Characterisation of novel molecular subtypes and corresponding pre-clinical models in AML, with implications for treatment response.

Abstract

Acutemyeloid leukemia (AML) is an aggressive cancer of the blood and bone marrow.Due to AML's molecular heterogeneity, there is a wide variation of treatment responseand survival rates among patients. This presents the need tobetter characterise the disease and identify novel molecular subtypes. There-use of published and validated prognostic predictive gene signatures presentsan invaluable in silico opportunity to uncover the biological mechanismsunderpinning treatment response in AML. 

Analysis was carried out using a primarydataset (AML-OHSU) and a validation dataset (TCGA-LAML). Both datasets had beenprocessed using Almac’s claraT platform. An automated analytical pipeline wasdeveloped to stratify patients via 167 uniquegene expression signatures categorised by 10 different Hallmarks of Cancer.Clusters identified as stable were subject to further downstream analysis toidentify clinical associations and trends in signature scores linked todifferences in survival. A drug sensitivity prediction pipeline was alsodeveloped and trained using IC50 and basal gene expression data available from NCI-60 and GDSC cell line databases. IC50 valueswere predicted in four independent geneexpression datasets. PredictedIC50s were evaluated for associations betweenpatient survival and response. 

The automated clustering pipeline analysed atotal of 1,245 stable clusters in the AML-OHSU dataset. Stable clusters weresubsequently processed via log rank analysis to identify significantdifferences (Log Rank p-values < 0.05) in survival outcome. A significantdifference in overall survival probability (Log Rank p-value: 0.035) was foundbetween Energetics clusters where K = 3. Energetics clusters with significantsurvival differences were validated in the TCGA dataset (Log rank p-value:0.019). The drug sensitivity pipeline found significant associations betweenpredicted IC50s and response in valproic acid (Cox PH p-value: 0.0194) and Fedratinib (Dunn’s testp-value: 0.02662) datasets when using celllines representative of the Energetics-associated cluster 3.