A comprehensive single-cell expression atlas of human AML leukemia-initiating cells unravels the contribution of HIF pathway and its therapeutic potential

Abstract

Relapse remains a major challenge in the clinical management of acute myeloid leukemia (AML), and is driven by rare therapy-resistant leukemia-initiating stem cells (LSCs) that reside in specific bone marrow niches. Hypoxia signaling keeps cells in a quiescent and metabolically relaxed state, desensitizing them to chemotherapy. This suggests the hypothesis that hypoxia contributes to AML-LSC function and chemoresistance and is a therapeutic target to sensitize AML-LSCs to chemotherapy. Here, we provide a comprehensive single-cell expression atlas (119,000 cells) of AML cells and AML-LSCs in paired diagnostic-relapse samples from risk-stratified patients with AML. The HIF/hypoxia pathway is attenuated in AML-LSCs compared with differentiated AML cells, but is enhanced when compared with healthy hematopoietic cells. Accordingly, chemical inhibition cooperates with standard-of-care chemotherapy to impair leukemogenesis, substantially eliminating AML-LSCs. These findings support the HIF pathway as a stem cell regulator in human AML, and reveal avenues for combinatorial targeted and chemotherapy-based approaches to specifically eliminate AML-LSCs.

Mutational signature decomposition with deep neural networks reveals origins of clock-like processes and hypoxia dependencies

“…In order to use our consensus hypoxia scores we used nine different hypoxic gene expression signatures: Harris, Semenza, Winter, Elvidge, Leonard, Percio, Buffa, Ragnum and Winter2. The first six signatures were obtained from [57] and the other three from [54,55,56], respectively. These hypoxic signatures are defined by sets of hypoxic genes containing between 36 and 172 genes, Supplementary Table 5.…”

“…However, there exist many different scores that provide a set of hypoxia-associated genes, and they contain different genes with very limited overlap. The most well-known hypoxia scores are Buffa, [54], Ragnum [55] and Winter [56], but other sets are also used [57]. Here we used a consensus approach to determine which mutational signatures are associated with hypoxia: Similarly to what is done with mutation callers, we say that a signature is significantly correlated with hypoxia if it is significantly correlated with at least two different hypoxia scores.…”

“…In total, we used nine different hypoxia scores. The three aforementioned, and also Harris, Semenza, Winter (a second one), Elvidge, Leonard and Percio [57]. In order to compute the hypoxia scores from the set of hypoxia-associated genes, we ranked gene expression data in the same way as described in Bhandari et al [24].…”

“…Of note, to quantify hypoxia the authors selected one from a set of gene expression-based hypoxia scoring schemes, described in [54]. However, we found that the multitude of existing hypoxia scores [55,54,56,57] are not very comparable with each other and therefore derived a consensus hypoxia score (Methods, Supplementary Table 5). Further, in addition to cancer type, patient age, patient sex and tumour purity, our regression model also controls for average tumour sequencing coverage (Supplementary Table 6).…”