Lipophilicity of Cationic Ligands Promotes Irreversible Adsorption of Nanoparticles to Lipid Bilayers

Abstract

A mechanistic understanding of how the surface properties of engineered nanomaterials influence their interactions with cells is essential for designing materials for applications such as bioimaging and drug delivery, as well as for assessing nanomaterial safety. Ligand-coated gold nanoparticles have been widely investigated because their highly tunable surface properties enable investigations into the effect of ligand functionalization on interactions with biological systems. Lipophilic ligands have been linked to adverse biological outcomes through membrane disruption, but the relationship between ligand lipophilicity and membrane interactions is not well understood. Here, we use a library of cationic ligands coated on 2-nm gold nanoparticles to probe the impact of ligand end group lipophilicity on interactions with supported phosphatidylcholine lipid bilayers as a model for cytoplasmic membranes. Nanoparticle adsorption to and desorption from the model membranes were investigated by quartz crystal microbalance with dissipation monitoring. We find that nanoparticle adsorption to model membranes increases with ligand lipophilicity. The effects of ligand structure on gold nanoparticle attachment were further analyzed using atomistic molecular dynamics simulations, which showed that increasing ligand lipophilicity promotes ligand intercalation into the lipid bilayer. Together, the experimental and simulation results could be described by a two-state model that accounts for initial attachment and subsequent conversion to a quasi-irreversibly bound state. We find that only nanoparticles coated with the most lipophilic ligands in our nanoparticle library undergo conversion to the quasi-irreversible state. We propose that initial attachment is governed by interaction between the ligands and phospholipid tail groups, while conversion into the quasi-irreversibly bound state reflects ligand intercalation between phospholipid tail groups and eventual lipid extraction from the bilayer. Systematic variation of ligand lipophilicity enabled us to demonstrate that the lipophilicity of cationic ligands correlates with nanoparticle-bilayer adsorption and suggests that changing the nonpolar ligand R group promotes a mechanism of ligand intercalation into the bilayer associated with irreversible adsorption.

Graphical Abstract

INTRODUCTION

Ligand-coated engineered nanomaterials (ENMs) have been used in bioimaging, drug delivery, and consumer goods, inspiring investigations into understanding how ENMs interact with biological interfaces.^1,2 In particular, establishing relationships between the properties of ENMs and their interactions with cellular membranes is essential for designing safe ENMs.^3,4 For example, interactions between ENMs and cellular membranes can result in lipid extraction^5,6 and membrane disruption,^7,8 events that can lead to cytotoxicity.^7,9,10 However, predicting such behaviors from ligand properties remains challenging, inhibiting ENM design. Toward this end, ligand lipophilicity—the property quantifying the partitioning of a ligand between aqueous media and lipid, which correlates strongly with hydrophobicity¹¹—has been found to dictate ENM interactions with biological interfaces.^12–15 However, the specific interactions between lipophilic ENM ligands and cellular membranes—and the degree to which ligand lipophilicity drives these interactions—remain unclear.

Ligand-coated gold nanoparticles (AuNPs) have been used as model ENMs because of their ease of fabrication and tunable surface chemistry.¹⁶ AuNPs can be synthesized at sizes commensurate with biomolecules (< 10 nm), enabling the study of interactions with biological interfaces at the same length scale.¹⁷ To probe interactions between ENMs and cell membranes, supported lipid bilayers (SLBs) have been used as model membrane systems.^18,19 For example, SLBs have been used to link nanoparticle size, core composition, and surface chemistry to increased cellular interaction, internalization, and cytotoxicity.^7,9 Quartz crystal microbalance with dissipation monitoring (QCM-D) has been used to monitor ENM behavior at the supported lipid bilayer-solution interface through measuring changes in mass and energy dissipation of material coupled to the motion of the sensor.²⁰ Changes in energy dissipation reflect the viscoelasticity of laterally homogeneous adlayers such as SLBs or to the stiffness of particle–surface contacts.^20,21 QCM-D has been used to deduce mechanisms of peptide incorporation into SLBs,²² as well as the kinetics of ENM interactions with surfaces.^23,24

While the kinetics of AuNP adsorption onto SLBs can be quantified via QCM-D, other approaches are needed to reveal the molecular-scale AuNP–bilayer interactions that lead to adsorption. As a result, computational models, such as classical molecular dynamics (MD) simulations, have been used to simulate AuNP–bilayer interactions at multiple length scales.^14,25–28 Previous work used atomistic MD simulations to study the free energy of AuNP insertion into the bilayer for AuNPs coated with varying ratios of neutral octanethiol and negatively charged 11-mercapto-1-undecanesulphonate (MUS).²⁹ Inclusion of more lipophilic ligands (viz. octanethiol) lowered the free energy of AuNP insertion into the lipid bilayer, suggesting that AuNP lipophilicity is critical to fusion with the lipid bilayer.²⁹ Subsequent studies found that AuNP insertion was driven primarily by the lipid membrane core shielding lipophilic ligands from the aqueous environment.^29,30 These studies, in conjunction with related computational results,¹⁴ suggest that the proportion and spatial distribution of lipophilic ligands on AuNP surfaces represent important parameters that modulate potential AuNP insertion into the lipid bilayer. However, the influence of ligand structure, specifically the degree of lipophilicity, on the interaction of AuNPs with lipid membranes remains to be determined.

In this work, we combine QCM-D experiments and atomistic MD simulations to systematically investigate the influence of ligand lipophilicity on AuNP interactions with single-component lipid bilayers. The adsorption of AuNPs to lipid bilayers and their long-term attachment stability increased with ligand lipophilicity. Simulations revealed a mechanism for the long-term stability observed for AuNPs with increased ligand lipophilicity: hydrophobic contacts between ligand lipophilic groups and the lipid tail groups drive the intercalation of more lipophilic ligands into the bilayer or lipid extraction from the bilayer. Kinetic analysis of the QCM-D data accounting for reversible and quasi-irreversible AuNP adsorption showed agreement with free energy calculations of AuNP–SLB systems. Our findings demonstrate that AuNPs with increased ligand lipophilicity have increased rates of conversion to a quasi-irreversibly bound state, providing design rules for developing ENMs with tailored membrane interactions.

RESULTS AND DISCUSSION

Ligand R Group Lipophilicity Governs AuNP Adsorption to Lipid Bilayers.

Figure 1a shows the library of cationic ligands used to functionalize the 2-nm diameter AuNPs employed in this study. As described in the Methods, ligand lipophilicity is expressed as the equilibrium partition coefficient between phosphatidylcholine liposomes and water (K_lip-w) calculated for each ligand R group: methyl (C₁), ethyl (C₂), butyl (C₄), benzyl (Bn), and decyl (C₁₀).¹¹ Larger log K_lip-w values correspond to a higher propensity for the ligand R group to partition into phosphatidylcholine liposomes from water. By modulating only the ligand R group, we systematically study the effects of R group lipophilicity while keeping the gold core size, charge, and ligand backbone constant. We determined AuNP hydrodynamic diameter and apparent zeta potential by dynamic light scattering and laser Doppler electrophoresis, respectively. Figure S8 shows that in the aqueous solution used for our experiments (10 mM NaCl buffered to pH 7.4 with 10 mM HEPES), the hydrodynamic diameters, and therefore degrees of aggregation, of the AuNPs in our library were comparable. The only statistically significant difference in hydrodynamic diameter was between the Bn- and C₁₀-AuNPs (p = 0.314), which we attribute to differences in the polarizability and flexibility of the ligand R groups (i.e., higher polarizability of the benzyl group leading to stronger van der Waals interactions, higher flexibility of the C₁₀ R group allowing it to minimize solvent exposure by folding back on itself). The apparent zeta potentials of the AuNPs in our library were statistically indistinguishable and did not differ from zero (p > 0.05, Figure S8d; zeta potential mostly between −10 and +10 mV).³¹ Given the near neutral apparent zeta potentials of the AuNPs, we do not expect electrostatics to dominate their interaction with bilayers. Similar libraries of AuNPs have been used to correlate cytotoxicity and hydrophobicity.¹⁵

Figure 1.

Experimental and computational systems used to study gold nanoparticle adsorption onto phospholipid bilayers. (a) Ligands are comprised of an alkane group (gray), an oligo(ethylene glycol) spacer group (green), and a cationic quaternary ammonium group (red) substituted with the indicated R group and two methyl groups. The five R groups used are displayed in red and labeled with their calculated log K_lip-w values in parentheses. (b) Schematic of the system used in quartz crystal microbalance experiments to measure nanoparticle adsorption to supported DOPC lipid bilayers. (c) Snapshot of 2-nm gold nanoparticle with C₁₀ ligands placed above a DOPC lipid bilayer. The color scheme is illustrated for each of the components at right. The DOPC lipids are comprised of a zwitterionic phosphatidylcholine head group and nonpolar acyl tails consisting primarily of aliphatic carbon atoms. Water is shown in grey.

As described in the Methods, SLBs were prepared from 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC) using the vesicle fusion method,³² after which AuNPs (10 nM) were flowed over DOPC lipid bilayers for 20 minutes (schematically represented in Figure 1b) followed by rinsing with AuNP-free solution for extended periods of time. We characterize AuNP-SLB interactions at two distinct time points: after 20 minutes flow—experimentally determined as sufficient time to attain an adsorption plateau—and after extended rinsing. We report QCM-D results as acoustic surface mass density (Γ) and the change in energy dissipation (ΔD). We obtained Γ from frequency changes using the Sauerbrey equation.²⁰

Figure 2a shows the acoustic surface mass density for the AuNPs after 20 minutes flow (Γ_max) and after rinsing (Γ_rinse). We find that C₂−, C₄−, Bn-, and C₁₀-AuNPs adsorb to the DOPC bilayer, whereas any adsorption of C₁-AuNPs was not detectable. For the series of AuNP ligands used, Γ_max correlates positively with log K_lip-w. For C₂-AuNPs, Γ_rinse = 0, indicating that these AuNPs reversibly adsorb to the lipid bilayer and rinse away using buffer. In contrast, Γ_rinse > 0 for C₄−, Bn-, and C₁₀-AuNPs, suggesting that a population of these AuNPs remain quasi-irreversibly bound to the bilayer. Given the similarities in particle core size, apparent zeta potential, ligand structure (with the exception of the R group), and hydrodynamic diameter, we expect that the nature of the ligand R group was the dominant contributor to AuNP–bilayer interactions; however, we cannot exclude that aggregation state had a minor influence in the case of the Bn-AuNPs. Increases in AuNP-bilayer interaction with increasing lipophilicity is consistent with established ideas of lipophilic ligand-mediated NP toxicity.¹⁵

Figure 2.

Influence of ligand lipophilicity on AuNP attachment to supported DOPC bilayers as determined by QCM-D. (a) Acoustic surface mass density (Γ) maximum and after rinse. (b) Dissipation factor (ΔD) maximum and after rinse. Error bars represent one standard deviation from four replicate QCM-D experiments.

Figure 2b shows change in energy dissipation after 20 minutes flow (ΔD_max) and after rinse (ΔD_rinse). The C₁-AuNPs do not effect a detectable dissipation change, consistent with the lack of observed mass attachment (Figure 2a). For C₂-AuNPs, zero Γ_rinse corresponds with nonzero ΔD_rinse. Without a quantifiable population of AuNPs adsorbed to the SLB, either small undetectable populations of AuNPs induce detectable viscoelastic changes to the bilayer, or AuNP adsorption induces a permanent change in the energy dissipation of the bilayer that persists after AuNPs have rinsed away. For C₄-, Bn-, and C₁₀-AuNPs, ΔD_max is statistically indistinguishable from ΔD_rinse (p < 0.05). A population of reversibly adsorbed AuNPs leave the bilayer upon rinsing without producing a detectable change in energy dissipation. From the QCM-D results, we hypothesize that the lipophilicity of C₂−, C₄−, Bn-, and C₁₀-AuNPs leads to spontaneous adsorption onto DOPC lipid bilayers and subsequent formation of a quasi-irreversibly bound state in a subset of the adsorbed AuNP population. Ligand lipophilicity appears to determine the degree of quasi-irreversible interaction, with the least lipophilic ligands leading to negligible quasi-irreversible adsorption. We employ classical molecular dynamics simulations to measure the free energy barriers to forming these states and to gain insight into the mechanism of quasi-irreversibly binding.

Contact between Lipophilic Ligand Groups and Lipid Tails Promotes Adsorption.

We first modeled interactions between C₁− and C₁₀-AuNPs and DOPC lipid bilayers to understand the molecular interactions that drive AuNP adsorption. These AuNPs were selected because the ligands represent the extremes of log K_lip-w in this study and the AuNPs exhibit distinct adsorption behavior: C₁-AuNPs do not adsorb to DOPC bilayers, whereas C₁₀-AuNPs adsorb quasi-irreversibly (Figure 2a). We first quantified potentials of mean force (PMFs) for C₁− and C₁₀-AuNP adsorption using umbrella sampling (US) simulations. The PMF measures the free energy as a function of a collective variable, z, which we define as the distance in the direction normal to the bilayer between the gold core and DOPC center-of-mass (Figure 1c) following past literature.^25–28 We calculated PMFs using initial simulation configurations generated by pulling the AuNP either from an initial position in water toward the bilayer (decreasing-z simulations) or from an initial position in the bilayer toward water (increasing-z simulations). These two methods were used to interrogate potential hysteresis associated with long timescale bilayer rearrangements.^33,34

Figure 3 compares decreasing-z and increasing-z PMFs for C₁− and C₁₀-AuNPs. Both sets of PMFs are comparable: the decreasing-z PMFs monotonically increase as z decreases, whereas the increasing-z PMFs exhibit free energy minima at positions near the water-bilayer interface (z ≈ 5 nm). Simulation snapshots indicate that the free energy minima correspond to configurations in which nonpolar lipid tail groups are in contact with lipophilic groups on the C₁− and C₁₀-AuNPs, suggesting that these favorable contacts lead to thermodynamically preferred adsorbed states for both AuNPs. These minima are consistent with the experimentally observed quasi-irreversibly bound states for C₁₀-AuNPs but do not explain the inability of the C₁-AuNPs to adsorb. However, the pronounced hysteresis between the increasing-z and decreasing-z PMFs suggests that using z as a collective variable does not capture potential free energy barriers that could inhibit adsorption, as previously observed in simulations of lipid insertion into a bilayer.³⁵

Figure 3.

Free energy as a function of the z distance between the AuNP and lipid bilayer. Potential mean force (PMF) versus z for C₁− and C₁₀-AuNPs when the gold core is (a) pulled towards (i.e. decreasing-z) and (b) away from (i.e. increasing-z) the DOPC lipid bilayer. Simulation snapshots show the last configuration from umbrella sampling simulations of C₁₀-AuNPs for different values of z. Water and chlorine atoms are omitted for clarity. Legends are the same for (a) and (b). (c) Number of hydrophobic contacts (c_h) versus z for both decreasing- and increasing-z simulations. Hydrophobic contacts are defined as the number of contacts between nonpolar groups in the ligands and in the DOPC tail groups. Error bars are reported as the standard deviation between two 20 ns blocks for each umbrella sampling window.

Unbiased Simulations Reveal Two Mechanisms Leading to Prolonged AuNP Adsorption to Bilayers.

We hypothesized that barriers to the formation of hydrophobic contacts between lipids and the AuNP could explain the differences between the increasing-z and decreasing-z PMFs; previous studies have also identified hydrophobic contacts as important for favorable AuNP-bilayer interactions.^30,36,37 Therefore, we calculated the total number of hydrophobic contacts (c_h) between the alkane and R groups of the ligands and the tail groups of DOPC. Figure 3c plots c_h as a function of z for both increasing-z and decreasing-z simulations. For both AuNPs, many hydrophobic contacts persist during the increasing-z simulations, indicating that hydrophobic contacts are highly favorable. Conversely, hydrophobic contacts are observed for only small values of z during the decreasing-z simulations, indicating the presence of hidden barriers that prevent contacts from forming. To confirm that hydrophobic contacts are important for adsorption, we performed 50-ns unbiased simulations initiated from configurations with different z and c_h values to determine if the C₁− and C₁₀-AuNPs desorb from the bilayer. Figure 4a shows that both C₁− and C₁₀-AuNPs remain adsorbed to the lipid bilayer if the initial value of c_h exceeds ~40 contacts, even for large values of z. These unbiased simulations are consistent with the hypothesis that hydrophobic contacts between the AuNP ligands and lipid membrane are important for adsorption but not captured by z alone.

Figure 4.

Unbiased simulations initiated from umbrella sampling trajectories. (a) Number of hydrophobic contacts (c_h) versus z for unbiased simulations initiated from increasing- and decreasing-z US configurations for C₁− and C₁₀-AuNPs. AuNPs are considered adsorbed if z < 6 nm for the last 10 ns of the unbiased simulation (filled markers) and desorbed if z > 6 nm (hollow markers). Points in the dashed blue box were for unbiased simulations initiated from increasing-z US configuration, all other points were for simulations initiated from decreasing-z US configurations. (b) Simulation snapshots after 50 ns of unbiased simulation for C₁− and C₁₀-AuNPs. The initial z values and final z values after 50 ns are labeled above the snapshots. Atoms in the DOPC head groups are omitted for clarity.

For unbiased simulations in which the AuNPs remain adsorbed, z increases until z ≈ 5 nm, thus reaching states consistent with the free energy minima obtained from the increasing-z PMFs. Figure 4b shows final snapshots from the unbiased simulations of C₁− and C₁₀-AuNPs for z ≈ 2 nm. The snapshots show two mechanisms that promote continued AuNP adsorption: (1) the C₁-AuNP remains adsorbed due to lipid extraction, where lipids are pulled away from the bilayer, and (2) the C₁₀-AuNP remains adsorbed due to both lipid extraction and ligand intercalation, where some C₁₀ ligands extend into the hydrophobic core of the bilayer. In both mechanisms, hydrophobic contacts drive the rearrangement of lipids to facilitate AuNP adsorption. C₁ ligands are less capable of facilitating ligand intercalation relative to C₁₀ ligands, because the C₁ hydrophobic chain length is shorter than C₁₀.

Ligand Intercalation within Bilayer Reduces Barrier for Forming Hydrophobic Contacts.

The results from the US and unbiased simulations suggest that hydrophobic contacts are important for AuNP adsorption but are not sampled when using z as a collective variable. We therefore performed US using c_h as a collective variable to calculate a corresponding PMF. We included Bn-AuNP in this analysis, which has intermediate ligand lipophilicity compared to the other ligands and adsorbs quasi-irreversibly (Figure 2a). Figure 5a shows PMFs for C₁-, Bn-, and C₁₀-AuNPs as a function of c_h. The PMF for the C₁-AuNP monotonically increases with c_h, indicating that the initial formation of hydrophobic contacts is unfavorable. The PMF for Bn-AuNP also monotonically increases with larger c_h values, but with a substantially lower slope than C₁-AuNP, which indicates that the more lipophilic Bn ligands reduce the free energy barrier to forming hydrophobic contacts. The PMF for C₁₀-AuNPs has an initial free energy barrier at c_h = 5 contacts and a local minimum at c_h ≈ 40 contacts, then increases monotonically at larger c_h values. The local minimum indicates a metastable state due to the formation of favorable hydrophobic contacts. This metastable state for C₁₀-AuNPs may lead to subsequent ligand intercalation or lipid extraction steps that result in the quasi-irreversibly bound states observed from experiments and the global minimum indicated by the increasing-z PMFs (Figure 3b).

Figure 5.

Free energy as a function of AuNP-bilayer hydrophobic contacts. (a) Potential of mean force (PMF) versus the number of hydrophobic contacts (c_h) for C₁−, Bn-, and C₁₀-AuNPs. Error bars are reported as the standard deviation between two 30 ns trajectories for C₁− and C₁₀-AuNPs and two 20 ns trajectories for Bn-AuNP in each umbrella sampling window. (b) Simulation snapshots with c_h = 5, 40, and 150 for C₁−, Bn-, and C₁₀-AuNPs. DOPC lipids that are within 0.35 nm of the ligand atoms are highlighted in cyan.

Figure 5b shows simulation snapshots of C₁−, Bn-, and C₁₀-AuNPs for various c_h values. For C₁-AuNPs, ligands contact the bilayer for c_h ≤ 30 contacts, then a single lipid molecule is extracted for c_h ≥ 40 contacts. At c_h = 150 contacts, the C₁-AuNP desorbs from the DOPC lipid membrane even after extracting two lipid molecules, suggesting that lipid extraction is not sufficiently favorable to promote adsorption without more substantial bilayer deformations (like those observed in Figure 4b). Conversely, the snapshots of the Bn- and C₁₀-AuNPs show that ligands intercalate within the bilayer. For the Bn-AuNP, multiple ligands intercalate within the bilayer with increasing c_h. We attribute the smaller slope of the Bn-AuNP PMF compared to the C₁-AuNP PMF to favorable intercalation. For the C₁₀-AuNP, a single ligand intercalates within the bilayer at c_h ≈ 40. This value of c_h corresponds to the local minimum in the PMF and is comparable to the threshold for stable adsorption identified from unbiased simulations (Figure 4a). Unlike the Bn-AuNP, multiple C₁₀ ligands only intercalate within the bilayer for large values of c_h (≈ 150 contacts). We further tested if ligand intercalation is sufficient to keep the C₁₀-AuNP adsorbed to the DOPC lipid bilayer. We performed four unbiased simulations of each of the three AuNPs initiated with c_h = 40 and found that the C₁₀-AuNPs remain adsorbed, whereas most C₁-AuNPs rapidly desorb and Bn-AuNPs desorb but at a slower rate than C₁-AuNPs (Supporting Information, Figure S22).

Taken together, the simulation results suggest that C₁−, Bn- and C₁₀-AuNPs can favorably adsorb to the bilayer if sufficient hydrophobic contacts are formed between the AuNP and the bilayer (Figure 3b). However, C₁-AuNPs can form hydrophobic contacts with the bilayer only via a lipid extraction mechanism that is associated with a large free energy barrier (Figure 5). This barrier thus inhibits adsorption in agreement with experimental measurements. Conversely, the more lipophilic R groups present in the Bn and C₁₀ ligands intercalate within the bilayer to promote the formation of hydrophobic contacts (Figure 5). Additional snapshots illustrating these two mechanisms are shown in the Supporting Information, Figure S20. Intercalation could facilitate the initial favorable adsorption of Bn- and C₁₀-AuNPs followed by longer timescale interconversion to a quasi-irreversibly adsorbed state associated with many hydrophobic contacts (Figure 2a). Furthermore, ligand intercalation for C₁₀-AuNPs could explain the experimentally observed bilayer viscoelastic change (Figure 2b).

Two-state Adsorption Kinetics Describes AuNP Reversibly and Quasi-Irreversibly Adsorbed States.

We constructed an analytical kinetic model to investigate the connection between the findings from MD simulations and the experimentally observed quasi-irreversible binding of AuNPs to lipid bilayers. QCM-D cannot directly measure the extent of ligand intercalation, but it does allow two distinct adsorbed states to be distinguished.^23,24,38 We define a three-state model to describe the kinetics of AuNP-SLB interactions which includes a (1) metastable, reversibly adsorbed state, which can undergo either further ligand intercalation, lipid extraction, or desorption; (2) quasi-irreversibly adsorbed state, which corresponds to a high number of hydrophobic contacts consistent with a global free energy minimum (Figure 3b); and (3) desorbed state, which could reflect barriers to forming initial hydrophobic contacts necessary for adsorption.

The two adsorbed states represent a reversibly adsorbed (Γ_α) and a quasi-irreversibly adsorbed (Γ_β) population of nanoparticles. At any given time, the total mass of adsorbed particles equals the sum of the masses of the reversibly and quasi-irreversibly adsorbed particles: Γ_total (t) = Γ_α (t) + Γ_β (t). Assuming first-order adsorption kinetics, the rate of mass adsorption (the first derivative of mass adsorption with respect to time) can be expressed as a function of the first-order rates of adsorption (k_a), conversion to quasi-irreversibly adsorbed state (k_β), and desorption (k_d). Equation (1) describes the adsorption kinetics, as modified from Zhang et al.²³

$$\frac{{\text{d}\Gamma }_{\alpha }}{\text{dt}}=k_{\text{a}}{m}_{\text{s}}\frac{{\Gamma }_{\text{max}}-{\Gamma }_{\text{total}}(t)}{{\Gamma }_{\text{max}}}-k_{\text{d}}{\theta \Gamma }_{\text{total}}(t)-k_{\beta }{\theta \Gamma }_{\text{total}}(t)$$ (1)

where m_s is the mass density of AuNPs in solution, and θ is the ratio of reversibly to total adsorbed particles, approximated as $\theta =\frac{{\Gamma }_{\text{rinse}}}{{\Gamma }_{\text{max}}}$. The adsorption capacity of the bilayer was taken as Γ_max. Particles that rinse from the bilayer were considered to have been in the reversibly adsorbed α state. Equation (2) describes the desorption of AuNPs from a SLB after AuNPs have been removed from solution.

$$\frac{\text{d}\Gamma }{\text{dt}}=-k_{\text{d}}{\Gamma }_{\text{total}}(t)$$ (2)

To determine the rate coefficient for conversion to the quasi-irreversible adsorption state (k_β), we consider the rate of change of Γ_β independently described by Equation (3).

$$\frac{{\text{d}\Gamma }_{\beta }}{\text{dt}}=k_{\beta }{\theta \Gamma }_{\text{total}}(t)$$ (3)

We have two distinct time points for Γ_β (t = 0 min and t = 20 min). By integrating over the total time of NP-bilayer exposure (t_max = 20 min), we determine k_β assuming first-order kinetics:

$$k_{\beta }=\frac{{\Gamma }_{\beta }}{{\Gamma }_{\text{total}}\theta t}$$ (4)

Figure 6 shows the rate coefficients k_a, k_β, and k_d for the different AuNPs. The rate coefficient for adsorption, k_a, increases monotonically with increasing ligand R group lipophilicity as expressed by log K_lip–w (Figure 6a). k_a is dictated by the free energy of binding, which is positively correlated to ligand lipophilicity. The rate coefficient for conversion from the reversible to the quasi-irreversible adsorbed state, k_β, appears to increase with lipophilicity for C₂−, C₄−, Bn- and C₁₀-AuNPs (Figure 6b). After initial adsorption, subsequent ligand intercalation into the bilayer and lipid extraction from the bilayer facilitates the formation of a quasi-irreversibly adsorbed state (Figure 4); therefore, we expect that k_β is dictated by the free energy of hydrophobic contact between ligand R groups and DOPC tail groups. As explored in the MD simulations, increasing ligand R group lipophilicity leads to a decreased free energy barrier for forming hydrophobic contacts (Figure 5). The rate coefficient for desorption of AuNPs in the reversibly adsorbed state, k_d, appears to be independent of ligand R group lipophilicity (Figure 6c), suggesting the mechanism for desorption remains constant between AuNPs of varying ligand lipophilicity. These findings indicate that selection of ligand R group lipophilicity could drive reversible or quasi-irreversible AuNP adsorption onto DOPC bilayers.

Figure 6.

The role of ligand end group lipophilicity on adsorption to and desorption from phospholipid bilayers. (a) Adsorption rate constant (k_a) calculated from AuNP adsorption with calculated desorption rate constant (k_d) values. (b) Rate constants for conversion to quasi-irreversibly adsorbed state (k_β) calculated from mass at maximum and after rinse. (c) Desorption rate constants (k_d) calculated from AuNP desorption. Error bars represent one standard deviation of four replicate QCM-D measurements. (d) Schematic showing hypothesized mechanism for preferential adsorption of C₁₀-AuNPs compared to C₁-AuNPs. C₁₀-AuNPs have a longer R group, denoted by the red lines. Alkane is shown as black lines and PEG is shown as green lines. The symbol and color for the gold core and lipid bilayer is the same as Figure 1b.

CONCLUSIONS

We explored the role of ligand lipophilicity on the adsorption of 2-nm diameter AuNPs to zwitterionic phospholipid bilayers using QCM-D measurements and atomistic MD simulations. The experiments indicated that AuNPs coated with lipophilic ligands can adsorb quasi-irreversibly to DOPC bilayers. The extent of conversion to the quasi-irreversibly adsorbed state scales with ligand functional group lipophilicity. The simulations revealed that AuNP adsorption depends on the number of hydrophobic contacts between the AuNP ligands and the phospholipid acyl chains, which can emerge either due to the intercalation of ligands within the bilayer or the extraction of lipids from the bilayer. Increasing ligand end group lipophilicity promotes intercalation within the bilayer rather than lipid extraction, which reduces the free energy barrier for forming hydrophobic contacts. Analytical modeling of experimental AuNP-bilayer interaction kinetics relates the MD findings to the formation of both reversibly and quasi-irreversibly adsorbed states.

Together, the experiments and simulations suggest a mechanism for AuNP adsorption that depends critically on ligand lipophilicity (Figure 6d). The first step for adsorption is hydrophobic contact between the ligand R group and DOPC tail groups. Since Bn and C₁₀ ligands have lipophilic R groups, they can intercalate within the bilayer to promote hydrophobic contacts after overcoming a ~30 kJ·mol⁻¹ free energy barrier. At long time scales, Bn- and C₁₀-AuNPs form a quasi-irreversible adsorbed state corresponding to a large number of hydrophobic contacts, consistent with the global free energy minimum in Figure 3b. Conversely, C₁ ligands are unable to make sufficient hydrophobic contacts to maintain adsorption due to the large barrier required for lipid extraction (Figure 5), resulting in desorption. Our results show that functionalized nanomaterials coated with lipophilic end group ligands could have enhanced adsorption and long-term stability on cell membranes. Tuning nanomaterial ligand lipophilicity may provide a means to enhance targeted nanodrug delivery through selective intercalation of ligands into cell membranes; alternatively, decreasing ligand lipophilicity may allow for creation of environmentally benign nanomaterials.

MATERIALS AND METHODS

Ligand Synthesis.

Figure S1 depicts the synthesis pathway for the ligands used in AuNP functionalization. Compound (1) was synthesized as described by Miranda et al.³⁹ Compound (1) (1.0 g, 2 mmol, 1 eq) was dissolved in 5 mL of ethyl acetate in a 20 mL vial. To each vial, we added 30 eq of the corresponding amine, and the entire solution was sealed properly and heated gently to 50 °C for 3–5 days. Afterward, the solvent was evaporated and the residue was washed three times by hexanes, heptanes, or a 1:1 mixture of hexanes and heptanes to obtain (2) as a yellowish oil-like liquid.

We dissolved 200 mg of (2) in 3 mL of dichloromethane under a nitrogen atmosphere. To the solution we added 20 eq of trifluoroacetic acid followed by addition of 1.2 eq of triisopropylsilane. The solution was stirred at room temperature overnight. Afterward, the solvent was removed, and the residue was washed with hexane or heptane three times and ether for three more times to obtain ligands as colorless or yellowish liquid. To validate the ligand structure, ¹H NMR spectroscopy of ligands was performed. Spectra are shown in the Supporting Information, Figures S2–S6.

Gold Nanoparticle Synthesis.

Gold nanoparticles were synthesized by the Brust-Schiffrin two-phase method as described in previous literature.^40,41 In brief, 1 g of HAuCl₄ was dissolved in 300 mL 1:1 water toluene. We added 2.1 g of tetraoctylammonium bromide directly with maximum stirring speed. We added 0.7 mL of pentanethiol dropwise until the whole solution turned into white. Then 2.0 g of sodium borohydride was dissolved in around 8 mL of water and immediately added into the white solution. After stirring overnight, the organic layer was separated and dried under reduced pressure in room temperature. The residue was precipitated in cold ethanol and re-dissolved in hexanes. The solution washed with acetonitrile 120 times until all TOAB was fully removed to obtain gold core.

Ligand Exchange Reaction.

Gold cores (40 mg) were dissolved in 4 mL dichloromethane under nitrogen atmosphere. To this solution we added 120 mg of the corresponding ligand in a mixed dichloromethane/methanol (2 mL/2 mL) solution in a dropwise manner under nitrogen and stirred the mixture for 72 h at room temperature. Solvents were removed under reduced pressure, and precipitations were washed with hexanes three times, and hexanes/dichloromethane mixture (1:1 v/v), or pure dichloromethane three times. The solid was suspended in ultrapure water, dialyzed for 3 days, and concentrated by ultracentrifugation. The concentration of gold nanoparticles was determined based on the absorption at 506 nm as previously reported by Haiss et al.⁴² Adsorption spectra and compiled absorbance at 506 for all AuNPs are shown in Supporting Information, Figure S7 and Table S1 respectively.

Characterization of AuNP Hydrodynamic and Electrokinetic Properties.

The hydrodynamic diameters and apparent zeta potentials of the AuNPs were determined by dynamic light scattering and laser Doppler electrophoresis, respectively (Malvern Zetasizer Nano ZS). Gold NPs (10 nM) were suspended in water or 10 mM HEPES 10 mM NaCl buffered at pH 7.4 for 30 minutes prior to measurement (Supporting Information, Figure S8). In preliminary experiments, we determined that the AuNPs remained colloidally stable in 10 mM NaCl for the duration of the QCM-D experiments. We therefore selected this salt concentration to study AuNP interaction with bilayers. Higher salt concentrations led to much more pronounced aggregation and destabilized the colloidal suspension.

Calculation of Ligand R group Lipophilicity.

We calculated the lipid–water partition coefficient (K_lip-w) of ligand R groups as a measure of ligand lipophilicity using the poly-parameter linear free energy relationship:¹¹

$$\text{log}{K}_{lip-w}=c+eE+sS+aA+bB+vV$$ (5)

where E, S, A, B, and V describe the ligand head group excess molar refraction, dipolarity/polarizability, H-bond acidity, H-bond basicity, and molar volume, respectively, and the corresponding lowercase letters are specific for the water–lipid partitioning system.¹¹ The constant c is obtained from the multiple linear regression used to establish the system descriptors. The chemical descriptors for the ligand R groups (E, S, A, B, and V) were obtained from the Helmholtz Centre for Environmental Research Linear Solvation Energy Relationship.⁴³ Values for the systems parameters c, e, s, a, b, and v were taken from equation 3 of Endo et al.¹¹ The values for the chemical descriptors and system parameters are compiled in Supporting Information, Table S2.

Vesicle Preparation and SLB Formation.

Solutions for all experiments employing lipids were buffered to pH 7.4 with 10 mM HEPES. Vesicles were formed from 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC; Avanti Polar Lipids, 850375) via vesicle extrusion^44,45. In short, chloroform was removed from DOPC by evaporation in a vacuum chamber for 1 h. We resuspended DOPC (2.5 mg·mL⁻¹) in 1 mM NaCl. To form vesicles, lipid suspensions were sonicated for 30 min followed by three freeze–thaw cycles (incubation in liquid nitrogen for 5 min followed by sonication at room temperature for 5 min). Vesicles were extruded 11 times through 50 nm polycarbonate filters. The DOPC vesicles had hydrodynamic diameters between 90 and 110 nm as determined by DLS and ζ-potentials of 0 to −6 mV as determined by laser Doppler electrophoresis. Vesicles were stored at 4 °C and used within 10 days.

Supported lipid bilayers were formed on SiO₂-coated QCM-D sensors (QSX203) in a Q-Sense E4 instrument (Biolin Scientific) by vesicle fusion.^20,45 In short, vesicles (0.125 mg·mL⁻¹) in 100 mM NaCl were flowed (0.100 mL·min⁻¹) over sensors freshly cleaned in an UV/ozone chamber for 20 min. The DOPC vesicles attained a critical surface concentration on the sensors after ~5 min, at which point the vesicles fused and ruptured and a stable bilayer was formed. After signal stabilization, the stable bilayer was rinsed for 10 min with 100 mM NaCl to remove any loosely adhered vesicles. Example frequency and energy dissipation traces for bilayer formation are shown in Supporting Information, Figure S9. The final frequency change for the bilayers was 25 ± 0.5 Hz and the dissipation factor was 0.2 (± 0.1) × 10⁻⁶, consistent with values previosuly reported for supported DOPC bilayers.⁴⁴

Nanoparticle Adsorption and Desorption Experiments.

After formation of stable DOPC bilayers on SiO₂-coated QCM-D sensors, we flowed 10 mM NaCl over bilayer until a stable baseline was achieved. We then introduced AuNPs (10 nM) at the same flow rate and in solution of the same composition until a stable plateau in frequency was attained (within 20 min). At this point, AuNP-free solution was introduced into the flow cell and the bilayer was rinsed until a stable baseline was observed. In preliminary experiments with C₁₀-AuNPs, we determined that a AuNP concentration of 10 nM and an exposure time of 20 min was sufficient for a plateau in frequency to be attained. Figure S10 shows changes in acoustic surface mass density and energy dissipation upon exposure of supported lipid bilayers to AuNPs decorated with each of the indicated ligands studied.

We calculated surface mass densities from frequency shifts (Δf_n) using the Sauerbrey equation:⁴⁶

$${\Gamma }_{\text{QCMD}}=-C\frac{\Delta f_n}{n}$$ (6)

where C is the mass sensitivity constant and n is the harmonic number.⁴⁶ Data for AuNP-bilayer interaction were determined to fall within the Sauerbrey regime, $4\times {10}^{-7}H{z}^{-1}>>\frac{-\Delta D_n}{\Delta f_{n_{/n}}}$ for a 5 MHz crystal.²⁰ All data taken between a bilayer baseline and final baseline fit the Sauerbrey regime. For all analysis the 5^th harmonic was used. We calculated the maximum surface mass density, Γ_max, at the end of the adsorption phase. We determined $\frac{\text{d}\Gamma }{\text{dt}}$ by taking the derivative of Γ with respect to time: a linear regression algorithm using 33 points (~30 s) centered around one point was used to calculate the derivative. To model the rate coefficients, k_a and k_d, a single-parameter optimized least squares model was used with equations 1 and 2 respectively. Adsorption data were fitted starting from the time the AuNP nanoparticle suspension had displaced AuNP-free solution in the flow chamber (indicated by a positive peak in $\frac{\text{d}\Gamma }{\text{dt}}$) and ending when the AuNP suspension had been displaced from the flow chamber by AuNP-free solution (indicated by a negative peak in $\frac{\text{d}\Gamma }{\text{dt}}$). The time period over which adsorption data were fitted (14 min) corresponds to the duration of AuNP flow minus the time of AuNP addition and removal from the flow cell. Desorption data were fitted for 8 min after observation of the negative peak in $\frac{\text{d}\Gamma }{\text{dt}}$ associated with displacement of the AuNP suspension from the flow chamber. The results of these fits are shown in Supporting Information, Figure S11 and Figure S12 for adsorption and desorption curves respectively. Goodness of fit was calculated as a non-linear R² value and reported in Supporting Information, Table S3.

System Setup for Classical MD Simulations.

Interactions of AuNPs with DOPC bilayers were modeled with classical MD simulations using Gromacs 2016.⁴⁷ The simulation workflow for developing the AuNP-DOPC systems is summarized in Supporting Information, Figure S13. AuNPs were modeled using a self-assembly protocol described previously,⁴⁸ which outputs 2-nm diameter AuNPs (226 gold atoms) with 83 ligands. AuNPs were modeled by using the INTERFACE force field⁴⁹ for gold atoms and the CHARM36/CGenFF force fields^50,51 for the ligands. The AuNPs were solvated with the TIP3P water model⁵² with sufficient chlorine counterions to ensure the system is charge-neutral then equilibrated for 5 ns at a temperature T = 300 K (controlled by a velocity-rescale thermostat) and pressure P = 1 bar (controlled by a Berendsen barostat). A 50 ns NPT simulation was subsequently performed at the same temperature and pressure, controlled by the same thermostat and Parrinello-Rahman barostat. The last configuration of the 50 ns NPT simulation was used to initiate AuNP-DOPC lipid membrane simulations.

The DOPC bilayer was generated using CHARMM36-GUI web-interface⁵³ with 196 lipids in each of the top and bottom leaflets. The dimensions of the DOPC bilayer were selected to avoid interactions between AuNPs due to the periodic boundary conditions (see Supporting Information, Figure S14). The bilayer was equilibrated with water at T = 300 K with semi-isotropic pressure coupling in the x-y dimensions at P = 1 bar, controlled by Nose-Hoover thermostat and Parinello-Rahman barostat. One AuNP was then inserted into the DOPC system such that the gold core center-of-mass was 6.9 nm away from DOPC center-of-mass and solvated with water (Figure 1c). The AuNP-DOPC lipid membrane system was then equilibrated for 10 ns in the NPT ensemble with the AuNP restrained at a 6.9 nm distance from the surface of DOPC with a harmonic potential of 2000 kJ·mol⁻¹·nm⁻² controlled by the Berendsen thermostat and barostat.

Increasing- and Decreasing-z US Simulations and Subsequent Unbiased Simulations.

We defined z as the center-of-mass distance between the gold core and DOPC bilayer. Initial configurations for the decreasing-z US simulations were generated by pulling the gold core towards the bilayer at 0.0005 nm·ps⁻¹ using a harmonic potential with a spring constant of 2000 kJ·mol⁻¹·nm⁻². Pulling was performed in the NPT ensemble at T = 300 K and P = 1 bar, controlled by the velocity-rescale thermostat and semiisotropic Berendsen barostat. Configurations from this trajectory were used to perform the decreasing-z US simulations with z varying from 1.3 nm to 6.5 nm. Initial configurations for the increasing-z US simulations were generated by pulling the gold core away from the bilayer starting from the final configuration of the z = 1.3 nm window from the decreasing-z US simulations. This pulling simulation used the same harmonic potential and pull rate as the decreasing-z pulling simulations. Configurations from this trajectory were used to perform increasing-z US simulations with z varying from 1.5 to 6.5 nm.

All US simulation windows were equilibrated for 500 ps in the NPT ensemble using the velocity-rescale thermostat and Berendsen barostat then simulated for 50 ns using the same thermostat and the Parinello-Rahman barostat. Some windows were extended to an additional 100 ns as discussed in the Supporting Information. The last 40 ns of the production simulation for each window were used to compute the PMF with the Weighted Histogram Analysis Method (WHAM).⁵⁴ Additional details on the US protocol and number of simulation windows are provided in the Supporting Information.

Unbiased MD simulations were performed using the 50 ns configuration from either a decreasing- or increasing-z US simulation window. The configuration was equilibrated for 500 ps with the z value restrained using the same spring constant as US simulations. Then, a 50 ns unbiased NPT simulation was performed at P = 1 bar controlled by the Parinello-Rahman barostat and T = 300 K controlled by the velocity-rescale thermostat.

Quantifying the Number of Hydrophobic Contacts.

The total number of hydrophobic contacts (c_h) was defined in Equation 7 by summing all possible contacts between alkane and R group atoms of the ligands (i) and tail group atoms of DOPC (i).

$$c_h=\sum _i\sum _j{s}_{ij}$$ (7)

Equation 8 defines S_ij as a continuous function that smoothly decays between 1 (corresponding to a hydrophobic contact) and 0 as a function of distance between atoms i and j (r^ij).

$$s_{ij}=\frac{1-{\left(\frac{r_{ij}}{r_o}\right)}^6}{1-{\left(\frac{r_{ij}}{r_o}\right)}^{12}}$$ (8)

r_o is defined as the cutoff when s_ij approaches zero and was set to 0.35 nm. Hydrogen atoms were not considered when quantifying the total number of hydrophobic contacts. Hydrophobic contacts were computed using PLUMED Version 2.5.1.⁵⁵

Hydrophobic Contact US Simulations.

We performed US simulations using c_h as the collective variable using PLUMED Version 2.5.1 in conjunction with Gromacs 2016.6. Initial AuNP-DOPC configurations used the last configuration from the decreasing-z US simulations at z = 5.1 nm, which has c_h = 0 for C₁− and C₁₀-AuNPs (Figure 3c). Initial configurations for US simulations were generated using a NVT pulling trajectory with a spring constant of 50 kJ·mol⁻¹·contacts⁻² (see Supporting Information, Table S7). US simulations were initiated using configurations from the pulling simulations with c_h values between 0 – 150 contacts in increments of 2.5 contacts, totaling up to 61 windows. A spring constant of 10 kJ·mol⁻¹·contacts⁻² was used for all US simulations. Each simulation window was equilibrated for 500 ps using the velocity-rescale thermostat then simulated for 80 ns using the same thermostat and the Parinello-Rahman barostat. The last 60 ns of the production simulation for each window were used to compute the PMF with WHAM.⁵⁴ For Bn-AuNPs, the last 40 ns of 50 ns production simulations for each window was used to compute the PMF, which was sufficient for convergence (see Supporting Information, Figure S19).

Simulation Parameters.

For all simulations, Verlet lists were generated using a 1.2-nm neighbor list cutoff. van der Waals interactions were modeled with a Lennard-Jones potential using a 1.2-nm cutoff that was smoothly shifted to zero between 1.0 and 1.2 nm. Electrostatic interactions were calculated using the smooth particle mesh Ewald method with a short-range cutoff of 1.2 nm, grid spacing of 0.12 nm, and fourth order interpolation. Bonds were constrained using the LINCS algorithm. Periodic boundary conditions were enabled in all directions.