Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2016 Oct 10;17(1):790.
doi: 10.1186/s12864-016-3143-y.

Mining Kidney Toxicogenomic Data by Using Gene Co-Expression Modules

Affiliations
Free PMC article

Mining Kidney Toxicogenomic Data by Using Gene Co-Expression Modules

Mohamed Diwan M AbdulHameed et al. BMC Genomics. .
Free PMC article

Abstract

Background: Acute kidney injury (AKI) caused by drug and toxicant ingestion is a serious clinical condition associated with high mortality rates. We currently lack detailed knowledge of the underlying molecular mechanisms and biological networks associated with AKI. In this study, we carried out gene co-expression analyses using DrugMatrix-a large toxicogenomics database with gene expression data from rats exposed to diverse chemicals-and identified gene modules associated with kidney injury to probe the molecular-level details of this disease.

Results: We generated a comprehensive set of gene co-expression modules by using the Iterative Signature Algorithm and found distinct clusters of modules that shared genes and were associated with similar chemical exposure conditions. We identified two module clusters that showed specificity for kidney injury in that they 1) were activated by chemical exposures causing kidney injury, 2) were not activated by other chemical exposures, and 3) contained known AKI-relevant genes such as Havcr1, Clu, and Tff3. We used the genes in these AKI-relevant module clusters to develop a signature of 30 genes that could assess the potential of a chemical to cause kidney injury well before injury actually occurs. We integrated AKI-relevant module cluster genes with protein-protein interaction networks and identified the involvement of immunoproteasomes in AKI. To identify biological networks and processes linked to Havcr1, we determined genes within the modules that frequently co-express with Havcr1, including Cd44, Plk2, Mdm2, Hnmt, Macrod1, and Gtpbp4. We verified this procedure by showing that randomized data did not identify Havcr1 co-expression genes and that excluding up to 10 % of the data caused only minimal degradation of the gene set. Finally, by using an external dataset from a rat kidney ischemic study, we showed that the frequently co-expressed genes of Havcr1 behaved similarly in a model of non-chemically induced kidney injury.

Conclusions: Our study demonstrated that co-expression modules and co-expressed genes contain rich information for generating novel biomarker hypotheses and constructing mechanism-based molecular networks associated with kidney injury.

Keywords: AKI networks; AKI pathways; Acute kidney injury; Cd44 ectodomain; Frequently co-expressed genes; Gene signature; Havcr1; Immunoproteasome; KIM-1; Kidney co-expression modules; Toxicogenomics.

Figures

Fig. 1
Fig. 1
Workflow used in this study to mine kidney toxicogenomic data
Fig. 2
Fig. 2
Iterative signature algorithm (ISA) parameter selection. a Number of modules generated for different combinations of gene and sample thresholds and for the merged results. b Percentage of modules enriched with gene ontology (GO) terms for different threshold values and merged results. “All,” merged result from all threshold combinations; %MGO, the percentage of modules enriched with GO terms; Nm, the number of modules generated
Fig. 3
Fig. 3
a Heat map view of modules clustered based on the module overlap score. b Activation of module clusters (MC1-16) for different phenotypes (P1-2) and chemical classes (C1-6). P1, chemical exposures that cause kidney-cortex, tubule, necrosis; P2, chemical exposures that cause kidney-tubule, regeneration; C1, chemical exposure known to cause hepatotoxicity; C2, fluoroquinolone antibiotics; C3, epithelial growth factor receptor kinase inhibitors; C4, estrogen receptor modulators; C5, high dose of statin drugs; C6, high dose of anti-lipidemic drugs (fibrates)
Fig. 4
Fig. 4
Receiver Operator Characteristics for the 30-gene signature, Havcr1, and Clu. The model uses early (1–5 days) transcription data to predict the future onset of kidney injury (at 28 days). The true positive rate is the rate of true predictions divided by true predictions and false positives; the false positive rate is the rate of false predictions divided by the false predictions and false negatives. The diagonal line indicates random predictions
Fig. 5
Fig. 5
Genes in the acute kidney injury (AKI)-relevant gene set mapped to the enriched Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways. The color of the gene nodes indicates the log2-fold change and the connecting lines represent membership in the KEGG pathways (P1-P12) listed in Table 3
Fig. 6
Fig. 6
Acute kidney injury (AKI)-relevant human protein-protein interaction sub-network. The protein nodes are distributed according to cellular localization. The size of the node represents the number of connections in the sub-network. The nodes are colored according to the average log2 fold-change ratio in chemical exposures that cause kidney necrosis. Proteins encoded by genes with average log2 fold-change ratios greater than 0.6 are shown in red. Proteins encoded by genes with average log2 fold-change ratios between 0.6 and −0.6 are shown in grey. Proteins encoded by genes with average log2 fold-change ratios less than −0.6 are shown in green. Orange stars denote hub proteins with >5 connections, green stars denote non-hub proteins with a high betweenness centrality (>0.09). The red star and dotted circle identify the highest interconnected region of the network associated with the immunoproteasome
Fig. 7
Fig. 7
Acute kidney injury-subnetwork genes associated with different anatomical regions of the kidney
Fig. 8
Fig. 8
Frequently co-expressed genes with Havcr1. The size of the node represents the number of times the gene was co-expressed with Havcr1 in the modules. The nodes are colored according to the average log2 fold-change ratio in chemical exposures causing kidney necrosis. Proteins encoded by genes with average log2 fold-change ratios greater than 0.6 are shown in red. Proteins encoded by genes with average log2 fold-change ratios between 0.6 and −0.6 are shown in grey. Proteins encoded by genes with average log2 fold-change ratios less than −0.6 are shown in green
Fig. 9
Fig. 9
Scatterplots of log2 fold-change ratio for the genes in the Havcr1-co-expression gene set from the DrugMatrix data (x-axis) and an external rat kidney ischemic injury data set (GSE58438) (y-axis) at a) 1 day and b) 5 days after ischemic injury. The gene set is highly correlated in both data sets at both 1 and 5 days after ischemic injury, indicating a similar response to both chemically and non-chemically induced kidney injuries

Similar articles

See all similar articles

Cited by 8 articles

See all "Cited by" articles

References

    1. Charlton JR, Portilla D, Okusa MD. A basic science view of acute kidney injury biomarkers. Nephrol Dial Transplant. 2014;29(7):1301–11. doi: 10.1093/ndt/gft510. - DOI - PMC - PubMed
    1. Xie HG, Wang SK, Cao CC, Harpur E. Qualified kidney biomarkers and their potential significance in drug safety evaluation and prediction. Pharmacol Ther. 2013;137(1):100–7. doi: 10.1016/j.pharmthera.2012.09.004. - DOI - PubMed
    1. Molitoris BA. Therapeutic translation in acute kidney injury: the epithelial/endothelial axis. J Clin Invest. 2014;124(6):2355–63. doi: 10.1172/JCI72269. - DOI - PMC - PubMed
    1. Stewart IJ, Tilley MA, Cotant CL, Aden JK, Gisler C, Kwan HK, et al. Association of AKI with adverse outcomes in burned military casualties. Clin J Am Soc Nephrol. 2012;7(2):199–206. doi: 10.2215/CJN.04420511. - DOI - PubMed
    1. Togel F, Westenfelder C. Recent advances in the understanding of acute kidney injury. F1000Prime Rep. 2014;6:83. doi: 10.12703/P6-83. - DOI - PMC - PubMed

Publication types

Feedback