Unexpected intracellular biodegradation and recrystallization of gold nanoparticles

Significance

While gold nanoparticles are at the core of an increasing range of medical applications, their fate in the organism has barely been studied so far. Because of their chemical inertness, common belief is that gold nanoparticles remain endlessly intact in tissues. We show that 4- to 22-nm gold nanoparticles are actually degraded in vitro by cells, with a faster degradation of the smallest size. Transcriptomics studies reveal the active role of cell lysosome into this biodissolution. Furthermore, we point out that the released gold recrystallizes into biopersistent nanostructures. Interestingly, these degradation products are similar to previously observed gold deposits in human tissues after gold salts treatment for rheumatoid arthritis, underlying a common metabolism between gold nanoparticles and ionic gold.

Abstract

Gold nanoparticles are used in an expanding spectrum of biomedical applications. However, little is known about their long-term fate in the organism as it is generally admitted that the inertness of gold nanoparticles prevents their biodegradation. In this work, the biotransformations of gold nanoparticles captured by primary fibroblasts were monitored during up to 6 mo. The combination of electron microscopy imaging and transcriptomics study reveals an unexpected 2-step process of biotransformation. First, there is the degradation of gold nanoparticles, with faster disappearance of the smallest size. This degradation is mediated by NADPH oxidase that produces highly oxidizing reactive oxygen species in the lysosome combined with a cell-protective expression of the nuclear factor, erythroid 2. Second, a gold recrystallization process generates biomineralized nanostructures consisting of 2.5-nm crystalline particles self-assembled into nanoleaves. Metallothioneins are strongly suspected to participate in buildings blocks biomineralization that self-assembles in a process that could be affected by a chelating agent. These degradation products are similar to aurosomes structures revealed 50 y ago in vivo after gold salt therapy. Overall, we bring to light steps in the lifecycle of gold nanoparticles in which cellular pathways are partially shared with ionic gold, revealing a common gold metabolism.

Gold has been used in various forms in medicine since antiquity (1). Therapeutic use of gold in modern medicine began in 1890 with the discovery that gold (Au^I) cyanide was bacteriostatic to the tubercle bacillus in vitro (2). This result initiated the development of chrysotherapy in modern medicine to treat rheumatoid arthritis (3, 4). The use of those therapies have declined since the 1980s due to long-term adverse effects and the development of alternative treatments (5, 6). Gold nanoparticles (GNPs), which were only exploited from the 1950s as a source of gamma ray (i.e., radioactive Au¹⁹⁸ seeds) for cancer treatment (7), raised new interest for their plasmonic and radiosensitizing properties in an expanding spectrum of biomedical diagnostic [photoacoustic imaging (8), 2-photon luminescence (9), surface enhance Raman spectroscopy (10)] and therapeutic application [photothermal therapy (11), photodynamic therapy (12), radiosensitization (13)] to cite but a few.

A wide variety of GNPs with different sizes, shapes, and coatings has been studied in vitro and in vivo so far to evaluate their distribution (14, 15) and toxicology (16–18). It is established that GNPs are internalized mostly in the liver and spleen by macrophages (19, 20) and sequestrated inside their lysosomes (21, 22), the organelles responsible for the degradation and recycling of exogenous (xenobiotics, bacteria) or endogenous compounds (defective organelles, protein aggregates). Lysosomes are characterized by an acidic environment (pH around 4.5), which is regulated by proton pumps, and contain specific enzymes, namely acid hydrolases. Lysosomal degradation processes have been described for metal or metal oxide nanoparticles, but only few concern GNPs (23–25). As a noble metal, gold is more inert, less sensitive to acidic environment, and less reactive than most other metals and metal oxide (26). Therefore, the current dogma is that the inertness of gold prevents biodegradation of gold implants or GNPs that could be left indefinitely intact in tissues. Sustained integrity of GNPs can indeed be an asset to maintain their nanoscale-related optical properties for a long time in the body. This inertness of gold would also prevent the release of potentially toxic bioactive uncontrolled ionic forms of gold. So far, to our knowledge, only 1 study from our group has reported in vivo degradation of 5-nm gold nanocrystals to 3-nm size, which was observed in mice spleen 3 mo after systemic administration of gold iron oxide heterostructures and dissolution of the iron oxide shell (27). Although numerous studies have been conducted in vitro on the influence of GNPs on cultured cells, only few exceed a few days, and no clue indicates GNP degradation on electron microscopy (28–32).

Here, we revealed the unexpected degradation of GNPs of different sizes in lysosomes up to 6 mo in fibroblasts and evidenced a surprising phenomenon of recrystallization and self-assembly of their degradation products. Ultrastructural observations of GNPs in cells were combined with transcriptomic analysis to unravel the biotransformations of GNPs kept in active lysosomes and to shed light on the mechanisms of lysosomal processing of gold species. To address this challenge, we implemented a methodology that enables long-term cell culture and high-resolution detection of GNPs and of their degradation products up to 6 mo after internalization in cells.

Primary human fibroblasts were chosen for their ubiquitous character in the body, their potential intended or accidental exposition to GNPs, and their low proliferation rate that enhances the residence time of nanoparticles in the same cell, enabling culture over several months and long-term follow-up of nanoparticles. Spherical GNPs (diameter [D] = 4, 15, and 22 nm) (SI Appendix, Fig. S1) covered with noncytotoxic citrates were chosen because of their narrow size distribution, easy synthesis, and wide use. GNP-labeled fibroblasts’ viability was maintained for 6 mo, with curbed proliferation to limit the dilution of GNPs on cell division and enable us to monitor their in situ biotransformation over time (SI Appendix, Fig. S2).

Results and Discussion

The 6-Mo Biotransformations of GNPs in Lysosomes.

The biotransformation of 4-nm GNPs in fibroblasts was monitored by transmission electron microscopy (TEM) observation on microtome sections of 70 nm at 1 d, 2 wk, 2 mo, and 6 mo after exposure to GNPs (Fig. 1). One day postlabeling, the original GNPs are densely packed into intracellular organelles showing the ultrastructural features of lysosomes (33). Two weeks after exposure to GNPs and up to 6 mo, we observed 2 types of electron-dense objects in lysosomes (dark and light orange arrows in Fig. 1 A–C): first, the very same GNPs as those observed at day 1 with a surface proportion decreasing with time and second, an increasing proportion of more diffuse structures (Fig. 1 D–K). Interestingly, this phenomenon of apparent GNPs transformation shows heterogeneous aspects. At the cell level, 2 wk after exposure to GNPs, some lysosomes are the seat of GNP transformations, while others show only unchanged GNPs. At the lysosome level, when diffuse structures are present, they are often, but not necessarily, in contact with a domain containing intact GNPs. The electronic contrast, spatial concentration, and aspect of the diffuse structures as well as the original GNPs do not seem to evolve over time on our study timescale. However, the surface proportion of the diffuse structures relative to the total surface occupied by electron-dense objects increases from 43% after 2 wk to 89% after 2 mo and did not change between 2 and 6 mo (Fig. 1L). The nature of these appearing diffuse structures was then further explored with high-resolution electron-based tools.

Fig. 1.

TEM observations on microtome sections of human fibroblasts exposed to 4-nm GNPs observed 1 d to 6 mo after GNPs incubation, evidencing the existence of dense and diffuse electron-dense areas. (A–C) Images of 1 cell 2 wk after GNPs exposure presenting 3 electron-dense areas identified as lysosomes constituted of dense areas (dark orange arrows) and diffuse areas (light orange arrow). (D–K) Representative lysosomes observed 1 d (D and H), 2 wk (E and I), 2 mo (F and J), and 6 mo (G and K) after exposure at 2 magnifications. L, Surface proportion of dense and diffuse areas from 1 d to 6 mo after exposure; 5 to 10 images have been analyzed for each condition, with at least 1 electron-dense area per image.

High-resolution scanning transmission electron microscopy (STEM) observations performed 2 wk after exposure to GNPs reveal that the diffuse structures are composed of discrete nanoparticles with a characteristic size of 2.5 $\pm$ 0.4 nm (Fig. 2 A–D). Interestingly, these individual elements are most often aligned along curved trajectories forming lash-like architectures with similar thickness (t = 5.3 $\pm$ 1.4 nm) and radius of curvature (R_c = 43.5 $\pm$ 13.2 nm). These dimensional characteristics are conserved for at least 6 mo (SI Appendix, Fig. S3). Occasionally, straight linear self-assemblies (length from 50 to 650 nm; thickness from 5 to 80 nm) are also observed (SI Appendix, Fig. S4). Furthermore, the size distributions of the nanoparticles composing the dense areas are shown to depend on their proximity to diffuse areas. In lysosomes presenting only dense areas, nanoparticles diameters are close to the original GNPs, remaining arounds 4 nm over time. Surprisingly, at the interface with the diffuse area, nanoparticles present a higher diameter that the original GNPs, with size increases from 4 to 10 nm after 6 mo, suggesting gold transfer, recrystallization, and growth at the surface of some particles concomitant with the dissolution of other particles and appearance of diffuse recrystallized degradation products (SI Appendix, Fig. S5). One more striking point is that no intermediate size was observed between the recrystallized cluster and the original GNPs, indicating an all-or-nothing degradation process.

Fig. 2.

STEM observations of human fibroblasts 2 wk after exposure to 4 nm GNPs showing that diffuse structures are 2D nanoleaves composed of self-assembled 2.5-nm nanoparticles. (A–C) STEM images of a cell domain at 3 different magnifications, showing the local structure of the diffuse areas previously described. (D) Diameter distribution of nanoparticles observed in dense or diffuse areas 2 wk after GNPs exposure measured on at least 115 objects. (E–G) STEM-HAADF images extracted from the tilt series acquired during tomography experiments. The tilt angle of the sample holder with respect to the electron beam is indicated in the top right corner of each image. The same 2 leaf-shaped nanostructures are indicated by arrows on each image. The resulting calculated 3D tomogram can be found in Movie S1.

The 3-dimensional (3D) character of the most common curved lash-like architectures was revealed by performing STEM using high-angle annular dark field (HAADF) at different tilt angles. As observed in Fig. 2 E–G, the projected shape of the diffuse structures varies depending on the tilt angle between the sample holder and the electron beam. If the lash-like structure is frequently observed on each frame of the tilt series, this 1-dimensional architecture is only observed in a narrow angular range for each individual nanostructure. When a nanostructure is out of this specific angular range, it looks more like a platelet with rough edges. The calculated 3D tomogram shown in Movie S1 reveals that lash-like architectures are in fact 2-dimensional (2D) curved objects, like nanoleaves, with various orientations. When these nanoleaves are parallel to the electron beam, they appear like high-contrast lashes because of their higher thickness along the optical axis.

High-resolution TEM also revealed similar crystalline atomic lattices both in the original GNPs and in the tiny nanoparticles self-assembled into leaves, confirmed by electron diffraction that shows a face-centered cubic lattice structure (a = 4.079 Å) identical to gold crystal (Fig. 3 A–G). However, the diffraction pattern obtained in the diffuse area of nanoleaves had less intense and broader peaks than the one in the areas composed of almost intact GNPs, which is compatible with the presence of smaller and fewer crystalline domains in diffuse structures.

Fig. 3.

High-resolution TEM observations, electron diffraction, and STEM-EDS performed on human fibroblasts 2 wk after exposure to 4-nm GNPs reveal the crystallinity of nanoparticles composing diffuse areas and a specific signal of sulfur. (A–D) TEM imaging of dense (A and B) and diffuse (C and D) area, presenting crystalline lattice highlighted in red. (E and F) The 2D diffraction patterns obtained on representative regions of dense (E) or diffuse (F) areas. (G) Scattered intensity obtained after radial integration of E and F. (H and K) STEM images of noncontrasted (H; red frame), diffuse (H; light orange frame), and dense (K; dark orange frame) areas analyzed by EDS. (I, J, and L) EDS spectra of noncontrasted (I), diffuse (J), and dense (L) areas and (Inset) enlargement of the 2- to 3-keV area, with Gaussian fits for gold, sulfur, and chloride peaks. The slight horizontal contrast modifications seen at the bottom of the STEM-HAADF image are due to sample charging.

Finally, elemental analyses performed by STEM energy dispersion spectroscopy (EDS) at the nanoscale (0.009 to 0.04 $\mathrm{\mu }$m² $\times$ 70 nm) unambiguously confirmed gold detection in electron-dense areas but in a significantly lower concentration in the diffuse nanoleaves (Fig. 3 H–L). Importantly, sulfur was also detected within diffuse structures with nanoleaves but not in proximity of untransformed GNPs (1.1 to 1.2 sulfur atoms for 1 gold atom).

Overall, the 6-mo monitoring of GNP-labeled cells reveals the appearance of gold structures made of 2.5-nm nanocrystals self-assembled into nanoleaves and associated with sulfur. Their occurrence increased over time at the expense of the original GNPs, which decreased in number. In the same time, the average size of original GNPs is found to increase from 4 to 10 nm in 6 mo when they coexist with nanoleaves in the same lysosomes.

These observations led us to conclude that the nanoleaves are products of GNPs degradation that results from gold ions release, recrystallization, and self-organization of de novo nanocrystals, probably mediated by sulfur species. Furthermore, the increase of the GNPs size at the interface with degradation products suggests a gold transfer and recrystallization process at the surface of the original GNPs in addition to the newly formed degradation products. This highlights nontrivial mechanisms of the degradation/recrystallization effect likely governed by surface effects increasing the reactivity of gold atoms.

Fibroblasts were then labeled with 15- and 22-nm citrate-coated GNPs to investigate if lysosomal degradation of GNPs and in situ recrystallization could also occur for larger sizes of particles. We indeed observed the very same nanoleaves composed of individual gold nanocrystals of 2.6 $\pm$ 0.4 nm appearing side by side with aggregates of 15- or 22-nm GNPs within lysosomes (Fig. 4 A–C and E–H and SI Appendix, Fig. S6). Such diffuse structures were less frequently observed and appeared later in comparison with cells labeled with 4-nm GNPs (Fig. 4D). These results prove that the biodegradation process described for 4-nm GNPs could be generalized to larger nanoparticles. They also strongly reinforce the hypothesis that such self-organized gold nanostructures are the result of size-dependent GNP lysosomal degradation taking place faster for small GNPs with higher surface/volume ratio and reactivity than for large GNPs (34).

Fig. 4.

TEM observations of human fibroblasts exposed to 22- or 15-nm GNPs reveal the presence of diffuse areas associated to degradation. (A–C) TEM images of representative lysosomes observed 1 d (A), 2 wk (B), or 2 mo (C) after 22-nm GNPs exposure, presenting dense areas (dark orange arrows) and diffuse areas (light orange arrow). (D) Surface proportion of dense and diffuse areas from 1 d to 2 mo after exposure to 22-nm GNPs; 5 to 10 images have been analyzed for each condition, with at least 1 electron-dense area per image. (E–H) Representative lysosomes observed 6 mo after 15-nm GNP exposure, presenting dense and diffuse areas observed at 2 magnifications.

GNPs Degradation Products Are Similar to Aurosomes Observed In Vivo after Gold Salts Treatment.

It is worth noting that similar self-assembled gold nanostructures have been observed before in at least 2 different contexts in vivo. In the previously published study on iron oxide-coated GNPs intravenously administrated in mice, the intralysosomal degradation of iron oxide crystals and the transformations of 5-nm gold crystalline core were observed sequentially (27). Organized assemblies of 3-nm GNPs forming plates or curved structures were also seen in spleen and liver from 3 mo postinjection of iron oxide@gold heterodimers (SI Appendix, Fig. S7). Although these structures were previously attributed to the erosion and further self-assembly of initial gold core, their compelling morphological similarities with the structures witnessed here in vitro strongly suggest a general mechanism of gold dissolution followed by in situ recrystallization to form 3-nm GNPs organized into suprastructures with assistance of endogenous lysosomal proteins.

This hypothesis is further supported by the second physiopathological condition in which similar gold nanoassemblies were observed in the so-called aurosomes. Based on electron microscopy observation by Ghadially (33), aurosomes were described early in the 1970s as lysosomes containing gold nanoassemblies formed in various organs after administration of gold salts for the treatment of rheumatoid arthritis. Striking resemblance between aurosomes and the presently highlighted GNPs degradation products is supported by the same filamentous (straight or curled), rod-like, or lamellar electron-dense profiles containing $<$5-nm gold particles within lysosomes (SI Appendix, Fig. S8). Dimensional analysis of the size of the particles composing these aurosomes, the radius of curvature, and average thickness of their ultrastructure shows high similarities with the nanoleaves observed in our study (SI Appendix, Fig. S9). The crystalline nature of the “granules” forming these morphologies was not fully established at that time (35). By contrast, the colocalization of gold and sulfur with Au/S atomic ratios from 1.5 to 2.2 and a gold oxidation degree +1 has been revealed by X-ray absorption spectroscopies (36). Importantly, aurosomes were produced from nonsulfur Au^III salts, suggesting that all or part of sulfur should be of endogenous biological origin (37). The same morphology and composition were found in numerous organs, in several species (rat, rabbit, and human), for different administration routes, and for various gold salts and colloidal gold, attesting to the generality of this biotransformation process (33). Overall, it is clear that the aurosome features (i.e., size, morphology, and chemical composition) are very similar to the degradation products of GNPs observed in this study. Hence, for the following parts, GNPs degradation products will be called aurosomes. This analogy leads us to the important conclusion that the intracellular fate of gold ions is the same in the long term whatever the origin of ions (gold salt, pure GNPs, or core-shell heterostructures) in vitro or in vivo. This conclusion can be of great importance to evaluate the toxicity of GNPs on the very long term and also, their clearance from the organism when gold is under its recrystallized form as aurosomes. If aurosomes proved to be an ultimate stable form of gold in cells and tissues, the concern of elimination or clearance of GNPs is translated to the elimination of aurosomes. Interestingly, renal elimination of gold species has been observed for patients treated with gold salts and presenting aurosome formation (38). This opens avenues to better understand and control GNP fate and clearance on the long term.

However, aurosome formation remains elusive as well as GNPs’ degradation biological mechanism. Taking advantage of our cellular model, we thought to perform a transcriptomic analysis to shed light on the pathways involved in GNP biotransformation.

Transcriptomics Analysis Reveals Long-Term Oxidative Response to GNPs Involving Metal Detoxification Pathways.

Transcriptomics analysis was performed on 18,537 genes using DNA microarray at 3 time points (1 d, 2 wk, 2 mo) after incubation or not of fibroblasts with 4-nm GNPs (39). Differences in gene expression between all samples were first analyzed by principal component analysis (40) (PCA) using the 500 genes with the highest variance-to-mean ratio as input (SI Appendix, Fig. S10). It appeared that 94% of the variance between samples could be captured with the first 4 dimensions, with good reproducibility between triplicates. The first dimension takes into account 73.5% of variance, showing that the main differences come from time and separating 1-d samples from later time points (Fig. 5A and SI Appendix, Fig. S11). The second dimension, recapitulating 13.5% of variance, highlights the differences between the 1-d control and GNP-treated samples. Finally, the third and fourth dimensions, with 4.6 and 2.4% of variance, respectively, discriminate samples obtained at 2 wk and 2 mo, and control and GNPs condition for these 2 time points (Fig. 5B).

Fig. 5.

Transcriptomics analysis reveals a time-dependent answer to 4-nm GNPs and the activation of 3 main pathways at the longer times of study (NADPH production, ROS detoxification, and metal chelation). (A and B) PCA of transcriptomics data performed with the 500 most variable genes. (C) Venn diagram of GNPs samples differential expression compared with unlabeled control at the same time (FDR q value lower than 10⁻³ only). Lists and arrows summarize highest hits of GSEA (FDR q value lower than 10⁻⁴ only). (D) Expression heatmap of up-regulated genes 2 wk after GNPs exposure. Only genes that are attributed to identified pathways are displayed (38 genes on 124). Expression was considered as significantly different for FDR q value below 0.01. CTL, control; TNF, tumor necrosis factor.

The dominant effect of time was evaluated on control samples by performing differential gene expression analysis using 1-d sample as reference and a false discovery rate (FDR) q-value threshold set at 0.01. According to these criteria, 4,129 genes were differentially expressed at 2 wk and/or 2 mo (SI Appendix, Fig. S12). Gene set enrichment analysis (GSEA) (41) was performed to identify the biological functions corresponding to these differentially expressed genes using an FDR q-value threshold at 0.01 (SI Appendix, Tables S1 and S2). It appeared that cell division and related functions (metabolism, cellular machinery, DNA repair) as well as diverse stress response (immune response, oxidative stress, unfolded protein response) were down-regulated at 2 wk and 2 mo, in line with previous measurements showing that cells were maintained at G0 phase (SI Appendix, Fig. S2).

In order to highlight differential gene expression due to GNPs, unlabeled control samples were used for each time point as reference, keeping an FDR q-value threshold at 0.01. In line with PCA result, the Venn diagram shows that 88% of the modified genes only concern the early time point (Fig. 5C). The most impacted biological functions were identified by GSEA, keeping an FDR q-value threshold at 0.01 (highest hits in Fig. 5C; SI Appendix, Table S3). At day 1 postlabeling, it reveals a down-regulation of cell division, DNA repair, and cellular machinery for GNP-labeled cells. These down-regulated functions were identified previously as a general nonspecific and fast response to stress, whether it is induced by xenobiotics or other stress factors (42). In contrast, gene set corresponding to oxidative stress, exogenous stress, and immune response was enriched at 1 d (SI Appendix, Table S3) and was still affected at 2 wk and 2 mo (highest hits in Fig. 5C; SI Appendix, Tables S4 and S5), revealing a long-lasting response of interest to understand GNPs degradation pathways.

Five up-regulated pathways are common to the 2 later time points: reactive oxygen species (ROS) pathway, xenobiotics and drug metabolism by cytochrome P450, glutathion (GSH) metabolism, interferon (IFN)-alpha response, and adipogenesis. Activation of ROS has already been reported for small GNPs (D $<$ 5 nm) and for different cell types (43). Likewise, the activation of cytochrome P450 pathway was already described in spleen and liver 1 and 2 mo after GNPs (D = 20 nm) injection in vivo (44). It suggests that cytochrome activation might be a late response to GNPs that cannot be detected at the early time points. GSH is a tripeptide responsible for the redox state of the cell and is implicated both in oxidative stress and in metal detoxification (45). The up-regulation of GSH metabolism has been reported previously for 5- and 13-nm GNPs 24 to 72 h after GNPs labeling (46, 47). Regarding IFN-alpha up-regulation, it is in line with the previously reported immunological activity of GNPs (48), although the mechanisms are still unclear.

Concerning the down-regulated pathways, ribosome metabolism is affected at both longer time points, and cell division is affected only at 2 mo. On the contrary, functions related to protein secretion and/or metabolism are up-regulated only at 2 wk. As several of these functions are linked to energy production (mammalian target of rapamycin complex 1 [mTORC1] signaling, glycolysis, gluconeogenesis, penthose phosphate pathway, fatty acid metabolism), it could reflect the need for energy to process GNPs degradation and recrystallization. Accordingly, these functions would return to their basal level at 2 mo, when most 4-nm GNPs are already transformed into aurosomes.

To identify the genes implicated in GNPs degradation, the significantly overexpressed genes at 2 wk were further studied in the light of their transcription factor using iRegulon, revealing a general scheme. Indeed, 80 of the 2-wk up-regulated genes (66%) contribute to nuclear factor erythroid 2 (NFE2) transcription factor pathway and 17 to metallothionein (MT) transcription factor 1 (MTF1) gene targets (SI Appendix, Table S6); 38 of these genes belong to the same cascades of reactions (49, 50), which enable us to identify 6 main gene sets: NADPH production, heme oxidase 1 (HMOX1), quinone, GSH and thioredoxin responses, and MTs production (Fig. 5D). Other up-regulated genes are not commented either because of a lack of information on their roles or because no strong relationship could be found between them (SI Appendix, Fig. S13). HMOX1, quinone, GSH, and thioredoxin responses are implied in ROS destruction through redox reactions that consume NADPH (SI Appendix, Fig. S14). It is important to underline that the well-known ROS detoxification pathway relying on superoxide dismutase (SOD), glutathione peroxidases (GPXs family), and catalase (CAT) are not up-regulated here, despite the likely production of ROS inside the cell. GSH as well as MTs appear to play a key role in the sustained response to GNPs as mediator of metal detoxification through chelation of metal via their sulfur functions. MTs are cystein-rich metal storage proteins able to bind different metals, including gold (51, 52). MTs up-regulation in presence of GNPs has been reported after exposure of Caco-2 cells with 5-nm GNPs (46). It was evidenced that 7 types of MTs were up-regulated after 72 h of exposure to GNPs but not after 24 h, while the amount of internalized gold was the same between the 2 time points, indicating once again a kinetics of response to GNPs.

Here, our transcriptomics results shed light on the chemical composition obtained per STEM-EDS, which reveals sulfurized compounds at the same location as degradation products/aurosomes. Either GSH or MTs could be responsible for this signal of sulfur. On one hand, GSH is known to be able to stabilize numerous gold clusters (53); on the other hand, as a storage protein, MTs are more likely to complex gold, with a higher stability owing to chelate effect, and were for a long time suspected to be part of aurosomes structures (36, 54). Moreover, as the iron storage protein ferritin has already be shown to act as a recrystallization matrix for metallic ions originated from nanoparticles, a similar process could take place in MTs for gold (55, 56).

Interestingly, excepting for HMOX1, most of the genes overexpressed at the later time points are not up-regulated at day 1. This reinforces the fact that the specific response to GNPs cannot be fully appraised with short time assessments. As mentioned before, the specific response to GNP also decreases from 2 wk to 2 mo, especially MTs, GSH, and NADPH production-related genes, in line with the almost complete degradation process at 2 mo.

To conclude, the transcriptomic analysis by DNA microarrays identified early nonspecific stress related to GNP uptake and a long-lasting response more likely linked to GNP degradation and recrystallization process. The latter is mostly directed by NFE2 transcription factor, which directs ROS elimination and possibly, metal binding. It suggests a degradation of GNPs by ROS followed by gold capture and recrystallization by MTs. To test this putative mechanisms, we thought of studying the following points: 1) the role of ROS production in GNPs degradation, 2) the ability of ROS to degrade GNPs, 3) the structure of MTs–gold complexes, and finally, 4) the impact of a metal chelator on the degradation structures.

ROS Production Activates GNPs Degradation and Recrystallization through MT Chelation.

We first attempt to link ROS production to GNPs degradation. ROS can be generated extracellularly or inside the lysosome to damage undesirable compounds by membrane protein assemblies called NADPH oxidase (NOX). NOX consumes NADPH and O₂ to produce O₂^·−, an unstable radical that will dismute, spontaneously or via SOD, in H₂O₂ and O₂ in acid medium. Then, through GPX action in the lysosome or CAT in the peroxisome, H₂O₂ is inactivated and converted into water. Somehow, in presence of metal ions or metallic nanoparticles, H₂O₂ can react with metals to form hydroxyl radical HO^· via Fenton reaction (57, 58). As the metal is oxidized during this reaction, NOX production of ROS could be implied in GNPs degradation into ionic gold. Nevertheless, NOX activity is regulated on a posttranslational level, and therefore, their activation could not be directly confirmed by transcriptional results (59).

Therefore, to assess NOX role in GNPs degradation, we use GKT137831, a NOX inhibitor that is active on 3 types of NOX (NOX1, NOX4, and NOX5) (60). This choice was motivated because NOX1 is the only protein of the NOX family that has been shown to be active in endosomes and lysosomes (61). After exposure to 4-nm GNPs, fibroblasts were cultured in culture medium supplemented with NOX inhibitor. After 2 wk, the proportion of degradation products (aurosome-like structures) was unequivocally decreased from 45 to 32% of the whole electron-dense gold-containing area compared with the control condition without NOX inhibitor (Fig. 6 A–C). This clearly indicates a role of NOX1 and possibly, of NOX4 and NOX5 in the intralysosomal degradation of GNPs.

Fig. 6.

GNPs degradation is mediated by NOX-created ROS followed by a probable recrystallization inside MTs and a self-assembly process impacted by BAL. (A and B) TEM observations of human fibroblasts 2 wk after exposure to 4-nm GNPs with (A) or without (B) the NOX inhibitor GKT137831. Dense and diffuse areas are indicated by dark orange and light orange arrows, respectively. (C) Proportion of dense and diffuse areas with or without GKT137831; 30 images or more have been analyzed, with at least 1 electron-dense area per image. (D–H) Liquid TEM snapshot performed in aqueous solution ([HCl] = 0.02 mol/L, [NaCl] = 1 mol/L) under constant radiation at 200 keV. Observations were performed on citrate-coated GNPs and gold nanorods (D = 25 nm, l = 11 nm, L = 45 nm). (I) Ultraviolet (UV)-visible spectroscopy performed steadily during MT filling with sodium aurothiomalate. The legend indicates the ratio MT:Au. (J) UV-visible signal evolution at 293 nm during MT filling. (K and L) High-resolution TEM imaging of MT–gold complex at the MT:Au ratio of 1:120 observed at 2 magnifications. (M–O) STEM observation of human fibroblasts 2 wk after exposure to 4-nm GNPs and cultured with BAL. Diffuse areas of 2 types were observed (N): an aurosome-like one (green frame and M) and a new one (blue frame and O).

We next evaluate the ability of ROS to oxidize and degrade the crystalline structure of GNPs. Tunable ROS generation coupled to the dynamic observation of nanoparticles morphology can be achieved by liquid TEM in water (62, 63). Indeed, the electron beam can serve both as the illumination source for real-time nanoparticle imaging and as an input energy to induce water radiolysis and subsequent generation of radicals, including OH^·. Liquid TEM conditions allowed our group to monitor the ROS-triggered degradation of multiwalled carbon nanotubes, recapitulating damages that were observed for weeks in living cells (62). In the case of GNPs, specific conditions of pH and chloride concentration are required to observe GNPs degradation as shown in Fig. 6 D–H. This in situ monitoring of GNPs dissolution reveals the ability of beam-generated ROS (most likely OH^·) to etch gold nanocrystals. Previous liquid TEM studies have shown that no degradation could be observed in basic conditions (pH = 12), in contrast to neutral and acidic environments (pH = 7 and 2, respectively) (63). Moreover, chloride ions are needed to observe the degradation process of GNPs at neutral pH and to speed up their degradation in acidic conditions. This is consistent with the role of chloride for stabilizing gold ions and lowering their redox potential. When extrapolating to the lysosome environment, the chloride content, evaluated to 0.06 M in lysosomes (64), is much lower than the conditions of GNP degradation in liquid TEM experiment (1 M). However intralysosomal medium includes other potential chelators of gold such as the upregulated GSH or MTs.

We propose that ROS production by NOX is responsible for intracellular GNP degradation inside the lysosome. The oxidation state of the released gold ions is not clear: Au^III oxidation potential is lower than Au^I but requires a 4-partners reaction, which seems improbable. Moreover, Au^I seems to be more stable than Au^III in biological medium (65).

After this first degradation step, the question arises of gold recrystallization and aurosome formation process. The transcriptomic analysis supports the implication of MTs and GSH that could both bind gold atom. Owing to their different chelating properties (20 cysteins per MT, 1 per GSH), we hypothesize that the released ionic gold would be preferentially sequestrated into MTs. MTs can bind 7 divalent cations (Zn²⁺, Cd²⁺, Hg²⁺) and 12 monovalent cations (Ag⁺, Cu⁺), but supermetallation has been also described, including for gold (66, 67). Indeed, MT–gold complexes can take the form of gold clusters, containing up to 35 atoms, stabilized by 1 or several MTs. To investigate the similarity between MT–gold complexes and aurosomes’ building blocks, we monitored the interactions of commercial MTs with sodium aurothiomalate salt on a large range of MT–gold ratios (Fig. 6 I–L). Ultraviolet-visible spectra show an absorption band at 293 nm that increases with the molar ratio MT:Au (Fig. 6 I and J). The absorption increase is linear up to the stoichiometric ratio of MT:Au = 1:12 and then, inflects markedly up to 1:144. This transition indicates the onset of the supermetallation regime, possibly coupled with gold aggregation into MT-stabilized clusters. TEM observations reveal gold clusters for the ratio of 1:120, whereas nothing could be observed at the stoichiometry 1:12 (Fig. 6 K and L). Among these clusters, about half of them show a crystal lattice, with a higher average diameter (D = 2.6 $\pm$ 0.5 nm for crystalline clusters, D = 1.9 $\pm$ 0.5 nm for noncrystalline ones) (SI Appendix, Fig. S15). This crystallinity suggests that MTs do not compose the core of the clusters but participate in their surface stabilization. The similar morphological features of MT–gold complexes and aurosome building units and the up-regulation of MTs in GNPs-labeled cells lead us to point out MTs as a credible candidate for the capture and subsequent recrystallization of gold into nanometric particles of well-defined size forming the aurosomes.

The last striking step of the aurosome formation regards the self-assembly of the recrystallized gold particles into nanoleaves with a particularly well-defined curvature. To our knowledge, this mechanism has never been explored so far. These assemblies could be mediated by an organic template that could explain why nanoleaves characteristics, such as their curvature radius, are highly conserved as MT–gold complexes present no tendency to self-organization in model medium. This potential template has to satisfy the following criteria: 1) to be present in lysosome of all tissues or cell types, 2) to keep its integrity in acidic conditions and despite acid hydrolases activity, 3) to present a curvature distribution comparable with aurosomes and to be stiff enough to endure mechanical confinement into lysosomal membranes, and 4) to be invisible in TEM after osmium and iron staining. To shed light on the self-assembly process, we use BAL (British anti-Lewisite; or dimercaprol), a dithiolated compound that was previously used for gold detoxification, in order to test if it could alter and/or disassemble the aurosome nanostructure. When BAL was added in the cell culture medium during aurosome formation, TEM reveals that 2 distinct types of self-assembled clusters-based ultrastructure could be formed (Fig. 6 M–O). If one type looks like previously described aurosomes and has similar features (D = 2.2 $\pm$ 0.4 nm, R_c = 33.8 $\pm$ 10.6 nm), the second type differs greatly from prior observations (D = 1.9 $\pm$ 0.3 nm, R_c = 10.6 $\pm$ 1.6 nm) (SI Appendix, Fig. S16). These 2 types of ultrastructures can surprisingly coexist in the same cell, or even the same lysosome, with equal proportions. Hence, if some features are conserved in presence of BAL, such as the cluster formation and the self-assembly into nanoleaves, others are slightly modified, like cluster size, or deeply affected, such as the resulting curvature radius of the self-assembly. As BAL is thiolated, it can interfere with either gold–sulfur bonds or disulfide bridges and results in a “denaturation” of the aurosomes that end up in a recrystallization pattern.

Overall, the degradation step seems to be mediated by NOX proteins that can produce highly oxidizing ROS in the lysosome, resulting in GNPs dissolution. In a second step, released ionic gold would then be crystallized and sequestrated, probably inside an MT-composed shell, owing to the morphology similarities between aurosome building blocks and MT–gold complexes. The nature of self-assembly template is much more puzzling, and we establish a list of criteria that it should fulfill according to our observations. It has also been shown to be affected by BAL, which created self-assembled patterns, possibly by disrupting disulfide or gold–sulfur bonds.

Conclusion

We have studied the biotransformation of GNPs in primary human fibroblast during 2 to 6 mo. First, TEM observations evidence the intracellular degradation of GNPs with a size-dependent dynamic. This degradation is induced by ROS, generated by NOX, that oxidized GNPs. This oxidation occurs together with an upregulation of NFE2 pathway, which creates a cell-protective environment. This result invalidates the current dogma of intracellular gold inertness and highlights the need for long-term studies to capture the slow process of nanoparticles degradation.

Second, we show that the released gold undergoes a biomineralization process and ends up in well-defined structures consisting of 2.5-nm crystalline particles self-assembled into nanoleaves. Reviewing of the literature dedicated to therapeutic gold salts evidences that similar structures, called aurosomes, have previously been described in vivo in different species and organs. It supports the original idea that ionic and crystallized gold have a common intracellular fate that could be named as gold metabolism. The morphological features of aurosomes and the transcriptomics results designate MTs as credible candidates to capture gold and form the ubiquitous crystalline cluster. If more understanding on the process of self-assembly is still needed, BAL, a thiolated chemical compound, has been shown to affect aurosome and to generate biomineralization patterns.

Fig. 7 summarized these 2 steps in GNPs lifecycle that could in the future be extended to other cell types (such as macrophages), in vivo, or to other persistent nanoparticles (such as quantum dots). Furthermore, this process can potentially impact other GNPs-related concerns, such as their toxicity or their elimination from the organism.

Fig. 7.

Proposed mechanism of GNPs degradation and recrystallization process. H⁺ and stoichiometric coefficients are not mentioned for clarity.

Materials and Methods

Cell Culture.

Deidentified human skin primary fibroblasts were obtained with consent from a healthy individual with no known metabolic disease at Necker–Enfants Malades Hospital. All experiments have been done with cells between 15 and 19 passages. All medium, antibiotics, and serum were purchased from Life Technologies (Thermo Fischer Scientific). Cells were cultivated in Dulbecco modified Eagle medium (DMEM) high glucose supplemented with 1% penicillin/streptomycin. Prior to GNPs internalization, 10% fetal bovine serum (FBS) was added to this medium, while only 2% was added after cell exposure to nanoparticles. Cells were maintained at 37 °C with 5% CO₂. For the culture with GKT137831 (Cayman Chemical), 2% FBS culture medium was supplemented with 50 $\mathrm{\mu }$M GKT137831 previously dissolved in water with 2% DMSO. For the culture with BAL (TCI Chemicals), 2% FBS culture medium was supplemented with 50 $\mathrm{\mu }$M BAL. In both cases, the medium was changed twice a week.

GNPs Uptake.

GNPs were diluted in Roswell Park Memorial Institute medium with glutamine containing 10% FBS for NPs stabilization. The cells were incubated at 80% of confluence with this medium during 24 h at 37 °C. The medium was then removed, and the cells were washed with phosphate buffer saline (PBS) and conserved for 2 h in 10% FBS medium; then, this medium was removed, and the cells were washed again in PBS. Finally, cells were conserved in 2% FBS DMEM for up to 6 mo. Medium was replaced once or twice a week.

DNA Microarray Preparation.

For RNA extraction, cells were washed twice with PBS, lysed, and treated using the NucleoSpin RNA Kit (Macherey Nagel) according to manufacturer’s protocol, including a DNAse treatment. The following steps concerning DNA microarray were performed by the platform Genomics (Institut Cochin, UMR8104, National Center for Scientific Research [France]/National Institute of Health and Medical Research [France]/Université Paris Diderot, Paris, France). After validation of the RNA quality with Bioanalyzer 2100 (using the Agilent RNA6000 nanochip kit), 75 ng of total RNA is reverse transcribed following the GeneChipWT Plus Reagent Kit (Affymetrix). Briefly, the resulting double-strand complementary DNA is used for in vitro transcription with T7 RNA polymerase (all of these steps are included in the wild-type complementary DNA synthesis and amplification kit of Affymetrix). After purification according to Affymetrix protocol, 5.5 $\mathrm{\mu }$g of Sens Target DNA is fragmented and biotin labeled. After control of fragmentation using Bioanalyzer 2100, complementary DNA is then hybridized to GeneChipClariom S Human (Affymetrix) at 45 °C for 17 h. After overnight hybridization, chips are washed on the fluidic station FS450 following specific protocols (Affymetrix) and scanned using the GCS3000 7G. The scanned images are then analyzed with Expression Console software (Affymetrix) to obtain raw data (cel files) and metrics for quality controls.

Electron Microscopy.

GNPs TEM images were obtained with a Tecnai 12 microscope operating at 80 kV (Platform Imagoseine, Institut Jacques Monod, UMR7592, National Center for Scientific Research [France]/Université Paris Diderot, Paris, France) after deposition of a droplet of NPs solution on a hydrophilized 400-mesh grid. For MT–Au complexes, 3 $\mathrm{\mu }$L of a solution of sodium aurothiomalate (9.9 mM) and MTs (83 $\mathrm{\mu }$M ) was deposed onto a 400-mesh grid 15 min after mixing. For TEM imaging, cells were washed with PBS, detached with 0.5% trypsin, and centrifuged. After 3 washes with PBS, cells were fixed with 5% glutaraldehyde (Sigma Aldrich) in a 0.1 mol/L sodium cacodylate buffer (Sigma Aldrich) for 1 h at ambient temperature, washed with cacodylate buffer, and kept in this buffer until inclusion. Samples were then contrasted with Oolong Tea Extract (0.5% in cacodylate buffer), osmium tetroxyde (1% in cacodylate buffer), and potassium cyanoferrate (1.5% in cacodylate buffer). They were then gradually dehydrated in ethanol (25 to 100%). The samples were then substituted gradually in propylene oxide, a mix of propylene oxide and Epon, and finally, embedded in pure Epon (Delta microscopie, Labège, France). Thin sections (70 nm) of samples were collected on 200-mesh grids. Observations were performed on either a Tecnai 12 microscope (Platform Imagoseine, Institut Jacques Monod, UMR7592, National Center for Scientific Research [France]/Université Paris Diderot, Paris, France) or a Hitachi HT7700 microscope (Platform Microscopie et Imagerie des micro-organismes, animaux et aliments 2 [MIMA2], UMR1313, Institut National de Recherche Agronomique [INRA]/AgroParis Tech, Jouy-en-Josas, France), both operating at 80 kV. Further TEM investigations, STEM-HAADF, high-resolution TEM, electron tomography, EDS, and electron diffraction were performed on an aberration-corrected JEOL ARM 200-F microscope operating at 80 kV (Laboratoire Matériaux et Phénomènes Quantiques, UMR7162, National Center for Scientific Research [France]/Université Paris Diderot, Paris, France).

Data Availability.

Transcriptomic data and script are available at the Zenodo repository, https://zenodo.org/record/3530617#.XfFNDht7ncs.