Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2017 Feb 8;8:14250.
doi: 10.1038/ncomms14250.

Reconciled Rat and Human Metabolic Networks for Comparative Toxicogenomics and Biomarker Predictions

Affiliations
Free PMC article

Reconciled Rat and Human Metabolic Networks for Comparative Toxicogenomics and Biomarker Predictions

Edik M Blais et al. Nat Commun. .
Free PMC article

Abstract

The laboratory rat has been used as a surrogate to study human biology for more than a century. Here we present the first genome-scale network reconstruction of Rattus norvegicus metabolism, iRno, and a significantly improved reconstruction of human metabolism, iHsa. These curated models comprehensively capture metabolic features known to distinguish rats from humans including vitamin C and bile acid synthesis pathways. After reconciling network differences between iRno and iHsa, we integrate toxicogenomics data from rat and human hepatocytes, to generate biomarker predictions in response to 76 drugs. We validate comparative predictions for xanthine derivatives with new experimental data and literature-based evidence delineating metabolite biomarkers unique to humans. Our results provide mechanistic insights into species-specific metabolism and facilitate the selection of biomarkers consistent with rat and human biology. These models can serve as powerful computational platforms for contextualizing experimental data and making functional predictions for clinical and basic science applications.

Conflict of interest statement

The authors declare no competing financial interests.

Figures

Figure 1
Figure 1. GENREs of human (iHsa) and rat (iRno) metabolism were reconciled for comparative analyses.
(a) Reactions from iHsa were transferred to iRno when GPR rules consisting of human genes could be replaced with equivalent GPR rules consisting of rat orthologs. A network-driven approach was developed to filter orthology annotations based on consensus from five databases: Homologene, KEGG, Uniprot, Rat Genome Database (RGD) and Ensembl. (b) iRno and iHsa were manually curated in parallel to capture species-specific reactions and to avoid introducing unverifiable differences based on genotype and phenotype information across various resources. This process, called network reconciliation, contributed to the identification and removal of several rat-specific reactions that are present in HMR2 and Recon 2. (c) Subsystem-level comparison of the knowledge gap that exists between rats and humans. Stacked bars represent the percentages of PubMed articles mapped to rat and/or human genes for all metabolic genes represented in a subsystem. PubMed articles referenced human genes more frequently than rat genes within all subsystems, but the knowledge gap was larger for pathways that included one or more human-specific reactions. ETC, electron transport chain; PPP, pentose phosphate pathway.
Figure 2
Figure 2. Reconciled GPR relationships between iRno and iHsa allow for varying degrees of redundancy.
(a) Comparison of the number of genes catalysing each reaction in iRno and iHsa. Gene-associated reactions capable of catalysis in both iRno and iHsa were classified as ‘shared' reactions (purple). Reactions associated with GPR rules in only one organism were classified as species-specific (red and blue for rat-specific and human-specific, respectively). Reactions present in both models that had no known GPR rule assignments were classified as non-enzymatic (grey). Each tile's colour intensity represents the (log-scaled) frequency of reactions in that bin. The annotated letters b,c refer to the individual tiles from which the reactions in b,c are binned. (b) Example of a shared reaction with balanced GPR rules in iRno and iHsa. This reaction, glutamate-cysteine ligase (EC 6.3.2.2), requires both a catalytic subunit (Gclc/GCLC) and a regulatory subunit (Gclm/GCLM) to join glutamine with cysteine and form γ-glutamyl-cysteine, a precursor of glutathione. This reaction was manually curated, because the original HMR2 did not contain information related to protein complexes in GPR rules. (c) Example of a shared reaction, adenosine aminohydrolase (EC 3.5.4.4), which is involved in purine degradation and can be catalysed by two redundant human isozymes or one rat enzyme.
Figure 3
Figure 3. Functional differences known to distinguish rat and human metabolism.
(a) Rats are capable of synthesizing vitamin C from limited substrates, providing an inherent resistance to scurvy. The rat-specific enzyme, Gulo, catalyses the last enzymatic step of the vitamin C synthesis pathway: L-gulonolactone oxidase (EC 1.1.3.8). The human orthologue of Gulo is a non-functional pseudogene (GULOP). (b) Gout formation is associated with accumulation and crystallization of urate, the end product of purine catabolism in humans. Rats are resistant to gout, because urate can be further degraded into allantoin, which is more soluble than urate. (c) Most mammals can synthesize the monosaccharide, Neu5Gc, which is known as a nonhuman sialic acid that is incorporated into glycoproteins. (d) The human-specific enzyme FUT3 synthesizes the Lewisa antigen (Lea) which is involved in Lewis blood group determination and is the precursor for the pancreatic cancer biomarker, CA19-9. (e) Metabolic tasks simulating the production of primary and secondary bile acids were consistent with bile acid species reported in a previous study that compared rat and human liver samples (X's indicate absence of production). (f) Summary of rat-specific and shared primary bile acid synthesis routes from cholesterol. Cyp3a18 was hypothesized as the critical enzyme enabling rats to produce rat-specific primary bile acids, which had not been previously described. (g) In the process of ‘bile acid recycling', primary bile acids secreted by the liver into the gut are transformed by bacteria and subsequently reabsorbed by the liver. Synthesis of secondary bile acids were accounted for in iRno and iHsa by including extracellular reactions associated with gut bacteria. Interplay between rat liver and gut metabolism was necessary for iRno to simulate the synthesis and secretion of the rat-specific bile acid, murideoxycholic acid (MDCA). CA19-9, carbohydrate antigen 19-9; LacCer, lactosylceramide; ManNAc, N-acetylmannosamine; Neu5Ac, N-acetylneuraminic acid; Neu5Gc, N-glycolylneuraminic acid.
Figure 4
Figure 4. TIMBR is a novel method for predicting treatment-induced biomarkers by integrating gene expression changes into metabolic networks.
(a) TIMBR calculates reaction weights using log2 fold changes of significantly (FDR<0.1) differentially expressed genes. For each reaction, log2 fold changes are averaged across isozymes after assigning a value of 0 to any insignificant changes. For reactions associated with protein complexes, the subunit with the largest value after averaging is selected. Summarized values are then transformed into larger (or smaller) reaction weights for representing relative expression between treatment and control conditions. (b) Caffeine-induced gene expression changes are displayed as a volcano plot for rat hepatocytes. (c) Optimization problem formulated by TIMBR to estimate the global network demand needed to produce a metabolite. The objective function minimizes the sum of all reaction fluxes (v) multiplied by TIMBR reaction weights (w). Treatment and control conditions were simulated separately for each potential biomarker under similar physiological lower-bound (lb) and upper-bound (ub) constraints that assumed steady-state reaction fluxes. The minimum required production rate for each metabolite was set to either a rate of 100 fmol per cell per hour or 90% of the maximum possible flux value, whichever was smaller. (d) Optimal caffeine-weighted (wtreatment) and control-weighted (wcontrol) flux distributions (vtreatment and vcontrol) for biomarker production of urea determined by integrating gene expression changes from (b) into iRno. Non-zero fluxes that were higher (purple), equal (grey) or lower (orange) relative to the other condition were displayed using MetDraw. Arrow thickness represents the inverse reaction weight as described in a. In this example, the global network demand (sum of weighted fluxes) was smaller in the treatment condition than the control condition, indicating that caffeine induced expression changes that were more consistent with the production of urea compared to controls. (e) Raw production scores in response to individual treatment strategies were calculated for each metabolite separately by comparing global network demands determined in c for the treatment and control conditions. TIMBR production scores represent these raw production scores normalized across all relevant metabolites with biomarker predictions.
Figure 5
Figure 5. Validation of caffeine-induced biomarker predictions for rat hepatocytes.
Comparison of rat production scores calculated by TIMBR in response to caffeine with previously reported average log2 fold changes in metabolite concentrations after caffeine exposure.
Figure 6
Figure 6. Comparative analyses of rat and human biomarker predictions and species-specific trends across perturbations.
(a) Heatmap of 16 metabolite biomarkers predicted to increase (purple) or decrease (orange) in response to 16 individual compounds. Metabolite production scores for rat (upper left triangle) and human (lower right triangle) hepatocytes were generated by integrating treatment-induced gene expression changes into iRno and iHsa using TIMBR. Rat and human production scores across all 286 metabolites were classified as positively correlated (FDR<0.1), uncorrelated or negatively correlated (FDR<0.1) for each individual compound. Compounds were ordered by correlation coefficients and metabolites were ordered by average production scores across all 76 compounds.BHB, β-hydroxybutryate; PGE2, prostaglandin E2. (b) Scatterplot comparing rat and human production scores for PGE2 across 76 compounds. Two antipyretic compounds with known cyclooxygenase inhibitor activities, acetaminophen and ibuprofen, were predicted to consistently decrease prostaglandin E2 production in both rat and human hepatocytes. (c) Scatterplot comparing rat and human production scores for urate across 76 compounds. Rat production scores for urate were consistently decreased by anti-gout medications that are known to reduce urate accumulation (colchicine, phenylbutazone, benziodarone and benzbromarone). Human production scores were also decreased for anti-gout compounds with the exception of benzbromarone. (d) Rat production scores in response to two xanthine derivatives, caffeine and theophylline, were strongly correlated. Biomarker predictions associated with glutamate, urate, glucose and urea were individually consistent across both compounds. (e) Human production scores in response to theophylline and caffeine were less correlated than rat production scores. Glutamate and urate were predicted to increase in response to theophylline and decrease in response to caffeine for human hepatocytes, whereas glucose and urea predictions were consistent across both compounds. Compared with rat production scores (d), urate, glucose and glutamate would be considered species-specific predictions. (f) Chemical structures for theophylline and caffeine differ by a single methyl group. Rat biomarker predictions did not indicate any potential differences between the two xanthine derivatives.In patients, theophylline is known to cause increased serum levels of urate.
Figure 7
Figure 7. Validation of iRno and iHsa biomarker predictions in response to theophylline.
(ad) Comparative biomarker predictions were experimentally validated in vitro using primary rat hepatocytes or an immortalized human hepatocyte cell line (HepG2). Extracellular changes in metabolite concentrations were measured after 24 h of treatment with theophylline (10 μM) or control (0 μM). To the left of each metabolite, predictions are summarized as elevated (up arrow), reduced (down arrow) or unchanged (equals sign). Below each measured metabolite, FDR-adjusted q-values are displayed for rat and human experimental comparisons between treatment (n=4) and control (n=4) sample concentrations using an unpaired two-sided Student's t-test. Individual points represent biological replicates from one rat and one human experiment. Of the eight predictions tested, seven were experimentally confirmed, while urea from HepG2 cells (d) was insignificant but trending in the expected direction. Horizontal lines represent average metabolite concentrations from two fresh media replicates and indicate that all metabolites were being produced on average with the exception of glucose in HepG2 cells. (e) Quantitative comparisons between model predictions and experimental results across metabolites. Positive and negative values represent elevated and reduced biomarkers based on TIMBR production scores (x axis) and average log2 fold changes between treatment and control (y axis) concentrations displayed in ad.

Similar articles

See all similar articles

Cited by 25 articles

See all "Cited by" articles

References

    1. Waters M. D. & Fostel J. M. Toxicogenomics and systems toxicology: aims and prospects. Nat. Rev. Genet. 5, 936–948 (2004). - PubMed
    1. Igarashi Y. et al. . Open TG-GATEs: a large-scale toxicogenomics database. Nucleic Acids Res. 43, D921–D927 (2015). - PMC - PubMed
    1. Sun J. et al. . Identification of a metabolic biomarker panel in rats for prediction of acute and idiosyncratic hepatotoxicity. Comput. Struct. Biotechnol. J. 10, 78–89 (2014). - PMC - PubMed
    1. Lin Z. et al. . Joint MS-based platforms for comprehensive comparison of rat plasma and serum metabolic profiling. Biomed. Chromatogr. 28, 1235–1245 (2014). - PubMed
    1. Mattes W. et al. . Detection of hepatotoxicity potential with metabolite profiling (metabolomics) of rat plasma. Toxicol. Lett. 230, 467–478 (2014). - PubMed

Publication types

Feedback