Cholesterol Hinders the Passive Uptake of Amphiphilic Nanoparticles into Fluid Lipid Membranes

Plasma membranes represent pharmacokinetic barriers for the passive transport of site-specific drugs within cells. When engineered nanoparticles (NPs) are considered as transmembrane drug carriers, the plasma membrane composition can affect passive NP internalization in many ways. Among these, cholesterol-regulated membrane fluidity is probably one of the most biologically relevant. Herein, we consider small (2–5 nm in core diameter) amphiphilic gold NPs capable of spontaneously and nondisruptively entering the lipid bilayer of plasma membranes. We study their incorporation into model 1,2-dioleoyl-sn-glycero-3-phosphocholine membranes with increasing cholesterol content. We combine dissipative quartz crystal microbalance experiments, atomic force microscopy, and molecular dynamics simulations to show that membrane cholesterol, at biologically relevant concentrations, hinders the molecular mechanism for passive NP penetration within fluid bilayers, resulting in a dramatic reduction in the amount of NP incorporated.


Synthesis of monodisperse amphiphilic AuNPs
Amphiphilic AuNPs with low core size dispersion (i.e. monodisperse AuNPs) were prepared applying few minor modifications to the procedure published by Yang et al. 2 This procedure is divided into two steps: 1) synthesis of monodisperse oleylamine (OAm)-coated AuNPs and 2) complete thiol-for-OAm ligand-exchange.
1) OAm-AuNP synthesis. All glassware was washed in aqua regia before use. A magnetic hotplate stirrer equipped with temperature control was used to stir and heat the reaction mixture.
Temperature gradients were avoided by immersing the reaction flask in a thermostated water bath. In a 100 ml 3-neck round-bottomed glass flask, HAuCl4·3H2O (0.5 mmol) was dissolved in 16 mL OAm and 20 mL n-octane. The flask was sealed with rubber septums, abundantly flushed with Ar to exchange all air, and placed in a sonicator bath for ca. 1 min before stirring; a static Ar atmosphere was then maintained throughout the synthesis. The reaction temperature was set to 45 °C. After temperature equilibration, a tBAB reducing solution (0.5 mmol in 4 mL of OAm) was quickly added to the gold-OAm mixture. After one hour, ca. 40 mL of EtOH was added to quench the reaction and facilitate NP precipitation. Final OAm-AuNPs were collected by centrifugation at 5000 g for 10 min.
Three more washing cycles in EtOH were performed (a small volume of DCM was used to redisperse the product). Final OAm-AuNPs were dried under high vacuum to obtain shiny dark powders.
2) Thiol-for-OAm ligand exchange. The ligand exchange was performed at room temperature.
In separate glass vials, ca. 30 mg of hydrophobic OAm-AuNPs and a large excess (0.1 mmol) of the thiol mixture were dispersed in DCM (5 mL and 20 mL, respectively). To obtain a final 2:1 MUS:OT molar ratio in the ligand-shell, the stochiometric thiol ratio in solution was set to 3:2. The thiol mixture was sonicated for several minutes before adding, under vigorous magnetic stirring, the NP dispersion to start the ligand-exchange. During the first 2 h, the exchange mixture was mildly sonicated for ca. 30 s every half hour; the stirring was then left overnight. The same sonication procedure was repeated 2 h before NP washing. After ~12-15 h, the exchange mixture was diluted in acetone (ca. 15 mL) and centrifugated (5000 g, 4 min). To remove all hydrophobic unbound ligands (OAm and excess OT), 6 washing cycles in organic solvent were performed before vacuum drying. Dried AuNPs were then redispersed in water (ca. 15-20 mL) and centrifuged (4000 g, 6 min) in a hydrated AMICON® ultra centrifugal tube (regenerated cellulose membrane, 10 kDa cutoff molecular weight) to filter off the excess of water-soluble MUS (this procedure was repeated 15x). The purified NP aqueous dispersion was freeze-dried to obtain a manageable NP powder. Before characterization, 2:1 MUS:OT AuNPs were suspended in water and diluted in the experimental buffer (PBS) when specified. In general, they showed remarkable long-term colloidal stability in both aqueous media and their dispersion did not require further manipulation before use.

Preparation of lipid vesicles with tunable bilayer stiffness
Increasing cholesterol amounts were added to DOPC to prepare fluid lipid vesicles with increased bilayer stiffness. Lyophilized DOPC was weighed in a glass vial, dissolved in CHCl3:CH3OH (2:1, v/v), and divided into aliquots. Stochiometric amounts of cholesterol -previously dissolved in the same solvent mixture -were added to DOPC aliquots to set the chol mol % to 17, 30, 33, 40, and 50 %. After gently mixing to homogenize each sample, the solvent was evaporated under a stream of N2.
After ca. 24 h under vacuum, the lipid films were then hydrated in PBS at a 2 mg/mL lipid concentration. To promote the formation of multilamellar vesicles, hydrated samples were sonicated in an ultrasonic bath for ca. 15 min at room temperature. Subsequently, always at room temperature, each lipid suspension was extruded 25 times using the Avanti Mini-Extruder (Avanti Polar Lipids) with a 0.1 μm pore diameter polycarbonate membrane (Nuclepore filters, Whatman). For fluorescence anisotropy assays, the vesicle bilayer was labeled by adding small aliquots of DPH to the organic DOPC/chol mixtures (1:1000 probe-to-lipid molar ratio 3 ). A DPH stock solution was previously prepared in CHCl3:CH3OH (2:1, v/v). Such relatively low DPH molar fraction was chosen to yield adequate signal to noise ratio and avoid any probe-induced perturbation of the bilayer structure. Upon hydration of labeled lipid films, DPH is known to intercalate within the lipid bilayer 4 .
All vesicle suspensions were stored at 4 °C and used within a few days; batches containing fluorescent probes were also protected from light till further use.
Estimated cholesterol content into the extruded DOPC bilayer. It is known that the vesicle phospholipid:chol composition is not equal to the lipid ratio before film hydration. In particular, the actual cholesterol content in the vesicle bilayer is always lower than the nominal content added in solution, and in general, the gap between the two values increases with increasing cholesterol. A robust and accurate determination of cholesterol content in DOPC/chol vesicles extruded as in this work has already been reported by Goñi et al. 5 Based on the incorporation efficiency reported in that study, the cholesterol content in the vesicles used in this work was estimated as shown in Table S1.

AuNP characterization
Electron microscopy analysis. Bright-field transmission electron microscopy (BF-TEM) characterization was perfomed on MUS:OT AuNPs using a FEI Tecnai Osiris operated at 200 kV.
After sonication in an ultrasonic bath, few drops of a diluted NP water dispersion were deposited onto an ultrathin carbon-coated Cu grid. A TEM image showing monodisperse spherical NPs is reported in Figure S1. The core size distribution (Table S2) was calculated by assuming spherical morphology and by automatically counting a few thousand (at least 1000) particles with ImageJ software.  Table S2. NP concentrations of 0.3 mg/mL and 0.2 mg/mL were used for size analysis in water and in buffer, respectively, whereas ζ-potential acquisitions were performed at 0.07 mg/mL in water and 0.1 mg/mL in buffer. In all cases, NP dispersions were sonicated in an ultrasonic bath for ca. 30 seconds before analysis.  Figure S2A) of sharp peaks generated by free molecules indicates that no unbound ligands were present 1 .
Subsequently, the Au core was etched by a large iodine excess to quantify the MUS:OT molar composition in the ligand shell. This procedure induces the decomposition of the gold core by releasing all thiol ligands as free disulfides 1 . A concentrated iodine solution was prepared by dissolving 15-20 mg of iodine in 800 µL CD3OD. After ca. 20 min of sonication to favor complete iodine solubilization, 800 µL of the etchant solution was added to NPs (4-5 mg). The NP-iodine mixture was sonicated in an ultrasonic bath for approx. 30 min before transferring the light orange supernatant to a 5 mm NMR tube to acquire the spectrum ( Figure S2B). No residual OAm trace was detected, thus confirming that the thiol-for-OAm ligand exchange was complete. After normalization on the number of nuclei and correction due to each contribution, the integral values indicated in Figure   S2B were used to calculate the MUS:OT molar ratio (full details of ligand ratio calculation are described in ref 1 ). Final ligand composition is reported in Table S2.

AFM measurements
AFM measurements were performed on SLBs in the absence of NPs, at room temperature and in liquid environment., Experiments were developed as follows: contact-based Quantitative Imaging (QI TM by JPK) was exploited to map the Young's modulus of SLBs with increasing membrane stiffness (Figure 2A, main text). Tapping mode imaging was also performed to obtain topographic images at higher resolution ( Figure S4).
Measurements were carried out in PBS using a Nanowizard III AFM (JPK Instruments) mounted on an Axio Observer D1 inverted optical microscope (Carl Zeiss). The drive frequency was 12-14 kHz, the scan rate 0.5-1.5 Hz.
contact-based Force Volume imaging was performed to investigate the cholesterol-induced variation in the SLB breakthrough force ( Figure S5). Measurements were carried out in water using a Multimode SPM equipped with "E" scanning head (maximum scan size 15 μm) and driven by a Nanoscope V controller (Digital Instruments-Bruker).
In both cases V-shaped silicon nitride cantilevers (DNP-10, cantilever C, Bruker, nominal spring constant: 0.24 N/m) were used. SLBs were allowed to cool to room temperature and finally rinsed gently with water. This step was necessary to remove undeposited vesicles from the liquid that may interfere with AFM measurements.  Statistical data analysis. The bilayer Young's moduli shown in Figure 1A (main text) are reported as mean value ± standard error. For each lipid composition, results were averaged on at least 30000 curves deriving from at least 3 QI images. The bilayer breakthrough forces shown in Figure   S5 are reported as mean values obtained from at least 3 Force Volume images. Errors were evaluated according to Student's statistics, assuming a confidence level of 95%.   (1). After tip-bilayer contact, the force starts to increase (2). The breakthrough force is the force value at which the tip penetrates the bilayer (3). Immediately after the breakthrough event, the force dramatically decreases as a consequence of bilayer rupture. Subsequently, the force increases further due to the interaction between the tip and the solid substrate (4).

Fluorescence anisotropy assays
According to Equation 1, the fluorescence anisotropy of bilayer-embedded DPH molecules increases when their translational and orientational freedom decreases due to a reduction in the degree of membrane fluidity. All experiments were performed using excitation and emission slits with a bandpass of 4.5 nm; λex and λem were set as 358 and 428 nm, respectively.
Statistical data analysis. Four replicates were performed for each experiment at varying membrane cholesterol. Uncertainties on mean values shown in Figure 2B (main text) were calculated using Student's statistics assuming a 95 % confidence level.

Computational methods
Computational model. In this study we used a coarse-grained (CG) model to represent the

Well-Tempered Metadynamics parameters. We used the Well-Tempered Metadynamics (WT-
MetaD) technique 18 to activate the process leading to the anchoring of the first MUS ligand on the opposite membrane leaflet and to measure the associated free energy barrier, following the same protocol adopted in previous works. 11,12 As collective variable, we used the distance along the z axis

QCM-D experiments
Dissipative QCM investigation was carried out in PBS using a QCM-Z500 microbalance (KSV Finland LLC) equipped with a thermostated flow chamber. To form homogeneous supported vesicle layers (SVLs), vesicles were deposited onto the surface of a gold-coated QCM sensors. These sensors contain a AT-cut disk-typed quartz crystal with standard resonance frequency of 5 MHz, a sensitivity coefficient (Cf) of 17.7 ng/(cm 2 ·Hz), and a piezoelectrically active surface of 78.5 mm 2 . Before use, the sensors were subjected to UV/Ozone for at least 10 min. The monitoring of higher (3 rd -11 th ) overtones was carried out for all experiments. In general, SVLs were used instead of more common SLBs 21,22 because vesicles containing increasing cholesterol content were unlikely to rupture on the sensor surface resulting in incomplete vesicle fusion. 23 The use of SLBs would therefore have led to poor reproducibility and comparability between results at different cholesterol percentages.
Furthermore, we already showed in our previous work 13 that spontaneous absorption of amphiphilic MUS:OT NPs is facilitated in the free bilayer of lipid vesicles compared to that of SLBs.
Sample preparation. Figure S8 schematizes the steps of a typical experiment involving SVLs and NPs. For SVL formation, freshly extruded vesicle suspensions ( Figure S3) were diluted in PBS (0.25 mg/mL) and then inserted into the QCM pre-chamber. After 10 min of equilibration at 22 °C, vesicle suspensions were injected into the QCM chamber. As shown in Figure 3A of the main text, SVL deposition occurred with different kinetics depending on the membrane cholesterol content.
Before NP addition, each SVL was gently rinsed with fresh PBS (thermostated at 22 °C) to remove the excess of undeposited vesicles as well as any vesicle loosely attached to the SVL surface ( Figure   4A, main text). In general, SVLs were highly stable after deposition and rinse. For NP addition, a fixed NP volume was taken from a stock solution in water (0.6 mg/mL) and diluted in PBS (2 mL) to achieve a lipid/NP molar ratio of ~1600 with respect to the vesicle concentration previously injected to form the SVL (0.25 mg/mL). This ratio is nominal since excess vesicles were rinsed away before NP addition ( Figure S8, step 2); the real lipid/NP value during SVL-NP incubation is therefore lower.
Each NP dispersion in PBS was then sonicated for a few minutes in an ultrasonic bath and finally inserted into the pre-chamber. After 10 min of equilibration at 22 °C, NPs were injected into the chamber and let to interact with the preformed SVL for at least 20 h at 22 °C ( Figure 4A, main text).
After a few hours with a sampling interval of 1 s, a point average every 10 s was selected. The following day, the NP-SVL complex was gently rinsed with PBS before the end of the recording. In all cases, no mass losses were recorded after buffer exchange ( Figure 4A, main text). Data processing. QCM-D results after SVL stabilization (and rinsing) were processed using the viscoelastic Voigt model described in ref 25 . This modeling provides information on membrane viscoelastic properties, including layer density, thickness, viscosity, and elasticity. In particular, data processing was carried out by fitting experimental Df and DD data (from at least 4 harmonics) to the Voigt model included in the KSV QCM Impedance Analyzer Software (version 2.0) 26 and using the SVL thickness (L, nm), density (ρ, g/cm 3 ), viscosity (η, Ns/m 2 ), and shear modulus of elasticity (G', MPa) as fitting parameters. In recursive fitting procedures, the SVL viscoelastic properties were obtained starting from viscosity and elastic modulus values derived from literature data 27 . Density and thickness results are reported in Figure 3B of the main text, whereas final viscosity and elastic modulus values are included in Figure S10.
After NP-SVL interaction, QCM-D results were processed in the assumption of a modified Sauerbrey model to include the change in viscosity typical of a viscoelastic layer ( Figure S10): where: is the overtone number at which the crystal is driven (in the field of acoustic waves, only odd harmonics are measured), instead of the canonical , Cf is the previously reported sensitivity coefficient of the quartz crystal sensor, and ᐃm is the change in mass per unit area of the piezoelectrically active surface (ng/cm 2 ).
By applying Equation 2 to ᐃf values after SVL rinsing and after maximum NP uptake, it was possible to calculate the mass changes due to the SVL formation (mSVL) and NPs (mNP), respectively ( Figure   4B, main text). To validate such data processing, the Voigt model was applied to determine the SVL thickness (L, nm) and density (r, g/cm 3 ) before and after interaction with NPs. The mass per unit area adsorbed onto the sensor surface after SVL formation (mSVL) and after maximum NP uptake (mNP) ( Figure 4B, main text) was then derived from thickness and density data (m = L r).
Statistical data analysis. At least two replicates were performed for each QCM experiment.
Uncertainties on the mass changes shown in Figure 4B of the main text were processed by averaging the results obtained from the overtones of different replicates and by applying the Student's statistics (95 % confidence level, N≥10). The same statistics was used to calculate the vertical error bars referring to the reduction (%) in NP uptake shown in the same figure (N=18). Only for the SVL viscoelastic properties shown in Figure 3B (main text) and Figure S10, error bars correspond to standard deviation.   In the initial part of the curve, which is thicker, the sampling interval corresponded to 1 s, while in the following part it was decreased to 10 s. The entire QCM experiment recorded at 17 mol % cholesterol is shown in Figure S12. A second NP addition -identical to the first one -was performed to test whether saturation of the SVL portion exposed to NPs had been reached. Whereas a clear decrease in resonance frequency was observed after the first NP injection, the signal remained unperturbed after the second. This control measurement suggested that the amount of NPs inserted in QCM experiments (see above) was sufficient to reach membrane saturation.
Quantification of the lipid/NP ratio. The calculation of lipid/NP molar ratios was made in the same assumptions considered for the viscoelastic model, i.e. assuming no water loss from the vesicles during NP adsorption and invariance of vesicle number (see main text). The density r (g/cm 3 ) and thickness L (nm) of the viscoelastic vesicle layers (see Figure 3B, main text) were used to calculate the total mass adsorbed onto the sensor before the SVL-NP interaction, assuming complete coverage of the sensor surface (0.78 cm 2 ). The mass change induced by NP uptake and extracted from the viscoelastic model was used to determine the mass -and thus the number -of the embedded NPs