Research Resource: A Reference Transcriptome for Constitutive Androstane Receptor and Pregnane X Receptor Xenobiotic Signaling

Abstract

The pregnane X receptor (PXR) (PXR/NR1I3) and constitutive androstane receptor (CAR) (CAR/NR1I2) members of the nuclear receptor (NR) superfamily of ligand-regulated transcription factors are well-characterized mediators of xenobiotic and endocrine-disrupting chemical signaling. The Nuclear Receptor Signaling Atlas maintains a growing library of transcriptomic datasets involving perturbations of NR signaling pathways, many of which involve perturbations relevant to PXR and CAR xenobiotic signaling. Here, we generated a reference transcriptome based on the frequency of differential expression of genes across 159 experiments compiled from 22 datasets involving perturbations of CAR and PXR signaling pathways. In addition to the anticipated overrepresentation in the reference transcriptome of genes encoding components of the xenobiotic stress response, the ranking of genes involved in carbohydrate metabolism and gonadotropin action sheds mechanistic light on the suspected role of xenobiotics in metabolic syndrome and reproductive disorders. Gene Set Enrichment Analysis showed that although acetaminophen, chlorpromazine, and phenobarbital impacted many similar gene sets, differences in direction of regulation were evident in a variety of processes. Strikingly, gene sets representing genes linked to Parkinson's, Huntington's, and Alzheimer's diseases were enriched in all 3 transcriptomes. The reference xenobiotic transcriptome will be supplemented with additional future datasets to provide the community with a continually updated reference transcriptomic dataset for CAR- and PXR-mediated xenobiotic signaling. Our study demonstrates how aggregating and annotating transcriptomic datasets, and making them available for routine data mining, facilitates research into the mechanisms by which xenobiotics and endocrine-disrupting chemicals subvert conventional NR signaling modalities.

As transcription factors regulated in part by interactions with small lipophilic molecules, members of the nuclear receptor (NR) superfamily are well-characterized targets for a variety of xenobiotics and endocrine-disrupting chemicals (EDCs) (1). Two NRs that have been shown to be particularly frequent mediators of xenobiotic action are the constitutive androstane receptor (CAR) (CAR/NR1I2) and pregnane X receptor (PXR) (PXR/NR1I3). Numerous hypothesis-driven research studies have constructed a complex model of xenobiotic signaling, involving interactions with multiple signaling pathways with intricate tissue- and promoter-selective specificities. CAR and PXR have overlapping functions in the regulation of gene expression by the xenobiotics acetaminophen (Acmphn) (Figure 1A) (2, 3), phenobarbital (Phenob) (Figure 1B) (4, 5), chlorpromazine (Chlorprom) (Figure 1C) (6), and pregnenolone-16α-carbonitrile (PCN) (Figure 1D) (7), whereas the pesticide contaminant 1,4-Bis[2-(3,5-dichloropyridyloxy)]benzene (TCPOBOP) (Figure 1E) and 6-(4-chlorophenyl)imidazo[2,1-b][1,3]thiazole-5-carbaldehydeO-(3,4-dichlorobenzyl)oxime (CITCO) (Figure 1F) are specific for CAR (8, 9).

Figure 1.

Chemical structures of CAR and PXR xenobiotic perturbants in transcriptomic datasets used to generate the reference transcriptome. CAR and PXR have overlapping specificity for Acmphn, Phenob, Chlorprom, and PCN, whereas TCPOBOP and CITCO are specific for CAR.

The universe of data points generated by the toxicotranscriptomics community has considerable collective value as a resource for identifying common transcriptional targets of CAR and PXR xenobiotic signaling pathways. Unfortunately, deficits in the curation and stewardship of these datasets have limited their reuse (10), to the extent that extracting a comprehensive list of xenotranscriptomic targets from the research literature is a far from straightforward task for the bench researcher. Reasoning that the xenobiotic research community might benefit from routine access to a reference CAR and PXR xenobiotic transcriptome based on unbiased, discovery-scale interrogation of xenobiotically perturbed model systems, we set out here to use the Nuclear Receptor Signaling Atlas (NURSA) Transcriptomine database (11) to generate this list. In addition to confirming experimentally documented components of the xenobiotic response, the list contains a number of prominent genes and cellular processes that conventional literature searches failed to associate with xenobiotic signaling, but that nevertheless shed light both on their known biology and suspected pathology of xenobiotics.

Materials and Methods

Dataset processing, annotation, and acquisition

Datasets were processed and annotated for inclusion in the NURSA database using a previously described protocol (11, 12). Briefly, expression data obtained from Gene Expression Omnibus (13) and TG-GATES (14) as investigator-provided summarized and normalized array feature expression intensities were extracted and processed in the statistical program R. To calculate differential gene expression for investigator-defined experimental contrasts, we used the linear modeling functions from the Bioconductor limma analysis package (11). P values obtained from limma analysis were not corrected for multiple comparisons. In cases where a given gene was represented on an array by more than one probeset, data from individual probesets were generated separately and fold change values were not pooled across array features. Experimental contrasts were annotated for RNA Source and regulatory molecule with reference to the original publication accompanying the dataset, and fold changes were crosschecked to ensure faithful recapitulation of the data points reported in the original manuscript. Datasets were reviewed and 22 corresponding to perturbation of CAR/NR1I2 or PXR/NR1I3, or small molecules characterized as direct agonists or antagonists, were selected for generation of the reference transcriptomic signature (Table 1). Figure 2 summarizes the data analysis and metadata biocuration pipeline.

Table 1.

Datasets Used to Generate the Reference CAR and PXR Xenobiotic Transcriptome

Dataset	DOI	Expts	Reference
Acmphn
Acute time course and dose-dependence analysis of the acetaminophen (Acmphn)-regulated transcriptome in rat kidney	10.1621/aYNzPYyURw	9	77
Acute time course and dose-dependence analysis of the acetaminophen (Acmphn)-regulated transcriptome in rat primary hepatocytes	10.1621/1vkREcPTii	9	78
Acute time course and dose-dependence analysis of the acetaminophen (Acmphn)-regulated transcriptome in rat liver	10.1621/dtsY1XDjLe	12	79
Acute time course and dose-dependence analysis of the acetaminophen (Acmphn)-regulated transcriptome in human primary hepatocytes	10.1621/p7F5N6o2kC	9	80
Chronic time course and dose-dependence analysis of the acetaminophen (Acmphn)-regulated transcriptome in rat kidney	10.1621/ha2dfkibBe	12	81
Chronic time course and dose-dependence analysis of the acetaminophen (Acmphn)-regulated transcriptome in rat liver	10.1621/ffXgOKigjs	12	82
Time course analysis of the acetaminophen (Acmphn)-regulated hepatic transcriptome in a variety of mouse strains	10.1621/j7SRGHu1mv	8	83
Transcriptomic profiling of acetaminophen (Acmphn)-, clofibrate (Clofib)- and lithocholic acid (Litho)-treated primary rat hepatocytes	10.1621/MIASPNjf4c	2	84
Chlorprom
Acute time course and dose-dependence analysis of the chlorpromazine (Chlorprom)-regulated transcriptome in rat primary hepatocytes	10.1621/UMT1Pv11IH	9	85
Acute time course and dose-dependence analysis of the chlorpromazine (Chlorprom)-regulated transcriptome in human primary hepatocytes	10.1621/RVje15d0l7	12	86
Acute time course and dose-dependence analysis of the chlorpromazine (Chlorprom)-regulated transcriptome in rat liver	10.1621/R8LBcqY0Kd	8	61
Chronic time course and dose-dependence analysis of the chlorpromazine (Chlorprom)-regulated transcriptome in rat liver	10.1621/cq2ujHYtTF	12	87
PCN
Analysis of the pregnenolone carbonitrile (PCN)-regulated transcriptome in mouse liver	10.1621/rVCNLZzcjj	1	88
Analysis of the pregnenolone carbonitrile (PCN)-regulated transcriptome in mouse duodenum	10.1621/9nnuQmZqWd	1	89
Phenob
Analysis of the phenobarbital (Phenob)-regulated transcriptome in mouse liver	10.1621/rVCNLZzcjj	1	88
Analysis of the phenobarbital (Phenob)-regulated transcriptome in HepaRG cells	10.1621/RdJpfZvFp6	6	62
Time- and dose-dependence analysis of phenobarbital (Phenob)-treated HepaRG cells and primary human hepatocytes	10.1621/l9Xpov1ODL	20	90
Time course analysis of peroxisome proliferator activated receptor (PPARα/Ppara) agonist-dependent transcriptomes in rat liver	10.1621/nvrYDqPZIP	2	91
Others
Analysis of constitutive androstane receptor (CAR/Nr1i3) ligand-dependent transcriptomes in wild type, CAR/Nr1i3 null mutant and human CAR/NR1I3 knockin mouse liver	10.1621/datasets0.01003	9	92
Analysis of the phenobarbital (Phenob)-regulated transcriptome in rat liver	10.1621/EvWGyqgO1Y	1	93
Analysis of the pregnane X receptor (PXR/Nr1i2)-regulated, β-secretase-dependent transcriptome in mouse liver	10.1621/gvx9xVfhUI	2	94
Time course transcriptomic analysis of the phenobarbital (Phenob)-regulated transcriptome in mouse liver	10.1621/1lIAjWjBMh	2	63

Digital object identifiers (DOIs) point to the dataset pages on the NURSA website.

Figure 2.

Transcriptomine data processing and metadata biocuration pipeline. GEO, Gene Expression Omnibus; AE, ArrayExpress. Datasets were processed and annotated for inclusion in the NURSA database using a previously described protocol (11, 12). Experimental contrasts were annotated for RNA Source and regulatory molecule with reference to the original publication accompanying the dataset, and fold changes were crosschecked to ensure faithful recapitulation of the data points reported in the original manuscript.

Generation of the reference xenobiotic transcriptome

To generate the reference transcriptome, we downloaded 22 Transcriptomine (11) expression profiling datasets that mapped to perturbations of the PXR or CAR xenobiotic pathways, corresponding to a total of 159 individual experiments and 25 801 significant (fold change ≥ ±2, P ≤ .05) fold changes (see Supplemental Dataset 1). Given that this list of genes represented human, mouse, and rat genes, we wished to map nonhuman genes to their human orthologs to provide for a unified, pan-species reference xenobiotic transcriptome. The NURSA data repository (12) maintains a collection of compilations of essential functions and properties of human, mouse, and rat genes, whose orthologous relationships are defined using NCBI's Homologene service (15). Human gene symbols corresponding to mouse and rat gene symbols in the list were replaced with their orthologous human gene symbols programmatically extracted from the NURSA database. Next, the pivot table function in Microsoft Excel was used to assign each human gene symbol a value corresponding to the frequency of its occurrence in the list, and symbols were then ranked in descending numerical order. Supplemental Dataset 2 contains the full reference xenobiotic transcriptome.

Protein annotation through evolutionary relationship (PANTHER) Gene Ontology (GO) biological process enrichment analysis

The PANTHER classification system (http://www.pantherdb.org/) combines gene function, ontology, and statistical analysis tools to support analysis of discovery-scale transcriptomic and proteomic datasets (16). The PANTHER Overrepresentation test (release 20160321) identifies statistically significant over- or underrepresentation of genes mapped to a specific biological process in a user's gene list with that of a reference gene list representing the universe of genes from the appropriate species. To identify biological processes that are selectively and consistently driven by exposure to xenobiotics, we carried out PANTHER Overrepresentation analysis on genes represented 10 or more times in the reference transcriptome (n = 575). Using the GO Hierarchy view from PANTHER, the most specific GO subclasses were selected for further analysis. These significantly overrepresented GO terms (Bonferroni corrected P ≤ .05) were ranked according to their fractional difference (number observed − number expected)/number expected). GO terms with a fractional difference more than or equal to 6.0 were used in Figure 1. Supplemental Dataset 3 contains all raw data pertaining to the PANTHER analysis.

Gene Set Enrichment Analysis (GSEA)

Using the Transcriptomine/NURSA API (17), we downloaded experiments for the endocrine disruptors groups Acmphn, Chlorprom, and Phenob. Due to the smaller numbers of experiments available for TCPOBOP, CITCO, and PCN, experiments involving these xenobiotics were not included in the analysis. For each experiment, we used the gene symbols and fold change values as inputs for the GSEA algorithm (18). Based on the compendium of biological pathways and processes compiled and curated by the Molecular Signature Database (MSigDB) (19, 20), we used the following gene set collections: Hallmark, KEGG, Reactome, Biocarta, and GO biological process, for a total of over 1900 gene sets. For each Transcriptomine experiment and each gene set, GSEA returned a Normalized Enrichment Score and an adjusted Q value. We considered gene set enrichments to be significant for Q < 0.25. Next, we collated enrichment profiles for all the experiment in a single matrix, and assigned a value of Normalized Enrichment Score = 0 for pathways and gene sets that did not reach significance. We then used ANOVA as implemented in the R software package (version 3.3.0) to determine pathways that have different enrichment profiles between distinct endocrine disruptor groups; significance was considered for P < .01. We used hierarchical agglomerative clustering with average linkage to identify cluster of pathways with similar characteristics, and used the Python (version 2.7.11) scipy (version 0.17.0) package to generate the heatmap. Supplemental Figure 1 contains a high resolution, fully annotated version of the heatmap.

Results and Discussion

CAR- and PXR-mediated xenobiotic signaling impacts myriad biological processes

We first ranked gene symbols based upon the frequency of their significant differential expression across 159 experiments involving perturbation with xenobiotics known to impact CAR- and PXR-mediated signaling pathways (Table 1 and Figure 1). Table 2 shows the top 50 ranked genes in the reference xenobiotic transcriptome and Figure 3 shows the top 10 most highly overrepresented GO Biological Process categories generated by the PANTHER analysis of genes with a reference transcriptome frequency (ƒ) of 10 or higher (Supplemental Dataset 2 and Supplemental Dataset 3). The robust overrepresentation of the ω-hydroxylase and epoxygenase P450 gene families, as well as components of the unfolded protein-response pathways are symptomatic of the cellular response to Acmphn and other sources of cellular stress (21, 22). Also prominent among highly ranked members of the gene list, and previously characterized targets of PXR and CAR in a variety of contexts (23), were 25 ABC transporters, 10 aldo-keto reductases, and 10 UDP-glucuronosyl-transferases, all of which have well-known roles in drug metabolism and detoxification (24, 25). In addition to these iconic cellular responses to xenobiotics are a number of transcriptional targets whose consistent and robust regulation by xenobiotic signaling is not visibly represented in the conventional research literature. The specific examples discussed here shed light upon several aspects of the known but mechanistically uncharacterized biology of xenobiotics in major physiological processes, and serve to illustrate the value of establishing a reference transcriptome for xenobiotic signaling.

Table 2.

Top 50 Genes in the NURSA CAR and PXR Xenobiotic Reference Transcriptome (Version 1)

Rank	Gene	Name	Rank	Gene	Name
1	SRXN1	Sulfiredoxin 1	25	FABP4	Fatty acid-binding protein 4
2	KLF6	Kruppel-like factor 6	25	GADD45B	Growth arrest and DNA damage inducible-β
3	DUSP6	Dual specificity phosphatase 6	28	ABCB1	ATP-binding cassette subfamily B member 1
4	POR	Cytochrome p450 oxidoreductase porcupine homolog (Drosophila)	28	IL1B	IL-1β
5	CYP2B6	Cytochrome P450 family 2 subfamily B member 6	28	LOX	Lysyl oxidase
5	G6PC	Glucose-6-phosphatase catalytic subunit	28	TIPARP	TCDD-inducible poly (ADP-ribose) polymerase
7	EGR1	Early growth response 1	28	TSKU	Tsukushi, small leucine-rich proteoglycan
7	IGFBP1	IGF-binding protein 1	28	ZFAND2A	Zinc finger AN1-type containing 2A
7	THRSP	Thyroid hormone responsive	34	EDNRA	Endothelin receptor type A
7	TRIB3	Tribbles pseudokinase 3	34	ONECUT1	One cut homeobox 1
11	PRLR	Prolactin receptor	34	PER2	Period circadian clock 2
12	HMOX1	Heme oxygenase 1	34	SOCS2	Suppressor of cytokine signaling 2
13	CCND1	Cyclin D1	34	SOCS3	Suppressor of cytokine signaling 3
14	CTGF	Connective tissue growth factor	39	AKR7A3	Aldo-keto reductase family 7 member A3
14	JUN	Jun protooncogene, AP-1 transcription factor subunit	39	CYP1A1	Cytochrome P450 family 1 subfamily A member 1
14	NFKBIZ	NFKB inhibitor-ζ	39	GCLC	Glutamate-cysteine ligase catalytic subunit
14	SOX4	SRY-box 4	39	GSTM3	Glutathione S-transferase μ 3 (brain)
18	ASNS	Asparagine synthetase (glutamine-hydrolyzing)	39	HSPA1B	Heat shock protein familyA (Hsp70) member 1B
18	FKBP5	FK506-binding protein 5	39	IDI1	Isopentenyl-diphosphate-δ isomerase 1
20	CHKA	Choline kinase-α	39	PDK4	Pyruvate dehydrogenase kinase 4
20	LPIN1	Lipin 1	46	ACOT2	Acyl-CoA thioesterase 2
20	SULT1E1	Sulfotransferase family 1E member 1	46	ATP1B1	ATPase Na+/K+ transporting subunit β1
23	CYP2C9	Cytochrome P450 family 2 subfamily C member 9	46	CCNB1	Cyclin B1
23	DUSP1	Dual specificity phosphatase 1	46	CISH	Cytokine inducible SH2 containing protein
25	CYR61	Cysteine-rich angiogenic inducer 61	46	CYP3A4	Cytochrome P450 family 3 subfamily A member 4

The full list of genes is shown in Supplemental Dataset 2.

Figure 3.

PANTHER GO biological process enrichment analysis identifies numerous cellular processes impacted by xenobiotic signaling. Biological processes enriched 9-fold or higher are shown (Supplemental Dataset 3 contains the full list of categories). The PANTHER Overrepresentation test identifies statistically significant over- or underrepresentation of genes mapped to a specific biological process in a user's gene list with that of a reference gene list representing the universe of genes from the appropriate species.

Oxidative stress

KLF6 (ƒ = 55, rank = 2) encodes a member of the Krüppel-like family of transcription factors, a total of 5 of which are represented in the reference xenotranscriptome. In addition to an implied tumor suppressive role in a variety of cancers (26, 27), it is induced in the livers of mouse models of nonalcoholic steatosis (28) and oxidative stress (29). Moreover, glomerular podocyte-specific depletion of KLF6 has been shown to result in mitochondrial dysfunction and apoptosis (30). Consistent with this, it has been previously identified as a DNA-binding partner of the arylhydrocarbon receptor (31). Its prominent position in the xenobiotic transcriptome suggests that it plays a key early role in marshalling of the transcriptional response to xenobiotic signaling.

Although most reactions involving CYP450 enzymes result in detoxification of xenobiotics, some reactions, such as the conversion of Acmphn to N-acetyl-p-benzoquinone imine, are thought to contribute to xenobiotic-induced oxidative stress and liver injury. The inhibition by N-acetyl-p-benzoquinone imine of thioredoxin reductase 1 (TXNRD1) (ƒ = 25, rank = 58) (32), a central component of cellular disulfide bond homeostasis, has been suggested to contribute to the development of oxidative stress in xenobiotic-exposed cells. It can be speculated therefore that the elevated ranking of TXNRD1 in the reference transcriptome reflects a compensatory mechanism to supplement depleted functional TXNRD1 levels to maintain appropriate redox homeostasis. Induction of NRF2/NFE2L2 has been identified as a potential compensatory factor in maintaining cellular redox homeostasis in hepatic-specific knockout mice (33), although its low ranking in the reference transcriptome suggests that this is a less than frequent event.

Carbohydrate metabolism

Reflecting the roles of PXR and CAR in homeostatic energy metabolism (34, 35) and previous reports of its regulation by PXR and/or CAR (36, 37), the gene encoding the last and rate-limiting enzyme of gluconeogenesis (catalytic α-subunit of glucose-6-phosphatase [G6PC]; ƒ = 40, rank = 5) was prominent among the reference xenobiotic targets. Also highly ranked was PGC-1/PPARGC1A (ƒ = 19, rank = 125), which encodes a transcriptional coregulator originally characterized as a cold-inducible coactivator of PPARγ/PPARG and subsequently implicated as a central component of cellular carbohydrate energy metabolism (38). Although the protein product has a well-established role in supporting NR-mediated transcriptional programs (39), the PPARGC1A gene itself has not been previously characterized as a transcriptional target of xenobiotic signaling. Its presence in the xenotranscriptomic signature is not entirely surprising given its regulation in response to other forms of stress, including cold and fasting (40). A plausible rationale for its consistent regulation by xenobiotic stimuli, and one lent credence by its reciprocal regulation of PXR/NR1I3 expression (41) and coactivation of genes encoding xenobiotic-metabolizing enzymes (42), is as part of a feed-forward loop supporting efficient clearance of elevated cellular xenobiotic concentrations. Interestingly, another reference xenobiotic target, SIRT1 (ƒ =10, rank = 474), encodes a FOXO1 deacetylase, and FOXO1 itself has been shown to contribute to transcriptional regulation of PCK and PGC1 (43). In concert these data point to the existence of a previously uncharacterized regulatory loop supporting a cellular shift towards energy conservation in response to xenobiotic stress.

Reproduction

A number of genes encoding key players in mammalian male and female reproduction are highly ranked in the reference xenotranscriptome. INHBA (ƒ = 19, rank = 125) and INHBB (ƒ = 11, rank = 396) encode the α- and β-subunits of the peptide hormone inhibin, which effects down-regulation of the synthesis and secretion of the gonadotropin FSH. In the case of the FGF pathway, which has been implicated in regulation of sperm development and maturation (44), genes encoding both ligand (FGF1: ƒ = 12, rank = 350; and FGF21: ƒ = 20, rank = 111) and receptor (FGFR2: ƒ = 10, rank = 474) components of the pathway were represented in the xenotranscriptome. Another highly ranked gene in the reference signature was that encoding the prolactin receptor (PRLR) (ƒ = 38, rank = 11), which mediates pleotropic cellular responses to prolactin. Consonant with the profound link between energy metabolism and reproduction, PRLR has recently been shown to potentiate insulin action (45). Interestingly, a search for PRLR in the Transcriptomine database identified 2 independent datasets establishing positive regulation of PRLR in response to SIRT1 overexpression (46). These data points are consistent with the insulin-sensitizing role of SIRT1 (47) and suggest that the coordination of energy metabolism with key reproductive processes might be critically exposed to the disruptive effects of xenobiotics and EDCs.

Inflammation

NFKBIZ (ƒ = 35, rank = 14) encodes a toll-like receptor signaling-inducible member of the ankyrin-repeat family (48) that has been recently identified as a key player in the development of inflammatory conditions, including psoriasis (49, 50) and asthma (51). Consistent with this, Transcriptomine displays data points documenting its induction in lipopolysaccharide (52)- and oxidative stress-derived extracellular vesicles (52)-treated bone marrow macrophages, and in colitic colonic mucosa (53). Given its role in the inflammatory response, there is a striking resonance between the prominent position of NFKBIZ in the xenobiotic reference transcriptome and emerging, albeit epidemiologically underdeveloped, evidence linking Acmphn to psoriasis (54) and asthma (55).

The dual specificity phosphatase family member DUSP6 (ƒ = 53, rank = 3) is a negative regulator of ERK1/2 in the context of FGF and ETS-1 signaling (56, 57) and has a well-characterized role in suppression of the immune response (58). DUSP6 has been shown to be robustly induced in response to treatment of nonsmall cell lung cancer cell lines with the epidermal growth factor receptor inhibitor CL-387785 (57) and in gastric cancer cell lines by the HGF inhibitory therapeutic PHA-665752 (59), indicating a potential role in cancer cell survival mechanism. Consistent with such a notion, siRNA depletion of DUSP6 has been shown to potentiate the cytotoxicity in cancer cell lines of CL-387785 and other cytotoxic agents (60). Collectively these results indicate a prominent role of DUSP6 in mediating the cellular response to exogenous chemicals and suggest that the CAR/PXR/xenobiotic pathway represents a potential therapeutic leverage point in a variety of cancers. Regulation of DUSP6 by xenobiotics is dynamic and a function of the duration of treatment (Supplemental Dataset 1), indicating the potential involvement of feedback mechanism in fine tuning and maintaining its cellular concentration at homeostatically appropriate levels.

GSEA of Acmphn, Chlorprom, and Phenob transcriptomes

Having established a reference transcriptome across all CAR and PXR xenobiotics, we next used GSEA to visualize commonalities and similarities in biological processes regulated by individual xenobiotics. Using the Transcriptomine API, we retrieved significantly differentially expressed genes for experiments in which Acmphn, Chlorprom, and Phenob were perturbants. To suppress differences between each of the treatments at the level of individual genes, we first represented each experiment as a collection of enriched pathways, based on GSEA (18) and a collection of over 1900 MSigDB pathways (19, 20), then used ANOVA parametric testing (P < .01) to determine significant pathways.

Figure 4 shows a GSEA-based heatmap of the Acmphn, Chlorprom, and Phenob transcriptomes. Although reasserting the existence of commonalities between the different xenobiotics with respect to regulation of gene expression, the results of this analysis also highlighted interesting differences in their transcriptional endpoints, both between individual xenobiotics as well as across different variables for a given xenobiotic. Interestingly, gene sets related to cell cycle, DNA repair, and E2F signaling were induced in Acmphn and suppressed by Chlorprom and Phenob, whereas oxidative phosphorylation and protein translation were induced by both Acmphn and Chlorprom and suppressed by Phenob. Conversely, Wnt Signaling and RNA Polymerase II/Transcription pathways were induced by Phenob, but suppressed by Acmphn and Chlorprom. Furthermore, Adipogenesis, Fatty acid metabolism, Peroxisome signaling, and Steroid Hormone Biosynthesis gene sets were induced in small subsets of both Acmphn and Chlorprom experiments and suppressed in the Phenob group, whereas Inflammation appeared to be induced by Phenob experiments but was not significantly altered by the other 2 xenobiotics. Although it is beyond the compass of this study to identify which of these effects are specifically CAR and/or PXR dependent, and not due to some non-NR-related mechanism of action, they do invite speculation as to the nature of this discrepancy in the context of NR signaling. They may reflect, for example, functional differences in the PXR and CAR transcriptional complexes assembled by Phenob and the other xenobiotics, or the relative contributions of PXR and CAR to Phenob-responsive transcriptional programs compared with those regulated by Acmphn and Chlorprom.

Figure 4.

GSEA heat map identifies convergence and divergence in the Acmphn, Phenob, and Chlorprom transcriptomes. Using the Transcriptomine API, we retrieved significantly differentially expressed genes for experiments in which Acmphn, Chlorprom, and Phenob were perturbants. To suppress differences between each of the treatments at the level of individual genes, we first represented each experiment as a collection of enriched pathways, based on GSEA and a collection of over 1900 MSigDB pathways, then used ANOVA parametric testing (P < .01) to determine significant pathways.

GSEA analysis also identified duration of exposure and species as important variables in determining the transcriptional impact of a given xenobiotic. A striking exemplar was the reversal in polarity of oxidative phosphorylation and protein translation gene set regulation effected by acute (Figure 4, bottom, A, and Supplemental Dataset 1 and Supplemental Figure 1, dataset 442, experiments 100084–88) (61) or extended (Figure 4, bottom, A, and Supplemental Dataset 1 and Supplemental Figure 1, dataset 442, experiments 100089–94) (61) exposure to Chlorprom. Similar discrepancies were observed in the Phenob datasets between experiments performed in human HepG2 cells (Figure 4, bottom, C, and Supplemental Dataset 1 and Supplemental Figure 1, dataset 215, experiments 854–859) (62) and mouse liver (Figure 4, bottom, D, and Supplemental Figure 1, dataset 172, experiments 679 and 680) (63). It bears pointing out that for the most part, the effects of these variables were less on-off in nature, and more yin-yang, in other words, the same sets of genes were regulated but in opposite directions.

A compelling trend across all 3 xenobiotics was the enrichment of KEGG gene sets relating to diseases of the central nervous system, including Parkinson's, Huntington's, and Alzheimer's diseases (Figure 4 and Supplemental Figure 1). This observation resonates strongly with previous research inferring connections between deficiencies in xenobiotic clearance exposure and central nervous system diseases (64–67) and is particularly interesting given the documented ability of Acmphn (68), Chlorprom (69), and Phenob (70) to traverse the blood brain barrier.

Conclusion

Despite the existence of numerous excellent reviews in the area of NR toxicotranscriptomics, identifying a validated list of transcriptional targets of xenobiotic signaling pathways can be a challenge for a researcher unfamiliar with the field, or even for those active in it. The goal of this study was to provide research communities in endocrinology, pharmacology, and related disciplines with a convenient reference of common and frequent transcriptional targets of xenobiotic signaling. Our reference xenobiotic transcriptome will have a variety of practical uses, for example, in the design of a panel of reporter genes to provide for screening of potential xenobiotics and EDCs, or to assess the impact of a given genetic perturbation on the integrity of PXR- and CAR-mediated xenobiotic signaling.

NURSA is continually growing its curated dataset holdings, and as a consequence, the reference xenobiotic transcriptome presented here is dynamic and will be reversioned as additional future relevant datasets are incorporated into the resource. At any given point in time therefore, it will represent the xenobiotic research community's “best guess” as to a reference xenobiotic transcriptomic signature. It should be borne in mind of course that the scope of our resource is currently largely restricted to NR signaling pathways, and transcriptomic datasets relevant to other signal transduction pathways are not currently represented in significant quantities. Future expansion of the resource to a broader cross-section of signaling modalities will most likely identify other signaling pathways that are significantly impacted by xenobiotic signaling.

A number of caveats and assumptions apply to this reference transcriptome. Our approach was predicated upon ranking genes according to the frequency with which their expression was significantly perturbed in transcriptome-wide experiments involving perturbants known to impact CAR/NR1I2 and PXR/NR1I3 signaling. Given the well-characterized context specificity of cellular signaling, which is influenced by diverse factors, including promoter epigenetic status and cell type, the reference transcriptome deemphasizes the direction or magnitude of individual fold changes, and focuses more on identifying those genes that are most consistently and reproducibly impacted across all species by perturbation of the CAR/PXR/xenobiotic network. Consequently, no specific mechanism, such as CAR or PXR dependence, direct or indirect regulation, or induction or repression is implied in the list. The GSEA analysis makes evident that although broad convergence exists across xenobiotics with respect to regulation of gene expression, factors such as small molecule structure, tissue context and others exert a considerable effect on the biological endpoints of exposure to xenobiotics. Such parameters are highly contextual, in that expression of a gene might be receptor dependent under one set of conditions but not another, for example, and their definition requires detailed bench characterization of individual promoter-cell-receptor-signal relationships that is beyond the scope of this study.

The breadth of biological processes impacted in the reference transcriptome reflects the numerous emergent human health indices in which xenobiotics and EDCs have been implicated (71). It should be borne in mind however that due to the currently limited number of studies in the database, the reference list was performed across different tissues and species. As a result, to gain an appreciation for the suppression or potentiation of human biological processes by xenobiotics, we inferred comparable regulation of human orthologs of mouse and rat genes. Although a well-established convergence exists between human and rodent metabolism (72, 73), the reliability of such extrapolation is not universally endorsed (74). As the resource grows with the incorporation of additional datasets, whether from large scale initiatives such as the excellent TG-GATES (14), or from individual investigators, the opportunity will arise for users to compare and contrast species-specific xenotranscriptomic responses.

The reference xenobiotic transcriptomic signature list is currently static and available as a Supplemental Dataset 2 accompanying this publication or for download from the NURSA website. Future work will provide for the 1-click availability of an interactive list that will be regularly updated by the addition of new CAR- and PXR-relevant xenobiotic datasets to the NURSA resource; these updated lists will be versioned and time-stamped so that the full complement of underlying datasets underlying any given list can be clearly identified. In addition to PXR- and CAR-cognate xenobiotics, NURSA curates datasets relevant to EDCs that target other signaling pathways, including the ER/estrogen signaling pathway. Future work will generate similar reference transcriptomes and provide for users of the site to readily identify commonalities and contrasts between the transcriptomic impact of diverse NR signaling pathways.

The reference transcriptome is perhaps best thought of as an informatic review, representing a convenient, high-level survey of the relative responsiveness to toxic exposure of the regulatory elements governing expression of mammalian genes. It bears emphasizing that despite the fact that all of the datasets that contributed to the reference transcriptome have been archived in public repositories such as Gene Expression Omnibus (13) and TG-GATES (14), they were largely opaque to conventional literature searches through PubMed and Google. The value in our study, and the wider NURSA biocurative and web tool development initiative, is in removing the barriers to their discovery, aggregation, annotation, reuse and citation by researchers (75). The National Institutes of Health's Big Data To Knowledge initiative (76) represents an expansive and coordinated effort to enhance its investment in discovery-scale datasets by investing in the infrastructure supporting their stewardship and accessibility to the community. We anticipate that as the ideals articulated in these principles find more widespread adoption, the assembly and maintenance of reference transcriptomes for cellular signaling pathways will accelerate the development of human therapeutics towards which the biomedical research community ultimately aspires.