implications for biological fusion

The conclusions emerging from our study of DNA-mediated fusion have several implications for the biological fusion reaction and lend support to several mechanistic proposals.

One, our data suggest there are multiple fusion pathways that can be accessed merely by bringing the vesicle and target membrane into close apposition upon DNA binding in the zippering orientation, and the kinetics of these processes imply a stochastic mechanism of fusion. This implies that the role of the biological fusion machinery is not just to encourage progression along a single trajectory toward content release, but rather to select from among several possible fusion pathways so that fusion will occur both properly and at the correct time, as recently proposed (5). In combination with our content mixing data (27) this also supports the idea that in biological fusion an inhibitory mechanism may exist in which docked vesicles are prevented from fusing until the correct moment (3). This inhibitory state could be achieved by holding the membranes and SNAREs a sufficient distance apart so they do not zipper until the correct moment, or by physically blocking the membranes from coming into close proximity following SNARE zippering or partial zippering.

Two, our data support the emerging proposals that very few fusion-mediating complexes (DNA, SNAREs, etc.), possibly even just one, may be required to achieve the membrane proximity that allows fusion to occur (6,8,14), although in the biological system it is likely that other fusion components are necessary to encourage the correct fusion pathway at the right time. In our data, we observed that fusion behavior and kinetics were relatively unchanged down to one DNA/vesicle on average, suggesting that additional complexes do not play a significant role in driving fusion.

Three, our results suggest that the hemi-fused intermediate is actually quite stable and can be a predominant endpoint. The hemi-fused intermediate has been observed in pure lipidic systems (49), and recent data suggests that hemi-fusion may produce a kinetically trapped fusion state for synaptic vesicle fusion as well (5,35), indicating that a stable hemi-fused structure can occur across a variety of model systems.

Four, in contrast to mechanistic hypotheses that propose the zippering of the SNAREs is directly transduced into membrane fusion (e.g. (45)), our data show a lag time between DNA zippering and membrane fusion and suggests a stochastic mechanism. This may be a limitation of not having a transmembrane anchor for our DNA-lipids or it could suggest an alternate fusion mechanism.

Finally, an exponential distribution of wait times with an average of several seconds might be acceptable for some biological fusion pathways, but it is clearly too slow for the exquisite timing needed for synaptic vesicle fusion (1). Indeed, slower-than-expected fusion has been a problem in nearly all in vitro membrane fusion systems to date and a variety of accessory proteins have been suggested to be the principal timing mechanisms (50). Ultimately, this highlights the importance of further experimentation to understand the molecular mechanism of vesicle fusion.