Characterization of drug-induced transcriptional modules: towards drug repositioning and functional understanding

Abstract

In pharmacology, it is crucial to understand the complex biological responses that drugs elicit in the human organism and how well they can be inferred from model organisms. We therefore identified a large set of drug-induced transcriptional modules from genome-wide microarray data of drug-treated human cell lines and rat liver, and first characterized their conservation. Over 70% of these modules were common for multiple cell lines and 15% were conserved between the human in vitro and the rat in vivo system. We then illustrate the utility of conserved and cell-type-specific drug-induced modules by predicting and experimentally validating (i) gene functions, e.g., 10 novel regulators of cellular cholesterol homeostasis and (ii) new mechanisms of action for existing drugs, thereby providing a starting point for drug repositioning, e.g., novel cell cycle inhibitors and new modulators of α-adrenergic receptor, peroxisome proliferator-activated receptor and estrogen receptor. Taken together, the identified modules reveal the conservation of transcriptional responses towards drugs across cell types and organisms, and improve our understanding of both the molecular basis of drug action and human biology.

Figure 1

Overview of the method. Workflow to identify and characterize drug-induced transcriptional modules across four microarray data sets from three human cancer cell lines and rat liver. Drug and gene sets of conserved drug-induced transcriptional modules (CODIMs) were characterized in detail using several annotation resources. These reliable CODIMs allow us to propose new MOA for marketed drugs and novel biological roles for poorly characterized genes which were validated experimentally.

Figure 2

Conservation of drug-induced transcriptional modules. (A) The number and proportion of transcriptional modules identified separately in each human cell line and rat liver that are codetected in multiple cell lines and/or organisms (gene overlap of modules were deemed significant with Fisher’s exact test, FDR-corrected P-value <0.01). Twenty-three CODIMs were defined from the connected components of the module network (using reciprocal best-hits only). (B) Functional characterization of conserved drug-induced modules. Connected modules (with data set-specific labels), enriched biological process (in yellow set, 61% of CODIM) and characteristic compound MOA (in red set, 44% of CODIM, in italics) of selected CODIMs are shown (BP, biological process; HDAC, histone deacetylase; WD40 repeat, β-transducin repeat). Graph inlets: conservation of modules across cell lines and species as measured by overlapping gene and drugs (Supplementary Figure 3).

Figure 3

Novel functional roles of uncharacterized genes as functional regulators of cellular cholesterol levels. (A) Heatmap of drug-induced gene expression changes (fold change) of CODIM2. Black boxes indicate gene membership of cell-line-specific modules. Hierarchical clustering of genes from CODIM2 shows two major components (labeled in blue and green) and three smaller groups (dark green, brown, light blue). (B) Projection onto the STRING protein–protein network reveals a densely interconnected component (enriched in blue set) corresponding to fatty acid and (chole-)sterol metabolism (Supplementary Table 5 and Supplementary Figure 7). The second component (enriched in green set) is enriched in ER-related processes, such as protein N-linked glycosylation, transport and unfolded protein response. Genes subjected to functional knockdown assays of LDL uptake and free cellular cholesterol levels are highlighted in red or orange depending on whether validated as a functional regulator or not, respectively (two or more siRNAs having consistent effect with absolute z-score >1 and FDR-corrected P-value <0.01, Supplementary Table 6). (C, D) Representative image of fixed cells taken by automated widefield fluorescence microscopy. Cells have been transfected for 48 h with siRNA targeting indicated gene, and experiments of (C) LDL-uptake assay with fluorescently labeled DiI–LDL and (D) staining with cholesterol binding dye filipin was performed. Total intensity of fluorescent signal in segmented spots/area has been quantified as indicated with arrows.

Figure 4

Prediction of novel cell cycle inhibitors based on CODIM1. (A) Heatmap: expression fold change in each cell line with shaded columns, indicating chemical treatments not present in the respective cell-line-specific module. Grey: compounds with a known effect on cell cycle (Supplementary Table 5). Although, six well-known cell cycle blockers were detected in all cell lines delineating a module core in CODIM1, 26 additional cell cycle blockers were only detected as part of this module in one or two of the three cell lines (shaded columns). This variability highlights the benefits of aggregating information across cell lines. (B) Three chemicals not previously known to inhibit cell cycle were examined in cell viability and proliferation assays. For sulconazole and vinburnine dose-dependent reduction of cell viability and proliferation were confirmed in two cell lines with IC50 values as indicated. For mephentermine, an effect on cell viability and proliferation could not be detected in either cell line (treat., treatment). (C) Control and drug-treated HL60 cells were attributed to different cell cycle phases according to their DNA content (PI, FACS analysis) for three treatment durations (24, 48 and 96 h). The error bars represent the s.e.m. Sulconazole (25 μM) treatment led to a marked increase in sub-G1 population across all time points (24, 48 and 96 h), which is indicative of an apoptotic cell population. Vinburnine (25 μM) induced a G2/M arrest within 24 h similar to Nocodazole (Supplementary Figure 9) and further treatment (48 and 96 h) resulted in increased apoptosis in HL60 cells.

Figure 5

Drug repositioning from drug-induced transcriptional modules. Three examples of cell-line-specific drug-induced modules that were enriched for pharmacological classes of (A) peroxisome proliferator-activated receptor activators (PC3-9), (B) α-adrenergic 2 agonists (HL60-17), (C) estrogen receptor-α modulators (MCF7-9; see Supplementary Tables 4 and 5 for a complete list of modules). For each module (as detected by the biclustering approach), a heatmap (A–C) was drawn to illustrate gene expression changes (fold change) under various drug treatments. Drugs in modules were characterized with respect to their molecular targets. Specific action on the main target associated with the respective module A–C is indicated by colored boxes. Novel drug–target relationships were inferred for drug members not previously known to modulate the main targets associated with these modules. Ten predicted modulators were experimentally tested (red labeled drugs in A–C). (D–G) Four of these predictions could by verified with in vitro binding assays (bold face). In three cases K_i values lower than 15 μM confirmed strong binding, whereas ERα affinity of nitrendipine (46 μM) was considered ambiguous (Lounkine et al, 2012; Supplementary Figure 10).