Immunophenotypic-Defined Stage of Leukemia Differentiation Arrest Identifies Oncogenic and Metabolic Signatures in AML

Upon differentiation, hematopoietic stem cells (HSC) give rise to multipotent progenitors (MPP) that retain the ability to produce all blood lineages but have lost their self-renewal capacity. MPP are then orientated towards either the lymphoid or myeloid lineages, developing into common myeloid progenitors (CMP) or lymphoid-primed multipotent progenitors which can still produce certain myeloid cell types. CMP can differentiate into either granulocyte-macrophage progenitors (GMP) or megakaryocyte-erythroid progenitors. GMP finally differentiate into granulocyte or monocyte progenitors (GP/MP). This study aimed to investigate whether acute myeloid leukemia (AML) samples can be sub-classified based on the stage of arrest in differentiation (Stage of Leukemia Arrest, SLA). Understanding which critical differentiation program is specifically altered could be of special interest to design new therapeutic strategies aimed at re-inducing the differentiation process in AML.

Based on CD34/CD117/CD13/CD33/MPO expression, we defined phenotypic signatures that identify all normal hematopoietic stem and progenitor cells (HSPC). Applied to leukemic hematopoiesis, these signatures allowed us to sub-classify 932 AML patients from Toulouse University Hospital (test cohort) and 142 AML patients from The Cancer Genome Atlas Network, published in NEJM in 2013 (control cohort). Globally, AML with an HSC phenotype (henceforth termed HSC-L) represented 1% of the cases, while MPP-L, CMP-L, GMP-L and GP/MP-L accounted for 15%, 30%, 22% and 30% of the cases, respectively.

To validate our phenotypic signature and SLA, we generated a transcriptomic signature of normal hematopoietic differentiation from three databases including normal HSPC assessment (GSE21973, GSE42414 and E-TABM-978). Unsupervised clustering and gene set enrichment analysis (GSEA) of a test cohort (n=40) and a control cohort (n=142) confirmed that AML shared a common transcriptomic signature with their SLA. To strengthen the relationship between SLA and HSPC, we then compared their energetic metabolism. Using metformin as an enhancer of glycolytic (GLY) metabolism (e.g. Pasteur effect), we constructed transcriptomic signatures for AML cell lines with either a GLY or with an oxidative (OX) metabolism.  We showed that, as expected normal HSC and MPP were enriched in GLY genes, whereas CMP and GMP were enriched in OX genes set. Similarly, HSC-L and MPP-L highly expressed genes related to GLY metabolism; CMP-L samples could be subdivided between GLY AML and OX AML; and GMP-L highly expressed genes related to OX metabolism.

Each AML subgroup defined accordingly to the SLA had a particular clinical presentation (Table). Interestingly, the least differentiated AML (HSC-L, MPP-L and CMP-L) were enriched in HSC and LSC genes sets. These AML samples blocked at early stage of hematopoiesis gave higher engraftment levels in NSG mice compared to GMP-L and GP/MP-L (21%, vs 5%, panel of 59 AML tested in 376 mice, p<0.0001). Finally, HSC-L/MPP-L/CMP-L were an independent factor of worse prognostic in the two cohorts of patients, for OS and EFS (cox regression model, p=0.009 HR=1.44 and p=0.007 HR=1.42, respectively).

We then sought to investigate the distribution of AML mutations according to SLA classification. We took advantage of the fully annotated TCGA database and found that 80% of HSC-L and 25% of MPP-L had RUNX1 mutations, 80% of GMP-L had either CEBPA mutation or RUNXL1-RUNX1/CBFB-MYH11 AML, and 80% of GP/MP-L had NPM1 mutation. The subgroups of CMP-L showed a greater molecular heterogeneity.

AML and normal HSPC shared phenotypic, transcriptomic and metabolic signatures. This is of special interested for differentiation therapy as specific critical dysregulated transcriptional factors (e.g. CEBPA) could be related to specific stage of arrest. Identifying each of them could allow to propose targeted differentiating therapy and maximize therapeutic responses.

	HSC-L	MPP-L	CMP-L	GMP-L	GP/MP-L
frequency	1%	15%	30%	22%	30%
Median WBC at diagnosis (G/L)	4	25	20	35	54
organ infiltration	NA	spleen	varied	lymph nodes	gengiva or lymph nodes
CFU-L (% of blasts)	0.7	8.1	9.6	8.7	6.9
metabolism	GLY	GLY	GLY/OX	OX	NA
oncogenic event	RUNX1m	RUNX1m/ TP53m	unknown	CEBPAm/CBF	NPM1c
BAALC/ERG/MN1 expression	high	high	mid	mid	low
stem cell signature	high	high	high	low	low
Median DFS (months)	5	22.9	13.7	41.3	>120
Median OS (months)	6.6	26.6	16.3	>120	83.3