Morphological profiling of environmental chemicals enables efficient and untargeted exploration of combination effects

Environmental chemicals are commonly studied one at a time, and there is a need to advance our understanding of the effect of exposure to their combinations. Here we apply high-content microscopy imaging of cells stained with multiplexed dyes (Cell Painting) to profile the effects of Cetyltrimethylammonium bromide (CTAB), Bisphenol A (BPA), and Dibutyltin dilaurate (DBTDL) exposure on four human cell lines; both individually and in all combinations. We show that morphological features can be used with multivariate data analysis to discern between exposures from individual compounds, concentrations, and combinations. CTAB and DBTDL induced concentration-dependent morphological changes across the four cell lines, and BPA exacerbated morphological effects when combined with CTAB and DBTDL. Combined exposure to CTAB and BPA induced changes on the ER, Golgi apparatus, nucleoli and cytoplasmic RNA in one of the cell lines. Different responses between cell lines indicate that multiple cell types are needed when assessing combination effects. The rapid and relatively low-cost experiments combined with high information content makes Cell Painting an attractive methodology for future studies of combination effects. All data in the study is made publicly available on Figshare. Highlights Assessment of combination effects of BPA, CTAB and DBTDL on four human cell lines Morphological profiling/Cell Painting captures dose and combination dependent effects BPA exacerbated morphological effects when combined with CTAB and DBTDL. Cell models of diverse origin are needed when profiling environmental chemicals


Introduction
Much effort is focused on understanding the potential hazard of environmental chemicals. These chemicals can be present in everyday products, such as food and cosmetics, and in our direct environment, such as pesticides, pharmaceuticals and synthetic chemicals 1 . Despite growing scientific evidence for the toxicity of chemical mixtures, little is known about combination effects of environmental chemicals and the possible health risks they pose. Current approaches for risk assessment of chemical mixtures are often based on concentration or response addition models of the individual component toxicities 2 . These methods neglect interactions between chemicals in the environment or at target sites, which can result in overall toxicity being stronger or weaker than predicted 2 . New experimental approaches to test and validate toxicity models to better capture combination effects are greatly needed.
The most widely used methodology to study the cellular response to chemicals in regards to its toxicity or mode of action are viability assays, or alternatively readouts to investigate the activation of specific pathways. Such assays include reactive oxygen species, DNA damage, ER stress, or NFKb activation, among others 1 and are commonly targeting specific individual biological pathways. It is not uncommon to perform several such assays to get a wider profile of relevant endpoints 3 .
Morphological cell profiling is increasingly being used in diverse scientific areas including toxicology and drug discovery, propelled by the ability to generate rich data at reasonable costs 4,5 . In contrast to traditional single endpoint assays, morphological profiling is untargeted and can be used to study multiple types of cell responses simultaneously. A popular method is Cell Painting 6 where cells are perturbed with e.g. exposure to a chemical compound or genetic alteration, stained with multiplexed dyes binding to different cell compartments, and imaged using automated microscopy in multiple channels. Image analysis can then be used to segment individual cells, followed by a range of measurements producing morphological profiles on a single-cell level. Contributing to the power and popularity of the method is also that the resulting images and morphological profiles have successfully been analyzed with unsupervised and supervised machine learning methods [7][8][9] .
Morphological profiling can be used to elucidate toxicity mechanisms and mode of action of a given chemical entity, which in turn allows for the generation of empirical hypotheses 10 . The methodology has been applied to experimental drugs, large chemical libraries, pseudo-natural products, and environmental chemicals across various cell lines 4,7,[11][12][13][14] . In safety assessment, cell-based assays offer the possibility to provide insights into the cellular responses to chemical exposure. Chemically-induced cellular responses can be cell-specific as well as organelle-specific and are often complex, consisting of initiation and activation of various biochemical processes that lead to a biological response. To this end, morphological profiling and Cell Painting are gaining popularity in screening and profiling applications. However, studies applying morphological profiling to investigate the effects of chemical combinations on cells have so far not been reported.
In this study, we analyze the individual effect as well as combination effects of three environmental chemicals, present in everyday objects and contexts, on four human cell lines using Cell Painting. Bisphenol A (BPA) is one of the most used plasticizers applied in the polymerization of polycarbonate plastics. BPA has been associated with multiple adverse health effects such as metabolic disorders 15 , cardiovascular diseases 16 , and cancers 17,18 . Due to its ubiquitous presence in consumer products such as food packaging, toys and drinking bottles, this chemical has a high potential to be co-exposed to other chemicals 19 . Dibutyltin dilaurate (DBTDL) is a widely used industrial chemical, serving as an antifouling coating and is used in pesticides and fungicides. It is a persistent chemical that can be deposited in the liver where it can affect liver functions 20 . Cetyltrimethylammonium bromide (CTAB) is a cationic compound that is used as an antibacterial and antifungal surfactant, used as a topical antiseptic in cleaning and cosmetic products. This compound has been shown to induce cytotoxicity and inflammatory toxicity in vivo and in vitro 21 .
The aim of this study is to establish that morphological profiling with Cell Painting combined with multivariate analysis is a suitable methodology that can be used to analyze the effect of combinations of chemicals on cells, demonstrated on three chemical substances that have widespread use.

Cell seeding and compound treatments
Using a Biotek Multiflo FX microplate dispenser, U-2 OS (1000 cells/well), A549 (800 cells/well), Caco-2 (1200 cells/well) and MCF7 (1200 cells/well) were seeded in a volume of 24µl in 384 multiwell plates (Costar; Falcon 384-well Optilux Flat Bottom plates, 353962). The plates were kept at room temperature for 20 minutes to aid homogeneous spreading. The plates were incubated overnight for 24 hours at 37 ℃ at 5% CO 2 atmosphere to allow for cell attachment.
The outer wells were excluded from experimentation to avoid edge effects.
The three selected environmental chemicals in this study included: BPA (Sigma-Aldrich; 239658), DBTDL (Sigma-Aldrich; 291234) and CTAB (Sigma-Aldrich; H9151). The compounds were tested individually, as well as in two and three-component mixtures in three concentrations (1µM, 6.45µM and 12µM). Six phenotypic reference compounds were included which are known to induce a distinctive phenotype in a variety of cells 22 . Specifically, these included: Etoposide (E1383), Fenbendazole (F5396), Metoclopramide (M0763), Berberine Chloride (B3251), Tetrandrine (T2695) and Fluphenazine dihydrochloride (F4765-1G). D-Sorbitol (S1876) was included as negative control and DMSO (D2438) was used as a compound vehicle (all purchased from Sigma Aldrich). All compounds were dissolved in DMSO to 10mM stock solutions, then a 5X source plate was prepared in Dulbecco's Minimum Essential Media (DMEM) using an automated liquid handler (OT-2; Opentrons, Brooklyn, NY). Compound conditions were distributed over the plates with four technical replicates and two biological replicates per cell line. From the 5X source plate, 6 µl compounds was transferred to the assay plates using a Viaflo 384 electronic pipette, reaching 1✕ of the final compound concentration (1µM, 6.45µM and 12µM). The cells were incubated for 24 hours at 37°C under 5% CO2. In order to optimally distribute the conditions over the plate and reduce the impact of positional effects, we produced plate layouts (Suppl. Fig. S1) with PLAID (Plate Layouts using Artificial Intelligence Design, https://github.com/pharmbio/plaid).

Cell Painting
The Cell Painting experiments closely followed the protocol described by Bray et al. (2016) 6 . In Phalloidin/Alexa Fluor 568 (Invitrogen; A12380) in 1X PBS. A total of 20 μl staining mixture was added to each well reaching a final well-concentration of 1μg/ml Hoechst, 15μg/ml Wheat germ agglutinin, 10μl/ml Phalloidin, 4μM SYTO 14 and 80μg/ml Concanavalin A, and was incubated for 20 minutes. Then, the plates were washed a final three times (80μl, 1x PBS), sealed and kept at 4°C prior to imaging. Plates were protected from light as much as possible.

Image acquisition and processing
Fluorescence microscopy was conducted using a high throughput ImageXpress Micro XLS Image analysis was divided into three steps: 1) quality control, 2) illumination correction and 3) segmentation and feature extraction. The quality control (QC) pipeline was run on the raw images to detect images with staining and/or imaging artifacts. Various quality measures were calculated on the raw images to represent a wide variety of artifacts. Images deviating more than five standard deviations from the median for FocusScore, MaxIntensity, MeanIntensity, PercentMaximal, PowerLogLogSlope and StdIntensity were flagged, inspected and removed if necessary. Blurred and oversaturated images were detected by calculating the PercentMaximal score and PowerLogLogSlope. Images with PercentMaximal values higher than 0. 25, or PowerLogLogSlope values lower than -2.3 were removed from the dataset. The remaining bright artifacts in the DAPI channel were identified using the Identify Primary Object module in CellProfiler. The artifacts and surrounding pixels were masked to avoid interference with segmentation. Intensity measures, object size and cell counts were visualized as plate heatmaps to detect plate and batch effects. Images with a cell count below 20 were removed from further analysis. To correct for uneven illumination, a polynomial illumination correction function was calculated for each plate and each image channel. Each image was then divided by the respective illumination correction image to correct for uneven illumination.
A feature extraction pipeline was built to segment three cell compartments (nuclei, cytoplasm, cells) and extract morphological features. Nuclei segmentation was performed on the Hoechst staining, by applying a gaussian blur followed by Otsu thresholding to segment the outlines of each nucleus. For U-2 OS, A549 and MCF7, the cell objects were segmented using watershed segmentation using minimum cross-entropy on the cytoplasmic RNA stain, using the nuclei as a Q2 and R2 values were calculated for the PLS-DA models, which were estimated by three-fold cross-validation repeated for 10 times. Induction scores were computed using z-score normalization relative to the DMSO controls per plate: x.zscore = (xmedian(x[subset]))/mad(x[subset]). The induction score was used as a measure for bioactivity, as described earlier 25

Evaluation of environmental toxicants by morphological profiling
In order to investigate the morphological effects the environmental toxicants CTAB, BPA and DBTDL (Fig. 1A) exert on cells, we used the Cell Painting assay 6 . In short, cell lines originally derived from different organ origin, represented by MCF7 (breast cancer), A549 (lung cancer), Caco-2 (colon cancer) and U-2 OS (osteosarcoma), were seeded on multiwell plates 24h prior to exposure to combinations of the environmental toxicants for another 24h. Subsequently, the Cell Painting protocol was performed, followed by high-content imaging, image analysis using Cell Profiler and CellPose 23,24 , and data analysis of the morphological features (Fig. 1B). To assess the reproducibility and sensitivity of the assay, we first analyzed the morphological changes induced by DMSO controls and six phenotypic reference compounds. Principal component analysis (PCA) of the morphological profiles in the DMSO condition separated each cell line in the morphological space, and are highly reproducible between biological replicates (Fig.   1D). Exposure to the phenotypic reference compounds induced dose-dependent changes that showed substantial similarity between cell lines (Suppl. Fig. S2). Next, we assessed the effect of exposing the four cell lines to the environmental toxicants CTAB, BPA and DBTDL. The cell lines responded differently to each of the treatments, which is depicted by the morphological changes observed upon the staining of the different cell compartments included in the Cell Painting assay (Fig. 1C). In particular, distinct morphological changes could be observed upon CTAB exposure on A549 and Caco-2 cells, characterized by intra-cytoplasmic vesicles, as well as increased intensity of mitochondrial staining upon DBTDL exposure for all cell lines (Fig. 1C). We generated morphological profiles of MCF7, A549, Caco-2 and U-2 OS cells exposed to vehicle (DMSO), CTAB, BPA, and DBTDL, at 1, 6 and 12 μM, for 24h (Suppl. Figs. S3 and   S4).

CTAB, BPA and DBTDL induce cell-specific morphological signatures
We used cell count to assess the effect of the chemical exposures at 1, 6 and 12μM ( Fig. 2A). The highest concentration of CTAB and DBTDL seem to induce toxicity in all cell lines, however DBTDL particularly affected U-2 OS cells. In order to quantitatively assess morphological changes, we calculated an induction score (IS) based on the z-score of the number of features altered by the exposure to each of the compounds, in respect to the vehicle DMSO (Fig. 2B). The IS clearly showed a dose-dependent increase in the alteration of the morphological features for DBTDL and CTAB, but to a much lesser extent for BPA.
Next, in order to explore what cellular compartments were altered upon exposure to the three environmental chemicals, we constructed radar plots by calculating the z-score of the extracted morphological features, grouped by cellular compartment (nucleus, endoplasmic reticulum (ER), nucleoli and cytoplasmic RNA, F-actin cytoskeleton and Golgi apparatus, and mitochondria), as well as correlation, neighboring and area features, for each cell line, chemical and dose (Fig. 2C) We used partial least-squares discriminant analysis (PLS-DA) on the extracted morphological features to create models for predicting the effects that each compound elicited on the cells. Similarly to the Induction Scores, the PLS-DA distance from DMSO ( Fig. 2D), indicated a dose-response for CTAB and DBTDL, which was not the case for BPA. The PLS-DA models predicted that CTAB and DBTDL induced distinct and reproducible morphological changes, indicated by the clustering of the different chemical doses, distanced from the DMSO control, as well as the high coefficient prediction Q 2 , especially for concentrations 6μM and 12μM (Fig. 2E and Supplementary Table S1). On the other hand, the low coefficient prediction Q 2 for BPA for most concentrations (Fig. 2E and Supplementary Table S1), indicated that the morphological changes induced by this chemical did not differ substantially from the DMSO control, as it is also seen by the close proximity of the clusters of the three BPA concentrations to DMSO (Fig. 2E). We next investigated the effect of the combination of CTAB, DBTDL and BPA. We exposed MCF7, A549, Caco-2 and U-2 OS cells to either double or triple combinations of the chemicals, at 1, 6 or 12μM. We used cell count to determine the toxicity resulting from combined exposure to the chemicals (Fig. 3A). Only at the highest concentration, either double or triple exposure of the compounds seemingly induced some degree of toxicity. With the exception of the combination of CTAB and DBTDL, BPA and DBTDL, as well as the triple CTAB+BPA+DBTDL combination, which induced toxicity in MCF7, Caco-2, and particularly on U-2 OS cells, at 6μM (Fig. 3A).

Cell Painting enables detection of combination effects
We calculated the Induction Scores (IS) for both double and triple combinations of chemicals (Fig. 3B). The IS indicated that combining DBTDL with either CTAB or BPA at 6μM, resulted in a reduction of the score compared to the exposure of the chemicals alone, pointing at a potential buffering effect (Fig. 3B). The combination of CTAB and BPA at 6μM and the combination of DBTDL and CTAB at 1μM, however, resulted in an increased IS compared to the single chemical exposures in U-2 OS, A549 and MCF7 cells, indicating a potential additive or synergistic effect. PLS-DA was then used on the extracted morphological features from the double and triple exposures. The PLS-DA distances from DMSO reflected a similar effect as identified by the Induction Score calculation, here we observed a shorter distance from DMSO when exposing the cells to DBTDL in combination to CTAB compared to the individual exposures (Fig. 3C).
To further explore the cell morphologies induced by combined exposure, the features were dimensionally reduced using non-linear uniform manifold approximation and projection (UMAP). (Fig. 3D, 3E). Morphological profiles induced by combined exposure to CTAB and BPA were dominated by the effects from CTAB in MCF7 cells (Fig. 3D).
Triple exposure to compounds CTAB, BPA and DBTDL, induced morphological changes that could be attributed to the cumulative changes from both CTAB and DBTDL ( Fig   3E).  4C).
An increase in global perturbation levels of the morphological features was observed by the addition of BPA, compared to the double combination of CTAB and DBTDL. Addition of BPA to DBTDL and CTAB, resulted in larger PLS-DA distances (Fig. 3B) and higher induction scores (Fig. 3C) across all four cell lines at 6μM and 12μM exposures. The radar plots display increased perturbation levels in MCF7 and Caco-2 cells upon addition of BPA ( Fig. 4D and E).

Discussion
The continued increase in the number and load of chemicals in our environment highlights the need for methods to systematically assess risks associated with exposure to combinations. Here, we employed morphological profiling as a new experimental approach to study combination effects of chemicals. We interrogated chemical-induced cellular morphologies of three environmental chemicals: CTAB, DBTDL and BPA exposed individually as well as in double and triple combinations, at three different doses, using four human cell lines.
Using the Cell Painting assay with single-cell measures of cell morphologies and multivariate data analysis, we were able to identify distinct changes upon exposure to both single chemicals and combinations of chemicals. We could effectively discern effects from individual doses for chemicals CTAB and DBTDL, as well as distinguish between single and combined exposure with CTAB, DBTDL and BPA. BPA did not result in significant morphological changes but showed additive or synergistic effects when co-exposed to CTAB and DBTDL.
Exposure to compounds CTAB and DBTDL followed a clear concentration-dependent morphological response across all four tested cell lines. In contrast, exposure to compound BPA resulted in a more subtle morphological response, which followed a non-monotonic dose-dependent response. This result could be explained by opposing effects induced by compound BPA binding to multiple receptors, receptor desensitization, negative feedback loops, or dose-dependent metabolism modulation, which are commonly reported for endocrine-disrupting chemicals 26 .
Previous research has shown that DBTDL can induce cell cycle alterations, DNA damage, and increased calcium influx, and at higher concentrations can lead to nuclei deformation, mitochondrial swelling, and apoptotic-like changes 27,28 . In concordance with these findings, we observed that high concentrations of DBTDL heavily affected the mitochondria in all cell lines, as well as the nuclei intensity and size of nuclei and cell compartments in U-2 OS and MCF7 cells, which might be an indication for apoptosis.
CTAB induced strong morphological changes in all cell organelles and was bioactive in at least three out of the four cell lines. This finding is in line with previous observations where CTAB present in gold nanorods has been shown to induce cell apoptosis and autophagy by activating reactive oxygen species (ROS) and leading to mitochondrial dysfunction 29 . One proposed mechanism of toxicity of compound CTAB, like other surfactants, is the formation of micelles, also known as nanobubbles, in aqueous solutions. These micelles can disrupt cell membranes leading to Ca 2+ influx and subsequent cell death and can induce inflammatory responses when exposed to human blood cells 30 .
Interestingly, even though compound BPA did not elicit a strong morphological change when exposed alone, an increase in perturbation level could be observed when combined with CTAB or when combined with both compounds DBTDL and CTAB together. Previously, co-exposure of BPA with other endocrine-disrupting chemicals, such as phthalates, have been reported to lead to synergistic and additive interactions, such as increased DNA damage in vitro, and significantly changed lipid profiles and glucose levels in vivo 31,32 . The increased bioactivity could also be explained by interactions between BPA and CTAB molecules; BPA can stabilize CTAB molecules in the micelle and at the air/water interface which might in turn increase the micelle-mediated toxicity by CTAB 33 .
Another important finding in this study is the observed cell-line specific effects upon treatment. We used breast cancer MCF7 cells, lung cancer A549 cells, colon cancer Caco-2 cells, and osteosarcoma U-2 OS cells, all of which differ on their genetic, metabolic or proteomic landscapes [34][35][36] . Thus, the diverse responses to the environmental chemicals by the different cell lines could be explained by the presence or absence of the specific chemicals' molecular targets. This finding emphasized the need to include biologically diverse cell lines for toxicity studies.
One surprising result in this study was the dose-dependent dampening or exacerbation of morphological effects for combined exposure of CTAB and DBTDL. Commonly used methods, such as additive and response-addition models only assume additive effects upon combined exposure, which further highlights the need for experimental tools to validate toxicity models. However, to develop a full picture of the combination effects between the chemicals, additional studies will be needed exploring a wider range of concentrations and using concentration-response matrices.
Morphological profiling has the ability to capture subtle morphological changes upon a given perturbation, before it reaches toxic levels. This indicates that we can detect cell stress upon exposure to contaminants with cell morphology profiling before it can be observed in e.g. cell viability assays. Further, compared to widely used receptor-based assays, morphological profiling captures a much wider set of changes at molecular and cellular levels integrated in a single assay. Indeed, analyses of Cell Painting profiles have earlier been shown to be predictive for multiple cell health outcomes such as proliferation, apoptosis, reactive oxygen species (ROS), DNA damage, cytotoxicity, and cell cycle phase 9,37 .
The untargeted and unbiased approach makes Cell Painting an attractive alternative to receptor based assays to test higher order environmental chemical mixtures. However, one possible shortcoming is assay sensitivity to detect changes at low concentrations.
In this study we observed only subtle changes at the 1μM concentration, whereas environmentally relevant concentrations of exposure of environmental contaminants are commonly much lower 1 . Possible explanations could be that assay variability and batch effects mask subtle morphological effects, that the multivariate analysis approach does not capture certain relevant changes, or that the biological activity does not manifest in a morphological change. In addition, the use of primary cell lines, or non-immortalised cell lines, might improve the detection of the subtle morphological changes that environmental toxicants may induce. Integration of additional profiling techniques, such as proteomics or transcriptomics, may shine more light on cellular responses.
Humans are potentially exposed to more than thirty thousand chemicals from a variety of sources, screening even a fraction of all possible combinations would be impossible 19 . Risk assessment of chemical mixtures is a multifaceted task, which should integrate experimental toxicity data, mechanism of action and in silico modeling 38 . Major bottlenecks for mixture models nowadays are that experimental toxicity data is scattered and insufficient to address complex mixtures. In this study we show that morphological profiling offers a reproducible and standardized approach, integrating multiple endpoints and cell lines, which could be used to systematically assess combination effects and has the potential to advance our understanding of health risks of chemical mixtures.