Technical-grade perfluorooctane sulfonate alters the expression of more transcripts in cultured chicken embryonic hepatocytes than linear perfluorooctane sulfonate

Abstract

Recently it was discovered that the perfluorooctane sulfonate (PFOS) detected in wildlife, such as fish-eating birds, had a greater proportion of linear PFOS (L-PFOS) than the manufactured technical product (T-PFOS), which contains linear and branched isomers. This suggests toxicological studies based on T-PFOS data may inaccurately assess exposure risk to wildlife. To determine whether PFOS effects were influenced by isomer content, we compared the transcriptional profiles of cultured chicken embryonic hepatocytes (CEH) exposed to either L-PFOS or T-PFOS using Agilent microarrays. At equal concentrations (10 µM), T-PFOS altered the expression of more transcripts (340, >1.5-fold change, p < 0.05) compared with L-PFOS (130 transcripts). Higher concentrations of L-PFOS (40 µM) were also less transcriptionally disruptive (217 transcripts) than T-PFOS at 10 µM. Functional analysis showed that L-PFOS and T-PFOS affected genes involved in lipid metabolism, hepatic system development, and cellular growth and proliferation. Pathway and interactome analysis suggested that genes may be affected through the RXR receptor, oxidative stress response, TP53 signaling, MYC signaling, Wnt/β-catenin signaling, and PPARγ and SREBP receptors. In all functional categories and pathways examined, the response elicited by T-PFOS was greater than that of L-PFOS. These data show that T-PFOS elicits a greater transcriptional response in CEH than L-PFOS alone and demonstrates the importance of considering the isomer-specific toxicological properties of PFOS when assessing exposure risk. Environ. Toxicol. Chem. 2011;30:2846–2859. © 2011 SETAC

INTRODUCTION

Perfluoroalkyl compounds (PFCs) are a synthetic class of compounds produced for use in nonstick, stain resistance, and other surface-tension-lowering applications. Because of their high production volumes and usage over the last 50 years and their inherent resistance to environmental and biological degradation, some PFCs have become persistent environmental contaminants 1. Structurally, PFCs are composed of a completely fluorinated carbon chain with a functional group at one terminal end of the molecule. Depending on the production method employed, the carbon backbone can be completely linear or in various branched arrangements. Perfluorooctane sulfonate (PFOS), the most prevalent PFC in wildlife tissues, was produced mainly by using an electrochemical fluorination process (ECF) that resulted in a technical product (T-PFOS) composed of approximately 60 to 70% linear isomer (L-PFOS), with the remaining 30 to 40% in different branched configurations 2.

Although most PFOS detected in the environment are thought to be of ECF origin, recent biomonitoring surveys found that the PFOS burden in wildlife, especially in animals at high trophic levels such as polar bears and fish-eating birds, had a much greater proportion of L-PFOS compared with manufactured T-PFOS 2–4. For example, Gebbink and Letcher 3 found that up to 98% of the PFOS burden measured in the eggs of herring gulls collected from the North American Great Lakes was in the linear form. Pharmacokinetic studies suggest that this may be due to the selective elimination of branched isomers and retention of L-PFOS, especially in the liver, where PFOS tends to accumulate 5–7, in exposed animals and in their prey 4.

Before the pharmacokinetic differences between PFOS isomers were recognized, most laboratory toxicology studies exposed test animals or cultured cells to T-PFOS. Effects from such studies include interference with lipid and carbohydrate metabolism, increased liver weight, hepatomegaly, disruption of the neuroendocrine system, and decreased reproductive success 8–11. For birds, exposure to T-PFOS can cause hepatic lesions, reduced hatching success, and reduced survivability of hatchlings 12–16. These experiments, however, might not accurately reflect natural exposure conditions experienced by wildlife because they do not consider PFOS isomer kinetics and therefore may lead to poor estimations of exposure effects. This is of particular relevance for conclusions drawn from in vitro experiments in which no selective accumulation of PFOS isomers occurs.

Several recent studies have shown that PFCs can elicit different molecular and physiological effects depending on the isomeric composition. Loveless et al. 17 observed increased weight loss and more robust lipid oxidation in rodents orally exposed to linear ammonium perfluorooctanoate compared with animals exposed solely to branched isomers. In contrast, Hickey et al. 18 found that genes involved in lipid and xenobiotic metabolism had a more profound transcriptional response in cultured chicken embryonic hepatocytes when exposed to T-PFOS than when exposed to L-PFOS.

The aim of the present study was to characterize further the differences in transcriptional response between L-PFOS and T-PFOS in a cell culture model used routinely in our laboratory for screening chemicals for transcriptional effects in avian species: cultured chicken embryonic hepatocytes (CEH). We used DNA microarrays to identify system-wide differences in transcriptome response to L-PFOS or T-PFOS exposure, with emphasis on pathways and functionally related genes associated with PFOS toxicity. We also examined the potential role of regulatory proteins in the CEH response to L-PFOS or T-PFOS.

MATERIALS AND METHODS

Chemicals

Technical-grade PFOS was obtained from Wellington Laboratories. The proportions of each isomer in this T-PFOS mixture were determined by gas chromatography–mass spectrometry and reported by O'Brien et al. 7. Briefly, T-PFOS comprised 65% linear and 35% branched isomers, most of which were mono(trifluoromethyl)-branched isomers. Linear PFOS (>98% purity) was also obtained from Wellington Laboratories. Working solutions for dosing CEH were prepared by dissolving T-PFOS or L-PFOS in dimethyl sulfoxide (DMSO). The preparation of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) solutions in DMSO has been described elsewhere 19. The TCDD exposures were used as positive controls for cytochrome P4501A4 (CYP1A4) and CYP1A5 induction for microarray validation purposes.

Preparation of hepatocyte cultures and dosing

Primary CEH cultures were prepared from the livers of day 19 chicken embryos as previously described 18. Briefly, 60 fertilized white leghorn chicken (Gallus gallus domesticus) eggs, obtained from the Canadian Food Inspection Agency, were incubated at 37.5°C and 60% humidity for 19 d. On day 19, embryos were euthanized by decapitation, and livers were removed and subsequently pooled. Hepatocytes were isolated from pooled livers by collagenase digestion then cultured in 48-well plates. Each well contained approximately 780 µg hepatocytes in 525 µl medium. Cultured cells were incubated for 24 h at 37°C and 5% CO₂. All procedures were conducted according to protocols approved by the Animal Care Committee at the National Wildlife Research Centre, Ottawa, Ontario, Canada.

Cells were dosed with a DMSO solvent control or working solutions of T-PFOS, L-PFOS or TCDD dissolved in DMSO and incubated for an additional 24 h. Cultures for microarray analysis were exposed to 10 and 40 µM of T-PFOS or L-PFOS and 0.03 and 1.0 nM TCDD (n = 5 per dose group). These concentrations were selected because they were the lowest levels at which transcriptional effects were observed in cultured CEH as reported elsewhere 18. Cultures for examining dose–response relationships using real-time RT-PCR were exposed to 1, 10, 20, 30, and 40 µM of T-PFOS or L-PFOS (n = 2–3 per dose group). After incubation, medium was removed, and cells were frozen on dry ice and stored at −80°C. The viability of cells exposed to T-PFOS, L-PFOS or TCDD was not affected by any of the concentrations tested based on the calcein-AM assay 18.

RNA isolation and quantification

Total RNA was isolated from CEH using RNeasy 96 kits (Qiagen), including the on-column DNase treatment according to the manufacturer's instructions. After isolation, a second DNase treatment was performed using a DNA-free kit (Ambion) as per manufacturer's instructions. RNA was quantified with a NanoDrop 2000 spectrophotometer (Thermo Scientific). RNA quality was assessed using a BioAnalyzer (Agilent Technologies). Samples with 260/280 ratios <1.7 and RIN <8.0 were not used for downstream applications. A reference pool of RNA for microarray hybridizations was prepared from equal parts of all samples used for microarray analysis.

Microarray hybridization

Complementary RNA (cRNA) for each microarray hybridization was prepared from 150 ng total RNA and hybridized to Agilent 4 × 44 k chicken whole-genome arrays (design 015068; Agilent Technologies) against cRNA prepared from 150 ng of the reference pool RNA. cRNA from experimental samples was labeled with Cy5 and reference cRNA with Cy3 using Agilent Quick Amp labeling kits according to the manufacturer's instructions. Labeled sample and reference cRNA (825 ng each) was fragmented and then hybridized to arrays for 17 h using Agilent Hybridization kits according to the manufacturer's directions. Arrays were washed and then scanned on an Agilent G2505B scanner at 5 µm resolution. Data were acquired using Agilent Feature Extraction software version 9.5.3.1.

Microarray data analysis

A reference design was used to analyze gene expression data 20. The design was blocked on the slide, because Agilent slides contain four arrays per slide. Background fluorescence was measured using the (-)3xSLv1 probes. Probes with median signal intensities less than the trimmed mean (trim = 5%) plus three trimmed standard deviations of the (-)3xSLv1 probe were flagged as absent. Data were preprocessed using R (http://www.R-project.org). The median signal intensities were normalized by the global LOWESS method using the transform.madata function in the microarray analysis of variance (MAANOVA) library 21. Ratio intensity plots, using complete and single linkages, were generated to identify arrays with poor quality. Differentially expressed genes were identified using the MAANOVA library. The analysis of variance (ANOVA) model included the main effect of treatment and block effect of the microarray. The F_s statistic 22, a shrinkage estimator, was used for the gene-specific variance components, and the associated p values for all the statistical tests were estimated using the permutation method (30,000 permutations with residual shuffling). These p values were then adjusted for multiple comparisons using the false discovery rate (FDR) approach 23. The least-squares means was used to estimate the fold changes for each pairwise comparison.

Principal component analysis (PCA) and hierarchical clustering were performed using GeneSpring GX version 11.0.2 (Agilent Technologies). Clustering was performed on both entities and conditions, using the Euclidian distance metric and the centroid linkage rule.

Before analysis via Ingenuity Pathway Analysis (IPA; ver. 8.6), the DAVID Gene ID Conversion Tool (http://david.abcc.ncifcrf.gov/) was used to optimize the number of microarray gene IDs that were mapped as human, rat, or mouse orthologs in IPA. The functional and canonical pathway analysis performed in IPA identified biological functions/diseases or pathways that were most significant to the dataset. Molecules from the data set that were recognized as human, mouse, or rat orthologs, met the fold change cutoff of 1.5 and FDR p value cutoff of 0.05, and were associated with biological functions and/or diseases in Ingenuity's proprietary Knowledgebase were considered for the analysis. A right-tailed Fisher's exact test was used to calculate a p value determining the probability that each biological function/disease or pathway assigned to that data set was due to chance alone.

For interaction network generation in IPA, differentially expressed genes that mapped into IPA were overlaid onto a global molecular network developed from information contained in Ingenuity's Knowledgebase. Networks of differentially expressed genes were then algorithmically generated based on their connectivity. In the generated network diagrams, genes are represented as nodes, and the biological relationship between two nodes is represented as an edge (line). All edges are supported by at least one reference from the literature, from a textbook, or from canonical information stored in the Ingenuity Knowledgebase. Nodes are displayed using various shapes that represent the functional class of the gene. The intensity of the node color indicates the degree of upregulation (red) or downregulation (green). White nodes are not differentially expressed and are added to the network as bridging/connector molecules.

Real-time RT-PCR

Total RNA (150 ng) was reverse-transcribed to complementary DNA (cDNA) using SuperScript II reverse transcriptase and random hexamer primers (Invitrogen Canada) according to the manufacturer's instructions with no-RT controls prepared in tandem. cDNA was diluted (1:20) and aliquoted for real-time RT-PCR. Real-time RT-PCRs were performed using a Stratagene Mx3000 or Eppendorf Mastercycler ep Realplex real-time PCR instrument. Assays for each gene of interest were duplexed with β-actin as the normalizing gene using TaqMan fluorogenic probes. It was previously demonstrated that β-actin mRNA expression is stable in CEH at the concentrations of PFOS used in the present study 24. This is further confirmed by our microarray expression data. Probes for genes of interest were labeled with FAM; β-actin probes were labeled with HEX (Mx3000) or JOE (Realplex). Reactions were performed using Stratagene Brilliant Q-PCR Core Reagent kits. Each reaction contained primers (Invitrogen), probes (Biosearch), 5 mM MgCl₂, 1× reaction buffer, 0.8 µM dNTPs, 8% glycerol, and 1.25 U SureStart Taq and brought to volume with diethylpyrocarbonate-treated water. Reactions performed in the Mx3000 had a total volume of 25 µl and contained 60 nM ROX reference dye and 5 µl diluted cDNA. Reactions performed in the Realplex instrument had a total volume of 12.5 µl and contained 2.5 µl diluted cDNA. The nucleotide sequence for each primer and probe and their final reaction concentrations are given in Supplemental Data, Table S1. The thermal profile included a 10-min activation step at 95°C, followed by 40 cycles of 30 s at 95°C and 1 min at 60°C. Fluorescence was measured after the 60°C step. Amplification efficiencies for each gene target were determined by standard curves and optimized by adjusting primer concentrations so that efficiencies were approximately 100% (±5%) and similar to the efficiency of β-actin (±5%). Relative gene expression values were calculated using the 2^−ΔCt equation 25 and expressed as fold induction relative to the DMSO control group. Data are reported as the mean and standard deviation of two to four replicates. The DMSO groups were n = 3. Because some dose groups had only n = 2, the normality assumption could not be justified. Statistical analysis therefore used a nonparametric approach. A one-way ANOVA was conducted on the ranks of the ΔCt values as described elsewhere 26, 27. This analysis was conducted in R. All pairwise comparisons with the control were conducted using the multcomp library 28, in which the Dunnett method 29 was used to adjust p values for multiple comparisons.

RESULTS AND DISCUSSION

Differentially expressed genes

Agilent 4 × 44k Chicken Gene Expression microarrays were used to identify changes in gene transcription in CEH exposed for 24 h to different isomeric mixtures of PFOS relative to a vehicle control group. Array hybridizations were performed using RNA from CEH exposed to 10 µM L-PFOS, 40 µM L-PFOS, and 10 µM T-PFOS (henceforth referred to as L10, L40, and T10, respectively) and the DMSO solvent control (n = 5 per dose group). Cells exposed to 40 µM T-PFOS had a statistically significant decrease in total RNA yield (identified by ANOVA, p < 0.05) and could therefore not be used for the microarray analysis. This was not observed in the independent culture used to investigate dose–response relationships. Therefore, it is unclear whether this effect was due to T-PFOS exposure.

Microarray data were submitted to the National Center for Biotechnical Information Gene Expression Omnibus following MIAME protocols. A total of 447 probe sets showed differential expression >1.5-fold (FDR p < 0.05) in CEH following exposure to either L-PFOS or T-PFOS relative to the DMSO control group. Among these, 334 genes were upregulated and 113 were downregulated. One hundred seven genes were dysregulated (up- or downregulated) by L-PFOS only (both doses), 178 affected only by T-PFOS, and 162 affected by both L-PFOS and T-PFOS (Fig. 1). The numbers of probes up- and downregulated in each treatment are summarized in Table 1. Perfluorooctane sulfonate T10 altered the expression of approximately 2.5 times as many probes as L10 and 1.5 times as many probes as L40. A detailed list of the probes differentially expressed by PFOS-treated cells is presented in Supplemental Data, Table S2.

Figure 1. Venn diagram illustrating the numbers of genes that were dysregulated (fold change >1.5, p < 0.05) by either 10 or 40 µM linear perfluorooctane sulfonate (L10 and L40, respectively) or 10 µM technical-grade perfluorooctane sulfonate (T10).

Table 1. Number of genes differentially expressed (fold change >1.5, p < 0.05) in chicken embryonic hepatocytes exposed to 10 or 40 µM linear or 10 µM technical-grade perfluorooctane sulfonate (L10, L40, or T10, respectively)

Treatment	No. of genes
	Upregulated	Downregulated
L10	104	26
L40	120	97
T10	278	62

The highest increase in expression was for myosin heavy chain 6 (Myh6), which was greater than threefold in all treatment groups. Typically, Myh6 is expressed only in ventricular and extraocular muscle 30. However, one EST, ChEST143b11, a possible splice variant consisting of Myh6 intron and exon sequence, has been detected in chicken liver tissue 31. The probe for this variant also showed differential expression, although significantly only in T10 (1.7-fold increase, gene ID CR353427). These results indicate that PFOS interferes with the transcriptional regulation of this gene locus. The role of Myh6 in the response to PFOS is unclear. It is a type II myosin usually associated with contractile activity. It may be involved in the contraction of bile canaliculi 32, which, if improperly regulated, can lead to intrahepatic cholestasis and interfere with lipid metabolism. Functional analysis revealed that several other genes associated with intrahepatic cholestasis were also dysregulated by PFOS (shown below under Functional analysis).

The largest increase in expression that was unique to L-PFOS was a 2.2-fold increase in the abundance of a transcript identified as LOC770996. This transcript encodes a protein that has 80% sequence identity to rat and mouse L-gulonolactone oxidase (GULO), an enzyme involved in L-ascorbate (vitamin C) production. Although not typically produced in chicken liver 33, L-ascorbate plays several important biological roles, including oxidative stress relief and collagen synthesis. This coincides with observations that PFOS can increase production of reactive oxygen species and induce related genes 34, 35. Two other genes downstream of L-ascorbate production were also affected by L-PFOS and T-PFOS (Supplemental Data, Fig. S1), providing further evidence that this pathway was affected by PFOS exposure. Why expression of LOC770996 was not affected by T-PFOS in the present study is unclear. It is possible that this particular gene is affected only by higher exposures to L-PFOS, which are not reached using the impure T-PFOS, or perhaps upregulation of this gene is somehow prevented by branched PFOS isomers.

The largest unique increase to T-PFOS was a 2.2-fold increase in the expression of type II iodothyronine deiodinase (Dio2), which converts thyroxine to its more active form, tri-iodothyronine. This is consistent with observations of reduced serum thyroxine in animals exposed to PFOS 36, 37. A significant increase in expression for a gene with high sequence similarity to thyroid hormone receptor beta 2 (gene ID X62642) was also observed only in the T-PFOS dose group. The fact that Dio2 and X62642 were perturbed only by T-PFOS, and not by L-PFOS, suggests that interference with active thyroid hormone levels may be instigated by the branched isomers of PFOS. Despite the presence of thyroxine in the culture medium (1 µg/ml) and changes in Dio2 and X62642 expression, an accompanying change in expression for other thyroid hormone receptor-controlled genes was not observed.

The largest overall decrease in expression was an approximately fourfold decrease in glutathione-S-transferase α3 (Gsta3) mRNA in all treatment groups. Downregulation of several other glutathione-S-transferases (GSTs) was observed, including the α4 and ω1 isozymes (Gsta4 and Gsto1, respectively). These genes belong to a class of enzymes that detoxifies reactive electrophilic compounds by catalyzing their conjugation to reduced glutathione (GSH). Liu et al. 34 observed a decrease in total GST activity in response to PFOS exposure in cultured tilapia hepatocytes as well as a decrease in GSH levels. The authors suggested that GST activity was likely diminished to help conserve unconjugated GSH as a response to the oxidative stress resulting from increased lipid metabolism. The decrease in GST mRNA expression in the present study supports this hypothesis. In addition to its role in detoxification, the protein encoded by Gsta3 also has steroid isomerase activity crucial for steroid hormone production, which consumes cholesterol 38. Suppression of Gsta3 expression may also serve to maintain cholesterol levels, which are known to decrease following PFOS exposure 39, 40. This is supported in the present study by the upregulation of several genes involved in cholesterol synthesis, including 3-hydroxy-3-methylglutaryl-CoA reductase (Hmgcr), cytochrome P4508B1 (Cyp8B1), and isopentenyl-diphosphate δ isomerase 1 (Idi1), although these three genes were affected only by T-PFOS, whereas Gsta3 was affected by both L- and T-PFOS. Because GSTs were downregulated in all treatments, their response may be independent of the presence of branched PFOS isomers. The largest decreases in expression detected that were unique to L-PFOS or T-PFOS were 1.9-fold and 1.7-fold decreases in the uncharacterized transcripts KLHDC8B and CN236254, respectively.

Microarray results were validated by real-time RT-PCR analysis for 15 PFOS-responsive genes. In addition to the PFOS samples, the transcriptional response of cytochrome P450 isoforms 1A4 and 1A5 in CEH exposed to TCDD was included for microarray validation purposes. Validation was performed using the same RNA samples used for microarray analysis (n = 2–4). Real-time RT-PCR results were directionally consistent with microarray data for 35 of the 48 conditions (Supplemental Data, Table S3), resulting in a validation rate of 73%. To supplement the microarray results, the expression of four genes was further examined in a separate independent cell culture using a more comprehensive range of exposure concentrations (1–40 µM) to investigate dose–response relationships (n = 2–3). The genes acyl-CoA synthetase long-chain family member 1 (Acsl1), acyl-CoA synthetase bubble gum family member 2 (Acsbg2), amphiregulin (Areg), and Gsta3 were selected for analysis because results for PCR validation closely matched the microarray data both in direction and in significance (see Supplemental Data, Table S3). Both Acsl1 and Acsbg2 genes are involved in fatty acid metabolism. Areg encodes a growth factor that is often associated with liver injury 41, whereas Gsta3, as previously mentioned, is involved in the oxidative stress response. The expression profiles for all four genes were similar to those observed from the microarray analysis (Fig. 2); however, the concentration at which significant effects were detected was higher (with the exception of Gsta3). This likely is due to the variability between the independent cell cultures and the lower samples size for cultures used to examine the broader range of doses. Both L-PFOS and T-PFOS downregulated Gsta3, which is consistent with the microarray analysis, although this effect was not apparent at higher concentrations of T-PFOS. Significant dose-dependent increases in the expression Acsl1, Acsbg2 and Areg were observed in cells exposed to T-PFOS, whereas expression was not affected by L-PFOS, as predicted from the microarray results.

Figure 2. Expression of Acsl1, Acsbg2, Areg, and Gsta3 in cultured chicken embryo hepatocytes following exposure to linear or technical-grade perfluorooctane sulfonate (L-PFOS or T-PFOS, respectively). Data represent the mean and standard deviation from two or three replicates.

PCA and clustering analysis

The expression profiles of all 20 microarray samples (five per dose group) were reduced to four principal components using the principal component analysis function in GeneSpring GX. Samples are plotted on the first three components in Figure 3. The first principal component (PC1) accounted for 57.6% of the variability among samples. A large proportion of differentially expressed transcripts (379/447), including all 74 that were affected by all three treatments, had absolute PC1 loadings greater than 0.5 (data not shown). This shows that PC1 largely characterized the difference between DMSO- and PFOS-treated samples. Accordingly, all PFOS-treated samples segregated from DMSO samples along the PC1 axis. The L10 samples were an intermediate distance from DMSO samples, whereas L40 and T10 clustered farthest from DMSO along PC 1 but were not separated, suggesting similar expression profiles between these two groups. The second principal component (PC2) accounted for 20.8% of intersample variability, and 115 transcripts had PC2 loadings greater than 0.5. Among these, 87% (100) were differentially expressed by either L-PFOS or T-PFOS, but not by both, demonstrating that PC2 describes mainly the variability between samples treated with different isomeric mixtures of PFOS. This is exemplified by the almost complete separation of the equal-dose samples, L10 and T10, along PC2. Again, L40 and T10 samples were not separated along PC2. The third and fourth principal components (PC3 and PC4) accounted for 11.1% and 10.5% of variability, respectively. However, no major trends in sample separation were observed along PC3 or PC4, suggesting that they captured residual intersample variability.

Figure 3. Expression profile of linear or technical-grade perfluorooctane sulfonate-treated (L-PFOS or T-PFOS, respectively) chicken embryo hepatocytes plotted on three principal components compared with the dimethyl sulfoxide (DMSO) control group. Analysis was based on the expression profiles of 447 dysregulated genes. [Color figure can be seen in the online version of this article, available at wileyonlinelibrary.com.]

Hierarchical clustering using the differentially expressed gene set was also performed in GeneSpring GX. The expression profile of each sample was clustered based on gene expression and treatment (Fig. 4). Cluster analysis separated DMSO-treated CEH from all PFOS-treated samples. Similar to PCA analysis, L10 and T10 clustered separately. However, hierarchical clustering could not separate L40 samples from the other treatment groups: two L40 samples clustered with L10, whereas the other three clustered with T10. Similar conclusions can be drawn from PCA and hierarchical clustering. Both methods show that T10 and L10 have distinct expression profiles. However, PCA could not distinguish L40 from T10, whereas clustering showed L40 to be intermediate between L10 and T10.

Figure 4. Hierarchical clustering of expression profiles of chicken embryonic hepatocytes exposed to linear and technical-grade perfluorooctane sulfonate (L-PFOS and T-PFOS, respectively) and dimethyl sulfoxide (DMSO) vehicle control. Clustering was based on 447 differentially expressed genes (fold change >1.5, p < 0.05). [Color figure can be seen in the online version of this article, available at wileyonlinelibrary.com.]

Mapping gene IDs to IPA

The DAVID Gene ID Conversion Tool was used to map microarray gene IDs to human, rat, or mouse orthologs in IPA. For all genes on the array, 12,615 of 42,034 genes (30%) were recognized by IPA as orthologs. Among all dysregulated genes, 197 of 447 (44%) were recognized as orthologs in IPA. Genes that mapped to IPA are identified in the list of differentially expressed genes (Supplemental Data, Table S2). Most unmapped IDs correspond to probes for hypothetical proteins and ESTs with unknown function. Several differentially expressed genes with known functions, however, did not map to orthologs in IPA, such as liver basic fatty acid binding protein (gene ID: LBFAPB) and thyroid hormone receptor β2 (gene ID: X62642), and therefore could not be included in downstream functional, pathway, and interactome analysis.

Functional analysis

Ingenuity Pathway Analysis was used to perform a functional enrichment analysis. The relevant functional categories are summarized in Table 2. The specific functions that were significantly enriched in each category are shown in detail in Supplemental Data, Table S4A–C. The top functional categories were lipid metabolism, hepatic system development and disease, cellular movement, and cellular growth and proliferation. Although a similar set of functional categories was enriched for both L-PFOS and T-PFOS, in almost all cases, the number of genes affected in each category was greatest in T10 followed by L40 and then L10. In general, the functional profiles are consistent with similar gene expression studies for mammalian and avian models 37, 42–44. Changes in genes associated with lipid metabolism and cellular proliferation are generally consistent with the activation of peroxisome proliferator activated receptors (PPARs), particularly the PPARα and PPARγ isotypes 45. However, one important difference from mammalian PFOS studies is the lack of response of genes involved in the peroxisomal β-oxidation of fatty acids that are transcriptionally regulated by PPARα. It is believed that much of the toxicity of PFOS is propagated through activation of PPARα in rodents. Differential expression of the β-oxidation genes acyl-CoA oxidase (ACOX), enoyl-CoA hydratase, bifunctional enzyme (BIEN), and peroxisomal 3-ketoacyl-CoA thiolase has been reported in rat liver and cultured rat hepatoma cells following PFOS exposure 42, 43. Other avian gene expression studies, however, reported no apparent correlation between PFOS exposure and β-oxidation 15, 44, 46, although a pair of studies performed in our laboratory reported T-PFOS-induced changes in β-oxidation genes in cultured CEH 18, 24. These changes were significant only after 36 h of exposure at ≥40 µM, which is a longer exposure time and higher concentration than were used in the present study. These results indicate that activation of PPARα may occur as a secondary response or only at high concentrations of PFOS. For 24 h of exposure to 10 µM T-PFOS, the results of Cwinn et al. 24 are in accordance with the present study, including a two to threefold change in liver basic fatty acid binding protein (LBFABP), which was suggested to be independent of PPARα. In the present study, most of the lipid metabolism functions that were enriched were involved not in β-oxidation but rather in the synthesis, modification, and transport of lipids, fatty acids, terpenoids, eicosanoids, steroids. and cholesterol. The differential expression of a similar gene set was also reported for rat hepatic cells exposed to PFCs 37, 42 and may represent PPARα-independent effects on lipid metabolism. The role of PPARα in the avian hepatocyte response to PFOS exposure is addressed further below under Interaction networks and potential regulatory molecules.

Table 2. Enriched functional categories for genes differentially expressed in chicken embryonic hepatocytes due to exposure to 10 or 40 µM linear or 10 µM technical-grade perfluorooctane sulfonate (L10, L40, or T10, respectively)

Rank	Category	L10		L40		T10
		p Value range	No. of genes	p Value range	No. of genes	p Value range	No. of genes
1	Lipid metabolism	2.82E-03–4.20E-02	11	9.93E-05–1.57E-02	18	5.71E-07–1.10E-02	26
2	Hepatic system development and disease	2.35E-03–2.30E-02	3	4.72E-06–1.02E-02	12	3.39E-05–1.10E-02	12
3	Cellular movement	1.85E-02–1.85E-02	4	5.76E-06–1.57E-02	15	4.72E-05–7.62E-03	18
4	Cellular growth and proliferation	2.81E-02–2.81E-02	1	1.41E-04–1.57E-02	23	2.23E-05–1.10E-02	37
5	DNA replication, recombination, and repair	4.73E-03–4.18E-02	1	1.38E-03–1.57E-02	10	1.13E-04–1.10E-02	12
6	Molecular transport	4.73E-03–4.63E-02	9	2.54E-04–1.57E-02	13	1.20E-04–1.10E-02	16
7	Cancer			6.76E-04–1.55E-02	24	1.44E-04–1.10E-02	31
8	Cell development and morphology	1.41E-02–1.41E-02	1	1.50E-04–8.19E-03	10	2.21E-03–1.10E-02	18
9	Carbohydrate metabolism	9.44E-03–4.18E-02	5	4.09E-04–1.57E-02	11	1.10E-03–1.10E-02	7
10	Cell-to-cell signaling and interaction	4.73E-03–2.34E-02	2	4.37E-04–7.88E-03	8	4.72E-03–1.10E-02	12
11	Amino acid metabolism	4.73E-03–4.63E-02	4	8.96E-04–1.57E-02	5	8.66E-03–1.10E-02	3
12	Tissue development and morphology	2.34E-02–2.34E-02	1	1.06E-02–1.37E-02	7	1.28E-03–3.83E-03	5
13	Small-molecule biochemistry	4.73E-03–4.18E-02	4	2.12E-03–1.57E-02	6	4.11E-03–1.10E-02	8
14	Cellular assembly and organization			7.88E-03–1.57E-02	2	2.43E-03–1.10E-02	3
15	Endocrine system development and function	4.73E-03–2.34E-02	3	7.88E-03–7.88E-03	2	3.22E-03–1.10E-02	3
16	Inflammatory response			4.52E-03–6.05E-03	10	6.75E-03–6.75E-03	7
17	Nucleic acid metabolism	4.73E-03–9.44E-03	2	7.88E-03–7.88E-03	3	1.10E-02–1.10E-02	4
18	Drug metabolism			7.88E-03–1.57E-02	2

The large number of dysregulated genes involved in cellular growth and proliferation, hepatic system disease, and DNA replication, as well as an enrichment of cancer-related genes in the L40 and T10 groups, is consistent with the hepatomegaly, hepatocellular adenoma, and hyperplasia that have been observed in mammals 10, 40, 47 and birds 12–14, 16 exposed to PFOS. Several genes in the hepatic system and disease category were also associated with intrahepatic cholestasis (Supplemental Data, Table S4A–C), which, as previously discussed, may be a result of Myh6 overexpression. Although cellular movement was a significantly enriched functional category, most genes from this group are involved in tumor cell migration and cell–cell signaling. Changes in other cell–cell interaction genes such as gap junction proteins, including gap junction protein β1 (Gjb1, also known as Cx32), cell adhesion molecule 1 (Cadm1), and other cell–cell signaling genes are in agreement with impaired gap junction intercellular communication observed in liver tissue and cultured cells 48.

Canonical pathway mapping

Ingenuity pathway analysis was used to map significantly dysregulated genes to canonical pathways in Ingenuity's Knowledgebase. Ingenuity pathway analysis revealed that exposure to L-PFOS and T-PFOS differentially regulated genes belonging to several pathways. Canonical pathways that were significantly affected (p < 0.05) are summarized in Table 3. A more detailed description of all pathways affected by each dose group and the genes that were affected is shown in Supplemental Data, Table S5A–C. Pathways that were most significantly disrupted by PFOS exposure included LPS/IL-1-mediated inhibition of RXR function, glutathione metabolism, LXR/RXR activation, and biosynthesis of steroids. Although these pathways were significantly enriched, the number of genes affected compared with the total number in each pathway is relatively low. This would suggest that neither of these pathways represents the main mechanism of action of PFOS. For example, although the LPS/IL-1-mediated inhibition of RXR function pathway was the most significantly affected, it was actually the genes downstream from this pathway, regulated by RXR's binding partners, CAR, PXR, and PPARs, that were disrupted and not genes involved in the inhibition mechanism (Supplemental Data, Fig. S2). The third most significantly enriched pathway was LXR/RXR activation. These results suggest that PFOS may exert its effect through RXR and its binding partners. However, binding studies have shown that PFOS does not activate human, rat, or mouse RXR 49. Perfluorooctane sulfonates may interfere with RXR heterodimer formation through its binding partners. Perfluorooctane sulfonate does show weak binding to both PPARα and PPARγ 49. Binding studies with CAR, PXR, and LXR would also clarify this issue.

Table 3. Significantly enriched (p < 0.05) canonical pathways for genes differentially expressed in chicken embryonic hepatocytes exposed to 10 or 40 µM linear or 10 µM technical-grade perfluorooctane sulfonate (L10, L40, or T10, respectively).

Ingenuity canonical pathways	L10		L40		T10		No. of genes in pathway
	p Value	No. of genes	p Value	No. of genes	p Value	No. of genes
LPS/IL-1-mediated inhibition of RXR function	3.16E-02	3	9.77E-04	6	1.95E-04	8	149
Glutathione metabolism	1.62E-02	2	2.95E-04	4	1.05E-02	3	42
LXR/RXR activation	2.39E-01	1	1.02E-03	4	3.89E-04	5	59
Biosynthesis of steroids			1.11E-01	1	5.13E-04	3	22
Metabolism of xenobiotics by cytochrome P450	8.71E-03	3	6.92E-04	5	3.16E-03	5	92
Aryl hydrocarbon receptor signaling	8.51E-02	2	8.51E-03	4	8.91E-04	6	105
Pentose and glucuronate interconversions			1.15E-03	3	3.02E-03	3	29
Pyruvate metabolism			4.68E-03	3	1.26E-03	4	51
Bladder cancer signaling			5.01E-03	3	1.38E-03	4	45
Cysteine metabolism	1.19E-01	1	1.86E-02	2	3.02E-03	3	28
NRF2-mediated oxidative stress response	2.19E-02	3	3.09E-03	5	5.13E-02	4	129
Xenobiotic metabolism signaling	2.19E-01	2	3.16E-03	6	1.62E-02	6	188
Ascorbate and aldarate metabolism	5.89E-02	1	4.37E-03	2	1.33E-01	1	16
Fatty acid metabolism	3.77E-01	1	4.27E-02	3	4.47E-03	5	100
Pancreatic adenocarcinoma signaling			1.74E-02	3	7.24E-03	4	71
Glycerolipid metabolism	2.36E-01	1	9.77E-03	3	2.40E-02	3	58
PXR/RXR activation	3.09E-02	2	1.07E-02	3	4.78E-01	1	60
Glycine, serine, and threonine metabolism			2.86E-01	1	1.12E-02	3	50
Bile acid biosynthesis	1.41E-02	2	3.63E-02	2	6.76E-02	2	47
Ovarian cancer signaling			1.55E-02	3	3.80E-02	3	68
Hepatic fibrosis/hepatic stellate cell activation			1.52E-01	2	1.58E-02	4	89
Galactose metabolism			1.58E-02	2	3.02E-02	2	30
Wnt/β-catenin signaling			5.08E-01	1	1.66E-02	4	90
Fructose and mannose metabolism			2.57E-02	2	4.68E-02	2	35
Taurine and hypotaurine metabolism	2.75E-02	1	4.57E-02	1	6.31E-02	1	9
Nicotinate and nicotinamide metabolism	2.95E-02	2			4.72E-01	1	59
Airway pathology in chronic obstructive pulmonary disease			3.09E-02	1	4.27E-02	1	4
p53 Signaling			3.95E-01	1	3.24E-02	3	64
IGF-1 Signaling	2.70E-01	1	9.55E-02	2	3.63E-02	3	67
Chronic myeloid leukemia signaling			4.09E-01	1	3.63E-02	3	67
Lysine biosynthesis	3.72E-02	1			8.32E-02	1	8
Melanoma signaling			2.03E-01	1	3.98E-02	2	29
Synthesis and degradation of ketone bodies	4.17E-02	1			9.33E-02	1	10
Chondroitin sulfate biosynthesis			2.09E-01	1	4.17E-02	2	30
Keratan sulfate biosynthesis			2.09E-01	1	4.17E-02	2	30
Cell cycle: G2/M DNA damage checkpoint regulation			2.09E-01	1	4.17E-02	2	30
Propanoate metabolism			4.79E-02	2	8.51E-02	2	49

Three of the four genes affected in the glutathione metabolism pathway were glutathione-S-transferases, which, as previously mentioned, may be downregulated in order to conserve unconjugated GSH. The fourth was glutaredoxin (Glrx), which was downregulated 1.7-fold in L40. Because Glrx catalyzes the reduction of disulfide bonds, consuming GSH in the process, this enzyme may also be downregulated in order to conserve GSH. We discussed above the upregulation of genes involved in cholesterol production, and pathway analysis demonstrated that all genes affected in the steroid biosynthesis pathway are upregulated and have upstream roles in the production of cholesterol. Again, these may be upregulated as a negative-feedback response to the decrease in cholesterol that often accompanies PFOS exposure. The impact on steroid biosynthesis genes was greatest in T10, suggesting that branched isomers have the greatest effect on these genes. As with the functional analysis, a general trend in the number of genes affected in each pathway was as follows: L10 < L40 < T10.

Interaction networks and potential regulatory molecules

Ingenuity pathway analysis was used to generate interaction networks with mapped orthologs of differentially expressed genes for all three treatment groups. Only direct interactions with gene products that occur in the liver were considered for interaction network construction. An example of one of these networks is shown in Figure 5. This example is one of two networks generated using T10 data. A high-resolution version of Figure 5, as well as all networks examined in the present study, can be found in Supplemental Data, Figures S3–S7.

Figure 5. One of two Ingenuity pathway analysis (IPA)-generated interaction networks for genes dysregulated by exposure to technical-grade perfluorooctane sulfonate, 10 µM. Red shapes represent genes that were significantly upregulated, and green shapes represent genes that were downregulated. Connecting lines represent interactions between genes that are documented in the Ingenuity Knowledgebase. White shapes are genes that were added to the network by IPA based on their connectivity to dysregulated genes. Shapes circled in orange appear in Table 4. [Color figure can be seen in the online version of this article, available at wileyonlinelibrary.com.]

All network entities that had direct interactions with four or more dysregulated genes were considered to have potential regulatory roles in the PFOS response. For example, in Figure 5, the nodes labeled HNF4A and PPARG have direct interactions with greater than four red or green nodes and were therefore considered to have potential regulatory activity. Ingenuity Pathway Analysis was then used to identify all possible direct or indirect interactions between potential regulatory molecules and all dysregulated genes in each dose group. All known interactions in the Ingenuity Knowledgebase were considered regardless of tissue type (an example for TP53 is shown in Supplemental Data, Fig. S8), and these data are summarized in Table 4. A detailed list of potential regulatory molecules and the genes with which they interact can be found in Supplemental Data, Table 6. The overall trend in the number of interactions between potential regulatory molecules and genes dysregulated by PFOS was similar to the trend observed for functional and pathway enrichment analysis. All potential regulatory molecules had the highest number of interactions with dysregulated genes in T10, followed by L40, then L10. This supports our hypothesis that, although L-PFOS and T-PFOS may affect similar regulatory machinery, the effect is much more prominent following T-PFOS exposure.

Table 4. Molecules from networks generated in Ingenuity Pathway Analysis that had interactions with four or more genes that were differentially expressed in chicken embryonic hepatocytes exposed to by 10 or 40 µM linear or 10 µM technical-grade perfluorooctane sulfonate (L10, L40, or T10, respectively; italicized entries are fold change >1.5; p < 0.05; NP = probe not present on microarray)

Molecule	Entrez gene name	Dose	Fold change	p Value	No. of interactions with DE genes
					Direct	Indirect	Total
TP53	Tumor protein 53	L10	NP	NP	3	1	4
		L40	NP	NP	6	7	13
		T10	NP	NP	12	9	21
MYC	v-myc Myelocytomatosis viral oncogene homolog (avian)	L10	1.16	0.422	4	3	7
		L40	1.41	0.060	9	4	13
		T10	1.33	0.078	13	5	18
HNF4A	Hepatocyte nuclear factor 4α	L10	1.09	0.686	5	1	6
		L40	−1.08	0.959	13	0	13
		T10	−1.10	0.587	17	1	18
CTNNB1	Catenin (cadherin-associated protein), β1, 88 kDa	L10	1.99	0.096	2	1	3
		L40	1.46	0.538	4	3	7
		T10	1.12	0.722	8	7	15
PPARG	Peroxisome proliferator-activated receptor gamma	L10	−1.01	0.849	4	2	6
		L40	1.06	0.677	6	2	8
		T10	−1.00	0.980	10	4	14
SP1	Sp1 transcription factor	L10	−1.14	0.468	3	1	4
		L40	−1.06	0.964	11	1	12
		T10	1.02	0.911	11	0	11
SREBF1	Sterol regulatory element binding transcription factor 1	L10	1.00	0.989	3	1	4
		L40	1.07	0.871	5	1	6
		T10	−1.05	0.646	11	1	12
CEBPB	CCAAT/enhancer binding protein (C/EBP), β	L10	−1.01	0.953	5	2	7
		L40	−1.01	0.998	8	1	9
		T10	1.05	0.749	8	2	10
CEBPA	CCAAT/enhancer binding protein (C/EBP), α	L10	NP	NP	3	0	3
		L40	NP	NP	10	0	10
		T10	NP	NP	8	0	8
HDAC1	Histone deacetylase 1	L10	1.40	0.188	1	0	1
		L40	1.50	0.085	6	0	6
		T10	1.58	0.019	8	0	8
NCOR1	Nuclear receptor corepressor 1	L10	1.17	0.687	2	0	2
		L40	1.12	0.977	7	0	7
		T10	1.80	0.111	7	0	7
ESRRA	Estrogen-related receptor α	L10	NP	NP	1	0	1
		L40	NP	NP	4	0	4
		T10	NP	NP	6	0	6
MYCN	v-myc Myelocytomatosis viral related oncogene, neuroblastoma derived (avian)	L10	−1.00	0.999	1	0	1
		L40	1.31	0.119	2	1	3
		T10	1.37	0.027	5	0	5

Many of the potential regulatory molecules identified through interactome analysis, such as tumor protein p53 (TP53), v-myc myelocytomatosis viral oncogene homolog (MYC), hepatocyte nuclear factor 4α (HNF4A), and catenin β1 (CTNNB1), play integral roles in the proper functioning of diverse cellular processes. The responsiveness of these proteins to PFOS exposure may represent secondary responses, although some evidence exists that PFOS may act more directly in some instances. Tumor suppressor protein TP53 is an important regulator of the cell cycle. It is not surprising that most PFOS-affected genes involved with cell cycle, proliferation, or cancer have direct or indirect interactions with TP53. It is unlikely that PFOS exerts its effect directly on TP53. A typical TP53 response involves differential p21/WAF1 expression, which was not observed, and is induced by DNA damage. Perfluorooctane sulfonate does not appear to be genotoxic 50–52, although Liu et al. 34 did report DNA damage at high PFOS exposures but concluded that it was likely apoptosis related. Recent evidence has shown that lipid peroxidation products can activate TP53 53, which may explain results from the present study. The fact that both Mdm2, a gene that inactivates TP53, and Cdkn2A, which modulates Mdm2 activity 54, were both overexpressed following PFOS exposure is further evidence that TP53 activity was perturbed.

The multifunctional transcription factor MYC is believed to regulate the expression of 10 to 15% of all cellular genes directly or indirectly and is one of the most frequently affected genes in a variety of cancers 55. This protein exerts effects on many cellular processes including growth, differentiation, and apoptosis in a TP53-dependent or -independent fashion 55. This being the case, many of the genes affected by PFOS that interact with MYC are also known to interact with TP53.

The transcription factor HNF4A is believed to be involved in the regulation of up to 40% of all genes expressed in the liver 56 and plays an indispensible role in hepatocyte development and lipid metabolism as evidenced in HNF4A knockdown studies [57]; HNF4A-null mice die during embryogenesis, whereas mature mice that lack HNF4A expression accumulate lipid in the liver and have reduced serum cholesterol and increased serum bile levels 57. Many of the genes that were dysregulated by PFOS and interact with HNF4A (see Supplemental Data, Table 6) are involved in lipid metabolism (acetyl-coenzyme A acetyltransferase 1 [Acat1]; Acsl1; aldo-keto reductase family 1, member B1 [Akr1b1]; apolipoprotein A-IV [Apoa4]; and Cyp8b1) or in cellular proliferation (cyclin G2 [Ccng2]; Gjb1; and inhibin, βA [Inhba]). Binding studies would be needed to determine whether PFOS interferes directly with HNF4A. Alternatively, HNF4A might be affected indirectly through altered fatty acyl-CoA thioester levels 58 that may result from PFOS interference in other regulatory mechanisms of lipid metabolism (for example, through interference with PPARs).

Catenin β1 is an important component of two cellular systems: Wnt signaling and adherens junctions. The Wnt signaling pathway is one of the most fundamental mechanisms that directs cell proliferation and fate during embryonic development and tissue homeostasis 59. Defects in proper Wnt signaling are often linked to birth defects and cancer and other diseases. Increased expression of one of the Frazzled Wnt receptors, Fzd4, and Wnt5B glycoprotein following PFOS exposure indicates that the Wnt signaling pathway may be disrupted by PFOS. Wnt signaling is also linked to TP53 and MYC activity, whereas Wnt5B protein has been shown to activate PPARγ 60. Catenin β1 is also a key component in the adherens junction complex, a cell–cell junction complex that mechanically attaches adjacent cells and relays signals. Interference with CTNNB1 function may be involved with impaired cell–cell communication, which has been demonstrated following PFOS exposure 48.

The hepatocyte response to PFOS may be mediated more directly through the lipid sensing PPARs. In interactome analysis, only PPARγ was present in any of the networks generated by IPA for both L-PFOS and T-PFOS. Analysis revealed that it had many direct interactions with differentially expressed genes, most of which were expression-type interactions (Fig. 6A). In contrast, PPARα did not occur in any of the interaction networks generated by IPA. When all possible interactions in the Ingenuity Knowledgebase with differentially expressed genes were considered (as was done for all potential regulatory molecules), PPARα had fewer interactions with genes that were dysregulated by PFOS than PPARγ (Fig. 6B). Only two of these were expression-type interactions, suggesting that PPARα had minimal influence on the expression profile of CEH in response to PFOS. Binding assays 49 and expression profile studies 37 support the hypothesis that PFOS may be a PPARγ agonist. These same studies, however, also suggest that PFOS is a more potent activator of PPARα in mammals. Given the lack of response of PPARα-regulated genes in the present study and other avian studies 15, 44, it is possible that PFOS is a more potent activator of PPARγ in the chicken. Binding assays using chicken PPAR isoforms would help in clarifying this issue.

Figure 6. Direct (solid line) and indirect (dashed line) interactions between genes that were differentially expressed following exposure to linear or technical-grade perfluorooctane sulfonate and peroxisome proliferator-activated receptor-γ (PPARγ; A) or peroxisome proliferator-activated receptor-α (PPARα; B). [Color figure can be seen in the online version of this article, available at wileyonlinelibrary.com.]

Another possible mechanism for the effects of PFOS exposure is activation of SREBP1, which enhances the transcription of genes involved in fatty acid and cholesterol metabolism. Cleavage activation of SREBP1 is upregulated in response to cellular cholesterol demand 61. This may represent a secondary response to PFOS exposure, which, as previously mentioned, can result in reduced cholesterol levels.

Among all potential regulatory molecules identified, the transcription of only one was affected and only by exposure to T-PFOS: HDAC1. This may represent a molecular effect that is specific to the branched isomers of PFOS and may be responsible for the increased transcriptional activity of T-PFOS over L-PFOS. Histone deacetylase (HDAC1) is involved in chromatin packaging and coordinates how the transcriptional machinery accesses DNA 62. Perturbation of Hdac1 transcription by T-PFOS may play a major role in the increased numbers of genes affected in this treatment group via effects on chromatin structure. Real-time RT-PCR expression analysis of an independent culture showed a significant dose-dependent increase in Hdac1 expression in response T-PFOS, which did not change in the L-PFOS treatment group (Supplemental Data, Fig. S9).

Given the diversity in the effects of PFOS exposure, it is unlikely that it has a single primary mode of action. The mechanism leading to PFOS toxicity is in all likelihood a complex interplay of various molecular events, many of which are proposed here. The results from the present study represent only a snapshot of the transcriptional response to PFOS after 24 h of exposure. Time-course experiments would be required to investigate more thoroughly the proposed mechanisms to obtain a better understanding of their order of activation, how they interact, and how they are affected by different isomer mixtures.

CONCLUSIONS

All PFOS released into the environment is believed to be the product of ECF processes that result in mixtures of linear and branched isomers. When in the environment, the isomeric proportions are altered, possibly through selective uptake and elimination through the food web or other biotic or abiotic transformation processes 5–7. Regardless of the mechanism, the PFOS burden in exposed wild animals, particularly in piscivores, has been skewed toward a much greater proportion of L-PFOS compared with the technical product 2, 3. Until recently, differences in isomer-specific kinetics through the environment were undetectable and therefore not considered in many laboratory studies.

In the present study, we showed that both L-PFOS and T-PFOS elicited a transcriptional response in genes related to lipid metabolism and transport, oxidative stress response, and cellular growth and proliferation. This transcriptional profile is generally typical for PFOS exposure studies using mammalian in vivo and in vitro systems 37, 42, 43. One major difference, however, is the lack of response in β-oxidation genes, suggesting that PPARα may not be a major target of PFOS in cultured CEH. Interactome analysis supported the important role of PPARγ in the PFOS response compared with PPARα. PFOS exposure also appeared to affect several systems that are fundamental to regular hepatocyte functions, such as TP53 signaling and HNF4A transcriptional regulation.

In almost all functional categories, pathways, and interaction networks examined in the present study, the impact of T-PFOS was much greater than that of L-PFOS. The reason for this difference is likely the increased number of molecule shapes presented by the branched isomers of T-PFOS. This broader selection of shapes may be able to interact with more cellular machinery, thereby interfering with more signaling cascades, activating more receptors and recruiting more transcription factors than the linear isomer of PFOS alone. Examples from the present study include interference with the TP53 pathway, PPARγ signaling, and HNF4A activity, all of which appear to be more affected by T-PFOS exposure than by L-PFOS exposure. We also suggest that the increased activity of T-PFOS may involve HDAC1-dependent chromatin packaging.

The present study demonstrates the importance of considering environmental and isomer-specific toxicological properties of PFOS when assessing possible exposure effects on wildlife, especially when an in vitro system is employed, and shows that the environmental relevance of studies using T-PFOS might have to be re-evaluated. Furthermore, the effects of different PFOS isomer mixtures remain poorly characterized in mammalian models. Research investigating effects of branched PFOS isomers in mammals should be prioritized, especially given that branched isomers appear to accumulate in human samples. Branched proportions as high as 50% 63 have recently been reported in human serum.

SUPPLEMENTAL DATA

Table S1. List of genes examined by real-time RT-PCR and their gene symbols, accession numbers, primer and probe sequences, and reaction concentrations (28 KB XLS).

Table S2. Detailed list of genes that were differentially expressed following exposure to L-PFOS or T-PFOS (107 KB XLS).

Table S3. Comparison of real-time RT-PCR results for changes in mRNA expression in cultured chicken embryonic hepatocytes for genes determined by microarray analysis to be significantly affected by exposure to L-PFOS or T-PFOS (30 KB XLS).

Table S4. Detailed list of functions and genes within each functional enrichment category for chicken embryonic hepatocytes (124 KB XLS).

Table S5. Detailed list of enriched canonical pathways for genes that were differentially expressed following exposure to PFOS (47 KB XLS).

Table S6. Detailed list of potential regulatory molecules and their interactions with genes that were dysregulated by exposure to PFOS (53 KB XLS).

Fig. S1. Exposure to L-PFOS (40 µM) alters the expression of GULO and others genes involved in L-ascorbate metabolism in CEH.

Fig. S2. LPS/IL-1 mediated inhibition of RXR function canonical pathway.

Fig. S3. IPA-generated interaction network for genes that were differentially expressed following 10 µm L-PFOS exposure.

Figs. S4 and S5. IPA-generated interaction network for genes that were differentially expressed following 40 µm L-PFOS exposure.

Fig. S6 and S7. IPA generated interaction network for genes that were differentially expressed following 10 µm T-PFOS exposure.

Fig. S8. All documented direct and indirect interactions from Ingenuity's Knowledgebase between TP53 and genes that were differentially expressed with exposure to L-PFOS 10 µM, L-PFOS 40 µM, or T-PFOS 10 µM.

Fig. S9. Expression of HDAC1 in subpooled CEH following exposure to L-PFOS or T-PFOS (11 MB PDF).

Acknowledgements

Funding was provided by two programs within Environment Canada, Chemicals Management Plan and Strategic Technology Applications of Genomics for the Environment. The authors declare that no conflicts of interest exist.