Critical role of surface chemical modifications induced by length shortening on multi-walled carbon nanotubes-induced toxicity
© Bussy et al.; licensee BioMed Central Ltd. 2012
Received: 1 August 2012
Accepted: 21 November 2012
Published: 27 November 2012
Given the increasing use of carbon nanotubes (CNT) in composite materials and their possible expansion to new areas such as nanomedicine which will both lead to higher human exposure, a better understanding of their potential to cause adverse effects on human health is needed. Like other nanomaterials, the biological reactivity and toxicity of CNT were shown to depend on various physicochemical characteristics, and length has been suggested to play a critical role. We therefore designed a comprehensive study that aimed at comparing the effects on murine macrophages of two samples of multi-walled CNT (MWCNT) specifically synthesized following a similar production process (aerosol-assisted CVD), and used a soft ultrasonic treatment in water to modify the length of one of them. We showed that modification of the length of MWCNT leads, unavoidably, to accompanying structural (i.e. defects) and chemical (i.e. oxidation) modifications that affect both surface and residual catalyst iron nanoparticle content of CNT. The biological response of murine macrophages to the two different MWCNT samples was evaluated in terms of cell viability, pro-inflammatory cytokines secretion and oxidative stress. We showed that structural defects and oxidation both induced by the length reduction process are at least as responsible as the length reduction itself for the enhanced pro-inflammatory and pro-oxidative response observed with short (oxidized) compared to long (pristine) MWCNT. In conclusion, our results stress that surface properties should be considered, alongside the length, as essential parameters in CNT-induced inflammation, especially when dealing with a safe design of CNT, for application in nanomedicine for example.
KeywordsCarbon nanotubes Macrophages Length Surface chemistry
Potential adverse effects of carbon nanotubes (CNT) on human health are of great concern, especially if we consider their increasing use in composite materials  and also their exploration as innovative solutions for biomedical applications [1–5]. Like other nanomaterials, the biological reactivity and toxicity of CNT were shown to depend on numerous physicochemical characteristics including length, diameter, structural defects, surface area, tendency to agglomerate, dispersibility in solution, presence and nature of catalyst residues, as well as surface chemistry [6–20].
Among those features, the length has been suggested to play a critical role in the CNT biological reactivity after inhalation. According to a well-established paradigm for high aspect ratio nanomaterials, CNT with length superior to that of phagocytic cells can induce an inflammatory response, which is an important event contributing to tissue remodeling and carcinogenesis. In a seminal study, Poland and coworkers  showed that ‘long’ multi-walled CNT (MWCNT) -the term ‘long’ meaning that a significant proportion of them was longer than 15 μm- induced acute and chronic peritoneal inflammation and also the formation of granulomas on the mesothelial lining in mice, while shorter MWCNT (with no reliable count obtained for CNT with a length > 15 μm) did not. The same group demonstrated that CNT length is also an important determinant of their retention in the pleural space and of their subsequent effects in terms of inflammation and fibrosis development in mice . However, a major drawback of such studies [21–23] was the use of different suppliers to provide the various MWCNT. Due to discrepancies in production methods, the CNT were therefore differing not only in length but also in many other physicochemical characteristics. Indeed, the authors reported larger diameters for longer CNT, and different contents in soluble metals between the different CNT studied were also described [21, 23]. As mentioned before, these physicochemical differences could in turn affect the CNT biological reactivity and subsequent toxicity, and thus should be considered alongside the variation in length to assess the toxicological profile of CNT.
On the basis of the length paradigm and the hypothesis that length is not the only parameter to consider when evaluating the cytotoxic effects of CNT, we designed a comprehensive in vitro study that aimed at comparing the biological effects, on murine macrophages, of two samples of MWCNT which differed in length but were of similar diameter and residual catalyst metal content. Both samples were specifically produced for our study following a similar synthesis process (i.e. aerosol-assisted CCVD Catalytic Chemical Vapor Deposition). Materials of the batch referred to as “short” (S-CNT) were obtained by reducing the length of pristine MWCNT (initially grown aligned as in a carpet for 10 min, and referred to as PS-CNT, where ‘P’ stands for ‘Precursor’) using “long lasting” (i.e. 7 weeks) soft ultrasonic treatment in water . The batch referred to as “long MWCNT” (L-CNT) were pristine CNT that were grown aligned for 2 minutes without further treatment . Along with length, other physicochemical features were extensively characterized by several material science methods, namely electron microscopies (transmission - TEM, and scanning - SEM), thermogravimetric analysis (TGA), X-ray diffraction (XRD) and X-ray photo-electron Spectroscopy (XPS), so as to evaluate in depth the physico-chemical differences between the two samples. We showed that modification of the length of MWCNT leads unavoidably to additional structural (i.e. defects) and chemical (i.e. oxidation) modifications that affect both CNT surface and residual catalyst iron nanoparticles. The biological response of murine macrophages to the two different MWCNT studied was then evaluated in terms of cell viability, pro-inflammatory potential and oxidative stress. Unexpectedly, we observed an enhanced pro-inflammatory and pro-oxidative response only with the short (oxidized) MWCNT, compared to the long (pristine) MWCNT, which can be also attributed to structural defects and surface oxidation -both introduced during the shortening process- rather than to the length reduction only.
TGA analysis showed that Fe content was about 5.8 wt.% for S-CNT and 4.8 wt.% for L-CNT.
Characteristics of the short and long CNT (S-CNT and L-CNT respectively)
Diameter (nm), mean (extremes)
-Mean length (μm)
- CNT <5μm (%)
- CNT <10 μm (%)
- CNT <15 μm (%)
- CNT < 20 μm (%)
- Iron (%) (TGA)
XPS (atomic %)
- Carboxylic functions
- OH function
MWCNT + Fe3O4, alpha iron and beta iron nanoparticles
MWCNT + Fe3O4, alpha iron and beta iron nanoparticles
Ratio of the amounts of iron oxide and CNT, normalized to that in L-CNT
Effect of CNT on cell viability
Pro-inflammatory and pro-oxidant effects of CNT
Cellular inflammatory response was analyzed by quantifying the mRNA expression and protein concentration of two pro-inflammatory cytokines, namely TNF-α and CXCL2.
Our study was aimed at critically analyzing the toxicity of short and long CNT in murine macrophages. Taken together, our results showed that short and long MWCNT elicited similar reduction in macrophage viability, but only short CNT induced marked dose-dependent pro-inflammatory and pro-oxidative responses. Moreover, thanks to a thorough CNT characterization, our data illustrate how reduction in dimension was inevitably accompanied by variations in other physicochemical characteristics and led us to the conclusion that these additional modifications might be as determinative as the length reduction itself to explain the differences in biological responses between the two samples.
While our aim was to control the physicochemical parameters by using two CNT samples synthesized via the same procedure (iron catalyzed CCVD) so as to control diameter and residual iron catalyst content, the length reduction treatment led to undesired but expected further physicochemical modifications in the shortened CNT sample: i) increase in iron oxide nanoparticles residues (two-fold increase of the iron oxide/CNT ratio), ii) increase in structural defects (increase of 1% of the sp3/sp2 ratio), together with iii) COOH and OH functionalization (4.4 atomic % of increase for surface O amount). Each one or all of these physicochemical modifications could contribute to the higher inflammatory and oxidative response of macrophages to S-CNT compared to L-CNT. First, even though the total iron content was similar in S-CNT and L-CNT (about 5%), XRD analysis showed that the amount of iron oxide (normalized to that of CNT) was two times higher in the former than in the latter CNT sample. Since iron nanoparticles were mainly localized inside the hollow core of the CNT, their oxidation in the S-CNT sample suggests that they were in contact with water during the shortening treatment, probably as a consequence of a breaking/tip opening process. More important, this also suggests that iron nanoparticles in S-CNT sample may have been later in contact with the biological medium, during cell exposure. Some published data indicate that the presence of bio-available metallic elements, such as iron, plays a critical role in CNT toxicity, through the induction of an intracellular oxidative stress [34, 35]. It is indeed well-known that transition metals may contribute to particle-induced reactive oxygen species (ROS) generation through mechanisms such as a Fenton reaction, leading, together with cell-derived ROS, to oxidative stress . The fact that two markers of oxidative stress were induced exclusively in the macrophages exposed to S-CNT, together with the protective effect of NAC on pro-inflammatory cytokine production, supports the idea that the induction of an oxidative stress was involved in the development of the inflammatory response observed in S-CNT exposed cells, and that the suspected bioavailable iron in S-CNT may be ascribed for the oxidative stress induction. In order to address the question whether iron from the residual catalyst particles was responsible for the oxidative and inflammatory effects of S-CNT, cells were incubated with the iron chelator desferrioxamine. These experiments did not provide any evidence for the iron contained in S-CNT to take part in the CNT-induced effects. Therefore, other modifications brought by the shortening process should be responsible for the specific induction of oxidative stress and inflammation with S-CNT.
The second major modification introduced by CNT shortening was the presence of structural defects. Fenoglio and collaborators  and Muller and coworkers  have demonstrated that MWCNT presenting a defective carbon framework induced higher inflammatory and genotoxic responses than MWCNT without these defects. The authors related these effects to the capacity of CNT to scavenge ROS; more defects were associated to an increased scavenging activity. We did not analyze ROS scavenging capacity of S-CNT, but if this capacity was present, one could expect an absence of oxidative stress, as shown with fullerenol C60(OH)22. Since exposure to S-CNT was associated with the expression of oxidative stress markers (HO-1 and SOD-2 mRNA), a ROS scavenging capacity of S-CNT consecutive to the presence of structural defects is unlikely. The difference in structural defects between the 2 CNT samples can therefore be ruled out to explain the difference in biological response.
Finally, the 3rd modification that was introduced in S-CNT and could be related to their inflammatory and oxidative effects is the presence of functional groups at their surface. Tabet and collaborators, using an approach consisting in embedding MWCNT with an acidic polymer, showed that acidic polymer-embedded MWCNT induced a higher inflammatory response (total cell number in broncho-alveolar lavage (BAL) fluid, production of TNF and CXCL-2) than pristine MWCNT or hydrophobic polymer-embedded ones . This was also associated with a higher cellular uptake of acidic polymer-embedded MWCNT by BAL macrophages, and the authors suggested that the COOH groups might play a role in the inflammation induced by such coated MWCNT. Similarly, Saxena and collaborators  showed that acid-functionalized SWCNT were more potent than pristine-SWCNT in inducing mouse lung epithelial cell cycle arrest and lung inflammation. They related these effects to a better dispersion resulting from their negative charges and leading to a high bioavailability of these CNT and/or to their negative charges. The same group of investigators demonstrated that acid-functionalization also enhanced cardiac toxicity of SWCNT after pulmonary exposure . Although these last two studies did not examine the effect of acid-functionalization on SWCNT cellular uptake or oxidative stress, they support the hypothesis of a role of COOH groups in the inflammatory and oxidative responses induced by the S-CNT. The absence of inflammatory and oxidative responses with the L-CNT without COOH groups agree also well with this hypothesis, and stress the paramount importance of surface properties/chemistry in determining biological impact of CNT.
In the present work, the number of cells internalizing CNT in vesicles was higher for S-CNT-exposed cells compared to L-CNT-exposed ones, with a similar number of CNT incorporated in each cell. As the initial aim of our study was to evaluate the role of CNT length in their cytotoxicity, we chose to prepare our samples without any additive to the culture medium. This led to sedimentation and the formation of aggregates for both CNT samples. As observed in the optical microscopy images, and thanks to observations at the early stages of the exposures, we couldn’t observe any obvious difference, which could have modified their internalization, between the 2 samples in the aggregation pattern or the way CNT sediment on cells over time. For both CNT, the length of internalized CNT was largely smaller than 15 μm, which is the currently proposed cut-off size over which CNT can induce an inflammatory response [21, 23]. These results are in accordance with data showing that CNT with length superior to 10 μm are poorly internalized . Interestingly, although the mean length of internalized S-CNT was higher than that of L-CNT, the length distribution of internalized CNT was quite large, with overlapping values for S-CNT (0.14 to 2.65 μm) and L-CNT (0.13 to 1.42 μm). Beside length, the physicochemical determinants ruling CNT cellular uptake are still poorly known. However, surface properties have been suspected to play a major role in the CNT-cell interaction and further internalization. Kostarelos and coworkers examined the uptake of a wide variety of functionalized CNT by different cell types  and concluded, as all CNT were internalized, that the nature of the functional groups on the CNT surface did not determine whether the CNT were internalized. However, these authors did not examine CNT displaying characteristics similar to those of the S-CNT studied in our work. In agreement with our findings, Tabet and coworkers  showed that variation in the nature of the polymer used to coat MWCNT was associated with differential internalization of CNT inside macrophages; hydrophilic acidic polymer-coated MWCNT were significantly more internalized than the hydrophobic polymer-coated ones. The penetration of CNT through the plasma membrane in a “nano-syringe”-like fashion has been theoretically demonstrated , and molecular dynamics simulations have confirmed that this phenomenon could be related to hydrophilic functionalities present on the surface of oxidized materials which spontaneously insert inside the cell membrane by a lipid-assisted mechanism [44, 45]. In our study, similar nano-syringe phenomena were also observed by TEM for S-CNT-exposed cells (Additional file 1: Figure S1b). Given that S-CNT present carboxylic and hydroxyl groups on their surface, such kind of phenomenon (lipid-assisted mechanism mediated by hydrophilic functionalities) could therefore explain the higher cellular incorporation of S-CNT.
No direct relationship between CNT internalization and inflammatory response has been clearly established yet [14, 46–48], but it would be tempting to explain these singular responses to S-CNT by the higher cellular uptake observed for these CNT. In order to evaluate this hypothesis, we took into account the percentage of CNT-positive cells (i.e. cells containing CNT) when interpreting the results of our cytokine and oxidative stress assays. The percentage of cells containing CNT within vesicles was 2.35 higher with S-CNT than with L-CNT (25.33% versus 10.77% for S-CNT and L-CNT respectively), while the number of CNT contained inside each vesicle was similar for both CNT. At the same time, TNF-α and CXCL-2 secretions were respectively 6 times and almost 17 times higher in response to S-CNT exposure than in response to L-CNT exposure (25.78 versus 4.31 ng/ml for TNF-α and 106.48 versus 6.38 ng/ml for CXCL-2, respectively). Therefore, the increase in inflammatory response (revealed by the amount of cytokines produced by cells exposed to S-CNT compared to L-CNT) was clearly more important than the difference in CNT content per cell between the two groups. Interestingly, a similar pattern was observed for HO-1 and SOD-2 mRNA expression. Taken together, these results strongly suggest that the enhanced inflammatory and oxidative responses to S-CNT were not only a consequence of a higher uptake of S-CNT by cells, but could also result from the material's intrinsic characteristics (length, surface features) that varied between S-CNT and L-CNT. Both S- and L-CNT induced a similar decrease in cellular viability, but, as discussed earlier, only exposure to S-CNT was associated with increased pro-inflammatory and pro-oxidative responses. Similar dissociation between cell mortality and inflammatory and/or oxidative response has been described in the literature [14, 49–51]. In the present case, the difference observed between the two CNT could be due to different cellular pathways targeted by S- and L-CNT, such as what has been described for MWCNT and asbestos in lung epithelial cells , or MWCNT embedded in different polymers . Another possibility could be a preferred interaction of proteins and/or DNA with L-CNT compared to S-CNT, further leading to false negative results when the amount of inflammatory proteins or DNA was quantified after cell incubation with L-CNT. However, internal controls in the experimental set-up allow us to rule out such possibility.
The absence of an inflammatory effect of L-CNT is all the more surprising since these CNT have a similar length distribution to those eliciting a clear inflammatory response (both in vitro and in vivo) in studies by Donaldson and coworkers (i.e. CNTlong1) [21, 23]. In those studies, the inflammatory reaction induced by long CNT has been related to a phenomenon called “frustrated phagocytosis”, which is characterized by macrophages not being able to eliminate long and rigid fibers because of incomplete engulfment . Experimental studies suggest that “frustrated phagocytosis” has a dramatic influence on the sustained generation of ROS , which in turn contribute to the secretion of inflammatory mediators [54, 55]. In our study, almost no frustrated phagocytosis was observed either for S-CNT or L-CNT. Given the complexity of in vivo environment, care should be taken when comparing in vitro and in vivo data. Indeed, frustrated phagocytosis is a critical issue for fiber toxicity, but an equivalent important issue is particle clearance [22, 56] that can hardly be evaluated by in vitro studies. To explain the difference between our findings with L-CNT and the results from Donaldson and coworkers with long CNT, one hypothesis is the use in our experiments of less rigid CNT that lead to less frustrated phagocytosis than rigid ones [16, 57]. We did not measure the rigidity of L-CNT, but our CNT were thinner than the ones used by Donaldson and coworkers (mean diameter 42 vs 85 nm respectively)  suggesting that they could be bent more easily, and therefore be less subjected to frustrated phagocytosis and more fully engulfed.
In conclusion, our results stress the difficulty to address the role of one single physico-chemical parameter at a time when dealing with CNT biological effects, even though a controlled synthesis procedure was used. Surface properties should be considered as essential determinants, alongside the length, in CNT-induced oxidative stress and inflammation, especially when dealing with the safe design of CNT for applications in nanomedicine.
Two MWCNT samples (S-CNT precursor and L-CNT) were produced by aerosol-assisted CCVD. The method is based on the catalytic decomposition of liquid hydrocarbons by pyrolysing mixed aerosols containing both the hydrocarbon and the metallic source which simultaneously and continuously fill the reactor . A solution composed of ferrocene dissolved in toluene (2.5 wt.% for PS-CNT and 5 wt.% for L-CNT) was used to synthesize the two CNT samples at 850°C. Following this procedure, samples are formed of aligned CNT carpets covering the reactor walls. The duration of the aligned growth of CNT was fixed at 10min for PS-CNT and only 2min for L-CNT. Once detached from the reactor walls by scrapping off, PS-CNT sample was treated in de-ionized water for 7 weeks using ultrasonic bath (25 kHz, 100% power) in order to shorten the CNT and reach a desired length distribution. The final dry sample of S-CNT was obtained by evaporating water in a fume hood.
Optical, scanning electron and transmission electron microscopies
Samples were observed using optical (Olympus BX 60 optical microscope coupled to a Color view digital camera), scanning electron (SEM, FEG-SEM; Carl Zeiss Ultra 55, field emission gun) and transmission electron (TEM) microscopies to evaluate the quality of the MWCNT (i.e. morphology, structure, and presence of synthesis by-products), and also to determine the length distribution. Morphology and thickness of the CNT carpets were investigated by SEM on cross sections of aligned CNT carpet grown on reference quartz substrate (PS- and L-CNT) which were fixed on the SEM sample holder with a carbon adhesive tape. Beam voltage was 5 kV, working distance 3 mm, and size aperture 30μm. We used SE2 or InLens electron detectors. To perform TEM analysis, CNT powder was dispersed in ethanol with US bath for less than 1 min. One droplet of this suspension was then deposited on a Cu grid covered with lacey carbon film. Grids were observed on a Philips CM12 TEM microscope operating at 120kV.
Thermogravimetric analysis (TGA 92–16, 18 SETARAM apparatus) was performed under flowing air at a temperature up to 1000°C (10°C min-1 heating ramp) to determine the sample initial iron content by measuring the remaining iron oxide weight.
X-ray diffraction (XRD) experiments were carried out in transmission geometry on a rotating anode generator. The Molybdenum Kα X-ray radiation was used as incident wavelength (λ = 0.711 Å) so that fluorescence from iron-based particles was relatively low. Collimator, sample, and detector were altogether placed in a vacuum chamber in order to minimize air scattering. Dry samples (i.e. MWCNT powders) were placed into glass capillaries. A two-dimensional phosphorescent imaging plate was used as the detector; the signal was then integrated angularly to obtain the wave-vector dependence of the scattered intensity.
X-ray induced photoelectron Spectroscopy
The surface chemical composition of both S-CNT and L-CNT samples was determined by XPS (X-ray induced Photoelectrons Spectroscopy) using a Kratos Analytical Axis Ultra DLD spectrometer with monochromatic Al Kα X-ray radiation (hν = 1486.6eV). C1s, O1s and Fe2p spectra were recorded at a take-off angle of 90° with a 700μm by 300μm slot aperture and 20eV pass energy. The energy scale of the instrument was calibrated by setting Au 4f7/2 = 84.0 eV, Ag3d5/2 = 368.7 eV. Data from 3 independent measures were acquired with Kratos Analytical Vision 2 software. Peak fitting was performed after Shirley baseline background subtraction  using Thermo Electron Software. A Lorentzian/Gaussian ratio of 70% was applied to sp2 carbon peak and 30% to other C1s, O1s, Si2p and Ti2p oxide peaks. The energy of sp3 carbon peak was fixed to 285.1 eV with a full width at half maximum (FWHM) of 1.5 eV. The atomic sensitivity factors used for semi-quantitative analysis were those given by Scofield  (C1s = 1.0, O1s = 2.93, Fe2p3/2 = 10.82 and Si2p =0.82, relative to C1s = 1.00).
Endotoxin contamination of CNT
S- and L-CNT samples were assessed for endotoxin contamination using the Limulus Amebocyte Lysate assay (Lonza), performed as per the manufacturer's instructions.
Cell culture and exposure to CNT
RAW 264.7 murine macrophages were purchased from the American Type Culture Collection (Manassas, VA). Cells were cultured in Dulbecco's Modified Eagle Medium (DMEM) supplemented with 10% heat-inactivated fetal calf serum and antibiotics (streptomycin, 10 mg/mL; penicillin G, 10000 IU/mL; and amphotericin B, 25 μg/mL) at 37°C in a humidified atmosphere of 5% CO2/95% air. Sub-confluent cells were then exposed for 6 or 24 hours to a scaling dose of MWCNTs (0.1-50 μg/ml; 0.2-20 μg/cm2) prepared by dispersion of the dry material sample in serum-free cell culture medium. For homogenization purpose, the MWCNT suspension was US bath-sonicated and vortexed just before cell stimulation. In a subset of experiments, cells were pretreated with the antioxidant N-Acetyl Cystein (NAC, 2 mM) 1 hour prior to CNT exposure, or with the iron chelator Desferrioxamine (DEF, 100 μM, as previously described ).
Morphology of cells exposed to CNT
Cell morphology was evaluated by optical microscopy after standard Harris haematoxylin-eosin/phloxin staining of cells exposed for 24h to 10μg/mL of CNTs.
Cellular uptake of CNT
Cells exposed for 24 h to 10 μg/ml of CNTs were analyzed by TEM as described previously . Briefly, cell monolayers were resin-embedded and then processed as to prepare semi-thin sections (2 by 2mm; 200nm thick) for ultra-structural cytology and ultra-thin sections (0.5 by0.5mm; 70nm thick) for TEM analysis.
The percentage of cells having internalized CNTs in vesicles was analyzed by optical microscopy on Toluidine blue-stained semi-thin sections of the cell monolayer (2 by2mm, prepared from TEM block specimen). For each condition of stimulation, 5 fields were selected from the top to the bottom across the semi-thin section. Analysis was performed blinded by 2 independent observers (SL and CB). The coefficient of variation for the measurement was <5%.
The calculation of the number and length of internalized CNT was performed on ultra-thin TEM sections. For each condition of stimulation, 15 fields were selected from the top to the bottom across the ultra-thin section, and a minimum of 50 cells per sample was observed. CNT number was evaluated in vesicles only (since CNT in cytoplasm are very difficult to observe), and CNT length was measured both in vesicles and free in the cytoplasm. Analysis was performed blinded by 3 independent observers (CB, MP and SL). The coefficient of variation for the measurement was <5%.
Cellular viability was assessed using 2 methods: WST-1 assay, and the quantification of DNA content. These tests were performed as previously described . Results were expressed as the means of at least 3 independent experiments, each of 6 replicates, given as the ratio of the mean for each condition to the mean of the control condition (cells exposed to DMEM). Since nanomaterials could interfere with cytotoxicity tests [62, 63], we performed the assays incubating dyes with nanotubes only (100 μg/ml of S-CNT or L-CNT) and then measured absorbance. No positive or negative interference of S-CNT or L-CNT with any assays was observed (data not shown).
Reverse transcription and quantitative PCR (Q-PCR)
Primers used for real-time quantitative PCR
The concentration of the proinflammatory cytokine TNF-α and the chemoattractant chemokine CXCL2 in culture supernatant was determined by ELISA (R&D Systems, Lille, France), as previously described . Interference of NP with ELISA assay was assessed by quantifying the amount of known concentrations of TNF-α or CXCL2 in presence or in absence of CNT. No interference was observed (Additional file 4: Figure S4). Results are expressed as pg/μg protein.
Each value is the mean ± Standard Error of the Mean (SEM) of at least 4 experiments performed in triplicate. Data were analyzed with the GraphPad Prism 4.0 software (La Jolla, CA, USA). Comparisons between multiple groups were performed by using Kruskall–Wallis’ non-parametric analysis of variance test followed, when a difference was detected, by two-by-two comparisons with the Mann–Whitney’s U test. P-values <0.05 were considered significant.
The authors would like to thank the TEM Team (CEA Saclay, DSV) where some experiments were performed, and also the Centre Commun de Microscopie Electronique de l’Université Paris Sud Orsay (CNRS UMR8080) for their help in the preparation of the biological specimen for TEM analysis. This work was supported by the Région Ile-de-France in the framework of C'nano IdF (NANOTUBTOX project), as well as for the MEG FEG instrument used, C'Nano-IdF is the nanoscience competence center of Paris Region, supported by CNRS, CEA, MESR and Région Ile-de-France.
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License ( http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.