Integrative analysis of cancer-related signaling pathways

Abstract

Identification and classification of cancer types and subtypes is a major issue in current cancer research. Whole genome expression profiling of cancer tissues is often the basis for such subtype classifications of tumors and different signatures for individual cancer types have been described. However, the search for best performing discriminatory gene-expression signatures covering more than one cancer type remains a relevant topic in cancer research as such a signature would help understanding the common changes in signaling networks in these disease types. In this work, we explore the idea of a top down approach for sample stratification based on a module-based network of cancer relevant signaling pathways. For assembly of this network, we consider several of the most established cancer pathways. We evaluate our sample stratification approach using expression data of human breast and ovarian cancer signatures. We show that our approach performs equally well to previously reported methods besides providing the advantage to classify different cancer types. Furthermore, it allows to identify common changes in network module activity of those cancer samples.

Keywords: cancer gene expression; expression signature; microarray analysis; modeling of signaling pathway; sample stratification.

Figure 1

Interconnected signaling network of cancer-related genes and modules. (A) Representation of the cancer relevant signaling network including all modules and contributing genes. (B) Exemplified schematic layout of the module grouping strategy; all genes in the network were associated with functional modules, e.g., genes encoding ligands, RTKs, or adaptor proteins of a given pathway were grouped into the corresponding ligand, RTK, or adaptor modules. In living cells, these modules contribute to activity of the pathways upstream signaling, here represented by a pathway upstream module. Functionally, the pathway upstream module of a given pathway activates or inhibits the activity of distinct downstream modules, here exemplified for activation of a MAPK and a PI3K/AKT module. These modules themselves consist of associated genes, e.g., encoding for ERK, RAS, or RAF isoforms in case of MAPK signaling or PI3K catalytic and regulatory isoforms as well as AKT, PTEN, and others. In our network, these downstream modules are again linked to events acting further downstream in the pathway, like activation of an AP1 or MYC transcription factor module.

Figure 2

Hierarchical clustering of a breast cancer sample set. (A) Heatmap representation of a complete linkage hierarchical bi-clustering of network module activity, based on expression data of 113 breast cancer samples of Lu et al. (2008) data set; the resulting dendrogram is annotated according to the clinical sample annotation using the color code as indicated in the legend; labelings are from top to down: labeling 4, tumor type; labeling 3, tumor grade; labeling 2, HER2 expression status; labeling 1, estrogen receptor (ER) expression status; regions analyzed in more detail in Figures 3A,B are indicated by black vertical bars. (B) Magnification of the dendrogram from (A) showing the cluster of mainly ER-negative, mainly HER2-negative, high grade ductal samples that is separated from all other samples. (C) Cluster dendrogram resulting from bi-clustering based on expression values of all individual genes represented in the network. (D–G) Cluster dendrograms resulting from bi-clustering of the breast cancer samples based on expression values of published discriminative gene signature sets generated specifically for breast cancer samples (D) Wang et al. (2005) (E) van’t Veer et al. (2002) a signature generated more generally to distinguish cancer samples from each other (F) Rhodes et al. (2004) and another breast cancer specific signature from Paik et al. (2004) (G).

Figure 3

Pathway activity analysis of a high grade breast cancer subcluster. (A,B) Regions of the heatmap in Figure 2A that appear to distinguish high grade ER-negative HER2-negative samples from all other samples in the Lu et al. (2008) data set. (C) Plot showing distribution of log10 values of all network module activity values (blue bars) for all breast cancer samples versus log10 gene-expression values (green bars) of all genes associated with the network. (D) Network diagram of the top 15% differentially expressed modules comparing the difference between average network module activity in high grade versus all other samples; node color and node label size are initialized with log2 difference of the average module and gene-expression values of high grade versus all other samples, respectively; a dark red color indicates log2 difference ≥5.04 as a cutoff for the largest 15% fraction of differentially expressed network modules or genes (max. 9.28).

Figure 4

Comparable network module activity differences in high grade breast cancer samples of an independent data set. (A) Dendrogram representation of a complete linkage hierarchical bi-clustering of network module activity-based on expression data of 145 breast cancer samples of the expO data set; the resulting dendrogram is annotated according to the clinical sample annotation using the color code as indicated in the legend; labelings are from top to down: labeling 4, tumor type; labeling 3, tumor grade; labeling 2, HER2 expression status; labeling 1, estrogen receptor (ER) expression status. A cluster of mainly high grade ER and HER2-negative samples analyzed further is indicated by a black vertical bar. (B) Magnification of module activities that appear to distinguish the cluster of high grade ER and HER2-negative samples from all other samples within the expO data set. (C) Plot showing distribution of log10 values of all network module activity values (blue bars) for all expO breast cancer samples versus log10 distribution of gene-expression values (green bars) of all genes associated with the network. (D) Network diagram of the top 15% differentially expressed modules comparing the difference between average network module activity in high grade versus all other samples; node color and node label size are initialized with log2 difference of the average module and gene-expression values of high grade versus all other samples; red color indicates expression above a log2 difference ≥2.964 (max. 7.742). (E,F) Comparison of log2 differences in (E) average PDGFR and (F) average INSR-pathway modules activity as calculated for the Lu (blue bars) and expO (orange bars) data sets. The thresholds for determination of the top 15% differentially activated modules within both data sets are indicated by the horizontal bars. (G) Comparison of log2 differences in average PDGFR module related gene-expression values as calculated for the Lu and expO data sets. (H) Comparison of log2 differences in average INSR module related gene-expression values as calculated for the Lu and expO data sets. For comparison of differences of individual gene expression with the differences in module activities, the thresholds for determination of the top 15% differentially activated modules within both data sets are indicated by the horizontal bars.

Figure 5

Hierarchical clustering of an ovarian cancer sample data set. (A) Heatmap representation of a complete linkage hierarchical bi-clustering of network module activity, based on expression data of 278 ovarian cancer samples derived from the Tothill et al. (2008) data set; the resulting sample distribution dendrogram is annotated according to the clinical sample annotation using the color code as indicated in the legend; labelings are from top to down: labeling 4, consolidated tumor grade; labeling 3, stage code low (I and II) or high (III and IV); labeling 2, cancer type LMP or malignant; labeling 1, primary tumor site ovary, fallopian tube, or peritoneum. Regions analyzed in more detail in Figure 6A are indicated by black vertical bars. (B) Magnification of the dendrogram from (A). (C) Cluster dendrogram resulting from bi-clustering based on expression values of all individual genes represented in the network. (D–G) Cluster dendrograms resulting from bi-clustering of the samples based on expression values of published discriminative gene signature sets generated specifically for ovarian cancer samples (D) Mok et al. (2009), (E) Denkert et al. (2009), as well as the signatures by (F) Rhodes et al. (2004), and (G) Paik et al. (2004).

Figure 6

Pathway activity analysis of the LMP ovarian cancer subcluster. (A) Regions of the heatmap in Figure 5 that appear to distinguish low malignancy potential (LMP) samples from all other ovarian cancer samples. (B) Plot showing distribution of log10 values of all network module activity values (blue bars) for all Tothill et al. (2008) ovarian cancer samples versus log10 distribution of gene-expression values (green bars) of all genes associated with the network. (C) Network diagram of the top 15% differential expressed modules and genes comparing the difference between average network module activity in LMP versus all other samples; node color and node label size are initialized with log2 difference of the average gene-expression values of LMP versus all other samples; red color indicates expression above a log2 difference ≥6.605 (max. 9.768).

Figure 7

Comparable network module activity differences between LMP and all other ovarian cancer samples of an independent data set. (A) Dendrogram representation of a complete linkage hierarchical bi-clustering of network module activity, based on expression data of 83 ovarian cancer samples of the Anglesio et al. (2008) data set; the resulting dendrogram is annotated according to the clinical sample annotation using the color code as indicated in the legend; labelings are from top to down: labeling 4, consolidated tumor grade; labeling 3, stage code low (I and II) or high (III and IV); labeling 2, cancer type LMP or malignant; labeling 1, primary tumor site ovary or metastasis. (B) Depiction of network module activities that appear to distinguish the cluster of LMP samples from all other samples within the Anglesio et al. (2008) data set. (C) Plot showing distribution of log10 values of all network module activity values (blue bars) for all Anglesio et al., ovarian cancer samples versus log10 distribution of gene-expression values (green bars) of all genes associated with the network. (D) Network diagram of the top 15% differentially expressed modules comparing the difference between average network module activity in LMP versus all other samples; node color and node label size are initialized with log2 difference of the average module activity and gene-expression values of LMP versus all other samples; dark red color indicates expression above a log2 difference ≥3.455 (max. 9.631). (E) Comparison of log2 differences in average AP1 and MAPK-Inhibitor module activity as calculated for the Tothill (blue bars) and Anglesio (orange bars) data sets. (F) Comparison of log2 differences in average AP1 module associated gene-expression values as calculated for the Anglesio et al. (2008) and Tothill et al. (2008) data sets. (G) Comparison of log2 differences in average MAPK-Inhibitor module associated gene-expression values as calculated for the Anglesio et al. and Tothill et al. data sets. For comparison of differences of individual gene expression with the differences in module activities, the thresholds for determination of the top 15% differentially activated modules within both data sets are indicated by the horizontal bars.