Deciphering the transcriptional regulation of microRNA genes in humans with ACTLocater

Abstract

Understanding the transcriptional regulation of microRNAs (miRNAs) is extremely important for determining the specific roles they play in signaling cascades. However, precise identification of transcription factor binding sites (TFBSs) orchestrating the expressions of miRNAs remains a challenge. By combining accessible chromatin sequences of 12 cell types released by the ENCODE Project, we found that a significant fraction (~80%) of such integrated sequences, evolutionary conserved and in regions upstream of human miRNA genes that are independently transcribed, were preserved across cell types. Accordingly, we developed a computational method, Accessible and Conserved TFBSs Locater (ACTLocater), incorporating this chromatin feature and evolutionary conservation to identify the TFBSs associated with human miRNA genes. ACTLocater achieved high positive predictive values, as revealed by the experimental validation of FOXA1 predictions and by the comparison of its predictions of some other transcription factors (TFs) to empirical ChIP-seq data. Most notably, ACTLocater was widely applicable as indicated by the successful prediction of TF → miRNA interactions in cell types whose chromatin accessibility profiles were not incorporated. By applying ACTLocater to TFs with characterized binding specificities, we compiled a novel repository of putative TF → miRNA interactions and displayed it in ACTViewer, providing a promising foundation for future investigations to elucidate the regulatory mechanisms of miRNA transcription in humans.

Figure 1.

Development of the ACTLocater method. (A) Summary of conserved accessible sequences upstream of 101 HMR intergenic miRNA units. The 101 regions upstream of miRNA units were arranged side by side. Within each region, conserved accessible sequences of 12 cell types were aligned according to their genomic coordinates. (B) Results of the leave-one-out cross-validation performed in 12 cell types. The horizontal axis shows the cell type left out. The vertical axis shows the percentage of conserved accessible sequences recovered by the remaining cell types. The average is marked by the dashed line. (C) Flowchart for the ACTLocater method. PSSMs—position specific scoring matrices. (D) The screenshot shows genome browser tracks for E-boxes (in human genomic sequence), conserved E-boxes (found in human, mouse and rat), verified E-boxes [two E-boxes verified as negative result and positive result in Ma et al. (24), respectively], accessible regions (accessible chromatin regions merged from 12 cell types) and placental mammal basewise conservation profile (35) for the 4-kb flanking region upstream of Hsa-miR-10b.

Figure 2.

Prediction and validation of FOXA1-targeted HMR intergenic miRNAs. (A) A boxplot shows the signal-to-noise ratios of FOXA1 predictions by the Conservation method and ACTLocater method, respectively. ***P < 0.001. (B) Examples of ChIP assays in MCF-7 cells. Primers against predicted regions were used to amplify DNA immunoprecipitated with FOXA1 antibody (lane 3) or control rabbit IgG (lane 2). Genomic DNA was used as a positive control (lane 1). Site P2 validated by previous studies (54–56) and site NC with no evidence of FOXA1 binding (54) were used as a positive and a negative control for ChIP enrichment, respectively. (C) Real-time RT–PCR and (D) western blotting analysis of FOXA1 from MDA-MB231 cells stably expressing empty vector or FOXA1. Error bars indicate s.e.m.; *P < 0.05. (E) Venn diagram of FOXA1-targeted HMR intergenic miRNA units identified by ACTLocater or previous genome-wide ChIP studies. miRNA units found only in the genome-wide ChIP studies were classified according to the chromatin accessibility and conservation of corresponding ChIP peak regions. (F) Fold changes of randomly selected and negative control miRNAs in FOXA1 overexpressing MDA-MB231 cells relative to the empty vector controls. miRNAs with multiple locus in the human genome are marked underlines. Hsa-miR-147a and hsa-miR-150-3p, which there were no FOXA1-binding sites associated with, were used as negative controls. Error bars indicate s.e.m.; two-sided one-sample Student’s t-test; *P < 0.05; **P < 0.01; ***P < 0.001. (G) Classification of predicted FOXA1-binding sites. Sites overlapped with ChIP peaks of H3K4me1, H3K4me3 or Pol II are marked as solid circles, otherwise are marked as unfilled circles. The plot shows the distances (absolute value) between sites and their corresponding TSSs. The TSS of the first miRNA unit is not available. **P < 0.01.

Figure 3.

Applicability assessment of ACTLocater. (A) Summary of MyoD-binding sites predicted with chromatin accessibility data of skeletal muscle and non-muscle cells, respectively. Sites are aligned according to the genomic coordinates. (B) Venn diagram of evidence supported MyoD sites. (C) Summary of Oct4-binding sites predicted with chromatin accessibility data of embryonic stem and non-stem cells, respectively. Sites are aligned according to the genomic coordinates. (D) Venn diagram of evidence supported Oct4 sites.

Figure 4.

ACTViewer: a database documented the ACTLocater predictions on available PSSMs. (A) A snapshot of the ACTViewer genome browser. The controls (at the top of the browser) position the browser over a specific region in the genome. Annotations are displayed as individual tracks along the genomic regions. (B) A pop-up window containing detailed information about a predicted HNF4A-binding site. (C, D and E) Snapshots of searching ACTViewer for target miRNAs of NF-κB (C), TFBSs associated with hsa-miR-101-1 (D) and c-Myc→hsa-miR-17 interaction (E). Search results for target miRNAs of NF-κB and TFBSs associated with hsa-miR-101-1 are partially shown.