Selection of the network components. (A) Non-transformed rat ovarian surface epithelial (ROSE) cells and their KRAS-transformed derivatives (RAS-ROSE) were subjected to gene expression analysis by interrogating Affymetrix microarrays and customized RAS target arrays (). Overall, 51 transcription factors were found to be upregulated in KRAS-transformed cells, of which 7 were chosen for further analysis (see text for details). (B) Expression of selected transcription factors in normal ROSE cells and their KRAS-transformed derivatives was investigated by RT–PCR and western blot. Pathway-specific expression regulation was assessed by treatment with the Mek1/2 inhibitor U0126 and the PI3K inhibitor LY 294002. Loading was controlled using Gapdh primers for RT–PCR and Histone H3 antibody for western blot. Source data is available for this figure in the .

Effects of transcription factor silencing on cell morphology and the transcriptional profile of RAS-ROSE cells. (A) Morphological characteristics of normal ROSE versus RAS-ROSE cells 48 h after treatment with scrambled siRNA-duplex (Sc), transfection reagents only (Mock) and two independent specific siRNAs targeting Fosl1, Hmga2, Klf6, JunB, Otx1, Gfi1 and RelA, respectively. Phase contrast, magnification: 450-fold. (B) Quantification of lateral processes of RAS-ROSE and transfectants, in which transcription factor expression was silenced by siRNA. (C) Effects of transcription factor silencing on the transcriptome of RAS-ROSE cells. Each transcription factor controlled expression of hundreds of regulated genes as indicated by hybridization of target RNA to probe sets on the microarray (cutoff for significant upregulation or downregulation: ±0.7 on log2 scale). Total number of probe sets indicating expression alterations as a response to transcription factor silencing: 5971 (3247 and 3054 upregulated and downregulated by knockdown, respectively; few probe sets indicated inconsistent target regulation among knockdowns); average number of probe sets indicating response to silencing of an individual transcription factor: 3300; number of probe sets indicating response to scrambled (sc) siRNA: 319. (D) Specificity of transcription factor silencing as determined by profiling of target genes. Transcription factor knockdowns induced broadly overlapping gene expression responses. Each probe set indicating expression alterations was classified according to the exact number of transcription factor knockdowns (1–7) affecting gene expression. Cutoff used for significant deregulation as in C. In total, expression levels of genes represented by 4832 (81%) probe sets were altered by two or more transcription factor knockdowns, while genes represented by only 1139 probe sets (19%) indicated specifically responded to one of the knockdowns. Source data is available for this figure in the .

Perturbation experiments. (A) Summary of the perturbation-response data of the signalling and gene-regulatory networks as determined by microarray, RT–PCR and western blot analysis. Rows in graph correspond to mRNAs and proteins analysed, columns correspond to experimental perturbations (overexpression of mutant RAS, inhibition of signalling kinases by small molecules and siRNA-mediated silencing of indicated transcription factors, respectively). Each square represents the log2-fold change for the perturbation-response combination (see Materials and methods). Blue squares denote responses indicating downregulation, red squares denote responses indicating upregulation (see ). (B, C). Relationship between transcription factors and target gene expression as determined by expression changes upon transcription factor knockdown. Upper heat maps of graphs B and C illustrate target gene expression patterns over seven knockdown conditions: Specifically, only probe sets related to target genes are shown whose expression patterns matched perturbation-response patterns of transcription factor protein and mRNA, and may thus represent direct targets or co-regulated genes, respectively. For comparison, lower heat maps of graphs B and C show patterns of transcription factor protein and mRNA expression, respectively (the area width of each transcription factor pattern in the heat map was adjusted to the number of matching probe sets and designations of transcription factors were placed underneath the brackets at the bottom of the lower panels). A probe set (gene expression pattern) was defined as ‘matching the transcription factor pattern', if the log-fold transcription factor mRNA or protein change cutoffs, defined as large (<1), small (0.3–1), or absent (<0.3), respectively, were accompanied by comparable expression changes of the target gene over all knockdown conditions (correlations and anti-correlations were considered; see for details). Out of 5971 probe sets indicating an mRNA expression change mediated by transcription factor knockdown, 947 matched the transcription factor protein pattern and 532 probe sets the mRNA patterns, respectively. Note that we included moderate differences between transcription factor and target gene patterns to accommodate high versus low affinity transcription factor targets and variability in microarray measurements. Source data is available for this figure in the .

Network model of hierarchical and regulatory interactions among the KRAS-dependent transcriptional network components and signalling pathways. (A) Result of modular response analysis (MRA) applied for mathematical calculation of hierarchical and regulatory interactions (cf. ). Each dot indicates an inferred regulatory interaction expressed as the local response coefficient (see legend on bottom of A). The rows and columns indicate the regulator and effector species, respectively. Some interactions were excluded a priori in the derivation of the model (panels 2, 6 and 8); others (indicated by boxes) were confirmed by work published previously. (B) Correlation between measured global response coefficients () and simulation results of the best-fit model (A). Off-diagonal lines indicate standard deviation of response coefficients. (C) Schematic representation of the network model depicted in A. Arrows point at a direct activating interaction between network components. The markings + or − placed at the top of a protein symbol represent activation or inhibition of transcription factors by cytoplasmic signalling; the markings + or − placed at the bottom of a protein symbol imply stimulation or inhibition of signalling activity (kinase phosphorylation) by downstream transcription factors.

Simulated and experimental double perturbations unveil incoherent regulation of Fosl1 protein by Otx1, and complex regulation of Hmga2 expression. (A) Strategy for assessing the post-transcriptional regulation of Fosl1 in RAS-ROSE cells: Ectopic Fosl1 overexpression eliminates endogenous transcriptional regulation (indicated by red X). (B) Inhibition of phospho-Erk strongly reduces Fosl protein level as shown by treatment of Fosl1 overexpressing cells with the Mek inhibitor U0126. Bars show log-fold change compared with control (mock transfection). (C) Model with refined post-transcriptional regulation. The MRA model selection procedure was applied after removing the negative influence of phosho-Erk on Fosl1 protein (indicated by red X; cf. B), and allowing other transcription factor proteins to regulate Fosl1 post-transcriptionally. The refined model includes post-transcriptional regulation by Gfi1, Otx1, Klf6 and Hmga2, while all other regulatory interactions of the original model in are retained. (D) The refined model reproduces incoherent regulation of Fosl1 mRNA (blue) and protein (red) after Otx1 knockdown. A set of independent gene silencing experiments was performed in RAS-ROSE cells to confirm incoherent regulation of Fosl1 mRNA and protein as described in . Bars show log-fold changes of Fosl1 mRNA and protein after Otx1 silencing compared with transfection of scrambled siRNA as control. (E) Post-transcriptional control of Fosl1 by Otx1 as revealed by double perturbation. Downregulation of Fosl1 protein by Otx1 knockdown was predicted and experimentally verified, even if Fosl1 was expressed ectopically. Overexpression was simulated by setting all local response coefficients for Fosl1 mRNA regulation to zero. Bars show log2-fold change following Otx1 knockdown relative to Fosl1 overexpression (OE) and transfection of scrambled siRNA. (F) Fosl1 is an upstream regulator of Hmga2. The model predicts a strong downregulation of Hmga2 mRNA after Mek inhibition, a modest increase after Fosl1 overexpression, and a modest decrease after combined Mek inhibition and Fosl1 overexpression. This pattern was verified in experiments based on Fosl1 overexpression and treatment with the Mek inhibitor U0126. Bars show average log2-fold changes of two independent experiments relative to DMSO-treated or mock-transfected controls. Source data is available for this figure in the .

Effects of transcription factor silencing on transformed phenotypes expressed in KRAS-transformed cells support hierarchical regulation in the gene-regulatory network. (A) Anchorage-independent proliferation (‘3-dimensional' (3D) growth in soft agar cultures). (B) Anchorage-dependent proliferation (‘2-dimensional' (2D) growth in adherent monolayer cultures) of RAS-ROSE cells 48 h after treatment with scrambled siRNA-duplex (Sc), transfection reagents only (Mock) and two independent siRNA duplexes targeting each transcription factor as determined by colorimetric XTT assay. Each bar indicates the alteration of cell growth relative to untransfected RAS-ROSE cells. Upstream transcription factors: Otx1, Gfi1, RelA and downstream transcription factors: Fosl1, Hmga2, Klf6 and JunB (see text). (C) Cell-cycle analysis of RAS-ROSE cells 48 h after treatment with scrambled siRNA-duplex (Sc), transfection reagents only (Mock) and two independent siRNA duplexes targeting each transcription factor. Distribution of cell-cycle phases G0/G1, S and G2/M was determined by FACS analysis after DNA staining of cells with propidium iodide (). Each bar shows the number of cells in G0/G1 phase relative to untransfected RAS-ROSE cells. (D) Representative images of scratch assays to determine migratory activity of RAS-ROSE cells following transcription factor silencing. Cells were transfected with scrambled siRNA, transfection reagents only (mock), and two independent siRNA duplexes targeting RelA. Pictures were taken before scratching the wound, at time 0 and 19, 24 and 36 h post scratch. Phase contrast, magnification 100-fold. Migration was strongly inhibited after RelA knockdown as compared with controls. (E) Quantitative analysis of wound closure by RAS-ROSE cells following transcription factor knockdown as compared with control cells. Bar plots show average and standard deviation of two biological and three technical replicates for two different transfected siRNAs normalized to scrambled control. Source data is available for this figure in the .