Toxicity of TiO2 Nanoparticles: Validation of Alternative Models

There are many studies concerning titanium dioxide (TiO2) nanoparticles (NP) toxicity. Nevertheless, there are few publications comparing in vitro and in vivo exposure, and even less comparing air–liquid interface exposure (ALI) with other in vitro and in vivo exposures. The identification and validation of common markers under different exposure conditions are relevant for the development of smart and quick nanotoxicity tests. In this work, cell viability was assessed in vitro by WST-1 and LDH assays after the exposure of NR8383 cells to TiO2 NP sample. To evaluate in vitro gene expression profile, NR8383 cells were exposed to TiO2 NP during 4 h at 3 cm2 of TiO2 NP/cm2 of cells or 19 μg/mL, in two settings—submerged cultures and ALI. For the in vivo study, Fischer 344 rats were exposed by inhalation to a nanostructured aerosol at a concentration of 10 mg/m3, 6 h/day, 5 days/week for 4 weeks. This was followed immediately by gene expression analysis. The results showed a low cytotoxic potential of TiO2 NP on NR8383 cells. Despite the absence of toxicity at the doses studied, the different exposures to TiO2 NP induce 18 common differentially expressed genes (DEG) which are involved in mitosis regulation, cell proliferation and apoptosis and inflammation transport of membrane proteins. Among these genes, we noticed the upregulation of Ccl4, Osm, Ccl7 and Bcl3 genes which could be suggested as early response biomarkers after exposure to TiO2 NP. On the other hand, the comparison of the three models helped us to validate the alternative ones, namely submerged and ALI approaches.


Introduction
Among the nanomaterials used in industries, titanium dioxide (TiO 2 ) nanoparticles (NP) are one of the most used. TiO 2 NP are mainly used as pigments for their brightness, high refractive index, opacity and antimicrobial properties. They are also useful in many cosmetics applications such as

In Vitro Cytotoxicity Study
Regardless of the tests and the doses (Figure 2), exposure of NR8383 cells to TiO2 NP for 24 h did not statistically decrease their viability. Indeed, compared to cells not exposed to NM-105 NP, the metabolic activity measured by the WST-1 test was constant and the 10% decrease observed after a TiO2 exposure of 100 μg/mL and 200 mg/mL was not statistically significant ( Figure 2A). Nevertheless, the release of LDH into the extracellular medium increased up to 30 % in a statistically insignificant manner for 100-μg/mL and 200-mg/mL TiO2 exposure ( Figure 2B). exposure) to NR8383 by WST-1 test and (B) loss of membrane integrity by LDH release measurement. Non exposed cells are negative control (C-) and positive control (C+) for LDH are cells exposed to the lysis buffer (Triton 5%) during 15 min before the measure. Data are presented as means ± standard deviation (SD) of the four biologic replicates.

Transcriptomic Analysis of Dysregulated Genes Following in Vivo, in Vitro Submerged, and ALI
Vitrocell Cloud ® Exposure (ALI) to NM-105 TiO2 NP.

In Vitro Cytotoxicity Study
Regardless of the tests and the doses (Figure 2), exposure of NR8383 cells to TiO 2 NP for 24 h did not statistically decrease their viability. Indeed, compared to cells not exposed to NM-105 NP, the metabolic activity measured by the WST-1 test was constant and the 10% decrease observed after a TiO 2 exposure of 100 µg/mL and 200 mg/mL was not statistically significant ( Figure 2A). Nevertheless, the release of LDH into the extracellular medium increased up to 30 % in a statistically insignificant manner for 100-µg/mL and 200-mg/mL TiO 2 exposure ( Figure 2B).

In Vitro Cytotoxicity Study
Regardless of the tests and the doses (Figure 2), exposure of NR8383 cells to TiO2 NP for 24 h did not statistically decrease their viability. Indeed, compared to cells not exposed to NM-105 NP, the metabolic activity measured by the WST-1 test was constant and the 10% decrease observed after a TiO2 exposure of 100 μg/mL and 200 mg/mL was not statistically significant ( Figure 2A). Nevertheless, the release of LDH into the extracellular medium increased up to 30 % in a statistically insignificant manner for 100-μg/mL and 200-mg/mL TiO2 exposure ( Figure 2B). loss of membrane integrity by LDH release measurement. Non exposed cells are negative control (C-) and positive control (C+) for LDH are cells exposed to the lysis buffer (Triton 5%) during 15 min before the measure. Data are presented as means ± standard deviation (SD) of the four biologic replicates.

Transcriptomic Analysis of Dysregulated Genes Following in Vivo, in Vitro Submerged, and ALI
Vitrocell Cloud ® Exposure (ALI) to NM-105 TiO2 NP. loss of membrane integrity by LDH release measurement. Non exposed cells are negative control (C−) and positive control (C+) for LDH are cells exposed to the lysis buffer (Triton 5%) during 15 min before the measure. Data are presented as means ± standard deviation (SD) of the four biologic replicates.

Transcriptomic Analysis of Dysregulated Genes Following In Vivo, In Vitro Submerged, and ALI Vitrocell
Cloud ® Exposure (ALI) to NM-105 TiO 2 NP As represented by the volcano plot of dysregulated genes (Figure 3), ALI rat cells and lung exposed to TiO 2 NP by inhalation showed similar number of differentially expressed genes (DEGs), 851 and 1477, respectively, with a fold change > 1.3 (Table 2), while the number of genes dysregulated of in vitro submerged conditions with a fold change > 1.3 was ten times higher with 9836 DEGs (Table 2, Figure 3A). With a fold change > 3, we found 1721 genes differentially expressed genes in the in vitro submerged cells exposed to NM-105, but only 69 DEGs in in vivo. Interestingly, with a fold change of 3, a higher proportion of downregulated genes were found with in vitro submerged condition compared to in vivo exposure which showed more upregulated genes than downregulated (Table 2). Finally, it is important to note that the transcriptomic study brings out only one DEG with a fold change > 3 in the case of ALI exposure, namely Myc ( Figure 3C). Indeed, ALI exposure showed less DEGs genes with only 202 DEG with a fold change > 1.5 vs. 780 DEGs for in vivo exposure and 7895 for in vitro submerged exposure (Table 2). As represented by the volcano plot of dysregulated genes (Figure 3), ALI rat cells and lung exposed to TiO2 NP by inhalation showed similar number of differentially expressed genes (DEGs), 851 and 1477, respectively, with a fold change > 1.3 (Table 2), while the number of genes dysregulated of in vitro submerged conditions with a fold change > 1.3 was ten times higher with 9836 DEGs (Table 2, Figure 3A). With a fold change > 3, we found 1721 genes differentially expressed genes in the in vitro submerged cells exposed to NM-105, but only 69 DEGs in in vivo. Interestingly, with a fold change of 3, a higher proportion of downregulated genes were found with in vitro submerged condition compared to in vivo exposure which showed more upregulated genes than downregulated (Table 2). Finally, it is important to note that the transcriptomic study brings out only one DEG with a fold change > 3 in the case of ALI exposure, namely Myc ( Figure 3C). Indeed, ALI exposure showed less DEGs genes with only 202 DEG with a fold change > 1.5 vs. 780 DEGs for in vivo exposure and 7895 for in vitro submerged exposure ( Table 2).     Figure S1) allows to identify five KEGG pathways for in vitro submerged cells exposed to TiO2 NP, including 'IL-17 signaling pathway', 'Chemokine signaling pathway' and 'Cytokine-cytokine receptor interaction' : rno04657, rno0462, rno05323, rno04060, rno05144 ( Figure  S1.A), emphasize the role of macrophages inflammation. No common pathway was identified in the analysis of in vivo and ALI DEGs. Interestingly, GO biologic process (BP) annotation showed 5 common biologic process between in vivo and ALI : 0080090, 0060255, 051172, 0051171, 0050896 ( Figure S1.B and S1.C); all are involved in metabolic and cellular processes, while only one is common between in vivo and in vitro submerged condition, namely 0048545 ( Figure S1.A and S1.B). No common BP for ALI and in vitro submerged conditions or for the three conditions were evidenced. More connections between the DEGs were found in vitro (22) and in vivo (16) than in ALI conditions (6), and for all the groups, the software indicated that the genes are at least partially biologically connected as a group ( Figure S1).

Common Dysregulated Genes between in Vivo, in Vitro, and ALI Exposures to TiO2 NP
Regarding the differences between the fold changes of in vitro submerged experiments, the results comparison was performed using a cutoff of 2.8 for in vitro submerged and 1.3 for both in vivo and in vitro ALI conditions with a p-value fixed to p < 0.05 for the three conditions. These cutoffs focus on the most deregulated genes in the three conditions and allow a better selective analysis for the comparison of the three exposures.  Figure S1) allows to identify five KEGG pathways for in vitro submerged cells exposed to TiO 2 NP, including 'IL-17 signaling pathway', 'Chemokine signaling pathway' and 'Cytokine-cytokine receptor interaction': rno04657, rno0462, rno05323, rno04060, rno05144 ( Figure S1A), emphasize the role of macrophages inflammation. No common pathway was identified in the analysis of in vivo and ALI DEGs. Interestingly, GO biologic process (BP) annotation showed 5 common biologic process between in vivo and ALI: 0080090, 0060255, 051172, 0051171, 0050896 (Figure S1B,C); all are involved in metabolic and cellular processes, while only one is common between in vivo and in vitro submerged condition, namely 0048545 ( Figure S1A,B). No common BP for ALI and in vitro submerged conditions or for the three conditions were evidenced. More connections between the DEGs were found in vitro (22) and in vivo (16) than in ALI conditions (6), and for all the groups, the software indicated that the genes are at least partially biologically connected as a group ( Figure S1).

Common Dysregulated Genes between In Vivo, In Vitro, and ALI Exposures to TiO 2 NP
Regarding the differences between the fold changes of in vitro submerged experiments, the results comparison was performed using a cutoff of 2.8 for in vitro submerged and 1.3 for both in vivo and in vitro ALI conditions with a p-value fixed to p < 0.05 for the three conditions. These cutoffs focus on the most deregulated genes in the three conditions and allow a better selective analysis for the comparison of the three exposures.
The Venn diagram of differentially dysregulated genes comparing TiO 2 NP exposition in the three conditions showed 18 common DEGs. 107 common DEGs were found between in vivo (lung) and in vitro submerged, 53 between in vivo (lung) and ALI exposures and 147 between in vitro submerged and ALI exposures ( Figure 4). The Venn diagram of differentially dysregulated genes comparing TiO2 NP exposition in the three conditions showed 18 common DEGs. 107 common DEGs were found between in vivo (lung) and in vitro submerged, 53 between in vivo (lung) and ALI exposures and 147 between in vitro submerged and ALI exposures ( Figure 4).

Comparison of Functional Annotations of Dysregulated Genes in Vivo, in Vitro, and ALI after Exposition to TiO2 NP
GSEA analysis showed two common gene sets dysregulated in the three exposure, related to inflammation and oncogenesis: the IL6-JAK-STAT3 signaling and genes regulated by the transcription factor MYC (MYC_TARGETS_V2) ( Table 3). Four common gene sets were found between in vivo and ALI exposure, among them genes implied in 'cell cycle regulation' (E2F_TARGETS, G2M_CHECKPOINT) and 'cell transformation' (EPITHELIAL_ MESENCHYMAL_TRANSITION). Interestingly, another group of genes regulated by the transcription factor MYC was found (MYC_TARGETS_V1), which can be clustered in a MYC_TARGETS group V1 and V2. Two common gene sets were found between in vivo and in vitro submerged exposures (UV_RESPONSE_DN and TNFA_SIGNALING_VIA_NFKB) and no common gene set was found between in vitro and in ALI conditions. Seventeen gene sets were found only in lung, which is not surprising because herein the cell types analyzed are heterogenous, notably composed of epithelial, immune and endothelial cells, whereas cultured cells are characterized by their homogeneity. These gene sets were related to inflammation, immune response and homeostasis. ALI exposure resulted in UV_RESPONSE_UP gene set which can be brought closer to those common between in vivo and in vitro submerged exposures, in a UV-RESPONSE dysregulation. Finally, three gene sets were found specifically for in vitro submerged samples (MITOTIC_SPINDLE, DNA_REPAIR and PROTEIN_SECRETION).

Comparison of Functional Annotations of Dysregulated Genes In Vivo, In Vitro, and ALI after Exposition to TiO 2 NP
GSEA analysis showed two common gene sets dysregulated in the three exposure, related to inflammation and oncogenesis: the IL6-JAK-STAT3 signaling and genes regulated by the transcription factor MYC (MYC_TARGETS_V2) ( Table 3). Four common gene sets were found between in vivo and ALI exposure, among them genes implied in 'cell cycle regulation' (E2F_TARGETS, G2M_CHECKPOINT) and 'cell transformation' (EPITHELIAL_ MESENCHYMAL_TRANSITION). Interestingly, another group of genes regulated by the transcription factor MYC was found (MYC_TARGETS_V1), which can be clustered in a MYC_TARGETS group V1 and V2. Two common gene sets were found between in vivo and in vitro submerged exposures (UV_RESPONSE_DN and TNFA_SIGNALING_VIA_NFKB) and no common gene set was found between in vitro and in ALI conditions. Seventeen gene sets were found only in lung, which is not surprising because herein the cell types analyzed are heterogenous, notably composed of epithelial, immune and endothelial cells, whereas cultured cells are characterized by their homogeneity. These gene sets were related to inflammation, immune response and homeostasis. ALI exposure resulted in UV_RESPONSE_UP gene set which can be brought closer to those common between in vivo and in vitro submerged exposures, in a UV-RESPONSE dysregulation. Finally, three gene sets were found specifically for in vitro submerged samples (MITOTIC_SPINDLE, DNA_REPAIR and PROTEIN_SECRETION). Table 3. Gene set enrichment analysis (GSEA) functional annotation of differentially expressed genes (DEG) in the three conditions.

p-Value in vitro Submerged
Common gene sets between the three exposures  Table 3. Cont.
GO-BP analysis of the 18 common DEGs between the three exposure methods highlight some groups functions: inflammation and immune response, cell migration, intracellular movements (Table 5A). Coherently, a KEGG analysis finds one pathway associated with inflammation: cytokine-cytokine receptor interaction pathway (RNO-04060) and Reactome analysis revealed five pathways implied in mitosis and transport: mitotic prometaphase (R-RNO-68877), separation of sister chromatids (RNO-2500257, RNO-2467813), kinetochore and actin functions (RNO-141444 and RNO-5663220) ( Table 5B,C). When projected to human model, the same dysregulated pathways were found (Supplementary Figure S2, Table S2).

Methodology
There are many studies about TiO 2 nanoparticles (NP) toxicity. Nevertheless, there are few publications comparing in vitro and in vivo exposures and even less comparing air-liquid interface exposure with other in vitro and in vivo exposures. Moreover, the Vitrocell Cloud ® system is an innovative system using cloud exposure by a nebulizer, which mimics the lung interface. In this original study we compare the classic in vitro method with Vitrocell Cloud ® and in vivo exposures to TiO 2 nanoparticles in a complete transcriptomic study. In the current state of our knowledge, it is the first publication comparing these three expositions methods. Identification of common markers of exposure or effect is relevant for the development of smart and quick tests of nanotoxicity. In addition, in order to respect the three "R" of the ethical approach outlined by Russel and Burch in 1959 to reduce, replace and refine the use of animal testing, it seems relevant to develop in vitro models [37]. It is obvious that the in vitro study of different cell lines represents a promising tool for the implementation of predictive devices for NP exposure [38]. Thus, the aim of this study was to identify and validate early specific markers of lung exposure to TiO 2 NP, by comparing in vitro, air-liquid interface (ALI) and in vivo exposures.
Knowing that nanoparticles can reach the alveoli [39], NR8383 cells are an appropriate model because they are alveolar macrophage precursors, which are the first implicated cells in the alveolar clearance of nanoparticles [40][41][42].
This cell line has already been studied and validated as a model in the field of nanotoxicology [20,43]. Indeed, NR8383 cells are relevant for their immune functions [21]. NR8383 cells were exposed under classic submerged conditions or through an ALI cloud device (Vitrocell Cloud System ® ), to 3 cm 2 /cm 2 (19 µg/mL of TiO 2 NP) during 4 h.
To limit the bias due to agglomeration of nanoparticles, TiO 2 NP were sonicated and vortexed before each treatment. However, typical submerged exposition methods do not take into account the cellular interactions, the role of alveolar surfactant, the pulmonary clearance and the differential deposition in the respiratory tract regarding the displacement of NPs in the air during the respiration [39,44,45]. Moreover, the surfactant plays an important role in the uptake of nanoparticles (Geiser, 2010). Therefore, we chose to compare classic in vitro method with ALI exposure using the same cell line, exposed with a nebulizer first to the surfactant and then to TiO 2 NP, at the same dose.
The in vivo study was carried out on Fischer 344 rats exposed by inhalation (nose only) of 10 mg/m 3 , 6 h/day, 5 days/week for 4 weeks to TiO 2 NP NM-105. Immediately after the last exposure of the rats, the genes expression was analyzed by transcriptomics and compared to in vitro results on NR8383 rat macrophages submerges and in air-liquid interface exposure.

Viability of NR8383 Cells Exposed to TiO 2
NR8383 cells were exposed to TiO 2 NP from 0.25 to 200 µg/mL for 24 h. The viability tests showed a little, although not significant, decrease of mitochondrial activity activity up to 10 % decrease for 100 and 200 µg/mL of NM105 in WST-1 test and for cell membrane integrity up to 30 % decrease for 100 and 200 µg/mL of NM105 in LDH test (Figure 2). The real cytotoxic potential of TiO 2 NP remains uncertain and depends on their size, agglomeration and crystalline composition: anatase seems to be less toxic than rutile form [46,47]. In addition, a review reports no reduction in viability for BEAS-2B cells exposed to NM-105 at 150 µg/mL [48]. Another study showed no loss of cell membrane integrity in a 3D human bronchial model exposed to TiO 2 NP [49]. At the opposite, Park et al. found a significant loss of 30 % viability, by MTT assay, in BEAS-2B cells exposed to 40 µg/mL of TiO 2 NP while Wiemann et al. found a significant loss of cell membrane integrity, by LDH assay, in NR8383 exposed to 90 and 180 µg/mL of NM-105 TiO 2 NP: 53.6 % and 69.2 %, respectively [50]. A high toxicity of TiO 2 NP was also found in RLE-6TN rat alveolar epithelial cells: IC50 was found to be 7 µg/mL following 24 h exposure [51].

Number of DEGs in Each Exposure Method
The number of differentially expressed genes was higher in in vitro submerged model than the others regarding both the number of genes and fold changes (1721 DEGs with a fold change > 3, Table 2), whereas ALI exposure resulted in less differences between exposed and control cells (1 DEG with a fold change > 3, Table 2). Given a fold change > 1.3, the number of genes dysregulated for in vitro submerged cells is ten times bigger than the number of genes dysregulated in vivo and in ALI conditions: 9836, 851 and 1477, respectively ( Table 2). These differences are probably due to the exposition mode and to the cell type model. Indeed, in vitro submerged exposure is the common method used in toxicology studies, but it showed some limits: the sedimentation of nanoparticles is heterogenous because of the aqueous behavior, agglomeration, interaction with the solvent and proteins, density, dispersion, dissolution and sedimentation [52][53][54]. Consequently, the cells can be heterogeneously exposed with some cells exposed in excess and some not exposed at all [53].
High numbers of DEG genes between for in vitro submerged cells was already observed in previous transcriptomics studies [55][56][57]. This is why we chose to focus on DEGs with a FC > 2.8 for in vitro submerged exposition.
ALI NR8383 cell culture with surfactant have showed less DEG genes than classic in vitro exposed cells without surfactant, this may be explained by the short exposure time and the surfactant layer above cells. Nevertheless, some studies showed that surfactant allows a better uptake of TiO 2 nanoparticles by alveolar macrophages. Pulmonary surfactant is composed of phospholipids and lipids (80% and 10%, respectively) and 10% proteins, among them, collectin proteins (SP-A, B, C, D) which rapidly opsonize the particles and modulate the immune response by increasing the macrophages phagocytosis [58,59]. This may suggest that the ALI exposition method with surfactant before TiO 2 exposure should have improved the uptake and the cell response and consequently the number of DEG compared to classic in vitro exposure without surfactant. But this is not the case in our study, and this implies that other mechanisms could play a role. Opsonization by surfactant proteins is a rapid process that occurs between 5 and 15 min [60], so the time course of opsonization does not seem to be involved in our results.
However, dispersion and settling of NP depend on the pH of surfactant, its composition and its concentration [61,62]. TiO 2 dispersion stability and agglomeration depend also on the electronic properties of the surfactant [63,64]. The formation of micelles could lead to the engulfment of NP [65][66][67] and TiO 2 NPs are more electronegative and agglomerate less in electrolytic surfactants [68,69]. Indeed, TiO 2 NP dispersion was shown to remain stable in surfactant, but NPs settles easily at the bottom in media without surfactant [61], probably reducing the contact of NPs with cells. As a confirmation, this hypothesis was already observed with other NPs, for example, cells treated four hours with SiO 2 NP in the presence of pulmonary surfactant have showed less internalization of NPs, and microscopic observations of the cells showed NPs with less agglomeration and floating above the cell culture [70].
To elucidate the influence of surfactant in macrophages response to TiO 2 NPs, it will be crucial to further study the composition of rat surfactant, the turbidity of this surfactant with TiO 2 , the deposition time of TiO 2 , and electronic microscopy imaging of NR8383 cells with surfactant exposed to NP TiO 2 NM-105 at different time points.
TiO 2 NP was a mixed phase nanocrystalline powder composed of approximately 80% anatase and 20% rutile, with an average primary particle diameter of 21.5 ± 7.2 nm (Table 1). TiO 2 aerosol was generated at a target concentration of 10 mg/m 3 . Taking into account the interspecies differences in terms of respiratory parameters and lung deposited dose, this concentration and exposure time relate to the 8 h weighted average occupational exposure of a worker at 0.3 mg/m 3 during most of its career, the NIOSH recommended exposure limit for ultrafine TiO 2 [71,72].
In a concomitant study realized with F344 rats, at the end of the inhalation period, the concentration of titanium in lung tissue was approximately 2 mg/lung in exposed rats [73]. In the lung, after inhalation, the particles are preferentially deposited in the proximal alveolar region (PAR). Then, the surface area of particles per unit area of PAR is approximately 5 cm 2 /cm 2 [73] while the threshold dose for the onset of inflammation is estimated at approximately 1 cm 2 /cm 2 [74]. The presence of a strong pulmonary inflammation at the end of the exposure, agrees with these data. Finally, the presence of macrophages loaded with TiO 2 NP in the alveoli and the lymph nodes associated with the lungs [22], suggests a conservation of the mechanisms of particle elimination, by the mucociliary escalator or by translocation from the alveolar region to the lymph nodes. These results demonstrate the key role of alveolar macrophages in the pulmonary response to nanoparticle inhalation and justify the comparison of in vivo response with that of cultured rat alveolar macrophages.

Transcriptomic Study
The analysis of the 20 most up-DEG and 20 most down-DEG in the three conditions highlights five KEGG pathways dysregulated for in vitro submerged cells exposed to TiO 2 NP including 'IL-17 signaling pathway', 'chemokine signaling pathway' and 'cytokine-cytokine receptor interaction'.
Interestingly, 'Il-17 signaling pathway' is commonly associated with the activity of T lymphocytes, but macrophages were already demonstrated to be involved in Il-17 mediated inflammation as suggested by the (i) upregulation of IL17 receptors in vitro, (ii) their production of unique profiles of cytokines and chemokines [75] and (iii) the activation of macrophages in a M1/M2 heterogeneous phenotype [76]. Five common biologic process (GO-BP) were found between in vivo and ALI exposures, only one between in vivo and in vitro exposures and none between ALI and in vitro exposures. The 5 common GO-BP are involved in metabolic and cellular processes, suggesting that cellular responses in ALI conditions are similar to those obtained for in vivo condition.
GSEA analysis highlighted two common gene sets dysregulated in all three exposure, related to inflammation and oncogenesis: the IL6-JAK-STAT3 signaling and genes regulated by the transcription factor MYC (MYC_TARGETS_V2) ( Table 3). IL6-JAK-STAT3 pathway is known to be implied in tumorigenesis [77] and play a role in cancer-associated inflammatory environment [78] and angiogenesis [79]. Genes regulated by the transcription factor Myc are mainly involved in cell growth, apoptosis and metabolism. The Myc gene is well known as a proto-oncogene that is over expressed in various types of cancers include lymphomas, lung carcinoma, breast carcinoma and colon carcinomas [80]. Myc is also known to be involved in the polarization of M2 tumor-associated macrophages and is targeted for new therapeutic strategies against cancer [81][82][83][84].
Interestingly, as opposed to what could be expected, Myc was the most downregulated gene in ALI exposures (Table 2). However, the role of this transcription factor is large and remain poorly understood. Some studies have shown that downregulation of Myc gene can induce cell cycle arrest and, in some case cell survival, differentiation or apoptosis [85][86][87]. As no cell death was observed after NM105 TiO 2 NP exposure in our study, we can suppose that the down regulation of Myc in NR8383 exposed in ALI experiments, may be associated with a cell cycle arrest and survival of NR8383. Cell survival and cell cycle states in these conditions will be further investigated.
To propose biomarkers of effect for NM-105 TiO 2 , we focused our analyses on the common dysregulated genes (DEGs) between the three exposure modes. 18 common DEGs were found, among them, only 4 are upregulated in the three conditions: Ccl4, Ccl7, Osm and Bcl3, while the other common DEG display a non-homogenous dysregulation. Indeed, among the 18 DEGs, 14 DEGs showed the same dysregulation profile following in vitro and ALI conditions: among them 12 DEGs were down regulated and 2 upregulated as noted in Table 4. Contrary those genes were downregulated in vivo. Only Ptpn13 was downregulated in vivo and in ALI conditions while upregulated in vitro (see Table 4). These discordant results can be explained by the experiments conditions. For in vivo study, transcriptomic analyses consider all the cellular perturbations induced by NP exposure as well as the interaction between the different cell types including macrophages, epithelial, interstitial and endothelial cells, after a subacute exposure (6 h/day, 5 days/week for 4 weeks) while only NR8383 cells are grown for a short 4 h exposure in both vitro tests. Moreover, surfactant was present in ALI and in vivo conditions, but not for in vitro submerged conditions. Consistently with our previous observations, common dysregulations for in vitro and ALI versus in vivo exposures revealed a negative regulation of mitosis and cell proliferation and a negative regulation of M2 polarization of macrophages (Table 4). These observations are in accordance with the hypothesis of a cell cycle arrest without any cytotoxicity of macrophages in vitro and in ALI conditions. It can be an early effect of TiO 2 NP on macrophages, while the upregulation of these genes for in vivo conditions can be due to other cell types in lung or to an adaptative regulation towards proliferation after a longer in vivo exposure.
Note that macrophages are ubiquitous innate immune cells and can be polarized in two distinct functional phenotypes, either in pro-inflammatory macrophages called M1 or in M2 macrophages that promote immune suppression and wound healing [88][89][90]. They also suggest a preferential differentiation in M2 macrophages in vivo compared to M1 that were found in vitro and in ALI conditions. It is well known that in cancer pathologies, M2 macrophages can be recruited and favor tumor growth by contributing to the tumor microenvironment; by the way, they are called tumor-associated macrophages (TAM). Furthermore, Ptpn13 is a tumor suppressor gene that is frequently inactivated in non-small cell lung cancer (NSLC) (in more than 70% of NSLC) [91], and was downregulated in our ALI and in vivo exposure to TiO 2 while upregulated in in vitro condition.
The four genes upregulated in the three exposure conditions are actor of the cytokine-cytokine receptor interaction KEGG pathway: Ccl4, CcL7, Osm and Bcl3 is implied in cell proliferation regulation (Table 4). Interestingly, Ccl4 was already identified as a candidate for biomarker for BCR pathway activation and prognostic in diffuse large B cell lymphoma [92] and as an inflammation-related diseases biomarker (Olink proteomics). Ccl7 (also called MCP3) plays a crucial role in cancer as it can promote tumor growth, tumor microenvironment, invasion and metastasis [93]. Indeed, overexpression of Ccl7 was associated with lung adenomas [94] and metastasis in colorectal cancer and renal carcinoma [95][96][97]. Moreover, Ccl7 expression is associated with the recruitment of tumor-associated macrophages (TAMs) [98]. Likewise, oncostatin M (Osm) is identified as an inflammation biomarker in periodontal diseases [99], sepsis [100], inflammatory bowel disease [101,102]. Other studies showed that Osm promotes cancer cell plasticity [103] and the polarization of pro-fibrotic M2-like macrophages [104]. Finally, Bcl3 (B-cell lymphoma 3 gene) is a proto-oncogene whose product is an inhibitor of NF-kappa B [105,106] and display anti-apoptotic functions. Bcl3 is involved in some cancers, the most known is his role in lymphoma, in which he regulates pro-survival and pro-inflammatory gene expression [107]. Bcl3 is also involved in solid tumor progression [108] like ovarian [109], cervical cancers [110,111], mammary cancer metastasis [112] and tumor progression in breast cancer [113]. In macrophages, Bcl3 was showed to be a key mediator of IL-10 macrophage polarization to an immunosuppressive phenotype as observed in TAM [88].

Cell Culture
NR8383 cells were obtained from the ATCC (American Type Culture Collection, Manassas, VA, USA). Cells were cultured in DMEM medium completed by 15% of fetal bovine serum (FBS), 100 U/mL of penicillin, 100 µg/mL of streptomycin, 0.25 µg/mL of amphotericin and 2 mM of L-glutamine at a 37 • C temperature and under a 5 % CO 2 atmosphere. For all tests, the density used was of 5 × 10 4 cells/mL.

In Vitro Cytotoxicity Study
Cell viability was assessed with the WST-1 assay and the measure of the lactate dehydrogenase (LDH) release in the extracellular medium. Non-exposed cells were considered as a control. Six technical and four biologic replicates were used per each condition. After a 24-h exposure to TiO 2 NP, cells were incubated with 5% of WST-1, during 2 h at 37 • C. Then, absorbance was measured at 450 nm, with a 690 nm as reference wavelength (iMarkTM, Bio-Rad). LDH dosage was carried out following manufacturer's recommendations. Briefly, after 24 h of exposure to TiO 2 NP, cells were incubated 30 min with 100 µL of mix of LDH buffer and substrate, at 20 • C. Then, 50 µL of stop solution were added and the absorbance was measured at 490 nm (iMarkTM, Bio-Rad). The positive control was the result of cells exposed to the lysis buffer (Triton 5%) during 15 min before the measure. Data are presented as means ± standard deviation (SD) of the four biologic replicates. Statistical differences were determined by a one-way analysis of variance (one-way ANOVA) followed by the post hoc Dunnett's test.

In Vivo Exposure
The animal experiments were performed according to European (Directive 2010/63/EC) and French (Décret n • 2013-118) legislations regarding the protection of animals used for scientific purposes. The INRS animal facility has full accreditation (authorization n • D54-547-10) from the French Ministry of Agriculture. This study was approved on October 14th, 2013 by the regional ethical committee (CELMEA n • 066) appointed by the Ministry of higher education and research (Authorization n • 00692.01).
The in vivo exposure was made at the National Institute for Research and Safety (INRS) and described previously [22]. Briefly, thirteen-week-old male Fisher F344 rats (Charles River Laboratories, France) were housed in standard environmental conditions (relative humidity: 55 ± 10 %; temperature: 22 ± 2 • C and 12/ 12 h light/dark cycle). Water and a standard laboratory animal diet (A04, Safe diet) were freely available. Two weeks before exposure, rats were gradually acclimatized to the restraining tubes. Animals were then exposed by nose-only inhalation either to filtered air (controls) or TiO 2 aerosol at a concentration of 10 mg/m 3 for 6 h/day, 5 days/week for 4 weeks. Immediately after the exposure period, the animals were anesthetized with pentobarbital (60 mg/kg) and exsanguinated through the abdominal aorta and lung tissues were collected. Gene expression was analyzed by transcriptomic analysis on the entire pulmonary accessory lobe. We reanalyzed the transcriptomic results obtained in vivo (GEO Accession: GSE99997) [22] in order to compare them with the results obtained in vitro and in ALI conditions.

Exposure at the Air-liquid Interface
Cells in air-liquid interface were exposed to aerosolized NM-105 nanoparticles using the Vitrocell cloud System ® . ALI exposure methodology used here is described in detail in another study [116]. Briefly, NR8383 were seeded in Transwell ® culture inserts (Polyester membrane, TC-treated, Corning, USA) with 300.000 cells/insert during 24 h before air-liquid interface exposure with the nebulization of pig surfactant and then NM-105 NP. After exposure to 3 cm 2 of TiO 2 NP/cm 2 of cell culture, cells were incubated 4 h in a 37 • C, under a 5% CO 2 atmosphere, and RNA extractions were performed.

Transcriptomic Study
Total RNA Extraction. In order to evaluate gene expression profile, total RNA was extracted from NM-105 exposed cells during 4 h at 3 cm 2 of TiO 2 NP/cm 2 of cell culture (same dose for ALI and submerged experiments), using RNA-Solv (R6830-02, USA). The unexposed cells were used as a control. The RNA concentration was determined by measuring the absorption at 260 nm using a spectrophotometer (Biotech-BioSpec-Nano, Shimadzu). The optimal purity of the RNA was ensured by the determination of 260/280 nm of an absorbance ratio A260/A280 > 1.8. RNA integrity was confirmed with the Agilent 2100 and RNA 6000 Nano LabChip bioanalyzer kits (Agilent Biotechnologies, Palo Alto, CA, USA). The threshold of 8 for RNA integrity number (RIN) was chosen as a cutoff to determine whether the extracted RNA was qualified or not.
Expression microarray hybridization. One hundred nanograms of RNA from each sample was labeled with cyanine 3-CTP using the low input quick amp labeling kit (Agilent Technologies). The labeled cRNAs were purified and hybridized on the Agilent G4853A Sure Print G3 Rat GE 8 * 60 K microarray chips (Agilent Technologies) that cover the entire rat transcriptome. The slides were then washed and scanned using the Agilent G2505C microarray scanner with a resolution of 3 µm. The data were extracted using Feature Extraction software version 11.0 (Agilent).
Bioinformatics analyses. First, the data were standardized using the GeneSpring software. The Student's test followed by the Benjamini-Hochberg correction and the filtering criteria were applied, to identify the genes whose level of expression was considerably modified. Genes with expression changes at a fold change (FC) > |1.5| compared to control with p < 0.05 were considered differentially expressed at a significant level. Genes were clustered into groups according to different criteria, such as the terms gene ontology (GO) biologic process and pathways (KEGG and Reactome) (see Table 3). Gene Set Enrichment Analysis (GSEA) functional annotation was done from MSigDB Collections. For biologic interpretation, we considered the GO biologic process and pathways with a p ≤ 0.05 value in the cluster with an enrichment score (Z-score) greater than 1.3 [117]. Venn diagram were realized thanks to GeneVenn online software [118]. The heatmap was realized with ClustVis software online [119].

Conclusions
This work suggests that Ccl4, Osm, Ccl7 and Bcl3, are early response biomarkers to TiO 2 NP. Interestingly, Ccl7 and Bcl3 genes were predicted as biomarkers of TiO 2 effects using a machine learning approach and mice in vivo datasets by our colleagues within the SmartNanotox project (preliminary results, Vadim Zhernovkov). This analysis permits us to identify five common GO-BP between in vivo and ALI exposures and only one between submerged and in vivo, suggesting that ALI better reflects the effect of TiO 2 NP exposure in vivo. It could be interesting, in a future work, to assess other nanoobjects, such as metal oxides or carbonaceous, to verify if these genes are modified too, suggesting a "nano" effect or if not, validating a specific titanium oxide response.
Thus, this work justifies the pertinence of our in vitro models, which, regardless of ethical considerations, are cost and time effective, as in vivo experiments could produce valuable results in 8 months and in vitro within two weeks.