Abstract
Acute leukemia of ambiguous lineage (ALAL) is a rare, poor-prognosis acute leukemia subtype that cannot be assigned to a single hematopoietic lineage. Although ALAL patients are typically treated with acute myeloid leukemia (AML) or acute lymphoblastic leukemia (ALL) regimens, optimal treatment choice is hindered by their lineage ambiguity. Therefore, we investigated the added value of transcriptomics for improving lineage assignment, currently based mainly on surface markers. First, we used an in-house pipeline to detect genetic lesions in RNA sequencing data (n = 30) with a sensitivity > 90% for small variants. Second, we compared ALAL gene expression profiles (GEPs) with representative AML (n = 145), B-ALL (n = 223), and T-ALL (n = 85) cases. In a principal component analysis (PCA), ALALs did not form a clear separate group, as most clustered with AML, B-ALL, or T-ALL. Accordingly, a machine learning classifier trained with GEPs of acute leukemias segregated 27/30 ALALs into myeloid-, B-, or T-lymphoid. These 27 cases harbored genetic abnormalities consistent with the classifier-assigned leukemia. Furthermore, deconvolution of ALAL GEPs revealed enrichment for signatures of normal hematopoietic cells corresponding to the leukemic type predicted by our algorithm. The classifier was also applied on an external ALAL cohort (n = 24), assigning 75% of the patients to a lineage matching their immunophenotypic and methylation profiles. In conclusion, integrative analysis of RNA sequencing data can accurately classify most ALAL cases as lineage-defined, while others show true transcriptional and epigenetic ambiguity driven by lesions like BCL11B. The pipeline and classifier developed here are valuable tools to improve ALAL diagnosis and guide therapeutic decisions.