Toxicity determinants of multi-walled carbon nanotubes: The relationship between functionalization and agglomeration

Graphical abstract


Introduction
After their discovery by Iijima [29], carbon nanotubes (CNT) have been increasingly used in advanced industrial applications. Indeed, due to their excellent physico-chemical, electrical and mechanical properties, they are applied in numerous technological fields such as polymer composites, microelectronics, energy storage and sensors [18]. CNTs are graphitic hollow filaments of variable lengths, up to several hundred micrometers, depending on the production method. Single-walled carbon nanotubes (SWCNT) are composed of a single cylindrical sheet of graphene, while multiwalled carbon nanotubes (MWCNT) consist of several concentric, coaxial, rolled up graphene sheets [44]. The CNT diameter typically ranges from 0.4 to 3 nm for SWCNT and from 1.4 to 100 nm for MWCNT [62]. Chemical vapor deposition (CVD) is the most frequently used method for CNT synthesis and is the prevalent technique for mass production of CNT due to its easy scaling-up [9]. Different preparations of MWCNT yield contrasting results about their bio-safety, as some preparations appear to be highly hazardous, while others seem harmless [13,24,47,51]. Thus, toxicity evaluation of these materials has to be taken on a case-by-case study, because CNT cannot be regarded as a simple chemical substance. Therefore, the investigation of MWCNT toxicity has to be designed according to their specific features and cannot adopt the same strategy of the conventional toxicology studies applied for general chemical compounds [42].
The conventional view point is that SWCNT exhibit significant cytotoxicity to human and animal cells through various mechanisms [12,33,40], whereas MWCNT are considered less active [32]. However, increasing evidences suggest that MWCNT can be inflammogenic and fibrogenic in rodents [11,20]. Moreover, a specific MWCNT preparation (MWCNT-7) has been recently classified as possibly carcinogenic to humans (Group 2B) by the International Agency for Research on Cancer, although other preparations of MWCNT were not classifiable respect to their carcinogenicity to humans [26]. These contrasting results can be attributed to the use of materials with different degree of purity and structural features [68], as well as to the conditions adopted for the in vitro studies [72] and the cell types tested for the assays [35]. Thus, despite several excellent reviews on the advancement in knowledge about CNT toxicity [2,16,19,36,37,46,63], the current understanding of the physico-chemical determinants of MWCNT toxicity is still incompletely settled [30].
In general, the toxicity of CNTs is attributed to their physicochemical characteristics, such as length, diameter, shape, purity, surface area and surface chemistry [37,38], which, in turn, are remarkably influenced by the synthetic route. CNT contamination by catalyst residues is unavoidable during their production, so that the impact of residual catalyst metals may be also important [38]. Among the various determinants known to influence the biological activity of CNT, also functionalization with carboxylic groups has been considered and investigated. However, contrasting results have been reported in this case too. Functionalization enhances toxicity in airway epithelial BEAS-2B cells [10,65], while on the contrary, it has been demonstrated to suppress bioactivity in other epithelial models [39] and in macrophages [28]. Also in vivo models have yielded diverging data, with mitigating [31,58] or enhancing [64] effects reported for different endpoints.
When suspended in biological media, MWCNT adsorb macromolecules that form a corona and modify the properties of the material [41]. Protein adsorption is heavily influenced by surface curvature and surface area. Surface curvature is directly related to the outer diameter [27,52], while surface area is influenced by outer diameter, pore volume and surface chemistry. In particular, surface chemistry influences the adsorption of macromolecules to CNT depending on: (1) the available specific surface area (functionalized CNT may present higher or lower surface area values according to the functional groups which are attached on their sidewalls); (2) the solution pH, in relation to CNT pKa value (at pH below pK ␣ adsorption is increased); (3) the ionic strength in the solution (ionic strength increases adsorption) [14]. However, the relationship between protein corona, functionalization and MWCNT biological effects is still under investigation.
In this work, we tested the hypothesis that functionalization with carboxyl or amino groups affects the formation of protein corona and, hence, the biological effects of MWCNT. To this purpose, we preliminary examined four MWCNT preparations, three synthesized in our laboratory via CVD method and one commercial, of comparable length but different diameter and surface chemistry, and compared their effects on cell viability with those exhibited by a benchmark MWCNT preparation obtained from the Joint Research Centre (JRC) repository (Ispra, Varese, Italy). One of the preparations exhibiting changes in biological activity attributable to functionalization was then further investigated, studying both protein corona and several toxicological endpoints.

Supply and synthesis of carbon nanotubes
A thermal chemical vapor deposition (CVD) reactor was used to synthesize MWCNTs. The reactor consists of a horizontal quartz tube with 3.4 cm inner diameter and 100 cm length housed in a three-zone 80 cm long cylindrical furnace. Synthesis of CNT was performed by two approaches. In the first case, camphor and ferrocene were used as carbon source and catalyst, respectively, while acetylene as carbon source and iron particles supported on Al 2 O 3 substrate as catalyst were used in the latter.
More specifically, for the preparation of the CNT1 sample (see Table 1), a Pyrex flask containing the reagent mixture, which consisted of camphor (96% purity in weight, Sigma-Aldrich, Athens, Greece) as carbon precursor and ferrocene (98% purity in weight, Sigma-Aldrich, Athens, Greece) as catalyst, in a 20/1 mass ratio, was connected to the tube nearby the nitrogen inlet. Nitrogen gas flow was used to carry the gas mixture of precursors toward the center of the furnace, where pyrolysis took place at 850 • C, and CNT were grown. For the preparation of CNT2 and CNT3 samples, the catalyst particles were placed on a ceramic boat inside the quartz tube and in the middle of the isothermal zone of the reactor. Firstly, a constant nitrogen flow rate was passed through the quartz tube to remove the air from the system, and, then, the reactor was heated at 700 • C under nitrogen flow. Subsequently, nitrogen was replaced by a mixture of acetylene/nitrogen.
Commercial MWCNT were obtained from Nanothinx S.A. (Patra, Greece). In particular, two types of commercial nanotubes were used in this study: NTX1 (pure MWCNT, REF) and NTX5 (MWCNT functionalized with −COOH groups, O-REF). The physical characteristics of these nanomaterials (according to the technical datasheet) are presented in Table 1. Additionally, a TEM image of the NTX1 nanomaterial is presented in Fig. 1e.
The NM-401 MWCNT preparation was obtained from the JRC Nanomaterials Repository hosting representative industrial nanomaterials (Ispra, Varese, Italy). This nanomaterial is classified as a representative test material (RTM) and includes a (random) sample from one industrial production batch sub-sampled into vials under reproducible (GLP) conditions, with the stability of the subsamples monitored. A detailed physico-chemical characterization of this material is provided in the specific JRC Report [54].

Carbon nanotubes purification and functionalization process
After the synthesis, the raw products were milled and exposed at atmospheric air flow at 400 • C for 1 h, aiming at the removal of amorphous carbon. Afterwards, they were purified with constant boiling of 5 M HCl in a Soxhlet extractor in order to remove the remaining metal particles. Finally, the purified CNTs were washed with distilled water and dried in oven. To activate the CNT surface with −COOH groups (O-CNT1, O-CNT2 and O-CNT3), an acid solution mixture of 6 M HNO 3 :H 2 SO 4 in a ratio of 1:3 was used. Then, the CNT/acid mixture (0.15 g CNTs/10 mL acid solution) was stirred for 48 h at 80 • C. The suspension was filtered, and the black powder was washed with ethanol, acetone and distilled water and, eventually, dried in oven.
For the preparation of −CONHCH 2 CH 2 NH 2 functionalized

. X-ray diffraction
The measurements were performed at room temperature with a Bruker D8 Advance Twin Twin X-ray diffractometer (Bruker Corporation, The Woodlands, TX 77381, USA) equipped with a Cu K␣ radiation source (wavelength = 1.5418 Å).

Scanning electron microscopy
The morphology of CNT was determined using a Nova NanoSEM 230 (FEI Inc., Hillsboro, OR, USA) microscope with W (tungsten) filament.

Transmission electron microscopy
TEM measurements were performed with a Tecnai G2 Spirit Twin 12 microscope (FEI Inc., Hillsboro, OR, USA) after the dispersion of CNT in distilled water.

Fourier transform infrared spectroscopy
FT-IR analysis was performed by using a ThermoScientific Nicolet 6700 Fourier Transform Infrared Spectrometer (Thermo Fisher Scientific Inc., Waltham, MA, USA).

Surface area analysis
Brunauer, Emmett and Teller (BET) analysis was performed with a Micromeritics TriStarII 3020 instrument. The specific area was determined by a 5-point BET measurement with UHP nitrogen as the adsorbate and liquid nitrogen as the cryogen.

Suspension Stability Index
A kinetic analysis of suspension stability was performed by monitoring the absorbance of the suspension at 550 nm for different times, using an UV-vis DU650 spectrophotometer (Beckman Coulter SpA, Milano, Italy). Typically, 1 mL of the MWCNT suspension at 128 g/mL was prepared, and absorbance readings were taken at 5 min intervals for 35 min.

Analysis of adsorbed proteins
To analyze proteins adsorbed to MWCNT, two aliquots of each preparation (REF, O-REF, A-REF), suspended at 1.28 mg/ml in PBS + 0.05% BSA, were diluted 1:10 in DMEM + 10% FBS or in plain DMEM. After an 1 h-incubation at 37 • C under continuous stirring, samples were rapidly centrifuged at 13,000 × g and the pellets were washed three times in bidistilled water. The pellets were then suspended in sample buffer (31.25 mM Tris-HCl, pH 6.8, 3% SDS, 10% glycerol 100 mM, DTT, 0.02% bromophenol blue) and boiled for 10 min. Aliquots were used for SDS-PAGE. After the electrophoresis, gels were washed in water and stained with silver staining (Cosmo Bio Co., Ltd., Tokyo, Japan, Cat. No. 423413) according to manufacturer's instructions. In parallel, the pellets of other two aliquots, also incubated in DMEM + 10% FBS or in plain DMEM, were suspended in water, and proteins quantified with a modified micro-Lowry procedure [5].

Cells and experimental treatments
Mouse peritoneal monocyte-macrophage cells (Raw264.7 line) were cultured in Dulbecco's modified Eagle's medium (DMEM) sup-plemented with 10% Foetal Bovine Serum (FBS), 4 mM glutamine, and antibiotics (streptomycin 100 g/mL penicillin, 100 U/mL). Human alveolar carcinoma epithelial cells (A549) were cultured in DMEM supplemented with 10% FBS, 2 mM glutamine, and antibiotics. Calu-3 lung adenocarcinoma cells [23] were cultured in EMEM supplemented with 10% FBS, 2 mM glutamine, 1 mM sodium pyruvate, and antibiotics. The three cell lines were purchased from the Cell bank of the Istituto Zooprofilattico Sperimentale della Lombardia e dell'Emilia (Brescia, Italy). Murine alveolar macrophages (MH-S), a gift of Prof. Dario Ghigo, University of Torino (Italy), were also originally provided by the Cell Bank of the Istituto Zooprofilattico Sperimentale della Lombardia ed Emilia-Romagna (Brescia, Italy) and cultured in RPMI1640 medium supplemented with 10% FBS, antibiotics, l-glutamine (2 mM) and ␤-mercaptoethanol (0.05 mM). Before and during treatments, all cultures were maintained in a humidified atmosphere of 5% CO 2 in air in 10-cm dishes.
For treatments, cells were seeded in 96-well dishes at a density of 1.5 × 104 cells/well for viability experiments and in 24-well dishes at a density of 1 × 105 cells/well for nitrite determination. MWCNT were added from stock solutions (2.56 mg/mL) prepared according to a modified Nanogenotox protocol (http://www.nanogenotox.eu/files/PDF/ web%20nanogenotox%20dispersion%20protocol.pdf) in which Caand Mg-free PBS replaced water + Bovine Serum Albumin (BSA, 0.05%) in order to maintain the physiological osmolality. Doses of materials were expressed in g/cm 2 of monolayer; a fixed proportion was maintained between the culture surface and the volume of incubation, so as to maintain the relationship between doses in g/cm 2 and doses in g/mL (dose in g/cm 2 × 1.6 = doses in g/mL) fixed.
No surfactant agent was employed, as we decided to allow aggregation in order to characterize CNT toxicological properties in real-life resembling conditions, i.e., occupational environment.

Cell viability
To evaluate the effects of the different preparations on cell viability, Raw264.7 and A549 cells -two lines representative of macrophages and airway epithelial cells -were treated with the MWCNT to be tested or with the benchmark NM-401 at the indicated doses, using for each preparation a separate dish in which it was tested in parallel with NM-401. Cell viability was measured after 24 h, 48 h, or 72 h of exposure to MWCNT using the resazurin assay [45]. Since it is known that nanomaterials could interfere with cytotoxicity tests, we incubated the dye with each MWCNT preparation at the maximal dose used in the viability experiments (80 g/cm 2 ) in the absence of cells, so as to evaluate autofluorescence; moreover, MWCNT were also added just before the reading, so as to assess possible fluorescence quenching. No interference was detected in either case (data not shown).

Medium nitrite concentration
Nitrite concentration, as a proxy for NO output, was determined in Raw264.7 and MH-S macrophages, which are known to be competent for NO production, through a fluorimetric approach, based on the production of the fluorescent molecule 1H-naphthotriazole from 2,3-diaminonaphthalene (DAN) in an acid environment [43]. After 48 h and 72 h of incubation with the materials, 100 L of medium were transferred to black 96-well plates with a clear bottom (Corning, Cambridge, MA). DAN (20 L of a solution of 0.025 mg/mL in 0.31 M HCl) was then added and, after 10 min at RT, the reaction was stopped with 20 L of 0.7 M NaOH. Standards were performed in the same medium from a solution of 1 mM sodium nitrite. Fluorescence ( ex 360 nm; em 430 nm) was determined with a multimode plate reader Perkin Elmer Enspire (Waltham, Massachussets, USA).

Real time PCR
Total RNA was isolated from Raw264.7 cells with GenElute Mammalian Total RNA Miniprep Kit (Sigma-Aldrich, Milan, Italy). After reverse transcription, aliquots of cDNA from each sample were amplified in a total volume of 25 L with the Go Taq PCR Master Mix (Promega Italia, Milan, Italy), along with the forward and reverse primers (5 pmol each) reported in Table 2 to evaluate the expression of different pro-inflammatory genes: inducible nitric oxide synthetase (Nos2), prostaglandin-endoperoxide synthase 2 (Ptgs2), interleukin 6 (Il6) and interleukin 1␤ (Il1b). Real-time PCR was performed in a 36-well RotorGeneTM3000, version 5.0.60 (Corbett Research, Mortlake, Australia). For all the messengers to be quantified, each cycle consisted of a denaturation step at 95 • C for 20 s, followed by separate annealing (30 s) and extension (30 s) steps at a temperature characteristic for each pair of primers (see Table 2). Fluorescence was monitored at the end of each extension step. Melting curve analysis was added at the end of each amplification cycle. The analysis of the data was made according to the relative standard curve method [7]. Expression data were reported as the ratio between each investigated mRNA and the Gapdh mRNA.

Confocal microscopy
The interaction between MWCNT and macrophages and, in particular, the internalization of the materials, was studied in confocal microscopy. Confocal analysis was carried out with a LSM510 Meta scan head integrated with an inverted microscope (Carl Zeiss, Jena, Germany). Samples were observed through a 40× (1.3 NA) oil objective. Image acquisition was carried out in multitrack mode, i.e., through consecutive and independent optical pathways. Vertical sections were obtained with the function Display-Cut (Expert Mode) of the LSM510 confocal microscope software (Microscopy Systems, Hartford, CT).
Reconstructions were performed from z-stacks of digital images (minimum 32 confocal sections, z-axis acquisition interval of 0.39 mm), processed with the Axiovision module inside 4D release 4.5 (Carl Zeiss, Jena, Germany), applying the shadow or the transparency algorithm.
For microscopy, phagocytosis-competent Raw264.7 macrophages were seeded on four-chamber slides at a density of 15 × 10 4 cells/cm 2 . After 24 h, monolayers were exposed to MWCNT and incubation prolonged for further 24 h. In the last 20 min of exposure, cells were transferred to serumfree medium supplemented with CellTracker TM Red CMPTX (8 M, Molecular Probes, Invitrogen) to label the cytoplasm; in the last 5 min 1,5-bis[2-(di-methylamino)ethyl]amino-4,8-dihydroxyanthracene-9,10-dione (DRAQ5 ® , 20 M, Alexis Biochemicals, San Diego, CA, USA) was also added to the incubation medium to counterstain the nuclei. After the experimental treatments, cell monolayers were rinsed in PBS and fixed with 3.7% paraformaldehyde (PFA) at room temperature for 15 min. After fixation, specimens were mounted on glass slides with fluorescence mounting medium (Dako Italia SpA).
Excitation at 488 nm and reflectance were used to visualize MWCNT (shown in white); excitation at 543 nm and emission recorded through a 580-630 nm band pass barrier filter were used to visualize the cytoplasm (shown in red); excitation at 633 nm and emission through a 670 nm long pass filter were recorded to visualize the nucleus (shown in blue).

Colony forming efficiency (CFE) assay
The CFE test is a clonogenic assay that measures the ability of a single cell to form a colony [25] and can be used to determine cytotoxicity induced by nanomaterials in any adherent cell models. Indeed, since it is a label-free test (non-colorimetric, nonfluorescent), CFE assay lowers the possibility of the occurrence of interference due to nanomaterials [50]. For the test, 10 2 A549 cells, were seeded in tissue culture dishes (BD Falcon 60 × 15 mm Style, BD Falcon, USA). After 24 h, culture medium was replaced with complete medium supplemented with MWCNT (2.5-5 g/cm 2 ), and the exposure prolonged for 72 h. After medium renewal, incu- Table 2 Primers used for gene expression studies.

Trans-Epithelial Electrical Resistance
Decrease in the Trans-Epithelial Electrical Resistance (TEER) indicates a perturbation of the barrier function of tight epithelial monolayers, and is a parameter used to evaluate epithelial damage by MWCNT [56,55,4,57] or other nanomaterials [22]. TEER was measured using an epithelial voltmeter (EVOM, World Precision Instruments Inc., Sarasota, FL, USA). For the experiments, Calu-3 cells, an airway epithelial cell line which forms a tight monolayer when cultured in a double chamber culture system, were seeded into cell culture inserts with membrane filters (pore size 0.4 m) for Falcon 24-well-multitrays (Cat. N • 3095, Becton, Dickinson & Company, Franklin Lakes, NJ, USA), at a density of 2.3 × 10 4 cells/cm 2 , and grown for 12d until a tight monolayer was formed (TEER > 1000 × cm 2 ). MWCNT were added in the apical chamber from a 1 mg/mL stock solution without changing the medium and TEER measured at the indicated times of treatment. TEER changes were expressed as the percentage of the initial value adjusted for control cell monolayers according to Eq. (1) [59]: Final TEER treated Final TEER control × Initial TEER control Initial TEER treated (1)

Statistics
Data are expressed as means ± standard deviation (mean ± SD). Statistical analyses were performed with one-way ANOVA with Tukey post hoc test. GraphPad Prism ® software version 6.00 (Graph-Pad Software Inc., San Diego, CA) was used. Results were considered significant at p < 0.05.

Reagents
Whenever not stated otherwise, all the reagents were provided by Sigma-Aldrich, Milan, Italy.

MWCNT characterization
After the synthesis, purification and functionalization of MWCNT, their structure, chemical composition and purity were investigated. Table 1 summarizes the features of the MWCNT preparations tested. In Fig. 1, TEM images of selected MWCNT samples are shown. Hollow filamentous structures were revealed for all samples and their specific characteristics were found to vary according to the different production method. In particular, the diameter of MWCNT ranges between 15 and 100 nm depending on the experimental conditions adopted for synthesis. In all cases, the length of CNT ranges between 10 and 30 m. Iron particles could be seen mostly encapsulated within the core of the MWCNT.
SEM images of the CNT preparations produced in our lab, is depicted in Fig. 2, along with a representative EDS analysis, XRD diagram and TGA graph of CNT3 sample. In Fig. 2A-D the SEM images and the corresponding EDS analysis revealed uniform diameter distribution and high purity. The main features of XRD patterns of CNT were close to those of graphite (Fig. 2E); a typical XRD pattern consisted of bands located near the (0 0 2), (1 0 0) and (1 1 0) reflections of graphite. The first peak at 2Â ∼ 26 • can be attributed to the (0 0 2) reflection of graphite while an asymmetric diffraction peak at 2Â ∼ 43 • was assigned to the (1 0 0) reflection of graphite, which is typically observed in MWCNT [48,49]. TGA was used in order to evaluate the thermal stability and the purity degree of the produced CNT (Fig. 2F). The initial weight loss of 1.5%, observed at temperatures up to 480 • C, could be attributed to the burning of amorphous carbon material. The residual weight at the end of the thermal oxidative curve (5.7%) corresponded to the iron catalyst particles. The differential thermogravimetric analysis (DTA) curve showed only one narrow peak at 560 • C indicating the high thermal stability in air atmosphere and the uniform graphitized structure of the CNT produced. The overall purity of CNT3 was around 92.8%. Table 1 includes the TGA data for all the MWCNT preparations. Purity range is that expected for MWCNT prepared with the methods adopted here [6]. There were no substantial differences among the various raw MWCNT preparations for EDX and XRD results (not shown). The specific surface area (SSA) values, obtained with BET (Table 1), indicated that the surface area value increases as the outer diameter of pristine materials decreases [6].
The functionalization process was studied via FT-IR spectroscopy and TG analysis. Specifically, FTIR spectra of pristine Thermogravimetric analysis was applied to quantify the functionalization degree of O-REF and A-REF samples [15]. From Fig. 3B, it is clear that the thermal degradation of the modified CNT is a multistage process. The weight loss occurs at temperatures up to 160-200 • C (∼1.3% for O-REF and ∼4.7% for A-REF) and is associated with the release of moisture which is physically or chemically absorbed. CNT-COOH and CNT-amino samples exhibit a first weight loss at 180-450 • C (∼4.4%) and 200-400 • C (∼8.7%) corresponding to the burning of carboxyl and −CONHCH 2 CH 2 NH 2 groups, respectively. The second loss (temperatures above 400-450 • C) is attributed to the CNT burning [15,17,71]. The approximate molar% functional groups on CNT were estimated [71] equal to ∼1.3 mol% for both cases, indicating the high conversion of the functionalization process.  It can be observed that after −COOH functionalization, MWCNT showed a slight increase in their specific area. This may be attributed to the oxidation treatment which creates sidewall openings and defects. On the other hand, the attachment of amino functional groups does not lead to a comparable increase, since the groups added are more bulky than −COOH ones, blocking CNT pores [6,8].
The surface charge of the MWCNT preparations can be estimated taking into account the Point of Zero Charge (PZC) theory [3]. According to this theory, pH is considered to be the most important factor which determines the surface charge of CNTs, with the surface charge negative when the solution pH is higher than the PZC value, while otherwise becomes positive. The PZC can be estimated to be 6.6 for the pristine REF    1, 3, 5) was different for the three MWCNT preparations. In particular, when MWCNT were incubated with serum, A-CNT adsorbed more BSA than the other two preparations but many other bands were fainter. More-  over, while few bands were clearly more abundant in O-REF, several others were less evident in functionalized than in pristine MWCNT.

Cell viability
The effect of the different MWCNT preparations (dose range 2.5-80 g/cm 2 ) on the viability of Raw264.7 macrophages was tested with the resazurin assay every 24 h up to a 72 h-exposure to the materials (Table 3). NM-401 MWCNT, which we have found endowed with marked cytotoxic effects toward Raw264.7 macrophages (unpublished results), were used as a benchmark. Significant effects on cell viability were observed only at 72 h of incubation. However, while NM-401 produced a clear cut dose-dependent decrease of cell viability, only moderate effects were observed for the other MWCNT. In particular, no decrease in viability was observed with MWCNT wider than 40 nm. On the other hand, MWCNT preparations with diameters between 15 and 40 nm produced a decrease in viability, although no material, but the benchmark NM-401, caused a viability loss larger than 50%. The direct comparison among the IC 20 values obtained with pristine MWCNT of different diameter indicated that cytotoxicity was roughly inversely proportional to diameter (REF = CNT3 > CNT2 = CNT1). Given that NM-401 are long, needle-like, while our samples are even longer, but entangled, we can speculate that cytotoxic properties are also associated with MWCNT shape, although this issue has not been further investigated.
Functionalization with -COOH groups did not change appreciably MWCNT effects on cell viability for CNT1 and CNT2, while IC 20 values of the functionalized O-CNT3 and O-REF were larger than those obtained for the corresponding pristine preparations CNT3 and REF, indicating a mitigating effect.
To ascertain if the mitigating effects of functionalization were specific for macrophages, we extended the study to the human epithelial airway cells A549, a cell model widely adopted in toxicological studies. On these cells, we studied three preparations (pristine, carboxyl-functionalized and amino-functionalized) of the thinnest MWCNT listed in Table 1. The results (Fig. 6) revealed that the pristine REF determined a clear cut dose dependent decrease of cell viability up to 35% at the maximal dose (80 g/cm 2 ). A much slighter, but still significant decrease was instead observed in the case of the COOH-functionalized preparation (up to 7% at the maximal dose). On the contrary, A-REF did not show a significant cytotoxic effect even at the highest dose tested (80 g/cm 2 ). As expected, the benchmark NM-401, used at the dose of 80 g/cm 2 , determined a marked decrease (around 50%) of cell viability. Thus, also in epithelial cells functionalization with either -COOH or amino groups mitigated the effect on cell viability.
To deepen the investigation of the effect of functionalization on cell viability, the different MWCNT preparations (2.5-5 g/cm 2 ) were studied with a CFE assay on A549 (Fig. 7). Only the REF preparation determined a significant decrease of colony number at the highest dose tested of 5 g/cm 2 , while the other MWCNT did not

Macrophage activation
Macrophages play a pivotal role in determining the final outcome of the interaction between the organism and the nanomaterials. In particular, if the exposure to the nanomaterial leads to macrophage activation, this triggers an inflammatory response. Fig. 8 reports data on the production of NO, a major inflammatory mediator produced by activated macrophages, and on the expression of pro-inflammatory genes in Raw264.7 cells exposed to MWCNT. Panels A and B report data on NO production, assessed from the nitrite concentration in the medium after

Discussion
Although relatively less studied than the toxicological properties of SWCNTs, the biological effects of MWCNT have been also repeatedly investigated in vitro and in vivo. According to the literature, various factors influence the biological effects of MWCNT and have the potential to modulate their biological effects [28]. In particular, a correlation has been reported between MWCNT toxicity, their size [28,34], contamination with metal catalysts [1], and surface functionalization [28].
However, it is difficult to find examples in literature where differences between two MWCNT preparations are limited to one parameter, allowing a clear cut assessment of its contribution to the toxicological properties of the nanomaterial. In this study, we have preliminarily compared a panel of MWCNT of comparable length (> 10 m), which differ for diameter (range 15-100 nm). Each preparation was available in pristine form or after COOHfunctionalization. Low acute cytotoxicity was observed with both types of materials in macrophages, although an increased effect was detected for thinner MWCNT. In general, while it is usually accepted that the longer are the MWCNT tested, the higher is their toxicity [28,34,61], contrasting reports exist on the role of diameter. According to some authors, the larger is MWCNT diameter, the higher is the toxicity reported on cells or with in vivo models [28], whereas the converse has been also described [21].
Using one of the MWCNT preparations endowed with significant effects on cell viability, we found that the presence of either carboxyl or amino groups mitigated the effects. The extension of the study to other endpoints indicated that the mitigating effect was not limited to decrease in cell viability of MWCNT. In particular, surface functionalization decreased the activating effects of MWCNT on macrophages, an endpoint related to the pro-inflammatory activity of MWCNT in vivo. This result is at variance with a recently published report [70] in which, however, the preparation of functionalized MWCNT used consisted in nanotubes much shorter than the pristine ones. Also the perturbation of the epithelial barrier, assessed through TEER measurements in airway epithelial cell monolayers, and the decrease in clonogenic activity were clearly more evident for pristine MWCNT than for functionalized derivatives.
For these endpoints, mitigation of toxic effects was observed with either carboxyl-or amino-functionalized MWCNT. As far as we know, while carboxyl-MWCNT have been repeatedly tested, the effects of amino-functionalization on MWCNT toxicity have never been investigated in depth. However, it is interesting that mitigation occurs independently of the net charge present on the MWCNT surface. However, we noted that the three materials more closely compared in this study, REF, O-REF and A-REF, also differed for their agglomeration tendency, with REF forming relatively small agglomerates, while large MWCNT agglomerates were evident in cell cultures treated with carboxyl or amino-functionalized counterparts (Fig. 8). Different tendency to agglomeration of the three preparations tested is consistent with the different sedimentation rates exhibited in both albumin supplemented saline solution and in complete, serum-supplemented culture medium (Fig. 4). These data are similar to the results of Hamilton et al. [28] who, comparing the effects of three preparations of pristine and COOHfunctionalized MWCNT, found that functionalized MWCNT formed larger agglomerates than pristine counterparts when suspended in cell culture medium. Interestingly, the effect was particularly evident for the MWCNT of a length comparable to that of the materials tested here. The presence of larger agglomerates has the obvious consequence of lowering the surface available for the interaction between cells and MWCNT. Thus, if it were expressed per surface area, it would be possible that the actual dose was actually lower  (Fig. 9), a difference of obvious importance in explaining the smaller pro-inflammatory activity of functionalized MWCNT.
The reason why functionalized MWCNT exhibited higher agglomeration tendency has not been specifically investigated here. However, from the data reported in Fig. 4, it is evident that the sedimentation of the MWCNT is profoundly influenced by the presence of proteins in the suspension medium, suggesting a role for the protein corona. Interestingly, recent data also indicate that COOH-functionalization has marked effects on the type of proteins adsorbed by MWCNT [60]. It is, therefore, likely that the greater agglomeration tendency exhibited by functionalized MWCNT in protein-rich media depends on their different interaction with proteins.
The role of agglomeration in MWCNT toxicity is still incompletely defined. In vitro experiments have indicated that agglomerated MWCNT are usually poorly toxic [69]. Greater toxicity of dispersed compared to agglomerated MWCNT has been confirmed with in vivo models [66], although it should be stressed out that the experimental condition adopted to improve dispersion in that paper, i.e. grinding, also substantially modifies the size of the materials.
Most of the contributions that investigate toxicity mitigation by MWCNT functionalization do not address the possible influence of surface modification on agglomeration tendency. A notable exception is represented by the recent contribution by Ursini et al. [65]. These authors have investigated the effects of pristine MWCNT, MWCNT-OH and MWCNT-COOH on two epithelial models (BEAS-2B and A549), showing different entry mechanisms and differential toxicity of the three preparations. The size of agglomerates, also determined in medium, was comparable for pristine and MWCNT-COOH. Therefore, these authors did not further consider agglomeration as a possible toxicity modulator. However, at variance with the materials used here, pristine MWCNT were much longer than MWCNT-COOH, thus rendering difficult a straightforward comparison between the two preparations. Conversely, Hamilton et al. [28], comparing MWCNT preparations of comparable length, measured the size of agglomerates formed by pristine and functionalized MWCNT in culture medium and found that COOH functionalization actually increases the size of agglomerates. However, the possible relationship between this behavior and the reduced bioactivity of functionalized MWCNT, exhibited both in vitro [28] and in vivo [58], was not been further discussed.
In conclusion, this contribution, while confirming that diameter is inversely related to the bioactivity of MWCNT preparations, indicates that the mitigating effects of carboxyl-or aminofunctionalization may be, at least in part, attributable to the greater tendency of functionalized MWCNT to form large agglomerates in protein-rich biological fluids.

Transparency document
The http://dx.doi.org/10.1016/j.toxrep.2016.01.011 associated with this article can be found in the online version.