Redox Processes Inform Multivariate Transdifferentiation Trajectories Associated with TGFβ-Induced Epithelial-Mesenchymal Transition

Abstract

Phenotype reprogramming during transforming growth factor β (TGFβ)-induced epithelial-mesenchymal transition (EMT) is an extensive and dynamic process, orchestrated by the integration of biological signaling across multiple timescales. As part of the numerous transcriptional changes necessary for EMT, TGFβ-initiated Smad3 signaling results in remodeling of the redox environment and decreased nucleophilic tone. Because Smad3 itself is susceptible to attenuated activity through antioxidants, the possibility of a positive feedback loop exists, albeit the time scales on which these mechanisms operate are quite different. We hypothesized that the decreased nucleophilic tone acquired during EMT promotes Smad3 signaling, enhancing acquisition and stabilization of the mesenchymal phenotype. Previous findings supporting such a mechanism were characterized independent of each other; we sought to investigate these relationships within a singular experimental context. In this study, we characterized multivariate representations of phenotype as they evolved over time, specifically measuring expression of epithelial/mesenchymal differentiation, redox regulators, and Smad transcription factors. In-cell western (ICW) assays were developed to evaluate multivariate phenotype states as they developed during EMT. Principal component analysis (PCA) extracted anti-correlations between phospho-Smad3 (pSmad3) and Smad2/Smad4, which reflected a compensatory up-regulation of Smad2 and Smad4 following cessation of TGFβ signaling. Measuring transcript expression following EMT, we identified down-regulation of numerous antioxidant genes concomitant with up-regulation of NADPH oxidase 4 (NOX4) and multiple mesenchymal phenotype markers. TGFβ treatment increased CM-H₂DCF-DA oxidation, decreased H₂O₂ degradation rates, and increased glutathione redox potential. Our findings suggest that the decreased nucleophilic tone during EMT coincides with the acquisition of a mesenchymal phenotype over too long a time scale to enable enhanced Smad3 phosphorylation during initiation of EMT. We further challenged the mesenchymal phenotype following EMT through antioxidant and TGFβ inhibitor treatments, which failed to induce a mesenchymal-epithelial transition (MET). Our characterization of multivariate phenotype dynamics during EMT indicates that the decrease in nucleophilic tone occurs alongside EMT, however maintenance of the mesenchymal phenotype following EMT is independent of the both the nascent redox state and continuous TGFβ signaling.

Introduction

More than seven decades since Waddington first introduced the concept of canalization, systems biology is still very focused on how to best characterize phenotype robustness, in which small perturbations from a developmental trajectory are protected against by a steep energetic landscape of descent towards a terminal state [1]. In the modern era, however, biological insight into the reversibility of differentiation processes through transdifferentiation and the reacquisition of pluripotency have expanded our view of how cells behave dynamically as they progress from one phenotype to another. The ability to measure many biomarkers in combination allows contemporary researchers to interrogate how cellular trajectories during transdifferentiation are driven or reinforced by various cellular programs. In particular, the influences of genetic reprogramming of a cellular state, such as redox potential, and external influences on this state can be investigated in parallel with traditional phenotype markers to ask how cellular oxidation influences the progression from one cell type towards another.

One such transdifferentiation process, epithelial-mesenchymal transition (EMT), occurs when epithelial cells loose certain phenotypic qualities (such as apical-basal polarity and basement membrane interaction) and acquire mesenchymal characteristics (such as cell migration and production of extracellular matrix components) [2]. The cytokine transforming growth factor beta (TGFβ) induces EMT and has been implicated in increased invasiveness of cancers and in the formation of metastases [3–5]. TGFβ signaling is enhanced within the tumor microenvironment through interactions with cancer-associated fibroblasts, immune cells, and extracellular matrix [6, 7]. Circulating tumor cells from breast cancer patients are enriched for mesenchymal markers and TGFβ signaling, implicating TGFβ-mediated EMT as a mechanism for entry into to the circulatory system [8]. Upon reentry into distant tissues, reversal of the mesenchymal-like phenotype into an epithelial phenotype (i.e. MET), may play a critical role in the establishment and progression of metastases [9]. A recent investigation of ovarian cancer, stratification of tumor phenotypes into epithelial, intermediate epithelial, intermediate mesenchymal, and mesenchymal states found increased 5-year progression-free survival in patients whose tumors were classified as epithelial or intermediate epithelial. Furthermore, of 43 classified cell lines, intermediate states were found to be more responsive to kinase inhibition [10]; thus, the timing of phenotype transition dynamics or characterization of intermediate phenotype states may have major implications for therapeutic strategies.

Numerous gene expression changes are initiated by TGFβ signaling as part of EMT. Upon binding of its cognate receptor, TGFβ quickly activates canonical and non-canonical signaling. Canonical signaling occurs through phosphorylation of Smad2 and Smad3 transcription factors, which bind Smad4 before translocation into the nucleus [11, 12]. Smad3 phosphorylation and its transcriptional activity are critical steps in TGFβ-mediated EMT [13, 14]. Smads have been linked to a number of critical steps involved in the formation of metastases. In one study, the formation of bone metastases by xenografted cancer cells relied on Smad4 [15]. In another, in vitro and in vivo metastatic processes were dependent on Smad3 and enhanced upon Smad2 knockdown [16], while, in yet another study, Smad2 elevation enhanced in vitro and in vivo pro-metastatic processes [17]. Thus, the pro-carcinogenic mechanisms of TGFβ rely on Smad signaling to carryout transcriptional remodeling, though in ways that may be cancer or cell-type specific.

In addition to induction of EMT, TGFβ can transform the regulation of the intracellular redox environment through a variety of mechanisms, such as the up-regulation of NADPH oxidase 4 (NOX4), which constitutively produces hydrogen peroxide (H₂O₂) [18, 19], increasing free intracellular iron, and the down-regulation of glutaredoxin-1 or reduced glutathione (GSH) levels [20–26]. The state of reduction equivalent capacity, capable of eliminating electrophiles such as ROS, has been defined as the nucleophilic tone of the cell and is determined by the expression of antioxidant system components. [27]. As such, decreased nucleophilic tone would impair the cell’s ability to clear ROS. Elevation of reactive oxygen species (ROS) during EMT can lead to direct activation or enhancement of a variety of redox sensitive signal transduction pathways [28] and H₂O₂ treatment itself has been shown to induce EMT in a TGFβ-dependent manner [29]. Antioxidant attenuation of Smad2/3 phosphorylation and transcription has been observed in a variety of cell types [30–34] and is attributed to the prevention of TGFβ-mediated EMT [23, 26, 31, 35]. Thus TGFβ signaling has both the capacity to modify the redox environment and also be subject itself to regulation by the redox environment.

Despite extensive studies devoted to the contribution of cellular oxidation on many individual biochemical processes involved in TGFβ-induced EMT, inclusion of redox markers in the characterization of the multivariate phenotype trajectory has never been performed in a systematic manner. We hypothesized that the previously reported cellular oxidation during TGFβ-mediated EMT [21, 23] reinforce TGFβ signaling in a feed-forward manner during EMT as well as contribute to maintenance of mesenchymal differentiation following EMT. Investigation of the aforementioned processes is fraught with the complexity arising from studying a transition that involves evaluating epithelial, mesenchymal, Smad signaling, and redox regulatory phenotypic characteristics as they evolve over time. To address the numerous interconnected regulators in this biological system, we developed a custom panel of markers for a multiwell in-cell western (ICW) assay [36] that could generate time-dependent samples for numerous epithelial, mesenchymal, TGFβ-specific, and redox markers. This data was compiled with other available information for multivariate modeling, specifically principal component analysis (PCA), to collapse features of high-dimension, temporal dynamics during EMT into latent variable space. This novel experimental and analytical approach allowed us to investigate whether cellular redox features are informative metrics of EMT transdifferentiation in A549 lung carcinoma cells. Using PCA we extracted a multivariate description of phenotype during the time course of EMT that was capable of interrogating how the cellular redox state may influence and relate to transdifferentiation between epithelial and mesenchymal states.

Materials and Methods

Cell Culture & Treatment Conditions

A549 lung carcinoma cells were obtained from American Type Culture Collection (ATCC; CCL-185) and maintained in high glucose DMEM with L-glutamine (Sigma D5796), 10% FBS (Sigma F4135) and penicillin (50 IU/ml)-streptomycin (50 µg/ml) (Cellgro 30-001-CI). Cells were plated in 96-well plates at density of 5,000 cells per well in growth media and maintained at 37°C and supplemented with 5% CO₂. The following day, the cells were serum starved with reduced serum (0.5% FBS) media for 24 hours prior to treatment. Cells were maintained and treated in 175 µl media per well. Cells were treated with a bolus of 200 pM TGFβ for the EMT time course, qRT-PCR, CM-H₂DCF-DA, and H₂O₂ degradation studies. For the GSH study, cells were seeded in T-75 flasks at a density of 5,000 cells/cm² and maintained in 15 ml of culture media. Culture media for the EMT intervention study was changed daily and consisted of either plain media, 100 pM TGFβ (Millipore, GF111), 10 µg/ml neutralizing anti-TGFβ antibody (R&D Systems, MAB240), 2 µM A8301 (Santa Cruz, sc-203791; mobilized in DMSO), 2 mM N-Acetyl-L-cysteine (NAC; Sigma-Aldrich, A9165), 2 µM ebselen (Alfa Aesar, J63190; mobilized in DMSO), or 0.2% v/v DMSO (Fisher Scientific, BP231). All experiments were the result of three independent biological replicate experiments.

In-Cell Western (ICW) Assay

Following treatment, cells were washed with PBS with Ca²⁺/Mg²⁺ and fixed with 100 µl 4% paraformaldehyde per well for 20 minutes at ambient temperature. The cells were permeabilized by washing five times with 50 µl 0.1% Triton X-100 solution for 5 min with gentle rotation at ambient temperature. The plates were blocked with 100 µl of blocking buffer consisting of 0.5x Rockland Blocking Buffer for Fluorescent Western Blotting (MB-070) in tris-buffered saline (TBS) for 1.5 hours with gentle rocking at ambient temperature. Following blocking, cells were immunostained with 35 µl of 1° antibody solutions (Table S3) overnight at +4°C with gentle rotation in blocking buffer supplemented with 0.1% Tween-20. The plates were then washed five times with TBS-T (TBS with 0.1% Tween-20) under gentle rotation at ambient temperature for 5 minutes each. Plates were stained with 45 µl Donkey anti-Rabbit antibody (1:800; LiCor, IRDye 800CW, 926–32213) and CellTag 700 (0.2 µM; LiCor, 926–41090) in blocking buffer supplemented with 0.2% Tween-20 for 1.5 hours. The plates were washed four times with TBS-T and once with TBS before being emptied and sealed for imaging. Signal intensities in the 700 nm and 800 nm channels were measured on stained plates via LiCor Odyssey system and analyzed in LiCor Image Studio (v2.1.10). For a given well, non-specific secondary antibody background staining (800_background) was subtracted from the raw 800 channel intensity (800_raw) to yield the 800 channel signal (800_signal) intensity. Please see the Supplementary Information for further details regarding the background subtraction methods. The loading control normalized signal intensity was determined by dividing 800_signal by the 700 channel signal intensity. To normalize individual signals across plates, the loading control normalized signal intensity was divided by the average loading control normalized signal intensity of the control condition from all of the plates (untreated A549 cells).

Quantitative Real-Time PCR

Cells were seeded in T-175 flasks at a density of 5,000 per cm² and serum starved (0.5% FBS) the next day. On the following day, cells were treated with 200 pM TGFβ for 48 hours. Cells were trypsinized and 10⁶ cells lysed and homogenized (QIAshredder, Qiagen, 79656). Next, RNA was isolated via RNeasy Mini kit (Qiagen, 74104), genomic DNA digested (Qiagen, RNase-free DNase Set, 79254), and the RNA concentration determined by NanoDrop. Isolated RNA (2 µg) was converted to cDNA (Qiagen, RT² First Strand Kit, 330401), prepared for amplification (Qiagen, RT² SYBR Green ROX qPCR Mastermix, 330522), loaded onto the Oxidative Stress Plus PCR Array (Qiagen, PAHS-065YC; primers detailed in Table S4), and amplified via thermocycling (Applied Biosystems, Step One Plus Real-Time PCR System; 40 cycles). Genes in the array were categorized as “antioxidant”, “pro-oxidant”, “undetermined significance”, or as a housekeeping gene upon review of the particular entry in the NCBI Gene Database for each gene. Following amplification, samples were analyzed as instructed by the Qiagen product materials, with the exception of using ANOVA with Fisher’s Least Squares Difference in place of the Student’s T-test to determine significance of up/down-regulation of transcripts following TGFβ treatment. P values < 0.0001 were arbitrarily set to 0.0001. Transcript Ct values were normalized to loading controls ACTB, GAPDH, HPRT1, B2M, and RPLP0 to obtain ΔCt values. Statistical analyses were performed on the ΔCt values. Next ΔCt values from TGFβ-treated samples were subtracted from untreated controls to obtain ΔΔCt values. ΔΔCt values were used to compute fold-change values (2^−ΔΔCt).

Multivariate Analysis

Umetrics SIMCA-P+ (v12.0.1.0) was used to perform principal component analysis (PCA). Fold-change ICW data was log-transformed and unit variance scaled for modeling purposes. Transcript expression of A549 cells during TGFβ treatment, over the course of 72 hours, was previously characterized by Keshamouni et al. using Affymetrix HG-U133_plus_2 microarrays [37]. Data were obtained from the NCBI Gene Expression Omnibus entry GSE17708, which were uploaded following the work by Sartor et al. [38]. A list of genes included in the analysis can be found in Table S1. Many of the transcripts were measured with multiple primers. When multiple primers were present, the average of primers was used to create a single averaged response for each variable at each time point. Transcript expression for each variable corresponds to log-transformed fold-change. Uninformative probes were removed to improve model fit. Data for the transcript-exclusive model were Pareto scaled while data for aggregated models were unit variance scaled. Pareto scaling improves the fit by scaling variable variance according to its standard deviation such that experimental noise is minimized and the model structure is more preserved and is suitable for quantitative data, such as microarray analyses [39, 40]. The significance of principal components was confirmed using cross-validation rules in SIMCA-P+. Details of fit, quality, and construction for each model can be found in the Supplementary Information.

CM-H₂DCF-DA Fluorescence

Following treatment, cells were trypsinized (CellGro, 25–053-CI) and resuspended in HBSS without phenol red (Thermo Scientific HyClone, SH30268) at a working concentration of 5×10⁵ cells/ml. Antioxidant pre-treatments of NAC or bovine catalase (Sigma-Aldrich, C1345) were applied for 1 hour prior to a 30 minute incubation with 10 µM CM-H₂DCF-DA (Invitrogen, C6827). Co-treatment with H₂O₂ (100 µM) occurred during the final 15 minutes of the CM-H₂DCF-DA incubation. The cells were then washed and co-stained with SYTOX Blue (Invitrogen, S34857, 1:1000) for live/dead discrimination. A BD LSR II flow cytometer was used to resolve SYTOX Blue (λ_ex=445, λ_em=473/10) and CM-H₂DCF-DA (λ_ex=488, λ_em=530/30) signals. Cells were gated by FSC/SSC to exclude cellular debris, then by FSC-A/FSC-H to exclude non-singular events, and finally by absence of SYTOX Blue staining, to exclude all non-viable cells. CM-H₂DCF-DA signals were then characterized by their geometric mean fluorescence intensity.

The CM-H₂DCF-DA fluorescence time course studies were performed in 96-well plates with 6 technical replicates per time point. Following TGFβ treatment, the plates were washed with Hank’s balanced salt solution with calcium and magnesium (HBSS) and incubated with 10 µM CM-H₂DCF-DA in 75 µl HBSS for 30 min at 37°C with 5% CO₂. The cells were then washed with 150 µl HBSS and suspended in 75 µl HBSS for fluorescence measurement on a BioTek Synergy 4 plate reader (λ_ex = 485/20 nm excitation, λ_em = 528/20 nm). Next, the cells were stained with 5 µg/ml Hoechst 33342 (AnaSpec, 83218) in 50 µl HBSS for 30 min at 37°C with 5% CO₂ (λ_ex = 350 nm, λ_em = 461 nm). Unstained wells were used to subtract background fluorescence from both signals. The CM-H₂DCF-DA fluorescence intensity was then normalized with respect to Hoechst 33342 staining for each well and normalized to the untreated control, giving the loading normalized CM-H₂DCF-DA signal.

Luminol Assay for Hydrogen Peroxide

Cellular H₂O₂ degradation rates were measured during the course of EMT using a luminol-based assay based on a method for determining the first-order kinetics of H₂O₂ turnover. A condition-specific rate constant was determined through sampling the supernatant for H₂O₂ at sequential time points following bolus addition of H₂O₂ [41]. Cells were treated with TGFβ for 1, 2, or 3 days in a 96 well plate. Following treatment, the cells were incubated with 20 µM H₂O₂ in HBSS for multiple time points up to 60 minutes. H₂O₂ present at the end of the time course was measured with a luminol (50 µM; Alfa Aesar, 3-Aminophthalhydrazide monosodium salt, L15205)/sodium hypochlorite (1 mM; Sigma-Aldrich, 239305)-based assay and compared against a titrated H₂O₂ standard. Cell densities were normalized with Hoechst 33342 staining. H₂O₂ degradation rates (k_deg) were calculated by fitting the H₂O₂ concentrations, incubation time, and cell density data to a curve of exponential decay derived from a model of H₂O₂ decay. The k_deg values from each assay were normalized to the untreated condition. More detailed information regarding the luminol assay can be found in the Supplementary Information.

Glutathione Concentration & Redox Potential Measurement

Cells were treated with 200 pM TGFβ in the presence of 10% FBS for 48 and 96 hours. Following treatment, cells were trypsinized and analyzed via Beckman Coulter Vi-Cell XR to measure cell density, viability, circularity, and diameter. The cells were transferred to a 96-well plate to measure the total GSH concentration (20,000 cells/well) and GSSG concentration (40,000 cells/well) compared to a standard curve using the Promega GSH/GSSG-Glo Assay (Cat. #V6611) with 3 technical replicates per condition. Cell shape was approximately spherical following trypsinization and spherical cell volumes calculated from diameters. Intracellular reduced GSH and GSSG levels were calculated from the GSH/GSSG-Glo Assay conditions. The redox potential was calculated according to the GSSG/2GSH half-cell reduction potential (E_GSH) with the electrochemical constants presented by Schafer and Buettner (ΔE°=−240 mV, pH 7.0, n=2, F=9.6485×10⁴ C mol⁻¹, R=8.314 J K⁻¹ mol⁻¹, T=310 K) [42].

Statistical Analysis

Statistical analyses of ICW, qRT-PCR, CM-H₂DCF-DA, H₂O₂ degradation, and GSH data sets were performed using GraphPad Prism version 6.0e, Mac OS X. Statistics were performed on data following normalization to loading controls. For graphical display and interpretation, data were normalized with respect to the untreated control condition.

Results

Highly reproducible EMT phenotype dynamics allows for visualization of transdifferentiation trajectories

Response of A549 cells to TGFβ treatment has been studied extensively; within 12 hours of treatment more than 2000 genes exhibit differential expression [43] while down-regulation of E-cadherin and up-regulation of mesenchymal proteins is apparent by 24 hours [44, 45]. To characterize multivariate phenotype states during the time course of EMT, we measured the expression of 8 proteins, spanning epithelial, mesenchymal, and Smad species, following a 200 pM bolus addition of TGFβ. In-cell western (ICW) assays [36] produce similar data as conventional western blotting but are more proficient in several ways, including increased precision and higher throughput [46].

Similar to western blotting data (Figure S1A–D), we observed decreased E-cadherin and increased aSMA, vimentin, and β-catenin expression (Figure 1A). However, the ICW assay yielded high precision results, reflected in the low experimental error. Individual Smad species displayed distinct expression profiles during treatment (Figure 1B). Smad2 expression remained relatively constant up to 72 hours before sharply increasing. Smad3 displayed a biphasic response, initially increasing 1.4-fold before decreasing 12 hours post-treatment. Smad4 levels were suppressed from 30 minutes through 8 hours before increasing expression at later times. Phosphorylation of Smad3 occurred within minutes of addition of TGFβ and remained sustained up to a 12-hour window (Figure 1C). Multivariate analysis enables the visualization of covariance among multiple variables, such as phenotypic markers [47]. We hypothesized that the aspects of differential dynamics displayed by Smad2 and Smad3 would be discernable using a multivariate modeling approach to characterize cellular phenotype during EMT. We sought to demonstrate the utility of pairing ICW data collection with PCA for multivariate characterization and analysis of phenotype dynamics during EMT. A major strength of PCA is that it simplifies large data sets by distilling high-dimensional data into principal components (PCs), which are latent variables, composed of a linear combination of response variables. The output of the model is a set of reduced-dimension PCs, which are composites of weighted response variables that exhibit similar “behavior”, as well as a mapping of model observations according to how they align, or project, along the nascent PCs.

Figure 1. Multivariate phenotype dynamics during TGFβ-mediated EMT.

[A-C] ICW analysis of protein expression over time following 200 pM TGFβ treatment in A549 cells. A) Down-regulation of E-Cadherin is apparent at later time points, as is the up-regulation of mesenchymal markers aSMA, vimentin, and β-catenin. B) The dynamic profile of each Smad transcription factor is unique. C) Smad3 is quickly phosphorylated upon the addition of TGFβ and remains elevated until the later time points. Normalized fluorescence values were analyzed using two-way ANOVA (p=0.05) with Dunnett’s multiple comparisons test. The values that differ significantly from the zero time point are enclosed within the shaded regions. Data are the result of 3 independent biological replicates (n=3) and were plotted as mean fold-change ± standard error of the mean. [D-F] Dimensional reduction through PCA modeling of ICW data resolves the relationships between response variables and the EMT time course. A possible 9 variables (8 proteins + time) are distilled into 2 PCs, or latent variables, composed of protein expression variables. D) Plots of the observation scores, from three technical replicates, trace a trajectory of phenotypic variation in multivariate space during EMT. The time points are resolved along a counter-clockwise path. E) Variable contributions to the formation of PCs are depicted by their loading weight along the PC axes, i.e. the coordinates (p1, p2) where p1 is the weight along PC1. Co-localized variables are more covariant while variables bisected by 0 along a PC are anticorrelated with respect to that PC. Some variables exhibit simple anticorrelation, E-Cadherin with Smad4, while others are contextually anticorrelated, e.g. β-catenin with Smad3 along PC1 but not PC2. F) The biplot contains both scores and loading data, scaled by their correlation along the PCs, enabling the analysis of relationships between response dynamics and the EMT time course. 16 and 96 hour samples are differentiated by a down-regulation of Smad3 and pSmad3 paired with up-regulation of Smad2 and Smad4 over time.

In the Scores plot (Figure 1D), which projects observations in latent variable space, we observed a clearly demarcated progression of time points starting in the lower right-hand quadrant and running counter clockwise into the lower left-hand quadrant. The counterclockwise trajectory of observations in the Scores plot moves along PC2 (up) before moving along both PC2 (down) and PC1 (left) at later time points (24+ hours); therefore we interpret PC1 as differentiating later time points from early or intermediate ones.

Contributions of each protein to the formation of PCs and their relationships to each other were characterized in the Loading plot (Figure 1E). E-cadherin had a relatively large weighting on PC1 but essentially none along PC2, indicating that E-cadherin contributed to the overall data variance captured by PC1 but very little by PC2. Smad4 loading, opposite E-cadherin, indicated an anticorrelated response profile to E-cadherin. Figure 1A–B suggests both E-cadherin and Smad4 were largely constant during the early time points before diverging at later time points. In this sense, their responses “behaved” similarly, contributing to the same PC, though with opposite signed weighting to account for their opposing nature (Figure 1E).

The Biplot (Figure 1F) superimposes observation data from the Scores plot and variable data from the Loading plot scaled by their correlation along the PCs. This serves as a tool to associate the relationships between variables and observations. Since all of the response variables, according to loading weights, contributed to the PCs, phenotype trajectories represent dynamic responses of numerous proteins. Close proximity of variables to observations indicates that high expression of the variables aids in differentiating those observations from the others. EMT was apparent by the anticorrelation between E-cadherin and mesenchymal markers while the anticorrelation between Smad3/pSmad3 and Smad2/Smad4 was aligned differently within latent variable space, capturing the difference in regulatory time scales. In this manner, the time course of EMT is resolved, according to multivariate phenotypic profiles, through superimposition of variables and time points within latent variable space.

Reciprocal regulation of antioxidants & NOX4 during EMT

Numerous studies have examined TGFβ-induced perturbation of isolated components involved in redox regulation, yet these studies neglect the complexity of the redox environment, which encompasses numerous reactive species and is subject to regulation by a variety of production and clearance mechanisms. The collective response of redox regulators has not been characterized. Using qRT-PCR, we examined an extensive array of antioxidant and pro-oxidant modulators of the redox environment following 48 hours of TGFβ treatment. The response of antioxidant and pro-oxidant genes were not uniform, with some increasing and some decreasing in each category (Figure 2). Numerous antioxidant enzymes, representing distinct antioxidant mechanisms, were down-regulated, while NOX4, was up-regulated by 50-fold.

Figure 2. Volcano plot of antioxidant and pro-oxidant gene modulation following TGFβ treatment.

A549 cells were treated with 200 pM TGFβ for 48 hours, after which the expression of antioxidant and pro-oxidant genes were compared against untreated controls via qRT-PCR. Genes demonstrating a greater than 2-fold change in expression and a p value less than 0.05 are labeled with the gene name. A comprehensive overview of the PCR array results are depicted in Figure S3. P values were determined by two-way ANOVA with multiple comparisons and the fold-change plotted as the mean ± standard error of the mean. Data are the result of 3 independent biological replicates (n=3).

Anticorrelation of antioxidants with mesenchymal phenotype markers during EMT

The data in Figure 2 represents a single point in time and the temporal relationship of the enzyme responses with respect to one another and to EMT are unclear. For example, the data does not distinguish between coordinated regulation, which would enable relative relationships in redox compartmentalization to be maintained, versus staggered dynamics, which would focus the pro-oxidant shift to particular redox couples or cellular compartments. In a previous study of TGFβ-mediated EMT in A549 cells by Keshamouni et al., Affymetrix microarrays and quantitative mass spectrometry were used to identify a very high degree of concordance between transcript and protein expression over a 3-day time course [48]. Using this dataset, we selected transcripts that were known to be differentiation markers, previously studied in TGFβ-induced EMT of A549 cells, or investigated elsewhere in this study. The changes in gene expression measured by PCR (Figure 2) were found to be correlated with those measured by Keshamouni et al. (Figure S5). Microarray transcript expression was analyzed by PCA to characterize the transcription dynamics of epithelial, mesenchymal, Smad, and redox species.

Similar to the ICW plot in Figure 1D, the Scores plot of the microarray PCA yielded a rotational phenotype trajectory through the four quadrants (Figure 3A). The variable relationships observed are evenly distributed across latent variable space (Figure 3B); however, PC1 captures the vast majority of model variance and is responsible for differentiating early from late observations as they progress from right to left within latent variable space while PC2 differentiates early/late from intermediate time points (Figure 3C). The combination of PC1 and PC2 enables resolution of the full time course. Closer inspection of the Loadings plot revealed both expected and unexpected results. High antioxidant expression at early time points (Figure 3B; purple diamonds) was anticorrelated with high mesenchymal marker expression at late time points (red triangles). NOX4 expression correlated with mesenchymal markers. Surprisingly epithelial markers (blue circles) did not covary during EMT; however, E-cadherin (CDH1) did covary with antioxidants and anticorrelate with mesenchymal markers. Notably, antioxidants and mesenchymal markers were distributed along both sides of PC2.

Figure 3. Antioxidant/mesenchymal anticorrelation during EMT.

PCA modeling of transcript expression during the time course of EMT, from the study by Keshamouni et al. [48], displays the transition of phenotype states over time. A) Observation scores trajectories from three technical replicates progress in a clock-wise manner from 0 to 72 hours of treatment. B) The model was populated by variables commonly used to define epithelial or mesenchymal differentiation as well as those pertinent to redox regulation and Smad transcription factors. The anticorrelation of antioxidants (purple diamonds) with mesenchymal markers (red triangles) is prominent along PC1. C) The biplot, incorporating information from both the scores (3A) and loading (3B) plots, relating the variables that define the phenotype with the time course on which they change. The path of the phenotype trajectory, predominantly along PC1, demonstrates that PC1 explains the majority of the model variance. The antioxidant/mesenchymal anticorrelation is a major contributor to explaining differentiation between untreated and EMT-transformed cells.

TGFβ treatment decreases nucleophilic tone

Covariance of antioxidant enzymes with epithelial markers and NOX4 with mesenchymal markers suggests that remodeling of the intracellular redox processes occurs over longer time scales, on the order of days. However, remodeling at the transcriptional level does not necessarily ensure that functional remodeling of the intracellular redox environment will ensue. To further investigate intracellular redox processes, we incubated TGFβ-treated A549 cells with the oxidation-activated fluorescein dye CM-H₂DCF-DA. Three days of TGFβ-treatment led to increased CM-H₂DCF-DA fluorescence (solid circles), which was enhanced in the presence of H₂O₂ (open squares; Figure 4A). Furthermore, pre-incubation with antioxidants NAC and catalase resulted in decreased CM-H₂DCF-DA fluorescence, matching levels of the untreated control (Figure 4B). In a time course study, mirroring the ICW analysis, significant increases in fluorescence of 2, 3.5, and 5-fold were observed at 48, 72, and 96 hour treated conditions compared to untreated controls (Figure 4C), indicating enhanced rates of CM-H₂DCF-DA oxidation in cells with more prolonged exposure to TGFβ.

Figure 4. ROS contribute to CM-H₂DCF-DA fluorescence during TGFβ-mediated EMT.

A) Flow cytometric analysis of CM-H₂DCF-DA (10 µM, 30 min) fluorescence in untreated and 200 pM TGFβ-treated A549 cells (solid circles) as well as with their counterparts receiving H₂O₂ co-administration (250 µM, 15 min; open squares). Samples with respective significant differences are demarcated (*). B) CM-H₂DCF-DA fluorescence following 1 hour pre-incubation with antioxidants NAC or catalase. Samples differing significantly from the 3-day TGFβ treated condition are enclosed within the shaded region. C) CM-H₂DCF-DA fluorescence time course during EMT measured by plate assay. Time points differing significantly from the untreated control (0 hours) are enclosed within the shaded region. All data are the result of 3 independent biological replicates (n=3) and plotted as the geometric mean ± standard error of the mean. Significance was determined by one-way ANOVA (p=0.05).

Owing to pronounced down-regulation of numerous antioxidant enzymes, we hypothesized that TGFβ-treated cells will exhibit a decreased rate of H₂O₂ clearance from the extracellular environment. Following a bolus addition of 20 µM H₂O₂, we measured the persistence of extracellular H₂O₂ using a luminol/sodium hypochlorite assay over a 60 minute time course (Figure 5A). The extracellular H₂O₂ concentration decreased over time and was used to calculate a relative rate of H₂O₂ clearance from the media. The H₂O₂ degradation rate (k_deg) was found to be decreased in TGFβ treated cells to ∼3/4 the rate of untreated controls (Figure 5B).

Figure 5. TGFβ treatment decreases nucleophilic tone.

A) Exogenous H₂O₂ was administered to cells in a 96-well plate and the H₂O₂ remaining in the media was measured at serial time points to determine the rate of H₂O₂ degradation (k_deg). B) Following TGFβ treatment (200 pM), the relative rate of H₂O₂ degradation from the media (k_deg) was measured with relative to the untreated control. C) The GSH and GSSG content of A549 cells were measured after multiple days of TGFβ treatment and used to calculate the half-cell reduction potential, or redox potential (E_GSH), of the GSSG/2GSH redox couple. Data are the result of 3 independent biological replicates (n=3) and plotted as mean ± standard error of the mean. Time points differing significantly from the untreated control, as determined by one-way ANOVA (p=0.05), are enclosed within the shaded regions.

Numerous regulators of GSH metabolism were among the antioxidant enzymes down-regulated following TGFβ treatment (Figure 2 & 3). Among them, glutamate-cysteine ligase catalytic subunit (GCLC) and glutamate-cysteine ligase regulatory subunit (GCLM) regulate GSH production, while glutathione reductase (GSR) mediates reduction of GSSG. [49] Therefore, we hypothesized that intracellular glutathione redox potential would become more oxidized. Calculation of the GSSG/2GSH redox potential necessitates knowledge of the absolute concentrations of GSH and GSSG. [42] Vi-Cell XR analysis revealed a 1% reduction in viability (Figure S7A), no significant difference in circularity (Figure S7B), and a 1.1 µm decrease in cell diameter (Figure S7C) in 96-hour TGFβ treated cells compared to untreated control. Cell volumes were computed from the cell diameters (Figure S7D) and used to determine intracellular concentrations of GSH and GSSG. Intracellular GSH and GSSG concentrations were measured following 0, 48 and 96 hours of TGFβ treatment and used to calculate the half-cell reduction potential, or redox potential, (E_GSH) of the GSSG/2GSH redox couple, which was observed to increase +6.8 mV (p < 0.05) following 96 hours of TGFβ treatment (Figure 5C).

Loss of nucleophilic tone coincides with EMT

We further investigated phenotype dynamics during the EMT time course by creating a multivariate model comprised of aggregated data from ICW protein expression, microarray transcript expression, CM-H₂DCF-DA oxidation, H₂O₂ degradation, and GSSG/2GSH redox couple studies. The majority of microarray transcript data were culled, but markers related to H₂O₂ production and degradation and regulation of the GSSG/2GSH redox couple were retained. A key strength of PCA is its ability to identify correlative relationships in complex, incomplete data sets, even those composed of data from a variety of methods of measurement and even nominal classifications.

Time point observations fell roughly along the typical rotational trajectory within the Scores plot with PC1 resolving early from late time points and PC2 resolving intermediate from early and late time points (Figure 6A). The 24-hour time point deviated slightly, which can be partially explained by the non-monotonic trend of pSmad3, Smad3, K_deg values. The Loadings plot is consistent with the ICW and microarray PCA models with the anticorrelation between epithelial/antioxidant markers and mesenchymal markers (Figure 1E; 3B; 6B). Many transcript/protein pairs are correlated. In fact, in a model composed of exclusively transcript/protein pairs across the entire time course (detailed in the Supplementary Information), correlation of transcript/protein dynamics is very evident.

Figure 6. Multivariate model of EMT dynamics, aggregating data from microarray, ICW, and redox assays.

PCA allows for the comparison of multiple types of data, obtained in a variety of experimental settings. In this aggregated model, transcript expression, protein expression, CM-H₂DCF-DA oxidation, H₂O₂ degradation rates (K_deg), and GSSG/2GSH redox couple data provide a multivariate description of phenotype. A) The scores follow the phenotype through a rotational trajectory. The single score trajectory reflects the loss of technical replicates upon data aggregation. B) Variable loading, and thus PC composition, is derived from multiple data sets, including microarray transcript (open markers), ICW (closed markers), and redox assay (cross markers) data sets. C) The biplot of the aggregated data PCA model, interrelating the information from the Scores and Loading Plots.

The Biplot enables the assessment of functional aspects of the redox environment within the context of EMT by combining the results of redox assays (black crosses) to be combined with transcript and ICW data (Figure 6C). The majority of model variance is explained by PC1, along which several trends are apparent. Indicators of nucleophilic tone (K_deg, GSH, total GSH, and GSSG levels) correlate with antioxidant and epithelial marker expression, indicating regulation of dynamics along similar time scales. Likewise, oxidizing shifts in CM-H₂DCF-DA and E_GSH data correlate with NOX4 expression and are anticorrelated with nucleophilic tone and antioxidants. In this manner, we observe that the loss of nucleophilic tone that follows TGFβ-treatment coincides with the transdifferentiation indicative of EMT.

Mesenchymal phenotype is stable to redox perturbation

Based on antioxidant antagonism of TGFβ signaling and the observed decrease in nucleophilic tone during EMT, we hypothesized that enhanced redox processes and continued TGFβ signaling serve to stabilize the mesenchymal phenotype following EMT. To determine whether the maintenance of a mesenchymal phenotype is dependent on redox processes and/or continued TGFβ input, we established a regimen of daily TGFβ supplementation, up to 4 days. After 2 days, TGFβ treatment media was exchanged in a set of intervention conditions with plain media or supplemented with neutralizing anti-TGFβ antibody, TGFβ signaling inhibitor (A8301), NAC, ebselen, or DMSO for an additional 2 days. Following treatment, we used ICW analysis to measure the response of epithelial, mesenchymal, antioxidant, and TGFβ signaling phenotype markers.

The responses profiles of the variables to this course of treatments were varied. Following 4 days of TGFβ treatment, E-cadherin, glutaredoxin-1, and catalase and the Smads underwent down-regulation while β-catenin, vimentin, Slug, pSmad3, and pErk1/2 exhibit up-regulation (Figure 7). Notably, at later time points, Smad2 and Smad4 levels increased, while pSmad3 levels decreased slightly, matching the trends of differential Smad regulation observed in the initial ICW time course (Figure 1B).

Figure 7. Phenotype profiles of antioxidant and TGFβ perturbed conditions following EMT.

A549 cells were treated up to 4 days, with daily replacement of TGFβ, to induce EMT. An intervention set received 2 days of TGFβ treatment followed by replacement of TGFβ media with plain, neutralizing anti-TGFβ antibody, TGFβ inhibitor A8301, NAC, ebselen, or DMSO-containing media for additional 2 days. The expression of epithelial/mesenchymal markers, antioxidants, Smads, and pErk1/2 were then measured via ICW assay. Normalized data are the result of 3 independent biological replicates (n=3) and plotted as mean ± standard error of the mean.

Given the high dimensionality of this data set, we again used PCA to aid our analysis the transdifferentiation and response trajectories. The rotational phenotype trajectory for EMT is evident as the solid line in the Scores plot (Figure 8A). Compared to the EMT phenotype resulting from a single bolus addition of TGFβ (Figure 1), the phenotype trajectory resulting from daily administration of TGFβ (Figure 8) shares similar temporal ordering of variable relationships, including anticorrelation of E-cadherin and antioxidants with mesenchymal markers and differential Smad regulation. The conditions from the intervention set (dashed lines) arise from the 2-day TGFβ treated samples. Examination of the Loading plot reveals familiar variable relationships, such as anticorrelation of E-cadherin and antioxidant markers with mesenchymal markers as well as anticorrelation of pSmad3 with Smad2/Smad4 (Figure 8B). The relationship between the variable loadings and the observation scores from all of the replicates are displayed in the Biplot (Figure 8C).

Figure 8. Multivariate analysis of phenotype profiles during EMT and following intervention with antioxidants or TGFβ inhibition.

Induction of EMT with daily 100 pM TGFβ treatment was intervened after 2 days with plain media, neutralizing anti-TGFβ antibody, A8301, NAC, ebselen, or DMSO. A) The scores from a representative replicate (replicate #2) are projected along the PCs. The rotational trajectory of EMT is traced in black (solid) while the intervention set (dashed) is distinguished from the other conditions. The phenotype trajectories replicates #1 and #3 can be seen in the biplot or found in the Supplementary Information B) Variable influence on PC composition reveals anticorrelation of E-Cadherin/Smad3/antioxidants with mesenchymal markers as well as of pSmad3/pErk1/2 with Smad2/Smad4. C) The differences in relative pSmad3/pErk1/2 and Smad2/Smad4 expression contribute to differentiation of intervention conditions from the 2-day TGFβ conditions while E-cadherin, antioxidant, and mesenchymal markers contribute relatively little.

The response of marker expression to cessation of TGFβ signaling within the intervention set, 2 days of TGFβ treatment followed by supplemented media, was quite varied (Figure 7) though the resultant phenotypes of each treatment condition were similar (Figure 8). Intervention of TGFβ treatment with supplemented media took place after 2 days of TGFβ treatment. Compared to the 2-day treated condition, those within the intervention set are notable for little to no change in expression of E-cadherin, β-catenin, vimentin, and Smad3 while glutaredoxin-1, Slug, pSmad3, and pErk1/2 expression decreases (Figure 7). However, catalase, Smad2, and Smad4 expression increases following cessation (Figure 7). PCA aids in visualizing these responses. The intervention conditions (dashed lines) follow a trajectory along PC2, remaining fairly constant along PC1 (Figure 8A). Variables with the largest response following intervention and cessation possess loading in latent variable space more heavily weighted along PC2 and less so along PC1 (Figure 8B, e.g. Smad4). In contrast, variables exhibiting little change following cessation of TGFβ signaling possess loading mostly along PC1 and less so along PC2 (Figure 8B, e.g. E-cadherin).

Discussion

Following TGFβ treatment, A549 cells undergo EMT, which is apparent by the down-regulation of E-cadherin and up-regulation of N-cadherin. Overall trends in these, and other markers, match changes in expression previously reported during TGFβ-mediated EMT in A549 cells [44, 45]; however, each of the markers possesses a distinct dynamic profile (Figure 1). Differential regulation of the Smads during TGFβ-mediated EMT has been previously reported [50, 51] and responses are cell-type specific. [52]. Broadly speaking, the differentiation markers and Smads suggest the phenotype profile is dynamic and that no two time-points are the same; therefore accurately determining which variables contribute to this differentiation is challenging even in this modest sized data set. Our implementation of the ICW assay has enabled greater resolution of the time course dynamics of Smad species along with epithelial and mesenchymal markers than is readily achievable using traditional western blotting approaches. The higher throughput afforded by ICW coupled with multivariate analysis enabled us to resolve the time-dependent nature of differential Smad regulation within the context of EMT.

Prior studies have identified TGFβ as a regulator of NOX4 and antioxidant enzyme expression. For example, NOX4 up-regulation by TGFβ is Smad3-dependent [21, 22]. Similarly, TGFβ, via Smad3, mediated the up-regulation of NOX4 as well as the down-regulation of catalase and manganese superoxide dismutase [31]. In other studies, TGFβ led to the down-regulation of antioxidant components, such as catalase, glutathione reductase, glutathione, glutathione synthetase [24, 53], glutamate cysteine ligase [54], and glutaredoxin-1 [25]. Significantly, these findings were observed in various cell types, with differing doses of TGFβ treatment, and at numerous times. Our findings of 50-fold NOX4 up-regulation concomitant with down-regulation of numerous key antioxidant enzymes (Figure 2) establishes clear anticorrelative changes at a single time point but alone does not discriminate between the possible synchronous expression changes that may coordinate remodeling of the redox environment or asynchronous regulation that may lead to competitive dynamics between oxidases and reductases.

The remodeling of phenotype following TGFβ treatment can be interrogated through a variety of methods, each of which, independently, has its own technical requirements and limitations. The high degree of concordance between transcript and protein expression shown here and reported by Keshamouni et al. [48], suggests that many of the transcript dynamics during EMT are likely to be reflected in the corresponding protein expression dynamics. Additionally, the union of these data sets enables the comparison high dimensional time course data from dissimilar modalities. A prominent feature of our PCA approach was the down-regulation of antioxidant expression on the same time scale (along PC1) during which mesenchymal expression is acquired during TGFβ treatment. Had the antioxidant and mesenchymal markers also been anticorrelated along PC2, it would have indicated that antioxidants were down-regulated before mesenchymal markers were up-regulated. Therefore, we find that the transcriptional down-regulation of antioxidant enzymes temporally coincides with the up-regulation of mesenchymal marker transcripts.

Because the set of epithelial markers (blue circles) did not covary in expression during EMT, some of these transcripts, assessed in isolation, may report a false characterization of phenotype. Indeed, a survey of several models of TGFβ-induced EMT identified numerous mixed or up-regulated epithelial markers [55]. As with the A549 cells, many of these cell lines were of cancerous origin. Thus, such seemingly contradictory findings may not reflect the biological response of normal epithelia to TGFβ, but may be a feature not uncommon amongst highly transformed cells.

We have demonstrated that, following the widespread down-regulation of antioxidants, functional aspects of the intracellular redox environment shift in a manner that would favor the stability and reactivity of electrophiles and reactive species. Increased CM-H₂DCF-DA oxidation indicates that oxidizing reactions are more favored following TGFβ treatment, which is corroborated by decreased CM-H₂DCF-DA fluorescence following pre-treatment with the antioxidants NAC and catalase (Figure 4B). Following TGFβ treatment, enhancement of several possible redox mechanisms, in isolation or combination, may account for the increased CM-H₂DCF-DA oxidation. Decreased k_deg rates (Figure 5B) reflect an impaired ability to eliminate electrophiles, such as H₂O₂, which can contribute to CM-H₂DCF-DA oxidation (Figure 4A). It is possible that decreased H₂O₂ transport rates across the plasma membrane could explain the decreased k_deg values; however, the equilibration of exogenous H₂O₂ across the plasma membrane occurs on the order of 1 second [56], while the half-life of H₂O₂ in the media was on the order of 10 minutes, indicating that membrane transport is not a rate-limiting step. A more likely mechanism of decreased k_deg values following TGFβ treatment is a decreased capacity for flux through the multiple antioxidant pathways capable of degrading H₂O₂. Finally, the more oxidizing E_GSH indicates the electrochemical capacity of the GSH redox buffer to clear electrophiles through reductive mechanisms is decreased. Therefore, the multifactorial effects of TGFβ on the redox environment result in decreased nucleophilic tone. [27] Further, our results indicate that the effect of TGFβ on a specific redox couple, such as GSSG/2GSH, may reflect perturbed regulation of numerous redox interconnected pathways, which are maintained in a non-equilibrium state [57]. For example, GSH [24, 53, 54], glutaredoxin-1 [23, 25], MnSOD, catalase, and NOX4 [31] are among the identified factors that have been found to be responsible for modulating the redox environment in response to TGFβ and the induction of EMT. Choosing a specific antioxidant enzyme, however, while ignoring others, and attributing its activity to an isolated effect within the redox environment is a gross over-simplification in the setting of TGFβ-mediated transcriptional reprogramming.

We observe the loss of nucleophilic tone to be coincident with change in phenotype that defines EMT (Figure 6). Furthermore, we observed the correlation of antioxidants with ferritin heavy chain (FTH1) expression (Figure 2, Figure 3). Within A549 cells, increased free intracellular iron and down-regulation of antioxidants have been reported as mechanisms leading to increased H₂DCF-DA oxidation following TGFβ treatment [20, 53, 54], though ROS production may be a contributory factor following EMT. NOX4 protein expression, and activity, producing H₂O₂, is known to vary proportionally with NOX4 mRNA transcript expression [19]. Given the widely reported lack of specificity of H₂DCF probes for specific ROS or redox reactions [58–60], and the observed dynamics of the vast number of factors that regulate the redox environment, attempts to identify specific factors responsible for CM-H₂DCF-DA oxidation would be futile. Increased NOX4, increased free iron, wide spread antioxidant down regulation, or any combination thereof are under dynamic control throughout EMT and can participate in processes that lead to CM-H₂DCF-DA oxidation.

Smads have been shown to exhibit redox regulation [30–34] and TGFβ can induce changes in the redox environment [20–26]. Therefore, it was possible that modulation of the redox environment precedes the differentiation along the epithelial/mesenchymal spectrum and alters Smad signaling in the process. Our study allowed for elucidation of the temporal relationship between reprogramming of the redox environment and the induction of EMT by TGFβ. The dynamics of CM-H₂DCF-DA oxidation and glutathione potential (Figure 5) are consistent with the results in Figure 3, indicating that decrease in nucleophilic tone occurs during the course of EMT, not preceding it. Further, the decreased tone occurs at too late a time to account for elevation of pSmad3 levels (Figure 1C). These findings relate the natural dynamic response of the redox environment to signaling and differentiation caused by TGFβ during EMT. Previous studies linking H₂O₂ treatment with the induction of TGFβ-mediated, pSmad3-dependent EMT in A549 cells relied upon exogenous administration of hyper-physiological levels of H₂O₂ [29]. Our findings negate a potential mechanism of TGFβ-mediated EMT via an induced feed-forward loop that would enhance Smad3 phosphorylation by decreased nucleophilic tone.

We observed anticorrelation of pSmad3 with Smad2 and Smad4 (Figure 8B), which was also a feature in the initial EMT time course study (Figure 1E). In agreement with previous studies [50–52], daily TGFβ treatment resulted in suppressed Smad3 levels (Figure 7H). Cessation of TGFβ signaling had no effect on Smad3, suggesting its expression is independent of active TGFβ signaling. Smad2 and Smad4, however, exhibited a strong up-regulation, suggesting that cessation of TGFβ signaling and pSmad3 activity may enable subsequent Smad2 and Smad4 up-regulation. The differential Smad activities operated independently from the state of differentiation and presence of exogenous antioxidants.

While we anticipated a relative loss of mesenchymal phenotype, via MET, for the intervention set conditions, the expression of E-cadherin, β-catenin, and vimentin remained largely unchanged. The failure of most of these conditions to exhibit a relative MET is notable for several reasons. NAC and ebselen treatments, common non-specific antioxidants, were not able to rescue an epithelial phenotype following EMT, nor were inhibitions of TGFβ. Each of the intervention conditions, except plain media and DMSO, are known to inhibit TGFβ signaling at the concentrations used and through independent mechanisms: neutralizing anti-TGFβ antibody [61], A8301 [62, 63], NAC [30, 32, 35, 64, 65], ebselen [31]. NAC [26, 35] and A8301 [62] inhibit TGFβ-mediated EMT, presumably through inhibition of TGFβ signaling and Smad3 phosphorylation. TGFβ signaling is decreased within the intervention set compared to the 2-day treatment condition, as indicated by decreased slug, pSmad3, and pErk1/2 expression (Figure 7). Smad expression exhibited vast remodeling, irrespective of maintenance of the mesenchymal phenotype or redox perturbation. Therefore, it appears that sustained TGFβ signaling and pSmad3 activity is not required to maintain a mesenchymal phenotype and that the state of differentiation following 2 days of TGFβ treatment is stable over the course of 2 days, even in the absence of continued TGFβ signaling or when augmented by exogenous antioxidants.

A8301-treated cells displayed a slight distinction from the rest of the intervention group in that in addition to displacement along PC2, its position also reversed along PC1 compared to the 2-day treated condition, suggesting a slight MET-type differentiation (Figure 8). Slight elevations in expression of markers correlated with an epithelial phenotype are consistent in the A8301 condition compared to the rest of the intervention set (Figure 7A–C, H). A8301, a TGFβ receptor kinase inhibitor, most strongly inhibits TGFβ signaling but it also significantly inhibits a number of other pathways, such as VEGFR, RIPK2, MINK1, p38a MAPK, PKD1, FGFR1, and CK1 [66]. Phospho-Erk1/2 expression is slightly suppressed in the A8301 condition, return to near baseline levels, while pErk1/2 levels remain slightly elevated compared to the untreated control (Figure 7K).

It is possible that pErk1/2 activity is sufficient to maintain the mesenchymal phenotype and that A8031 activity, directly or indirectly, inhibits pErk1/2 and in doing so induces MET. MAPKs and a number of other signaling pathways are components of non-canonical TGFβ signaling [67]. However, as neutralizing anti-TGFβ antibody and exogenous antioxidants failed to induce MET, it appears that neither canonical nor non-canonical TGFβ signaling is responsible for maintenance of the mesenchymal phenotype. Still, pErk1/2 may serve to maintain phenotype. MEK inhibition has been demonstrated as a means to prevent TGFβ-mediated EMT, implicating a mechanism of Erk activation in the induction of EMT [68–70]. Alternatively, MEK inhibition was also demonstrated to be ineffective in the prevention of EMT but found to be critical for the induction of FGF1-mediated reversal of EMT (MET) [71]. Similar findings have been described for other non-canonical pathways [68, 72–74]. The collaboration of multiple signaling pathways, sometimes referred to as cross-talk, in the propagation of signals are likely to be critical components in the determination of cell fate rather than auxiliary pathways. Such scenarios highlight the complex, multivariate, non-linear nature in which biological systems are controlled.

A549 cells are an immortalized lung carcinoma cell line [75], in which the responses to TGFβ have been studied extensively. Studies of more focused aspects of phenotype (e.g. EMT, antioxidant down-regulation, or NOX4 up-regulation) are critical for our understanding of molecular mechanisms of transdifferentiation but underappreciate the scope and scale in which TGFβ-mediated transformation occurs. Here we demonstrate that such characteristics do not operate in isolation and that the response of A549 cells to TGFβ involves remodeling of the redox environment, resulting in decreased nucleophilic tone, induction of EMT, and perturbation of Smad transcription factor expression.

Under normal developmental conditions, EMT is a highly coordinated and regulated effort, subject to control by multiple extracellular and intracellular signaling pathways, including MAPKs [3]. Our results suggest that in the A549 cell line, the induction of TGFβ-mediated EMT is intrinsically different from the reversal of EMT via MET. A549 cells are, in a sense, primed for response to TGFβ as they possess an activating Ras mutation [76], which is a key enabler of TGFβ-mediated EMT [77]. Similarly, in mouse mammary epithelial cells, transformation via H-Ras conferred the ability to undergo TGFβ-mediated EMT and maintain the resultant mesenchymal phenotype to untransformed cells [73]. In the same cell lines, glutaredoxin down-regulation by TGFβ was found to be MAPK-mediated and its overexpression prevented induction of EMT [23] underscoring the interconnectedness of EMT and the redox environment. While the decreased nucleophilic tone acquired during EMT may not serve to stabilize the mesenchymal phenotype, the altered redox environment may play a role in other cellular behaviors such as enhanced cell motility [21, 78] or apoptotic resistance [79, 80]. Cannito et al. have presented an extensive review of the numerous redox mechanisms that have been identified within the context of various models of EMT [28]. Transformed cells may be primed for EMT through perturbed intracellular signaling or altered redox states. Judicious attribution or negation of such mechanisms for a particular cell type would require extensive study.

The A549 cell line is a transformed carcinoma cell line that exhibits phenotypic deviations from healthy primary lung epithelia. A549 cells exhibit an intermediate phenotype, displaying some mesenchymal characteristics alongside the epithelial phenotype [10]. Additionally, antioxidant enzymes are highly expressed [81, 82]. An impaired Keap1-Nrf2 interaction in A549 cells has resulted in a much greater nucleophilic tone compared to non-malignant cells [83]. Therefore our observations about the multivariate nature of EMT should be best understood in the context of highly transformed cells responding to TGFβ, as might occur within the tumor microenvironment. In such transformed tissues, reversion of a TGFβ-influenced mesenchymal phenotype is not likely to be as simple as blockade of TGFβ signaling. Thus administration of treatments using non-specific antioxidants, small molecule inhibitors, or biologics that specifically target TGFβ signaling may be insufficient; however, such treatments may play key roles in combination therapies that attempt to restore normal function through systems level approaches. More robust characterization of multivariate phenotype dynamics will allow for parsing of covariant phenotypic programs, which will improve our ability to modulate specific cellular behaviors and responses.

Conclusions

The phenotypic response of A549 cells to TGFβ treatment is an extensive and dynamic process with relevance to carcinogenesis and other pathologies. The precision and reproducibility afforded by ICW techniques allowed us to construct a multivariate representation of phenotype dynamics during EMT using PCA. This model demonstrated the validity of the approach, robustness of the transdifferentiation trajectories, and helped generate additional research questions. We examined the scope of redox remodeling during EMT and found that multiple antioxidants enzymes are down-regulated while the oxidase NOX4 is up-regulated by TGFβ on a time scale that matches the acquisition of a mesenchymal phenotype. Increased CM-H₂DCF-DA oxidation, decreased H₂O₂ degradation, and elevated GSSG/2GSH redox potentials provided additional functional evidence of decreased nucleophilic tone in parallel with the acquisition of the mesenchymal phenotype. Following EMT in A549 cells, the mesenchymal phenotype was stable in the presence of the antioxidants NAC and ebselen as well as TGFβ inhibition through neutralizing antibody and the small molecule inhibitor A8301. Additionally, we observed differential Smad dynamics operating independently of epithelial/mesenchymal differentiation. This novel approach enabled the investigation of the dynamics of multivariate phenotype states as they developed over time. This investigation yielded a new perspective on the state of intracellular redox environment within the context of EMT.

Highlights.

• Decreased nucleophilic tone during EMT does not precede mesenchymal differentiation.

• Multiple TGFβ-induced redox transcriptional dynamics are coordinated.

• Redox markers strongly correlate with A549 transdifferentiation.

• The mesenchymal phenotype following EMT is resistant to antioxidant perturbation.

Abbreviations

(TGFβ)

Transforming Growth Factor β

(EMT)

Epithelial-Mesenchymal Transition

(MET)

Mesenchymal-Epithelial Transition

(PCA)

Principal Component Analysis

(PC1)

Principal Component 1

(PC2)

Principal Component 2

(ICW)

In-Cell Western

(qRT-PCR)

Assay Quantitative Real-Time Polymerase Chain Reaction

(NOX4)

NADPH Oxidase 4

(H₂O₂)

Hydrogen Peroxide (H₂DCF), Dichlorodihydrofluorescein

(k_deg)

H₂O₂ Degradation Rate

(GSH)

Reduced Glutathione

(GSSG)

Oxidized Glutathione

(E_gsh)

GSSG/2GSH Half-Cell Reduction Potential;