Graphene microsheets enter cells through spontaneous membrane penetration at edge asperities and corner sites

Fig. 1 shows the CGMD results for graphene and few-layer graphene interacting with lipid bilayers. In the first simulation set, a small, rhombic, monolayer graphene flake with edge length of 6.4 nm is placed initially at a distance about 4 nm above and parallel to a square patch of lipid bilayer with 992 lipid molecules and 67,817 water molecules in a cubic box with edge dimension 24 nm (Fig. 1A). Periodic boundary conditions are imposed in all three dimensions. Under thermal fluctuations, the graphene flake undergoes Brownian motion, including rapid vibration, rotation, and migration in the vicinity of the bilayer. Spontaneous piercing into the bilayer is observed to begin as soon as the flake finds a configuration with one of its sharpest corners oriented nearly orthogonal to the membrane (Fig. 1 B and C). Piercing is facilitated by the attractive interactions between graphene and the tail groups of lipids, but occurs only after the tip of the penetrating corner touches the hydrophobic core of the bilayer. In our CG simulations, the lipid membrane can be fully penetrated through by the graphene flake, accompanied by the tilt of the graphene flake to maximize its coverage with the membrane interior. The small graphene flake in our simulation eventually ends up embedded in the bilayer due to its small dimensions (Movie S1). The simulation of a rhombic graphene flake with two different corner angles (and ) demonstrates that orthogonal piercing of the sharpest corner of graphene has the lowest energy barrier and is the most preferred entry pathway. The edge planes of graphene have a complex chemistry and are typically decorated with hydrophilic oxygen functional groups, whereas in cell culture medium graphene may exhibit adsorbed proteins that can also reduce apparent hydrophobicity. The set of GFN samples used in our bioimaging studies has C/O ratios ranging from 10 to 32 by X-ray photoelectron spectroscopy (31) with the oxygen atoms presumed to be on edge and defect sites. Also, the amount of adsorbed protein was measured to cover 3–46% of the GFN surface area, depending on conditions and sample (SI Text). To investigate these effects, further CGMD simulations of bilayer interaction with graphene flakes of different shapes and surface chemistry (different corner/edge functionalizations and hydrophilic/hydrophobic properties) confirm that small graphene sheets tend to penetrate into the cell via spontaneous piercing at their sharpest hydrophobic corner. A series of calculations with different interaction parameters between graphene and lipid molecules are also performed to demonstrate that the corner entry mode is robust within a broad range of dissipative particle dynamics (DPD) parameters (SI Text). We will show shortly that the nearly orthogonal orientation of a sharp graphene corner with respect to the bilayer minimizes their interactive free energy and is the thermodynamically preferred configuration even before penetration begins.

We also performed CGMD simulations of large few-layer graphene sheets interacting with lipid bilayers (Fig. 1 F–H). We began by studying an ideal, atomically smooth, infinite graphene edge interacting with lipid bilayers, but did not observe penetration (SI Text, Fig. S6A, and Discussion). We realized that the edges of experimentally fabricated graphene exhibit atomic-scale roughness (Fig. 1E) as seen by atomically resolved scanning tunneling microscopy (32–34), and most few-layer graphene sheets also show very rough edges (see Fig. 4), as well as terraced or beveled edge structures that become successively thinner toward one of the two faces (35, 36). Fig. 1F shows a model terraced edge structure created on a five-layer FLG flake interacting with a bilayer. The simulation system consists of a patch of bilayer with 2,016 lipid molecules and 133,052 water molecules in a cubic box with dimensions of 24 nm × 48 nm × 24 nm. The plate-like FLG, which has a ragged edge topography mimicking those observed in experiments, is composed of five atomic layers with an equilibrium interlayer distance of 0.34 nm and placed initially at a distance about 3 nm above and orthogonal to the bilayer. Each graphene layer is assigned a different color. The first two and last two layers are set to be symmetrical with respect to the midlayer. To mimic part of a much larger structure for which Brownian motion is limited, the top edge of the FLG is clamped and periodic boundary conditions are imposed in all three dimensions of the simulation box. The lipid bilayer undergoes Brownian motion in the vicinity of the large graphene edge for 2.19 μs under the confinement of a harmonic potential. The latter is then removed and the bilayer membrane is set free to interact with the ragged graphene edge. It is seen that the FLG penetrates the bilayer despite its size, starting with localized piercing at sharp protrusions along the edge. The penetrated portion of the membrane then propagates along the whole edge, resulting in full penetration (Movie S2). In this process, the energy barrier to penetration is overcome by local piercing at sharp corners along the nominally flat edge, and the full penetration is driven by the attractive interaction between the graphene and the tail groups of lipids once initial piercing is successful. We have tested the robustness of this entry mode by carrying out further CGMD simulations on monolayer or few-layer graphene flakes with an isolated protrusion or a terrace or by initiating contact near a corner or a locally folded edge (SI Text and Figs. S2–S7). The simulation results all show similar pathways that are initiated by localized piercing at an atomically thin graphene feature, which requires only thermal energy to overcome the small barrier, followed by spreading and complete penetration driven by hydrophobic forces between the graphene and the bilayer core. We believe this entry mechanism may be generic for cell uptake of all 2D hydrophobic nanomaterials with atomic-scale thickness. The importance of localized corner piercing for entry initiation can also be demonstrated by carrying out simulations on ideal, atomically smooth, infinite graphene edges without the irregular features observed in real samples. In this case, our simulations show that the lipid bilayer is repelled away from the graphene edge due to a combination of a strong entry energy barrier and entropic interactions between the bilayer and an atomically smooth graphene edge (SI Text and Fig. S6A). This last simulation was done to further demonstrate the importance of local piercing as the initiating event, but does not correspond to any known biological exposure scenario, because it would require cell contact with a uniform, atomically smooth, horizontally aligned, long-length graphene edge structure that is difficult to achieve in practice.

Our simulations demonstrate local orthogonal piercing at a sharp corner or asperity that initiates graphene penetration of lipid bilayers. To better understand this behavior, we use the thermodynamic integration (TI) technique (37) to calculate the free energy of the system as a function of two orientation angles of a rhombic graphene flake when one of the sharp corners of the flake is fixed at a distance of 0.4 nm above the bilayer (the detailed implementation of the TI method is given in SI Text). Here is the angle between the long diagonal axis of the flake and the bilayer within the graphene plane and is the angle between the vectors normal to the graphene plane and the membrane plane (Fig. 1I). The free energy associated with orthogonal corner piercing is set to 0 as a reference value. Not surprisingly, the orientation , which corresponds to the graphene flake parallel to the bilayer plane, shows the highest free energy because it induces the most severe confinement of thermal motion in this configuration. We further normalized the calculated free energy by its peak value at the parallel configuration . The plotted normalized free energy in Fig. 1I demonstrates that the orthogonal orientation exhibits the lowest free energy due to its weakest confinement on the thermal motions of both membrane and graphene (thereby maximizing the entropy of the system).

Our CGMD simulations suggest that localized corner piercing plays a critical role during the initial stage of cell uptake of graphene. To determine the energy evolution associated with such initial piercing events, further simulations are conducted on an all-atom model of corner piercing of a monolayer graphene flake across a bilayer patch of 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) lipid in a box of water molecules. The all-atom simulations are of two types. Type I simulations are designed to test whether there exists a positive driving force for piercing in an all-atom MD simulation. As shown in Fig. 2A, a triangular graphene flake initially placed in a corner-piercing configuration across the bilayer was observed to spontaneously move downward, penetrating further into the bilayer (Movie S3). In type II simulations, the energy barrier associated with corner piercing is calculated by SMD simulations (38), in which a graphene corner is pulled across the bilayer by a virtual spring (see SI Text for more details). Fig. 2B shows the graphene–bilayer interaction energy calculated from the SMD simulations as a function of the penetration distance. The SMD calculations confirm that the energy barrier is small, only ∼5k_BT, for the graphene corner to pierce through the top hydrophilic head region of the bilayer. Shortly after this point, the total interaction energy starts to decrease due to favorable interactions between the lipid tails in the core of the bilayer and an ever-increasing area of immersed graphene. Thus, both our CGMD and all-atom simulations reveal that corner piercing involves a small energy barrier comparable to thermal energy and is essentially a spontaneous process.

Both CGMD and all-atom simulations indicate that the lipid bilayer structure remains essentially intact upon graphene insertion except that the hydrophobic tails of lipids are somewhat straightened due to strong adhesion onto the side surfaces of graphene. In a continuum description, the energy change accompanying piercing can be expressed in terms of four variables: h_H and h_T (thicknesses of the head and tail groups in the lipid monolayer, as shown in Fig. 2C) and γ_H and γ_T (interaction energy densities between one side surface of graphene and head and tail groups of lipids relative to that between solvent and graphene). For a graphene corner with an internal angle of 2α, the rate of energy change during piercing can then be written as a piecewise function of the penetration depth h as (SI Text) Our all-atom simulation shown in Fig. 2A indicates that there is a positive driving force for the last regime of piercing, i.e., . Under this condition, it can be seen that the energy increases in the first regime, , but decreases in both regimes and . The energy peak occurs when the graphene tip lies in the hydrophobic core, , at a point defined by . This gives the energy barrier for piercing as at a critical penetration depth of .

The surface interaction energy densities γ_H and γ_H may be estimated as , if they are considered to be of the same order of magnitude as those between the hydrocarbon tail groups and water (39). Taking and shows that the energy barrier for corner piercing is at a critical penetration depth of . These numbers are in excellent agreement with the direct SMD simulation results shown in Fig. 2B.

It may be tempting to make an analogy between graphene corner piercing into a lipid bilayer described here and mechanical piercing by stress concentration at sharp corners. However, we emphasize that the former is an essentially spontaneous process, whereas the latter is driven by an applied force.

Figs. 3 and 4 present confocal fluorescent and ex situ electron micrographs that confirm the MD predictions of edge/corner-first penetration and cell entry of few-layer graphene microsheets. Lung epithelial cells and keratinocytes are representative of the epithelial lining of the respiratory tract and the skin, respectively (40), and form flat, single-cell monolayers in vitro (Fig. 3). Polarized epithelial cells have a well-organized microtubular cytoskeletal network whereas macrophages have a subcortical distribution of actin filaments that can be visualized using indirect immunofluorescence confocal microscopy (Fig. 3 A–C, Insets). Plate-like graphene microsheets are internalized by human lung epithelial cells (Fig. 3 A and B) and macrophages (Fig. 3C) and can be visualized within the cytoplasm using confocal imaging. It is interesting that the graphene basal planes show preferential orientation parallel to the basolateral cell surface attached to the substrate. Plate-like graphene microsheets physically disrupt the cytoskeletal organization of both lung epithelial cells (Fig. 3 A and B) and macrophages (Fig. 3C). In thin sections using transmission electron microscopy (TEM), some graphene microsheets can be visualized within cytoplasmic vacuoles inside macrophages (Fig. 3D) and lung epithelial cells (Fig. 3E). The overall structure and integrity of subcellular organelles are preserved as revealed by TEM (Fig. 3 D and E), confirming our in vitro assays that show preserved cell viability (SI Text and Fig. S8).

The imaging protocols used in Fig. 3 are useful to show graphene internalization and orientation, but do not reveal the entry mode. We therefore carried out shorter time exposures to capture the uptake process, using ex situ field-emission SEM of target cells with the outer membrane enhanced by osmium tetroxide postfixation. This imaging was carried out both with and without critical point drying to check for drying artifacts. Fig. 4 shows high-resolution images of cell surfaces exposed to graphene after 5 or 24 h. Edge-first or corner-first penetration of the membrane is seen in all cases and for all three target cell types. Fig. 4 C and D shows particularly clear cases of membrane penetration that appears to have initiated at an asperity or protrusion on the graphene edge (Fig. 4C) or initiated at a graphene corner (Fig. 4D). Note that the edges of these sheets are observed to be highly irregular (Fig. 4 A, C, and D) and provide numerous sites for initial penetration as described in the modeling (SI Text and Fig. S9). Internalization of these plate-like graphene microsheets ranging in lateral dimension from 0.5 nm up to 5 μm (Table S1) did not compromise cell viability at the doses and time points used in these in vitro studies. More generally, however, inhaled nanoparticles can be associated with adverse health effects (41) and Schinwald et al. (7) reported that graphene nanoplatelets induced granuloma formation and lung inflammation following pharyngeal aspiration in mice. Intracellular uptake and cytoplasmic localization of plate-like graphene nanomaterials may interfere with cytoskeleton organization (Fig. 3 A–C) and normal physiological functions including polarized secretion (40), barrier formation (42), and cell migration during differentiation and repair of epithelial injury (43). Internalization of nanoparticles by macrophages has been shown to disrupt phagocytosis and clearance of particles and microbes from the lungs (44, 45). Schinwald et al. (7) provide evidence that graphene nanoplatelets are not readily cleared from the lungs and induce release of proinflammatory mediators from macrophages. Other studies report that graphene nanomaterials are biocompatible (46, 47), and much more work is needed before graphene material health risks can be fully assessed.

In summary, coarse-grained molecular dynamics, all-atom steered MD, analytical modeling, and live-cell and ex situ bioimaging were used to investigate the fundamental mechanisms of graphene interactions with lipid bilayers. The simulations reveal direct bilayer penetration that begins with localized piercing at sharp corners or at protrusions along graphene edges followed by propagation along the edge to achieve full penetration. For a small graphene flake, Brownian motion and entropic driving forces in the near-membrane region first position the flake orthogonal to the bilayer plane, which then leads to spontaneous corner piercing. All-atom steered molecular dynamics simulations track the free energy evolution during corner piercing and reveal only a small energy barrier, comparable to k_BT. Interestingly, in the absence of sharp corners or edge protrusions, the cell membrane has a high intrinsic energy barrier against penetration by long graphene edge segments even though they are atomically thin. Such uniform, atomically smooth, horizontally aligned, long-length graphene edges are rare, however, so in practice cell penetration is spontaneous due to the presence of atomic- or nano-scale edge roughness that essentially eliminates the energy barrier. Experimental imaging studies confirm graphene penetration of cell membranes in a dominant edge-first or corner-first mode for each of three cell types studied: lung epithelial cells, keratinocytes, and macrophages. The experiments also show penetration and successful uptake of GFN flakes as large as 5–10 μm in lateral dimension, which supports the model prediction that penetration activation barriers are not intrinsically length dependent, because of initiation at local sharp features. Once the initial energy barrier for spontaneous membrane penetration has been overcome, we hypothesize that interaction between the hydrophobic basal surfaces of graphene microsheets with the inner hydrophobic region of the plasma membrane promotes cellular uptake. Hydrophobic surfaces are considered to represent damage-associated molecular patterns (DAMPs) that nonspecifically activate the innate immune response (48). Hydrophobic cellular surfaces (49) and surface functionalized nanoparticles (50) are more readily internalized and initiate more potent innate immune responses than weakly charged, hydrophilic surfaces. By this mechanism, we hypothesize that graphene microsheets that penetrate into hydrophobic lipid domains may be recognized as DAMPs by target cells that are the first line of defense against particles and microbes deposited on the skin or on the epithelial lining of the lungs following inhalation. The ability of graphene microsheets with large lateral dimension to penetrate and enter cells, documented here both experimentally and through simulation, may lead to cytoskeletal disruption, impaired cell motility, compromised epithelial barrier function, or other geometric and steric effects that deserve further study.

The coarse-grained simulations in this paper are based on DPD, which is a Lagrangian method derived from coarse graining of molecular dynamics (51) widely used as a mesoscopic simulation method for biomembrane systems (52–55). The lipid bilayer membrane is represented by the H₃(T₅)₂ coarse-grained model (SI Text and Fig. S1) (51). The hydrophilic lipid heads and hydrophobic lipid tails are shown as red and yellow beads, respectively. A unit cell of CG graphene consists of three beads with the nearest-neighbor distance of 0.4 nm and internal angle of 60°. The interactions among the beads are chosen to keep elastic properties of CG graphene consistent with experimental results (2, 56). The interaction parameters between graphene and lipid molecules are calibrated against parameters from all-atom MD simulations and also varied in a range to test the robustness of key observations. The simulations are performed in the number-volume-temperature (NVT) ensembles with the time step taken as and carried out by using the software package LAMMPS (57). More details of the method are given in SI Text.

The all-atom MD simulations were performed in NAMD (58) and visualized in VMD (59). The graphene and the POPC lipid bilayer membrane were generated using VMD Graphene Builder and Membrane Builder, respectively. The graphene and lipid bilayer system was then fully hydrated (adding water molecules), using VMD. The CHARMM36 force field (60) with added parameters for the graphene and TIP3 water model was adopted in the simulations. The simulations were composed of two equilibration steps and one production step. Because the lipid membranes generated by the membrane plug-in of VMD were far from the equilibrium state, an equilibrium run of 0.5 ns was carried out in which everything except lipid tails was fixed. This first step allowed the lipid tails to melt into a fluid-like configuration. In the second step, the systems were equilibrated for about 2 ns in the number-pressure-temperature (NPT) ensemble at temperature of 310 K and pressure of 1 atm. Next, the production runs were performed as the third step. In all three steps of simulation, the out-of-plane degrees of freedom of the carbon atoms of the graphene flakes were restricted using harmonic constraints. Further details of the all-atom simulations are given in SI Text.

Experiments were conducted on three cell types: murine macrophages, lung epithelial cells, and primary human keratinocytes. The cells were exposed to commercial few-layer graphene microsheet samples with a range of lateral dimensions (0.5–10 μm) and layer numbers (4–25). The FLG samples were dispersed using 1,2-dipalmitoyl-sn-glycero-3-phosphocholine (DPPC) (Avanti Polar lipids)/albumin before addition into cell culture medium (Fig. S10). One hundred microliters of graphene flakes in a 5-mg/mL ethanol stock solution were added to 62 μL of a 40-mg/mL DPPC/ethanol solution and vortexed for 5 s followed by addition of 838 μL PBS containing 2% (vol/vol) albumin to produce a 500 μg/mL (ppm) graphene stock solution containing 2.5 mg DPPC and 2% albumin/PBS. The amount of adsorbed protein that results from application of this protocol was measured in separate control experiments and found to comprise 3–46% of the GFN surface area, depending on conditions and sample (SI Text). For keratinocytes, the graphene stock solution was diluted to 10 μg/mL graphene in dermal cell basal medium containing keratinocyte growth supplements, whereas RPMI 1640 served as a dilution medium for macrophages and lung epithelial cells. Further details of the cultivation of three cell types are given in SI Text.