Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2011 Jul;7(7):2118-27.
doi: 10.1039/c1mb05014a. Epub 2011 May 19.

Relating Protein Adduction to Gene Expression Changes: A Systems Approach

Affiliations
Free PMC article

Relating Protein Adduction to Gene Expression Changes: A Systems Approach

Bing Zhang et al. Mol Biosyst. .
Free PMC article

Abstract

Modification of proteins by reactive electrophiles such as the 4-hydroxy-2-nonenal (HNE) plays a critical role in oxidant-associated human diseases. However, little is known about protein adduction and the mechanism by which protein damage elicits adaptive effects and toxicity. We developed a systems approach for relating protein adduction to gene expression changes through the integration of protein adduction, gene expression, protein-DNA interaction, and protein-protein interaction data. Using a random walk strategy, we expanded a list of responsive transcription factors inferred from gene expression studies to upstream signaling networks, which in turn allowed overlaying protein adduction data on the network for the prediction of stress sensors and their associated regulatory mechanisms. We demonstrated the general applicability of transcription factor-based signaling network inference using 103 known pathways. Applying our workflow on gene expression and protein adduction data from HNE-treatment not only rediscovered known mechanisms of electrophile stress but also generated novel hypotheses regarding protein damage sensors. Although developed for analyzing protein adduction data, the framework can be easily adapted for phosphoproteomics and other types of protein modification data.

Figures

Figure 1
Figure 1
Overview of the systems approach for the integrative analysis of gene expression and protein adduction data. A) Mapping data to network. Protein-protein interaction (PPI) and protein-DNA interaction data are modeled in an integrative network. mRNAs and proteins corresponding to the same gene are modeled separately in the PPI layer and the mRNA layer, with transcription factors (TFs) connecting these two layers. Transcription factors, non-transcription factor proteins, and mRNAs are represented as square, round, and triangle nodes, respectively. Gene expression data and protein adduction data are mapped to the network. In the mRNA layer, up-regulated genes, down-regulated genes, and genes with no significant change are colored in red, green, and yellow, respectively. In the PPI layer, adducted proteins are highlighted with blue circles around the nodes. B) Transcription factor inference. Over-representation analysis is used to identify transcription factors that are responsive to the treatment. Responsive transcription factors are colored in red, while non-responsive transcription factors are colored in yellow in the PPI layer. C) Walking the interactome. Random walk with restart is used to score all proteins in the PPI network for their network proximity to the responsive transcription factors. Based on the scores, each node in the PPI layer is colored with a gradient from red to yellow (high score to low score). D) Significance evaluation. A global null score distribution for all nodes and a local null score distribution for each node are estimated by scores generated from randomly created transcription factor sets (Rdm_1 through Rdm_n) and used to evaluate the significance of the real scores for each node. E) Subnetworks that constitute of significant proteins are defined as responsive signaling networks. Overlaying protein adduction data on the inferred signaling networks allows the detection of candidate stress sensors and associated regulatory mechanisms.
Figure 2
Figure 2
Evaluation of the random walk approach for signaling network inference using known pathways. A) Using a restart probability of 0.5, transcription factors in known pathways were used to predict signaling proteins in the same pathway based on loose (L) and stringent (S) cutoff levels. Hypergeometric test was used to evaluate the quality of prediction for each pathway. Empirical cumulative distribution functions (cdfs) of the hypergeometric p-values demonstrate high level of overlap between random-walk based predictions and corresponding pathways (solid curves). In comparison, much lower level of overlap was observed between random predictions and real pathways (dotted curves) as well as between real predictions and randomly selected signaling proteins (dashed curves). The black line indicates hypergeometric p-value of 0.01. B) The proportion of significant predictions (p < 0.01) for a wide range of restart probabilities (0.1-0.9).
Figure 3
Figure 3
Inferred HNE-responsive signaling network. The interaction network of the 199 proteins identified based on the loose cutoff level, which includes 887 interactions. The 29 proteins identified based on the stringent cutoff level are shown in larger node size. Transcription factors, HNE targets, and other proteins are labeled with blue, red, and black colors. Cell cycle proteins, RNA splicing proteins, proteins involved in both processes, and proteins involved in neither process are colored in pink, cyan, green, and purple, respectively.
Figure 4
Figure 4
DNA damage response network. The network map was curated in the WikiPathways database. Cyan nodes represent proteins in the inferred HNE-responsive signaling network. Red and green nodes represent proteins that are up or down-regulated at mRNA level in response to HNE treatment.
Figure 5
Figure 5
Possible mechanisms for protein adduction mediated transcriptional regulation. A) CCNA2 adduction-mediated repression of genes related to DNA replication, cell cycle, and RNA splicing. B) RUVBL2 adduction-mediated activation of genes related to regulation of transcription and homeostatic process. mRNAs, transcription factors, and signaling proteins are represented by triangle, square, and round nodes, respectively. Node color represents the log fold change at the mRNA level based on the microarray gene expression data, as indicated in the color scale bar. Different types of interactions are represented by different edge styles, as indicated in the figure legend.

Similar articles

See all similar articles

Cited by 17 articles

See all "Cited by" articles

Publication types

MeSH terms

LinkOut - more resources

Feedback