Integrated analysis of regulatory and metabolic networks reveals novel regulatory mechanisms in Saccharomyces cerevisiae

Abstract

We describe the use of model-driven analysis of multiple data types relevant to transcriptional regulation of metabolism to discover novel regulatory mechanisms in Saccharomyces cerevisiae. We have reconstructed the nutrient-controlled transcriptional regulatory network controlling metabolism in S. cerevisiae consisting of 55 transcription factors regulating 750 metabolic genes, based on information in the primary literature. This reconstructed regulatory network coupled with an existing genome-scale metabolic network model allows in silico prediction of growth phenotypes of regulatory gene deletions as well as gene expression profiles. We compared model predictions of gene expression changes in response to genetic and environmental perturbations to experimental data to identify potential novel targets for transcription factors. We then identified regulatory cascades connecting transcription factors to the potential targets through a systematic model expansion strategy using published genome-wide chromatin immunoprecipitation and binding-site-motif data sets. Finally, we show the ability of an integrated metabolic and regulatory network model to predict growth phenotypes of transcription factor knockout strains. These studies illustrate the potential of model-driven data integration to systematically discover novel components and interactions in regulatory and metabolic networks in eukaryotic cells.

Figure 1.

A schematic drawing of the approach used in this work for model-based analysis of growth phenotyping and gene expression data to identify new network components and interactions. The approach combines in silico modeling of genome-scale metabolic and regulatory networks with analysis of in vivo data obtained by gene expression and growth phenotyping experiments. Specific mispredictions of either gene expression changes or growth phenotypes are identified and used as inputs for systematic model expansion. The primary data types used for model expansion are ChIP-chip data on protein–DNA interactions and the presence of known TF-binding motifs on promoters. The result of the expansion is a model that includes new regulatory interactions that allow improved prediction of expression changes and growth phenotypes of knockout and overexpression strains.

Figure 2.

Comparison between expression changes in TF knockout and overexpression strains predicted by iMH805/775, in vivo observed expression changes, and promoter occupancy for the corresponding TFs derived from ChIP-chip and motif data. (A) Overlaps between the three data sets shown in the form of Venn diagrams. The numbers refer to the number of genes in each category out of the total of 750 metabolic genes. (B) Interpretation of each of the segments in the Venn diagrams shown in A. (C) Scatterplots of the gene expression changes in TF knockout/overexpression strains (log₂ ratios between knockout strain and wild type) and the corresponding promoter occupancy scores derived from ChIP-chip data (−log₁₀ of the P-value reported in Harbison et al. [2004). The genes that were predicted to change in expression by the iMH805/775 model are indicated by blue circles. Genes with significant gene expression change or promoter occupancy are colored using a color scheme similar to the one used in the Venn diagrams in A.

Figure 3.

Results for the three regulatory network expansion scenarios (A,B,C) using a combination of the iMH805/775 network and a provisional regulatory network derived from ChIP-chip and TF-binding motif data. Each pie chart indicates the fractions of potential target genes that can be reached from the TF through regulatory cascades containing one to five steps as well as the targets that cannot be reached in five or less steps from the TF. For each strain, the numbers of potential novel targets are indicated in parenthesis. (A) Expansion using ChIP-chip and motif data for the 55 TFs in iMH805/775 assuming that each TF can only act as a repressor or activator depending on its known type of activity. (B) Expansion using the same ChIP-chip and motif data, but allowing each TF to act either as repressor or activator. (C) Expansion using ChIP-chip and motif data for all 203 TFs studied in Harbison et al. (2004) allowing each TF to act either as repressor or activator.

Figure 4.

(A) Measured (upper corner) and predicted (lower corner) maximum specific growth rates (1/h) for transcription factor deletion strains on different carbon sources. The in silico predictions were made using the iMH805/837 model. The color scheme indicates the magnitude of the growth rate from low (dark) to high (light). The squares with numbers in bold/italics indicate strain-condition combinations with significant mispredictions discussed in the Supplemental material. (B) Comparison of experimentally measured and predicted growth rates after the improvements to the model discussed in the Supplemental material were done.