A transcriptomic continuum of differentiation arrest identifies myeloid interface acute leukemias with poor prognosis

Jonathan Bond,Vahid Asnafi,Christophe Roumier,Guillaume Hypolite,Sylvie Castaigne,Francine Garnache-Ottou,Nicolas Boissel,Mohamed Belhocine,Anne Roggy,Alexander Abdulkader Kheirallah,Elizabeth Macintyre,Claude Preudhomme,Elisa Laurenti,Ludovic Lhermitte,Patrick Villarese,Aleksandra Krzywon,Hervé Dombret

doi:10.1038/s41375-020-0965-z

Abstract

Classification of acute lymphoblastic and myeloid leukemias (ALL and AML) remains heavily based on phenotypic resemblance to normal hematopoietic precursors. This framework can provide diagnostic challenges for immunophenotypically heterogeneous immature leukemias, and ignores recent advances in understanding of developmental multipotency of diverse normal hematopoietic progenitor populations that are identified by transcriptional signatures. We performed transcriptional analyses of a large series of acute myeloid and lymphoid leukemias and detected significant overlap in gene expression between cases in different diagnostic categories. Bioinformatic classification of leukemias along a continuum of hematopoietic differentiation identified leukemias at the myeloid/T-lymphoid interface, which shared gene expression programs with a series of multi or oligopotent hematopoietic progenitor populations, including the most immature CD34+CD1a−CD7− subset of early thymic precursors. Within these interface acute leukemias (IALs), transcriptional resemblance to early lymphoid progenitor populations and biphenotypic leukemias was more evident in cases originally diagnosed as AML, rather than T-ALL. Further prognostic analyses revealed that expression of IAL transcriptional programs significantly correlated with poor outcome in independent AML patient cohorts. Our results suggest that traditional binary approaches to acute leukemia categorization are reductive, and that identification of IALs could allow better treatment allocation and evaluation of therapeutic options.

Highlights

Successful management of acute leukemia is underpinned by accurate diagnostic classification, which provides a basis for treatment allocation, risk stratification and implementation of targeted therapies [1]
Unsupervised hierarchical clustering (HC) analysis of the expression data revealed that T-acute lymphoblastic leukemia (T-ALL) and acute myeloid leukemia (AML) samples largely formed two distinct groups (HC cluster 1 and HC cluster 2, Fig. 1a)
In keeping with modern concepts of a hematopoietic progenitor framework that comprises a spectrum of differentiation potential, integrated transcriptional analysis of AMLs and T-ALLs revealed a continuum of leukemic developmental arrest

Summary

Introduction

Successful management of acute leukemia is underpinned by accurate diagnostic classification, which provides a basis for treatment allocation, risk stratification and implementation of targeted therapies [1]. Knowledge of the molecular landscape of leukemia has increased enormously over the past decades, contemporary classification remains heavily predicated on simple immunophenotypic resemblance to either myeloid or lymphoid normal hematopoietic precursors [2]. While this system has historically been successful, some leukemia categories provide specific diagnostic and therapeutic challenges. There is little consensus on the best treatment approaches for these patients, and prognosis is usually poor [3,4,5] This framework poses difficulties for some cases of T-acute lymphoblastic leukemia (T-ALL) and acute myeloid leukemia (AML). Immature T-ALLs are frequently chemoresistant and require intensive treatment [10, 14, 16], while M0-AML cases have poor outcomes compared to other AML subgroups [17, 18], so it is clinically important to consider whether improved classification of these cases might allow better therapeutic choices

Methods

Results

Conclusion