A Data-Driven Transcriptional Taxonomy of Adipogenic Chemicals to Identify White and Brite Adipogens

Background: Chemicals in disparate structural classes activate specific subsets of the transcriptional programs of peroxisome proliferator-activated receptor-γ (PPARγ) to generate adipocytes with distinct phenotypes. Objectives: Our objectives were to a) establish a novel classification method to predict PPARγ ligands and modifying chemicals; and b) create a taxonomy to group chemicals on the basis of their effects on PPARγ’s transcriptome and downstream metabolic functions. We tested the hypothesis that environmental adipogens highly ranked by the taxonomy, but segregated from therapeutic PPARγ ligands, would induce white but not brite adipogenesis. Methods: 3T3-L1 cells were differentiated in the presence of 76 chemicals (negative controls, nuclear receptor ligands known to influence adipocyte biology, potential environmental PPARγ ligands). Differentiation was assessed by measuring lipid accumulation. mRNA expression was determined by RNA-sequencing (RNA-Seq) and validated by reverse transcription–quantitative polymerase chain reaction. A novel classification model was developed using an amended random forest procedure. A subset of environmental contaminants identified as strong PPARγ agonists were analyzed by their effects on lipid handling, mitochondrial biogenesis, and cellular respiration in 3T3-L1 cells and human preadipocytes. Results: We used lipid accumulation and RNA-Seq data to develop a classification system that a) identified PPARγ agonists; and b) sorted chemicals into likely white or brite adipogens. Expression of Cidec was the most efficacious indicator of strong PPARγ activation. 3T3-L1 cells treated with two known environmental PPARγ ligands, tetrabromobisphenol A and triphenyl phosphate, which sorted distinctly from therapeutic ligands, had higher expression of white adipocyte genes but no difference in Pgc1a and Ucp1 expression, and higher fatty acid uptake but not mitochondrial biogenesis. Moreover, cells treated with two chemicals identified as highly ranked PPARγ agonists, tonalide and quinoxyfen, induced white adipogenesis without the concomitant health-promoting characteristics of brite adipocytes in mouse and human preadipocytes. Discussion: A novel classification procedure accurately identified environmental chemicals as PPARγ ligands distinct from known PPARγ-activating therapeutics. Conclusion: The computational and experimental framework has general applicability to the classification of as-yet uncharacterized chemicals. https://doi.org/10.1289/EHP6886


A Data-Driven Transcriptional Taxonomy of Adipogenic Chemicals to Identify White and Brite Adipogens
Stephanie Kim, Eric Reed, Stefano Monti, and Jennifer J. Schlezinger Table of Contents   Table S1. Chemical information. Table S2. Mouse (M) and human (H) primer sequences for reverse transcription qPCR.             LG100268 LG268 153559-76-3 Sigma Aldrich SML0279 1x10-7 1x10-7 Yes Activates PPARγ through RXR > 98% (Cesario et al. 2001) LG100754 LG754 *NA = Not Available a Toxicity was assessed by microscopic inspection. The concentration was reduced by two fold until a non-toxic, maximal concentration was identified. Maximal concentrations were determined in an N=4. b "Yes" indicates there is experimental evidence of modification of PPARγ activity, including PPARγ binding assays, coactivator recruitment or computational modeling (definitive ligands), PPARγ-driven reporter assays (at least 25% of the rosiglitazone-induced maximum), expression of PPARγ target genes and/or differentiation of 3T3 L1 or multipotent stromal cells into adipocytes in the absence of a known PPARγ ligand. Chemicals that changed the expression of PPARγ (e.g., PCB126 and DDE) were not considered to be ligands or modifiers. "No" indicates that the chemical was chosen based on the fact that it is known to be a specific ligand of another receptor. "No evidence" indicates that this chemical has not been tested for PPARγ activation but is structurally dissimilar to known classes of PPARγ ligands. "Potential" indicates that the chemical was identified in other screening approaches or by the ToxPi designed to identify chemicals in the ToxCast dataset that have potential to be PPARγ ligands/modifiers. c Abbreviations: AhR, aryl hydrocarbon receptor; ER, estrogen receptor; GR, glucocorticoid receptor; PR, progesterone receptor; PXR, pregnane X receptor; RAR, retinoic acid receptor; RXR, retinoid X receptor  Table S3. Metabolic parameters included and excluded in human transcriptome analysis.

PARAMETERS INCLUDED PARAMETERS EXCLUDED
Fat free mass % Body mass index (kg/m2) Fasting Plasma parameters Waist-to-hip ratio   (Table S1). On days 3, 5, and 7 of differentiation, the medium was replaced and the cultures re-dosed. Following 10 days of differentiation and dosing, cells were analyzed for lipid accumulation by Nile Red staining (Data are from Figure 1) and gene expression by 3'DGE. Each point represents the mean data for each chemical, (n=2-4). The least squares linear model estimate is shown in blue.

Figure S3. Lipid accumulation in differentiated and treated 3T3-L1 pre-adipocytes in the absence of dexamethasone.
Confluent 3T3 L1 cells were differentiated using a standard hormone cocktail for 10 days, with the exception of using no dexamethasone. During differentiation, cells were treated with vehicle (Vh, 0.2% DMSO, final concentration), rosiglitazone (positive control, 100 nM) or test chemical (Table S1). On days 3, 5, and 7 of differentiation, the medium was replaced and the cultures redosed. Following 10 days of differentiation and dosing, cells were analyzed for lipid accumulation by Nile Red staining. A) Nile Red staining induced by individual chemicals. Nile Red fluorescence was normalized by subtracting the fluorescence measured in naïve pre-adipocyte cultures within each experiment and reported as "Naïve Corrected RFU." Data are presented as mean ± SE (n=4). Statistically different from Vh-treated (highlighted in green) (*p<0.05, ANOVA, Dunnett's). B) Correlation between lipid accumulation induced in 3T3 L1 cells differentiated in the presence (data are from Figure 1) and absence of dexamethasone. Numerical data are provided in Excel File 3. Pearson's r = 0.5429 (p<0.0001).

Figure S4. Lipid accumulation in differentiated and treated OP9 pre-adipocytes.
Confluent OP9 cells were differentiated using a standard hormone cocktail for 10 days, with the exception of using 125 nM dexamethasone. During differentiation, cells were treated with vehicle (Vh, 0.2% DMSO, final concentration), rosiglitazone (positive control, 100 nM) or test chemical (Table S1). On days 3, 5, and 7 of differentiation, the medium was replaced and the cultures redosed. Following 10 days of differentiation and dosing, cells were analyzed for lipid accumulation by Nile Red staining. A) Nile Red staining induced by individual chemicals. Nile Red fluorescence was normalized by subtracting the fluorescence measured in naïve pre-adipocyte cultures within each experiment and reported as "Naïve Corrected RFU." Data are presented as mean ± SE (n=4). Statistically different from Vh-treated (highlighted in green) (*p<0.05, ANOVA, Dunnett's). B) Correlation between lipid accumulation induced in 3T3 L1 cells differentiated in the presence of dexamethasone (data are from Figure 1) and in OP9 cells. Numerical data are provided in Excel File 3. Pearson's r = 0.5768 (p<0.0001).

Figure S5. Performance comparison of random forest methods.
Boxplots of performance estimates for repeated 10-fold cross validation for each of the four random forest methods considered for classifying PPARγ activity modifying compounds from high-throughput gene expression profiles of chemically treated 3T3 L1 cells. Abbreviated metrics shown include: area-under the curve (AUC), balanced accuracy (Bal. Acc.), F1-score (F1). Besides AUC, for all performance metrics besides, an appropriate classification threshold for predicting PPARγ activity modifying compound labels in each test set was estimated based on out-of-bag voting of their corresponding training set. Sensitivity and specificity are the proportions of identified known PPARγ activity modifying compounds and known non-PPARγ activity modifying compounds, respectively, out of the total number of each label in the data. Bal. Acc. is the mean of sensitivity and specificity. Precision is the proportion of accurately identified known PPARγ activity modifying compounds out of all predicted PPARγ activity modifying compounds. F1 is the harmonic mean of sensitivity and precision. Finally, AUC is the integral of sensitivity and specificity across every possible classification threshold. The distribution of each method/metric combination is based on 10 repetitions of cross validation (i.e. N=10). The full set of performance estimates for each repetition, performance metric, and classification procedure considered are shown in Excel File 1 (Performance Comparison). The midline, box limits, and whiskers show the median, upper/lower quartile, and minimum/maximum of each distribution cutoff at a distance of 1.5 * the interquartile range from their closest box limit, respectively. Individual points indicate values which extend beyond the whisker limits. The horizontal gap in each plot indicates that, while the range of possible estimates for each performance metric is between 0.0 and 1.0, all estimates fell between 0.45 and 1.0.

Figure S6. Classification Results (Distributions of individual genes).
Confluent 3T3 L1 cells were differentiated using a standard hormone cocktail for 10 days. During differentiation, cells were treated with 0.1% DMSO, final concentration (vehicle), test chemical, or were untreated (naive). On days 3, 5, and 7 of differentiation, the medium was replaced and the cultures re-dosed. Following 10 days of differentiation and dosing, cells were analyzed for gene expression by 3'DGE. The labels, "Yes", "No", and "Potential", indicate test chemicals predetermined to be known PPARγ activity modifying compounds, known non-PPARγ activity modifying compounds, and Potential PPARγ activity modifying compounds, respectively, based on previous studies (See Table S1). Rpl13 and Cidec demonstrated the greatest predictive value for classifying PPARγ activity modifying compounds based on their Gini importance estimates from random forest modeling (Breiman 2001).  , TBBPA (20 μM) and TPhP (10 μM). On days 3, 5, and 7 of differentiation, the adipocyte maintenance medium was replaced and the cultures re-dosed. Cells were incubated for a total of 10 days of differentiation. To assess cell number, cells were stained with Janus green stain. Absorbance in experimental wells was normalized by dividing by the absorbance measured in naïve pre-adipocyte cultures within the experiment and reported as "Relative Cell Density." Numerical data are provided in Excel File 3. Data are presented as means ± SE (n=8). Statistically different from Vh-treated (**p<0.01, ANOVA, Dunnett's). Gene expression levels were normalized to the geometric mean of the expression levels of B2m and Rn18s and expressed as "Relative Expression" in comparison to naïve, pre-adipocyte cultures using the Pfaffl method. Numerical data are provided in Excel File 3. Data are presented as mean ± SE of n=6 independent experiments. Statistically different from Vh-treated (highlighted in green) (*p<0.05, **p<0.01, ***p<0.001, ****p<0.0001, ANOVA, Dunnett's). Figure S9. Spearman correlation analysis of lipid accumulation (Nile Red) and gene expression. Confluent 3T3 L1 cells were differentiated using a standard hormone cocktail with 1 nM dexamethasone for 10 days and analyzed for adipocyte differentiation by staining for lipids with Nile Red (shown in Figures 6 and 11) and gene expression (shown in Figures 7 and 11).
Figure S10. Cell number analyses in the differentiated and quinoxyfen and tonalide treated 3T3-L1s and human primary preadipocytes. (A) Confluent 3T3 L1 cells were differentiated using a standard human adipocyte hormone cocktail for 10 days. During differentiation, cells were treated with Vh (0.1% DMSO, final concentration), rosiglitazone (Rosig, 200 nM), quinoxyfen (Quino, 10 μM) or tonalide (Tonal, 4 μM). On days 3, 5, and 7 of differentiation, the adipocyte maintenance medium was replaced and the cultures re-dosed. Cells were incubated for a total of 10 days of differentiation. Data are presented as means ± SE (n=8). (B) Confluent primary human preadipocytes were differentiated using a standard hormone cocktail for 14 days. During differentiation, cells were treated with Vh (0.1% DMSO, final concentration), rosiglitazone (Rosig, 4 μM), quinoxyfen (Quino, 4 μM) or tonalide (4 μM). On days 3, 5, 7, 10, and 12 of differentiation, the medium was replaced and the cultures re-dosed. Following 14 days of differentiation and dosing, cultures were analyzed for relative cell density using JANUS Green staining. Absorbance in experimental wells was normalized by dividing by the absorbance measured in naïve pre-adipocyte cultures within the experiment and reported as "Relative Cell Density." Data are presented as mean ± SE (n=3, each n is from adipocytes from an individual). Statistically different from Vh-treated (**p<0.01, ANOVA, Dunnett's).