Interactions of Choline and Geranate (CAGE) and Choline Octanoate (CAOT) Deep Eutectic Solvents with Lipid Bilayers

Mixtures between choline and geranic acid (CAGE) have previously been shown to insert into lipid bilayers. This may be useful for the transdermal delivery of larger pharmaceuticals, however, little is known about the mechanism of activity. By comparing the interactions between CAGE and lipid bilayers with those of a less‐active, yet closely‐related analogue, choline octanoic acid (CAOT), a chemical basis can be investigated. Overall, six systems are studied here by neutron reflectivity, where d54‐1,2‐dimyristoyl‐sn‐glycero‐3‐phosphocholine (DMPC) solid‐supported phospholipid bilayers are first formed on SiO2 substrates before exposure to the deep eutectic solvent (DES). Components of the DES could be identified within the bilayer by exploiting contrast variation and selective deuteration. CAGE is shown to be a mild disruptive agent, free to insert and diffuse across the bilayer, preserving much of the bilayer integrity. Experiments identify co‐mingling of geranate ions inhibits the efficient packing of lipid tails, increasing hydration across the bilayer. Conversely, CAOT is found to both exchange and remove lipid molecules to achieve incorporation, inducing swelling and the formation of solvent patches. It appears these behaviors derive from the structures of the anions and thus amphiphilicity of the DES, laying the foundations for the rational design and optimization of these candidates toward transdermal delivery.


Introduction
Functional and highly-tailorable, ionic liquids (ILs) and deep eutectic solvents (DESs) are often considered more sustainable than their traditional counterparts given that they can be bioderived, exhibit low toxicity, and can be readily recycled.Many of these solvents are also amphiphilic and it is common to find inherent interactions with phospholipid membranes. [1]n particular, mixtures between choline (Ch) and geranic acid (GA), coined CAGE (Figure 1A), have been found to be potent in this regard.Able to transport large molecules, including proteins, [2] antibodies, [3] DNA, and even insulin [2,4] across a range of biological tissues, [3,5] CAGE could facilitate the transdermal delivery of many currently challenging pharmaceuticals.CAGE has already been found to improve the bioavailability of active pharmaceutical ingredients (APIs) through a range of mechanisms, including reduced mucosal viscosity, [5b] increased stability of the API against pH, improved polymorphic retention, and solubility. [6]It is also bactericidal, [7] penetrating deep into the skin, and has already been formulated into a commercial sanitizer. [8]ot much is known about its mechanism of activity, but it appears that a 1:2 Ch:GA ratio is optimal for the DES to effectively penetrate skin. [4]Mechanistic studies on other materials have shown that IL and DES systems disrupt membranes through complex pathways, including nanoparticle nucleation on the membrane, [9] exchange with lipid molecules, [10] stiffness perturbation, [9,11] and solvent patch formation. [10]Found to relate to a structure-function relationship, [12] activity with membranes has thus far been primarily rationalized by the lipophilicity of the IL or DES used. [9,10]Moreover, while there have been some attempts to gain insights through computational methods, [9,12] these models are limited, leading us to look to neutron scattering studies; a complementary and increasingly popular technique to identify and locate DES species within lipidic mesophases during interaction. [11,13]13a] In that study, at concentrations of 100 mm and above, changes in scattering length density (SLD) were found that related to a ≈8-10% volume occupancy of the bilayer by the IL.The IL was also found to be evenly distributed across both bilayer leaflets, suggesting that it could diffuse across the bilayer while preserving its integrity.As might be expected, the most polar IL domains were found associated with the lipid head groups, and the least, with tails.Although DMPC lipids are inherently limited in their ability to capture the complexities of real skin and differ significantly from the structure of sebaceous or epidermal lipid compositions, [14] the work of Benedetto et al. demonstrates the valuable mechanistic insights to be gained from using relatively simplistic membrane models.
Many of these previous studies, however, have focused on noncomparable materials, such as those comprising a hard anion like the study described above, and so learning is not directly translatable.This prevents rational optimization or design of new IL/DES systems toward specific membrane-related effects, such as transdermal drug delivery.Instead, we propose steady and systematic chemical comparisons to CAGE to robustly evaluate the potential structure-function relationship of this activity. [15]ere, CAGE will be compared to a less effective, yet closelyrelated analogue, a mixture between Ch and octanoic acid (OA), [16] referred to here as CAOT (Figure 1B), which lacks the carbon-carbon double bonds and branching methyl groups in the anion.There are two primary physiochemical differences between CAOT and CAGE which can be compared.These relate to the rotational freedom, and hence steric rigidity, of the anionic aliphatic chain, as well as the overall hydrophobicity of the mixture, where octanoic acid has been found to be sparingly soluble in water, [17] while GA is insoluble without choline. [18]In this case, choline may be acting as a hydrotrope, [19] increasing the solubility of the fatty acid in mixtures with water, highlighting the importance of evaluating structure-function relationships.Neutron reflectivity was used to examine the interactions of CAGE and CAOT with solid-supported lipid bilayers (SLBs), alongside contrast variation, in an attempt to locate the DES within the bilayer.This will begin to identify potential mechanistic differences that enable the unique and powerful membrane interactions of this DES.

DES Structure and Behavior
CAGE forms a rich diversity of nanostructures, both in pure [20] and hydrated forms, [18] driven by hydrogen bonding and nanosegregation between charged and hydrophobic domains.In alternative IL/DES systems, others have indicated that these nanostructures may have a role in mediating lipid membrane disruption. [9,10]Hence it is of interest to characterize the structure of CAGE under the conditions in which it will be applied.
Here, both CAGE and CAOT in IL (1:1 ratio; choline:'ate) and DES (1:2 ratio; choline:'ate:'acid) forms were mixed with water, with concentration calculated in respect to choline, for example, 1:2 100 mm CAGE would be composed of 100 mm Ch and 200 mm GA.The minimum active concentration of CAGE was desired to ease interpretation of eventual NR results.To determine this, DMPC supported vesicle layers (SVLs) were exposed to increasing concentrations of aqueous 1:2 CAGE (0.3, 10, and 100 mm) until a response was recorded on a quartz crystal microbalance with dissipation (QCM-D) (Figure S4, Supporting Information).7a] This was also near the concentration at which aqueous 1:2 CAGE mixtures appeared as turbid emulsions (Figure 2A), evidencing a possible structure-function relationship.
To investigate this further, 100 mm aqueous mixtures of CAGE and CAOT in both 1:2 and 1:1 ratios were examined by dynamic light scattering (DLS) (Figure 3).Upon agitation, suspensions of both 1:2 CAGE and CAOT appeared as milky emulsions (Figure 2B), stable for ≈3 h.DLS indicated that these consisted of large structures, ≈400-2000 nm in diameter.Conversely, the 1:1 materials in water had a clear appearance with a slight grey tint (Figure 2B), and smaller nanostructures, ≈70-200 nm in diameter, were identified.This size difference is likely due to the ratio of the materials, as there are more hydrophobic species to accommodate in the emulsion droplets for the 1:2 DES.CAOT produced the most monodisperse structures, both in 1:2 and 1:1 cases, whereas CAGE showed greater deviation between runs.This may be because geranic acid is slightly less soluble in water [18] than octanoic acid, [17] and therefore the DES produces less stable structures.This in turn may influence the propensity for interaction of the DES with lipidic interfaces in polar solvents.Differences in the ratio of the cross-section of the acid group compared to the size of its hydrophobic "tail" in CAGE versus CAOT may also have an effect on the size or shape of these aggregates via interfacial stabilization of the droplets.SAXS/SANS studies are underway to fully resolve the structures formed across a range of concentrations of these and other CAGE analogues.
To evaluate the interaction of the DES systems with lipidic mesophases, QCM-D was conducted using both SLBs on silica interfaces (5 MHz), as well as SVLs on gold (10 MHz) (Figures 4 and 5).This allows some insight into how the DES interacts with the bilayers.By monitoring the mass gained at the interface upon introduction of the DES, affinity for the lipidic layers can be compared, and from losses, signs of disruption discerned.
When SLBs were incubated with 1:2 CAGE suspensions (Figure 4A,C), it was found that CAGE adsorption plateaued around 80 Hz, corresponding to 1400 ng cm −2 according to the Sauerbrey Equation (see QCM-D in the Experimental Section; Equation ( 6)) after 5 min, with little to no change in adlayer rigidity.In contrast, 1:2 CAOT suspensions initially deposited a 2100 ng cm −2 adlayer, before proceeding to lose mass over 10 min, equilibrating around 1100 ng cm −2 .While this technique cannot discern exactly what this mass loss was comprised of, whether lipid, whole DES, one or more of its constituents or a mixture of these, mechanisms of disruption can still be inferred.Previous computational studies into IL-induced membrane disruption have identified deposition of IL species until a critical threshold, where IL insertion and bilayer swelling occurs, before bilayer budding and eventual disruption. [9]The threshold adhesion seen here for CAOT could feasibly represent something similar.
Interestingly, a large drop in dissipation (5 × 10 −6 ) after incubation with CAOT was also observed, denoting a significant stiffening of the bilayer.Stiffness can increase upon IL exchange or incorporation into the lipid bilayer, [11] particularly when occurring asymmetrically, for example, to the outer leaflet only.This has been found to be predictive of the greatest disruptive effect, [9] whereas symmetrical insertion is more likely to retain the integrity of the bilayer.
SVLs were also incubated with the same DES mixtures (Figure 4B,D), as disruption can be further identified by any additional losses relating to the solvent contained within the vesicles upon rupture. [21]For example, CAGE deposited only 790 ng cm −2 on an SVL compared to 1400 ng cm −2 on the SLB.Furthermore, there was a longer timeframe of equilibration.Whereas CAGE adsorption to a SLB reached plateau within 5 min, in the SVL case, a plateau had still not been reached after 25 min.This suggests that there are competing mechanisms between CAGE adhesion and the removal of molecules from disrupted vesicles or the vesicles themselves.
When incubating 1:2 CAOT with SVLs, 1320 ng cm −2 were initially deposited before again, a rearrangement was observed, similarly over a longer timeframe compared to the SLB (20 vs 10 min), reaching a plateau around 90 ng cm −2 .Stiffness was again seen to increase, in line with asymmetric DES insertion.Some have found that other ILs lead to the removal of mixed IL-lipid micelles, leaving behind intact mixed IL-lipid bilayer leaflets. [10]This could be the origin of the dynamic adhesion mechanism for CAOT, explaining the resultant stiffness increase.Overall, our results suggests that 1:2 CAOT is more disruptive to DMPC layers than 1:2 CAGE, possibly owing to the increased hydrophobicity of CAGE and hence propensity to diffuse across the bilayer without interference.
Neither 1:1 CAGE nor 1:1 CAOT facilitate transdermal delivery, [4] and interestingly these systems had very similar responses to each other (Figure 5).When incubated with an SLB (Figure 5A,C), both 1:1 CAGE and CAOT rapidly deposited around 3200 ng cm −2 .Faster rates of deposition were also seen with the SVL (Figure 5B,D) compared to the 1:2 materials.This could be due to the decreased hydrophobicity of the 1:1 materials, where there is a lower volume percentage of hydrophobic DES aggregates present in solution, and so protection from the polar solvent is instead sought through insertion into the lipidic phases.Overall, no trend in stiffness was observed in the SLBs, whereas the SVLs became much less stiff, in line with the presence of a soft adlayer.
From both the DLS and QCM-D results, it appears that the structure of the DES anion, as well as the ratio to cations, has an important role in determining the overall DES structure and disruption behaviors.However, as 1:2 CAGE was found to be the least disruptive to DMPC structures, the extent of distortion to lipid vesicles and bilayers may not be the best indicator of capacity for transdermal delivery.Given that QCM-D alone cannot determine vertical insertion or distribution of the DES, these effects may be better probed by neutron reflectivity.

Evaluation of Solid-Supported Phospholipid Bilayers by Neutron Reflectivity
NR data were collected on the FIGARO beamline at the ILL, as described in the Experimental Section.Data were analyzed using Motofit and Refnx [22] software packages where a "slab" model was used to generate a SLD profile.Initially, the bare Si blocks were characterized in D 2 O and fit using a single-layer model representing a SiO 2 layer on the surface of the Si block (Figure S7 and Table S7, Supporting Information).SLD values for Si and SiO 2 were fixed, and that of the bulk solvent only allowed to deviate slightly from that calculated for the known solvent compositions to account for incomplete solvent exchange within the flow cell.Characterization was limited to three out of the total six blocks due to beamtime restrictions.Where fit parameters were obtained for the SiO 2 layer, these were held constant when fitting later adlayers such as the lipid bilayer.For those SiO 2 layers not directly characterized, values similar to those that were characterized were used.
After initial characterization, Si blocks were exposed to small unilamellar vesicle (SUV) suspensions which then collapse to form d 54 -DMPC bilayers.The model used here to fit the NR data consisted of five layers: the SiO 2 layer, an inner phosphatidylcholine head group leaflet, two identical tail group layers, and an outer head group leaflet, assuming a symmetrical bilayer.Additionally, head and tail thicknesses, d Heads and d Tails respectively, were constrained as to not exceed the maximum thickness for a fully extended, all-trans conformation of DMPC lipids based on Tanford equations; ≈14 Å for the tails and 12 Å for the head groups. [23]The roughness of these layers were also held equivalent, given the chemical coupling between head and tail groups.SLDs for head and tail groups were held constant using values from Hall et al. [23] An additional constraint, that the mean molecular area of the head and tail groups should be equivalent, was used to enforce a physically realistic bilayer structure.This was applied using the following equations adapted from Campbell et al. [24] and Hall et al. [23] To illustrate, the hydrated cross-sectional area of the head groups, A Heads,Hyd , can be calculated using Equation (1).
where, V m,Heads represents the partial specific molecular volume of the head groups (277.2Å 3 ) [23] and d Heads , the fitted thickness of the head groups.Similarly, the area of the tail groups, A Tails , may be calculated from Equation (2), with the exception that the solvent volume fraction within the tails, X Solvent,Tails , has been included to account for incomplete lipid coverage.
where, V m,Tails is the specific molecular volume of the tail groups (789.6 Å 3 [23] ) and d Tails , the fitted thickness of the tail group.As the mean molecular area of the head and tail groups must be equal, the volume fraction of "dry" lipid head groups, X Heads , may be calculated as allowing for calculation of the solvent volume fraction, X Solvent,Heads within the head group layers.
Figure S8, Supporting Information, presents overlaid data pertaining to all six DMPC bilayers prepared, where SUV collapse appears to have yielded consistent bilayer structures on the six different silicon blocks used during the experiment.Parameters for this single model, based on the constraints outlined above, are detailed in Table S8, Supporting Information.This was found to closely describe the majority of the lipid bilayers, particularly at low Q where there was the least uncertainty.The area per molecule of the lipids was found to be 58 Å 2 , in agreement with previous findings. [23]However, although these parameters satisfy Equations ( 1)-( 4), it can be seen that they are unaccounted for deviations between the six bilayers.Primarily, these differences were found to be described by three variables: the SLD of the bulk solvent, due to incomplete solvent exchange within the cell, substrate (SiO 2 ) parameters, and the hydration of the head groups.
To precisely fit individual bilayers, Equations ( 1)-( 4) ultimately proved too constraining, relating to the achievable resolution of layer thicknesses from fitting NR data alone.To explain further, for any individual contrast, the minimum real space dimension that can be resolved is given by Equation (5).
However, smaller structures may yet contribute partial features to the NR data.By utilizing multiple interfacial contrasts in combination with the a priori assumptions detailed above, the structural contribution of these components may be addressed within a useful level of uncertainty.Hence, fit parameters are reported to be within the nearest Å.The average error estimation within the fitting software for the thickness of these layers was between ± 0.5-1 Å.Using thickness parameters with a 0.5 Å uncertainty, based on Equation (2), would carry an error to the solvent fraction of ±1 − d±0.5 d or ± 1 2d , which is ≈± 3.6% in the tail hydration and ± 5.6% in the head group hydration based on the current model (Table S8, Supporting Information).As 0.5 Å differences in thickness cannot be resolved under the current Qrange, despite three contrasts, it seems unreasonable to constrain the hydration of these layers accordingly.Hence, these boundaries of uncertainty provide a more reasonable, yet still conservative (<10% variance from assumptions), constraint upon the model, which indeed was found to satisfy all bilayers (Tables 1  and 5).The majority of bilayers were fit with a tail thickness of 14 Å, 2-5% hydrated, and a head group thickness of 9 Å, between 40-43% hydrated.Indeed, the bilayer presenting the highest hydration of head groups at 52% (Table 5) was the only bilayer to have an increased head group thickness (10 Å) where a higher   extent of hydration would be expected to maintain the equivalent cross-sectional area, primarily dictated by the 14 Å thick tail slab.

Structural Effects on Phospholipid Bilayers upon Interaction with CAGE
1:2 CAGE suspensions (100 mm) were incubated with lipid bilayers for 30 min before being rinsed with clean solvent followed by characterization in three contrasts that were fitted simultaneously: D 2 O, Si-matched water (SiMW), and H 2 O.As before, solvent SLDs were allowed to vary slightly from those calculated to allow for incomplete solvent exchange within the cell.This will be partially correlated to the other fitted variables, but is only expected to introduce minor uncertainties as a result.CAGE from both fully hydrogenous hCh-hGA and deuterated dCh-hGA systems induced lasting changes to the bilayer (Figures 6 and 7).SLDs calculated for the individual DES components can be found in Table S6, Supporting Information, and full fit parameters for the bilayer pre-and post-CAGE incubation in Tables S9 and S10, Supporting Information.Purposefully keeping the model as constrained as possible, alongside solvent SLD, the distortions seen after CAGE incubation could be accommodated in the fit by releasing only four variables: the SLDs and hydrations of the lipid head and tail groups (Tables 1 and 2).Interestingly, the data could be fitted even when the layer thicknesses were kept fixed, suggesting the presence of CAGE does not have a significant structural impact on the bilayer, consistent with QCM-D results.Incorporated components of the DES could be identified within the bilayer through selective deuteration.For example, it was found that choline inserted into the d 54 -DMPC head group region as the h-Ch sample evidenced a decrease in the SLD of the head group layer and the d-Ch sample an increase, whereas the hydrogenated geranate would be expected to cause a decrease in the SLD in both cases, as was seen in the tail region.These SLD changes were found evenly distributed across both bilayer leaflets, suggesting CAGE is free to insert into and diffuse across the bilayer while preserving its integrity.7a] Volume occupancy can be similarly examined by calculating the percentage difference of the fitted SLD between the calculated SLDs of the mixed-layer components.For example, for the dCh-hGA sample (Table 2), the SLD of the lipid head group region post incubation was 2.4 × 10 −6 Å −2 , and hence found to be comprised of 90% vol.DMPC head groups (SLD = 2.14 × 10 −6 Å −2 ) and 10% vol.dCh (SLD = 4.7 × 10 −6 Å −2 ).Although differences were seen in the extent of interaction between the samples (10-11% for dCh-hGA, and 3-4% volume occupancy for hCh-hGA), this was possibly due to the effects of deuterating Ch.Similar isotope effects are beginning to receive attention elsewhere [25] and appear particularly important when interactions are mediated by hydrogen bonding. [26]In both cases the ratio of the volume occupancy between choline and geranate was found to be 1:1, indicating geranic acid was removed during the rinsing step.As it has been shown that 1:2 CAGE is necessary for effective transdermal delivery, and QCM-D results indicated a difference in behavior between the 1:1 and 1:2 materials, it may be that geranic acid is a mediating participant in the insertion mechanism.As done by Benedetto et al., [11] by comparing the relative molecular volumes of DMPC (1066.8Å 3 ) [23] to those estimated for CAGE by DFT calculations (see Section 4, Figure S18, and Table S22, Supporting Information) it is possible to use the volume occupancy to evaluate that there were between 1.9-2.5 hCh-hGA molecules per ten d 54 -DMPC molecules by number, and 6.3-6.9 dCh-hGA molecules per ten d 54 -DMPC molecules by number, in line with previous findings. [11]ydration changes offer further information on the system.In both cases, the hydration of the tail groups increased proportionally to the volume % occupied by the DES, 3% for hCh-hGA, and 9% for dCh-hGA.The hydration of the head groups, however, instead decreased by the same extent.If the increased tail hydration was due to removal of whole lipid molecules (i.e., a reduction in coverage) an increase in head group hydration would be required to maintain symmetry (Equations ( 1)-( 4)).However, the decoupling of the head and tail hydrations suggests that choline, associated with head groups, does so by displacing the water molecules already between them.This means there was no disturbance to the volume occupied by the dry lipid head groups, possibly aiding to maintain the structure of the bilayer.
Increases to the tail group hydration without a reduction in coverage means that the packing between the lipid tails must have been disrupted.This would be an expected prerequisite for transdermal delivery, allowing for solvent and molecular transport.As it has been found that 1:2 CAGE appears optimal for transdermal delivery, although lacking information in the xy plane, it is tempting to speculate that these regions of increased hydration align to create channels bridging the entire bilayer.
An alternative model was also evaluated, where SLD and hydration changes were constrained to the outer leaflet only.An example output is shown in Figure S9 and Table S14, Supporting Information.It is noted that this model not only produced a worse fit, unable to capture the downturns present in all three contrasts at higher Q (≈0.2 Å −1 ), but required increased hydrations of both head and tail regions.This suggests an unreasonable structure, where lipid coverage has been reduced in the outer leaflet only, leaving lipid tail groups on the inner leaflet solventexposed.The ratio volume occupancy between choline and geranate was found to be 1:1.4,corroborating the tendency of these models to find a 1:1 CAGE ratio remaining in the bilayer, regardless of constraints.
While contrast variation has helped to improve the models, still we can only be moderately confident in this interpretation in the absence of further data.By comparing these results against an analogue known to be less active, CAOT, it is hoped that variant interaction mechanisms may be distinguished.

Structural Effects on Phospholipid Bilayers upon Interaction with CAOT
1:2 CAOT suspensions (100 mm) also interacted with the dDMPC bilayers, producing a variety of responses.Figure 8 and Table 3 present the data relating to CAOT from hCh-hOA, where full fit parameters can be found in Table S11, Supporting Information.Here, it appears that CAOT is arranged within the bilayer as CAGE was, with 14.4% choline volume occupancy in the lipid head groups, and 14.7% octanoate in the lipid tails.Again this was found to be in a 1:1 ratio, evenly distributed across both bilayer leaflets.An alternative model, where only the variables associated with the outer leaflet were unconstrained, was also attempted (Figure S10 and Table S15, Supporting Information), however this resulted in a significantly poorer fit, again with an unreasonable bilayer structure.
CAOT was found to occupy the bilayer to a greater extent than CAGE, corroborating the stronger interaction indicated by QCM.Hydration changes, however, were no longer decoupled, where both hydrations of the head and tail groups increased by 11% (Table 3).As this occurred evenly across both bilayer leaflets, this suggests the removal of lipid patches or a reduction in coverage.Patch formation in bilayers upon interaction with ILs has been reported elsewhere, [10] and here, is a further indication of a behavioral change compared to CAGE.Interestingly, a persistent swelling of the bilayer (+10 Å) was also identified, apparent from the small shift of the Keissig fringe in the data to lower Q in the reflectivity pattern of the D 2 O contrast from Q = 0.05-0.06Å −1 to Q = 0.04-0.05Å −1 .Alternative models, where bilayer thickness was constrained, produced the poorest fits (Figures S11 and S12; Tables S16 and S17, Supporting Information).Corner plots, describing the slight correlation between lipid tail thickness, SLD, and hydration, can be found in Figure S3B, Supporting Information.
Although useful for comparison, dCh-hOA appeared to produce an anomalous result.It was noted during preparation that this DES was the only sample not to form a turbid suspension.It is possible that this specific isotopic contrast resulted in a different macrostructure due to effects of deuteration on hydrogen bonding.Furthermore, QCM-D results from this mixture (Figure S14, Supporting Information) uniquely showed a rapid stripping of the vesicle layer from the interface.The response of the bilayer was challenging to fit by freeing bilayer variables alone (see Figure S13 and Table S18, Supporting Information), and required that it became highly hydrated (≈48-85%).A structured monolayer was also found not to produce a satisfactory fit.Instead, an arbitrary organic layer at the SiO 2water interface was used to fit the data, albeit with a high degree of uncertainty (Figure S15 and Table S19, Supporting Information).This was found to be 18 ± 12 Å thick, with an SLD of 5.84 ± 3 × 10 −6 Å −2 , 60 ± 18% hydrated, somewhat representative of a diffuse, isotropic mixed dDMPC-dCh-hOA layer.
Results from hCh-dOA and dCh-dOA interacting with the dDMPC bilayer are shown in Figures 9 and 10, and Tables 4  and 5, respectively.These samples induced a different response in the bilayer compared to hCh-hOA, again possibly pertaining to hydrogen-bonding-mediated isotope effects arising from the deuterated acid.Here, unconstraining the hydration of the bilayer led to only minor changes which did not significantly alter the fit, and so were held constant to reduce the number of fitted variables.This suggests that any insertion by these DESs was achieved through exchange with the lipid molecules, and as tail hydration did not increase, we do not expect any propensity for transdermal delivery.
In both cases, models with only the outer leaflet unconstrained produced satisfactory fits.The SLD of the lipid tails was found to reduce to between 5.2-5.8 × 10 −6 Å −2 .This was also the case when using models where both leaflets were free to deviate (Figures S16 and S17; Tables S20 and S21, Supporting Information).This was unexpected as these values are far below the SLD of d 15 -OA (6.21 × 10 −6 Å −2 ), the anticipated deuteration of which was checked by 1 H NMR (Figures S5 and S6, Supporting Information).Although the DES samples are hydroscopic, aliphatic chains do not undergo H/D exchange.Hence, we are forced to consider the drop in tail SLD to be induced by choline, and the placement of dOT elsewhere in the structure.
Although penetration of choline into the tails is a reasonable suggestion, this would mean the d 9 -portion of d 9 -Ch would no longer contribute to the SLD in the region of the lipid head groups.Yet, we observe a rise in head group SLD upon incubation with dCh-dOA, indicating the presence of a deuterated species in this region.An alternative solution is that only choline is contributing to the SLD changes across the entire bilayer.Here, it may be possible that the polar NC 3 H/D 9 portion of choline is tethered to the head group region, while the less polar CH 2 CH 2 OH portion is to be found within the tails.Indeed, segregating choline this way provides SLDs attaining the closest 1:1 volume occupancy ratio between the lipid head and tail groups.This also implies that the charged portion of choline is tethered to the bilayer surface, aligned perpendicular to the solvent interface.Hence, octanoate must be likewise aligned, protruding from the bilayer-solvent interface.As the SiO 2 surface would block such a structure on the inner leaflet, the bilayer was fit with only the outer leaflet unconstrained, producing the fits in Figures 9 and 10, preferred over those with both leaflets unconstrained, despite producing slightly better fits at high Q (Figures S16 and S17; Tables S20  and S21, Supporting Information).The additional dOT layer was found to be 5 ± 2 Å thick, and highly hydrated at 92-94%.Although not very thick, this would make sense given the hydrophobic C 8 chain of octanoate would more likely be lying atop the head group region, opposed to extended into solution.
Simplified schematic representations of the fit models for CAGE from h/dCh-hGA and CAOT from h/dCh-dOA and hCh-hOA can be found in Figure 11.Overall, CAOT demonstrated a much more dynamic mechanism of bilayer disruption, as was suggested by QCM-D results, where it is possible that initial penetration of the bilayer by choline (Figure 11C) is followed by removal of lipid patches, allowing rearrangement and diffusion of octanoate into the tail regions of both bilayer leaflets (Figure 11D).
Interestingly, it seems CAGE had much less of a disruptive effect on the bilayer, as was again suggested by QCM-D results.This allowed CAGE to interact with the bilayer without   perturbation to its structural integrity, that is, without the removal or exchange of lipid molecules.Here, insertion of the anionic component was found to disrupt packing, increasing the hydration of the tails (Figure 11B), and speculatively, this is what conveys the transdermal delivery properties of CAGE in comparison to other IL/DES systems.Regardless, clear behavioral differences in interaction, incurred by chemical alterations to the anion, have been demonstrated, likely driven by a mixture of amphiphilic balance and rigidity of the aliphatic chain.Changes in the length or shape of the anion could also now be explored using similar experiments.As octanoate occupies slightly less volume than geranate (Table S22, Supporting Information) it is possible that the variant volume ratio to choline is also having an effect upon DES packing and hence distribution within DMPC bilayers.

Conclusions
Through techniques such as NR and QCM-D, this study has successfully identified the interactions of CAGE and CAOT with lipid bilayers, allowing for mechanistic differences to be discerned.1:2 aqueous CAGE suspensions (100 mm) were mildly disruptive to DMPC vesicles and bilayers.Post-exposure, no dimensional changes to the bilayer could be identified under NR, where instead the DES appeared free to diffuse across, and insert into, both bilayer leaflets.Here, co-mingling of geranate increased the hydration of the lipid tail groups without reduction in surface coverage, hence indicating distortion to the packing of the bilayer.Interestingly, despite using a 1:2 ratio during incubation, only a 1:1 choline to geranate ratio was found inserted into the bilayer after rinsing.1:1 CAGE was not expected to be effective as a transdermal delivery agent, and behaved similarly to 1:1 CAOT under QCM-D, appearing to rapidly adhere to, but largely not disrupt the bilayer.Hence, in 1:2 CAGE systems, geranic acid may only be playing a mediatory role in the insertion mechanism.Future experiments should seek to probe this aspect further.
Uniquely, behavioral differences between the DESs have been induced by chemical alteration to the anion.Previous studies, focusing on IL/DES systems with hard anions, have typically concentrated on the implications of the organic cation; not necessarily translatable to species such as CAGE.Hence, it was beneficial to compare CAGE to a more chemically similar species.Unlike CAGE, 1:2 aqueous CAOT suspensions (100 mm) appeared to have a multi-step disruption mechanism.QCM-D experiments showed that the DES initially had a stronger adhesion to the bilayer than CAGE, followed by some form of structural rearrangement that resulted in increased stiffness.NR also showed multiple states, including a bilayer with choline inserted into the outer leaflet only, as well as the insertion of the anionic octanoate into both leaflets, accompanied by swelling and the formation of solvent patches.It could appear from these results that CAOT was more disruptive to lipid membranes than CAGE, and previously, this might have been interpreted to mean CAOT would be more likely to be an effective transdermal delivery agent.However, through NR, we have shown that those DESs that have only a mild effect upon the bilayer structure, can still induce the increased hydrations needed to allow for solvent transport and transdermal delivery.
Through a high-fidelity technique like NR, and a relatively simple comparison, already aspects pertaining to the chemical identity of the anion have been discovered to contribute to the interaction with lipid bilayers.It is likely these differences were driven by a mixture of shape, amphiphilic balance, and rigidity of the aliphatic chain.In the interest of examining interactions more realistically representative of those with skin, it is important that lipid species beyond phosphatidylcholines like DMPC, or mixtures of these, are also investigated.The characteristics of DMPC, such as the zwitterionic head group or saturated tails, may well affect the extent of penetration by the DES.An extended series of systematic and sequential experiments should now be explored, with the aim to understand the structure-function relationship of these biologically-active DESs, potentially unlocking a new class of functional pharmaceutical carriers.

Experimental Section
Materials: The lipid species, d 54 -DMPC, was purchased from Merck and not purified further before use.Similarly, choline bicarbonate (80% aqueous), geranic acid (85% technical grade), and octanoic acid (≥99%) were purchased from Merck and used without further purification unless specified.Deuterated compounds were purchased from Cambridge Isotope Laboratories and used as received.
Mono and dibasic sodium phosphate (purity ≥99%) were purchased from Acros Organics and not purified further.A 50 mm phosphate buffer solution (PBS) was prepared by combining 0.1 m aqueous solutions of monobasic sodium phosphate (2.65 mL, 2.65 × 10 −4 mol) and dibasic sodium phosphate (47.35 mL, 4.735 × 10 −3 mol) with 50 mL H 2 O (18.2 ΩM).NaCl (1.1688 g, 0.02 mol) was then added, resulting in a 50 mm PBS (200 mm NaCl) stabilized at pH = 8.0, representative of physiological conditions.PBS was only used in the vesicle deposition step for SVLs on gold in QCM.
DES Preparation: Briefly, to a solution of d 9 -choline hydroxide in water or choline bicarbonate in water and ethanol was added dropwise a solution of geranic or octanoic acid in ethanol, as appropriate.This was then stirred for 2 h before excess solvent was removed by reduced pressure resulting in a viscous liquid that was then dried under vacuum for at least 24 h.Full experimental details and yields can be found in the Supporting Information.
SUV Preparation: SUVs were prepared following an established thinfilm methodology.First, DMPC was dissolved in chloroform (1 mL, 10 mg ml −1 ), and dried under N 2 (g) in a vial.During drying, the vial was rotated by hand to promote the formation of a thin film.Caution was taken to ensure that no material losses occurred at this point, as this was effectively determinant of the final concentration.The resulting film was then dried in a vacuum oven for 2 h to ensure complete removal of chloroform.The resulting lipid film and 2 mL of the appropriate solvent were then heated to 10 °C above the T g of DMPC (≈24 °C), before mixing (5 mg mL −1 ).While warm, the mixture was briefly vortexed to create a turbid suspension.Suspensions were then sonicated for ≈1 h until clear with a slight blue tint.This must be done in a bath sonicator as to avoid contamination from titania nanoparticles that shed from probe sonicators.Suspensions were periodically dipped into an ice bath to protect against possible damage from overheating.Clarity signified the loss of particles greater than 100 nm which scatter light, indicating the formation of SUVs (20-100 nm) as confirmed by DLS and SAXS (Figure S1, Supporting Information).
SLB Preparation: SLBs were prepared by SUV collapse.Immediately prior to use, SUV suspensions were again bath sonicated to clarity, and subsequently centrifuged (16 000 × g) for 15 min to remove any larger or multilamellar vesicles.13c] Suspensions were then injected over clean SiO 2 substrates under slow continuous flow (0.5 mL min −1 , 30 min) to allow time for vesicle collapse, before rinsing with clean water.SLB formation was confirmed by QCM-D (Figure S2, Supporting Information) where ≈530 ng cm −2 were deposited over 5 min with little to no change observed in adlayer stiffness.
SVL Preparation: SVLs were instead prepared by incubating SUVs in PBS (50 mM phosphate, 200 mm NaCl) with Au-coated substrates for ≈30 min before rinsing with clean buffer.After which, a stable SVL spontaneously forms, again confirmed through QCM-D (Figure S2, Supporting Information).Here, 880 ng cm −2 were deposited, causing an ≈2.5 × 10 −6 rise in dissipation, associated with the decreased stiffness of a vesicle adlayer. [27]The mass ratio of DMPC deposited between the SBL and SVL was consistently found to be 1:1.6,further supporting the successful formation of these layers. [27]LS: DLS was conducted using a Malvern Zetasizer Nanoseries, using disposable plastic cuvettes.Measurements were taking using backscattering ( = 173°) and  = 663 nm.In all cases, at least five sets of measurements were taken, each of at least 12 runs, to ensure satisfactory cumulant fits.
QCM-D: QCM-D was conducted using an OpenQCM Q-1 sensor module with either SiO 2 -coated (5 MHz, OpenQCM) or Au-coated (10 MHz, OpenQCM) quartz chips (13.9 mm diameter).Pristine substrates were first rinsed with ethanol, acetone, and water before cleaning under UV/ozone for 15 min before being rinsed with milli-Q water prior to experiments.Cleanliness was determined by contact angle with water: <15°for SiO 2 surfaces and >80°for Au surfaces.The sensors were specified with a frequency stability with temperature of ±20 KHz at 23 °C.Mass deposited, Δm, may be calculated using the Sauerbrey equation (Equation (6)) assuming rigid, poorly hydrated layers.
where Δf is the change in resonant frequency (Hz), n the overtone, and C the chip sensitivity, where C = 17.7 and 4.4 ng Hz −1 cm −2 for a 5 and 10 MHz chip, respectively.Clean chips were first sealed into the flow cell and calibrated in air, before incubation with clean solvent.Once a stable baseline had occurred, experiments were commenced.
DFT Calculations: DFT calculations of 1:2 CAGE and 1:2 CAOT structures were performed using Gaussian16. [28]Optimization, energy, and volume calculations were run at the B3Lyp/6311+g(d,p) level.All structures were visualized and checked using frequency analysis.To estimate molecular volume, volume calculations were performed in Gaus-sian16.The volume calculation for each structure was performed ten times and the average of these used as an estimate of the molar volume (Table S22, Supporting Information).
Neutron Reflectivity: Experiments were conducted at the Institut Laue-Langevin (ILL), France, on the FIGARO neutron reflectivity (specular) beamline.Si blocks (Crystran; 8 by 5 cm, aiming for 3 Å roughness) were cleaned under UV/ozone for 20 min, rinsed with milli-Q water, and then sealed in solid/liquid flow cells with an approximate volume of 1 mL, prior to experiments.Temperature was controlled by steel plates flanking the flow cell, regulated at 27 °C to be above the T g of DMPC.For contrast, SiMW was prepared by mixing 38% volume D 2 O with H 2 O. Full experimental protocols can be found in the Supporting Information.
Resulting multidimensional time-of-flight data were converted to reflectivity curves using the COSMOS software available at the ILL. [29]Models were fitted using the Markov-Chain-Monte-Carlo (MCMC) algorithm within Refnx, [22] using prior bounds (Tables S2-S5, Supporting Information) and at least 8000 steps with 200 walkers.Fit uncertainty was quantified through the use of Bayesian inference of the posterior distribution of solutions within Refnx and fits were subsequently plotted in the Motofit package within Igor Pro. [30]Again, full details can be found in the Supporting Information.
Statistical Analysis: Uncertainties in the neutron reflectivity data derived from counting statistics, instrument set up, and detector characteristics were calculated during data reduction using the data reduction software COSMOS, [29] available at the ILL.Fit parameters for the models used to fit NR data were inferred from the Bayesian MCMC algorithm within the Refnx [22] software package and are expressed as mean ± standard deviation (SD) in the tables above and in the Supporting Information.

Figure 4 .
Figure 4. A-D) QCM-D response of a A) SLB on SiO 2 (5 MHz) and B) SVL on Au (10 MHz) to incubation with 100 mm aqueous 1:2 (Red) CAGE and (Blue) CAOT mixtures with corresponding dissipation measurements (C,D).Data shown are representative traces selected after at least two repeats.

Figure 5 .
Figure 5. A-D) QCM-D response of a A) SLB on SiO 2 (5 MHz) and B) SVL on Au (10 MHz) to incubation with 100 mm aqueous 1:1 (Red) CAGE and (Blue) CAOT mixtures with corresponding dissipation measurements (C,D).Data shown are representative traces selected after at least two repeats.

Figure 6 .
Figure 6.Contrasts of hCh-hGA.A,C) Neutron reflectivity data (points) with overlaid fits (solid lines) and B,D) corresponding SLD profiles of a dDMPC bilayer A,B) before and C, D) after incubation with hCh-hGA in (Red) D 2 O, (Green) SiMW, and (Blue) H 2 O. Fit parameters are given in Tables1 and S9, Supporting Information.Uncertainties in (A,C) show the statistical errors in reflectivity calculated during data reduction (see Section 1.3, Supporting Information).

Figure 7 .
Figure 7. Contrasts of dCh-hGA.A,C) Neutron reflectivity data (points) with overlaid fits (solid lines) and B,D) corresponding SLD profiles of a dDMPC bilayer A,B) before and C,D) after incubation with dCh-hGA in (Red) D 2 O, (Green) SiMW, and (Blue) H 2 O. Fit parameters are given in Tables 2 and S10, Supporting Information.Uncertainties in (A,C) show the statistical errors in reflectivity calculated during data reduction (see Section 1.3, Supporting Information).

Figure 8 .
Figure 8. Contrasts of hCh-hOA.A,C) Neutron reflectivity data (points) with overlaid fits (solid lines) and B,D) corresponding SLD profiles of a dDMPC bilayer A,B) before and C,D) after incubation with hCh-hOA in (Red) D 2 O, (Green) SiMW, and (Blue) H 2 O. Fit parameters are given in Tables 3 and S11, Supporting Information.Uncertainties in (A,C) show the statistical errors in reflectivity calculated during data reduction (see Section 1.3, Supporting Information).

Figure 9 .
Figure 9. Contrasts of hCh-dOA.A,C) Neutron reflectivity data (points) with overlaid fits (solid lines) and B,D) corresponding SLD profiles of a dDMPC bilayer A,B) before and C,D) after incubation with hCh-dOA in (Red) D 2 O, (Green) SiMW, and (Blue) H 2 O. Fit parameters are given in Tables4 and S12, Supporting Information.Uncertainties in (A,C) show the statistical errors in reflectivity calculated during data reduction (see Section 1.3, Supporting Information).

Figure 10 .
Figure 10.Contrasts of dCh-dOA.A,C) Neutron reflectivity data (points) with overlaid fits (solid lines) and B,D) corresponding SLD profiles of a dDMPC bilayer A,B) before and C,D) after incubation with dCh-dOA in (Red) D 2 O, (Green) SiMW, and (Blue) H 2 O. Fit parameters are given in Tables 5 and S13, Supporting Information.Uncertainties in (A,C) show the statistical errors in reflectivity calculated during data reduction (see Section 1.3, Supporting Information).

Table 1 .
Selected fit parameters for dDMPC bilayer that change after hCh-hGA incubation.Remaining parameters are given in full in TableS9, Supporting Information.

Table 2 .
Selected fit parameters for dDMPC bilayer that change after dCh-hGA incubation.Remaining parameters are given in full in TableS10, Supporting Information.

Table 3 .
Fit parameters for dDMPC bilayer that change after hCh-hOA incubation.Remaining parameters are given in full in TableS11, Supporting Information.

Table 4 .
Selected fit parameters for dDMPC bilayer that change after hCh-dOA incubation.Remaining parameters are given in full in TableS12, Supporting Information.

Table 5 .
Fit parameters for dDMPC bilayer that change after dCh-dOA incubation.Remaining parameters are given in full in TableS13, Supporting Information.