Combined flow-focus and self-assembly routes for the formation of lipid stabilized oil-shelled microbubbles

Lipid and polymer stabilized microbubbles are used in medicine as contrast agents for ultrasound imaging and are being developed for the delivery of water soluble drugs to diseased areas of the body. However, many new therapeutics exhibit poor water solubility or stability, which has led to the requirement for the development of effective hydrophobic drug delivery systems. This study presents a new method to produce microbubbles coated with an oil layer capable of encapsulating hydrophobic drugs and suitable for targeted, triggered drug release. This new method utilizes highly controllable flow-focusing microfluidics with lipid oil nanodroplets self-assembling and spreading at gas–aqueous interfaces. Oil layer inside microbubbles were produced with diameters of 2.4±0.3 μm (s.d., 1.6 μm) and at concentrations up to 106 bubbles per milliliter. The mechanism of oil layer inside microbubble assembly and stability were characterized using methods including contact angle measurements, quartz crystal microbalance with dissipation monitoring and fluorescence resonance energy transfer imaging. A method for making microbubbles with an oil layer for encapsulating hydrophobic drugs could lead to more effective drug-delivery systems. Many therapeutic drugs exhibit poor stability or solubility in water, driving a need for effective delivery systems for these hydrophobic drugs. Stable microbubbles that are capable of delivering water-soluble drugs to diseased areas of the body are therefore attracting much attention. This led Stephen Evans and colleagues at the University of Leeds, United Kingdom, to develop a method for producing microbubbles containing a layer of oil that can encapsulate hydrophobic drugs. The team's approach utilizes highly controllable flow-focussing microfluidics with lipid-oil nanodroplets that self-assemble and spread at gas-aqueous interfaces, making them promising for use as taregtted, individualized therapeutic agents for various diseases. It also combines both diagnostic and therapeutic capabilities.


INTRODUCTION
It has been estimated that up to 40% of all newly developed drugs have poor water solubility. 1,2 Many drugs also show poor stability in water. Thus, while many of these drugs have shown good results in vitro their efficacy is reduced in vivo and in some cases have been show to increase toxicity toward healthy tissue 2 . For example, drugs such as Combretastatin A4, a vascular disrupting agent proposed for the treatment of colorectal cancer, have proven difficult to deliver effectively due to their hydrophobicity 3 , while drugs such as the anti-cancer agent Decitabine, a DNA methyltransferase inhibitor, have been shown to undergo rapid decomposition in water to inactive products 4 . A number of hydrophobic drug encapsulation systems have been investigated previously, such as porous nanoparticles 1,5 , sub-micron oil emulsions 6 , and micelles 7 . The release mechanisms of these systems can be either passive, for example, due to drug leakage from the carrier, or active, due to environmental-triggered release, for example, in response to local temperature or pH 8,9 . However, such mechanisms often lead to unpredictable drug release profiles when applied in vivo 10 . Therefore, a hydrophobic drug delivery system with a controlled, externally activated, triggered release mechanism could offer a considerable advantage over existing technologies.
Lipid and polymer-shelled microbubbles (MBs) have been used as contrast enhancing agents in ultrasound (US) imaging for over 30 years 11 . The compressibility of their gas core combined with its high acoustic impedance mismatch against surrounding blood plasma leads to a large scattering cross-section so that MBs provide a higher echogenicity than tissue interfaces [12][13][14] . This makes MBs a powerful tool in diagnostic US imaging of the cardiovascular system and organs. MBs have more recently been shown to enhance therapeutic delivery 15 , making them promising 'theranostic' agents, with both 'diagnostic' and 'therapeutic' capabilities. US destruction pulses can be used to trigger the release of encapsulated/attached drug payload by destroying the MB vehicle at the target. Alternatively, stable cavitation of MBs may be used to drive sonoporation of the nearby cells and stream fluid around the MBs, thereby aiding drug uptake, and MB destruction can lead to 'microjets,' which are able to pierce endothelial lining allowing improved efficacy of drug injection 15,16 .
To date, these dual functional MBs have been formed in a number of designs. MBs have been loaded with DNA/RNA in the lipid shell [17][18][19] . They have been decorated with nanobeads 18 and vesicles 14 for coating with and the encapsulation of hydrophilic species, respectively. Examples of loading have included quantum dots, luciferin, and propidium iodide 20,21 . With a simple lipid or polymer shell structure, MBs have also been co-delivered with hydrophilic species and drugs, for example, to overcome delivery across the blood-brain barrier 22 and enhance transdermal transport of insulin and heparin 23 .
MBs are conventionally produced by sonication, mechanical agitation, or microfluidics, typically through the use of a gas with low water solubility and an aqueous solution of lipids or polymers. MBs are required to be smaller than 8 μm to avoid the risk of blood flow complications such as arterial embolisms 24,25 . For the production of more complex MB architectures, such as those with drug payloads and molecular targeting, microfluidic approaches offer greater control over MB size, reduced wastage of drug cargo and targeting agents, and ease of cleaning the final product 14,26 . Microfluidic-based MB preparation typically utilizes flow-focusing (FF) geometries, where a gas phase is pinched off at a nozzle to form a MB, by a continuous aqueous phase containing a surfactant agent, usually lipids. Hettiarchi and Talu have previously shown high levels of size control, producing lipid-coated, perfluorocarbon core 1-5 μm MBs with polydispersity indices as low as 2-5% [27][28][29] . We have also developed an on-chip 'microspray regime' for MB formation, which can produce high concentrations of MBs (410 9 MB mL − 1 ) with an average size of 1.5 μm 14,30 .
One possible route for the incorporation of drugs onto the MB architecture is to thicken the MB shell with an oil layer, giving an oil layer inside MB (OLI-MB). These MBs would allow hydrophobic drugs to be held within the oil component, while retaining their US properties for imaging and active release. Unger et al. 31 were the first to incorporate an oil layer into MBs, which they termed acoustically active lipospheres (AALs). Since then AALs have been reproduced and further characterized by other groups. AALs have been produced through the mechanical agitation of lipid solution, oil (primarily soybean or triacetin), and a gas such a perfluorobutane (C 4 F 10 ), resulting in MBs of 2.9 μm average diameter with a thick oil shell capable of high drug loading per MB 31 , but large MB-MB variability with sizes ranging from 1 to 10 μm and drug loading being inconsistent between vehicles 32,33 . Oil layers can range from 500 to 1000 nm and 300 to 700 nm for triacetin and soybean oil, respectively, with some vehicles containing no gas at all 33,34 . It is thought that this large MB to MB variability is due to the production method's chaotic nature, and lack of control over gas encapsulation. In terms of their US capabilities, AALs' thick oil layer is thought to lead to viscous dampening 34 . Compared to non-oil-MBs, AALs fragmentation by US is significantly more difficult, with a high mechanical index of 0.67-2.0, and a fivefold higher pulse length being required for MB destruction 32,34 . During tests of AALs on cells, 30% of the cells were found to die from the US alone 35 . In 2009, Lentacker et al. 35 concluded that it remained difficult to efficiently release drugs from AALs using medically relevant US. Colombo and Edirisinghe demonstrated the production of~50 μm multi-layered OLI-MBs through the use of coaxially arranged needles, providing more monodisperse MBs 36,37 . However, this method is in part limited by the needles' inner and outer diameters, often leading to MBs that are larger than that required for medical applications. Lee and Zhang have previously demonstrated methods for OLI-MB and multi-layered MB production using a three-phase FF system and a system of two subsequent two-phase FF nozzles, respectively 10,25,38 . However, control of three different phases on-chip is inherently difficult and often leads to poor MB yields. As depicted in Figure 1, we propose a new, simple method for the production of MBs with an internalized thin oil layer between the gas core and lipid shell, OLI-MBs. Here we produce OLI-MBs using a FF system but include an emulsion of oil nanodroplets stabilized by a lipid monolayer shell, termed lipid oil nanodroplets (LONDs), into the aqueous phase. The LONDs undergo adsorption, rupture, and spreading at the gas-water interface, leading to the formation of a tens of nanometers thick oil layer around the MB. By including the LONDs in the aqueous phase, the microfluidics is brought back to a twophase system, permitting easier control over MB production and potentially allowing a broader range of hydrophobic solvents to be used for delivery. A combination of contact angle measurements and quartz crystal microbalance with dissipation (QCM-D) monitoring was used to identify the mechanisms of OLI-MB formation while fluorescence microscopy, including fluorescence resonance energy transfer (FRET), was used to confirm the final OLI-MB structure.

MATERIALS AND METHODS Materials
The lipid used in this work was 1-palmitoyl-2-oleoyl-sn-glycero-3phosphocholine (POPC), purchased from Avanti Polar Lipids (Alabaster, Al, USA). Squalane oil (Sigma-Aldrich, Dorset, UK) was used as the hydrophobic carrier phase. Phosphate-buffered saline (PBS) was prepared by dissolving one 5 g tablet of PBS powder (Life Technologies, Carlsbad, CA, USA) in 500 mL of MilliQ water and filtered through a 200 nm cellulose membrane syringe filter (Sartorius Stedim, Epsom, UK) prior to use.

Contact angles measurement and surface energy
Contact angle measurements of oil droplets at the water/lipid-air interface were performed. These were used to calculate the interfacial free energies of an OLI-MB, and to estimate the total interfacial energy of a specific OLI-MB structure compared to other conformations of oil and air in water in the presence of lipids.
A circular Teflon dish (6 cm diameter) was filled with MilliQ to allow a meniscus to form above the top of the dish walls. This meniscus was aspirated off before the dish was refilled. This was repeated a minimum of five times to remove any surface contaminants. This set up was analogous to a Langmuir Blodgett (LB) trough. POPC lipid, solubilized in chloroform at 1 mg mL − 1 , was added gradually to the water surface. This gave surface concentrations of either no lipid or enough lipid that the surface pressure was increased to just before the lipid layer collapsed at the dish edge, which was assumed to be at the collapse pressure. The sessile drop technique was used for contact angle measurement, using a First Ten Angstroms instrument, by depositing a droplet of oil at the air-water/lipid interface using a square cut needle. Images of the oil droplet were analyzed in ImageJ to measure both the left and right contact angles. At least five droplets were measured for each lipid monolayer condition. These contact angles were used to calculate the water/lipid-oil surface tension and total surface energies, described below.

LOND production
LONDs were formed via two methods, either by tip sonication or a two-stage ultra-fine homogenization process (in a similar method to that presented by Mico et al. 39 ). 3,3′-Dioctadecyloxacarbocyanine perchlorate (DiO), and dioctadecyl-3,3,3′,3′-tetramethylindocarbocyanine perchlorate (DiI) (Sigma-Aldrich, Dorset, UK) were encapsulated into the oil as drug mimics, and for optical quantitative purposes, by sonication for 30 min at room temperature, at a concentration of 1 mg mL − 1 .
LOND formation via tip sonication. An aliquot of 10 mg POPC lipid was dissolved in 1:1 chloroform-methanol, dried under nitrogen in a glass vial for 30 min, and then re-dispersed in 930 μL  LOND sizing and concentration measurements were performed using a Zetasizer and NanoSight (Malvern Instruments, Malvern, UK). LONDs were typically produced at~170 nm diameter and 10 13 LONDs per milliliter.
QCM-D: LOND spreading mechanics QCM-D was used to study LOND interactions with hydrophobic solid surfaces in an aqueous environment as a mimic to the airwater MB interface.
Octadecanethiol (Sigma-Aldrich, Dorset, UK) was used for coating gold-coated QCM-D crystals (Q-Sense AB, Stockholm, Sweden). Prior to the experiments, the crystals were thoroughly cleaned by UV-ozone cleaning for 20 min, followed by immersion in Decon 90 and then MilliQ while in a sonication bath, and finally vigorous agitation in isopropanol. The crystals were kept in each solution for about 10 min. This process was followed by an N 2 dry and again by UV-ozone for 20 min. Hydrophobic self-assembled monolayers were formed on the gold-coated crystals by immersion of the crystals in octadecanethiol in ethanol (1 mg mL − 1 ) for 12 h, after which they were rinsed with ethanol, to remove excess octadecanethiol, and MilliQ. For all QCM-D experiments, water contact angle measurements were performed on a control QCM-D crystal to confirm hydrophobicity, using a First Ten Angstroms instrument with MilliQ droplets.
A Q-Sense E4 (Q-Sense AB, Stockholm, Sweden) was used to monitor the interaction between the LONDs and the QCM-D crystal. Crystals were allowed to equilibrate for at least 1 h in degassed PBS, flowing at 14 μL min − 1 . The 5th, 7th, 9th, and 11th overtones were recorded and used in the analysis. The LONDs were introduced in to the flow chambers at a dilution of 10 12 LONDs per mL in PBS, at 140 μL min − 1 . The flow was then stopped for~2 h until no further frequency or dissipation changes were observed. PBS was used to rinse the system, in order to remove any unattached material (140 μL min − 1 ). The Sauerbrey and the Voigt-viscoelastic models were fitted to the above harmonics to calculate the mass adhered to the crystal during the experiment and estimate the thickness of the oil-lipid layer formed.
Microfluidic OLI-MB production Microfluidic FF devices were designed in Leeds and fabricated by Epigem Ltd (Redcar, UK) in PMMA and SU-8, as described previously and shown schematically in Figure 1a 9 . LOND solution was diluted to 10 12 LONDs per mL in 10% v/v glycerol, 4 mg mL − 1 NaCl. LOND solution flow rates of 15-50 μL min − 1 were delivered via a syringe pump (Aladdin, World Precision Instruments, Sarasota, FL, USA), passing first through a 200 nm filter (Sartorius Stedim, UK) to remove any oil that was not successfully homogenized or from LONDs that had destabilized over time. Flow of the LOND solution in the absence of MB production did not result in the formation of larger agglomerates but rather yielded oil emulsion comparable to the initial LOND solution, indicating their relative stability due to the lipid shell. Indeed, we have found these solutions to be stable for over 6 weeks 39 . C 4 F 10 gas (F2 Chemicals, Preston, UK) pressures of 20-34.5 kPa were regulated by a digital gas flow controller (Alicat Scientific, Tucson, AZ, USA). At the nozzle, the gas was pinched off and stabilized as a MB by the LOND solution. The MB solution contained 10 μL of liquid perfluorohexane (C 6 F 14 ) per 1 mL of precursor MB solution to extend MB lifetime by intercalation of the C 6 F 14 into the MB shell 40 . High frame rate imaging (FASTCAM SA-X, Photron, San Diego, CA, USA) between 10 and 50 kfps, performed on an inverted light microscope (Eclipse Ti-U, Nikon, Tokyo, Japan), was used to observe single MB pinch-off and flow immediately after the nozzle during production. For size and concentration measurements, 50 μL of the MB solution, taken from the middle of the solution immediately after gentle agitation to prevent preconcentration due to bubble buoyancy, was placed in a 120 μm deep, BSA Dorset, Sigma-Aldrich, UK) coated, glass chamber and observed under dark field (Eclipse Ti-U, Nikon, Tokyo, Japan) with an oil immersion × 100 objective.
For the determination of MB lifetime, three 50 μm deep, BSAcoated, wax sealed, glass observation chambers were prepared and the MBs were observed using dark-field microscopy, as a function of time. The MB diameter and concentration were measured using ImageJ.

OLI-MB washing
Microfluidic cell-traps were used to trap and wash larger OLI-MBs to confirm the structure of the MBs 41 . The traps were made from PDMS, and were designed to catch 45 μm MBs, to allow higher resolution imaging of OLI-MB structures. Prior to any use, the microfluidic devices were washed with 70% ethanol to remove any air, followed by 1 mL MilliQ and 1 mL of 1 mg mL − 1 BSA in PBS at 100 μL min − 1 to reduce non-specific binding of excess LONDs to the cell-trap walls. The MB sample was then introduced to the device at 5 μL min − 1 over 20-30 min (allowing enough MBs to be trapped), before being washed by PBS at 10 μL min − 1 for 30 min. All observations were made using a Confocal Laser Scanning Platform Leica TCS SP8 on a DMi8 microscope (Leica Microsystems, Milton Keynes, UK) with a 488 nm excitation laser and a 561-582 nm emission window. LAS X software was used for image capture and processing.

FRET LOND interactions on MBs
FRET was used to infer LOND destablization and LOND to LOND mixing at the MB gas-aqueous interface. Separate preparations of LONDs were made containing either a FRET donor (DiO) or a FRET acceptor (DiI) dye 42,43 . These solutions were mixed 1:1 and then diluted to 10 12 LONDs per ml in PBS. This mixed LOND solution was then used to produce MBs microfluidically, as above. MB samples were loaded into 50 μm deep, BSA-coated glass-viewing chambers. Fluorescence was then observed on the above confocal system with a × 40 objective, using a 488 nm excitation laser, and 500-520 nm and 561-582 nm emission windows for the donor and acceptor, respectively. In order to identify the enhanced FRET fluorescence, FRET donor LOND + FRET acceptor LOND samples were compared to control donor + non-fluorescent/ empty and empty + acceptor samples. As in Equation (1), the intensity at the MB surface in both control images was removed from the donor + acceptor image to reveal the FRET signal 44,45 .
where FRET f is the final FRET image, FRET i is the initial FRET (donor +acceptor) image, and I is the intensity.

RESULTS AND DISCUSSION
Contact angle measurement and surface energy calculation Figure 2 shows a schematic of a droplet of oil at the air-water (modified with lipid) interface, and images of droplets at the airwater interface in the absence and presence of an adsorbed POPC lipid layer. The squalane contact angle decreased from 25°to 6°in the presence of POPC. The surface tension values of squalane reported in the literature vary between 25 and 30 mN m − 1 (Refs. [46][47][48][49][50]. While that of the lipid modified air-water interface is 26-27 mN m − 1 for POPC at the collapse pressure [51][52][53][54] . The oilwater interfacial tension was calculated using the balance of the interfacial tensions for oil in contact with a water sub-phase and air spectator phase (see Equation (s1) from supplementary information). This was found to be 55 mN m − 1 without lipid and ⩽ 1.9 mN m − 1 with POPC. During OLI-MB formation, we form the MBs through flow focus 'pinch-off' of the gas in the presence of an aqueous solution containing LONDs with an average diameter of~170 nm, as shown schematically in Figure 2c(i). If we assume a process in which LONDs adsorb then rupture to form a uniform oil layer at the water/gas interface (vide infra), then we expect of the order of 350 LONDs per MB. With this assumption, we can estimate the total surface energies for the initial starting state (Figure 2c(i)) and the different possible end configurations; (Figures 2c(ii) and (iii)). In going from the initial configuration (Figure 2c(i)) to the desired lipid-coated oil-shell surrounding a MB (Figure 2c(ii)) there would be a 2.6-fold reduction in the total surface free energy. In contrast, the scenario in which the LONDs agglomerate into a larger oil droplet, and the MB remains uncoated (Figure 2c(iii)), would lead to a similar total surface free energy to the initial configuration. The conclusion drawn here is that, since the OLI-MB structure was calculated to have the lower total surface free energy, it is energetically preferential as a final structure, and so formation of the structure is plausible.

QCM-D: LOND spreading mechanics
To mimic the LOND/MB interaction, we performed QCM-D of LOND interactions with model hydrophobic surfaces. Figure 3 shows the changes in frequency (ΔF) and dissipation (ΔD) on interaction of LONDs with a hydrophobic QCM-D crystal. Initially, there was a rapid, negative increase in ΔF accompanied by a rapid, positive increase in ΔD, indicative of the absorption of LONDs and the formation of a heterogeneous, non-uniform surface layer, which couples strongly to the aqueous phase. Over the following 10 min ΔF decreased and reached a plateau. Simultaneously, the ΔD decreased to a stable value of~15 × 10 − 6 . This behavior is similar to that of bilayer formation via vesicle adsorption and rupture on hydrophilic surfaces [55][56][57] . We propose a similar twostage mechanism to explain our observations, (i) the attachment of LONDs (possibly partial wetting of the surface) followed by (ii) the slower spreading of droplets on the surface, leading to the detachment of entrapped water and possibly some oil. At the onset of washing the chambers, fresh LONDs were introduced to the surface, leading to a smaller increase in ΔF and ΔD, followed by the PBS wash, which resulted in the signal recovering to the 'before wash' level and further remaining stable. Changes in frequency provided an estimated value for the thickness of the oil layer formed during LOND rupture to be between 10 and 20 nm. This result gave an estimate of the order of 70 LONDs per MB. Recalculating the previous total surface free energies in respect to this still showed a 2.5-fold reduction when forming an OLI-MB from LONDs.
Microfluidic OLI-MB production Figure 4a shows microfluidic OLI-MB production. Optimized production required low gas pressure , and aqueous flow rates between 20 and 25 μL min − 1 , to prevent coalescence of the pinched off gas. This could be due to the slower adsorption and rupturing of LONDs at the gas-aqueous interface to stabilize the MBs, as depicted in Figure 1a. Production rates were~5 × 10 3 OLI-MBs per s. Figures 4b and c show off-chip imaging of the OLI-MBs and an associated size distribution. While OLI-MBs were routinely produced on-chip at 9.0 ± 1.8 μm diameter (s.d., 1.4 μm), off-chip imaging showed a repeatable reduction in MB size to~0.3 times their initial size, giving OLI-MBs of 2.4 ± 0.3 μm (s.d., 1.6 μm) and a concentration of 6.2 ± 0.8 × 10 6 MBs mL − 1 . This size change was due to gas leakage from the interior of the MB after production. A MB's diameter effectively stabilizes when the lipid shell concentrates from an equilibrium state to being close packed at the collapse pressure. This effect has been shown previously on DPPC: DPPA:DPPE-PEG5000 MBs formed using FF devices, where MBs shrank to 0.39 times initial size 58 . Figure 5 shows the change in OLI-MB concentration and diameter over time. There was a sharp initial decrease in concentration over the first 90 min, after which the bubble concentration appeared stable for the duration of the measurement. During this period, the MB population showed a modest increase in size, presumably due to Ostwald ripening 29 . Figure 6 shows a trapped OLI-MB after the washing step to remove excess LONDs from the surrounding solution. The persistent fluorescent halo suggested that the oil component of the OLI-MBs was stable against shear forces and the removal of excess oil and lipid from the surrounding solution. All MBs observed in these traps after washing steps were recorded with some degree of fluorescent shell, indicating that the OLI-MB samples had been formed without any non-oil-MBs.
FRET LONDs: observing OLI-MB oil layer Figure 7a shows a schematic for a FRET experiment to demonstrate the coalescence of LONDs and the mixing of their contents. Figure 7b shows the fluorescence signatures expected for three potential outcomes of LONDs at a MB's gas/water interface. Figure 7c shows the experimental acceptor emission following the interaction of LONDs containing donors and acceptors. FRET was shown to occur at the MB surface. As detailed by Figures 7a and b, observation of FRET signal provides evidence that the LONDs did adsorb and rupture at the MB interface, where oil and loaded fluorophores from both LOND populations mixed together. When compared to the QCM-D results for LONDs at a water-hydrophobic solid (as opposed to gas) interface, these results appear to show a similar mechanism for the formation of a continuous oil layer around the gas core, resulting in OLI-MBs.

Additional work
Tripropionin OLI-MBs. Squalane is a non-polar oil with a very low water solubility 59 . Tripropionin is a partially water soluble triglyceride (3.07 mg mL − 1 at 37°C) used as flavoring agent 60 . By expanding the above work to form OLI-MBs from tripropionin LONDs, solubilization of a larger range of non-polar to partially polar drugs in to the oil phase would be possible. As such, the microfluidic OLI-MB production and wash steps have been repeated using tripropionin (Sigma-Aldrich, UK) LONDs. (See Supplementary Figure S1) Tripropionin OLI-MBs have been produced at 2.1 ± 0.1 μm diameter (s.d., 0.8 μm), though with a lower concentration of 10 4 OLI-MBs per milliliter.

CONCLUSION
We have presented a new approach to preparing MBs with an internalized oil component utilizing simple flow focussed microfluidics coupled with self-assembly. Squalane OLI-MBs were produced on-chip with diameters of 9.0 ± 1.8 μm that subsequently shrank and stabilized to a diameter of~2.4 ± 0.3 μm (s.d., 1.6 μm), without any OLI-MBs shown above 8 μm. The final OLI-MB size has made this formation route of practical use for producing these MBs for clinical application. Off-chip OLI-MBs had a concentration of 6.2 ± 0.8 × 10 6 OLI-MBs per mL. The mechanism of formation was via the adsorption and spreading of LONDs at the hydrophobic-hydrophilic interface during and shortly after MB 'pinch-off'. This process was mimicked using QCM-D, where the formation of a uniform oil layer was confirmed from changes in the frequency and dissipation. The continuous nature of the oil shell was confirmed using FRET imaging of MBs formed from two LOND species (one containing a donor species and the other an acceptor species). We have also shown that this is a general approach that can be used with different oil types. These architectures hold potential for the encapsulation and delivery of poorly water soluble and water-stable drugs. For potential clinical application of OLI-MBs, it may be advantageous to selectively separate the OLI-MBs from the excess LONDs they are formed in. This is possible based on the buoyancy of the different particles 61 .