Network Analysis Reveals the Molecular Bases of Statin Pleiotropy That Vary with Genetic Background

Abstract

Many approved drugs are pleiotropic: for example, statins, whose main cholesterol-lowering activity is complemented by anticancer and prodiabetogenic mechanisms involving poorly characterized genetic interaction networks. We investigated these using the Saccharomyces cerevisiae genetic model, where most genetic interactions known are limited to the statin-sensitive S288C genetic background. We therefore broadened our approach by investigating gene interactions to include two statin-resistant genetic backgrounds: UWOPS87-2421 and Y55. Networks were functionally focused by selection of HMG1 and BTS1 mevalonate pathway genes for detection of genetic interactions. Networks, multilayered by genetic background, were analyzed for key genes using network centrality (degree, betweenness, and closeness), pathway enrichment, functional community modules, and Gene Ontology. Specifically, we found modification genes related to dysregulated endocytosis and autophagic cell death. To translate results to human cells, human orthologues were searched for other drug targets, thus identifying candidates for synergistic anticancer bioactivity. IMPORTANCE Atorvastatin is a highly successful drug prescribed to lower cholesterol and prevent cardiovascular disease in millions of people. Though much of its effect comes from inhibiting a key enzyme in the cholesterol biosynthetic pathway, genes in this pathway interact with genes in other pathways, resulting in 15% of patients suffering painful muscular side effects and 50% having inadequate responses. Such multigenic complexity may be unraveled using gene networks assembled from overlapping pairs of genes that complement each other. We used the unique power of yeast genetics to construct genome-wide networks specific to atorvastatin bioactivity in three genetic backgrounds to represent the genetic variation and varying response to atorvastatin in human individuals. We then used algorithms to identify key genes and their associated FDA-approved drugs in the networks, which resulted in the distinction of drugs that may synergistically enhance the known anticancer activity of atorvastatin.

Keywords: chemical genetics; epistasis; network analysis; pleiotropy; statins; synthetic genetic array; synthetic lethality; yeast.

FIG 1

Statins inhibit the synthesis of HMGCR and downstream products in the mevalonate pathway. Statins are competitive inhibitors of HMGCR encoded by HMG1 and HMG2 in yeast and the HMGCR gene in humans. A critical step in the mevalonate pathway is mediated by the enzyme geranylgeranyl diphosphate synthase (encoded by BTS1 in yeast and GGPPS1 in humans), where the main ergosterol/cholesterol-synthesis pathway branches off to synthesize other fundamental cellular components for isoprenylation of small GTPases. Genes in blue are yeast genes, and genes in gray are their human orthologues. Red asterisks in yeast genes indicate oxygen-dependent steps of the pathway. Human genes in orange at the end of the cholesterol pathway are less conserved with yeast and do not correspond to the yeast gene to the left.

FIG 2

Flow diagram for the methods used to identify interactions and pathways behind atorvastatin pleiotropy in three genetic backgrounds. Single deletion mutant array libraries (A) in the S288C genetic background (depicted in blue outline) and in the recently created UWOPS87 and Y55 genetic backgrounds (depicted in yellow and purple) through a backcross methodology (18) were used to generate genome-wide double deletion mutants (deletion mutant genes depicted as empty circles) as models to investigate the atorvastatin pleiotropy in three genetic backgrounds (B). About 25,800 double deletion mutants in 1,536-colony format (384 quadruplicate colonies per agar plate) were created, treated with atorvastatin, and screened to identify fitness defects that would reveal epistatic interactions as measured by decreased colony size. Atorvastatin-hypersensitive double mutants were then validated in serial dilution spot assays and used as input to create genetic (GIN) and protein-protein (PPIN) interaction networks (C). GINs and PPINs were multilayered in one network (D) per genetic background and subjected to network topology analyses. The network centrality metrics pinpointed bottleneck and hub genes of high biological relevance. The communities of genes identified through network modularity (E) were analyzed through a KEGG enrichment analysis to distinguish key metabolic pathways. Human orthologues of the key yeast genes were used in a search for drug enrichment (F) to identify potential combination therapies to enhance the anticancer activity of atorvastatin.

FIG 3

Atorvastatin sensitivity confers similar synthetic sickness/lethality in strains with HMG1 deleted and varies in strains with BTS1 deleted in three genetic backgrounds. (A) Haploid cells deficient of HMG1 or BTS1 and their wild types in three genetic backgrounds were pinned on increasing concentrations of atorvastatin in serial dilution and incubated for 2 days at 30°C. (B) Violin plot distributions of average fitness of 12,900 strains as measured by colony sizes (n = 4) of the xxxΔ and hmg1Δ xxxΔ mutants as well as (C) xxxΔ and bts1Δ xxxΔ mutants, where positive scores represent increased fitness and negative scores represent decreased fitness. The red dashed lines indicate the score cutoff values selected for validation in independent assays for double deletions that did not overlap the xxxΔ single deletions. Venn diagrams visualize the overlap in the number of genes below the cutoff lines. Statistical differences were evaluated by Student’s t test (*, P < 0.05; **, P < 0.01; ***, P < 0.001).

FIG 4

Four hmg1Δ xxxΔ double deletion mutants were hypersensitive to atorvastatin treatment in three genetic backgrounds, while others depend on genetic background. Haploid cells derived from SGA analyses and gene deletion libraries were pinned on SC with or without supplementation of atorvastatin in serial dilution and incubated for 2 days at 30°C. Shown here are deletions of genes that enhanced sensitivity to atorvastatin treatment in the (A) S288C, (B) UWOPS87, and (C) Y55 genetic backgrounds. The WT/hmg1Δ panel refers to either the nonmutated wild types (WT) for the xxxΔ strain panels or the hmg1Δ single deletion mutants for the hmg1Δ xxxΔ double deletion strain panels. Gene deletions in boldface indicate interactions overlapping in three genetic backgrounds.

FIG 5

Eight bts1Δ xxxΔ double deletion mutants were hypersensitive to atorvastatin treatment in at least two genetic backgrounds, while others depend on the genetic background. Haploid cells derived from SGA analyses and gene deletion libraries were pinned on SC with or without supplementation of atorvastatin in serial dilution and incubated for 2 days at 30°C. Shown here are deletions of genes that enhanced sensitivity to atorvastatin treatment in the (A) S288C, (B) UWOPS87, and (C) Y55 genetic backgrounds. WT/bts1Δ refers to either the nonmutated wild type (WT) for the xxxΔ strain panels or the bts1Δ single deletion for the bts1Δ xxxΔ double deletion strain panels. Gene deletions in boldface indicate interactions overlapping in three genetic backgrounds; asterisks indicate interactions overlapping in two genetic backgrounds.

FIG 6

Multilayer networks derived from atorvastatin-sensitive hmg1Δ xxxΔ interactions. GINs (layer 1), PPINs (layer 2), and the edges between them were integrated in a multilayer network using TimeNexus. Edges between layers connect overlapping nodes in the two layers, and the genes linking these edges are shown in the periphery of circular networks. Darker nodes in multilayer networks are validated atorvastatin-sensitive interactions.

FIG 7

Multilayer networks derived from atorvastatin-sensitive bts1Δ xxxΔ interactions. GINs (layer 1), PPINs (layer 2), and the edges between them were integrated in a multilayer network using TimeNexus. Edges between layers connect overlapping nodes in the two layers, and the genes linking these edges are shown in the periphery of circular networks. Darker nodes in multilayer networks are validated atorvastatin-sensitive interactions.

FIG 8

Network topology centrality analyses of multilayer networks identify key HMG1/BTS1 interactors for atorvastatin sensitivity. Centrality measurements (degree, closeness, and betweenness) were calculated for each gene and visualized in a 3D plot. UBI4 was excluded because, due to its highly interactive nature, it skewed all the other nodes to one corner of the plot, obscuring the relevance of other genes. High-centrality genes RIM15, CDC28, TLG2, and BRE5 are enclosed in orange rectangles.

FIG 9

Metabolic pathway enrichment of modules in multilayer networks for atorvastatin sensitivity. Bubble plots show enrichment for each of the modules (named for their genetic background) identified through community analysis for HMG1 (top panel) and BTS1 (bottom panel) interactions. The size of the bubbles is relative to the enrichment score for each pathway, while the intensity of the colors is relative to the adjusted P value. The x-axis labels show the genetic background followed by the number of modules. Numbers missing in the sequence are modules without significantly enriched pathways.

FIG 10

Atorvastatin treatment in the UWOPS87 genetic background increases the survival integral of double mutants. Cells were grown in triplicate with and without atorvastatin. Cultures were left growing at 30°C for 2 weeks, and growth was measured every second day for a 2-week period via hourly measurements of optical density. YODA was used to calculate the surviving cell percentage. Data are shown as the mean ± standard deviation (SD) (n = 3). *, P ≤ 0.05, **, P ≤ 0.01, and ***, P ≤ 0.001, by Student’s t test relative to the vehicle control.

FIG 11

Human orthologues of yeast interactions reveal drugs/compounds to test for synergy with atorvastatin. Human orthologues of validated genes and bottleneck genes were processed via an enrichment analysis for signature genes in the Drug Signature Database. Bubble plots represent the human orthologues (y axis) that were enriched for drugs/compounds (x axis). The color of each bubble is determined by the adjusted P value (AdjP), and the size of bubble reflects a score computed by running the Fisher exact test for random gene sets to determine the deviation from the expected rank, where bigger bubbles represent greater enrichment.

FIG 12

Proposed integration of mechanisms. Atorvastatin inhibits components of the actin cytoskeleton, which in turn inhibits actin-mediated endocytosis and induces UPR. Atorvastatin inhibits aging pathways, which also results in the dual induction of UPR and autophagy. Hence, atorvastatin is an indirect inhibitor of endocytosis and indirect activator of UPR and autophagy. Red blunt-headed arrows point to pathways inhibited by atorvastatin. Blue arrows and the blue blunt-headed arrow point to pathways that are inhibited or induced, respectively. Dashed pink arrows and the dashed blunt-headed arrow point to inhibition or induction, respectively, of pathways via indirect mechanisms of atorvastatin.