Identification of differentially expressed genes and networks related to hepatic lipid dysfunction

Abstract

A range of chemical exposures that resulted in the specific pathology of hepatic lipid dysfunction in rats were selected from DrugMatrix, a publicly available toxicogenomic database. Raw microarray data collected from these exposures were further analyzed using bioinformatic tools to generate a differentially expressed genes (DEGs) dataset associated with hepatic lipid dysfunction. Further analysis of the DEGs dataset resulted in 324 upregulated genes, and 275 genes that were down regulated. Meanwhile, 36 genes were either up regulated or down regulated in different chemical treatments. All identified genes were uploaded in the web application for Database for Annotation, Visualization and Integrated Discovery (DAVID) for gene ontology enrichments and to identify Kyoto Encyclopedia of Genes and Genome (KEGG) pathways. Some of the identified pathways included glycolysis/gluconeogenesis, steroid hormone biosynthesis, retinol metabolism, and metabolism of xenobiotics by cytochrome P450. The same DEGs dataset was also analyzed using Ingenuity Pathway Analysis (IPA) software. IPA identified several pathways including PXR/RXR activation, Aryl hydrocarbon receptor signaling, and xenobiotic metabolism signaling. Furthermore, the generated DEGs lists were uploaded into NCATS BioPlanet platform. Some of the identified pathways were related to fatty acid omega oxidation, lipid and lipoprotein metabolism, and adipogenesis. The enrichment and clarification of the pathways and biological networks obtained from the DEGs dataset provide prior knowledge on the underlying biological key events and molecular mechanisms for the computational development of putative adverse outcome pathways (AOPs) for hepatic lipid dysfunction as a precursor to hepatic steatosis.

1. Introduction

The National Research Council report entitled “Toxicity Testing in the 21^st Century: A Vision and Strategy” set the foundations for a significant shift in chemical toxicity assessment away from in vivo testing toward in vitro methods based on pathways of biological function (Council, N. R, 2007). The goal of this shift was to advance principles that could leverage increasing knowledge surrounding toxicological effects toward the challenge of increasing efficiency, reducing cost, and decreasing the number of animals used to determine potential of substances to adversely affect humans and the environment. Specifically, the report highlights the application of high-throughput toxicity testing (HTT) data to more apical adverse outcomes in the framework of adverse outcome pathways (AOPs). AOPs provide a framework to systematically investigate the biological pathways starting from molecular initiating events (MIEs), passing through key events (KEs) leading to an adverse outcome (AO) (Edwards et al., 2016). Once developed, AOPs can be used to screen chemicals based on their interactions with key events, identify biomarkers for adverse outcomes, investigate relevance of biological key events among different species, and provide mechanisms to construct biologically-based dose-response quantitative relationships.

Development of AOPs is a process that will require integrating data sets into plausible biological mechanisms consisting of key events and relationships between them (Villeneuve et al., 2014a, 2014b). The data may come from different sources such as experimentations, literature and/or query of gene-gene interaction pathways using publicly available data sets. Several public microarray databases, such as DrugMatrix and TG-gates, are available which can be analyzed to generate gene expression data linked to adverse health outcomes (Igarashi et al., 2015; Svoboda et al., 2019). Gene expression data can be further analyzed using pathway enrichment methods obtained from databases, such as the Kyoto Encyclopedia of Genes and Genomes (KEGG) and Reactome to elucidate mechanisms and key biological events related to disease (Fabregat et al., 2018; Kanehisa and Goto, 2000).

An example of employing bioinformatic tools to a publicly available toxicogenomic database, DrugMatrix, was performed to obtain gene lists and enriched pathways related for hepatic fibrosis. AbdulHameed et al. (2014) employed computational bioinformatic tools such as rank product analysis and hierarchical clustering to obtain differentially expressed genes related to hepatic fibrosis. They continued to link gene expressions data with protein-protein interaction (PPI) networks to overcome limitations that may arise using pathway analysis. This combination of bioinformatic tools and gene expression datasets provide a computational framework for the identification of putative key molecular events and biomarkers that can be associated with adverse health outcomes in the framework of an AOP. The purpose of this paper is to query DrugMatrix database using computational tools based on methods by AbdulHameed et al. (2014) to identify differentially expressed genes, biological pathways, and putative key events that are related to lipid homeostasis dysfunction in liver as a possible precursor for hepatic steatosis adverse outcome.

Hepatic steatosis is considered a major liver disease among the U.S. population (Noureddin and Rinella, 2015). The main pathological feature of hepatic steatosis is excessive lipid accumulation, primarily triglyceride, within the cytosol of the hepatocyte (Angrish et al., 2016). The two main forms of hepatic steatosis are alcoholic liver disease (ALD) and nonalcoholic fatty liver disease (NAFLD). A variety of exposures can induce NAFLD, such as high-calorie high-fat diets, and environmental chemicals such as pesticides, solvents, polychlorinated biphenyls, dioxins, fungicides, herbicides, and insecticides (Al-Eryani et al., 2015; Savolainen et al., 1993).

2. Methods

2.1. Overall workflow using DrugMatrix database

DrugMatrix is a publicly available database provided by the National Institute for Environmental and Health Sciences (NIEHS) which compiles Affymetrix gene chip data from experiments conducted on Sprague Dawley rats (Svoboda et al., 2019). These data include experimental descriptions (containing dose, duration, compound, etc.) and histopathological responses from 637 chemicals treatments including drugs, biochemicals and standard toxicants.

To identify the differentially expressed genes (DEGs) associated with hepatic lipid dysfunction, raw Affymetrix Microarray data (CEL files) related to liver were first obtained from DrugMatrix database. CEL files for chemical treatments that specifically resulted in hepatic lipid dysfunction, as defined by macrovesicular lipid accumulation, were then selected. After removing outliers and preprocessing the CEL files, a set of differentially expressed genes were identified using Bioconductor rank product methods. The set of genes were then compiled via both gene ontology (GO) database and enrichments methods to obtain and visualize biological pathways and putative key events related to hepatic lipid dysfunction. Fig. 1 shows the overall workflows used in this study.

Fig. 1.

The workflow for the application of bioinformatic tools to DrugMatrix. The microarray data were downloaded from Drug Matrix website. Series of statistical and analytical tests were performed on them to identify the differentially expressed genes (DEGs). These DEGs were then uploaded in DAVID for gene ontology enrichment. The DEGs were also enriched using DAVID for KEGG pathways, IPA for canonical pathways and NCAST BioPlanet for a catalog of different types of pathways from various sources. These pathways were then visualized using Cytoscape 3.4.0.

2.2. Identifying the treatment conditions related to hepatic lipid dysfunction

Initially, any chemical treatment in the database related to hepatic lipid dysfunction histopathology observations in the liver was selected. To maximize the likelihood that genes modulations resulting from hepatic lipid dysfunction, only treatments with average severity score greater than one, and p-value less than or equal to 0.05 were included. Applying these criteria resulted in a total of 21 chemical treatments encompassing 13 different chemicals. Furthermore, some chemical treatments were excluded from analysis because their CEL files were considered outliers as illustrated the preprocessing step later. One chemical treatment for Clotrimazole 178 mg/kg for a 5-day exposure had only one CEL file replicate which resulted in its dismissal from the rank product analysis further on. The final 12 unique treatment conditions from 6 different chemicals used to derive the DEGs are reported in Table 1.

Table 1.

Chemical treatment conditions associated with hepatic lipid dysfunction from DrugMatrix database.

Chemical	Dose (mg/kg)	Duration (days)	ROUTE	Histopathology	Average severity
CARBON TETRACHLORIDE	400	3	ORAL GAVAGE	Centrilobular	1.667
CARBON TETRACHLORIDE	400	7	ORAL GAVAGE	Centrilobular	3.000
CARBON TETRACHLORIDE	1175	1	ORAL GAVAGE	Centrilobular	1.333
CARBON TETRACHLORIDE	1175	3	ORAL GAVAGE	Centrilobular	2.000
CARBON TETRACHLORIDE	1175	7	ORAL GAVAGE	Centrilobular	3.667
CLOTRIMAZOLE	89	3	ORAL GAVAGE	Nonzonal	1.333
CLOTRIMAZOLE	89	5	ORAL GAVAGE	Nonzonal	2.000
CLOTRIMAZOLE	178	5	ORAL GAVAGE	Nonzonal	2.500
ECONAZOLE	334	5	ORAL GAVAGE	Periportal	1.667
FLUCONAZOLE	394	5	ORAL GAVAGE	Nonzonal	3.000
LETROZOLE	250	5	ORAL GAVAGE	Centrilobular	1.667
SAFROLE	488	5	ORAL GAVAGE	Centrilobular	1.667

All the chemical treatments identified in Table 1 are directly or indirectly related to hepatic steatosis. Carbon tetrachloride is known to cause fatty liver disease (Becker et al., 1987). Clotrimazole, Econazole and Fluconazole are antifungal medications of the imidazole class. These chemicals have been associated with activation of PXR genes (Martin et al., 2007; Sinz, 2013). Activation of PXR is one of the mechanisms that can lead to accumulation of lipids in the liver, a mechanism that is shared with ethanol induced fatty liver disease (Choi et al., 2018). Letrozole is an aromatase inhibitor which is used in the treatment of hormonally-responsive breast cancer after surgery. Inhibition of estrogen synthesis in postmenopausal women undergoing treatment with aromatase inhibitors could increase the risk of NAFLD (Lee et al., 2019). Safrole is a member of the methylenedioxybenzene group, of which many compounds are used as insecticide synergists. Liver changes produced by safrole include hepatic cell enlargement, which was usually focal and resulted in the formation of nodules, and bile duct proliferation (Hagan et al., 1965).

2.3. Preprocessing raw data for hepatic lipid dysfunction

Initially, raw data from selected chemical treatments were mapped to controls by matching experimental conditions. To obtain gene expression data from both treated and control conditions, CEL files were processed using the package affy and the function rma from the Bioconductor library (Gautier et al., 2004; Gentleman, 2008). Next, quality control check using the package arrayQualityMetrics was performed to identify 17 outliers. Robust multi-array averages was then performed to normalize the data using the function mas5calls (AbdulHameed et al., 2014; Gentleman, 2008; Kauffmann et al., 2009). The resulting dataset was then filtered using the non-specific filtering function in the genefilter package to obtain only arrays which corresponded to Entrez ID’s (the nomenclature for gene expression used for primary text search and retrieval system defined by the National Center for Biotechnology Information) using PubMed database (Gentleman, 2008). The probe sets, the short DNA sequences representing an array or a gene sequence, were also filtered for at least 20% of the replicates for each array (“Present” calls) (AbdulHameed et al., 2014; Gentleman, 2008). Application of the rank preprocessing methods to the data generated an expression set (Expressionset) with 191 observations.

2.4. Rank product analysis and identification of DEGs

Rank Product analysis was performed on the filtered ExpressionSet to generate a set of differentially expressed genes associated with hepatic lipid dysfunction in rats. The Rank Product method is a non-parametric statistical test that converts scores from a fold change matrix to ranks (Breitling et al., 2004). These ranks indicate the likelihood that a gene will be upregulated or downregulated given a condition using the Bioconductor Package RankProd (RP).

The function RP, part of the RankProd package version 3.4.0, was applied to obtain the Rank Product statistics. Following the recommendations used in the Bioconductor Package, random seed was set to 123 to account for slight variations which occur during each permutation and to ensure reproducible results. After calculating Rank Product statistics, topGene function was used to generate a list of upregulated and downregulated genes for each of the treatment conditions reported in Table 1. To compile significantly differentially expressed genes, genes which had a p-value less than or equal to one, and percentage of false positive predictions (pfp) < 0.05 were selected (AbdulHameed et al., 2014). When mapped to treatment conditions, a list of up-regulated and down-regulated Probe IDs were generated. These Probe ID’s were converted to rat genes using the rat2302 database from the biomaRt package in the Bioconductor library (Durinck et al., 2009). These genes were then matched to human homologues by mapping rat Entrez ID to human Entrez ID’s using the biomaRt database of the Bioconductor package. Bioconductor packages org.Hs.eg.db and org.Rn.eg.db were used to identify human genes and rat genes annotations, respectively. To ensure correct mapping of the gene annotation in the database, AnnotationDbi package was also used. Finally, the DEGs tables are created by grouping all the Probe IDs that are either up-regulated or down-regulated and were generated from two or more treatment conditions. In some cases, genes were both up-regulated and down-regulated depending on the chemical treatment conditions. These genes at tabulated as “common” genes.

2.5. Pathway enrichment analysis

The pathways enrichment analysis was performed using three separate programs. NCATS BioPlanet, a publicly available web-based resource supported by National Institutes of Health (Huang et al., 2019), Ingenuity Pathway Analysis (IPA) software (Ingenuity Systems, Redwood City, CA), and Integrated Discovery (DAVID) web-based application (Huang da et al., 2009). DAVID application was used to perform KEGG pathway analysis and GO classification. IPA analysis was used to perform canonical pathway analysis and visualization. NCAST BioPlanet catalogues all publicly available pathways and their annotations from several sources such as KEGG, Reactome, NetPath, Wiki pathways, NCI-Nature, and BioCarta. For this reason, NCAST BioPlanet was used to investigate a broader range of genes and pathways.

The IPA pathway analysis necessitated the use of fold-change data for genes that are up-or down-regulated. The fold-changes were calculated using a log-ratio matrix of genes following methods by AbdulHameed et al. (2014). First the average intensity of the replicates of a chemical exposure condition was calculated, then the log-ratio of fold change between the treatments and their corresponding controls for each gene was computed. This calculation was carried out two ways: (1) using the average of the fold change across all the chemicals (AllFC) or (2) using fold changes for each different treatment conditions (SepFC).

The canonical pathways analysis is used to predict pathways that are changing based on gene expression and are most significant to the data set. The significance of the canonical pathways is based on two parameters: (1) a ratio of the number of DEGs from the data set that map to the pathway divided by the total number of genes that map to the canonical pathway and (2) a p value calculated using Fischer’s exact test (FDR < 0.05) determining the probability that the association between the DEGs in the data set and the canonical pathway is not due to chance alone (Savil et al., 2008). For the continuity purposes, the Fisher’s exact test p-value was changed to Benjamini-Hochnerg p-value (BH p-value) and only the ones with −log(BH p-value) > 1.3 which is equivalent to FDR < 0.05 were selected. Another IPA utility for the generation of biological pathways is based on interactions between genes. This analysis is performed using a generated z-score in the software which enables the predictions about upstream or downstream processes. Z-scores account for the directional effect of one molecule on another molecule or on a process, and the direction of change of molecules in the dataset. Z-scores can be reported as positive, negative, zero or not applicable when no activity pattern is available.

An additional method to develop the biological pathways is to introduce the derived DEGs lists to the web-based application DAVID. Again, to ensure the significance of the resulting pathways, only the ones with the Benjamini-Hochnerg FDR < 0.05 were selected. The corresponding significant KEGG pathways and biological process GO terms were then saved to be visualized using Cytoscape 3.4.0 software (Supplemental Material Figs. 1 and 2). A similar method was used to obtain biological pathways using NCAST BioPlanet. The DEGs were first uploaded on the web-based interface. The resulting significant pathways were then narrowed down using a cutoff value of 0.05, (equivalent to p < 0.05 in Fisher exact test).

3. Results

3.1. Hepatic lipid dysfunction relevant differentially expressed genes

The analysis of the chemical treatments associated with hepatic lipid dysfunction in DrugMatrix resulted in the identification of 599 hepatic lipid dysfunction relevant genes, of which 324 genes were significantly up-regulated, 275 genes were significantly down-regulated, and 36 genes were both up-regulated by some treatment conditions and down-regulated by another.

Several genes identified from the database can be directly linked to biological events related to lipid synthesis, metabolism and homeostasis in the liver. Many of the cytochrome P450 family genes were differentially expressed due to hepatic lipid dysfunction. Cytochrome P450 enzymes are involved in the metabolism of xenobiotics, drugs, cholesterol, steroids, and lipids. Other genes such as lipoprotein lipase (LPL), and very low-density lipoprotein receptor (VLDLR) are involved in metabolism and transport of fatty acids. The LDL receptor related protein 1 (LRP1) and VLDLR were both down-regulated. Aldo-keto reductase family (AKR1C2, 1C4, 7A2) and hydroxysteroid beta dehydrogenase (HSD17B6, HSD3B1, HSD3B2) are also involved in lipid and steroid metabolism and are down-regulated. Glucose-6-phosphatase, catalytic subunit (G6PC) is involved in glycolysis and is also down-regulated. All results for identified genes and the treatment conditions related to their regulations are reported in Supplementary File 1. The corresponding genes fold changes were also computed and are given in Supplementary File 2.

3.2. Pathway enrichment analysis

The combined list of all DEGs from DrugMatrix were uploaded in DAVID and the gene ontology (GO) for biological process-terms were enriched. Examples of GO terms that are related to hepatic steatosis were lipid, steroid, cholesterol, lipoprotein, and acyl-CoA metabolic processes, cholesterol homeostasis and lipid hydroxylation. The complete list of the GO terms is reported in Supplementary File 3. The relationship between GO terms, DEGs and chemical treatments can also be shown using heatmaps. Specifically, the fold changes of genes related to lipid metabolic process (GO0006629) are depicted in relationship to chemical treatments in Fig. 2.

Fig. 2.

A heatmap showing the relationship between genes, and chemical treatment conditions using the gene ontology for lipid metabolic processes (GO0006629). Some human homolog genes can share different rat gene probe ids. (*) The second occurrence of the same gene. (**) the third occurrence of the same gene.

Uploading the gene list in DAVID resulted in 45 KEGG pathways. To only account for the significantly enriched pathways that are related to lipid homeostasis, 9 pathways with the Benjamini-Hochberg FDR < 0.05 were selected (Table 2a). The top 4 pathways are chemical carcinogenesis, metabolism of xenobiotics by cytochrome P450, steroid hormone biosynthesis and metabolic pathways. The up-regulated and the down-regulated DEGs were also uploaded in DAVID separately. The up-regulated genes recognized 28 KEGG pathways of which 15 were significant. Only three up-regulated pathways can be related to hepatic lipid dysfunction (Table 2b). The down-regulated genes were used to obtain 24 KEGG pathways of which only 6 were significant and related to hepatic lipid homeostasis (Table 2c). The full list of all pathways are reported in Supplementary File 4.

Table 2a.

KEGG pathway enrichment analysis for all DEGs.

KEGG Pathway	Count	Molecules	BH
Chemical carcinogenesis	18	GSTA1, CYP3A4, GSTA2, CYP3A5, CYP3A7, CYP1A1, GSTA5, SULT2A1, CYP2C18, CYP2C8, ADH1C, ADH1B, ADH6, ADH1A, CYP3A7-CYP3A51P, CYP3A43, GSTM2, CBR1	1.640E-06
Metabolism of xenobiotics by cytochrome P450	17	GSTA1, CYP3A4, GSTA2, CYP3A5, CYP1A1, GSTA5, SULT2A1, ADH1C, ADH1B, ADH6, ADH1A, AKR1C2, GSTM2, CBR1, AKR1C4, AKR7A2, AKR7A3	1.776E-06
Steroid hormone biosynthesis	13	CYP3A4, HSD3B2, CYP3A5, AKR1C2, CYP17A1, AKR1C4, HSD3B1, CYP3A7, CYP1A1, HSD17B6, SRD5A1, COMT, CYP3A7-CYP3A51P	9.884E-05
Metabolic pathways	77	ACOX2, CYP3A4, TM7SF2, CYP3A5, TUSC3, CYP3A7, GNPDA1, CYP2C18, ADH1C, ACOT2, ADH1B, ACOT1, ADH1A, ALAS1, AKR1C4, ALAS2, TRAK2, TKFC, HDC, RGN, PIK3C2G, CYP1A1, OTC, HAL, ALOX15, G6PC, G6PD, SDS, ALDH1B1, AKR1B10, RRM2, HAO2, PKLR, GAA, ABAT, GPAM, OAT, AOC3, PRODH, HSD3B2, HSD3B1, SRM, LOC102724788, ENPP3, HMGCS1, CTPS1, ADH6, ASNS, COMT, CMPK2, ALDH1A1, ISYNA1, CBR1, IVD, PLA2G12A, CYP26B1, FASN, HSD17B6, PAFAH1B1, MTMR7, PLA2G16, CES1, CYP2C8, ACLY, CYP3A7-CYP3A51P, PCK1, ACSM3, CYP17A1, GCK, PHGDH, PRODH2, HIBCH, RDH16, LIPC, IDI1, CYP8B1, ACSM5	2.908E-04
Bile secretion	13	ABCG8, FXYD2, SLCO1A2, ABCG5, NCEH1, AQP9, SULT2A1, SLC22A8, ABCC3, ABCC4, CA2, SLC10A2, ABCB4	4.453E-04
Drug metabolism - cytochrome P450	12	GSTA1, CYP3A4, GSTM2, GSTA2, CYP3A5, GSTA5, FMO1, CYP2C8, ADH1C, ADH6, ADH1B, ADH1A	1.577E-03
p53 signaling pathway	11	STEAP3, CDKN1A, CCND1, RRM2, GADD45G, RPRM, MDM2, GADD45B, CCNG1, IGFBP3, GADD45A	5.031E-03
Glutathione metabolism	9	GSTA1, GPX2, GSTM2, GSTA2, GSR, G6PD, GSTA5, SRM, RRM2	1.193E-02
Glycolysis/Gluconeogenesis	9	G6PC, GCK, ALDH1B1, PKLR, ADH1C, ADH6, ADH1B, ADH1A, PCK1	4.935E-02

Table 2b.

KEGG pathway enrichment for up-regulated DEGs.

KEGG Pathway	Count	Molecules	BH
Chemical carcinogenesis	10	CYP3A4, CYP3A43, CYP3A5, CBR1, CYP3A7, CYP1A1, SULT2A1, GSTA5, CYP2C8, CYP3A7-CYP3A51P	1.169E-03
Bile secretion	8	FXYD2, SLCO1A2, NCEH1, SULT2A1, ABCC3, ABCC4, CA2, ABCB4	1.065E-02
Steroid hormone biosynthesis	7	CYP3A4, CYP3A5, CYP17A1, CYP3A7, CYP1A1, COMT, CYP3A7-CYP3A51P	2.182E-02

Table 2c.

KEGG pathway enrichment for down-regulated DEGs.

KEGG pathway	Count	Molecules	BH
Metabolic pathways	52	CYP3A4, ACOX2, TM7SF2, CYP3A5, CYP3A7, CYP2C18, ADH1C, ADH1B, ADH1A, AKR1C4, ALAS2, TKFC, RGN, PIK3C2G, OTC, HAL, G6PC, ALOX15, ALDH1B1, RRM2, AKR1B10, PKLR, HAO2, ABAT, GPAM, OAT, PRODH, AOC3, HSD3B2, HSD3B1, LOC102724788, ENPP3, HMGCS1, ADH6, IVD, FASN, HSD17B6, PAFAH1B1, MTMR7, PLA2G16, CES1, ACLY, CYP3A7-CYP3A51P, PCK1, ACSM3, CYP17A1, GCK, PRODH2, RDH16, CYP8B1, IDI1, LIPC	1.579E-06
Chemical carcinogenesis	13	GSTA1, CYP3A43, CYP3A4, GSTA2, GSTM2, CYP3A5, CYP3A7, CYP2C18, ADH1C, ADH1B, ADH6, ADH1A, CYP3A7-CYP3A51P	2.889E-06
Steroid hormone biosynthesis	11	CYP3A4, HSD3B2, CYP3A5, AKR1C2, CYP17A1, AKR1C4, HSD3B1, CYP3A7, HSD17B6, SRD5A1, CYP3A7-CYP3A51P	7.384E-06
Metabolism of xenobiotics by cytochrome P450	12	GSTA1, CYP3A4, GSTM2, GSTA2, CYP3A5, AKR1C2, AKR1C4, ADH1C, AKR7A2, ADH6, ADH1B, ADH1A	6.167E-06
Drug metabolism - cytochrome P450	10	GSTA1, CYP3A4, GSTM2, GSTA2, CYP3A5, FMO1, ADH1C, ADH6, ADH1B, ADH1A	1.633E-04
Glycolysis/Gluconeogenesis	9	G6PC, GCK, ALDH1B1, PKLR, ADH1C, ADH6, ADH1B, ADH1A, PCK1	1.032E-03

The same combined list and its corresponding fold change for each gene (AllFC) was uploaded in IPA and the ingenuity canonical pathways were observed. The total of 78 pathways were recognized, but only 15 were significant based on −log (B–H p-value) > 1.3, 8 of which were related to lipid homeostasis. Table 3a lists these pathways and their associated genes. The up-regulated and down-regulated gene lists were also uploaded separately in IPA. For up-regulated genes 51 ingenuity canonical pathways were recognized of which 6 were significant and related to lipid homeostasis (Table 3b). For down-regulated genes 37 ingenuity canonical pathways were recognized of which only the bile acid biosynthesis is considered significant and may be related to lipid dysfunction. The complete list of all ingenuity canonical pathways is given in Supplementary File 5.

Table 3a.

Ingenuity canonical pathway enrichment for all DEGs.

Ingenuity canonical pathways	Ratio	Molecules	BH
Estrogen Biosynthesis	0.108	POR,CYP1A1,AKR1C4,CYP2C8	3.715E-04
PXR/RXR activation	0.069	ALDH1A1,ALAS1,ABCC3,CYP2C8	1.047E-03
Acetone Degradation I (to Methylglyoxal)	0.111	POR,CYP1A1,CYP2C8	1.445E-03
Bile Acid Biosynthesis, neutral pathway	0.167	AKR1C4,CYP8B1	5.129E-03
LPS/IL-1 Mediated Inhibition of RXR FUNCTION	0.0219	ALDH1A1,ALAS1,ABCC3,CYP2C8	1.820E-02
Xenobiotic Metabolism signaling	0.0163	CYP1A1,ALDH1A1,ABCC3,CYP2C8	3.981E-02
Aryl Hydrocarbon receptor signaling	0.0231	CYP1A1,ALDH1A1,HSPB1	4.074E-02
Hepatic Cholestasis	0.0214	ABCB4,ABCC3,CYP8B1	4.677E-02

Table 3b.

Ingenuity Canonical pathway enrichment for up-regulated DEGs.

Ingenuity canonical pathways	Ratio	Molecules	BH
PXR/RXR Activation	0.069	ALDH1A1,ALAS1,ABCC3,CYP2C8	8.511E-05
Acetone Degradation I (to Methylglyoxal)	0.111	POR,CYP1A1,CYP2C8	1.514E-04
Estrogen Biosynthesis	0.0811	POR,CYP1A1,CYP2C8	3.020E-4
LPS/IL-1 Mediated inhibition of RXR function	0.0219	ALDH1A1,ALAS1,ABCC3,CYP2C8	8.913E-04
Xenobiotic Metabolism signaling	0.0163	CYP1A1,ALDH1A1,ABCC3,CYP2C8	2.455E-03
Aryl Hydrocarbon receptor signaling	0.0231	CYP1A1,ALDH1A1,HSPB1	4.571E-03

Several of the canonical pathways that were generated using IPA are selected and visualized using Cystoscope 3.4.0. The selection of these processes was performed based on 1) their significance using the BH < 0.05 criterion, and 2) their biological impact on hepatic lipid homeostasis. Fig. 3 illustrates the relationships between significant pathways that are shared via common genes. For instance, the activation of HSPB1 gene is involved in Aryl hydrocarbon receptor signaling which shares the ALDH1A1 gene with either PXR/RXR activation or LPS/IL-Mediated inhibtion of RXR function. ALDH1A1 gene regulations impact aldehyde dehydrogenase as the second enzyme of the major oxidative pathway of alcohol metabolism (Li et al., 2018) while PXR activation has been implicated with increase influx of lipids from blood into liver via receptor CD36 (Li et al., 2018; Zhou et al., 2008). LPS/IL inhibition of RXR function is further linked to hepatic cholestasis which can be related to steatosis (Shipovskaya and Dudanova, 2018). Although the relationships in Fig. 3 are restricted by chemical treatment conditions, and criteria set by the bioinformatic analysis, they can be a starting point for identifying key events and pathways that may be related to hepatic lipid homeostasis or its downstream effects.

Fig. 3.

The Ingenuity canonical pathways for all genes generated using AllFC visualized by Cytoscape 3.4.0. Only the significant pathways with BH < 0.05 that are related to lipid dysfunction are selected for this illustration.

The combined list (AllFC) was also uploaded in NCAST BioPlant Enrichment tab and all the pathways were observed. A total of 43 significant pathways were identified using the cutoff value of 0.05. However, only 21 pathways that are specifically related to hepatic lipid dysfunction are given in Table 4. In addition to cytochrome P450 drug metabolism, they include fatty acid omega oxidation, lipid and lipoprotein metabolism, FOXM1 transcription factor network and adipogenesis. The complete list of the NCAST BioPlanet identified pathways is given in Supplementary File 6.

Table 4.

BioPlanet Pathways for all DEGs.

BioPlanet pathway name	Pathway source	Count	P-value
Biological oxidations	Reactome	139	1.52E-03
Metabolism	KEGG	1615	1.81E-07
Phase I of biological oxidations: functionalization of compounds	Reactome	69	1.81E-07
Drug metabolism: cytochrome P450	KEGG	83	1.83E-06
Metapathway biotransformation	Wiki Pathways	174	3.15E-06
Steroid hormone biosynthesis	KEGG	59	7.12E-05
Cytochrome P450 pathway	Wiki Pathways	61	5.60E-04
Fatty acid omega oxidation	Wiki Pathways	15	5.90E-04
Lipid and lipoprotein metabolism	Reactome	489	8.10E-04
Linoleic acid metabolism	KEGG	35	1.12E-03
p53 signaling pathway	BioCarta	139	1.52E-03
Oxidative stress-induced gene expression via Nrf2	BioCarta	21	2.29E-03
Cytochrome P450 metabolism of xenobiotics	Reactome	15	3.49E-03
HIF-1 transcriptional activity in hypoxia	NCI-Nature	66	3.49E-03
Ethanol oxidation	Reactome	10	5.78E-03
Statin pathway	Wiki Pathway	29	8.99E-03
Ghrelin-mediated regulation of food intake and energy	Manual Curation	13	1.32E-02
TSH regulation of gene expression	NetPath	97	1.47E-02
Bile acid and bile salt metabolism	Reactome	27	2.71E-02
FOXM1 transcription factor network	NCI-Nature	41	3.74E-02
Adipogenesis	Wiki Pathways	133	4.48E-02

4. Discussion

Several databases, such as TG-GATEs and DrugMatrix, are publicly available where bioinformatics tools can be used to identify gene modulations in response to chemical treatment in rats and other species in vivo and in vitro. The identification of these genes in relationship to specific pathological evidence provide a computational format to build networks to assist in developing putative AOPs de novo or verify existing ones. In this effort, genes that are involved in the hepatic lipid dysfunction were identified using DrugMatrix. Once identified, the set of genes and their biological pathways can be used to develop key events for a putative AOP for an adverse outcome related to hepatic lipid dysfunction such as hepatic steatosis.

Hepatic steatosis is an outcome of disruptions of lipid homeostasis in hepatocytes through a complex network of biological events that can be generalized in 4 apical key events: hepatic fatty acid (FA) uptake, de novo FA and lipid synthesis, FA oxidation, and lipid efflux (Angrish et al., 2016). Some of the genes that were identified using the DrugMatrix database for hepatic lipid dysfunction can be directly linked to the apical events for hepatic steatosis. For instance, acyl_CoA thioestrase (ACOT1, 2, 9) is a key and highly regulated enzyme that is required for the biosynthesis fatty acyl-CoA (Paton and Ntambi, 2009). Other genes were identified that are involved in metabolism and transport of fatty acids include lipoprotein lipase (LPL), and Very low-density lipoprotein receptor (VLDLR). These genes are also involved in RXR activation which is highly regulated by fatty acid synthesis (Steineger et al., 1997). Other genes that were identified are also involved in lipid and steroid metabolism. Among them are Aldo-keto reductase family (AKR1C2, 1C4, 7A2), and hydroxysteroid beta dehydrogenase (HSD17B6, HSD3B2). Other identified down-regulated genes including apolipoprotein L (APOL1, 2, 3, 4), acetyl-CoA acetyltransferase 2 (ACAT2), and acyl-CoA oxidase 2 (ACOX2) which are all involved in fatty acid and lipid metabolism and transfer. Several of CYP450 genes were also identified using the database. Several biological pathways identified using the database can be linked to hepatic lipid homeostasis such as glycolysis/gluconeogenesis, fatty acid oxidation, and adipogenesis, and PXR/RXR activation. The collection of the identified genes and their associated biological pathways and key events is an initial step toward the computational development of a putative AOP for hepatic steatosis.

The application of bioinformatic methods to DrugMatrix provides information that can be useful for identifying biological pathways related to an adverse outcome. However, the information is limited by several factors that are related to both the data and computational methods. The identification of significant pathways is restricted to the criteria based on the BH score being < 0.05. Using this criterion will bias the results toward pathways that may not necessarily be associated with the pathological end point of concern in view of the large number of identified DEGs and the complex biological interactions between them. Additionally, the decision to use data with high a pathological severity score will result in gene modulations that may be related to several downstream effects of lipid dysfunction such as cholestasis, and inflammation-related immuno response pathways. There is also the possibility of identifying several gene modulations and pathways directly linked to chemical effects such as metabolic processes or cytotoxicity responses. Some of these limitations can be overcome by increasing the number of selected chemical treatments associated with hepatic lipid homeostasis using a more integrative approach of utilizing all data from several databases combined with literature information and targeted in vitro and in vivo experiments.