Engineering thermoresponsive phase separated vesicles formed via emulsion phase transfer as a content-release platform

Elucidation of cholesterol insertion efficiency into phase-transfer vesicles enables the rational design of phase-separated membranes as thermally-responsive platforms for artificial cell construction.


Introduction
Giant Unilamellar Vesicles (GUVs) are increasingly being used as functional units in biotechnology, as articial cell chassis, as cell-mimics for the study of cellular processes, and as model membranes to study membrane biophysics. [1][2][3][4][5][6]35 Comprised of a lipid bilayer encasing a small uid volume, vesicles offer a biomolecular fabric for the insertion of membrane proteins while also providing a microcompartment for encapsulating material cargo. Part of their appeal is the rich phase behaviour they possess, which can be taken advantage of to add functionality, for example enabling the creation of smart and responsive systems. However, taking advantage of a membrane's inherent structural and biophysical properties is predicated on achieving accurate control of lipid composition.
A major enabler in the construction of functional GUVs has been the development of emulsion phase transfer (EPT) methods for their production, 5,7-10 where water-in-oil droplets are used as templates around which a bilayer is assembled. This method is attractive as unlike traditional methods (such as electroformation) it allows the encapsulation of large, charged chemical species (a category which many biomolecules fall under), requires no specialised equipment, is amenable to translation into high-throughput microuidic formats, 6,8,[11][12][13] and is uniquely capable of producing asymmetric GUVs. 14,15 However, there is an open question surrounding the degree of incorporation of different amphiphiles into membranes formed by EPT, an issue which must be addressed if compositiondependent features are to be utilised, for example in the construction of vesicles with controlled surface structure.
One widely-studied membrane phenomenon is the lateral heterogeneity of membranes containing multiple lipid components. These membranes can separate into coexisting phases (or domains) that possess different degrees of molecular order and have distinct compositions. 16 Single component membranes show a transition between a gel phase (L b ) and uid phase (L a ) at a dened melting temperature (T m ), a process that is associated with increased membrane permeability due to the formation of membrane defects. 17,18 In two component mixtures of high-and low-melting point lipids, these phases can coexist. When cholesterol is added to form ternary mixtures, coexisting liquid disordered (L d ) and liquid ordered (L o ) domains can form. The L o phase is enriched in cholesterol and saturated lipid (e.g. DPPC) and a L d phase composed of mostly unsaturated lipid (e.g. DOPC). [19][20][21] While both the L o and L d phases exhibit rapid lateral diffusion within the plane of the membrane, the lipid chains within the L o phase have a signicantly higher degree of order than those in the L d phase. Transition from a phase separated to a mixed state can be achieved by heating above a characteristic mixing temperature (T mix ) for a particular lipid composition (which can differ from the melting temperature (T m ) of individual lipids). 22 Despite being extensively studied in GUVs formed via electroformation, there remains a scarcity of literature on phase separation in GUVs constructed from water-in-oil emulsions. Hamada et al. reported the use of the EPT technique to prepare asymmetric GUVs with microdomains. 23 However, the work merely conrmed domain formation was possible and did not probe any further. A droplet-shooting and size-ltration (DSSF) method also showed that phase separation was possible, but did not undertake a comprehensive investigation. 24 Finally, a vesicle formation technique, 'continuous droplet interface crossing encapsulation' (cDICE) has recently been investigated for its potential to incorporate charged lipids and sterols. 25 Although cDICE tolerated charged lipids well, the method was highly inefficient at incorporating cholesterol (<1% efficiency), so L o /L d domains could not be formed.
Further investigations of the phase behaviour of GUVs formed via the EPT technique are needed, specically regarding the extent to which the concentration of lipids during preparation matches that in the nal GUVs. Resolving this issue is critical to assessing the suitability of this technique for unravelling fundamental principles in membrane biophysics and generating responsive micro-containers. Importantly, adding the rich membrane phase behaviour exhibited by lipid systems to the toolkit at our disposal to introduce functionality to vesicles formed via EPT will be vital in developing targeted, active structures. This is key given that GUVs generated via this method underpin many applications of GUVs as microreactors, articial cells, responsive so microsystems, and as tools in biotechnology. 2,[26][27][28]36 Here we characterise the cholesterol insertion efficiency of GUVs formed via EPT for the rst time by studying the phase behaviour of ternary lipid bilayers composed of different ratios of DOPC/DPPC/cholesterol. We compare the observed phase behaviour to previous studies using electroformation to suggest a 28-50% cholesterol incorporation efficiency, but still show that the expected membrane structures can be observed using EPT GUVs by compensating for the cholesterol loss. We demonstrate that heating different ternary vesicle compositions through the T mix , either globally using a heating stage, or locally using through a laser-assisted heating setup, leads to release of encapsulated uorophore and peptide cargo. Finally, we show that release temperature can be tuned by the user through rational design of vesicle composition. In this way we couple the encapsulation advantages associated with EPT to the phase behaviour of multi-component membranes, to engineer structurally controlled, thermoresponsive vesicles which release cargo when domains are mixed.

Results and discussion
The incorporation efficiency of cholesterol in EPT vesicles GUVs were formed using the centrifuge-driven EPT technique, where lipids are incorporated into the oil phase (Fig. 1A). 28 12 different types of GUVs were made from oils containing DOPC/ DPPC (1 : 1 molar ratio) and increasing amounts of cholesterol (0-90 mol%). Incorporation of rhodamine-PE (at 1 mol% of total phospholipid concentration), which partitions into the DOPC-rich disordered phases, allowed visualization of L d domains via uorescence microscopy. Within these samples we observed the full range of phase behaviour observed previously in electroformed vesicles ( Fig. 1B-D).
We investigated whether EPT GUVs exhibited similar phase behaviour as those prepared by electroformation, based on results obtained in previous studies. 21 Typically, at RT electroformed vesicles with equal amounts of DOPC and DPPC phase separate into non-circular gel phase domains within a uid continuous phase when the cholesterol mole fraction is less than 10 mol%. Between 10-45 mol% of cholesterol, L o /L d coexistence is seen (characterised by the presence of circular uid domains). NMR has revealed there may also be threephase coexistence regions, although these were not observed with uorescence microscopy. 20 Above the 45 mol% cholesterol threshold, a single uniform liquid phase is present. This behaviour is summarised in the phase diagram in Fig. 1B and graph in Fig. 1C.
In our experiments (indicated as coloured circles and boxes Fig. 1B-D) we obtained gel/liquid phase separation in vesicles where 0-35 mol% cholesterol was present in the initial lipid-in-oil mixture. From 37-80 mol%, each vesicle composition displayed circular domains characteristic of L o /L d phase separation. Finally, at 90 mol% cholesterol, a single liquid phase was obtained. We therefore estimate the gel/liquid to L o /L d transition occurs between 35-37 mol% cholesterol, and the L o /L d to single liquid transition occurring between 80-90 mol% cholesterol in the originally dissolved lipid mixture. Comparing these transitions to those obtained with electroformation gives us a cholesterol incorporation efficiency of 28% in the gel/liquid to L o /L d transition region and 44-50% at the L o /L d to single liquid transition region. As with electroformed GUVs, we found that increasing cholesterol concentration was associated with an increasing area fraction of dark domains in our EPT GUVs. 29 Whilst there is clearly reduced cholesterol insertion (relative to the phospholipid insertion) in EPT vesicles compared to electroformed vesicles, the incorporation percentages reported here are signicantly higher than the <1 mol% reported by Blosser et al. using the cDICE method. 25 In addition, the EPT GUVs showed very low compositional dispersity (ESI Fig. SI1 †), with more than 95% of vesicles generated displaying the same phase behaviour as the rest of the set (a minimum of 100 GUVs for each composition observed). This suggests that the membrane composition of EPT GUVs can be accurately controlled simply by compensating for the relative partitioning of cholesterol by adding an increased concentration to the initial lipid mixture.
To ensure that the absolute concentration of cholesterol in oil doesn't affect GUV formation, 1 : 1 : 3 DOPC : DPPC : Chol vesicles were produced from four different lipid-in-oil solutions ranging from 0.2 mg ml À1 to 80 mg ml À1 (ESI Fig. S2 †). We observed L o /L d domains in each case, despite there being a 400fold difference in absolute cholesterol level. This demonstrates that in the concentration regime we use, absolute cholesterol concentration does not have an effect on nal GUV composition. We note that below above 80 mg ml À1 lipid did not completely dissolve in the oil phase, and below 0.2 mg ml À1 vesicle yield was signicantly compromised, hence these being the upper and lower bounds of the concentrations tested. The  32 The lipid composition shown are those of the initial lipid film, and not necessarily of the GUVs themselves. (C) A comparison of ternary GUV phase behaviour between EPT (left) and electroformed vesicles (right). The observed phase behaviour is shown as a function of the mol% cholesterol in the initial lipid film. Although the same phase behaviours are observed in each case, due to reduced cholesterol incorporation, increased cholesterol is needed to reach key phase changes ($36 vs. 10%, and $80-90 vs. 40% respectively). (D) Selected fluorescence microscopy images of GUVs with increasing mol% cholesterol. Each composition contains 1 mol% rhodamine-PE in order to visualise domains. In all images, the dark regions correspond to the more ordered phase (L o or gel) and the light regions correspond to the L d phase. Scale bar ¼ 5 mm, experiments conducted at RT. ability to generate GUVs from such low lipid concentrations bodes well for wider adoption of the phase transfer technique in downstream applications.
The cholesterol incorporation observed here is likely to be the result of cholesterol's low amphipathic character. Unlike phospholipids, which possess both a polar headgroup and lipophilic acyl chains, cholesterol is predominantly a hydrophobic molecule, which possesses only a minimal polar hydroxyl head group. This leads to an increased propensity for the cholesterol to remain dissolved in the oil phase as opposed to partitioning to the oil-water interface.

Thermally-triggered release from phase-separated GUVs
To demonstrate the utility of EPT generated phase-separated GUVs as temperature sensitive microcontainers, we developed an assay where 1 : 1 : 3 (molar ratio) DOPC/DPPC/cholesterol GUVs were loaded with a uorescent dye (calcein, which has an emission wavelength sufficiently spectrally separated from the rhodamine lipid to image independently). These GUVs were subjected to four temperature cycles through T mix (>32 C) and back down to 21 C using a heated microscope stage and cargo release monitored (Fig. 2). This particular lipid composition was chosen as it is well studied, 21 and has phase transition signicantly above room temperature, so phase transition would only occur when actively heated.
Aer one cycling event, 25% of GUVs were found to have released their entire internal dye cargo. Through subsequent cycles, this proportion could be increased, with 51% of GUVs lacking calcein aer four cycles ( Fig. 2A). As a control, leakage through GUVs composed of DOPC (not phase separated) showed an average leakage of only 14% aer 4 cycles. These statistics were obtained from 5 runs, with a minimum of 40 vesicles analyzed in each run. When vesicles leaked, they also lost contrast when imaged by phase contrast as the internal and external solutions equilibrated (Fig. 2B).
To conrm that complete calcein release can be obtained from ternary GUVs, we applied 20 heat cycles to ternary and DOPC vesicles ( Fig. 2C and D). The ternary vesicle population displayed almost complete calcein release (>95% vesicles showing release), whilst $60% of DOPC GUVs still contained calcein, conrming the robustness of the release mechanism.
We note that in this case vesicles were incubated in 0.5 M sucrose during the thermal cycle, and subsequently transferred to 0.5 M glucose solution. For this reason, they were visible under phase contrast microscopy. Complete content release from a ternary vesicle population as shown here gives another indication to the homogeneity of the ternary vesicles formed through EPT.
Our data suggests that the bilayer leakage depends on the GUVs undergoing a phase transition, which we attribute to the biophysical phenomena of packing defects at domain interfaces. 30 As the T mix of this phase transition in ternary systems can be controlled through lipid composition, we also produced calcein-loaded 1 : 1 : 3 DOPC : Egg Sphingomyelin (EggSM) : -Chol vesicles containing L o /L d domains as shown in Fig. 2D. As the T mix for this composition is ca. 45 C, release should occur only when heating beyond this temperature. 21 We therefore heat cycled both 1 : 1 : 3 DOPC : DPPC : Chol and DOPC : EggSM : Chol vesicles to 30 C, 40 C, and 50 C, and monitored calcein release (Fig. 2E). As expected, the DOPC : DPPC : Chol vesicles leaked at both 40 C and 50 C, whereas calcein release from DOPC : EggSM : Chol vesicles only occurred aer heating to 50 C, further indicating that the T mix of the ternary system controls content release. These results also highlight how the thermoresponsive property of ternary GUVs can be tuned by the user for use in triggered release applications.
Aer demonstrating the potential to tune release of a dye, we then exploited the ability of GUVs to encapsulate biomolecules. We used the model peptide biomolecule, MOCAc-RPKPVEN-vaWRK(Dnp)-NH 2 (NFF-3). NFF-3 contains an N-terminal 7methoxycoumarin uorophore quenched via resonance transfer by a C-terminal 2,4-dinitrophenyl group. 37 As the peptide contains a trypsin cleavage site between the glutamic acid and norvaline residues, proteolysis of NFF-3 results in liberation of uorescence, which can be monitored spectroscopically (ESI Fig. S3A †). We used the uorogenic nature of NFF-3 to design a peptide release assay, where the leakage of peptide can be monitored through adding trypsin to the external solution (ESI Fig. S3B †). One heat cycle was applied to peptide-loaded 1 : 1 : 3 DOPC : DPPC : Chol and DOPC vesicles. Trypsin was then added to external solution, and the change in uorescence quantied against untreated vesicle controls. As shown in ESI Fig. S2C, † appreciable peptide release from ternary vesicles was observed (22%, S.D. ¼ 7%, n ¼ 9), whilst negligible release occurred from DOPC vesicles (2.5%, S.D. ¼ 4%, n ¼ 4).
One important property of the release system that is still to be characterised is the size limit of the transient pores generated in ternary vesicles. Although packing defects in dimyristoylphosphatidylcholine vesicles with an effective molecular weight cut-off of 900 Da have been recorded previously, 30 here we show release of peptide molecules with a weight of 1674 Da. This release may be due to partial unfolding of the linear polypeptide chain of NFF-3 during cargo release (resulting in smaller effective molecular weight), or alternatively, indicate that the packing defects generated in ternary systems possess larger sizes compared to gel-uid transitions observed in single lipid systems.
Using heat to trigger leakage in L o /L d ternary vesicles represents a proof-of-concept that could be applied to a host of applications involving the controlled release of material; however, one drawback to the methods employed above is that the response of bulk GUV populations was monitored, as opposed to individual GUVs.

Phase changes and cargo release in individual GUVs through laser-induced local heating
In order to further investigate the phenomenon of content release, we induced phase changes in selected vesicles by employing a tightly focused laser beam from an optical tweezers set-up to locally heat individual GUVs. This was achieved by using a single-beam optical trap, which locally heats each targeted vesicle. We applied a 1.1 W (at trap) infrared laser to individual selected phase separated GUV. This leads to an increase in temperature which was sufficient to raise the GUV above the T mix (estimated temperature increase due to laser $21 C) 38 as shown in Fig. 3A. Upon applying the laser, the domains mixed to form a uniformly uorescent liquid phase within 5 seconds. Once in the mixed phase, the trap was switched off to allow the vesicle to cool, leading to immediate demixing and the reappearance of coexisting liquid phases.
While monitoring the rapid transition between coexisting and uniform phases, uorescent images were taken in the calcein channel to monitor the uorescence of the GUV internal content. Between being in a mixed state and demixing into separate coexisting liquid phases release of calcein was observed (Fig. 3A). This release process was monitored over a period of 30 seconds; during this time, L o domains were coalescing to form larger domains. Multiple cycles of laserinduced localised heating were then carried out on individual GUVs with or without domains (Fig. 3B). Aer a single cycle, the internal uorescence of phase separated GUVs decreased by $66% (n ¼ 10), whilst the uorescence of single-phase GUVs showed no change. Two further heating cycles were then performed, and aer the third cycle, the phase separated GUVs showed almost complete loss of uorescence, whilst minimal change was observed inside single phase vesicles.
As with bulk GUV populations, we hypothesise that leakage is due to packing defects along the increased domain line interface during the phase transition. When GUVs were fully phase separated into two distinct L o and L d phases (half-moon topology) the domain line interface is relatively low when to compared with the demixing domain 'ripening' stage. During this stage, where many nano/microdomains are growing through collision and coalescence, the line interface between domains at its highest. Packing defects between domain boundaries are thus most prominent here which leads to the formation of transient gaps between the domains through which material can leak out. This phenomenon has been reported previously in gel/liquid systems. 18,31,39,40

Conclusions
We have shown that GUVs made using EPT exhibit the full range of phase behaviour observed in electroformed GUVs and can be used as thermosensitive microcompartments that can be individually targeted and triggered to release their cargo. We detail the phenomenon of membrane leakage when coexisting uid membranes are taken through the T mix and back. Previous work in this area has focused on single component systems that transition between gel and uid phases. To our knowledge, this is the rst work to conrm that any emulsion-based method can replicate the full range of phase behaviour previously displayed in ternary vesicles, 24,25,32 and the rst time that a laser has been used to induce phase transitions and content release in phase separated GUVs. Vesicles that display lateral phase separation offer advantages over single component systems for cargo release applications in that the transition temperature (and thus release temperature) can be nely controlled by tuning the lipid composition, as shown through the replacement of the saturated DPPC lipid with EggSM (Fig. 2). We demonstrated the versatility of our release platform through the encapsulation and triggered the release of both the uorescent dye calcein as well as the uorogenic peptide NFF-3.
A second key advantage of using ternary systems lies in the high level of structural control. Membrane organisation could serve as a foundation for controlled patterning of membranebound molecules in sensing or communication applications; by employing membrane phase separation, the transition temperature becomes a switch not just for content release, but also the controlled mixing of membrane-bound species anchored to different domains. These switchable systems could also be exploited in responsive cell-mimics where they can behave as thermal transducers to initiate down-stream chemical events within an articial cell.

Materials and methods
GUV generation via emulsion phase transfer DOPC, DPPC, rhodamine-PE were purchased from Avanti Polar Lipids. All other reagents were purchased from Sigma Aldrich. All experiments were performed in DI water and at 21 C unless otherwise stated. A protocol for the EPT method was developed based upon previous methods reported in the literature. 7,28,33 DOPC/DPPC/cholesterol lipids lms containing 1 phospholipid mol% rhodamine-PE were made by rst dissolving lipids in chloroform at appropriate ratios, followed by removal of chloroform under a stream of nitrogen and placing under vacuum for a minimum of 60 minutes. The lipid lm was dissolved in mineral oil (2 mg ml À1 ) by sonicating for 30 minutes at 40 C. An emulsion of droplets in oil was made by vortexing 20 mL of a 0.5 M sucrose aqueous solution in 200 mL of lipid-in-oil for 20 seconds. In calcein loaded-GUV experiments aqueous solution also contained 1 mM calcein. This solution was then layered above 250 mL of 0.5 M glucose in an Eppendorf to form a water/oil column, with an emulsion present in the oil phase. Droplets were driven through the interface to form GUVs by centrifugation at 9000g for 30 minutes, resulting in a pellet at the bottom of the tube. The supernatant was extracted and the pellet resuspended in 0.5 M glucose by pipetting up and down ve times.

Optical and uorescence microscopy of GUVs
Vesicles were visualised with a Nikon Eclipse TE2000-E inverted microscope, and recorded on an Andor Zyla, sCMOS based Once the laser was turned on the sample was locally heated and the ordered and disordered liquid domains mix into a uniform state (above the T m $ 32 C). After this point the laser was switched off to allow the GUV to cool and the domains to demix and eventually coalesce. Imaging the same vesicles in the calcein channel revealed leakage once the GUV was heated above T m and allow to cool ($30 seconds). Scale bars ¼ 10 mm. (B) Graph showing average mean fluorescence of phase separated GUVs (black squares) or single phase DOPC GUVs (red circles) (diameter 7-12 mm) after successive cycling events. Error bars ¼ 1 S.D., n ¼ 10.
camera. During uorescence imaging samples were illuminated with a mercury arc lamp, the TRITC and FITC lter were used to visualise uorescent rhodamine-PE (200 ms exposure) and encapsulated calcein dye (100 ms exposure) respectively. Fluorescence exposure was kept at a minimum to minimise photooxidation of unsaturated DOPC lipid.

Thermally responsive dye release
Thermally responsive dye release assays were conducted on GUVs formed from 1 : 1 : 3 DOPC : DPPC : cholesterol or DOPC : EggSM : cholesterol lipid lm. 20 mL of GUV solution was placed between two cover slips, which was in turn placed on a water bath controlled heating stage located on the microscope. Vesicles were cycled back and forth through the L o /L d phase transition temperature and calcein release observed.
When vesicles were taken through 20 heat cycles to achieve full content release, this was done in bulk, by placing an Eppendorf containing vesicles on a heating rack placed on a hot plate set at 40 C. The sample was le for 5 minutes, then removed and allowed to cool back to RT for 5 further minutes. This was repeated 20 times. In these experiments, vesicles contained 0.5 M sucrose both internally and externally, so they would not sediment. They were then transferred to the microscope for viewing, where they were diluted 1 : 9 in 0.5 M glucose in order for them to settle on the glass slide.
In the laser-induced heating experiments, 20 mL of GUV solution was placed on a Nikon TE2000-U inverted epi-uorescence microscope combined with a custom optical trapping system. The latter has been described and characterised in previous publications 34 and comprises a single Gaussian beam from an Ytterbium bre laser source (20 W at 1070 nm) which is directed to the microscope objective (60Â 1.4 NA) through a custom optical train. Individual vesicles were heated past the phase transition by applying a focused laser to the centre of the vesicle for 5 seconds.

Triggered release of peptide biomolecules
MOCAc-RPKPVENvaWRK(Dnp)-NH 2 (NFF-3) peptide was encapsulated (15 mM) within 1 : 1 : 3 DOPC : DPPC : Chol GUVs produced as detailed above. An additional purication step was added when preparing these vesicles; aer resuspension of the vesicle pellet in 0.5 M glucose, the suspension was centrifuged at 6000g for 10 minutes, before resuspension in 0.5 M sucrose to avoid gradual sedimentation of the vesicles during spectroscopic measurements. Vesicles were exposed to a single heat cycle consisting of heating at 40 C for 15 minutes before leaving the vesicles at room temperature for 30 minutes. TPCKtreated bovine trypsin (T1426 SIGMA, Sigma-Aldrich) was then added to external solution (0.1 mM, prepared in 40 mM HEPES, 0.5 M sucrose pH 7.4) and the uorescence emission of proteolysed NFF-3 recorded (l ex/em ¼ 320/405 nm) using a Horiba Jobin-Yvon Fluoromax 4 Spectrouorometer (Horiba, Japan). Fluorescence emission was compared against the emission of freshly prepared vesicle samples with added trypsin (used as a control to obtain background levels of uorescence of free peptide in solution). All vesicles samples were then lysed by adding Triton X-100 detergent to a nal concentration of 0.03 v/ v%, mixing and leaving for 1 hour. Then, uorescence emission was then recorded again to obtain values corresponding to '100% leakage'. In order to calculate the percentage of peptide leakage, the raw data was processed as follows: Peptide leakage ð%Þ ¼ À FI heat À À FI bg Â N ÁÁ À FI lysed À À FI bg Â N ÁÁ where FI heat , FI bg and FI lysed are the uorescence intensities of the peptide sample aer heating, the background uorescence intensity derived from the control sample and the uorescence intensity of the heat cycled sample aer lysis with Triton X-100. N is the normalisation factor: where FI lysed sample and FI lysed control are the uorescence intensity of the sample and the control aer lysis with Triton X-100. N takes into account small differences in peptide concentration in the cuvettes between the control and sample. N can be determined in this manner due to the linear relationship between uorescence intensity and peptide concentration in the concentration range used in these experiments (ESI Fig. S4 †).

Conflicts of interest
There are no conicts to declare.