Continuous in vitro exposure of intestinal epithelial cells to E171 food additive causes oxidative stress, inducing oxidation of DNA bases but no endoplasmic reticulum stress.

The whitening and opacifying properties of titanium dioxide (TiO2) are commonly exploited when it is used as a food additive (E171). However, the safety of this additive can be questioned as TiO2 nanoparticles (TiO2-NPs) have been classed at potentially toxic. This study aimed to shed some light on the mechanisms behind the potential toxicity of E171 on epithelial intestinal cells, using two in vitro models: (i) a monoculture of differentiated Caco-2 cells and (ii) a coculture of Caco-2 with HT29-MTX mucus-secreting cells. Cells were exposed to E171 and two different types of TiO2-NPs, either acutely (6-48 h) or repeatedly (three times a week for 3 weeks). Our results confirm that E171 damaged these cells, and that the main mechanism of toxicity was oxidation effects. Responses of the two models to E171 were similar, with a moderate, but significant, accumulation of reactive oxygen species, and concomitant downregulation of the expression of the antioxidant enzymes catalase, superoxide dismutase and glutathione reductase. Oxidative damage to DNA was detected in exposed cells, proving that E171 effectively induces oxidative stress; however, no endoplasmic reticulum stress was detected. E171 effects were less intense after acute exposure compared to repeated exposure, which correlated with higher Ti accumulation. The effects were also more intense in cells exposed to E171 than in cells exposed to TiO2-NPs. Taken together, these data show that E171 induces only moderate toxicity in epithelial intestinal cells, via oxidation.


Introduction
Titanium dioxide (TiO 2 ) has been used as a white pigment and opacifying agent for many years. Its many applications include as an additive in paints, plastics, paper and food. As a food additive, the European nomenclature refers to TiO 2 as E171, and its use has been authorized since 1969 thanks to evidence indicating low intestinal absorption, solubility and toxicity (Amenta et al., 2015;Jovanovic, 2015). The characteristics of E171 were reevaluated in 2016 by the European Food Safety Authority (EFSA AP), which concluded on extremely low absorption of orally-administered TiO 2 , regardless of particle size (EFSA AP, 2016). Based on the existing literature at that time, no acceptable daily intake value could be defined.
According to the European recommendations defining nanomaterials (2011/696/EU), E171 is not a nanomaterial. However, it can contain up to 43% of TiO 2 nanoparticles (TiO 2 -NPs) Peters et al., 2014;Weir et al., 2012;Yang et al., 2014), and these nanoparticles may have toxic effects (Grande & Tucci, 2016;McCracken et al., 2016;Shi et al., 2013). E171 is used in a wide range of food products, in particular confectionery and chewing gums (Peters et al., 2014;Weir et al., 2012), with between 0.02 and 9.0 mg/g TiO 2 found in some food and personal care products, of which 5-10% corresponds to TiO 2 -NPs (Peters et al., 2014). Based on these data, human daily exposure through ingestion was initially estimated to range between 0.035 mg/kg of body weight (b.w.)/day (Frohlich & Roblegg, 2012) and 5 mg/person (Powell et al., 2010), with differences depending on the age of the individual: 1-2 mg TiO 2 /kg b.w. for US children under 10 years of age, and 0.2-0.7 mg TiO 2 /kg b.w. for other US consumers (Weir et al., 2012). More recently, these estimations were refined, and exposure was reported to range between 0.2 and 0.4 mg/kg b.w. in infants and the elderly, and 5.5-10.4 mg/kg b.w. in children, depending on the exposure scenario (EFSA AP, 2016). Within the Dutch population, daily exposure has been estimated at 4.2 mg/kg b.w. for children between 2 and 6 years of age, 1.6 mg/kg b.w. for 7-69 year-old consumers, and 0.74 mg/kg b.w. for older consumers (Rompelberg et al., 2016). Based on these refined estimations of intake, an extensive review of the existing literature assessing the risks associated with ingested TiO 2 -NPs and toxicokinetic data concluded that a risk for the liver and reproductive organs could not be excluded .
While the impact of inhaled TiO 2 has been extensively studied and reviewed (Johnston et al., 2009;Shi et al., 2013;Skocaj et al., 2011), the oral route has been less well-documented (McCracken et al., 2016;Skocaj et al., 2011). When ingested by human volunteers, E171 was found to be absorbed in the gastrointestinal tract at low levels (Pele et al., 2015). In rats, it impaired intestinal immune homeostasis (Bettini et al., 2017), and its consumption was linked to colorectal cancer (Bettini et al., 2017;Urrutia-Ortega et al., 2016). However, no acute or sub-chronic toxicity was observed in rats ingesting TiO 2 particles with the same size distribution and crystalline structure as E171 (Warheit et al., 2015), although toxic effects were reported with TiO 2 -NPs of different sizes and crystal phases. These effects were observed on the cardiovascular system Wang et al., 2007Wang et al., , 2013, the kidneys (Al-Rasheed et al., 2013;Wang et al., 2007), the liver (Wang et al., 2007) or reproductive organs (Jia et al., 2014).
The mechanisms behind these effects are not well known. Inhaled TiO 2 -NPs cause oxidative stress in lung cells which drives inflammation, genotoxicity, cytotoxicity and impairs the autophagic process (Johnston et al., 2009;Shi et al., 2013). Ingested TiO 2 -NPs may cause oxidative stress, and/or may be associated with damage to DNA and cytotoxicity depending on their crystal structure, their size and the differentiation state of cells (Brun et al., 2014;Dorier et al., 2015;Gerloff et al., 2012;Tay et al., 2014). TiO 2 extracted from chewing gum was reported to increase levels of reactive oxygen species (ROS) in cells without causing any cytotoxicity to intestinal or gastric cells . Pretreatment of E171 with simulated digestive fluids increased intracellular accumulation and inhibited growth in undifferentiated Caco-2 cells but not differentiated Caco-2 cells (Song et al., 2015).
The present study aimed to identify the mechanisms behind the intestinal toxicity of E171. The impact of E171 was assessed on two in vitro models of the ileum: (i) differentiated Caco-2 absorptive cells and (ii) a mucus-secreting model composed of Caco-2 cells (70%) co-cultured with HT29-MTX mucus cells (30%). HT29-MTX cells are adapted from the HT29 cell line, by growth in medium containing methotrexate (MTX) (Lesuffleur et al., 1990). At post-confluence, these cells are totally differentiated into mucus-secreting cells, with appearance of mucus droplets in the culture and expression of MUC1, MUC2, MUC3, MUC4 and MUC5C glycoprotein and mucins (Lesuffleur et al., 1993). Mucus secretion is then maintained even after reversion to MTX-free medium. Both cell models were either acutely exposed to E171 for 6 h or 48 h, or repeatedly exposed three times a week during their differentiation (21 days). These exposure regimes were designed to mimic acute exposurewhen the food bolus reaches the intestineor the effects of repetitive consumption of TiO 2 on the continuously renewing intestine. Cells were also exposed to two different TiO 2 -NPs, (i) mixed anatase/rutile TiO 2 -NP, measuring 24 nm in diameter (JRCNM01005a from the European Commission Joint Research Center (JRC)), which has been widely used in toxicity studies and (ii) A12, a 12-nm pure anatase TiO 2 -NP, which had a composition close to that of the E171 tested. Cell viability, cell redox status and DNA integrity were monitored, and endoplasmic reticulum (ER) stress was assessed as it can be triggered by oxidative stress and has been associated with inflammatory bowel diseases (Deuring et al., 2011). The biological responses observed correlated with intracellular accumulation of Ti.

Chemicals and reagents
Unless otherwise indicated, all chemicals were purchased from Sigma-Aldrich and were >99% pure. Cell culture medium and fetal bovine serum (FBS) were purchased from Thermo Fisher Scientific (Waltham, MA).

Particle dispersion and characterization
E171 food additives were purchased as powders from commercial websites. P25 TiO 2 -NP is referenced as JRCNM01005a (previously NM105) in the nanomaterials library at the European Joint Research Center (Ispra, Italy). A12 was synthesized in our laboratories (Pignon et al., 2008) and was used in our previous studies (Brun et al., 2014;Dorier et al., 2015;Jugan et al., 2012;Simon-Deckers et al., 2008). The specific surface area (SSA) of all TiO 2 forms was measured according to Brunauer, Emmett and Teller (BET), and their crystalline phase was determined by X-ray diffraction. Particle diameters were measured by transmission electron microscopy (TEM). Powders were suspended in ultrapure sterile water (10 mg/mL), then sonicated for dispersion using an indirect cup-type sonicator (Cup Horn; Bioblock Scientific, Vibracell 75041, Illkirch, France) in continuous sonication mode for 30 min, 4 C, at 52.8 W (80% amplitude), as determined by calorimetry (Taurozzi et al., 2011). The zeta potential and agglomeration state of suspensions were monitored using a Malvern NanoZS (Worcestershire, UK).
For acute exposure experiments, cells were grown to confluence and culture was continued for 21 days post-confluence to promote differentiation, changing culture medium three times a week. Differentiated cells were exposed for 6 h, 24 h or 48 h to E171 or TiO 2 -NPs (1-200 lg/mL) in complete cell culture medium. Cells were then harvested ( Figure S2). In the repeated exposure regimen, cells were seeded at day 0; the cell culture medium was replaced three times a week with fresh medium containing 1-100 lg/mL E171 or TiO 2 -NPs, from day 1 to day 21 post-seeding. Exposure medium containing particles was thus changed a total of eight times ( Figure S2). Cells were not subcultured during this period. As calculated previously (Brun et al., 2014), the exposure concentrations exceeded the estimated human daily exposure to TiO 2 approximately 10,000-fold. After repeated exposure, cell differentiation status was checked; tight junctions ( Figure S3A) and microvilli ( Figure S3B) showed typical morphologies and mucus secretion was equivalent in exposed cells vs. control cells ( Figure S3C-E).

Quantification of intracellular Ti accumulation
In exposed cells, Ti content was measured by inductively coupled plasma mass spectrometry (ICP-MS). After exposure, cells were rinsed four times with PBS to remove particles that may be weakly bound to cell membrane before harvesting in 125 lL of ultrapure water. Cells were lysed by microwave-assisted acid decomposition, as previously described (Dorier et al., 2015). Briefly, 6 mL of 24% (v/v) ultrapure grade H 2 SO 4 was added to cellular suspensions, which were then heated for 30 min at 1100 W, samples were then diluted in ultrapure grade HNO 3 (1% v/v) and analyzed on a Nexion 300X ICP-MS (Perkin Elmer, Waltham, MA) equipped with a concentric nebulizer and operated in standard mode. Calibration curves were obtained using a certified ionic Ti ICP-MS standard. Concentrations of 47 Ti, 48 Ti and 49 Ti were analyzed; but final results were only interpreted based on 47 Ti as S content (from H 2 SO 4 ) interfered with 48 Ti and 49 Ti readings.

Cell viability assessment
In the acute exposure regimen, cytotoxicity was assessed using the WST-1 assay (Roche, Basel, Switzerland). Cells were grown in 96-well plates, and exposed to 0-200 lg/mL of particles for 6 h or 48 h. Exposure medium was then replaced by 100 lL of WST-1 diluted to 1:10 (v:v) in cell culture medium, and plates were incubated for 90 min at 37 C. To avoid particle interference with the assay, plates were centrifuged for 5 min at 200Âg, to settle particles at the bottom of wells. Supernatant (50 lL) was then transferred to a clean plate and absorbance was measured at 438 nm using a SpectraMax M2 spectrophotometer (Molecular Devices, Sunnyvale, CA). Absorbance was normalized by subtracting absorbance readings for control (untreated) wells. Potential interference was checked for as previously described (Brun et al., 2014); no interference was detected. In the repeated exposure regimen, cell viability in the Caco-2 monoculture was assessed using the trypan blue exclusion assay. Viable and non-viable cells were counted using an automatic cell counter (Countess automatic cell counter, Invitrogen, Carlsbad, CA). Cytotoxicity was assessed for the Caco-2/HT29-MTX coculture model based on propidium iodide (PI) exclusion. Cells were incubated with PI before counting using a FacsCalibur analyzer (BD Biosciences, Franklin Lakes, NJ) equipped with CXP software (Beckman Coulter, Brea, CA). Mean fluorescence was analyzed using Flowing Software 2.5.1 (http://www.flowingsoftware.com/). In both exclusion assays, measurements were performed in triplicate for each condition; results are expressed as the mean percentage of viable cells after normalization relative to the untreated control.

Quantification of reactive oxygen species
Intracellular ROS content was quantified using 2 0 ,7 0 -dichlorodihydrofluorescein diacetate acetyl ester (H 2 -DCF-DA) (Invitrogen, Carlsbad, CA). Cells were grown in 12-well plates and exposed to particles, then washed twice with PBS, incubated for 30 min with 80 lM H 2 -DCF-DA at 37 C, and harvested by scraping. Fluorescence intensity was measured with excitation at 480 nm and emission at 530 nm, on a SpectraMax M2 (Molecular Devices, Sunnyvale, CA). The emission intensity was normalized for protein concentration, as determined using Bradford reagent. Potential interference of TiO 2 particles was tested by measuring DCF fluorescence, either in cells exposed to particles without H 2 -DCF-DA, or in particle suspensions (10, 50 and 100 lg/mL) to which H 2 -DCF-DA was added. No interference was detected (Table S1).

Genotoxicity
DNA strand breaks, alkali-labile sites and formamidopyrimidine DNA glycosylase (Fpg)-sensitive sites were assessed using the alkaline version of the comet assay and its Fpg-modified version (Tice et al., 2000). Cells were grown in 12-well plates, exposed to particles, then harvested and stored at -80 C in 85.5 g/L sucrose, 11.76 g/L sodium citrate, 50 mL/L DMSO, pH 7.6. Cells were then embedded in 0.5% low-melting-point agarose and spread on glass slides pre-coated with 1% agarose. After cell lysis by immersion in 2.5 M NaCl, 100 mM EDTA, 10 mM Tris, pH 10, 10% DMSO overnight at 4 C, slides were rinsed in 0.4 M Tris-HCl pH 7.4. Three slides were incubated with Fpg buffer only and another three slides were incubated with Fpg enzyme prepared in Fpg buffer (0.05 U/lL, Interchim, Montluc¸on, France), for 45 min at 37 C. Slides were cooled on ice, before immersion in cold migration buffer (NaOH 300 mM, EDTA 1 mM) for 30 min to allow DNA to unwind. Electrophoresis was performed at 25 V, 300 mA for 30 min. Slides were neutralized by washing three times in 0.4 M Tris-HCl, pH 7.4, then stained with Gel Red (VWR). %tail DNA was measured on 50 nuclei using Comet IV software (Perceptive Instruments, Bury St Edmunds, UK). As positive control for comet-Fpg, A549 cells were exposed to 1 lM riboflavin for 20 min at 37 C and then UVA-irradiated (010.0 J/cm 2 ). As positive control for comet alkaline assay, control (unexposed cells) slides were exposed to 50 mM H 2 O 2 . Experiments were performed in three independent replicates. Results were normalized for %tail DNA in control (unexposed) cells, and expressed as mean fold-change-± standard error calculated over the three independent experiments.

Gene expression
To assess ER stress, levels of classical ER stress markers were monitored based on mRNA expression. The markers analyzed were IRE-1, PERK and ATF6, which encode ER stress sensors that initiate the unfolded protein response (UPR) signaling cascade; and GRP78/ BiP, which encodes a chaperone that controls protein quality and activates signaling molecules in the ER. We also monitored mRNA levels for ATF6 and XBP1, which encode proteins mediating the UPR mechanism of action, and CHOP, which is involved in ERstress-induced apoptosis (Chaudhari et al., 2014;Zhang & Kaufman, 2008). Oxidative stress was assessed based on the mRNA expression of antioxidant enzymes, catalase (CAT), superoxide dismutases 1 and 2 (SOD1, SOD2) and glutathione reductase (GSR). Gene expression was assessed by reverse transcriptionquantitative PCR (RT-qPCR). mRNA was extracted using GenElute TM mammalian total RNA Miniprep kit (Sigma, St. Louis, MO) and stored at -80 C. Purity was checked using a spectrophotometer (NanoDrop) to measure absorbance at 260 nm (abs260), and determine abs260/abs280 and abs260/abs230 ratios. mRNA was reverse-transcribed using SuperScript III Reverse Transcriptase (Invitrogen, Carlsbad, CA). Total cDNA concentration and purity were determined. MESA Blue qPCR Mastermix for SYBR Assay (Eurogentec, Li ege, Belgium) with ROX reference was used to perform the quantitative PCR in a MX3005P thermocycler (Stratagene, San Diego, CA). The following thermal cycling steps were applied: 95 C for 5 min, then 95 C for 15 s, 55 C for 20 s and 72 C for 40 s for 40 cycles. For the dissociation curve: 95 C for 1 min, 55 C for 30 s and 95 C for 30 s. Primer sequences are indicated in Table  S2; their efficiency was determined to be between 1.8 and 2.2.
Relative expression values were calculated as 2 ÀDDCq , where DCq is the difference between the cycle threshold (Cq) value for target and reference genes and DDCq the difference between the DCq in exposed cells vs. control (unexposed) cells. CYCLOA, CYCLOB, S18 and GAPDH were used as reference genes. Variability of expression for reference genes was assessed using Bestkeeper (Pfaffl et al., 2004). Gene expression levels were determined and statistically analyzed using the Relative Expression Software Tool (REST2009, Qiagen, Hilden, Germany) (Pfaffl et al., 2002).

Statistical analysis
Unless indicated otherwise, statistical significance was assessed by applying two tests: the Kruskal-Wallis (non-parametric one-way analysis of variance on ranks) and the Mann-Whitney U-test, a paired comparison. Statistical analysis was performed using Statistica 8.0 (Statsoft, Chicago, IL). A p value <.05 was taken as the threshold for statistical significance.

Results
Physico-chemical characterization of the particles making up E171 samples and NPs Five batches of commercially-available E171 were purchased and characterized (Table S3). The characteristics of all batches were comparable; two batches were pure anatase, whereas the other three batches were mainly anatase with traces of rutile (<1%). The average primary diameter of the particles ranged between 100 and 136 nm, as measured on TEM images. The percentage of particles with diameter <100 nm, i.e., NPs, ranged from 30% to 55%, in the number size distribution. We selected E171A for the tests to assess E171 toxicity because it was sold in large quantities. This product was composed of round particles ( Figure 1(A)) with a broad size distribution centered on 119 ± 65 nm (Figure 1(B)). Particles with diameters <100 nm accounted for 47% of the product (Figure 1(B)), and X-ray diffraction analysis indicated that it was pure anatase (Figure 1(C,  D)). The SSA for this product was 9.4 m 2 /g. After high-energy sonication in water, a hydrodynamic diameter of 415.4 ± 69.5 nm was determined. This diameter increased to 739.3 ± 355.3 nm after dilution in cell culture medium due to particle agglomeration (Table 1). Polydispersity of the suspension was high, with polydispersity indexes (PdI) of 0.48 in water and 0.64 in exposure medium.
P25 and A12 NPs were previously characterized and used in our laboratory (Brun et al., 2014;Dorier et al., 2015;Jugan et al., 2012;Simon-Deckers et al., 2008); their main characteristics are summarized in Table 1. A12 was pure anatase with a diameter of 12 ± 3 nm. P25 was mixed anatase/rutile (86%/14%) TiO 2 -NP, with a mean diameter of 24 ± 6 nm. The hydrodynamic diameter of A12 was 85.0 ± 2.9 nm in water and increased to 447.9 ± 0.3 nm when diluted in cell culture medium. For P25, the hydrodynamic diameter was 157.6 ± 1.0 nm in water and 439.9 ± 6.7 nm in cell culture medium. PdI were low for both NPs, suggesting that the suspensions were homogeneously dispersed and stable.

Intracellular accumulation and/or membrane absorption of Ti
One of the aims of this study was to compare the response of two cell models to three types of particles, therefore, we first compared their intracellular accumulation. Ti content was measured by ICP-MS in cells after exposure to 50 lg/mL of A12, P25 or E171, either as acute doses for 6 h or 48 h, or repeatedly for 21 days (Table 2). ICP-MS gives an overall quantification of Ti in lysed cells, but it cannot discriminate between Ti accumulated inside the cell and Ti bound to the cell membrane. Whatever the exposure conditions, Ti accumulation in Caco-2 cells was slightly higher than in Caco-2/HT29-MTX cells, but the difference was not statistically significant. TEM images of Caco-2/HT29-MTX cells exposed for 21 days to E171 are provided in Figure S4, showing agglomerates of E171 accumulated in the cytoplasm, entrapped in a vesicle.
In contrast, there were differences depending on the type of particle. Thus, E171 accumulated more than P25, which accumulated more than A12 (E171 ) P25 > A12). Accumulation was also influenced by the exposure time, with systematically greater accumulation in repeatedly-exposed cells, compared to acutelyexposed cells. In the Caco-2 cell model, accumulation was also significantly higher after 48 h of exposure compared to 6 h of exposure.

Impact of exposure to TiO 2 particles on cellular viability
Since particles accumulated in cells, they could potentially affect cell viability. We therefore assessed cytotoxicity in cell cultures, following exposure. No effect on cell viability was observed in either cell model in the tested conditions (Figure 2).

Oxidative stress
Oxidative stress is reported to be the main mechanism of toxicity induced by TiO 2 -NPs (Johnston et al., 2009;Shi et al., 2013). For this reason, the redox balance was evaluated in both cell models after exposure to E171 or TiO 2 -NPs.
First, intracellular ROS content was quantified using the H 2 -DCF-DA assay. In Caco-2 cells exposed to an acute dose of E171 for 48 h, intracellular ROS content was significantly higher than in untreated cells, at all concentrations tested (Figure 3(A)). A similar increase was observed in Caco-2/HT29-MTX cells at all three timepoints (6 h, 24 h, 48 h, acute exposure), and ROS levels were found to increase in a dose-dependent manner (Figure 3(B)). Similarly, in both cell models, intracellular ROS levels were higher in repeatedly-exposed cells than in untreated cells (Figure 3(C,D)). Cells repeatedly exposed to E171 did not display higher ROS levels than acutely-exposed cells. TiO 2 -NPs also caused intracellular accumulation of ROS, although to a lesser extent than E171.
To confirm that exposure to E171 and TiO 2 -NPs induced oxidative stress, the expression of genes encoding antioxidant enzymes -CAT, GSR and SOD1, SOD2was measured by RT-qPCR. In general, modulation of gene expression was not significantly different in acutely-exposed Caco-2 and Caco-2/HT29-MTX cells, although in Caco-2/HT29-MTX exposed for 48 h to A12 a difference was noted (Table S4). In repeatedly-exposed cells, gene expression was unchanged in the Caco-2 monoculture (Figure 4(A)), whereas in repeatedly-exposed Caco-2/HT29-MTX, CAT and GSR were downregulated upon exposure to 10 lg/mL E171, and GSR and SOD1 were downregulated in response to 50 lg/mL E171 (Figure 4(B)). Exposure to 10 lg/mL A12 caused CAT and SOD1 downregulation; P25 only induced downregulation of SOD1 (Table S4). Finally, the mRNA expression level for NRF2a transcription factor controlling the expression of some of these antioxidant enzymeswas measured; no significant change was observed in any of the exposure conditions tested (Table S4).

Oxidative damage to DNA
Oxidative stress can cause oxidative damage to DNA, which can be quantified using the comet assay in its alkaline and Fpg-modified versions. These assays probe DNA strand breaks and alkalilabile sites (alkaline version of the assay) and Fpg-sensitive sites including the 8-oxo-dGuo oxidized base (Fpg-modified version of the assay). In cells exposed to E171 in the acute exposure scenario, no significant DNA damage was observed (Table S5). In contrast, in repeatedly-exposed cells, a significant increase in %tail DNA was observed, in the Fpg-modified version of the comet assay only, in Caco-2 cells exposed to 10 or 50 lg/mL E171, and in Caco-2/HT29-MTX cells exposed to 50 lg/mL E171 (Table 3). Exposure to TiO 2 -NPs did not cause oxidative damage to DNA in our assays (Table 3, Table S5). 0.23 ± 0.05 a Primary diameter (diam.) was measured on TEM images (the diameter of 200-500 particles was manually measured), specific surface area (SSA) was determined using the Brunauer, Emmett and Teller (BET) method.
Crystalline structure (cryst. struct.) was identified by X-ray diffraction. Hydrodynamic diameters (HD), mean ± standard deviation, corresponds to the Z-average values given by the Nanosizer; PdI: polydispersity indexes.
HD measurements were performed in ultrapure water (water) or in cell culture medium (med.). The zeta potential (f) was measured in cell culture medium before cell exposure.

Endoplasmic reticulum stress
ER homeostasis is strictly controlled by the cellular redox status (Malhotra & Kaufman, 2007). ER stress is classically assessed based on expression levels of markers of the UPR. The mRNA expression of five of these markers, which are typically up-regulated in ER stress conditions, was measured in cells exposed to E171 and TiO 2 -NPs to assess ER homeostasis. Unexpectedly, acute exposure to E171 for 48 h induced a significant decrease in the expression of ATF6 in Caco-2 cells ( Figure 5(A)). Similarly, in Caco-2/HT29-MTX cells, CHOP, ATF6, GRP78 and IRE-1 were downregulated after exposure to E171 (Figure 5(B)). Repeated exposure of Caco-2 cells to E171 induced downregulation of CHOP and GRP78 (Figure 5(C)), whereas in Caco-2/HT29-MTX cells it led to downregulation of sXBP1 ( Figure 5(D)).
Some of these markers were also downregulated in cells exposed to TiO 2 -NPs. Thus, in Caco-2 cells exposed to A12, whatever the conditions, and in Caco-2/HT29-MTX repeatedly exposed to P25, CHOP expression was decreased. In both cell models repeatedly exposed to P25, GRP78 was decreased, and in Caco-2 cells repeatedly exposed to P25, ATF6 levels were lower than in control cells (Table S6).

Discussion
The objective of the present study was to shed some light on the mechanisms behind the potential toxicity of E171 on intestinal cells. Two in vitro models of the human ileum were used, the first one was a culture of enterocytes alone, while the second one consisted of enterocytes and mucus-secreting cells. Both models were exposed to E171 or two types of TiO 2 -NPs in two different exposure regimes: acute (6-48 h), or repeated (three times a week for 3 weeks) as a simulation of chronic exposure that would result from daily consumption of food containing this additive. The cellular responses to this exposure were monitored.
In terms of toxicological impact of E171, the results of this study indicate that: (i) E171 is only moderately toxic to the two cell models tested; (ii) repeated exposure triggers an increased cellular response compared to acute exposure; (iii) cells respond more to E171 than to TiO 2 -NPs; (iv) the two cell models respond similarly to E171; and (v) the main mechanism driving the toxicity of E171 is oxidative stress.
The exposure conditions used were harsh, with high concentrations and long exposure times. But nevertheless, the overall impact of E171 was modest. The only responsive endpoint identified was oxidative stress, including moderate intracellular accumulation of ROS and downregulation of expression levels for antioxidant enzymes. That the level of oxidative stress induced was damaging was confirmed by the increase in the level of Fpgsensitive DNA lesions observed, primarily 8-oxo-dGuo. These lesions are considered mutagenic, because 8-oxo-dGuo can pair with adenine and cytosine with equal efficiency (Carriere et al., 2017). Therefore, the accumulation of 8-oxo-dGuo observed in this study could cause the cell transformation reported by others upon chronic in vitro exposure of lung cells to TiO 2 -NPs . Our results thus indicate that the main mechanism behind E171 toxicity is oxidative stress, as already described for TiO 2 -NPs in lung and intestinal models (Johnston et al., 2009;McCracken et al., 2016;Shi et al., 2013).
In contrast to the effects on DNA, exposure to E171 only induced minor changes in the expression levels of UPR markers, leading to an overall decrease. Consequently, ER stress was not triggered in our conditions. This finding is in contradiction with previous reports related to in vivo and in vitro exposure of lung models to TiO 2 -NPs (Yu et al., 2015a,b). This apparent discrepancy can be explained by differences in exposure conditions and cell models (lung vs. intestinal), by the physico-chemical characteristics of the TiO 2 particles, and by the doses to which cells were exposed (higher in Yu et al., 2015a,b). In the present study, the overall response of cells to TiO 2 -NPs (mostly anatase) was the same as the overall response to E171 (pure anatase), although it was less intense. This could be surprising because NPs with smaller diameter and larger SSA generally trigger a more intense 40.13 ± 10.24 b,d a Ti concentrations (lg/L) measured by ICP-MS in Caco-2 and Caco-2/HT29-MTX cells exposed to 50 lg/mL of A12, P25 or E171 for 6 h, 48 h or repeatedly for 21 days. Statistical significance, p < .05. b Exposed vs. control (CTL). c 6 h vs. 48 h. d 48 h vs. 21 days. Caco-2 vs. Caco-2/HT29-MTX: none of the differences were statistically significant. The experiment was performed once, with three independent replicates. n/a: not available. (B) cell viability in cells repeatedly exposed to 10 lg/mL of A12, P25 or E171 ('A12', 'P25' and 'E171') or to 50lg/mL of E171 ('E171 high'), as determined using trypan blue (Caco-2) or IP (Caco-2/HT29-MTX) exclusion assays. % of control (Ctl) (unexposed cells), mean ± standard deviation, n ¼ 3.
response than larger ones with smaller SSA. However, here the cellular response correlates well with intracellular accumulation of Ti, itself well correlated with the size of TiO 2 agglomerates in exposure medium. Agglomerates of E171 are larger and consequently settle down more quickly than agglomerates of TiO 2 -NPs, leading to higher cell exposure level which explains higher cell response. Repeated exposure to E171 systematically induced more intense cellular responses than acute exposure. This difference in response was not unexpected as in the repeated exposure regimen, cells received 10 or 50 lg/mL of E171 three times a week for 10 µg/mL 50 µg/mL * * * * Figure 4. Expression of antioxidant enzymes altered in cells repeatedly exposed to E171. RT-qPCR was performed in Caco-2 (A) and Caco-2/HT29-MTX (B) models after repeated exposure for 21 days to 10 or 50 lg/mL of E171, and compared to the respective control (unexposed Caco-2 monoculture and unexposed Caco-2/ HT29-MTX coculture). The genes tested were catalase (CAT), glutathione reductase (GSR), superoxide dismutases 1 and 2 (SOD1, SOD2) and nuclear factor (erythroidderived 2)-like 2 (NRF2). Results are expressed as fold-increase vs. control (unexposed cells), mean ± standard deviation, Ã p < .05, n ¼ 3. Table 3. Chronic exposure to TiO 2 induces DNA damage a .
3 weeks, which corresponds to eight times more particles than in the acute exposure condition and indeed intracellular accumulation of Ti was higher in repeatedly-exposed cells than in acutelyexposed cells. It should be noted that the cells were not subcultured during the 3 weeks of exposure, and therefore the same cells were present from the beginning of the exposure period to the end, progressively accumulating particles. The higher response in repeatedly-exposed cells could also be explained by the fact that, in the repeated exposure regimen, cells are exposed throughout the whole process of their differentiation, and thus on the first day of exposure, they are undifferentiated. Previous studies reported that undifferentiated cells are more sensitive to NPs and accumulate more NPs than differentiated cells (Gerloff et al., 2013;Song et al., 2015). E171 could accumulate more intensively during the first week of the repeated exposure regimen, before Caco-2 cells differentiate. Another possible explanation would be that repeated exposure to TiO 2 impairs or delays the process of cell differentiation. Consequently exposed cells would remain undifferentiated for a longer period of time than control cells, which would lead to increased accumulation of particles. However after 21 days of repeated exposure to TiO 2 , the morphology of tight junctions and microvilli is similar in exposed cells and in control cells. Moreover qualitative inspection of the Caco-2/HT29-MTX coculture, after repeated exposure to 10 or 50 lg/mL of TiO 2 particles, shows that mucus secretion is similar in repeatedly-exposed cells. These observations suggest that the differentiation process is not drastically affected. Cellular responses to continuous vs. acute exposure to NPs have also been compared in other studies. Some of these studies report similar or higher responses in continuously-exposed cells compared to acutely-exposed cells (Annangi et al., 2015;Armand et al., 2016a,b;Kocbek et al., 2010;Vales et al., 2015), while others indicate lower or different types of responses in continuously-exposed cells (Annangi et al., 2016;Aude-Garcia et al., 2016;Comfort et al., 2014). A diminished and altered response is always reported for cells continuously exposed to sub-lethal concentrations of NPs that are prone to dissolution or physico-chemical transformation, such as ZnO-NPs (Annangi et al., 2016) or Ag-NPs (Aude- Garcia et al., 2016;Comfort et al., 2014). This effect can be explained by the fact that NP dissolution would lead to the release of toxic metal ions from cells, thus limiting their toxic impact. Transformed NPs show altered surface reactivity, surface coating and/or agglomeration states. Cells chronically exposed to NPs that are prone to transformation would thus be continuously exposed to a mixture of intact and transformed NPs, and the impact of such exposure undoubtedly differs from the impact of exposure to intact NPs alone. In contrast, continuous exposure to insoluble or inert particles, such as TiO 2 , could lead to cell overload that may impair normal cellular function and eventually lead to cell death. This raises the question of whether an acute in vitro exposure regime can appropriately mimic the chronic in vivo exposure situation. As demonstrated previously for inhalation exposure, acute in vitro experiments rarely reproduce the effects observed in vivo (Sayes et al., 2007;Warheit et al., 2009). Attempts to increase the predictive capacity of in vitro models consist in extending the exposure time, reducing the exposure concentration and applying NPs to cells only for a limited amount of time per day, as would be the case in in vivo protocols. Comfort et al. (2014) developed an in vitro model of chronic exposure to Ag-NPs, mimicking the protocols used for inhalation exposure, by exposing HaCat cells for 14 weeks to low concentrations of Ag-NPs (pg/mL) for 8 h per day, five days per week, and sub-culturing them weekly. Chronically exposed cells showed higher stress than acutely exposed cells, but caused no impact on cell viability. No similar development has been performed to test the effect of oral exposure. We previously compared the in vitro response of acutelyexposed intestinal models to the response observed in mice receiving TiO 2 -NPs orally (Brun et al., 2014). One of the conclusions of this previous study was that in vitro cells were more resistant to TiO 2 -NPs than the cells in the mouse gut, i.e., the same type of response was observed (including altered cell junctions, leading to altered intestinal permeability), but the responses to in vitro and in vivo exposures differed in intensity (Brun et al., 2014). Daily consumption of additive-containing food consists in ingesting the additive once a day, incorporated in a food matrix. As the food matrix passes through the stomach, it is transformed and some elements are removed. The additive would certainly have become coated with some components of this transformed matrix by the time it reaches the intestine. Due to peristaltic movements, the additive then passes through the intestine before it is eliminated. However, some of the additive could be trapped in the intestinal mucus for longer periods of time. In the worstcase scenario, additive trapped in the intestinal mucus would be in continuous contact with epithelial cells. In vitro, a repeated exposure scenario where cells are exposed to the food additive three times a week for 3 weeks is therefore a good approximation of this worst-case in vivo scenario. Consequently, even if refinements may be required, such as lower E171 concentrations, we believe that the in vitro exposure regimen used in this study is a good approximation for the in vivo situation.
Finally, the response of both cell models to E171 exposure was fairly similar, although for some endpoints the response was slightly more intense in Caco-2 cells than in the coculture model, which differs from Caco-2 monoculture due to the presence of mucus. Since the coculture is composed of 70% Caco-2 cells, it is not surprising that its response only minimally differs from that of the monoculture (100% Caco-2 cells). The slightly lower sensitivity of the mucus-containing model could be explained by a lower exposure level, because the mucus could trap some particles and prevent them from reaching cell membranes. Depending on their surface charge, particles can be repulsed by the mucus layer, accumulate inside it, or cross it (Frohlich & Roblegg, 2012). In the present study, the two models accumulated similar amount of Ti, consequently E171 and TiO 2 -NPs can cross the mucus layer. The lower sensitivity of the coculture therefore rather relies on a lower intrinsic sensitivity of the coculture model. The mucus layer can be protective by a mechanism which does not involve physical hindrance of particle interaction with cells. Potentially, the mucus provides cells with essential nutrients that enhance their defensive capacity. Moreover, some mucins and glycoproteins, especially MUC1, can modulate epithelial cell inflammatory signaling (McGuckin et al., 2011). MUC1 also protect adenocarcinoma cells from apoptosis, after exposure to genotoxic agents (McGuckin et al., 2011). The coculture model, which produces MUC1, would therefore respond differently to the well-documented pro-inflammatory potential of TiO 2 particles (Shi et al., 2013) than the monoculture model, and would potentially be protected from their DNA-damaging action.

Conclusions
The results of this study demonstrate that E171 is only moderately toxic to two intestinal cell models in vitro. This toxicity is mainly due to oxidative stress, which results in intracellular ROS accumulation and oxidation of DNA bases; no ER stress or effects on cell viability were observed. Both cell models responded similarly to E171, with a more intense cellular response observed in cells repeatedly exposed to E171 over 21 days, compared to cells exposed to an acute dose for 6 h, 24 h or 48 h. The toxic effects therefore increase as TiO 2 accumulates in intestinal cells, and the presence of a mucus coating on intestinal cells only slightly mitigates these effects. These findings underline the necessity to use repeated exposure regimens rather than acute exposure to correctly assess the mechanisms behind particle and nanoparticle toxicity.