Genomic aberrations and transcription signatures to profile metabolism

A first layer of understanding the role of metabolism in disease biology is offered by the detection of genetic mutations in metabolic enzymes and other related molecules. Genomic analysis is an integral part of the diagnostic workup in acute leukemias, although the vast majority of affected genes are unrelated to metabolism. Nevertheless, with the emergence of next-generation sequencing (NGS) and primarily through projects such as the cancer genome atlas (TCGA),⁷² BeatAML,⁷³ and BeatAML2.0,⁷⁴ researchers were able to identify IDH1 and 2 mutations as recurrent aberrations in AML.⁷⁵ The neomorphic activity of mutant IDH1 and 2 leads to the production of the oncometabolite 2-HG, which supports leukemia progression by epigenetic remodeling.^{14, 15, 17} To interfere with this pathway, small molecule inhibitors of mutant IDH1 (ivosidenib) and IDH2 (enasidenib) were successfully developed and FDA-approved in 2018⁷⁶ and 2017.⁷⁷ While mutations in IDH1 and 2 have a clear impact on disease biology by generating an epigenetically active oncometabolite, they remain, to date, the only therapeutically relevant genetic aberrations in metabolic enzymes in acute leukemias.

Genomic analysis by PCR has been applied to quantify mitochondrial DNA (mtDNA) content in leukemic cells.^{78, 79} Changes in mtDNA content can reflect alterations in mitochondrial biogenesis, function, and metabolic adaptation in cancer cells. After the isolation of genomic DNA from cell samples, mtDNA content is determined by a quantitative real-time PCR using primers for a mitochondrial gene (e.g., cytochrome B, CYTB).⁷⁹ The relative amount of mtDNA can be defined as a ratio between the expression of this gene and the expression of a nuclear-encoded gene (such as pyruvate kinase, PKLR, or hemoglobin subunit beta, HBB).⁷⁹ As mitochondrial metabolism is a central therapeutic vulnerability in AML,⁸⁰ and AML cells often exhibit metabolic plasticity, increased mtDNA content could indicate a higher reliance on OXPHOS and mitochondrial respiration for energy production. Given this, patients with elevated mtDNA levels may be particularly sensitive to mitochondria-targeted therapies.

NGS is further employed to characterize transcriptional programs. Bulk RNA-sequencing data from large cohorts of patient samples are available through TCGA,⁷² the BeatAML,⁷³ and BeatAML2.0 trials,⁷⁴ the Therapeutically Applicable Research to Generate Effective Treatments (TARGET) AML projects (https://www.cancer.gov/ccg/research/genome-sequencing/target) for pediatric leukemia, and others.^81-83 While these investigations provide helpful information about the expression of metabolic genes across a large number of samples, they reflect mRNA levels in an ill-defined mixture of cells, frequently comprising both malignant and benign cell populations from peripheral blood and/or bone marrow. Advances in the single-cell RNA sequencing (scRNA-seq) field now allow the assessment of metabolic signatures at the single-cell level.⁸⁴ Individual cells are captured using methods like microfluidics or fluorescence-activated cell sorting. Cells are lysed to capture mRNA, and reverse transcription is performed to generate cDNA using unique molecular identifiers and cell barcodes that later allow the distinguishing of transcripts from different cells. The cDNA is amplified and prepared into sequencing libraries that are subjected to NGS. During data analysis, cells are clustered based on the similarity of their transcriptional profiles. ScRNA-seq offers the advantage of studying the transcriptional metabolic states at a single-cell level and in rare populations that might be too small to be interrogated by other methods. It can also be combined with cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq), a method in which oligonucleotide-labeled antibodies are used to identify surface proteins with sequencing as a read-out.⁸⁵ Furthermore, there is an increasing number of commercially available platforms for spatial transcriptomics, facilitating a spatially resolved assessment of gene expression.⁸⁶ Nevertheless, while genomic NGS is broadly established in clinical practice, scRNA-seq and spatial transcriptomics applications are currently limited to research settings due to the required substantial expertise in sample preparation and data analysis, and the relatively high cost.

Taken together, genomic and transcriptomic information delivers the first layer to understanding the biological relevance of metabolism (Figure 3). One major limitation, in particular of gene expression studies, lies in the fact that mRNA levels of a specific transporter or enzyme do not necessarily correlate with the protein expression or actual metabolic activity of a pathway. A significant part of metabolism regulation occurs at the level of enzyme activity and metabolite abundance, making technologies that detect these parameters indispensable.

Protein expression analysis

Metabolism-related protein expression adds another layer to the characterization of cellular metabolic activity. While traditionally, the expression of nutrient transporters and enzymes has been analyzed by immunoblotting, multiple novel techniques have been developed, including mass spectrometry (MS)-based proteomics, flow cytometry, and cytometry by time-of-flight (CyTOF) (Figure 4). These methods are complementary to each other as they all have distinct benefits and limitations concerning coverage, sensitivity, bulk versus single-cell analysis, and spatial resolution.

MS and, in particular, tandem mass spectrometry (MS/MS) is an analytical tool that allows the simultaneous identification and quantification of thousands of proteins, metabolites, or lipids in complex biological samples. Inside a mass spectrometer, electrostatic and electromagnetic fields are used to guide and separate analyte molecules. Therefore, these molecules must be electrically charged. Proteins are typically digested to peptides using proteases such as trypsin, whereas metabolites and lipids are mostly analyzed in their native form. For the analysis of biofluids and extracts of blood cells or cultured cells, chromatographic separation before MS is typically used for improved sensitivity and specificity. Liquid chromatography (LC) is a highly versatile technique that can be used for any soluble compound. For studying protein, metabolites, and lipids, electrospray ionization (ESI) is almost exclusively used to couple LC to MS. Spatial separation can be seen as an alternative to chromatographic separation. Matrix-assisted laser desorption ionization (MALDI) and desorption ESI (DESI) are examples of ionization techniques that allow for spatially resolved analysis of proteins or small molecules in tissue slices and are typically coupled to high-resolution mass spectrometers. This technology is also called imaging MS because images of the distribution of metabolites or lipids in tissue can be generated.^{87, 88} In comparison, chromatography-coupled MS allows higher confidence in the identification of analytes, whereas imaging MS delivers the additional spatial information.

Once inside the mass spectrometer, molecules are separated by their mass-to-charge ratio (m/z), and their relative abundance is counted in a detector. Of note, one molecule usually generates multiple characteristic m/z signals corresponding to different isotopologues (the molecule containing different number of heavy isotopes), different adducts (the molecule forming a transient complex with strong ions such as sodium, potassium, ammonium, chlorine, formate, or acetate), or different fragments (if the molecule breaks during ionization). Correctly assigning the different m/z signals to analytes is one of the challenges of MS-based analysis. Different MS analyzers exist, such as quadrupoles, ion trap, and time-of-flight (TOF) instruments, and their use depends mostly on the specific application.⁸⁹ Non-targeted MS experiments aim at identifying and (relatively) quantifying all proteins, metabolites, or lipids in a biological sample, while targeted approaches such as selected reaction monitoring quantify pre-defined analytes of specific interest.

While the majority of MS proteomics data to date have been generated from bulk cells, there have been significant developments in the field of single-cell proteomics by MS (scMS) in recent years.^90-92 The extent of coverage (about 1500 proteins/cell) and the relatively low throughput speed (up to several hundred cells per day) are major limitations of the current use.^{90, 93} Nevertheless, we expect that as technologies advance, scMS will gain importance in profiling metabolic states in single cells. Finally, imaging-based spatial proteomics platforms such as co-detection by indexing (CODEX),⁹⁴ imaging mass cytometry (IMC),⁹⁵ multiplexed ion beam imaging (MIBI),⁹⁶ and others, offer first insights into spatially resolved proteomic information within tissues at the single-cell level.

Another technique allowing the quantification of metabolism-related proteins is flow cytometry. Flow cytometry is a standard method in hematological diagnostics laboratories used to analyze the expression of cell surface and intracellular molecules. Single-cell suspensions from cell culture or tissue samples are labeled with fluorescence-coupled antibodies or fluorescent compounds before being injected in the flow cytometry instrument. The cell suspension is focused by the sheath fluid so that one cell at a time passes through a laser beam. Initially, forward scatter (FS) and side scatter (SS) are measured, reflecting cell size and granularity. Subsequently, fluorescence detectors measure the emitted fluorescence from stained cells, thus allowing quantification of the amount of labeling. Classic examples of metabolic markers studied by flow cytometry include the fatty acid transporter CD36, the amino acid transporter CD98, the glucose transporter 1 (GLUT1), and the transferrin receptor 1 (CD71). One particular challenge hampering the use of flow cytometry in metabolic research is the paucity of antibodies able to detect the expression of metabolite transporters and metabolic enzymes reliably. A recent development in this area is Met-Flow, a high-parameter flow cytometry method utilizing antibodies against ten essential metabolic proteins that are critical and rate-limiting in their respective pathways⁹⁷ (Table 1). Anabolic flux can be analyzed by measuring FA synthesis and levels of an arginine metabolism protein.⁹⁷ In contrast, catabolic flux analyses include the quantification of proteins involved in glycolysis, the PPP, TCA cycle, OXPHOS, and FAO.⁹⁷ Advantages of flow cytometry include the broad availability of instruments in clinical diagnostic units, the single-cell resolution with high throughput (up to 10,000–20,000 cells/second), and the relatively simple analysis procedure. We expect further developments to address the scope of antibodies and enable studies of additional pathways.

Cytometry by Time-of-Flight (CyTOF) represents another powerful tool in protein expression analysis, enabling high-dimensional single-cell analysis using metal isotope-labeled antibodies to detect proteins. Compared to traditional flow cytometry, the usage of metal isotype-tagged antibodies has only minimal signal overlap and allows the simultaneous measurement of up to 60 markers per cell.⁹⁸ After being stained with tagged antibodies, single-cell suspensions are introduced into a mass cytometer. They are first nebulized into small droplets containing one cell per droplet. These droplets are then introduced into the inductively coupled plasma where they are vaporized, atomize, and ionized, breaking the cellular contents down into a cloud of metal ions.⁹⁹ The resulting ion cloud per cell is then passed to the time-of-flight analyzer (TOF), which measures the relative abundance of each metal isotope mass correlating to the expression level of the targeted proteins.¹⁰⁰ Similar to flow cytometry, CyTOF is suitable to perform single-cell analyses of patient cells and tissues, including blood. General sample handling as well as staining protocols are comparable. However, CyTOF has the significant benefit that it uses metal isotope-labeled antibodies instead of fluorophore-labeled antibodies, which eliminates the need of fluorescence compensation and the occurrence of spectral overlap.¹⁰⁰ However, the throughput with CyTOF is lower compared to flow cytometry with less than 1000 cells/second.

Metabolite quantification

Genomic mutations, transcriptional signatures, and protein expression provide valuable information about the metabolic state of a cell. However, metabolic reactions are highly dynamic, and their reaction rates depend not only on the availability of enzymes but also on their subcellular localization, post-translational modifications, and concentration of substrate metabolites. Critical information is provided by directly quantifying metabolite concentrations in cells and biological fluids. Metabolomics analyses covering a broad range of metabolites are routinely performed with two methods: nuclear magnetic resonance (NMR) spectroscopy and MS. The latter is typically coupled to liquid chromatography (LC-MS) or gas chromatography (GC-MS) for better sensitivity and specificity. Additionally, individual metabolites can be quantified by flow cytometry or enzymatic assays with colorimetric, fluorometric, or bioluminescent read-outs. While NMR, MS, and enzymatic assays can be employed in the analysis of cells/tissues, cell culture supernatants, and biofluids, the use of flow cytometry is limited to cells and tissues.

In NMR spectroscopy, the spectra of active nuclei, very frequently ¹H, are recorded to determine intra-molecular proton ratios and infer information about the molecules. NMR spectroscopy has several advantages, such as being a highly reproducible, easily quantifiable method that requires little to no sample treatment or chromatographic separation. These advantages render it a highly suitable technique for large-scale or multi-center studies.¹⁰¹ Nevertheless, significant disadvantages limit the use of NMR spectroscopy compared to MS-based methods. Most importantly, NMR spectroscopy has a lower separation power, leading to a significantly (10–100×) lower sensitivity compared to MS in complex samples such as biofluids or cell extracts. Consequently, it yields information only for a lower number (typically tens) of highly abundant metabolites.^102-104 MS can be used not only for the detection of proteins (see “Protein expression analysis” section) but also to detect and quantify small polar and non-polar (lipid) metabolites. In addition to imaging MS and LC-MS, gas chromatography (GC) coupled to MS is suitable for volatile compounds, small polar compounds (up to about 200 Da, such as amino acids, TCA intermediates, glycolytic intermediates), and fatty acids. For GC, electron impact (EI) ionization is commonly used as the ion source for MS coupling. For a detailed overview of MS metabolomics sample preparation, measurement, and data analysis, we refer to the following sources.^105-107 The great physiochemical diversity of metabolite classes and their large dynamic range present a significant challenge in fully describing the metabolome in a single sample. Hence, adapted methods for sample extraction, chromatography, and MS have been developed to preferentially capture specific metabolite or lipid classes. Metabolomics analyses can be performed either in an untargeted or a targeted manner. In untargeted metabolomics, all MS signals meeting particular criteria are recorded in an unbiased manner, and annotation to specific molecules is attempted at the analysis stage by matching the exact mass, the chromatography retention time, and potentially other parameters to those of known molecules and/or standard compounds. Untargeted metabolomics is used as a discovery approach as it holds the potential to identify previously unsuspected metabolic changes. However, only a fraction (typically 5%–10%) of the recorded MS signals can indeed be annotated to a known molecule. In targeted metabolomics, a list of metabolites of interest is predefined, and signals matching the characteristics of these metabolites are recorded. This method frequently offers a higher sensitivity and confidence in the identification. However, as the selection of metabolites for analysis is based on prior knowledge, this can introduce a bias and result in overlooking critical metabolic pathways that were previously unknown. Finally, metabolomics analyses can capture not only the steady-state levels of metabolites but also the activity of metabolic pathways. The latter is achieved by stable-isotope tracing. Synthetically produced molecules containing, for instance,¹³C (instead of ¹²C) or ¹⁵N (instead of ¹⁴N) atoms (e.g., glucose, amino acids, and many more) can be fed to cells, and the incorporation of ¹³C/¹⁵N atoms into downstream metabolites provides information about the relative activity of the respective metabolic pathways.¹⁰⁸ In a basic stable-isotope tracing experiment, isotope labeling can be analyzed by both NMR and MS.¹⁰⁹ Labeling can be performed not only by adding the stable isotope tracers to cell culture media, but also by infusing them to living organisms, which provides information about dynamic metabolic flux rates in vivo.¹⁰⁹ While the majority of in vivo stable-isotope tracing studies have been done in animal models, infusion of stable-isotope tracers has also been done in humans since the 1930s, when Hevesy (later awarded the Nobel Prize in Chemistry for the development of radioactive tracers) used deuterium oxide to estimate the size of the whole-body water pool and the rate of water elimination.¹¹⁰

In practice, the acquisition of high-quality MS data remains a challenge for non-expert laboratories. Many metabolic reactions occur very rapidly, making it hard to capture the in vivo metabolic state precisely. Cold and quick handling of the samples is critical to maintaining metabolic states as close as possible to the in vivo situation. Moreover, MS provides information on a bulk cell population and, at least for classical approaches, a relatively high number of cells (>100,000 is required). Nevertheless, new developments in the field allow us to perform metabolomics from rare cell populations, including HSCs, with an input as low as 5000 cells.^{111, 112}

MS imaging and MALDI-TOF, as methods allowing to perform spatial metabolomics, have undergone substantial advances in the last few years, but their application to the hematological field is impeded by several significant obstacles. As metabolites are prone to rapid degradation and transformation, spatial metabolomics relies on obtaining fresh snap-frozen tissue without preceding incubation or washing step. However, classical bone marrow biopsies require decalcification to enable tissue sectioning. Moreover, metabolite identification remains a considerable challenge.⁸⁷ Further improvements in the MS imaging field will be necessary to allow spatial metabolomics from the bone marrow. Such advances will be highly informative, given the complexity and heterogeneity of cell populations in the bone marrow niche and our current lack of understanding of the metabolic cross-talk between them.

Besides the high-resolution technologies of NMR spectroscopy and MS, flow cytometry-based methods for the detection of some metabolite classes have been developed. For instance, ROS can be detected by incubating cells with cell-permeant dyes (such as CM-H2DCFDA, CellROX [ThermoFisher Scientific], and others) that are weakly fluorescent while in a reduced state and exhibit bright, photostable fluorescence upon oxidation by ROS.¹¹³ Another example of probes enabling metabolite quantification by flow cytometry are BODIPY dyes—chemicals containing a boron difluoride group and a dipyrromethene group, which strongly absorb UV light and re-emit it in very narrow frequency spreads.¹¹⁴ BODIPY derivatives have been used in the context of leukemia for the quantification of neutral lipids, fatty acids, and lipid peroxidation.¹¹⁵ Some ions can also be quantified by flow cytometry, such as iron, calcium, and others.

Finally, the abundance of individual metabolites can be quantified in enzymatic assays coupled to a colorimetric, fluorometric, or bioluminescent read-out. In these assays, the concentration of a metabolite of interest is quantified by a chemical reaction yielding a product for which absorption, fluorescence, or bioluminescence signal intensity can be measured and is proportional to the concentration of the metabolite of interest. Such assays exist for some common metabolites, such as glucose, ATP, lactate, glutamine, glutamate, glutathione, creatine, creatinine, and others. While the implementation is simple and requires only standard laboratory equipment, a relatively high amount of input material is necessary due to the comparatively low sensitivity of these assays. Another disadvantage is that the majority of these assays can measure the abundance of only one to two metabolites per sample.

In conclusion, measuring metabolite abundance is now possible via NMR spectroscopy, MS, flow cytometry, and enzymatic assays (Figure 5). While MS-based bulk and spatial metabolomics is the most sensitive method, it requires specialized equipment and expertise in sample preparation and data interpretation. For this reason, MS metabolomics is predominantly used in basic research. Flow cytometry and enzymatic assays are easier to implement but cover a much narrower range of metabolites. A combination of these technologies delivers broad information on steady-state metabolite content, metabolic flux, and spatial distribution.

Functional read-outs

Functional read-outs provide information about dynamic cellular metabolic processes and build the final dimension of metabolism studies. Enzymatic assays are one classical approach in which the activity of an enzyme can be assessed by measuring the conversion of a specific substrate, for instance, using a spectrophotometric or a fluorometric assay. More recent technologies allow the monitoring of the activity of whole metabolic pathways, such as glycolysis or OXPHOS. Here, we discuss the use of extracellular flux assessments and flow cytometry for functional metabolic characterization (Figure 6). Both of these technologies are suited to exploring the metabolic activity of cells or cell suspensions obtained from tissues. Furthermore, we discuss the possibility of measuring metabolism using the OmniLog analyzer and tracking metabolic functions on the organismal level in mice using metabolic cages.

The extracellular flux assay is now a standard method for evaluating cellular metabolic states, both in vitro and ex vivo. Seahorse real-time cell metabolic analysis (Seahorse XF Analyzer, Agilent) provides two key real-time measurements: glycolytic activity, assessed by the extracellular acidification rate (ECAR), which reflects lactic acid and bicarbonate accumulation in the extracellular environment, and mitochondrial respiration, measured by the oxygen consumption rate (OCR), which indicates extracellular oxygen levels.¹¹⁶ To perform a typical extracellular flux assay, cells are plated in a monolayer onto specialized cell culture microplates. The other integral part of the assay is the proprietary solid-state biosensor cartridge. The biosensor cartridge contains two fluorophores: one is quenched by oxygen, and the second one is sensitive to protons (hydrogen protons H⁺, providing information on glycolysis). The biosensor cartridge plate is placed onto the cell culture microplates containing the cells to be analyzed. Inside the instrument, the fiber optics of the instrument emit light, which can excite the embedded fluorophores. The sensors can then detect the changes in fluorescence, which are a result of the change in the total levels of oxygen and hydrogen protons. During the measurement cycle, the sensor cartridge creates a transient microchamber in each well of the cell microplate. Through this process, the Seahorse XF analyzer can detect real-time changes in oxygen consumption or hydrogen production by the analyzed cells within minutes.¹¹⁷ The Seahorse assay further offers the opportunity to measure the dynamic metabolic changes in cells upon drug challenge. The cartridge contains four compound injection ports, and drugs plated in these ports can be added to the cells at selected time points.¹¹⁸ In the case of a classical Mitochondrial Stress test, four drugs are injected: oligomycin, which inhibits the ATP synthase (ETC complex V), carbonyl cyanide-p-trifluoromethoxyphenylhydrazone (FCCP), a proton gradient uncoupler, and a combination of rotenone and antimycin A, which inhibit ETC complex I and III, respectively. By observing the changes in OCR, parameters such as basal respiration, maximal oxygen consumption, spare respiratory capacity, and non-mitochondrial OCR can be calculated and give information about the mitochondrial respiration activity and capacity of cells.^{117, 118} Similarly, the Glycolysis Stress test uses the addition of glucose, oligomycin, and the glycolysis inhibitor 2-deoxyglucose (2-DG) to investigate the glycolytic rate of the cell.¹¹⁸ With this assay, key parameters such as the basal glycolytic rate, the glycolytic capacity (ECAR, following oligomycin treatment), glycolytic reserve (as the difference between basal glycolysis and maximal glycolytic capacity), and non-glycolytic acidification (following 2-DG, injection, indicative of proton production from non-glycolytic sources) can be quantified. These quantifications give important information regarding the cells' capacity to fulfill energy requirements via glycolysis and their metabolic flexibility in response to stress.¹¹⁸

Besides determining rates of respiration and glycolysis, bioenergetic profiling of cells with the Seahorse XF analyzer can also be utilized to probe cells for their specific preference of different fuels or substrates in the Mito Fuel Flex assay. It is designed to measure how cells utilize key mitochondrial fuels—glucose (pyruvate), glutamine (glutamate), and long-chain fatty acids—by measuring the OCR in response to specific inhibitors. These include UK5099 (which inhibits glucose oxidation by blocking the mitochondrial pyruvate carrier MPC), BPTES (glutamine oxidation inhibitor), and etomoxir (FAO inhibitor). By sequentially blocking these pathways, the assay determines fuel dependency (how reliant cells are on a specific fuel), capacity (the ability to use a fuel when alternatives are inhibited), and flexibility (the ability to switch between fuels). A lack of flexibility indicates a strong reliance on a particular pathway to sustain basal mitochondrial respiration.¹¹⁹

Additionally, it is also possible to analyze respiration in isolated mitochondria from different cells and tissues, such a brain, heart, liver, and others.^{120, 121} Apart from investigating overall respiration in isolated mitochondria, one can also assess the activity of single ETC complexes by providing specific substrates for these complexes, even on frozen tissue material.¹²² As rates of oxygen consumption by cells can be influenced by oxygen diffusion into and out of the transient microchamber, it is furthermore possible to conduct the Seahorse XF assays in hypoxia chambers (with ≥3% oxygen).^{121, 123} Extracellular flux assays provide a dynamic insight into the real-time metabolism of cells with a moderately high cell input and can be flexibly adjusted to address different scientific questions by varying the chemical compounds the cells are treated with. Disadvantages of the technology include the limited insight into individual metabolic pathways, the necessity to obtain a homogenous cell population, and the lack of single-cell resolution.

Another assay type focuses on mitochondrial metabolism by employing fluorescent dyes, which can be visualized by flow cytometry or microscopy. Fluorescently labeled positively charged molecules (for instance, MitoTracker^TM, ThermoFisher Scientific) are attracted to the negative charge of the mitochondria and covalently bind to healthy mitochondria components in living cells. Such fluorescent probes can be used to investigate mitochondrial mass, distribution, and morphology by flow cytometry or microscopy. In contrast to this covalent binding, another fluorescent dye based on the tetramethylrhodamine methyl ester (TMRM) can be used to investigate changes in the mitochondrial membrane potential specifically. This cationic dye binds in a reversible and directly proportional to the mitochondrial membrane potential manner. Therefore, it can be used as a dynamic real-time indicator of mitochondria health and function (e.g., in drug response treatments).¹²⁴ Besides functional readouts concerning overall mitochondrial health, another crucial aspect of mitochondrial metabolism is the production of mitochondrial superoxide generated as a byproduct of OXPHOS. With the help of cell-permeable fluorescent dyes, which are designed to accumulate in the mitochondria selectively and to react there with the produced superoxide to produce a fluorescent signal, oxidative stress can be monitored.^{113, 125, 126} The advantages of these technologies include that they require a low cell number, offer a single-cell resolution, and allow to study heterogeneous cell mixtures when cell type-specific antibodies are added.

Single-cell energetic metabolism by profiling translation inhibition (SCENITH) is an emerging flow cytometry-based assessment of metabolic activity with single-cell resolution.¹²⁷ This technology is based on the fact that the energy state of a cell is closely reflected by its protein synthesis rate. By measuring protein translation (via quantifying the incorporation of puromycin), researchers can gain information about the metabolic state of a cell. In the original study, the authors treated cell suspensions with 2-deoxyglucose (2-DG) to inhibit glycolysis, oligomycin to inhibit the ATP synthase, or a combination of both.¹²⁷ By quantifying protein synthesis via flow cytometry, four functional metabolic parameters were calculated.¹²⁷ Glucose dependence was estimated as the proportion of protein synthesis, and thus energy production, dependent on glucose oxidation. Mitochondrial dependence was quantified as the proportion of protein synthesis dependent on OXPHOS activity. Glycolytic capacity was defined as the maximum capacity to sustain protein synthesis when mitochondrial OXPHOS is inhibited. Fatty acid and amino acid oxidation capacity was defined as the capacity to use fatty acids and amino acids as sources for ATP production when glucose oxidation is completely inhibited.¹²⁷ By combining protein translation measurement with a panel of flow cytometry antibodies, SCENITH allows gaining information about the metabolic state at the single-cell level even of rare cell subpopulations in a heterogeneous suspension.¹²⁷ Another advantage is that only a relatively small cell number is required and that the read-out can be done on any flow cytometer. Nevertheless, similar to extracellular flux assays, SCENITH provides an overview of the energetic capacities and dependencies of a cell without offering an insight into individual metabolic pathways. More recently, CENCAT (cellular energetics through noncanonical amino acid tagging) was introduced¹²⁸ as an alternative to SCENITH. While SCENITH measures protein translation by quantifying puromycin incorporation,¹²⁷ puromycin can be toxic via the induction of ER stress.¹²⁹ In CENCAT, the authors proposed using click labeling of alkyne-bearing noncanonical amino acids to quantify protein synthesis without potential cellular toxicity.¹²⁸

The Biolog OmniLog phenotype microarray analyzer is a fully automated high-throughput microarray system that is designed to perform metabolic profiling mainly in microorganisms. Using a 96-well plate-based colorimetric detection system, it can assess the utilization of different sources, including carbon and nitrogen, and, therefore, measure the cellular respiration and metabolic activity. This technology detects changes in cellular respiration by measuring the reduction of a tetrazolium-based dye, which generates a colorimetric signal corresponding to metabolic activity.¹³⁰ So far, this technique has been used to study antibiotic susceptibility or nutrient utilization in microorganisms,^130-132 and has not been employed for functional metabolic investigations of hematopoietic and leukemic cells.

While these aforementioned techniques are suited to study metabolism ex vivo, metabolic cages can provide information about in vivo metabolic functions of rodents at the broader organismal level. Metabolic cages are designed in a way that food and water intake are measured, and urine and feces samples are collected separately. Additional parameters, such as O₂ consumption and CO₂ production, can also be recorded in some systems. While metabolic cages are used in pharmacodynamic and pharmacokinetic studies (such as Brunner et al.¹³³), the specifics of this tool have been shown to elicit stress and alter physiological metabolism.^{134, 135} Animal welfare considerations further limit the use of metabolic cages.

Altogether, extracellular flux assays and flow cytometry-based metabolic analyses are common approaches to probe cell metabolism upon metabolic stress. By measuring the cellular responses to metabolic inhibition, we can obtain information about the native metabolic cell state.

Functional, proteomic, and genomic metabolism studies identify mitochondrial metabolism as a vulnerability in AML

Mitochondrial metabolism represents one of the best-characterized metabolic vulnerabilities in AML, as evidenced by multiple studies combining different technologies. Lagadinou et al. found that human primary leukemic stem cells (LSCs)-enriched populations were metabolically dormant with low levels of OCR and ECAR compared to AML blasts from the same patients, as assessed by Seahorse extracellular flux assays.¹³⁶ Interestingly, LSC-enriched populations also had a significantly decreased capacity to upregulate the glycolytic machinery when mitochondrial respiration was blocked with oligomycin.¹³⁶ Moreover, the authors showed that LSCs expressed high levels of BCL-2 and that BCL-2 inhibition decreased OCR and induced cell death.¹³⁶ These data are in line with an earlier work by Skrtic et al., in which tigecycline, an inhibitor of mitochondrial translation, reduced OCR in the human AML cell line TEX, and selectively killed LSCs compared to their nonmalignant counterparts.¹³⁷ In another study, Sriskanthadevan et al. showed that AML cells had low spare respiratory capacity (by Seahorse extracellular flux assays), in particular with a low reserve in ETC complexes I, III, IV, and V.¹³⁸ These studies represent early examples of the utilization of the extracellular flux assay technology to discover metabolic profiles specific to AML cells. OXPHOS upregulation has also been described as a mechanism of chemotherapy resistance. Farge et al. utilized 25 patient-derived xenografts (PDX) to study the metabolic profile of residual leukemic cells (RLCs) after cytarabine treatment.⁵⁷ Interestingly, RLCs had an increased mitochondrial mass (estimated by MitoTracker flow cytometry staining) and mitochondrial membrane potential (by TMRM labeling).⁵⁷ Transcriptome analysis confirmed a high OXPHOS signature in RLCs recovered from cytarabine-treated compared to vehicle-treated PDX mice.⁵⁷ Furthermore, leukemic cell lines with low and high mitochondrial ATP production and OCR were injected in immunodeficient mice, which were treated with cytarabine. While low OXPHOS cell lines had a high sensitivity to cytarabine, high OXPHOS cell lines showed resistance.⁵⁷ These and other studies paved the way for multiple translational efforts aiming at targeting mitochondrial metabolism in AML.^{139, 140}

Interestingly, AML patients show heterogeneity in their mitochondrial metabolism, which might be related to disease outcome. Jayavelu et al. performed proteogenomic analyses of 252 AML samples, in which they integrated information from protein expression, gene expression, mutation, and cytogenetic profiles.¹⁴¹ They identified five proteomic AML subtypes, including the high-risk Mito-AML, which was only detectable in the proteome but not by analyzing genomic mutations or transcriptional signatures.¹⁴¹ Characterized by high expression of mitochondrial proteins, this subtype conferred poor patient outcomes, reduced remission rate, and shorter overall survival upon treatment with induction chemotherapy.¹⁴¹ The authors could demonstrate that this Mito-AML subtype was hypersensitive to mitochondrial complex I inhibitors, highlighting a potential therapeutic target.¹⁴¹ In another study, quantitative proteomics revealed a higher expression of ETC complex I and V proteins in human LSCs compared to AML blasts and healthy hematopoietic stem and progenitor cells.¹⁴² More recently, a proteomic analysis of 47 adult and 22 pediatric patients with AML at serial time points during disease progression found higher expression of mitochondrial ribosomal proteins and ETC complex subunits at relapse compared to primary diagnosis.¹⁴³ These studies give examples of the use of proteomics to define metabolic features and subtypes of disease and support the notion that mitochondrial metabolism defines aggressive AML subgroups.

The quantification of mtDNA has been proposed as a relatively easy-to-measure proxy of mitochondrial activity. Elevated mtDNA content was first reported by Boultwood et al. in patients with AML (n = 25) and chronic myeloid leukemia (n = 9).⁷⁸ More recently, in a cohort of n = 502 de novo AML patients, the authors identified a subset of patients with increased mtDNA content (n = 163) compared to an age- and sex-matched healthy control cohort.¹⁴⁴ Notably, high mtDNA was associated with worse clinical outcomes and reduced sensitivity to cytarabine treatment ex vivo.¹⁴⁴ Mechanistically, AML blasts with increased mtDNA content displayed increased mitochondrial mass and membrane potential measured by flow cytometry along a proteomic signature indicative of a higher ETC complex I activity. Blocking complex I activity with metformin reduced the mtDNA content of AML blasts in vitro (n = 40), synergized with cytarabine, venetoclax, and FLT3 inhibitors in vitro, and prolonged the survival of K562-bearing leukemic mice.¹⁴⁴ While high mtDNA content appeared detrimental in this setting, the opposing effects have been observed in acute promyelocytic leukemia (APL).⁷⁹ Overall, these studies suggest that the quantification of mtDNA might allow the identification of specific biological disease subtypes. Further studies will be required to disentangle the biological importance and therapeutic consequences of high mtDNA in distinct disease types.

Identification of metabolic vulnerabilities in AML by metabolomics

The advances in MS-based metabolomics have opened up significant opportunities in translational medicine, including the characterization of disease states, the identification of new biomarkers, and targeted drug development. While we discuss only few examples of studies based on metabolome analysis of primary human AML cells here, a detailed summary of published datasets is provided in Table S1.

Several landmark studies uncovering metabolic vulnerabilities in AML with the use of LC-MS have been published by the Craig Jordan lab. For instance, global metabolic profiling from 15 primary AML patients found that 16 amino acids were enriched in LSCs compared to blasts.¹⁴⁵ LC-MS tracing experiments with stable isotope-labeled amino acids revealed significantly faster uptake, in particular of glutamine and glutamate, in LSCs.¹⁴⁵ Functional experiments showed that limiting amino acid uptake selectively targeted LSCs by impairing their OXPHOS.¹⁴⁵ In another study, primary AML cells were cultured ex vivo in media depleted of individual amino acids.¹⁴⁶ Depletion of arginine, glutamine, and especially cysteine reduced cell viability of both blasts and LSCs.¹⁴⁶ Cysteine depletion led to impaired glutathione synthesis with subsequently reduced glutathionylation of succinate dehydrogenase A (SDHA), a key component of ETC complex II.¹⁴⁶ As a consequence, cysteine-starved AML cells had reduced OXPHOS activity, lower ATP rates, and higher cell death.¹⁴⁶ Characterizing changes in metabolite abundance also offers insights into the therapeutic mechanisms of action of anti-leukemic drugs. For instance, MS analysis of AML patient samples collected pre- and 24 h post-treatment with azacitidine and venetoclax showed significant decreases in glutathione and the TCA intermediates αKG, malate, and fumarate, and an increase in succinate.¹⁴⁷ These metabolic perturbations suppressed OXPHOS and specifically targeted LSCs, building the foundation for the clinical efficacy of azacitidine and venetoclax.¹⁴⁷ Further studies investigated the metabolic profiles of refractory or relapsed AML. Global metabolomics analysis of LSCs isolated from n = 6 patients with de novo AML and n = 6 patients with relapsed/refractory (R/R) AML after treatment with venetoclax/azacitidine showed distinct metabolic profiles in LSCs from patients with R/R AML.¹⁴⁸ In particular, 10 amino acids had significantly higher abundance in R/R samples compared to de novo AML, supporting the hypothesis that venetoclax/azacitidine acts by disrupting amino acid metabolism, and elevating amino acid levels might contribute to therapy resistance.¹⁴⁸ Interestingly, LSCs from R/R AML patients also had higher levels of the NAD⁺ precursor nicotinamide.¹⁴⁸ Targeting nicotinamide phosphoribosyltransferase (NAMPT), the rate-limiting enzyme in nicotinamide metabolism, selectively eradicated R/R LSCs.¹⁴⁸ Similarly, the upregulation of FAO induced resistance to azacitidine/venetoclax.¹⁴⁹ Inhibition of FAO by deletion or pharmacological blockade of the fatty acid transporter CPT1A reduced viability and OXPHOS of LSCs from patients with venetoclax/azacitidine-resistant primary AML.¹⁴⁹ Interestingly, high FAO rates were correlated with RAS mutations in some patients, underscoring the link between oncogenic mutations and metabolic profiles.¹⁴⁹ Another such example is ceramide metabolism. Dany et al. found higher levels of ceramides in human AML blasts from patients with FLT3-ITD mutations compared to patients with wild type FLT3.⁶⁸ Furthermore, multiple other studies have also demonstrated that recurrent oncogenic mutations and karyotype aberrations elicit specific metabolic profiles in leukemic cells or patient plasma, which might be exploited as therapeutic vulnerabilities.^150-152 These are examples of how MS metabolomics contributes to a better understanding of disease biology and treatment responses.

While the vast majority of studies has been performed ex vivo, evaluating dynamic metabolic functions by stable-isotope tracing (see “Metabolite quantification” section) in vivo offers the advantage of obtaining information about metabolic processes as they occur in the complex physiological or pathophysiological environment of the living organism. Stable-isotope tracing studies for cancer have mostly been performed in patients with solid tumors and were recently reviewed.¹⁰⁹ Notably, one study performed intravenous infusion of ¹³C glutamine into healthy volunteers (n = 7), patients with monoclonal gammopathy of undetermined significance (n = 11), and patients with multiple myeloma (n = 12).¹⁵³ The participants underwent bone marrow biopsy 60 minutes after the infusion, and CD138⁺ (plasma) cells as well as CD138^- cells (remainder of bone marrow mononuclear cells) were sorted and analyzed by GC-MS or LC-MS.¹⁵³ The authors observed a higher ¹³C fractional enrichment of TCA cycle metabolites in malignant CD138⁺ (relative to CD138⁻ cells from the same patient) compared to premalignant CD138⁺ plasma cells.¹⁵³ While stable-isotope tracing studies provide very valuable information, several caveats and challenges need to be accounted for.¹⁰⁹ For instance, tracer infusions can measure nutrient contributions to metabolites but usually rely on a single biopsy time point, making it impossible to quantify actual metabolic fluxes. Accounting for tumor heterogeneity, potentially by implementing technologies with spatial resolution, and the development of computational frameworks to integrate the generated data are further priorities for the field.¹⁰⁹

Inferring metabolic states from transcriptional data

While metabolomics analyses are frequently possible only for expert laboratories, platforms for transcriptome profiling, as well as extensive public datasets, are more readily available. Thus, it is of significant interest whether metabolic states can be inferred from transcriptional profiles. While the activity of metabolic pathways cannot be measured by the expression of individual genes, computational and machine-learning methods are being developed.

One excellent example is a recent study on sphingolipid metabolism in AML.¹⁵⁴ Sphingolipid metabolism has been recognized as a metabolic vulnerability in AML (see “Metabolic regulation in healthy and diseased hematopoietic cells” section). In this study, Paudel et al. quantified 33 sphingolipid metabolites in 213 primary AML samples, 30 human AML cell lines, and 6 healthy CD34⁺ bone marrow samples by MS.¹⁵⁴ Within AML samples, the authors found two clusters: cluster 1 with proportionally less hexosylceramides (Hex) and more sphingomyelins (SM) and cluster 2 with proportionally more Hex and less SM. Patients with a Hex^lowSM^high profile had twice the failure rate after induction therapy of patients with a Hex^highSM^low profile. Transcriptomic analysis by bulkRNA-sequencing was performed for 33 primary AML samples and 30 cell lines. Based on 60% this cohort, the authors developed a vector machine stratifier of Hex-SM profiles using the 284 most variable and differential genes. The classifier showed a predictive performance on the remaining 40% of the cohort with an AUC of 0.91. Furthermore, when patients from the TCGA⁷² and BeatAML2.0⁷⁴ cohorts were stratified for their Hex-SM profiles based only on transcriptional data, patients inferred as Hex^lowSM^high had significantly worse outcomes than those predicted to be Hex^highSM^low. Interestingly, Hex-SM profiles were not predictable by genetic mutations.¹⁵⁴

Several studies highlight possibilities for defining metabolic states of immune cells from single-cell transcriptome analyses.^{155, 156} For instance, Fernandez-Garcia et al. could define metabolic states of CD8⁺ T cells as cells transitioned through the activation/differentiation cascade by single-cell RNA-seq.¹⁵⁵ The authors used gene set variation analysis (GSVA) to determine per-cell pathway activity scores in the transition from an unactivated to an activated state.¹⁵⁵ This approach identified several well-known hallmarks of T cell metabolism, such as a substantial increase in aerobic glycolysis¹⁵⁷ and amino acid¹⁵⁸ uptake upon activation. Importantly, the study also uncovered novel aspects of T cell metabolism by recognizing a role for asparagine synthetase expression in effector T cell differentiation, function, and anti-tumor responses.¹⁵⁵ Moreover, Wagner et al. developed Compass, an algorithm that characterizes cellular metabolic states based on single-cell transcriptomics and employed it as a tool to uncover metabolic signatures in non-pathogenic and pathogenic Th17 cells.¹⁵⁶ Efforts to model metabolic states from single-cell transcriptomics have been recently reviewed by Hrovatin et al.¹⁵⁹

Flow cytometry-based assessments of metabolism and their application in hematology

Flow cytometry has versatile applications for the assessment of metabolic profiles. It can be used to assess the expression of metabolic proteins (see “Protein expression analysis” section), to measure metabolites, such as fatty acids or ROS (see “Metabolite quantification” section), and to evaluate functional metabolic properties, such as mitochondrial membrane potential and glucose/mitochondrial metabolism via SCENITH (see “Functional read-outs” section). The broad availability of flow cytometers and the necessary reagents as well as the relative simplicity of the analysis and the high throughput, make these techniques widely used tools in hematological and immunological research.

For instance, primary AML samples showed substantial heterogeneity in their redox staining profile, with a ROS-low fraction characterized by quiescence, lower cell cycle activity, and higher engraftment capacity in mouse xenograft models compared to a ROS-high AML cell fraction from the same patients.¹³⁶ ROS-low AML cells also maintained engraftment capacity in serial transplantation experiments, identifying them as LSC, compared to the blast-enriched ROS-high population.¹³⁶ This observation allowed a relatively simple identification and isolation strategy for the LSC-enriched fraction in primary patient samples. Furthermore, SCENITH was utilized to describe the metabolic signature of LSCs in AML.¹⁶⁰ In an analysis of CD34⁺ samples from n = 29 AML patients, Forte et al. observed that CD34⁺ cells relied predominantly on glucose oxidation as measured by high glycolytic capacity and glycolytic dependence, and low mitochondrial dependence and FAO/amino acid oxidation capacity.¹⁶⁰ However, the metabolism of the more immature CD34⁺CD38^low/− fraction was marked by an increase in mitochondrial dependence, in line with the notion that mitochondrial metabolism represents a therapeutic vulnerability in AML.¹⁶⁰ SCENITH-derived metabolic profiles also showed a correlation with risk groups according to the ELN 2022 classification.¹⁶¹ For example, samples from the favorable group had a lower mitochondrial dependence (8.2%) compared to samples from the intermediate and adverse risk groups (22.3% and 23.5%, respectively).¹⁶⁰ Altogether, these studies demonstrate how flow cytometry-based assessments of metabolism can be implemented to define hematopoietic cell subset biology but also as a non-invasive, easy-to-implement approach for metabolic risk stratification in hematological cancers. We expect that the development of new antibodies and compounds will facilitate the studies of individual metabolic pathways and increase the scope of these studies.

Definition of disease and risk prognosis by biofluid metabolomics

Several studies have attempted to characterize the serum metabolome of patients with AML and other hematological diseases (Table S1). Both NMR spectroscopy and LC-MS have been utilized in these studies. NMR spectroscopy was used to measure metabolites in 116 human serum samples, including patients with AML, non-Hodgkin's lymphoma, chronic lymphocytic leukemia, and healthy controls.¹⁶² A total of n = 50 metabolites were assigned, and their concentrations in the respective disease group were used to infer the metabolic profiles associated with a specific disease type. Interestingly, more aggressive malignancies (such as AML) caused more profound changes in the serum metabolome.¹⁶² Significant differences in the serum metabolome were also found in a study of 183 patients with de novo AML and 232 age- and sex-matched healthy controls.¹⁶³ In particular, differences in glycolysis/gluconeogenesis, TCA cycle, biosynthesis of proteins and lipoproteins, and metabolism of fatty acids and cell membrane components were observed.¹⁶³ Similar results were obtained in an NMR spectroscopy analysis of serum samples from n = 32 patients each with AML, ALL, aplastic anemia, and healthy controls.¹⁶⁴ Simonetti et al. investigated the serum and urine metabolome from >100 patients with AML and healthy controls.¹⁶⁵ They found that both the serum and the urine metabolome (assessed by NMR spectroscopy) provided efficient discrimination between patients and healthy controls with an accuracy of 83% and 85%, respectively, and the integration of serum and urine data yielded an even higher (90%) accuracy. Thirteen metabolites were significantly altered in the biofluids of AML patients, compared to controls, including elevated levels of lactate, polyunsaturated fatty acids, pyruvate and succinate in AML patients and reduced glutamine, citrate, glycine, and creatinine. Importantly, the authors also leveraged analysis of intracellular metabolites in AML cells (CD34⁺ and CD33⁺), normal umbilical cord blood CD34^+, and normal peripheral blood CD33⁺ cells by LC-MS. By integrating intracellular and biofluid metabolomics, they observed particularly prominent aberrations in polyamine, purine, ketone bodies, and polyunsaturated fatty acid metabolism in AML. Moreover, metabolome-based clusters showed enrichment for specific mutations, such as NPM1. Within NPM1-mutated samples, two subgroups were identified. One was enriched for mutations in cohesin/DNA damage-related genes and had increased serum choline + trimethylamine-N-oxide and leucine. This study highlights how the integration of metabolome data from cells and biofluids with genomic and transcriptomic analyses can define disease subgroups that might be particularly susceptible to metabolic pathway targeting.¹⁶⁵ Additional studies exploring the metabolome of biofluids of patients with hematological diseases are referenced in Table S1.

CyTOF and spatial proteomics uncover immune cell metabolic profiles based on protein expression

Similar to flow cytometry, CyTOF allows the quantification of the expression of proteins related to metabolism in combination with analysis of phenotypic identity at the single-cell level. CyTOF offers the advantage of minimal signal overlap, generally allowing for a higher number of analyzed parameters within the same panel. Hartmann et al. combined CyTOF with spatial analysis by multiplexed ion beam imaging by time-of-flight (MIBI-TOF) to establish single-cell metabolic regulome profiling (scMEP) with the analysis of metabolic phenotypes in conjunction with spatial organization.¹⁶⁶ Initially, the authors assessed the performance of over 110 commercially available antibodies by immunohistochemistry, CyTOF, and MIBI-TOF. Of these, they selected a panel of 41 antibodies covering glycolysis, amino acid metabolism, fatty acid metabolism, the PPP, the TCA cycle and ETC, mitochondrial dynamics, biosynthesis, signaling, and transcription. Next, they studied the scMEP of human CD4⁺ and CD8⁺ T cells upon in vitro activation and benchmarked those against classical read-outs such as extracellular flux assays. Mass cytometry-based scMEP recapitulated multiple hallmarks of T cell metabolic remodeling, such as increased glycolysis upon activation (as measured by increased expression of the glucose transporter GLUT1, hexokinase 2, phosphofructokinase 2, and lactate dehydrogenase A). Simultaneously, TCA/ETC proteins such as citrate synthase and ATP synthase were elevated, and this was in line with increased levels of basal OXPHOS measured using the Seahorse analyzed. Notably, the authors could transfer the technology to spatial assessments with MIBI-TOF and analyzed the spatial architecture and heterogeneity of immune cells in colon tissue from patients with colorectal carcinoma (n = 4) and non-malignant control sections from independent healthy donors (n = 3) using a panel of 36 antibodies.¹⁶⁶ Similarly, Levine et al. employed a metabolic CyTOF panel covering T cell phenotypic markers, signaling molecules, nutrient transporters, and enzymes involved in the TCA cycle, ETC, and FAO to study the metabolic profiles of mouse CD8⁺ T cells during activation and memory development.¹⁶⁷ Effector T cells showed an elevated expression of GLUT1, GAPDH, and the amino acid transporter CD98, compatible with an activated glycolytic signature, while memory T cells upregulated the expression of the fatty acid transporter CPT1A. Furthermore, using L. monocytogenes infection as a model of in vivo T cell activation, the authors identified a group of early activated T cells that comprised up to 20% of the total T cell population and had simultaneously the highest expression of GLUT1 and GAPDH, but also of oxidative markers such as CTP1A and ATP5a.¹⁶⁷ Taken together, these technologies illustrate that CyTOF-based metabolic profiling recapitulates known metabolic processes in activated immune cells and allows to address questions such as population heterogeneity, which cannot be captured by extracellular flux assays, bulk proteomics, or metabolomics. The combination with imaging-based technologies also paves the way to a better understanding of immune cell metabolism heterogeneity in complex tissues.

Metabolic signatures in allogeneic HSC transplantation

While the majority of studies we have discussed so far concern hematological diseases outside of an allogeneic hematopoietic stem cell transplantation (allo-HCT) setting, several studies could demonstrate the translational relevance of metabolic profiles in shaping and predicting patient transplantation outcomes. A recent study was performed on pre-transplantation serum samples derived from 92 consecutive allo-HCT recipients with AML or MDS.¹⁶⁸ Using LC-MS non-targeted metabolomics, the authors could identify significant differences in metabolites in the recipients, which could be correlated to the occurrence of severe transplantation-related mortality (TRM). Comparing recipients who suffered from TRM-associated factors (early extensive fluid retention, pre-transplantation inflammation, and aGVHD) to those without, they could identify significant differences in overall metabolism. Specific metabolic changes could be associated with each of the three TRM-associated factors mentioned earlier, as evidenced by increased levels of amino acid metabolites, altered levels of xenobiotics and lipid metabolites, specifically in the group of patients who developed aGVHD. The authors could further identify patient subsets with an increased risk of 2-year TRM, using unsupervised hierarchical cluster analyses to subclassify allotransplant recipients based on their metabolomic profiles.¹⁶⁸ These findings suggest that integrating metabolomic analysis into clinical practice could enhance patient monitoring and improve long-term transplant success rates.

Other examples concerning the role of metabolomics (also as potential predictive biomarkers) in transplantation outcomes include studies on the occurrence of chronic GVHD (cGVHD) after allo-HCT. Research on serum and plasma metabolites comparing samples from patients with or without cGVHD could demonstrate significantly different levels of metabolites, mainly belonging to amino acid and lipid metabolism, establishing associations between the patients' metabolome and post-transplant outcome.^169-171 A similar association could be seen investigating serum pre-transplant profiles of patients with or without post-transplant capillary leak syndrome.¹⁷² These studies collectively suggest a potential for using distinct metabolic signatures as predictive biomarkers for early diagnosis, risk assessment, and treatment monitoring. More studies and detailed information on this topic can be obtained from Table S1.

Besides using systemic metabolomics as predictive biomarkers, recent studies also highlight how immune-related metabolic alterations can influence post-transplant outcome. Studying fecal samples after allo-HCT and early administration of azithromycin (an antibiotic shown to increase relapse risk), a complex network of bacterial taxa and metabolites, among others, associated with relapse occurrence and remission status could be identified.¹⁷³ Furthermore, distinct enterotypes could be associated with the status of exhausted T cells, showing how an antibiotic can potentially influence the microbiome and modify immune cells, and thereby promote disease relapse.¹⁷³ Similarly, the accumulation of oxidative DNA damage measured by increased concentrations of 8-hydroxydeoxyguanosine (8-OHdG) in serum and reconstituting T cells after allo-HSCT was linked to increased metabolic stress, leading to premature exhaustion of the T cells and impaired anti-leukemic activity, which ultimately increased the risk of relapse and reduces survival.¹⁷⁴ Taken together, these findings highlight the role of both systemic and immune-related metabolic alterations in post-transplant outcomes and how they can potentially be used as biomarkers or therapeutic targets for improving the outcomes of allo-HCT recipients.