Genome-Wide Mapping and Subtype Classification of Long Non-Coding RNA in Acute Myeloid Leukemia

Abstract

Background: Most efforts to characterize acute myeloid leukemia (AML) have focused so far on genetic and epigenetic aberrations, which can ultimately lead to altered protein-coding gene function. The roles of long non-coding RNAs (lncRNAs), which orchestrate cell physiology and act as key regulators of the AML oncogenic state, remain uncharacterized globally.Material and Methods: We performed a genomic analysis of GENCODE lncRNAs in 179 clinically annotated cases of de novo AML, using The Cancer Genome Atlas (TCGA) RNA-Seq data. In addition, we described global correlations between lncRNAs and the expression of cis- and trans-acting genes. We also established lncRNA-based subtype classification based on distinct signatures and then correlated that classification with fusion transcripts and genetic alterations.Results: Using stringent criteria (RPKM ≥1 in at least 10% of AML), we identified 2,913 expressed lncRNAs and used an integrative analysis to predict those that are potential regulators of AML oncogenic state. The expression of 1,935 (66.4%) lncRNAs showed positive correlations with the mRNA expression of their neighboring genes, while only 14 (0.4%) of the lncRNAs showed negative correlations. Gene ontology analysis using GREAT revealed enrichment of cis-neighboring genes in the PML body gene set (p=8.2x10-7). Unsupervised clustering of lncRNA-based expression showed five robust molecular clusters (C1 to C5), which were highly correlated with the mRNA-based classification. Of those, three clusters (C1, C2 and C5) were tightly associated with recurrent fusion transcripts; cluster C1 (n=16) was composed exclusively of promyelocytic leukemias, while cluster C5 (n=45) was enriched for MLL-rearranged cases (24.4%), and cluster C2 (n=31) was enriched for MYH11-CBFB or RUNX1-RUNX1T1 (55.2%) rearranged cases. Importantly, cluster C4 (n=30), which includes cytogenetically normal leukemias, was highly enriched for NPM1 (p=2.3x10-11) and FLT3 (p=1.6x10-4) mutations; conversely, cluster C3 (n=53) was highly enriched for recurrent copy-number alterations as well as RUNX1 (p=0.001) and TP53 somatic mutations (p=0.004). We further discovered a core of 37 lncRNAs significantly associated with a MLL-signature and 16 lncRNAs with a NPM1-mutated signature.Conclusion: This study describes the first genome-wide mapping and characterization of lncRNAs in AML and proposes a robust lncRNA-based classification. This classification may serve in defining core lncRNAs that orchestrate key oncogenic states in the different clinical subtypes. DisclosuresNo relevant conflicts of interest to declare.