Rapid and efficient testing of the toxicity of graphene-related materials in primary human lung cells

Graphene and its derivative materials are manufactured by numerous companies and research laboratories, during which processes they can come into contact with their handlers' physiological barriers—for instance, their respiratory system. Despite their potential toxicity, these materials have even been used in face masks to prevent COVID-19 transmission. The increasingly widespread use of these materials requires the design and implementation of appropriate, versatile, and accurate toxicological screening methods to guarantee their safety. Murine models are adequate, though limited when exploring different doses and lengths of exposure—as this increases the number of animals required, contrary to the Three R's principle in animal experimentation. This article proposes an in vitro model using primary, non-transformed normal human bronchial epithelial (NHBE) cells as an alternative to the most widely used model to date, the human lung tumor cell line A549. The model has been tested with three graphene derivatives—graphene oxide (GO), few-layer graphene (FLG), and small FLG (sFLG). We observed a cytotoxic effect (necrosis and apoptosis) at early (6- and 24-h) exposures, which intensified after seven days of contact between cells and the graphene-related materials (GRMs)—with cell death reaching 90% after a 5 µg/mL dose. A549 cells are more resistant to necrosis and apoptosis, yielding values less than half of NHBE cells at low concentrations of GRMs (between 0.05 and 5 µg/mL). Indeed, GRM-induced cell death in NHBE cells is comparable to that induced by toxic compounds such as diesel exhaust particles on the same cell line. We propose NHBE as a suitable model to test GRM-induced toxicity, allowing refinement of the dose concentrations and exposure timings for better-designed in vivo mouse assays.


Results
Characterization of nanomaterials. Figure 1A shows standard high-resolution transmission electron microscopy (HRTEM) images for GO, FLG, and sFLG. The size distribution of the graphene flakes shows completely different lateral sizes depending on the type of material ( Fig. 1B and 1C), with an average length of 1.18 µm ± 994 nm for GO, 300 ± 23 nm for FLG, and 36.04 ± 15 nm for sFLG. Thermogravimetric analysis (TGA) (Fig. 1D) of GO, FLG, and sFLG was performed under a nitrogen atmosphere. The weight loss at a temperature of 600 °C-corresponding to the oxygen-containing groups on the graphene layers-was 57.30%, 4.81%, and 33.30% for GO, FLG, and sFLG, respectively. The significant mass loss of GO and sFLG between 100 and 300 °C was expected to decompose functional groups (-OH, -COOH, and -C-O-C) 30,31 that are not found on FLG. Raman spectroscopy is illustrated in Fig. 1E, indicating the presence of the D band (1350 cm -1 , related to some defects in the carbon rings), G band (1580 cm -1 , associated to sp 2 carbon bonds in the hexagonal structure), and 2D band (2700 cm -1 , related to the number of graphene layers and the quality of carbon rings) 32 . For carbon nanomaterials, two main parameters need to be considered in Raman spectra: the intensity ratio between the D and G bands (I D /I G, ), to quantify the density of defects in graphene 33 ; and the shape of the 2D band, to determine the number of layers (N G ) 34 . The I D /I G values obtained for the nanomaterials were 0.94, 0.42, and 1.34 for GO, FLG, and sFLG, respectively. GO and sFLG showed the highest I D /I G values due to these having a more significant amount of defects than FLG, which is consistent with the TGA results. The increased D band in sFLG is related to the small size of graphene layers compared to the number of functional groups at the edges. At the same time, GO shows a low intensity in the 2D band related to higher structural defects of its carbon rings 35 . In the case of FLG and sFLG, it was possible to calculate the average number of layers-three in each case 34 . Elemental analysis of GO, FLG, and sFLG (Fig. 1F) yielded a percentage of 48.37% oxygen in the GO sample, 6.53% in FLG, and 9.19% in sFLG-…results which are consistent with those obtained with other characterization techniques. The nanomaterial powders were re-dispersed in the different culture media (DMEM with/ without FBS and completed BEGM) at 5 µg/ml ( Supplementary Fig. 1A-C) and the colloidal stability of the nanomaterials was studied through UV-Vis absorption spectroscopy for 24 h, (see "Methods"). Supplementary Table 1 shows the average sedimentation at 2 h and 24 h for all the different nanomaterials. sFLG is the nanomaterial with the lowest sedimentation in all the different culture media after 24 h, which can give an idea about the delivered dose in each treatment. It is also important to note that although the sedimentation of GO after 2 h depends on the culture media, after 24 h there are no significant differences in the sedimentation of this nanomaterial in DMEM with FBS or in completed BEGM. Same results are observed for FLG and sFLG. These data can be explained due to the fact that BEGM incorporates complements that are similar to those found in FBS, such as different proteins or BPE (bovine pituitary extract).  36,37 . Necrosis is an uncontrolled mode of cell death involving loss of membrane integrity, which leads to activation of inflammation in vivo 38 . Previous works have shown that GRM-induced  www.nature.com/scientificreports/ toxicity involves necrosis in different cell types and organs 28,37 , including lung tumor cells 39 . However, the toxicity of GRMs remains undetermined in normal, primary epithelial cells.
When exposure to the different GRMs was extended up to seven days, necrosis drastically increased for all compounds. Compared to their respective controls, a 5 µg/mL dose of GO, FLG, and sFLG increased necrosis significantly. In particular, 5 µg/mL sFLG induced 27.5% of necrosis (Fig. 2C). Exposure to 100 µg/mL GO was the most harmful, damaging more than 50% of cells (Fig. 2C).

Graphene induces apoptosis in primary human lung cells.
Apoptosis is a type of programmed cell death essential for maintaining cell homeostasis. It is characterized by specific morphological nuclear changes such as condensation and fragmentation and the appearance of apoptotic bodies 36,40 . Apoptosis, as necrosis, is one of the main mechanisms of GRM-induced cell death 36 . Our results indicate a similar trend to that observed for necrosis, although percentages of apoptotic cells were consistently lower than necrotic ones ( Supplementary  Fig. 2). In cells exposed for 6 h, a significant increase in apoptosis induced by 0.5 µg/mL FLG and sFLG was noted, reaching 7.7 and 6.8%, respectively (Supplementary Fig. 1). Although the effect did not seem to be dose-dependent, percentages increased to 10-12% for higher GRM concentrations (5-100 µg/mL). The same trend was observed at 24 h ( Supplementary Fig. 2B) and seven-day exposures ( Supplementary Fig. 2C)-the latter with apoptosis percentages above 20% at high concentrations (50-100 µg/mL). These results indicate that GRMinduced toxicity causes NHBE cells to die preferentially by physical damage rather than programmed cell death.
Cytotoxic effect of graphene in A549 lung tumor cells. A549 is the lung cell line most widely used to assess the toxicity of nanomaterials, including graphene 16,41,42 . First, we compared the morphological features between A549 and NHBE cells without observing differences in cell size and morphology ( Supplementary  Fig. 3A, B), indicating two phenotypically similar cell types. Then, we evaluated the toxicity of increasing doses of GO, FLG, and sFLG in A549 cells exposed for 24 h, comparing the results with those observed on NHBE cells (Fig. 3). GRMs induced a dose-dependent increase in necrosis, although the values were less than half those of NHBE cells (shaded bars) at doses between 0.05 and 5 µg/mL. This difference was reduced at higher concentrations (50-100 µg/mL) (Fig. 3A). A similar trend was observed in apoptosis, which was significant only for 50-100 µg/mL (Fig. 3B). A549 cells grow in a different culture medium than NHBE cells, which includes 10% fetal bovine serum (FBS). Previous research has shown that the presence of FBS in the medium can reduce There is no effect on A549 cells with GRM concentrations of 0.05-5 µg/mL (Fig. 4A). A reduction was detected for GO at 50 µg/mL and for all GRMs at 100 µg/mL, although always lesser than those values observed for NHBE cells (Fig. 4A). No differences were observed in the number of A549 cells cultured in medium with FBS and medium without FBS and exposed to 5 µg/mL of the different GRMs ( Supplementary  Fig. 4C). A seven-day exposure to GRMs profoundly impacted NHBE cell viability, significant for low doses of 0.5 µg/mL GO and sFLG. For doses of 5 µg/mL GO, FLG, and sFLG, there was a decrease of 86.2%, 81,1%, and 81,7%, respectively, which was even more significant for higher doses (50-100 µg/mL of GRMs reduced cell viability up to 90%) (Fig. 4B). Interestingly, for a seven-day exposure the effect on A549 cells was only observed at a concentration of 100 µg/mL (Fig. 4B).
Graphene alters cytosolic and mitochondrial Ca 2+ and reactive oxygen species in NHBE cells. The next step was to examine the underlying mechanisms through which GRMs can induce cell death.
Based on the results detailed above, experiments were performed in NHBE and A549 cells incubated for 24 h with a 5 µg/mL dose of GO, FLG, and sFLG. Cell morphology was determined as a standard measure of cell wellness status 45 . No morphological alterations in the width/length ratio were found ( Supplementary Fig. 6A), although cell size decreased slightly in response to sFLG ( Supplementary Fig. 6B). Calcium homeostasis and oxidative stress were then examined, as these are key processes related to graphene toxicity 3,28 . The free cytosolic Ca 2+ level increased by 20% in NHBE cells treated with all GRMs but showed no change in lung tumor A549 cells (Fig. 5A). At the same time, there was a similar increase in mitochondrial Ca 2+ for NHBE cells treated with FLG and sFLG-an effect not found in A549 cells (Fig. 5B).
One of the main mechanisms through which graphene generates toxicity is by increasing oxidative stress 46 5C). No effect was observed in A549 lung tumor cells (Fig. 5C). On the other hand, O 2 levels were not altered by exposure to GRMs, neither on NHBE nor on A549 cells (Fig. 5D). Again, these results suggest that primary lung cells are more sensitive than the tumor cell line.  [47][48][49] . Once it had been demonstrated that these cells were susceptible to GRM-induced cytotoxicity, our results were compared with existing data on the effect of other toxic compounds-i.e., cigarette smoke extract and diesel exhaust particles 25 , 26 . After performing a database search, data on NHBE cell necrosis and apoptosis were compared to that extracted from research studies that used similar methodologies in terms of mode of exposure and incubation times. This comparison allowed us to establish that a 5 µg/mL dose of GO, FLG, and sFLG is as toxic as low concentrations of cigarette smoke extract 25 or diesel exhaust particles 26 , whereas 50 µg/mL doses, especially in the case of FLG, damage cells in a similar magnitude to the highest doses of the compounds found in the literature 23,25,26 . Their toxicity was only exceeded by exposure to cigarette mainstream smoke 23 (Fig. 6).

Discussion
In recent years, many potential graphene applications have emerged across different research and innovation fields 3,7,8,50,51 . The growing interest in this material has led to an increase in its production-and, consequently, in human exposure to it. Many of these applications-e.g., face masks, sensors, and smart clothes-involve daily use and thus continuous exposure [52][53][54] . In order to create safe-by-design protocols, it is essential to study how graphene and GRMs interact with different human biological barriers, especially those that will come into direct contact with them 2,3 . Therefore, assessing how graphene interacts with the respiratory system, especially the interaction with the first chain of defense, the respiratory epithelium. These studies are crucial, for example, for setting occupational exposure limits. On the other hand, it is necessary to establish standardized criteria for this kind of studies 1,3 . The scientific community must conduct multiple studies, evaluating the potential impact of different GRMs at different doses and exposure times. In addition, it is necessary to define the most appropriate www.nature.com/scientificreports/ biological model to conduct these studies 3 . Finally, for an adequate toxicity assessment, different GRMs should be well-characterized through standardized protocols 55 .
The major potential routes of graphene into the body are inhalation, ingestion, and dermal adsorption 3 . Exposure to graphene is variable during its production process, involving direct interaction with the respiratory tract if adequate personal protective equipment is not used 56 . Concerns about the toxic effect of graphene on the lungs also extend to its integration into everyday products such as face masks 52 and biomedical applications such as intranasal immunization 57 . Moreover, different studies on the biodistribution of graphene have demonstrated  www.nature.com/scientificreports/ the presence of graphene in the lung after intravenous 58,59 , oral 60 , and intraperitoneal administration 61,62 . This suggests that the lung could also be damaged when other administration routes are used. Different studies have evaluated the pulmonary toxicity of graphene in murine models in recent years, with contradictory results 3,63-68 . This is because the impact of graphene depends on its different physicochemical characteristics, concentration, and exposure time 3 . Bussy et al. recently observed GO inhalation could induce lung granulomas that persist up to 90 days after exposition 69 . This suggests that in vivo studies must evaluate its long-term effects. However, this is not very common. On the other hand, the in vivo studies published to date, evaluating different conditions and scenarios, required very large numbers of mice. To ensure the 3Rs principle and reduce costs and time, it is essential to refine the in vivo exposure conditions prior to conducting the experiments by using standardized in vitro toxicity assessment protocols. However, the choice of cellular models for in vitro study is a crucial issue that should not be taken lightly 70,71 .
In this work, we propose a model using primary normal human bronchial epithelial (NHBE) cells, which have been used previously to study particle-generated lung toxicity 23,25,26,72 . The gold standard to study grapheneinduced lung toxicity is the lung tumor cell line A549 73,74 . Tumor cell models are cost-efficient, easy to use and provide an unlimited material supply. However, they do not have the same characteristics as normal cells, particularly regarding the composition and net charge of the plasma membrane or the oxidative stress response-all of which are critical for interacting with GRMs 3 . Indeed, some studies using A549 cells showed no toxicity after exposure to high doses (≥ 50 µg/mL) of graphene, indicating that this cell line is highly resistant to grapheneinduced toxicity [75][76][77] .
Therefore, the use of the NHBE model offers a more realistic scenario for toxicity assessment. In this work, we have proposed a series of simple and reproducible toxicity determination procedures for identifying variations in cell viability, from slight to acute effects. The results indicate that low doses of different GRMs significantly increased NHBE cell death, an effect not observed in A549 cells (Figs. 2, 3, 4). Both cell lines, with similar morphological characteristics ( Supplementary Fig. 3), showed different behaviour in response to GRMs. This effect could be enhanced by differences in the composition of the culture media of both models, especially by the presence of FBS in the culture medium of A549 cells, which may be associated with a higher protein corona in the graphene, therefore lower cytotoxicity 43 . Although the medium of NHBE cells lacks FBS, it incorporates high concentrations of different protein complements, producing the protein corona. However, to avoid this possible effect, the toxicity of the different GRMs (5 µg/mL) was studied in A549 cells grown in FBS starvation without observing significant differences. On the other hand, the results obtained in A549 cells were similar to those reported in previous works 46,65 . Differences were only due to the intrinsic characteristics of tumoral cells A549-i.e., membrane dynamics and resistance to oxidative stress 78 .
However, to avoid underestimating the real impact of GRM-based toxicity on lung cells and the cell model used, it is also crucial to combine different approaches. Studies published to date quantifying cytotoxicity by classical methods may underestimate the real in vitro cytotoxic impact of GRMs. Our study observed necrosis and apoptosis in cells exposed for seven days ( Fig. 2C; Supplementary Fig. 2C) to 5, 50, and 100 µg/mL doses were much higher, since it was related to a very small proportion of surviving cells (Fig. 4). The substantial increase in cell death at seven-day exposures led us to focus our attention on a 24-h exposure time-which is also the standard exposure time in toxicity studies. Moreover, our study further evaluated other indirect parameters of cell damage, such as alteration in Ca 2+ homeostasis and ROS levels. We observed that low doses of GRMs altered these parameters only in NHBE cells (Fig. 5).
Our study assessed the toxicity of three well-characterized GRMs with different lateral sizes and oxidation degrees. Regarding necrosis, 5 and 50 µg/mL GO (more oxidized) generated an immediate and acute increase in this parameter compared to FLG and sFLG, which was maintained over time ( Supplementary Fig. 7). On the other hand, the size of the graphene was determinant in cytotoxicity at long times and low doses, as suggested by the high toxicity effect of seven-day sFLG exposure. This result could also be due to the fact that sFLG showed the lowest sedimentation after 24 h ( Supplementary Fig. 1), which could imply a higher interaction with the cells. This difference was not observed at higher doses since the level of cytotoxicity generated was extremely high. This trend was not observed regarding apoptosis, highlighting the importance of combining different approaches to assess toxicity in the same study.
It has been fully demonstrated that small particles harm the lung 79 , and graphene is no exception. The toxicity of many of these particles has been studied previously using the NHBE cell line. Therefore, to put our results into context, we compared graphene-induced toxicity levels in NHBE cells with those of other toxic particles analyzed using the same cell model. The toxicity levels induced by 5 µg/mL doses of GO, FLG, and sFLG were comparable to those generated by low doses of toxic compounds such as DEPs 26 and cigarette smoke extracts 25 . For 50 µg/mL doses (particularly FLG), toxicity levels were similar to those induced by high doses of DEP compounds or electronic cigarette smoke extracts 23,25,26 . For example, DEPs are generated by diesel engines, one of the most important sources of anthropogenic particulate matter emissions. These particles generate cytotoxicity in various cells, including NHBE [80][81][82] . Remarkably, different studies show that exposure to even low doses of these toxic compounds has a detrimental effect on human health 23,25,26,72 . The results obtained in our study allow us to conclude that, for NHBE cells, a 5 µg/mL dose of GRMs (considered as low) generated toxicity after 24 h of exposure, and a dose of 50 µg/mL was as toxic as higher doses of other, well-studied toxic nanoparticles.

Conclusions
The management of graphene derivatives for their integration into everyday applications such as face masks can involve regular direct contact between the nanomaterials and the lung barrier. For this reason, it is essential to design accurate, fast, and easy-to-use screening protocols to (1) assay the toxicity of current and potential novel GRMs prior to their use in commercial applications, and (2)  www.nature.com/scientificreports/ preparation and handling at research laboratories and companies. For the first time, the present work evaluates the harmful effect of different, well-characterized GRMs in a 2D model of primary human bronchial epithelial cells. This model allowed us to ascertain that the toxicity of several materials such as GO, FLG, and sFLG could be underestimated when using the current standard model, the lung tumor cell line A549. Indeed, our results indicated that lung cytotoxicity is proportional to the size and oxidation degree of the compound, with GO being the most toxic one tested-as lethal as cigarette compounds or DEPs even at low doses of 5 µg/mL. The use of primary, non-immortalized, and non-tumorigenic cells can provide a more accurate assessment of the interaction between GRMs and human lung cells-providing essential information for further testing in animal models, thus allowing the fulfillment of the Three R's principle.

Methods
GO synthesis. GO was kindly provided by Grupo Antolin (Burgos, Spain). Before its use, the material was washed to eliminate acid traces until the pH of the GO aqueous suspension was ∼5 in several cycles of Milli-Q water addition, re-dispersion, and centrifugation (4000 rpm, 30 min). The final suspension was lyophilized at a temperature of − 80 °C and pressure of 0.005 bar to obtain powdered GO.
FLG and sFLG synthesis. FLG and sFLG were prepared by ball milling treatment using melamine 83 and glucose 84 as exfoliating agents, respectively, using a Retsch PM 100 planetary mill in both cases. Briefly, for FLG, graphite (7.5 mg SP-1 graphite powder, purchased from Bay Carbon, Inc.) and melamine (22.5 mg, Sigma-Aldrich, ref. M2659) were mixed in a 25 mL stainless steel jar with ten stainless steel balls (1-cm diameter) and treated at 100 rpm for 30 min at room temperature and air atmosphere. After that, the resultant solid was dispersed in 20 mL of water for further dialysis at 70 °C, changing the washing water periodically (five changes every 120 min, including one overnight). Finally, the dispersion was left for five days to allow the sedimentation of graphite; the supernatant was extracted and lyophilized at a temperature of − 80 °C and pressure of 0.005 bar.
For sFLG, graphite (75 mg SP-1 graphite powder, purchased from Bay Carbon, Inc.) and D-glucose (4.5 g, purchased from Panreac) were mixed in a 250 mL stainless steel jar with 15 stainless steel balls (2-cm diameter). The jar was introduced in the planetary ball-milling machine at room temperature and air atmosphere for 4 h. The obtained solid was dispersed in 100 mL of water for further centrifugation (1500 rpm for 15 min) to remove nonexfoliated graphite and partial glucose. The supernatant was dialyzed at 70 °C to remove the glucose, changing the washing water periodically (seven changes every 90 min, including one overnight). The resulting dispersion was left to rest for five days at room temperature and air atmosphere. Then, the supernatant was lyophilized at a temperature of − 80 °C and a pressure of 0.005 bar. The colloidal stability in different culture media was studied using a UV-vis-NIR spectrophotometer (UV-Vis Cary 5000) with 1 cm quartz cuvettes 85 . The concentration of the nanomaterials was determined from the optical absorption at 386 nm for GO and at 660 nm for FLG and sFLG, during 24 h at different intervals, and using the calibration lines reported in Supplementary Table 2 , Epinephrine, Transferrin, Insulin, Retinoic Acid, Triiodothyronine, and Gentamicin/Amphotericin-B (CC-4175, Lonza). The growth media was changed every 48-72 h. When cells exceeded 45% confluence, the volume of the medium was doubled. Once cells reach 75-85% confluence, cells were re-seeded at 100,000 cells/T25 flask. Cells were passaged every seven days or when 85% confluency was reached. We used Clonetics ReagentPack (CC-5034, Lonza) for cell subculture with HEPES Buffered Saline Solution, Trypsin/ EDTA, and Trypsin Neutralizing Solution. Cells were maintained at 37 °C in a 5% CO 2 atmosphere. All experiments were performed between passages 1-5.

Morphological analysis.
For morphological analysis A549 and NHBE cells were seeded in a 6-well plate (50.000 cells/well), after 24 h phase contrast images were acquired using an inverted microscope. Cell area, width and length were analysed using ImageJ (N > 50 cells).
Determination of Ca 2+ and mitochondrial Ca 2+ in single cells. The intracellular Ca 2+ levels were quantified using the probe Fluo-4 (#F23917; Thermo Fisher). Cells were seeded in 96-well plates (10.000 cells/ well) and incubated for 24 h with 5 µg/mL of GO, FLG, or sFLG. Cells were then washed with PBS (5 min twice) and loaded for 30 min with 1 µM Fluo-4. After a brief washout, cells were imaged using a fluorescence microscope Nikon TiU (20 × objective) and analyzed using ImageJ 1.53. The results show the relative fluorescence units (RFUs) normalized vs. control levels (n = 3). Levels of mitochondrial Ca 2+ were quantified as described in earlier studies 86 . Briefly, cells were seeded in 96-well plates (10.000 cells/well) and incubated for 24 h with 5 µg/mL of GO, FLG, or sFLG. Cells were then loaded with 1 µM Calcein-AM (#C1430; Thermo Fisher). Cytosolic Ca 2+ fluorescence (Calcein AM) was quenched with 1 mM CoCl 2 . After washing in fresh medium, images were acquired using a Cytation 5 Reader (Biotek) (20 × objective) and analyzed using ImageJ 1.53 (n = 3).

Determination of O 2
and H 2 O 2 in single cells. The level of intracellular reactive oxygen species was quantified in living cells using MitoSox (#M36008; Thermo Fisher) for O 2 and H 2 DCFDA (#C6827; Thermo Fisher) for H 2 O 2 . Cells were seeded in 96-well plates (10.000 cells/well) and incubated for 24 h with 5 µg/mL of GO, FLG, or sFLG. Cells were then washed with PBS (5 min twice) and loaded 30 min with 1 µM MitoSOX and 2.5 µM H 2 DCFDA. After 30 min, the excess dye was washed off with PBS (5 min once). For H 2 O 2 quantification, cells were incubated at 37 °C DMEM in darkness for 30 min. Images were acquired using a Cytation 5 Reader (Biotek) (20 × objective) and analyzed using ImageJ 1.53. The results show relative fluorescence units (RFUs) normalized vs. control levels (n = 3).
Statistics. Statistical analysis was performed with GraphPad Prism 8 (San Diego, CA, USA). To determine the statistical significance between control cells and GRM-treated cells we used Student t-test or one-way ANOVA (*p < 0.05; **p < 0.01, ***p < 0.001; ****p < 0.0001), followed by a Bonferroni's post-hoc test. All graphs were designed with GraphPad Prism 8 (San Diego, CA, USA). Dara are presented as mean ± standard error of the mean (SEM) of three independent experiments.

Data availability
The data sets used and/or analyzed during the current study are available from the corresponding author on request.