The metabolic reprogramming in acute myeloid leukemia patients depends on their genotype and is a prognostic marker

Abstract

Leukemic cells display some alterations in metabolic pathways, which play a role in leukemogenesis and in patients' prognosis. To evaluate the characteristics and the impact of this metabolic reprogramming, we explore the bone marrow samples from 54 de novo acute myeloid leukemia (AML) patients, using an untargeted metabolomics approach based on proton high-resolution magic angle spinning-nuclear magnetic resonance. The spectra obtained were subjected to multivariate statistical analysis to find specific metabolome alterations and biomarkers correlated to clinical features. We found that patients display a large diversity of metabolic profiles, according to the different AML cytologic subtypes and molecular statuses. The link between metabolism and molecular status was particularly strong for the oncometabolite 2-hydroxyglutarate (2-HG), whose intracellular production is directly linked to the presence of isocitrate dehydrogenase mutations. Moreover, patients' prognosis was strongly impacted by several metabolites, such as 2-HG that appeared as a good prognostic biomarker in our cohort. Conversely, deregulations in phospholipid metabolism had a negative impact on prognosis through 2 main metabolites (phosphocholine and phosphoethanolamine), which could be potential aggressiveness biomarkers. Finally, we highlighted an overexpression of glutathione and alanine in chemoresistant patients. Overall, our results demonstrate that different metabolic pathways could be activated in leukemic cells according to their phenotype and maturation levels. This confirms that metabolic reprogramming strongly influences prognosis of patients and underscores a particular role of certain metabolites and associated pathways in AML prognosis, suggesting common mechanisms developed by leukemic cells to maintain their aggressiveness even after well-conducted induction chemotherapy.

Graphical abstract

Figure 1.

Specific metabolic profile of AML patients according to cytologic subtypes of leukemic progenitors (FAB classification). (A) Score plot of the OPLS-DA model built with HRMAS NMR data of all patients (n = 54) and the different AML subtypes. Based on the model factors (R2Y = 0,456; Q2 = 0,306), we see a tendency to separate the patients’ profiles according to their cytologic subtypes. Some patients appear to have common features. (B) Score plot of the same OPLS-DA model as panel A, excluding the IDH mutated patients (n = 41). The model is very robust and predictive (R2Y = 0,845; Q2 = 0,601). The distribution of the metabolic profiles is influenced by the oncometabolite 2-HG.

Figure 2.

Specific metabolic profile of AML patients according to the mutational status of NPM1. (A) Score plot of the OPLS-DA model built with HRMAS-NMR data of NPM1 wild-type patients (n = 30) and NPM1-mutated patients (n = 17). The model is robust and predictive (R2Y = 0,726; Q2 = 0,498). We see a clear separation between the 2 groups. (B) Levels of Glu, PC, PCr/Cr, and tau in NPM1 wild-type patients compared with NPM1-mutated patients. Mean value ± standard error of the mean (SEM). Student t test, *P < .05, **P < .01.

Figure 3.

Specific metabolic profile of AML patients according to the mutational status of FLT3-ITD. (A) Score plot of the OPLS-DA model comparing FLT3 wild-type patients (n = 32) vs FLT3-ITD patients (n = 13). The model is very robust and predictive (R2Y = 0,913; Q2 = 0,713). The separation between FLT3-ITD patients and FLT3 wild-type patients is clear. (B) Levels of Gln, PCr/Cr, scyllo-I, Ala, Cho, PE, and τ in FLT3 wild-type patients compared with FLT3-ITD patients. Mean value ± SEM. Student t test, *P < .05, ***P < .001.

Figure 4.

Specific metabolic profile of AML patients according to the mutational status of IDH. (A) Score plot of the OPLS-DA model comparing IDH wild-type patients (n = 36) vs IDH-mutated patients (n = 13). The model is robust and predictive (R2Y = 0,896; Q2 = 0,822), showing a clear separation between IDH-mutated and IDH wild-type patients. (B) 1D loading plot of IDH wild-type and IDH mutated patients, showing the main differences between the 2 groups. (C) Levels of Ala, Glu, GSH, PC, PE, 2-HG, and scyllo-I in IDH wild-type patients compared with IDH-mutated patients. Mean value ± SEM. Student t test, *P < .05, **P < .01, ***P < .001.

Figure 5.

Changes in metabolites levels according to the immunologic status of patients. (A) Levels of Asp and 2-HG metabolites according to the CD34/MPO status in M0, M1, and M2 subtypes, corresponding to immature AML. (B) Levels of Gln and Pro according to the CD14/CD300e status in M4 and M5 subtypes, corresponding to more mature AML. Mean value ± SEM. Two-way analysis of variance test, ***P < .001.

Figure 6.

Specific metabolic profile of AML patients according to their prognosis (ELN classification). (A) Score plot of the OPLS-DA model comparing adverse (n = 18) vs favorable (n = 21) prognostic groups (R2Y = 0.751; Q2 = 0.472). (B) Levels of GSH and Asp in favorable group compared with adverse group. Mean value ± SEM. Student t test, *P < .05, **P < .01. (C) The prediction of the patients with intermediate risk in the OPLS-DA model built with adverse and favorable subgroups shows a distribution of the half of these patients in favorable risk (n = 8) and the other half in adverse risk (n = 7).

Figure 7.

Specific metabolic profile of AML patients according to their response to chemotherapy induction. (A) Score plot of the OPLS-DA model comparing chemosensitive patients (n = 31) vs chemoresistant patients (n = 7). The model is very robust and predictive (R2Y = 0.865; Q2 = 0.641), showing well-separated groups. (B) 1D loading plot of chemosensitive (CS) and chemoresistant (CR) patients. (C) Levels of Gln, scyllo-I, Ala, GSH, and τ in chemosensitive patients compared with chemoresistant patients. Mean value ± SEM. Student t test, *P < .05, **P < .01.