Human airway construct model is suitable for studying transcriptome changes associated with indoor air particulate matter toxicity

Abstract In vitro models mimicking the human respiratory system are essential when investigating the toxicological effects of inhaled indoor air particulate matter (PM). We present a pulmonary cell culture model for studying indoor air PM toxicity. We exposed normal human bronchial epithelial cells, grown on semi‐permeable cell culture membranes, to four doses of indoor air PM in the air‐liquid interface. We analyzed the chemokine interleukin‐8 concentration from the cell culture medium, protein concentration from the apical wash, measured tissue electrical resistance, and imaged airway constructs using light and transmission electron microscopy. We sequenced RNA using a targeted RNA toxicology panel for 386 genes associated with toxicological responses. PM was collected from a non‐complaint residential environment over 1 week. Sample collection was concomitant with monitoring size‐segregated PM counts and determination of microbial levels and diversity. PM exposure was not acutely toxic for the cells, and we observed up‐regulation of 34 genes and down‐regulation of 17 genes when compared to blank sampler control exposure. The five most up‐regulated genes were related to immunotoxicity. Despite indications of incomplete cell differentiation, this model enabled the comparison of a toxicological transcriptome associated with indoor air PM exposure.

anthropogenic particles and fibers (eg, vehicle exhaust, combustion emissions, house dust, and tobacco smoke). Most of these types of inhalation exposure agents have been linked to indoor air problems in buildings. Exposure indoors always constitute a combination of agents, and it is likely that in many cases, synergistic effects of heterogeneous agents instead of a single component are contributing to the observed adverse health effects. 1 For this reason, measuring the combined effects of multiple exposure agents rather than individual components is desirable in indoor environmental assessments.
Animal testing has been historically used for studying toxicity caused by microbial growth, 2 coal combustion derived fine particulate matter, 3 and dust chemicals 4 present in the indoor environment. However, the respiratory system of rodents is anatomically and physiologically different from humans. 5 Consequently, results from animal experiments are not directly comparable with the health effects observed in humans. Additionally, 3Rs (ie, replacement, refinement, and reduction) legislation and ethical considerations have restricted the use of animals in toxicological studies, resulting in replacement of in vivo methods with more affordable and easily implementable in vitro approaches. 5 The epithelial layer of the respiratory system is the first barrier with which inhaled particles interact within the human body. Therefore, new cell culture models in inhalation toxicology aim to mimic the human respiratory epithelium. [6][7][8] Human cell culture models are tissue-engineered from either primary cells (ie, represent the tissue of origin) or secondary cell lines (ie, carcinoma-derived or virus-transformed). Co-cultures of different secondary cell lines have also been utilized for several years in particulate matter (PM) exposure studies. 6,9 Immortalized secondary cell cultures derived from the bronchus and alveolar tissues are relatively common in inhalation toxicology. For instance, secondary human lung epithelial cells have been used for indoor PM 10 toxicity studies 10 and studying pro-inflammatory responses of spores and hyphae. 11 In secondary cell culture models, different types of cells are co-cultured in either submerged or air-liquid interface (ALI) configurations.
However, the nature of secondary cell-based systems has changed from their original state due to their carcinoid origin or immortalization process. 12 Moreover, secondary cell line cultures are quite simple compared with genuine three-dimensional human lung tissues, which consists of more than 40 distinct cell types 13 with different functions.
Consequently, comparing the study results from cell lines to those obtained from healthy human lung cells is not straightforward. 12 Scientists have highlighted that the use of more complex cell culture or tissue models would enable us to study cell signaling and interaction in a more realistic way. 7,8 In addition to rather complex and costly developments such as different micro-tissue chips, for example lung-on-a-chip, 14 stem cell-derived models, 15 and perfused organs, 16,17 companies have begun marketing primary cells extracted from the human nasal cavity, bronchi, and alveoli after surgical procedures or post-mortem. These respiratory primary cells can be further differentiated to human airway tissue consisting of multiple cell types. For example, in vitro normal human bronchial epithelial (NHBE) cell model has been employed for toxicological testing 18 and for inhalation modeling studies. 19 However, human airway construct models can be used only once in a single exposure study due to their restricted division cycles. These cultures are also relatively expensive and laborious to maintain and differentiate.
The human bronchial epithelium is characterized as being a mucociliary phenotype featuring basal, goblet, Clara, ciliated, and intermediate cells. 18,[20][21][22] Goblet or bronchiolar exocrine cells and secretory ducts protect the respiratory epithelium by forming and secreting lung surfactant. Moreover, lung surfactant protects cells via mucociliary clearance against inhalable particles and pathogens (eg, fungi, bacteria, and viruses). In toxicological studies, lung surfactant has been substituted by using water or physiological salt solutions containing phospholipids [23][24][25] and surfactants. 26,27 In addition, some commercially available products, such as Survanta ® and Curosurf ® , have been used. 28,29 Synthetic lung lining fluid (LLF) typically consists of water, phospholipids, fatty acids, sterols, proteins, and antioxidants that have been identified in human-derived lung surfactant. 30,31 The aim of our study was to optimize a human airway construct model for studying indoor air PM toxicity using synthetic LLF to mimic the real human airway exposure more realistically. A commercially available toxicology transcriptome kit for 386 genes associated with toxicological responses was utilized for RNA sequencing. These measurements were paired with conventional toxicological and microscopical assessments, as well as with size-resolved monitoring of airborne PM and microbial determinations.

| Sampling and monitoring indoor air particulate matter
Indoor air particulate matter (PM) samples were collected from a non-complaint residential apartment located in an urban area in Eastern Finland by using NIOSH BC251 two-stage bioaerosol cyclone sampler, 32

Practical Implications
• Human airway constructs cultured from primary bronchial epithelial cells are suitable for studying the toxicological effects of exposure to indoor air particulate matter.
• The tested experimental design provided enough highquality RNA for transcriptome analysis and showed activation of particularly immunotoxicological processes by indoor air PM.
• This airway epithelial model is a promising tool for comparing different indoor environments in environmental health research, also because it is well linkable to corresponding microbiome characterization of the exposure.
Safety and Health, NIOSH). An integrated sample was collected over the period of 7 days, intermittently for 12 hours each day, with total sampling time of 84 hours. The NIOSH sampler was operated at 10 L/min. Samples were stored at −20ºC before use.
Airborne PM for microbial determination was collected using Button Inhalable Aerosol Sampler 33 (SKC Inc.) onto polytetrafluoroethylene (PTFE) filter membranes (pore size 0.45 µm) (Merck Millipore) at 4 L/min daily during the same period. Filter samples were stored also at −20ºC until processing. Size-resolved (PM 2.5 to PM 10 ) particle count and estimated mass concentrations (using a device-specific assumption of particle mass) were monitored with Lighthouse Optical PM3016-IAQ particle counter (Lighthouse Worldwide Solutions) during sampling. A proxy of the total particle mass in the samples was based on the device-derived particle mass concentrations in the room air during sampling and the sampled air volume.

| Synthetic lung lining fluid (LLF)
Composition of synthetic LLF is presented in Table 1. All reagents used in synthetic LLF were purchased from Sigma-Aldrich.
The recipe for synthetic LLF was adjusted based on previous studies. 30,[34][35][36] First, non-water-soluble reagents were dissolved in acetonitrile (Sigma-Aldrich), albumin, and uric acid in HBSS, and rest of the water-soluble reagents in mQ-water. Reagents were combined at room temperature and then mixed continuously (+37°C) for approximately five hours. After mixing, acetonitrile was evaporated under gaseous nitrogen flow in room temperature and pH adjusted to 7.25. Synthetic LLF was aliquoted and stored in −70°C before use.

| Culturing human airway constructs
Human airway constructs were cultured and differentiated from Transepithelial electrical resistance (TEER) of the tissue barrier was measured using EndOhm chamber and EVOM2 ™ resistance meter (World Precision Instruments), and cell cultures were evaluated using light microscope to assess differentiation and wellbeing of the cells.

| Analyses and RNA extraction
Transepithelial electrical resistance (TEER) was measured after the exposure to assess the integrity of the cell barrier. Chemokine inter-

| RNA quality assessment
Integrity and quantity of RNA was verified by Caliper GX RNA LabChip (Perkin Elmer) and Qubit RNA BR system (Thermo Fischer Scientific), respectively. Samples with 100 ng of high-quality RNA (RNA score >8, DNase treated) were used for expression profiling by RNAseq.

| Library preparation for sequencing
QIAseq Targeted RNA panel Human Molecular Toxicology Transcriptome (Qiagen) of 386 genes associated with toxicological responses was used for first-strand synthesis, molecular barcoding, genespecific amplification, sample indexing, and library preparation for targeted RNA sequencing according to instructions by manufacturer.

| Sequencing of RNA amplicons
Illumina HiSeq2000 platform was used for sequencing. All the samples were pooled in one lane of HiSeq flow cell with capacity of 238 million 100 bp reads.

| Data analysis
The data were analyzed in GeneGlobe Data Analysis Center. In brief, after uploading the RNAseq data in FASTQ format the data were trimmed according manufacturer's instructions and aligned to GRCh38 reference genome using STAR RNA read mapper. After processing the alignments, molecular tags (unique molecular identifiers, UMIs) were counted for 386 genes including gDNA controls, reference genes, and a summary of data quality.
For further data analysis, the Secondary QIAseq Targeted RNA Panel Data Analysis Software (Qiagen) was used. After data normalization (trimmed mean of M, edgeR) 37 and defining the groups for comparison, this software analyzed unique molecular index (UMI) counts to calculate changes in gene expression. Gene expression data were analyzed using Volcano plot analysis with minimum fold regulation of two (Student's t test, P < .05).

| Quantitative and qualitative analyses of microbiota
PTFE filter samples collected with Button Inhalable Aerosol samplers were processed as previously described. 38 In brief, after an ini-

| Microscopy
Sample preparations and transmission electron microscopy (TEM) imaging were undertaken at the SIB Labs (University of Eastern Finland, Kuopio, Finland).

| Light microscopy (LM)
Semi-thin resin sections of 1.0 µm were cut from TEM blocks and stained with toluidine blue. 48

| Statistical analyses
Statistical analyses and figures were done using GraphPad Prism 8.1.1 (GraphPad Software). The normality of the data was tested with the Shapiro-Wilk test, 50 P < .05. For comparing TEER, chemokine IL-8 secretion, and protein secretion of exposed airway constructs, controls (LLF, growth medium, and LPS exposed) were compared using ordinary one-way ANOVA, Bonferroni's multiple comparisons test, 51 P < .05 and PM exposed cells using two-way ANOVA, Bonferroni's multiple comparisons test, P < .05.

| Particulate matter concentrations and microbial content of PM samples
Particle concentrations in indoor air as well as microbial levels and composition were monitored during individual days of the sampling period. Based on the particle count and flow rate, we estimated the total particle mass in 7-day integrate indoor PM sample used in the exposure. ing days, linked to activity patterns in the monitored apartment ( Figure   S1). PM concentrations, and bacterial and fungal levels as well as bacterial species richness varied considerably between sampling days (Table S1 and Figure S2). This supports the chosen approach to collect and analyze an integrated sample of 12-hour sampling periods from seven subsequent days. The bacterial content of the 7-day integrated sample was dominated by Staphylococcus and other genera that are likely of human origin.

| Light microscopy and TEM imaging
In light microscopy images, we located two nucleus layers, mucin granules, and inter-cellular gaps ( Figure 1A). In PM exposed cells, we observed loosened inter-cellular gaps between cell layers and signs of cytolysis (ie, the dissolution or disruption of cells, especially by an external agent) throughout the cell layers following PM exposure ( Figure 1B).
In TEM images, the NHBE cell layer was determined to be 12 µm thick consisting of two nuclear layers (Figure 2A). Compared with control ( Figure 2A), we observed multiple signs of cell distress after PM exposure ( Figure 2B).
Additionally, we observed microvilli, adherens junctions, and desmosomes in TEM images ( Figure 3). No cilia, basal bodies anchoring cilia, glycocalyx around cilia/microvilli, pseudostratified morphology or maturation of columnar-shaped ciliated and goblet cells were visible. We did observe 500 nm long microvilli without basal bodies and a 9 + 2 axoneme organization, which are characteristic of fully differentiated NHBE tissues ( Figure 3). 18

| TEER measurements, protein secretion, and chemokine IL-8 secretion
Human airway constructs were secreting surfactant after 10 days in air-liquid interface (ALI), which is typical for lung epithelial tissue. However, chemokine IL-8 production of LLF, growth medium, and LPS exposed cells did not differ from each other significantly (ordinary oneway ANOVA, Bonferroni's multiple comparisons test, P < .05) ( Figure 5).
Chemokine IL-8 secretion was consistently, but not statistically significantly increased due to PM exposure in different doses compared with control ( Figure 6A). NHBE barrier resistance did not significantly decrease after indoor PM exposure ( Figure 6B) and neither did the total protein concentration increase in apical wash after PM exposure ( Figure 6C).

| Gene expression profiling
Multiple genes were up-and down-regulated upon PM exposure  (Table 2). In addition, one oxidative stress-related gene (OGG1) was up to eightfold down-regulated compared with control ( Table 2). in T-cell activation 55 and CD8A, which is associated with antigen processing and presentation, 56 signaling pathways of cell surface receptors, 57 differentiation of cytotoxic T cells, 57 and immune response. 58 The fifth of the most up-regulated genes, CYP1A1, is associated with immunotoxicity 52 and xenobiotic metabolism. 59 It is also known to be up-regulated after exposure to polycyclic aromatic hydrocarbons, 60 which could be relevant in indoor settings due to exposure to indoor combustion particles (eg, from candles or fireplaces) or alternatively due to outdoor aerosol infiltration to indoor environment. In contrast to these five genes, OGG1 was clearly down-regulated. OGG1 is associated with protection against oxidative stress, 61 acute inflammation responses, 62 and DNA repair. 63 Since PM was sampled from a non-moisture-damaged, complaint-free apartment, we did not expect to see highly toxic effects on airway constructs.

| D ISCUSS I ON
Nevertheless, PM exposure induced up-and down-regulation of several genes associated with toxicological processes.
Synthetic lung lining fluid was not acutely toxic for airway constructs, making it a good candidate for carrier buffer to be used in future exposure studies. Arguably, the cells lose their direct contact with air during the exposure experiment in our model. However, by using synthetic lung lining fluid as a carrier buffer for the PM, we can mimic the real-life situation where a film of similar fluid normally covers the cells of the airways and the secretion of the mucus is further increased during inflammation or irritation. We noted that LLF composition and subsequently its effects on airway constructs

F I G U R E 8
The number of up-and down-regulated genes (fold regulation, P < .05, fold difference ≥ 2) in different cell function categories. 52 Airway constructs exposed to 1:8 (estimated total particle mass 7.4 µg/cm 2 ) indoor PM for 24 h may vary due to challenges in standardizing the LLF preparation process. Consequently, we highly recommend testing toxicity of each batch before applying synthetic LLF on airway cells. We observed also indications that the composition and toxicity may vary due to multiple thawing cycles and extended storing even when frozen.
Without further and detailed testing, maximum of 2 months of storage in −70°C and one thawing cycle is recommended based on our observations. Due to long storing and multiple thawing cycles, LLF tends to crystallize and form separate phases, therefore becoming potentially harmful for the cells. showed that the exposed constructs react reproducibly to indoor PM, indicating that our model is useful in studying indoor air PM toxicity.
The particle count and microbial content of intermittently occupied indoor spaces is known to fluctuate. 75 collected particulate matter. However, we were able to estimate the total particle mass based on the airborne particle count and the flow rate during the sampling. Even though this provides only crude estimates of the mass concentrations, the levels (12-98 µg/mL) corresponded the doses used in earlier studies where cultured alveolar epithelial cells were exposed to urban air inhalable PM (25-300 µg/ mL). 79 Our study describes the characteristics of indoor PM from one apartment, which serves well our aim to test the usefulness of this experimental model in a real-life situation. However, the choice to analyze a pre-defined panel of genes of toxicological relevance instead of a complete transcriptome limited the scope of conclusions we were able to make based on the data. Nevertheless, we are confident that the selection of the genes in this panel represents different toxicological pathways well. Moreover, when working with a very limited number of samples, a targeted panel is likely a more sound option than collecting non-targeted screening data that increase uncertainty in data interpretation. The approach presented here will enable us to study associations between toxicological responses and microbiota characteristics in different types of indoor environments in the future. Follow-up of this work will aim at comparing pulmonary toxicity caused by indoor PM sampled from multiple non-moisture-damaged and moisture-damaged houses (REMEDIAL consortium).

| CON CLUS ION
Here, we demonstrated that human airway constructs are highly useful in studying indoor air PM toxicity, while our work acknowledges that obtaining fully differentiated airway constructs remains challenging. The studied indoor PM exposure levels, with PM collected from a non-complaint residential home, were not acutely toxic for airway constructs, but induced up-and down-regulation of several genes associated with toxicological responses. Most up-regulated genes were related to different functions of immune defense. This exposure model enables the comparison of transcriptome related to indoor air PM exposures and can be matched with microbiota determinations to study interactions with the toxicological responses.

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.

AUTH O R ' S CO NTR I B UTI O N S
Nordberg, Täubel, and Huttunen were involved in the design of the study and contributed substantially to collecting and interpreting data and writing the manuscript. Jalava helped planning and synthesizing lung lining fluid for exposure studies. Tervahauta contributed to planning the acquisition of transcriptome data. BéruBé contributed to interpreting microscopy images. Hyvärinen contributed to conception and design of the study as the leader of the REMEDIAL consortium. All authors revised and approved the final manuscript.