Abstract
Acute Myeloid Leukemia (AML) is a hematological malignancy with heterogeneous genetics and clinical course. Studies have shown that there is a strong correlation between xenograft in vivo engraftment potential and clinical outcome. Furthermore, such studies have conducted gene expression profiling to identify gene signatures associated with engraftment potential, leukemic stem cell (LSC) property, and prognosis. Most of these analyses were limited to protein coding genes. More recently, advances in next generation sequencing (NGS) have created a paradigm shift on our perception of transcription that, contrary to previous consensus, the human genome is largely transcribed and most of these transcripts do not code for proteins (ncRNA). In the current work, we conducted a comprehensive functional genomics study to identify long ncRNA (lncRNA) that significantly predict in vivo engraftment potential and LSC properties of AML and evaluate their functional and prognostic relevance.
We first sorted normal granulocyte-macrophage progenitor (GMP), lymphoid-primed multi-potential progenitor (LMPP) and CD34-subpopulations from leukemic bulk of 15 AML patient samples. Sorted samples were then transplanted into the non-obese diabetic/severe combined immunodeficient Gamma (NSG) mouse model. Only the subpopulations showing robust leukemic engraftment (> 1%) were classified as LSC populations. In parallel, we performed total RNA based RNA-seq using Ovation® Single Cell RNA-Seq System (NuGEN Technologies Inc.) and evaluated both protein coding genes and lncRNAs.
In line with previous studies, we observed that the engraftment potential primarily resides within the CD34+ subpopulations (GMPs and LMPPs). LMPPs/GMPs from 8 AML patients engrafted (defined as LSC populations), while subpopulations from the other 7 patients failed to engraft (non-LSC fraction). Using our RNA-seq data, we first conducted a global principal component analysis (PCA) based between group analysis (BGA) based on lncRNA expression levels. Interestingly, we noted that the LMPPs and GMPs primarily cluster based on their engraftment potential rather than cell type. In two subsequent independent analyses, we employed weighted correlation network analysis (WGCNA) and top scoring pairs (tsp) to identify lncRNAs that show significant correlation to engraftment percentages and could also categorically classify samples based on their engraftment status. Results of the WGCNA analysis showed that 830 lncRNAs are statistically significantly correlated to engraftment percentages of subpopulations (p < 0.05). In the tsp analysis, by taking all lncRNAs in the dataset in a pairwise fashion, we identified two lncRNAs that could statistically significantly classify samples into 'LSC' (engrafters) and 'non-LSCs' (non-engrafters) (mean classification error ~ 0.19 using bootstrap analysis; statistically significant with p-value < 0.0001 using Monte Carlo Simulation). Of interest here, the two lncRNAs were also in the list of significantly correlated lncRNAs based on the separately conducted WGCNA. Combining the two analyses and using a 3d plot, we looked at the expression levels of two lncRNAs and engraftment percentages of the subpopulations and were able to see that the GMP/LMPP LSCs formed a distinct cluster. The non-LSCs with GMP or LMPP phenotype were distantly clustered from their engrafting counterparts while all the CD34-negative non-LSCs clustered with no overlap to the other groups. Furthermore, a guilt-by-association analysis was carried out to identify potential functional link of the candidate lncRNAs by assessing their correlation to functionally annotated genes. 'JAK-STAT', 'Hematopoietic lineage', and 'Toll-like receptor signaling' were some of the significant pathways (p < 0.05; FDR < 0.25). We also tested some published AML LSC and engraftment associated signatures. The lncRNAs showed negative correlation to the Eppert et al. (2011; Nat. Med.) LSC signature and set of lineage affiliated genes (Goardon et al. 2011; Cancer Cell) including FLT3, NOTCH1, and RUNX1. Taken together, our approach shows that the two lncRNAs we identified sufficiently recapitulate the underlying engraftment potential of AMLs and predict LSC property with significant accuracy. We conclude that these findings highlight the necessity to focus on lncRNAs as key players in clinical and functional studies of AML.
Disclosures
Buske: Celltrion, Inc.: Consultancy, Honoraria. Mulaw:NuGEN: Honoraria.
Predicting the Functions of Long Noncoding RNAs Using RNA-Seq Based on Bayesian Network
