Functional Module Connectivity Map (FMCM): a framework for searching repurposed drug compounds for systems treatment of cancer and an application to colorectal adenocarcinoma

Abstract

Drug repurposing has become an increasingly attractive approach to drug development owing to the ever-growing cost of new drug discovery and frequent withdrawal of successful drugs caused by side effect issues. Here, we devised Functional Module Connectivity Map (FMCM) for the discovery of repurposed drug compounds for systems treatment of complex diseases, and applied it to colorectal adenocarcinoma. FMCM used multiple functional gene modules to query the Connectivity Map (CMap). The functional modules were built around hub genes identified, through a gene selection by trend-of-disease-progression (GSToP) procedure, from condition-specific gene-gene interaction networks constructed from sets of cohort gene expression microarrays. The candidate drug compounds were restricted to drugs exhibiting predicted minimal intracellular harmful side effects. We tested FMCM against the common practice of selecting drugs using a genomic signature represented by a single set of individual genes to query CMap (IGCM), and found FMCM to have higher robustness, accuracy, specificity, and reproducibility in identifying known anti-cancer agents. Among the 46 drug candidates selected by FMCM for colorectal adenocarcinoma treatment, 65% had literature support for association with anti-cancer activities, and 60% of the drugs predicted to have harmful effects on cancer had been reported to be associated with carcinogens/immune suppressors. Compounds were formed from the selected drug candidates where in each compound the component drugs collectively were beneficial to all the functional modules while no single component drug was harmful to any of the modules. In cell viability tests, we identified four candidate drugs: GW-8510, etacrynic acid, ginkgolide A, and 6-azathymine, as having high inhibitory activities against cancer cells. Through microarray experiments we confirmed the novel functional links predicted for three candidate drugs: phenoxybenzamine (broad effects), GW-8510 (cell cycle), and imipenem (immune system). We believe FMCM can be usefully applied to repurposed drug discovery for systems treatment of other types of cancer and other complex diseases.

Figure 1. Flowchart of methodology.

Figure 2. Function-function networks for colorectal adenoma.

Condition specific function-function networks (FFNs) were generated from gene-gene networks (GGINs), shown in Figure S2, by reduction. Nodes in an FFN are functional modules (FMs), which are gene sets in the corresponding GGIN forming over-represented Gene Ontology terms. FMs containing less than 70 genes are not shown. The diameter of a node scales with the logarithm of the number of genes in the node. The color shade of a node indicates the number of intra-node gene-gene interactions per gene. The thickness of the edge indicates the number of inter-node gene-gene interactions.

Figure 3. Enrichment score versus fold-change for CMap drugs.

Enrichment score (ES) was obtained by querying the CMap with gene set (size indicated by vertical bar) determined using varying fold-change (FC) threshold. A drug is considered beneficial for the treatment for colorectal adenoma if ES <−0.5, harmful if ES >0.3, and neutral otherwise. (A) Screening by IGCM procedure. Querying gene set was complete set of differentially expressed genes (DEGs) identified from gene expression arrays of colorectal adenoma cohort (versus control) using the SAM algorithm with fixed FDR <0.01. (B) Screening by FMCM procedure. Querying gene sets were functional modules obtained by partition of over-represented Gene Ontology terms in GSToP filtered DEGs.

Figure 4. Function-drug association map (FDAM) for colorectal adenoma.

Nodes in the map are functional modules (FMs; gene sets) and drugs obtained by querying CMap using individual FMs. Drug-function links indicate beneficial (green) or harmful (red). Only drugs beneficial to at least one FM are included.

Figure 5. Accuracy and reproducibility in drug prediction.

(A) Accuracy is the sum of true positive (predicted beneficial and known anti-tumor agent) and true negative (predicted harmful and known cancer-inducing agent) over sum of predicted beneficial and harmful drugs. IGCM results are in black, and FMCM, in red and cyan. Specificity is given in Figure S5. (B) Reproducibility is the measure of agreement between the selected drugs in two runs using different subsets of microarray data (Materials and Methods). Results shown are averaged over 45 pair-wise comparisons of selected drugs. The five towers on the left are IGCM results for given threshold FC value. The eight towers on the right are FMCM results (FC >2) for the 8 functional modules. Size of querying gene set is given by line in red.

Figure 6. Enrichment scores of known anti-cancer drugs.

(A) irinotecan, (B) thapsigargin, (C) 8-azaguanine, and (D) vorinostat. CMap querying gene sets are shown on the horizontal axis. The first five entries from left are whole DEG sets selected by SAM using FDR = 0.01 and FC ranging from 3.0 to 5.0. The rest are the eight functional module selected by GSToP with FC = 2.0. Star indicates permutation p-value<0.005.

Figure 7. Viability test of colon and breast cancer cells treated with single drug.

Tests were conducted on eight drugs: phenoxybenzamine (PB), GW-8510, phthalylsulfathiazole (PS), etacrynic acid (EA), ginkgolide A (GA), triflusal (TF), imipenem (IM), and 6-azathymine (6-AT), with concentrations of 0, 0.1, 1, 10, and 30 µM. (A) Viability of MCF7 on treatment of the eight drugs. (B) Viability of five cell lines on treatment of GW-8510. Colon cancer cells HCT116, RKO, SW403 and SW620, and the breast cancer cell MCF7, were treated with single drug for 5 days. After 5 days, proliferation activities of these cells were detected by Alamar Blue assay.

Figure 8. Clustering of genomic profiles of drug-treated cancer cell lines HCT116 and MCF7.

(A) Individual gene approach (IGA). (B) Gene-set approach (GSA). Cell lines were treated with three drugs: GW-8510, phenoxybenzamine (PB), and imipenem. Entries marked “cmap” were microarray drug treatment genomic profiles of MCF7 taken from the CMap. Others were from drug treatment microarray experiments (Affymetrix U219 (PrimeView) platform) conducted for the present study, where the same experimental protocol used in CMap were followed: averaged over three dosages of 10 M, 11.8 M, 13.4 M; treatment time 6 hours after overnight culture. Heatmaps were results of two-way hierarchical clustering.

Figure 9. Overlap of candidate repurposed drug sets curated from colon adenoma and colorectal cancer data sets.

Numbers in brackets correspond to those given in the first column of Table 1, which lists the drug set for colon adenoma. The overlap includes 9 of the 13 drugs in Table 1 with degrees not less than 3, and 5 of the 8 drugs selected for cell viability tests marked by “*”.