Network-based identification of microRNAs as potential pharmacogenomic biomarkers for anticancer drugs

Abstract

As the recent development of high-throughput technologies in cancer pharmacogenomics, there is an urgent need to develop new computational approaches for comprehensive identification of new pharmacogenomic biomarkers, such as microRNAs (miRNAs). In this study, a network-based framework, namely the SMiR-NBI model, was developed to prioritize miRNAs as potential biomarkers characterizing treatment responses of anticancer drugs on the basis of a heterogeneous network connecting drugs, miRNAs and genes. A high area under the receiver operating characteristic curve of 0.820 ± 0.013 was yielded during 10-fold cross validation. In addition, high performance was further validated in identifying new anticancer mechanism-of-action for natural products and non-steroidal anti-inflammatory drugs. Finally, the newly predicted miRNAs for tamoxifen and metformin were experimentally validated in MCF-7 and MDA-MB-231 breast cancer cell lines via qRT-PCR assays. High success rates of 60% and 65% were yielded for tamoxifen and metformin, respectively. Specifically, 11 oncomiRNAs (e.g. miR-20a-5p, miR-27a-3p, miR-29a-3p, and miR-146a-5p) from the top 20 predicted miRNAs were experimentally verified as new pharmacogenomic biomarkers for metformin in MCF-7 or MDA-MB-231 cell lines. In summary, the SMiR-NBI model would provide a powerful tool to identify potential pharmacogenomic biomarkers characterized by miRNAs in the emerging field of precision cancer medicine, which is available at http://lmmd.ecust.edu.cn/database/smir-nbi/.

Keywords: breast cancer; metformin; miRNA; network-based inference; pharmacogenomics.

Figure 1. General diagrams of the SMiR-NBI model for miRNA-mediating cancer pharmacogenomic studies

(A) The biological diagram of miRNA pharmacogenomics. MiRNAs can be up-/down-regulated by an anticancer drug, then directly target the downstream genes to mediate its anticancer responses. (B) The workflow of miRNA pharmacogenomics. The predictive Small Molecule-miRNA Network-Based Inference (SMiR-NBI) model was built by network-based inference algorithm, based on the curated heterogeneous network connecting small molecules and miRNAs (SM-miR network). The miRNA-mediating mechanism-of-action (MOA) of anticancer responses was annotated by bioinformatics analyses on the miRNA-gene function network.

Figure 2. Building a SM-miR network connecting small molecules (SM) and miRNAs and network topological analyses

(A) Global diagram of the known SM-miR network containing 2,447 up-/down-regulations between 154 small molecules and 618 miRNAs; (B) The degree distribution for the top 100 small molecules and miRNAs; (C) Module 4-6, with clustered small molecules and their co-regulated miRNAs.

Figure 3. The receiver operating characteristic (ROC) curves of the SMiR-NBI model

The areas under the ROC curves (AUC) were labeled with means and standard errors using 100 times of 10-fold cross validation for predicting potential miRNAs to a given small molecule (magenta line) or predicting potential small molecules to a miRNAs of interest (steel blue line) via our previously developed network-based inference framework [24].

Figure 4. Identification of potential pharmacogenomic biomarkers for breast cancer, natural products and non-steroidal anti-inflammatory drugs (NSAIDs) via the SMiR-NBI model

(A) The heatmap showed the predicted scores (color keys) of the SMiR-NBI model for 183 differentially expressed miRNAs in breast cancer from The Cancer Genome Atlas (see Methods) and 17 anti-breast cancer drugs. (B) The representative miRNA pharmacogenomic pathway for natural products included 13 common miRNAs regulated by at least 3 different natural products. (C) The representative miRNA pharmacogenomic pathway for NSAIDs contained 10 hub genes with the highest degrees in NSAIDs-regulating miRNA-gene subnetwork and 27 miRNAs targeting at least 3 hub genes. Regulations between NSAIDs and miRNAs were denoted by different colors.

Figure 5. Discovery of new miRNAs mediating treatment responses for tamoxifen and metformin in MCF-7 or MDA-MB-231 cell lines via qRT-PCR assays

(A) The qRT-PCR assay for expression change of 6 predicted miRNAs among the top 10 predicted candidates in the MCF-7 cells treated with 100 nM, 500 nM or 1 μM tamoxifen respectively. (B) The qRT-PCR assay for expression change for 11 predicted miRNAs among the top 20 predicted candidates in the MCF-7 cells treated with 1 mM, 5 mM or 10 mM metformin respectively. (C) The qRT-PCR assay for expression change of 12 predicted miRNAs among the top 20 predicted candidates in the MDA-MB-231 cells treated with 1 mM, 5 mM or 10 mM metformin respectively. *p < 0.05, **p < 0.01 and ***p < 0.001 were determined by t-test. Error bars represent standard errors (s.d., n = 3).

Figure 6. The discovered metformin-miRNA-target gene regulatory network

(A) The whole network included 23 up-regulated miRNAs and 16 down-regulated miRNAs with their 619 target genes. (B) and (C) The down-regulation subnetwork (B) and up-regulation subnetwork (C) for metformin identified by bioinformatics analyses. Regulatory details by metformin were represented by different colors of miRNA nodes.