Freeze-thaw cycles induce content exchange between cell-sized lipid vesicles

Our method for generating GUVs is based on emulsion transfer in microtiter plates [29, 30]. In contrast to common protocols, we both generated and imaged GUVs in glass bottom microtiter plates, enabling high-throughput parameter optimization. Since the same 96-well plate is used for all experimental steps including GUV generation, freeze-thawing and subsequent imaging, pipetting steps are minimized.

Zoom In Zoom Out Reset image size

First, POPC (1-Palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine, Avanti Polar Lipids, Inc.) was dissolved in chloroform at a concentration of 8 mM. The solution (75 μl) was then transferred to a 4 ml glass vial and the chloroform was evaporated under argon flow (15 min), followed by removing residual solvent traces by applying a vacuum (1 h). After addition of 1.5 ml mineral oil (Sigma-Aldrich, M5904), the vial was sonicated for 1 h at 40 °C, and incubated overnight at room temperature to dissolve the lipids in the oil.

The glass bottom microtiter plates (Greiner Bio-One, 96-well glass bottom SensoPlate™) were passivated with Pluronic F-127 (Sigma-Aldrich): after plasma-cleaning the microtiter plate, 50 μl of 20 mg ml⁻¹ Pluronic F-127 were added to each well of the plate intended to hold GUVs and incubated for 30 min at room temperature, followed by a wash with 900 mM glucose solution (50 μl). 100 μl of the same glucose solution was then pipetted into each well (outer aqueous phase containing future GUVs) and POPC-mineral oil solution (40 μl) was layered on top, followed by incubation for 30 min at room temperature to form a monolayer at the water–oil interface (figure 1(a)).

To create two distinct fluorescently labelled populations of GUVs, we prepared internal GUV solutions (900 mM sucrose) with DNA oligonucleotides (26 bases (13 repetitions of 5'-CA-3'), IDT) either labelled with a 5' Cy5 fluorophore (4 μM) or a 5' 6-FAM fluorophore (8 μM).

To generate the water-in-oil emulsion for the emulsion transfer, 5 μl of each inner phase solution (each containing one of the two oligonucleotides) was added to 250 μl of the POPC-mineral oil solution in a 0.5 ml micro test tube (figure 1(b)). Droplets were formed by shear forces through mechanical agitation (i.e. repeatedly rubbing each tube for about 1 min across the array of cavities on a micro test tube rack) (figure 1(c)), then 40 μl of both emulsions (80 μl in total) were immediately pipetted into each well of the microtiter plate (figure 1(d)). The microtiter plate was subsequently centrifuged for 3 min at 250 g (Eppendorf Centrifuge 5804 R) in a rotor designed to hold microtiter plates (Eppendorf A-2-DWP) to generate GUVs (figure 1(e)).

All steps that involve handling of the hygroscopic mineral oil, except for the last centrifugation step, are carried out in a glove box under nitrogen flow and a humidity of 3%–8% to minimize the amount of water in the organic phase. While this measure generally increased vesicle yields drastically, it proved to be crucial in warm summer months with indoor humidities of more than 50%, when we sometimes did not see any vesicle formation if samples were handled at ambient conditions.

For the experiments in figures 4, S6 and S7 membrane labels were used in the following mole fractions of membrane label to lipids: 0.013 mol% DOPE-ATTO488 (1, 2-dioleoyl-sn-glycero-3-phosphoethanolamine labelled with ATTO488) (ATTO-TEC GmbH); 0.017 mol% DiD (DiIC₁₈(5)) and DiI (DiIC₁₈(3)) (Thermo Fisher Scientific). The labelled dextrans (3000 MW and 40 000 MW) were both used in a concentration of 0.08 mg ml⁻¹.

Three different FT protocols were used, enabling fast freezing and fast thawing (condition A), slow freezing and slow thawing (condition B) and fast freezing followed by slow thawing (condition C). We only observed very few intact vesicles after slow freezing and fast thawing (condition D; figure S3), and therefore did not include this condition in our further investigations and analysis.

Prior to freezing, the microtiter plates containing GUVs were sealed with a sealing film (Carl Roth Rotilabo^®-sealing film for micro test plates) and were positioned at a slight angle for 5–10 min, to allow the GUVs to accumulate on one side of the well.

For condition A, a 10 cm × 6 cm × 6 cm block made of stainless steel was placed in a −80 °C freezer. After the metal block reached the temperature of the freezer, the microtiter plate was placed on top and pressed firmly against it for about 1 min. To allow direct contact with the glass bottom, the metal block must be smaller than the bottom frame of the microtiter plate. After 30 min, the plate and the metal block were taken out of the freezer. The plate was removed from the first metal block and placed onto a second block preheated to 60 °C on a hot plate. After 5 min of thawing at 60 °C, the plate was transferred to the microscope for imaging (see figure 2(b)).

Zoom In Zoom Out Reset image size

For condition B (figure 2(c)), the microtiter plate was placed into a −20 °C freezer at a slight angle for 30 min, and subsequently into a −80 °C freezer for another 30 min. Afterwards the GUVs were thawed at room temperature for 30 min.

Condition C is a combination of the previous two. The GUVs were frozen for 30 min at −80 °C on a metal block and subsequently thawed for 30 min at room temperature (see figure 2(d)).

Imaging was performed with an LSM 780/CC3 confocal microscope (Carl Zeiss, Germany) equipped with a C-Apochromat, 40x/1.2 W objective. We used PMT detectors (integration mode) to detect fluorescence emission (excitation at 488 nm for 6-FAM and 633 nm for Cy5) and record confocal images. To capture many GUVs in high resolution, tile scans were performed by recording a defined number of adjoining single images of the sample (the 'tiles'), moving the sample using motorized stages with high precision. Following the scanning process, the mosaic image is assembled using the stitching algorithms provided by the Zeiss software.

Fluorescence images were analysed using custom software written in MATLAB (MATLAB R2014b, The MathWorks, Inc.) adapted from Ranasinghe et al [31]. Briefly, vesicles were detected (i.e. regions of interest identified) in each field of view across two images sequentially acquired for each colour channel ('red' and 'green') by converting both images to a binary image (mask) using a threshold intensity set to be the mean single pixel intensity of the bandpass-filtered images plus 0.5–2 times this value. This factor was determined empirically by visual inspection of the binary image obtained and comparison with the original data. In the next step, objects (vesicles) are detected by tracing continuous regions of ones in the binary image. Signals are assumed to correspond to single vesicles if the area of the region is >10.8 μm² (250 pixel). The mean intensity for the centres of each region/vesicle (i.e. the originally detected region minus a shell of 5 pixel) are then calculated from both original images for vesicles detected in either of the two channels. Finally, the intensities are plotted in scatter plots coloured according to data density (figures 2(i)–(l)) and the ratio of red/green intensity was calculated for each vesicle.

MATLAB code for analysing image data is available online.

To study the effect of FT-cycling on vesicle content mixing, we first developed a method to simultaneously generate GUVs of differing content via emulsion transfer in microtiter plates, thus greatly improving experimental throughput and parallelizing the screening for optimal conditions. Briefly, two distinct populations of GUVs were obtained by (i) preparing water-in-oil droplets from two solutions, each containing a different fluorescently labelled oligonucleotide. These two emulsions were then (ii) mixed and (iii) passed through a lipid-lined oil–water interface upon which GUVs were formed (figure 1).

We found the surface passivation of the microtiter plate chambers to be a crucial factor in our experiments. We tested β-casein (2 and 5 mg ml⁻¹), PLL-g-PEG (0.5 and 2 mg ml⁻¹) and Pluronic F-127 (1–50 mg ml⁻¹) as passivation agents in combination with various subsequent washing protocols. Vesicle generation in β-casein-passivated chambers was successful, but only very few vesicles remained intact after the FT cycles. PLL-g-PEG seemed promising while working with vesicles that did not contain oligonucleotides, but after incorporating labelled oligonucleotides, the experiments failed. Pluronic F-127 at high concentrations (10–50 mg ml⁻¹) was the only passivating agent to work under relevant conditions. In contrast to the other passivation methods, we also observed the highest overall vesicle yield and no aggregation on the glass surface.

With our optimized microtiter plate protocol in hand, we investigated how the yield of content mixing upon freeze-thawing vesicles depended on the heating and cooling rates of the FT process. We tested three different FT protocols with variations in heating and cooling rates. Firstly, the sample was rapidly frozen at −80 °C and rapidly thawed at 60 °C, in both cases maximizing heat exchange by placing the plate on a metal surface (condition A: fast freezing/fast thawing). Secondly, the sample was cooled slowly to −20 °C before cooling it further to −80 °C, then slowly thawed at room temperature (condition B: slow freezing/slow thawing). Thirdly, we combined rapid freezing (step 1 from condition A) with slow thawing (step 2 from condition B) such that fast freezing is followed by slow thawing (condition C). The remaining combination (condition D: slow freezing/fast thawing) is not covered in our detailed analysis, since it resulted in almost no vesicles after the FT process (figure S3).

We applied these FT-cycles to microtiter plates containing mixtures of GUVs encapsulating 26 bp DNA oligos either labelled with Cy5 or with 6-FAM, and recorded two-colour images of the GUVs by confocal microscopy, both before (figure 2(e)) and after the FT cycle (figures 2(f)–(h)). We then used an automated, custom image analysis routine to extract the fluorescence intensities for individual GUVs from the image data.

For conditions A–C, we observed that the majority of GUVs contained both DNAs after a FT cycle, whereas before freezing, two distinct populations of vesicles could clearly be distinguished (figures 2(i)–(l)). The efficiency of this mixing of contents varied between the different FT protocols. This content mixing is accompanied by an overall decrease in GUV fluorescence intensity in both colour channels (figures 2(i)–(l), figure S1).

Under condition A (fast/fast), the ratios of green to red intensities were shifted compared to the control for many GUVs, indicating content mixing. However, two populations of GUVs which either contain more Cy5- or 6-FAM-DNA can still be distinguished (figures 2(j) and (n)), indicating that the mixing is incomplete. Only a small population of GUVs shows approximately equal intensities for both dyes, and hence contains approximately equal amounts of oligonucleotides (figures 2(f), (j) and (n)).

Similarly, condition B (slow/slow) resulted in the majority of GUVs containing a highly uneven ratio of fluorescence signals, and thus an uneven ratio of DNA concentrations (figures 2(k) and (o)). Compared to condition A, however, the overall intensity of these vesicles (and hence the concentration of their contents) was lower (figure S1).

Under condition C (fast/slow), content mixing was by far the most efficient. Most GUVs displayed approximately equal intensity levels for both dye-labelled DNAs (figures 2(l) and (p)), in stark contrast to the first two conditions. In addition, the combined fluorescence intensities from both DNAs contained within the GUVs were similar to those measured for GUVs subjected to condition A (figure S1). Protocol C did not seem to change vesicles numbers noticeably when we visually inspected the data, whereas GUV numbers decreased for protocols A, B and D with almost no intact GUVs remaining for condition D. Thus, condition C not only results in the most complete content mixing but also retains the most GUVs and the most content (labelled oligos) within those GUVs.

We note that after a FT cycle, the GUV size distribution is shifted to slightly smaller vesicle sizes for all tested conditions (figure S2), indicating that the FT cycle induced some degree of vesicle fragmentation.

While fast freezing and slow thawing of vesicles encapsulating differently labelled DNA oligos leads to efficient content mixing, the overall concentration of encapsulated DNA decreased after mixing. We thus investigated whether this noticeable content loss is due to exchange with the surrounding phase during the FT step. To quantify the extent of this exchange, we freeze-thawed GUVs containing 6-FAM-labelled oligos, which were surrounded by an outer phase containing Cy5-labelled oligos (figure 3(a)). While only background-level Cy5 signals were detected inside vesicles before freeze-thawing, the level increases notably after freeze-thawing for the tested conditions (figures 3(b) and (c)). Although this suggests some internalization of the outer phase during the FT process, the intensity of Cy5 measured inside GUVs after FT was still much lower than that measured in the outer phase (on average 18% of outer phase), whereas the average intensity of the encapsulated DNA dropped to ∼43% of the initial value for FT condition C (fast freezing, slow thawing) (figure 3(b)). Increased uptake of surrounding fluid is observed for condition B (figure S5(b)), which is consistent with the larger decrease of fluorescence intensity observed for content mixing between GUVs under this condition (figure 2(k)).

Zoom In Zoom Out Reset image size

If content exchange between GUVs were to happen via content release and uptake (and diffusion of the content through the GUV-surrounding phase), we would expect that content exchange is more efficient at high GUV densities, as dilution of the content would be minimized. Conveniently, our FT-protocol involves tilting of the microtiter plates to an angle of 45° before freeze-thawing, which results in a GUV density gradient across the well (figure S5(a)). In areas with high GUV density, the immediate surrounding phase constitutes only a small fraction of the total volume. We therefore tested whether GUV content composition after freeze-thawing is dependent on the location of the GUVs within the well, and hence the local GUV density. In line with our hypothesis, figures 3(b) and (d) show that vesicles in high-density areas (left) keep more of their original content (6-FAM, green) and encapsulate less of the surrounding phase (Cy5, red) than vesicles in low-density areas (right). Thus, GUV density determines the degree GUV content loss and encapsulation of the surrounding phase.

To gain further insights into the mechanisms responsible for content exchange between vesicles, and to specifically test whether this exchange could, at least partially, be mediated by fusion and subsequent fission of vesicles, we performed our FT-cycling with two populations of GUVs with different membrane labels. The lumen of one GUV population also contained a fluorescently labelled cargo of various molecular sizes to allow for simultaneous monitoring of size-dependent content mixing. We generated the GUVs according to the protocol shown in figure 1, but used different lipid-in-oil solutions in the different steps to generate two GUV populations: one population with DOPE-ATTO488 in the membrane, but no labelled content; and a second population with DiD-labelled membranes and TMR-labelled 40 kDa dextran cargo (see figure 4(a) for a depiction of the protocol). Importantly, we found that if the time between mixing the two types of emulsions and the centrifugation step is reduced to a minimum, no significant amount of foreign membrane is incorporated into the membrane of the respective other population of GUVs.

Zoom In Zoom Out Reset image size

When we freeze-thawed the GUVs according to protocol C, surprisingly we observed content mixing, while maintaining two populations of distinctly membrane-labelled vesicles (figure 4(b) and (c)). These data suggest that content exchange is possible even for large molecules up to 40 kDa without a significant number of membrane fusion and fission events. Figure S6 shows more images of the experiment. To validate these findings further, we performed a series of two-colour experiments in which we labelled the membrane of at least one population of GUVs with various membrane dyes (DOPE-ATTO488, DiD and DiI) and also used a variety of labelled cargo molecules (dextran 40 000 MW, dextran 3000 MW and the 6-FAM labelled Oligo from previous experiments). A comparison of confocal images before and after freeze-thawing for these experiments is shown in figure S7. In all cases, we saw a homogenization of contents, but could still distinguish the two populations of vesicles by their distinct membrane labels, or lack thereof.

We conclude from these experiments that GUVs mostly retain their membrane identity during the FT process, and that the cause for content exchange between vesicles is not fusion (and subsequent fission) of vesicles.

To investigate vesicle fate throughout the FT process in more detail, we attempted to follow the GUVs during the thawing process by time-lapse microscopy. Although imaging frozen samples throughout the entire thawing process proved unattainable with our experimental equipment, we were able to follow vesicle shape changes after thawing the sample for 20 min at room temperature. At this stage, the content of the complete sample well was fully thawed. However, numerous elongated vesicles could be observed, as well as their fission into smaller spherical GUVs over the following 5 min (figure S8(a)). While we assume that these events are not required for the content exchange between vesicles, the observations explain the shift to smaller vesicles after freeze-thawing (figure S2). Dynamic shape changes like these can be induced by vesicle deflation through osmotic water extraction. To explore whether changes in osmolarity during the FT process are a possible cause, we analysed the osmolarity in horizontal sections from wells that were frozen according to protocol C. We found a gradient of osmolarity with higher concentrations at the bottom of the well (supplementary figure S8(b)) which could be a cause for the GUV destabilization and subsequent fission.