Inflammatory and Oxidative Stress Responses of an Alveolar Epithelial Cell Line to Airborne Zinc Oxide Nanoparticles at the Air-Liquid Interface: A Comparison with Conventional, Submerged Cell-Culture Conditions

The biological effects of inhalable nanoparticles have been widely studied in vitro with pulmonary cells cultured under submerged and air-liquid interface (ALI) conditions. Submerged exposures are experimentally simpler, but ALI exposures are physiologically more realistic and hence potentially biologically more meaningful. In this study, we investigated the cellular response of human alveolar epithelial-like cells (A549) to airborne agglomerates of zinc oxide (ZnO) nanoparticles at the ALI, compared it to the response under submerged culture conditions, and provided a quantitative comparison with the literature data on different types of particles and cells. For ZnO nanoparticle doses of 0.7 and 2.5 μg ZnO/cm2 (or 0.09 and 0.33 cm2 ZnO/cm2), cell viability was not mitigated and no significant effects on the transcript levels of oxidative stress markers (HMOX1, SOD-2 and GCS) were observed. However, the transcript levels of proinflammatory markers (IL-8, IL-6, and GM-CSF) were induced to higher levels under ALI conditions. This is consistent with the literature data and it suggests that in vitro toxicity screening of nanoparticles with ALI cell culture systems may produce less false negative results than screening with submerged cell cultures. However, the database is currently too scarce to draw a definite conclusion on this issue.


Introduction
Exposure to airborne particles has been linked to adverse health effects including pulmonary in�ammation, thrombosis, neurodegeneration, and cardiovascular disease [1][2][3]. A number of studies have indicated that particles with diameters below 100 nm have a more pronounced effect than larger particles, implying that nanoparticles (or ultra�ne particles) are more toxic on a mass basis [3][4][5][6].
Zinc is an ubiquitous transition metal associated with industrial emissions (e.g., mining and smelting of zinc) that typically appears in the form of zinc oxide (ZnO) in ambient particulate matter (PM) [7][8][9]. ZnO is known as an occupational hazard, since inhalation of high concentrations of ZnO formed during welding activities can lead to metal fume fever [10,11] associated with a marked upregulation of proin�ammatory markers in the lung [11][12][13]. In addition to these inadvertently generated ZnO nanoparticles, there is a variety of ZnO nanostructures, which have shown great potential for nanotechnological products including manufacturing and pharmaceutical applications [14,15]. However, there is increasing concern that the desirable technological characteristics of nanosized ZnO may be countervailed by increased health and environmental risks due to toxic effects that do not occur for bulk ZnO. While the enhanced toxicity potential of nanoparticles is at least in part due to their inherently large surface-to-mass ratio [4,6,16,17], there is also evidence that some metal particles trigger additional toxicological pathways making them more toxic (per surface area) than many other particle types (e.g., carbon, polystyrene) [18].
Cell-based in vitro toxicity assays are widely used to assess the toxicity of nanoparticles. ese toxicological in vitro studies are typically performed using cell cultures grown under submerged conditions, where the toxin/stressor is dissolved or suspended (nonsoluble nanoparticles) directly in the cell culture medium covering the cells. While this approach is experimentally simple, submerged cell exposures have two main limitations. First, the particle dose interacting with the cells is typically unknown since the particle fraction reaching the cells can neither be readily measured nor always be calculated from the hydrodynamic properties of the particles (size, density, shape) [19]. is problem is especially pronounced for particles smaller than about 100 nm, when diffusion becomes the dominant transport mechanism [20], leading to loss of particles to lateral walls. e second limitation is that submerged cell-culture conditions represent an unrealistic and arti�cial environment for alveolar epithelial cells in the lungs. In vivo exposure through inhalation involves deposition of PM onto the lung epithelium, that is, the cells are exposed to inhaled air (airborne PM) from one side while being in contact with the blood circulation from the other side. Since submerged cell systems are completely covered with cell culture medium (see Figure 1(b)), in vivo exposure conditions can be mimicked more realistically by exposing epithelial cells at the air-liquid interface (ALI) (Figure 1(a)). Various ALI exposure systems have been introduced [21][22][23][24][25][26][27][28], but it is unclear whether the enhanced experimental complexity of the ALI exposures compared to submerged exposures is �usti�ed. For that reason, we compared the cellular response to nanoparticles aer ALI and submerged exposure.
One of the most widely accepted paradigms of particle toxicity states that particles induce in�ammation via oxidative stress and subsequent activation of redox-sensitive transcription factors [29]. Nel and colleagues re�ned and expanded this concept into the hierarchical oxidative stress paradigm [30,31] suggesting the transition from an antioxidant defense response (tier1) to in�ammation (tier2) and �nally to cytotoxicity (tier3), if the induced stress is strong enough. Proin�ammatory responses mediated by oxidative stress have been proposed to be not only crucial but also the most sensitive readout for particle toxicity [30]. We therefore measured three proin�ammatory cytokines (interleukin-8 (IL-8), IL-6, and granulocyte macrophage colony-stimulating factor (GM-CSF)) and three oxidative stress markers (heme oxygenase 1 (HMOX1), superoxide dismutase (SOD-2), and glutamate-cysteine synthetase, catalytic subunit (GCS)) by qRT-PCR.
In this study the �rst ALI exposure of human epitheliallike cells (A549) to airborne agglomerates of ZnO nanoparticles is presented. e dose-and time-dependent cellular responses of the cells were compared aer ZnO exposure under submerged and ALI conditions at two dose levels (0.7 and 2.5 g/cm 2 ) and two time points (0 h or 2 h aer incubation). From the exposure-speci�c in vitro toxicity data, we deduced corresponding lowest observed effect levels (LOELs) and compared them with similar studies available in the literature.

Materials.
Common laboratory chemicals were purchased from Sigma-Aldrich (Tauirchen, Germany). e particle exposure experiments were performed with commercially available powder of ZnO nanoparticles (NPs) (Alfa Aesar, Ward Hill, MA, USA ID 43141) with primary particle diameters between 24 and 71 nm (manufacturer information) and a measured BET surface area of 13 ± 2 m 2 /g, which agrees with the manufacturer speci�cations (15-45 m 2 /g) within experimental uncertainties.

Cell Culture.
In this study, the alveolar epithelial-like cell line (A549) from a human lung adenocarcinoma (obtained from ATTC, Manassas, VA, USA) representing the alveolar type II phenotype [32] was used.
For ALI exposure (Figure 1(a)), A549 cells were seeded on perforated Anodisc membranes (Whatman, Maidstone, UK; aluminum oxide, diameter: 47 mm, pore size: 0.2 m) with about 1.6 × 10 5 /cm 2 cells and cultivated in 25 mL Petri dishes under submerged conditions for 9 d at 37 ∘ C in DMEM/F12/L-Glut/15 mM HEPES buffer (Invitrogen, Germany) containing 100 U/mL penicillin, 100 g/mL streptomycin, and 10% FCS. Aer 9 d a con�uent layer with a cell density of approximately 5.1 × 10 5 /cm 2 was obtained. 1 hour prior to particle exposure, the cells were transferred to the ALI, by taking the six cell-covered membranes from the Petri dishes and placing them in two cell exposure chambers (described below) using the same culture medium as above but without FCS. is arrangement allows nourishment of the cells with a cell culture medium through the perforated membrane from the bottom and exposure to airborne particles from the top. Immediately aer ALI exposure, the cells were washed with PBS and gently scrapped off the membranes aer adding trypsin/EDTA (for RT-PCR). For reasons discussed below, one of the cell-covered membranes in each ALI exposure chamber was incubated for a postincubation period of 2 h (submerged in 3 mL medium at 37 ∘ C) prior to determination of the biological endpoints.

2.3.
Exposure at the ALI. e ZnO powder was aerosolized with a commercially available venturi-type dry powder disperser (Model SAG 410, TOPAS, Leipzig, Germany) optimized for output stability by taking the following measures: (i) the metal venturi nozzle was replaced by a ceramic nozzle to avoid chemical and mechanical erosion, (ii) the particle reservoir and the inlet of the venturi nozzle were permanently �ushed with dry nitrogen instead of �ltered ambient air to minimize clogging due to moisture effects, (iii) the scraper in the reservoir was modi�ed to allow for permanent stirring of the powder especially at the bottom of the reservoir, (iv) the aerosol output was passed through a buffer volume to remove extremely large particles (sedimentation) and smoothen �uctuation in ZnO NP concentration, and (v) particle growth due to coagulation was minimized by diluting the aerosol (1 : 1) with compressed �ltered air directly aer generation.
A detailed description of the ALI exposure chamber used here was provided by Bitterle et al. [21]. Brie�y, ZnO aerosol was generated at a �ow rate of 1.5 L/min with the generator described above and evenly distributed to two cell exposure chambers (one for particle exposure or control) holding three cell-covered Anodisc membranes each. e two chambers were operated in parallel using symmetric �ow splitters with the control atmosphere (clean air) being obtained by �ltration with a PALL �lter (BB50TE, PALL, Newquay, UK). Each chamber was supplied with 0.25 L/min aerosol-laden air (or �ltered air for control), which was directed at a radially symmetric stagnation point �ow pro�le over the cell-covered membrane. is design assures spatially uniform particle deposition onto the cells at a deposition fraction, which was experimentally determined to be almost constant (2% of the particles in the sample �ow) over a broad particle size range of about 50 to 500 nm [33], due to the compensating effects of diffusional and gravitational deposition [34]. e air �ow was conditioned to 37 ∘ C and 99.5 % relative humidity. A more detailed description of the ALI exposure chamber is provided by Bitterle et al. [21]. e particle number size distribution was measured immediately downstream of the exposure units with a scanning mobility particle sizer (SMPS, model 3080, TSI, St. Paul, MN, USA, combined with a TSI model 3025A condensation particle counter). By maintaining a constant particle concentration (within about ±20%) during the 3 h exposure time, the cell-delivered particle dose increased linearly with time. Aer 3 h the �nal dose was reached and the biological parameters were evaluated at this time point (referred to as 0 h) and aer an additional 2-hour postincubation time at 37 ∘ C under submerged conditions (referred to as 2 h).

Exposures under Submerged Conditions.
For ZnO exposures under submerged conditions, the culture medium in each well was replaced with serum-free medium into which NPs of 0.7 and 2.5 g/cm 2 of well area were given by adding the appropriate volume of a freshly prepared 1 mg ZnO/mL H 2 O stock suspension (vortexed and sonicated twice for 1 min intermittently immediately prior to application). e size distribution of the ZnO NPs in suspension was determined with a dynamic light scattering sizer (DLSS) (HPPS 5001 Malvern Instruments Ltd., Worcestershire, UK). As shown in the Results section, gravitational settling was sufficient for all particles to reach the cells within 1 h. us, the �nal particle dose was delivered to the cells aer a 1 h exposure time. e biological parameters are reported relative to control conditions (incubated cell cultures without ZnO) either directly aer the exposure time (referred to as 0 h) or aer an additional 2 h postincubation time (referred to as 2 h).

2.�. qRT��CR Measurements o� �roin�ammatory and �xi� dative Stress Markers (mRNA Expression).
Gene expression levels of interleukin-8 (IL-8), IL-6, granulocyte macrophage colony-stimulating factor (GM-CSF), the antioxidant enzyme heme oxygenase 1 (HMOX1), superoxide dismutase (SOD-2), and glutamate-cysteine synthetase, catalytic subunit (GCS), were measured by RT-PCR with SYBR green. Aer exposure, cells were lysed and homogenized in a buffer containing guanidine isothiocyanate and total RNA was isolated using a RNeasy kit according to the method recommended by the manufacturer (Quiagen, Germany). To detect cytokine mRNA expression, RNA was reverse-transcribed into cDNA using the First-Strand cDNA Kit (Pharmacia, Germany). For PCR ampli�cation, the above-mentioned cDNA served as template and 3 L was added together with the speci�c 5 ′ and 3 ′ primers to the Absolute QPCR SYBR Green Mixes from ABgene (ermo Fisher Scienti�c, Germany). Quantitative PCR was performed in a TaqMan instrument (TaqMan ABI Prism 7700 Sequence Detector System; Perkin-Elmer, Germany) offering the advantage of fast and realtime measurement of �uorescent signals during ampli�cation. e housekeeping gene glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was used as internal reference to normalize the RNA levels of the genes being studied.

Viability Assay.
Cell viability was measured with the cell proliferation reagent WST-1 (Roche Applied Sciences, Germany). e WST-1 reagent is a ready-to-use solution which was added to the cells at a concentration of 100 L/mL. Light absorbance was measured aer 30 min incubation at 37 ∘ C at 450 nm (iEMS Reader MF, Lab Systems).

Statistical Analysis.
Results are presented as geometric mean and geometric standard error of the mean of at least four separate experiments ( = 4-7), since the data are not normally but close to log normally distributed. Data  Equivalent mobility diameter (nm) Normalized volume (or light intensity) Submerged Air-liquid interface Air-liquid interface (fitted) F 2: Typical ZnO particle size distribution during ALI submerged (SUB) exposure conditions, respectively. When comparing the particle size distributions during ALI and submerged exposures, one has to consider that different sizing instruments were used. As discussed in the text, both SMPS (ALI) and DLSS (submerged) measure the particle mobility diameter. Furthermore, the volume distribution (normalized to the maximum volume level) is approximately equal to the (normalized) light intensity distribution in the size range between about 315 to 1250 nm (highlighted by the grey shaded area), which encompasses most of the size regime of interest here. e ALI size distributions showed volume-weighted median diameters of about 335±40 nm and a width of 1.77±0.05 (geometric standard deviations). For the submerged conditions (SUB), dynamic light scattering measurements (DLSS) showed ZnO aggregates of about 900 nm (mobility diameter) with a less pronounced (∼20%) secondary mode near 350 nm. us it is evident that the average ZnO agglomerates were considerably larger during submerged than during ALI exposure condition. For comparison with other studies, the number-weighted size distribution of the ZnO particles during ALI exposure had a count median diameter of about 140 nm (data not shown).
comparisons were carried out using the Kruskal Wallis test (Statgraphics plus 5.0), a nonparametric one-way analysis of variance (ANOVA).
0.05 was considered as statistically signi�cant.

Particle Size
Distribution. e measurement of particle size distributions in different media, such as air and liquid during ALI and submerged exposure, respectively, requires the use of different measurement techniques, here scanning mobility sizing with an SMPS (ALI) and dynamic light scattering using a DLSS (submerged). e SMPS counts individual particles, which are size-selected based on their migration speed in an electric �eld [35]. e DLSS determines the mobility diameter from the time-dependent �uctuations of the scattered light intensity signal from an ensemble of suspended particles [36]. While both instruments measure the mobility diameter ( -axis of size distribution as depicted in Figure 2), the SMPS counts individual particles and the DLSS reports a signal proportional to the light intensity of a given particle [35]. Since the light intensity signal cannot be directly related to particle number concentration, the SMPS number distribution was converted into effective volume (or mass) distribution, which can be related to the scattered light intensity as described below. Accurate performance and comparability of both instruments was validated with NISTtraceable (National Institute of Standards and Technology, USA) reference particles.
For ALI exposures, the SMPS measurements of the ZnO aerosol revealed a count median (mobility) diameter (CMD) and geometric standard deviation of 141 ± 12 nm and 1.77 ± 0.05, respectively. Since the CMD is larger than the diameter of the primary ZnO NPs (24-71 nm), it is evident that the ZnO aerosol mainly consists of agglomerated (nonspherical) structures. For the two dose levels studied here, the mean and standard deviation of the number concentration was (3.5 ± 0.45) × 10 5 and (9.5 ± 0.9) × 10 5 particles/cm 3 . is corresponds to average mass concentrations of 10.1 mg/m 3 and 30.4 mg/m 3 , respectively, with an almost constant mass (or volume) median diameter (±standard deviation) of 335 ± 40 nm. As mentioned above, for comparison of SMPS and DLSS, data the SMPS data were converted from numberinto mass-based (volume-based) size distribution taking into account the nonspherical shape of the ZnO particles as follows. First, the number size distribution was converted into volume distribution (assuming spherical particle shape for now) and �tted as lognormal distribution using the Hatch-Choate equations for consistent conversion of the count median into mass median diameter. Integration of the volume distribution yields the total volume. For correct conversion into particle mass accounting for the nonspherical particle shape, the volume is multiplied by the effective density of 4.6 g/cm 3 [37], which was experimentally determined by dividing the gravimetrically determined particle mass by the (spherical) particle volume determined from the SMPS data (see Figure 2). e fact that the effective density of the ZnO aerosol is smaller than the bulk density of ZnO (5.6 g/cm 3 ) is consistent with the agglomerated structure of the ZnO particles [37]. e relatively small difference between effective and bulk density indicates that the particles have a relatively compact (sphere-like) structure.
Although exposures under submerged conditions were performed with the same ZnO particles as used for ALI exposures, ZnO particles suspended in the cell culture medium were more agglomerated and hence larger than those dispersed in air (ALI exposure). As seen from the DLSS size distribution depicted in Figure 2, the ZnO particles, suspended in cell medium for 30 min, displayed a minor mode near 350 nm (about 20% of total mass) and a more pronounced mode near 900 nm (∼80% of total particle mass). For accurate comparison of the DLSS sizing data, one has to relate the volume-based size distribution derived from the SMPS data with the light intensity values of the DLSS. For particle sizes near the wavelength of the light source (between about and ), it has been shown that the light-intensity-to-volume ratio is almost constant [38]. us for the Malvern DLSS ( 33 nm), we can assume that the normalized light intensity distribution (normalized to the maximum of the intensity spectrum) as shown in Figure 2 is approximately equal to the normalized volume (or mass) distribution obtained from the SMPS in the size range between 315 and 1250 nm (shaded area in Figure 2), which covers most of the size range of interest for the present study. Outside this range, the light intensity level is systematically lower than the corresponding volume level [38].
In summary, it is evident that the ALI size distribution was dominated by ZnO agglomerates with a volume median diameter near 350 nm. While this mode is also seen during submerged exposures, most of the particle mass (∼80%) resides in a mode near 900 nm indicating that suspending the ZnO particles in cell culture medium for 30 min leads to enhanced particle size due to agglomeration effects.

Particle Dosimetry.
For reliable comparison of the cellular dose-response relationship under ALI and submerged culture conditions, the biological response should be correlated to the cell-delivered particle dose ( ) normalized to the cell-covered surface area. Here is given by where is the particle mass passing or �oating over the cell layer during the exposure, Dep is the deposition efficiency (fraction of particles depositing onto the cell layer), and cell is the area covered by the exposed cells.
For the ALI exposure, is calculated from , where is the average mass concentration (ZnO mass per volume air; 10.1 and 30.4 mg/m 3 for the low and high dose level, resp.), 0. 5 L/min is the volumetric �ow rate passing over the cell layer, and 3 h is the exposure time. With cell 1 . cm (per culture membrane) and Dep = 0.02 [21,33], we �nd from (1) that ,ALI 0.7 and 2.2 g ZnO/cm for the low and high doses, respectively. With the speci�c BET surface area of 13 m /g, this corresponds to BET surface area doses of 0.09 and 0.29 cm ZnO/cm , respectively.
For submerged exposures, we substitute by in (1), where is the ZnO particle mass concentration in the stock suspension (here 1 mg/mL) and is the volume of the ZnO stock suspension (here 1.4 or 5 L) added to the culture medium (1 mL). Under the assumption that all particles contained in the medium will deposit on the cells within 3 h (i.e., Dep = 1; will be �usti�ed below) we �nd from (1) that ,sub 0.7 and 2.5 g ZnO/cm (with cell .0 cm ) or 0.09 and 0.33 cm ZnO/cm , respectively.
As a �usti�cation for Dep = 1 under submerged conditions, the following aspects were considered [39]: (I) Is the particle deposition dominated by sedimentation or diffusion (the latter would result in loss of particles to the lateral walls and hence Dep < 1) and (II) if sedimentation dominates particle deposition, is the exposure time long enough for all particles (even the ones near the top of the cell culture well) to reach the cells at the bottom of the well? To address these issues, we calculated the gravitational settling velocity and the mean diffusional displacement speed of the particles in water to be 56 mm/h and 0.033 mm/h, respectively [20], where we assumed an average particle diameter of 900 nm (see Figure 2). Since the ratio of gravitational to diffusional displacement speed is about 1700 for 900 nm particles with a density of 4.6 g/cm 3 (ZnO), sedimentation is the dominant deposition mechanism; that is, negligible particle loss to lateral walls is expected. Secondly, for a sedimentation speed of 56 mm/h and a 50 mm depth of the cell culture medium (1 mL of medium; 2 cm cross-sectional area of well), all ZnO particles are expected to deposit onto the cells within about 1 h.
As a caveat we note that due to differences in deposition kinetics during ALI and submerged exposures (as described above), the �nal dose was delivered to the cells a�er 3 h and 1 h, respectively. [32,40]. First, cell viability was determined to exclude the possibility that the observed effects of ZnO NPs on gene expression levels are negatively biased due to cytotoxic effects. No effects on cell viability were seen for the ZnO concentrations investigated here for both ALI (viability in % of (submerged) unchallenged control: 93.5% ± 2.1, at ≤2.2 g/cm 2 ) and submerged conditions (94% ± 7.99, at ≤2.5 g/cm 2 ). Hence, the cellular response to ZnO exposure was not signi�cantly hampered by reduced cell viability and there was no signi�cant reduction in cell viability due to exposure of the cells to the air-liquid interface.

Biological Endpoints. Several biological endpoints were investigated aer ZnO exposures of the A549 cells representing human alveolar epithelium type II cells
ALI exposure of A549 cells to ZnO NPs caused elevated levels of mRNA coding for IL-8, GM-CSF, and IL-6 as shown in Figure 3(a) (le panel). IL-8 showed a signi�cant increase with increasing dose and time. e time response of GM-CSF was similar to that of IL-8, but no signi�cant dose effect was observed. IL-6 was increased for all ALI exposure scenarios, but no signi�cant dependence on dose or time was observed. On the other hand, the oxidative stress markers HMOX1 and SOD-2 showed no signi�cant increases in mRNA expression except for GCS mRNA which was slightly, but statistically signi�cantly increased for both time points of the high dose level (1.8-fold and 3-fold increased at 0 h and 2 h, resp., see Figure 3

(b), le panel).
Under submerged conditions, the expression levels of all proin�ammatory markers were lower than those under ALI conditions (Figure 3(a), right panel). For IL-8 only the high dose showed a signi�cant induction of 1.9-fold and 3.7-fold at the two time points, and IL-6 was increased (4-fold) for the high dose at 2 h. Out of the three oxidative stress markers, a signi�cant expression was observed only for HMOX1 (2.7fold) aer 2 h (Figure 3(b), right panel).
In summary, compared to the submerged conditions, ALI exposure showed slight, but statistically signi�cant enhancements in mRNA expression of IL-8, GM-CSF, and IL-6 for all dose levels and time points as well as for the high dose level of GCS (both time points). e only case where the submerged exceeded the ALI response was HMOX1 (high dose, 2 h). us the ALI exposure system was generally more sensitive to mRNA induction than the submerged exposure assay especially for the proin�ammatory markers.

Discussion
To the best of our knowledge, the data presented here represents the �rst in vitro measurements of the cellular response to an exposure of airborne agglomerates of ZnO particles at the ALI. We model inhalation exposure to ZnO NPs with the widely used A549 cell line. A549 cells represent human alveolar epithelial type II cells [32,40], which are considered the defenders of the alveoli because they are important producers of cytokines [41] and metabolically more active than type I pneumocytes [42,43]. Consequently, A549 cells are widely regarded as a valid model cell system for pulmonary particle toxicity studies [44,45]. e data presented here are consistent with positive dose-response correlations. Focusing on ALI conditions �rst, we �nd that the IL-8 response (2 h postincubation time) is enhanced for the higher concentration with a 94.5% con�dence level ( . 55). A positive dose-response is also found for GCS (2 h postincubation time; 95% con�dence level). On the other hand, no signi�cant response is found for HMOX1 and SOD2 for any of the concentrations used here, since none of these parameters is upregulated. Furthermore, the lack of a dose-response correlation for GM-CSF and IL-6 might be due to reaching the saturation levels already for the lower concentration. Similar considerations can be conducted for the submerged cell culture data. A positive dose-response is seen for IL-8 (both time points), IL-6, and HMOX-1 aer 2 h postincubation time. All other parameters are not changed. While historically in vitro particle toxicity studies have been performed with submerged cell systems, various ALI exposure systems have recently been introduced in an attempt to mimic more realistically the exposure conditions during particle inhalation [21][22][23][24][25][26][27][28]. Further advantages of ALI exposures include the preservation of the physicochemical characteristics of the airborne particles (e.g., particle agglomeration and/or particle-medium interactions such as partial dissolution of ZnO in cell culture medium are avoided [46]), the synergistic effects between particulate and gaseous compounds can be investigated (e.g., relevant for combustion emissions) and the biological complexity can be more adequately represented (e.g., surfactant coating can be added to of alveolar epithelial cells). Last but not least, it is typically technically simpler to determine the cell-delivered particle dose under ALI than submerged conditions [47,48]. Some of the recently introduced ALI exposure systems utilize aerosolized nanoparticle suspensions instead of dry airborne nanoparticles [28]. While these systems allow for cell exposure at the ALI, partial dissolution and possibly agglomeration of the nanoparticles cannot be ruled out with these systems.
Although ALI exposures have become more widely used, there is very little quantitative information on whether and how the exposure type (ALI and submerged) affects the cellular response. A summary of the currently available studies is listed in Table 1 and will be discussed below. In Table 1 all but one investigator utilizes gene expression analysis instead of protein determination as toxicological readout. Gene expression is commonly preceding protein expression; however, the latter can additionally be regulated at the posttranscriptional level. In some cases, protein expression without associated gene expression can occur. However, this is not the case for any of the markers listed in Table 1. Hence, both protein and gene expressions are suitable for toxicity studies. e advantages of gene expression analysis by qPCR include higher sensitivity than the measurement of protein levels, simultaneous quanti�cation of several markers and higher cost efficiency. For these reasons, gene expression analysis was used in the present study to screen for representative markers of different acute response pathways related to in�ammation and oxidative stress.
As mentioned above, the cellular response to nanoparticles depends on numerous aspects including cell type, preand postprocessing of the cells, state of cell differentiation, particle dose, deposition kinetics, and physicochemical particle characteristics. Matching all of these aspects is very difficult, if not impossible, since, for instance, the state of cell differentiation is inherently different under submerged and ALI culture conditions [49][50][51] and particle deposition rates onto the cell system may vary signi�cantly for submerged and ALI conditions, since they depend on agglomeration state and carrier medium of the particles (air or liquid). Hence, any study on cellular response under ALI versus submerged cell conditions should provide as much details on these aspects as possible as is done in the following section.
Exposures were performed at two dose levels with no statistically signi�cant difference for the two exposure types (low: 0.7 g/cm 2 ; high: 2.2 g/cm 2 (ALI) and 2.5 g/cm 2 (submerged)). In addition, the cell type (A549) was identical for both exposure types and the cell densities (cells per cm 2 ) were similar at the time of exposure. Preprocessing of the cells was different because it had to be adapted to the two different exposure conditions, whereas postprocessing of the cells was identical. ere have been several studies indicating differences in cell differentiation due to transfer of the cells from submerged to ALI culture conditions even at the �rst investigated time points between a few hours and 1 d [49][50][51]. However, we assume that in the present study the state of cell differentiation was similar, since cells were kept under submerged conditions except for a brief period of time during ALI exposure (1 h prior to exposure; 3 h during exposure). is is supported by the fact that we found no statistically signi�cant differences in cell viability as well as IL-8 mRNA and HMOX-1 mRNA aer transfer of the cells to ALI conditions. is is an important aspect, since with an already strained antioxidant defense, as, for instance, reported by [49], cells may be more susceptible to the effects of the particle exposure. Other important aspects for ALI-submerged comparisons are related to the particle characteristics. For ALI conditions, the count median and mass median particle diameters were 141 nm and 335 nm, respectively, and the rate of particle deposition onto the cells was constant (within 20%) during the 3 h exposure time by keeping the sample �ow and the ZnO aerosol concentration constant. Under submerged conditions, the size of the ZnO particles increased from a mass median diameter of about 350 nm to about 900 nm within about 30 min due to agglomeration, which results in an approximately 3-fold deposition rate; that is, the entire ZnO dose is delivered to the cells with about 1 h. Hence, differences in deposition kinetics may affect the comparison of the two exposure scenarios. Since the total number of primary particles in the medium is not changing with the agglomeration state, agglomeration does not affect the number of primary particles or the surface area dose delivered to the cells. However, agglomerate size may in�uence the biological response of the A549. Furthermore, ZnO is partially soluble in aqueous media [30]. Hence, the Zn 2+ /ZnO ratio may be different under ALI and submerged conditions with higher Zn 2+ /ZnO ratios to be expected under submerged conditions due to the relatively high dissolution of ZnO in the cell culture medium.
In spite of some experimental differences between submerged and ALI exposures, as described above, it is instructive to compare the ZnO dose-response curves observed under ALI and submerged conditions and relate these �ndings to similar data sets for other particle and pulmonary cell types from the literature. is can be done by determining the dose range in which the lowest observed effect levels (LOELs) occurred. If none of the two dose levels investigated here showed a statistically signi�cant response, the LOEL lies above the highest dose level (>2.5 g/cm 2 ). If the low dose showed no response, but the high dose did, then the LOEL falls in the range of 0.7-2.5 g/cm 2 . If both dose levels showed a response, then the LOEL is below <0.7 g/cm 2 .
As seen from Table 1, our data indicate that four biological parameters (mRNA levels of IL-8, GM-CSF, IL-6, and GCS) showed lower LOELs and hence elevated response levels under ALI conditions. e results for two of the six investigated parameters (HMOX1, SOD-2) were inconclusive, since the investigated dose regime was not broad enough to discern differences in LOEL. Similar results were reported by other studies with pulmonary cell lines and primary cells reported in the literature (Table 1). Volckens and colleagues [26] exposed primary human bronchial epithelial cells to concentrated coarse ambient particulate matter. Applying our LOEL scheme to their data indicates that the mRNA levels of IL-8 and HMOX1 were more pronounced under ALI conditions, while no conclusive result was found for COX-2 mRNA expression. Holder and colleagues [24] investigated Diesel exhaust particles with a human bronchial epithelial cell line (16HGE14o). As seen from Table 1, they found no conclusive result for IL-8 protein levels but state that a much smaller dose was required to induce similar IL-8 expression levels. We contend that this claim is not substantiated by their data, since in contrast to submerged exposures their IL-8 response under ALI conditions was not statistically signi�cantly different from unity (according to their own statistical analysis). It is important to note that the currently available data on submerged versus air-liquid exposure comparisons is limited, but diverse. As seen from Table 1 the data were generated with both immortalized and primary cell cultures as well as with different particle types, various biological endpoints, and different pre-/postprocessing protocols. ese differences are likely to result in exposuredependent differences in the state of cell differentiation, particle-cell interaction, and deposition kinetics. In spite of this heterogeneity, none of the currently available studies has identi�ed a biological parameter, which responded to be more sensitive to particle challenge under submerged exposure conditions than under ALI conditions. While it cannot be inferred that this is true for all possible biological endpoints, the currently available data suggest that air-liquid interface exposures are a more "conservative" toxicity test than submerged systems, that is, ALI cell systems are likely to lead to less false negatives.
To put the particle dose levels typically used for in vitro toxicity testing into perspective, it is instructive to consider T 1: Comparison of the lowest observed effect levels (LOELs) for nanoparticles exposure of cells exposed at ALI and submerged conditions. F 3: Comparison of the effect of ZnO on proin�ammatory and oxidative stress markers in A549 cells following exposure at the ALI and under submerged (S�B) conditions. (a) mRNA expression of proin�ammatory cytokines (IL-8, GM-CSF, and IL-6) was measured with RT-PCR either directly aer (0 h aer incubation) or two hours aer the exposure (2 h). (b) Same as (a), but for oxidative stress markers (HMOX1, SOD-2, and GCS). e postincubation of the cells aer ALI exposure was also performed under submerged conditions. e mRNA values were normalized to glyceraldehyde-3-phosphate dehydrogenase (GAPDH) levels and expressed as the fold increase over control (the control level was set to unity) which was �ltered air and pure medium for ALI and S�B, respectively. e data show the geometric means and geometric standard error of the mean based on 4 to 7 independent experiments. Due to differences in the deposition kinetics described in the experimental section, the �nal dose was delivered to the cells aer 3 h (ALI, open bars 0.7 g/cm 2 and solid bars 2.2 g/cm 2 ) or 1 h (submerged, open bars 0.7 g/cm 2 and solid bars 2.5 g/cm 2 ). e symbol ( * ) indicates signi�cant differences from control levels at , and ( * * ) at . e symbol (#) indicates mRNA values which are statistically different from the corresponding submerged mRNA levels (differences are 2.8 to 12-fold ( )).

Reference
that the currently recommended Occupational Safety and Health Administration (OSHA) standard for ZnO fume (and many other occupational dusts) is 5 mg of ZnO fume per cubic meter of air (mg/m 3 ) averaged over an eight-hour-perday work shi. Assuming an accumulated breathing volume of 3 m 3 in 8 h, a lung surface area of 140 m 2 , an alveolar deposition efficiency of 10-50% depending on particle size, and negligible clearance from the alveolar regime within 24 h [52], the OSHA standard corresponds to a daily alveolar surface dose of 1.1-5.4 ng/cm 2 , which is about 3 orders of magnitudes smaller than what was deposited during the ZnO ALI exposures performed here (0.7-2.2 g/cm 2 ). Furthermore, it can be seen from Table 1 that the LOEL during submerged exposures is typically between 1 and 65 g/cm 2 , which is in the range of the expected lifetime dose (4-18 g/cm 2 ) under worst case conditions represented by a worker exposed to the OSHA (ZnO) dust limit (5 mg/m 3 ) for 5 days per week, 50 weeks per year for 45 years (where we assumed that only about 30% of the lung-deposited particles remain in the alveoli due to alveolar clearance mechanisms). As the in vitro dose is deposited onto the cells within a few hours instead of 45 years, this does not represent a realistic in vivo exposure scenario. In spite of these unrealistically high dose levels, in vitro cell tests are useful for pharmacological and toxicological prescreening of substances and studies of cellular response mechanisms, but lower cellular doses may be desirable. Table 1 suggests that an incremental progress may be possible with ALI cell systems. In combination with other measures such as the use of multicell cocultures instead of single-cell cultures, this may lead to signi�cantly more realistic in vitro dose rates in the future.

Conclusion
In this study the in vitro response of pulmonary epithelial cells to different types of (nano-)particles was compared for air-liquid interface (ALI) and submerged exposure conditions. e scarce data pool on this issue was expanded by presenting the �rst ALI data on airborne agglomerates of ZnO nanoparticles using alveolar epithelial-like type II cells (A549). For ZnO, the lowest observed effect levels (LOELs) of the proin�ammatory markers (mRNA gene expression of IL-8, IL-6, and GM-CSF) were lower under ALI than under submerged conditions, while no signi�cant response was observed for most of the oxidative stress markers (HMOX1, SOD-2, and GCS). ese �ndings are consistent with the few previous comparative studies on this issue indicating that toxicity testing with the conventional submerged systems may yield more false negatives than the more recently developed ALI systems. e dose levels used here and in similar studies reported in the literature are in the range of an entire lifetime dose of occupational dust received by a heavily exposed worker. e ability to induce cellular responses at somewhat lower and hence more realistic dose levels under ALI conditions may provide biologically more meaningful data than those obtainable with the conventional submerged exposures. Further advantages of ALI cell systems include the biologically more realistic exposure scenario (cells in the lungs are exposed under ALI-like not submerged conditions), the absence of inadvertent modi�cations of the particle properties in the cell culture medium (e.g., agglomeration, partial dissolution), and the possibility of direct dose measurement (e.g., quartz crystal microbalance). Depending on the application, these aspects may outweigh the larger experimental complexity of ALI exposures. However, quantitative comparisons of the cellular response under ALI and submerged culture conditions are still very limited. us, further studies are needed to address these issues.