In vitro activity of a panel of per- and polyfluoroalkyl substances (PFAS), fatty acids, and pharmaceuticals in peroxisome proliferator-activated receptor (PPAR) alpha, PPAR gamma, and estrogen receptor assays

Abstract

Data demonstrate numerous per- and polyfluoroalkyl substances (PFAS) activate peroxisome proliferator-activated receptor alpha (PPARα), however, additional work is needed to characterize PFAS activity on PPAR gamma (PPARγ) and other nuclear receptors. We utilized in vitro assays with either human or rat PPARα or PPARγ ligand binding domains to evaluate 16 PFAS (HFPO-DA, HFPO-DA-AS, NBP2, PFMOAA, PFHxA, PFOA, PFNA, PFDA, PFOS, PFBS, PFHxS, PFOSA, EtPFOSA, and 4:2, 6:2 and 8:2 FTOH), 3 endogenous fatty acids (oleic, linoleic, and octanoic), and 3 pharmaceuticals (WY14643, clofibrate, and the metabolite clofibric acid). We also tested chemicals for human estrogen receptor (hER) transcriptional activation. Nearly all compounds activated both PPARα and PPARγ in both human and rat ligand binding domain assays, except for the FTOH compounds and PFOSA. Receptor activation and relative potencies were evaluated based on effect concentration 20% (EC₂₀), top percent of max fold induction (pmax_top), and area under the curve (AUC). HFPO-DA and HFPO-DA-AS were the most potent (lowest EC₂₀, highest pmax_top and AUC) of all PFAS in rat and human PPARα assays, being slightly less potent than oleic and linoleic acid, while NBP2 was the most potent in rat and human PPARγ assays. Only PFHxS, 8:2 and 6:2 FTOH exhibited hER agonism >20% pmax. In vitro measures of human and rat PPARα and PPARγ activity did not correlate with oral doses or serum concentrations of PFAS that induced increases in male rat liver weight from the National Toxicology Program 28-d toxicity studies. Data indicate that both PPARα and PPARγ activation may be molecular initiating events that contribute to the in vivo effects observed for many PFAS.

Keywords: Adverse outcome pathway; Bioactivity; GenX; Mechanism of action; Molecular initiating event.

Figure 1.

Binding assay activation of human and rat PPAR alpha (hPPARα and rPPARα) by PFAS, fatty acids and pharmaceuticals. Panels separated by test chemical subgroups including PFAS carboxylates (A, B); PFAS sulfonates and sulfonamides (C,D); fluorotelomer alcohols, fatty acids, and pharmaceuticals (E,F). Highest concentration tested was dependent on cytotoxicity. Data points were fit with four-parameter logistic regression using GraphPad Prism with model constraints top=100 and bottom=0.

Figure 2.

Scatterplots of human PPAR alpha (hPPARα) versus rat PPAR alpha (rPPARα) area under the curve (AUC) value (panel A) and hPPARα versus human PPAR gamma (hPPARγ) AUC values (panel B). Data points identified by compound and color coded by PFAS subgroup). Solid blue line represents linear regression for PFAS carboxylate and sulfonates because all fluorotelomer alcohols and sulfonamides were negative (no AUC, assigned values of 0.01 or 0.1 for visualization). Black dotted line represents a 1:1 line between x and y axes.

Figure 3.

Scatterplot of in vitro human PPAR alpha (hPPARα) effective concentration 20% (EC₂₀) and in vivo male rat liver weight effective dose 20% (ED₂₀, mg/kg/d, red circles) or EC₂₀ (serum μM concentration, blue triangles). Liver weight data from NTP 28-day studies of PFAS sulfonates and carboxylates (NTP 2019a, 2019b).

Figure 4.

Binding assay activation of human and rat PPAR gamma (hPPARγ and rPPARγ) by PFAS, fatty acids and pharmaceuticals. Panels separated by test chemical subgroups including PFAS carboxylates (A, B); PFAS sulfonates and sulfonamides (C,D); fluorotelomer alcohols, fatty acids, and pharmaceuticals (E,F). Highest concentration tested was dependent on cytotoxicity. Data points were fit with four-parameter logistic regression using GraphPad Prism with model constraints top=100 and bottom=0.

Figure 5.

Transcriptional activation of human estrogen receptor (hER) by PFAS with percent of maximum fold induction >5%. Highest concentration tested was dependent on cytotoxicity. Data points were fit with four-parameter logistic regression using GraphPad Prism with model constraints top=100 and bottom=0.