Changes in lung epithelial cell volatile metabolite profile induced by pro-fibrotic stimulation with TGF-β1

Volatile organic compounds (VOCs) have shown promise as potential biomarkers in idiopathic pulmonary fibrosis. Measuring VOCs in the headspace of in vitro models of lung fibrosis may offer a method of determining the origin of those detected in exhaled breath. The aim of this study was to determine the VOCs associated with two lung cell lines (A549 and MRC-5 cells) and changes associated with stimulation of cells with the pro-fibrotic cytokine, transforming growth factor (TGF)-β1. A dynamic headspace sampling method was used to sample the headspace of A549 cells and MRC-5 cells. These were compared to media control samples and to each other to identify VOCs which discriminated between cell lines. Cells were then stimulated with the TGF-β1 and samples were compared between stimulated and unstimulated cells. Samples were analysed using thermal desorption-gas chromatography-mass spectrometry and supervised analysis was performed using sparse partial least squares-discriminant analysis (sPLS-DA). Supervised analysis revealed differential VOC profiles unique to each of the cell lines and from the media control samples. Significant changes in VOC profiles were induced by stimulation of cell lines with TGF-β1. In particular, several terpenoids (isopinocarveol, sativene and 3-carene) were increased in stimulated cells compared to unstimulated cells. VOC profiles differ between lung cell lines and alter in response to pro-fibrotic stimulation. Increased abundance of terpenoids in the headspace of stimulated cells may reflect TGF-β1 cell signalling activity and metabolic reprogramming. This may offer a potential biomarker target in exhaled breath in IPF.


Introduction
Idiopathic pulmonary fibrosis (IPF) is a chronic progressive lung disease associated with significant morbidity and mortality [1]. There is an urgent need for novel biomarkers to inform both diagnosis and prognosis. Volatile organic compounds (VOCs) offer a promising source of biomarkers and have been shown to discriminate between patients with IPF and healthy controls [2,3]. Exhaled breath VOCs may originate from host metabolic processes (endogenous) or from non-host sources (exogenous), including the environment, diet and drugs, and the microbiome [4], and so identifying their source presents a significant challenge in understanding their utility as biomarkers. One method of addressing this is to measure VOCs in in vitro disease-models in which confounders can be minimised. Whilst headspace sampling has been performed in such models for lung cancer and pulmonary infection [5,6], studies assessing VOCs associated with cellular models of lung fibrosis are limited [7].
Immortalised cell lines offer a starting point for establishing consistent sampling methodology and testing variability related to controlled perturbation. Two cell lines that are frequently used in mechanistic pulmonary fibrosis research are A549 and MRC-5 cells. A549 cell line are adenocarcinoma cells that have been established as a model for type 2 alveolar epithelial cells (AEC2) in vitro [8]. AEC2s play an important role in the development of IPF through abnormal response to epithelial injury [9]. They undergo epithelial-mesenchymal transition (EMT), differentiating into a fibroblast morphology which contributes to fibroblast expansion and extracellular matrix (ECM) deposition [10]. MRC-5 cells are human lung fibroblasts derived from a 14-week foetus [11]. In IPF, fibroblasts proliferate and become chronically activated, differentiating into myofibroblasts and leading to excess ECM production [12].
Transforming growth factor-β (TGF-β) is a key cytokine in the development of pulmonary fibrosis, inducing fibroblast proliferation and differentiation, EMT, AEC2 apoptosis and inhibiting the breakdown of ECM [13]. In vitro, TGF-β1 induces EMT in A549 cells and differentiation of MRC-5 cells into myofibroblasts [14,15]. These pro-fibrotic processes are known to alter metabolic pathways [16,17], and could lead to changes in detectable VOCs produced by cells in vitro.
The aim of this study was to identify VOCs that could distinguish A549 and MRC-5 cells from culture media and each other (a cell-specific volatile signature), and identify VOCs altered by pro-fibrotic stimulation with TGF-β1.

Cell culture
A549 (from passage no. 104) and MRC-5 cells (passage no. 17) kindly donated from the Blaikley lab (University of Manchester, UK) were grown to 80% confluence before sub-seeding into 75 cm 2 (T75) polystyrene culture flasks with vented caps (Corning, Glendale, AZ, USA). A549 cells were seeded at 5 × 10 4 cells ml −1 and MRC-5 cells were seeded at 2 × 10 5 cells ml −1 . Cells were maintained in Dulbecco's modified eagle's medium (DMEM, Gibco, NY, USA) supplemented with 10% foetal bovine serum (FBS, Gibco, NY, USA). Antimicrobials were not added to the cell culture medium. This was to reduce background VOCs and the chances of lowgrade bacterial contamination being masked. Cells were maintained in an incubator at 37 • C and 5% carbon dioxide (CO 2 ). Cell media was replenished every 48 h.

TGF-β1 stimulation
Recombinant human TGF-β1 (R&D systems, MN, USA) was reconstituted as per the manufacturer's instructions. Cells were stimulated with TGF-β1 at a concentration of 5 ng ml −1 for 48 h, as per the protocol described by Kim et al [15]. Cells were transferred to serum-free media for 24 h prior to stimulation.

Headspace sampling-experimental set up
We adapted the sampling method developed by Ahmed et al [18] actively sampling headspace using a low flow pump to draw headspace gas through a sorbent-packed sampling tube (i.e. dynamic headspace collection). The sampling circuit is shown in figure 1. The culture flask cap was removed, and a polytetrafluoroethylene (PTFE) stopper introduced into the top of the flask in its place. A hollow stainlesssteel tube was passed through the centre of the stopper through which a PTFE capillary tube was fed. This was placed in the headspace above the cell culture media. The tube was connected to a stainless-steel Tconnector with the capillary tube passing through the centre and a Tenax GR stainless steel sorbent tube (Markes International, Rhondda Cynon Taff, UK) connected to the other port which acted as an ambient air VOC-filter. The capillary tube then connected via an adaptor to another Tenax GR sorbent tube which was used for sampling. PTFE tubing connected the sampling tube to a low flow sampling pump (ACTI-VOC PLUS Markes International, Rhondda Cynon Taff, UK) to draw headspace gas into the sampling tube. The headspace gas within the flask was then displaced by filtered air which was drawn through the sorbent tube connected to the T-connector. All sampling was performed in a class II biological safety cabinet and started within 60 s of the culture flasks being removed from the incubator. The sampling method was optimised to collect a total headspace volume of 500 ml using a flow rate of 100 ml min −1 . The vented culture flask caps were exchanged for nonvented caps 24 h prior to headspace sampling to preconcentrate headspace in the flask. After collection, the sorbent tubes were sealed at both ends with caps and refrigerated at 4 • C for a maximum of 48 h before analysis.

Headspace sampling-cell volatile signature
Cell culture flasks containing A549 cells (n = 15), and MRC-5 cells (n = 15) were grown over 96 h to approximately 80% confluence. Cells were sampled along with empty T75 flasks (n = 15) and flasks containing just media supplemented with 10% FBS (n = 15). All flasks were kept in the incubator for the same period. Media was replenished for all cell flasks containing media at 48 h. Headspace sampling circuit. A PTFE stopper (1) with a hollow steel tube running through is attached to a T75 cell culture flask (2). This is attached to a T-connector (3) which holds a Tenax GR tube which acts as an air inlet filter (4) and has a PTFE capillary tube running through the centre which passes through the stopper and enters the headspace. The capillary tube connects to a Tenax GR sampling tube (5). This connects via PTFE tubing to the ACTI-VOC PLUS pump (6) which is operated outside the safety cabinet. See also S1 and S2 in supplementary material. PTFE = polytetrafluoroethylene.

Headspace sampling-volatile response to TGF-β1
A549 (n = 18) cells were cultured in media supplemented with 10% FBS for 48 h before being switched to serum-free media. Twenty-four hours later, TGF-β1 (5 ng ml −1 ) was added to nine of the flasks for 48 hours. Eighteen further flasks underwent the same protocol but did not contain cells. This process was then repeated with MRC-5 cells.

Thermal desorption-gas chromatographymass spectrometry (TD-GC-MS)
Samples were analysed within 48 h of collection. Sorbent tubes were dry purged with N 2 to remove excess water at a flow rate of 50 ml min −1 for 8 min. Before primary desorption each sample tube was automatically injected with an internal standard (1 ppmv, 4-bromofluorobenzene in N 2 , Thames Restek, High Wycombe, UK). The tubes were then desorbed using a TD

Statistical analysis
Data were pre-processed using the eRah package in R (v. 4.2.1) [19]. This consisted of peak picking, alignment, and integration. This produced a sample x feature (n x p) data matrix for further processing. Missing data were imputed using one-fifth the lowest recorded value for each feature. Data were logtransformed and normalised to the internal standard abundance. A data matrix of identified features listed by retention time was produced with normalised peak intensity recorded for each sample. Both univariate and multivariate analysis was performed independently to identify differences in detected features between test groups. Univariate statistical analysis was performed using independent t-test to compare test groups and provide an overview of the number of significant features between groups. Correction for false discovery was performed using the Benjamini-Hochberg procedure [20]. A p-value of less than 0.05 was considered statistically significant. Multivariate analysis using principal component analysis (PCA) and a supervised method, sparse partial least-squares discriminant analysis (sPLS-DA), was performed to compare experimental groups and identify discriminant features. The sPLS-DA model parameters were tuned using five-fold cross validation. The top discriminant features between groups were identified using the variable importance of projection (VIP) score. Comparisons were made between A549, MRC-5 cells and media-only samples, as well as between stimulated and unstimulated cells in each cell line.
Statistical analysis was performed using MetaboAnalyst 4.0 [21]. Figures and individual feature analysis was performed using GraphPad Prism (GraphPad Software, version 8, San Diego, CA, USA). VOCs were identified from the top discriminant features using the Masshunter Qualitative Analysis software package (Agilent Technologies, Cheadle, UK) and the NIST mass spectral library (v.2014) and calculated retention indices at MSI level 2 in accordance with Metabolomics Standards Initiative (MSI) minimum reporting standards [22].

Cell volatile signature
In total, 929 mass spectral features were detected; 64 features were significantly different between A549 cells and media-only samples (27 higher in A459 cells) and 48 features were different between MRC-5 cells and media, (18 higher in MRC-5 cells). Nineteen features were different between the two cell lines (10 higher in A549 cells). All detected features were included in multivariate analysis. Unsupervised multivariate analysis using PCA suggested a strong effect of the day of sampling but not separation of experimental groups (figure S3, supplementary material). Supervised analysis using sPLS-DA was subsequently performed and demonstrated discrimination between groups as illustrated in figure 2. Table 1 lists the putative identities of the top VOCs which differentiated groups from each other ranked by VIP score. Identification was made to MSI level 2 where possible [22]. Scatter plots for these discriminant VOCs are shown in figure 3.

Volatile response to TGF-β1
Twenty-five feature abundances were different between A549 cells stimulated with TGF-β1 and unstimulated cells, 16

Discussion
We have reported VOCs associated with the basal activity of two lung cell lines and in response to stimulation with TGF-β1, a key mediator in lung fibrosis [13]. TGF-β alters metabolic pathways in both AEC2 and lung fibroblasts [23]. This is partly mediated through the activity of the enzyme prolyl 4-hydroxylase subunit alpha 3 (P4HA3), which is essential for proline metabolism and collagen synthesis, a hallmark of fibrosis [24]. TGF-β upregulates P4HA3, inducing EMT in A549 cells and stimulating collagen production in fibroblasts [17,25]. TGF-β also stimulates glycolysis in both A549 and MRC-5 cells [26,27]. Glutaminolysis is increased in fibroblasts on stimulation with TGF-β while oxidative phosphorylation is reduced [28,29]. These metabolic changes may be responsible for changes in VOCs in cells lines upon stimulation with TGF-β1.
3,3-dimethyl-2-pentanol, a secondary alcohol, was increased in A549 cells following stimulation which may reflect increased alcohol dehydrogenase activity as a result of glycolysis and proline metabolism. Likewise, ethyl 2-methylbutyrate, a fatty acid ester, was decreased in MRC-5 cells following stimulation, perhaps reflecting reduced fatty acid oxidation as glycolysis is increased. Dodecanal, another by-product of lipid metabolism, was also lower in MRC-5 cells following stimulation. There appeared to be a general reduction in VOC abundance in MRC-5 cells upon stimulation with TGF-β1, potentially reflecting cell differentiation and consumption of metabolites. A similar trend towards reduced metabolite abundance has been noted in MRC-5 cells exposed to silica nanoparticles [30].
Release of terpenoids from cell lines was consistently demonstrated following profibrotic cytokine stimulation. Isopinocarveol produced through the monoterpenoid biosynthesis pathway [31], was higher in A549 cells stimulated with TGF-β1 compared to unstimulated cells; sativene, a sesquiterpenoid, was elevated in stimulated A549 cells and 3-carene, a monoterpene, was higher in stimulated MRC-5 cells. Increased levels of terpenoids may reflect TGF-β-induced metabolic reprogramming in EMT leading to increased glycolysis and terpenoid production [27]. There is evidence that terpenoids alter TGF-β activity through Smad signalling [32], a key pathway in development of fibrosis [33]. Limonene may have use as a clinical biomarker of fibrosis as it has been found to be elevated in the breath of patients with liver cirrhosis [34]. There is also interest in the therapeutic effects of terpenoids and they may have anti-inflammatory and antifibrotic properties [35]. The VOC profile of basal lung cell activity is yet to be fully determined. A549 cells are the most commonly utilised cell line in headspace studies and some have reported VOCs produced by these cells compared to cell-free media [36][37][38][39][40][41][42]. However, few studies have described the volatile profile of MRC-5 cells [43][44][45]. We found that nonanal and benzaldehyde were dominant discriminant VOCs with both being reduced in cells lines compared to mediaonly samples. Both have been commonly identified in A549 and MRC-5 headspace sampling previously [36-38, 40, 43, 44]. Our results support previous findings that both VOCs are lower in cell lines compared to media [37,38,40,42], although some studies have noted elevated levels [36,38]. Both nonanal and benzaldehyde are aldehydes and reduced levels may reflect cellular aldehyde dehydrogenase activity [46]. However benzaldehyde is also a known artefact from sorbent tubes and its presence in headspace samples may be a contaminant. This could explain why higher levels of benzaldehyde were seen in unstimulated A549 cells in the fibrotic-signature experiments in contrast to reduced levels seen in cell-signature experiments. 2-ethylhexanol is another VOC which is consistently reported in A549 and MRC-5 headspace studies [37][38][39][40]45]. We found that it was higher in cell lines compared to media, consistent with previous reports [37,38]. Several VOCs differentiated A549 cells from both MRC-5 cells and media-only samples. This included ethyl acetate which was higher in A549   cells compared to MRC-5. This has previously been found to be elevated in A549 cells compared to media [39]. Acetophenone, o-xylene and an unknown feature (VOC 610) were all reduced in A549 cells in comparison to MRC-5 and media-only samples. O-xylene has been previously reported to be lower in A549 cell cultures in comparison to media, consistent with our results [37,39]. Ethyl 2-methylbutyrate was higher in MRC-5 cells in comparison to both A549 cells and media-only samples. Ethyl 2-methylbutyrate is a fatty acid ester, compounds which have been implicated in the activation and proliferation of lung fibroblasts in vitro [47]. There are several limitations to this study. Due to the nature of cell culture work, there is a high risk of contamination of VOCs from exogenous sources. This was apparent in PCA analysis which was confounded by date of sampling. Plasticware used to grow cells produces a high number of VOCs, particularly derivatives of benzene, such as styrene, toluene and ethylbenzene [39,48,49]. Glassware represents an alternative for use in headspace experiments that is inert and does not produce VOCs [49]. However, cell growth tends to be slower on glassware in comparison to plastic culture flasks [50]. We capped cells with a non-vented cap for 24 h prior to sampling as there was increased VOC yield. This may have led to an accumulation of exogenous contaminant VOCs as well as biologically relevant VOCs. In addition, the caps used in cell culture sampling are known to be a high producer of contaminant VOCs [48]. We performed active sampling outside of the incubator which meant that humidity, temperature, and CO 2 levels could not be regulated. These changes can alter VOC composition [51]. The other drawback of our method is the risk of bacterial contamination through sampling, meaning that samples can only be taken at a single timepoint. This means that repeated measurement of VOCs over a period of cell growth cannot be attempted. However, to assess the impact of sampling on bacterial contamination, we returned several flasks back to the incubator after sampling and did not witness any subsequent contamination. We did not include any objective measurements of cellular change to confirm the pro-fibrotic response to TGF-β1. This may have been useful to allow correlation with VOC abundance. Finally, all VOC identification was made at MSI level 2 and identification is putative. Targeted analysis supported by chemical standards would improve identification confidence. The addition of online VOC analysis, for example using proton-transferreaction mass spectrometry, would also be beneficial in reducing the impact of VOC degradation and contamination associated with offline techniques such as GC-MS. Likewise the integration of additional datasets, such as proteomics and transcriptomics, would also improve confidence in the biological relevance of terpenoids and other identified volatiles.  Media-only and media + TGF-β1 samples also shown. Mean and 95% confidence interval lines shown. Significant differences between samples highlighted. * p < 0.05; * * p < 0.001; * * * p < 0.0001. VOC = volatile organic compound; TGF = Transforming growth factor.

Conclusion
We identified VOCs in the headspace of two lung cell lines used to model pulmonary fibrosis: AECs (A549) and human lung fibroblasts (MRC-5). We found VOCs which discriminated cells from culture media and from each other. The profile of VOCs changed in cells in response to stimulation with the pro-fibrotic cytokine TGF-β1. This included an increase in several terpenoids which may reflect an association with TGF-β1 cell signalling.

Data availability statement
The data cannot be made publicly available upon publication because they are not available in a format that is sufficiently accessible or reusable by other researchers. The data that support the findings of this study are available upon reasonable request from the authors.

Funding
This study was supported by a grant from the North West Lung Centre Charity. W A and S J F are supported by the NIHR-Manchester Biomedical Research Centre. JFB is a MRC transition support fellow (MR/T032529/1) and is supported by grants from Asthma+Lung UK and also the NIHR-Manchester Biomedical Research Centre.
The authors have no relevant financial disclosures.