Graphene nanoflakes protein corona composition
Graphene can be produced starting from graphite, by a large variety of techniques, even though reliable production of biologically dispersed single-layer graphene samples in high quality and yield is still a challenge33. Top-down approaches for graphene nanoflakes production have focused on the separation of graphite planes using, e.g., ultrasonic or shear exfoliation in organic solvents or water-based surfactant solutions34,35,36. Liquid-phase exfoliation (LPE) of graphite is a common method to obtain graphene water dispersions, and the use of biomolecules as a means to produce biocompatible graphene dispersions has recently attracted increasing attention37,38,39,40. A common choice has been to disperse graphene (and other carbon materials), making use of single proteins. However, because our strategy is to explore biological interactions, we employ the full-protein portfolio (complete serum) as a model for realistic biological exposure scenarios. This leads to presentation of appropriate biological recognition motifs.
In this work, highly stable colloidal dispersions of graphene nanoflakes in aqueous solutions were prepared by 1–4 h of bath sonication of natural flake graphite in complete serum and phosphate-buffered saline (PBS). Different kinds of serum and concentrations were investigated. This process, described in more detail in Methods, resulted in a mixture of graphite and graphene with different thicknesses and lateral sizes. The dispersions were therefore subjected to a size selection by centrifugation (Supplementary Fig. 1a), and unbound proteins in excess were removed by high-speed centrifugation. As an alternative to the size selection, different sonication times can also be used to tune the size distribution of the flakes (Supplementary Fig. 1d).
During this exfoliation process, a layer of protein adsorbs onto the graphene flakes, surrounding them and stabilizing the highly hydrophobic graphene surface in water. The composition and orientation of this protein layer, with a high affinity for the graphene surface, represent the final biological identity of the graphene nanoflakes, and the key biological motives presented at the periphery will be eventually interacting with the surface of cells. The protein composition was resolved by proteomic analysis. Here, we report the results obtained using human serum (HS) at different concentrations (Fig. 1a and Table 1), while the results related to the use of fetal bovine serum (FBS) can be found in Supplementary Fig. 2–4 and Supplementary Data 1 and 2.
The proteins surrounding the flakes were denatured (procedure described in the Methods section) and separated from graphene flakes using 1D sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE), as reported in Fig. 1a. The concentrations of graphene dispersions of the same size distribution were normalized using the extinction values at 800 nm. At this wavelength, the extinction coefficient of graphene is size independent and also not influenced by the presence of proteins, as shown in Supplementary Fig. 5 and directly correlates to the graphene concentration41.
Figure 1a shows the protein profile for graphene nanoflakes exfoliated with different concentrations of HS. Two reference controls have been included in the gel to assure that (1) no proteins are left in the supernatants after the last washing step and that (2) no protein aggregates that might form during the sonication procedure are left in the graphene dispersion which would otherwise affect the protein analysis.
To better resolve the protein composition on the graphene nanoflakes, mass-spectrometry (MS) analysis was performed, as detailed in the Methods section. Three bioinformatics analyses were used for comparison purposes (more details in the Methods section). Table 1 presents a list of proteins highly abundant on the graphene nanoflakes surface when exfoliated with 100% v/v HS, as identified by MS and analyzed by MaxQuant. For better clarity, only the most abundant proteins (matched with the other two analyses described in the Methods section) are reported in Table 1. The complete list of proteins analyzed by MaxQuant as normalized spectral counts (NSpC) can be found in Supplementary Data 1 and 2.
The pie charts in Fig. 1b represent the statistical distribution of proteins (organized by class) on the graphene flakes when the concentration of serum is varied. From these charts and Table 1, the following considerations can be made: (i) human serum albumin (HSA) seemed to have a strong affinity for the graphene flakes under these exfoliation conditions, and this has also been found in the literature for graphene oxide (GO) and carbon nanotubes42,43,44,45,46; (ii) lipoproteins and immunoglobulins played a major role when HS was used for exfoliation, while, as shown in Supplementary Fig. 3, the presence of hemoglobin and opsonin proteins is remarkable when FBS is used; and (iii) the serum concentration seems to slightly affect the overall corona composition, especially in the case of HS. Some of the proteins that we found to be strongly bonded to graphene nanoflakes (albumin, immunoglobulins, complement, and apolipoproteins) are also reported to have good affinity for other carbon-based nanomaterials such as carbon nanotubes45,46,47,48; however, the data at the present stage do not allow to conclude a general trend for carbon-based materials in a competitive environment such as full serum. Apolipoprotein A-I, the major component of high-density lipoprotein (HDL), was found to be highly abundant on the nanoflakes exfoliated with HS. It is also interesting to note that apolipoprotein B100 (the major component of low-density lipoproteins, LDL) was nearly absent even though it is commonly found in the corona of several nanoparticle types, such as silica and amine-modified polystyrene nanoparticles49, 50. However, we now understand that the composition of the corona may reflect little on the abundance of specific recognition motifs present in endogenous HDL, and therefore consider that question in more detail later in the paper.
HDL is a complex of small lipoproteins containing an outer-shell layer of phospholipids, free cholesterol stabilized by apolipoproteins (apoA-I for the 70%), and a hydrophobic lipid core of cholesterol esters and triglycerides51, 52, and plays a major role not only in the lipid metabolism (cholesterol efflux) but also in the innate immunity53. Therefore, this result is even more interesting, considering the reported affinity between graphene (and GO) surface and lipids. It has been reported that both the materials can promote the formation of supported lipid bilayers54, self-organization of phospholipids on their surface55, or vesicles in lipid layers56. Moreover, they can promote the disruptive extraction of phospholipid molecules from the lipid bilayers by phospholipid interaction onto its own surfaces57.
The presence of apolipoprotein A-I could suggest the binding of intact HDL lipoprotein complexes onto the nanomaterial, but whatever the source, depending on how the apolipoprotein A-I is presented, combined with the flat shape of graphene nanoflakes, suggests that they might be biologically recognized as HDL complexes, though without the capacity to modify its structure that HDL possesses58. Various studies in the past have suggested that HDL elasticity and the capacity to adopt such flattened structure could affect cellular uptake, as well as interendothelial transport59. Whether such capabilities could be transferred onto graphene flakes in the relevant media is unknown, but it will at least be of interest to understand the nature of the dispersion and the biological presentation.
Graphene nanoflakes dispersion characterization
To characterize the obtained graphene nanoflakes dispersions, Raman spectroscopy, atomic force microscopy (AFM), differential centrifugal sedimentation (DCS), and ultraviolet/visible (UV/Vis) extinction spectroscopy were used, as shown in Fig. 2 and in Supplementary Fig. 5 and 6. Raman spectroscopy is one of the most common characterization methods for graphite and graphene60,61,62. Figure 2a depicts the Raman spectrum for graphene nanoflakes dispersion (dried droplet), showing the characteristic D, G, and 2D bands typical of graphene/graphite. Graphene nanoflakes edges activate the D band at ~1350 cm−1 (in the absence of other plane defects), while the number of graphene layers modify the shape and intensity of the 2D band (~2750 cm−1). Monolayer graphene presents a single narrow 2D peak with twice the intensity of the G band (~1580 cm−1)60,61,62. The overall spectral pattern (D/G intensity and shape and relative intensity of the 2D band) is consistent with the presence of few-layer graphene41. More spectra and their I2D/IG ratios are shown in Supplementary Fig. 7.
AFM confirmed the presence of few-layer graphene nanoflakes (Fig. 2b; Supplementary Fig. 6). Lateral size distribution ranged between 100 and 800 nm, and a main population of nanoflakes around 200–300 nm was found after the analysis (Fig. 2c).
Although the size distribution given by DCS is not formally correct in the case of nonspherical particles, still it can produce very consistent and repeatable results even when applied to graphene63, allowing to distinguish different populations and to detect aggregation.
DCS and AFM showed comparable results for size distributions with a maximum peak around 250 nm for graphene exfoliated in 100% v/v HS after 2 h of ultrasonication (Fig. 2c). These size-distribution data were in agreement to what has been found for graphite exfoliation in surfactants64, 65. By further AFM analysis presented in Supplementary Fig. 6, thicknesses up to 25 nm are found. Taking into account the thickness of the protein layer, and the typically observed overestimated measured AFM thickness of liquid-exfoliated nanoflakes compared to the theoretical thickness64, 66,67,68, the number of graphene layers was estimated to range between 2 and 10 layers.
Despite the clear challenge for such a complex suspension, the ζ potential for the dispersions at pH = 7 was measured, resulting in −28 mV. In such complex mixtures, the meaning and validity of zeta potential should be considered with some caution. A more pertinent point is that the effects that stabilize the nanomaterials in the presence of proteins (either in serum, or isolated hard-corona complexes) are believed to be similar to those that stabilize the protein solution: a mixture of charge (not usually the primary enabling mechanism for dispersion) and hydrogen bonding, as well as entropic interactions (e.g., water ordering) and other forces are believed to play a significant role.
It is frequently observed that changing the media composition can impact the colloidal stability of the nanomaterial; therefore, it is essential to assess the nanomaterial stability in a given biological media over time to translate the material toward in vitro testing. The stability of graphene nanoflakes was tested in the biological medium serum-free minimum essential medium (MEM) and supplemented with 50% v/v of serum (complete MEM or CMEM).
DCS measurements were performed after 24 h of incubation at 37 °C in MEM and CMEM, and any shaking was performed during the incubation or prior to the analysis, to be closer to in vitro exposure conditions. In serum-free MEM after 24 h, a deposit of flakes can be clearly noticed, and almost nothing can be measured in the supernatant. After few seconds of vortexing, the sample can be measured, but it results in larger aggregates (Fig. 2d, black curve). In 50% v/v serum-supplemented medium (CMEM), the dispersion remained very stable over 24 h, without the need for vortexing, as can be seen in Fig. 2d (blue curve). After 48 h, the flakes can be seen to sediment on the bottom of an Eppendorf™ LoBind microcentrifuge tube, but after few seconds of vortexing, they can be measured at DCS, resulting in the same stable distribution (Fig. 2d, red curve). We can therefore conclude that the protein exfoliated graphene is stable up to 24 h in serum- supplemented medium, and this is the first time that the colloidal long-term stability is demonstrated for graphene biological dispersion.
Protein functionality and availability of key epitopes
It is now widely accepted that the slowly exchanging biomolecular part of nanomaterial–biomolecular complexes determines the biological identity of the nanomaterial via the presentation of key recognition motifs in the corona that interact with relevant receptors28. The peripheral surface of the particles is therefore dependent on the conformation and orientation of biomolecules within the protein corona, which can be studied by methods currently under development25, 27, 28. Most of these methods use recognition (such as antibodies) and reporting functions (say gold nanoparticles (GNPs) or QDs) that recognize sites close to the reported recognition domains for relevant receptors, but in any case, they are capable of giving information on the general interface organization. Prior to the experiment, we made sure that the effect of prolonged sonication did not affect the protein conformation. To this aim, we exploited tryptophan fluorescence emission and circular dichroism as an indicator of the tertiary structure for some proteins. The results reported in Supplementary Fig. 8 showed a negligible effect of bath sonication on the proteins tertiary structure for the conditions used in this paper.
Preliminary evidence on the availability of exposed epitopes of interest on the graphene nanoflakes surface was obtained by immuno dot-blot assay, as described in Methods. The results are reported in Fig. 3a, and expectations from macroscopic proteomics (the predominance of apolipoprotein A-I in the exfoliated flakes and the absence of apolipoprotein B100) were confirmed using monoclonal antibodies anti-apoA-I and anti- apoB100.
Since the exfoliated graphene produces a black spot if adsorbed onto the polyvinylidene fluoride membranes (PVDF), making the fluorescence of the secondary antibody undetectable, a different strategy was used for this immunoassay. The dark color of graphene was therefore exploited as a signal, and a dark circle can be clearly seen when the interaction with antibodies occurs (Fig. 3a). An empty PVDF membrane was used as a control, and no adsorption of graphene flakes onto the membrane was found after incubation (Fig. 3a). A second control was represented by a functionalization with monoclonal antibody anti-apoB100, and the results confirmed the specificity of the interaction.
To understand the likelihood of biological recognition of exfoliated graphene nanoflakes (which after all occurs on a particle-by-particle basis), we aimed to map out the relevant recognition motifs at the surface of graphene by using recently reported immunomapping methods25, 27. In this paper, we used two different immunoprobes (called immunogold, IG or immuno-quantum dots, IQDs) that consisted of GNPs or QDs with a nominal diameter of about 4 nm functionalized with a specific monoclonal antibody; in this case, again the monoclonal anti-apoA-I was able to recognize amino acids 113–243 of apoA-I of human origin (proxy for the HDL receptor-binding domains)69.
It should be noted that the application of these approaches for graphene is more challenging than for many other nanomaterials to which it is routinely applied, given the peculiar size, thickness, and shape variety of the nanoflakes, but it can still be carried out in order to collect a series of evidences. Moreover, significant care must be taken in preparing these antibody constructs to avoid non-specific binding. The graphene nanoflakes exfoliated with HS were exposed to a large excess of IG and IQDs (detailed in the Methods section), and then washed an increasing number of times, following the procedure described elsewhere25, 27. Negative controls (Fig. 3d; Supplementary Fig. 9b) suggest that the only IG or IQDs remaining are those bound by biologically mediated recognition.
Thus, as shown in Fig. 3b, first, we do recognize a shift in the main peak in DCS after incubation with immune probes, suggesting in situ antibody recognition across the whole distribution of particulates.
In the case of IG mapping, as shown in Fig. 3c and Supplementary Fig. 9, direct imaging by transmission electron microscopy (TEM) showed the presence of 4-nm GNPs retained on the surface of the flakes after washing. As a negative control (IG_cntrl), the nanoflakes were also exposed to GNPs functionalized with HSA and very little non-specific adsorption was found (Fig. 3d; Supplementary Fig. 9b).
To confirm these data, IQDs were also used to detect the availability of apoA-I functional epitopes, as described in Supplementary Information. The emission spectrum (Supplementary Fig. 10) showed a fluorescence emission peak corresponding to the emission of the QDs retained after washing. The spectrum was normalized over the emission of the negative control (QDs functionalized with albumin), therefore, it is representative only of specific interactions.
All these evidences suggested the widespread surface presentation of the relevant apoA-I epitopes, able to promote specific recognition by, for instance, suitable cell receptors.
The estimation of an average number of exposed epitopes per particle is extremely challenging in the case of graphene flakes. The wide distribution of the flakes in terms of size and shape made it very difficult to reasonably estimate the number of particles. Also, the drying effect on the TEM grid made it very difficult to isolate a statistically significant number of single flakes. However, based on the samples mass concentration, and the QDs emission intensity, it was possible to estimate about 1 × 1014 functional apoA-I epitopes per mg of exfoliated graphene nanoflakes, as detailed in Supplementary Fig. 10.