Introduction

Acute myeloid leukemia (AML) is a rare but lethal malignancy of precursors of myeloid stem cell lineage, which includes platelets, red blood cells (RBCs), and white blood cells (WBCs) but excludes B and T cells, leading to clonal proliferation and overproduction of neoplastic myeloid stem cells1,2. AML has an overall five-year survival rate of approximately 25% due to its aggressive clinical course3,4. Adults with AML typically experience a range of symptoms, including exhaustion, bruising or bleeding, fever, and infections, which are indicators of failing bone marrow5. AML is more prevalent in adults than in children. In contrast, acute lymphoblastic leukemia (ALL) is the most common form of leukemia in children and the second most frequent type in adults6. Approximately half of the leukemia cases are acute, with ALL being the most common cancer in children. In contrast, the incidence of AML is around 3 to 8 cases per 100,000 people annually worldwide, with more than 80% of these cases occurring in adults7,8. The survival rate for leukemia over five years is 54%. However, for children under 15, the five-year relative survival rate is 89% for ALL and 60% for AML7.

Currently, a combination of morphological examination, immunohistochemistry (IHC), and cytogenetic analysis of a bone marrow aspirate—an invasive and painful procedure is used to diagnose leukemia. The gold standard for these diagnostic methods, which are performed on bone marrow aspirates, is trephine biopsy. Patients must undergo an invasive, painful, and expensive procedure known as a bone marrow biopsy9. Therefore, there is an urgent need to prioritize the development of sensitive, specific, and early diagnostic and prognostic biomarkers for non-invasive diagnosis10. Such advancements could provide new therapeutic options and guidelines to improve disease prognosis.

Metabolomics is the study of systemic changes in metabolism associated with chemical, biological, or disease processes. These metabolic alterations are analyzed using techniques such as liquid chromatography-mass spectrometry (LC-MS), gas chromatography-mass spectrometry (GC-MS), and nuclear magnetic resonance (NMR) spectroscopy. While LC-MS has gained prominence—accounting for 70% of published metabolomics studies, NMR retains distinct advantages in this field. Notably, in 2021 alone, more than 926 papers on NMR-based metabolomics were published, marking the highest annual output to date and underscoring its expanding role in research. This trend highlights NMR’s continued growth and its significant potential to advance scientific discovery11.

Nuclear magnetic resonance (NMR) is widely used to analyze biological fluids such as serum, plasma, and transcellular fluids12,13, as well as intact tissue samples14,15. NMR enables quantitative analysis of metabolite levels and provides detailed insights into metabolite properties, making it a powerful tool for metabolomics. A key advantage of NMR is its ease of sample handling, which simplifies experimental workflows16. In the study of disease-specific metabolite variations and biomarker validation, NMR-based metabolomics has been predominantly employed in conjunction with robust statistical methods and validated using independent sample cohorts17. This approach is particularly valuable because metabolomics generates highly accurate data about biological systems18, with applications spanning precision medicine, drug toxicity mechanism analysis, and the diagnosis and treatment of blood disorders, including AML17,19,20.

AML remains a significant clinical challenge due to its aggressive nature, low survival rate, and reliance on invasive diagnostic procedures. The urgent need for non-invasive, accurate, and early diagnostic tools has driven scientific interest toward metabolomics, particularly NMR-based approaches. NMR offers significant advantages, including high reproducibility, minimal sample preparation, and reliable quantification of metabolites, positioning it as a powerful tool for identifying disease-specific metabolic signatures. This study leverages NMR-based serum metabolomics to delineate metabolic alterations in AML patients compared with healthy individuals, aiming to identify reliable biomarkers for diagnosis and prognosis. These findings may contribute to the way of precision medicine strategies, and improved clinical outcomes for AML patients.

Materials and methods

Population recruitment and sample collection

The required sample size was calculated to be 56 subjects per group using a two-sided test formula, with a study power of 80% (Z1–β = 1.28) and a 95% confidence level (Z1–α = 1.96). The anticipated mean glycine levels were 220.4 ± 32.2 in controls and 202.7 ± 42.9 in cases, based on Liu et al. (2019b). Applying these parameters to the sample size formula:

$${\text{N = ((}}Z_{1} - \beta + Z_{1} - \alpha )^{2} \times (\sigma _{1}^{2} + \sigma _{2}^{2} ))/(\mu _{1} - \mu _{2} )^{2}$$

Adult patients diagnosed with acute myeloid leukemia (AML) of both genders were recruited, including individuals before and after their first follow-up bone marrow examination during standard chemotherapy treatment (cytarabine [ara-C] combined with daunorubicin or idarubicin).

A total of 112 participants were included in the study: 56 AML patients (with samples collected at two time points—pre-treatment and post-treatment) and 56 age- and gender-matched healthy controls. Overall, 168 samples were analyzed and categorized into three groups: : Group 1-AML diagnosed cases without treatment (56), Group 2-AML cases with standard treatment (56), and Group 3-healthy controls (56).

To minimize confounding variables, controls were selected from the same familial background as AML patients, ensuring shared genetic and lifestyle factors. Individuals with comorbidities (e.g., renal, cardiac, or hepatic failure) or those receiving concurrent medications were excluded. This design aimed to control for diet, medication use, and pre-existing health conditions that could confound metabolomic data, ensuring observed metabolic changes were AML-specific. The study adhered to the Declaration of Helsinki guidelines. All participants provided written informed consent, and confidentiality was rigorously maintained. Compliance with the General Data Protection Regulation (GDPR) ensured the secure handling of personal and medical data.

Patients diagnosed with AML (based on peripheral blood analysis, bone marrow examination, cytochemistry, immunohistochemistry, and flow cytometry) were recruited from Jinnah Hospital, Lahore, by the inclusion and exclusion criteria.

Metabolomics analysis by nuclear magnetic resonance (NMR) technique

Sample preparation for NMR

A volume of 200 µL of the homogenized serum was combined with 400 µL of deuterated water (D₂O) to facilitate nuclear magnetic resonance (NMR) analysis. This mixture was then centrifuged for 5 min at 6,000 rpm to separate any particulate matter and ensure a clear solution for NMR measurement. Following centrifugation, 550 µL of the supernatant was carefully transferred to a 5 mm NMR tube, ready for spectroscopic analysis.

NMR process

The proton NMR (1H-NMR) spectra were obtained using a 600-MHz Bruker Avance AV-600 spectrometer (Bruker Biospin, Rheinstetten, Germany), equipped with a z-gradient probe, at a temperature of 300.0 K. For each serum sample, three one-dimensional (1D) experiments were conducted to gather comprehensive spectral data. The first experiment employed a standard 1H-NMR spectrum with a Nuclear Overhauser Effect Spectroscopy (NOESY) technique, which included water suppression to eliminate the overwhelming signal from water. This spectrum was recorded using a 90° flip angle, a recycle delay of 4 s, a spectral width of 10 kHz, and a total of 32,768 data points. A receiver gains of 362 was utilized, with a mixing time of 0.1 s, culminating in a total of 280 scans along with four dummy scans to enhance sensitivity. The second experiment consisted of a diffusion-edited 1H-NMR spectrum, also incorporating water suppression to isolate signals from larger molecules, which often carry significant biochemical information. The third experiment was a T2-edited 1H-NMR spectrum utilizing the Carr–Purcell–Meiboom–Gill (CPMG) pulse sequence (Fig. 1). Post-acquisition, the obtained NMR spectra were manually corrected for phase and baseline distortions using AMIX version 3.9 (Bruker Biospin, Rheinstetten, Germany).

Fig. 1
figure 1

1D1H-NMR spectrum with labeled peaks corresponding to specific metabolites. Identified metabolites include glycerol, cholesterol, lipids (LDL, VLDL), various amino acids, organic acids, carbohydrates, and other small molecules. Numbers in parentheses indicate peak assignments.

Full size image

The raw NMR data were imported using MestReNova 14.3.1, which supports data from spectrometers like Bruker 600 MHz. Zero filling was applied to increase digital resolution by adding extra data points, improving peak clarity. Apodization functions such as exponential, Gaussian, or Lorentzian were used to enhance peak shapes and reduce noise. Fourier transformation converted the FID from the time domain to the frequency domain, yielding a spectrum with chemical shifts. Phase and baseline corrections ensured proper peak alignment and accurate quantification. Chemical shift calibration aligned the spectrum with a known reference, and TSP was used as an internal standard. Signal deconvolution separated overlapping peaks, and peak picking identified significant peaks for analysis. Finally, the data was divided into smaller regions (bucketing) for multivariate analysis, enabling better pattern recognition and metabolite quantification.

Statistical analysis

The data were entered and analyzed using the Statistical Package for Social Sciences (SPSS) version 26. Qualitative variables were described as frequencies and percentages. The Shapiro–Wilk test was used to assess the normality of the data, while Levene’s test was applied to determine homogeneity of variance. Normally distributed quantitative variables were described mean ± SD, whereas, non-normally distributed variables were presented as median (interquartile range, IQR). Parametric tests (independent sample t-test) and non-parametric tests (Mann–Whitney U-test) were employed to compare variables between the two groups. Correlation analyses were conducted to evaluate the association between parameters in patients and controls. A p-value < 0.05 was considered statistically significant.

Multivariate statistical techniques, including Principal Component Analysis (PCA) and Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA), were performed using SIMCA 14.0 (Umetrics, Sweden) to discriminate between AML patients and controls. Additional statistical analyses using SPSS and GraphPad Prism (version specified if known) included descriptive statistics, Receiver Operating Characteristic (ROC) curves, and Area Under the Curve (AUC) calculations to evaluate diagnostic/prognostic accuracy. A linear regression model (SPSS) was applied to explore relationships between key metabolites in AML patients and controls, providing insights into their potential role in disease diagnosis, progression, and treatment response.

Results

A total of 112 subjects were recruited for this study, comprising 56 adult AML cases and 56 healthy controls. Each group consisted of 40 males (71%) and 16 females (29%). The comparison of demographic data revealed that hematological parameters differed significantly between cases and controls (p < 0.05), as shown in Table 1. Additionally, a comparison between pre-treatment and post-treatment AML patients demonstrated a significant reduction in TLC (p < 0.05) and an increase in platelets among post-treatment cases (Table 2).

Metabolomics in controls/aml patients (pre- and post-treatment)

In this study, a wide range of metabolites were analyzed in AML patients and healthy controls. Most metabolites were normally distributed, while cholesterol, glucose, TMAO, proline, glycine, glycerolipids, citrulline, methionine, glycoproteins, serine, pyruvate, glycerol, alpha-glucose, methyl-histidine, and formate exhibited non-normal distributions. Table 3 highlights the significant imbalance of metabolites between cases and controls (p < 0.05). Similarly, Table 4 illustrates statistically significant differences in metabolite levels between pre-treatment and post-treatment groups (Table 4).

Comparison between remission and non-remission groups

The study found that non-remission patients exhibited significantly higher levels of lipids, 2-hydroxybutyrate, adipic acid, glutamine, glutamate, and the citric acid cycle intermediate ketoglutarate (p < 0.05), suggesting dysregulated energy metabolism. Additionally, elevated citrate levels in non-remission patients (p < 0.05) indicated persistent disease activity and altered tricarboxylic acid (TCA) cycle metabolism. Although albumin levels were higher in the non-remission group (15.73 vs. 11.84, p = 0.06), potentially reflecting enhanced liver function, improved nutritional status, or compensatory responses to metabolic/inflammatory demands, this difference did not reach statistical significance. Notably, choline (22.24 vs. 15.34, p = 0.01) and serine (24.46 vs. 17.41, p < 0.001) levels were significantly elevated in non-remission patients (Table 5).

Diagnostic/prognostic accuracy

The area under the curve (AUC), sensitivity, specificity, and p-value of key biomarkers were analyzed to evaluate their ability to distinguish between disease status and healthy controls. Top-performing diagnostic markers, including creatine, lipids, cholesterol, glycoproteins, and citrulline, demonstrated high accuracy (AUC > 0.8). Moderate performers such as leucine, isoleucine, valine, and carnitine showed acceptable discriminatory power (AUC 0.7–0.8), while pyruvate, alpha-glucose, and glycerol lipids performed poorly (AUC < 0.7), with creatine emerging as the most robust diagnostic marker (Fig. 2A). For prognostic accuracy, moderate markers such as glutamate, glutamine, ketoglutarate, and threonine exhibited reasonable utility (AUC ≈ 0.7). Poor prognostic markers, including alanine, arginine, carnitine, valine, and lipids, displayed limited discriminatory power (AUC 0.61–0.67) (Fig. 2B).

Fig. 2
figure 2

Metabolic overlay: (A) ROC (Receiver Operating Characteristic) curve illustrates the diagnostic performance of several biochemical markers for AML (Acute Myeloid Leukemia) diagnosis; (B) ROC curves in this Fig. assess the diagnostic performance of four biochemical markers—Citrate, Choline, Threonine, and Glutamate-in distinguishing between remission and non-remission groups.

Full size image
Table 1 Comparison of demographic and hematological parameters between normal controls and AML patients.
Full size table
Table 2 Comparison of hematological parameters in AML patients pre- and post-treatment.
Full size table
Table 3 Relative intensities of metabolomics data of controls group and patients group.
Full size table
Table 4 Relative intensities of metabolomics data of AML groups (pre- treatment and post-treatment).
Full size table
Table 5 Comparison of relative intensities of metabolomics between AML patients after treatment (remission/non-remission).
Full size table

Screening of potential diagnostic biomarkers

To evaluate the potential diagnostic model for AML, logistic regression analysis was employed. This method demonstrates strong discriminative and predictive capabilities and is widely used to screen epidemiological risk factors. The diagnostic logistic regression model was assessed using Receiver Operating Characteristic (ROC) curve analysis, achieving an AUC of 0.98, which underscores its excellent discriminatory power and the utility of biochemical markers in diagnosis (Fig. 3A). Coefficients and absolute coefficients for glutamate and choline were calculated, highlighting their contribution to predicting prognosis. A separate logistic regression model for AML prognosis, incorporating citrate, glutamate, and choline, was evaluated via ROC analysis, demonstrating robust discrimination between remission and non-remission cases (Fig. 3B).

Fig. 3
figure 3

(A) ROC curveshowing the performance of a logistic regression model for AML diagnosis usingacetate, creatine, and lactate as predictors (AUC = 0.98). (B) ROC curve depicting themodel’s ability to predict prognosis/treatment response based on citrate, glutamate, andcholine levels (AUC = 0.89).

Full size image

Hierarchical clustering analysis

The hierarchical clustering dendrogram illustrated the grouping of samples based on biochemical marker similarities, revealing two distinct clusters: probable AML patients (red) and controls (blue). The clear separation between AML patients and controls reflects significant biochemical differences, supporting the markers’ diagnostic value for disease identification and monitoring (Fig. 4A). In remission vs. non-remission comparisons, clustering analysis showed that patients in remission clustered together, indicating biochemical homogeneity. This suggests specific markers may predict prognosis or treatment response. Shared metabolic profiles among remission patients further emphasize their potential as prognostic tools (Fig. 4B).

Fig. 4
figure 4

(A, B) Dendrograms illustrating the hierarchical clustering of samples based on biochemical markers. The vertical axis represents dissimilarity between clusters (measured as distance), while the horizontal axis shows individual samples. Group labels (red for Group 1, AML patients, and blue for Group 0, controls) indicate sample distribution. The clustering shows a distinct separation between groups, suggesting strong differences in biochemical markers between patients and controls.

Full size image

Correlational studies for metabolites

Heatmap analysis revealed strong correlations between biochemical markers (e.g., hydroxybutyrate, leucine, valine, and lactate), reflecting altered metabolic profiles in AML patients. Positive correlations between energy production and amino acid metabolism were observed, whereas negative correlations suggested shifts in glycolytic activity. A positive association between lipids and cholesterol was identified, indicative of dyslipidemia. These findings provide insights into disease mechanisms and potential biomarkers (Fig. 5A). In remission vs. non-remission groups, heatmaps highlighted distinct metabolic profiles. Strong positive correlations between amino acids and hydroxybutyrate implied energy balance, while glutamine-glutamate linkages and negative correlations with glycolytic activity underscored metabolic pathway variations (Fig. 5B).

Fig. 5
figure 5

(A) Heatmap illustrating correlations between biochemical markers in patient and control groups. Positive correlations appear in red, negative in blue, with intensity indicating correlation strength. The diagonal represents perfect self-correlation (coefficient = 1), aiding in identifying biomarker relationships relevant to disease mechanisms and prognosis. (B) Heatmap showing biochemical marker relationships in remission and non-remission groups. Red indicates positive correlations, blue indicates negative, with the diagonal representing self-correlation.

Full size image

PCA (principal component analysis)

PCA visualized sample clustering between AML patients and controls, reducing dimensionality while preserving variance. Key metabolites contributing to variability, such as valine, isoleucine, leucine, and lipids, were identified as potential biomarkers (Fig. 6A). These metabolites exhibited similar contributions to principal components, suggesting correlated metabolic pathways. Metabolites distant from the origin (e.g., cholesterol) significantly influenced variability. Correlations among metabolites accounted for sample separation, with high concentrations of key metabolites driving group differences (Fig. 6B).

Fig. 6
figure 6

(A) PCA scatter plot showing some separation between the three groups based on metabolomics profiles. Group 1 (green circles) and Group 2 (blue squares) cluster closely, while Group 3 (red triangles) shows greater variability with outliers. Separation along the first principal component (t1) indicates significant metabolic differences. (B) PCA loading plot illustrating metabolite distribution across two principal components (p1 and p2) from SIMCA analysis.

Full size image

PLS-DA (partial least squares discriminant analysis)

The PLS-DA scatter plot delineates three groups: controls, AML patients, and follow-up AML patients. Group 1 (controls) clusters tightly, indicating high metabolic similarity, while Group 2 (AML patients) exhibits moderate variability. Group 3 (follow-up AML patients) displays greater variability along the t1 axis, suggesting distinct metabolomic profiles. Hotelling’s T² distribution demonstrates homogeneity in Groups 1 and 2, with the first component explaining 28.6% of the variance and the second component contributing 21.7%, collectively accounting for 50.3% of the total variance. Group 3 shows distinct metabolic processes and heightened variability, implying potential differences in biological or disease status compared to the homogeneous Groups 1 and 2 (Fig. 7A). Analysis of the PLS-DA model in SIMCA highlights the metabolites contributing to the separation observed in the score plot, with the w*c1 and w*c2 axes explaining 28.6% and 21.7% of the variance, respectively (Fig. 7B). To validate the model, permutation analysis was performed using PLS-DA, confirming that the model is not overfit (Fig. 7C).

Fig. 7
figure 7

(A) PLS-DA scatter plot in SIMCA representing sample distribution based on metabolomics profiles. (B) PLS-DA loading plot showing metabolite contributions to group separation. Axes wc1 and wc2 account for 28.6% and 21.7% variance, respectively. Alpha-glucose, methyl histidine, and choline strongly influence separation, while clustered metabolites (valine, leucine, isoleucine) suggest common pathways. Metabolites near the origin (citrate, methionine) contribute less variance. (C) Permutation plot validating PLS-DA model robustness with R2 (green circles) and Q2 (blue squares) over 20 permutations and 7 components. Higher R2 and Q2 values indicate better model performance and minimal overfitting.

Full size image

Discussion

Cancer metabolism is one of the prompt proclivities of biology research. It depends on the fact that metabolic activities are changed in cancer cells compared to those in normal ones. As a heterogeneous disease, every type of cancer has its metabolic characteristics13.

The present study compared the metabolite levels between acute myeloid leukemia (AML) patients and healthy individuals. Our study highlights key metabolic disruptions that may have occurred due to underlying pathology. Metabolomic profiling revealed significantly altered lipid, amino acid, and carbohydrate metabolism pathways in AML patients compared to healthy controls, underscoring the disease’s impact on systemic metabolism. This study also demonstrates a significant alteration in cholesterol and total lipid levels in AML patients (p < 0.001), linked to disrupted synthesis or uptake pathways, possibly due to the high-energy demands of rapidly increasing leukemic cells. Lipid dysregulation, driven by the Warburg effect and altered lipid biosynthesis, reflects metabolic reprogramming in cancer cells21,22. Post-treatment lipid levels are partially restored23,24, indicating a gradual normalization of lipid metabolism25. AML patients exhibited higher acetate levels, suggesting disruption in lipid metabolism, especially fatty acid oxidation, indicating reliance on acetate as an alternative energy source26. This reinforces that acetate aids cancer metabolism by providing energy. Post-treatment patients showed partial recovery in acetate levels, suggesting normalization of lipid metabolism as leukemia burden decreases, aligning with prior observations27.

Branched-chain amino acids (BCAAs) play a crucial role in AML by supporting anabolic processes, protein synthesis, and energy production, which are essential for leukemic cell growth. Significant decreases in branched-chain amino acids (BCAAs) levels among AML patients (leucine: p < 0.001; valine and isoleucine: p = 0.000; isoleucine: p = 0.01), as well as glutamate and glutamine levels, compared to healthy controls, was observed, reflecting their reliance on BCAAs for anabolic processes. This study supports previous research indicating that AML cells heavily rely on BCAAs for anabolic processes, protein synthesis, and energy production28,29,30,31,32. Glutamine depletion emphasizes its role in AML patients, highlighting the importance of glutamine metabolism in AML pathophysiology. Targeting glutamine metabolism, particularly through inhibiting glutaminase enzyme, could selectively impair leukemic cells29. Post-treatment recovery of amino acid levels was minimal, reflecting that chemotherapy effects may persist, or ongoing low-level disease activity33. Interestingly, contrary studies in solid tumors, like those by Katagiri et al. (2018), often show elevated BCAA levels, indicating the context-specific nature of amino acid metabolism in different cancers34.

In AML, glucose metabolism is crucial for supporting the high energy demands of leukemic cells, primarily through enhanced glycolysis. Pyruvate, a key glycolytic product, is further utilized in oxidative phosphorylation or converted to lactate, driving energy production and supporting cell survival in the tumor microenvironment. AML patients exhibit elevated glucose and pyruvate levels, indicating a glycolytic metabolic shift, known as the Warburg effect, with glucose levels higher than healthy controls (p < 0.001). Leukemic cells undergo metabolic reprogramming for ATP production and anabolic intermediates35. Elevated levels of lactate and pyruvate in AML patients support the Warburg effect, where pyruvate is converted to lactate36. AML cells show increased reliance on glycolysis, despite the presence of oxygen, which aligns with evidence indicating enhanced glycolysis and lactate production37. However, contradictory findings highlight the metabolic heterogeneity within AML. Jones et al. described that a certain subpopulation of AML cells, particularly chemotherapy-resistant ones, rely more on oxidative phosphorylation (OXPHOS)38. This metabolic flexibility aids the survival of AML cells in a dynamic environment, suggesting dual-pathway targeting could improve therapeutic efficacy.

The tricarboxylic acid (TCA) cycle plays a critical role in AML by supporting cellular energy production through oxidative phosphorylation. AML disrupts the TCA cycle, with elevated citrate and reduced ketoglutarate levels indicating oxidative phosphorylation impairment38. Mutated AML-producing 2-hydroxyglutarate inhibits α-ketoglutarate-dependent dioxygenases, causing severe metabolic alterations, necessitating genetically specific interventions39. Post-treatment AML patients show increased ketoglutarate and citrate levels, suggesting partial recovery of TCA cycle function and a shift towards oxidative phosphorylation and mitochondrial metabolism40. Successful therapies reduce leukemic burden and promote metabolic homeostasis in surviving AML cells, focusing on glycolytic shift and disruption of the TCA cycle. A combination of therapies targeting both glycolysis and mitochondrial pathways may be essential for effective treatment and prevention of relapse in AML due to metabolic heterogeneity.

Metabolic differences in AML patients and healthy individuals offer insights into the altered biochemical landscape, aiding in diagnostic and therapeutic strategies, distinguishing AML from non-cancerous states, and potentially enabling earlier detection. Metabolic intermediates in diagnosing and predicting AML treatment outcomes are promising, but their utility varies due to the complexity and heterogeneity of AML metabolism.

In AML, creatine plays a pivotal role in maintaining cellular energy homeostasis by replenishing ATP through its conversion to phosphocreatine via the action of creatine kinase. The Creatine, with an AUC of 0.92, is a potent diagnostic marker for AML, reflecting the increased energy demands of leukemic cells41. Its role in ATP regeneration and energy buffering is well-documented in cancer metabolism, particularly in high-proliferation cancers, with studies linking creatine kinase pathway dysregulation to its biomarker potential42. Creatine levels in certain solid cancer subtypes can be suppressed due to alternative metabolic pathways, necessitating context-specific applications in AML diagnostics due to the varying cancer types43.

Acetate, a precursor for acetyl-CoA in lipid biosynthesis and the TCA cycle is an important metabolite in cellular energy production and biosynthesis, has strong diagnostic utility in AML patients due to its use as an alternative carbon source for fatty acid synthesis26. Acetate elevation in AML patients is due to cancer cells using it as an alternative carbon source for fatty acid synthesis, adapting to nutrient scarcity27. Acetate levels in solid cancers like glioblastoma and breast cancer fluctuate, potentially reducing sensitivity in distinguishing cancerous from non-cancerous cells, necessitating acetate combination with other markers for consistent results26.

In AML, lactate is a key metabolite produced during aerobic glycolysis, a metabolic pathway often upregulated in leukemic cells. The lactate, a product of the Warburg effect, is another marker that exhibited high diagnostic potential in this study. Elevated lactate levels reflect the metabolic shift of AML cells toward anaerobic glycolysis, a hallmark of many cancers44. This reliance on glycolysis, even in the presence of oxygen, allows for rapid ATP generation and provides the biosynthetic precursors necessary for cell growth. Lactate’s role in acidifying the tumor microenvironment further supports cancer cell survival by promoting immune evasion and angiogenesis45. It was observed that in some cancers, such as pancreatic or colorectal cancers, lactate levels alone may not provide sufficient diagnostic sensitivity due to the heterogeneous nature of tumor metabolism. In AML, while lactate is elevated due to the Warburg effect, not all AML subtypes rely exclusively on glycolysis, with some subsets, particularly treatment-resistant ones, shifting toward oxidative phosphorylation (OXPHOS)44. This metabolic flexibility could reduce lactate’s effectiveness as a standalone biomarker, suggesting that its use should be context-specific.

While numerous studies have shown that many cancers, including solid tumors, share common metabolic features such as enhanced glycolysis (Warburg effect), amino acid reprogramming, and elevated lipid synthesis46,47. AML demonstrates distinct metabolic dependencies—for example, a flexible shift between glycolysis and oxidative phosphorylation35, a unique reliance on glutamine metabolism48, and interactions with bone marrow adipocytes that support leukemic cell survival via fatty acid transfer and oxidation49. These features distinguish AML from many solid tumors and underscore the disease-specific nature of its metabolic reprogramming.

When we consider the diagnostic potential of combining multiple metabolites like creatine, acetate, and lactate into a diagnostic panel could improve sensitivity and specificity, leveraging their unique diagnostic strengths for improved overall diagnostic accuracy. Metabolic heterogeneity in AML suggests a multi-metabolite panel for consistent detection, with ROC curve analyses showing higher AUC values (0.98) when considering creatine, acetate, and lactate. Yang et al.‘s study on ALL patients identified a metabolomic diagnostic panel of unsaturated lipids, choline, and histidine, revealing that myeloid and lymphoid cell lineages differ in metabolism50. Metabolites, when compared between remission-seeking AML patients, can serve as prognostic indicators, providing insights into disease progression and treatment response, and potentially aiding in personalized AML treatment strategies.

Citrate (AUC = 0.82) is reflected as a prognostic marker in AML, indicating its role in mitochondrial function and the TCA cycle, and its potential role in energy production and lipid synthesis51. Glutamate, derived from glutaminolysis, is negatively associated with remission in AML cells, indicating its role in sustaining proliferation through glutamine metabolism, a marker of poor prognosis52. AML cells rely on glutaminolysis for TCA cycle fuel, indicating poor prognosis. Elevated glutamate levels are linked to cancer cells’ reliance on glutamine in nutrient-deprived environments. Choline, a crucial prognostic marker in AML, is involved in cell membrane synthesis and signalling. Elevated choline levels post-treatment are associated with remission, as upregulation of choline metabolism supports increased membrane demand for cell proliferation50.

Combining multiple metabolites, such as citrate, choline, and glutamate, can improve prognostic precision and predictive accuracy in AML (AUC = 0.89), highlighting the importance of integrated metabolic panels. The prognostic value of metabolites suggests personalized treatment for AML patients with elevated citrate, choline, and glutamate levels. Targeting these pathways could improve treatment efficacy and allow real-time feedback for timely adjustments to therapeutic regimens.

Metabolomics profiling plays a crucial role in distinguishing AML from healthy individuals, diagnosing AML, and predicting treatment responses. While differences in metabolites between AML and ALL are suggested, further studies are needed. Comparisons with solid cancers reveal similarities, emphasizing the need for context-specific interpretation and further research to refine metabolomics marker specificity for diagnosis and treatment.

Metabolomics offers a promising, non-invasive approach to complement genetic testing in AML diagnostics by enabling early detection, treatment response monitoring, and personalized care. However, its routine clinical adoption faces challenges such as high costs, lack of standardization, limited accessibility, and the need for specialized expertise. While established genetic markers like FLT3-ITD and NPM1 remain central to AML management, metabolomics provides functional insights into disease physiology. At present, it should be viewed as a complementary tool rather than a standalone solution. Future research should focus on key metabolic pathways such as one-carbon metabolism, the tryptophan-kynurenine axis, sphingolipid, and branched-chain amino acid metabolism to uncover novel biomarkers and enhance integrated multi-omics strategies for AML diagnosis and prognosis.

The transition of metabolomics from research to clinical practice is becoming increasingly feasible due to technological advancements, improved data analysis tools, and growing recognition of its diagnostic and prognostic potential. However, several challenges persist, including the need for standardized protocols, reproducibility across laboratories, and regulatory validation. Integration into clinical workflows will require simplified, automated platforms and clear clinical interpretation guidelines. Despite these challenges, the success of pilot studies and emerging clinical trials suggests that metabolomics, especially in diseases like AML, holds significant promise for clinical translation in the future.

Conclusion

It is concluded that significant metabolomic changes in AML are mainly observed in lipid metabolism, amino acid metabolism, and glycolysis. Also, AML patients unveiled remarkably reduced serum cholesterol and total lipid levels, elevated acetate, reduced branched-chain amino acids, and persistent glucose and pyruvate disruptions post-treatment. Metabolomics profiles, including acetate, creatine, and lactate, revealed strong diagnostic accuracy (AUC = 0.98), while the panel of citrate, glutamate, and choline offered solid prognostic potential (AUC = 0.89). These results emphasize the value of metabolomics profiling as a potent modality for establishing AML diagnosis and predicting treatment outcomes, paving the way for enhanced patient care through directed and tailored approaches.

Limitations

The current study has some limitations:

  1. 1.

    The lack of an independent external validation cohort is one of the study limitations, which limits how broadly the results may be applied.

  2. 2.

    The limited investigation of the underlying biochemical pathways is another drawback. This could limit the scope of biological interpretation and the mechanistic comprehension of the reported metabolomic changes.

  3. 3.

    In spite of the use of rigorous internal validation procedures like permutation testing and cross-validation, these approaches are not able to completely replace the robustness offered by external replication in a variety of populations.