Elevated levels of mitochondrial/OXPHOS genes correlate with poor prognosis of AML patients

Overall survival (OS) or the time from randomization to death is considered the gold standard endpoint in oncology clinical trials. To in silico test OXPHOS genes as candidate biomarkers for AML patients, we downloaded the list of human genes encoding proteins involved in OXPHOS (n = 200) from GSEA MSig database and exploited publically available transcriptional profiling datasets of AML patients (n = 734) with 10 years of follow-up using K-M plotter database [36]. Indeed, K-M analysis revealed that elevated transcript levels of NADH:ubiquinone oxidoreductase subunit A6 (NDUFA6), NADH:ubiquinone oxidoreductase subunit A8 (NDUFA8), NADH:ubiquinone oxidoreductase subunit A9 (NDUFA9), NADH:ubiquinone oxidoreductase subunit B5 (NDUFB5), NADH:ubiquinone oxidoreductase subunit B8 (NDUFB8), NADH:ubiquinone oxidoreductase subunit C1 (NDUFC1), NADH:ubiquinone oxidoreductase subunit S6 (NDUFS6), succinate dehydrogenase complex flavoprotein subunit A (SDHA), cytochrome B5 type A (CYB5A), solute carrier family 25 member 12 (SLC25A12), electron transfer flavoprotein subunit beta (ETFB), carnitine palmitoyltransferase 1A (CPT1A), phytanoyl-CoA 2-hydroxylase (PHYH), translocase of inner mitochondrial membrane 9 (TIMM9), cytochrome c oxidase subunit 7A2 (COX7A2), cytochrome c oxidase copper chaperone (COX11), cytochrome B5 reductase 3 (CYB5R3) and voltage dependent anion channel 2 (VDAC2) were significantly associated with poor OS in AML patients (log-rank P value < 0.05 and FDR = 1%) (Supplementary Table 1 and Fig. 1A-G). Notably, NDUFA6, SDHA, CYB5A, SLC25A12, ETFB and CPT1A remained significant when running a multivariate analysis including gender and treatment (untreated and chemotherapy treatment).

Fig. 1

Elevated levels of mitochondrial/OXPHOS mediators correlate with poor prognosis of AML patients

A-G Kaplan-Meier survival plots of A NADH:ubiquinone oxidoreductase subunit A6 (NDUFA6), B NADH:ubiquinone oxidoreductase subunit C1 (NDUFC1), C succinate dehydrogenase complex flavoprotein subunit A (SDHA), D cytochrome B5 type A (CYB5A), E solute carrier family 25 member 12 (SLC25A12), F electron transfer flavoprotein subunit beta (ETFB) and G carnitine palmitoyltransferase 1A (CPT1A)) in AML patients with a follow-up threshold of 120 months based on KM plotter AML dataset. Samples were divided into low (blue) and high (red) expression cohorts for each gene. Hazard ratio (HR) and log-rank P values are illustrated in each plot. H HR and log-rank P values of the prognostic potential of indicated genes in AML patients based on Gene Expression Profiling Interactive Analysis (GEPIA) database

Event-free survival (EFS) is defined as the time from randomization to an event (as disease progression, treatment discontinuation or death). Yin and colleagues reported that EFS provided more precise assessment of the efficacy of drug therapy in AML patients [39]. Indeed, heightened levels of NDUFA6, NDUFC1, SDHA, CYB5A, SLC25A12, ETFB and CPT1A were associated with poor EFS in AML patients (Table 1), Next, we sought to validate the prognostic value of the above-mentioned mitochondrial/OXPHOS mediators for AML patients utilizing the Gene Expression Profiling Interactive Analysis (GEPIA) tool [37]. Genome-wide transcriptomics (RNA-Seq) datasets used by GEPIA is based on the UCSC Xena project [37]. GEPIA survival analysis confirmed that increased expression of NDUFA6, NDUFC1, SDHA, CYB5A, SLC25A12, ETFB and CPT1A significantly correlated with shorter OS of AML patients (Fig. 1H).

Table 1 Hazard ratios (HR) and log-rank P values of the assessed biomarkers in AML patients (n = 525) for event-free survival (EFS). Follow up threshold: 120 months

Furthermore, we took advantage of the BeatAML2 dataset which comprises real-world cohort of ~ 940 biospecimens with genomic, transcriptomic, and clinical annotations obtained from young (< 45 years) and older patients with de novo, transformed, or therapy-related AMLs [38]. Notably, upregulated levels of NDUFA6, SDHA, SLC25A12, ETFB and CPT1A were significantly associated with worse OS in AML patients (Table 2).

Table 2 Hazard ratios (HR) and log-rank P values of the assessed biomarkers in AML patients (BeatAML dataset, OHSU, Cancer Cell 2022) for overall survival (OS). Follow up threshold: 120 months

Next, we exploited the Human Protein Atlas database to inspect the subcellular localization and expression of the assessed markers (Supplemental Table 2). Indeed, NDUFA6 and NDUFC1 encodes for accessory subunits of the mitochondrial complex I which promotes the transfer of electrons from NADH to the ETC [23, 45]. SDHA encodes for a major catalytic subunit of mitochondrial complex II [46]. In the citric acid cycle, complex II promotes the oxidation of succinate to fumarate [46]. CYB5A is a flavoprotein reductase which catalyzes electron transfer from NADH to target substrate [47]. SLC25A12 functions as a mitochondrial aspartate/glutamate carrier [48, 49]. ETFB is an electron-transfer-flavoprotein which transfers electrons between flavoprotein dehydrogenases and flavoprotein ubiquinone oxidoreductase [50]. CPT1A is the first rate-limiting enzyme of fatty acid oxidation which promotes the mitochondrial uptake of fatty acids [51]. In line with our analysis, SLC25A12, ETFB and CPT1A were reported to predict the prognosis of AML patients [49, 52].

In addition, time-dependent ROC and area under time dependent ROC curve (AUC) analysis revealed that the AUC values of NDUFA6 were 0.56, 0.54 and 0.72 at 12, 48 and 100 months respectively (Figure.S1A). The AUC values of SDHA were 0.58, 0.56 and 0.56 whereas those of SLC25A12 were 0.50, 0.49 and 0.55 at 12,24 and 48 months respectively (Figure.S1B-C). ETFB had AUC values of 0.56, 0.56 and 0.59 whereas CPT1A had AUC values of 0.51, 0.49 and 0.53 at 1,2 and 4 years respectively (Figure.S1D-E).

Elevated levels of NDUFA6, SDHA, CYB5A, SLC25A12, ETFB and CPT1A are differentially associated with poor prognosis in AML patients with distinct mutations

Restricting K-M OS survival analyses to AML patients with distinct mutations revealed that upregulated levels of NDUFA6, SDHA and CPT1A were associated with unfavourable prognosis in mutated IDH1 AML patients (Table 3). Higher levels of NDUFA6 and CPT1A significantly correlated with poor OS of AML patients with IDH2 mutation. Elevated levels of NDUFA6, SDHA, CYB5A, SLC25A12, ETFB and CPT1A were associated with shorter OS of NPM1 and FLT3-ITD mutated AML patients (Table 3). Upsurged levels of SDHA, ETFB and CPT1A were linked to unfavourable prognosis of AML patients with FLT3-TKD mutation. Increased expression of SLC25A12 and CPT1A correlated with poor OS of AML patients with CEBPA mutation (Table 3).

Table 3 Hazard ratios (HR) and log-rank P values of the assessed biomarkers associated with poor prognosis in patients with mutated AMLsSLC25A12, ETFB and CPT1A are significantly overexpressed in AML biospecimens compared to healthy bone marrow-derived mononuclear cells

Next, we questioned whether the above-mentioned biomarkers are differentially expressed in primary human AML versus normal biospecimens. Exploiting BeatAML2 dataset revealed that there were no statistically significant differences in the mRNA levels of NDUFA6, SDHA, and CYB5A in biospecimens obtained from AML patients compared to healthy BMNCs (Fig. 2A-C). However, the transcript levels of SLC25A12, ETFB and CPT1A were significantly higher in AML compared to healthy BMNCs (Fig. 2D-F).

Fig. 2

Solute carrier family 25 member 12 (SLC25A12), electron transfer flavoprotein subunit beta (ETFB) and carnitine palmitoyltransferase 1A (CPT1A) are significantly overexpressed in AML biospecimens compared to healthy bone marrow-derived mononuclear cells (BM MNC)

A-F Transcript levels of the indicated genes (A: NADH:ubiquinone oxidoreductase subunit A6 (NDUFA6), B: succinate dehydrogenase complex flavoprotein subunit A (SDHA), C: cytochrome B5 type A (CYB5A), D: solute carrier family 25 member 12 (SLC25A12), E: electron transfer flavoprotein subunit beta (ETFB) and F: carnitine palmitoyltransferase 1A (CPT1A)) in AML biospecimens as well as healthy bone marrow-derived mononuclear cells obtained from BeatAML.2 dataset (38). *: P value ≤ 0.05 compared to healthy bone marrow-derived mononuclear cells

AML patients with higher levels of CYB5A, SLC25A12 or CPT1A had significantly higher levels of circulating as well as engrafted blasts compared to low-expressing cohorts

Monitoring the percent of blasts in the peripheral blood as well as bone marrow is critical for the diagnosis, prognosis, and therapy of AML patients [2, 53, 54]. Accordingly, we inspected whether AML patients with low and high levels of biomarkers as NDUFA6, SDHA, CYB5A, SLC25A12, ETFB and CPT1A have differential leukemic burden in their peripheral blood and bone marrow biospecimens. To this end, we categorized AML patients (BeatAML2, OHSU, Cancer 2022 dataset) according to the mRNA expression of the assessed genes (z-scores relative to all samples – log RNA-Seq RPKM) into low (z score ≤ -1) and high (z score ≥ 1) expressing cohorts. Statistical analyses of the clinical attributes of low- and high-expressing cohorts are summarized in Supplemental Table 3. Notably, CYB5Alow AML cohort had significantly lower percent of blasts in both the peripheral blood as well as bone marrow compared to CYB5Ahigh AML cohort (Fig. 3A-B). Similar findings were also noted in SLC25A12low versus SLC25A12high and in CPT1Alow versus CPT1Ahigh AML cohorts (Fig. 3C-F). It is worth mentioning that the current regimen and biospecimen type of SLC25A12low cohort (69% bone marrow aspirate and 31% peripheral blood) were statistically significant from that of SLC25A12high cohort (45.13% bone marrow aspirate, 7.01% leukapheresis and 47.79% peripheral blood) (Supplemental Table 3). Otherwise, the age at biospecimen collection, ethnicity category, sex, specimen type and current regimen were not statistically different among other low- and high-expressing cohorts (Supplemental Table 3).

Fig. 3

The percent of AML blasts in the peripheral blood as well as the bone marrow of AML patients with low and high expression levels of the indicated genes

Based on the mRNA expression levels of the indicated genes (mRNA expression z-scores relative to all samples – log RNA-Seq RPKM) in AML patients (OHSU dataset, Cancer Cell, 2022) [38], cBioPortal tool was exploited to inspect the percent of circulating blasts as well as blasts in the bone marrow biospecimens of AML patients with low (z score ≤ -1) and high (z score ≥ 1) expression levels of the assessed genes. A, B The percent of circulating blasts (A) and blasts in the bone marrow (B) of cytochrome B5 type A (CYB5A)low and CYB5Ahigh AML patients (BeatAML.2 dataset [38]). C, D The percent of circulating blasts (C) and blasts in the bone marrow (D) of solute carrier family 25 member 12 (SLC25A12)low and SLC25A12high AML patients. E, F The percent of circulating blasts (E) and blasts in the bone marrow (F) of carnitine palmitoyltransferase 1A (CPT1A)low and CPT1Ahigh AML cohorts

Prevalence of genetic mutations in AML patients with low and high levels of mitochondrial/OXPHOS genes

Given the prevalence of genetic mutations in AML, we examined the correlation between the transcriptional levels of the forementioned genes (low versus high expressing cohorts) and AML mutations using the cBioPortal tool [41]. Intriguingly, NPM1 and serine/arginine-rich splicing factor 2 (SRSF2) mutations were higher in SDHAlow and CPT1Alow AML cohorts respectively (Fig. 4A-B). In contrast, FLT3-ITD, NPM1 and IDH1 mutations were more common in CPT1Ahigh AML patients (Fig. 4B).

Fig. 4

Frequency of genetic mutations in AML patients with low and high levels of the indicated genes

A, B Mutation frequencies of the indicated genes in AML patients with low and high expression levels of succinate dehydrogenase complex flavoprotein subunit A (SDHA) (A) and carnitine palmitoyltransferase 1A (CPT1A) (B). *: Adjust P value ≤ 0.05

Functional prediction and pathway enrichment analysis of mitochondrial/OXPHOS genes in primary human AML biospecimens

We examined the genes which are co-expressed with mitochondrial/OXPHOS genes using the cBioPortal tool [38, 41]. The expression of NDUFA6 positively correlated with the upregulation of several genes including succinate dehydrogenase complex iron sulfur subunit B (SDHB), cysteine rich with EGF like domains 2 (CRELD2), adaptor related protein complex 2 subunit sigma 1 (AP2S1), HIG1 hypoxia inducible domain family member 1A (HIGD1A), cytochrome C oxidase subunit 7B (COX7B), proteasome maturation protein (POMP), myosin 1C (MYO1C), mitochondrial ribosomal protein S15 (MRPS15), NADH: ubiquinone oxidoreductase subunit B9 (NDUFB9), calmodulin binding transcription activator 1 (CAMTA1), G protein regulated inducer of neurite outgrowth 1 (GPRIN1) and BolA family member 3 (BOLA3). Gene Ontology (GO) enrichment analysis of NDUFA6 co-expressing genes showed the enrichment of distinct biological processes (BP) as mitochondrial ATP synthesis coupled electron transport, cellular respiration, oxidative phosphorylation, positive regulation of cell migration by vascular endothelial growth factor signalling pathway, positive regulation of vascular endothelial growth factor signalling pathway and mitochondrial respirasome assembly (Fig. 5A).

Fig. 5

Gene ontology of the biological process of mitochondrial/OXPHOS co-expressed genes in AML patients

A-E) Bar chart of top enriched terms from the GO_Biological_Process_2023 gene set library which are significantly co-expressed with NADH: ubiquinone oxidoreductase subunit A6 (NDUFA6) (A), succinate dehydrogenase complex flavoprotein subunit A (SDHA) (B), cytochrome B5 type A (CYB5A) (C), electron transfer flavoprotein subunit beta (ETFB) (D) and carnitine palmitoyltransferase 1A (CPT1A) (E) in biospecimens obtained from AML patients (OHSU Cancer Cell 2022 dataset). The cut-off for Spearman’s correlation is > 0.5 and adjusted P value ≤ 0.05. The top 10 enriched terms for the input gene set are displayed based on the -log10 (P value), with the actual P value illustrated next to each term. The term at the top has the most significant overlap with the input query gene set

On the other hand, the expression of SDHA positively correlated with the expression of acetyl-CoA acetyltransferase 1 (ACAT1), SUMO1 activating enzyme subunit 1 (SAE1), ATP synthase membrane subunit C locus 3 (ATP5MC3), TNF receptor associated protein 1 (TRAP1), guanine monophosphate synthase (GMPS) and protein phosphatase 1 catalytic subunit gamma (PPP1CC). GO BP analysis of SDHA co-expressing genes highlighted the enrichment of isoleucine metabolic process, purine nucleibase biosynthetic process, ketone body metabolic process and positive regulation of protein sumoylation (Fig. 5B).

GO BP enrichment analysis of CYB5A co-expressed genes revealed their association with aromatic compound biosynthetic process, aldehyde biosynthetic process, protein localization to basolateral plasma membrane, maintenance of apical/basal cell polarity, peptidyl-cysteine modification, protein oxidation, oligosaccharide-lipid intermediate biosynthetic process, and dolichol metabolic process (Fig. 5C).

ETFB co-expressed genes were rather enriched with biological processes as protein insertion into ER membrane by stop-transfer membrane-anchor sequence, cellular respiration, mitochondrial electron transport, NADH to ubiquinone and energy derivation by oxidation of organic compounds (Fig. 5D).

GO enrichment analysis of CPT1A co-expressed genes revealed the enrichment of BP as positive regulation of protein localization to chromosome, telomeric region, nucleosome disassembly, tRNA aminoacylation for protein translation, regulation of nucleotide-excision repair, regulation of mitotic metaphase/anaphase transition and chromatin organization (Fig. 5E).

Differential dependency of AML cells on mitochondrial/OXPHOS genes

Based on the above-mentioned findings, we questioned whether human AML cells are dependent on the evaluated mitochondrial/OXPHOS genes. At genome-wide scale, the CRISPR–Cas9 system objectively identifies genes which are indispensable for the proliferation and survival of cancer cells including AML [44, 55]. To this end, we exploited the Cancer Dependency Map database (which comprises large-scale functional genomics profiling using CRISPR loss-of-function screens) to investigate the potential dependency of AML cells on mitochondrial/OXPHOS genes. Dempster and colleagues developed Chronos model which addressed several limitations associated with other models including sgRNA efficacy, variable screen quality and copy number bias [43]. Negative Chronos scores indicate slower growth of cancer cells [43].

Indeed, AML cells exhibited heterogeneous responses to CRISPR-mediated genetic pertubation of OXPHOS genes which encode for subunits of the OXPHOS complexes: I (NDUFA6 and NDUFC1) and II (SDHA) (Fig. 6A-C). Of note, AML cell lines with FLT3-ITD mutations (as MV4-11, MOLM13 and MOLM14) [5] were more vulnerable to the genetic depletion of NDUFA6, NDUFC1 and SDHA (Fig. 6A-C). In contrast, modest dependencies of AML cells were observed for CRISPR KO of CYB5A, SLC25A12, ETFB and CPT1A (Fig. 6D-G). Overall, these findings highlight the therapeutic potential of targeting mitochondrial/OXPHOS dependent/addicted AML cells.

Fig. 6

Differential dependencies of AML cells on mitochondrial/OXPHOS genes

A-G Dot plot depicting the gene dependency effect (CRISPR, DepMap Public 23Q4+ Score, Chronos) of AML cells on the indicated mitochondrial/OXPHOS genes (A: NADH : ubiquinone oxidoreductase subunit A6 (NDUFA6), B: NADH : ubiquinone oxidoreductase subunit C1 (NDUFC1), C: succinate dehydrogenase complex flavoprotein subunit A (SDHA), D: cytochrome B5 type A (CYB5A), E: solute carrier family 25 member 12 (SLC25A12), F: electron transfer flavoprotein subunit beta (ETFB) and G: carnitine palmitoyltransferase 1A (CPT1A)) plotted on the X-axis versus their corresponding mRNA expression level (log2(TPM+1) plotted on the Y-axis

Heightened expression of glutathione peroxidase 4 (GPX4) is associated with higher percent of circulating and engrafted blasts of AML patients and DNMT3a mutant AML cells are dependent on GPX4

Glutathione peroxidase 4 (GPX4) is included in the geneset of HALLMARK_OXIDATIVE_PHOSPHORYLATION downloaded from GSEA MSigDB. As an antioxidant, GPX4 promotes the reduction of hydrogen peroxide, organic hydroperoxides and lipid hydroperoxides [56]. Km plotter database and GEPIA survival analyses showed controversial results in terms of the prognostic value of GPX4 expression in AML patients. Unlike Km plotter, the survival analyses of GEPIA and BeatAML.2 datasets demonstrated that elevated GPX4 levels are significantly associated with worse OS of AML patients (Fig. 7A and Supplemental Table 2). The AUC values of the ROC curve of GPX4 were 0.51, 0.52 and 0.62 at 2, 4 and 6 years respectively (Figure.S1F). GPX4high AML cohort had significantly higher percent of blasts in both the peripheral blood as well as bone marrow compared to GPX4low AML cohort (Fig. 7B-C).

Fig. 7

Increased expression of glutathione peroxidase 4 (GPX4) is associated with higher percent of circulating and engrafted blasts as well as poor prognosis of AML patients and DNMT3a mutant AML cells are dependent on GPX4

A Kaplan–Meier survival curve depicting the hazard ratio (HR) and log-rank P values of the prognostic potential of GPX4 gene in AML patients based on Gene Expression Profiling Interactive Analysis (GEPIA) database. B, C The percent of circulating blasts (B) and blasts in the bone marrow (C) of GPX4low and GPX4high AML patients (BeatAML.2 dataset (38)). D Dot plot depicting the gene dependency effect (CRISPR, DepMap Public 23Q4 + Score, Chronos) of AML cells on GPX4 plotted on the X-axis versus their corresponding mRNA expression level (log2(TPM + 1) plotted on the Y-axis

Several patient-derived AML cell lines were evidently susceptible to CRISPR KO of GPX4 (Fig. 7D). Notably, OCIAML2 and OCIAML3 AML cell lines were preferentially responsive to the genetic ablation of GPX4. It is worth noting that OCI-AML2 and OCI-AML3 are the only known human AML cell lines with the DNMT3A mutation. Altogether, these findings underscore the therapeutic potential of targeting GPX4 for AML therapy.