A data-driven network model of primary myelofibrosis: transcriptional and post-transcriptional alterations in CD34+ cells

Abstract

microRNAs (miRNAs) are relevant in the pathogenesis of primary myelofibrosis (PMF) but our understanding is limited to specific target genes and the overall systemic scenario islacking. By both knowledge-based and ab initio approaches for comparative analysis of CD34+ cells of PMF patients and healthy controls, we identified the deregulated pathways involving miRNAs and genes and new transcriptional and post-transcriptional regulatory circuits in PMF cells. These converge in a unique and integrated cellular process, in which the role of specific miRNAs is to wire, co-regulate and allow a fine crosstalk between the involved processes. The PMF pathway includes Akt signaling, linked to Rho GTPases, CDC42, PLD2, PTEN crosstalk with the hypoxia response and Calcium-linked cellular processes connected to cyclic AMP signaling. Nested on the depicted transcriptional scenario, predicted circuits are reported, opening new hypotheses. Links between miRNAs (miR-106a-5p, miR-20b-5p, miR-20a-5p, miR-17-5p, miR-19b-3p and let-7d-5p) and key transcription factors (MYCN, ATF, CEBPA, REL, IRF and FOXJ2) and their common target genes tantalizingly suggest new path to approach the disease. The study provides a global overview of transcriptional and post-transcriptional deregulations in PMF, and, unifying consolidated and predicted data, could be helpful to identify new combinatorial therapeutic strategy. Interactive PMF network model: http://compgen.bio.unipd.it/pmf-net/.

Figure 1

Experimental design and analysis procedures flow chart. miRNA and gene expression data in CD34+ cells of PMF patients and healthy controls were considered; CTR BM and CTR PB controls were considered separately in parallel analyses, and results were merged at the final steps. MAGIA integrated analysis outputted the subset of predicted miRNA-target relations supported by expression data; these were used to enrich KEGG pathway-derived miRNA-gene networks based on pathways annotations and on validated miRNA-target relations. Topological pathway analyses by Micrographite identified most modulated paths, showing significant gene/miRNA expression variations and changes in relations strength, then a non-redundant comprehensive meta-pathway was derived; iterative analysis of the meta-pathway identified the most modulated network for each comparison. Magia² analysis identified mixed TF-miRNA-gene circuits. Results of parallel comparisons were merged in an integrated network.

Figure 2

PMF network model. (a) PMF network integrating pathway-derived miRNA-gene networks deregulated in PMF and mixed TF-miRNA-gene circuits discovered by reverse engineering of expression data. The network (see Supplementary Figure 3 for a larger version) gives a non-redundant and comprehensive picture of most modulated paths in the two PMF vs CTR comparisons, of the impact of miRNAs on pathway genes, and of connected TF-miRNA-gene mixed circuits discovered in the study. Genes are reported as round rectangles, transcriptional factors as diamonds and miRNAs as triangles. Node colors represent the fold-change (FC) of the gene expressions in the PMF vs CTR BM (node inner color) and PMF vs CTR PB (node border color). The type of edges depends on the type of interaction: arrow for activation, T arrow in case of inhibition, and arrow line for miRNA-target predicted/supported interactions. The light blue shade indicates the part of the network resulting from miRNA and gene topological pathway analysis. The yellow shade indicates mixed TF-miRNA-gene circuits, inferred by Magia² analysis, connected to the path-derived network. (b) Cluster analysis of expression profiles of miRNAs and genes included in the final PMF network do not show clustering of PMF patients by mutation. Samples are colored according to the carried mutation as shown in the legend (3N indicates triple negative). Sample clustering was obtained according to Euclidean distance and complete clustering. See Supplementary Figure 4 for the corresponding heatmap. (c) Summary of most deregulated pathways represented in the miRNA-gene network, and connections thereof. See http://compgen.bio.unipd.it/pmf-net/ for an interactive, searchable and zoomable version of the PMF network model.

Figure 3

Details of the PMF network model showing deregulated miRNAs and genes that participates to specific connected pathways, linked in turn to biological processes and functions germane to the disease. (a) Akt signaling; (b) Rho GTPases, CDC42, PLD2 and PTEN; (c) HIF-1a pathway; (d) Calcium signaling.